Konstantin K.

Likharev
Essential Graduate Physics
Lecture Notes and Problems

Exercise Problems
with Model Solutions

Part EM:
Classical Electrodynamics

Supplement to the open-access materials available online at mirror sites

http://commons.library.stonybrook.edu/egp/
https://essentialgraduatephysics.org/
https://sites.google.com/site/likharevegp/

DISCLAIMER:
These solutions are provided without warranty.
They have not been fully verified by independent instructors, and may contain unintended errors/typos.

Last corrections: May 28, 2024

© K. Likharev 2013-2024; open access under the http://creativecommons.org/licenses/by-nc-sa/4.0/ license

Essential Graduate Physics EM: Classical Electrodynamics

Table of Contents

Chapter 1. Electric Charge Interaction…………….…….…………………………...……. 3

Chapter 2. Charges and Conductors…………………………….…………………………. 22

Chapter 3. Dipoles and Dielectrics………………….……………….…………………….. 99

Chapter 4. DC Currents……………... ………………….……………………………….…137

Chapter 5. Magnetism……………………………………………………………….………159

Chapter 6. Electromagnetism……………….…………………………………………….…200

Chapter 7. Electromagnetic Wave Propagation……………….………………..…………. 250

Chapter 8. Radiation, Scattering, Interference, and Diffraction………………….………... 313

Chapter 9. Special Relativity……………...…………………………………..……..….…. 354

Chapter 10. Radiation by Relativistic Charges……………………….………………..…... 405

Problems with Solutions Page 2

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 1. Electric Charge Interaction

Problem 1.1. Calculate the electric field of a thin, long, straight filament, electrically charged
with a constant linear density , by using two approaches:
(i) directly from the Coulomb law, and
(ii) from the Gauss law.
Solutions:
(i) From the translational and axial symmetries of the dE cos
problem, we can conclude that E(r) = nE(), where  is the 
dE
shortest distance between the observation point and the
filament, i.e. its 2D radius.1 Let us select the plane of (ρ 2  z 2 )1 / 2
drawing so it contains both the filament and the observation ρ 
point, and take the line of filament for the z-axis – see the
figure on the right. Then, according to the linear 0 dz z
superposition principle, the field’s magnitude may be calculated as
z   z   z  
ρ
E d    dEρ   dE cos   dE ,
z   z   z   ρ 2
 z2 
1/ 2

where dE is the magnitude of the elementary contribution to the field, created by a small segment dz of
the filament, with the electric charge dz. According to Eq. (1.7) of the lecture notes,
1 1
dE  dz ,
4 0 ρ  z 2
2

so the total field2


ρ dz   d 
E ρ    ρ    . (*)
4 0 
2
 z2 
3/ 2
2 0 ρ 0 1   
2 3/ 2
2 0 ρ

(ii) Taking a round cylinder of radius  and length l, with its axis on the filament, for the
Gaussian volume, we ensure that on its round walls, the electric field E() is constant and normal to the
volume’s boundary, and the field flux through the cylinder’s “lids” is zero. As a result, Eq. (1.16) yields
l
2 ρ lE ρ   ,
0
immediately giving the same result (*).

1 Let me hope that the difference between the fonts used for the 2D radius  and the volumic density  of the
electric charge is sufficient to avoid confusion. (Similar notation is used in all solutions where both these notions
are encountered.)
2 See, e.g., MA Eq. (6.5c). Actually, this integral may be easily worked out by the substitution   tan , giving

d = d/cos2  d (1 + tan2)  d (1 + 2), so d/(1 + 2)3/2 = d/(1 + 2)1/2  cos d  d(sin ).

Problems with Solutions Page 3

Essential Graduate Physics EM: Classical Electrodynamics

We see that for this highly symmetric problem, both approaches are straightforward, but the
Gauss law makes calculations simpler.

Problem 1.2. Two thin, straight, parallel filaments separated by distance d

carry equal and opposite uniformly distributed charges with a linear density  –   
see the figure on the right. Calculate the force (per unit length) of the Coulomb
interaction of the filaments. Compare its functional dependence on d with the
Coulomb law for two point charges, and interpret their difference. d

Solution: Using the result of Problem 1 with the notation replacement   d, and Eq. (1.6) of the
lecture notes, we get
F qE (d ) 2
  E ( d )  .
l l 2 0 d
So, the force drops with distance as 1/d, rather than as 1/r2 for point charges. This difference may
be interpreted just as Eq. (1.23) was interpreted in Sec. 1.2 of the lecture notes: the farther is one of the
filaments from a fixed elementary charge of the counterpart filament, the more elementary charges dQ =
dz are “visible” to it under a modest angle to the normal direction, thus mitigating the drop of the field
induced by each of them. Such different scaling of the total magnitude of the same interaction in systems
of different dimensionality is very typical for physics at large.

Problem 1.3. Calculate the electric field of the following spherically symmetric charge
distribution: (r) = 0exp{–r}.
Solution: The solution of this problem is essentially given by Eq. (1.20) of the lecture notes:
Qr
E r   ,
4 0 r 2
so what remains is just to spell out the charge Qr inside the sphere of the radius r:
r r
Qr    r'  d r'  4   r' r' dr'  4 0  exp r' r' dr'.
3 2 2

r' r 0 0

Perhaps the easiest way to calculate the last integral is to notice that it may be represented as the second
partial derivative of another (elementary) integral over the parameter :
r r
2 2  1  2   2 r 2  
 exp r' r' dr'  2 0
                exp r ,
2
exp r' dr' 1 exp r 1
 3   1 r
0 2     2  
so, finally:
0   2 r 2  
E r   1
   1   r   exp r .
 0 3 r 2   2  
Note that at large distances (r >> 1/), the field decreases as 1/r2, i.e. as that of a point charge,
because of a fast (exponential) drop of the charge density (r).

Problems with Solutions Page 4

Essential Graduate Physics EM: Classical Electrodynamics

Problem 1.4. A sphere of radius R, whose volume had been charged with a constant density , is
split with a very narrow planar gap passing through the sphere’s center. Calculate the force of mutual
electrostatic repulsion of the resulting two hemispheres.
Solution: Since the gap is very narrow, we may neglect its effect on the distribution of the total
electric field E inside the sphere, and use Eq. (1.22) of the lecture notes:
r
Er   n r E (r ), with E r   ,
3 0
where r is the distance of the observation point from the sphere’s center 0 – see the figure on the right.
Acting on an elementary volume dV = d3r of the sphere’s material, with the elementary charge dQ =
d3r, this electric field produces a radially-directed force of magnitude
z
r  2r 3
dF  E r dQ  d r 
3
d r.
3 0 3 0
d 3r
 r
Due to the axial symmetry of the problem, the net (repulsive)  R
Coulomb force F acting on each hemisphere has to be normal to the gap’s
plane – in the figure on the right, has to be directed along the z-axis. Hence 0
only the z-component of the elementary force dF,
 2r
dFz  dF cos   cos  d 3 r ,
3 0
where  is the polar angle of the radius vector r of the elementary volume d3r (see the figure above),
can contribute to the net force F. As a result, we may use the standard spherical coordinates, with the
origin in the sphere’s center, to calculate the net force magnitude as
2  /2
2 R
2 1 R4   2R4
F  Fz   dFz   d  cos sin  d  r dr  2 
3
.
z 0
3 0 0 0 0
3 0 2 4 12  0

Another way to represent this result is to express it via the charge Q = (2/3)R3 of each hemisphere:
1 3 Q2 1 Q2 2
F  , with Ref  R,
4 0 4 R 2 4 0 Ref2 3
showing that the effective distance between two point charges Q with the same interaction is Ref  1.155
R – a very reasonable value.
One may worry whether the above calculation properly excludes the effect of the electric field of
a hemisphere on itself, since the field we have used is produced by the sphere as a whole. A proper
response to this concern is that due to the 3rd Newton law, the internal forces between elementary
charges of the same hemisphere compensate each other, so these forces are automatically canceled at the
integration over the hemisphere’s volume:

 E
z 0
full d 3r    E
z 0
other  E self  d 3 r   E
z 0
other d 3r  0   E
z 0
other d 3r  F .

Problems with Solutions Page 5

Essential Graduate Physics EM: Classical Electrodynamics

Problem 1.5. A thin spherical shell of radius R, which had been charged with a constant areal
density , is split into two equal halves with a very thin planar cut passing through the sphere’s center.
Calculate the force of the mutual electrostatic repulsion of the resulting hemispheric shells, and compare
the result with that of the previous problem.
Solution: A very thin cut does not alter substantially the electric field distribution, so it has the
same radial direction and spherically-symmetric distribution E(r) = nrE(r) as for the uncut shell. The
function E(r) may be readily found using the Gauss law applied to spheres with r < R and r > R:
 0, for r  R,
E r   
 R /  0 r , for R  r.
2 2

Hence the force dF applied to an elementary area dA of the shell is normal to it, and F
its magnitude equals3
E R  0   E R  0  2 dF
dF  E dQ  dA  dA .

2 2 0
Due to the axial symmetry of the problem, only the “vertical” components
(meaning the direction normal to the cut plane – see the figure on the right) of these
elementary forces may contribute to the total force F acting on each hemisphere –
for example, the top one:
2  /2
 dF  2 dF 2   2R2
F z 0  dA  z d r  z0 dA cos d r  R 0 d  cos  sin  d 
2 2
. (*)
0
2 0 2 0

For a fair comparison of this result with the solution of the previous problem, let us assume that
the radius R and the total charge Q of each hemisphere are the same in both cases:
2 3
2R 2  R  Q.
3
In this case, Eq. (*) becomes
1 1 Q2 1 Q2
F  , with Ref  2 R  1.414 R ,
4 0 2 R 2 4 0 Re2f
while for a volume-distributed charge, the similarly defined effective distance is close to 1.155R. This
difference is natural, because in the case of thin hemisphere shells, the elementary charges of the
counterpart shells are, on average, farther from each other.

Problem 1.6. Calculate the spatial distribution of the electrostatic potential created by a straight
thin filament of a finite length 2l, charged with a constant linear density , and explore the result in the
limits of very small and very large distances from the filament.

3 A strict proof of the correctness of such field averaging will be given in the model solution of Problem 2.1.

Problems with Solutions Page 6

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Due to the limited (only axial) symmetry of the problem, z, z'
applying the Gauss law to it is not very productive, so let us resort to the direct l
summation (actually, integration) of the component charge fields. Let us select r
z
the reference frame so that the filament coincides with segment [-l, +l] of the z-
axis, and the observation point r has Cartesian coordinates {, 0, z} – see the r'
figure on the right. Then the field source point r' has coordinates {0, 0, z'}, and 0 ρ
Eq. (1.38) of the lecture notes, integrated across the filament’s cross-section,
reads
l
 l dz'  l dz'
 r  
4 0 l r  r' 4 0 l  z  z' 2  ρ 2 1 / 2

 .

By introducing the dimensionless variable   (z' – z)/, we may reduce the integral to a table one:4
( l  z ) / ρ ( l  z ) / ρ
 d 
 r   ln  2  1  
1/ 2

4 0   2
1 
1/ 2

4 0 ( l  z ) / ρ
( l  z ) / ρ
(*)

 
l  z   ρ 2

2 1/ 2
 l  z 
4 0
ln

l  z 2  ρ 2  1/ 2
 l  z 
.

For observation points very close to the filament and not very close to any of its ends, the
denominator of the fraction is much smaller than its numerator. Expanding these expressions in small
parameters /(l  z) and keeping only the leading terms, we get the result obtained in the solution of
Problem 1:
 2l  z  
 r   ln 2  ln ρ  f z , for z  l , and ρ  l , l  z .
4 0 ρ / 2l  z  2 0
On the other hand, at large distances from the filament, the numerator and denominator of the
fraction approach each other. Expanding both expressions in the Taylor series in the small parameter l/r,
where r  (z2 + 2)1/2 is the 3D distance from the filament’s middle point, in the first nonvanishing
approximation we get
 1 l / r   2l   2l
 r   ln  ln1    , for r  l ,
4 0 1  l / r 4 0  r  4 0 r
i.e. Eq. (1.35) for the point charge q = 2l.

Problem 1.7. A thin planar sheet, perhaps of an irregular shape, carries an electric charge with a
constant areal density .
(i) Express the electric field’s component normal to the plane, at a certain distance from it, via
the solid angle  at which the sheet is visible from the observation point.
(ii) Use the result to calculate the field in the center of a cube with one face charged with a
constant density .

4 See, e.g., MA Eq. (6.2a).

Problems with Solutions Page 7

Essential Graduate Physics EM: Classical Electrodynamics

Solutions:
dE dE z
(i) Let us place the coordinate origin 0 at the charged plane’s point z
that is closest to the field observation point (at a distance z from the plane) –
see the figure on the right. Then, according to Eq. (1.7) of the lecture notes,
r
the normal (z-) component of the electric field component induced by charge d 2 ρ' 
element dQ = d2 (where  is the 2D radius vector within the plane) equals
 
1 dQ  d ρ 2
0 ρ
dE z  dE cos   cos   cos  , (*) d2ρ
4 0 r 2
4 0 r 2
where r = (2 + z2)1/2 is the distance between the elementary area d2 and the observation point, while
cos = z/r. But as the figure above shows, the product d2 cos equals d2’ – the projection of the
elementary area d2 on the plane normal to the vector r. In turn, the ratio d2’ /r2 is just d – the solid
angle at which the elementary area d2 is visible from the field measurement point.5 As a result, Eq. (*)
may be rewritten simply as

dE z  d ,
4 0
and its integration over the charged sheet yields the requested result:

Ez  . (**)
4 0
(ii) Since all six faces of a cube are visible from its center at equal angles , and their sum has to
be equal to the full solid angle 4, each  is equal to 4/6, and Eq. (**) yields
 4 
Ez   .
4 0 6 6 0
Due to the obvious symmetry of the system, the field cannot have other Cartesian components, so we
may also write

E nz .
6 0

Problem 1.8. Can one create, in an extended region of space, electrostatic fields with the
Cartesian components proportional to the following products of the Cartesian coordinates {x, y, z}:
(i)  yz, xz, xy ,
ii   xy, xy, yz ?
Solution: Let us calculate the curl of both supposed fields, using the definition of that operator:6
 E E y E x E z E y E x 
E   z  ,  ,  .
 y z z x x y 

5 Note that this calculation is similar to the one made in the proof of the Gauss law in Sec. 1.2 of the lecture notes
– see the transition from Eq. (1.13) to Eq. (1.14).
6 See, e.g., MA Eq. (8.5).

Problems with Solutions Page 8

Essential Graduate Physics EM: Classical Electrodynamics

For field (i), we get E  {x – x, y – y, z – z}  0, while for field (ii), E  {z – 0, 0 – 0, y –
x} vanishes only on one line: x = y, z = 0. However, according to the homogeneous Maxwell equation
(Eq. (1.28) of the lecture notes), the curl of the electrostatic field has to equal zero at any point where it
exists; hence field (i) can exist in a region of finite size, while field (ii) cannot.
The fact that field (i) has zero divergence as well, i.e. requires (r)  0 within the region of its
existence, does not prevent it from being realistic, because the field may be created by electric charges
outside of that particular region. Note also the electric field (ii) may be induced by a certain magnetic
field changing (linearly) in time – see Sec. 6.1 of the lecture notes, but such E cannot be called an
electrostatic field – and this was the term used in the assignment.

Problem 1.9. Distant sources have been used to create different uniform electrostatic fields in
two half-spaces:
E , at z  0,
Er  r  R  n z    z
 E , at z  0,
except for a transitional region of scale R near the origin, where the field is E 
perturbed but still axially symmetric. (As will be discussed in the next
chapter, this may be done, for example, using a thin conducting membrane
with a round hole of radius R in it – see the figure on the right.) Prove that 2R
such field may serve as an electrostatic lens for charged particles flying along E 
the z-axis, at distances  << R from it, and calculate the focal distance f of
this lens. Spell out the conditions of validity of your result.
Solution: Let us select the orientation of the mutually perpendicular axes x = cos and y =
sin so that for the particular flying particle, y = 0. At the axis of this axially-symmetric system (i.e. at
 = 0), the partial derivatives Ex/x and Ey/y have to be equal due to the symmetry, so the Maxwell
equation (1.27), E = 0, yields
E x E z
2 ρ 0  ρ 0 .
x z
Since at the axis, Ex has to vanish due to the same symmetry, the Taylor expansion of the function Ex(x,
y) at x = 0 starts with the terms linear in x and y, so at small x and y = 0:
E x x E z
Ex x  R x ρ 0  ρ 0 .
y 0 x 2 z
According to this expression, at sufficiently small x, the field Ex is small, so the particle’s
deviation from its initial trajectory during its flight through the perturbed field region (of the length z ~
R) is negligible. However, during this flight, the particle picks up the transverse momentum created by
the field:

dt x E z x
p x   Fx dt  q  E x dt  q  E x dz   q  ρ  0 dz   q E   E   .
dz 2v z  z 2v z
(Here the change of particle’s longitudinal velocity vz during its flight through the hole was neglected –
which is legitimate if the length of the uniform field region is much larger than R, as implied by the
problem’s conditions.) If this momentum, and hence the acquired radial velocity vx = px/m, are directed

Problems with Solutions Page 9

Essential Graduate Physics EM: Classical Electrodynamics

toward the axis, which is true at the proper sign of the combination q(E+ –E-)/vz,7 it would lead to the
particle’s crossing the system’s axis at the z-distance
x x 2mv z2 4T
f  vz  vz   ,
 v x   p x / m  qE   E   qE   E  
where T = mvz2/2 is the particle’s kinetic energy. The most important feature of this result is that the
initial coordinate x of the particle has dropped out of it. This means that all particles with any small x
(and, by the axial symmetry, all particles flying sufficiently close to axis z and parallel to it) will be
directed to the same focal point f. This is the key property of any lens – in this case, the electrostatic one.
It is remarkable that this result does not depend on the exact distribution of the field in the
transitional region. (For a thin-membrane implementation shown in the figure above, this distribution
may be calculated analytically – for example, using the oblate spheroidal coordinates to be discussed in
Sec. 2.4 of the lecture notes.) The result is valid if two strong conditions,
T
R  f  ,
qE 
are satisfied. The first of them has allowed us to treat x as a constant during the integration of the force
Fx over time, while the second one, to neglect the change of vz during the focusing process. Note,
however, that a violation of the second condition would not ruin the focusing effect; it would only make
the expression for f more involved and dependent on the specific distribution of the field at z ~ f.

q r3
Problem 1.10. Eight equal point charges q are located in the corners of a
cube of side a. Calculate all Cartesian components Ej of the electric field, and
their spatial derivatives Ej/rj’, in the cube’s center, where rj are the Cartesian r2
coordinates oriented along the cube’s sides – see the figure on the right. Are all
0
of your results valid for the center of a planar square, with four equal charges at r1
its corners? a
Solution: Per Eq. (1.33) of the lecture notes, the Cartesian components of
the field and their derivatives may be expressed via spatial derivatives of the electrostatic potential:
 E j  2
Ej   ,  . (*)
r j rj' r j r j'
Due to the cube’s symmetry, at its center, the whole vector E should vanish. (Indeed, if it did not, the
vector would be directed toward either some face or some corner of the cube, but that would violate the
equivalence of all faces and corners.) Hence, according to the first of Eqs. (*), all field components, i.e.
the partial derivatives of the scalar potential have to vanish at the center. So, if we take this point for the
origin (r = 0, i.e. r1 = r2 = r3 = 0), the Taylor expansion8 of the function (r1, r2, r3), besides an arbitrary
constant (0, 0, 0), should start with quadratic terms. Due to the cube’s symmetry with respect to
coordinate swaps, the coefficients at these terms have to be independent of the coordinates:

7 If the sign is opposite, the system disperses the parallel particle beam, but in a highly ordered way – with the
distance from the axis proportional to the initial distance x. Such a system still may work as a lens, though a
diverging (“negative”) one, with f < 0.
8 See, e.g., MA Eq. (2.11b).

Problems with Solutions Page 10

Essential Graduate Physics EM: Classical Electrodynamics

A 3 2 B
 
3
  r1 , r2 , r3     0, 0, 0   r j  r r j j'  O r j3 , (**)
2 j 1 2 j , j '1
j ' j

where A and B are constants:

 2 E j  2 E j
A 2 r 0  r 0 , while B  r 0  r 0 for any j  j' . (***)
r j r j r j r j' r j'

But because of the system’s symmetry, the potential has to be also invariant with respect to the
change of sign of any single coordinate, rj  –rj, which geometrically corresponds to its mirror
reflection in the common plane of two other coordinate axes.9 As Eq. (**) shows, this is only possible if
B = 0. Moreover, since there is no charge in the vicinity of the cube’s center, the potential has to satisfy
the Laplace equation (1.42), in the Cartesian coordinates reading
3
 2

j 1 r j
2
0. (****)

Comparing this requirement with the first of Eqs. (***), we get A = 0 as well, so all the derivatives
Ej/rj’ at r = 0 have to equal zero.
For the field in the center of a plane square, with four similar charges in its corners, the situation
is somewhat different, due to the reduced symmetry. The electric field E at the center (and hence all its
Cartesian components) still has to equal zero, because its in-plane component (say, E12) cannot be
directed toward any side of the square, and its normal component E3 cannot be directed to either side of
the plane, due to the r3  –r3 symmetry. However, the potential may not remain the same if the
coordinate r3 directed normally to the square’s plane is swapped with one of the in-plane coordinates (r1
or r2). As a result, Eq. (**), still with B = 0 due to the mirror symmetry, has to be generalized as

 r1 , r2 , r3    0,0,0  
2

A 2
 A'
r1  r22  r32  O r j4 ,
2
 
and the Laplace equation (****) imposes only the following requirement: 2A + A’ = 0, i.e.
E1 E 2 1 E 3
  .
r1 r2 2 r3
This relation is similar to the one used for the solution of Problem 9 (where it was true due to the
axial symmetry of the system).

Problem 1.11. By a direct calculation, find the average electric potential of a spherical surface of
radius R, created by a point charge q located at a distance r > R from the sphere’s center. Use the result
to prove the following general mean value theorem: the electric potential at any point is always equal to

9Actually, because of this symmetry, the higher terms of the expansion (**) should start with O(rj4), rather than
O(rj3).

Problems with Solutions Page 11

Essential Graduate Physics EM: Classical Electrodynamics

its average value on any spherical surface with the center at that point while
containing no electric charges inside it. r q

Solution: Using the evident axial symmetry of the problem (see the figure on
the right) and Eq. (1.35) of the lecture notes, we get: r'
 
1 1 1 q
 ave 
4   ( )dΩ 
4
2   ( ) sin  d  
0
2 0
4 0 r'
sin  d , 
R
0
where r’ is the distance between the considered point charge and the observation
point:
(r' ) 2  R 2  r 2  2 Rr cos  .
The integral may be readily worked out by the introduction of a new variable   cos (so that sin d
= –d):
1
d
 ave 
1 q

2 4 0 1 R 2  r 2  2 Rr 1 / 2

1 q 1
2 4 0 r' Rr
R 2  r 2  2 Rr   11
1/ 2


1 q 1
2 4 0 Rr

R 2  r 2  2 Rr 
1/ 2

 R 2  r 2  2 Rr 
1/ 2
   4q 0
r  R  r  R 
2 Rr

q 1 / r , for R  r ,
 
4 0 1 / R, for r  R.
We see that for our case r > R, the average potential of the sphere indeed coincides with the
potential’s value in its center. Now the proof of the mean value theorem becomes straightforward by
using the linear superposition principle: since the relation in question,
q
  ave , for r  R,
4 0 r
holds for each point charge located outside the sphere, it is also true for any system of such charges.

Problem 1.12. Two similar thin, circular, coaxial disks of radius R, R 

separated by distance 2d, are uniformly charged with equal and opposite 2d
areal densities  – see the figure on the right. Calculate and sketch the
distribution of the electrostatic potential and the electric field of the disks 
along their common axis.
Solution: Let us start by calculating the electrostatic potential of one
of the disks, with a constant areal charge density , at the axial point z
separated by distance z from the disk’s plane – see the figure on the right. r
In the polar coordinates {, } within the plane of the disk (with angle    
referred to an arbitrary horizontal axis}, the elementary disk area is dd, 0 R
and its electric charge is dd, so Eq. (1.38) of the lecture notes takes
the following form:

Problems with Solutions Page 12

Essential Graduate Physics EM: Classical Electrodynamics

2
 R
ρdρ
 z    d  ,
4 0 0 0
r
where r is the distance between the elementary charge’s location and the observation point on the disks’
common axis. As the figure above shows, r = (2 + z2)1/2, so the function under the integral is
independent of , and it may be easily worked out:

d ρ2  z2 
 
ρR
 R
  
 z   2 
ρ dρ
  ρ   ρ2  z 2 
1/ 2 ρR
 R2  z2   z .
1/ 2

0 ρ  z  
4 0 2 2 1 / 2
4 0 ρ 0
2
z 2 1/ 2 2 0 ρ 0 2 0

Now by using this formula and the linear superposition principle, we may readily write down the
expression describing the potential created by both disks, at a distance z from the center of the system
(in this new reference frame, the disk center positions are d):
 2 
  R  z  d 
2
1/ 2
 zd 
 1
  .
2 0  2

 R   z  d  
2 1/ 2
 z  d  0.5
R/d  5
1
 )
This function is plotted in the figure on the right for 0 .2
three different values of the R/d ratio. For small values of  d /  0 0
this ratio, the potential is clearly separated into two peaks,
of opposite polarity, created by each disk. On the other  0.5
hand, at R >> d, the result tends to the one for two infinite
planes, with  between the disks being a linear function of
z, with the slope corresponding to the electric field – see the 1
2 1 0 1 2
model solution of Problem 15 below. z/d
Now the electric field at the axis (which has only
one, vertical component, due to the axial symmetry of the problem) may be calculated by the partial
differentiation of the electrostatic potential – see Eq. (1.33) of the lecture notes:

   zd zd 
Ez     2  sgn  z  d    sgn  z  d  .
z 2 0 
 R   z  d 2  1/ 2
R 2
 z  d  
2 1/ 2


In the figure on the right, this function is plotted for 1

the same three values of the R/d ratio as the potential in the
figure above. The plots show that at R/d = 5, the field
0.5
between the disks is already pretty uniform, though its )
magnitude is still noticeably smaller than that (/0) E z
R / d  0.2
between two infinite planes. Note also that the ratio R/d  /  0 0
does not affect the electric field’s jump by /0 as the 1
observation point crosses a disk – just as it has to be per Eq.  0.5
(1.24) of the lecture notes. 5
On the technical side, this solution illustrates again 1
the advantage of calculating the electrostatic potential (a 2 1 0 1 2
z/d
scalar function) first, and only then the electric field from it.

Problems with Solutions Page 13

Essential Graduate Physics EM: Classical Electrodynamics

Problem 1.13. The electrostatic potential created by an electric charge distribution is

1 1   r
  r   C    exp  ,
r 2r0   r0 
where C and r0 are constants, and r  r is the distance from the origin. Calculate the charge
distribution in space.
Solution: According to the Poisson equation (see Eq. (1.41) of the lecture notes), the charge
density may be calculated as
 r    0 2 r  .

Since the given potential distribution is spherically symmetric, (r) = (r), the general expression for the
Laplace operator of such scalar function in spherical coordinates10 reduces to
1 d  2 d 
 2 r   r , (*)
r 2 dr  dr 
so by performing a straightforward differentiation, we get a very simple, exponential charge density
distribution:
C  r
 r    0 3 exp  . (INCOMPLETE!)
2r0  r0 
Note, however, that since Eq. (*) is invalid at r = 0, this point has to be explored separately. At r
 0, the given potential distribution tends to C/r, and as Eq. (1.35) of the lecture notes shows, such
potential is created by a point charge q = 40C, located at the origin. As a result, the complete solution
of our problem is
 1  r   1  r 
 r    0 C 4 r   3 exp    q  r   exp   ,
 2r0  r0   8r03  r0 
with the total charge11
 1  r 3   1

 r 2 
Q    r  d 3 r  q1  3 
exp    d r 
  q  1 
 8r 4  exp  r dr 
 8r0
3
 r0    0 0  r0  
 1 
 q1   exp   2 d   0 .
 20 
Just for the reader's reference, such (r) is an approximate model for the charge distribution in a
neutral atom, describing screening of the positive nuclear charge q = Ze by Z negatively charged
electrons. Such an approximation, as well as a more accurate Thomas-Fermi model of the screening,12
takes into account the quantum-mechanical properties of electrons only partly, and as a result, works
reasonably well only for heavy atoms (Z >> 1).

10 See, e.g., MA Eq. (10.9).

11 The last step of this calculation uses the table integral MA Eq. (6.7d) with n = 2.
12 See, e.g., either QM Problems 8.30-8.31 or SM Problems 3.12-3.13.

Problems with Solutions Page 14

Essential Graduate Physics EM: Classical Electrodynamics

Problem 1.14. A thin, flat, rectangular sheet of size ab is electrically charged with a constant
areal density . Without an explicit calculation of the spatial distribution (r) of the electrostatic
potential induced by this charge, find the ratio of its values in the center and in the corners of the
rectangle.
Hint: Consider partitioning the rectangle into several similar parts and using the linear
superposition principle. y
Solution: Selecting the Cartesian coordinates as shown in the figure
on the right, we may use Eq. (1.38) of the lecture notes to express the b 
potential at the origin (i.e., in one of the rectangle’s corners) as
 a b 1 a A
0   dx  dy
4 0 0 0  x  y 
2 2 1/ 2

4 0
I, b/2

where I is the following dimensionless integral:

1 b/a
1 0 a x
I   d a/2
0
 d
0  2
 
2 1/ 2
,

with   x/a,   y/a.

Now let us calculate what potential would be induced, at the same point, only by the adjacent
quarter of the rectangle, of size (a/2)(b/2), marked with darker shading in the figure above:
 a/2 b/2
1
 0' 
4 0  dx  dy  x
0 0
2
 y2 
1/ 2
.

Evidently, this result may be reduced to the same dimensionless integral I:

 a / 2
0'  I,
4 0
so even without the calculation of the integral13 we see that 0’ = 0/2.
Next, we may note that due to the linear superposition principle, the potential A at the center of
the rectangle may be represented as a sum of four potentials induced by such quarters in their corners.
Due to the system’s symmetry, each of these partial potentials is equal to 0’, so, finally,
 A  40'  20 .
This solution illustrates the power of scaling arguments, broadly used for symmetric systems in
all fields of physics.

Problem 1.15. Calculate the electrostatic energy per unit area of a system of two thin, parallel
planes with equal and opposite charges of a constant areal density , separated by distance d.

13 Just for reference, this integral may be worked out analytically: I 

b
ln

a2  b2 1/ 2
a
 sinh 1
b
.
a b a

Problems with Solutions Page 15

Essential Graduate Physics EM: Classical Electrodynamics

Solution: From the similar problem solved in the lecture notes (see Fig. 1.4 and its discussion), it
is clear that the electric field everywhere should be normal to the planes and constant within each of
three ranges:
 E  , at z  d / 2, z
 E 
E  n z  E0 , at  d / 2  z   d / 2, d / 2
 E , at  d / 2  z, E0
 
d / 2
where the z-axis and its origin are selected as shown in the figure on E 
the right, and E–, E0, and E+ are some scalar constants. These constants
may be calculated by applying the Gauss law to three pillboxes with all three possible combinations of
different lid positions, giving three equations:
E+ – E0 = +/0, E0 – E– = –/0, E+ – E– = 0.
An (easy) solution of this system of equations yields14
 .
E  E  0, E0   (*)
0
Thus, the field exits only between the planes, producing the electrostatic potential
z
  z , for z  d / 2,
   z 0   E ( z' )dz'   z 0  
 0 (d / 2) sgn( z ), for z  d / 2,
0 
so the potential difference (voltage) between the planes is
 d  d  
V         d  ,
 2  2  0 C0
where C0 = 0/d is the specific capacitance of this simple system (which is equivalent to a plane
capacitor – for more, see Chapter 2).
Now using Eq. (*) in Eq. (1.65) for the potential energy of the field, we get
d / 2
U 0  0 E 02 2 C V2
  E dz  d d 0 .
2

A 2 d / 2
2 2 0 2
(We will repeatedly run into the last expression in Chapter 2.) Note that at fixed  (of any sign) the
energy grows with d. This is natural, because the oppositely charged planes attract each other, so
following the force (reducing d) corresponds to the way toward the potential energy’s minimum.

E ext
Problem 1.16. The system analyzed in the previous problem (two
thin, parallel, oppositely charged planes) is now placed into an external,  
uniform, normal electric field Eext = /0 – see the figure on the right. Find d
the force (per unit area) acting on each plane, by two methods: 

14 An even simpler way to get Eq. (*) is to employ the linear superposition principle by summing up the fields of
the same magnitude (1.23), E = /20, induced by each of the planes. Due to the difference in plane charge
signs, the fields add up inside the gap between the planes but cancel each other outside of this gap.

Problems with Solutions Page 16

Essential Graduate Physics EM: Classical Electrodynamics

(i) directly from the electric field distribution, and

(ii) from the potential energy of the system.
Solutions:
(i) In order to calculate the force F acting on the top plane, we have to neglect the field of this
plane itself,15 summing up the external field directed up and equal to /0, and the field of the bottom
plane, directed down and equal to /20 – see Fig. 1.4 and Eq. (1.24) of the lecture notes. The net field
equals /20 and is directed up, so the force per unit area is
F q  2
  , (*)
A A 2 0 2 0
and is directed up. A similar calculation for the bottom plane yields a force of the same magnitude but
directed down.
(ii) Using the linear superposition principle, we may calculate the total field in each spatial
region by adding the external field to that of the planes, which was calculated in the previous problem:
 0, for z   d / 2,   1, for z   d / 2,
     
E  nz  nz    1, for  d / 2  z   d / 2,  n z  0, for  d / 2  z   d / 2,
0 0   0 
 0, for  d / 2  z ,   1, for  d / 2  z.
Using these values, let us calculate the potential energy U of the field from Eq. (1.72) of the lecture
notes, artificially limiting the integration volume to a pillbox of an arbitrary area A, with some fixed
thickness d0 > d:
2
U 0    2
   d 0  d   const  d,
A 2 0  2 0
where “const” means a quantity independent of d. Now the vertical force (per unit area) exerted at the
top plane may be calculated as F/A = –(U/A)/d, giving the same result (*) as the first method. The
same is true for the lower plane whose vertical coordinate is (const – d), so the ratio U/A should be
differentiated over (–d) rather than d.

Problem 1.17. Explore the relationship between the Laplace equation (1.42) and the minimum of
the electrostatic field energy (1.65).
Solution: Let us consider a small variation (r) of the electrostatic potential inside a charge-free
volume V limited by a closed surface S, such that the variation vanishes at the surface, but is otherwise
arbitrary. Let us calculate the corresponding variation of the electric field energy (1.65):
  
U    0    d 3 r   0       d 3 r   0       d 3 r   0       d 3 r . (*)
2

 2 V  2 V V V

15Different elementary charges of the same plane do Coulomb-interact, but the elementary forces between them
are directed along the plane, and cancel at summation, due to the 3rd Newton law.

Problems with Solutions Page 17

Essential Graduate Physics EM: Classical Electrodynamics

(The last step of the above calculation has exploited the main rule of the variational calculus that the
operations of small variation and the usual differentiation are interchangeable.)16 Now let us work out
the resulting 3D integral by parts – just as was done at the derivation of Eq. (1.65) of the lecture notes,
i.e. by applying the divergence theorem17 to the vector function f = , and then differentiating this
product by parts:18

    d r        d 3 r        d 3 r     2 d 3 r .

2
n (**)
S V V V

Due to our condition S = 0, the surface integral on the left-hand side of this relation has to equal zero,
while the first volume integral on the right-hand side equals that in the last form of Eq. (*). Hence, Eqs.
(*) and (**) may be combined to give
U   0    2 d 3 r .
V

Since the variation  is arbitrary, this expression shows that the only way for the potential
energy of the field to have the lowest possible value (just as it does in any stable equilibrium in classical
mechanics), and hence the variation U to equal zero, is to have the Laplace operator of the field equal
zero, i.e. to satisfy the Laplace equation (1.42) in all charge-free spatial regions.
Note that this fact may be used to evaluate approximate (say, just guessed) solutions of the
Laplace equation in situations when finding the exact one is difficult. Indeed, if several such trial
solutions have been suggested, the one giving the lowest energy (1.65) has a good chance of being the
closest to the genuine one. Moreover, if we have guessed a family of such approximate solutions
depending on a parameter, the best of them may be found by varying this parameter to minimize the
corresponding energy (1.65).19

Problem 1.18. Prove the following reciprocity theorem of electrostatics:20 if two spatially-
confined charge distributions 1(r) and 2(r) create, respectively, electrostatic potentials 1(r) and 2(r),
then
 1 r 2 r d r    2 r 1 r d r .
3 3

Hint: Consider the integral  E1  E 2 d 3 r .

Solution: By applying Eq. (1.33) of the lecture notes to E1(r), let us rewrite the integral
mentioned in the Hint as
 E1  E2 d r    1  E2 d r .
3 3

Now we may use the rule of spatial differentiation of a vector-by-scalar function product21 to continue as
follows:

16 This rule has a simple geometric meaning – see, e.g., CM Sec. 2.1, in particular Fig. 2.2.
17 See, e.g., MA Eq. (12.2).
18 See, e.g., MA Eq. (11.4a).
19 Such variational method is especially popular in quantum mechanics – see, e.g., QM Sec. 2.9 and on.
20 This is only the simplest one of the whole family of reciprocity theorems in electromagnetism – see, e.g., Sec.
6.8 of the lecture notes.
21 See, e.g., MA Eq. (11.4a), with f =  and g = E .
1 2

Problems with Solutions Page 18

Essential Graduate Physics EM: Classical Electrodynamics

  1  E 2 d 3 r   1   E 2  d 3 r    1E 2  d 3 r.

Next, we may use the inhomogeneous Maxwell equation (1.27) in the first integral on the right-hand
side, and the divergence theorem22 to transform the second integral to that of (1E2)n over some very
distant closed surface S that may be taken for the limit of our spatial integration. As a result, our expression
becomes
1
   d r    E  d
3 2
r.
0 1 2
S
1 2 n

Since per the problem’s conditions, the charge (and hence the field) distributions are space-confined, we
may select the surface S so distant that the surface integral in the above expression is negligible,23 and
we get
1
 E1  E 2 d r   0  1 2 d r .
3 3

Now repeating the same calculation with swapped indices, we arrive at the reciprocity theorem.
Note that if some parts of these two charge distributions reside on some surface(s) S, and may be
well described by surface charge densities 1(r) and 2(r) (as is very instrumental, for example, in
systems with good conductors, to be discussed in Chapter 2), the reciprocity theorem may be rewritten
as
 1 r 2 r  d r    1 r 2 r d r    2 r 1 r  d r    2 r 1 r  d r ,
3 2 3 2

V S V S

where 1(r) and 2(r) are the remaining "genuinely-volume" parts of the distributions. (In this form, it is
sometimes called the "Green's reciprocity theorem".)

Problem 1.19. Calculate the energy of the electrostatic interaction of two spheres, of radii R1 and
R2, each with a spherically symmetric charge distribution, separated by distance d  R1 + R2.
Solution: According to Eq. (1.55) of the lecture notes, applied sequentially to each of the spheres,
the energy Uint of their interaction may be calculated in any of two ways:24

U int    r  r d
3
1 2 r, (*)
sphere 1
U int    r  r d
3
2 1 r. (**)
sphere 2

22 If you still need to, see MA Eq. (12.2).

23 Indeed, we may take S in the shape of a spherical shell, with the radius R much larger than the spatial scales of
distributions 1(r) and 2(r). At such large distances, their fields are close to those of point charges, so according
to Eqs. (1.6) and (1.35), 1  Q1/R and E2  Q2/R2, so the integral over surface S, of area A  R2, is decreasing
with R as 1/R – or even faster if any of the net charges Q1,2 vanishes.
24 Their equivalence may be strictly confirmed using the reciprocity theorem (see the previous problem) even if

the charge distributions 1,2(r) partly or fully overlap – even though in our current case, they do not.

Problems with Solutions Page 19

Essential Graduate Physics EM: Classical Electrodynamics

But as was discussed in Sec. 1.2 of the lecture notes (see Eqs. (1.19)-(1.20)), the electric field,
and hence the electrostatic potential , created by a spherically symmetric charge distribution (r)
outside of its boundary, coincides with that of the full charge Q of the sphere concentrated in its center.
Applying this fact to, for example, the first sphere, we get
Q1
1 r   1( point ) r   , where Q1    r d
3
r, for r  r1  R1
4 0 r  r1
1
sphere 1

where r1 is the center of sphere 1. Since, per the problem’s conditions, all points r inside sphere 2 do
satisfy the last condition, we may apply this expression for 1(r) to Eq. (**), getting

U int    r  r d 3 r .
( point )
2 1
sphere 2
This result means that the interaction energy cannot depend on whether the charge of sphere 1 is
distributed as given or concentrated at point r1, i.e. should be invariant with respect to the replacement
1 r   Q1 r  r1  .
Making this replacement in Eq. (*), we get
U int  Q1 2 r1  . (***)
Now let us use the same argument to express the potential created by the spherically symmetric
distribution 2(r):
Q2
 2 r    2( point ) r   , where Q2    2 r d 3 r , for r  r2  R2 ,
4 0 r  r2 sphere 2

where r2 is the position of the center of sphere 2. By applying this formula to point r1 (which, by the
problem’s conditions, satisfies the condition  r – r2  > R2) and plugging the result into Eq. (***), we
finally get
QQ
U int  1 2 , with d  r1  r2 .
4 0 d
In plain English, the spheres interact as if their electric charges were concentrated in their
centers. Note, however, that this result, by its derivation, is only applicable at d  R1 + R2 when the
charge distributions 1(r) and 2(r) do not overlap – even partly.

R1 R2
Problem 1.20. Calculate the electrostatic energy U of a (generally, thick) Q
spherical shell, with charge Q uniformly distributed through its volume – see the figure
on the right. Interpret the dependence of U on the inner cavity’s radius R1, at fixed Q
and R2.
Solution: Calculating the only (radial) component E of the electric field E =
nrE(r) (for example, by using the Gauss law), we get

Problems with Solutions Page 20

Essential Graduate Physics EM: Classical Electrodynamics

 0, for 0  r  R1 ,
E r  
Q
4 0 r 
2

 
  r 3  R13 / R23  R13 ,  for R1  r  R2 ,
 1, for R2  r ,
so Eq. (1.65) for the electrostatic energy yields
0 0 
 Q 0 
2
 R2  r 3  R 3  2 dr  dr 
   
2  0
U E r 2
dr  3
4 E r r2
dr  2
4     3 1
 2   2
 4 0  R  R2  R1  r
 3
2 2   1 R2 r 

Q 2    3   3  d d  1 / 5   3  9 / 5 5   6  1,
1 2 
Q2
   
8 0 R2   1   3   2 1  2  8 0 R2
        f  , with f  
  1 3
2
 
where   R1/R2  1.
The figure on the right shows a plot of the function f(),
i.e. of the normalized electrostatic energy as a function of R1 at 1.15
fixed Q and R2. At  = 0, i.e. for a solid sphere, the function
reaches its maximum f(0) = 6/5 (so U is reduced to Eq. (1.66) of f   1.1

the lecture notes, with R = R2). On the other hand, at   1, i.e.

1.05
for a very thin spherical shell, it tends to f(1) = 1, giving
Q2 1
U  U min  . 0 0.2 0.4 0.6 0.8
8 0 R2 
This behavior is very natural because the elementary charges of
the sphere repulse each other and “try” to go apart as far as possible, in particular by increasing its inner
radius.

Problems with Solutions Page 21

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 2. Charges and Conductors

Problem 2.1. Calculate the force (per unit area) exerted on the surface of a conductor by
an electric field normal to it. Compare the result with the electric field’s definition given by Eq. (1.6) of
the lecture notes, and comment.
Solution: Let us make a natural assumption that the density u(r) of the electrostatic energy, given
by Eq. (1.65) of the lecture notes, at the conductor’s surface is a smooth function of the surface’s
position. Then, at a small displacement x of the surface, in the direction along the field vector E,25 the
change of the full energy (1.65), per unit area, may be calculated as
U 1 0   1 
    E 2 d 3 r   0 E 2 V  0 E 2 (x) ,
A A  2  2 A 2
where V is the volume outside the conductor (filled with the field). Hence the force exerted by the field
is directed out of the conductor (which is therefore attracted by the field – of any polarity), and equal to
F 1 U  0 2 1
  E  E , (*)
A A x 2 2
where  is the surface charge density – see Eq. (2.3).
The apparent contradiction (the extra factor ½) between this result and the electric field’s
definition (1.6) reduced to a unit surface area (F/A = E) may be interpreted as follows. The field equals
E on one side of the surface charge layer but vanishes on its other side (inside the conductor), so the
average field inside the layer is Eave = E/2. Hence the force per unit area may be calculated as F/A =
Eave = (0E)(E/2), in agreement with Eq. (*).

S4
Problem 2.2. Electric charges QA and QB have been put on two QB S3
conducting concentric spherical shells – see the figure on the right. What S2
is the full charge of each of the surfaces S1-S4? QA S1
Solution: Let us start with applying the Gauss law (1.16),
Q
E d r
2
(*)
S
n
0
(where Q is the full charge inside the closed surface S), to any surface
located inside the inner sphere and containing the surface S1 inside it –
see the inner dashed line in the figure above. As was discussed in Sec. 2.1 of the lecture notes, in statics,
the electric field inside any conductor should be zero, so the left-hand side of Eq. (*) for that surface
vanishes, indicating that the full charge of surface S1 equals zero: Q1 = 0. Hence the full charge QA of the
first sphere should reside completely on its external surface: Q2 = QA.

25 A small shift normal to the vector E evidently cannot change u(r) at all.

Problems with Solutions Page 22

Essential Graduate Physics EM: Classical Electrodynamics

Now let us consider a similar Gaussian volume with the boundary inside the outer conducting
sphere, i.e. containing surfaces S1, S2, and S3. Similarly, the left-hand side of Eq. (*) equals zero, and the
Gauss law gives
Q  Q2  Q3
0 1 .
0
With our prior results for Q1 and Q2, this means that Q3 = –QA. Hence the charge residing on the
external surface S4 of the same sphere is
Q4  QB  Q3  QB  (Q A )  Q A  QB .

C1 C2
Problem 2.3. Calculate the mutual capacitance between the
terminals of the lumped-capacitor circuit shown in the figure on the right. C0
Analyze and interpret the result for major particular cases.
Solution: When solving the lumped-capacitor circuit problems by C2 C1
prescribing certain charges to the capacitor plates, it is necessary to
remember that the net charges of each isolated conducting “island” of the circuit and of both plates of
each capacitor have to equal zero. Indeed, since for the capacitance calculation, we are only interested in
the charges induced by the voltage V applied to the circuit, we may take the total charge of each
insulated island at V = 0 for zero, and because of its isolation, the net charge of the island cannot change
when the voltage is applied. On the other hand, the lumped capacitor model is only valid if the electric
field outside the capacitors is negligible, and hence the net charge of each capacitor has to be negligibly
small – see Fig. 2.5 of the lecture notes.
For example, for the elementary example of two capacitors Q Q Q Q
connected in series (see the figure on the right), with one isolated island,
these requirements immediately imply that all plate charges should have C1 C2
the same magnitude. From this charge pattern, and Eq. (2.26) applied to
each capacitor, we get V1 = Q/C1 and V2 = Q/C2. Since these voltages V1  Q / C1 V2  Q / C2
across each capacitor26 are, by definition, just the differences between the V  V1  V2
electrostatic potentials of their plates, at the connection in series, they just
add up (see the same figure again), giving V = V1 + V2 = Q(1/C1 + 1/C2) for the voltage between the
external terminals. Now considering the whole circuit as a “black box” between the terminals, we have
to define its (mutual) capacitance C as the ratio of the charge Q passed through the terminals at the
circuit’s charging, to the resulting voltage V between them. So, in this easy case, this definition yields
the well-known answer:
1
Q  1 1  1 1 1
C      , i.e.   ,
V  C1 C 2  C C1 C 2
with an evident generalization to the case of an arbitrary number of capacitances connected in series.
The reader has to excuse me for the detailed discussion of this apparently trivial example.
Indeed, without a clear understanding of this general approach, the solution of more complex problems,
such as our current problem, may be difficult because it cannot be reduced to chains of capacitances
connected either in parallel or in series. For this problem, taking into account the capacitor and island

26 They are sometimes called “voltage drops”, though it is more logical to call them “potential drops”.

Problems with Solutions Page 23

Essential Graduate Physics EM: Classical Electrodynamics

neutrality and the circuit symmetry, we arrive at the general  Q1  Q1  Q2  Q2

charge distribution shown on the top panel of the figure on the
right, with the total inserted charge C1 C2
 Q1  Q2
Q  Q1  Q2 , (*) C0
 Q1  Q2
but with Q1 and Q2 still unknown.
C2 C1
Now using Eq. (2.26) to calculate the voltage drop across
each capacitor, we arrive at the voltage pattern shown on the  Q2  Q2  Q1  Q1
bottom panel on the figure on the right, where the arrow
Q1 / C1 Q2 / C2
directions show the assumed voltage polarities, correct if all
specified capacitor charges (Q1, Q2, and Q1 – Q2) are positive.
For these voltage drops, we may write two relations: the obvious C1 C2
equality for the voltage between the terminals: Q1  Q2
Q1 Q2 C0
C0
V   , (**)
C1 C 2 C2 C1
and the condition that the net voltage drop along any of the
closed loops of the circuit is zero:27 Q2 / C2 Q1 / C1
Q1 Q2 Q1  Q2
   0. (***)
C1 C 2 C0
Solving the system of two linear equations (*) and (***) for Q1,2, and plugging the result into Eq. (**),
we finally get V = Q/C, where
2C C  C1  C 2 C 0
C 1 2 . (****)
C1  C 2  2C 0
Let us start the analysis of this result from the case C0 = 0. In this particular case, Eq. (****) is
reduced to
2C1C 2 1 1 1
C , i.e. C  2C' , where   .
C1  C 2 C' C1 C 2
This result is natural, because in this case the capacitor C0 cannot accommodate any charge, and may be
replaced with an open circuit. As a result, our full system becomes just a parallel connection of two
similar branches, each being just a connection in series of the capacitors C1 and C2, with the branch
capacitance C’.
Note that the last result (or rather its particular form C = C1 = C2) is also valid in the case of an
arbitrary C0, but equal C1 and C2. Indeed, in this case, the circuit is symmetric with respect to the central
horizontal line in any of the figures above, so there is no voltage drop between its internal notes.

27 In the formal theory of lumped-element circuits, this condition is called the 2nd (or “loop”) Kirchhoff rule, with
the 1st (or “node”) Kirchhoff rule saying that the algebraic sum of the charges/currents flowing into each circuit
node (wire connection) should equal zero. (For our circuit, the latter rule yields Eq. (*), obtained above from the
reasoning similar to that used for the proof of the rule in the general case.) The Kirchhoff rules will be discussed
in Chapters 4 and 6 of the lecture notes.

Problems with Solutions Page 24

Essential Graduate Physics EM: Classical Electrodynamics

In the opposite limit, C0   (practically meaning that this capacitance is much larger than both
C1 and C2), Eq. (****) yields
C"
C , where C"  C1  C 2 .
2
This is also natural, because (as the last figure above shows) a very large capacitance C0 makes the
potential difference between two internal nodes of the circuit negligible, so it becomes just a connection
in series of two similar groups of capacitors C1 and C2, connected in parallel within each group.
Finally, if one of the branch capacitances (say, C2) tends to 0, Eq. (****) is reduced to
C1C 0 1 1 1 1
C , i.e.    .
C1  2C 0 C C1 C 0 C1 C1
This result becomes clear if we redraw our circuit for this particular case,
by replacing both zero capacitors C2 with open circuits. The resulting C0
system, shown in the figure on the right, is just a connection in series of 3
capacitors: C1, C0, and C1 again, correctly described by the last formula for C1
C. (The result for C1 = 0 is similar, with the obvious index replacement.)
So, the circuit under analysis, despite its relatively simple topology,28 becomes either the
capacitors’ connection in series or their connection in parallel in different ultimate cases, but generally it
is neither.

C1 C1 C1
Problem 2.4. Calculate the capacitance between the 
terminals of the semi-infinite lumped-capacitor chain shown in the
figure on the right, and find the law of the applied voltage’s decay C2 C2 C2
along the system. Analyze and interpret the result. 
Solution: Since all links of the chain are similar, it is
sufficient to analyze just a couple of them, say with numbers j and
(j + 1), where j is the “distance” from the input terminals.  Q j  Q j  Q j 1  Q j 1
 
Assigning the electric charges to the capacitor plates, with the C1 V j C1 V j 1
island and capacitor neutrality rules discussed in detail in the  Q j  Q j 1
previous problem’s solution, we arrive at the charge and voltage
distribution pattern shown in the figure on the right. As it shows,  Q j  Q j 1 C C2
2
we may describe the circuit by Eq. (2.26) spelled out for two
different capacitors of the system:  
Q j 1
V j  V j 1  , Q j  Q j 1  C 2V j . (*)
C1
Due to the linearity of these relations, we may expect both variables V and Q to decay
exponentially with the “distance” j from the system’s terminals:

28With a galvanometer instead of the capacitor C0, such circuits (of either capacitors or resistors) are broadly used
for sensitive differential measurements and have deserved a special name, the Wheatstone bridges.

Problems with Solutions Page 25

Essential Graduate Physics EM: Classical Electrodynamics

V j 1 Q j 1
V j  Q j  e  j , so that   e     1 .
Vj Qj
Indeed, using the last relations to eliminate Vj+ 1 and Qj+1 from Eqs. (*), we get a system of two linear
homogenous equations for the same two variables Vj and Qj:
Q j
1   V j  , 1   Q j  C 2V j . (**)
C1
These equations are compatible (and hence the assumed solution is valid for all j) if the determinant of
this system equals zero:
1   / C1  C 
 0, i.e. if  2   2  2    1  0 .
 C2 1  C1 
The needed root  < 1 of this quadratic equation29 is
1/ 2
C C C2 
  1  2   2  2 2  . (***)
2C1  C1 4C1 
Plugging it back into any of Eqs. (**), for example the first one, we get the sought-for capacitance
between the input terminals:
1/ 2
Q1 Q j 1 Q j  C 22  C
C    C1 1      C1C 2 
 
  2 . (****)
V0 Vj Vj  4  2

Let us consider what our results (***) and (****) yield in two ultimate cases. If C1/C2  0, then
C2
  1, C  C1 .
2C1
This result means that the voltage between the input terminals does not propagate noticeably beyond the
first capacitor C1 of the chain, because the next, relatively large capacitance C2 effectively “grounds” the
voltage.
In the opposite limit, C1/C2  , Eq. (***) yields
1/ 2 1/ 2
C  C 1 
  1   2   1, so that   ln   2   1 .
 C1    C1 
According to the definition of , this means that the voltage applied to the chain’s terminals penetrates
deep into the chain, decaying at the “distance” N  1/  (C1/C2)1/2 >> 1. For the capacitance between
the terminals, in this limit, Eq. (****) gives
C  C1C 2 
1/ 2
 NC 2 .
The last form of this result may be interpreted by saying that the input voltage penetrating by “distance”
N >> 1 into the chain “sees” N capacitors C2 which are effectively connected in parallel.

29 The second root ’ = 1/ > 1 of the equation describes an exponential growth of Vj and Qj with j, and could be
relevant if our chain had a finite length.

Problems with Solutions Page 26

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.5. A system of two thin conducting plates is A1

located over a ground plane as shown in the figure on the right,
A2
where A1 and A2 are the areas of the indicated plate parts, while d’
and d” are the distances between them. Neglecting the fringe effects, 1
calculate: d'
2
(i) the effective capacitance of each plate, and d"
(ii) their mutual capacitance.
Solutions: Let the plates have potentials 1 and 2 (relative to the ground). They create uniform
vertical electric fields in each of the three regions of the system. Integrating Eq. (1.33) of the lecture
notes, we may readily find that the field in the gap (of thickness d’ and area A2) between the plates is E’
= (1 – 2)/d’, that in the gap (of the same area A2, but thickness d”) between plate 2 and the ground is
E” = 2/d”, and, finally, in the clear gap (of thickness d’ + d” and area A1 – A2) between plate 1 and the
ground, the field is E0 = 1/(d1 + d2). Now applying the Gauss law (1.16) to a flat pillbox drawn around
each plate,30 we may calculate their total electric charges:
 A  A2 A   A  A2 A2  A
Q1   0  A1  A2 E 0  A2 E'    0  1 1  2 1   2    0  1  1   0 2  2 ,
 d'  d" d'   d'  d" d'  d'
(*)
  2 1   2  A2 1 1 
Q2   0 A2 E"  E'    0 A2      0 1   0 A2    2 .
 d" d'  d'  d' d" 
These relations enable us to answer both posed questions.
(i) To calculate the effective capacitance of each plate, we should find its charge when the
counterpart plate is grounded.31 From Eqs. (*) we readily get
 A  A2 A2   A  A2 A2 
Q1  0   0  1  1 , i.e. C1ef   0  1  , (**)
2  d'  d" d'   d'  d" d' 
1 1  1 1 
Q2  0   0 A2    2 , i.e. C 2ef   0 A2    . (***)
1  d' d"   d' d" 
The first of these results, Eq. (**), may be readily understood, 1
in the language of the equivalent circuit shown in the figure on the
C12
right, as the parallel connection of the “direct” capacitance,
2 C1
A1  A2 C2
C1   0 ,
d'  d"
ground
of the clear gap between plate 1 and the ground plane, with the direct
capacitance between the plates:

30 Applying the Gauss law to plate 1, we should remember that the field above it has to be vanishingly small
because any appreciable field in that region (of volume ~A13/2) would be of prohibitive energy cost. Another
explanation of the same fact is that all charges of plate 1 are strongly attracted to their opposite-sign images in the
ground plane (see Sec. 2.6 of the lecture notes), so may reside only on the lower surface of the plate.
31 The effective capacitances may be also calculated from the reciprocal capacitance matrix p by using Eq. (2.33)
jj’
and its counterpart for the second plate, but in our simple case, this approach is less straightforward.

Problems with Solutions Page 27

Essential Graduate Physics EM: Classical Electrodynamics

A2
C12   0 .
d'
On the other hand, Eq. (***) describes the parallel connection of the same C12 with the direct
capacitance,
A
C2   0 2 ,
d"
between plate 2 and the ground.
(ii) To calculate the mutual capacitance between the plates, we have to take Q2 = –Q1  Q, and
calculate V  2 – 1 from Eqs. (*). The result of this simple calculation may be represented in the form,

Q  A  d'  d" d'  1   1 1 

1

C    0   
2
    C12     ,
V  d'  A2  A1 A2    C1 C 2 
and understood from the same equivalent circuit (see the figure above) as the parallel connection of C12
with the serial connection of C1 and C2.
These facts illustrate how convenient the language of equivalent circuits is. However, note again
that this approach is valid only if the system under analysis may be partitioned into a set of independent
field regions represented as lumped circuit elements – see Fig. 2.5 of the lecture notes and its discussion.

x
Problem 2.6. A wide and thin film carrying a uniformly
t
distributed electric charge of areal density  is placed inside a
similarly wide plane capacitor, whose plates are connected with a E

wire – see the figure on the right. Neglecting the fringe effects, d
calculate the surface charges of the plates and the net force exerted on 0 E
the film (per unit area).
Solution: Let us pursue the most methodical (but not the shortest) way to solve the problem:
solving the Poisson equation (1.41) for the electrostatic potential . Since, due to the problem’s
symmetry,  may depend on just one Cartesian coordinate (say, x), normal to the planes of the electrode
surfaces and the thin film, the equation is reduced to
d 2 
   x  d  . (*)
dx 2
0
Here the origin of x is taken at the bottom plate’s surface, so at the top plate’s surface, x = t – see the
figure above. Since the plates are connected with a conducting wire, in statics, their potentials are equal.
Taking this value for zero, we may write the boundary conditions for the function (x) as
 0    t   0 . (**)
The solution of this boundary problem is straightforward. Indeed, per Eq. (*), in each of the two
gaps between the film and the plates, d2/d2x = 0, so  has to be a linear function of x:

c   E  x, at 0  x  d ,
 x   
 c   E  x, at d  x  t ,

Problems with Solutions Page 28

Essential Graduate Physics EM: Classical Electrodynamics

where E are the corresponding values of the electric field, E = –d/dx – see the figure above. Integrating
Eq. (*) over an infinitesimal interval [d – 0, d + 0], we get
d d 
x  d  0  x  d  0   E  E   .
dx dx 0
The boundary conditions (**) give two more equations for the constants c and E:
c   0, c  E t  0 .
Finally, the potential has to be continuous, so both expressions for (x) have to give the same value at
the boundary x = d:
 d   c   E  d  c   E  d . (***)

Solving this simple system of four equations, we get32

 td  d
E   , E  . (****)
0 t 0 t
Due to the fundamental relation (2.3), these formulas immediately yield the surface charges of the
capacitor’s plates:
td d
 bottom plate   0 E    ,  top plate   0 E    .
t t
(As a sanity check, bottom plate + top plate +  = 0, as it should be.33)
Next, plugging Eqs. (****) for the fields E into Eq. (1.65) of the lecture notes, we may calculate
the electrostatic potential energy of the system (per unit area):
 d  2  t  d    t  d  d
2
U 0 2 2
 
2
 E  t  d   E 2 d    t  d     d  .
A 2 2 0  t   t   2 0 t

Now the force exerted on the film (also per unit area) may be found just as the (minus) gradient of its
potential energy – see, e.g., Eq. (1.32). In our geometry, the only nonvanishing component of the
gradient is vertical, so F = Fnx, with34
F  U / A  2 t  2d
  .
A d 2 0 t
This result shows that the film is attracted to the nearest plate, so at the middle between the
plates, at d = t/2, the net force vanishes, i.e. the film is in equilibrium. Since at this point (and actually at
any position d)

32 Alternatively (and easier), these relations for the fields E may be obtained by applying the Gauss theorem to
three pillboxes with various positions of their lids – see the derivation of Eq. (1.23) in the lecture notes. This
simple additional exercise is highly recommended to the reader.
33 Even if the net electric charge of the capacitor’s plates is different from zero, it resides on their exterior surfaces
– cf. the solution of Problem 1.
34 A shorter but less obvious way to obtain the same result is to say that the effective electric field E acting on
ef
the film (so that F/A = Eef) is equal to the average of the fields on its two sides: Eef = (E+ + E–)/2.

Problems with Solutions Page 29

Essential Graduate Physics EM: Classical Electrodynamics

 2U 2
   0,
d 2  0t
this equilibrium is unstable.35

Problem 2.7. A very small conductor (possibly, of an

irregular shape) with self-capacitance C is located at distance R
r from the center of a conducting sphere of radius R – see the r C
figure on the right. In the first approximation in C, find the
reciprocal capacitance matrix of the system. Use the matrix to
calculate its potential energy and the force of the conductor
interaction for two cases:
(i) the conductor charges Q are equal, and
(ii) the conductor potentials  are kept equal.
Solution: Marking the variables describing the sphere with index “s”, and those describing the
small conductor with index “c”, we may rewrite the definition of the reciprocal capacitance matrix given
by Eq. (2.19) of the lecture notes, with an account of Eq. (2.20), as
s  p s Qs  pQc ,
(*)
 c  pQs  p c Qc .

Using the same argumentation as at the derivation of Eq. (2.29), we may argue that the relatively weak
interaction of these two conductors, described by the off-diagonal matrix element p, cannot affect its
diagonal elements ps and pc substantially, so we may write
1 1 1 1
ps   , pc   ,
C s 4 0 R C 4 0 a
where Eq. (2.17) and the estimate (2.18) are used to define the small conductor’s size parameter a, with
a << r, R. Also, as was argued in Sec. 2.2, if the small conductor is electroneutral (Qc = 0), its potential
has to be approximately equal to that created at its location by other charges of the system; in our case,
according to Eq. (1.35), this means that
Qs 1
c Qc 0  pQs  , i.e. p  .
4 0 r 4 0 r
Now that we have the reciprocal capacitance matrix established, we may readily answer all posed
questions.
(i) If Qs = Qc = Q, then Eq. (2.24) yields the following potential energy of the system:
 pS pc  Q2  1 1 1 
U Q  2
p     .
 2 2  4 0  2 R r 2a 

35 If the last point is not absolutely clear, please revisit the discussion of the fixed point stability in CM Sec. 3.2.

Problems with Solutions Page 30

Essential Graduate Physics EM: Classical Electrodynamics

Of the three terms in the parentheses, only the middle one depends on the distance r between the sphere
and the small conductor, and hence determines the force of their interaction (repulsion):
U Q2
F   .
r 4 0 r 2
So, quite naturally, in our limit a << r, R, the conductors interact as point charges.
(ii) If s = c = , then Eq. (*), together with the matrix elements listed above, gives the
following simple system of two linear equations for the two charges of the conductors:
1  Q s Qc  1  Qs Qc 
   ,    .
4 0  R r  4 0  r a 
Solving it, we get
Rr r  a  ar r  R 
Qs  4 0 , Qc  4 0 , (**)
r 2  Ra r 2  Ra
From here, we can immediately calculate the repulsion force:
Qs Qc Rar r  a r  R  Ra  R 
F   4 0 2  4 0 2 2 1   . (***)
4 0 r 2
r  Ra 
2 2
r  r
where the last approximation is valid if a is smaller than not only r and R but also than r2/R. 36
With the potential energy, we have to be more careful than in Task (i), because Eq. (2.24) is only
valid for two independently-fixed charges which, in particular, would not depend on the distance r, as
they do for our system – see Eq. (**) again. The simplest way to calculate U for our current case of
equal potentials is to use the already found charges Qs and Qc to calculate the self-capacitance of our
total system, keeping only the terms of the 0th and 1st order in small a:
Rr r  a   ar r  R    R 
2
Qs  Qc
C   4 0  4 0  R  a 1    ,
 r 2  Ra   r  
and then use the last of Eqs. (2.15) to find the energy:
C 2   R 
2

U   2 0  R  a 1    .
2

2   r  
With our assumptions a << r, R, r2/R, the second term in the square brackets is much smaller
than the first one, and for some purposes may be neglected, but since only this term depends on the
distance r between the sphere and the small conductor, we need to keep it if we want to use this
expression for the sanity check of the above result for the repulsion force:
 r  R    
2
U R R2 2R  R 
F   2 0 2 a  2  2
a  1  2    2 0 2 a  2 1  ,
r 
0
r r r 2 r r2  r  r
i.e. exactly Eq. (***).

36It is curious that the force is not a monotonic function of the distance r, but has a maximum at r =
(3/2)R.

Problems with Solutions Page 31

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.8. Use the Gauss law to calculate the mutual capacitance of
the following two-electrode systems, both with the cross-section shown in Fig. ba
2.7 of the lecture notes (reproduced on the right):
(i) a conducting sphere in the center of a spherical cavity inside another 0 a
conductor, and
(ii) a long conducting round cylinder on the axis of a cylindrical cavity
inside another conductor, i.e. a coaxial cable. (In this case, we speak about the
capacitance per unit length).
Compare the results with those obtained in Sec. 2.2 using the Laplace
equation.
Solutions:
(i) Applying the Gauss Law to a sphere of a radius r in the range a < r < b, we get the following
result for the magnitude of the electric field E = E(r)nr:
Q
E (r )  ,
4 0 r 2
where Q is the total charge of the inner conductor. Integrating this result along the radius, we find that
the voltage V  (b) – (a) between the electrodes is
b b
Q dr Q 1 1
V   E (r )dr  r    ,
a
4 0 a
2
4 0  a b 
so the mutual capacitance
1
Q 1 1 ab
C  4 0     4 0 ,
V a b ba
in full agreement with Eq. (2.56) of the lecture notes.
(ii) An absolutely similar calculation, but applied to a cylinder of radius , with charge   Q/l
per unit length, yields
 C 2 0
E (r )  ,  ,
2 0  l ln (b / a)
in full agreement with Eq. (2.49).

Problem 2.9. Calculate the electrostatic potential’s distribution around two 0

V
barely separated conductors in the form of coaxial round cones (see the figure on   
the right), with voltage V applied between them. Compare the result with that of a 2
similar 2D problem, with the cones replaced with plane-face wedges. Can you
calculate the mutual capacitances between the conductors in these systems? If not,    V
can you estimate them? 2
Solution: In the spherical coordinates {r, , }, with the z-axis directed
along the common symmetry axis of the cones, the conductors’ surfaces are
described by the relations  = 0 and  =  – 0, so the values V/2 of the electrostatic potential at these

Problems with Solutions Page 32

Essential Graduate Physics EM: Classical Electrodynamics

surfaces do not depend on the coordinates r and . So, it is natural to look for the solution of this
boundary problem in the form  = (). In this case, the Laplace equation is reduced to37
d  d 
 sin    0, for 0     .
d  d 
Integrating this equation sequentially twice, we get
d d d sin d d
sin   c1 , i.e. d  c1 , so     c1   c1   c1  ,
d sin  sin  sin 
2
1 2
where   cos. The last integral is easy,38 and we get
c1 1   c 1  cos  
    ln  c2  1 ln  c2  c1 ln tan  c2 .
2 1  2 1  cos  2
With the proper choice of the two integration constants c1,2, we may indeed satisfy the boundary
conditions on the surfaces of both conductors:
0 V  0 0 V
  0   c1 ln tan  c2   ,     0   c1 ln tan  c 2  c1 ln tan  c2   .
2 2 2 2 2
Solving this system of two equations for the constants c1,2, we finally get the potential distribution:
V ln tan  / 2 
    , for  0       0 . (*)
2 ln tan  0 / 2 
This is a solution, and hence the (unique) solution of this boundary problem.
A similar 2D problem, on the potential distribution between two conducting
cylindrical wedges, with the cross-section shown in the figure on the right, may be V
 
solved using the polar coordinates. Using the same argumentation as in the above cone 2
problem, we may expect the electrostatic potential to be the function of the polar angle
alone. (For the purposes of comparison with the just solved cone problem, I will 0
denote this angle, referred to the polar axis, by the same letter .) Then the 2D Laplace
equation is reduced to a very simple form,39
d 2
 0, V
d 2  
with the linear solution 2
    c1  c 2 .

37See, e.g., MA Eq. (10.9) with /r = / = 0.

38As a math reminder, it may be worked out by representing the function under the integral as the sum of two
simple singularities (“single poles”):
1 1 1
  .
1 2
21    21   
39 See, e.g., MA Eq. (10.3) with / = /z = 0, and   .

Problems with Solutions Page 33

Essential Graduate Physics EM: Classical Electrodynamics

Selecting the constants c1,2 to satisfy the same boundary conditions, (0) = +V/2, ( – 0) = –V/2, we
get
V   2
    , for  0       0 . (**)
2   2 0
The plots of the 3D function (*) and the 2D 0.5
function (**), each for the same two representative 2D 2D
values of the angle 0, are shown in the figure on the 3D 3D
right. They show that the 3D function’s gradient (and
hence the electric field) is more localized near the 
conducting electrodes, though for larger values of 0, V
approaching /2, the difference is insignificant, i.e. the 0
3D function () is also virtually linear. 
0 
16
Finally, to calculate the mutual capacitance
between the electrodes (for any dimensionality), we 7
need to calculate the surface charge density on, say, the 0 
16
top electrode: 0.5
0 0.5 1
  /
   0 E n     0 ,
0 n  0
where n is the linear coordinate normal to the surface. In both the 3D case and the 2D case, this
coordinate is proportional to the distance from the system’s center, so  is inversely proportional to this
distance.40 For example, in the 3D case of two conducting cones,41
c d  
 , with c   0 .
r d  0
Even without calculating the coefficient c, we see that the integral determining the total charge Q of the
upper cone,
 
Q   d 2 r  2  rdr  2c  dr ,
S 0 0

strongly diverges at the upper limit. Physically, this means that to calculate a finite value of the system’s
mutual capacitance C = Q/V, we need to specify how exactly their spatial extent is limited.
The 2D system (of two conducting wedges) features a similar divergence not only at large but
also at small distances, though in both cases, rather mildly. Indeed, for it:42

40 Note that the 1/r-type divergence of the electric field and the surface charge density near the systems’ centers is
quantitatively different from the one discussed in Sec. 2.6 of the lecture notes for a single conducting
corner/wedge – see Eq. (2.123). The difference is due to the fact that in the latter case, the field was due to some
remote sources, rather than the oppositely biased counterpart wedge.
41 See, e.g., MA Eq. (10.8) with /r = / = 0.
42 See, e.g., MA Eq. (10.2) with / = /z = 0 and   .

Problems with Solutions Page 34

Essential Graduate Physics EM: Classical Electrodynamics

c d   V
 , with c   0    0 ,
 d 0   2 0
where  is the 2D radius in the cross-section’s plane, so the charge per unit length of the system,
  
Q 1 d 2 0V d
  d 2 r  2  d  2c     .
l lS 0 0
   2 0 0

So, the calculation of a finite mutual capacitance between the wedges requires a specification of not only
their exact shape at large distances but also of the exact geometry of the gap between them. However,
due to the weak (logarithmic) character of this singularity, a fair estimate of the capacitance may be
given as
C 2 0 
 ln max ,
l   2 0  min
where the two values of  are, respectively, the scales of the wedges’ size and of the gap’s width.

Problem 2.10. Calculate the mutual capacitance

0
between two rectangular planar electrodes of area A = al,
with a very small angle 0 between them – see the figure on 0
0 0  a
the right.
Solution: As was discussed in Sec. 2.6 of the lecture notes, the variable separation in polar
coordinates shows that the Laplace equation has a partial solution expressed by the second of Eqs.
(2.110):
  c0  s0 .
If the polar coordinates’ center coincides with the point 0 in which inner surface planes cross (see the
figure above), i.e. if the surfaces correspond to fixed values of the angle  (which differ from each other
by 0), the boundary conditions on these surfaces are also satisfied by this solution, provided that s00 =
V, where V is the voltage between the electrodes, so the following potential distribution,

 V  const , (*)
0
is the solution of our boundary problem, provided that
 0  a  0  a, l ,
so that the fringe effects are minor, and boundary conditions on outer surfaces are not important.
Now the inner surface charge densities may be calculated from the basic Eq. (2.3):

   0 .
n
At the surfaces of constant , /n = (1/)/, where  is the distance from the point 0, and the
upper/lower sign corresponds to the upper/lower plate, so

Problems with Solutions Page 35

Essential Graduate Physics EM: Classical Electrodynamics

1  1 V
    0   0 .
   0
Integrating these densities over the plate surfaces, we get their full charges:
0  a 0  a
V d V 0  a
Q     d r  l
2
   d   0 l    0 l ln ,
0 0 0  0 0

so the mutual capacitance

Q l  a l  a 
C  0 ln 0  0 ln1  . (**)
V 0 0  0   0 
In the limit of a very short capacitor, with a << 0 (but still 0 << a/0, l/0), whose plates are
essentially parallel, this expression may be approximated as
l a A
C  0  0 ,
0 0 d
where d  00 is the distance between the plates, thus reducing our result to the familiar Eq. (2.28).
On the other hand, at 0 << a, the potential distribution (*) and hence Eq. (**), in the form
l  a 
C  0 ln  ,
0  0 
are valid for an arbitrary opening angle 0 – cf. the solution of the previous problem for the 2D system.

Problem 2.11. Using the results for a single thin round disk, obtained in R
Sec. 2.4 of the lecture notes, consider a system of two such disks at a small d
distance d << R from each other – see the figure on the right. In particular,
calculate:
(i) the reciprocal capacitance matrix of the system,
(ii) the mutual capacitance between the disks,
(iii) the partial capacitance of one disk, and
(iv) the effective capacitance of one disk
– all in the first nonvanishing approximation in d/R << 1. Compare the results (ii)-(iv) and interpret their
similarities and differences.
Solutions:
(i) Due to the symmetry of this system with respect to the disk swap and the common property
expressed by Eq. (2.20) of the lecture notes, the general linear relations (2.19) are reduced to
1  p'Q1  pQ2 ,
(*)
 2  pQ1  p'Q2 .

Problems with Solutions Page 36

Essential Graduate Physics EM: Classical Electrodynamics

At d << R, approximate expressions for the remaining two coefficients p’ and p may be found from the
two particular problems already solved in the lecture notes. Indeed, if both disks carry equal charges of
the same sign: Q1 = Q2 = Q/2, then there is virtually no field inside the gap separating them, while the
field outside is virtually the same as that of a single disk charged with charge Q. According to Eq.
(2.69), in this case
Q
1   2  .
8 0 R
On the other hand, in this case, Eqs. (*) give
p  p'
1   2  Q.
2
Comparing these two expressions, we get
1
p  p'  . (**)
4 0 R
On the other hand, if the disk charges are equal but opposite, say Q1 = –Q2 = Q, then the system
is nothing more than a plane capacitor, with a uniform field inside the gap and a negligible field outside
it, so according to Eq. (2.28),
V Q Qd Qd
1   2     .
2 2C 2 0 A 2 0 R 2
Comparing this result with the prediction of Eqs. (*) for this case,

1   2  ( p'  p )Q,
we get one more equation for p and p’:
d
p'  p   p  p' . (***)
2 0 R 2
Solving the system of two linear equations (**) and (***), we get

1 d 1  2d  1 d 1  2d 
p'    1  , p   1  .
8 0 R 4 0 R 2
8 0 R  R  8 0 R 4 0 R 2
8 0 R  R 

(ii) With these expressions, the mutual capacitance, defined by Eq. (2.26) with p1 = p2 = p’,
1
C ,
2 p '  p 
becomes
 R 2
C 0
d
– admittedly, the expression already used above.
(iii) Both partial capacitances of the disks, generally defined as C1 = 1/p1 and C2 = 1/p2, in our
symmetric case are equal to 1/p’, and in the first approximation, are independent of d << R:
C1  C 2  8 0 R .

Problems with Solutions Page 37

Essential Graduate Physics EM: Classical Electrodynamics

(iv) On the contrary, the effective capacitances of the disks, which may be calculated from Eq.
(2.34), are much larger and inversely proportional to the distance between the disks:
 0 R 2
C1e,f2  C   C1, 2 .
d
This big difference between the partial and effective capacitance (typical for all strongly coupled
systems of conductors) is very natural. If one disk is charged and the second one is not, the latter disk
acquires almost the whole potential of the charged counterpart,43 and virtually does not affect the field
distribution in space. However, if the second disk is grounded (i.e. set to  = 0), it kills the electric field
everywhere besides the narrow gap between the charged disk and the effective ground.

y
Problem 2.12.* Calculate the mutual capacitance (per
unit length) between two cylindrical conductors forming a
system with the cross-section shown in the figure on the right, R
in the limit t << w << R.
Hint: You may like to use the elliptic coordinates t
mentioned in Sec. 2.4 of the lecture notes. They are defined by  0
0
the following equality:
x
x  iy  c cosh (   i ). (*) w
where c is a constant.
Solution: Since the complex function z = c cosh w
defining these orthogonal coordinates is analytic, the Laplace   V
equation in them is simple – see Sec. 2.4 of the lecture notes:44
 2 2 
 2    0 . (**)
   2 
According to Eq. (*), the lines of constant  on the [x, y] plane are ellipses with horizontal and
vertical semi-axes ccosh and csinh, respectively. At   0, the ellipse degenerates into a straight
horizontal segment –c < x < +c, while at  >> 1, the ellipse is virtually a circle of the radius  = (c/2)e.
As a result, if we select the axes x and y as shown in the figure above, and take c = w/2, the boundary
conditions on the conductor surfaces may be satisfied, at t << w << R, by a potential distribution ()
independent of :
 4R 
 (0)  0,   ln   V , (***)
 w 
where V is the voltage between the conductors. Hence the boundary problem is satisfied with a function
(), provided that it obeys the 1D Laplace equation following from Eq. (**):

43 Electronic engineers call such electrodes “floating”.

44 Just for the reader’s reference, the Laplace operator in these coordinates is
1  2 2 ,
2    
c 2 (cosh 2   cos 2  )   2  2 
in compliance with Eq. (**).

Problems with Solutions Page 38

Essential Graduate Physics EM: Classical Electrodynamics

d 2
 0.
d 2
Per this equation,  () is just a linear function c1 + c2. Selecting the two constants c1,2 to satisfy the
boundary conditions (***), we get

  V .
ln(4 R / w)
To calculate the surface charge of the central conductor by using the general Eq. (2.3) of the
lecture notes, we need to calculate the (normal) electric field at its surface (  0), where Eq. (*), with
c = w/2, yields x = (w/2)cos and y  (w/2)sin:
 V
En   y  0,  ,
y w / 2 x w / 2 ( w / 2) ln(4 R / w) sin 
so the two-surface charge density
V V
  2 0 E n  2 0  2 0 .
( w / 2) ln(4 R / w) sin  
( w / 2) ln(4 R / w) 1  (2 x / w) 2 
1/ 2

Integrating this distribution over the strip’s width, we get its total charge (per unit length)
w / 2 w / 2
Q 2 0V dx 2 0
   dx    V,
l w / 2 
( w / 2) ln(4 R / w)  w / 2 1  (2 x / w) 2 
1/ 2
ln(4 R / w)
so the mutual capacitance per unit length is
C Q/l 2 0
  .
l V ln(4 R / w)
Comparing this result with Eq. (2.49), we see that the only difference between the capacitance of
this system and that of the round coaxial cable with a = R and b = w is minor: just a different numerical
factor of the order of 1 under the logarithm of a large argument.

Problem 2.13. Calculate the mutual capacitance (per unit length) between two similar, long,
parallel wires, each with a round cross-section of radius R, whose axes are separated by distance d > 2R.
Explore and interpret the result in the limits R  0 and R  2d.
Hint: You may like to use the 2D orthogonal bipolar coordinates {, } defined by the following
relations with the Cartesian coordinates {x, y}:
sinh  sin 
xa , ya , with      ,       .
cosh   cos  cosh   cos 
In these coordinates, the Laplace operator is
1  2 2 
 2  2 cosh   cos    2  .
a    2 
Solution: Using the given definition of the bipolar coordinates, it is elementary to prove that the
lines of constant  are circles of radii  = a/sinh, centered at the points with x = x  a coth and y = 0,

Problems with Solutions Page 39

Essential Graduate Physics EM: Classical Electrodynamics

and the lines of constant  are also circles but with the radii  = a/sin, centered at the points with x = 0
and y = y = a cotan – see, respectively the solid and dashed lines in the figure below. All circles of
constant  pass through the focal points x = a and are orthogonal to the -circles at all points. As a
result, if we use the coordinate system shown in this figure, and select the parameters a and 0 to satisfy
the following relations:
y
a d
R ,  a coth  0 ,
sinh  0 2    / 2
 0
then the two -circles with  = 0 coincide  0    0    0
with the circular borders of the cross-sections  0
of our conductors – see the colored circles in R R
x
the same figure. Solving this simple system 0
of equations, we get V V
 
 d  2 
1/ 2 2 2
d 1 d   0
a     R   ,  0  cosh
2
.
 2   2 2R
With this choice, the boundary
conditions for the electrostatic potential  as a 2a
function of the bipolar coordinates  and  d
become very simple, especially if we assume that our system of two conductors is voltage-biased in the
symmetric way shown in the figure above:
V
   0 ,     , for any  .
2
Since, per the expression for 2 provided in the Hint, the Laplace equation in these coordinates is also
very simple:45
 2  2
  0,
 2  2
the solution of our boundary problem is obviously given by a linear function of  alone, namely

 V .
2 0
What remains is to calculate the electric charge (per unit length), of one (say, the right) wire, by
using the basic Eq. (2.3):
Q
  0  E n dr , (*)
l R

where the integral is over the boundary of the wire’s cross-section, i.e. over the circle with  = 0 =
const. Generally, this may be done directly, by expressing En as –/n and calculating the differential
dn as a function of d on this circle, but that must be done very carefully because the change d affects

45 This should be not too surprising, because the above relations between the bipolar and Cartesian coordinates
may be represented by an analytic complex function: z  ia cotan(w/2), where z  x +iy and w   + i – see the
discussion of conformal mapping in Sec. 2.4 of the lecture notes.

Problems with Solutions Page 40

Essential Graduate Physics EM: Classical Electrodynamics

not only  but also x. A simpler approach is to notice that the integral on the right-hand side of Eq. (*)
is the full flux of the electric field through the wire’s surface. Since the free-space region between this
surface and the symmetry plane, x = 0 (on which  = 0 as well) has no electric charges, per the Gauss
law, the incoming flux has to be equal to the outcoming flux through that plane. Hence we may write
 
Q 
  0  E x x  0 dy   0  x x 0 dy .
l  

The advantage of this expression over Eq. (*) is that at x  0 (and hence   0), the relations between
the bipolar and Cartesian coordinates simplify:
 sin 
xa , ya ,
1  cos  1  cos 
so
 
Q      sin   V V
  0    a  da   0  d   ,
 1  cos    const  1  cos   2 0 0
0
l   

giving us the following final expression for the mutual capacitance between the wires (also per unit
length):
C Q  0  0
   .
l lV 0 cosh d / 2 R 
1

According to this result,

C  1 lnd / R , for R  d ,
  0   (**)
l R t 1 / 2 , for t  d  2 R  d .
The first of these expressions may be derived in the simple way discussed for the two spheres in
Sec. 2.2 of the lecture notes – see Fig. 2.4 and the accompanying text. Namely, assuming that the charge
Q of the right wire is known, and neglecting the left wire effects, we may readily calculate (say, using
the Gauss law) its electric field at an arbitrary point  of the [x, y] plane:
Q ρ  d / 2n x
ER  .
l 2 0  ρ  d / 2 n x 2
Similarly, the field of the left wire (carrying the opposite charge, –Q/l per unit length) is
Q ρ  d / 2 n x
EL   ,
l 2 0  ρ  d / 2n x 2
so on the x-axis,
Q  1 1 
EL  ER  n x   .
2 0 l  x  d / 2 x  d / 2 
Now we can calculate voltage V between the wires approximately by integrating the net field E  EL+ ER
along the free-space segment of the x-axis between the wires:

Problems with Solutions Page 41

Essential Graduate Physics EM: Classical Electrodynamics

 d / 2 R Q  d / 2 R  1 1 
V   x   d / 2 R   x   d / 2 R   
 d / 2 R
 E L  E R  x dx     dx
2 0 l  d / 2  R  xd /2 x  d /2
Q  R d R Q d
  ln  ln  ln , for R  d ,
2 0 l  d  R R   0 l R

thus confirming the first of Eqs. (**) and providing its interpretation.
y
In the second limit, 2R  d, the wires are so close to each other (see the figure
on the right) that each elementary fragment of area dA = ldy may be considered as a
plane capacitor of thickness
ty
t ( y )  d  2R 2  y 2 
y2 y2
, with t  t 0  d  2 R,
1/ 2
 d  2R 
t
R R
whose capacitance is approximately obeying Eq. (2.28) of the lecture notes: dC = 0 x
0dA/t(y). The summation of all such elementary capacitances, effectively connected in
parallel, gives the integral:46
  1/ 2
C dC dy dy R
  0   0    0   , for t  d ,
l l 
ty  t  y / R
2
t 
thus confirming (and explaining) the second of Eqs. (**). This result is of significant practical
importance, in particular, for the so-called twisted-pair transmission lines – to be discussed in Section
7.5 of this course.
Note that the bipolar coordinates allow a straightforward generalization of our result to the
system of two wires with different radii (say, R1 and R2):
C 2 0
 .
1
 
l cosh d  R12  R22 / 2 R1 R2
2

So, these coordinates are very useful. The same is true for their two simple 3D generalizations, the so-
called toroidal coordinates and bispherical coordinates, that may be obtained by additional rotations
about, respectively, the -axis and the -axis. The reader is encouraged to explore these options.

Problem 2.14. Formulate the 2D electrostatic problems that may be simply solved using each of
the following analytic functions of the complex variable z  x + iy:
(i) w = ln z,
(ii) w = z1/2,
(iii) w = z + 1/z,
and solve these problems.

1/2
46 See, e.g., MA Eq. (6.5a). Note that our integral is converging at  y  ~ (Rt) << R, thus giving a posteriori
justification for the used approximation for the function t(y).

Problems with Solutions Page 42

Essential Graduate Physics EM: Classical Electrodynamics

Solutions: For a 2D boundary problem of electrostatics to be readily solvable with a conformal

mapping, the borders of the conductor’s cross-sections should coincide with lines of either constant real
or constant imaginary part of the complex function w.
(i) Plugging the definitions w  u + iv and z  x + iy into the given equality w = ln z, and
separating the real and imaginary parts, we get

u  ln x 2  y 2  , v  tan -1 .
1/ 2 y
x
This means that on the [x, y] plane, the lines of constant u are concentric circles, while those of equal v
are straight lines extending from the origin to infinity. As a result, for this conformal map, two examples
of easily solvable boundary problems are:
Problem A. Calculate the field distribution between two concentric cylinders of radii a and b,
held at different potentials. (This is exactly the coaxial cable problem already solved in Sec. 2.3 of the
lecture notes).
Problem B. A similar problem for two cylindrical wedges, barely
separated at the origin – see the figure on the right. (This is exactly the second V
 
part of Problem 9.) 2

The solution of Problem A: since the analytic function u(x, y) satisfies the
Laplace equation, and the conductors’ surfaces are equipotential, any line of 
constant u should be equipotential, so the potential  may depend only on u.
Hence the Laplace equation is reduced to d2/d2u = 0; its solution is a linear
function of u:

  c1u  c 2  c1 ln x 2  y 2
1/ 2

 c 2  c1 ln ρ  c 2 ,  
V
2
where constants c1,2 may be readily calculated from the potentials fixed on the
concentric conducting cylinders. This is of course just the result given by Eq. (2.43), which was obtained
in Sec. 2.3 of the lecture notes by using a more general approach.
For Problem B, a similar argumentation yields the same result as was obtained in the solution of
Problem 9 (with the notation replacement   /2 –):
y
  c1v  c 2  c1 tan -1  c 2  c1  c 2 ,
x
which is valid in the free space between the conducting wedges.
(ii) In the same fashion, separating the real and imaginary parts of the function w = z1/2, we get

   
1/ 2 1/ 2
1
u   x 2  y 2 
1/ 2 
x ,
1
v   x 2  y 2 
1/ 2 
x , (*)
2  2 
where the signs in both formulas should be changed simultaneously.47 As the second of these relations
shows, the lines of constant v on the [x, y] plane are parabolas:

47 As evident from Fig. 2.9b of the lecture notes, which shows the reciprocal map z = w1/2, in our current
problem, each point {x, y} is mapped on two points {u, v}.

Problems with Solutions Page 43

Essential Graduate Physics EM: Classical Electrodynamics

10
y2
x  2  v2 ,
4v
v 3
in the limit v  0 giving the straight segment 0 < x < + – v2
y
see the figure on the right. (As the first of Eqs. (*) shows, the v 1
lines of constant u are similar, with the opposite direction of v0
parabola bending.) 0

This mapping may be used, for example, to find the

field distribution around the edge of a thin conducting sheet
occupying the half-plane x > 0, y = 0. For this problem,
d2/dv2 = 0, with the obvious solution

 
1/ 2  10
1
   10 0 10
1/ 2 x
  c1v  c 2  c1  x 2  y 2 x  c2 .
2 
The condition of a fixed potential (say,  = 0) of the conducting half-plane corresponding to v = 0 gives
c2 = 0, so finally
 
1/ 2

  c1  x 2  y 2   x  ,
1 1/ 2 
(**)
2 
where the constant c1 is determined by the potential at the (distant) external electrode creating the
electric field. This means that the figure above shows the equipotential lines corresponding to  = c1v.
Near the surface of the half-plane conductor (x > 0, y << x), Eq. (**) yields
y
  c1 ,
2x
so both the electric field at the sheet surfaces,
 c1
En   y 0  ,
y 2x
and the surface charge density  = 0En have the same strong (non-integrable) divergences at the edge,
as in Problem B of Task (i), discussed above. (See the discussion in the model solution of Problem 9.)
(iii) The given complex function,
1 1  1   1 
w  u  iv  z   x  iy   x 1  2   iy 1  2
2 
,
2 
(***)
z x  iy  x y   x y 
maps the left and right parts of the x-axis (with, respectively, x  0 and 0  x) on the corresponding
“outer” parts (with  u   1) of the u-axis:

1   1  u  , for 0  x  ,

w y 0  x  , so 
x    u  1, for    x  1,

while in both cases, v = 0. The remaining “inner” part of the u-axis (with  u   1) is covered by mapping
of the unit-radius circle on the [x, y] plane. Indeed, taking the complex argument z in the polar form
exp{i}, for the particular value  = 1, we get

Problems with Solutions Page 44

Essential Graduate Physics EM: Classical Electrodynamics

w  1  exp i   exp i   2 cos  , so u  2 cos  , v  0 .

Also, both vertical semi-axes z = iy are mapped onto the whole v-axis:
 1 1
w x 0  iv  i  y   , so v  y  , u  0 .
 y y
Finally, for all very distant points, the mapping is identical:
w  z , in particular v  y, for z  x 2  y 2  1/ 2
 1 .
The figure below summarizes the mapping.

plane w y plane z
v

 1


1 0 1 u 0 x

Now let us consider the following boundary problem: a function  satisfies the 2D Laplace
equation on the whole [u, v] plane, with the following boundary condition imposed at large distances:
  cv, for v   ,
where c is a constant. Its evident solution, valid on the whole plane, is also  = cv, in particular giving 
= 0 on the whole horizontal axis v = 0. Plugging in the expression for the function v(x, y) following from
Eq. (***), we see that the resulting function,
 1 
  cy 1  2 ,
2 
(****)
 x y 
gives the solution of the 2D Laplace equation on the [x, y] plane, with the corresponding boundary
conditions:
  cy, for y  ;

  0, 
for y  0 and ρ  x 2  y 2 1/ 2
 1.

But with the proper choice of the constant (c = –E0), and rescaling of x and y to an arbitrary
radius R, this is just the conducting-cylinder problem solved by another (variable-separation) method in
Sec. 2.6 of the lecture notes. Indeed, the functional form of Eq. (****) coincides with Eq. (2.117), with
the x and y swapped to conform with the coordinate choice accepted in that solution – see Fig. 2.15.
So, for this particular function, just for function (i), the conformal mapping does not give us any
new results. However, in physics, having discretion in the choice of the solution methods is not only
aesthetically pleasing but sometimes practically fruitful, because different approaches may suggest
generalizations to different groups of more complex problems.

Problems with Solutions Page 45

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.15. On each side of a cylindrical volume with a rectangular cross-section ab, with no
electric charges inside it, the electric field’s component normal to the side’s plane is constant, and also
equal and opposite to that on the opposite side. Calculate the distribution of the electric potential inside
the volume, provided that the magnitude of the normal components on the sides of length b equals E.
Suggest a practicable method to implement such potential distribution.
Solution: First of all, let us calculate the normal component E’ of the field on the other pair of
walls. Since the volume is free of charge, the Gauss law (1.16) applied to its unit length yields
b
 2 Eb  2 E'a  0, i.e. E'  E
. y
a E y   Eb / a
b/2
Hence, in suitable Cartesian coordinates (see the figure on
the right), we get the following 2D boundary problem for E x   E Ex  E
the distribution of the electrostatic potential (x, y):
a 0 a x
  
2
 
2
a b 2 2
 2  2   0, for x  , y  ;
 x y  2 2 b/2
E y   Eb / a
  b
  E,  E .
x x  a / 2, y b / 2 y y b / 2, x a / 2 a

There are two ways to solve this problem. The regular (“pedestrian”:-) way is to use the standard
variable separation method,48 with due respect to the evident symmetry of our system (in the above
coordinates, taking (0, 0) = 0, we have (–x, y) = (x, –y) = (x, y), so /x = 0 at y = 0, and /y = 0
at x = 0), and then use the summation formula49

 1n cos n   2   2  1 ,

n 1 n2

4 

3
for   1 ,

to bring the result to the following very simple form (in particular, independent of b):
x2  y2
E
. (*)
a
Though this exercise is highly recommended to the readers, they could notice that our boundary
problem is satisfied by the field that was already calculated in Sec. 2.4 of the lecture notes – see Eqs.
(2.76)-(2.77) for the case V = 2Ea. This means, in particular, that the corresponding potential
distribution (*) may be created in practice, for example, by using the quadrupole electrostatic lens
shown in Fig. 2.9a.

Problem 2.16. Complete the solution of the problem shown in Fig. 2.12 of the lecture notes
(reproduced below), by calculating the distribution of the surface charge on the semi-planes. Can you
calculate the mutual capacitance between the semi-planes (per unit length of the system)? If not, can you
estimate it?

48See Sec. 2.5 of the lecture notes.

49It may be readily proved by the Fourier expansion of its right-hand side on the interval –1    +1, repeated
with period  = 2 over all the -axis.

Problems with Solutions Page 46

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Near the electrode’s surface (y  0, x2 >

2
t ), Eq. (2.83) of the lecture notes yields
V  y2  1 1 
 1
sin 1    
  4x  x  t x  t  V / 2 V / 2
V Vy t t x
  .
2  ( x  t 2 )1 / 2
2

From here, the only non-vanishing (vertical) component

of the field is
 V
E n y  0   y  0   ,
y  x 2  t 2 
1/ 2

so, according to Eq. (1.24) of the lecture notes, the full (double-side) surface charge density is
2 0V
  2 0 E n .
 x 2  t 2 
y  0 1/ 2

Its divergence at x  t is not too strong (it is integrable), just as in the thin-disk problem solved
in Sec. 2.4, but in contrast to that problem, the total charge of each semi-plane (per unit length) is still
infinite, because the corresponding integral,

 
 
2 V 2 0V
ln x  x 2  t 2 
Q dx 1/ 2 x
   dx  0  x  x t , (*)
l t
 t
2
t 
2 1/ 2 
diverges at the upper limit, if only very slowly (logarithmically). Hence in order to calculate the mutual
capacitance per unit length, C/l  Q/lV, one needs to make one more step from this idealized model
toward reality – for example, take into account a large but finite width w of each semi-plate. Even
without solving the resulting problem exactly, we can use Eq. (*) to make a semi-quantitative prediction
of the result by truncating the integral at x = w >> t:
C Q 2 0

l lV



ln x  x 2  t 2  1/ 2
 x  wt
x t 
2 0

ln
2w
t
.

V / 2
Problem 2.17. A straight, long, thin, round-cylindrical conducting pipe
has been cut, along its axis, into two equal parts – see the figure on the right.
(i) Use the conformal mapping method to calculate the distributions of R
the electrostatic potential created by voltage V applied between the two parts,
both outside and inside the pipe, and of the surface charge.
(ii)* Calculate the mutual capacitance between the pipe’s halves (per
unit length), taking into account a small cut width 2t << R.
Hints: In Task (i), you may like to use the following complex function:
V / 2
Rz
w  ln ,
Rz
while in Task (ii), you may use the solution of the previous problem.

Problems with Solutions Page 47

Essential Graduate Physics EM: Classical Electrodynamics

Solutions: y
(i) Let us consider the plane of the pipe’s cross-section as the plane V / 2 
of the following complex variable:
z  x  iy  ρe i , 
where {x, y} are the Cartesian coordinates, and {, } are the polar 0 x
coordinates, introduced as shown in the figure on the right – so, x = cos R
and y = sin. Let us analyze the conformal mapping performed by the
function given in the first Hint. For the points on the pipe’s walls ( = R,  V / 2
i.e. z = Rei):
R  Re i 1  e i e  i / 2  e i / 2 2 cos / 2  
w  ln  ln  ln  ln   ln   i tan ,
R  Re i 1  e i e i / 2  e i / 2  2i sin  / 2  2
so, by using the standard notation w = u +iv (where u and v are real), we get

e w  e u e  iv  e u cos v  i sin v   i tan
.
2
Since the right-hand side of this equality is purely imaginary, so should be the expression before it,
which is only possible if cosv = 0, i.e. at the horizontal lines sinv = 1, e.g., v = /2.50 On these two
lines, the above relation yields, respectively,

 e u  tan .
2
Since u is real by definition, e–u may be only positive, so as  is increased from 0 to + (i.e. as
we move, on the z-plane, along the pipe’s top wall), the corresponding point on the plane w moves from
the right to left along the horizontal line v = +/2. Similarly, as  is decreased from 0 to –, the point in
plane w moves, in the same horizontal direction, along the line v = –/2 – see the figure below.

plane w plane z
   / 2 y  
v
    0
 y
  0

  V / 2  /2 V / 2

 0
0 u x
V / 2
  V / 2  / 2

  0 
    0
   / 2 y  

50 Due to the 2-periodicity of z as a function of , and the function w(z) being logarithmic, the whole mapping
has the 2-periodicity along the v-axis, so we may limit our analysis to the segment –  v  + shown in the
figure.

Problems with Solutions Page 48

Essential Graduate Physics EM: Classical Electrodynamics

Absolutely similar analyses show that the function w also maps (see the figure above):
– the horizontal axis of the z-plane (y = 0): onto that (v = 0) of the plane w,
– the vertical half-axes of the y-axis (with  = /2): onto the respective segments of the u-axis,
0  v  + and –  v  0, and
– any horizontal segment of the z-plane, with y >> x  R: onto a horizontal segment on plane
w, with v approaching sgn(y). (Indeed, if y  , then
R  x  iy 1  ( x  R ) / iy R
e w  e u cos v  i sin v     1  2i ,
R  x  iy 1   x  R  / iy y
so cosv  –1, while sgnv = sgny.)
To summarize, the function w maps the pipe’s interior on the horizontal strip –/2  v  +/2
(shaded darker in the figure above), and two parts of its exterior, separated by the x-axis, onto two strips:
+/2  v  + and –  v  –/2 (shaded lighter).
Now we may return to our boundary problem. Since all the boundary conditions imposed on the
electrostatic potential  , and the unnecessary but convenient requirement  = 0 at  = 0, are independent
of u (see the labels on the left side of the figure above), and the function w is analytic, a linear function
(v) = c1v + c2 satisfies the Laplace equation inside each strip and, at the appropriate choice of the
coefficients c1,2 (specific for each strip), the boundary conditions as well. With the proper selection of
c1,2, we get
   v , for   / 2  v   ,
V  (*)
    v, for   / 2  v   / 2,
 
   v , for    v   / 2,
(It is straightforward to use the left-side labels in the figure above to verify that this solution indeed
satisfies all the boundary conditions.)
What remains is to express v via the coordinates on the z-plane. For that, we may use the fact
that if w = lnF  ln(Fexp{iarg(F)})  ln(F) + iarg(F), then v  Im(w) = arg(F). In our case, we can
use this identity with F = (R + z)/(R – z), so

v  Im w  arg
R  x  iy
 arg
R  x  iy R  x  iy   arg R 2  x 2  y 2  2iRy .
R  x  iy R  x  iy R  x  iy   R  x 2  y 2
The denominator of the last fraction is, by construction, real, and does not affect its argument, so
2 Ry 2 Rρ sin 
tan v   2 ,
R x y
2 2 2
R  ρ2
and taking into account the -periodicity of the tangent function, all Eqs. (*) may be rewritten in a single
simple form:
V 2 Rρ sin 
  ρ,    tan 1 2 . (**)
 R  ρ2
The left panel of the figure below shows the general pattern of the equipotential surfaces (or
rather their cross-sections by the plane normal to the pipe’s axis) given by this formula. Close to the

Problems with Solutions Page 49

Essential Graduate Physics EM: Classical Electrodynamics

pipe, the surfaces naturally align to it; they merge in the gaps between the half-pipes, where the electric
field is concentrated.

2 0.5
ρ / R  1 .1
0 .9
1 1 .5
0 .5
 3 .0
y V 0 .1
0 0
R

1

2  0.5
1 0 1
2 1 0 1 2  /
x/R
To illustrate this behavior in more detail, the right panel of the figure below shows the potential
 as a function of  for several fixed values of , both larger than R (red lines) and smaller than R (blue
lines). These plots show, in particular,51 that small deviations from the pipe’s surface to either side lead
to similar deviations of the potential from its piece-constant form at the surface. This fact implies that
the magnitude of the electric field on the pipe’s surface,  En  = / at  = R, and hence the surface
charge density  = 0En is the same, for the same , on both (the external and the internal) surfaces of
the pipe. Indeed, a direct differentiation of Eq. (**) over  gives the same result:
 0V
    . (***)
R sin 
(ii) The charge density (***) exhibits strong (non-integrable) divergences at   0 and   ,
i.e. near the gap between the two halves of the pipe; for example, at   0,
 0V  0V
     , for y  R ,
R  y
so the total charge Q of each half of the pipe, and hence the capacitance C = Q/V of the system (even per
unit length along the z-axis) are formally infinite. This is an artifact of neglecting the cut’s width 2t in
the above calculation. In the limit 2t << R, finite Q and C may be found by comparing the above
solution with that of the previous problem (on the field distribution between two planar, thin electrodes
separated by a cut of width 2t), which overlap in a vicinity of the gap.
Indeed, with the proper notation replacement x  y, the solution of the previous problem gives
the following result for the single-side charge density:
 0V
    sgn  y  , for y  t , (****)
 y 2  t 2 
1/ 2

51 For an analysis of the potential’s behavior far outside of the pipe, see the model solution of the next problem.

Problems with Solutions Page 50

Essential Graduate Physics EM: Classical Electrodynamics

which gives the same result as Eq. (***) within a relatively broad range of angles:
t
   1 .
R
Hence Eq. (****) may be used instead of Eq. (***) at y  R << R. As a result, the total charge of
the upper half of the pipe (per unit length), may be calculated by the separation of the corresponding
integral into two parts:52

Q
 /2  0  /2 
 2    Rd  4 R     d  4 R      d      d  ,
l  
sin  0 t/R t / R 0 
where the separation boundary 0 is arbitrary within the solutions’ overlap region. The inequality t/R <<
0 allows us to use Eq. (****) in the first integral, while the inequality 0 << 1 makes the approximation
y = R valid in it. On the other hand, due to the former relation, we may use the uncorrected expression
(***) for the charge density in the second integral:

Q  0V  0 d 1
 /2
d 
l
 4R
 
 t/R R  t2
2 2
 
1/ 2

R  sin  
.
 0 
The remaining calculations are simple. Using variable replacements   (t/R)cosh, so d =
(t/R)sinhd in the first integral, and   cos, so d/sin = –d/sin2  –d/(1 – 2) in the second one,
we get
   
Q 4 0V  0 d  4 0V  0 1  
  / 2   / 2
sinh d 1  1
   t / R cosh 2   11 / 2   1   2    t / R 2    1   1    
    d      d 
l
 0   0 
4 0V   0 1 1    / 2  4 0V  R 0 1 1  cos  0 
   ln    cosh 1  ln  .
   t / R
2 1
 0
   t 2 1  cos  0 
Due to the condition 0 >> t/R, cosh-1(R0/t) may be approximated as ln(2R0/t)  ln(4R/t) + ln(0/2),
while due to the condition 0 << 1, the logarithm in the last term reduces to ln(4/02)  –2ln(0/2). As a
result, the terms with 0 cancel (as they should, due to the qualified-arbitrary choice of this separation
boundary), and we finally get
Q 4 0V 4 R C 4 4R
 ln , i.e.  0 ln .
l  t l  t
My earnest advice to the reader is to review the used trick of matching (“stitching”) two
approximate solutions that share a broad overlap region because it is used in many problems of not only
physics but also other quantitative sciences and engineering.

52The front factor of 2 takes into account equal densities () of the charge at the inner and outer surface, while
the next step uses the obvious mirror symmetry of  with respect to the vertical axis x = 0 ( = /2).

Problems with Solutions Page 51

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.18. A gap of constant width w between two grounded y

conducting semi-spaces is closed, from one side, with a conducting
plunger biased with voltage V, so that the cross-section of the system looks  0  0
like the figure on the right shows. Use the variable separation method to
calculate the distribution of the electrostatic potential within the gap.
Solution: Let us use the convenient Cartesian coordinates shown in
the figure on the right. In this case, the variable separation may be carried
out similarly to how this was done in Sec. 2.5 of the lecture notes for a  V
rectangular box, with just a few changes. First, now we are dealing with a 0 x
w w
2D rather than 3D problem, so Eq. (2.85) does not need the factor Z(z).  
Second, with our current choice of the origin, the solutions X(x) of Eq. 2 2
(2.89), that satisfy the boundary condition  = 0 on the gap’s walls, are not the sine functions but rather

X  x   cos kx, with k  2n  1, where n  1, 2, 3. (*)
w
Finally, the corresponding function Y(y) satisfying both the Laplace equation and the condition   0 at
y  , is exp{–ky} with the same k as in Eq. (*), so the solution of the boundary problem is given by
the following series:

 2n  1x  2n  1 y  w w
 ( x, y )   c n cos exp  , for   x   , 0  y.
n 1 w  w  2 2
The coefficients cn should be found from the still unused boundary condition  (x, 0) = V:

 2n  1x w w
V   c n cos , for  x .
n 1 w 2 2
Multiplying both sides of this equation by cos[(2n’ – 1)x/w] and integrating the result over x from –w/2
to +w/2, we get
w / 2
 2n'  1x w / 2
 2n'  1x  2n  1x
V  cos dx  c n  cos cos dx .
w / 2
w w / 2
w w

Now, elementary integrations yield cn = 4V(–1)n-1/(2n – 1), so, finally,

4V 
 1n 1 cos  2n  1x exp  2n  1 y . 0.01



n 1 2n  1 w

 w


0.8 0.03
The figure on the right shows the calculated
potential as a function of x, for several values of y. As  0 .1
0.6
might be expected, near the voltage-biased plunger, the
V
electrostatic potential closely follows its value V, 0.4
besides small regions near the side walls where it 0 .3
transitions to their (zero) value of . However, far from 0.2
the plunger, its effect rapidly decreases, and the y/w 1
potential drops exponentially, as described by the first
0
term of the series, which dominates at large y: 1  0.5 0 0.5 1
x /(w / 2)

Problems with Solutions Page 52

Essential Graduate Physics EM: Classical Electrodynamics

4V x  y 
 cos exp , for y  w .
 w  w
V / 2

Problem 2.19. Use the variable separation method to calculate the a a

electrostatic potential’s distribution inside a very long thin-wall metallic box with
a quadratic cross-section, cut and voltage-biased as shown in the figure on the
right. (Assume that the cut’s width is negligibly small.)
a a
Solution: The easiest way to solve this problem is to avoid using the
Cartesian coordinates implied by the figure in the assignment, with one of the V / 2
coordinate axes directed along the cut, but instead let the coordinate y
axes pass through the middle of each wall. The figure on the right a V / 2
shows the resulting configuration, turned clockwise by /4 to make 
2
the axes horizontal and vertical – just to comply with tradition and
make the forthcoming formulas more intuitive.
0
The second useful move is to use the linear superposition
principle to break the desired potential into two parts, with its a/2 a/2 x
components satisfying different boundary conditions:
  h  v , a

where V / 2 2
 V / 2, for x   a / 2, 0, for x   a / 2,
h   v   (*)
0, for y   a / 2,  V / 2, for y   a / 2.

The advantage of this separation is that the “horizontal” potential h is evidently an odd function
of x and an even function of y, while the “vertical” potential has the opposite symmetry. Let us start with
the former of these functions, carrying out the variable separation as for the 3D box in Sec. 2.5 of the
lecture notes and in the previous problem, but taking into account that the zero boundary conditions at y
= a/2 make it more natural to use trigonometric functions along this coordinate:

 h   c n sinh
2n  1x cos 2n  1y , (**)
n 1 a a
immediately satisfying these conditions and both mirror symmetries of the function. As usual for the
variable separation method, the coefficients cn may be found from the requirement that this solution
describes the fixed potential +V/2 on the wall with x = +a/2:53

2n  1 2n  1y  V .
c
n 1
n sinh
2
cos
a 2
Multiplying both sides of this equation by cos(2n’ – 1)y/a and then integrating them over the segment –
a/2  y  +a/2, we get

53 The boundary condition on the opposite wall, with x = –a/2 will be then satisfied automatically, due to the
antisymmetry of the function (*) with respect to the inversion x  –x.

Problems with Solutions Page 53

Essential Graduate Physics EM: Classical Electrodynamics

2n  1  a / 2 cos 2 2n  1y dy  V  a / 2 cos 2n  1y dy, 2 1 V
n
c n sinh
2 
a / 2
a 2 
a / 2
a
giving c n 
n sinh n  ½ 
..

As a result, Eq. (**) becomes

2V 
 1n 2n  1x cos 2n  1y .
h  
 n 1 2n  1 sinh n  ½ 
sinh
a a
(***)

Now we may notice that since at the swap x  y, the boundary conditions (*) are also swapped,
the same has to be true for the functions h and v, so for the latter function we may use Eq. (***) with
this replacement, and the full solution of our problem becomes
2V 
 1n  2n  1x cos 2n  1y  cos 2n  1x sinh 2n  1y  .
h  
 n 1 2n  1 sinh n  ½  
sinh
a a a a 

The figure on the right shows the equipotential y
surfaces described by this result, drawn with equal steps
 = V/20. They naturally concentrate near the cut gaps
where the electric field increases becoming formally
infinite in these two corners of the cross-section.
If for some reason we need to express the
potential as a function of the coordinates (say, {x’, y’})
implied by the problem’s assignment, we may always
x
use the above solution with the substitutions
x'  y'  x'  y'
x , y ,
2 2
corresponding to the back (counterclockwise) rotation of
the coordinate axes. (If you want just to visualize the
solution, it is easier just to turn your head by /4
clockwise, and have one more look at the figure on the
right :-)

Problem 2.20. Solve Problem 17(i) by using the variable separation method, and compare the
results.
y
Solution: Introducing the polar coordinates as shown in the figure V / 2 ρ
on the right (i.e. just as in the model solution of Problem 17) and guided by
Eq. (2.112) of the lecture notes and the symmetry of the problem, we may
look for the solution of the outer problem (for   R) in the form 

bn 0 x
  ,     sin n .
n 1 ρn R
Indeed, the inconsequential coefficient a0 in Eq. (2.112) may be taken  V / 2
equal to zero for convenience. The logarithmic term in that expression,
proportional to b0, would describe the effect of the net charge of the pipe, which we obviously do not

Problems with Solutions Page 54

Essential Graduate Physics EM: Classical Electrodynamics

have. The terms proportional to an with n > 0 would diverge at    and thus should be excluded, and
the terms proportional to cosn would give even-function contributions into the potential, which in our
problem should be an odd function of .
The boundary conditions at  = R yield the following system of equations for the coefficients bn:

bn V   1, for 0     ,
R sin n   R,    
n 1
n
2   1, for     2 .
Solving this system in the way usual for the Fourier series, i.e. by multiplying both parts of the equation
by sinn’ with an arbitrary n’ > 0, integrating them over any 2-long interval (for example, 0    2),
and using the orthogonality of the functions sinn with different n:
2

 sin n' sin n d  

0
nn' , for n  0 ,

we obtain

V V n V n 0, for n even,
bn  R n  sin n d  R 1  cos n   R 
 0
n n  2, for n odd,
so denoting the odd values n = 1, 3, … as 2m – 1 (with m = 1, 2, …), we finally get
2 m 1
2V 1 R

 ρ,       sin 2m  1 , for ρ  R . (*)
 m 1 2m  1  ρ 
The solution of the internal problem (for   R) is similar, besides that instead of terms
proportional to bn/n in Eq. (2.112), we need to keep the terms ann that do not diverge at   0:

  ρ,     a n ρ n sin n .
n 1

Since the boundary conditions at  = R are similar for both problems, the above result for bn is valid for
the normalized coefficients an as well (with the replacement bn/Rn  anRn), so the final result is very
similar to Eq. (*):
2 m 1
2V  1  ρ
 ρ,       sin 2m  1 , for ρ  R . (**)
 m 1 2m  1  R 
Numerical plots of Eqs. (*) and (**) show that they yield results identical to the explicit
expressions obtained in the solution of Problem 17(i).54 Moreover, these series are equally (if not more)
convenient for the analysis of potential behavior far from the pipe’s walls. Indeed, as Eqs. (*) and (**)
show, at both  >> R and  << R, the series terms decay fast with n, so potential is determined by their
first terms, with m = 1, i.e. is proportional to sin. In the former case,
2VR
 ρ,    sin  , for ρ  R ,
ρ

54Their exact equivalence may be also proven analytically by the Fourier expansion of Eqs. (**) of the model
solution of Problem 17(i) – an optional exercise highly recommended to the reader.

Problems with Solutions Page 55

Essential Graduate Physics EM: Classical Electrodynamics

the potential corresponds to the electric field of magnitude E  1/2. As will be discussed in Chapter 3
of this course, such a field may be interpreted as the one created by a straight line of similar electric
dipoles. On the other hand, near the cylinder’s axis, we get,
2V
  ρ,    ρ sin  , for ρ  R , (***)
R
meaning that the electric field here is uniform. Indeed, since the product sin is just y (see the figure
above), Eq. (***) gives
2V 2V
 y, so that E     n y  const.
R R

Problem 2.21. Use the variable separation method to calculate the potential distribution above
the plane surface of a conductor with a strip of width w singled out with very thin cuts and biased with
voltage V – see the figure below.
y
 w/ 2  w/2
 0  V  0 x

Solution: Let us introduce the Cartesian coordinates as shown in the figure above, and separate
the variables similarly to Eq. (2.85) of the lecture notes, besides that in this 2D problem, the partial
functions Z(z) are evidently constant and may be taken for 1:
 k  X k x Yk  y  ,
so Eq. (2.87) may be rewritten as
1 d2Xk 1 d 2Yk
   k 2  const .
X k dx 2 Yk dy 2
This notation for the separation constant is convenient in our current case because since the electrostatic
potential has to tend to zero at y  , only exponentially decaying solutions of the above equation for
Yk(y),
Yk  y   e  ky ,
with k > 0, are acceptable.55 With that, the possible solutions of the above equation for the functions
Xk(x) are coskx and sinkx. Taking into account the mirror symmetry of the system with respect to the
plane x = 0, only the first of them are acceptable:
X k  x   cos kx .
Now comes the largest difference between this problem and the one discussed in Sec. 2.5(i) of
the notes: due to the absence of any period of the system in any direction,56 the spectrum of possible

55Strictly speaking, complex values of k, with Rek > 0, are also acceptable, but as will be shown below, the full
solution of the problem may be constructed without involving imaginary parts of k.
56 The problem analyzed in the lecture notes (see Fig. 2.13) may be considered a periodic one, with the
rectangular volume repeated infinitely in all three directions.

Problems with Solutions Page 56

Essential Graduate Physics EM: Classical Electrodynamics

values of k is continuous rather than discrete. As a result, the sum (2.84) now becomes an integral over
k; and with our current eigenfunctions Xk(x) and Yk(y), it is

  x, y    c k cos kx e ky dk . (*)
0

The continuous set of coefficients the ck (essentially, a continuous function of k) may be found from the
boundary condition at y = 0 and (due to the problem’s symmetry) x  0:

V , for 0  x  w / 2,
c k cos kx dk    x,0    (**)
0 0, for w / 2  x  .
Since this is just the expansion of the function on the right-hand side into the Fourier integral, the way of
calculating ck is well known: we need to multiply both parts of Eq. (**) by cosk’x with an arbitrary k’,
and then integrate the result over x. Changing the order of integration on the left-hand side, we get
   w/ 2

 ck dk  cos k'x cos kx dx    x,0cos k'x dx  V

0 0 0
 cos k'x dx .
0
(***)

The integral over x on the left-hand side is just a sum of two delta functions,57
 
1 
0 cos k'x cos kx dx  4 Re expi k  k' x  expi k  k' xdx  2  k  k'    k  k' ,
while that on the right-hand side is elementary:
w/ 2
1 k'w ,
 cos k'x dx  k' sin
0
2
so for any sign of k’, Eq. (**) reduces to
 V k'w
c k'  sin .
2 k' 2
Plugging the resulting expression for ck’ (with the index k’ replaced with k) into Eq. (*), we get

2V kw dk
  x, y    sin cos kx e ky . (****)
 0
2 k
In this particular case, the integral over k may y / w  0.03
be worked out analytically58 but generally, such an 0.8 0.1
integral form of the result is typical for variable-
separation solutions. Plots in the figure on the right  0.6 0.3
show the potential  as a function of x, for several
V
fixed values of y. We can see that close to the 0.4
conductor’s surface, the voltage-induced potential
bump is almost rectangular, with a width of x  w 0.2
1.0
and a height approaching V, but at large distances
from the surface, the bump is lower and more spread 0
3.0
 1.5 1  0.5 0 0.5 1 1.5
out, with a width x of the order of y >> w. x/w
57 See, e.g., MA Eq. (14.4).
58 See, e.g., the model solution of Problem 42.

Problems with Solutions Page 57

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.22. The previous problem is now modified: the cut-out and voltage-biased part of the
conducting plane is now not a strip, but a square with side w. Calculate the potential distribution above
the conductor’s surface.
Solution: This problem is conceptually very similar to the previous one, besides that now the
potential’s distribution is three-dimensional (just like in the problem solved in Sec. 2.5 of the lecture
notes), so its solution is essentially a synthesis of the two known solutions, and the explanations below
will be very short. Let us use the Cartesian coordinates with the axes x and y in the plane of the
conductor’s surface and parallel to the cut square’s sides, and the z-axis directed normally to the plane,
toward the free space, with its origin in the square’s center. Then the variable separation yields
 
  x, y, z    d  d c cos x cos  y e z , (*)
0 0

where the separation constants are related by Eq. (2.88) of the lecture notes:

   2   2 
1/ 2
0.
The boundary condition at the conductor’s surface is
  V , for x  w / 2 and y  w / 2,
0 0
d d c cos  x cos  y    x , y ,0   
0, otherwise .
It is clearly satisfied by the product c = Vf()f(), where the function f obeys the condition

 1, for 0  s  w / 2,
 f   cos s d  
0 0, for w / 2  s  .
But such a function was already calculated in the previous problem:
2 w
f    sin ,
 2
so, finally, the solution (*) may be spelled out as

 
 
w d w d
  x, y , z  
4V
 2  sin
2
cos x
  sin 2
cos  y


exp   2   2  1/ 2
z .
0 0

This result is plotted, in the figure on the right,

as a function of one of the horizontal coordinates (x), z / w  0.03 y0
for the counterpart horizontal coordinate (y) fixed at 0.8
0.1
zero, and the same values of the vertical coordinate z as
were used for the similar plots in the model solution of  0.6
the previous problem. (Note the difference in the V
vertical coordinate’s notation.) 0.3
0.4
The comparison of these two figures shows
that, quite naturally, the voltage-induced potential 0.2 1.0
bump decays, with height, faster (at z >> w, much
3.0
faster) in our current case of a voltage-biased square 0
1 0 1
x/w

Problems with Solutions Page 58

Essential Graduate Physics EM: Classical Electrodynamics

than in the previous case of a biased infinitely long strip.

V V V
Problem 2.23. Each electrode of a large plane capacitor is     
cut into parallel long strips of equal width w, with very narrow gaps 2 2 2
between them. These strips are kept at alternating potentials, as
shown in the figure on the right. Use the variable separation method w d
to calculate the electrostatic potential distribution in space, and
V V
explore the limit w << d.   V   
2 2 2
Solution: Due to the symmetry of the problem, the
electrostatic potential should vanish at:
(i) the horizontal symmetry plane in the middle between the planes, and
(ii) any vertical plane passing between the strips.
Selecting these planes (or rather their traces on the plane of the drawing) for the coordinate axes, we see
that each of the functions X(x) and Y(y) forming any particular solution k = XY has to be odd. Since the
functions X(x) also have to be periodic, with the period x = 2w, they may be taken in the form of sinkx
with k = n/w, where n = 1, 2, 3…. Hence, in order to satisfy the Laplace equation, the corresponding
functions Y(y) have to equal sinhy, so the full solution takes the form

nx ny
 ( x, y )    n sin sinh . (*)
n 1 w w
The coefficients n should be found from the boundary conditions on the electrodes. Since Eq.
(*) already ensures the proper symmetry and periodicity of the solution, it is sufficient to require that it
fits the boundary value on just one strip (say, 0 < x < w, at y = +d/2, where  = +V/2):
V 
nx nd
   n sin sinh .
2 n 1 w 2w
Applying the reciprocal Fourier transform formula (or just multiplying both sides of the last equation by
sin(xn’/w) and integrating the result over x from 0 to w), we readily get
 2V  nd 
1

 n   n  sinh 2w  , for n odd,
0, for n even.

The plots in the figure below show the potential’s distribution between the plates for two values
of the ratio d/w and for three distances y from the central plane. For any d, close to the electrodes, the
potential follows the square-wave profile dictated by the applied voltages. However, if d >> w, this is
true only very close to their surfaces, while in most of the volume, the potential is relatively small and
close to the simple sine function given by the first term of the series (*):
2V sinh y / w x
 ( x, y )  sin .
 sinh d / 2w w
The solution in the outer regions (at  y  > d/2) is very similar, with the multiplier sinh(ny/w)
replaced with sgn(y)exp{–n y/w}.

Problems with Solutions Page 59

Essential Graduate Physics EM: Classical Electrodynamics

0.5 0.5
d  0.5w y / d  0.48 d  2w
0.40 y / d  0.48
0.3 0.3
) ) 0.4
V 0.1
0.2 0.1
0.2
0 0
 0.1  0.1

 0.3  0.3

 0.5  0.5
1  0.5 0 0.5 1 1  0.5 0 0.5 1
x/w x/w

Problem 2.24. Complete the cylinder problem started in Sec. 2.7 of z

the lecture notes (see Fig. 2.17, reproduced on the right), for the cases when   V ( , )
the top lid’s voltage is fixed as follows: l
(i) V = V0J1(11/R) sin, where 11  3.832 is the first root of the
Bessel function J1();
(ii) V = V0 = const. 0 R
y
For both cases, calculate the electric field at the centers of the lower x
and upper lids. (For Task (ii), an answer including series and/or integrals is
 0
acceptable.)
Solution: The general solution to this problem was found in Sec. 2.7 of the lecture notes – see
Eq. (2.139):
 
    z
 (  ,  , z )   J n  nm c nm cos n  s nm sin n sinh nm , (*)
n  0 m 1  R  R
where nm is the mth root of the Bessel function Jn(). This solution already satisfies not only the Laplace
equation inside the cylinder but also the boundary conditions on its sidewall and bottom lid; hence the
coefficients cnm and snm may be found from the boundary condition on the top lid:
 (  ,  , l )  V (  ,  ). (**)
(i) In this case, the function V(, ) has only one azimuthal harmonic proportional to sin, and
only one radial harmonic, proportional to J1(11/R), so only one of the coefficients cnm, snm is different
from zero:
 l
s11 sinh 11  V0 ,
R
and Eq. (*) reduces to a simple solution:
 11   sinh 11 z / R 
  V0 J 1   sin  .
 R  sinh 11l / R 

Problems with Solutions Page 60

Essential Graduate Physics EM: Classical Electrodynamics

Since J1(0) = 0, the potential equals zero along the symmetry axis of the system, so in this case
there is no normal electric field En = –/z in the center of either lid. However, since the potential
changes within the top lid’s plane,
     
V  V0 lim  0 J 1  11  sin   V0 11 sin   V0 11 y ,
 R  2R 2R
the field in its center does have a horizontal component directed along the y-axis:
11 V0
Ey  0  V0  1.916 .
2R R
(ii) Here, the top lid’s potential does not depend on the angle , and hence only one angular
function (with n = 0) fits this boundary condition, so Eq. (**) reduces to an axially-symmetric form:

    l
V0   c0 m J 0  0 m  sinh 0 m .
m 1  R  R
Multiplying both parts of this equation by J0(x0m/R), integrating the result over  from 0 to R, and
using Eq. (2.141) of the lecture notes, we get
2V0 R
 
c0 m  2 2  J 0   0 m   d .
R J 1  0 m sinh  0 m l / R  0  R
(Actually, the last integral may be worked out analytically, giving a more explicit result:
2V0
c0 m  ,
 0 m J 1 ( 0 m ) sinh  0 m l / R 
but I did not require my SBU students to accomplish this astonishing mathematical feat :-)
According to Eq. (*), and due to the fact that Jn(0) = 0 unless n = 0 (see, e.g., Fig. 2.18 in the
lecture notes), the potential’s distribution along the symmetry axis of the cylinder is contributed only by
the terms with n = 0:

 z
 ( z )   c0 m sinh 0 m ,
m 1 R
so the (purely vertical) field in the center of the bottom lid (located at z = 0) is
d ( z ) 
 0m
En bottom  Ez z 0  z 0   c0 m , (***)
dz m 1 R
while the field at the top lid (z = l) is
d ( z ) 
 0m  0m l
En top  Ez z l  z l   c0 m cosh .
dz m 1 R R
For any l > 0, each term of the last series is larger than the corresponding term of Eq. (***), so the
field’s magnitude at the bottom lid is always lower – physically, this fact is due to the screening effect
of the cylinder’s sidewall. However, at l/R  0, cosh(0ml/R)  1 for all leading terms in the series, and
this difference is negligible.

Problems with Solutions Page 61

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.25. For the infinitely long periodic system z

sketched in Fig. 2.21 of the lecture notes (reproduced on the right), R
assuming that t << h, R:   V / 2
h
(i) calculate and sketch the electrostatic potential’s
distribution inside the system for various values of the ratio R/h, and   V / 2
(ii) simplify the results for the limit R/h  0. t
Solutions:   V / 2
(i) Due to the problem’s 2h-periodicity along the z-axis, its
particular solutions should be proportional to linear combinations of
sinkz and coskz, with 2kh = 2m, where m = 1, 2, 3…, so the spectrum of possible values of the
parameter k is restricted to k = m/h. Moreover, if the origin of the z-axis is selected at one of the gaps
between the rings, the solution should be an odd function of z, so for the internal problem (  R) we
may write

 m  mz
  , z    c m I 0   sin .
m 1  h  h

This particular modified Bessel function of the first kind has zero index  due to the evident axial
symmetry of the problem – see Eq. (2.128); the function’s argument results from the discreteness of the
variable separation constant k:   k = m/h. Finally, we had to drop the modified Bessel functions of
the second kind from our solution because they diverge at   0 (see the right panel of Fig. 2.22) and
thus cannot be used to represent the finite electrostatic potential inside the system.59
The coefficients cm in the above series should be found from the boundary condition on the
conducting rings; since the proper periodicity is already incorporated into our solution, it is sufficient to
require that

 mR  mz V
 R, z    c m I 0   sin  , for 0  z  h.
m 1  h  h 2
Multiplying both sides of this expression by sin(m’z/h) and integrating the result from 0 to h, we get
1
V   mR   1, for m odd,
cm  I    
m   h 
0
0, for m even,
so with the substitution m = 2n – 1 (where n = 1, 2, 3,…), we get
2V 
1  (2n  1) z I 0  (2n  1)  / h
  , z    2n  1 sin . (*)
 n 1 h I 0  (2n  1) R / h
The figure below shows the plots of this result for the cylinder’s axis ( = 0), on one period of its
z-dependence, for several values of the ratio R/h. At R >> h, the result is dominated by the first term of
the series (with m = 1), which is proportional to 1/I0(R/h), so at R >> h, according to Eq. (2.158), the
potential’s variation amplitude equals (8R/h)1/2Vexp{–R/h} << V.

59Note that these functions have to be used in the solution of the external problem (  R), which is similar to that
discussed above, besides the replacement of I0(m/h) with K0(m/h).

Problems with Solutions Page 62

Essential Graduate Physics EM: Classical Electrodynamics

On the contrary, at R << h, the potential 0.5

follows that of the closest ring electrode at most of R / h  0.1
the axis, besides narrow (z ~ R) intervals centered
to gaps between the electrodes. The exact behavior 0.3
of the potential in these intervals (for example, in 
the close vicinity of the origin,  z  ~ R << h) is of V 0 1.0
considerable practical interest, because it is also
pertinent to a system of just two long coaxial
cylinders, with voltage V between them, which
may serve as an effective electron lens.60
(ii) In order to simplify Eq. (*) in the limit
 0.5
R/h  0, we may notice that in the range z ~ R, 1 0 1
each term of the sum is much smaller than 1, so it z/h
may be well approximated with an integral:
2V

1  (2n  1) z I 0  (2n  1)  / h  V

sin kz I 0 (k )
  , z    sin dn  0 k I 0 kR  dk , for z ~ R  h,
 0 2n  1 h I 0  (2n  1) R / h  
where k  (2n – 1)/h. This integral converges both 0.5
at the lower limit (because at k  0, sin(kz)/k  z), 0.4  / R  0.9
and at the upper limit (because at kR  , the 0.3 0.5
function I0(kR) grows exponentially – see, e.g., Eq. 0.2
0.0
(2.158) and/or the left panel of Fig. 2.22), and is  0.1
convenient both for numerical calculations (for V 0
example, see the figure on the right) and for  0.1
asymptotic behavior analyses, because it does not  0.2
depend on h.  0.3
 0.4
 0.5
2  1.5 1  0.5 0 0.5 1 1.5 2
z/R

Problem 2.26. A long round cylindrical conducting pipe is V / 2 V / 2

split, with a very narrow cut normal to its axis, into two parts that 2R
are voltage-biased as the figure on the right shows. Use two
different approaches to calculate the force exerted by the resulting z
field upon a charged particle flying along the pipe close to its axis.
Can the system work as an electrostatic lens? z0

Solution:
Approach 1. First, we may notice that the distribution of the electrostatic potential everywhere
inside the pipe is given by the solution of the previous problem in the proper limit h/R  :

60 The convenience of such a lens in comparison with the one discussed in Problem 1.9 is that in it, the
accelerating electric field is confined to a limited length of z ~ R.

Problems with Solutions Page 63

Essential Graduate Physics EM: Classical Electrodynamics


V sin kz I 0 (k )
  , z  
 0 k I 0 kR 
dk , (*)

where  is the distance from the pipe’s axis, and z is the 

distance along the axis, referred to the cut’s plane – see R
the figure in the assignment. The figure on the right
shows the corresponding equipotential surfaces drawn
with equal steps V/20; note that the field near the pipe’s
axis is spread by distances z ~ R from the cut’s plane z
= 0. (See also the last figure in the model solution of the
previous problem.)
Since the integral in Eq. (*) converges at kmax ~  R 0 R z
1/min[ R,  z ], for the points close to the axis (with  <<
R), we may expand the modified Bessel function I0(k)
in the Taylor series in its small argument and truncate it
to the leading terms. Per the first of Eqs. (2.159) of the
lecture notes, I0() = J0(i), so we may use Eq. (2.133) to
get
2
 i   2
I 0    1     1  , for   1 .
 2 4
This accuracy is sufficient to calculate both nonvanishing components (longitudinal and radial) of the
force F = –q exerted by this field on a particle of charge q:61
 
 qV cos kz qV  z  cos 
Fz  q
z

 0 I 0 kR  dk   R f z  R , with f z    0 I 0   d ,


V sin kz 1   k 2  qV  z

sin 
F   q  q  1   dk   f  , with f     d .
 0 k I 0 kR   I 0  
 
  4  R 2 R  R  0

The figure on the right shows the integrals 3

participating in these formulas as functions of the
normalized distance   z/R from the cut. Not 2
surprisingly, it shows that both forces are fz f
substantial only at distances z ~ R near the cut and 1
that the longitudinal field is an even function of z,
while the radial field is its odd function.62 As a 0
result, if the initial longitudinal velocity of the
particle is so high that its acceleration or 1
deceleration (depending on the sign of the qV
product) by the longitudinal field is negligible, it 2
does not acquire an appreciable net radial velocity
v necessary for focusing. Note that this potential-
3
3 2 1 0 1 2 3
z/R
61 See MA Eq. (10.2) with / = 0.
62 Note that the radial force is also much smaller, due to the pre-integral factor /2R << 1.

Problems with Solutions Page 64

Essential Graduate Physics EM: Classical Electrodynamics

step situation is very different from the field-step arrangement that was the subject of Problem 1.9, in
which the induced radial velocity and hence the focusing are direct effects.
However, the focusing still happens even in the system under analysis, due to the following
secondary effect. Due to the longitudinal acceleration/deceleration of the charged particle by this field,
the time periods of its passage through the negative and positive “bumps” of the radial force F (see the
figure above again) become different, so their time (rather than length) integrals become different,
giving the particle a non-zero net momentum p = Fdt and hence a non-zero final velocity v = p/m in
the radial direction. Since this velocity has the same proportionality to the initial distance  of the
particle from the axis as the force F itself, the time t = /v and hence the length f = vzt of its crossing
the axis are independent of , so a parallel beam of such particles with different initial values of  << R
is focused at one point z = f. The calculation of this focal distance in the first approximation in small 1/vz
is conceptually straightforward but leads to bulky formulas, and I have to leave this task for an
appropriate special-topic course.
Approach 2. The structure of Eq. (*) used in the previous approach is essentially an artifact of
the z-periodicity of the system considered in the previous problem. For our current system, it is more
natural to use the expansion in a series over the Bessel functions Jn(), as this was done at the beginning
of Sec. 2.7 of the lecture notes for the system shown in Fig. 2.17. However, at that problem’s solution,
we have seen that such expansions are very convenient for the functions turning to zero at  = R. For our
current problem, we may introduce such a function as follows:

~   V / 2, for z  0, ~    R, at any z  0,

  so that   0 for 
  V / 2, for z  0,  z  , at any ρ  R.
One more boundary condition for this function follows from the continuity of the full potential at z = 0:

 z  0  z  0 ~
  z  0  
~
V / 2   z  0  ~
 V / 2  0, so  z  0
~
 z  0  V .
~
The (minor) inconvenience of this approach is that since the function  so defined experiences a jump
equal to V at the plane z = 0, it does not satisfy the Laplace equation exactly on that plane. However, this
~ ~
drawback is compensated by the evident antisymmetry of this function:   , z     , z  , so, in
particular, the above boundary condition yields
~ V
 z  0   , (**)
2
and it is sufficient to solve this boundary problem only for z  0.
Separating the variables  and z may be done as in Sec. 2.4 but with the appropriate choice for
the solution of Eq. (2.126) and with the due account of the axial symmetry of our current problem ( = n
= 0). So, instead of the double series (2.139), we get a single series:
~ 
   z
 z 0   cm J 0   0 m  exp   0 m  ,
m 1  R  R
where 0m is the mth zero of the function J0() – see the upper row in Table 2.1 of the lecture notes. This
solution, with arbitrary coefficients cm, satisfies the Laplace equation and all boundary conditions
besides Eq. (**); to satisfy it, we need to have

Problems with Solutions Page 65

Essential Graduate Physics EM: Classical Electrodynamics


  V
c
m 1
m J 0   0m    .
 R 2
Let us multiply both sides of the last equality by (/R)J0(0m’/R), integrate the result over the interval 0
 s  /R  1, and then use the orthonormality condition (2.141) with n = 0. For the only nonvanishing
term of the expansion, we get
1
1 V
c m J 1  0 m     J 0  0 m s  sds.
2

2 20
The integral on the right-hand side of this equation may be readily worked out using the recurrence
relation (2.143) with n = 1:
d
J 0    J 1  , so sJ 0 s 0m   1 d sJ 1  0 m s  .
d  0 m ds
~
As a result, we get cm = –V/0mJ1(0m), and with the account of the definition and the symmetry of  , the
following final solution:
 1  J   / R  exp  0 m z / R
  V sgn  z     0 0 m . (***)
 2 m 1  0 m J 1  0 m  
Though this formula looks very different from Eq. (*), it gives exactly the same results at each
point! Such duality (when it can be found) is very useful in physics, because it may enable using the
form best suitable for our needs. In our particular case, Eq. (*) may be preferable for numerical
computation at an arbitrary point, because it does not require the calculation of “all” (in practice, a
dozen or so) zero points 0m. On the other hand, Eq. (***) reveals the asymptotic behaviors of the
potential much better. In particular, it shows that at large distances from the cut, the difference between
the potential on the pipe’s symmetry axis and on its walls scales as exp{–01 z /R}  exp{–2.405 z /R}.
That formula also allows the calculation of the forces F and Fz in the form of series rather than the
integrals obtained in Approach 1.63 However, these formulas also give the same final results as our first
approach (see the plots above) and of course, lead to the same conclusions concerning the charged
particle focusing.

Problem 2.27. Use the variable separation method to calculate the potential distribution inside
and outside of a thin spherical shell of radius R, with a fixed potential distribution on it: (R,,) = V0
sin cos.
Solution: According to Eq. (2.182) of the lecture notes, the surface potential is proportional to the
product P11(cos)F1(), i.e. to a single spherical function with l = 1 and n = 1. Per Eq. (2.184), inside the
sphere, the particular solution of the Laplace equation, proportional to this function, can use only radial
functions alrl, which do not diverge at r  0, while for the outer problem, we may only use functions
bl/rl+1, which do not diverge at r  . As a result, the general axially symmetric solution (2.172) of the
Laplace equation in the spherical coordinates is reduced, in our case, to

63 This simple calculation is a simple but useful additional exercise, highly recommended to the reader.

Problems with Solutions Page 66

Essential Graduate Physics EM: Classical Electrodynamics

a1 r , for r  R,

 (r , ,  )  sin  cos    b1
 r 2 , for R  r.
Finding the constants a1 and b1 from the boundary conditions on the shell (r = R), we get
(r / R), for r  R,
 (r , ,  )  V0 sin  cos   
( R / r ) , for R  r.
2

Note that the first of these results may be very simply expressed in Cartesian coordinates,
V0
 x, for r  R,
R
so the electric field inside the sphere is uniform: E = –(V0/R)nx.

Problem 2.28. A thin spherical shell carries an electric charge with areal density  = 0cos.
Calculate the spatial distributions of the electrostatic potential and the electric field, both inside and
outside the shell.
Solution: Taking into account the axial symmetry of the problem, the potential distributions both
inside (in) and outside (out) of the shell may be represented as the series given by Eq. (2.172) of the
lecture notes, with only the terms that do not diverge in the corresponding space region:
 
bl
in   al r l P l (cos  ),  out  a 0'   P l (cos  ).
l 0 l 0 r l 1
We may select the arbitrary constant a0’ to equal zero (conveniently corresponding to   0 at r  ),
while the coefficients a0 and b0 equal zero because the corresponding terms would describe the Coulomb
field of shell’s net charge Q = d2r (which our particular charge distribution does not have), so
 
bl
in   al r l P l (cos  ),  out   P l (cos  ).
l 1 l 1 r l 1
The internal and external fields are related by two boundary conditions. The first of them is
potential’s continuity (necessary to avoid an infinite electric field), giving us a set of homogenous linear
relations
b
al R l  ll1 , for l  1, 2, ... (*)
R
The second boundary condition may be obtained by applying the Gauss law to a small, flat pillbox
drawn around a small part of the shell:
  out in   
 r  r     0 cos  .
  rR 0 0
For all Legendre polynomials but the first one (with l = 1), this condition also gives homogeneous linear
relations between al and bl. Such a system of two homogeneous equations is satisfied by the trivial
solution al = bl = 0, and since the solution of this electrostatic problem is unique, this is the genuine

Problems with Solutions Page 67

Essential Graduate Physics EM: Classical Electrodynamics

solution. The only exception is the relation for l = 1, with Pl(cos )  cos , which does have a finite
right-hand side:
b 
 2 13  a1   0 .
R 0
Solving this equation together with Eq. (*) for l = 1, namely a1R = b1/R2, we get
0  R3
a1  , b1  0
3 0 3 0
so, finally,
0  R
in  r cos  , out  0 2 cos  .
3 0 3 0 r
The first result describes a uniform electric field inside the shell:
0
E in   nz ,
3 0
which the second one, a dipole field, corresponding to the dipole moment p = (4/3)0Rnz, outside it:
 0  2 cos  sin  
E out  nr  n 3  .
3 0  r 3
r 
These results are particular manifestations of the general relations to be discussed in Sec. 3.1 of
the lecture notes: of Eqs. (3.24) and (3.13), respectively.

Problem 2.29. Use the variable separation method to solve the z

problem already considered in Sec. 2.10 of the lecture notes: calculate the   V / 2
potential distribution both inside and outside of a thin spherical shell of R
radius R, separated with a very thin cut along the central plane z = 0 into
two halves, with voltage V applied between them – see Fig. 2.32, partly t  R
reproduced on the right. Analyze the solution; in particular, compare the
field at the z-axis, for z > R, with Eq. (2.218).
Hint: You may like to use the following integral of a Legendre   V / 2
polynomial with an odd index l = 2n – 1 = 1, 3, 5, ….: 64

1
I n   P 2 n 1   d 
1 1  3  5 3 
             n    1
n 1 2n  3!! .
0
n!  2   2   2   2  2n 2n  2!!
Solution: Due to the axial symmetry of the problem, we may look for the solution of the Laplace
equation in the form given by Eq. (2.172) of the lecture notes. Moreover, due to the symmetry of the
applied potentials, in that series, we may take a0 = b0 = 0, so the term with l = 0 may be dropped, giving

 bl 
 r ,     al r l  P l cos   .
l 1  r l 1 

64As a reminder, the double factorial (also called “semifactorial”) operator (!!) is similar to the usual factorial
operator (!), but with the product limited to numbers of the same parity as its argument – in our particular case, of
odd numbers in the numerator and even numbers in the denominator.

Problems with Solutions Page 68

Essential Graduate Physics EM: Classical Electrodynamics

For the internal problem (potential at r  R) we have to set all coefficients bl to zero because,
otherwise, the corresponding terms would diverge at r  0. For the coefficients an, the boundary
conditions at r = R yield the following system of equations:

V  
 a R P cos    R,   2 sgn 2   ,
l 1
l
l
l for r  R . (*)

On the contrary, for the external problem (r  R), these are the coefficients an that have to vanish
to avoid the divergence at r  , while the coefficients bn have to be found from a system of equations
similar to Eq. (*):

bl V  
 l 1
P l cos     R,    sgn    , for r  R . (**)
l 1 R 2 2 
Due to this similarity, both systems may be solved similarly: by multiplying both sides by the
product sin Pl’(cos) with an arbitrary index l’ > 0, integrating the results over the segment 0    ,
and then using the orthonormality condition (2.171). The results are, naturally, also similar:
1
b V 2l  1
sgn  P l   d ,
2 2 1
al R l r  R  ll1 r  R 
R
where   cos. Since, according to Eq. (2.169) (see also Fig. 2.23) of the lecture notes, the Legendre
polynomials with even indices are symmetric functions of their argument, while those with odd index
are antisymmetric, the last integral vanishes for even l, while for odd l = 2n – 1 (where n = 1, 2, …), it
may be rewritten as
1
b2 n 1  1  1
r  R  2 n r  R  V  2n    P 2 n 1  d  V  2n   I n ,
2 n 1
a 2 n 1 R
R  20  2
where In are the numbers specified in the Hint. Since these numbers drop very rapidly with the growth of
n and also alternate their signs, the series giving the potential distribution inside the shell,
  2 n 1
 1  r 
 r ,    a 2 n 1 r 2 n 1
P 2 n 1 cos    V   2n    I n P 2 n 1 cos  , for r  R , (***)
n 1 n 1  2  R 
and outside it,
  2n
b  1  R 
 r ,     22nn1 P l cos    V   2n    I nP 2 n 1 cos  , for r  R , (****)
n 1 r n 1  2  r 
converge reasonably fast even for the points rather close
to the shell (r  R).
For example, the figure on the right shows the 0.4  0
resulting potential distribution along the radius r, plotted 
for two values of the polar angle . For  = 0 (i.e. along 0.3 4

the symmetry axis z), and r  R, the result exactly
coincides with that given by Eq. (2.218) of the lecture V 0.2
notes, obtained by the Green’s function method.
0.1
Besides giving the result for any point of the
system, Eqs. (***) and (****) are very convenient for
0
0 1 2 3
r/R

Problems with Solutions Page 69

Essential Graduate Physics EM: Classical Electrodynamics

obtaining analytical asymptotic expressions for the important cases r << R (the shell center’s vicinity)
and r >> R (points far from the pipe). Indeed, in both cases, good approximations are given by the first
terms of these series (with n = 1), so since I1 = ½, we get

3  r / R , for r  R,
 r ,   V cos   
4 R / r 2 , for r  R.
The first of these expressions yields the electric field near the shell’s center,
3V
E r 0   r R  nz ,
4R
while the second one generalizes Eq. (2.219) of the lecture notes for arbitrary values of the polar angle.

z
Problem 2.30. Calculate, up to terms O(1/r2), the long-range electric field  V
induced by a split and voltage-biased conducting sphere – similar to that discussed d
in Sec. 2.10 of the lecture notes (see Fig. 2.32) and in the previous problem, but
with the cut’s plane at an arbitrary distance d < R from the center – see the figure on R
0
the right.
Solution: Applying the variable separation method just as was done in the
model solution of the previous problem, for the field outside of the sphere65 we may
 0
write a similar expression:

b
 r ,     l l 1 P l cos  , for r  R , (*)
l 0 r

and calculate the expansion coefficients bl in a similar way, from the following evident generalization
of Eq. (**) of that solution:

bl 0, for - 1  cos  d / R,
R P l cos     R,   V  
l 0
l 1
1, for d / R  cos  1.
As in that problem, multiplying both parts of this equation by Pl’(cos) and integrating the result over
the surface of the sphere, we may use the orthonormality condition (2.171) to get
1 1
bl 2
     P l   d .
R l 1 2l  1 1
 P l   R , cos 1
 d   V (**)
d/R

(Here   cos, so d/R = cos0, where 0 is the polar angle of the sphere surface’s cut line.)
By using Rodrigues’ formula (2.169), the last integral may be expressed via the Legendre
polynomials of d/R. However, we are only asked about the field at large distances, r >> R. This field,
with the requested accuracy O(1/r2), is given by the first two terms of the expansion (*):
b0 b b b
 r ,   P 0 cos    12 P1 cos    0  12 cos  , for r  R , (***)
r r r r

65For the field in this region, it obviously does not matter whether the conducting sphere is solid, or it is a
spherical shell.

Problems with Solutions Page 70

Essential Graduate Physics EM: Classical Electrodynamics

so we only need to calculate two coefficients, b0 and b1. For them, Eq. (**) yields simple results:
1 1
1 1
b0  RV  d  V R  d ,
2 2
3 3
b1  R 2V  d  V R 2  d 2 ,
2 4
 
d/R d/R

so Eq. (***) gives

1 R  d 3 R2  d 2
 r ,   V  V cos  , for r  R .
2 r 4 r2
At d = 0 (i.e. at the cut’s plane passing through the sphere’s center), the first term becomes
(1/2)VR/r and is appropriately twice less than that of an uncut sphere kept at potential V, while the
second term coincides with Eq. (2.219) of the lecture notes, and with the solution of the previous
problem. On the other hand, at d  R (i.e. in the limit on an uncut, grounded sphere), both terms vanish
– as they should.
Note also that all the above calculations, and hence their results, are valid for any sign of d, with
the only limitation –R  d  +R.

Problem 2.31. Calculate the field distribution in the simple electrostatic lens that was the subject
of Problem 1.9, provided that the separation of the two field regions is provided by a thin conducting
membrane, with a round hole of radius R.
Hint: You may like to use the fact that the general axially symmetric solution of the Laplace
equation in the oblate ellipsoidal coordinates (see Eqs. (2.59)-(2.60) of the lecture notes) may be
represented in the following variable-separation form:

    p nP n i sinh    q nQ n i sinh  P n cos   ,
n 0

where pn and qn are constants, Pn are the Legendre polynomials (2.169), which are sometimes called the
Legendre functions of the first kind, while Qn are the Legendre functions of the second kind (briefly
mentioned, in a different context, in Sec. 2.8) that may be defined by the following recurrence relations:
1 1   2n  1 n 1
Q 0    ln , Q 1    Q 0    1, Q n  2    Q n 1    Q n  2  .
2 1 2 n n
Solution: We may expect that far from the hole its effects are negligible, and hence the
electrostatic potential tends to that of the corresponding external field, i.e. to –Ez, where the z-axis
coincides with the symmetry axis of the system, with z = 0 at the middle of z
the hole – see the figure on the right. Hence if we take the conducting
membrane’s potential for zero, and represent the full solution in the form E
~
  E  z    r , for z  0, R
 r    ~ 0
  E  z    r , for z  0,
~ E
then the hole-induced perturbations   have to satisfy the Laplace equation
(each one, in the corresponding semi-space) and the following boundary
conditions:

Problems with Solutions Page 71

Essential Graduate Physics EM: Classical Electrodynamics

~ ~
     

~
  
~
 z 0  0,  
 z  z  z 0  E   E  , ~ r R  0.

(*)
rR   rR
(The second of these equalities is the requirement for the z-component of the full electric field to be
continuous inside the hole, where there are no electric charges; the continuity of the in-plane component
of the field is already ensured by the first of these conditions.)
Since at z  0 and r < R, the coordinate  of the oblate ellipsoidal coordinates (2.59) tends to
zero as well, we may approximate z as R cos, so (/z)z = 0 = (1/Rcos)(/) = 0. As a result, the
boundary conditions (*) may be recast as
~ ~
     

~ ~

      0  0, 
    
  E   E  R cos  ,
~
 
      0. (**)
  0
(The change of the sign at the second of the derivatives and also the modulus sign are due to the fact that
~
at the same point of the plane  = 0, the factor cos has opposite signs for the functions   because they
are defined for opposite signs of z  cos.) We see that these conditions, which should be satisfied for
all values of , include this variable only via one function, cos. But according to Eqs. (2.169)-(2.170),
this is just a specific Legendre polynomial P1(cos), which is orthogonal to all other Pn(cos). Hence, in
the series mentioned in the Hint, all coefficients pn and qn with n  1 should equal zero, so
~
    p1  P1 i sinh    q1  Q 1 i sinh  P1 cos  
  i sinh  1  i sinh  
  p1  i sinh   q1   ln  1 cos 
  2 1  i sinh  
   1   
  p1  i sinh   q1  sinh   tan 1    1  cos  .
   sinh  2   
(The last step has used the fact that the modulus of the complex function under the logarithm equals 1,
so the logarithm is equal to i multiplied by the function’s phase.)
At   , each term of the sum in the last expressions diverges as sinh, i.e. does not satisfy the
last of the boundary conditions (**); however, if we take (p1)/(q1) = i/2, these divergences cancel and
the resulting solutions,
~   1    1 
   q1   sinh   sinh   tan 1    1 cos   q1   sinh  tan 1  1 cos  ,
2  sinh  2    sinh  
tend to zero as (sinh)-2 at   . With this, the two first boundary conditions (**) give a simple
system of two linear equations for the two coefficients (q1). Solving them, we get the following final
solution:66
R  1   E  z , for z  0,
  E   E   sinh  tan 1  1 cos   
  sinh    E  z , for z  0.

66It is very instructive to compare this compact solution with the much more cumbersome derivation of the same
result, using the cylindrical-coordinate expansion, described in Sec. 3.13 of Jackson’s Classical Electrodynamics.

Problems with Solutions Page 72

Essential Graduate Physics EM: Classical Electrodynamics

The two panels of the figure below show this distribution of the electrostatic potential and the
corresponding distribution of the electric field along the symmetry axis of the lens (where sinh = z/R,
 cos  = 1), for a particular ratio of the external field strengths. Naturally, the plots describe a smooth
crossover, on a length of the order of R, between the two values of the external field (shown by the
dotted lines). In particular, at the center of the system ( =  = 0),
E  E R
E ,   E   E   .
2 
0.5 1.5
 E z E  / E   0 .3 E  / E   0 .3

E
0 1

 E
E / R E
 0.5 0.5

 E z E
1 0
1 0 1 1 0 1
z/R z/R

Note, however, that the field outside of the hole, i.e. at  > 0, and z = 0 (i.e.  = /2, cos = 0),
experiences a jump, due to the membrane’s surface charge induced by the field. At  >> 1, i.e. far from
the hole, this jump tends to E+ – E–.

Problem 2.32. A small conductor (in this context, usually called  V

a single-electron island) is placed between two conducting electrodes,
C
with voltage V applied between them. The gap between the island and   i
one of the electrodes is so narrow that electrons may tunnel quantum- " island"
mechanically through this “junction” – see the figure on the right. Q  -ne
Neglecting thermal excitation effects, calculate the equilibrium charge n tunnel
C0 junction
of the island as a function of V.
Hint: To solve this problem, you do not need to know much  0
about the quantum-mechanical tunneling between conductors, besides that such tunneling of an electron,
together with energy relaxation of the resulting excitations, may be considered a single inelastic (energy-
dissipating) event.67 At negligible thermal excitations, such an event takes place only if it decreases the
total potential energy of the system.

67 Strictly speaking, this statement, implying negligible quantum-mechanical coherence of the tunneling events, is
correct only if the junction transparency is so low that its effective electric resistance R is much higher than the
fundamental quantum unit of resistance, RQ  /2e2  6.5 k – see, e.g., QM Sec. 3.2. However, this condition
is satisfied in most experimental tunnel junctions.

Problems with Solutions Page 73

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Let us assume that the electric field induced by the island’s charge Q does not extend
beyond the space between the two electrodes. Then, regardless of the island’s geometry, according to the
linear superposition principle, the electrostatic potential is of the island has to be a linear function of Q
and the applied voltage V. For the convenience of what follows, we may write this linear relation in the
form
Q C
Q  C 0is  C is  V , i.e.  is    ext , with  ext  V and C   C  C 0 , (*)
C C
where the coefficients C0 and C have the physical sense of the mutual capacitances between the island
and the system’s electrodes (see the figure above), so C of the full self-capacitance of the island at V =
0. (Note that Eq. (*) is valid even if the electric fields associated with the capacitances C0 and C are not
spatially separated in full, and hence the lumped capacitor model cannot be used from the very
beginning – cf. Fig. 2.5 of the lecture notes.)
Since the first term in the potential is given by Eq. (*) is proportional to Q, the corresponding
electrostatic energy has to be calculated by using Eq. (1.61) of the lecture notes. On the other hand, the
energy associated with the externally-induced potential ext, which does not depend on Q, is described
by Eq. (1.54). As a result, the total electrostatic energy of the system (or, more exactly, its part
dependent on Q) is
Q2
U Q    Q ext . (**)
2C 
In order to use this result, it is convenient to introduce the notion of “external charge”,
Qext   ext C   CV ,
because in this notation, Eq. (**) takes a very simple form:
Q 2 QQext Q  Qext 
2

U Q      const . (**)
2C  C 2C 
The plot of this expression as a function of Q is a quadratic parabola with the minimum at Q = –
Qext (see the left panel of the figure below) and if Q could take any real value, the system’s static
equilibrium would follow this minimum. However, in reality, Q is a multiple of the fundamental charge
e 1.610-19 C: Q = –ne, where n is the number of electrons acquired by the island since it had been
electrically neutral.68 As a result, the system’s equilibrium corresponds to the value Q = –en closest to –
Qext, and the change of n by 1 (via tunneling of a single electron into or from the island) takes place
when Qext crosses any of the critical points e(n + ½) – see the right panel in the figure below.
The simplest interpretation of this Coulomb staircase pattern is that at small C, the n extra
electrons in the island, whose negative charge is not compensated by the positive charge of the atomic
lattice, repel each other strongly, so the island has only as many of them as necessary to compensate the
externally-induced polarization charge Qext.

68 It is crucial that this discreteness does not pertain to Qext, which is just a normalized external voltage and hence
is not quantized – at least on this scale. In other words, Qext is just the charge of polarization of the island by
voltage V, which may take any real values – see the discussion in Sec. 2.9 of the lecture notes.

Problems with Solutions Page 74

Essential Graduate Physics EM: Classical Electrodynamics

Q e 3e
2 2
U Q 
0 Qext  CV
Q  Qext
e

 en  1  en Q  2e

The most counter-intuitive aspect of this effect is that it may take place in rather macroscopic
conducting islands – with a size up to a few microns, with zillions (e.g., ~1010) of background electrons.
However, for that, the temperature has to be low enough, to make the energy scale of the masking
thermal fluctuations, kBT, much smaller than the relevant scale of U, namely e2/2C. As a result, the
Coulomb staircase (or rather its footprint on the rf impedance of the system) was observed
experimentally and explained theoretically only in the 1970s.69 Presently, this system, called the single-
electron box, is considered a generic device, from which quite a few other single-electron devices may
be derived – see, e.g., the next problem.

 V
Problem 2.33. The system discussed in the previous problem is now
generalized as the figure on the right shows. If the voltage V’ applied between C
the two bottom electrodes is sufficiently large, electrons can successively " island"
tunnel through two junctions of this system (called the single-electron Q  Q1  Q2
transistor), carrying some dc current between these electrodes. Neglecting
Q1 Q2
thermal excitations, calculate the region of voltages V and V’ where such a C1 C2
current is fully suppressed (Coulomb-blocked).
  V'  0
Solution: Repeating the argumentation of the previous problem, for the
island’s electrostatic potential we may write a similar expression:
Q C C
is    ext   ext' , with  ext  V ,  ext'  1 V' , and C   C1  C 2  C , (*)
C C C
where the constants C1, C2, and C have the sense of mutual capacitances between the island and the
corresponding electrodes of the system – see the figure above.
Now, writing the expression for the system’s potential energy, we may follow the argumentation
of the model solution of the previous problem, but should take into account that Eq. (1.54) of the lecture
notes, used in that solution, is valid only if the charge qk has been moved into the system under analysis
from a region with negligible potential. If that initial potential k-ini is different from zero, then we
should generalize that relation as
U ext   q k  ext rk    k ini  .
k

69 J. Lambe and R. Jaklevic, Phys. Rev. Lett. 22, 1371 (1976).

Problems with Solutions Page 75

Essential Graduate Physics EM: Classical Electrodynamics

Indeed, the expression in the square brackets is just the work of external forces, and hence the potential
energy’s increase, at a transfer of a unit charge between the points with potentials (k)ini and ext(rk). In
our current problem, ext is actually the sum (ext + ext’) defined by Eq. (*),(k)ini = V’ for the charge Q1
brought into the island by the electrons tunneling through the left junction, and (k)ini = 0 for the charge
Q2 brought in through the right junction. (The total charge of the island is evidently Q = Q1 + Q2, see the
figure above.) As a result, for the total electrostatic energy of the system, we get
Q2
U Q1 , Q2    Q1  ext   ext'  V'   Q2  ext   ext'  ,
2C 
Now by using Eqs. (*) for ext and ext’, we may rewrite this expression in the form
Q1  Q2  Qext 2 C C C 
U Q1 , Q2     1 Q2  2 Q1  V'  const, where Qext  CV , (**)
2C   C C 
which is the proper generalization of Eq. (**) in the model solution of the previous problem.
In order to use this expression for finding the dc current threshold, we should take into account
the discreteness of the electron charges passed through the junctions,
Q1   n1e, Q2   n 2 e ,
and compare U(–en1, –en2) with all four values of energy, U(–en1  e, –en2) and U(–en1, –en2  e),
resulting from tunneling of an additional electron through either junction, in either direction. If any of
these events leads to an increase of U, the state is stable, and dc current cannot flow through the system.
An elementary calculation yields four stability conditions, which may be summarized as follows:
 1  Qext  C1V'   1
n  e     n  e,
 2 Qext  C 2  C V'   2

where n  n1 + n2, i.e. of the total number of charge-uncompensated electrons in the island. This result
may be understood easier by plotting the boundaries for these conditions on the plane of applied
voltages V and V’ – see the figure below. Only if the point {V, V’} is inside one of the so-called
Coulomb diamonds of this periodic diagram (which continues infinitely to both sides of the horizontal
axis),70 with the indicated value of n, the charge state is stable, with no persistent sequential tunneling of
electrons. This stable state is said to be due to the Coulomb blockade of current. Note that the largest
magnitude of the blockade threshold is e/C, clearly indicating its single-electron nature.

V'
3e e / C e 3e
  
2 2 2
n  1 n0 n  1
Qext  CV
e

2  e / C

70 In the particular case V’ = 0, this pattern is reduced to the Coulomb staircase analyzed in the previous problem.

Problems with Solutions Page 76

Essential Graduate Physics EM: Classical Electrodynamics

This result means that the gate voltage V may control the average (“dc”) current in this system,
so it may be indeed used as a transistor, based just on the Coulomb repulsion of electrons rather than on
any specific material properties. Invented in 1985, and experimentally demonstrated for the first time in
1987, single-electron transistors are used in experimental physics, mostly for sensitive electric
measurements at low temperatures. The transistor may be further modified to obtain other Coulomb-
blockade devices, in particular, fundamental dc current standards and nonvolatile memory cells with
digital bits stored in the form of single electrons. Unfortunately, the tough temperature-to-size trade-offs,
and uncontrollable blockade threshold offsets by randomly located charged impurities have so far
prevented the practical use of these devices in digital electronics.71
Note that similar devices with superconducting electrodes enable the controllable transfer of not
only discrete single electrons, but also discrete single Cooper pairs between two or more
superconducting Bose-Einstein condensates, and also an electrostatic control of supercurrents. However,
due to the quantum coherence of the condensates, any quantitative discussion of these devices requires
quantum mechanics.72

Problem 2.34. Use the charge image method to calculate the full surface charges induced in the
plates of a very broad, externally-unbiased plane capacitor of thickness D by a point charge q separated
from one of the electrodes by distance d. Suggest at least one alternative method to obtain the same
result.
Solution: The system of image charges that satisfies the corresponding Poisson equation and the
boundary conditions ((0) = (D) = 0) for this problem has already been discussed in Sec. 2.6 of the
lecture notes – see Fig. 2.28c, reproduced below in an extended version. Here the red balls denote point
charges +q (including the real one), while the blue ones, charges –q.

D
q
z
d d d d

 2D 0 2D 4D
As a result, the total charge induced in any electrode surface (say, the one located at z = 0) may
be calculated as was discussed in Sec. 2.9:

   q
Q  d r   0 z 0 z z 0 d r    0 z 0 z j  j z 0 d r    0   2d , (*)
2 2 2

z 0 0 
z j ,  4 0  2  z 2j 
1 / 2 z 0

where  is the distance from the z-axis, and zj is the position of an image belonging to the jth pair of
adjacent charges, centered at 2jD. Per Eq. (2.194) (but also as evident from the figure above),
z j  2 jD  d .

71 For more discussion of these issues, see the literature cited in Sec. 2.9 of the lecture notes.
72 See, e.g., QM Sec. 2.8.

Problems with Solutions Page 77

Essential Graduate Physics EM: Classical Electrodynamics

The differentiation and integration in Eq. (*) may be swapped with the summation, making two first of
these operations easy; however, the final analytical summation is not too pleasing.
A simpler way to calculate Q is to apply the Gauss law to the system of charges (including the
actual charge q) shown in the figure above. Let us first consider the set of only the “positive”73 charges
(+q) inside a round cylinder with its symmetry axis coinciding with the z-axis, a very large radius R >>
D, and the following special choice of position of the lids (both parallel to capacitor planes): one just to
the right of the surface of our interest, i.e. at z = +0, while the other lid located exactly in the middle
between the actual charge and the first “positive” charge image on the right of it, i.e. at
d  2 D  d 
z  L   Dd.
2
Due to the condition R >> D, the electric field E+ of the “positive” charges at the lateral (curved) wall of
the Gaussian cylinder is virtually uniform, normal to the z-axis, and is the same as of a continuous
charge line with uniform linear density + = q/2D. Hence, according to the (easy) solution of Problem
1.1, E+ = +/20, so its flux through the wall is
  L q
 E 
  R  D
 n d 2r 
0

2 D 0
D  d  .

Generally, the electric flux through the cylinder’s lids should be calculated more carefully
because some parts of them are at distances of the order of d ~ D from the nearest charges. However,
with our choice of lid positions, the flux through the lid located at z = L+ equals zero due to the
symmetry, while, per Eq. (2.3), the flux through the lid with z = +0 is proportional to the surface charge
Q+ of the electrode, induced by the “positive” charges:
Q
  E  n d r   ,
2

z 0
0
where the minus sign is due to the fact that we are calculating the flux coming out of the considered
cylinder. Now taking into account that inside this cylinder we have just one “positive” charge q, the
Gauss law for it reads
q Q q
 E n d r  2D 0 D  d    0   0 ,
2

giving
q d
Q    1   .
2 D
Now we may repeat this calculation for the system of “negative” charges (-q), with the
replacement of L+ with the exact middle between two such charges, for example
 d  2 D  d 
L   Dd ,
2
the replacement of + with – = –q/2D, and taking into account that there are no “negative” charges
inside this new cylinder, with +0  z  L–. The result is similar:

73 I am taking this word in quotes because our calculation is correct for any sign of the actual point charge q.

Problems with Solutions Page 78

Essential Graduate Physics EM: Classical Electrodynamics

q d
Q    1   ,
2 D
so the total surface charge Q = Q+ + Q– of the surface is74
 d
Q   q 1   . (**)
 D
This result may be obtained much more easily by using the reciprocity theorem whose proof was
the task of Problem 1.18. Indeed, let us take our problem’s situation (with grounded
capacitor electrodes, the point charge q between them, and the surface charge on the D
left electrode equal to Q) for the charge and potential distributions number 1, and
those in the same capacitor and biased with voltage V as shown in the figure on the q
right but without the charge q, for distributions number 2. Then 1(r) = 0 at z = 0 and
d F
z = D, and 2(r) = 0 at 0 < z < D, while, per Eq. (2.39), 2(r) = V(1 – z/D), where z is
the distance of the point r from the left electrode. So in the reciprocity theorem, we Q
have to take QV
 d
 1 r  2 r d r  QV  qV 1  D ,   2 r 1 r d r  0,
3 3

and the stated equality of these integrals immediately yields Eq. (**).  V   0

Finally, one more way to obtain the same result is by using the following simple reasoning. Let
us assume that our plane capacitor, with the point charge inside, is connected to a battery fixing some
voltage V independent of the charge position – see the figure above. Then the uniform external electric
field E = V/D creates a position-independent force of the magnitude F = qE = qV/D, which is applied to
the charge and directed normally to the electrode surfaces. On a small displacement d, this force
performs work W = Fd = qVd/D. On the other hand (“from the point of view of the battery”), the
same work W has to be equal to VQV, where QV is the charge that is transferred through the battery
as a result of this displacement. But this charge has nowhere to go rather than to change the surface
charge of the capacitor plate: QV = Q. Comparing these two expressions for W, we get Q =
q(d/D), i.e. Q = qd/D + const. Since the surface charge Q has to tend to –q when the point charge
approaches the surface (d  0) and hence becomes fully compensated by the closest surface charge, the
constant in the last expression has to equal –q, so we return to Eq. (**).
An additional task for the reader: explain why the last two derivations of this result are not
entirely independent.

Problem 2.35. Use the charge image method to calculate the potential energy of the electrostatic
interaction between a point charge placed in the center of a spherical cavity that had been carved inside a
grounded conductor, and the cavity’s walls. Looking at the result, could it be obtained in a simpler way
(or ways)?

74 With this result for Q on hand, the surface charge Q’ of the counter-electrode (located at z = D) may be most
simply calculated by applying the Gauss law to a similar cylinder, but with its lids inside the opposite capacitor’s
plates, so the electric field on them vanishes. This yields Q + Q’ + q = 0, finally giving Q’ = –qd/D.

Problems with Solutions Page 79

Essential Graduate Physics EM: Classical Electrodynamics

Solution: First, let us calculate the force F exerted by the

conductor on a point charge q’ shifted by some distance d’ < R from
the cavity’s center – see the figure on the right. As it follows from
the discussion of an example in Sec. 2.9 of the lecture notes (see Fig. q' q
2.29a and its discussion), both the Poisson equation inside the cavity
0 d' R d
and the boundary conditions at the conductor’s surface may be
satisfied by adding, to the unperturbed Coulomb field of the real
charge, that of an image charge q placed at distance d > R from the
center, with q and d related to q’ and d’ by Eqs. (2.198). Solving
these equations for q and d (now functions of the given q’ and d’), we get
R2 d R
d , q   q'  q' .
d' R d'
From here, the magnitude of the force F (directed toward the image charge, i.e. the point of the
conducting surface, which is nearest to the real charge) is
qq' 1 q' 2 R 1 q' 2 Rd'
F (d' )    .
4 0 d  d'  2
4 0 d' R 2 / d'  d' 2
4 0 R 2  d' 2 2

As two sanity checks, the force vanishes at d’  0:

q' 2 d' d'
F (d' )   0, at  0,
4 0 R 3 R
in accord with the problem’s symmetry, while at d’  R, it tends to the expression
q' 2 1
Fplane R     , where   R  d' ,
4 0 2 2
which we would get for the charge at the distance  from a conducting plane.
Now the required interaction energy U may be defined as the difference between the energy of
the (charge + sphere) system and the sum of energies of these two objects when they are far from each
other. Hence, U may be calculated as the (minus) integral of the interaction force on a charge’s path
from infinity to the cavity center. Our results for the force F are only valid inside the sphere, but we may
circumvent this problem by extending the integration from the center to some distance  << R from the
surface, and subtracting from the result the potential energy (2.191) provided by at that point by a
conducting plane (reader, explain why):
R 
q' 2  R  R 2 x 1  q' 2
U (0)    F ( x)dx  U plane R     
0
4 0
 
 0 R 2  x 
2
dx  
4 
 
8 0 R
. (*)

According to Eq. (2.192), the magnitude of this energy is twice larger than what a conducting
plane would give at the distance R from the charge. The reason for such difference is clear: for a
spherical cavity, all induced surface charges are at the same distance (R) from the initial charge, while
only the closest charges on the conducting plane are at such distance, so the effect of their attraction is
smaller.

Problems with Solutions Page 80

Essential Graduate Physics EM: Classical Electrodynamics

A much simpler way to get Eq. (*) is to use Eq. (1.65) for the electric field energy, limiting the
integral to the cavity interior, r < R (since there is no field in the conductor). The divergence of the
integral at r  0 disappears if U is understood as above, i.e. as the result of the difference between the
charge’s field in the cavity and that in the free space:

 0   0  
U  Ein cavity d r 
2 3
E 2
d 3 
 in free space   2
r   Ein2 cavity  Ein2 free space  d 3 r   Ein2 free space d 3 r  .
2  cavity  
all space   rR rR 
According to the Gauss law, if a charge is in the center, its field at r < R is not affected by the conductor,
so the first integral in the last expression vanishes, and we get
2
0 0  q'  q' 2
U  E d r      4r dr  
2 3 2
,
2 R  4 0 r 2  8 0 R
in free space
2 rR

i.e. the same result as given by Eq. (*).

One more way to obtain the same result is to first calculate the small energy increment U due to
a small addition q << q to the charge, brought from infinity to the initial charge’s location, i.e. the
center of the sphere, by replacing the conducting sphere with a thin spherical shell with a uniformly
distributed charge q. (Physically, such replacement of the conductor by the charge of its external surface
is equivalent to the subtraction made at the previous approach.) Since, the due to the spherical
symmetry, the shell’s charge does not create an electric field inside it,
   
q qq  q2 
U    Fshell dr  q  Eshell dr  q  Eshell dr  q  dr        .
0 0 R R 4 0 r 2 4 0 R  8 0 R 
Integrating the result from q = 0 to q = q’, we again arrive at Eq. (*).
z
r1
Problem 2.36. Use the method of charge images to find
the Green’s function of the system shown in the figure on the
right, where the bulge on the (otherwise, plane) surface of a r2
conductor has the shape of a semi-sphere of radius R.
Solution: Let a (real) charge q1 be at point r1 in the free-
0 r4 R 
space part of the system. Then, according to Eq. (2.198) of the
 0
lecture notes, all the boundary conditions may be satisfied using
r3
the three charge images shown in the figure on the right, with
2 2
R R R R
q 2   q1 , r2    r1 ; q3  q1 , ρ 3  ρ1 , z 3   z1 ; q 4  q3 , r4    r3 ;
r1  r1  r3  r3 
where j are the horizontal components of the radius vectors rj with the origin in the sphere’s center. As
a result, the Green’s function may be represented simply as
4 q j / q1
G (r, r1 )   .
j 1 r  rj

Problems with Solutions Page 81

Essential Graduate Physics EM: Classical Electrodynamics

However, when spelled out in the most suitable, cylindrical coordinates where r1 = (12 + z12)1/2,
this compact expression becomes somewhat bulky, because each denominator looks like


r  r j   2   2j  2  j cos   j   z  z j  
2 1/ 2
.

Problem 2.37.* Use the spherical inversion expressed by Eq. (2.198) of the lecture notes to
develop an iterative method for a more and more precise calculation of the mutual capacitance between
two similar conducting spheres of radius R, with their centers separated by distance d > 2R.
Solution: There are several ways to develop the requested iterative process; the following one
uses perhaps the most natural logic. At each step of the process, let us keep the total charge of the right
sphere equal to +Q, and that of the left sphere equal to –Q, and keep the surface of each sphere
equipotential (at potentials L and R, correspondingly), while calculating more and more exact values of
these potentials, and hence that of the mutual capacitance
Q Q
C  , (*)
V R  L
by replacing the genuine distribution of the charge on the surface of each sphere with more and more
extended systems of point image charges. At the 0th iteration, the distributed charges of sphere surfaces
are replaced with single point charges qR(0) = +Q and qL(0) = –Q, located at their centers. (This
approximation correctly describes the field distribution at very large distances, i.e. it is asymptotically
correct at d/R  .) In this approximation, the potentials of the sphere surfaces are
q R(0) Q q L(0) Q
 R(0)   ,  L(0)   ,
4 0 R 4 0 R 4 0 R 4 0 R
so their mutual capacitance,
Q
C (0)   2 0 R,
 ( 0)
R   L(0)
does not depend on the distance d.
In order to make the 1st iteration, let us calculate the corrected value of L using the fact proved
in Sec. 2.9 of the lecture notes: if the field is induced by a point charge qR(0) located at a distance d > R
from the sphere's center, then the potential of
the sphere's surface equals zero if its genuine Q L1 R1  Q
surface charge is replaced with a point charge
q'L = –(R/d)qR(0) = –(R/d)Q, located at the q L1 q' L q' R q R1
2
distance d' = R /d from the center – see Eqs.
0 d' R d
(2.198) and the figure on the right. Now,
following our general commitment, let us d  d'
correct the central point charge qL to such a 1st iteration
value qL(1) that the sum of all charges inside the
left sphere is equal to –Q:
R  R
q L(1)  q' L  Q, giving q L1  Q  q' L  Q  Q  Q1   .
d  d
This charge still keeps the surface potential of the left sphere constant, but changes its value to

Problems with Solutions Page 82

Essential Graduate Physics EM: Classical Electrodynamics

q L(1) Q  R
 (1)
  1   .
4 0 R 4 0 R  d 
L

Generally, we would need to calculate a similar adjustment of the charge qR and the potential R, but due
to the symmetry of our problem, the results are evident:
 R Q  R
q R1  q L1  Q1  ,  R(1)   L(1)  1   .
 d 4 0 R  d 
Now we may calculate the mutual capacitance in the 1st approximation:
Q 1
C (1)   2 0 R ; (**)
 (1)
R  L
(1)
1 R / d
this expression shows that C increases as the distance d between the spheres is reduced – as it should.
For the next, 2nd iteration, we have to use the same approach to correct the values of L and R,
by taking into account the corrected values of the fields of the two central charges, qL(1) and qR(1), as well
as the existence of the image charges q'L and q'R. According to Eq. (2.198), the latter step requires the
introduction of new image charges, q”L and
q”R, located at different distances, d" = R2/(d –  Q L
( 2)
R2   Q
2 2 2 2 2
d') = R /(d – R /d)  R d/(d – R ) > d', from the
centers of the respective spheres – see the q L2  q' L2 q" L q" R q' R2 q R2 
figure on the right.
0 d' d" R d
This process may be continued
similarly, by adding, at each iteration, one more 2nd iteration
image charge inside each sphere, and then
readjusting the prior charges and the surface potential. However, even the initial iteration (**) gives
surprisingly accurate results if the ratio R/d is not too close to the convergence limit (R/d)max = ½; for
example, even for d as small as 4R (i.e. with the gap between the spheres as small as the sphere’s
diameter), its error is close to 0.5%, and only at d = 3R, it increases to slightly more than 2%.
An additional task for the reader: by using the approach described at the end of the model
solution of Problem 13 (but adjusted for the 3D geometry of our current problem), calculate an
expression for C, which would be asymptotically correct in the limit t  d – 2R << R.
And just for the reader’s reference: by using the bispherical coordinates briefly mentioned in the
solution of Problem 13, C may be calculated in the form of an explicit series:

C  2 

 sinh ln x  x 2  1   ,
1/ 2
d
sinh n ln  x  x  1 
0 1/ 2
where x  1
n 1
2 2R

Problem 2.38.* A conducting sphere of radius R1, carrying an electric R1

charge Q, is placed inside a spherical cavity of radius R2 > R1, carved inside
another bulk conductor. Calculate the force exerted on the inner sphere if its R2
center is displaced by a small distance  << R1, R2 – R1 from that of the cavity – 
see the figure on the right. Q

Problems with Solutions Page 83

Essential Graduate Physics EM: Classical Electrodynamics

Solution: As it follows from the solution of the previous problem, this problem could be solved
(for an arbitrary  < R2 – R1) by building up a sequence of point image charges at the following
distances from the cavity’s center (here the term “positive” means a charge with the same sign as Q):
positive charges negative charges
d   (charge Q), D  R22 /  ,
d'    R12 / D  d , D'  R22 / d' ,
d"    R12 / D'  d , D"  R22 / d" ,  ,
with the concurrent calculation/readjustment of their magnitudes. However, this solution would be in the
form of an algorithm rather than an explicit series. Another possible analytical approach is using the
bispherical coordinates, but for our current case of two unequal radii R1 and R2, it leads to calculations
and results much more cumbersome than the last formula cited in the previous problem’s solution.
However, for our assignment, i.e., for  << R1, R2 – R1, a simple explicit result may be obtained –
for example, in the following way. In the reference frame whose origin coincides with the center of the
cavity, and the z-axis directed along the shift , the surface of the inner sphere is described by the
following relation:
x' 2  y' 2   z'  δ   R12 ,
2

where the prime signs distinguish a point on the sphere’s surface from an arbitrary point in space. In the
spherical coordinates, we may rewrite this relation as

r' sin  cos  2  r' sin  sin  2  r' cos    2  R12 , giving r' 2 sin 2 θ  r' cos θ  δ   R12 .
2

In the limit /R1  0, the last expression reduces to the first-order approximation
r'  R1   cos  . (*)
Now let us calculate the potential distribution (r, ) between the spheres, taking into account its
initial (radially-symmetric) part plus the first nonvanishing perturbation, which is, per Eq. (*),
proportional to cos. This means that in the general axially symmetric solution (2.172) of the Laplace
equation we may drop all terms with l > 1:
b  b 
 r ,   a 0  0   a1 r  12  cos  , with a1 , b1   . (**)
r  r 
Plugging this distribution into the appropriate boundary conditions,
 r' ,    R1   cos  ,     0 ,   R2 ,    0 ,
(where 0 is the so far unknown potential of the inner sphere, and the potential of the outer conductor is
taken for zero), we get two equations,75

75 Taking into account the variation, cos; of radius r1 in the terms with a1 and b1 would give us additional terms
of the order of (cos)2 and higher, which may be ignored in our linear approximation.

Problems with Solutions Page 84

Essential Graduate Physics EM: Classical Electrodynamics

b0  b  b b   b 
a0    a1 R1  12  cos   a 0   0  0 2 cos     a1 R1  12  cos    0 ,
R1   cos   R1   R1 R1   R1 
b  b 
a 0  0   a1 R2  12  cos   0,
R2  R2 

which should be satisfied at all . This requirement gives us a system of four equations for four
unknown coefficients a and b:
b b b b b
a0  0  0 , a1 R1  12  0 2 ; a 0  0  0, a1 R2  12  0 . (***)
R1 R1 R1 R2 R2
The equations for a0 and b0, and hence their solution,
R1 R1 R2
a 0  0 , b0   0 ,
R2  R1 R2  R1
do not depend on a1 and b1. Before calculating a1 and b1, let us express the potential 0 of the inner
sphere via its charge Q, by applying the Gauss law to a spherical surface of some radius r within the
range r1 < r < R2:
 2 Q
S En d r  S r d r   0 .
2

For the potential distribution given by Eq. (**), the surface integral equals

 b0  b1   2 2  b0  b1   R2  R1
S  r 2   a1  2 r 3  cos  d r  2r 0  r 2   a1  2 r 3  cos  sin d  4b0  40 R1 R2 ,
so the relation between 0 and Q is (in the linear approximation in  only!) the same as at  = 0:76
Q R1 R2  1 1 
0  , where C  4 0  4 0   
C R2  R1  R1 R2 
– cf. Eq. (2.56) of the lecture notes. Presently, we need this result only to express the coefficient b0 via
the given charge Q:
RR Q
b0  0 1 2  .
R2  R1 4 0
Using this expression and solving the remaining two equations (***), we get
1 Q 1 R23 Q R23
a1  b0 3   , b1  b0 3  ,
R2  R2
3
4 0 R2  R13
3
R2  R23 4 0 R23  R13
so the spatial distribution of the radial electric field is

76This fact may be also proved by the following reasoning. The capacitance C between the spheres has to be a
smooth function of , which may be represented as the Taylor series C() = C(0) + (dC/d)=0 + …. However,
due to the evident mirror symmetry of the system with respect to the plane  = 0, the change of the sign of 
cannot result in a change of C, and hence (dC/d)=0 = 0.

Problems with Solutions Page 85

Essential Graduate Physics EM: Classical Electrodynamics

 b0  b  Q 1 1  R23  
Er    2    a1  2 13  cos    2  
3 
1  2 3 
  cos   .
r r  r  4 0
 r R2  R1 
3
r  
Now let us use this field distribution to calculate Er at the shifted surface of the inner sphere, i.e.
at r = r’ = R1 + cos . The small difference between r’ and R1 may be taken into account only in the
largest, -independent term of the above expression for Er, and only in the first, linear approximation:
1 1 1  2 / R1  cos  1 2
   2  3  cos  .
r' 2
R1   cos   2
R12
R1 R1
(In all other terms, the difference between r’ and R1 may be ignored, because they are already
proportional to .) As a result, we get

Q  1 2 1  R23   Q  1 3 cos  
E r r r'   2  3  cos   3 
3 
1  2   cos   
3 
 2  3 .
3 
4 0  R1 R1 R2  R1  R1   4 0  R1 R2  R1 
Per the solution of Problem 1, the normal force fn per unit area of the inner sphere may be calculated as
2 2
dFn  0 2   Q   1 3 cos  
 E r r r'  0    2  3  .
3 
dA 2 2  4 0   R1 R2  R1 
In the first, linear approximation in small cos:
2
dFn  0  Q   6 R12 cos 
   1  .
3 
dA 2  4 0 R12   R 3
2  R1 
Due to the axial symmetry of the problem, the net force may have only one (z-) component:

dF dF
F  Fz   z d 2 r   n cos  d 2 r  2R12  f n cos  sin d .
r1
dA r1
dA 0

At the integration, the net contribution of the unperturbed, radially-symmetric normal forces vanishes,
while the first-perturbation term yields a very simple final result:
2 2
0  Q  6 R12    Q  6 2 Q2 
3 
F  2R 2
  3 cos 2  sin d  2 0   3  .
 2 
2  4 0  R2  R1 3 4 0 R2  R1
1
2  4 0 R1  R2  R1 0 3 3 3

As the simplest sanity check, at R1  0, when the inner sphere turns into a point charge, the
result is reduced exactly to the corresponding limit of the byproduct of the solution of Problem 35:
Q2 
F ,
4 0 R23
obtained there by using the charge image method. Another check may be obtained in the opposite limit,
R1  R2, by calculating F using a combination of Eq. (2.28) with the approach demonstrated at the end
of the model solution of Problem 13. (This additional simple exercise is highly recommended to the
reader.)

Problems with Solutions Page 86

Essential Graduate Physics EM: Classical Electrodynamics

Problem 2.39. Within the simple models of the electric field screening in conductors, discussed
in Sec. 2.1 of the lecture notes, analyze the partial screening of the electric field of a point charge q by a
planar conducting film of constant thickness t << , where  is (depending on the charge carrier
statistics) either the Debye or the Thomas-Fermi screening length – see, respectively, Eqs. (2.8) or
(2.10). Assume that the distance d between the charge and the film is much larger than t.
Solution: Due to the condition d >> t, the gradient of the electrostatic potential inside the film is
dominated by its component along the axis (say, z) normal to the film’s plane, so we still may use the 1D
equation (2.7):
 2 1
 2,
z 2

even if  is relatively slowly (on distances of the order of d) changing in the plane of the film as well.
Integrating both sides of this equation over the film’s thickness t, we get
t / 2
    1
z

z
 2
  dz ,
t / 2
(*)

where the indices  refer to the two surfaces of the film. This equation shows that if t << , the film can
create only a small difference (of the order of t/2) of the derivatives, and hence just a negligible (of
the order of t2/2) relative variation of the potential inside it. As a result, Eq. (*) yields two boundary
conditions: z
    t q or q"
     ,   2. (**)
z z 
this point “sees”
Now, inspired by the charge-image analysis of the d both charges
basic problem shown in Fig. 2.26 of the lecture notes, we may t q and q’
try to describe the field at z > 0 as a superposition of the fields
by the real charge q and by its image q’ located at the same 0 
distance d from the film but on its opposite side – see the d
figure on the right. However, now q’ is not necessarily equal this point “sees”
to (–q); indeed, if t  0, all the film’s effects, in particular the q' charge q” alone
charge q’, should vanish.
In addition, in contrast to the conducting half-space case, we should expect to find some field
remnants on the other side of the film, at z < 0. This field cannot be contributed by the image charge q’
because it would provide the potential’s divergence at its location. Thus, in that half-space, we should
try to use the real point source only, but with a re-normalized charge q” rather than the genuine charge q
– see the figure above. As a result, we may look for the potential’s distribution in the form77
 q q'
 , for z  0,
 ( , z) 
1
 2


   z  d 
2
 1/ 2
 2
 z  d  
2 1/ 2

4 0  q''
, for z  0.

  2   z  d 2 
1/ 2

77As will be discussed in Sec. 3.4 of the lecture notes, absolutely the same solution (though with different
constants q’ and q”) describes the electric field’s penetration into a half-space filled with a linear dielectric.

Problems with Solutions Page 87

Essential Graduate Physics EM: Classical Electrodynamics

(In the “coarse-grain” picture described by this formula, with the size scale given by d >> t, the film’s
thickness is negligible.)
Plugging this solution into the boundary conditions (**), we see that they are indeed satisfied (so
this is indeed the unique solution of our boundary problem), provided that the effective charges q’ and
q’’ obey the following two relations:
q  q'  q'' , q  q'   q"  td2 q'' .

Solving this simple system of two linear equations, we get, in particular, the following expression for the
q”/q ratio – which is, obviously, a relevant measure of the field’s penetration behind the film:
q" 1
 .
q 1  td / 22
As the sanity check, q” tends to q at t  0, and to zero at t   – as it should. However, the
most interesting feature of this result is that the ratio q”/q starts to drop as soon as t is increased to ~2/d
<< , i.e. the field behind the film may be well (though not exponentially well) screened even by a film
much thinner than the screening length . Very unfortunately, the fact of such efficient screening of the
electric field by thin conducting films (and of the similarly efficient shielding of the magnetic field by
thin ferromagnetic and superconducting films78) is overlooked even in some very popular textbooks.

Problem 2.40. Prove the following expansion of the simplest Green’s function (2.204) into a
series over the Legendre polynomials:
l
1 1   r 
    P l (cos  ),
r  r' r l 0  r 

where r> is the largest of the two scalars r   r   0 and r’   r’   0, while r< is the smallest of them.
Solution: Let us direct the z-axis along the radius vector r’. Then the Green’s function (2.204),
considered as a function of r alone, has to be axially symmetric and its Legendre-polynomial expansion
is given by the general Eq. (2.172):

1  b 
   al r l  l l 1  P l (cos  ), for r'  n z r' , (*)
r  r' l 0  r 

where in our case, the polar angle  is also the angle between the radius vectors r and r’. To find the
coefficients al and bl, let us consider the particular case when the point r is positioned on the z-axis as
well: r = nzr. In this case, Eq. (*) is reduced to

1  b 
   al r l  l l 1  . (**)
r  r' l 0  r 
Let us first consider the case r < r’, and Taylor-expand the left-hand side of this relation in r at
point r = 0:

78 See, e.g., Secs. 5.6. 6.3, and 6.4 of the lecture notes.

Problems with Solutions Page 88

Essential Graduate Physics EM: Classical Electrodynamics

1 
1 l  1 
   r .
l
(***)
r'  r l 0 l! r l  r'  r  r 0
Calculating the involved derivatives one by one,
0  1  1 1
0     ,
r  r'  r  r 0 r'  r  r 0 r'
1  1  1 1
1     2,
r  r'  r  r 0 r'  r  r 0 r'
2

2  1   1 1 2 1 2
     3 ,,
r 2  r'  r  r 0 r r'  r  r 0 r'  r  r 0 r'
2 3

we see that
l  1  l!
l    l 1 ,
r  r'  r  r 0 r'
so the expansion (***) becomes
 l
1 rl 1  r
  l 1     , for r  r' .
r'  r l 0 r' r' l 0  r' 
For the opposite case r’ < r, we may perform a similar expansion in r’, getting
 l
1 r' l 1   r' 
  l 1     , for r'  r.
r  r' l 0 r' r l 0  r 
Comparing these two expressions with Eq. (**), we see that
1 / r' l 1 , for r  r' , 0, for r  r' ,
al   bl   l
0, for r'  r ,  r' , for r'  r.

This equality, together with Eq. (*), proves the expansion given in the assignment.
Just for the reader’s reference: for the general direction of the z-axis, this Green’s function may
be represented as an expansion over the spherical harmonics (mentioned in Sec. 2.8):
l
4 1  r 
 Y θ','  Y θ, .
 l
1 *
r  r'

r
 
l  0 2l  1  r
 l
m
l
m

 m  -l

Problem 2.41. Use the expansion that was the subject of the previous problem to confirm the
analysis, in Sec. 2.9 of the lecture notes, of the system shown in Fig. 2.29: a grounded conducting sphere
of radius R and a point charge q located at distance d > R from its center.
Solution: For the confirmation, let us calculate the potential of the sphere itself, induced by the
suggested system of charges. In the spherical coordinates used in Fig. 2.29, the distance r’ = d of the
point-charge source from the origin is larger than r = R of any point on the sphere’s surface. So, in the
expansion derived in the previous problem’s solution, we have to take r> = d and r< = R, and it yields

Problems with Solutions Page 89

Essential Graduate Physics EM: Classical Electrodynamics

l
1 1  R
    P l (cos  ) .
r  r' d l 0  d 
Hence, according to Eq. (1.37), the electrostatic potential induced by the point charge q (alone, without
the effect of the sphere’s charge) on the sphere’s surface is
 l
q R
q     P l (cos  ).
4 0 d l 0  d 
On the other hand, using the same expansion for the image charge q’ = –(R/d)q located at
distance d’ = R2/d < R from the center, so we have to take r> = R and r< = d’, getting

q'  l
 d'   
 R / d q   R 2 / d d 
l
q 
R
l

 q'   
4 0 R l 0  R 
 P l (cos  )   
4 0 R l 0  R
 P l (cos  )  
4 0 d
   P l (cos  ) .
l 0  d 

We see that two contributions exactly cancel:   q + q’ = 0 for any , thus satisfying the requirement
to have  = 0 at any point of the grounded sphere.
This example shows that the Legendre-polynomial expansion of the Green’s function, proved in
the previous problem’s solution, is in full compliance with the spherical inversion.

Problem 2.42. Suggest a convenient definition of the Green’s function for 2D electrostatic
problems, and calculate it for:
(i) the unlimited free space, and
(ii) the free space above a conducting plane.
Use the latter result to re-solve Problem 21.
Solution: As was discussed at the beginning of Sec. 2.10 of the lecture notes, the Green’s
function G(r, r’) for 3D electrostatic problems is defined as the partial solution (in the specific geometry
of the problem) of the Poisson equation (1.41) for the electrostatic potential at point r, provided that the
right-hand side of the equation is proportional to the 3D delta function (r – r’) – see Eq. (2.208):

 2 G (r, r' )  4 (r  r' ) .

Now, if the field is two-dimensional (say, is a function of only the coordinates x and y), it is described
by the same Poisson equation with /z = 0, so the charge distribution is also z-independent:
 2 2    x, y 
 2   2    2  2     ,
 x y  0
where  = {x, y} is the 2D radius vector. Hence the most natural definition of the 2D Green’s function
G(, ’) is the solution of this equation with the similar, but 2D delta-functional right-hand side:79

 2 G ρ, ρ'   4 ρ  ρ'  .

79 As was noted in Sec. 2.10, numerical coefficients in such definitions is a matter of convention, not affecting the
final solution of any particular problem.

Problems with Solutions Page 90

Essential Graduate Physics EM: Classical Electrodynamics

Physically, this solution is the potential created by a thin straight line, normal to the [x, y] plane,
with the “unit” linear charge density  (in the sense that /40 = 1). This formulation allows us to
readily calculate this Green’s function for simple geometries.
(i) In the unlimited free space, we know the electric field induced by a thin line of charges with
linear density  – see Problem 1.1: E = nE, with

E ,
2 0 ρ
where  is the shortest distance from the field observation point to the line of charges, i.e. just the 2D
distance between the observation point and the point (taken for the 2D origin) where the charged line,
parallel to the z-axis, pierces the [x, y] plane. From here, the electrostatic potential is
 
    Edρ   ln ρ  const    2 ln ρ  const  .
2 0 4 0
Hence the 2D Green’s function is
2
G (ρ, ρ' )  2 ln ρ  ρ'  const   ln ρ  ρ'  const ,

where  and ’ are the 2D radius vectors of the observation point and the charged line’s trace (“cross-
section”), respectively.80
(ii) For a charged line parallel to a grounded conducting plane y = 0 and crossing the [x, y] plane
at the point ’ = {x’, y’}, both the Poisson equation and the boundary condition, (x,0) = 0, may be
satisfied by the introduction of an image line with the charge density –, crossing the [x, y] plane at the
point ” = {x’, –y’}. Hence the Green’s function of the system is
2 2
   
G (ρ, ρ' )   ln ρ  ρ'  ln ρ  ρ''   ln ( x  x' ) 2  ( y  y' ) 2  ln ( x  x' ) 2  ( y  y' ) 2 . (*)

For the particular system that was the subject of Problem 21 (see the figure below), all we need
from the Green’s function is the normal component of its gradient at the surface.
y
 w/ 2  w/2
 0  V  0 x

A straightforward differentiation of Eq. (*) yields

80An additional simple exercise for the reader: use the same approach as in the model solution of Problem 41 to
prove that this Green’s function may be represented as the following series:
n
1ρ

G (ρ, ρ' )  2 ln ρ   2    cos n   '   const ,
n 1 n  ρ 

where > is the largest, and < is the smallest of  and ’, while  and ’ are the polar angles of the 2D vectors 
and ’. This expansion may be very useful for the solution of some 2D problems.

Problems with Solutions Page 91

Essential Graduate Physics EM: Classical Electrodynamics

G G 4y
S  y '0  .
n' y' ( x  x' ) 2  y 2
Plugging this expression into the 2D version of the main formula (2.210) of the Green’s function theory
(without the first term describing the spatially-distributed charge), we get
w / 2
1 G (ρ, ρ' ) V 4y
 (ρ) 
4
k 
k n'
dr' 
4  ( x  x' ) 2
 y 2
dx' .
lk w / 2

This integral may be readily worked out using the substitution   (x – x’)/y, giving81
( x  w / 2) / y
V d V  1 x  w / 2 x  w / 2
  
 ( x  w / 2) / y   1 
2  tan
 y
 tan 1
y 
. (**)

This formula gives exactly the same result as the integral (****) in the model solution of
Problem 21. (See the plots in that solution.) Actually, Eq. (**) may be obtained by working out that
integral, but the above Green’s-function solution is much simpler. Note also alternatively the problem
could be solved using the 3D Green’s function given by Eq. (2.211) of the lecture notes, but this would
cost us one more integration (along the z-axis).

Problem 2.43. A conducting plane is separated into two parts with a very narrow straight cut, and
voltage V is applied between the resulting half-planes – see the figure below. Use the Green’s function
method to find the distribution of the electrostatic potential and the electric field everywhere in the
space. Compare the result with Eq. (2.83) of the lecture notes. In hindsight, could the problem be solved
in an even simpler way (or ways)?
y

  V / 2 0   V / 2 x

Solution: Let us use the coordinates shown in the figure above, with the z-axis directed along the
cut. For our 2D geometry (/z = 0), and in the absence of free charges, the basic formula (2.210) of the
Green’s function theory is reduced to
1 G (ρ, ρ' )
 ( ) 
4
 
k
k
n'
d' ,
lk

where, for our choice of coordinates,  = {x, y}, while G(, ’) is the 2D Green’s function, and the
integration should be extended along all boundaries of conductors’ cross-sections, numbered with index
k. For a semi-space limited by a conducting plane, the Green’s function and its normal derivative have
been calculated in the solution of the previous problem:

  
G (ρ, ρ' )  2 ln ρ  ρ'  2ln ρ - ρ''   ln ( x  x' ) 2  ( y  y' ) 2  ln ( x  x' ) 2  ( y  y' ) 2 , 

81 See, e.g., MA Eq. (6.3d).

Problems with Solutions Page 92

Essential Graduate Physics EM: Classical Electrodynamics

G 4y
y ' 0  ,
y' ( x  x' ) 2  y 2
so for our current problem, we get82
0 
(V / 2) 4y (V / 2) 4y V x
 
4  ( x  x' )  y
2 2
dx' 
4  ( x  x' )
0
2
y 2
dx' 

tan 1
y
.

But this tan-1 is just (/2 – ), where  is the usual polar angle measured from the positive
direction of the x-axis and defined on the interval [-, +], so our result may be rewritten as
V  
    . (*)
 2 
Now we may readily find the electric field. According to MA Eq. (10.2) with / = /z = 0, the field
has only one (azimuthal) component
1  V
E   
ρ  ρ
sgn  , where ρ  x 2  y 2  1/ 2
.

Just as one could expect, the field increases (indeed, diverges) as we approach the infinitely narrow gap
that holds the non-zero voltage V.
Now looking back at Eq. (*), we see that this very simple result might be obtained in several
other ways. First, it is the limiting case, for t  0, of Eq. (2.83) of the lecture notes. Second, it is a
particular case (for  = 0) of Problem B formulated and analyzed in the model solution of Problem 14(i).
Finally, the solution of the latter problem, and Eq. (*) as well, could be just guessed (and then verified)
from the independence of the boundary conditions of , and the simple form of the Laplace equation for
any function ():
1 d 2
0.
ρ 2 d 2
This equation tells us that  should be a linear function of the angle in each of the two sectors where it is
valid: 0 <  < +, and – <  < 0.

Problem 2.44. Use the last result of Problem 42 and one of the conformal mappings discussed in
Sec. 4 to find one more solution of Problem 18.
Solution: As was discussed in Sec. 2.4 of the lecture notes, the analytic complex function (2.81),
w
z  sin , (*)
2
maps the interior of the gap –w/2  u  +w/2, 0  v <  (just with both coordinates normalized to w/2)
on the complex plane w = u + iv,83 onto the upper half-plane, y  0 on the complex plane z = x + iy – see

82See MA Eq. (6.3d).

83I hope that my usage of different fonts excludes any chance of confusion between the complex function w and
the real parameter w.

Problems with Solutions Page 93

Essential Graduate Physics EM: Classical Electrodynamics

Fig. 2.11 and its discussion. Hence it maps the boundary problem considered in Problem 18 onto that
solved in Problem 42 (and, in a different form, in Problem 21). What remains is just to plug into Eq. (**)
of that solution the function (*) spelled out for its real and imaginary parts, with the scale (w/2) for both
coordinates:
w u v w u v
x  sin cosh , y  cos sinh ,
2 w w 2 w w
and then make the replacements u  x and v  y – just in order to comply with the notation used in the
model solution of Problem 18. The resulting formula,
 x y x y 
 sin cosh  1 sin cosh  1
V w w w w
   tan 1  tan 1 .
 x y x y 
cos sinh cos sinh
 w w w w 
looks very different from the final formula of the solution of Problem 18,
4V 
 1n1 cos  2n  1x exp  2n  1 y ,



n 1 2n  1 w

 w


but they give exactly the same results for every point {x, y} within the free-space region –w/2  x 
+w/2, 0  y < .

Problem 2.45. Calculate the 2D Green’s function for the free spaces:
(i) outside a round conducting cylinder, and
(ii) inside a round cylindrical hole in a conductor.
Solutions:
(i) Taking into account the solution of Task (i) of Problem 42,

G ρ, ρ'  unlimited space  2 ln ρ  ρ'  const

(where  = {, } is the 2D radius vector of a point), and inspired by the solution of a similar 3D
problem (see Eqs. (2.197)-(2.198) of the lecture notes), let us try to look for the solution of our current
problem in a similar form:

G ρ, ρ'   2 ln ρ  ρ'  2a ln ρ  ρ"  b   ln ρ  ρ'  a ln ρ  ρ"  b ,

2 2
(*)

where ” is the charge image point whose distance from the cylinder’s ρ   ρ,  
axis is related to that of the actual source point ’ (see the figure on the
right) by the inversion relation similar to the first of Eqs. (2.198):

R2
ρ"  . 0 ρ" R ρ'
ρ'
The coefficients a (physically, the relative magnitude of the image charge)
and b are still to be determined.

Problems with Solutions Page 94

Essential Graduate Physics EM: Classical Electrodynamics

Per the solution of Problem 42, Eq. (*) satisfies the 2D Laplace equation (at   ’, ”) for any a
and b, so these coefficients may be found (and thus the solution (*) proved) by requiring the Dirichlet
boundary condition
G ρ, ρ'   0, for ρ  R,  ,
to be satisfied at any point at the boundary, i.e. for any polar angle . Counting  from the direction
from the center toward the points ’ and ” (so ’ = {’, 0} and ” = {”, 0}), and applying basic
trigonometry to the figure above, we may spell out this condition as

  
 ln ρ' 2  R 2  2 R'' cos   a ln ρ' 2  R 2  2 R' cos   b 
 R 2 
2
 R2  
 2 2

  ln ρ'  R  2 R' cos   a ln    R 2  2 R  cos    b
 ρ'   ρ'  
 R2 2 
  
  ln ρ'  R  2 R' cos   a ln  2 R  ρ' 2  2 R' cos    b  0 .
2 2

 ρ' 
It is evident that this condition is satisfied if a = 1 and b = –ln(R2/’2)  2ln(’/R), so, finally, the
Green’s function is
R2 ρ' ρ  ρ"
G ρ, ρ'    ln ρ  ρ'  ln ρ  ρ"  ln 2  2 ln
2 2
. (**)
ρ' R ρ  ρ'
In terms of electrostatics, the result a = 1 may be interpreted as the image charge (or rather its
linear density) having, in contrast to the 3D cases, the same magnitude as the real linear charge density,
though still the opposite sign. This equality quenches the logarithmic divergence of the Green’s function
(pertinent to the unlimited free space) at large distances.
(ii) Repeating the solution of Task (i) for this geometry, we see
that Eq. (**) is valid for the “inner” problem as well, but now with ’ < ρ   ρ,  
R < ” (see the figure on the right). In particular, on the central line of 
the cylindrical hole (i.e. at  = 0), 0 ρ' R ρ"
ρ" R
G 0, ρ'   2 ln  2 ln  0 .
R ρ'

Problem 2.46. Solve Problem 17(i) using the Green’s function method.
Solution: For our problem, the 2D form of Eq. (2.210), with no space charge and the half-pipe
potentials specified in the problem's assignment, yields

1 V  0  G ρ, ρ' 
 ρ   sgn   R     
 ' '  R Rd' , (*)
4 2  0  
where G(, ’) is the 2D Green’s function of the thin, hollow, grounded pipe, and the sgn function is
due to the opposite signs of the scalar product of the vector ’ by the outer normal n to the surface for
the external problem ( > R, ’n > 0, see the left panel of the figure below) and the internal problem (
< R, ’n < 0, shown on the right panel of that figure).

Problems with Solutions Page 95

Essential Graduate Physics EM: Classical Electrodynamics

ρ
ρ' ρ"
ρ
ρ" ρ'
 
' '
0 0
R R

As was shown in the solution of the previous problem, the Green’s function has the same form in
both cases:
ρ' ρ  ρ"
G ρ, ρ'   2 ln ,
R ρ  ρ'

where the charge image point ” is located on the same radius as the source point ’, at the distance "
= R2/’ from the center. In the notation used in the figure above,

  
G ρ, ρ'  2 ln ρ'  ln ρ 2  ρ" 2  2 ρρ" cos'     ln ρ 2  ρ' 2  2 ρρ' cos'     const . 
Taking into account that at ’ = R, the derivative d”/d’ = –(R/’)2 equals (–1), this expression yields

G ρ, ρ'  2 2 R 2  2 ρR cos'     2 ρ2  R2

 1   
ρ' ρ'  R
R  ρ 2  R 2  2 ρR cos'     R ρ 2  R 2  2 ρR cos'   

Now using MA Eq. (6.3c),84 with

 
a  b  ρ 2  R 2  2 Rρ   ρ  R  ,
2
a 2
 b2 
1/ 2
  ρ 2
 R2   2 ρR  
2 2
1/ 2
 ρ2  R2 ,
we get
G ρ, ρ'  4  ρR '   
 ρ' ρ'  R d'  sgn   R  tan -1  tan .
R  ρ  R 2 

At the substitution of this expression into Eq. (*), the product of two front sgn functions yields +1, and
we get
V  -1  ρ  R '    '  π  ρR '    '  0 
 ρ   tan  tan   tan -1  tan  
2π   ρ  R 2  '  0  ρ  R 2  '  π 

V   ρR   ρR  
 -1
2 tan  cot an   2 tan 1  tan   π sgn sin  ,
2π   ρ  R 2   ρ  R 2  

This is already the final answer, and its numerical plots coincide with those shown in the model
solution of Problem 17(i). However, for the analytical comparison of the results, we may rewrite it as

84 This formula is applicable here because in our case, a2 – b2 = (2 – R2)2  0, i.e. a2  b2.

Problems with Solutions Page 96

Essential Graduate Physics EM: Classical Electrodynamics

V   ρR  ρR 
 ρ      '  sgn sin  , with tan α  cot an , tan α'  tan ,
 2  ρR 2 ρR 2
so by using the well-known formula for the tangent of a sum, we get
tan   tan ' ρR 1   
tan   '     cotan  tan 
1  tan  tan ' ρ  R 1   ρ  R 2 /  ρ  R 2  2 2
ρ 2  R 2 sin 2 / 2   cos 2 / 2  ρ2  R2
  ,
4 ρR sin / 2 cos/ 2  2 ρR sin 
and our result may be rewritten in the form derived in the solution of Problem 17(i):
  V
 -1  ρ  R  π
2 2
V  2 ρR sin 
 ρ   
tan    sgn sin    tan -1 2 .
π   2 ρR sin   2  π ρ  R2
V / 2
An advantage of the Green’s function approach, as well as of the variable
separation method, over the conformal mapping is that they may be more readily used 
in less symmetric situations – for example, in a similar problem with the pipe cut into 0
two fragments as shown in the figure on the right, where  is an arbitrary angle rather
than exactly  as in the problem solved above. (Carrying out such a calculation, using
both methods, is a very useful additional exercise, highly recommended to the reader.) V / 2

Problem 2.47. Solve the 2D boundary problem that was discussed in Sec. 2.11 of the lecture
notes (Fig. 2.34) by using:
(i) the finite difference method with the finer square mesh h = a/3, and
(ii) the variable separation method.
Compare the results at the mesh points, and comment.
Solutions:   V0
(i) Due to the problem’s symmetry, there are only two types of internal
A A
nodes in this mesh: A and B (see the figure on the right), so we need only two
finite difference equations, each describing a 5-point “cross” (see Fig. 2.33b)
B B
with its center in each of these points. For these two cases, Eq. (2.221) yields:
a/3
- points A: 0   A  V0   B  4 A  0, a/3
- points B: 0   B   A  0  4 B  0.  0

Solving this simple system of two linear equations, we get

3 1
 A  V0  0.375 V0 ,  B  V0  0.125 V0 .
8 8
As a simple sanity check, the potential in the cross-section’s center, which is located in the
middle of diagonal segments AB, may be estimated by the linear interpolation of these results, giving

Problems with Solutions Page 97

Essential Graduate Physics EM: Classical Electrodynamics

A  B 3 / 8  1/ 8 1
 center   V0  V0 ,
2 2 4
i.e. the exact value – see Sec. 2.11 of the lecture notes.
(ii) In full analogy with the solution (2.95) of the similar 3D problem, obtained by the variable
separation method in Sec. 2.5 of the lecture notes, the solution of this 2D problem is

nx ny
 ( x, y )   c n sin sinh ,
n 1 a a
where the coefficients cn have to be found from the boundary condition on the top lid (y = a):

nx
V0   c n sin sinh n.
n 0 a
Performing the reciprocal Fourier transform (i.e., multiplying both parts of this equation by sin(n’x/a),
and integrating each of them over x from 0 to a), we get

nx 4V0 / n sinh n, for n  2m  1,

a
1
cn  V0  sin dx   with m  1, 2, 3,...
(a / 2) sinh n 0 a  0, for n  2m,

thus proving Eq. (2.224) of the lecture notes. In the a/3 mesh nodes, this (exact) formula yields the
values
4 
sin 2m  1 / 3sinh2 (2m  1) / 3
 A  V0   0.38072 V0 ,
 m 1 (2m  1) sinh (2m  1)
4 
sin2 2m  1 / 3sinh (2m  1) / 3
B  V0   0.11928 V0 .
 m 1 (2m  1) sinh (2m  1)
Note that though these series are formally infinite, virtually all their values are provided by just a
few first terms, because the exponential character of the sinh functions at large values of their arguments
ensures a very fast convergence of the sums.
Comparing the results provided by the two methods, we see that the finite difference method’s
error, even with this rather coarse mesh, is just a few percent.

Problems with Solutions Page 98

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 3. Dipoles and Dielectrics

Problem 3.1. Prove Eqs. (3.3)-(3.4) of the lecture notes, starting from Eqs. (1.38) and (3.2).
Solution: In Sec. 3.1 of the lecture notes, the first two terms of the multipole expansion (3.3)
have already been derived from the general Eq. (1.38),
1  r' 
 r    d 3 r' ,
4 0 r  r'
so we need only to calculate the third (quadrupole) term from the corresponding term on the right-hand
side of Eq. (3.2):
1 1 3  2 f R 
 q r     
4 0 
 r' 3
r'  0 r' j r' j' d r' , (*)
2! j,j' 1 r j r j'

for our case f(R) = 1/R, where R = R and R  r – r’. Spelling out the function f via the Cartesian
coordinates Rj = rj – rj’ of the vector R,
1 / 2
 3 2
f R     r j  r' j   ,
 j 1 
makes the first differentiation (/rj) in Eq. (*) elementary:
3 / 2
f R   3 2
    r j  r' j   r j  r' j  .
r j  j 1 
However, the result of the second differentiation (/rj’) depends on whether the indices j and j’
coincide, because if they do, then we have to differentiate rj as well:
5 / 2 3 / 2
 2 f R    f  3  3 2  3 2
    r j  r' j   2 r j  r' j  r j'  r' j'    jj'   r j  r' j  
r j r j'  
r j'  r j  2  j 1   j 1 
3R j R j'  jj'
  3,
R5 R
so the particular value participating in Eq. (*) is

 2 f R  3r j r j'  jj' 3r j r j'   jj' r 2

r'  0    ,
r j r j' r5 r3 r5
and may be moved out of the integral over the r’ space. The resulting expression for the quadrupole
potential,
q r  
1 1 3
5 
4 0 2r j , j '1
 
3r j r j'   jj' r 2   r'  r' j r' j' d 3 r' , (**)

though correct, is still not quite convenient for most applications: it is beneficial to move the account for
the special character of the diagonal components of the sum into the integration over r’.

Problems with Solutions Page 99

Essential Graduate Physics EM: Classical Electrodynamics

The standard way of doing that is shown in Eq. (3.4) of the lecture notes; plugging that
expression for Qjj’ into the last term of Eq. (3.3), we may rewrite the latter as

 r r   r' 3r' r' 

3
11
 q r     jj' r' 2 d 3 r' ; (***)
4 0 2r 5
j j' j j'
j , j '1

let us prove that Eqs. (***) and the already proved Eq. (**) are equivalent. The off-diagonal terms under
both sums are clearly similar (they differ only by the location of the factor 3), so the sums over off-
diagonal elements are equal as well. Hence we need to analyze only the sums of the diagonal elements:

  3r r   jj' r 2   r'  r' j r' j' d 3 r'    3r j2  r 2   r'  r' 2j d 3 r'

3 3
in Eq. (**) : j j'
j  j' 1 j 1
3
 3 r j2   r'  r' 2j d 3 r'  r 2   r'  r' 2 d 3 r' ;
j 1

 r r   r'  3r r   jj' r 2  d 3 r'   r j2   r'  3r' 2j  r' 2  d 3 r'

3 3
in Eq. (* * *) : j j' j j'
j  j '1 j 1
3
 3 r j2   r'  r' 2j d 3 r'  r 2   r'  r' 2 d 3 r' .
j 1

We see that the results are identical, thus completing the required proof.

Problem 3.2. A thin ring of radius R is charged with a constant linear density . Calculate the
exact distribution of the electrostatic potential along the symmetry axis of the ring, and prove that at
large distances, r >> R, the three leading terms of its multipole expansion are indeed correctly described
by Eqs. (3.3)-(3.4) of the lecture notes.
Solution: Due to the axial symmetry of the system, the
r'
calculation of its electrostatic potential from Eq. (1.38), integrated over
the cross-section of the ring, is elementary (see the figure on the right):
R 2
 z2 
1/ 2

R
 dr'  2
Rd' r

4 0 
ring

r  r' 4 0  R
0
2
 z2 
1/ 2 0 z
 2R R
  f  ,
4 0 R 2  z 2 1 / 2
2 0 z
where
2
R
f    1   
1 / 2
   , .
z
Let us expand this function f() into the Taylor series in the argument  at point  = 0:
d 1 d2 1 3
f    1    1   1 / 2 1   1 / 2
1 / 2
 0   0    0  2  ...  1     2  ... .
d 2 d 2
2 8

Problems with Solutions Page 100

Essential Graduate Physics EM: Classical Electrodynamics

At large distances from the ring, where  is small, the function f() is well approximated by these three
leading terms, and we get
 R 1  R  3 R  
3 5

        , at z  R . (*)
2 0  z 2  z  8  z  

This expression should be compared with the first three terms of the quadrupole expansion (3.3):
1 1 1 3
1 3 
 r    Q 3
  rj p j  5 r r Qjj '  , (**)
4 0  r
j j'
r j 1 2r j,j' 1 
where
Q   ρr'  d 3 r' , p j   ρr'  r' j d 3 r' , Qjj'   ρr' 3r' j r' j'  r' 2 jj' d 3 r' ,
and (r’) is the electric charge density (per unit volume). In the standard polar coordinates, with the
origin in the ring’s center (see the figure above), the integration over the ring’s length yields
2
Q    Rd'  2R ;
0
2 2 2
p x    R cos ' Rd'  0, p y    R sin ' Rd'  0, p z    0  Rd'  0;
0 0 0
2 2
Qxx    3R 2 cos 2 '  R 2 Rd'  R 3 , Qyy    3R 2 sin 2 '  R 2 Rd'  R 3 ,
0 0
2 2
Qzz    0  R 2 Rd'  2R 3 , Qxy  Qyx    3R 2 sin ' cos ' Rd'  0,
0 0
2 2
Qxz  Qzx    3  0  R cos ' Rd'  0, Qyz  Qzy    3  0  R sin ' Rd'  0 .
0 0

As a sanity check, the quadrupole moment matrix calculated above has zero trace (the sum of
their diagonal elements) – as it should, for any system, by the very definition of the tensor Qjj’:

     
3 3 3
Tr Q    Qjj     r  3r j2  r 3 d 3 r    r  3r j2  r 2 d 3 r    r  3r 2  3r 2 d 3 r  0 .
j 1 j 1 j 1

Now plugging these results into Eq. (**) written for z-axis, i.e. with r1 = r2 = 0, and r = r3 = z,
1 1 1 1 
 z    Q  2 p z  3 Qzz  ,
4 0  z z 2z 
we get
1 1 1 1 3 1  2R R 3 
 z    2R  2  0  3 2R     3  .
4 0  z z 2z  4 0  z z 
Comparing this expression with Eq. (*), we see that the quadrupole approximation exactly
describes the three leading (including one vanishing) terms of that expansion. Evidently, for this simple

Problems with Solutions Page 101

Essential Graduate Physics EM: Classical Electrodynamics

geometry, the exact calculation is simpler, but in more complex cases, the multipole expansion may be
the only way to carry out an (approximate) analytical calculation.

Problem 3.3. In suitable reference frames, calculate the dipole and quadrupole moments of the
following systems (see the figures below):
(i) four point charges of the same magnitude but alternating signs, placed in the corners of a
square;
(ii) a similar system but with a pair charge sign alternation; and
(iii) a point charge in the center of a thin ring carrying a similar but opposite charge uniformly
distributed along its circumference.

i  q a q ii  q a q iii 
R

a a a a Q

Q
q a q q a q

Solutions: For point charges, the definitions of the dipole and quadrupole moments of a system
(see Eq. (3.4) of the lecture notes) take the following forms

p j   q k r j k , (*)
k

Qjj'   q k 3r j r j'  r 2 jj' k , (**)

k

where the index k numbers the charges, the indices j, j’ = 1, 2, 3 mark the Cartesian coordinates, and the
prime sign of the radius vector components is dropped for brevity. Let us y
apply these formulas to systems (i) and (ii). q 2  q q1  q
a
(i) For this system, in the reference frame and with the charge 
2
numbering shown in the figure on the right, Eq. (*) yields
a a
px  q 1  1  1  1  0, py  q 1  1  1  1  0, pz  0 , a/2 0 a/2 x
2 2
a
where the order of terms in the parentheses follows the charge numbers 
2 q  q
and z = 0 for all of them. So, the dipole moment p completely vanishes. q 3  q 4

Similarly, Eq. (**) yields

2 2

Qxx  q  1  1  1  1  0, Qyy  q  1  1  1  1  0,

a a
2 2
2 2

Qzz  q   2  2  2  2  0, Qxy  Qyx  q  3  3  3  3  3qa 2 ,

a a
2 2

Problems with Solutions Page 102

Essential Graduate Physics EM: Classical Electrodynamics

with all off-diagonal components involving the z-coordinate being equal to zero as well, so the
quadrupole moment may be represented by the following off-diagonal matrix:
0 3 0
 
Q  qa  3 0 0  .
2

0 0 0
 
(ii) For this system, in a similar reference frame (see the figure on y
the right): q2  q q1  q
a

a a 2
p x  q 1  1  1  1  0, p y  q 1  1  1  1  2qa, pz  0 ,
2 2
so the system has a nonvanishing dipole moment p = 2qany, and a/2 0 a/2 x
2 2 a
Qxx  q  1  1  1  1  0, Qyy  q  1  1  1  1  0, 
a a
q3  q 2 q  q
2 2 4
2 2

Qzz  q   2  2  2  2  0, Qxy  Qyx  q  3  3  3  3  0,

a a
2 2
with all off-diagonal components involving the z-coordinate being equal to zero as well. So in this case,
it is the quadrupole moment that completely vanishes, while the dipole moment does not.
(iii) Generally speaking, for this system, we need to combine contributions from both the central
point charge +Q and the distributed charge of the ring, with the linear density  = –Q/2R. However, in
a reference frame with its origin in the center of the ring, the former charge (now with r = 0) yields no
contribution to either pj or Qjj’, and we may focus on the ring alone. Directing the axes so that x = Rcos,
y = Rsin, and z = 0, and hence r2 = x2 + y2 = R2 for all points of the ring, we get d3r = Rd, so
2 2
p x  R  R cos  d  0, p x  R  R sin  d  0, pz  0 ,
0 0

(i.e. the dipole moment p is completely vanishing), and

2 2
Qxx  R   3x  r d  R   3R 2 cos 2   R 2 d  R 3   1 2   QR 2 ,
2 2 3 1
0 0 2  2
2 2
Qyy  R   3 y 2  r 2 d  R   3R 2 sin 2   R 2 d  R 3   1 2   QR 2 ,
3 1
0 0 2  2
2 2 2 2
Qzz  R  0  r d  R  R d  QR ,
2 2 2
Qxy  Qyx  R  3xyd  R  R 2 sin  cos d  0,
0 0 0 0

with all other components involving the z-coordinate being equal to zero as well. So, the quadrupole
tensor has the following diagonal matrix:
  1/ 2 0 0
 
Q  qR  0
2
 1/ 2 0 .
 0 0 1 


Problems with Solutions Page 103

Essential Graduate Physics EM: Classical Electrodynamics

As a sanity check, the sums of the diagonal components of all the quadrupole tensors calculated
above equal 0, as they should – see the solution of the previous problem.

Problem 3.4. Calculate the dipole and quadrupole moments of a thin spherical shell of radius R,
carrying an electric charge with the areal density  = 0cos. Discuss the relation between the results
and the solution of Problem 2.28.
Solution: As was discussed in Sec. 1.1 of the lecture notes, we may describe the areal charge
density  of a thin spherical shell by taking the bulk density (r) equal to (r – R), so after their
integration over r’ from R–0 to R+0, the two last Eqs. (3.4) become

p j    r'  r j'd 2 r' , Qjj'    r' 3r j'r j''  r' 2 jj' d 2 r' ,
where both integrals are over the surface of the shell. With the natural choice of the coordinate origin in
the center of the sphere, these integrals become
2 

 ...d r'  R  ...d'  R  d'  ...sin 'd' ,

2 2 2

4 0 0

so for the given charge density  = 0cos, we get

2 
cos ' 
p x    0 R 3   d'  cos ' sin 'd'  0,
2

 y
  0
sin '  0
2 
(*)
4
p z   0 R  d'  cos ' sin 'd' 
3
 0R3,
2

0 0
3
2 
  
Q xx    0 R 4  d'   3cos2 '  sin 2 '  1 cos ' sin 'd'  0,
2

 yy 
  0 0   sin '  
(**)
2 
  2  
Qzz   0 R 4  d'   3cos2 '  cos 2 '  1 cos ' sin 'd'  0,
0 0 
sin '  
and all off-diagonal elements equal zero as well, because they are proportional to the azimuthal integrals
of either sin (for Qxz and Qzx), or cos (for Qyz and Qzy), or sincos (for Qxy and Qyx).
These results are in full accord with the result of Problem 2.28 for the exact field of the sphere
outside it:
 R
  0 2 cos  , for r  R.
3 0 r
Indeed, a comparison of the last formula with Eq. (3.7) of the lecture notes shows that this is the field of
a dipole with moment p = nzpz, where pz is given by the second of Eqs. (*). For such a purely dipole
field, all higher multipole moments starting from the quadrupole one, have to vanish – just as Eq. (**)
exemplifies.

Problems with Solutions Page 104

Essential Graduate Physics EM: Classical Electrodynamics

Problem 3.5. For a regular cubic lattice of similarly oriented identical dipoles, calculate the
electric field it creates at the location of each dipole.
z
Solution: In a Cartesian coordinate system with the origin at the
location of one of the dipoles, and the axes directed along the lattice sides a
(see the figure on the right), the dipole coordinates are
x jkl  aj , y jkl  ak , z jkl  al , p
where j, k, and l are the integers numbering the lattice nodes, with j = k = a
l = 0 at the origin. Now we may use the last form of Eq. (3.13) and the 0 y
linear superposition principle to calculate one of the Cartesian a
components (say, along the x-axis) of the field created at the location of x
the dipole at the origin by all other dipoles of the lattice:

1  3 j ( jp x  kp y  lp z )  p x ( j 2  k 2  l 2 )
E x 0    ,
4 0 a 3 j , k ,l   ( j 2  k 2  l 2 )5 / 2
with the point j = k = l = 0 excluded. The sums of all cross-terms, proportional to the products jk and jl,
vanish due to the system’s symmetry with respect to the inversion of any coordinate, so we get

E x 0  
1 
3 j   j 2 2
 k2  l2  p
 .
4 0 a 3
j  k
j , k ,l  
2 2
l 
2 5/ 2
x

Since all the partial sums participating in this expression are equal,
  
j2 k2 l2
 2 2 2 5/ 2 
j , k ,l   ( j  k  l )
 2 2 2 5/ 2 
j , k ,l   ( j  k  l )
 2 2 2 5/ 2 ,
j , k ,l   ( j  k  l )

we get E x(0) = 0. But, due to the system symmetry with respect to the coordinate swap, the same result
is valid for all other components of the dipole field. Hence E(0) = 0, so (due to the similarity of all the
dipoles of the system), at the location of each dipole, the field of all other dipoles vanishes.
Historically, this counter-intuitive result, first derived by H. Lorentz, served as the basis of his
theoretical explanation of the initially-empirical Clausius-Mossotti formula (3.52) – see Problem 15
below.

Problem 3.6. Without carrying out an exact calculation, can you predict the spatial dependence
of the interaction between various electric multipoles, including point charges (in this context,
frequently called electric monopoles), dipoles, and quadrupoles? Based on these predictions, what is the
functional dependence of the interaction between homonuclear diatomic molecules such as H2, N2, O2,
etc., on the distance between them when the distance is much larger than the molecular size?
Solution: According to Eq. (1.35) of the lecture notes, the electrostatic potential m of a
monopole is proportional to 1/r. Since, according to Eq. (1.31), the potential energy U of a monopole in
an external potential is proportional to the potential itself, the potential energy of monopole-monopole
interaction is inversely proportional to the distance between them:
1
U mm  .
r

Problems with Solutions Page 105

Essential Graduate Physics EM: Classical Electrodynamics

This conclusion is of course elementary, but the above reasoning establishes the framework that
may be readily extended to higher multipoles. For example, according to Eq. (3.7), the potential d of a
dipole scales as 1/r2. Hence its interaction with a monopole
1
U dm  2 .
r
This conclusion is confirmed by Eqs. (3.15) with Eext = Em = –m  1/r2.
The same Eq. (3.15) may be used to calculate the dipole-dipole interaction, now taking Eext = Ed
= –d  1/r3, so
1
U d d  3
r
– the result quantified by Eq. (3.16).
The general trend is perhaps already clear to the reader, but let us follow it one step further.
According to the multipole expansion whose proof was the subject of Problem 1, the quadrupole
potential q decreases with distance as 1/r3, so its interaction with a monopole scales as
1
U qm  .
r3
Again using Eq. (3.15), we may conclude that its interaction with a dipole is proportional to Eq = –q
 1/r4, i.e.
1
U q d  4 .
r
Now we may formulate the general rule, which covers all the above cases: the interaction
between two multipoles as a function of the distance r between them acquires one more factor 1/r each
time when the order of any of the interacting multipoles is increased by 1. On this basis, we may safely
predict that the interaction between two quadrupoles drops with the distance as
1
U q q  .
r5
The formal proof of this statement and the exact expression for Uq-q may be obtained by the
corresponding extension of Eq. (3.14) – the additional exercise highly recommended to the reader.
Due to their symmetry, homonuclear diatomic molecules do not have a dipole moment but may
have a nonvanishing quadrupole moment. Indeed, a crude but fair model of their charge distribution is
given by a nonuniform but symmetric linear density (z) = (–z), where z is the molecule’s symmetry
axis, with the origin in its center. For such a density, the dipole moment vanishes, because its only
significant component,
p z     z zdz ,

is proportional to an integral of an odd function in symmetric limits, i.e. is equal to zero. Using the
above results, and those of the previous problem, one may conclude that the long-range intermolecular
interaction is proportional to 1/r5.
Note, however, that molecules’ quadrupole moments are typically so small that this interaction is
rather insignificant. An even more important factor diminishing the importance of such interaction is
that, according to Eq. (3.3), the quadrupole potential has a zero angular average, so if the mutual

Problems with Solutions Page 106

Essential Graduate Physics EM: Classical Electrodynamics

orientation of the interacting molecules is random (as it is, for example, in the gas phase), the
quadrupole interaction is averaged out. This is why another long-range interaction (called the London
dispersion force), with the potential energy proportional to 1/r6, is usually much more important. As
was mentioned in Sec. 3.1 of the lecture notes, this effect results from the interaction of two mutually
induced (rather than spontaneous) dipole moments, due to their quantum fluctuations.85 Due to this
mutual correlation of the induced dipole moments, this effect is not averaged out at the random
orientation of the molecules and always has the same sign, corresponding to their mutual attraction.

Problem 3.7. Two similar electric dipoles, of a fixed magnitude p, located at a fixed distance r
from each other, are free to change their directions. What stable equilibrium position(s) they may take as
a result of their electrostatic interaction?
Solution: Directing the z-axis along the line connecting the dipoles, we may use the last form of
Eq. (3.16) of the lecture notes for the potential energy of the dipole-dipole interaction:
1 p1x  p 2 x  p1 y  p 2 y  2 p1z  p 2 z
U int  .
4 0 r3
Plugging into this relation the expressions for Cartesian components of both dipole moments via the
polar and azimuthal angles of their orientation,
p jx  p sin  j cos  j , p jy  p sin  j sin  j , p jz  p cos  j , where j  1, 2 ,

we may rewrite the interaction potential as U int = Cf, where C  (1/40)p2/r3 is just a constant, while f
is a function of the angles j and j:

f  sin 1 cos 1 sin  2 cos  2  sin 1 sin 1 sin  2 sin  2  2 cos 1 cos  2
 sin 1 sin  2 cos   2 cos 1 cos  2 ,

where   1 – 2 is the mutual shift of the azimuthal angles of the dipole moment vectors. (The last
form of the function f reflects the obvious fact that the interaction is invariant with respect to any
simultaneous rotation of the vectors about the line connecting them.)
The stable equilibrium orientations of the dipoles correspond to the minima of this function of
three arguments. In order to find them, let us use the system’s symmetry. Since the dipoles are similar,
and the dependence of each p on its  is monotonic, the equilibrium angles 1 and 2 should be either
equal or -complementary, providing us with the following two options:

Option A : 1   2   , giving f  f 0  ,    sin 2  cos   2 cos 2   cos   2  cos   cos 2  ,

Option B : 1     2   , giving f   f 0  ,  .

On the relevant segment 0    , the function f0(, ) has two equal minima, f0(0, ) = f0(, )
= –2, and one maximum, f0(/2, ) = cos . Since  cos   is always less than 2, this means that the
function f has two minima, at  = 0 and  = 0 (when the azimuthal angle  is inconsequential). So, the

85 A quantitative discussion of this interaction may be found, e.g., in QM Chapters 3, 5, and 6.

Problems with Solutions Page 107

Essential Graduate Physics EM: Classical Electrodynamics

system has two similar stable equilibrium states, in which both dipoles are aligned with each other (p1 =
p2), and are together oriented in either direction of the line connecting them.

Problem 3.8. An electric dipole is located above a grounded infinite conducting plane. Calculate:
(i) the distribution of the induced charge in the conductor,
(ii) the dipole-to-plane interaction energy, and
(iii) the force and the torque exerted on the dipole.
z
Solutions:


(i) The problem may be solved by the introduction of a dipole image, d p'
at the same distance d below the plane, and with the same dipole moment
magnitude p as the original dipole, but reflected in the vertical plane normal to
that containing the dipole moment vector – see the figure on the right. The
simplest way to come up with this fact is to represent the dipole in the 0 x
approximate form of two point charges, (+q) and (–q), slightly displaced along 
the direction of the dipole moment vector, and then construct the dipole image
d p''
from the mirror images of these point charges in the conducting plane.
However, so far this is just a guess, not a proof; let us prove this fact.
The net field of these two dipoles evidently satisfies the proper Poisson equation in the upper
half-space, so the only thing we have to prove is that it also satisfies the boundary condition ( = 0) on
the plane’s surface. Let us generalize Eq. (3.7) of the lecture notes to the system of two dipoles (calling
them p’ and p”, see the figure above), with the following Cartesian components:
p x'   p x"  p sin θ, p y'  p y'  0 , p z'  p z"  p cos θ , (*)

and located, respectively, at points r’ and r” with the following Cartesian coordinates:
x'  x''  0, y'  y''  0, z'   z''  d .
(Here x is the coordinate within the vertical plane that contains the vectors p’ and p”, i.e. in the plane of
the above drawing.) In these coordinates, the generalization yields

1  (r  r' )  p' (r  r" )  p"  p  ( z  d ) cos   x sin  ( z  d ) cos   x sin  

     2  3/ 2 
.
4 0  r - r'

3
r - r"
3
 4 0 
 x  y 2  ( z  d ) 23/ 2

x 2  y 2  (z  d )2 


This expression shows that the potential indeed vanishes everywhere on the surface (z = 0), thus proving
our guess.
Now the induced surface charge density may be calculated from Eq. (2.3) of the lecture notes as

   0 z 0 ,
z
giving
p (2d 2  x 2  y 2 ) cos   3dx sin 
 .
2 (x 2  y 2  d 2 )5 / 2

Problems with Solutions Page 108

Essential Graduate Physics EM: Classical Electrodynamics

(ii) The potential energy of interaction between the actual dipole and its image (i.e. the
conducting plane) may be calculated using Eqs. (3.16), with the additional factor ½ because the image
dipole is induced by the actual one – see the discussion following that formula in the lecture notes.
Taking r = 2d, and using Eqs. (*), we get

U int  
1 p2
8 0 (2d ) 3
 
1  cos 2  . (**)

Note that for any angle  , the interaction energy is negative, with its magnitude increasing at d  0, i.e.
an electric dipole is always attracted to a conductor.
(iii) Now we can use Eq. (**) to calculate the force and the torque acting of the dipole. As should
be clear from the symmetry of that expression (namely, its independence on the horizontal position of
the dipole), the force has only one nonvanishing component,
U int
Fz  
d

1 3p2
4 0 16d 4

1  cos 2   0 , 
so the force is directed toward the plane. The torque vector also has only one Cartesian component,
normal to the plane of the drawing:
U 1 p2
 y   int   sin 2 .
 4 0 16d 3
(Alternatively, this result may be obtained from Eq. (3.17) of the lecture notes.) The torque disappears at
 = 0, , and /2, i.e. in all positions in which the dipole moment is aligned with the field created by its
image. Of these configurations, only the former two,  = 0 (the dipole directed up), and  =  (directed
down), are stable with respect to dipole rotation, because they correspond to the minima of the
interaction energy (**).86

R
Problem 3.9. Calculate the net charge Q induced in a grounded  0
r
conducting sphere of radius R by a dipole p located at point r outside the sphere p
0
– see the figure on the right.
Solution: The simplest way to solve this problem is to recall that
according to Eq. (3.9) of the lecture notes, the dipole p may be represented as a limiting case of a couple
of equal but opposite charges q displaced by the vector a = p/q, at a  0 but q  , so that qa = p =
const. For our problem’s geometry, in the first approximation in the small parameter a/r << 1, the
distances of the components of such a pair from the sphere’s center are
a
d  r  cos  , R
2
d q 
where  is the angle between the vectors r and p – see the figure on a
the right. Now applying Eq. (2.199) to each charge of the pair, we get 0 d q

86 In principle, such bistable systems of spontaneous molecular dipoles may be used as binary memory cells,
potentially scalable to atomic size,. However, to be practicable, such cells would need atomic-scale devices for
their state change and readout.

Problems with Solutions Page 109

Essential Graduate Physics EM: Classical Electrodynamics

R R  1 1 
Q q q  Rq   .
d d  r  a / 2 cos  r  a / 2 cos  
The limit of this expression at a/r  0,
Rqa Rp Rr p
Q 2
cos   2 cos   ,
r r r3
gives the solution of our problem.

Problem 3.10. Use two different approaches to calculate the energy of interaction between a
grounded conductor and an electric dipole p placed in the center of a spherical cavity of radius R, carved
in the conductor.
Solution: Let us use the spherical coordinates {r, , }, with the origin in the center of the
sphere, and the z-axis directed along the dipole moment vector p.
Approach 1. As we know from the solution of Problem 2.28, if a thin spherical layer of radius R
is charged with areal density
   0 cos  , (*)
it creates, outside it, a purely dipole field corresponding to the dipole moment p = nz(4/3)0R3. Hence,
if we select 0 to satisfy the condition p + p = 0, i.e. take
3 3
0   p, so    p cos  , (**)
4R 3 4R 3
the net field of these two dipoles vanishes at r  R, thus satisfying the boundary condition (R) = 0.
Now let us use one more result of the solution of Problem 2.28, namely that the internal electric
field induced by the distribution (*) is uniform and equal to
0
E in   nz .
3 0
With the 0 given by the first of Eqs. (**), this field of dipole-induced surface charges is
p
E in  .
4 0 R 3
So, according to Eq. (3.15b), the dipole-cavity interaction energy is
p2
U  .
8 0 R 3
Approach 2. The model solution of Problem 2.35 had the following by-product: the force F of
attraction of a point charge q, displaced from the center of a spherical cavity by distance z << R, is
directed to the nearest point of the surface, and is equal to
q2
F z   z. (***)
4 0 R 3

Problems with Solutions Page 110

Essential Graduate Physics EM: Classical Electrodynamics

Let us represent our dipole p as a result of a gradual separation of two point charges +q and –q, shifting
each of them from the center of the cavity, along the z-axis, to the final positions a/2, where a = p/q <<
R – see Eq. (3.9) of the lecture notes. Physically, the force (***) is exerted by the surface charge induced
by this point charge. In our case, there are two such distributions (induced by each of the two charges),
and each of them is exerted on both charges, so the full work of the forces at the dipole formation is
a/2 p / 2q 2
q2 q21 p  p2
W  4  F z  dz  4 0 zdz  4    .
0 4 0 R 3 4 0 R 3 2  2q  8 0 R 3
(Note that this calculation does not take into account the forces of the direct interaction of the dipole
charges, because they are responsible for the built-in internal energy of the dipole itself, and in
particular, do not depend on R.)
Since at the beginning of the build-up, the dipole did not exist, the potential energy of the
interaction at the end of the build-up interaction is just minus the calculated work, giving the same result
as in Approach 1.

Problem 3.11. A plane separating two halves of otherwise free space is densely and uniformly
(with a constant areal density n) filled with electric dipoles, with similar moments p oriented normally to
the plane.
(i) Use two different approaches to calculate the electrostatic potential at distances d >> 1/n1/2 on
both sides of the plane.
(ii) Give a physical interpretation of your result.
(iii) Use the result to calculate the potential distribution created in space by a spherical surface of
radius R, densely and uniformly filled with radially oriented dipoles.

Solutions:
(i) At the stated condition d >> 1/n1/2, the dipole set may be treated as a 2D continuum, with the
polarization
P  np  z   n z np  z  , (*)

where the z-axis is normal to the plane, whose position is taken for z = 0. This approximation allows us
to proceed using either of the following approaches.
Approach 1. Plugging Eq. (*) into Eq. (3.27) of the lecture notes, we get
np r  r' np z np z
 (r )   nz    z  d 3 r'   d 2 r'    d 2 r' ,
4 0 r  r'
3
4 0 r  r'
3
4 0 2
z 
2 3/ 2

where  is the in-plane component of the vector (r’ – r), and the integration is over the whole plane.
This integral may be readily worked out in polar coordinates:
sgn  z  sgn  z 
 
z z d
  ρ' d 2 r'  2  ρ'dρ'  2   2 2  2 sgn  z ,
2
 z2 
3/ 2
0  ρ' 2
 z2 
3/ 2
2 0   1 3/ 2
2

finally giving

Problems with Solutions Page 111

Essential Graduate Physics EM: Classical Electrodynamics

np
 (r )  sgn z  . (**)
2 0
As usual, the potential is only defined up to an additive constant, but such an addition would not affect
the finite leap of the potential across the dipole layer:
np
    z  0     z  0   . (***)
0

Approach 2. Due to the symmetry of this problem with respect to any translation parallel to the
plane, the electrostatic potential may depend only on the z-coordinate, so the Poisson equation for the
macroscopic electrostatic potential takes the form
d 2 
  ef ,
dz 2
0
where ef is given by Eq. (3.30) of the lecture notes, which in our 1D case is reduced to
dPz
 ef   .
dz
Integrating the Poisson equation over z once, and using Eq. (*), we get
d 1 Pz  np
     z dz   c1    z   c1 ,
dz 0 0 0
where the physical sense of the constant c1 is the (minus) electric field outside of the interface. Per the
Gauss law, the dipole layer cannot create such a field because its net charge density  equals zero, so in
the absence of other field sources, c1 = 0. Now integrating the last equation again over an infinitesimal
interval across the plane (so the contribution of the integral of c1 is negligible), we return to Eq. (**).
(ii) The physical sense of the potential’s leap (***) at z = 0 becomes clear from the simple
cartoon in which each dipole p is represented by two equal and opposite charges q separated by a small
distance vector a = p/q – see Eq. (3.9) of the lecture notes. Because of that, the dipole layer we are
discussing may be faithfully represented by two slightly separated charge layers, with equal and
opposite areal densities  = nq. (Such a system is frequently called the double electric layer.) As we
know from the numerous discussions of such a system in Charters 1 and 2, the electric field induced by
the charges between these layers equals Ez = –nq/0, so the potential difference it creates is  = –Eza =
nqa/0  np/0, thus returning us to Eq. (***).
(iii) According to the Gauss law, since the sphere carries no net charge, the electric field not only
inside it but also outside it has to be zero, so the electrostatic potential should be constant in each of
these regions:
  const, for r  R,
 r    in
  out  const, for r  R,
with the origin of r in the center of the sphere. Very close to any smooth surface, it may be well
approximated with a plane, so the boundary condition (***) is valid for two points very close to the
sphere’s surface, giving
np
 out   in  .
0

Problems with Solutions Page 112

Essential Graduate Physics EM: Classical Electrodynamics

With the usual convention () = 0, this result gives

 np /  0 , for r  R,
 r    (****)
 0, for r  R.
All the above results, based on the approximation of the dipole system with a 2D continuum, are
valid if the average distance n–1/2 between the nearest dipoles is much smaller than any other linear scale
of the problem. For the case of the sphere, this role is played by the distance d =  r – R  of the
observation point from the sphere’s surface and its radius R, so Eq. (****) is valid if

nR 2 , nr  R   1 .
2

Actually, these conditions of continuity are applicable to any smooth interface, provided that R is
understood as the smallest of its two curvature radii.

Problem 3.12. Prove Eq. (24) of the lecture notes.

Hint: You may like to use the basic Eq. (1.9) to spell out the left-hand side of Eq. (24), change
the order of integration over r and r’, and then contemplate the physical sense of the inner integral.
Solution: The first two steps suggested in the Hint yield
1 r  r'  1 r'  r 
 Er  d r   d r  d r'  r'     d 3 r'      .
3 3 3 3
d r r' (*)
V
4 0 V r  r'
3
 4 0 V
3
r'  r 

A comparison of the contents of the square brackets with the same Eq. (1.9) shows that this is just the
electric field Ec(r’) that would be created by a distributed charge with a constant (for the purposes of the
r-space integration) density (r’) in the considered volume V. By the condition of Eq. (24), this volume
is a sphere of any radius R large enough to contain all the charges, so the r’-space integration in Eq. (*)
may be limited to the region with r’ < R. For the observation points satisfying this condition, the field
Ec(r’) was calculated in Sec. 1.2, and we may use for it the vector form of Eq. (1.22) with the
appropriate change of notation (r  r’):
 r' 
E c r'   r' , for r'  R .
3 0
As a result, Eq. (*) is reduced to
1
 Er  d r    E c r' d 3 r'     r' r'd
3 3
r' .
V r ' R
3 0 r ' R

But according to Eq. (3.6), the last integral is just the total electric dipole moment p of our system, so
the last equality coincides with Eq. (24).

Problem 3.13. A sphere of radius R is made of a material with a uniform spontaneous

polarization P0. Calculate the electric field everywhere in space – both inside and outside of the sphere,
and compare the result for the internal field with Eq. (3.24) of the lecture notes.
Solution: According to the general Eq. (3.31), the effective charges with the density ef  –P
create the electric field exactly as the stand-alone charges – which are absent in this problem. Since the

Problems with Solutions Page 113

Essential Graduate Physics EM: Classical Electrodynamics

polarization P is uniform both inside the sphere (P = P0) and outside it (P = 0), ef is different from zero
only on the sphere’s surface, forming the effective surface charge density ef that may be found by the
integration of Eq. (3.31) through an infinitesimal interval of the outer normal n to the surface:
n  0 n  0
 ef  
n  0
ef dn      P dn  P cos ,
n  0
0

where  is the angle between the vector P0 and the local normal to the surface. With the z-axis directed
along the vector P0,  becomes the usual polar angle of the spherical coordinates with the origin in the
sphere’s center. But this means that the spatial distribution of the electrostatic potential  and the electric
field E = – in this problem are exactly the same as were calculated in the solution of Problem 2.28,
with the replacement 0  P0:
Pr P R3
in  0 cos  ,  out  0 2 cos  . (*)
3 0 3 0 r
Comparing the second of these expressions with Eq. (3.7) of the lecture notes, we see that the
electric field outside of the sphere coincides with that of an electric dipole with the moment
4 3
p R P0  VP0 , (**)
3
equal to the sum of all elementary dipoles within the sphere – the result that could be conjectured even
before the formal solution of the problem. On the other hand, the first of Eqs. (*) describes a uniform
electric field inside the sphere, with
P
E in   0 , (***)
3 0
and Din  0Ein + P0 = (2/3)P0. (Note that the vectors E and D have opposite directions, once again
demonstrating the radical difference between these two macroscopic fields.)
Note also another interesting way to derive Eq. (***). In Sec. 3.4 of the lecture notes, we have
obtained the following result for the internal field and polarization of a sphere made of a linear dielectric
(with P  E) , placed into a uniform external field E0 – see Eqs. (3.65):
3  1
E in  E0 , P  3 0 E0 . (****)
 2  2
The field Ein may be considered a sum of the external field E0 and the field Eself created by the
polarization of the sphere itself. For the latter field, Eqs. (****) yield
3  1 P
E self  E in  E 0  E0  E0   E0   ,
 2  2 3 0
i.e. the equality similar to Eq. (***). This is natural because the field induced by a polarization of some
volume does not depend on whether this polarization is induced or spontaneous.
One more note: by using Eq. (**), Eq. (***) may be recast as
p 1 p
E in  
3 0 V
, so E
V
in d 3r  
3 0
,

in agreement with the general Eq. (3.24) whose proof was the subject of the previous problem.

Problems with Solutions Page 114

Essential Graduate Physics EM: Classical Electrodynamics

Problem 3.14. Calculate the electric field at the center of a cube made of a material with the
uniform spontaneous polarization P0 of arbitrary orientation.
Solution: As was discussed in the previous problem, the effect of a uniform polarization of a
volume is equivalent to its surface’s charging with the areal charge density
 ef  P  n , (*)

where n is the unit vector outer-normal to the surface. Let us orient the coordinate axes along the cube’s
sides and consider, for example, the two faces normal to the x-axis. For them, Eq. (*) yields constant
charge densities (ef)x = Px, where Px  Pnx is the x-component of the vector P0 Obviously, these two
face charges give equal contributions to the electric field in the center of the cube, directed normally to
those faces, i.e. along the x-axis. The calculation of the magnitude of the field induced by one cube’s
face carrying a constant charge density  was one of the tasks of Problem 1.7; the result was E =
/60. Replacing  with ef, doubling the result (to reflect the equal contributions of both charged faces),
and taking into account the field’s direction (from the positive charge to the negative one), we see that
the x-normal faces yield
P
Ex   x nx ,
3 0
with no contributions to other field components because of the system’s symmetry. Now repeating this
analysis for the other two face pairs, we may summarize our result as

E  Ex  Ey  Ez  
1
Px n x  Py n y  Pz n z    P0 , (**)
3 0 3 0
for any polarization’s orientation and independently of the cube’s size.
Note that this is exactly the same result as was obtained in the solution of the previous problem
for the spontaneously polarized sphere. Also, by multiplying both sides of Eq. (**) by the volume V of
the cube, and taking into account that its total dipole moment of the cube is p = P0V, we may rewrite our
result in the form of Eq. (3.24) of the lecture notes (whose proof was the task of Problem 13):
p
E
V
in d 3r  
3 0
,

showing that the validity of this relation extends beyond the spherical geometry.

Problem 3.15. Derive the Clausius-Mossotti formula (see Eq. (3.52) of the lecture notes) by
combining Eq. (3.24) with the result of the solution of Problem 5.
Solution: Due to the linear superposition principle, the actual (microscopic) field Em(r) inside a
dipole medium may be represented as the sum,
E m r   E ext  E self r , (*)

where Eext is the field of the distant stand-alone charges (say, those located outside of the considered
sample)87 and the field Eself of the medium itself. On the atomic/molecular scale, the former of these

87 Per the discussion following Eq. (3.33) of the lecture notes, Eext is just the “would-be” field D/0.

Problems with Solutions Page 115

Essential Graduate Physics EM: Classical Electrodynamics

fields may be considered uniform, while the latter of them may vary from point to point very
substantially.
First, let us average Eq. (*) over a macroscopic sphere of volume V so large that, near its center,
the field of the outer molecules is negligible, applying Eq. (3.24) to the last term, Eself(r):88
1 P
 E m r  d 3 r  E  E ext  .
VV 3 0
Here E is the usual average (macroscopic) electric field, while P is the average polarization of the
medium. In an isotropic linear dielectric, they are related by Eq. (3.45), P = ( - 1)0E, so the last
displayed relation may be rewritten as
P P
 E ext  . (**)
ε 0   1 3 0
On the other hand, we may apply the same Eq. (*) to the center of one of the molecules, taking it
for r = 0:
E m 0   E ext  E self 0.

Modeling molecules as point dipoles, we may associate Em(0) with the field participating in Eqs. (3.48)-
(3.49), so this equality becomes
P
 E ext  E self 0. (***)
αn
Now let us apply, to the last term, the result of Problem 5 (obtained for a simple but representative
model of the dipole lattice): Eself (0) = 0. With this, the combination of Eqs. (**) and (***) immediately
yields the Clausius-Mossotti relation
n /  0
 1  . (****)
1  n / 3 0
Note that this derivation substantially depends on the variation of the field Eself(r) from point to
point – in particular, on the relation Eself(0)  Eself(r). Quantitatively, it is based on modeling molecules
with point dipoles, explaining why Eq. (****) does not work well for many condensed materials in that
the effective size of the constituent molecules is close to the average distance between them.

Problem 3.16. Stand-alone charge Q is distributed, in some way, within the volume of a body
made of a uniform linear dielectric with a dielectric constant . Calculate the total polarization charge
Qef residing on the surface of the body, provided that it is surrounded by free space.
Solution: Let us apply the macroscopic Gauss law (see Eq. (3.34) of the lecture notes) to two
closed surfaces – one (Sin) immediately inside the body’s surface and one (Sout) immediately outside it,
so the stand-alone charge inside them is the same (Q):

D d r  Q, D d r Q.
2 2
n n (*)
Sin Sout

88As the solution of the previous problem illustrates, this relation is not limited to the spherical geometry, which,
due to its isotropy, represents the average of Eself pretty fairly.

Problems with Solutions Page 116

Essential Graduate Physics EM: Classical Electrodynamics

The inner surface is inside the dielectric, so at all its points, D = E, i.e. Dn = En (where   0 is the
dielectric’s permittivity), while at the outer surface, D = 0E, i.e. Dn = 0En. Plugging these relations into
Eqs. (*), we get
Q Q
 En d r  ,  En d r  .
2 2
(**)
Sin  Sout 0

But according to the “microscopic” Gauss law (see Eq. (1.16) of the lecture notes), applied to the very
slim volume between these two surfaces,
Qef
 En d r   En d r  ,
2 2

Sout Sin 0

because this volume, by the problem’s conditions, does not contain any stand-alone surface charges but
may have a surface polarization charge Qef. Plugging in the integrals from Eqs. (**), we get
Q Q Qef     1
  , giving Qef  Q 1  0   Q 1   .
0  0     
Note that this result does not describe the part of the polarization (“effective”) charge with the
density ef = –P that may be distributed in the volume of the body. (This density vanishes only if the
polarization inside the body is uniform: P = const.)

Problem 3.17. In two separate experiments, a thin plane sheet of a linear dielectric with  =
const is placed into a uniform external electric field E0, in two different ways:
(i) with the sheet’s surfaces parallel to the electric field, and
(ii) with its surfaces normal to the field.
For each case, find the electric field E, the electric displacement D, and the polarization P inside
the dielectric, sufficiently far from the sheet’s edges.
Solution: The same reasoning that was used in Sec. 3.4 of the lecture notes to analyze free-space
slits in dielectrics, based on the continuity of E and Dn at the “major” (larger) surfaces of the sheet,
gives the following results:
(i) for the sheet parallel to the electric field:89
E  E0 , and hence D  E   0 E 0 , P    1 0 E 0 ;
(ii) for the sheet normal to the field:
D E0  1
D  D0   0E0 , and hence E   , P  0E0 .
 0  
We see that in the latter case, both fields, D and E, are  times weaker than in the former one.

89Note that this result was also derived in Sec. 3.5 from the condition of the minimum of the Gibbs potential
energy (3.78).

Problems with Solutions Page 117

Essential Graduate Physics EM: Classical Electrodynamics

Problem 3.18. A fixed dipole p is placed in the center of a spherical cavity of radius R, carved
inside a uniform linear dielectric. Calculate the electric field distribution everywhere in the system.
Hint: You may start with the assumption that the field at r > R has a distribution typical for a
dipole. However, be ready for surprises.
Solution: Following the Hint, let us look for the electric potential’s distribution in the form
 p' cos 
1  r2 , for r  R,
 
4 0  p cos    (r ,  ), for r  R,
 r2
where (r,) is an axially symmetric function that satisfies the Laplace equation (added to accommodate
the promised surprise). With this assumption, the boundary conditions at r = R take the form
p' cos  p cos 
  const :    ( R,  ),
R2 R2
(*)
  2 p' cos  2 p cos  
   const :     ( R,  ).
n r R 3
R3 r
Representing the function (r,) as a series over the Legendre polynomials (see, e.g., Eq. (2.172) of the
lecture notes), we see, from the boundary conditions, that only the first term of that expansion, a1rcos  ,
may have a nonvanishing magnitude. For the coefficients p’ and a1, Eqs. (*) turn into a system of two
linear equations:
p' p 2 p' 2p
2
 2  a1 R,   3   3  a1 ,
R R R R
with the solution
3p 2(  1) p
p'  , a1   .
2  1 (2  1) R 3
Hence the electric field outside the spherical cavity indeed corresponds to a single dipole, but
with a different dipole moment p’ = 3p/(2 + 1), while that inside the sphere is the sum of the original
dipole field and (here is the promised surprise) an additional uniform field
1 2  1 p
E in  .
4 0 2  1 R 3
As a sanity check, in the limit   , this field approaches the -independent value that was
calculated in the model solution of Problem 10. On the other hand, in the limit   1 (say, of no
dielectric at all), the uniform field vanishes.

Problem 3.19. A spherical capacitor (see the figure on the right) is   ,  

filled with a linear dielectric whose permittivity  depends on the spherical
angles  and , but not on the distance r from the system’s center. Derive an
explicit expression for its capacitance C. a

ba

Problems with Solutions Page 118

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Let us prove that the boundary problem for the field distribution inside the capacitor is
satisfied by the following radially-directed fields:
E  n r E r , D  n r   ,  E r  .
Since the curl of such a vector E equals zero,90 the first macroscopic Maxwell equation for E, given by
Eq. (3.36) of the lecture notes, is satisfied. On the other hand, the macroscopic Maxwell equation (3.32)
for D is satisfied if D = 0. For the selected form of D, this equation is reduced to91
d 2
dr

r E r   0 . 
Its straightforward integration yields the same result,
c
E r   12 ,
r
as for the capacitor without the dielectric (i.e. with  = 0 = const), which was analyzed in Sec. 2.3(iii) of
the lecture notes, so the boundary conditions for the corresponding electrostatic potential,
r
c1
 r    a    E r dr   a  
 c2 ,
a
r
may be satisfied by the same choice of the constants c1,2 as in that case.
Hence, our proof has been accomplished, and now we may use Eq. (2.53) to write the expression
for the electric field on the surface of any conductor – say, the inner one:
1
V 1 1
E a   2    .
a a b
Only at this stage, the difference between the current and dielectric-free cases cuts in, because the areal
density of stand-alone charges on the surface of the conductor has to be calculated by using Eq. (3.54)
rather than Eq. (2.3):
1
V 1 1 ,
 a  D a     ,  E a     ,   2   
a a b
so the capacitance is
1 1  2
Q 1 a2 1 1 1 1
C a   ad r  V r a a d   a  b  4   , d   a  b   sin d  d   ,  .
2
(*)
V V r a 0 0

For the particular case  = const, this double integral is equal to 4, and we immediately get an
evident generalization of the (easy) solution of Problem 2.8(i):
1
1 1 ab
C  4     4 . (**)
a b ba
Evidently, Eq. (*) may be interpreted as just a result of the connection, in parallel, of many elementary
spherical sectors d = sindd, each with a capacitance dC described similarly to Eq. (**):

90 See, e.g., MA Eq. (10.11) with f = f = 0 and / = / =0.

91 See, e.g., MA Eq. (10.10) with f = f = 0.

Problems with Solutions Page 119

Essential Graduate Physics EM: Classical Electrodynamics

ab
dC    ,   d. .
ba

Problem 3.20. A spherical capacitor similar to that considered in the previous problem is now
filled with a linear dielectric whose permittivity depends only on the distance from the center. Obtain an
explicit expression for its capacitance, and spell it out for the particular case (r) = (a)(r/a)n.
Solution: As was discussed at the end of Sec. 3.4 of the lecture notes, in the case of spatial-
dependent permittivity (r), we may find the distribution of the electrostatic potential (defined by Eq.
(1.33), E = –) by solving the Maxwell equation (3.32) for D together with the relation (3.46) valid for
linear dielectrics:
  D     r E     r     . (*)

In our spherically-symmetric case,  may depend only on the distance r from the capacitor’s
center (which we will take for the spherical coordinates’ origin), so in the region a < r < b where  = 0,92
this equation is reduced to93
1   2  
 2  r  r    0 .
r r  r 
Its first integral is, obviously,
 c
 2 ,
r r  r 
where c is an integration constant. This constant may be expressed via the inner conductor’s charge Q by
applying the macroscopic Gauss law (3.34) to a sphere slightly larger than it:
 c
4a 2 Da  0  4a 2  a   4a 2  a   4c  Q ,
r r a 0
a  a 
2

so our first integral becomes94

 Q
 .
r 4r 2  r 
Now we may integrate this equation again to get a formal solution for the potential’s distribution
in the dielectric layer,
r
Q dr
  
 r  a    for a  r  b,
4 a r  r 
2

and in particular, for the voltage between the electrodes and hence their mutual capacitance:

92 As a (hopefully, unnecessary) reminder, in Eq (*),  is the density of stand-alone charges not bound
into the dielectric layer’s dipoles.
93 See, e.g., MA Eqs. (10.8) and (10.10), with / = / = 0.
94 Admittedly, this formula could be obtained immediately via the application of the same Gauss law to a sphere
of radius r. However, since this problem is my first (and the only) example of using Eq. (*) with a continuous
function (r), I wanted to demonstrate a more general approach to its solution, which is applicable to less
symmetric geometries as well.

Problems with Solutions Page 120

Essential Graduate Physics EM: Classical Electrodynamics

b b
Q dr Q dr
V   a    b    r  r  , so C   4  r  r  . (**)
4 a
2
V a
2

This expression may be interpreted as a result of the connection in series of many elementary
quasi-plane capacitors of thickness dr and area A(r) = 4r2, with their reciprocal capacitances obeying
Eq. (3.55): d(C-1) = dr/A(r)(r) – see the discussion of Eq. (3.58) in the lecture notes.
For the particular case given in the assignment, the integration is elementary, giving


n  1 1  a / b n 1 ,
C  4 a  a  
 for n  1,
1 lnb / a , for n  1.
As a sanity check, in the particular case of a uniform dielectric case (r) =  = const, i.e. n = 0, this result
coincides with the formula obtained (for the same case) in the solution of the previous problem:
1 1
C  4    .
a b

Problem 3.21. A uniform electric field E0 has been created (by distant 
external sources) inside a uniform linear dielectric. Find the electric field’s E0
change created by carving out a cavity in the shape of a round cylinder of R
radius R, with its axis normal to the external field – see the figure on the
right.
Solution: Introducing the usual polar coordinates, we can use the
general solution (2.112) of the Laplace equation. Based on our experience
with using it for the problem shown in Fig. 2.15 of the lecture notes, we may
immediately look for the solution of this equation in the following form:
 b 
   R  a1  cos  ,    R    E 0   1  cos  ,
 
where the coefficient a1 has the physical sense of the uniform field inside the cavity (with the minus
sign). Using the boundary conditions of continuity of  and /n (and hence of /) on the cavity
surface ( = R), we get the following system of two linear equations,
b1  b 
a1 R   E 0 R  , a1     E 0  12 ,
R  R 
for two unknown coefficients, a1 and b1. Solving the system, we get:
2  1
a1   E 0 , b1   E0 R 2 .
 1  1
As a result, the electrostatic potential’s distribution may be represented as
2
  R  E 0  cos  ,
 1

Problems with Solutions Page 121

Essential Graduate Physics EM: Classical Electrodynamics

  1 R2    1 R2 
  R   E 0     cos    E 0 x 1  ,
  1     1 x  y
2 2

where x is the Cartesian coordinate along the initial field.
As a (necessary for us all :-) sanity check, at  = 1 (uniform space with no dielectric), the
potential distribution is the same at both  > R and  < R:
   0   E 0  cos  ,
and corresponds to the uniform field E0 = E0nx.
In the general case, the electric field, E = –, is:

2  1  R2 ( y2  x2 ) 2 R 2 xy 
E  R  E 0n x , E   R  E0 n x  E 0 n x  n .
 1   1  x 2  y 2 2 y x 2  y 2 2 
From here, the electric field’s change from its original value E0 is:

 2   1  1  R2 ( y2  x2 ) 2 R 2 xy 
E   R    1 E 0 n x  E 0n x , E   R  E 0 n x  n 2 
.
 1   1   1  
x2  y2
2
 y

x2  y2  
It is curious that in the limit   , the internal electric field increases by exactly E0nx, i.e. by
100% of its initial value, while D drops dramatically (by a factor of ) as a result of the cavity carving.
The reader is invited to interpret this fact in the light of the thin-gap “experiments” discussed in Sec. 3.4
of the lecture notes – see Fig. 3.9.

Problem 3.22. Similar small spherical particles, made of a linear dielectric, are dispersed in free
space with a low concentration n << 1/R3, where R is the particle's radius. Calculate the average
dielectric constant of such a medium. Compare the result with the apparent but wrong answer
  1    1 nV , (WRONG!)
(where  is the dielectric constant of the particle's material and V = (4/3)R3 is its volume), and explain
the origin of the difference.
Solution: Due to the given condition nR3 << 1, we may neglect the mutual electrostatic
interaction between the particle dipole moments p induced by an external electric field E0, so for each p,
we may use Eq. (3.64) of the lecture notes:
 1
p  4 0 R 3 E0 .
 2
Now we may calculate the average polarization of the medium by just adding all moments in a unit
volume:
 1
P  np  4nR 3  0E0 .
 2
By applying Eq. (3.37) to the average values of P and , we get

Problems with Solutions Page 122

Essential Graduate Physics EM: Classical Electrodynamics

 1 3
  1  4nR 3  1    1 nV . (*)
 2  2
The formula shows that the apparent result cited in the assignment is valid only asymptotically
when the last fraction on its right-hand side of Eq. (*) approaches 1, i.e. in the limit   1 when the
particle material's polarization is very weak. The reason for this difference is clear from the solution of
the single sphere polarization problem in Sec. 3.4, and in particular from Fig. 3.11b and Eq. (3.65): due
to the non-uniform distribution of the electric field in the vicinity of the sphere (caused by its
polarization), the field inside the sphere (albeit uniform) differs, by exactly the factor 3/( + 2), from the
external field far from the sphere. The comparison shows that generally, the average dielectric constant
depends not only on the concentration, volume, and material of the dispersed particles but also on their
shape and orientation in space.
Note also in the limit   , the correct result (*) approaches Eq. (3.51) for a diluted medium of
metallic spheres, which was derived in the lecture notes in a similar way.

Problem 3.23. A straight thin filament, uniformly charged with linear density , is positioned
parallel to the plane separating two uniform linear dielectrics, at a distance d from it. Calculate the
electric potential’s distribution everywhere in the system.
Solution: Using the macroscopic Gauss law (3.34) for a straight filament with the linear charge 
in a uniform medium with dielectric constant , for its radial-oriented fields we readily get
 D  , y
D , so that E  
2  2 0
 
where  is the distance between the observation point and the closest point of
the filament. (In the coordinate system shown in the figure on the right, 2 = d
x2 + (y – d)2.) Hence its electrostatic potential is
1
  1 0
 ln   const  ln 2  const. (*) 2 '
x
2 0 4 0 x   y  d 2
d
Now repeating the argumentation used in Sec. 3.4 of the lecture notes
to solve the problem shown in Fig. 3.12, but replacing the point charges q, q’, q'
and q” with linear charge densities , ’, and ” (see the figure), and the
basic Coulomb law  = q/40r with Eq. (*), we get the following 2D analog of Eq. (3.66):
1  1 q' 
   ln 2  λ' ln 2 2 
 const, for y  0,
1  1  x  ( y  d ) 2
x  ( y  d ) 
  (**)
4 0  " 1
ln 2  const, for y  0,
  2 x  ( y  d ) 2

where the charge densities ’ and ” still need to be found from the boundary conditions at the interface
(y = 0). Using Eqs. (**) to calculate the corresponding tangential and normal components of the fields
on both sides of the interface,

Problems with Solutions Page 123

Essential Graduate Physics EM: Classical Electrodynamics

1 x    λ' / 1 , for y  0,

E  E x   
2 0 x 2  d 2 " /  2 , for y  0,

1 d   λ' / 1 , for y  0,

En  E y    
2 0 x 2  d 2  " /  , for y  0,
 2

and requiring them to satisfy Eqs. (3.37) and (3.56) for all x, we get the following simple system of
linear equations:
  ' "
 ,   '  " ,
1 2
which gives
1   2 2 2
'  , ''  . (***)
1   2 1   2
Formulas (**) and (***) give the full solution of our problem. Note that despite its different
dimensionality, it yields essentially the same results for the image charges as for the system discussed in
the lecture notes. Indeed, in the particular case considered in Sec. 3.4: 1 = 1 and 2  , Eqs. (***) give
exactly the same results for the ratios ’/ and ”/ as Eq. (3.68) gives for the ratios q’/q and q”/q. Of
the new results given by the more general Eqs. (***), note an interesting dependence of the image
charge’s sign on the difference 1 – 2: if this difference is positive, the density ’ of the charge
observed from the points with y > 0 has the same sign as the real one.
Let me add one more (hopefully, unnecessary) note: there is no need to be concerned about the
divergence of Eqs. (**) at large distances from the filament because they still yield finite values of the
electric field E = –, and hence of the field energy density u  E2. Indeed, such divergence of the
electrostatic potential with distance is typical for many applications of this concept (starting from the
uniform electric field’s description) and is an artifact of the assumption of the infinite spatial extent of
the system under analysis – see, e.g., the discussion of Eqs. (2.49)-(2.50) in the lecture notes.

Problem 3.24. A point charge q is located at a distance d > R from the center of a sphere of
radius R, made of a uniform linear dielectric with permittivity .
(i) Calculate the electrostatic potential’s distribution in all the space, for an arbitrary ratio d/R.
(ii) For large d/R, use two different approaches to calculate the interaction force and the energy
of interaction between the sphere and the charge, in the first nonzero approximation in R/d << 1.
Hint: Task (i) cannot be carried out using the method of charge images, so you may like to use
the expansion of the function 1/ r – r’  in the series over the Legendre polynomials, whose proof was
the subject of Problem 2.40.
Solution: Aligning the z-axis, with its origin in the sphere’s center, along the direction toward the
point charge, let us look for the potential in the following natural form:
 
  al r Pl cos  , for r  R,
l

q
 r    l 0
 (*)
4 0  1  bl
 Pl cos  , for R  r.
 r  r' l 0 r l 1

Problems with Solutions Page 124

Essential Graduate Physics EM: Classical Electrodynamics

Here the regular parts of the field follow the corresponding (non-diverging) parts of the general solution
(2.172) of the Laplace equation for our axially-symmetric case, while the singular term, with r’ = nzd,
takes care of the point charge’s field in its vicinity. Following the Hint, we may represent this term as a
similar expansion:
l
1 1   r 
    P l (cos ),
r  r' r l 0  r 

where r> is the largest of the two scalars r and r’  d, and r< is the smallest of them. Plugging it into Eq.
(*), we get the following expressions for the potential and its normal derivatives on the sphere’s surface:
 
  a l R Pl cos  , for r  R  0,
l

q  l 0
   
4 0   R b 
r R l

  l 1
 ll1  Pl cos  , for r  R  0,
 l  0  d R 
 
  a l l R l 1 Pl cos  , for r  R  0,
 q  l 0
  
r
rR
4 0   l R
l 1
l  1bl 
   Pl cos  , for r  R  0.
 l 0  d l 1
R l  2 
According to the boundary conditions (3.37) and (3.56), in our case we should have
 
 r  R 0  r  R 0 ,  r  R 0  r  R0 ,
r r
where   /0. Due to the orthogonality of the Legendre polynomials, these equalities should hold for
each l of the series. This requirement gives us the following pair of linear equations for each pair of the
coefficients al and bl:
Rl bl lR l 1 l  1bl
al R  l 1  l 1 ,
l
al lR  l 1 
l 1
,
d R d R l 2
which may be readily solved, giving
2l  1 1 l   1 R 2l 1
al  , bl   . (**)
l  l  1 d l 1 l  l  1 d l 1
With these results plugged in, Eqs. (*) give the full solution of Task (i).
The left figure below shows the resulting general pattern of the equipotential planes (more
exactly, their intersection with a plane passing through the sphere’s center and the point charge) for a
particular case d/R = 1.5 and  = 10. This pattern clearly shows the partial expulsion of the electric field
from the high- dielectric, which becomes full at   , i.e. for a conductor – see Fig. 2.29 of the
lecture notes.
The right figure below shows a more quantitative description of this trend: plots of the
potential’s distribution along the symmetry axis of the system for a few values of  (and for the same
charge position d = 1.5R as in the left figure). Note that the potential at the center of the sphere does not
depend on ; indeed, Eqs. (*) and (**) show that (0) = (q/40)a0 = q/40d, i.e. exactly the potential
the charge would produce at that point in the absence of the sphere.

Problems with Solutions Page 125

Essential Graduate Physics EM: Classical Electrodynamics

z  r cos 

0.8

  30
 0.6 10
4 0 R 3
0   r sin  0.4
1

0.2 d
R
0
2 1 0 1 2 3 4
z/R

(ii) With the solution of Task (i) on hand, we may readily calculate all global characteristics of
the charge-to-sphere interaction. In particular, according to the basic Eq. (1.60), the sphere-to-charge
interaction energy U may be calculated as
q
U  s r d ,
2  0
where s is the potential induced by the sphere at the point charge’s position. This potential is given by
the second line of Eq. (*) less the singular term describing the field of the charge itself. Since according
to Eq. (**), b0 = 0,95 at r  , this potential is dominated by the dipole term with l = 1:
q b1 q  1 R3
s r d    ,
 0 4 0 d 2 4 0   2 d 4
so we get
1  1 q2 R3
1
U  , for R  d . (***)
4 0 2   2 d 4
Since for all realistic materials,  > 1, this formula describes the mutual attraction of the sphere and the
charge – for any sign of q. The attraction force may be calculated as
U 1  1 q2 R3
F   2 , for R  d . (****)
d 4 0   2 d 5
The last formula may be also obtained without any appeal to the exact solution. Indeed, at R <<
d, the point charge’s field E = q/40d2 at the sphere’s location is nearly uniform on the scale of its
radius R << d. From the problem solved in Sec. 3.4 of the lecture notes, we know that such a uniform
field induces, in a dielectric sphere, a dipole moment of the magnitude p = 40ER3( – 1)/( + 2),
directed along the initial electric field – see Eq. (3.64). Hence we can use the general formula for the

95This is natural because the corresponding term with l = 0, proportional to 1/r, describes the far field of the
sphere’s net electric charge – which it does not have.

Problems with Solutions Page 126

Essential Graduate Physics EM: Classical Electrodynamics

radial component of the dipole field, Er = 2pcos /40d3 (see the first form of Eq. (3.13) of the lecture
notes), with  = 0, to calculate the interaction force:
2 2  1 2  1 q
F  qE r  q pq 4 0 R 3 Eq 4 0 R 3 .
4 0 d 3
4 0 d 3
 2 4 0 d 3
  2 4 0 d 2
This is the same result as in Eq. (****). Now we may readily obtain Eq. (***) by either
integrating the calculated force over d from  to r, or just by using Eq. (3.15b) of the lecture notes for
the interaction energy of a dipole with the field that has induced it.

Problem 3.25. Calculate the spatial distribution of the electrostatic potential

induced by a point charge q located at distance d from a very wide parallel plate, of
D
thickness D, made of a linear dielectric – see the figure on the right.
q
Solution: Let us direct the z-axis normally to the plate so that the radius vector
of the charge is r = {0, 0, d}, as shown in the figure below, and examine what system  d
of charge images may describe the field everywhere in the system, while satisfying the
boundary conditions (3.35) and (3.37) on both surfaces of the plate.

q 3  q1   q 0  q 2 
       
Q3  Q1  Q0  Q2 
q' q z
 4 D  d    2 D  d  d d 2D  d

z  D z0

From Sec. 3.4 of the lecture notes (see Fig. 3.12 and its discussion) we know that the conditions
on the plate’s right surface (z = 0) may be satisfied by using, besides the actual charge q, just two of its
images: the charge
 1
q'   q (*)
 1
located at z = –d and describing the additional field in the free space (at z  0), and the charge
2
q Q0  (**)
 1
(in Sec. 3.4, called q”) co-located with the actual charge (z = +d) but describing the field inside the
dielectric (at z  0).
Now let us try to satisfy the boundary conditions on the opposite surface of the plate, located at z
= –D, by introducing an image charge Q1 located at z = –(2D + d), i.e. at the same distance (D + d) from
this surface as the charge Q0 (see the figure above) and describing the additional field inside the
dielectric. The most important point here (and in this solution as a whole) is that the value (**) of the
charge Q0 cannot be adjusted, because the charge Q1 (as well as all other image charges we will
introduce later) gives a different spatial dependence of the potential inside the dielectric, so their net
potential is

Problems with Solutions Page 127

Essential Graduate Physics EM: Classical Electrodynamics

1  Q0 Q1 
1 r    2  1/ 2 
, for  D  z  0 , (***)

4 0     z  d 2 
1/ 2
 2
 z  2 D  d 
2
 

(where   (x2 + y2)1/2 is the distance of the field observation point r from the axis z), so any adjustment
of Q0 would irreparably violate the boundary conditions at z = 0.
However, to balance the field (***) in the dielectric, and hence on the right of the border z = –D,
with that in the free space left of the border (i.e. at z < –D), we need a new contribution that would have,
at the border, the same dependence on . Such contribution may be provided by another charge image
located at the same distance (D + d) from that border as Q0 and Q1. This charge cannot be co-located
with Q1, because that would violate the Laplace equation at that point, z = –(2D + d), which does not
have a real charge. Hence, this new image charge (let us call it q0) has to be co-located with the charge
Q0, i.e. sit at the same point z = d (see the figure above), creating the contribution
1 q0
1 r   , for z   D , (****)

4 0  2   z  d 2 1/ 2

to the potential.96 Now applying the boundary conditions (3.35) and (3.37) at z = –D to Eqs. (***) and
(****), we get a system of equations somewhat similar to Eqs. (3.67),
Q0  Q1  q 0 ,  Q0  Q1   q 0 ,

which allows us to express the new image charges via the already determined Q0:
 1 2
Q1  Q0 , q0  Q0 .
 1  1
(As an intermediate sanity check, at  = 1, i.e. with no dielectric plate at all, Q1 = 0, and q0 = Q0 = q – as
it should be because in this case, the field everywhere is just that of the real point charge q.)
Now we should revisit the surface z = 0, and compensate the field contribution by the charge Q1
in the dielectric (see Eq. (***) above) by an absolutely similar introduction of two new charge images
located at the same distance, 2D + d, from that boundary: charge Q2 at z = 2D + d and describing a new
contribution to the field in the dielectric, and charge q1 co-located with Q1, which describes a new
contribution (besides those from q and q’) to the field in the free space outside the dielectric. The
boundary conditions relating Q1, Q2, and q1 are exactly the same as the above relations for Q0, Q1, and
q0, and lead to similar expressions:
2
 1   1 2 2   1
Q2  Q1    Q0 , q1  Q1  Q0 .
 1   1  1  1  1
Similarly returning to the surface z = –D, then to surface z = 0, etc. (see arrows in the figure
above, showing the sequence of new charge image introduction), we get an infinite system of image
charges

96 Here and later, I will use lower-case letters q to denote any charge image describing a contribution to the field
in a free space region (necessarily on the plate’s side opposite to that charge’s location), while capital-case letters
Q are reserved for the charges giving contributions into the field inside the plate.

Problems with Solutions Page 128

Essential Graduate Physics EM: Classical Electrodynamics

n n
  1 2 2    1 
Qn    Q0 , and q n  Qn    Q0 , for n  0 ,
  1  1  1  1
located at z = nD + d = 2mD + d for even n = 2m, and at z = –(n + 1)D – d = –2mD – d for odd n = 2m –
1 (with m = 0, 1, 2,…). From this series, using Eq. (**) for Q0, and taking into account the contributions
from the actual charge q and the “irregular” image charge q’ given by Eq. (*),97 we may readily
construct the potential distribution in the system:
2m
q 4 
  1 1
 r   2   , for z   D ,
4 0   1 m 0    1   2
  z  2mD  d  
2 1/ 2

  1 1 

 r  
q 2    1
2 m 1
  1 2
 
   z  2mD  d 2 
1/ 2 

   
4 0   1 m 0    1  1   m 0 
, for  D  z  0 ,
 


    z  2mD  d 
2 2 1/ 2
 

 1  1 1 

q
 2

   z  d  
2 1/ 2

  1  2   z  d 2 1 / 2  

 r    , for 0  z .
4 0  4 
  1
2 m 1
1 
   12     1 
 m 1 
 2   z  2mD  d 2 1/ 2 

The figure on the right shows plots 1
of this potential as a function of z, for the
particular values  = d = D, and three
representative values of the dielectric 0.8
 1
constant . (The necessary sanity check is 3
that at both borders, the potential  is 0.6 10
continuous, while its derivative /z jumps 
by a factor of .) The plots show that the q / 4 0 D
0.4
increase of  suppresses the electric field
not only behind the dielectric plate but also
inside and in front of it, i.e. in the vicinity 0.2
of the actual point charge. (According to
Eqs. (3.66) and (3.68) of the lecture notes, plate
0
the last effect is even stronger for a 3 2 1 0 1 2 3
dielectric half-space, which may be viewed z / D
as the particular case of our current problem for D  .) It is also instructive to compare these plots
with the right figure in the solution of the previous problem, which is a spherical analog of this one.

97This charge (the only one with the sign opposite to that of the actual charge q) drops out of the regular sequence
{Qn, qn} because this is the only charge that may be located inside the dielectric plate, and hence cannot have a
co-located partner (say, Q’), which would give a diverging contribution into the field inside the plate.

Problems with Solutions Page 129

Essential Graduate Physics EM: Classical Electrodynamics

As an additional exercise, I can recommend the reader to solve this problem also by the variable
separation method, using the following (generally useful) expansion:

1 k z
  J 0 k  e dk ,
 2
z 
2 1/ 2
0

where J0() is the Bessel function of the first kind – see Sec. 2.7 of the lecture notes.

Problem 3.26. Discuss the physical nature of Eq. (3.76) of the lecture notes. Apply your
conclusions to a material with a fixed (field-independent) polarization P0(r), and calculate the electric
field’s energy of a uniformly polarized sphere (see Problem 13 above).
Solution: In a dipole media with a field-independent number n of elementary dipole moments p
per unit volume, P = np, per the general Eq. (3.33), the electric energy density’s variation, given by
Eq. (3.76), equals
u  E    0 E  P    0 E  E  E  P  u 0  nE  p,
where u0  (0/2)E2 is the energy density of the electric field as such – see Eq. (1.65) of the lecture notes.
But the product Ep is just the elementary energy transfer from the electric field to the internal degrees
of freedom of the elementary dipole p, resisting its polarization.98 (For example, for the simplest model
of the dipole: a point charge q deviating from its equilibrium position by distance r, with p = qr, this
product is Ep = Eqr = Fr, where F = qE is the force exerted by the field.) Hence, the deviation of
the electric field energy U from the free-space value (1.65) is due to the additional energy, possibly of
non-electric origin, stored inside elementary dipoles – for example, polarized atoms or molecules.
For a medium with a field-independent polarization P0(r) (for example, an electret or a saturated
ferroelectric), the internal energy of the elementary dipoles does not depend on the electric field (P =
0), so it is reasonable to associate with the field only the energy density u0. Hence, the full electric field
energy may be calculated as an integral,
0
U  U0  E
2
d 3r , (*)
2
over the whole space where the electric field is significant – both inside and outside the volume filled
with the polarized medium. However, in some cases it may be technically easier to calculate the energy
using the equivalent Eq. (1.60), with the charge density replaced by the effective charge (3.30):
1 1
U   ef r  r d 3 r      P0 r  r  d 3 r , (**)
2 2
because this integration may be limited to the polarized medium – moreover, if P0 is spatially
independent inside it, then just to its surface.
For example, for Problem 13 (the uniformly polarized sphere), where the effective charge resides
on the sphere’s surface S, with the areal density ef = P0cos, Eq. (**) expression yields

98 This expression is in agreement with Eq. (3.15a), despite its apparently different sign. For example, if the
dipole moment increases by p in a fixed external field Eext, then the energy of its interaction with the field
described by Eq. (3.15a) decreases by pEext, and hence its internal energy increases by that amount.

Problems with Solutions Page 130

Essential Graduate Physics EM: Classical Electrodynamics

2 
1 R 2 P0
2 S  d  cos 
U  ef  r R d r 
2
r R sin d .
2 0 0

Plugging in the potential’s distribution over the surface of the sphere, calculated in the solution of that
problem,
PR
 r  R  0 cos  ,
3 0
we get
R 2 P0 P0 R  P02 R 3 2  P02 R 3 2 2P02 R 3
2 3 0 0 6 0 0 0
U cos 2
 sin d   d  cos 2
 sin d   2  .
6 0 3 9 0
A straightforward integration by using Eq. (*) and the results obtained in the model solution of
Problem 13 (a good additional exercise for the reader), confirms this result, with one-third of the energy
U localized inside the sphere and two-thirds, outside it.

Problem 3.27. Use Eqs. (3.73) and (3.81) of the lecture notes to calculate the force of attraction
of a plane capacitor’s plates, for two cases:
(i) the capacitor is charged to voltage V, and then disconnected from the battery,99 and
(ii) the capacitor remains connected to the battery.
Solution: The attraction force may be calculated as
U
F ,
d
where d is the distance between the plates, and U is the appropriate potential energy. For case (i), when
the capacitor is disconnected from the battery, the only relevant energy is that of the capacitor itself, so
we have to use Eq. (3.73). In this particular case, with the uniform field E = V/d inside the volume Ad
between the plates, and a negligible field outside it, this formula gives
2
  V  AV 2
U  E Ad 
2
  Ad  ,
2 2d  2d
where A is the capacitor’s area.
Before the partial differentiation over d, however, we have to take into account that as d is
varied, so is the voltage V. In order to spell out this dependence, we may use the fact that since the
capacitor is disconnected from the battery, its electric charge Q is not affected by d, so using the well-
known expression for the capacitance C (see Eq. (3.55) of the lecture notes), we get

99“Battery” is a common if misleading term for what is usually either a single electric accumulator or a single
galvanic element. (The last term stems from the name of Luigi Galvani, a pioneer of electric current studies.
Another term derived from his name is the galvanic connection, meaning a direct connection of two conductors,
enabling a dc current flow – see the next chapter.) The term “battery” had to be, in all fairness, reserved for the
connection of several electric accumulators or galvanic elements in series – as was pioneered in 1800 by L.
Galvani’s friend Alexander Volta.

Problems with Solutions Page 131

Essential Graduate Physics EM: Classical Electrodynamics

2
Q Qd A  Qd  Q2
V   , and hence U     d.
C A 2 d  A  2A
(Actually, for the particular case of no dielectric filling, i.e.  = 0, this expression in the form
U 2
 d,
A 2 0
where  = Q/A is the areal density of the plate’s charge, was already obtained in the solution of Problem
1.15.) Now the differentiation over d yields

F
  Q2  Q2
 d   
CV 2  AV 2 . (*)
d  2A  2A 2A 2d 2

This result (which coincides with the solution of Problem 2.1, with the natural replacement 0  ) may
be verified by a direct calculation of the force per unit area, F/A = E/2, with the factor of ½ resulting
from the fact that the force exerted on one plate is due to the electric field E/2, produced by the other
plate alone.
However, in case (ii), when the voltage V is fixed by the battery, the use of the “ordinary”
potential energy (3.73) would give a wrong result:
  AV 2  AV 2
F     , (WRONG!) (**)
d  2d  2d 2
which differs from the correct result (*) by the sign. The reason for this error is that if the capacitor stays
connected with the battery during the virtual change of d, we have to take into account the total energy
of the (capacitor + battery) system, or at least of its part depending on d. This is exactly the definition of
the Gibbs potential energy UG, given (for the electrostatic field’s case) by Eq. (3.81). Since according to
that expression (besides an inconsequential constant), UG = –U, there is no need to do the calculations
again, and we may just revert the sign in the wrong Eq. (**), turning it to the correct Eq. (*).
So, the attraction force between the plates, at the same voltage between them, does not depend
on whether the capacitor is still connected to the battery or not.
Let me hope that this simple example illustrates the difference between the potential energies U
and UG very clearly.

Problem 3.28. A slab made of a linear dielectric is partly

inserted into a plane capacitor – see the figure on the right.
 d
Assuming the simplest (cylindrical) geometry of the system,
calculate the force exerted by the field on the slab, for the same two
cases as in the previous problem. x ax
Solution: As it follows from the analysis of a similar system
in Sec. 3.4 of the lecture notes (see Fig. 10a), the insertion of the stub, with its lateral surface normal to
the capacitor plates, does not perturb the electric field uniformity inside the capacitor. Also, since the
capacitor plates are equipotential, the electric field in both parts of its volume is the same: E = V/d,
where V is the voltage across it.

Problems with Solutions Page 132

Essential Graduate Physics EM: Classical Electrodynamics

Now we are ready to proceed to the requested calculations. In the first case, when the capacitor is
disconnected from the battery, the relevant energy U is given by Eq. (3.73), and we can (neglecting the
fringe field effects) calculate it as the sum of energies of the two parts of its interior:
 E 2  E2 
U  U x  U a x   x  a  x  0 bd , (*)
 2 2 
where a is the complete width of the capacitor (see the figure above) and b is its other planar size (in the
direction formal to the plane of its drawing). If we calculated the force Fx from this expression as –
U/x, we would get a wrong (negative) result because if the capacitor is disconnected from the battery,
E also depends on x. Hence we should first calculate what is indeed constant at the variation of x – the
full electric charge Q of the capacitor, which is also the sum of the contributions from its two parts:

Q  Q x  Qa  x  b x x  a  x  a  x   b xD x  a  x Da  x   b xE  a  x  0 E , (**)

so
Q
E .
bx   0 a  x 
Plugging the last relation into Eq. (*), we get
Q2 d
U .
2 bx   0 a  x 
(As a sanity check, each term in the square brackets, after its multiplication by b/d, is the capacitance of
the corresponding part of the system, so this expression for U describes the energy (2.15) of the parallel
connection of these partial capacitors, as was discussed in Sec. 3.4.) Now we are ready to calculate the
requested force:
U Q 2    0 d
Fx    ;
x 2 bx   0 a  x 2
plugging in Q from Eq. (**), we get
U    0 E 2
Fx    bd . (***)
x 2
Since for any realistic dielectric,  > 0, this result shows that Fx > 0, i.e. the force pulls the slab
into the capacitor – see the figure above. Note also the product bd is just the area of the interface
between the slab and the free space inside the capacitor, so we may rewrite our result in the form
Fx    0 E 2
 ,
Ainterface 2
hinting that it may be applicable not only to this particular geometry. (As will be discussed in Chapter 9,
this turns out to be true.)
Now proceeding to the case when the applied voltage V and hence the applied field E = V/d are
fixed, here the appropriate potential energy is the Gibbs energy UG. For our particular case when, per
Eq. (3.45), the polarization of the dielectric material of the slab does not depend on x but is proportional
to E: P = ( – 0)E, it is beneficial to derive this energy from Eq. (3.86):

Problems with Solutions Page 133

Essential Graduate Physics EM: Classical Electrodynamics

1    0 E 2 xbd
U G    PEd r  const  
3
,
2 0, x  2

so an elementary differentiation over x yields the same Eq. (**).

Two major conclusions may be drawn from this solution. The first one is (in hindsight :-) almost
trivial, especially after the solution of the previous problem: the force exerted on the slab does not
depend on whether the capacitor is still connected to the battery or not if the electric field in it is still the
same. Another fact to remember is that for some calculations, the Gibbs energy UG is much more
convenient than the “usual” electrostatic energy U.

Problem 3.29. For each of the two capacitors shown in Fig. 3.10 of the lecture notes, calculate
the electric force exerted on the interface between two different dielectrics, in terms of the fields in the
system.
Solution: As the discussions in Sec. 3.5 of the lecture notes and the two previous problems show,
the force does not depend on which electric field parameter of the capacitor is considered fixed: its full
charge Q or the voltage V across it, so let us use the first assumption. In this case, the relevant potential
energy of the system, whose gradient (with the minus sign) is the force, is given by Eq. (3.73) of the
lecture notes. Reviewing the arguments that were used in Sec. 2.2 to derive Eq. (2.15),
Q2
U ,
2C
we see that it remains valid for capacitors with linear dielectrics,100 so we can use the results for C
obtained in Sec. 3.4.
For the capacitor with the dielectric interface normal to the
capacitor’s plates (see the figure on the right), whose capacitance is given
by Eq. (3.57), we get d 1 2
Q2 Q 2d Q2d
U   ,
2C 2 1 A1   2 A2  2b 1 x1   2 x2 
x1  A1 / b x 2  A2 / b
where b is the linear size of the capacitor in the direction normal to the
plane of the drawing. Now differentiating this expression over x  x1 = const – x2, we obtain

U U U Q 2  1   2 d  1   2 E 2 bd , Fx Fx  1   2 E 2
Fx       i.e.   ,
x x 2 x1 2b 1 x1   2 x 2 2 2 Ainterface bd 2
where E = V/d = Q/Cd is the electric field – common for both parts of the capacitor. Note that the result
obtained in the previous problem is just the particular case of this one, for 1 = 0 and 2 = 0.
In contrast, as was already discussed in Sec. 3.4, in the capacitor
with the dielectric boundary parallel to the electrodes (see the figure on d1 1
the right), the field common for both dielectrics is D =  = Q/A, so it is d2 2

100
A
This is an important qualification – see the discussion in Sec. 3.5.

Problems with Solutions Page 134

Essential Graduate Physics EM: Classical Electrodynamics

more natural to express the final result via this field. Using Eq. (3.59) of the lecture notes to write
Q 2 Q 2  d1 d 2  ,
U    
2C 2 A   1  2 
and taking y  d2 = const – d1, we get
U U U Q 2  1 1  D2  1 1  Fy D2 1 1 
Fy             A, i.e.    .
y d1 d 2 2 A   1  2  2  1  2  A 2  1  2 
In the limit 2  , corresponding to the second material turning into a conductor, i.e. to a plane
capacitor of thickness d  d1, this result is reduced to that of Problem 27, with  = 1.
Note also for both considered systems, the force exerted on the interface is directed toward the
dielectric with the lowest permittivity, i.e. the electric field always “tries” to pull in the material with the
higher dielectric constant, pushing out that with the lower .

Problem 3.30. One half of a conducting sphere of radius R, R

carrying electric charge Q, is submerged into a half-space filled with a 0 Q
linear dielectric with permittivity  – see the figure on the right. Calculate  0
the electric force exerted on the sphere by the dielectric.
Solution: Repeating the arguments given in the model solution of
Problem 19 for a similar (and actually, more general) system, we see that
the electric field outside the sphere is purely radial:
c
E r   2 , for R  r ,
r
with the same constant c in the free space and in the dielectric. This constant may be related to the
charge Q by applying the macroscopic Gauss law (3.34) to a sphere of radius r  R:
Q Q
D d
2
r  2r 2  0 E r   2r 2 E r   2  0   c  Q, so c  , i.e. E r   .
2  0    2  0    r 2
r
S

According to the solution of Problem 27 (see also Problem 2.1), this field, normal to the
conductor’s surface, exerts on it the following elementary radial force directed outward:

dFr E 2 R   0 , in the free - space part,

 
dA 2  , in the submerged part.
At the integration over the sphere’s surface, only one Cartesian component (normal to the dielectric’s
surface) of each dFr gives a nonvanishing contribution to the net force exerted on the sphere:

dFz 2 dF dF
Fz   d r   r cosd 2 r  2R 2  r cos sin d
S
dA S
dA 0
dA

E 2 R   2 E R    0
 /2 
 2
 1 Q2    0
 2R  0 0 cos sin d   / 2cos sin d   2R 2  2  2    8 R 2    2 .

2

2  0

Problems with Solutions Page 135

Essential Graduate Physics EM: Classical Electrodynamics

So, for all realistic dielectric media, with  > 0, the field pulls the sphere toward the dielectric
half-space (regardless of the sign of the sphere’s charge). This is exactly what we could expect from the
solution of the three previous problems.
Curiously, in our current case, the force as a function of   0 has a maximum at  = 30, i.e. at 
= 3. The reason for this behavior is that at   , the effective capacitance of the lower (dielectric-
embedded) part of the sphere increases, leading (at fixed charge Q) to the drop of its potential, i.e. to the
decrease of the electric field everywhere in the system, and hence of the attraction force it exerts.

Problems with Solutions Page 136

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 4. DC Current
R1 R1 R1
Problem 4.1. DC voltage V0 is applied to the open end of a 
semi-infinite chain of lumped Ohmic resistors, shown in the figure on
the right. Calculate the voltage across the jth link of the chain. V0 R2 R2

Solution: This problem is a close analog of Problem 2.4 for a 

similar chain of lumped capacitances. As in its solution, since all links of the chain are similar, it is
sufficient to analyze just a couple of them – say, with numbers j and (j + 1), where j is the “distance”
from the open end. Denoting the currents through the “series” resistors R1 and the voltages across the
“parallel” resistors R2 as shown in the figure on the right, the 1st
Kirchhoff law (4.7a) for the left of the two displayed circuit Ij I j 1
nodes gives  
Vj R1 V R1 V j 1
I j  I j 1  , (*) j

R2
R2 R2
nd
while the 2 Kirchhoff law (4.7b) for the displayed loop reads
V j  V j 1  R1 I j 1 . (**)  

These relations are absolutely similar to Eqs. (*) of the model solution of Problem 2.4 (with the
natural replacements Q  I and C  1/R), so we may use the same arguments to look for the solution of
the infinite system of equations (*) and (**), written for j = 1, 2, …, in a similar form:
V j 1 I j 1
V j  I j  e  j , so that     e    1.
Vj Ij
Indeed, using the last of these relations to eliminate Vj+ 1 and Ij+1 from Eqs. (*) and (**), we get a system
of two linear homogenous equations for the two variables Vj and Ij:
Vj
1    I j  , 1   V j  R1I j .
R2
These equations are compatible (and hence the assumed solution is valid for all j) if the determinant of
the system equals zero:
1  1 / R2  R 
 0, i.e. if  2   2  1    1  0 .
 R1 1  R2 

The needed root  < 1 of this quadratic equation is

1/ 2
R R R2 
  1  1   1  1 2  . (***)
2 R2  R2 4 R2 
Since our definition of Vj (see the figure above) complies, for j = 0, with the boundary value V0,
the requested voltage across the jth link of the chain is

Problems with Solutions Page 137

Essential Graduate Physics EM: Classical Electrodynamics

V j  V0 j  V0 e   j ,

where  is given by Eq. (***).

In the limit R1/R2  , this relation yields
R1
  1 .
2 R2
This result means that the applied voltage does not propagate noticeably beyond the first resistance R1 of
the chain, because the next, relatively small resistance R2 effectively “grounds” the voltage across it –
see the figure above. In the opposite limit, R1/R2  0, Eq. (***) yields
1/ 2 1/ 2
R  1 R 
  1   1   1, so that   ln   1   1 .
 R2    R2 
According to the definition of the parameter , this strong inequality means that the applied voltage
penetrates deep into the chain, decaying at the “distance” of the order of N  1/  (R2/R1)1/2 >> 1.

Problem 4.2. It is well known that properties of many dc current sources (e.g., batteries) may be
reasonably well represented as a connection in series of a perfect voltage source and an Ohmic internal
resistance. Discuss the option, and possible advantages, of using a different equivalent circuit that would
include a perfect current source.
r
Solution: The well-known representation mentioned in the I
assignment is shown in the figure on the right. Here the circle denotes
a perfect voltage source – an imaginary two-terminal circuit element   V  V
that sustains a fixed voltage V, regardless of what is connected to it.
101  
(Naturally, it is impossible to implement such an ideal element
separately from the battery’s internal resistance r; indeed, by short-circuiting its terminals, we would get
infinite current at the non-zero voltage V, i.e. an infinite power.) Let us apply the 2nd Kirchhoff law
(4.7b) to the loop formed by this equivalent circuit and the (unspecified) external circuit to write the
relation between the voltage V at the battery’s terminals and the current I flowing through the battery:102
V  V  rI . (*)
This is just an analytical representation of the equivalent circuit shown above (and hence of the battery’s
properties), and is certainly well known to the readers from their undergraduate studies.
The fact which attracts much less attention is that the following equivalent form of Eq. (*),
V V
I I  , with I  , (**)
r r
may be graphically represented with another equivalent circuit – see the figure below. Here the circle
with an arrow is a common notation of a perfect current source defined as an imaginary two-terminal

101 This is exactly the formal definition of the electromotive force V. (Note that in the circuit theory, it is
frequently denoted as either E or E.)
102 This relation, in the limit R  0, explains why the perfect voltage source is sometimes called the “battery
i
with zero internal resistance”.

Problems with Solutions Page 138

Essential Graduate Physics EM: Classical Electrodynamics

I
circuit element that sustains a certain current I (in our case, equal to V/r),
regardless of the circuit it is connected to. Evidently, such an ideal element is V r
also impossible to implement separately, without the parallel connection of a
I  V
r
finite resistance r, because it would push the same current I even through
empty space, creating infinite voltage and hence supplying infinite power.)
Due to the mathematical equivalence of Eqs. (*) and (**), the two equivalent circuits shown
above always give the same final result, but each of
them may be more or less convenient, depending
V1 , r1 V2 , r2 V1 r1 V2 r2
on the problem. For example, if we need to  
calculate the current in the circuit shown on the    
first panel of the figure on the right, it is more 
R I ? R I
convenient to use the first equivalent circuit of each
battery, with its perfect voltage source and internal
resistance in series. Indeed, as the second panel of that figure shows, its application reduces the full
circuit to a connection, in series, of the net e.m.f. (V1 + V2) to the total resistance (r1 + r2 + R),
immediately giving the final result:
V  V2
I 1 .
r1  r2  R
However, using this equivalent circuit for the solution of the V1 , r1
(infamous :-) problem shown in the lower figure on the right is much V1
I1 
less convenient, and requires using the heavy artillery of the r1
Kirchhoff laws (4.7). On the other hand, using the perfect-current-   r1
V2 , r2
source alternative immediately reduces the full circuit to a simple
connection in parallel of an ideal source of the total current (I1 + I2) V2
 I2 
and the total Ohmic conductance (1/r1 + 1/r2 + 1/R) of the three   r2
resistors, so the requested voltage across the circuit is calculated very r2
R
similarly to the current in the first problem:
R
I1  I2 V1 / r1  V2 / r2
V   . V ?
1 / r1  1 / r2  1 / R 1 / r1  1 / r2  1 / R V
(The current through the “load” R is of course just V/R.)


Problem 4.3. Prove the following Rayleigh-Lorentz-Carson I1  r  V
 1
reciprocity relation: the results of the two separate experiments shown
schematically in the figure on the right, with an arbitrary Ohmic conductor
with four electrodes/terminals, are related as I1V2 = I2V1.
 I2
Hint: Try to apply the same approach as was used to prove Green’s
V  r 
reciprocity relation of electrostatics in Problem 1.18, but with proper 2
modifications.
Solution: Let us consider the integral almost similar to that as in the solution of Problem 1.18:

J    r  E1 r   E 2 r d 3 r ,
V

Problems with Solutions Page 139

Essential Graduate Physics EM: Classical Electrodynamics

where V is the whole volume of the conductor, (r) is its Ohmic conductivity, and E1(r) and E2(r) are
the electric field distributions in the two experiments under consideration. First, let us apply the Ohm
law (4.8) to E1(r) and Eq. (1.33) to E2(r); then the integral may be rewritten as

J   j1  E 2 d 3 r    j1   2 d 3 r ,
V V

where j1(r) is the current distribution in the first experiment and 2(r) is the potential’s distribution in
the second experiment. Using the same integration by parts as in the solution of Problem 1.18, we get

J    2   j1 d 3 r      2 j1  d 3 r.
V V

Since we are considering dc (i.e. time-independent) conductivity, the first term vanishes due to
the continuity relation – see Eq. (4.6). Now using the divergence theorem to transform the second term
to a surface integral (again, just as this was done in the solution of Problem 1.18), we get
J     2 j1 n d 2 r ,
S

where S is the full surface of the conductor, and n is the outer normal to it. Nonvanishing contributions
to this integral are given only by the interfaces SL of two current-carrying electrodes in the first
experiment (shown in the figure above as two left bold points). On each of these interfaces, 2 is
constant because, in the second experiment, they carry no current. Marking these potential’s values by
upper indices , we get
J   2   j1 n d 2 r   2   j1 n d 2 r.

SL 
SL

But these two integrals are nothing else than the full currents flowing out of the conductor, equal to (–I1)
for the interface +, and (+I1) for the electrode –, so
J   2 I 1   2 I 1 .  I 1V2 .

Now repeating the same calculation with the indices 1 and 2 swapped, and requiring the two
results to be equal, we get the reciprocity relation I1V2 = I2V1.103
Note that this relation is conditioned by the Ohmic (linear) conductivity of the sample.

Problem 4.4. Calculate the resistance between two large uniform


Ohmic conductors separated by a very thin, plane, insulating partition
with a circular hole of radius R in it – see the figure on the right.
Hint: You may like to use the same oblate spheroidal coordinates 2R
that were discussed in Sec. 2.4 of the lecture notes.
Solution: Selecting the parameter R of the degenerate ellipsoidal
coordinates defined by Eqs. (2.59) of the lecture notes,

103This simple proof was suggested by J. Carson only in 1930 (mimicking an earlier work on elasticity
theory by Lord Rayleigh), well after it had been derived from a more general reciprocity relation proved
in 1896 by H. Lorentz, by using the full set of Maxwell equations – see Chapters 6 and 7.

Problems with Solutions Page 140

Essential Graduate Physics EM: Classical Electrodynamics

x  R cosh  sin  cos  ,

y  R cosh  sin  sin  ,
z  R sinh  cos  ,
to be equal to the hole’s radius, and placing the coordinate origin into its center, we see that the partition
(outside the hole) corresponds to  = /2 = const, while for the conductor points slightly above the
partition, z ≈ R (/2 – ) sinh, so
  1 
z 0  z 0  .
n rR z r R R sinh  
This means that if the potential is a function of  alone, the boundary condition of having no current
flow through the partition is automatically satisfied. (Due to the mirror symmetry of the problem with
respect to the partition plane, the same conclusion is valid for the lower semi-space as well.) From the
disk capacitance problem’s solution in Sec. 2.4 of the lecture notes, we already know that the Laplace
equation may be also satisfied with the potential distribution (2.64):
  c1 tan 1 (sinh )  c 2 . (*)

For the sake of notation simplicity, let us accept the (anti)symmetric boundary conditions at
infinity:
V V
   sgn( z )   sgn( ), at z  , i.e. at    ,
2 2
where V is the voltage applied between the distant points of the conductors; in this case, the solution (*)
takes the simple form
V
   tan 1 (sinh ).

Note that according to this formula, the contribution of distant parts of the conductor (with r >> R) to the
total voltage drop is negligible; this is why the exact shape of the external electrodes is not important,
provided that they are not too small and are not too close to the hole.
What remains is to calculate the total current
 2
I   j n d 2 r    d r,
S S
n
where S is an arbitrary surface crossed by the whole current flow. (Due to the current continuity, the
result does not depend on the choice of the surface.) The easiest choice is evidently the plane z = 0 –
more exactly, its fraction inside the hole, r < R. Near the plane, the coordinate relation for z is reduced
to z ≈ R cos, while our solution, to just  ≈ –(V/), so –/z = V/Rcos. Since at this plane, the
distance of a point from the center is   (x2 + y2)1/2 = Rsin, so Rcos = (R2 – 2)1/2, we get
 R
 2  d d ( 2 )
R R
V
I   
z 0
z
d   2 
0
z
d  2
  R cos 
0
 V  R
 0
2
 2 
1/ 2
 2VR ,
 R

so the hole’s resistance V/I is just 1/2R.

Problems with Solutions Page 141

Essential Graduate Physics EM: Classical Electrodynamics

This Maxwell (or “Maxwell-Rayleigh”) formula for resistance is frequently memorized as the
ratio of the conductor’s resistivity 1/ to the hole’s diameter 2R, without any numerical coefficients.

Problem 4.5. A very narrow plane crack inside a round 

conducting wire of radius R does not reach its surface by a small distance j
w – see the figure on the right. Assuming that the Ohmic conductivity 
of the wire’s material is otherwise constant, calculate the electric
resistance of the obstacle in the first approximation in small w/R << 1.
Hint: You may like to use the same elliptic coordinates as were w
2R
employed at the solution of Problem 2.12.
Solution: Due to the condition w/R << 1, the local distribution of
the electric potential  created by the current flowing around the y
crack’s edge may be considered a function of just two Cartesian
coordinates x and y shown in the figure on the right. In this  0   const
approximation, the function should satisfy the 2D Laplace equation,   const
 
2 2  0
  0, 0 w x
x 2 y 2
and the following boundary conditions:

 0, for x  0,
x
(due to the absence of the current flow jx = –/x through the wire’s surface);

 0, for y  0 and w  x,
y
(due to the absence of the current flow jy = –/y through the crack). Also, due to the evident mirror
symmetry of our problem with respect to the crack’s plane, we may take (x, –y) = –(x, y), giving
  0, for y  0 and 0  x  w.
Now let us compare these conditions with the definition of the elliptic coordinates (which were
already encountered in Sec. 2.4 of the lecture notes) with their scale parameter equal to w:104
 x  w cosh  cos ,  
x  iy  w cosh   i , i.e.  with      ,      .
 y  w sinh  sin , 2 2

(In these coordinates, the lines of constant  are ellipses, while those of constant  are hyperbolas, with
the focus at the point {x = w, y = 0} – see the figure above.) The comparison shows that all above
boundary conditions are satisfied by  being a function of just one variable, .105 Since w = w cosh z is

104 For our current purposes, this variable range, which allows  to change sign, is more convenient than the
similarly legitimate (and more common) choice 0    , 0    .
105 This immediately means that the hyperbolas  = const, orthogonal to the equipotential lines  = const at each
point (again, see the figure above), coincide with the current flow lines.

Problems with Solutions Page 142

Essential Graduate Physics EM: Classical Electrodynamics

an analytic function, the Laplace equation in the curvilinear orthogonal coordinates  and  has the same
simple form as in the Cartesian coordinates x and y – see the discussion in Sec. 2.4. Plugging  = ()
into this equation, we immediately see that it this function has to be linear, so with our problem’s
symmetry, we may take simply
  C . (*)
For relatively large distances from the origin, i.e. at    >> 1, the 2D radius

  x 2  y 2   
1/ 2 1/ 2
 w sinh 2   cos 2 

tends to w  sinh   (w/2)e  , so     ln(2/w) and   sgn(y) ln(2/w), i.e.

2 C ρ
  C sgn  y  ln , j     sgn  y n  , for  1.
w  w
Since this formula may be used only if  is still much smaller than the wire’s radius R (to keep our 2D
Cartesian approximation legitimate), this radial current density, multiplied by the arc (/2) of its spread
on the [x, y] plane, has to be equal to the linear density J of the current flowing along the wall. This
requirement gives106
2J 2J 2 ρ
C , i.e.    sgn  y  ln , for  1.
  w w
Hence the voltage drop across the crack region is
4J 2  max
V    min     max   ln ,
 w
where max >> w is the cutoff distance of our 2D analysis, which is of the order of the wire’s radius R.
Taking max = cR, where c ~ 1, for the crack’s contribution to the wire’s resistance, we get
V V 2 2cR 2  2R 
  2 ln  2  ln  ln c  .
I 2RJ  R w  R  w 
(Since R >> w, the choice of the constant c ~ 1 has a very weak effect on the result.)
Perhaps the most striking feature of this result is its very weak (logarithmic) dependence on the
width w; this weakness is due to the effectively 2D geometry of this problem – to be compared with the
essentially 3D geometry of the previous problem.

Problem 4.6. Calculate the effective (average) conductivity ef of 

a medium with many empty spherical cavities of radius R, carved at
random positions in a uniform Ohmic conductor (see the figure on the R
right), in the limit of low density n << R–3 of the cavities.
Hint: You may like to use the analogy with an electric-dipole
medium – see, e.g. Sec. 3.2 of the lecture notes.

106An alternative way to obtain this result is to calculate J from the same Eq. (*) by integrating jy = –/y over
the interval 0  x  w at y  0 (i.e.   0), where x  w cos and y  w sin.

Problems with Solutions Page 143

Essential Graduate Physics EM: Classical Electrodynamics

Solution: As was calculated in Sec. 4.3 of the lecture notes, a single spherical cavity cut in a
conductor with a uniform current flow of density j0 creates outside it an additional dipole electric field
described by the second term in Eq. (4.26)
j R3
 d   0 2 cos  .
 2r
According to Eq. (3.7), this field corresponds to the following electric dipole moment:
R3
p  4 0 j0 .
2
If the concentration of the cavities is low (nR3 << 1), their dipole fields do not affect each other and
hence add up independently to form the following average electric polarization:
nR 3
P  np  4 0 j0 .
2
Our analysis of electrically polarized media in Sec. 3.2 has shown that the polarization (divided
by 0) is effectively subtracted from the field created by external charges. In particular, Eq. (3.33) may
be rewritten as
P D
E  E0  , with E 0  , (*)
0 0
where E is the macroscopic electric field, and the vector D is defined by the external charge density 
via Eq. (3.32). Since the Sec. 3.2 arguments concerning the effect of polarization did not use any
assumptions about the dipole media’s nature, Eq. (*) may be used for conductors as well. In this case, E0
is the field created by an external e.m.f., which would be equal to E at P = 0, i.e. E0 = j0/. This means
that in our case, the average field is
P j0 nR 3 j
E  E0    4 j0  0 ,
0  2  ef
where ef is the effective conductance we were looking for:

 ef    (1  2nR 3 )   .
1  2nR 3

The last (approximate) equality is valid because our analysis is limited to the small cavity
concentrations, nR3 << 1, when the correction to  is relatively small. Of course, the very fact of the
conductivity reduction, i.e. the negative sign of the difference   ef –  due to the cavities and the
proportionality of the small ratio / to the dimensionless product nR3 could be predicted from
handwaving arguments, so the main result of our calculation is the particular (and rather large)
numerical coefficient 2 before the product.

Problem 4.7. In two separate experiments, a narrow gap, possibly of irregular width, between
two close, perfectly conducting electrodes is filled with some material: in the first case, a uniform linear
dielectric with an electric permittivity , and in the second case, a uniform conducting material with an
Ohmic conductivity . Neglecting the fringe effects, calculate the relation between the mutual
capacitance C between the electrodes (in the first case) and the dc resistance R between them (in the
second case).

Problems with Solutions Page 144

Essential Graduate Physics EM: Classical Electrodynamics

Solution: According to Eqs. (3.53) and (4.14), the electrostatic potential distribution (r) inside
both materials satisfies the same (Laplace) differential equation. The boundary conditions on the
electrode surfaces ((r1) = 1 = const, and (r2) = 2 = const) are also the same. (If fringe effects are
negligible, the boundary conditions on the lateral surfaces of the materials are not important.) Thus the
potential’s distribution (r) in both cases is the same, and defines, in the first case, the electrode surface
charge density (3.54):

σ s   ,
n
and in the second case, the current density (4.8) on the surfaces:

jn   .
n
As a result, the ratio s/jn, and hence the ratio of the full charge Q to the full dc current I, and hence the
ratio of the mutual capacitance C  Q/V to the dc conductance G  1/R  I/V, all are independent of the
exact potential distribution (and hence of the electrode shape):
C Q σs 
  
G I jn 
According to this formula, the product RC ( C/G) depends only on one material parameter ratio:
  0
RC   .
 
Note that both these effects (of the surface charge accumulation and Ohmic conductivity) may
coexist in the same material, and in this case, the RC product is equal to its relaxation time constant r –
see Eq. (4.10) and its discussion.
I
Problem 4.8. Calculate the voltage V across a uniform, wide
resistive slab of thickness t, at distance  from the points of the t  
V ?
injection/pickup of the dc current I passed across the slab – see the
figure on the right.
Solution: Following the discussion in Sec. 4.3 of the lecture I
notes, both the Laplace equation for the electric potential  inside the
conductor (related to the dc current density as j = E = –) and the boundary condition at its surface
outside of the injection/pickup points (jn = –/n = 0) may be satisfied z
by complementing the actual slab with an infinite stack of imaginary slabs,
with a periodic pattern of mirror images of the injection/pickup points – t
see the figure on the right, where the points’ colors code the current sign. 
0
With the coordinate frame selected as shown in this figure, the 
points of current injection (shown red) are located at zk = 2kt, while points slab
t
of current pickup (shown blue) are at zk’ = (2k – 1)t, where k are integers. 
Since the current’s magnitude at each injection/ejection point is 2I (as it
follows, e.g., from Fig. 4.8 of the lecture notes, with d  0), it creates, in  2t
the slab stack, the following spherically symmetric distributions of the

Problems with Solutions Page 145

Essential Graduate Physics EM: Classical Electrodynamics

current density and the electric field:

(1) k 2 I jk (1) k I
j r 

k


r  rk , E r  
k


 r  rk  ,
4 r  rk
3
 2 r  rk
3

and hence the following spherically symmetric distribution of the potential:

(1) k I (1) k I
k ( z)   .
2 r  rk 2 (kt  z ) 2   2 
1/ 2

Summing up these contributions for all k, for the top point of voltage pick-up (z = 0), we get
I  (1) k
   .
 2 k   k t  2
2 2 1/2

From the last figure above, it is evident that in this symmetric situation, the potential at the lower pick-
up point is equal and opposite, – = –+, so the voltage V is
I  (1) k
V      2  .
    k  k 2 t 2  ρ 2 1/2
This result is shown (on the most 100
appropriate log-log scale) by the solid red line in I
the figure on the right. At close distances, /t  0, 
the sum is dominated by the term with k = 0, so the
voltage does not depend on t and increases as
I/, reflecting the potential’s divergence near 10

the current’s injection and pickup points. (This V

divergence would disappear at an account for a
non-zero size of the electrodes.) I / t 3/ 2
2 I   
  exp  
In the opposite limit /t >> 1, the voltage  t 
1
 
1/ 2
 t 
decreases with distance as –1/2exp{–/t}. Note
that this law may be readily conjectured from even
the first step of the variable separation approach to
this problem – which is left for the reader’s 0.1
0.01 0.1 1
additional exercise. Indeed, since  has to be a 2t-  /t
periodic function of z, its leading spatial harmonic is proportional to exp{ik1z} with k1 = /t. Hence, in
order to satisfy the Laplace equation, its dependence on  >> t (when this z-harmonic dominates) is
proportional to the Bessel function K0(k1)  –1/2exp{–k1}, with the same k1.

Problem 4.9. Calculate the distribution of the dc current’s y   sin 

density in a thin, round, uniform resistive disk, if the current is R
inserted into some point at the disk’s rim, and picked up in the I I I'  I
center.
Solution: The image method enables us to solve 0 d d' x   cos 
analytically even a more general problem when the current I is
inserted into an arbitrary point of the disk, at distance d from the

Problems with Solutions Page 146

Essential Graduate Physics EM: Classical Electrodynamics

center – see the figure on the right. Indeed, following the discussion at the end of Sec. 4.3 of the lecture
notes, let us look for the solution of the problem in the form of the current in an unlimited thin layer
made of the same material as the disk, with an additional current I’ inserted into a point at a certain
distance d’ from the center, on the same radius. According to Eq. (4.14) of the lecture notes, the
distribution of the electric potential in the disk (i.e. at  < R), outside of the current injection and pickup
points, has to satisfy the same 2D Laplace equation as in 2D problems of electrostatics – for example,
Problem 2.46. Guided by the model solution of that problem, we may guess that the proper location of
the additional current’s injection is in the inversion point with d’ = R2/d and that the image current I’
exactly equals the actual injection current I.
Taking into account that, according to the solution of Problem 2.45, the Green’s function of the
2D Poisson equation for the free space is proportional to ln   – ’ , this solution has the form

 ρ   C 0 ln   C d ln  2  d 2  2 d cos    
1/ 2 1/ 2
 C d' ln  2  d' 2  2 d' cos  , (*)

where the argument of each log function is the distance of the observation point, with the polar
coordinates {, }, from one of the current injection/pickup points – see the figure above. The constants
C should be selected so that the total current flowing out of each contact equals the corresponding
insertion/pickup current. For example, the first term, describing the current pickup point from the center
of the disk, gives the linear current density
 tC 0
J  tj  tE  t  t n   n ,
 
where t is the disk’s thickness, so the above requirement,

J n dr  2J    I ,
around  0
yields
I
C0  . (**)
2 t
The constants in two other terms of Eq. (*) are evidently similar by magnitude but opposite by sign: Cd =
Cd’ = –C0, because they describe the insertion of similar currents I.
With these constants, Eq. (*) satisfies the 2D Laplace equation for our problem and the boundary
conditions at the current injection/pickup points, and the last step is to prove that it also satisfies the
boundary condition at the disk’s rim:

J n   R  t   R  0 .

The proof is straightforward; indeed, according to Eq. (*), in an arbitrary point of the disk,
1   
 
C 0   
1
 1
 
ln   ln  2  d 2  2 d cos   ln  2  d' 2  2 d' cos 
2 2

1/ 2 

1   d cos    d' cos 
  2  2 .
   d  2 d cos    d' 2  2 d' cos 
2

Now taking  = R, and using the inversion rule d’ = R2/d, we get

Problems with Solutions Page 147

Essential Graduate Physics EM: Classical Electrodynamics

1  1 R  d cos  R  ( R 2 / d ) cos 
 R   2 
C 0  R R  d 2  2 Rd cos  R 2  ( R 2 / d ) 2  2 R( R 2 / d ) cos 
1 R 2  Rd cos  d 2  Rd cos  
 1     0.
R  R  d  2 Rd cos  d  R  2 Rd cos  
2 2 2 2

So, Eq. (*) with proper constants (again, Cd = Cd’ = –C0 = –I/2t) is indeed the correct electric
potential’s distribution. Now we may simplify it for our case of current injection at the disk’s rim, d = d’
= R, getting a very simple expression:

 ρ  
I
2t
 
ln   ln  2  R 2  2 R cos  
I

ln 2
2t   R  2 R cos 
2
. (***)

From here, the current density components are

 I  2  R2
J   t   ,
 2   2  R 2  2 R cos  
1  I 2 R sin 
J   t  .
  2   R 2  2 R cos 
2

The figure on the right shows the color-coded contour

plot of the distribution (**), with the boundaries between the
different colors showing the equipotential surfaces, so the
current density vectors are normal to them. The plot shows
that the current flows from the injection point on the rim
mostly straight to the pickup point in the center.

I I
Problem 4.10. DC current is passed between two point
electrodes connected to a wide, thin, uniform resistive sheet – see the
figure on the right. Use the solution of the previous problem to prove,
without much new calculation, that cutting a round hole in the sheet
(outside of the current injection/extraction points) doubles the voltage V
between any two points on its border.
Solution: As was discussed in the model solution of the
previous problem, the electric potential due to the point injection of current I into a uniform thin sheet
(with its extraction at infinity) equals Cln+, where + is the distance of the observation point from the
injection point, and C  I. The extraction of the same current, at a different point, may be described by a
similar expression with the equal but opposite current and the distance – from the extraction point. So,
the full potential distribution in the sheet, in the absence of the hole, is107
 without th e hole  C ln    ln   . (*)

107At this point, we do not need the assumption of the current pickup/injection at infinity anymore,
because at large distances, the contributions of the real injection and extraction of the same current
cancel each other.

Problems with Solutions Page 148

Essential Graduate Physics EM: Classical Electrodynamics

From the same solution, the boundary condition jn = 0 at a circular border between the sheet and
an insulator (like that inside the hole we are discussing) is satisfied by addition, to any initial potential of
the type Cln, of the potential Cln’ where ’ is the distance from the field source point inverted in a
circle, plus a constant. (This constant depends only on the positions of the source and the circle, but not
on that of the potential’s measurement point, so it does not contribute to the voltage between any two
points.) So, the total potential, in our case, is
 with the hole  C ln    ln  '   ln    ln  '   const . (**)

Next, as was discussed in Sec. 4.3 of the lecture notes, the distribution of  in such a uniform
resistive sheet, outside of the current injection/extraction points, is governed by the same 2D Laplace
equation as in cylindrical electrostatic systems. But from the (easy!) calculation in the solution of
Problem 2.45, we know that adding, to the potential Cln, the image potential –Cln’ turns the net
potential C(ln – ln’) into a constant (also not contributing to the voltage), at any point of the circle’s
border. Hence, on that border (only!) Cln ’ = Cln + const. Plugging this equality, written for both field
source points , into Eq. (**), and comparing the result with Eq. (*), we get
 with the hole  2 without th e hole  const ' , (***)

where the new constant is also the same for all points on the circle’s border. Hence, the hole’s cut
doubles the potential difference (i.e. the voltage) between any two points of its border.
Finally, note that this equality remains valid for any areal distribution of the injected and
extracted dc currents, provided that they are all outside of the hole’s interior. Indeed, any such
distribution may be represented as a sum of point pairs of elementary source/drain currents dI flowing
into elementary areas d2. Our result (***) is valid for each such pair, and hence (due to the linear
superposition principle, applicable here due to the Ohm law’s linearity) for the whole distribution as
well.

Problem 4.11. The rim of a hemispherical thin shell, of radius R I

and thickness t << R, made of a uniform Ohmic conductor, is connected
to a plane ground electrode. Calculate the distribution of the ' R
electrostatic potential created in the shell by a dc current I injected into
it through a small-size electrode located at a polar angle ’ < /2 from
the symmetry axis – see the figure on the right.
Hint: You may like to use the variable substitution   tan(/2) to map the hemisphere onto a
unit circle.
Solution: Due to the small thickness of the shell, the distribution of the electrostatic potential in it
(outside of the current injection point) obeys the 2D Laplace equation108
     2 
sin   sin    0, for 0    .
     2 2
With the substitution suggested in the Hint, it becomes

108 See, e.g., MA Eq. (10.9) with /r = 0.

Problems with Solutions Page 149

Essential Graduate Physics EM: Classical Electrodynamics

     2
    0. (*)
     2
But this is exactly the Laplace equation in the polar coordinates.109 Hence, on the planar circular area of
a unit radius: 0    tan(/2)  1, onto which the suggested substitution maps the initial hemisphere
with 0    /2, the scalar potential  obeys the Laplace equation as well.
In the initial geometry, the ground electrode attached to the hemisphere’s rim keeps it at a
constant potential, which we may take for zero. Then, since this rim is mapped onto the circle’s rim,
with  = 1, Eq. (*) should be solved with the boundary condition =1 =0. But the Green’s function of
this boundary problem, describing a delta-functional source of the electric field, was calculated in
Problem 2.45:
 ρ  ρ" 
G ρ, ρ'   2 ln ' ,
 (**)
 ρ  ρ' 
where  is the 2D radius vector of the observation point, ’  tan(’/2) is
such vector of the source point, while the vector ” with the length ” = ρ
1/’  1 is the result of the source point’s inversion in the circle with a 
unit radius – see the figure on the right, in which the source point’s
0 ρ' 1 ρ"
azimuthal angle is taken for zero. Hence, by using the well-known
geometric expression for the length of the triangle’s side opposite to
angle , and then returning to the original variable , we may write the
solution of our problem as
 
  c ln  ' 2   1 / '   2 / '  cos  
2
 ρ  ρ" 2 2
  c ln ' 2
 
 ρ  ρ'
2    2  ' 2  2 ' cos  
  (***)
  '  1  2 ρρ' cos  
2 2
 tan  / 2  tan θ' / 2  1  2 tan  / 2 tan θ' / 2  cos  
2 2
 c ln  2   c ln  .
   '  2 ' cos    tan  / 2  tan θ' / 2  2 tan  / 2  tan θ' / 2 cos  
2 2 2

What remains is to express the constant c via the injected current I. This may be done, for
example,110 by noticing that the potential’s distribution would not change if the current is injected not
into one point of the hemisphere (this is physically impossible anyway), but into an infinitesimal area
that is mapped onto a similarly small circle with the center at the point ’.111 Hence we may calculate c
by comparing the limit of the first form of Eq. (***) at   ’,


  c ln ρ  ρ'
2
 c ln' 2
ρ'  ρ"
2
  2c ln ρ  ρ'  const ,
with Eqs. (*)-(**) of the solution of Problem 9: near the point of injection of current I into a thin film,

109 See, e.g., MA Eq. (10.3) with /z = 0.

110Another way to get the same result (recommended to the reader as an additional exercise and as a
sanity check) is to calculate the linear density of the current flowing through the hemisphere’s rim, J = jt
= tE = –(t/R)(/) at  = /2, from Eq. (***), integrate the result over the length 2R of the rim,
and then require the integral to equal I.
111The fact that the corresponding area on the initial hemisphere is an ellipse rather than a circle and is
proportional to R2 is irrelevant here, due to the logarithmic character of the potential’s distribution.

Problems with Solutions Page 150

Essential Graduate Physics EM: Classical Electrodynamics

I
 ln  ρ  ρ'   const,
2 t
immediately giving c = I/4t.

Problem 4.12. A rectangle of area lw is cut from a uniform I

resistive sheet of thickness t << l, w. Use two different approaches to
 ,t V ?
calculate the voltage V between its two adjacent corners, induced by the dc l
current I that is passed between the two other corners – see the figure on
I
the right. w
Hint: Besides the charge/current image method, you may like to
consider using the variable separation method, with due respect to the current injection/extraction points.
Solution: Extending the solution of Problem 9, we see
that the real situation may be replaced with an infinite
resistive sheet with a rectangular mesh of current
injection/extraction points, shown in the figure on the right. l
Here each red dot means an injection of current 4I, while a
actual
blue point, an extraction of a similar current. (The factor 4 is l sheet V
due to the merger of 4 equal currents, due to the same-sign
w w w
reflection in both sides of the square – cf. Fig. 4.8 of the l
lecture notes.) Indeed, as it is clear from the figure, the net
current, which is the vector sum of the currents radially
spreading from each injection point, does not cross any
border of the actual sheet (shown by hatching), thus satisfying
the boundary conditions. In our 2D case, each of these currents creates the following current density:
I k (ρ  ρ k ) 4 I (ρ  ρ k )
j k (ρ)   ,
2 t ρ  ρ k 2
2 t ρ  ρ k 2
and hence the following electric field:
j k (ρ) 2 I (ρ  ρ k )
E k (ρ)   ,
  t ρ  ρ k 2
which may be represented as the (minus) gradient of the following electrostatic potential:

 k ρ   
2I
 t
ln ρ  ρ k  const  
I
 t
ln ρ  ρ k  2
 const.
As usual, the constant in the potential is unimportant; indeed, it drops out from the voltage V, i.e.
the potential difference between the two points we are interested in, at the corners of the sheet – see the
figure above again. As the figure shows, for these points,
ρ  ρk
2

 w 2 k   l 2 k'  ,
2 2

where k and k’ are the integers listing, respectively, the “horizontal” and “vertical” distances, in the
mesh cell units, between the corner of interest and the current injection/extraction points. Now using the
linear superposition principle to sum up the contributions of all the images, we get

Problems with Solutions Page 151

Essential Graduate Physics EM: Classical Electrodynamics

2I 
(2k  1) 2 r 2  (2k'  1) 2
V 
 t

k , k ' 
ln
(2k  1) 2 r 2  (2k' ) 2
, (*)

where the ratio r  w/l is the “form factor” of our rectangle.

Unfortunately, for large values of r, the double series in Eq. (*) is very inconvenient for practical
calculations, because it converges very slowly, requiring a careful balance of the sum truncation limits at
even very large numbers of k and k’. (Such poor convergence of the sums over multiple charge/current
images is typical for 2D geometries.)
In order to improve the situation, we may use an alternative approach, for example, the variable
separation method. (As we know from Sec. 2.5, for our current rectangular geometry it does not require
any special functions.) The only new112 issue we should be careful about is the description of the
currents’ injection and extraction at the corners of our rectangular sheet. This may be done, for example,
by artificially moving these points by an infinitesimal distance along y
one of the borders – the mathematical trick that obviously does not
change the physical current distribution. Which of the borders is  l
preferable, depends on our choice of the basic functions. 2
For example, in the coordinate frame shown in the figure on the
0
right, let us first use naturally periodic trigonometric functions along w x
the x-axis, by taking the partial solution of the 2D Laplace equation
l
 2  2 l l 
  0, for 0  x  w,   y   , 2
x 2 y 2 2 2
in the following form – cf., e.g., Eq. (2.95) of the lecture notes:113
 n  c n cos k n x  s n sin k n x  sinh k n y, for k n  0.
In this case, it is beneficial to shift the current injection/extraction points slightly along the “horizontal”
(x-) sides of the rectangle, keeping the boundary conditions on its “vertical” (y-) sides purely
homogeneous:

j x x 0, w  0, and hence x 0, w  0
x
These two conditions immediately give sn = 0, kn = n/w, so the full solution takes the form

nx ny  
n nx nl
  c0 y   c n cos sinh , giving y l / 2  c0   c n cos cosh ,
n 1 w w y n 1 w w 2w
where the first terms describe an x-independent partial solution of our Laplace equation. Now we may
represent the injection of current I into the point {x = , y = +l/2},114 with 0 <  << l, by the requirement

112 Indeed, in Chapters 2 and 3 the variable separation method was discussed for problems described by the
Laplace equation, while our current problem is formally described by the Poisson equation, if with a delta-
functional right-hand side.
113 Another y-function satisfying the Laplace equation, cosh k y, obviously does not satisfy our problem’s
n
symmetry – or rather asymmetry: (x,–y) = –(x, y).
114 The extraction of the current from the mirror point at the border y = –l/2 is already taken care of by the
assumed y-asymmetry of our solution.

Problems with Solutions Page 152

Essential Graduate Physics EM: Classical Electrodynamics

I  I
jy y l / 2     x   , and hence y l/ 2   x    .
t y t
Now employing the orthogonality of the functions cos(nx/w) in the usual way, in the limit   0, we
get
I I 2
c0  ; cn  , for n  0,
tw tw n cosh nl / 2w
so the final solution is
I y 2  1 nx sinh ny / w 
    cos . (**)
t  w  n 1 n w cosh nl / 2w 
This form of the result is very convenient for small and  l
moderate values of the ratio r  w/l; for example, the figure on 2
the right shows the equipotential lines for the case of the square
(r = 1). The lines are naturally concentric near the left-side
corners, i.e. the current injection and extraction points. Note also
that the lines are much denser on the side connecting these
corners, while only a small part of them reaches the opposite 0
w
side of the rectangle. This means that the voltage between the
corners of that side (which is the main subject of this problem)
should be rather small – much smaller than the value V0 =
(I/t)(l/w) given by the first term of Eq. (**). (That value is
reached only in the limit r  w/l  0 when the current is l
virtually uniformly distributed across the shorter side of the  2
rectangle.) This conclusion is quantified by Eq. (**), which
yields
I  1 4   1 n 
n
V   x 0   x 0     tanh  . (***)
y  l / 2 y l / 2 t  r  n 1 n 2r 

The single sum in this result converges better than the double sum in Eq. (*),115 in particular giving the
proper limit V  V0  (I/t)/r at r  0, but still rather poorly for large values of r. Moreover, even
getting an asymptotic expression for V in the limit r   from Eq. (***) is not straightforward.
This fact justifies using one more approach to this problem: also the variable separation, but with
exponential/hyperbolic rather than trigonometric functions along the x-axis:116

   c n coshk n  x  w sin k n y,
n 1

where the shift of the x-argument of the cosh function by w immediately enforces the proper boundary
condition at the current-free side of the rectangle:

j x x  w  0, and hence xw  0 .
x

115 For the particular case r = 1, illustrated by the last figure above, Eq. (***) yields V  0.221(I/t).
116 Note that in this case, adding a special partial solution for kn = 0 is unnecessary, because the term proportional
to y may be described by its extension over the functions sin kny with n  0.

Problems with Solutions Page 153

Essential Graduate Physics EM: Classical Electrodynamics

At this approach, we better make the boundary conditions on the x-sides homogeneous as well:

j y y   l / 2  0, and hence yl / 2  0 ,
y
immediately getting the following eigenvalue spectrum: cos(knl/2) = 0, i.e. kn = (2n – 1)l, so

 2n  1 x  w  2n  1 y
   c n cosh sin ,
n 1 l l
  
 2n  1w  2n  1 y
x 0   c 2n  1sinh
n sin .
x l n 1 l l
The price to pay for such simplicity is the necessary shift of the current injection/extraction
points from the corners to close points y+ = l/2 –  and y– = –l/2 + , with  << l, so the boundary
condition becomes
I  l   I  l 
j x x 0    y    , i.e. x 0     y    , for y  0 .
t  2  x t  2 
(Again, the boundary condition for y  0 has already been taken care of by the properly asymmetric
form of the potential as a function of y.) Now repeating the standard procedure of finding the
coefficients cn, in the limit   0, we get the following final solution:
I 4 
 1n cosh 2n  1 x  w / l   2n  1 y

t 
 2n  1
n 1 sinh 2n  1w / l 
sin
l
,

giving the following expression for the sought-for voltage:


I 8 1
V  x 0
y  l / 2
 x 0
y l / 2

t 
 2n  1sinh 2n  1 r  .
n 1

Though these expressions much differ from Eqs. (**) and (***), they give exactly the same results, in
addition converging much faster for larger values of
r. Moreover, the last formula makes it elementary to 100
obtain the following asymptotic expression for the
limit r  , when the first term dominates the sum:
I 16
V exp r.
t  10
The figure on the right summarizes our results for
the function V(r), showing its plot (red line) and two V
1/ r
asymptotes (dashed lines). It is interesting that the I / t
crossover between these two limits takes place at a 16
1 exp r 
relatively small ratio r ~ 0.3. 
Finally, note that the calculated voltage is
always directly proportional to the so-called sheet
resistance R  1/t. This notion is very popular in
experimental and engineering practice because it 0.1
0 0.2 0.4 0.6 0.8
characterizes the average lateral conductive
r  w/l

Problems with Solutions Page 154

Essential Graduate Physics EM: Classical Electrodynamics

properties of a thin sheet or a thin film, even if it is nonuniform at the distances of the order of t. The
usual notation of this constant (alternatively called the “resistance per square”) is due to the fact that
according to Eq. (4.21), R is the resistance of a square of any size, cut of a thin resistive sheet, provided
that the current is uniformly distributed in it – say, flows between two electrodes contacting opposite
sides of the square.

Problem 4.13.* The simplest reasonable model of a vacuum diode consists of two parallel planar
metallic electrodes of area A, separated by a gap of thickness d << A1/2: a “cathode” that emits electrons
into the gap, and an “anode” that absorbs the electrons arriving from the gap at its surface. Calculate the
dc I-V curve of the diode, i.e. the relation between the average current I flowing between the electrodes
and the dc voltage V applied between them, using the following simplifying assumptions:
(i) due to the effect of the negative space charge of the emitted electrons, the current I is much
lower than the emission ability of the cathode,
(ii) the initial velocity of the emitted electrons is negligible, and
(iii) the direct Coulomb interaction of electrons (besides the space charge effect) is negligible.
Solution: Due to the system’s symmetry, the average current is uniformly distributed over the
electrode area, and directed along the axis (say, x) normal to their surfaces, so we may integrate the
relation j = v, where  is the space charge density, over the area to calculate the total current:117
I  Aj x   A  x v x (x) . (*)
Due to the same symmetry, the electrostatic potential  also depends only on the same coordinate:  =
(x). Since the direct electron-electron interaction is negligible, the total energy of each electron,
consisting of its electrostatic energy q(x) = –e(x) and the kinetic energy mev2/2, is conserved during its
motion from the cathode to the anode, i.e. does not depend on x. Since the initial velocity of electrons is
also negligible, v = nxvx everywhere in the gap, and taking the cathode’s electrostatic potential for zero,
we may write this energy conservation law as
me v x2  x 
 e x   0 . (**)
2
Now we may use Eqs. (*) and (**) to eliminate vx, and hence express the local charge density as
a function of the local value of the electrostatic potential: (x) = –I/Avx(x) = –I/A[2e/me(x)]1/2. Plugging
this expression into the Poisson equation (1.41), with 2 = d2/dx2 for our 1D geometry, we get the
following equation for the distribution of the electrostatic potential along the x-axis:
1/ 2
d 2 C I  me 
 1/ 2 , with C    .
dx 2
  0 A  2e 
For the first integration of the equation, we may use the very common118 trick of multiplying and
dividing the second derivative by d:

117 The minus sign in this expression reflects the usual convention to take the anode on the right of the cathode, so
the electrons, with their negative charges q = –e, move from left to right (vx > 0), and their current flows to the
left. At this notation, I  0, while (x) is negative.
118 For example, this trick is used in classical mechanics for the simplest derivation of the work-energy principle –
see, e.g., CM Eq. (1.20).

Problems with Solutions Page 155

Essential Graduate Physics EM: Classical Electrodynamics

d 2 d  d  d  d  d 1 d  d  2 
         ,
dx 2 dx  dx  d  dx  dx 2 d  dx  
so the equation takes the form
 d  2  d
d     2C 1/ 2 .
 dx   
Now we may integrate both parts of this relation, getting
2
 d  d
   2C  1 / 2  4C  const .
1/ 2
(***)
 dx  
Now comes the (only :-) nontrivial point of the solution. Per assumption (i), if the electric field E
= –d/dx on the cathode’s surface was substantially negative (with the force F = –eE, acting on each
electron, directed towards the anode), it would extract a much larger current than what we are
calculating. On the other hand, if the field was substantially positive, it would suppress the emission
altogether, chasing the emitted electrons back to the cathode. Hence E has to be at the very emission
threshold, and a good approximation may be done by taking E = 0 on the cathode’s surface. In this
approximation, the integration constant in Eq. (***) equals zero. Extracting the square root of both parts
of the resulting relation (with the physically justified positive sign),
d d
 2C 1 / 2 1 / 4 , i.e.  2C 1 / 2 dx ,
dx  1/ 4
we may readily integrate its last form, taking into account the second boundary condition, on the anode’s
119
surface: (d) = V, where the cathode surface’s position is taken for x = 0. The integration yields
d
V d
4

0
1/ 4
 V 3 / 4  2C 1 / 2  dx  2C 1 / 2 d .
3 0

After squaring both sides of this relation and using the definition of the constant C:
1/ 2
16 3 / 2 I  me 
V  4Cd 2  4   d 2,
9  0 A  2e 
we obtain the requested dc I-V curve:
1/ 2
4  2e  A 3/ 2
I   0   V .
9  m e  d2
This is the famous Child-Langmuir equation (frequently called the “V3/2 law”),120 one of the
theoretical foundations of all vacuum electronics. Please note its substantial difference from Ohm’s law
I = V/R; physically, this difference is due to the effect of the negative space charge of the electrons,
which suppresses the electric field on the cathode’s surface to E  0 – and hence the current to the values

119 The same integration, but with an arbitrary upper limit (on the interval 0  x  d) yields the following spatial
distribution of the electrostatic potential:   x4/3, so the space charge distribution is   –1/2  x–2/3.
120 It was first derived in 1911 by C. Child for ionic currents, and in 1913 by I. Langmuir for electrons.

Problems with Solutions Page 156

Essential Graduate Physics EM: Classical Electrodynamics

much lower than the cathode’s emission ability. The V3/2 law is very well obeyed by many vacuum
electronic devices unless the applied voltage is so large that it dissolves the electron “cloud”.
Note that very similar space-charge effects may take place in semiconductor structures, but
because of the frequent scattering of electrons by impurities, the charge carriers (electrons and holes) do
not move ballistically (as was assumed in the above calculation) but rather drift through the material,
dissipating their energy at frequent collisions. Because of that, the quantitative result of the
(conceptually, similar) theory is different – see the next problem.

Problem 4.14.* Calculate the space-charge-limited current in a system with the same geometry
as in the previous problem, and using the same assumptions besides that now the emitted charge carriers
do not fly ballistically, but drift in accordance with the Ohm law, with the conductivity given by Eq.
(4.13) of the lecture notes:  = q2n, with a constant mobility .121
Hint: In order to get a realistic result, assume that the medium in which the charge carriers move
has a certain dielectric constant  unrelated to the carriers.
Solution: Let us combine the Ohm law for the current, with the conductivity  taken in the
suggested form,
I  Aj  A  x E x   Aq x E  x  , (*)
where (x) = qn(x) is the space charge density, with the 1D version of the Maxwell equation (3.53):
dE   x 
 . (**)
dx  0
For that, we may, for example, express (x) from Eq. (*) and plug it into Eq. (**), getting the following
differential equation for the electric field’s distribution:
dE I I
E  , i.e. EdE  dx. .
dx Aq 0 Aq 0
Integrating both sides of the last equation, we get
E2 I
 x  const .
2 Aq 0
Now by using exactly the same argumentation as in the previous problem to conclude that the
proper boundary condition at the emission surface (x = 0) is E = 0, we see that the integration constant
equals zero, and taking the appropriate (negative) sign before the square root,122 get
1/ 2
d  2I 
  E   x  . (***)
dx  Aq 0 
Integrating this equation with the boundary conditions  = 0 at x = 0 and  = V at x = d, we obtain

121 As was mentioned in Sec. 4.2 of the lecture notes, the approximation of a constant (in particular, field- and
charge-density-independent) mobility is most suitable for semiconductors.
122 In our system with, say, positive anode voltage V, the electric field should be directed from the anode to the
cathode, i.e. (for our coordinate choice) should be negative.

Problems with Solutions Page 157

Essential Graduate Physics EM: Classical Electrodynamics

1/ 2 d 1/ 2
 2I   2I  2 3/ 2
V    x dx   
1/ 2
d .
 Aq 0  0  Aq 0  3
This is the final result, which is usually represented in the reciprocal form,
9 A
I   0q 3 V 2 ,
8 d
and called the Mott-Gurney law (first derived in 1948). Note that it gives a current dependence on the
applied voltage, I  V2, which significantly differs from both the integral form of the Ohm law, I  V
(valid in semiconductors only at a negligible space charge accumulation) and the Child-Langmuir
formula, I  V3/2 (valid only at the ballistic motion of charge carriers).
As an interesting by-product, we may readily use Eq. (**) and then Eq. (***) with x = d to
calculate the full space charge (per unit area):
1/ 2
Q
d d
dE  2 Id 
    x dx   0  dx   0 E d    0   .
A 0 0
dx  Aq 0 
Now plugging in our final result for the current I, we get
Q 3V 3
  0   C 0V ,
A 2d 2
where C0 = 0/d is the capacitance (also per unit area) of the same device when it is empty of charge
carriers – see Eq. (3.55). This increase of the effective capacitance (by 50%) is natural, because the field
lines starting on the surface charges of the anode, which in an empty device would continue all the way
(d) to the cathode, now end up on the closer spatially-distributed charges.

Problem 4.15. Prove that the distribution of dc currents in a uniform Ohmic conductor with a
given voltage distribution along its surface corresponds to the minimum of the total energy dissipation
rate (“Joule heat”).
Solution: According to Eq. (4.40) of the lecture notes, the total Joule heat power (i.e. the electric
energy dissipation rate) in a uniform Ohmic conductor may be expressed as

P     2 d 3 r . (*)
V

According to Eq. (4.14), the distribution of the scalar potential  in a uniform Ohmic conductor (with 
= const) satisfies the Laplace equation, just as in free space. Hence, its distribution is the same as in a
similar free-space volume V with the same boundary conditions S. But as we know from Chapter 1 (see
the model solution of Problem 1.17), the latter distribution corresponds to the minimum of the
electrostatic energy
 
U  0  E 2 d 3 r  0    d 3 r ,
2

2V 2V
i.e. to the minimum of the same integral as in Eq. (*).

Problems with Solutions Page 158

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 5. Magnetism

Problem 5.1. DC current I flows around a thin wire loop bent into the form of a plane equilateral
triangle with side a. Calculate the magnetic field in the center of the loop.
Solution: According to Eq. (5.18) of the lecture notes, for our y
planar system, the magnetic field dB of each wire element dr’, at an
observation point r in the same plane as the current, is directed
normally to the plane, so all dB add up as scalars. Due to the
system’s symmetry, the total field is six times that induced by one-
half of each side. So, introducing the Cartesian coordinates as shown a a
in the figure on the right, with r = {0, h, 0} and r’ = {x’, 0, 0}, we B
r 
may use Eq. (5.18) to write
I r  r'
0 I a/2
hdx' 0 I a / 2h
d  /6 h
B6
4 0 x' 2
h 
2 3/ 2
6
4h 0  2
1  3/ 2
,
0 r' a/2 x
where   x’/h, while h = (a/2) tan(/6) = a/23 is the height of the triangle – see the figure above.
Working out the last integral,1 we finally get
0 I a / 2h  I 3 3 3 0 I 9 0 I
B6 6 0   .

4h a / 2h   1
2 1/ 2
4h 

2
3 1 

1/ 2
 
4 h 2 a
 
As a sanity check, it is useful to rewrite this result in the form
0 I 2
B , with R ef  h  1.21h ,
2 Ref 3 3
where per Eq. (5.24) of the lecture notes, Ref is the effective radius of the wire loop. Looking at the
figure above again, we see that our result, Ref  1.2h, makes sense.

Problem 5.2. A circular wire loop, carrying a fixed dc current, is

placed inside a similar but larger loop, carrying a fixed current in the same I1
direction – see the figure on the right. Use semi-quantitative arguments to I2
analyze the mechanical stability of the coaxial and coplanar position of the
inner loop with respect to its possible angular, axial, and lateral displacements.
Solution: The stability may be discussed using a semi-quantitative analysis of magnetic forces.
Indeed, since according to the basic Eq. (5.1) of the lecture notes, the currents flowing in the same
directions attract each other, and this force decreases with the distance d between them, the interaction
of the closest parts of two loops plays the most important role.

1 See, e.g., MA Eq. (6.2b).

Problems with Solutions Page 159

Essential Graduate Physics EM: Classical Electrodynamics

Hence, a small axial displacement of the inner ring from dF dF

the outer ring’s plane results in a rise of a net force F dragging it
back – see the top panel of the figure on the right, where colors F
code the current direction – in or out of the plane of the drawing.
Very similarly, an angular displacement of the inner loop dF
causes the elementary forces dF exerted on it to create a
mechanical torque pulling it back to the coplanar position – see
the middle panel of the figure.
dF
The analysis of the lateral displacement is a bit more
subtle. According to Eq. (5.7) of the lecture notes, the attraction dF F dF
of two currents (per unit length) scales approximately as 1/d,
2

where d is the distance between them. At a lateral displacement

d   d 
of the loop by a small distance  << d (see the bottom panel of
the figure on the right), the nearest distances d for one half of the ring decrease by ~, and for the other
half, increase by the same amount. However, the function f(d)  1/d grows faster at the argument’s
decrease than drops at its increase:
1 1 1   2 
f (d  )    1   2  ,
d   d 1   / d  d  d d 
so we may expect the net force to be directed toward the nearest part of the outer ring. Hence, the
coaxial and coplanar position of the inner loop is stable with respect to its axial and angular
displacements,3 but is unstable with respect to its lateral displacements.

Problem 5.3. Two planar, parallel, long, thin conducting strips of width w, I
separated by distance d, carry equal but oppositely directed currents I – see the
figure on the right. Calculate the magnetic field in the plane located in the middle w I
d
between the strips, assuming that the flowing currents are uniformly distributed
across the strip widths.
Solution: Due to the linear superposition principle, we may calculate the total field B of a strip as
a vector sum of elementary fields dB induced by elementary currents dI y ρ  x, y
= (I/w)dw flowing in each elementary segment dw << w, d of the strip’s d
widths, considering it as a thin wire. The magnetic field of such a wire
dI 
was calculated in Sec. 5.1 of the lecture notes (see Eq. (5.20) and its
 dB
derivation), and we need just to rewrite it in a form more convenient for ρ'  x' , y'
our current purposes. Directing the z-axis along the current (see the
figure on the right), we get 0 x

2 As a reminder, Eq. (5.7) is strictly valid only for infinite straight wires, but as we know, for example from Eq.
(5.24), the results for other similar geometries scale almost similarly.
3 Eq. (5.53) of the lecture notes allows easy confirmation of these two conclusions using potential energy
arguments, but I will defer this task until the conclusion of the final discussion of the magnetic field energy (or
rather energies) in the next chapter.

Problems with Solutions Page 160

Essential Graduate Physics EM: Classical Electrodynamics

0 y  0 y  y'
dB x  dB sin   dI  dI ,
2d d 2  x  x' 2   y  y' 2
 x  x  x'
dB y  dB cos    0 dI  0 dI ,
2d d 2  x  x' 2   y  y' 2

where  = {x, y} is the 2D radius vector of the field observation

y
point, while ’ = {x’, y’} is that of the source (wire segment). I
Now we should integrate this expression over both d /2
current-carrying strips. Selecting the reference frame in the
natural way shown in the figure on the right, so the mid-plane in
question corresponds to y = 0, we see that due to the problem’s  w 0 
w x
symmetry, the integral of dBy vanishes, while the “horizontal” 2 2
component of the field is twice that of a single strip: d /2
I
0 x'  w / 2
d /2
B x  2
2  x  x'   d / 2
x'   w / 2
2 2
dI
0

0 I d w / 2 dx' d /w3
 
 w 2  w / 2 x  x'   d / 22
2  0.2

 0 I  1 x  w / 2 x  w/ 2
  tan  tan 1 . Bx  0.4 1
w  d /2 d /2 
 0 I / w
The figure on the right shows the plots  0.6
of Bx as a function of x, for several values of 0.3
the d/w ratio. If d >> w, the field is relatively  0.8 0.1
low, distributed along axis x over a broad
interval x ~ d >> w, and may be well 0.03
1
approximated by the sum of fields of two thin 1  0.5 0 0.5 1
wires. x/w
On the other hand, as the distance between the strips is reduced, the field becomes localized
within the gap between them. (At d/w  0, the so-called “fringe fields” at x > w/2 become negligible.)
Moreover, the field becomes uniform and tends to an x- and d-independent value:
0 I
. Bx  
w
A simple explanation of this important result and its discussion are the subjects of the next
problem.

Problem 5.4. For the system studied in the previous problem, but now only in the limit d << w,
calculate:
(i) the distribution of the magnetic field in space,
(ii) the vector potential of the field,
(iii) the magnetic force (per unit length) exerted on each strip, and

Problems with Solutions Page 161

Essential Graduate Physics EM: Classical Electrodynamics

(iv) the magnetic energy and self-inductance of the loop formed by the strips (per unit length).
Solutions:
y
(i) As was shown in the previous problem, in the limit d/w 
0, the magnetic field is localized in the gap between the strips and is I
uniform. Applying the Ampère law, given by Eq. (5.37) of the lecture B
notes, to the contour shown with the dashed line in the figure on the d
right (which shows the cross-section of the system), we get the same x
result as was obtained in the previous problem’s solution, I
I w
B   Bx   0 , (*)
w
in a much simpler way, and for arbitrary y between –d/2 and +d/2.
(ii) According to Eq. (5.28) of the lecture notes, in our system, the vector potential has to be
directed along the z-axis, i.e. along the current, and be independent of z. From the structure of that
formula, we may also argue that at d << w and well inside the gap, A should not depend on x either
(because the strip edges are not “visible” from those points). Hence we may look for the vector potential
in the form
A  A( y )n z .
Calculating the curl of such a vector,
A
  A  nx ,
y
and requiring it to be equal to the vector B = –Bnx described by Eq. (*), we get
I
A   0 yn z  const.
w
This result could be also obtained by the direct integration of Eq. (5.28) along x’ and z’. (Alternatively,
we can integrate Eq. (5.51) written for each current component dI = Idx’/w, along the x’-axis.)
Note, however, that a uniform magnetic field, like our B = –Bnx, cannot “tell” one transverse
coordinate (say, y) from the other one (z), and remains the same for different distributions of the vector
potential, for example4
I
A   0 zn y  const ,
w
or a linear combination of these two functions. This is one more manifestation of the gauge invariance of
the magnetic field with respect to any transformation A  A +  – see Eq. (5.46).
(iii) As it follows from Eq. (5.1), the total magnetic force exerted on each strip is directed along
the y-axis and corresponds to the strips’ repulsion. To calculate its magnitude, it would be wrong to plug
Eq. (*) into Eq. (5.15), because an elementary current (like a point electric charge) does not exert force
on itself.5 As evident from the system’s symmetry (and as may be readily checked) using the Ampère
law, the field created by a single strip is twice lower:

4Note that this distribution, of course, does not satisfy Eq. (5.28).
5 Of course, different components Jdx of the current in a strip do exert nonvanishing (x-directed) forces on each
other, but their sum over the strip equals zero.

Problems with Solutions Page 162

Essential Graduate Physics EM: Classical Electrodynamics

I
B1   0 ,
2w
so, according to Eq. (5.15), the force magnitude (per unit length)
F 1 I2
  jB1 d 3 r  IB1   0 .
l l strip 2w
This expression may be also represented as an integral, over the strip’s area, of the positive
pressure of the magnetic field:
F 0 I 2 B2
P     u,
lw 2 w 2 2 0
where u = U/Vgap = U/lwd is the magnetic energy’s density – see Eq. (5.57b).
(iv) Since the full magnetic field is uniform inside the gap (with the cross-section area dw) and
vanishes outside of it, the magnetic energy per unit length is just6
U 1 B2 3 B2 d I2 Vgap
  d r dw   0 u .
l l gap 2  0 2 0 w 2 l

From this result and Eq. (5.72), we immediately get the following expression for the loop’s self-
inductance (also per unit length):
L d
 0 .
l w
The same result may be also calculated as the ratio (/l)/I, where  is the magnetic flux in the
gap between the strips:
 1 I
  Bn d 2 r  Bt   0 t .
l l gap w

This simple formula, which shows a clear way toward the reduction of current loop inductances
(bring the counterpart conductors as close to each other as possible!), is very important for applications
in electronics because self-inductances may play a negative role in high-performance integrated circuits,
reducing their maximum speed/bandwidth.
z

Problem 5.5. Calculate the magnetic field distribution near the center d /2
of the system of two similar, plane, round, coaxial wire coils, carrying equal R I
but oppositely directed currents – see the figure on the right. 0 y
x
d /2
I
6 This result poses an additional question for the reader: how can the strip repulsion (i.e. their “desire” to increase
d, and hence the volume Vgap and the product U = uVgap) be compatible with the general trend for any system to
decrease its potential energy? As a reminder, for the electrostatic force exerted on a conductor, this paradox does
not exist because the force direction corresponds to negative pressure – see, e.g., the solution of Problem 2.1.
Thinking about this issue may be a good preparation for the general discussion, in Sec. 6.2 of the lecture notes, of
the relation between the two magnetic energies given by Eqs. (5.53) and (5.54).

Problems with Solutions Page 163

Essential Graduate Physics EM: Classical Electrodynamics

Solution: In order to find the field B(x, y, z) exactly on the system’s axis, we may combine two
versions of Eq. (5.23) of the lecture notes, applied to each of the loops:
 0 I  R2 R2 
B0,0, z     3/ 2  z
n ,

2  R 2  (d / 2  z ) 2  3/ 2
 R 2  (d / 2  z ) 2  
where z is the axis directed along the system’s symmetry axis, with the origin in its center – see the
figure above. Per this formula, at the center of the system (z = 0), the field vanishes, while for small (z
<< R, d) but nonvanishing deviations from that point, it may be found from the linear term of the Taylor
expansion of its right-hand side:

 0 I   R2 R2  3 0 IdR 2
B0,0, z    2   z n  zn z .

2 z  R  (d / 2  z ) 2  3/ 2
 R 2  (d / 2  z ) 2
3/ 2

 z 0
z
2 R 2  ( d / 2) 2   5/ 2

Obviously, this expression may be rewritten as

B z 3 0 IdR 2
 .
z
r 0

2 R 2  ( d / 2) 2  5/ 2

Now, in the Cartesian coordinates, the general Eq. (5.29) reads

B x B y B z
   0.
x y z
Due to the axial symmetry of our system, we may argue that at z = 0, the first two partial derivatives
have to be equal to each other. Hence, the last two relations yield
B x B y 1 B z 3 0 IdR 2
   ,
x
r 0
y
r 0
2 z
r 0

4 R 2  ( d / 2) 2  5/ 2

so in the linear approximation in small x, y, and z we may write

3 0 IdR 2
B  x, y , z    xn  yn y  2 zn z , for r  d , R .

4 R 2  ( d / 2) 2  5/ 2 x

Such a quadrupole magnetic field pattern may provide trapping of a particle with zero net
electric charge but a non-zero total spin (and hence a nonvanishing magnetic moment), within a certain
range of its initial velocity. A brief discussion of magnetic traps, and references to additional literature
on the subject, may be found in Sec. 9.7 of the lecture notes.
z

Problem 5.6. The two-coil system considered in the previous problem d /2

now carries equal and similarly directed currents – see the figure on the right. R I
Calculate what should be the ratio d/R for the second derivative 2Bz/z2 to 0 y
x
equal zero at z = 0.7
d /2
I
7This Helmholtz coils system, producing a highly uniform magnetic field near its center, is broadly used in
physical experiment.

Problems with Solutions Page 164

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Just as in the model solution of the previous problem, by using Eq. (5.23) for the field
of each loop, we get
3 / 2
0 I R2  0 IR 2  2 d  
2

Bz      R   2  z   .
2  R 2
 (d / 2  z ) 2  3/ 2
2  
Note that at an arbitrary ratio d/R, the first derivative
5 / 2
  2  d  
2
B z  0 IR 2  d
z

2
  3 2  z   R   2  z   ,
   
vanishes in the center of the system (at z = 0) – just as one might expect from the system’s symmetry.
Now calculating the second derivative, we get
7 / 2
 2 Bz  0 IR 2  2  d   d
2
  2  d
2
 
2

z 2
z 0 
2
   2    2    2  
R    z   5  z  R    z  z 0
   
7 / 2
 2  d 2  2  d 2 
  0 IR  R  4    R    
2

  2     2  
This expression evidently turns to zero at
2
d 
R 2  4   0, i.e. at R  d .
2
This is exactly the relation used for the Helmholtz coils’ design. Note that due to the system’s
symmetry, the third derivative, 3Bz/z3, vanishes at z = 0 as well (this one for any d/R), so the deviation
of the field from its value in the center,
3/ 2
R2 4 0 I
Bz  0 I   ,
z 0
d R R 2
 (d / 2) 
2 3/ 2
d R
5 R

grows very slowly, as (z/R)4, with the deviation from that point.

Problem 5.7. DC current of a constant density j flows along a round j

cylindrical wire of radius R, with a round cylindrical cavity of radius r cut in it. The r  R
cavity’s axis is parallel to that of the wire but offset from it by a distance d < R – r
d 0
(see the figure on the right). Calculate the magnetic field inside the cavity.
Solution: We may formally represent this current distribution as an algebraic
sum of the following two components:
– a current of the constant density j inside an uncut wire of the given radius R, and
– a current of the density (–j) inside the cavity of the radius r.
Applying the Ampère law to the first (axially-symmetric) component, just as this was done in Sec. 5.2 of
the lecture notes, we get the result expressed by Eqs. (5.38). For this field inside the wire, with j() =
const, the first of those formulas yields.

Problems with Solutions Page 165

Essential Graduate Physics EM: Classical Electrodynamics

0 j
B1 ρ   ρ, for ρ  R,
2
where  is the 2D radius vector of the field measurement point, referred to the center of the wire’s cross-
section. With the account of the azimuthal direction of the field, this result may be represented in the
following vector form:
 j
B 1 ρ   0 n j  ρ, for ρ  R,
2
where nj is the unit vector directed as the current is – in the figure above, into the plane of the drawing.
Now we may carry out an absolutely similar calculation for the second component of the current,
which is also axially symmetric but with the symmetry axis offset from the origin by a vector d (see the
figure above), so the Ampère contour C has to be centered to that point. The result is
0 j
B 2 ρ    n j  ρ  d , for ρ  d  r ,
2
so the net field inside the cavity is
0 j
Bρ   B 1 ρ   B 2 ρ  
n j  d.
2
Rather counter-intuitively, the field is uniform! It is directed normally to the center displacement
vector d, and its magnitude is equal to that of the component field B1 at the center of the cavity.
I
Problem 5.8. Calculate the magnetic field’s distribution along the axis of
a straight solenoid (see Fig. 5.6a of the lecture notes, partly reproduced on the R
right) of a finite length l, and a round cross-section of radius R. Assume that the N l
solenoid has many (N >> 1, l/R) wire turns, uniformly distributed along its I
length.
Solution: Because of the high wire turn density n  N/l >> 1/R, and
hence a virtually circular shape of each turn, the solenoid’s field may be represented as a sum of fields
of N circular current loops. The field induced by a single loop was calculated in Sec. 5.1 of the lecture
notes – see Eq. (5.23). Generalizing that relation to an arbitrary position (z’) of the loop, we get
0 I R2
B1  .
2 R 2
 ( z  z' ) 2  3/ 2

Now we can use the linear superposition principle to find the field induced by the solenoid as a whole:
0 I N /2
R2 0 I N / 2
R2
BN     ,
2 k  N / 2 R 2
 (z  zk ) 2  3/ 2
2 k  N / 2 R 2
 ( z  kl / N ) 2  3/ 2

where the index k numbers wire turns, starting from the middle of the solenoid. For N >> 1, this sum is
well approximated by the corresponding integral
N / 2 l / 2
0 I R2 0 I R2
BN 
2  R
N / 2
2
 ( z  kl / N ) 2  3/ 2
dk 
2
n  R
l / 2
2
 ( z  z' ) 2  3/ 2
dz' .

Problems with Solutions Page 166

Essential Graduate Physics EM: Classical Electrodynamics

This is a table integral,8 giving finally:

 0 In z  z' z' l / 2  0 In  z l/2 z l/2 

BN    2  1/ 2 
.
2 R 2
 ( z  z' ) 2 
1/ 2 z'  l / 2

2  R  ( z  l / 2) 2 1/ 2
 R 2  ( z  l / 2) 2  

The figure on the right shows typical 2

field profiles given by this formula. For points
l / R  10
well inside a long solenoid (l >> z, R), the 1.5
expression in the figure brackets tends to 2, so B 3
N
our result approaches the simple Eq. (5.40): BN 1
= 0In. Note, however, that a close approach to  0 In
1

this value requires rather large values of the l/R 0.5

ratio, i.e. very long solenoids. (In practice, this
handicap may be remedied by using either the 0
 10 5 0 5 10
toroidal solenoids sketched in Fig. 5.6 of the
lecture notes, or the straight solenoid’s filling z/R
with a high- magnetic material – see Sec. 5.6.)
On the other hand, well outside of the solenoid, i.e. for z >> R, l, our result yields
 0 InR 2 l  0 2m N
BN   , where m N  R 2 IN  Nm1 ,
2 z
3
4 z 3
i.e. the dipole field (5.99), with the dipole moment mN, which is N times larger than that (m1 = AI) of a
single current loop – see Eq. (5.97).

Problem 5.9. A thin round disk of radius R, carrying an electric charge of a constant areal density
, rotates about its axis with a constant angular velocity . Calculate:
(i) the magnetic field on the disk’s axis,
(ii) the magnetic moment of the disk,
and relate these results.
Solutions:
(i) The rotating disk may be fairly represented as a set of narrow elementary rings of radius  and
width d << , each carrying a ring current of the magnitude dI = Jd = d and thus creating an
elementary field dB described by Eq. (5.23) of the lecture notes (with the notation replacements B  dB,
I  dI, R  ):
 0 dI 2  0  3 d
dB   ,
2  2  z 2 3/ 2 
2  2  z 2 3/ 2   
where z is the distance of the field observation point from the ring’s plane. Since the fields of all the
elementary rings have the same direction (along the disk’s axis), we may sum up their contributions to
the total field as scalars:

8 See, e.g. MA Eq. (6.2b).

Problems with Solutions Page 167

Essential Graduate Physics EM: Classical Electrodynamics

 R  R
 0 R  3 d  0 z d
B
 0
 dB  2 0  2
z 
2 3/ 2

2 2 
 0   1
3/ 2
,

where   2/z2. The last integral may be readily worked out as

 d   1 d  d d d
 2  1
2 2  2
   1    
1/ 2
  ,
3/ 2
  1 3/ 2
  13/ 2
  11/ 2
  13 / 2   1 1/ 2
  11 / 2
and we finally get
 R
 0 z  2 2 / z 2  2    R 2  2z 2 
B  2 2 1/ 2 
     z . (*)
2 
2   / z  1  
 0
0
 2 R 2  z 2
1 /

2
 
(ii) Very similarly, the magnitude of the magnetic moment m of the system (evidently, also
directed along the symmetry axis z) may be calculated by summing up the elementary ring contributions
given by Eq. (5.97):
dm  AdI   2 dI    3 d ,
and getting
 R R
 R 4
m  dm      d 
3
. (**)
 0 0 4

In order to relate the above results, let us find the limit of Eq. (*) at large distances, z >> R:
 1  R 2 / 2z 2   R2  R 2 3R 4   R4
B   0 z   1    z 
 1  1  2  4   1   0 . (***)
 1  R 2 / z 2 
1/ 2 0 2 3
  2 z  2 z 8z   8 z

On the other hand, for the observation point on the magnetic dipole’s axis (r = zm/m), the general
expression for the magnetic dipole’s field, following from Eq. (5.99) of the lecture notes, is
 0 2m
B .
4 z 3
With the m given by Eq. (**), this expression coincides with that Eq. (***), i.e. our two results
do match.

Problem 5.10. A thin spherical shell of radius R, with charge Q uniformly distributed over its
surface, is rotated about its diameter with a constant angular velocity . Calculate the distribution of the
magnetic field everywhere in space. ω
Solution: As viewed from the lab reference frame, the rotation creates dI

ring-like surface currents with the linear density J = v, where  = Q/4R is the
2

(constant) surface charge density, v = Rsin’ is the linear velocity, and ’ is the R
'
polar angle of the elementary current’s location, measured from the rotation axis
– see the figure on the right, showing a part of the shell’s cross-section. As was
discussed in Sec. 5.4 of the lecture notes, at large distances r >> R, each 0
elementary ring current

Problems with Solutions Page 168

Essential Graduate Physics EM: Classical Electrodynamics

Q Q
dI  JRd '  R sin  ' R d '  sin  'd ' ,
4R 2
4
creates a magnetic field with the same dipole distribution (5.99), with the elementary dipole moment
(5.97):
2 Q 1
dm  A'dI   R sin  '  sin  'd '  QR 2 sin 3  'd ' .
4 4
Due to the similarity of these distributions, the long-distance magnetic field of the whole shell is also
similar, with the magnetic dipole moment m that may be calculated by summing dm of all elementary
currents, i.e. integrating dm over the segment 0  ’   – see the figure above:9

1 1 4 1
m   dm  QR 2  sin 3  'd '  QR 2  QR 2 .
4 0
4 3 3
After having solved so many problems on uniform spheres and thin spherical shells in this
course, the reader may readily guess that the dipole field distribution (5.99),
1 3r r  m   mr 2 m 2n r cos   n  sin 
H  , (*)
4 r 5
4 r3
is valid not only at large distances, but for any r > R, and that the field inside the shell is uniform:
ω
H H 0  H 0 n r cos   n  sin  , for r  R , (**)

where H0 is some constant.
Indeed, as we already know, the fields (*) and (**) do satisfy the stationary Maxwell equations
(5.36) in the current-free regions. Hence, due to the uniqueness of the solutions of linear boundary
problems, it is sufficient to prove that, at a certain choice of the scalar constant H0, these fields also
satisfy the boundary conditions at the shell, i.e. at r = R. These conditions,

H r  R 0  H r  R 0  J, Hr r  R 0  Hr r  R 0  0,

follow from, respectively, Eqs. (5.116) and (5.119) of the lecture notes. For our fields (*) and (**), with
the dipole moment m = QR2/3 and the linear current density J = (Q/4R2)Rsin calculated above, the
conditions take the following form:

 1 1 sin   Q  1 1 2 cos  
 QR 2 3    H 0 sin    R sin  ,  QR 2   H 0 cos    0.
 4 3 R  4R 2  4 3 R3 
One can readily see that if
Q
H0  ,
6R
both boundary conditions are satisfied for any , so we have indeed found the (unique) solution of the
problem.

9 The integral over d’ may be readily worked out using the usual variable replacement   cos’, so that
sin3’d’ = –sin2’d(cos’) = –(1 –  2)d.

Problems with Solutions Page 169

Essential Graduate Physics EM: Classical Electrodynamics

Problem 5.11. A sphere of radius R, made of an insulating material with a uniform electric
charge density , rotates about its diameter with a constant angular velocity . Calculate the magnetic
field distribution inside the sphere and outside it.
Solution: We may represent the sphere as a set of thin spherical shells, each with radius R’  R,
thickness dR’, and charge dQ = dV =  (4R’2dR’). As was discussed in the model solution of the
previous problem, such a shell creates outside it a purely dipole magnetic field, with the spatial
distribution
1 3r r  dm   r 2 dm
dH  ,
4 r5
where dm is the shell’s magnetic dipole moment directed along the axis of rotation, with the magnitude
1 4
dm  dQR' 2  R' 4 dR' ,
3 3
and inside it, a uniform field,
dH  dH 0 n r cos   n  sin  ,
of magnitude
dQ 2
dH 0    R'dR' .
6R' 3
As the result of the addition of these elementary fields, the total field outside the sphere (for r 
R) has the same dipole distribution, with the total dipole moment m whose magnitude m that may be
calculated by the scalar addition of the elementary moments dm:
4 4
R
m   dm    R' 4 dR'  R 5 . (*)
3 0
15
In order to calculate the field inside the rotating sphere, i.e. at distances r < R from its center, we
should sum up the dipole field induced by all thin shells with R’ < r, giving the total dipole moment
4
m  r 5 ,
15
and the uniform magnetic fields created by all thin shells with R’  r, whose magnitude is the scalar sum
of all shell contributions:
R
2
3
1

H 0   dH 0    R'dR'   R 2  r 2 .
3

r

Using an alternative form of the dipole field representation,10

1 2n r cos   n  sin 
H m , (**)
4 r3
these two contributions may be readily summed up, giving

10 See, e.g., the first form of Eq. (3.13) of the lecture notes.

Problems with Solutions Page 170

Essential Graduate Physics EM: Classical Electrodynamics

1 4 2n cos   n  sin  1
H
4 15
r 5 r
r 3
3
 
  R 2  r 2 n r cos   n  sin  

r

15
    
n r 5R 2  3r 2 cos   n  6r 2  5R 2 sin  , for r  R . 
As useful sanity checks: on the sphere’s surface (i.e. at r = R), the result gives a purely dipole
field:
R 3
H r R  n r 2 cos   n  sin  ,
15
which coincides with the field following from Eqs. (*) and (**) for r = R. On the other hand, in the
sphere’s center (r = 0), the field is naturally directed along the rotation axis:
R 3
n r cos   n  sin    R
3
H r 0  ω.
3 3

Problem 5.12. A conducting sphere with no total electric charge is rotated about its diameter with
a constant angular velocity , in a uniform constant external magnetic field B directed along the rotation
axis. Assuming that the sphere’s contribution to the magnetic field is negligibly small, calculate the
stationary distribution of the electric charge density inside the sphere and on its surface, and the
electrostatic potential both inside and outside the sphere. Quantify the above assumption.
Solution: By definition, a conductor is a medium with some charge carriers (e.g., electrons in
metals) free to move under the effect of the Lorentz force:
F  q E  v  B 
– see, e.g., Eq. (5.10) of the lecture notes. In a stationary state with no currents, this force has to vanish,
so the charges have to self-distribute in a way to make their electric field E balance the magnetic field’s
effect:
E  v  B .
where the charge’s velocity v in the lab reference frame is due to the sphere’s rotation as a whole:11 v =
r, so
E  ω  r   B .

Here r is the observation point’s radius vector with the origin at any static point; for what follows, it is
convenient for us to place this origin in the sphere’s center.
In our simple case, the directions of the vectors  and B coincide, so by taking this direction for
the z-axis, decomposing the radius vector as r = nzz + n (where nnz, so nz  n = 0), and using the
bac minus cab rule,12 we get

 
E   n z    n z z  n ρ ρ   n z B   n z  n ρ  n z  ρB  n ρ  Bρ.

11 If there is any doubt about this expression for v, please consult CM Eq. (4.9).
12 See, e.g., MA Eq. (7.5).

Problems with Solutions Page 171

Essential Graduate Physics EM: Classical Electrodynamics

Hence, the field E is normal to the rotation axis and proportional to the observation point’s distance 
from it. Now we may use the Maxwell equation (1.27) to calculate the corresponding distribution of the
electric charge density :13
   0   E   0 B   n ρ ρ   2 0 B .
Hence the volumic density of the induced electric charge inside the sphere is constant.
The distribution of the electric potential inside the sphere may be readily calculated directly from
E, by using Eq. (1.33):14
  E  n  Bρ .

Since our E is axially symmetric and does not have any z-component, we may take (r) = (), so
according to MA Eq. (10.2),  = n d/d, and this equation becomes merely
d
 B , i.e. d  Bd ,
d
which is very easy to integrate:
Bρ 2 Br 2
  0   0  sin 2  , for r  R . (*)
2 2
Here the integration constant 0 has the sense of the potential’s value at the rotation axis (and in
particular, in the sphere’s center), and  is the usual polar angle. In particular, on the sphere’s surface (r
= R) the potential is
BR 2 BR 2
  0 
2
sin    0 
2

2

1  cos 2  , for r  R . (**)

At r > R, where  = 0, the potential satisfies the Laplace equation, which has to be solved with
the axially symmetric boundary condition (**). Hence its solution is also axially symmetric and has to
obey Eq. (2.172). After dropping all the terms diverging at r   (which clearly should vanish in our
case) and also the term proportional to b0/r (which describes the Coulomb field E  1/r2 of the net
charge of the sphere, in our case equal to zero), and taking the potential’s value a0 at infinity for zero
(just for simplicity), it reads

b
 r ,     l l 1 P l cos  , for r  R, (***)
l 1 r

where Pl() are the Legendre polynomials. In order to find the remaining coefficients bl, it is
instrumental to express Eq. (**) as a linear combination of these polynomials as well. This is easy: from
the third line of Eq. (2.170), we have
2
1   2  1  P 2   ,
3
so by using the first line of Eq. (2.170) as well, Eq. (**) may be rewritten as

13 The last step uses MA Eq. (10.4) with f = , and f = fz = 0.

14 Note that in this case, Eq. (1.46) of the lecture notes, also obtained for a constant charge density inside a sphere,
is inapplicable because its derivation used the additional assumption of the spherical symmetry of the function
(r), which is clearly invalid in this case since the field E does not obey this symmetry.

Problems with Solutions Page 172

Essential Graduate Physics EM: Classical Electrodynamics

BR 2
1  P 2 cos    0  BR
 BR 2
2
  0  P 0 cos    P 2 cos   .
3  3  3
Now requiring Eq. (***) to match this expression at r = R, we get
BR 2 b1 b2 BR 2
0   0,  0,   , and bl  2  0,
3 R2 R3 3
so the external field of the sphere is a pure electric-quadrupole one:
BR 5 BR 5
 3
P 2 cos     3
3 cos
 1 , for r  R .
2
 (****)
3r 6r
What remains is to calculate the density  of the surface charge of the sphere. For that, we may
use the evident generalization of Eq. (2.3) for the case when the electric field has nonvanishing normal
components on both sides of the surface:
   
   0  r  R 0  r  R 0  .
 r r 
With our Eqs. (*) and (****), this formula yields
 0BR
   0BR  3 cos 2   1  1  cos 2    3  5 cos   .
1  2

2  2
As a sanity check, the surface integral of this density,

 0BR 1
8
Qs    d r  2R   sin  d  2R
2 2 2

2
 
2  3  5 2 d 
3
 0BR 3 ,
rR 0 0

is equal and opposite to the volume integral of the bulk density of charge,
8
Qb   d r  2 0B  ρd 3 r    0BR 3 ,
3

rR rR
3
as it has to be for the sphere with zero total charge.
Finally, from the solution of the two previous problems, we know that the rotation of bulk and
surface charges Qs and Qb of a sphere induces their own (generally, differently distributed) magnetic
field Hind ~ Q/R. In our current case, this field’s magnitude is of the order of 02BR2, so
Bind H v2
~  0 ind ~  0  0 R   max
2
,
B B c2
where vmax  R is the largest linear velocity of the sphere’s points. Hence the used assumption Bind <<
B is valid for any non-relativistic rotation.

Problem 5.13.* The simplest version of the famous homopolar (or “unipolar”) motor15 is a thin
round conducting disk, placed into a uniform magnetic field normal to its plane, with dc current passed
between the disk’s center and a sliding electrode (“brush”) on its rim – see the figure below.

15 This device was invented by Michael Faraday in 1821 (i.e. well before his much-celebrated work on
electromagnetic induction) and is frequently called the Faraday disk.

Problems with Solutions Page 173

Essential Graduate Physics EM: Classical Electrodynamics

(i) Express the torque rotating the disk via its radius R, the magnetic
B
field B, and the current I.  
(ii) If the disk is allowed to rotate about its axis, and the motor is I

driven by a battery with e.m.f. V, calculate its stationary angular velocity ,   
neglecting friction and the electric circuit’s resistance. V
(iii) Now assuming that the current’ path (battery + wires + contacts I
+ disk itself) has a non-zero resistance R, derive and solve the equation for
the time evolution of , and analyze the solution.
Solutions:
(i) Applying Eq. (5.8) of the lecture notes to an elementary area of the disk, we get the
elementary Lorenz force
dF  J ρ   Bd 2 ρ ,
where J is the linear density (measured, e.g., in A/m) of the current in the disk, and  is the 2D radius
vector in the disk’s plane. Relative to the disk’s center (i.e. the rotation axis), this elementary force
creates the following elementary rotating torque:16
dτ  ρ  dF  ρ  J ρ   B d 2  .
Since the vector dF is perpendicular to the vector B, i.e. lies in the disk’s plane, and so is vector , all
elementary torque vectors d are directed along the disk’s symmetry axis, and hence ad up as scalars.
Moreover, the radial component of dF, proportional to the angular component Jn of the current, does
not contribute to the net torque , so its magnitude is17
   J  ρ Bd 2  , (*)
S

where J is the radial component of the current density, and S is the disk’s area.
Superficially, it may look like that in order to complete the calculation of the torque, we need to
know in detail the distribution of the current density J (or at least of its radial component J) over the
disk’s area – for example, the one calculated in the solution of Problem 4.9 in the approximation of point
injection and pickup of the current. However, this is not so. Indeed, spelling out Eq. (*) in the form
R 2
  B  d  d J  ρ  ,
0 0

we may notice that since d is an elementary arc of a circle of the radius , the internal integral is just
the total current I flowing out from the disk’s center. Due to the charge conservation (see the discussion
in Sec. 4.1 of the lecture notes), this current cannot depend on , and we may complete the integration as
follows:
R
R2 .
  BI  d  BI (**)
0
2

16 See, e.g., CM Eq. (1.34).

17 Formally, this result may be obtained by using the well-known “bac minus cab rule” – see MA Eq. (7.5):
dτ / d 2   ρ  J  B   J ρ  B   Bρ  J   0  BJ  .

Problems with Solutions Page 174

Essential Graduate Physics EM: Classical Electrodynamics

The background reason for this surprising simplicity will be discussed below.
(ii) The simplest (though somewhat formal) way to fulfill this task is to apply the energy-work
principle to an elementary time interval dt:
 I 2 
dW  VdQ  VIdt  dTrotation  d   ,
 2 
where I is the disk’s moment of inertia for rotation about its axis.18 Dividing both sides of this equation
by dt, and differentiating its right-hand side, we get
VI  I .
But according to the basic equation of rotation dynamics,19 the product I (i.e. the time derivative of
the disk’s angular momentum) should be equal to the applied torque , so using Eq. (**), we get
R2 2V
VI    BI , giving    0  . (***)
2 BR 2
At first glance, this result is counter-intuitive. Since we have neglected the circuit’s resistance,
why could not the non-zero e.m.f. lead to an infinite growth of the current I, and hence of the torque, and
hence an infinite speed-up of the disk’s rotation? Some light on the underlying physics may be shed by
the following simple model. Let the disk’s conductivity be due to certain classical particles with charge
q each. Then the disk’s rotation with angular velocity  results in the particles’ motion (in the lab
reference frame, i.e. relative to the magnetic field’s source) with the velocity v = , and hence to the
radially-directed Lorentz force (5.10) with the magnitude F = qvB = qB. The ratio F/q = B may be
viewed as the magnitude of some additional, rotation-induced electric field Eind; it is easy to check that
for any sign of q, this field is directed against the current I. As a result, the total additional e.m.f.,
R R R
R2
Vind    E ind  dρ   Bdρ  B  dρ  B , (****)
0 0 0
2
induced by the rotation between the disk’s axis and the brush contact, has the polarity opposite to the
battery’s e.m.f. V. So, the physics of Eq. (***) is that at  = 0, Vind exactly compensates V.20
(iii) The last discussion allows for a ready generalization of the result to the case of nonvanishing
resistance R of the current-supplying circuit. Indeed, by writing the 2nd Kirchhoff law (4.7b) with the
account of the rotation-induced e.m.f.,

18 See, e.g., CM Secs. 4.2-4.3.

19 See, e.g., CM Eqs. (4.38).
20 This result is actually more general than the explored simple model and is one of the manifestations of
Faraday’s law of electromagnetic induction – to be discussed in Sec. 6.1 of the lecture notes. Indeed, it is easy to
check that the product R2/2, which participates in Eq. (****), is just dA/dt, where A is the area swept by the
circle’s radius, so Eq. (****) may be rewritten as Vind = d(BA)/dt. Since, for a uniform magnetic field B normal to
the disk’s plane and directed against axis z, BA is just minus magnetic flux  through area A, Eq. (***) agrees
with the induction law in its best-known form – see Eq. (6.1). (Note, however, that the exact specification of what
exactly A is in this case, and hence the derivation of Eq. (****) from the Faraday law, require some care.) This
discussion shows that the Faraday law is not something entirely new, and is at least consistent with the laws of
magnetostatics. (For more discussion of this fact, see Sec. 6.2 of the lecture notes.)

Problems with Solutions Page 175

Essential Graduate Physics EM: Classical Electrodynamics

V  Vind  IR ,
using the I expressed from Eq. (***) with   I , and Vind from Eq. (****), we get
BR 2 2
V   I R.
2 BR 2
This linear, first-order differential equation for  may be rewritten in the standard form
 d
1  t 0    0 ,
 dt 
where t0 is the following time constant:
4IR
t0  2 4 ,
B R
and 0 is the stationary value of , given by Eq. (***). This equation may be readily solved, giving
t / t0 t / t
  0 1  e 0  .
 t    0 e
 
This formula describes an exponential transient process, with the time constant t0, from the initial
rotation velocity (0) to its stationary value (***). This time constant is proportional to the circuit’s
resistance R, making it clear why in Task (ii) of the problem (where R and hence t0 were neglected) we
received value (**) as a constant, without any apparent transient process.
Finally note that the same device, if driven by an external torque and with the battery V replaced
with an electric load, may serve as a homopolar generator of dc e.m.f. (***).21 Since it has only one
current loop, for most practical purposes it is less efficient than the usual multi-coil generators but has
certain technical advantages for producing short, very large pulses of current (in some cases, of several
mega-amperes!) by using the gradually accumulated kinetic energy of massive flywheels, for such
special purposes as driving railguns – see, e.g., Problem 6.7 below.

Problem 5.14. The reader is hopefully familiar with the classical Hall effect in the usual
rectangular Hall bar geometry – see the left panel of the figure below. However, the effect takes a
different form in the so-called Corbino disk – see the right panel of the figure below. (Dark shading
shows the electrodes, with no appreciable resistance.) Analyze the effect in both geometries, assuming
that in both cases, the conductors are thin and planar, have a constant Ohmic conductivity  and a
charge carrier density n, and that the applied magnetic field B is uniform and normal to the conductors’
planes.
I I
I I  R2
w  B
R1
B

21 It was also invented by M. Faraday and is sometimes called the “Faraday wheel”.

Problems with Solutions Page 176

Essential Graduate Physics EM: Classical Electrodynamics

Solution: In the Hall bar geometry, the current I is uniformly distributed across the sample, with
the density j = I/wt, where t is the bar’s thickness. As was discussed in Sec. 4.2, in an Ohmic conductor,
this density equals qnv, where v is the average “drift” velocity of the charge carriers (on top of their
random thermal and/or quantum motion), so v = I/wtqn. This velocity gives rise to an average magnetic
component of the Lorentz force (5.10), with magnitude F = qvB, and directed normally to the current
and the magnetic field, i.e. across the Hall bar. As the current is turned on, this force causes
accumulation of the charges at the lateral sides of the bar, which stops when their electric field E
becomes large enough to compensate for the magnetic force: E = –vB at each point of the bar. So, at
a stationary motion of the charge carriers along the current’s direction, the electric field has to be
uniform, hence providing a transverse voltage of magnitude V = RHI, where RH is the so-called Hall
resistance
V wE w v B B
RH     .
I I I qnt
Note the following important features of this well-known result: first, RH does not depend on w
and l, i.e. on the Hall bar’s size, and, second, in contrast to the usual Ohmic resistance (see Eqs. (4.9)
and (4.13) of the lecture notes) it is sensitive to the sign of q. These features make the measurement of
RH the standard technique for finding out the sgn(q) and the charge carrier density n in semiconductors.
Now addressing the Corbino disk: due to the axial symmetry of its geometry, the radial
component j of the current density has to be uniformly distributed over the angle, so at distance  from
the center,
I j I
j  , and v    ,
2 t qn 2 tqn
causing an azimuthally-directed average magnetic force of the following magnitude:
IB
F  q v  B  .
2 tn
This field is equivalent, in its action on the carriers, to an azimuthal electric field
F IB
E   .
q 2 tqn
However, in this axially-symmetric geometry, the resulting azimuthal motion of the charge carriers does
not lead to their accumulation, and hence continues even in the stationary state, giving an azimuthal
(circular) current with the following density:
IB
j   E  .
2 tqn
Calculating the full circular current,
R2
B R
I   t  j d  I ln 2 ,
R 2 qn R1
1

we see that the I/I ratio is independent of the disk’s thickness t, and, according to Eq. (4.13), also of the
charge carrier density:

Problems with Solutions Page 177

Essential Graduate Physics EM: Classical Electrodynamics

I  c R2
 ln ,
I 2 R1
where  = m/q2n is the effective scattering time participating in the Drude formula (4.13), and c 
qB/m is the cyclotron frequency of the same charge carriers if they move ballistically in the same
magnetic field. (For more about c, see Sec. 9.6 of the lecture notes.)
For typical conductors in practicable magnetic fields, I/I << 1, and the magnetic field produced
by the circular currents (which was ignored in the above solution) is much smaller than B.

Problem 5.15. A wire with a round cross-section of radius a has been bent into a round loop of
radius R >> a. Prove the formula for its self-inductance, which was mentioned at the end of Sec. 5.3 of
the lecture notes:
cR
L   0 R ln , with c ~ 1.
a
Solution: The general expression (5.54) for the magnetic energy U, and Eq. (5.72) for the self-
inductance of a loop, L = U/(I2/2), allow us to separate L into a short-distance contribution Ls from the
region with  r – r’ ~ a, and a long-distance contribution Ll from the region outside of that range. For the
latter, dominating contribution, we may use the Neumann formula (5.60). For the self-inductance of a
loop, it reduces to
 dr  dr' r
Ll  0   , for r  r'  a , (*)
4 l l r  r'
where both integrals (in the space of the vectors r and r’) are over the same r  r'
R 
loop’s length. For our geometry (see the figure on the right),  r – r’ =
2Rsin(/2) and drdr’ = Rd dr’cos, where  is the angle between the
radius vectors r and r’, with the origin at the loop’s center. Due to the axial
symmetry of this geometry, the integration over one of the variables (say, r’) r'
may be performed while keeping that angle  fixed and gives

 0 R  cos  d a
4  sin  / 2 
Ll  , with   .
R
Due to the last condition, we have to restrict the last integration to the angles outside a certain
interval near the point  = 0 – say, a symmetric interval –/2    +/2, where  must be within the
limits a/R <<  << 1. In this case, the integral is symmetric in , and may be readily calculated

0 R
2

cos d
 0 R 
 
 
1  2 cos 2  / 2 sin  / 2  d  / 2
 cos( / 4)
2  1 d
2

Ll 
4  / 2 sin  / 2   / 2 1  cos 2  / 2
 0 R 
 0 1 2
 cos( / 4)
 1 1  1 1     cos( / 4)
 0 R 
 0
 21     21     2 d   2 ln 1    2    0
   
 1 1  cos / 4     8 
  0 R  ln  2 cos    0 R ln tan   2 cos  .
 2 1  cos / 4 4    4

Problems with Solutions Page 178

Essential Graduate Physics EM: Classical Electrodynamics

If we take  = Ca/R, where C is a constant: 1 << C << R/a, the above result reduces to
 8R 
Ll   0 R ln  2 .
 Ca 
Since, at R >> a the magnetic field inside the wire and its close vicinity cannot be affected by its
bending, the short-range contribution Ls to the full inductance of the loop may be fairly estimated using
Eq. (5.80) of the lecture notes, by replacing b with a certain C’a, where 1 << C’ << R/a, i.e. C’ ~ C, and
integrating the result along the loop:
 1
Ls   0 R ln C'  .
 4
We may see that this contribution is much smaller than Ll, so the approximate character of its evaluation
does not affect the (very reasonable) accuracy of the final result:
 8C'R 1
L  Ll  Ls   0 R  ln  2  . (**)
 Ca 4
(Here I have kept the term ¼ separate, because, as was already mentioned in Sec. 5.3 of the lecture
notes, it is due to the magnetic field inside the wire and hence disappears in the important case when a
high frequency of the field prevents its penetration into the wire’s bulk.) This expression is equivalent to
the formula given in the assignment, with
8C'  1
c exp  2   ~ 1 ,
C  4
so our task has been accomplished. (A more careful evaluation of Ls shows22 that C’ = C, so in Eq. (**),
these factors cancel, giving c = 8exp{–7/4}  1.39 for stationary or low-frequency fields, and c = 8exp{–
2}  1.08 in the high-frequency limit.)
Please note an important byproduct of our solution: since at a  0, the short-range contribution
does not depend on the long-range loop’s geometry, a reasonably good approximation of the self-
inductance of a thin-wire loop of an arbitrary shape is formally given by its long-distance contribution
(*) alone, but with a more mild cut-off:  r – r’ > a.

Problem 5.16. Prove that:

(i) the self-inductance L of a current loop cannot be negative, and
(ii) each mutual inductance coefficient Lkk’, defined by Eq. (5.60) of the lecture notes, cannot be
larger than (LkkLk’k’)1/2.
Solutions:
(i) Since, according to Eq. (5.57), the magnetic energy U of any system cannot be negative, Eq.
(5.72) for a single current loop,
L
U  I2,
2
cannot give negative results for any current I, which is only possible if L  0.

22 See, for example, W. R. Smythe, Static and Dynamic Electricity, 3rd ed., McGraw-Hill, 1968.

Problems with Solutions Page 179

Essential Graduate Physics EM: Classical Electrodynamics

(ii) Let us apply a similar argument to the more general Eq. (5.59). It also cannot give a negative
result for any combination of currents, in particular when only two of them, Ik and Ik’ with k’  k,23 are
different from 0, and hence this relation (taking into account Eq. (5.61), Lkk’ = Lk’k) is reduced to a form
similar to Eq. (5.62):
L L
U  kk I k2  Lkk ' I k I k '  k 'k ' I k2' . (*)
2 2
This means, in particular, that if we fix one of the currents (say, Ik’), U as a function of the other current
(Ik) cannot be negative even at its minimum, where
U
 Lkk I k  Lkk ' I k '  0 .
I k
Plugging the resulting value, Ik = –Lkk’Ik’/Lkk, back into Eq. (*), we get
2
L  Lkk' I k '  L I L  L2kk '  Lk 'k ' 2
U min  kk    Lkk' kk' k ' I k'  k 'k ' I k2'  1   Ik' .
2  Lkk  Lkk 2  Lkk Lk 'k '  2
Now the requirement Umin  0 immediately gives the required proof of the inequality Lkk’  (LkkLk’k’)1/2.

Problem 5.17. Calculate the mutual inductance of two similar thin- h

wire square-shaped loops offset by distance h in the direction normal to their
planes – see the figure on the right. a a
Solution: This problem may be most simply solved using the Neumann formula – see Eq. (5.60)
of the lecture notes:
 dr  dr2
M  L12  L21  0   1 , (*)
4 l l r1  r2
1 2

where the indices 1 and 2 number the wire loops (in any order), and both contour integrals should follow
the same (but arbitrary) direction. Due to the evident symmetry of the system, this integral falls into a
sum of four equal components Ms, each corresponding to the integration along only one side of one of
the squares. Moreover, since the scalar product dr1dr2 vanishes for the sides normal to each other, and
equals dr1dr2 for the parallel sides, Ms may be represented as

Ms 
0
4
 
f a, h   f a, h 2  a 2 , 
where f(a, d) is the same double integral as in Eq. (*), but limited to just two parallel sides separated by
distance d. (One may loosely think about (0/4)f(a, d) as the mutual inductance of these two wire
segments, even though this notion is strictly defined only for two closed current loops.)
In order to calculate the function f(a, d), it is convenient to use x2
the Cartesian coordinates shown in the figure on the right; in them,
d

0 a x1

23 If k = k’, the correctness of (ii) is obvious because Lkk’  Lkk cannot be larger than (LkkLk’k’)1/2 Lkk, i.e. itself.

Problems with Solutions Page 180

Essential Graduate Physics EM: Classical Electrodynamics

a a
1
f a, d    dx2  dx1
0 0 x  x 
1 2
2
d2  1/ 2 .

The inner integral is easy: by taking (x1 –x2)/d  , and then   sinh, so d/d = cosh and (2 + 1)1/2
= cosh cancel, we get
a ( a  x2 ) / d a  ( a  x2 ) / d a
d
f a, d    dx2    dx2  d   dx2 sinh 
1
 (a  x ) / d
2

0  x2 / d  2
1  1/ 2
0   x2 / d 0  x2 / d
a
 a  x2 x 
  sinh 1  sinh 1 2  dx2 .
0 d d
This expression falls into a sum of two similar integrals, which may be readily worked out by parts:
d
1 1 1
 1

 sinh d   sinh    d sinh    sinh     2  1 1 / 2   sinh     1 .
1 2 1/ 2
 
 
The result is
 a

f a, d   2a sinh 1  d  d 2  a 2
d
1/ 2 
 .
100

The red line in the figure on the
2a / h 
right shows the final result for the mutual 10 3

inductance M
 0 a / 4
  
0 ca
8 ln
M  4M s  f a, h   f a, h 2  a 2 , 1 h

as a function of the h/a ratio. As the )
distance h between the loops (with a fixed 0.1
size a) is increased, the mutual inductance
coefficient decreases. At h/a << 1, this
trend is only logarithmic because it is
0.01
governed by the field very close to the 0.01 0.1 1 10
wires: h/a
2  0 a  2a  2  0 a ca
M   ln  2  2  sinh 1 1  ln , with c  0.461, at h / a  0
  h   h
– see the lower dashed line in the figure above. (The weak divergence at h  0 disappears for wires of
any non-zero thickness.)
On the other hand, at large distances between the loops, the mutual inductance decreases fast:
0a 4
M 
, at h / a   (**)
2h 3
– see the upper dashed line. This behavior may be readily understood: at a large distance h >> a from a
loop carrying current I, its field approaches that of a magnetic dipole m = IA, in our current case with
area A = a2 – see Eq. (5.97) of the lecture notes. According to Eq. (5.99), on its symmetry axis, the
dipole field is directed along the axis and has the magnitude

Problems with Solutions Page 181

Essential Graduate Physics EM: Classical Electrodynamics

 0 2m  0 2 Ia 2
B  .
4 h 3 4 h 3
Piercing the counterpart loop, of the same area A = a2 and with the plane normal to its lines, the field
creates the magnetic flux
 2 Ia 4
  Ba 2  0 ,
4 h 3
so the ratio /I, i.e. the mutual inductance of the loops (see, e.g., Eq. (5.67) and Fig. 5.7) is indeed given
by Eq. (**).
Note that this approach (the decomposition of M into side-pair components) may be used for the
expression of the mutual inductance of other systems of rectangular loops, including those offset
sidewise within their common plane, via the same function f(a, d).

Problem 5.18.* Estimate the values of magnetic susceptibility due to

(i) orbital diamagnetism, and
(ii) spin paramagnetism,
for a medium with negligible interactions between the induced molecular dipoles. Compare the results.
Hints: For Task (i), you may use the classical model described by Eq. (5.114) of the lecture notes
– see Fig. 5.13. For Task (ii), assume the ordering of spontaneous magnetic dipoles m0, with a fixed
magnitude m0 of the order of the Bohr magneton B, similar to the one sketched for electric dipoles in
Fig. 3.7a.
Solutions:
(i) According to Eq. (5.114), the angular velocity of the torque-induced precession of the vector
L (in this case called the Larmor frequency),
2 2 qB 
ΩL     B B,
T L 2L sin  /( dL / dt ) 2me 
does not depend on the angle  between these two vectors. 24 This motion creates the average current
ΩL q2 e2
I  qf L  q  B B
2 4 me 4 me
(note its independence of the sign of q) and hence the induced magnetic moment may be estimated as25

Δm  I  2 
e2 B 2
4me
 
e2 B 2
4me

x  y2 , 

24 As a useful intermediate sanity check,  =  B is exactly the result given by quantum mechanics for the
L B
atomic energy level splitting by a magnetic field, due to the electron’s orbital motion – see, e.g., QM Sec. 6.4.
25 In this model, the basic dipole moments m are assumed to be disordered, so their net contribution to the
magnetization vanishes.

Problems with Solutions Page 182

Essential Graduate Physics EM: Classical Electrodynamics

where  is the distance of the electron from the precession axis, taken for axis z. Now let us add such
contributions from Z electrons of an atom, assuming that the initial orientations of their rotation planes
are random, and hence the averages of all Cartesian coordinate squares are equal:
r2 2 2
x 2
 y 2
 z 2
 , so that x 2  y 2 
r .
3 3
With this assumption, for the orbital molecular susceptibility we get the following Langevin formula:
Δm Δm  nZe 2 2
m  n n  0 r ,
H B / 0 6me
where n is the number of atoms per unit volume.
Amazingly enough, this is exactly the result given by quantum mechanics,26 with the only
difference that in it, r2 should be understood as the simultaneously statistical and quantum-mechanical
average, taking into account the spread of the electrons’ wavefunctions. Thus the susceptibility may be
crudely estimated by taking r2 equal to the square of the Bohr radius rB  402/mee2.27 This
substitution yields
 0 nZe 2 rB2 2
m ~  Z nrB3  2 , (*)
6me 3
where, in the last expression, the parameter
1/ 2
e2 e2 0c e2  0  1
       1
4 0 c 4 4   0  137
is the fine-structure constant, which describes the relative strength of electromagnetic interactions on the
quantum-relativistic scale.28 It is evident that even for the heaviest atoms (Z ~ 102) in their condensed-
matter form (where n ~ rB–3), this magnetic susceptibility is still small: m << 1, i.e. the orbital
diamagnetism is always weak.
(ii) As was discussed in Sec. 5.5 of the lecture notes, we may expect the full alignment of
spontaneous magnetic dipoles to be achieved at m0B ~ kBT, resulting in the induced magnetic moment
m = Zm0 per atom, and hence the saturation of the magnetization at the level M0 = nZm0. Hence it is
reasonable to expect that in a weak magnetic field, the magnetization scales as29
m0 B nZm02 B
M ~ M0  , for M  M 0 ,
k BT k BT
corresponding to the paramagnetic (positive) susceptibility

26 See, e.g., the solution of QM Problem 6.14.

27 See, e.g., QM Secs. 1.1 and 3.6. As a mnemonic rule, the expression rB may be obtained by equating the scales
of the electrostatic (e2/40rB) and quantum-kinetic (2/merB2) energies in Bohr’s model of the hydrogen atom,
without any numerical coefficients. (Note also that each of these fractions gives the atomic energy scale EH 
(e2/40)2me/2  27.2 eV, called the Hartree energy.)
28 In this course, we will run into the fine-structure constant again in Sec. 10.4; it will also be repeatedly discussed
in the QM part of the series. For its more exact value, see Appendix UCA: Selected Units and Constants.
29 This assumption is confirmed by quantitative statistical analyses – see, e.g., SM Sec. 2.4.

Problems with Solutions Page 183

Essential Graduate Physics EM: Classical Electrodynamics

M M  0 nZm02
m    .
H B / 0 k BT
This result obeys the so-called Curie law, m  1/T (also valid for e in paraelectrics).
In order to estimate this susceptibility, we may take m0 ~ B  e/2me, getting
2
 0 nZ  e 
m ~ 
E
   Z nrB3  2 H .
k BT  2m e  k BT
 
For room temperatures (T ~ 300K), the ratio EH/kBT is close to 103, so comparing this estimate with Eq.
(*), we see that the spin paramagnetism is much stronger than the orbital diamagnetism,30 so the latter
effect prevails only in materials with zero net atomic spins – see Table 5.1 in the lecture notes.

Problem 5.19.* Use the classical picture of the orbital (“Larmor”) diamagnetism, discussed in
Sec. 5.5 of the lecture notes, to calculate its (small) contribution B(0) to the magnetic field B felt by an
atomic nucleus, treating the electrons of the atom as a spherically symmetric cloud with an electric
charge density (r). Express the result via the value (0) of the electrostatic potential of the cloud and
use this expression for a crude numerical estimate of the ratio B(0)/B for the hydrogen atom.
Solution: As was discussed in Sec. 5.5, a relatively low external magnetic field B causes the
torque-induced precession of a classical particle’s orbit around the field’s direction, with the angular
velocity
q
Ω B. B
2m0 
In the (admittedly, somewhat eclectic but still very useful) model, in which  r 
jr   r
the quantum-mechanical spread of electron positions at B = 0 is represented by a
time-independent classical charge density (r), the rotation induces loop currents
(see the figure above) with density j(r) = v(r)(r), where v(r) = r.31 The 0
magnetic field correction B due to these currents may be calculated using Eq.
(5.14). For the origin-located nuclei (with r = 0), and with r’ replaced with r (just for notation
simplicity), that formula yields
0 r   r  0 q  r 
B0     jr   3 d 3 r   0  Ω  r   r d 3r   B  r   r d 3r .
4 r 4 r 3
4 2m0 r3
The vector (Br)r lies in the mutual plane of the vectors r and B may be represented as a vector
sum of two components: –Br2sin2 directed against the vector B (where  is the angle between the
vectors B are r – see the figure above) and another one, normal to this vector. For any axially symmetric
charge distribution (r), the integral of the latter component over the azimuthal angle vanishes. Hence,

30 This estimate is invalid in metals with their conduction electrons forming a degenerate Fermi “sea”. In this
case, the diamagnetic susceptibility, calculated within a simple model of independent Fermi particles, called the
Landau diamagnetism, equals exactly 1/3 of the so-called Pauli paramagnetism due to the spin orientation of the
same electrons – see, e.g., the solutions of SM Problems 3.10 and 3.11.
31 If you are uncomfortable with the last relation, please revisit a classical mechanics course, e.g., the discussion
at the beginning of CM Sec. 4.1 of this series, in particular Eq. (4.9).

Problems with Solutions Page 184

Essential Graduate Physics EM: Classical Electrodynamics

for any sign of the moving particles’ charge (which determines the sign of not only q, but also of ), the
vector B(0) is directed exactly against B and hence describes a diamagnetic response of the orbital
motion. The expression for its relative magnitude,
B0   r  3  
 q  q
 0  sin 2  d r 0 2  sin 3 d   r  rdr ,
B 4 2m0 r 4 2m0 0 0

may be further simplified takes an especially simple form if the charge distribution is spherically
symmetric: (r) = (r). Indeed, in this case, the integral over  is may be readily worked out by the
usual substitution   cos (so d = –sin d and sin2 = 1 – 2):
 1
d   1   2 d  .
4
 sin
3

0 1
3
The final result,
B0

 q 4
 r  rdr ,
4 m0 3 0
 0 (*)
B

may be readily related to the value (at the same central point) of the electrostatic potential (r) induced
by this charge distribution. Indeed, by applying Eq. (1.38) of the lecture notes to the point r = 0, and
again replacing r’ with r, for the spherically symmetric charge density we get the same integral:
1  r  1

 0   d 3r  4   r rdr ,
4 0 r 4 0 0

so the comparison with Eq. (*) yields

B0   q 1 q
 0 0  0    0 .
B 3 m0 3 m0 c 2
This equality enables a simple estimate of the diamagnetic correction in the hydrogen atom: for
electrons, with q = –e and m0c2 = mec2  0.51 MeV,  (0)  may be crudely estimated as e/40rB  EH/e,
where rB is the Bohr radius, and EH  27 eV is the Hartree energy unit,32 so
B 0  1 E H
~ 2
 2  10 5 .
B 3 me c

Problem 5.20. Calculate the self-inductance of a toroidal z

I I
solenoid with a round cross-section of radius r ~ R (see the figure r
on the right), with N >> 1, R/r wire turns uniformly distributed 
along the perimeter, and filled with a linear magnetic material of 0 R 
permeability .
Solution: In Sec. 5.2 of the lecture notes, the magnetic
field inside such a solenoid without magnetic filling was calculated – see Eq. (5.41). According to the
discussion at the beginning of Sec. 5.6, since all the field is concentrated inside the solenoid filled with
the magnetic material, we may just multiply that result by the ratio /0:

32 See, e.g., either the solution of the previous problem or QM Sec. 1.1, in particular, Eqs. (1.9) and (1.13).

Problems with Solutions Page 185

Essential Graduate Physics EM: Classical Electrodynamics

NI
B ,
2
where  is the distance from the z-axis. Now we can calculate the magnetic flux piercing one wire loop:

NI r R  r 2  z 2  NIr 1 a  1   2 
r R ( r 2  z 2 )1 / 2 1/ 2 1/ 2
NI d
 1   Bn d r   dz   0 R  r 2  z 2 1 / 2  0 a  1   2 1 / 2
2
 ln dz  ln d ,
A
2 r R ( r 2  z 2 )1 / 2


where a  R/r > 1 and   z/r. This is a table integral33 equal to [a – (a2 – 1)1/2], so, finally,

 
 1  NI R  R 2  r 2 
1/ 2
.
Just as in the long solenoid discussed in Sec. 5.2 of the lecture notes, the whole flux  piercing
all N turns of the whole wire is N times larger. As a result, the solenoid’s inductance

L
 1 N
I

I

 N 2 R  R 2  r 2 1/ 2
. (*)

In the limit r << R, we may expand this expression into the Taylor series in small ratio r/R and,
in the first nonvanishing approximation, get
r2 N
L  N 2  n 2 lA, with n  , l  2R, and A  r 2 .
2R l
We see that in this limit, the result coincides with the inductance of a long straight solenoid, calculated
in Sec. 5.6 – see Eq. (5.143). It is curious that in the opposite limit (r = R), Eq. (*) also acquires a very
simple form,
L  N 2 r .

Note that these results, as well as Eq. (5.142) of the lecture notes for the straight solenoid, may
be used for relatively small values of N only if  >> 0; otherwise, we have to account for the “stray”
magnetic field spilling out between separated wire turns.

Problem 5.21. A long, straight, thin wire carrying current I runs I

1 d
parallel to a sharp plane boundary between two uniform, linear magnetic
media – see the figure on the right. Calculate the magnetic field
everywhere in the system, and the force (per unit length) exerted on the 2
wire.
Solution: If the linear magnetic material surrounding the wire was purely uniform ((r) = const),
its field lines would be circular, with the magnitude described by Eq. (5.20) of the lecture notes, adjusted
as prescribed by Eq. (5.115). This result could be represented in the following vector form:
I I
H ρ   n  , Bρ   n  , (*)
2 ρ  ρ 0 2 ρ  ρ 0

33 See, e.g., MA (6.13).

Problems with Solutions Page 186

Essential Graduate Physics EM: Classical Electrodynamics

where  is the 2D radius vector of the observation point within a plane normal to the wire (e.g., the
plane of the figure above), 0 is that of the wire, and n is the unit azimuthal vector in this plane, which
is normal to the vector  – 0. However, in our current case, with 1  2, such a solution would not
satisfy the second of the boundary conditions (5.118)-(5.119),
H   const, Bn  const , (**)
at the interface between the two materials.
Inspired by the solution of similar electrostatic and dc-current y
problems in Chapters 2-4 of the lecture notes (see, in particular, the d I, I"
problem illustrated by Fig. 3.12), we may try to look for the solution of n' n
our current problem in the form of a sum of two contributions of the 1 ρ
type (*): one from the actual current I (for observation points in the
lower half-space, replaced by another value, I”) and the other from an 0 x
image current I’ passing at the same distance d from the interface, but 2
on its opposite side – see the figure on the right. With the Cartesian d I'
coordinates selected as shown in this figure, the assumed solution
(analogous to Eq. (3.66) of the similar electrostatics problem) is
 I I'
 n  2 ρ  ρ  n ' 2 ρ  ρ' , for y  0 ,
H ρ    0
(***)
I"
n  , for y  0 ,
 2 ρ  ρ 0

where 0 = {0, +d}, ’ = {0, –d}. At the interface, i.e. at  = {x, 0}, this solution yields

 I d I' d
 2 x 2  d 2  2 x 2  d 2 , for y  0 ,
Hx  
 I" d
, for y  0 ,
 2 x 2  d 2

 1 I x 1 I' x
 2 x 2  d 2  2 x 2  d 2 , for y  0 ,
By  
  2 I" x
, for y  0 .
 2 x  d 2
2

These solutions satisfy the boundary conditions (**) if I’ and I” are selected so that

I  I'  I" , 1 I  I'    2 I".

Solving this simple system of two equations, we get
1   2 2 1
I'   I, I"  I.
1   2 1   2
These relations, together with Eqs. (***), give the solution of the first task of the problem. Now
we may use them to calculate the magnetic field at the position 0 of the actual wire (excluding its own,
diverging contribution):

Problems with Solutions Page 187

Essential Graduate Physics EM: Classical Electrodynamics

I' I' I 1   2
H ρ 0   n '  n '  n ' .
2 ρ 0  ρ' 4d 4d 1   2

At this point, the unit vector n’ (see the figure above) is opposite to nx, so the field is horizontal, with
I 1   2 I 1   2
H x ρ 0   , i.e. B x ρ 0   1 .
4 d 1   2 4 d 1   2
Per the basic Eq. (5.15), the force exerted on the wire is vertical, with
Fy I2   2
 I z B x  IB x  1 1 .
l 4 d 1   2
For the important particular case of a wire stretched in free space, parallel to a linear magnetic with the
permeability , this formula yields
Fy I2  
 0 0 .
l 4 d 0  
Hence the current, regardless of its sign, is attracted by a paramagnetic material (with  > 0)
and repulsed by a diamagnetic one.

Problem 5.22. Solve the magnetic shielding problem similar to that H0

discussed in Sec. 5.6 of the lecture notes, but for a spherical rather than b
cylindrical shell, with the same central cross-section as shown in Fig. 5.16, a
partly reproduced on the right. Compare the efficiency of those two shields,
for the same permeability  of the shell, and the same b/a ratio. 0
H int  ?
Solution: Guided by the general axially symmetric solution (2.172) of
the Laplace equation in the spherical coordinates, we may look for the scalar
potential of the magnetic field in the form

  H 0 r  b1' / r 2  cos  , for b  r ,


 m   a1 r  b1 / r 2 cos  ,

for a  r  b,

  H int r cos  , for r  a ,

which differs from Eq. (5.127) of the lecture notes (for the cylindrical shield) only by the power of the
radius factor in the denominators. Plugging this solution into the boundary conditions, m = const and
m /r = const, for both interfaces (r = b and r = a), we get the following system of four equations:
 H 0 b  b1' / b 2  a1b  b1 / b 2 ,  a a  b / a    H a,
1 1
2
int

 0  H 0  b1' / 2b 3 H 0   a1  b1 / 2b 3 ,   a  b / 2a     H
1 1
3
0 int ,

for four unknown coefficients a1, b1, b1’, and Hint. Solving the system, we get, in particular:

Problems with Solutions Page 188

Essential Graduate Physics EM: Classical Electrodynamics

H int  s 1 (   2  0 )(    0 / 2)
 , where  s  . (*)
H 0  s  a / b 3 (   0 ) 2

Comparing the magnetic material factors  for the spherical (s) and cylindrical (c) geometries
(see Eq. (5.129) of the lecture notes),
 s (  2)(  1 / 2)  2  (5 / 2)  1 
  2 , where   ,
c (  1) 2
  (4 / 2)  1 0
we see that s > c for any paramagnetic shell material (with  > 0), but for the most important case of
soft ferromagnets with  >> 0, this difference is minor: s  c  1. More importantly, since a < b,
the factor (a/b)3 in the denominator of Eq. (*) is smaller than the factor (a/b)2 in the denominator of Eq.
(5.129). Hence, the spherical shields are more effective – if not always practically acceptable because of
the sample placement convenience and fabrication costs. A popular compromise is to use short cylinders
with thick lids that may be closed after the sample has been placed inside.

Problem 5.23. Calculate the magnetic field’s distribution around a spherical permanent magnet
with uniform magnetization M0 = const.
Solution: Due to the absence of stand-alone currents in the region of our interest, the field H in
the problem may be represented by Eq. (5.120) of the lecture notes, H = –m, where the scalar
potential m satisfies the Laplace equation both inside the sphere and outside it. Due to the axial
symmetry of the problem with respect to the direction of the constant magnetization M0, which may be
conveniently taken for the polar axis, the general solution of the Laplace equation may be represented in
the form similar to Eq. (2.172) of the lecture notes. Setting the arbitrary constant a0 to zero, and taking
into account that due to zero divergence of the vector H, the constant b0 vanishes as well, this solution is
reduced to
 al r l  bl / r l 1 , for r  R ,
 m   P l (cos  )   l
l 1 al'r  bl' / r l 1 , for R  r ,

where R is the sphere’s radius, and Pl() are the Legendre polynomials (2.169). Next, since the field is
produced by the sphere itself, it should vanish at large distances from it, so all coefficients al’ must equal
zero. Similarly, the potential inside the cylinder has to be finite, so all coefficients bl have to vanish as
well. As a result, the above expression for m is reduced to
 al r l , for r  R ,
 m   P l (cos  )  
l 1  bl' / r l 1 , for R  r.
In order to find the coefficients al and bl’, we have to write boundary conditions at the sphere’s
surface (r = R). From Eq. (5.118) for the tangential component of field H, we get
 
bl'
 al R l P l (cos )  
l 1 n 1 R l 1
P l (cos  ) . (*)

For the normal component of the field, we have to use the general Eq. (5.118), Bn = const (rather than
Eq. (5.119), which is only valid for linear magnetics) because B is a linear function of H (equal to 0H)

Problems with Solutions Page 189

Essential Graduate Physics EM: Classical Electrodynamics

only outside the sphere, at r  R. At r  R we should use the general relation (5.108), B = 0(H + M), for
our geometry giving Bn = 0(Hn + Mn) = 0(-m/r + Mn). For the fixed, uniform polarization M0, Mn =
M0cos  M0P1(cos), so the second boundary condition takes the form

 
 
bl'
 0   lal R l 1P l (cos θ )  M 0P1 (cos θ )   0  (l  1) P l (cos θ ) . (**)
 l 1  l 1 R l 2
Due to the linear independence of the functions Pl(cos), Eqs. (*) and (**) fall apart into a set of
independent couples of linear equations for each pair {al, bl’}, but only for l = 1, such a pair of equations
is inhomogeneous, i.e. gives a nonvanishing solution for coefficients a1, b1’. (All other coefficients may
equal zero, and due to the uniqueness of the Laplace equation’s solution, they have to equal zero.)
Solving the system of two linear equations for l = 1, we get
M0 M 0 R3
a1  , b1'  .
3 3
From here, the scalar potential of the magnetic field is
M M 0 R3
 m r  R  0 r cos  , m rR  cos  .
3 3r 2
The first of these expressions describes a uniform magnetic field with H = –M0/3 and B = 0(H
+ M0) = (2/3)0M0. (Note that the fields H and B have different directions!) The second formula shows
that at r > R, the field exactly coincides with that of a magnetic dipole with the moment
4 3
m R M 0  VM 0 ,
3
i.e. the sum of all elementary dipoles within the sphere – the result that could be conjectured even
without the formal solution of the problem.
Note that this problem and its solution are very similar to Problem 3.13 on a sphere with fixed
electric polarization. Note also that alternatively, the result may be obtained by the method discussed in
the next problem – though in this particular case, that would involve a more bulky calculation.

Problem 5.24. A limited volume V is filled with a magnetic material with field-independent
magnetization M(r). Write explicit expressions for the magnetic field induced by the magnetization and
its potential, and recast these expressions into the forms that are more convenient when M(r) = M0 =
const throughout the volume.
Solution: As was discussed in Sec. 5.6 of the lecture notes, Eq. (5.132) for the magnetic field,
  H    Μ ,
is mathematically identical to the Maxwell equation Eq. (1.27) for the electric field:

 E  .
0
As was discussed in Sec. 1.2 of the lecture notes, the last expression is equivalent to Eq. (1.9):

Problems with Solutions Page 190

Essential Graduate Physics EM: Classical Electrodynamics

1  r'  r  r' 3
Er    d r' , (*)
4 V
 0 r  r' 3
for the field created by the electric charge contained inside the volume V. Hence, if the only source of
the field H in a system is the magnetization of some volume V, we may use this analogy to write
1  m r'  r  r' 3
H r    d r' , where  m r'     0   M r'  . (**)
4 V
 0 r  r' 3
This means that in the free space outside of the volume V,
1 r  r' 3  r  r'
Br    0 H r   
4 V
 m r' 
r  r'
3
d r'  0
4     Mr'  r  r'
V
3
d 3 r' .

However, per Eq. (5.108), inside volume V, the field B is different:

0 r  r'
Br    0 M r   H r    0 M r       Mr'  r  r' d 3 r' .
4 V
3

Next, as we know from Sec. 1.3 of the lecture notes, Eq. (*) is equivalent to Eq. (1.38),
1  r'  1
 r    d 3 r' ,
4 V
 0 r  r'
for the electrostatic potential defined by Eq. (1.33):
E   .
This means that the magnetic field potential, defined by the similar Eq. (5.120),34
H   m ,
may be calculated as
1  m r'  1 1 1
 m r    d 3 r'      Mr'  r  r' d
3
r' . (***)
4 V
 0 r  r' 4 V

As was also discussed in Sec. 5.6, if M(r) = M0 = const inside the volume V, and M(r) = 0
outside it, the effective magnetic charge density m = –0M is different from zero (and formally
infinite) only on the surface S of the volume. Hence it may be characterized by the effective surface
charge density m, which may be calculated by the integration of m along the local coordinate n normal
to the surface, from a point just inside the surface to a point just outside it (i.e. out of the volume V):
ns 0 ns 0 ns 0
M n
m   m dn    0    M dn     0
n
dn   0 M n n 0   0 M 0 cos  ,
s
ns 0 ns 0 ns 0

where  is the angle between the direction of the vector M0 and the outer normal n to the surface.35 This
means that the effective surface charge is positive at the points where the vector M0 is directed out of the

34 As a (hopefully, unnecessary) reminder, this expression is only valid in the absence of a vortex-like field
induced by stand-alone currents – the condition satisfied in our case of the field induced by magnetization M(r).
35 Admittedly, this is just another form of Eq. (5.134) of the lecture notes.

Problems with Solutions Page 191

Essential Graduate Physics EM: Classical Electrodynamics

magnetic material, negative in the opposite case, and equals zero at the m  0
points where M0 is parallel to the surface – see the figure on the right. m  0 
M0
In this case, Eq. (**) takes the form
M0 r  r' 2
H r    cos r'  d r' , (****)
4 S r  r'
3

 0
and Eq. (***), the form m  0 m

M0 1
m r    cos r'  d 2 r' .
4 S r  r'

Problem 5.25. Use the results of the previous problem to

M0 R
calculate the distribution of the magnetic field H along the axis of a
l
straight permanent magnet of length 2l and a round cross-section of
radius R, with a uniform magnetization M0 parallel to the axis – see l 0 l z
the figure on the right.
Solution: Due to the axial symmetry of the problem, the magnetic field on the axis has to be
directed along it: H = nzHz. This is why it is sufficient to use the z-component of Eq. (****) of the model
solution of the previous problem with cos = 1:
M0 z  z'
H z z     d 2 r' .
4 S r  r'
3

where the integration is extended only over the magnet’s flat ends. Taking the magnet’s center for z = 0
(so the positions of its ends are z = l, as shown in the figure above), we get

M0  R
2d
R
2d 
H z z     z  l    z  l  ,
4  
0 z  l   
2 2
 3/ 2
0  z  l  2
 
2 3/ 2

where  is the distance of the source point r’ from the symmetry axis. These two integrals may be
readily worked out by similar substitutions:   (z  l)2 + 2, so in both cases 2d = d:

M0   d 
( z l ) 2  R 2 ( z l ) 2  R 2
 d 
H z z    z  l 
 2  3/ 2   z  l   2  3 / 2 .
4  ( z l ) ( z l )
 
M  z  l  z  l  
 0   sgn  z  l    sgn  z  l .
 
2   z  l 2  R 2 1
/ 2

 z  l 2  R 2 1/ 2


This function is plotted in the figure below for three representative values of the R/l ratio. Note,
first of all, the reversal of the direction of the field H at each end of the magnet – the result which should
not be too surprising after looking at Eq. (5.134) of the lecture notes and the solution of the very similar
Problem 23 for a sphere. Note that the magnitude of H is always below M0 at all points, so the field B =
0(H + M) is still directed along the magnetization vector: Bz > 0 for all z.

Problems with Solutions Page 192

Essential Graduate Physics EM: Classical Electrodynamics

At R << l, H is nearly a sum of Coulomb- 1

like fields of two equal and opposite point
“magnetic charges” qm = 0M0A = 0M0R2. In
0.5
the opposite limit, the field inside the magnet is
H z z 
nearly constant: Hz  –M0, while outside it is very
M0 R / l  0.1
small, and drops further only at distances z ~ R 0
>> l. In one more limit, z >> R, l, the result is 1
reduced to
M 0 R 2l  0.5
Hz  ,
z3 10
M 0 R 2 l  0 2M 0 2R 2 l  1
Bz   0  . 2 1 0 1 2
z3 4 z3 z /l
This is just the field (5.99) of a magnetic dipole, with the dipole moment m = M0V, where V =
2R2l is the magnet’s volume, on its axis (where r = nzz, m = nzm).

Problem 5.26. A flat end of a long straight permanent magnet, similar to that considered in the
previous problem but with an arbitrary cross-section of area A, is stuck to a flat surface of a large sample
of a linear magnetic material with a very high permeability  >> 0. Calculate the normally directed
force needed to detach them.
Solution: As was discussed in Sec. 5.6 of the lecture notes, the fields inside ultra-high-
materials are well described by the approximation H = 0, so according to the definition (5.120) of the
scalar magnetic potential, we may take m = 0 inside its bulk and on its surface, forming the boundary
condition for the distribution of m outside of the sample. But this condition, and Eq. (5.121) governing
the distribution, are completely similar to that for the distribution of the electrostatic potential  outside
a similarly shaped conductor.
As we know very well (see, e.g., Sec. 2.9), such a distribution, due to a stand-alone point electric
charge q at a distance d from the plane surface of a conductor, is described by the conductor’s
replacement with the additional charge q’ = –q located at the same distance d from the surface’s plane,
but on the opposite side of it. Hence, the same is true for the magnetic field of each effective magnetic
charge qm outside a flat surface of a high- sample. This fact may be used to calculate the magnetic field
distribution in many problems with various systems of charges qm.
As was discussed in the same Sec. 5.6, in our current case of a flat end of a permanent magnet
with a constant magnetization M0 normal to the end’s surface, the effective magnetic charges form a
uniform sheet of area A, with the areal density m = 0M0. Hence the magnetic image charge is
distributed on a similar sheet, on the other side of the surface, with density –m. But this situation is
exactly similar to the attraction of two opposite poles of permanent-magnet rods – see Fig. 5.18 and its
discussion. Hence the force of attraction between a permanent magnet and a linear high- magnetic
material (and the force necessary for their detachment) is given by the same Eq. (5.136):
 0 M 02 A
F  .
2

Problems with Solutions Page 193

Essential Graduate Physics EM: Classical Electrodynamics

This also means that the crude estimate of this force given in the lecture notes,  F /A ~ 4105 Pa,
is also valid for typical refrigerator magnets because the refrigerator walls are usually made of steel, and
the magnetic permeability of most varieties of this material satisfy the requirement  >> 0 well. (The
main exception are the so-called stainless steels – alloys of iron and more than 10% chromium, which is
almost non-magnetic, with   0. However, such steels, broadly used in physical experiment because
of their unique chemical and thermal properties, are relatively expensive and hence rarely used outside
of special application fields.)

Problem 5.27. A permanent magnet with a uniform magnetization M0 has the form of a spherical
shell with an internal radius R1 and an external radius R2 > R1. Calculate the magnetic field inside the
shell.
Solution: According to Eqs. (5.134) and (5.137) of the lecture notes (see also the solution of
Problem 24), an abrupt leap of magnetization on the external surface of the shell is equivalent to the
effective magnetic charge with the areal density m = 0M0cos, where  is the angle between the radius
vector r of the point and the direction of the magnetization vector M0. As we know from the solution of
Problem 2.28, a similar distribution,  = 0cos, of the electric charge density on a spherical surface of
radius R would induce, inside this surface, a uniform electric field E of magnitude
0
E , for r  R,
3 0
directed opposite to the ray  = 0. But due to the similarity of the Poisson equations (1.27) and (5.133)
for the electric and magnetic fields,36 with the replacement
   
 m    M, i.e.  m,
0 0 0 0
we may immediately translate this result of electrostatics to magnetostatics, giving us the following
uniform magnetic field induced by the external surface of our shell:
M0
H2  , for r  R2 . (*)
3
For a uniform sphere of the radius R2, this would be the final result; however, in our current case
of an empty shell, the abrupt back leap of the magnetization on its inner surface gives an equal and
opposite field component
M
H1  0 , for r  R1 ,
3
so inside the shell, the net field vanishes:
H  H1  H 2  0, and B   0 H  0, for r  R1  R2 .

Note, however, that the field inside the shell’s material, i.e. at R1 < r < R2, does not vanish, being
a sum of a purely dipole field of its inner surface and the uniform field (*) of the external surface.

36 This similarity was already discussed in detail in the model solution of Problem 24.

Problems with Solutions Page 194

Essential Graduate Physics EM: Classical Electrodynamics

Problem 5.28. A very broad film of thickness 2t is permanently magnetized normally to its plane,
with a periodic checkerboard pattern, with the square of area aa:
 x y 
M z  t  n z M  x, y ,
with M  x, y   M 0 sgn  cos cos  .
 a a
Calculate the magnetic field’s distribution in space.37
Solution: Due to the absence of stand-alone currents in the system, we may describe the magnetic
field in it by using the scalar potential m defined by Eq. (5.120) of the lecture notes:

H   m . (*)
According to Eq. (5.121), in each of the three uniform z-regions (z < –t, –t < z < +t, and t < z), the
potential has to satisfy the Laplace equation. Taking into account the symmetry of the problem (which
requires in particular that H(–z) = H(z), so if m(0) is set to zero, then m(–z) =m(z)), and its aa –
periodicity in the x-y plane, it is natural to look for the solution in the following variable-separated form
(very similar to that in Eq. (2.95) of the lecture notes):

nx my sgn z  exp  nm z / exp  nm t, for z  t ,
 m  x, y , z    c nm cos cos  (**)
n , m 1 a a  sinh  nm z sinh  nm t , for z  t.

(The denominators in the last operands are set up to have the first boundary condition at the film’s
surfaces, the potential’s continuity, satisfied for any choice of the coefficients cnm.) Indeed, the
substitution of any term of this series to the Laplace equation shows that it is satisfied if the separation
constants nm obey the relation similar to Eq. (2.93):

 nm 
a
n 2
 m2 
1/ 2
. (***)

According to Eq. (*), the distribution (**) corresponds to the following normal component of the
magnetic field on the film’s top surface:38

 m 
nx my 1, for z  t  0,
H z  x, y , t  0      nm nmc  cos cos  
z z t 0 n ,m 1 a a  cosh  nm t / sinh  nm t , for z  t  0.

On the other hand, Eq. (5.134) applied to the film’s surface at z = t, gives us the following second
boundary condition:
H z  x, y , t  0   H z  x , y , t  0   M  x , y  ,

so, for our particular magnetization pattern M(x, y), the coefficients cnm have to be found from the
following system of equations:

nx my  cosh  nm t  x y 
c  nm cos cos 1    M 0 sgn  cos cos  .
sinh  nm t 
nm
n , m 1 a a   a a 

37 This problem is of evident relevance for the perpendicular magnetic recording (PMR), which presently
dominates high-density digital magnetic storage technology.
38 The z-asymmetry of the solutions (**) takes care of the boundary condition at z = –t automatically.

Problems with Solutions Page 195

Essential Graduate Physics EM: Classical Electrodynamics

As usual, let us multiply both parts of this equation by a function from the same orthogonal set,
in this case, cos(n’x/a)cos(m’y/a) with arbitrary integer n’ and m’, and then integrate them over one of
the x-y periods of the structure, say over the square 0  x, y  a. On the left-hand side, only terms with n
= n’ and m = m’ survive, and we get39
 cosh  nm t  a 2 a a
nx my  x y 
c nm  nm 1    M 0  dx  dy cos cos  sgn  cos cos 
 sinh  nm t  4 0 0
a a  a a
4a 2 n m
 M 0 2 sin sin ,
 nm 2 2
so the coefficients cnm depend on a only implicitly, via nm:
16 sinh  nm t n m
c nm  M0 sin sin . (****)
 nm
2
 nm sinh  nm t  cosh  nm t  2 2
(Note that the last two factors make cnm nonvanishing only if both integers n and m are odd, and also
provide its sigh alternation for even n and m, speeding up the convergence of the series (**).)
Formulas (*)-(****) give a complete solution of the field distribution problem. For example, the
left panel in the figure below shows the resulting Hz as a function of z at the center of one of the
checkerboard squares (say, x = y = 0), for several values of the film thickness.

1 0.5
z t
x y0  0.05 y0
t 0.15
t / a  0.5 ta
0.25 0.3
Hz Hz
0.1
M0 M0
0 0

1  0.5
1 0 1 1 0 1
z/a x/a
The jumps of the field by M0 at the film surfaces, and the resulting reversal of its sign,
qualitatively similar to that in the previous problem, are clearly visible. From the point of view of digital
magnetic recording application, these plots imply that to avoid a significant drop of the magnetic-field
“readout signal”, the film’s half-thickness t should be larger than at least ~0.3a, where a2 is the recorded
bit area.

39The integration on the right-hand side is easy using the fact that within our integration area, the sgn function is
equal to (+1) if both x and y are either between 0 and a/2, or between a/2 and a, and to (–1) in the complementary
regions.

Problems with Solutions Page 196

Essential Graduate Physics EM: Classical Electrodynamics

The right panel of the figure shows Hz as a function of x at y = 0, for several values of z, for a
sufficiently thick film (t = a). Both panels show that in order to avoid a substantial signal loss, the
magnetic field sensing device (in modern technology, a GMR sensor) has to be rather close to the film’s
surface. Indeed, the distance (z – t) = 0.3t already causes more than a two-fold drop of the field’s swing
from its maximum value M0.

Problem 5.29.* Based on the discussion of the quadrupole electrostatic lens in Sec. 2.4 of the
lecture notes, suggest permanent-magnet systems that may similarly focus particles moving close to the
system’s axis, for the cases when each particle carries:
(i) an electric charge,
(ii) no net electric charge, but a spontaneous magnetic dipole moment m of a certain orientation.
Solutions:
(i) As was discussed in Sec. 2.4, a quadrupole lens is a cylindrical system exerting the force F
with Cartesian components
Fx  cx, Fy  cy (*)

(where c is a constant), on a particle moving along the z-axis. The V / 2

electrostatic system of four hyperbolically-shaped electrodes, having this V V
property in all space between them, is shown in Fig. 2.9 of the lecture notes.  y 
2 2
However, it is clear that any cylindrical system of four electrodes
0 x
with a similar symmetry of their voltage biases, for example, the one shown
in the figure on the right, has the property (*) near its center (i.e. at x, y 
0). Indeed, due to the mirror symmetry of the system relative to the planes x V / 2
= 0 and y = 0, the electrostatic potential  at the central line x = y = 0 should
equal zero and have vanishing partial derivatives /x and /y (and, actually, all odd derivatives as
well). Hence the Taylor expansion of the potential at that point has to start from quadratic terms, with
equal but opposite coefficients for x and y:

  x, y    x 2  y 2   O 4 , at   x 2  y 2 
1/ 2
 0,

with  ~ V/a2, where a is the spatial scale of the system’s cross-section. (As a sanity check, the leading
term of this expansion40 satisfies the Laplace equation 2 = 0.) Now calculating the force F = qE = –
q exerted on a particle with an electric charge q, we get Eqs. (*) with c = 2q.
Now the analogy between the electrostatic field induced by surface electric charges, and the
magnetic field induced by permanent magnets with the magnetization normal to their axis-facing
surfaces (see, e.g., the discussion following Eq. (5.132) of the lecture notes and the model solutions of
the four previous problems) implies that a similar distribution of the magnetic potential,

 m  x, y    m x 2  y 2 , at   0 , (**)

40 Note that this term coincides with the exact potential distribution (2.75) in the hyperbolic-electrode system.

Problems with Solutions Page 197

Essential Graduate Physics EM: Classical Electrodynamics

with m  M0, may be obtained, for example, in the system shown in the
M0
figure on the right. The magnetic field in the free space between the
magnets, corresponding to this potential, equals B = –0m, so its y X
Y
Cartesian components are
M0 M0
B x  2 0 m x, B y  2 0 m y . 0 x

As a result, the Cartesian components of the magnetic Lorentz force F =

qvB exerted on a charged particle, moving along the z-axis, are M0
Fx  qv z B y  2  0 qv z  m y, Fy  qv z B x  2 0 qv z  m x .

This result is different from Eqs. (*) not only by the signs but also by field components’
proportionality to different coordinates. However, relative to the Cartesian coordinates {X, Y} turned by
angle /4 relative to {x, y}, the force has the required properties. Indeed, writing the relations between
the two pairs of coordinates, which are evident from the figure above,
x y x y X Y X Y
X  , Y ; so x  , y ,
2 2 2 2
we may use them to recast Eq. (**) as

 m   m x 2  y 2  
m
2
 X  Y  2

  X  Y   2 m XY ,
2

so
B X  2 0 mY , BY  2 0 m X , (***)
and the expressions for the Lorentz force components,
FX  qv z BY  2 0 qv z  m X , FY  qv z B X  2 0 qv z  mY ,
have the same functional form as in Eq. (*), with the coefficient c = 20qvzm.
(ii) According to Eq. (5.102) of the lecture notes, the force exerted on a magnetic dipole m in the
external magnetic field B is
F   m  B  . (****)
For the quadrupole magnetic field (***), this force is coordinate-independent, i.e. does not have the
required focusing properties. A clue in the search for a suitable system may be obtained by rewriting Eq.
(**) in the complex form:
 m   m Rez 2 , with z  x  iy  e i ,

where  is the polar angle on the [x, y] plane. Now let us consider a different magnetic potential

 m  Rez 3     3 cos3    ,
where  is some complex constant, and  is its argument. Such a function provides the extra power of z,
necessary to make the force (****) proportional to the particle’s deviation from the system’s axis.
Indeed, calculating the radial component of the force (****) with m = const,41 we get

41 See, e.g., MA Eq. (10.2) – to be used twice.

Problems with Solutions Page 198

Essential Graduate Physics EM: Classical Electrodynamics

F  6 0     m  cos3     m sin 3   .

This expression shows that the component is indeed proportional to the distance  of the particle’s
trajectory from the system’s axis, so it does provide beam focusing, with the focal distance depending
both on the angle  and the dipole moment’s orientation.
In addition, according to the discussion in Sec. 2.4 of the lecture notes, any analytic complex
function f(z), in particular z3, does satisfy the Laplace equation – and so does its real part. Hence, the
above distribution m(, ) may be obtained using some permanent M0
magnet system. The form of this distribution implies that the system
should be symmetric with respect to the rotation by  = 2/3, rather than M M0
0
by   2/2 for the quadrupole system shown in the figure above, i.e.
needs 3 rather than 2 pairs of magnets of alternating polarity – for
example as shown in the figure on the right. v
Such sextupole (also called “hexapole”) magnet systems are
actually used in experiment, in particular, for focusing beams of neutrons M 0 M0
with their non-zero dipole magnetic moments.
M0

Problems with Solutions Page 199

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 6. Electromagnetism

Problem 6.1. Prove that the electromagnetic induction e.m.f. Vind in a conducting loop may be
measured as shown on two panels of Fig. 6.1 of the lecture notes:
(i) by measuring the current I = Vind/R induced in the loop closed with an Ohmic resistor R, or
(ii) using a voltmeter inserted into the loop.
Solutions:
(i) The issue is not quite trivial, because the resistance is defined (see Sec. 4.2) as the ratio V/I,
where V is a charge-induced voltage, in its turn defined as the difference of electrostatic potentials  at
the resistor’s ends – see Eq. (2.25). On the other hand, for the vortex field Eind of electromagnetic
induction, such representation is generally impossible. Moreover, the Faraday law does not prescribe a
unique spatial distribution of the field; rather it only specifies its contour integral (6.2) or, equivalently,
the field’s curl at each point – see Eq. (6.5). The exact particular distribution Eind(r) depends not only on
that the time evolution of the inducing magnetic field B(r, t), but also on the geometry and properties of
the system.
Note, however, that for the “usual” (charge-induced, potential) field E, we may represent the
voltage drop at a resistor as the integral
B
V   E  dr
A

along any path connecting the resistor’s terminals A and B. Now using the differential form (4.8) of the
Ohm law, we may write
jr   dr
B B

A E  dr  A  r  .
The results of Sec. 4.3 show that the normalized law,
jr 
f r   , (*)
I
of the dc current distribution in a conductor is uniquely defined by Eq. (4.6), j = 0, equivalent to f =
0, with the boundary conditions fn = 0 on the conductor’s surface, and f = 0 on its interfaces with the
external electrodes, so we may use Eq. (*) to express the above integral as
f r   dr
B B

 E  dr  IR,
A
with R  
A
 r 
, (**)

the last equality giving an equivalent definition of the sample’s resistance.

Now moving on to electromagnetic induction, we may argue that since the current distribution
law f(r), and hence the resistance R, are governed by relations including j alone, they do not depend on
the nature of the applied electric field. Hence in the particular case of a conductor, it is its geometry that
dictates the field’s distribution inside it:

Problems with Solutions Page 200

Essential Graduate Physics EM: Classical Electrodynamics

jr  I
E ind   f r  , (***)
 r   r 
leaving Eq. (6.2) to govern only the total magnitude I of the current. (This internal distribution, in turn,
dictates that of the field Eind(r) outside the conductor, following the conductivity hierarchy that was
discussed in Sec. 4.3.)
Now let us consider a closed resistive loop that may be represented as the result of a merger of
the endpoints A and B. Though such a closed loop does not have external electrodes, the condition j = 0
discussed above is still valid on an arbitrary cross-sectional surface normal, in each point, to the vector j.
(Such surface may be understood as a galvanic connection of two ultimately thin external electrodes.)
Hence, all the above relations are valid for such a closed loop as well, so integrating Eq. (***) along it,
and then using Eq. (**), we get
f r   dr
Vind   E ind dr  I   IR ,
C C
 r 
thus giving the required proof of the relation I = Vind/R.
(ii) The prevailing electrodynamic species of voltmeters42 are essentially sensitive galvanometers
that measure the weak current I induced inside them by the voltage under measurement, recalibrated into
the voltmeter readout as
V  Rint I ,
where Rint is the voltmeter’s internal resistance. As it follows from the solution of Task (i), for our
system (the voltmeter inserted into the loop) we may write
Vind
I ,
Rint  Rext
where Rext is the resistance of the loop itself. Combining the two displayed formulas, we get
Rint
V Vind .
Rint  Rext
This result shows that if the internal resistance of the voltmeter is appropriately high, Rint >> Rext,
it indeed measures the induced e.m.f. faithfully. The situation, however, changes if a voltmeter does not
interrupt the induction loop, but rather is connected in parallel with its part – see, for example, the next
problem.

A
Problem 6.2. The flux  of the magnetic field that pierces a
resistive ring is being changed in time, while the field outside of the C C2
1
ring is negligibly low. A voltmeter is connected to a part of the ring, Φ(t ) V
as shown in the figure on the right. What would the voltmeter show?
Solution: A naïve application of the Faraday induction law B
(6.2) to contour C1 (shown with the external dashed line in the figure

42The alternative, electrostatic voltmeter species, will be briefly discussed in SM Sec. 6.5. Those voltmeters are
typically slow and not used for the measurement of such transient effects as electromagnetic induction.

Problems with Solutions Page 201

Essential Graduate Physics EM: Classical Electrodynamics

above) would give

V   E ind  dr  
. [WRONG!]
C1
The fallacy of this approach becomes evident if we notice that it might be equally applied to contour C2,
giving a totally different answer:
V   E ind  dr  0 . [WRONG!]
C2
The reason for these errors is the lack of account for the Ohmic voltage drops on the current-carrying
parts of the ring.
The simplest way to get the correct answer is to use the equivalent lumped circuit language,
which is broadly used in electrical engineering and was briefly discussed in Secs. 4.1 and 6.6 of the
lecture notes. First, note that in the absence of the voltmeter, the system is similar to that analyzed in the
previous problem, so according to its solution, the current Ir in the ring may be expressed as
Vind Vind
Ir 
R

R1  R2
, with Vind  E
ring
ind  dr ,

where R1 and R2 are the resistances of the two parts of the ring between the contact points A and B, i.e.
running along the contours C1 and C2, respectively. This result may be obtained from the equivalent
circuit shown on panel (a) of the figure below, where each circle means a perfect voltage source, i.e. an
imaginary two-terminal device that maintains the voltage equal to the specified e.m.f. (in this case, Vind)
between its terminals, regardless of the external circuit – see the model solution of Problem 4.2.
(a) (b) C1 (c)
Vind V1 V2 V1 V2 V2
    

    C2 
R1 R2
R1 R2 R1 R2 Rint

Ir
Ir Ir Iv
Since the integral defining Vind may be always broken into two parts, corresponding to the two
fragments of the ring,
B A
Vind  V1  V2 , where V1   E ind  dr,
A
V2  E
B
ind  dr ,
(along C1 ) (along C2 )
the equivalent circuit (a) may be transformed into another one, shown on panel (b). The latter circuit
gives the same current as the former one (hence the term “equivalent”) but has the advantage that it may
be readily generalized to the case of the voltmeter connected between points A and B – see panel (c).
The only nontrivial component of the last circuit is the additional source of the same e.m.f. V2 in the
voltmeter branch, which is necessary to satisfy the Faraday induction law for the contour C2:

E ind  dr  V2  V2  0 .
C2

Problems with Solutions Page 202

Essential Graduate Physics EM: Classical Electrodynamics

Note that there is nothing unphysical in that compensating e.m.f. source, because the induced
electric field Eind is not localized in the magnetic field’s region (only its curl is – see Eq. (6.5) of the
lecture notes), and may extend to the whole space. (For example, if the ring and the magnetic field
distribution in our current problem are axially symmetric, so should the field Eind,43 and Eq. (6.2) is
satisfied with the following solution

Φ
E ind   n ,
2

where  is the distance from the system’s axis, and the sign depends on the magnetic field’s direction.)
This field extends to infinity and induces the e.m.f. not only in the ring but also in the voltmeter’s wires,
so in contour C2, they exactly compensate each other.
Now writing the usual Kirchhoff laws for the circuit (c) (see the discussion in Secs. 4.1 and 6.6)
and solving the resulting simple system of equations for the voltmeter current Iv, we get a somewhat
bulky result, which, in the most important limit Rint >> R1,2 (indeed, only in this case, a galvanometer
may be called a voltmeter) reduces to the simple form44
Vind R2
Iv  , for Ri  R ,
Rint R1  R2
so the voltmeter readout is
R2
V  I v Rint  Vind  Vind .
R1  R2
Note that if the voltmeter was connected to the same points A and B, A
but physically located on the other side of the ring (see the figure on the right),
the measurement result would be different:
V' Φ(t )
R1
V'   Vind , B
R1  R2
the minus sign being valid if the positive and negative terminals of the
voltmeter are still connected to the same points of the ring. This sign difference enables the popular and
spectacular lecture demonstration, in which two voltmeters placed on opposite sides of an induction ring
and connected simultaneously to the same points with the same polarity, show readouts of opposite
signs.
Such experiments, as well as the above analysis, clearly show that at the Faraday induction (and
generally in electromagnetism) such a notion as “voltage between points A and B” depends on the way
of its measurement, and is not uniquely defined without such specification. This is a natural result of the
vortex nature of the induced electric field, which (in contrast to the electrostatic field) cannot be
represented as a gradient of a certain scalar function such as  – whose difference the voltage would be.

43 Strictly speaking, this symmetry is violated by the voltmeter’s wires; this perturbation may be minimized by
running wire fragments mostly along concentric circles.
44 Actually, the calculation may be much simplified by the replacement of the connection in series in each branch
with the parallel connection of the same resistor and a perfect current generator – see the model solution of
Problem 4.2.

Problems with Solutions Page 203

Essential Graduate Physics EM: Classical Electrodynamics

Problem 6.3. A weak constant magnetic field B is applied to a small, axially-symmetric

permanent magnet, with its dipole magnetic moment m directed along its axis and rapidly rotating about
the same axis, with an angular momentum L. Calculate the electric field resulting from the field’s
application, and formulate the conditions of your result’s validity.
Solution: According to Eq. (5.101) of the lecture notes, the magnetic field exerts the torque
τ  mB
on the dipole. Per the basic law of rotational dynamics,45 the torque makes the dipole’s angular moment
evolve as L  τ . Since, per our problem’s conditions, the vectors L and m maintain the same direction
(along the symmetry axis of the magnet), we may write m = L.46 As a result, the dipole moment
changes in time as
  L  τ  m  B .
m (*)

According to basic kinematics,47 this equation describes an additional, relatively slow rotation
(the so-called torque-induced precession) of the vector m, and hence of the magnet’s symmetry axis,
about the direction of the magnetic field’s vector B, with the angular velocity
Ω  B , (**)
thus maintaining the angle  between the vectors m and B at its initial value. Directing the z-axis along
the magnetic field, we may describe this rotation as

 
m  m z  m  t , with m z  n z m cos  , m  t   m sin  Re n x  in y  exp i t   0 , (***)

where 0 is the constant phase, which is also determined by the initial conditions (and the selected time
origin).
Due to the Faraday induction, this rotation induces, in the space around the magnet, an electric
field E changing in time periodically, with the same frequency . To calculate the field, we may use the
first of Eqs. (6.7) of the lecture notes, which, in the absence of an electrostatic potential , reduces to
A
E .
t
According to Eq. (5.90), the vector potential of the magnetic dipole’s field is
0 m  r
A , (****)
4 r 3
so the electric field at a fixed point r is
0 m
 r
E .
4 r 3
Now using Eqs. (*)-(****), we finally get

45 See, e.g., CM Eq. (1.33).

46 In atomic and nuclear physics, the scalar coefficient  is frequently a fundamental constant (see, e.g., Eq. (5.95)
of the lecture notes), called the gyromagnetic ratio. It will be repeatedly discussed in the QM part of this series.
47 See, e.g., CM Sec. 4.1, in particular Eq. (4.8).

Problems with Solutions Page 204

Essential Graduate Physics EM: Classical Electrodynamics

 0 B  m   r  0 n z  m  t  r  0 Ren y  in x  r expi t   0 

E   B  Bm sin 
4 r3 4 r3 4 r3
.
0 n x z  n z x  cost   0   n y z  n z y sin t   0 
 Bm sin  .
4 r3
So, the oscillation of any Cartesian component of the induced electric field in time is exactly
sinusoidal, and the amplitude of these oscillations drops with the distance as 1/r2. This result is only
valid in the quasistatic approximation i.e. at r << c/; as will be extensively discussed in Chapter 8 of
the lecture notes, at larger distances, Eq. (****) needs to be amended, leading to the electromagnetic
radiation of waves with their electric and magnetic fields proportional to 1/r at r >> c/.48
Another condition of quantitative validity of our result follows from the theory of torque-induced
precession of such “symmetric tops”:49 the applied field B should be sufficiently low to keep the
precession-induced addition I to the basic angular momentum L negligibly small:
m
I   L, i.e. B  ,
 2I
where I is the principal moment of inertia of the magnet’s rotation about any axis normal to its
symmetry axis.

Problem 6.4. The similarity of Eq. (5.53) obtained in Sec. 5.3 without any use of the Faraday
induction law, and Eq. (5.54) proved in Sec. 6.2 using it, implies that the law may be derived from
magnetostatics. Prove that this is indeed true for a particular case of a current loop being slowly
deformed in a fixed magnetic field B(r).
Solution: Consider a thin wire loop with current I, placed  (d 2 r )
in a magnetic field B – see the figure on the right. According to
Eq. (5.21) of the lecture notes, the magnetic force exerted by the dr
field upon a small fragment dr of the wire is I
r B
dF  I (dr  B)   I B  dr  ,
S
where dr is the vector tangential to the loop’s contour and
directed along the current I. Now let the wire be slightly (and C
slowly) deformed so that this particular fragment is displaced by
a small distance r. (Let me hope that the figure above makes the difference between the elementary
vectors dr and r absolutely clear.)
In order to keep the wire’s acceleration (and hence the kinetic energy of the system) negligibly
small, some other external forces should balance the force dF, providing an equal and opposite force.

48See, in particular, Eqs. (8.124)-(8.126). Note that in quantum mechanics, the torque-induced precession as
such does not lead to radiation; only quantum transitions between the quantized precession states (with
different quantized values of mz = mcos ) lead to radiation with the frequency  – see, e.g., QM
Chapter 9. However, for the values of L much larger than Planck’s constant, the final conclusions of the
classical and quantum theories coincide.
49 See, e.g., CM Sec. 4.5 – in particular, Eqs. (4.73) and (4.85) and their discussion.

Problems with Solutions Page 205

Essential Graduate Physics EM: Classical Electrodynamics

With the account of this opposite direction, the work of these external forces at the displacement r, and
hence the change of the magnetic potential energy U of the system, is
 (dU )   dF  r  Ir  B  dr  .
Let us apply to this mixed product the operand rotation rule of the vector algebra50 in such a way that the
vector B comes out of the vector product:
 (dU )  IB  dr  r  . (*)

But the magnitude of the vector product in the parentheses is nothing more than the area (d2r)  (dA)
swept by the wire’s fragment at the deformation (see the figure above again), while its direction
coincides with the unit vector n = (dr/dr)(r/r) normal to this elementary area dA. The scalar
multiplication of the vector B by this unit vector is equivalent to taking the B’s component Bn along this
normal. Hence, integrating Eq. (*) over all the wire length, we get the following result for the variation
of the total magnetic energy of the system:
U  I  Bn (d 2 r ) .
C

If B is fixed, i.e. does not change at the loop’s deformation, the variation sign may be moved out from
the integral, and we get
U  I , (**)
where  is the magnetic flux through the loop.
Now let the work W necessary for this energy change come from a generator of an external
voltage Vext, inserted somewhere in the loop. In order for the system to stay in quasi-equilibrium during
the slow deformation of the loop, this voltage should counter-balance the electromagnetic induction’s
e.m.f.: Vext = –Vind. The work of this voltage at the transfer of charge Q = It during a small time
interval t is
W  Vext Q  VindQ  Vind It .

Requiring this work to be equal to the potential energy’s change (**) it causes, we arrive at the Faraday
induction law (6.2) – for this particular case only. As was discussed in sec. 6.1 of the lecture notes, the
law is actually much broader.

Problem 6.5. Could Problem 5.2 (i.e. the semi-quantitative analysis of

the mechanical stability of the system shown in the figure on the right) be I1
solved using potential energy arguments?
I2
Solution: Since in this problem, the currents are assumed to be fixed,
the stable equilibrium of the system corresponds to the minimum of the Gibbs potential energy given by
Eqs. (6.17) and (6.20) of the lecture notes. In our magnetic-material-free case ( = 0), it reads51

50See, e.g., MA Eq. (7.6).

51 Note that the sign in this expression (and hence the conclusions of this analysis) would be opposite if not
currents in the loops but individual magnetic fluxes in each of them were fixed, as they may be if the loops are
superconducting – see, e.g., Sec. 6.5 of the lecture notes.

Problems with Solutions Page 206

Essential Graduate Physics EM: Classical Electrodynamics

1
UG   B
2
d 3r . (*)
2 0
Using the linear superposition principle, we may represent the vector B at each point as the sum B1(r) +
B2(r), where each field is proportional to the corresponding current. Plugging this expression for B into
Eq. (*), we may rewrite it as
U G  U 1  U 2  U int ,
where Uk  Ik2, and only the last term
1
U int    B r   B r d
3
r, (**)
0 1 2

describing the energy of magnetic interaction of the loops, depends on their mutual orientation.
This energy is evidently smallest (and hence corresponds to the system’s equilibrium) if the
directions of the fields B1 and B2 coincide in the regions where they are strongest, namely near the
common center of the loops.52 In particular, any turn of the inner loop by angle  from the common axis
decreases the magnitude of Uint by a factor ~cos and hence increases the energy, so the stable
equilibrium corresponds to  = 0. (In the limit of a relatively small radius of the internal loop, when it
may be well approximated by its dipole moment m (5.97) whose direction coincides with that of the
field B1 at the center of the ring, this conclusion is directly confirmed by Eq. (5.100) of the lecture
notes.)
Similarly, if the inner loop is displaced, from the plane of the outer loop, in either direction along
the common axis, the “field overlap” integral in Eq. (**) decreases, so the stable equilibrium
corresponds to its coplanar position.
However, the analysis of the system’s stability (or rather instability :-) with respect to the lateral
mutual displacement of the rings, by using the energy arguments, is somewhat less apparent than that
using the interaction force arguments – see the model solution of Problem 5.2.

Problem 6.6. Use energy arguments to calculate the pressure exerted by the magnetic field B
inside a long uniform solenoid of length l and a cross-section of area A << l2, with N >> l/A1/2 >> 1
turns, on its “walls” (windings), and the forces exerted by the field on the solenoid’s ends, for two cases:
(i) the current through the solenoid is fixed by an external source, and
(ii) after the initial current setting, the ends of the solenoid’s wire, with negligible resistance, are
connected, so it continues to carry a non-zero current.
Compare the results, and give a physical interpretation of the direction of these forces.
Solutions:
(i) If the current in the solenoid is maintained by an external current source, its equilibrium
corresponds to the minimum of the total potential energy of the system (solenoid + source), i.e. of the

52Strictly speaking, the individual fields B1 and B2 are the largest in close vicinities of the corresponding wires,
but since here the field lines are closed circles, i.e. the regions with equal and opposite values of the scalar product
B1B2 are virtually equal in volume, they give a negligible net contribution to Uint.

Problems with Solutions Page 207

Essential Graduate Physics EM: Classical Electrodynamics

Gibbs energy UG of the solenoid. In a long solenoid (of length l >> A1/2), with dense winding (N >>
l/A1/2 >> 1), the internal field’s H magnitude is given by Eq. (5.141):
NI
H ,
l
While outside the solenoid, it is virtually zero outside the solenoid. Hence we may use Eq. (6.20) to get
B2 H 2 N 2 I 2 A
UG   V  Al   . (*)
2 2 2l
This expression shows that to minimize UG (i.e. to maximize its magnitude), the field “tries” to
increase the area A of the solenoid and decrease its length l. These effects may be quantified by the
calculation, respectively, of the pressure on the “walls” (windings),53
U G 1 U G N 2 I 2 B 2
P  l  const    , (**)
V l A 2l 2 2
and the effective force applied to its ends:
U G N 2 I 2 A B2
F     A. (***)
l 2l 2 2
The directions of these forces may be readily interpreted using the basic Eq. (5.1). Indeed, the
lateral force on a current-carrying “wall” of the solenoid is dominated by
the oppositely-directed current in the opposite wall, corresponding to their P I
repulsion. (This is especially apparent if the cross-section has an elongated
shape – see the figure on the right.) On the other hand, the force in the
direction along the solenoid’s axis, exerted on each wire turn, is dominated P I
by its adjacent turns, with the same current direction, corresponding to the attraction of the turns.
(ii) If the solenoid is disconnected from the current source, its equilibrium corresponds to the
minimum of its own potential energy, U, which may be calculated using Eq. (6.15):
B2 N 2 I 2 A
U V  .
2 2l
This expression differs from Eq. (*) only by the sign, and it may look that now the results (**) and (***)
would be opposite. However, this is not so. Indeed, if the wire’s resistance is negligible, it cannot sustain
any voltage, in particular Faraday’s e.m.f. (6.2) in the closed loop formed by the connection of the
wire’s ends, and hence the magnetic flux  in the loop, rather than the current I, remains constant at the
solenoid’s shape variations. (As was discussed in Sec. 6.4-6.5, if the wiring is superconducting, this flux
conservation is perfect, but even in realistic large, high- magnets with “normal” wires, the flux
relaxation time  = L/R may be in hours, much longer than needed for force measurements.)
Now using Eq. (5.143) of the lecture notes for the self-inductance L of the solenoid,
 N 2 A
L   n 2 Al  ,
I l

53 As will be shown in Sec. 9.8 of the lecture notes, the last expression is very general.

Problems with Solutions Page 208

Essential Graduate Physics EM: Classical Electrodynamics

we may eliminate the current from the above expression for U:

LI 2  2  2l
U   .
2 2 L 2N 2 A
From here, we may readily calculate the pressure on the walls as
U 1 U 2 B2
P  l  const    ,
V l A 2N 2 A 2 2
and the longitudinal force as
U 2 B2
F     A,
l 2 N 2 A 2
getting exactly the same results as given by Eqs. (**) and (***).
Note that this problem, for a system with a uniform magnetic field (a long solenoid), is a
virtually exact analog of Problems 3.27-28 for systems with a uniform electric field (planar capacitors),
with a similar general conclusion: the (correctly calculated :-) forces do not depend on whether the
system is connected to an external source or not, despite using different potential energies for the
analyses of these cases. Let me hope the solutions of these three problems make the notion of the Gibbs
potential energy, and the conditions of its application, even more clear.

Problem 6.7. The electromagnetic l

railgun is a projectile launch system
V I
F, v
consisting of two long parallel conducting  (a)
rails and a sliding conducting projectile

shorting the current I fed into the system by a I
powerful source – see panel (a) in the figure v t (b)
of the right. Calculate the force exerted on the w
projectile, using two approaches:
(i) by a direct calculation, assuming that the cross-section of the system has the simple shape
shown on panel (b) of the figure above, with t << w, l, and
(ii) by using the energy balance (for simplicity, neglecting the Ohmic resistances in the system),
and compare the results.
Solutions:
(i) Due to the condition t << l, w, the magnetic field induced by the shorting currents in the
projective itself is negligible in comparison with that induced by the current in the rails. As was
calculated in the solution of Problem 5.4, the latter magnetic field is directed normally to the rails, and
its magnitude, in a particular cross-section of the projectile, is
I x 
Bx    .
w
Here x is the distance (along the rails) from one end of the projectile to its other end, so I(0) = I, and I(l)
= 0. According to Eq. (5.8) of the lecture notes, the force dF exerted by this field on a small fragment dx

Problems with Solutions Page 209

Essential Graduate Physics EM: Classical Electrodynamics

of the projectile, which carries the shorting current dI = J(x)dx (where J is its linear density), is directed
along the rails, always away from the current source (see the figure above), and has the magnitude
t
dF  B x tdI  J x B x tdx  J  x I  x dx ,
w
so the total force exerted on the projectile is
x l
t l
F  dF 
x 0
w  J x I x dx .
0

Since the current’s conservation requires that dI(x)/dx = –J(x), we get the following simple result,

t l dI  x  t  I 2  x  
x l x l
t t I 2
F   I  x dx    I x  dI x       , (*)
w 0
dx w x 0
w  2  x 0 w 2
valid regardless of the distribution of the shorting currents J(x) inside the projective (which may be due
to, for example, its nonuniform conductivity).
Note that according to another result of Problem 5.4 (in an evident way generalized to the case 
 0), the first fraction in the last form of Eq. (*) is just the self-inductance L0 of the rail system (or if
you like, the mutual inductance between the rails) per unit length, so our result may be represented in a
simpler form:
L0 I 2
F . (**)
2
As we will see in a minute, this simplicity is not occasional.
(ii) At negligible resistance of all components of the system, the only voltage V between the rails
at the current source’s location is the Faraday induction e.m.f. Vind = –d/dt. So, the elementary work
done by the source against the e.m.f. during an elementary time interval dt is dW = –VinddQ = –VindIdt =
Id. If the projectile is at a distance X >> l from the current source, the flux  = LI may be well
approximated as L0XI, so we get
dW  Id  Id L0 XI   L0 I 2 dX  L0 XIdI . (***)

On the other hand, the change of the magnetic energy (5.71) of the loop during the same time interval is
 LI 2   L0 XI 2  1
dU  d    d    L0 I 2 dX  L0 XIdI ,
 2
 2   2 
so Eq. (***) may be rewritten as
1
dW  dU  L0 I 2 dX .
2
Due to the conservation of the total energy of the system, the last term has to describe the mechanical
work dWm = FdX of the magnetic field on the projectile, so for the force F, we again get Eq. (**),
showing that this expression is independent of the particular cross-section of the system.
In the modern (so far, experimental) railgun systems, the (pulsed) driving current, typically
supplied by a capacitor battery’s discharge, may be as high as ~5106 A, so Eq. (**), with L0 ~ 0 ~ 10-6
H/m, shows that the force may be above 107 N, providing a projectile’s acceleration approaching ~106

Problems with Solutions Page 210

Essential Graduate Physics EM: Classical Electrodynamics

m/s2 (i.e. ~105 g!). Even much higher accelerations may be achieved in railguns with plasma projectiles,
which are being explored as possible reaction triggers in inertial-confinement nuclear fusion systems.

Problem 6.8. A uniform static magnetic field B is applied along the axis of a long thin pipe of
radius R and wall thickness  << R, made of a material with Ohmic  ,
conductivity . A sphere of mass M and radius R’ << R, made of a linear M , R'
R B
magnetic material with permeability  >> 0, is launched, with an initial
velocity v0, to fly ballistically along the pipe’s axis – see the figure on the v0
right. Use the quasistatic approximation to calculate the distance the sphere
would pass before it stops. Formulate the conditions of validity of your result.
Solution: According to Eqs. (5.125) of the lecture notes, taken in the limit  >> 0, the magnetic
field distortion created by such a sphere outside it coincides with the field of a point magnetic dipole
with the moment54
4R' 3
m B, (*)
0
positioned in the sphere’s center. In the quasistatic approximation, the vector potential created by the
dipole at a fixed point r may be found from Eq. (5.90) with the duly shifted origin:
 0 m  r  r' t 
Ar, t   ,
4 r  r' t  3
r
where r’ is the radius vector of the effective dipole, i.e. of the r  r'
R
sphere’s center. According to this formula, for our axially r' v
symmetric geometry (see the figure on the right), the vector z' (t ) m z
potential is purely azimuthal, A = An, and its magnitude at the
pipe’s wall is
 mR
A wall  0 .

4 z  z' t 2  R 2 3 / 2 
Now we may use Eq. (5.65) to calculate the magnetic flux through a cross-section of the pipe:

 0 mR 2 1
z, t    A wall  dr  2RA wall  .
R'
2 z  z' t  2
 R2 
3/ 2

According to Faraday’s induction law (6.2), the rate of change of this flux in time,

 z , t   0 mR 2   1  dz'  mR 2


 3z  z' t  
 v , (**)
  0
t
 
2 z'  z  z' t 2  R 2 
3/ 2  dt

2
 
 z  z' t 2  R 2 
5/ 2 


54 Note that the moment is automatically aligned with the applied magnetic field, so all following results are
robust with respect to the sphere’s rotation – unless it is so fast that a finite re-magnetization speed does not allow
m to follow its quasi-stationary value (*).

Problems with Solutions Page 211

Essential Graduate Physics EM: Classical Electrodynamics

(where v  dz’/dt is the velocity of the sphere) causes an equal but opposite e.m.f. Vind. Due to the axial
symmetry of our system, the induced electric field’s magnitude does not depend on the azimuthal angle:
Vind z, t    z , t  1
E ind r, t   n  Eind  z , t , with Eind  z , t    .
2R t 2R
In the thin wall of the pipe, this field creates circular eddy currents with the linear density
  z, t 
J  z , t   n  J z , t , with J  z , t   j  z , t   Eind  z , t    .
2R t
These loop currents, in turn, induce an additional axially symmetric magnetic field BJ. For any
point z” at the system’s axis, this field is directed along the axis, and its magnitude may be calculated by
integration, using Eq. (5.23) of the lecture notes with the proper replacement z  z” – z:
J  z , t dz

0 R 2
B J z" , t     z"  z  .
2 
2
 R2 
3/ 2

According to Eq. (5.102) of the lecture notes, this axially-oriented field exerts on the effective magnetic
dipole m = nzm of the sphere located at point z’(t), the drag force F = nzF with :55
B J  z" , t   R2 
 1
F m
z" z"  z' t 
m 0
2  z"  z"  z 

2
 R2  3/ 2 z"  z' t   J z, t  dz
3z  z' t  3z  z' t    z , t 

0 R 2  0 R 
m  z  z' t  J z , t dz  m   dz .
2 
2
R 
2 5/ 2 4   z  z' t   R
2
 2
5/ 2
t

Plugging in Eq. (**), we get56

 0 m 2 v R 3  z  z' 2t  2 5 dz   9  0 m 2  v 14 I , with I   2 5 d ,

 2 2 
9
F 
8 
  z  z' t   R
8  R    1  
where   [z – z’(t)]/R. The dimensionless integral I may be worked out, for example, using the
following variable replacement:  = tan, so d = d/cos2, while 1/(2 + 1) = cos2, giving:
 / 2  / 2  / 2
I  sin  cos  d   1  cos  cos  d   cos   cos 8  d .
2 6 2 6 6

 / 2  / 2  /2

Now the function under the integral may be represented as a sum of expressions proportional to power-
free trigonometric functions, using a repeated application of the well-known trigonometric identities57

55 Here the condition R’ << R, specified in the assignment, is essential.

56 An alternative way to obtain this expression for the drag force F is to use the above result for Eind(z, t) and Eq.
(4.39) of the lecture notes for the Joule power density, p = Eind2, to calculate the total power P as an integral of
p over pipe wall’s volume. After that, we may argue that P should be equal to the rate, –Fv, of the sphere’s
mechanical energy loss. The drawback of this approach is that it does not reveal the physical origin of the drag
force as explicitly as the calculation made above.
57 See, e.g., MA Eqs. (3.2d) and (3.3).

Problems with Solutions Page 212

Essential Graduate Physics EM: Classical Electrodynamics

1 1 3 1
cos 2 a  cos 2a, and cos 3 a  cos a  cos 3a .
2 2 4 4
The final result of this (a bit tedious but elementary) calculation is
5 1 1 1 1
cos 6   cos 8    cos 2  cos 4  cos 6  cos 8 .
128 32 32 32 128
At the integration over our interval –/2    +/2, all terms of this expression but the first one vanish,
so the integral is simply
5
I ,
128
and taking into account Eq. (*), the drag force F may be rewritten as58

 0 m 2 4 5  45 R' 3 B 2 4 .

2
9
F  v, with  
8 R 128 64 R
Since the force is proportional to the sphere’s velocity, the equation of its motion along the
pipe’s axis,
Mv  v ,

is linear, so it may be readily integrated to give an exponential law, v(t) = v0exp{-t/T}, with the time
constant T = M/. From here, the distance passed by the sphere tends to
 
 t Mv0 64 Mv0 R 4
d   vt dt  v0  exp dt  v 0T   . (***)
0 0  T   45 2 R' 6 B 2
Just to get some feeling of this result, a practicable 3-Tesla magnetic field would stop a 1-cm-radius
steel ball (with a mass ~0.3 kg), launched with the initial speed of 1 m/s into a copper pipe of a 10-cm
radius with 1-mm-thick walls after it has passed just d  87 cm.
There are two major conditions of validity of Eq. (***). First, our solution assumed that the
induced current j is uniformly distributed over the pipe wall’s thickness . This assumption is valid only
if  is much smaller than not only R but also the skin depth, s = (2/0)1/2, at the main frequency
components,  ~ v/R  v0/R, of the transient process of the current’s rise and fall, i.e. if
1/ 2
 R 
    . (****)
 v0 
The second condition is that the magnetic flux J of the field created by the induced current J
through the cross-section R2 of the pipe remains much smaller than , so it may be neglected (as it
was) at the calculation of the derivative (**). The above results show that this is true at a condition
stronger than Eq. (****):
 0v0  1 .
Physically, this is the condition that the L/R time constant of the pipe (equal to 0R/2) is much
shorter than the time scale R/v0 of the current’s rise and fall. In our numerical example, even this

58  is a usual notation for such a drag coefficient – see, e.g., CM Sec. 5.1.

Problems with Solutions Page 213

Essential Graduate Physics EM: Classical Electrodynamics

stronger condition is well satisfied. (An additional task for the reader: think about possible ways to take
the L/R time constant into account if it is not negligible.)

Problem 6.9. A planar thin-wire loop with inductance L, resistance R, and area A is launched to
fly ballistically from field-free space into a region where the magnetic field B is constant. Calculate the
final change of the kinetic energy of the loop, assuming that the time of its entry into the field region is
much shorter than the relaxation time constant L/R and that the loop cannot rotate.
Solution: As was stated in Sec. 6.1 of the lecture notes, and proved in the model solution of
Problem 1, the current I in a loop with an Ohmic resistance R obeys the relation IR = Vind, where Vind = –
d/dt is the Faraday induction e.m.f. (2), and  is the full magnetic flux piercing the loop’s area. In a
loop with a substantial self-inductance L, we need to take into account the contributions to  made not
only by the external magnetic field Bext but also by the current itself – see Eq. (5.68) and its discussion:59
   ext  LI , with  ext   B ext n d 2 r .
A
As a result, the flux (and hence the current) may be calculated from the following equation:
d   d L.
  IR  ext R, i.e.      ext , where   (*)
dt L dt R
In our current problem, during the short time interval t <<  of the loop’s entry into the field,
the external flux ext leaps from 0 to BAcos, where  is the angle between the field B and the vector
normal to the loop’s plane. As Eq. (*) shows, during this time, the flux  cannot change substantially,
so the leap of ext has to be compensated by that of the current I from zero to
BA
I0   cos  .
L
This current gives the loop the magnetic moment m (5.97), with the magnitude m = I0A = –
BA cos/L and the direction normal to the loop’s plane. Hence the energy of its interaction with the field
2

may be found from Eq. (5.100), with the additional factor ½ due to the field-induced character of the
moment m:60
m  B B 2 A2
U0    cos 2  .
2 2L
This positive potential energy can only come from the initial kinetic energy T of the loop,
causing its reduction:
B 2 A2
T  U 0   cos 2  . (**)
2L
(If the initial value of T is smaller than this entry cost, the loop is reflected back from the field region.)
Inside the constant-field region, the flux ext through a non-rotating loop stays constant, and the linear
differential equation (*) may be readily integrated, giving the well-known current and flux relaxation
law:

59 A different sign before the term LI in Eq.(6.78) was due to the specific choice convenient for the discussion of
the Josephson effect – see the accompanying footnote.
60 Note that this result is independent of the mutual orientation of the loop’s plane and its velocity.

Problems with Solutions Page 214

Essential Graduate Physics EM: Classical Electrodynamics

  t   t
   ext 1  exp    , I  I 0 exp   .
     
This gradual relaxation leads to the dissipation (irreversible reduction) of the potential energy U, with no
accompanying change of its kinetic energy T. (Indeed, at the motion of any closed current loop in a
constant magnetic field, the net Lorentz force (5.8) vanishes, so the loop’s kinetic energy cannot
change.) Hence, Eq. (**) gives the final answer to our problem.

Problem 6.10. AC current of frequency  is passed through a long uniform wire with a round
cross-section of a radius R comparable with the skin depth s. In the quasistatic approximation, find the
current’s distribution across the cross-section, and analyze it in the limits R << s and s << R. Calculate
the effective ac resistance of the wire (per unit length) in these two limits.
Solution: The current distribution at the skin effect obeys an equation similar to Eq. (6.23) for the
magnetic field B. Indeed, let us stop halfway through the derivation of that equation – see the second
form of Eq. (6.23),
B 1
    j,
t 
and take the curl of both sides. According to the Maxwell equation j = H = (B/), the left-hand
side of the resulting relation is equal to j/t, and since in the quasistatic approximation, j = 0, its
right-hand side equals 2j/, 61 so we get
z
j 1 2 R
  j.
t 
Since in our axially symmetric situation, j = n j(, t), a j j
z
similar equation is valid for the scalar function j(, t), so spelling out  sin 
the Laplace operator in the cylindrical coordinates,62 for the Fourier 0
amplitude of this function, we get the ordinary differential equation  cos 
B
1 1 d  dj 
 i j     .
  d  d 
By replacing  with the dimensionless coordinate   , with 2 
i  –2i/s2, the equation may be recast as
d 2 j 1 dj
  j   0.
d 2  d
This is the Bessel equation (2.130) with n = 0, and hence its general solution is
j (  )  c1 J 0 ( )  c 2Y0 ( )  c1 J 0 ( )  c 2Y0 ( ) . (*)

The argument  =  is now a complex function, but as was mentioned just after Eq. (2.158) of the
lecture notes, this does not affect the general properties of the Bessel functions discussed in Sec. 2.7. In

61 See, e.g., MA Eq. (11.3).

62 See, e.g., MA Eq. (10.3) with /z = / = 0.

Problems with Solutions Page 215

Essential Graduate Physics EM: Classical Electrodynamics

particular, Eqs. (2.132), (2.135), (2.142), (2.143), and (2.152) all remain valid for an arbitrary complex
argument. Hence, according to Eq. (2.152), the function Y0() diverges at   0, so the constant c2 in
Eq. (*) should equal 0. The other constant, c1, may be calculated by requiring the total current through
the wire’s cross-section to have a certain complex amplitude I:
R R
2c1 R
I   2  j d  2c1  J 0 ( ) d  J 0 ( )d .
0 0 2 0

This integral may be readily calculated by using Eq. (2.143) with n = 1:

2c1
I  J 1 ( )0kR  2c1 R J 1 (R) ,
 2

so, finally, 0.8
J 0 ( )
j (  )  I .
2RJ 1 (R) 0.6
j
Plots in the figure on the right show
I / R 2
the radial distribution of j, i.e. of the real 0.4  s / R  1 .0
amplitude of the current’s density, for several
values of the ratio s/R. In the limit R << s, 0.2 0 .3
i.e.  R <<1, we may use the Taylor 0 .1
series (2.132) to expand the functions J0() 0.03
0
and J1() near the origin. In the lowest 0 0.2 0.4 0.6 0.8
nonvanishing approximation, J0()  1 and /R
J1()  /2, so
I  I
j (  )    2  const ,
2R (R / 2) R
i.e. the current is uniformly distributed over the wire’s cross-section. (As the figure above shows,
actually, the inequality R << s should not be strong to ensure a virtually uniform current distribution.)
In the opposite limit, s << R, i.e. 1 <<   R, we may use the asymptotic expansion
(2.135) to get
1/ 2 1/ 2
 2     2    
J 0 ( )    cos       cos 1  i    ,
    4      s 4 
1/ 2 1/ 2
 2      2    
J 1 (R)    cos R       sin 1  i    .
 R   4 2   R   s 4 
Now using the well-known formulas MA Eq. (3.5) (which may serve as the definition of the hyperbolic
functions), and the asymptotic relations sinh b  cosh b  (1/2)exp{b} at b >> 1, we get
1/ 2
 1         1 
1/ 2
R   R  
J 0 ( )    exp  exp i  , J 1 (R)    i exp  exp i   ,
 2   s     s 4   2R   s     s 4 
so, finally,

Problems with Solutions Page 216

Essential Graduate Physics EM: Classical Electrodynamics

i  R  R
j (  )   I  exp  exp i .
2 R   s  s 
1/ 2

At (R – ) ~ s << R, this solution coincides with that following from Eqs. (6.32)-(6.34) of the lecture
notes, with the replacement x  R – . This is natural because in this limit, on the current decay scale s
<< R, the surface’s curvature is negligible.
Next, the simplest way to define the effective ac resistance R() of the wire is to write the same
relation for the average Joule power of electric energy loss,
R   I 2
P  ,
2
as follows from Eq. (4.40) of the lecture notes in the dc limit. (The factor ½ is due to the averaging over
the period of sinusoidal oscillations of the current.)63 For this power, per unit length of the wire, the time
averaging of Eq. (4.39) gives the following expression:
P 1
 j   
2
 d 2 ,
l 2 A

so we can calculate the effective resistance per unit length as

R   1 j    2 j   
2 R 2
1
l

 
A
I
d   2
 
0
I
d .

Plugging in the current distribution limits calculated above and integrating, we get

R   1 1 / R, for R   s ,
 
l  R 1 / 2 s , for  s  R .
In the first of these expressions, we may readily recognize Eq. (4.21) for the dc resistance of the whole
wire, with the cross-section area A = R2. The second of these results is the same as the dc resistance of
a thin hollow tube, with radius R, thickness s, and hence the cross-section area A’ = 2Rs. Both results
fit the physical picture of the current distribution – see the figure above.
Note that since s  –1/2, in the high-frequency limit (s/R  0), the ac resistance of metallic
wires grows as 1/2,64 providing serious problems for telecommunication technologies – see the
discussion in Sec. 7.9.

Problem 6.11. A long round cylinder of radius R, made of a uniform Ohmic conductor with
conductivity  and magnetic permeability , is placed into a uniform ac magnetic field Hext = H0cost
directed along its symmetry axis. Calculate the spatial distribution of the magnetic field’s amplitude and,

63 Another (perhaps less obvious) definition giving the same result is R()/l  Re[E(R)/I], where E(R) is the
complex amplitude of the ac electric field on the wire’s surface.
64 The fraction R()/R(0) = R/2  1/2 is sometimes called the Rayleigh resistance ratio.
s

Problems with Solutions Page 217

Essential Graduate Physics EM: Classical Electrodynamics

in particular, its value on the cylinder’s axis. Spell out the last result in the limits of relatively small and
large R.
Solution: This is just an axially symmetric version of the skin-effect problem solved in Sec. 6.3
of the lecture notes. Due to the symmetry, the magnetic fields (both B and H) at any point are directed
along the cylinder’s axis,

Hr, t   n z Re H    e it , 
where  is the distance for the axis (while the induced electric field and hence the eddy current density
have only azimuthal components). As a result, the Laplace operator of the only component of the
magnetic field takes a simple form:65
1 d  dH 
2H  n z  ,
 d  d 
and Eq. (6.23) yields the following equation for its complex amplitude inside the cylinder:
1 1 d  dH  
 i H     , for   R .
  d  d 
This is exactly the same equation as was obtained for the current density’s amplitude j in the solution
of the previous problem, and we may borrow its solution (also giving a finite value at  = 0):

H     cJ 0   ,

where J0 is the zero-order Bessel function of the first kind, while

1 i
   i 1 / 2   .
s
The numerical plots of H as a function of  are also similar to those for j in the model
solution of the previous problem; however, their vertical scaling is different because in our current case,
the constant c should be calculated from the boundary condition H(R) = H0, giving
J 0  
H     H 0 .
J 0 R  1

Now by using the fact that J0(0) = 1 (see, e.g., either Eq.
(2.132) or Fig. 2.18 of the lecture notes, with n = 0), for
the field on the cylinder’s axis, we get simply H  0 
1 H0
H  0  H 0 .
J 0 R 
The figure on the right shows (in the appropriate
semi-log scale) the modulus of this value, as a function
of R normalized by the skin depth s. For small values 0.1
of the R/s ratio, the result may be well approximated by 0 1 2 3 4
using the leading two terms of the expansion (2.132): R /s

65 See, e.g., MA Eqs. (10.3) and (10.6) with f = f()nz.

Problems with Solutions Page 218

Essential Graduate Physics EM: Classical Electrodynamics

H  0  1 1 1
   ,
H0 J 0 R  1  R / 2  2
1  iR 2 / 2 s2
so at R/s  0,
H  0
4
1 1 R 
  1     1 .
H0 1  R 2
/ 2 s2 
2 1/ 2 8  s 

In this approximation, the applied field is virtually uniform everywhere in the cylinder because the
diamagnetic effect of the eddy currents is small.
In the opposite limit, we may reuse the large-argument asymptote of function J0, which was
employed in the model solution of the previous problem, to get
H  0
1/ 2
1/ 2  R  R   R R
 2R exp   2    exp   0, for  .
H0  s   s   s  s
In this limit, the eddy currents shield most of the cylinder’s bulk from the applied external magnetic
field, which therefore penetrates only into a relatively thin skin layer, exactly as into a conducting semi-
space – see Sec. 6.3 of the lecture notes.

Problem 6.12.* Define and calculate an appropriate spatial-temporal Green’s function for Eq.
(6.25) of the lecture notes, and then use this function to analyze the dynamics of propagation of the
external magnetic field that is suddenly turned on at t = 0 and then kept constant:
 0, at t  0,
H  x  0, t   
H 0 , at t  0,

into an Ohmic conductor occupying the semi-space x > 0 – see Fig. 6.2.
Hint: Try to use a function proportional to exp{–(x–x’)2/2(x)2}, with a suitable time
dependence of the parameter x and a properly selected pre-exponential factor.
Solution: As was discussed in Sec. 2.10 of the lecture notes, the Green’s function approach uses
the linearity of the corresponding differential equation, giving an explicit expression of the linear
superposition principle. For a time-independent but inhomogeneous equation, such as the Poisson
equation (1.41), the superposition is applied to the sum of elementary right-hand sides distributed over
the space – see Eq. (2.203). In that case, the Green’s function is purely spatial.66 On the other hand, for a
time-dependent but homogeneous equation, such as Eq. (6.25):
H 1 2H
 , for x  0 ,
t  x 2
the superposition principle may be applied to the spatial distribution of the initial conditions:

66 Similarly, Green’s functions of ordinary inhomogeneous differential equations describing the time dynamics
alone are purely temporal – see, e.g., CM Sec. 5.1.

Problems with Solutions Page 219

Essential Graduate Physics EM: Classical Electrodynamics


H  x, t    G  x, t ; x' , t'  H  x' , t' dx' . (*)
0

With this definition, the Green’s function G, considered as a function of x and t, is just the particular
solution of Eq. (6.25), corresponding to delta-functional initial conditions:
G x, t ; x' , t'   H  x, t , for H  x, t'     x  x' , with x'  0 . (**)

First, let us calculate the Green’s function for the case when Eq. (6.25) is valid in infinite space
(– < x < +), by using the provided Hint. By selecting the pre-exponential coefficient to keep the
function properly normalized,67
   

 Gx, t; x' , t' dx   Gx, t' ; x' , t' dx 

  
 H x, t'  dx    x  x'  dx  1 ,


we get the well-known expression68

1  x  x' 2 
G x, t ; x' , t'   exp .
2 1 / 2 xt , t'   2xt , t'  2 
Now plugging this formula into Eq. (6.25), we see that it is indeed the solution of the equation, provided
that x(t, t’) satisfies the following ordinary differential equation:

t
x 
1
x
, i.e.

t
 
x 2  2  const .

With the requirement x  0 at t  t’, following from the initial condition (**), the solution of this
equation is obviously
2t  t'   2t  t'  
1/ 2

x  
2
, so xt , t'     .
   
Since the Green’s function depends on the difference (t – t’) only, in all formulas below I will take t’ =
0, and use shorthand notations G(x, t; x’)  G(x, t; x’, 0) and x(t)  x(t, 0). As Eq. (**) tells us, the
physical sense of x(t) is the gradually increasing width of the spatial distribution of the magnetic field
induced by a delta-functional spatial “pulse” of the external field at t = 0.
Now we may use the equation’s linearity, and the mirror-image themes discussed in Chapters 2-5
of the lecture notes, to construct the Green’s function for the semi-space x > 0, which obeys the
boundary condition G(0, t; x’) = 0:

1    x  x' 2    x  x' 2   2t 
1/ 2

G x, t ; x'   exp   exp 2 

, with xt     , (***)
2 1 / 2 xt    2xt 2   2xt     

67 This requirement follows from the fact that integrating both parts of Eq. (6.25) from x = – to x = +, and
assuming that the field’s gradient has a final spatial extent, i.e. that H/x  0 at x  , we get

d
dt 
H ( x, t )dx  0. .

It describes, for example, the Gaussian (or “normal”) probability distribution with variance equal to (x) .
68 2

Problems with Solutions Page 220

Essential Graduate Physics EM: Classical Electrodynamics

and hence automatically generates solutions (*) that satisfy this solution.
Next, due to the eventual decay of the eddy currents in time, the constant applied magnetic field
cannot be prevented from its eventual penetration into the conductor: H(x, t)  H0. Since the Green’s
function (***) tends to zero at t  0, we better reduce our problem to finding some function H’(x, t)
with the same trend. Evidently, this may be done by representing the field as the difference
H  x, t   H 0  H' ( x, t ) ,
with the initial value H’(x, 0) = H0. For this auxiliary field, Eqs. (*) and (***) yield

H0 
   x  x' 2    x  x' 2 
H'  x, t   H 0  G  x, t ; x' dx' 
2 1 / 2 xt  0   2xt 2 
exp   exp 2 
dx' . (****)
0  2xt  
Note that the integration limits automatically exclude the non-physical mirror-image delta function at x’
< 0, used to construct the Green’s function (***).
This integral may be readily expressed via the so-called error function

exp  2 d ;
2
erf   
 1 / 2 0
however, for most practical purposes (such as plotting, asymptote analysis, etc.), the explicit integral
form (****) is preferable. Note that by introducing
the following normalized variables: field h  H/H0
and coordinate   x/x(t) = x/(2t/)1/2, we may
rewrite our result in the form of a universal function 0.8

    ξ' 2   0.6
exp    H
1

  2   h
h   1  0     ξ' 2  d ' , H0
2 1 / 2  exp  
0.4

  
  2  0.2

plotted in the figure on the right, so the field’s profile 0

in the actual (dimensional) space just stretches with 0 0.5 1 1.5 2
time as x(t)  t1/2.   x / x t 

Problem 6.13. Solve the previous problem using the variable separation method, and compare the
results.
Solution: Let us look for the solution of the same Eq. (6.25),
H  2 H
  , for x  0 ,
t x 2
in the usual variable-separated form, but (just as in the solution of the previous problem) singling out the
eventual uniform distribution of the field:

Problems with Solutions Page 221

Essential Graduate Physics EM: Classical Electrodynamics

H ( x, t )  H 0   Tk t X k  x  , (*)
k

so the second term of Eq. (*) is equal to H0 at t = 0 and has to vanish at t  . Plugging a particular
term of this series into the equation, and dividing both sides by TkXk, we get
1 dTk 1 d2Xk
   const  k 2 .
Tk dt X k dx 2
Solving the resulting simple differential equations for Tk and Xk, we get69
 k2 
Tk  a k exp  t , X k  bk sin kx ,
  
with (at this stage) arbitrary ak and bk. Since the segment on which our equation is valid (0  x  ) is
semi-infinite, the spectrum of the eigenvalues k is continuous, so plugging these solutions into Eq. (*),
we need to replace the summation over k with the corresponding integration:

 k2 
H  x, t   H 0   c k sin kx exp  t dk , where c k  a k bk . (**)
   
What remains is to find the function ck from the initial condition:


c

k sin kx dk  H 0 , for x  0. (***)

As usual, we may calculate ck from this equation by multiplying both sides by a similar fundamental
function but with an arbitrary argument, in our case sink’x, and integrating them over x. Changing the
integration order on the left-hand side, we get
  

 dk ck  dx sin kx sin k'x  H 0  sin k'x dx .

 0 0

Since the expression under the inner integral on the left-hand side is an even function of x, the integral
may be transformed as
 
1
0 sin kx sin k'x dx  2 sin kx sin k'x dx
1  i k k' x 
  
dx   e i k  k' x dx  .
1
  cosk  k 'x  cosk  k 'x  dx  Re   e
4  4   
Per MA Eq. (14.4), these integrals are equal to, respectively, 2(k  k’), so our equation yields


ck'  ck'   H 0  sin k'x dx   H 0 cos k'x x
x0
.
2 0
k'
As Eq. (**) shows, ck has to be an odd function of k, so (ck’ – c-k’) = 2ck’. Also, the upper limit on
the right-hand side of the last equality may be ignored because the solution should not be sensitive to a

69Other fundamental function candidates, exp{+k2t/} for Tk(t) and coskx for Xk(x), do not satisfy our boundary
and initial conditions H(x, ) = H(0, t) = H0.

Problems with Solutions Page 222

Essential Graduate Physics EM: Classical Electrodynamics

very slow quenching of the right-hand side of Eq. (***) at x  . As a result, changing the index
notation from k’ to k, we get
H
ck  0 ,
k
so Eq. (**) yields
 1  sin kx  k2    2  sin kx  k2  
H  x, t   H 0 1   exp t dk   H 0 1   exp t dk  .
   k       0 k    
Introducing dimensionless variables   x/(2t/)1/2, and   k(2t/)1/2 (so that kx = ), and h 
H/H0, we may rewrite this result in a parameter-free form

2 sin   2
h  h   1   exp   d , (****)
 0
  2 
which shows that the field distribution profile in the conductor is universal, and all its time evolution is
reduced to the increase of its spatial scale (2t/)1/2. Indeed, the numerical plot of this result exactly
coincides with that shown in the model solution of the previous problem, despite the rather different
analytical formulas. This fact is not quite surprising, because it is straightforward (but still recommended
for the reader) to show that Eq. (****) is just as the Fourier-integral expansion of the result derived in
the previous problem.

z R'
Problem 6.14. Calculate the average force exerted by ac current I(t)
of amplitude I0, flowing in a planar round coil of radius R, on a conducting
sphere with a much smaller radius R’ (which is still much larger than the
skin depth s at the ac current’s frequency), located on the loop’s axis, at 0 R
distance z from its center – see the figure on the right.
Solution: In close vicinity of the sphere, due to the condition R’ << R, I t 
the current’s field may be considered locally uniform. On the other hand, the condition s << R’ allows
us to ignore the effects of ac field penetration into the sphere, and thus to use the coarse-grain (ideal-
diamagnetic) фззкщчшьфешщт B(t) = 0 inside the sphere. As a result, the instantaneous distribution of
the applied ac field outside the sphere is similar to that of a dc field around a superconducting sphere –
see Eq. (6.58) and Fig. 6.3 of the lecture notes. As was discussed there, this field is a sum of the locally-
uniform field H(t) created by the current at the point of the sphere's location, and a purely dipole field
corresponding to the magnetic dipole moment given by Eq. (5.125) with  = 0:

mt   2R' 3 H t  .

The negative sign means that the dipole moment is directed opposite to the applied field H(t) – which is
natural for the ideal diamagnetism.
The field H(t) = B(t)/0 is given by Eq. (5.23) of the lecture notes:
I t  R2
H t   n z ,

2 R2  z2 3/ 2

Problems with Solutions Page 223

Essential Graduate Physics EM: Classical Electrodynamics

The energy of interaction between an external field and a fixed magnetic dipole is given by Eq. (5.100)
of the lecture notes, U = –mB; however, in our case, the dipole is not fixed but is induced by the field
(m  B = 0H), so we need to multiply this expression by the usual factor ½, getting

mt   Bt  mt   0 H t   R' 3 R 4

U t      0 R' 3 H 2 t   0 I 2 t  2 .
2 2 4 R  z 2 3
The time average of this energy over the period of the sinusoidal ac current of amplitude I0 is
 R' 3 R 4
U  I2
.
8
0 0
R 2
 z2 
3

The energy is positive and decreases with distance z, so the average force exerted on the sphere,
U 3 R' 3 R 4
F    0 I 02 z,
z 4 R2  z2  
4

has the same sign as z, i.e. corresponds to repulsion.

As a matter of principle, such repulsion of a sphere located above the loop (i.e. with z > 0) may
be used for balancing the sphere's weight mg if it is lower than the maximum value of the force,
3 3
3 71 / 2  R'   R' 
Fmax F   0 I 02    1.522  10 3  0 I 02   ,
z  zopt  R / 7 4 8 4
R R
because one (the highest) of the two stationary points z corresponding to the equilibrium condition F =
mg is stable with respect to vertical perturbations. However, this point is unstable with respect to lateral
displacements of the sphere – or of a small levitated object of any shape (see, e.g., the solution of
Problem 5.2); as a result, levitation requires more complex geometries.
A more practical problem of such levitation is the energy losses in the skin layer (see Eq. (6.36)
of the lecture notes); the losses may be avoided using superconductivity, but this approach involves
costly deep refrigeration. As a result, we still have not yet achieved the old dream of magnetically-
levitated ground transportation systems.

Problem 6.15. A small planar wire loop carrying a fixed current I is located relatively far from a
planar surface of a superconductor. Within the coarse-grain (ideal-diamagnet) description of the
Meissner-Ochsenfeld effect, calculate:
(i) the energy of the loop-superconductor interaction,
(ii) the force and torque acting on the loop, and
(iii) the distribution of supercurrents on the superconductor’s surface.
Solution: As was discussed in Sec. 6.4 of the lecture notes, in the ideal-diamagnet (coarse-grain)
approximation, the field outside the superconductor has to satisfy the boundary condition (6.59):
Bn = 0. (*)

Problems with Solutions Page 224

Essential Graduate Physics EM: Classical Electrodynamics

The field of a wire loop in unlimited free space, at distances much larger than the loop size, may
be described in the magnetic-dipole approximation. An elementary generalization of Eq. (5.99) of the
lecture notes to an arbitrary position r’ of the loop’s center is
z

 0 3 r - r' r - r'   m'   m' r - r'
2

B' (r )  , (**) m'

4 r - r'
5 r'

where m’ is its magnetic moment – in our case, with the

magnitude m’ = IA, where A is the loop area. Note that a
simultaneous flipping of signs of the same Cartesian components d
of the vectors m and (r – r’) reverses the sign of this component B''
of the field, but not its other components. From this symmetry, it r
is clear that the boundary condition (*) may be satisfied by 0 x
representing the total magnetic field outside the superconductor
as a sum of this field B’ and a similar field B” of a mirror-image B'
magnetic dipole m” that is located and oriented as shown in the d
figure on the right. Selecting the coordinate axes as shown in this
figure (with the x-axis in the same vertical plane as the vector
m’), we may write
r" m"
x''  x'  0, y''  y'  0, z''   z'  d ,
m" x  m' x  m sin  , m" y  m' y  0, m" z  m' z  m cos  .

Note that the image dipole orientation is exactly opposite to that in Problem 3.8, due to the
different boundary conditions. Hence we may reuse the solution of that problem, just minding the signs
and the fundamental constants, to get the following results.
(i) The interaction energy,
0 m 2
U int
1
2
1
  m'  B'' (r' )   m' x B" x  m" z B" z  x  y 0, z d 
2 8 (2d ) 3

1  cos 2  ,

corresponds to the repulsion of the dipole from the superconductor, at any dipole’s orientation. This is
very natural, because a superconductor pushes out the external field lines, and hence repels their source.
Such repulsion is the basis of all magnetic levitation projects (including the “maglev” trains) using
superconductivity – see the brief discussion in the model solution of the previous problem.
(ii) The force and the torque may be readily calculated from Uint:
U int  0 3m 2 U int  0 m 2
F  Fz  
d

8 8d 4

1  cos 2   ,  y  


8 8d 3
sin 2 .

Note that the system’s stable equilibrium with respect to rotation is at  = /2, with the magnetic
moment parallel to the surface.
(iii) The linear density of the surface supercurrent may be calculated from its universal relation
(6.38) with the magnetic field just outside the surface:
J  n z  H   n z  B /  0 ,

Problems with Solutions Page 225

Essential Graduate Physics EM: Classical Electrodynamics

where B is the tangential component of the magnetic field (in our coordinates, B = Bxnx + Byny) near
the surface. By spelling out Eq. (**) in Cartesian components, we may readily find the net field of the
two dipoles, B = B’ + B” in an arbitrary point r = xnx + yny of the surface (z = 0):
 0 m n x 2 x  y  d sin   3dx cos   n y 3 y  x sin   d cos  
2 2 2

B  B'  B"  2B'  ,

2 (x 2  y 2  d 2 )5 / 2
so, finally,

J
 2 2 2

m n x 3 y  x sin   d cos    n y 2 x  y  d sin   3dx cos 
.

2 (x 2  y 2  d 2 )5 / 2
In the case when the current loop plane is parallel to the superconductor’s surface (sin = 0, cos
= 1), the current lines are concentric circles, with   (x2 + y2)1/2 = const, and its magnitude,
m 3d
J  ,
2  2  d 2 5 / 2
does not depend on the azimuthal angle. The direction of this vortex current is opposite to that of that in
the loop, thus fulfilling its function of shielding the superconductor’s interior from the penetration of the
loop’s field.

Problem 6.16. A straight uniform magnet of length 2l, cross-section 2l

area A << l2, and mass m, with a permanent longitudinal magnetization M0,
is placed over a horizontal surface of a superconductor – see the figure on M0
the right. Within the ideal-diamagnet description of superconductivity, find g
the stable equilibrium position of the magnet.
Solution: As was discussed in Sec. 5.6 of the lecture notes, the magnetic field induced by a long
and thin (A << l2) permanent magnet, with the magnetization M0 parallel to its length, is equivalent to
that of two point magnetic charges qm = 0M0A located at its ends, each producing a Coulomb-like
radial magnetic field
q r  r'  M A r  r' ,
Br   m  0 0
4 r  r' 3
4 r  r'
3

where r’ is the charge’s (i.e., the magnet end’s) position. In the coarse-grain (ideal-diamagnet) picture of
the Meissner-Ochsenfeld effect, the magnetic field cannot penetrate into the superconductor. Thus the
boundary problem for a magnetic point charge near the  qm
superconductor’s surface, in this approximation, is generally  l
similar to that of an electric point charge near a conductor’s  qm l
surface and may be solved exactly as was discussed in Sec. 2.9, y d
i.e. by the introduction of its mirror image. However, the image
charge sign should be similar (rather than opposite) to the initial d d
charge, because it should compensate the normal (rather than 
tangential) component of the field at the surface – see Eq. (6.59). d
Hence, in our case, we get the system of two actual and  qm
two image magnetic charges shown in the figure on the right.
Introducing the vertical coordinate y of the magnet’s center and  qm

Problems with Solutions Page 226

Essential Graduate Physics EM: Classical Electrodynamics

the angle  of its slope as shown in that figure, we may readily express the distances d and d between
the charges as functions of these two generalized coordinates:70
d   2 y  l sin  ,

d 2  2 y   2l cos    4 y 2  l 2 cos 2  .
2 2

Continuing to use the analogy with electrostatics, in particular with Eq. (2.192) with the replacements q
 qm and 0  0,71 we may write the following expression for the potential energy of the system as72
q m2
1  1 1 2 q m2  1 1 1 
U   
   mgy     1/ 2 
 mgy
4 0 2  d d d  8 0 
 2 y  l sin   2 y  l sin   y  l cos 
2 2 2
 

q m2  y 1  y  q m2  0 M 02 A 2
  2 2  1/ 2 
 2 , with a 
2
mg  mg .
8 0  y  l sin 
2
 y  l cos 2 
2 2
  a  8 0 8

For any fixed y within the geometrically possible region y > lsin, the function U() has two
similar minima at  = 0 and , i.e. at sin2 = 0 and cos2 = 1. Perhaps the easiest way to see that is to
expand the expression in the square brackets (which alone depends on the angle) in the Taylor series at
small , keeping only two leading terms:
y 1 1 1
  
y  l sin 
2 2
2

y 2  l 2 cos 2  1/ 2

y 1  l sin 2  / y 2
2
 
y2  l2  1  l
1/ 2 2

sin 2  / y 2  l 2 
1/ 2

1 1  2 2 1 1 
  1/ 2 
l   3  3/ 2 
.
 y y  l
2 2
    y 
2 y  l2
2
 
Since the second term in the last square brackets is always smaller than the first one, the potential energy
always grows with 2.
Hence, the stable stationary position of the magnet is horizontal.73 Its stationary height y over the
superconducting surface should be calculated from the requirement dU/dy = 0 (at  = 0 or ). For
arbitrary values of the only dimensionless parameter a/l of the problem, the equation for y, resulting
from this condition,
1 y 1
  2, (*)
y 2
y 2  l 2  a
3 / 2

70 The magnet’s potential energy (and hence its equilibrium) is evidently unaffected by its arbitrary translational
displacement along the surface and arbitrary rotation about the vertical axis.
71 See Eqs. (5.137)-(5.139) and the accompanying discussion.
72 We may ignore the potential energies of interaction of the magnetic charge pairs belonging to the same magnet,

because these energies do not depend on the magnet’s position, i.e. on coordinates y and . (The longitudinal
forces resulting from these interactions are compensated by the mechanical stress forces in the rigid magnets.)
73 Note that this conclusion is in agreement with the solution of Task (ii) of the previous problem. Indeed, in the
particular case y >> l (see the balance of this solution), the magnet may be approximated as a magnetic dipole –
just as a small current loop.

Problems with Solutions Page 227

Essential Graduate Physics EM: Classical Electrodynamics

may be solved only numerically. However, this equation 10

may be readily used for plotting the reciprocal function,
a(y), and then just transposing the plot – see the solid red
curve in the figure on the right. Moreover, Eq. (*) is easily
solvable analytically in two limits, giving: y
l
a, for a  l , 1
y (**)
 3 21 / 4 la 1 / 2 , for l  a,
– see the two dashed lines in the figure on the right. The
exact solution describes a smooth crossover between these
asymptotes, taking place at a/l  1.2.
Physically, the first line of Eq. (**) corresponds to 0.1
0.1 1 10
the limit of relatively weak magnetic interaction, when a/l
gravity presses the magnet very close to the superconductor
surface, and only the closest magnetic poles (“magnetic charges”) interact substantially. On the other
hand, if the magnetic interaction is strong enough, the magnet’s repulsion by the superconducting
surface brings it up so high (y >> l) that its field near the surface is close to that of a dipole with moment
m = VM0 = 2lAM0. As a result, the second line of Eq. (**) may be calculated from Eqs. (5.99) and
(5.100) of the lecture notes with this value of m and an additional factor of ½ due to the induced nature
of the dipole’s image.

Problem 6.17. A plane superconducting wire loop of area A and 0


inductance L may turn, without static friction, about a horizontal axis 0
(in the figure on the right, normal to the plane of the drawing) passing
through its center of mass. Initially, the loop had been horizontal (with  B
= 0) and carried supercurrent I0 in such a direction that its magnetic
dipole vector was directed down. Then a uniform magnetic field B,
directed vertically up, was applied. Using the ideal-diamagnet description of the Meissner-Ochsenfeld
effect, find all possible equilibrium positions of the loop, analyze their stability, and give a physical
interpretation of the results.
Solution: Due to the magnetic flux quantization (see (6.62) of the lecture notes and its
discussion), the total flux through the loop,74
   ext  LI  AB cos   LI ,
has to be constant and equal to its value,
   LI 0 ,
before the external field’s application. Combining these two relations, we get

74 Here, just as Eq. (6.78), the sign before the term LI is selected in such a way that the magnetic moment m
created by current I > 0 is directed opposite to the field B. However, this choice (if consistent) does not affect the
energy U and hence the final results.

Problems with Solutions Page 228

Essential Graduate Physics EM: Classical Electrodynamics

 ext AB cos 
I  I0  I B , with I B   .
L L
This means that the vertical component mz of the magnetic moment (5.97) of the loop may be
represented as the sum of two parts: m0 = –AI0cos created by the pre-existing current, and mB = –AIB
cos created by the current IB induced by the external field. Such decomposition of m is important
because while the former part’s contribution to the potential energy U of the loop may be calculated
using Eq. (5.100):
U 0  m0 B  ABI 0 cos  ,

the latter part’s contribution should be multiplied by the usual factor ½:75
1 1 A 2 B 2 cos 2 
U B   m B B  AI B cos  B  .
2 2 2L
As a result, the total potential energy of the loop is76
A 2 B 2 cos 2 
U  U 0  U B  ABI 0 cos   .
2L
Since this is a system with just one degree of freedom (), its equilibrium positions correspond to
the extrema of the function U(), in which dU/d = 0, while their stability is determined by the sign of
the second derivative of this function: if d2U/d2  0, then the stationary position is stable. For our
function, the equilibrium requirement gives the following equation:
dU A 2 B 2 sin  cos   AB cos  
  ABI 0 sin     AB sin   I 0    0. (*)
d L  L 
In relatively weak fields, when
AB  LI 0 , (**)
the expression in the parentheses of the right-hand side of Eq. (*) is positive for all values of , so this
equation has only two physically distinguishable roots, both with sin = 0:
 0  0, and 1   ,
both corresponding to the horizontal position of the loop. Now, calculating the second derivative,
d 2U d  A 2 B 2 sin  cos   d  A 2 B 2 sin 2 
   ABI 0 sin       ABI 0 sin   
d 2 d  L  d  2L 
A 2 B 2 cos 2  AB cos 2 
  ABI 0 cos     AB I 0 cos   ,
L  L 
we see that if the low-field condition (**) is satisfied, the second term in the last parentheses cannot
affect the sign of the whole expression, so it is negative for 0 and positive for 1. This means that the

75 Cf. Problem 9 whose first part is very similar to this one, because on a time scale t << L/R, a resistive loop
behaves similarly to a superconducting one.
76 Since the rotation axis passes through the loop’s center of mass, gravity does not give any contribution to U.

Problems with Solutions Page 229

Essential Graduate Physics EM: Classical Electrodynamics

initial horizontal position of the loop, with  = 0 = 0, is unstable, while the flipped state, with  = 1 = ,
is its only stable position.
However, in higher fields, when the condition (**) is not valid, the second term in the
parentheses (at  = 1 = , equal to AB/L), overrules the first term (at  = 1 = , equal to – I0), so the
flipped horizontal state becomes unstable as well. Instead, at such fields, Eq. (*) acquires two other
roots,
 LI 
 2,3   cos 1   0  .
 AB 
Since
 AB cos 2 2,3 
d 2U
d 
2  3, 4   AB I 0 cos  2,3 

L

   AB  I 0 cos  2,3 
AB
L
2 cos 2  2,3  1

   
 LI AB   LI 0 
2
  1
 AB  I 0 0  2    1   A 2 B 2  L2 I 02 ,
 AB L   AB    L
both states are stable as soon as they exist (at AB > LI0).
The physics of this curious behavior is as follows. In weak fields, the role of the induced current
IB is small, and the applied field just tries to flip the loop, so its magnetic moment vector m  m0 would
be aligned with its vector B. However, in high magnetic fields, the effects related to IB dominate, and in
the limit B >> LI0/A, the field tries to turn the loop’s plane parallel to the vector B: 2,3  /2 (cos2,3
 0). This fact may be interpreted using the discussion at the end of Sec. 6.2 of the lecture notes, even
though it is quantitatively valid only for cylindrical geometry. As was shown there, the external
magnetic field tends to fill all available space. In the case of a superconducting ring with I0  0, the
Meissner-Ochsenfeld effect pushes the field out of a volume of the order of A3/2cos, so the system
“tries” to minimize this volume by making cos  0.

Problem 6.18. Use the London equation to analyze the penetration of a uniform external
magnetic field into a thin (t ~ L) planar superconducting film whose plane is parallel to the field.
Solution: According to the definition B = A of the vector potential, in this 1D situation, the
field’s magnitude B is just the spatial derivative of the vector potential’s magnitude, and hence obeys an
equation similar to the 1D version of Eq. (6.56) of the lecture notes:
d 2B 1
 2 B,
dx 2
L
where the x-axis is normal to the film. The general solution of this linear equation is a linear
combination of the functions exp{x/L}, or alternatively, of the functions sinh(x/L) and cosh(x/L). In
our current problem, the magnetic field’s distribution should be symmetric relative to the plane passing
through the middle of the film’s thickness. This is why, by selecting this plane for x = 0, we get
x
B ( x )  const  cosh .
L
The constant in this relation may be found from the usual macroscopic (not the coarse-grain!) boundary
condition (5.117) applied to film surfaces: H(t/2) = H0 = B0/0, i.e. B(t/2) = (/0)B0. (For most

Problems with Solutions Page 230

Essential Graduate Physics EM: Classical Electrodynamics

superconductors, the difference between the ratio (/0) and 1 is negligible, but I still keep it spelled out
for clarity.) As a result, we get
 cosh( x /  L )
B ( x )  B0 .
 0 cosh(t / 2 L )
This result shows that if the film is thin (t << L), the field is almost uniformly distributed over
its thickness. In the opposite limit, the field only penetrates into two L-scale layers at the film’s surfaces
and is shielded from the film’s interior by the equal and opposite supercurrents flowing in these layers.

Problem 6.19. Use the London equation to calculate the distribution of the supercurrent density j
inside a long straight superconducting wire with a circular cross-section of radius R ~ L, carrying
current I.
Solution: In this cylindrical and axially symmetric geometry, the solution of the London equation
(see Eq. (6.56) of the lecture notes) may be looked for in the form A(r) = A()nz, where the z-axis
coincides with that of the wire, and  is the distance of the observation point r from that axis. With the
well-known expression for the Laplace operator in cylindrical coordinates,77 the equation becomes78
1 d  dA  1
  A.
 d  d   L2
With the introduction of the natural dimensionless argument   /L, this equation becomes
d 2 A 1 dA
  A  0,
d 2  d
which is the modified Bessel equation (2.155) with  = 0. Its general solution is

A  a1 I 0 ( )  a 2 K 0 ( ) ,

so the supercurrent density j (which, within the London gauge, is proportional to A – see Eq. (6.55) of
the lecture notes) is
     
j    c1 I 0    c 2 K 0   .
L  L 
Since the modified Bessel function of the second kind, K0(), diverges at   0 (see, e.g., either
Eq. (2.157) or just the right panel of Fig. 2.22) while j has to stay finite at all points, the coefficient c2
has to equal 0. The remaining coefficient c1 may be readily related to the total current:
R /L
R R
  
 j  d   2  j  d  2 c1  I 0 
2
 d  2  L2 c1  I  d  I .
L
0
A 0 0  0

77See, e.g, MA Eq. (10.3) with / = /z = 0.

78Note that this equation, and the balance of the solution, are very similar to those in Problem 10, with the “only”
exception that the complex parameter  given by Eq. (6.30) is now replaced with the real parameter 1/L.

Problems with Solutions Page 231

Essential Graduate Physics EM: Classical Electrodynamics

The last dimensionless integral may be calculated using the recurrence relation (2.143) (valid for any
Bessel functions), in our current case with n = 1. As a result, we get
R /L
R /L I
c1  I 2 2
 I  d  I 2 L2 I 1 ( )]0  ,
2 L RI 1 ( R /  L )
L 0
0
so, finally,
I I 0 ( /  L )
j .
2 L R I 1 ( R /  L )
If the wire is very thin (R << L), we may use the first of Eqs. (2.157) to write I0(/L)  1,
I1(R/L)  R/2L, so j  I/R2 = const, i.e. the supercurrent is uniformly distributed over the wire’s cross-
section. In the opposite limit L << R, we may use the first of the asymptotic formulas (2.158) to reduce
our result to
1/ 2
I  R  exp{ /  L } I  R
j    exp .
2 L R    exp{R /  L } 2 L R /  1/ 2
 L 
This expression shows that appreciable supercurrent only flows in a L-thin sheet at the wire’s surface –
exactly like in the plane-surface problem – see Eq. (6.57).

w  t ,  L
Problem 6.20. Use the London equation to calculate the
inductance (per unit length) of a long uniform superconducting  I d   L
strip placed close to the surface of a similar superconductor – t ~  L
see the figure on the right, which shows the structure’s cross-
section.
Solution: DC supercurrent does not require any electric field to sustain it, so it is free to self-
distribute over the strip’s cross-section to minimize the total energy of the system. In particular, in our
case when t, L << w, all the current in the strip has to be localized at its surface facing the bulk
superconductor (which plays the role of a magnetic ground plane), to minimize the magnetic field
outside the gap t, because any substantial field in those areas would decrease only on distances of the
order of w and hence give contributions with a very large spatial weight ~ w2, into the magnetic energy
per unit length:
U B 2 (r ) 2
 d r. (*)
l 2 (r )
Next, for a gap of a constant thickness t, the linear density J of the supercurrent should be
constant across the gap’s width w, because otherwise the magnetic flux (also per unit length),

l 
 B( x)dx  J , (**)

of the field B(x) it creates in the gap would change across the width. (Here x is the Cartesian coordinate
across the gap.) This is impossible because the thick superconducting electrodes prevent the magnetic
field lines’ from escaping the gap. So, J = I/w = const.
The last (but not least) fact we need is that to prevent the magnetic field in the gap from
spreading into the ground plane, its surface has to carry an equal and opposite supercurrent J. (This fact

Problems with Solutions Page 232

Essential Graduate Physics EM: Classical Electrodynamics

is the basis of dc current transformers, which are important components of superconductor electronic
circuits.79)
Now the problem may be readily solved using Eq. (6.57) of the lecture notes. Indeed, the integral
in Eq. (**) is the sum of three integrals: one over the gap (giving the contribution Bt = 0Ht = 0Jt =
0It/w), and two similar integrals through the London penetration layers of the top electrode and the
ground plane:
 
 x I L
0 B ( x ) dx  B ( 0) 0 exp  L dx  H 0 L  J L  w .
As a result, for the inductance L per unit length, we get

L  1 1   It I L  t 
   B( x)dx   0  2    0 ef , with t ef  t  2  . (***)
l Il I I w w  w 0 L
This expression is similar to the solution of Problem 5.4(iv), except for the effective increase of
the gap between the strips due to the magnetic field’s penetration into the superconductors. In practical
superconductor integrated circuits, the insulating gap thicknesses t may be from a few nanometers to
~100 nm, while L is of the order of 100 nm, so this penetration effect is quite noticeable (and even may
be used for the London depth measurement).
Note also that in contrast to the coarse-grain (ideal-diamagnet) description of superconductors,
which is equivalent to taking  = 0 inside these materials, the London equation takes supercurrents into
account explicitly as “stand-alone” currents, so the magnetic permeability participating in the above
result describes only the effects of localized atomic currents. In typical superconductors, these effects
are relatively weak (  0),80 so the effective gap width tef is very close to just (t + 2L).
One more important note: it might be tempting to derive Eq. (***) by calculating the magnetic
field energy (*) and requiring it to equal L/I2/2. However, such a calculation gives a wrong result for tef,
without the factor of 2 before the second term. The reason for this discrepancy is somewhat subtle: since
the Cooper pairs move in superconductors without dissipation, we cannot neglect their kinetic energy.
Indeed, according to the London equation (6.55), the effective velocity of the Cooper pair condensate is
j q
v   A.
qn p m
For the Meissner-Ochsenfeld effect in a 1D geometry, described by Eq. (6.57), the vector potential
definition A  B yields A2 = L2B2  (m/q2np)B2, so the kinetic energy of Cooper pairs per unit
volume,
2 2
T mv 2 m q  2 m q  m B2 ,
 np  np   A  np   B 2

V 2 2 m 2  m  q 2 n p 2
is exactly equal to that of the magnetic field. Naturally, its account properly doubles the contribution to
the total energy from the London penetration layers.

79 See, e.g., A. Kadin, Introduction to Superconducting Circuits, Wiley, 1999.

80 In particular, because ferromagnetism generally suppresses superconductivity, so the two effects rarely coexist
in the same material.

Problems with Solutions Page 233

Essential Graduate Physics EM: Classical Electrodynamics

Problem 6.21. Calculate the inductance (per unit length) of a b c

superconducting cable with the round cross-section shown in the figure on the a
right, in the following limits:
(i) L << a, b, c – b, and I
(ii) a << L << b, c – b.
I
Solutions: Due to the system’s axial symmetry, the direction of the
magnetic field is normal to the direction  from the system’s axis, and its magnitude is a function of  
   alone. Hence the magnetic energy (per unit length of the cable) may be calculated just as
U 1 
l

2 0  B 2 (  )d 2  
0  B 2 (  ) d ,

where the integral should be extended to all distances at which B is substantial. Note that in order to use
this formula for the evaluation of the inductance in the usual sense of the word (i.e. the mutual
inductance between the internal and external conductors), the algebraic sum of their currents (i.e. the net
current in the cable) has to be zero, so according to the Ampère law (5.37), the field vanishes at  > c.
(i) In the limit L << a, b, c – b, we can neglect the magnetic field’s penetration into both
conductors of the cable,81 and use Eq. (5.20) of the lecture notes for the field between them:
0 I
B    .
2
An elementary integration gives
2b
U  b 2   0 I  d 0 I 2 b
l
 
0 a
B (  ) d   
 0  2  
a


4
ln ,
a
so, according to Eq. (5.71), the self-inductance per unit length is
L U /( I 2 / 2)  0 b
  ln .
l l 2 a
Comparing this result with Eq. (5.79) of the lecture notes, we see that this inductance is
somewhat lower than that of a normal-conductor cable (at relatively low frequencies), due to the
negligible magnetic field penetration into the conductors. Note, however, that the above expression is
valid for normal-conductor cables as well, provided that the frequency of the current is so high that the
skin-depth s (6.33) is much smaller than all transverse dimensions of the cable.
(ii) As Eq. (6.56) shows, in the limit a << L, the supercurrent is uniformly distributed across the
inner conductor, with density j = I/a2, just as in a uniform normal conductor. Thus we may repeat the
calculations of Sec. 5.3, which have led to Eq. (5.79), with the integration from 0 to b only, and get
Lm U /( I 2 / 2) 1  b 
    0 ln   ,
l l 2  a 4

81
This essentially means that in this limit, we are using the coarse-grain (ideal-diamagnetic) description of the
Meissner-Ochsenfeld effect.

Problems with Solutions Page 234

Essential Graduate Physics EM: Classical Electrodynamics

i.e. the same answer as for a normal-metal cable with (c – b)  0 – see Eq. (5.80). However, this is not
the end of the story. As was discussed in the solution of the previous problem, in superconductors, the
magnetic energy has to be summed up with the kinetic energy T of the Cooper-pair condensate.
Moreover, in our current case of a very thin inner conductor (a << L, i.e. Aa  a2 << L2), T is much
larger than U. Indeed, for the uniformly distributed supercurrent, the kinetic energy per unit length is
2 2
T mv 2 m  j  m I 
  m I ,
2
 Aa np  Aa n p   Aa n p 
l 2 2  qn p  2  qn p Aa 
 q 2 np Aa 2

so the so-called kinetic inductance associated with this effect,

Lk T /( I 2 / 2) m  L2  L2
        ,
l l np q 2 Aa Aa a 2

is much larger than the magnetic inductance Lm/l ~ 0.

Problem 6.22. Use the London equation to analyze the magnetic field shielding by a
superconducting thin film of thickness t << L, by calculating the penetration of the field induced by
current I in a thin wire that runs parallel to a wide planar thin film, at a distance d >> t from it, into the
space behind the film.
Solution: This problem is conceptually very close to that of Problem 2.39 on electrostatic
screening, so the reader is advised to review its solution first.82 The only substantial novelty of the
magnetic field is the vector character of its potential A. However, as was discussed in Sec. 5.2 of the
lecture notes,83 the potential created by the current I in a straight wire has only one Cartesian
component, directed along the wire, and it is clear that a superconducting film parallel to it cannot
change this alignment. (The proof of the last fact is simple: if A had a transverse component, would it
be directed clockwise or counter-clockwise?) So, in our geometry, the London equation (6.56) is valid
for the magnitude A of the only (say, z-) component of A = Anz.
Hence, we may return to the same reasoning as was used in the model solution of Problem 2.39.
Namely, due to the imposed condition t << L, the gradient of the vector potential A in the film is
dominated by its component along the axis (say, y) normal to the film’s plane, so the London equation
(6.56) is well approximated by its 1D version
2 A 1
 A,
y 2  L2
even if A is relatively slowly (on distances of the order of d >> t) changing in the plane of the film.
Integrating both sides of this equation over the film’s thickness t, we get
t / 2
A A 1
y
  2
y  L  Ady ,
t / 2
(*)

82 The current problem is more realistic, because (as was discussed in Sec. 6.4) the London penetration depth L is
typically of the order of 10–7-10–6 m, so superconducting films of thickness t << L are quite realistic.
83 See, e.g., the discussion leading to Eq. (5.49).

Problems with Solutions Page 235

Essential Graduate Physics EM: Classical Electrodynamics

where the indices  refer to opposite film’s surfaces. This relation shows that if t << L, the film can
create only a small change (of the order of At/L2) of the derivative A/y through its thickness, and
hence just a negligible variation (of the order of At2/L2) of the vector potential inside it. As a result, Eq.
(*) yields two boundary conditions for the potential’s distribution outside the film:
A A t
A  A  A,   2 A. (**)
y y  L
Now, inspired by the success of numerous charge- y
image analyses in Chapters 2-5, we may try to describe the
magnetic field (and hence its potential A) in the semi-space I or I"
where the actual current I resides (say, y > 0) as a this point “sees”
superposition of the fields induced by that current and its d both currents
image I’ running in parallel, at the same distance d from the t I and I’
film, but on its opposite side – see the figure on the right.
0 x
In addition, to describe the field’s remnants at the
other side of the film (at y < 0), we may use the effective d
this point “sees”
current I” co-located with the actual current. As a result, by current I” alone
using Eq. (5.51) of the lecture notes for each current’s  I'
potential in unlimited space, we may look for the potential
distribution in the form84

A( x, y )   0  

  I ln x   y  d 
2

2 1/ 2

 I' ln x 2   y  d  
2 1/ 2
, for y  0,
(***)
 
2  I" ln x 2   y  d 2 
1/ 2
, for y  0.
Plugging this solution into the boundary conditions (**), we see that they are indeed satisfied (so
Eq. (***) is indeed the unique solution of our boundary problem) if the effective currents I’ and I’’ obey
the following relations:
I  I'  I'' , I  I'   I"  td2 I'' ,
L
which are similar to those for image charges in the corresponding electrostatic screening problem. As a
result, the solution of this simple system of equations for I’ and I’’ is also similar. In particular, for the
ratio I”/I, which is an adequate measure of the field’s penetration behind the film, we get
I" 1
 .
I 1  td / 2 L2
As in electrostatics, the most interesting feature of this result is that the ratio starts to drop as
soon as t is increased to ~L2/d << L, i.e. the field may be well (though not exponentially well) screened
even by a film much thinner than the penetration depth – if d is large enough. Another formulation of the
same fact, which may be more revealing in some problems, is that a thin (t << L) superconductor film
has a certain characteristic lateral size scale

84 As was discussed in Sec. 3.3 of the lecture notes, absolutely the same solution (though with different constants
q’ and q”) describes the electric field penetration into a half-space filled with a linear dielectric.

Problems with Solutions Page 236

Essential Graduate Physics EM: Classical Electrodynamics

2 L2
  ,
t
usually called the perpendicular penetration length, which characterizes its diamagnetic properties. In
particular, it is easy to use Eqs. (5.28) and (6.55) to show (the additional task highly recommended to the
reader) that a supercurrent run through a strip of width w is uniformly distributed over its width only if w
<< . This is one of the reasons why in applications, superconducting strips are typically run very close
to a superconducting ground plane, which enforces the uniform current distribution even in wide thin
films – see, e.g., Problem 20 above.

Problem 6.23. Assuming that the magnetic monopole does exist and has a magnetic charge qm,
calculate the change I of current in a superconducting loop due to a passage of a single monopole
through its area. Evaluate I for a monopole with the charge’s value conjectured by P. Dirac, qm = nq0 
n (2/e) with an integer n, and compare the result with the magnetic flux quantum 0 (6.62). Review
your result for a similar passage of a single quasi-monopole magnetic charge formed at one of the ends
of a permanent-magnet needle – see Fig. 5.19 of the lecture notes and the accompanying discussion.
Hint: To simplify calculations, you may consider the monopole’s passage along the symmetry
axis of a round ring of radius R, made of a superconducting wire with a cross-section’s area A satisfying
the conditions L2 << A << R2.
Solution: As was discussed in Sec. 5.6 of the lecture notes (see Eq. (5.138) and its derivation),
the magnetic charges are usually defined so that if a single charge qm is located at point r’, it creates the
magnetic field
qm
Br   3
r  r' . (*)
z
4 r  r'
Let us direct the z-axis along the line of a slow passage of 
qm
the monopole, i.e. along the symmetry axis of the considered z'
superconducting ring, with the origin on its plane – see the figure
on the right. Then, taking into account the given condition A << R2,  I I
the flux of the field (**) through the ring’s area may be calculated
as85 0 ρ R
R
 ext   Bn d r  2π  B z ρdρ 
qm R  z'  ρdρ

2

S 0
2 0  ρ  z' 2 3 / 2
2

q m R d  ρ 2  z' 2  qm 1 ρ  R qm  z' 
 z'    z'    sgn  z' . (**)
4 0  ρ 2  z' 2 3 / 2 2  ρ 2  z' 2 1/ 2 ρ  0 2   R 2  z' 2 1 / 2 

Due to the strong condition A >> L2, one might assume that due to the Meissner-Ochsenfeld
effect, the full magnetic flux (6.78) through the ring,
   ext  LI

85Here the observation’s point distance  from the axis z = 0 is typeset in Roman font, to avoid any chance of
confusion with the electric charge density , typeset in Italics.

Problems with Solutions Page 237

Essential Graduate Physics EM: Classical Electrodynamics

(where I is the supercurrent in the ring, with the positive direction indicated in the figure above, and L is
the ring’s self-inductance) should be constant during the monopole’s passage, the current’s change
(referred to its initial value at –z’ >> R when ext = 0) should be

 ext q m  z' 
I    2  sgn  z' . (***)
L 2 L   R  z' 2 1 / 2 
Here we have arrived at a contradiction: the
 ext
function ext(z’), and hence the calculated current I(z’),
are discontinuous at z’ = 0, experiencing jumps ext = –  qm / 2
qm and hence I = – qm/L – see the figure on the right.  ext  q m
This is not some technical error in our calculation.
Indeed, if the magnetic charge is slightly below the ring’s 0 z'
plane, its field lines piercing the ring’s area are directed
up, so their flux is positive and approaches ½ of the total  qm / 2
flux full = qm of all field lines originating from the point
charge. As soon as the charge crosses the plane, the signs of the field and the flux reverse, with ext
keeping its magnitude – exactly as Eq. (***) predicts.86 So, our math is right but for physics, such an
instant jump of such a measurable variable as current is unacceptable.
This problem is actually much broader: in 1931, Paul Dirac noticed that a magnetic monopole’s
passage even through usual matter would provide an instant phase shift of any quantum-mechanical
wavefunction met on its way. In order to make this disrupting effect less drastic, Dirac suggested that
the monopole charges qm may be multiples of
2
q0  ,
e
because in this case, the shifts of the wavefunction phases would be multiples of 2, with no observable
consequences for ordinary matter. For our case of a ring I
made of a usual superconductor with the effective
supercurrent carrier charges equal to –2e rather than –e,
the jump ext induced by such qm (with the replacement n 0 2 n 0

e  2e) would be a multiple of 20, where 0 is the flux L L
quantum (6.62), making the new quantum state of the
superconducting Bose-Einstein condensate in the ring
compatible with the same current I. Assuming that during 0 z'
this hypothetical instantaneous jump, the current would be n 0
continuous, we arrive at the function I(z’) shown with the 
L
solid line in the figure on the right, where the dashed line
shows the unphysical result (****). The resulting change of the current by 2n0/L (where n is some
unknown integer) would be sufficiently large to be measured, for example, by a SQUID magnetometer –
see Sec. 6.5.

86 By the way, from this reasoning, it is clear that the flux jump’s magnitude  = –qm does not depend on the
loop’s shape and on the exact trajectory of the magnetic monopole. (The reader is challenged to use Eq. (*) for a
more formal proof of this fact, by employing the same approach as in the solution of Problem 1.7.) This jump also
remains the same if the monopole moves with a very high (relativistic) velocity.

Problems with Solutions Page 238

Essential Graduate Physics EM: Classical Electrodynamics

Actually, magnetic monopole searches using such superconducting loop setups were carried out
in the 1980s by several groups, and one observed event (compatible with n = 4) was even claimed – but
then never confirmed by either that or any other research team.
Note that the above contradiction does not appear if the ring’s plane is pierced by an end of a
thin ferromagnetic needle, despite the fact that the magnetic field it creates around is also described by
Eq. (*), in this case, with qm = 0M0A – see Fig. 5.19a of the lecture notes and the accompanying
discussion. Indeed, as was shown in the solution of Problem 5.25, the magnetic field inside the thin
needle is nearly uniform: H = –M0, B = 0H = –0M0. As a result, in this case, the jump of the flux (**)
due to the needle’s field outside it is exactly compensated by the equal and opposite jump of the flux due
to the field inside the needle. This means that the function I(z’) behaves as the last figure above shows
(with the gradual change by qm/L = 0M0A/L) without any hypothetical assumptions. The passage,
through the same ring, of the second end of the needle, which carries an equal and opposite magnetic
charge (see Eq. (5.138) again), induces the opposite change of the current, returning it to the initial
value.

Problem 6.24. Use the Ginzburg-Landau equations (6.54) and (6.63) to calculate the largest
(“critical”) value of supercurrent in a uniform superconducting wire with a cross-section area much
smaller than L2.
Solution: As was discussed in the solution of Problem 19, in a wire with a cross-section so small,
the current density, and hence the gauge-invariant combination ( – iqA) participating in the
Ginzburg-Landau equations, are uniformly distributed along the cross-section, so Eqs. (6.54) yields
qn p 2   q 
j  jz     A , (*)
m  z  
where A is the (only) Cartesian component of the vector potential, directed along the wire’s length,
which is taken for the z-axis. As was discussed at the derivation of Eq. (6.55) of the lecture notes, in this
geometry, we may (just for the notation simplicity) select the gauge  = const, so Eq. (*) yields
m  L2
A 2
j 2
j.
q 2 np  
Plugging this expression into the second form of Eq. (6.63), taken
in the same gauge, we get the following equation:
4

 1 
2

1   .
2
 q 2
 4 2
 
2 L
j    (**) 0.1
  
At j  0, its solution gives the unperturbed value (equal to
1) of the modulus of the wavefunction  of the Cooper pair 0.05
condensate, but an increase of current suppresses it – see the figure
on the right, which shows the right-hand side of Eq. (**) as a
0
function of 2. As an elementary differentiation shows, the 0 0.2 0.4 0.6 0.8
function reaches its maximum at 
2

Problems with Solutions Page 239

Essential Graduate Physics EM: Classical Electrodynamics

 1   
2
2 2 4 2 2  2 4
  , with    1     0.148 .
3 max
3  3  27
From here, the maximum (“critical”) value of the supercurrent density is
1/ 2
 4  
jc    . (***)
 27   q  L2
Increasing j above this value destroys the superconductivity, driving the wire into its “normal”
state, with  = 0. Note that an external magnetic field H, applied to a bulk superconductor, has a
qualitatively similar effect. Indeed, as Eqs. (6.38) and (6.57) of the lecture notes show, it induces a
surface supercurrent with a linear density J ~ H,87 distributed within a layer of thickness L, i.e. with the
areal density j ~ H/L. When this density approaches the critical value (***), i.e. at
  0
H ~ Hc  , i.e. B ~ Bc  H c   . (****)
2 q  L 2 q  L 2 2 L 
the superconductivity in the surface layer starts to be suppressed. As the detailed solutions of the
Ginzburg-Landau equations show,88 the behavior in higher fields may be rather complicated and
depends on the relation between the superconductor’s characteristic length L and . In the so-called
type-I superconductors with L < /2, the superconductivity (at least in cylindrical geometries) is
suppressed, at H = Hc, simultaneously in the whole sample. However, in the type-II superconductors
that have /2 < L, the magnetic field may penetrate into the superconductor in the form of separate
Abrikosov vortices – see the discussion in Sec. 6.5 of the lecture notes.89 Moreover, in the case /2 <<
L, such vortices may penetrate the superconductor even at much lower fields – see Eq. (6.32) of the
lecture notes and also the next problem.

Problem 6.25. Use the discussion of a long straight Abrikosov vortex, in the limit  << L, in
Sec. 6.5 of the lecture notes, to prove Eqs. (6.71)-(6.72) for its energy per unit length and the first
critical field.
Solution: As it follows from the model solutions of Problems 20-21, the total energy of the
vortex is the sum of its magnetic field energy:
1
2 
U B 2 d 3r ,

and the kinetic energy of the Cooper-pair condensate:

2
m m  j  3  L2
T   np v 2 d 3 r   np   d r m
 j 2d 3r  j
2
d 3r ,
2 2  qn  2q 2 n p 2
 p 

87 As Eq. (6.38) shows, in a cylindrical geometry (such as shown in Fig. 5.15 or Fig. 6.2) the relation J = Hext is
exact; however, in the general case, there is an additional geometric factor in their relation.
88 See, e.g., Chapters 1 and 5 in M. Tinkham, Introduction to Superconductivity, 2nd ed., McGraw-Hill, 1996.
89 For such superconductors, the above result for the critical current is valid for wires with cross-section areas

smaller than not only L2 but also  2.

Problems with Solutions Page 240

Essential Graduate Physics EM: Classical Electrodynamics

where, for the last step, Eq. (6.46) of the lecture notes was used. Plugging into these expressions,
respectively, Eqs. (6.68) and (6.70), we get the following energies per unit length of the vortex:
 2 
U 1 1        02  2
 2  B 2  d   0 2  K   L K 0  d ,
4 L2 
 B d r
2 2
 2
 d 
2 2  ~   2 L
0

~
ξ
l  
 2 
T  L2  L2  0      02  2
 2  j  d    ~ K   L K 1  d ,
4 L2 
 j d r
2 2 2 2
L
 1
2
 d 
 2 L
2
l 2 2 ~  
so the total energy of the vortex per unit length, i.e. its tension, is


 K    K  d ,
U T  02
T   2
0 1
2

l 4 L2 

where  ~ /L << 1. (This lower cutoff is due to the fact, that Eqs. (6.68) and (6.70) are only valid at
distances much larger than the coherence distance  – see the discussion in Sec. 6.5 of the lecture notes.)
Such integrals are well-known:90
2
 K n  d 
2

2
K 2
n    K n1  K n1  ;
taking into account that, according to Eq. (2.150), K–1() = –K1(), for our particular case they give
 
 2 2 
 K    K   d    
K 0    2 K 12    K 0  K 2    . 
2 2
0 1
  2 
As Eq. (2.158) and Fig. 2.22 of the lecture notes show, on the upper limit all these functions vanish.
Now using the second of Eqs. (2.157) for their approximate values on the lower limit, we get

2 L
 K    K12  d 1
2
 K 0  K 2    ln  ln ,
 
0
 2
thus proving Eq. (6.71):
 02 L
T  ln .
4 2
L 
A more detailed theory,91 involving an explicit solution of the Ginzburg-Landau equation (6.53)
in the vortex core (i.e. at distances at  ~  << L from its axis), confirms this result, offering just a small
correction: ln(L/)  ln(L/) + 0.50.
In order to calculate the first critical field Hc1, let us use Eq. (6.17) of the lecture notes for the
Gibbs potential energy, corrected for the kinetic energy T of the Cooper-pair condensate by the
replacement U  U + T  Tl, for a single vortex inside a superconductor:
UG
 T   H ext  Bd 2 r ,
l

90 This equality, valid for any Bessel functions, follows from the recurrence relations (2.142).
91 C. R. Hu, Phys. Rev. B6, 1956 (1972).

Problems with Solutions Page 241

Essential Graduate Physics EM: Classical Electrodynamics

where the integral may be limited by the cross-section of the vortex. In the cylindrical geometry we are
considering, the field B of the vortex is directed along the external field Hext, which is constant. Hence
we may take Hext out of the integral, and the remaining integral of B is just the quantized magnetic flux
0 of the vortex; hence
UG
 T  H ext  0 .
l
It is energy-favorable for a vortex to enter the superconductor if this value is lower than UG in the
absence of the vortex (with our choice of the arbitrary constant, this value is zero), i.e. if Hext > Hc1,
where
T 0 
H c1   ln L ,
 0 4 L2

thus proving Eq. (6.72) of the lecture notes.
Note that the condition Hext > Hc1 makes the vortex entrance possible but does not guarantee it,
because it may be prevented by a substantial potential barrier at the surface, which is fully suppressed
only when the field reaches the (much higher) value Hc estimated in the previous problem:
 0 
Hc   ~ L H c1  H c1 .
2 q  L 2 2 q  L 
However, certain experimental tricks (such as using slightly non-cylindrical geometries, e.g., some
surface roughening) may enable reproducible vortex entry and exit at fields very close to Hc1.

Problem 6.26.* Use the Ginzburg-Landau equations (6.54) and (6.63) to prove the Josephson
relation (6.76) for a small superconducting weak link, and express its critical current Ic via the Ohmic
resistance Rn of the same weak link in its normal state.
Solution:92 Let the weak link size scale a be much smaller than both the Cooper pair size  and
the London penetration depth L. The first of these relations (a << ) makes the left-hand side of the
second form of Eq. (6.63), which scales as 2/a2, much larger than 1, i.e. much larger than its right-hand
side. On the other hand, according to the discussion in Sec. 6.4 of the lecture notes, the second relation
(a << L) allows us to neglect magnetic field effects, and hence drop the term (–qA) from the parenthesis
in Eq. (6.63), reducing it to just our familiar Laplace equation, but now for the complex wavefunction:

 2  0 .
Since weak coupling of the superconducting electrodes cannot change  in their bulk, this
differential equation should be solved with the following simple boundary conditions:
  e i1 , for r  r ,
 1
 (r )  
i 2
  e , for r  r2 ,

92 This simple and elegant calculation belongs to L. Aslamazov and A. Larkin, JETP Lett. 9, 87 (1969).

Problems with Solutions Page 242

Essential Graduate Physics EM: Classical Electrodynamics

where r1 and r2 are some points well inside the corresponding superconductors, i.e. at distances much
larger than a from the weak link. It is straightforward to verify that the (unique) solution of this
boundary problem may be expressed as follows,
i1 i 2
 (r)   e f (r )   e 1  f (r),
via the real function f(r) that satisfies the Laplace equation and the boundary conditions
1, for r  r1 ,
f (r )  
0, for r  r2 .
The function f(r) depends on the weak link’s geometry and may be rather complicated, but we do
not need to know it to get the most important result. Indeed, plugging this solution into Eq. (6.54), with
the term –qA again ignored as being negligibly small, we get
qnp
j f sin  .
m
Integrating this relation over any cross-section S of the weak link, we arrive at B. Josephson’s result
(6.76),
I  I c sin  .
with the following critical current:
qn p T 
Ic  
m  f 
S
n d 2r . (*)

This result may be readily expressed via the resistance of the same weak link in the “normal”
(non-superconducting) state, say at T > Tc. Indeed, as we know from Sec. 4.3, the distribution of the
electrostatic potential  at normal conduction also obeys the Laplace equation, with boundary conditions
that may be taken in the form
V , for r  r1 ,
 (r )  
 0, for r  r2 .

Comparing the boundary problem for the function (r) with that for the function f(r), we get  = Vf. This
means that the gradient f that participates in Eq. (*) is just (–E/V) = (–j/V). Hence the integral in that
formula is just –I/V = –1/Rn, where Rn is the resistance of the weak link in its normal state. As a
result, we get the expression
qnp 1
Ic  ,
m Rn
showing that the IcRn product is independent of the link’s geometry, though it does depend on
temperature – vanishing, together with np, at T  Tc.
The microscopic theory of superconductivity (which is beyond the framework of this series)
confirms this conclusion and shows that well below the critical temperature, IcRn of such short weak
links is of the order of (0)/e – typically, a few mV.

Problems with Solutions Page 243

Essential Graduate Physics EM: Classical Electrodynamics

Problem 6.27. Use Eqs. (6.76) and (6.79) of the lecture notes to calculate the coupling energy of
a Josephson junction and the full potential energy of the SQUID shown in Fig. 6.4c.
Solutions: The elementary work of an external voltage V applied to a lumped two-terminal
device that carries current I = dQ/dt (where Q is the total electric charge moving through the device)
may be calculated as dW = VdQ = VIdt. Combining this expression with Eqs. (6.76) and (6.79), we get
 d I I
dW  VIdt  I c sin  dt  c sin  d   c d cos   .
2e dt 2e 2e
Hence the work needed to drive the junction’s phase to a certain value , i.e. its coupling (“Josephson”)
energy UJ is93

I
U J     dW   E J cos   const, where E J  c .
2e
Note that the function UJ() is highly nonlinear, and quadratic only at   << 1 (when Eq. (6.76) is
reduced to I  Ic):94
2 E I2
U J    E J  const  2J  const .
2 Ic 2
In a SQUID, the Josephson energy UJ should be added to the magnetic energy Um = LI2/2 of the
superconducting loop, so the energy of the system may be represented as
LI 2    ext  I c 2
2

U  E J cos   const   cos  const , (*)

2 2L 2e 0
where at the second step, the first of Eqs. (6.78) was used. If the magnetic flux  is understood as the
generalized coordinate of the system, this function U = U() represents its effective potential energy
(despite the “kinetic energy” terminology for its component UJ), so its minima correspond to stable
states of the system.95 The figure below shows this potential energy profile, plotted for several values of
the external flux, and two characteristic values of the normalized LIc product,   2LIc/0.
An elementary analysis of Eq. (*) shows that the fixed-point requirement U/ = 0 yields the
stationary relation (ext) given by Eqs. (6.77)-(6.78) and shown in Fig. 6.6 of the lecture notes. The
negative-slope branches of those plots correspond to the maxima of the energy U (with 2U/2 < 0)
and hence are unstable. As the plots show, a SQUID with  < 1 always has just one stable state , at 
<< 1 closely following ext. On the contrary, a SQUID with  > 1 may have several (at  >> 1, many)
stable states, corresponding to quasi-quantization of the flux.
The flexibility of this U() profile to the change of parameters ext and  is broadly used in
superconductor electronics; for example, it was used for the first experimental observation of the

93 Physically, UJ is just a form of the kinetic energy of the Cooper-pair condensate, similar to that discussed in the
solutions of Problems 6.19, 6.20, and 6.22, but for the specific case of a Josephson junction, with its highly
nonlinear dynamics.
94 By the way, the last expression shows that the non-magnetic (kinetic or Josephson) inductance of the junction

for small currents is LJ  EJ/Ic2  /2eIc. This formula is broadly used in superconductor electronics.
95 Since U is the potential energy of the system in a fixed magnetic external field, which plays the role of the
generalized external force, its more fair name is the Gibbs potential energy – see, e.g., Sec. 6.2.

Problems with Solutions Page 244

Essential Graduate Physics EM: Classical Electrodynamics

quantum interference of two macroscopic flux states,96 while the bistability possible at  > 1 is used in
virtually all modern digital devices and circuits based on the Josephson effect.97

  0.3   3.0
4 4

U U
EJ EJ
2 2
 
 
 
0.5
0  ext 0.25 0
 0.0 0.75 1.0
0
 0.5 0 0.5 1 1.5  0.5 0 0.5 1 1.5

 / 0  / 0
Finally, note that for a full analysis of the system’s dynamics, the potential energy (*) has to be
augmented by the electrostatic energy CV2/2 of the Josephson junction. According to Eq. (6.79), this
energy equals
2 2
C     d 
    ,
2  2e   dt 
and hence may be considered the kinetic energy T of the SQUID. The sum (T + U) is the full energy of
the SQUID, which is conserved in the absence of dissipation. In many applications, however, such
dissipation, mostly due to the Ohmic current of “normal” (unpaired) electrons through the Josephson
junction, flowing at V  0, is essential for a fair description of the system’s dynamics.

I j 1 Ij I j 1
Problem 6.28. Analyze the possibility of wave propagation
Vj V j 1
in a long uniform chain of lumped inductances and capacitances –
see the figure on the right. L L L
 C C 
Hint: Readers without prior experience in electromagnetic
wave analysis may like to use a substantial analogy between this link j  1 link j link j  1
effect and mechanical waves in a 1D chain of elastically coupled particles.98
Solution: Applying the 1st Kirchhoff law (the “node rule”, see Eq. (6.88a) of the lecture notes) to
the jth link of the chain, numbered as the figure above shows, we get
I j 1  I j  CV j  0 , (*)

96 J. Freedman et al., Nature 406, 43 (2000).

97 See, e.g., A. Kadin, Introduction to Superconducting Circuits, Wiley (1999).
98 See, e.g., CM Sec. 6.3.

Problems with Solutions Page 245

Essential Graduate Physics EM: Classical Electrodynamics

where the last term on the left-hand side is the time derivative of the electric charge Qj = CVj of the jth
capacitor. On the other hand, the 2nd Kirchhoff law, i.e. the “loop rule” (6.88b), applied to the contour
containing the jth inductance and two adjacent capacitors, yields
V j  V j 1  LI j  0 , (**)
where Eq. (6.85) was used for the voltage drop on the inductance L that carries current Ij(t).
Generally, the system of differential equations (*) and (**) for a long chain, i.e. for many
functions Vj(t) and Ij(t), may have rather complex solutions, but for our task, it is sufficient to look for
the case when they are sinusoidal functions of time, with a constant phase shift (say, ) between
adjacent links:99
I j  Re I  exp i j  t , V j  Re V exp i j  t  . (***)

Plugging this solution into Eqs. (*) and (**), and then canceling the common factor exp{i(j-t)}, we
get the following system of two linear homogeneous equations for the complex amplitudes I and V:
 
I  e i  1  iCV  0,

V 1  e i   iLI
   0.
The equations are compatible if the system’s determinant equals zero:

e  i  1 iC
 0.
i
iL 1 e
This condition yields the following dispersion relation between  and :100
 2
 2   max
2
sin 2 , where  max  .
2 LC 1 / 2
So, the LC-chain indeed supports the process described by Eqs. (***) with a real parameter ,
i.e. sinusoidal traveling waves,101 if their frequency is below max. (If the chain is excited at a higher
frequency, the solution (***) is still valid but with purely imaginary values of , and describes an
exponential decay of the excitation magnitude as the link number j grows, i.e. as the observation point
moves further from the excitation point.)

Problem 6.29. A sinusoidal e.m.f. of amplitude V0 and I j 1 Ij I j 1

frequency  is applied to an end of a long chain of similar lumped Vj V j 1
resistors and capacitors, shown in the figure on the right. Calculate
the law of decay of the ac voltage amplitude along the chain.  R C R C R 
st
Solution: Applying the 1 Kirchhoff law (the “node rule”, link j  1 link j link j  1

99 See, e.g., CM Eq. (6.26) and its discussion.

100 A similar dispersion relation for mechanical elastic waves plays an important role in the theory of solids – see,
e.g., CM Sec. 6.3 and SM Sec. 2.5.
101 For a chain with a spatial period d, the exponents in Eq. (***) may be rewritten in a more usual form as

exp{i(kz – t)}, with k  /d, where the z-axis is directed along the chain.

Problems with Solutions Page 246

Essential Graduate Physics EM: Classical Electrodynamics

see Eq. (6.88a) of the lecture notes) to the jth link of the system, numbered as the figure above shows, we
get the same equation as in the previous problem:
I  I  CV  0 .
j 1 j j (*)

On the other hand, the 2nd Kirchhoff law (the “loop rule”, Eq. (6.88b) of the lecture notes), applied to the
contour containing the jth resistor and two adjacent capacitors, yields a somewhat different result:
V j  V j 1  RI j  0 . (**)

Taking into account that the chain is driven by a sinusoidal e.m.f. of frequency , we may look
for a solution of the system of Eqs. (*) and (**) in the same form as in the previous problem:

I j  ReI 0 exp j  it, V j  ReV0 exp j  it . (***)

Plugging this solution into Eqs. (*) and (**), and then canceling the common factor exp{–j –it}, we
get the following system of two linear, homogeneous algebraic equations for two complex amplitudes I0
and V0,
I 0 e  1  iCV0  0,
V0 1  e    RI 0  0
The equations are compatible if the system’s determinant equals zero:
e  1 i C
 0, i.e. 2cosh   1  iRC  0.

R 1 e
This equation yields102 the following expression for the constant , which is now a complex number:


  cosh -1  1  2i   cosh 1 1   2 2 
1/ 2

   i cos 1
1    
2
1
2 1/ 2
 
.

where   RC/4. If we are not interested in the phase shift described by the second term of this
expression, we may use the first term to write the following law of decay of the real amplitude Vj of the
ac oscillations in jth link in a semi-infinite chain:
10
V j  V0 exp Re   j

 
 V0 exp  cosh 1 1   2 2 
1/ 2

  j .
Re 
1
The figure on the right shows the dependence of the
decay factor Re on frequency – or rather on the dimensionless
combination . At relatively low frequencies ( << 1), Re
 (2)1/2 << 1, i.e. the characteristic depth (expressed in the
0.1
number of links) d  1/Re of the oscillations’ penetration into 0.01 0.1 1 10 100

102The simplest way to get the last form of this formula is to rewrite the first one as cosh  cosh(Re+ iIm) 
cosh(Re) cos(Im) + i sinh(Re) sin(Im) = 1 + 2i, require the real and imaginary parts of both sides to be
separately equal, and then solve the resulting simple system of two equations for Re and Im.

Problems with Solutions Page 247

Essential Graduate Physics EM: Classical Electrodynamics

the chain, is much larger than 1. However, as the frequency  is increased well beyond ~1/, i.e. beyond
~1/RC, we get Re >> 1, i.e. the ac excitation essentially does not go beyond the first link of the chain.
Due to this property, electrical engineers call such a circuit, used most typically with just a few links, the
low-pass RC filter.

Problem 6.30. As was discussed in Sec. 6.7 of the lecture notes, the displacement current concept
allows one to extend the Ampère law to time-dependent processes as

 H  dr  I
C
S  
t S
Dn d 2 r .

We also have seen that this generalization makes the integral H  dr over an Q Q
external contour, such as the one shown in Fig. 6.10, independent of the choice C
of the surface S limited by the contour. However, it may look like the situation is I I
different for a contour drawn inside a capacitor – see the figure on the right.
Indeed, if the contour’s size is much larger than the capacitor’s thickness, the S
magnetic field H created by the linear current I on the contour’s line is virtually
the same as that of a continuous wire, and hence the integral H  dr along the
contour apparently does not depend on its area, while the magnetic flux Dnd2r d
does, so the equation displayed above seems invalid. (The current IS piercing this
contour evidently equals zero.) Resolve this paradox, for simplicity considering an axially-symmetric
system.
Solution: Let us consider a circular contour C of radius  < R, where R is the radius of circular
parallel plates of the capacitor. If  >> d, we can use Eq. (5.20) of the lecture notes for the magnetic
field HI induced by the current I in the wire, so HIdr along the contour C is indeed equal to I. However,
at dI/dt 0, the system also has currents flowing in the capacitor plates in the radial direction, necessary
to change the surface density  = Q/R2 of the full charges Q of the plates in time. For example, the
left plate’s charge residing outside a circle of radius  equals (R2 – 2), so the linear density J() of
the surface currents in the plate may be found from the charge conservation requirement:

2J (  ) 
d
   
 R2   2    R2   2
d
,  giving J 
R 2
  2  d
.
dt dt 2 dt
The magnetic field between the plates obeys Eq. (6.38), so it is also tangential to the circular contour C,
opposite in direction to that of straight wire, and has the magnitude

H J    J   
R 2
  2  d R 2   2  dQ R 2   2 
  I.
2 dt 2R 2 dt 2R 2
As a result, the net integral of the field is
R  2
2

2
 H  dr   H
C C
I  dr   H J  dr  I   H J
C C
dr  I  2
2R 2
I  2 I,
R
so it exactly matches the integral of the displacement current over the surface area 2:

Problems with Solutions Page 248

Essential Graduate Physics EM: Classical Electrodynamics

  d d 1 dQ 2  2
t S
Dn d r   d r 
dt S
2 2
d r
2
  2
2
  2 I ,
t S dt R dt R
thus resolving the paradox.

Problem 6.31. A straight, uniform, long wire with a circular cross-section of radius R, made of an
Ohmic conductor with conductivity , carries dc current I. Calculate the flux of the Poynting vector
through its surface and compare it with the Joule rate of energy dissipation.
Solution: As was discussed in Sec. 4.3 of the lecture notes, in such a uniform, long wire, the dc
current is distributed uniformly, with a constant density
I
j .
 R2
Per the differential form of the Ohm law, the electric field’s distribution inside the wire is also uniform,
j I
E  ,
  R 2
and is directed, as the current density, along the wire’s length.
The magnetic field distribution in such a wire was (easily) calculated in Sec. 5.2; for the wire’s
surface, the second of Eqs. (5.38) yields
I
H .
2R
Since the direction of the vector H is tangential to the wire’s cross-section (see Fig. 5.5 of the lecture
notes), i.e. perpendicular to the vector E, the Poynting vector (6.114) on the wire’s surface is directed
normally to it, into the wire’s bulk, and has the magnitude
I I I2
S  EH   .
 R 2 2R 2 2R 3
Hence the power flow into a wire’s fragment of length l is
I2
PPoynting  2RlS  l.
 R 2
On the other hand, according to the Joule law (4.39), the total power of energy dissipation in
such a fragment is
2
j2  I  1 I2
PJoule  pV  R l   2  R 2 l 
2
l.
  R    R 2
So, we have arrived at a simple result, PJoule = PPoynting, thus confirming the natural picture of the
continuous power flow in the system – from the current source to the electromagnetic field outside the
wire, then through the wire’s surface into its bulk, and then to heat.

Problems with Solutions Page 249

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 7. Electromagnetic Wave Propagation

Problem 7.1. Find the temporal Green’s function of a medium whose complex permittivity ()
obeys the Lorentz oscillator model given by Eq. (7.32) of the lecture notes, by using:
(i) the Fourier transform of the underlying Eq. (7.30), and
(ii) the direct solution of that equation.
Hint: For the Fourier-transform approach, you may like to use the Cauchy integral.1
Solutions:
(i) The general relation between () and the Green’s function G() is given by Eq. (7.26b) of
the lecture notes:

 ( )   0   G ( )e i d . (*)
0

Though the Green’s function G() does not have a physical sense for  < 0, mathematically nothing
prevents us from extending the integral (*) to the whole axis – <  < +, by taking G() = 0 for  < 0.
Now we can write the reciprocal transform:

1
  ( )   e d
iω
G ( ) 
2
0


For the particular form (7.32) of the function (), this equality becomes

nq 2 1
G ( )   e i d . (**)
2 m  ( 0   )  2i
2 2

In order to calculate this integral, let us first represent the denominator as a product –( – +)(
– –), where  are the values of the complex variable  at which the denominator turns to zero, i.e. the
roots of the corresponding quadratic equation. An elementary calculation yields

ω    0'  i , 
where  0'   02   2 
1/ 2
.

Now we may rewrite the fraction under the integral in Eq. (**) as a sum of two simple singularities
(poles):
1 1 a b
   .
( 0   )  2i
2 2
(  ω  )(  ω  )   ω    ω 
The constants a and b may be calculated from the requirement for the two last expressions to be
identical; this gives
1
a  b   .
2 0'
As a result, Eq. (**) turns into

1 See, e.g., MA Eq. (15.2).

Problems with Solutions Page 250

Essential Graduate Physics EM: Classical Electrodynamics

q2  d 
 
i d i
4 m 0'     ω  
G ( )    e  e .
  ω 

Since at  > 0, the factor e–i, considered a function of

the complex argument   ’ +i”, tends to zero at    ω"
and ” < 0, each integral in the last displayed formula may be  0' 0  0'
replaced by the one over the closed contour C on the plane [’,  '
”], shown with the dashed line in the figure on the right. This ω  ω ω  ω
integral may be worked out using the Cauchy integral formula:
C ω 
dω
 f (ω) ω  Ω  2 if (Ω) .
C

The formula is valid if the function f() is analytical (as our e–i is), and the pole  is located within
the area encircled by the contour C. This is correct in our case because both poles  are in the negative
semi-plane – see the figure above. However, the Cauchy theorem implies the positive
(counterclockwise) integration’s direction along a closed contour, while our direction is negative, so we
have to change the signs before both integrals. As a result, we get

G ( ) 
nq 2
4 m 0'

iω 
2 i e   e iω 
m 0'

nq 2 
e sin  0' .

(ii) In Sec. 7.2 of the lecture notes, Eq. (7.32) was derived from the Lorentz oscillator model: a
set of independent classical oscillators obeying the differential equation (7.30):
d 2x dx q
2
 2   02 x  E (t ) .
dt dt m
Its temporal Green’s function g() is just the solution of this differential equation for t =  and E = (),
while the Green’s function for the electric polarization of the media with n independent similar
oscillators per unit volume is G() = nqg(), so we may write
d 2G dG nq 2
 2   2
G   ( ) .
d
0
d 2 m
At  > 0, this equation is homogeneous,
d 2G dG
 2   02 G  0 , (***)
d 2
d
and has a simple, well-known solution2

G ( )  cc cos  0'  c s sin  0'  e  , (****)

where 0’  (02 – 2)1/2 is the same as in the first approach.

2 If this is not evident, please consult CM Sec. 5.1.

Problems with Solutions Page 251

Essential Graduate Physics EM: Classical Electrodynamics

In order to find the constants cc and cs, we need to formulate the appropriate initial conditions for
Eq. (***). For that, let us integrate both parts of the initial (inhomogeneous) equation for G() over a
small time interval of width 2  0, centered to point  = 0:

 

 dG dG  nq 2
      0   
2
     d      2 G    G    Gd .
 d  
m
Due to the causality principle, both the Green’s function and its derivative should equal zero for all
moments t < t’, i.e.  < 0. Moreover, G() should be a continuous function, because is proportional to
the oscillator’s coordinate, which cannot change instantly even if the acting force is infinite at some
moment. On the other hand, under such an infinite force pulse, the derivative dG/d may experience a
finite jump,3 so dG/d  = + may be different from zero. Since in the limit   0, all other terms on the
left-hand side vanish, we get the following initial conditions for the homogeneous equation (***)
dG nq 2
G (0)  0, ( 0)  .
d m
Applying them to the above general solution of this equation, we get cc = 0, cs = nq2/m0’, so
nq 2 
G ( )  e sin  0' ,
m 0'
i.e. (luckily :-) the same result as in the first approach.

Problem 7.2. The electric polarization of some material responds to an electric field step4 in the
following way:
 
P t    1 E 0 1  e  t /  ,
0,
if E t   E 0  
for t  0 ,
for 0  t ,
 1,
where  > 0 and 1 are some constants. Calculate the complex permittivity () of this material, and
discuss a possible simple physical model giving such dielectric response.
Solution: Let us calculate the dielectric response of an arbitrary system described by Eq. (7.23)
of the lecture notes to such an electric field step:
  0,
 for t    t
Pt    E t   G   d  E0    G   d  E 0  G   d .
0 
0 1, for   t  0

Comparing this response with that given in the problem's assignment, we see that the temporal Green's
function of this system has to obey the following equation:

1  e t / ,
t

 G  d  
0
1 for t  0.

3 If in doubt about this point, please recall any of the standard collision problems in classical mechanics, solved
assuming a finite momentum transfer between two bodies during an infinitesimal time interval.
4 This function E(t) is of course proportional to the well-known Heaviside step function (t) – see, e.g., MA Eq.

(14.3). I am not using this notion here just to avoid confusion between two different uses of the Greek letter .

Problems with Solutions Page 252

Essential Graduate Physics EM: Classical Electrodynamics

Differentiating both sides of this relation over t, we get

G t    1
d
dt
 
1  e t /  1 e t / ,

for t  0 .

Now it is straightforward to use Eq. (7.26b) to calculate


 1   / i 1
     0   G  e i
d   0   e e d   0  .
0
 0 1  i
Comparing this expression with Eq. (7.32) that follows from the Lorentz oscillator model (7.30),
we see that they coincide in the low-frequency limit  << 0 if we take
nq 2 
1  and τ  , with   m 02 and   2m 0 .
 
In other words, such a dielectric response is pertinent, for example, to a medium whose independent
electric dipoles p = qx are due to the heavily overdamped charge displacements x(t),5 obeying the
following equation of motion:
x  x  qE t  ,
i.e. having negligible inertia.

Problem 7.3. Calculate the complex permittivity () of a material whose temporal Green’s
function, defined by Eq. (7.23) of the lecture notes, is

G    G0 1  e  / , 
with some positive constants G0 and . What is the difference between this dielectric response and the
apparently similar one considered in the previous problem?
Solution: With the given G(), Eq. (7.26) of the lecture notes yields
 
e  /
   1  i 

     0  G0  1  e  / e i d   0  G0   e
  .
0  i i  1 /   
  0
We see that the exponential factor is uncertain at the upper limit, making the first term uncertain. This
technical problem may be treated in the following standard way: adding to  (before the integration) a
small imaginary part i, with  > 0,6 and then (after the integration) taking the limit  0:

     0  G0 lim  0  1  e  / e 
   d  

G G
0

i   i 
  0  2 . (*)
i 1  i     i
0
0

5 See, e.g., CM Sec. (4.1).

6 This sign correctly reflects the properties of any stable physical system, by providing a gradual suppression of
its response as the time interval between the cause and effect tends to infinity.

Problems with Solutions Page 253

Essential Graduate Physics EM: Classical Electrodynamics

Concerning the second question, the temporal Green's function of a system is, by definition, its
response to a delta-functional pulse (in our case, of the electric field), rather than to a step-like excitation
specified in the previous problem, so the Green's functions and hence all dielectric properties of the
systems considered in this and in the previous problem are rather different. Indeed,the function given
by Eq. (*)also approaches the complex permittivity (7.32) of the Lorentz oscillator model but at very
high rather than very low frequencies.

Problem 7.4. Use the oscillator model of an atom, given by Eq. (7.30) of the lecture notes, to
calculate its average potential energy in a uniform, sinusoidal ac electric field, and use the result to
calculate the potential profile created for the atom by a standing electromagnetic wave with the electric
field amplitude E(r).
Solution: For any isotropic linear model of a charged particle’s motion, its oscillations induced
by a sinusoidal electric field E(t) = Ex(t)nx with


E x (t )  Re E e it   12 E e it  c.c.,


are also sinusoidal and directed along the field: r(t) = x(t)nx, with


x(t )  Re x e it   12 x e it  c.c.


This motion creates the oscillating dipole moment p(t) = qr(t), also directed along the applied
field, and hence the potential energy of its interaction with the applied field may be calculated using Eq.
(3.15b) of the lecture notes:7
1
2
q
8
  q
 
U (t )   p(t )  E(t )   E e it  c.c. x e it  c.c.   E x e 2it  E x*  c.c. .
8

Time averaging of the energy over the period of the applied field kills the terms proportional to
exp{2it}, giving
q
 q
U   E x*  c.c.   Re E x* .
8 4
  
In particular, for the oscillator model (7.30), the complex amplitude x is given by Eq. (7.31):
q E ,
x 
 
m  0   2  2i 0
2

so the average energy is

q2  1  q2  2   02
U  E E* Re  2  E E *
. (*)
4m   0 
  2
 2i 0

 4 m
 
 2
  0 
2 2
 2 
 0  2

This expression shows that the potential energy is negative at  < 0 and positive in the opposite
case. In a more general model (7.33) corresponding to several oscillators with different resonance

7 Note that a non-zero net charge of the atom (if present) also gives another, polarization-independent contribution
(1.31), U0(r, t) = q(r, t), to its potential energy in the applied field, but for a sinusoidal field, the electric potential
 is also a sinusoidal function of time, so the time average of this part of the interaction vanishes.

Problems with Solutions Page 254

Essential Graduate Physics EM: Classical Electrodynamics

frequencies j, such behavior repeats near each of them – see, e.g., the red line in Fig. 7.5 of the lecture
notes.8
Proceeding to the second part of the problem, we should take into account that according to Eqs.
(5.10) and (7.6)-(7.8), the interaction of non-relativistic charged particles, moving with velocity v << c,
with any electromagnetic wave is dominated by the wave’s electric field. In addition, the linear size of
typical atoms and molecules is of the order of 10-10 m and hence much smaller than the wavelength ,
all the way up to extremely high frequencies corresponding to X-rays. Hence, for a standing wave, we
may use Eq. (*), derived for a uniform electric field, with E replaced with E(r), where r is the atom’s
position. This expression shows that in a standing wave with  < 0, the particle is repulsed from
electric field maxima and hence is pushed to an adjacent field’s minimum (“node”), while in the
opposite case ( > 0) it is pushed to the adjacent maximum of the standing wave.
As Eq. (*) shows, the effect is especially large if the oscillator’s damping is low ( << 0), as
typical for atoms in gases, and the field’s frequency is close to one of the atom’s eigenfrequencies 0,
but still out of the narrow interval 0  0. Note, however, that the effect exists even for free particles,
which may be described by the Lorentz oscillator model with 0 = 0 = 0 when Eq. (*) is reduced to
q2
U E E* , (**)
4m 2

showing that the particle is pushed toward the electric field’s minimum.9
On the other hand, in the limit , 0 << 0, in which Eq. (7.35) is valid, Eq. (*) yields
q2  0   0
U  E E*   E E*  0, (***)
4m 0
2
n
so the atom is attracted to the regions where the field has its maximum. (In the last expression, (0) is the
dc electric susceptibility of a dilute medium with n particles per unit volume; in this form, Eq. (***)
also holds for atomic clusters and small but macroscopic dielectric particles.) This effect is widely used
in experimental practice for optical trapping of the particles, for example, small biological samples at
the focus of paraxial (e.g., Gaussian) laser beams – the so-called optical tweezers.10

Problem 7.5. The solution of the previous problem shows that a standing electromagnetic wave
may exert a time-averaged force on an otherwise free non-relativistic charged particle. Reveal the
physics of this force by writing and solving the equations of motion of such a particle in:
(i) a linearly-polarized monochromatic plane traveling wave, and
(ii) a similar but standing wave.
Solutions:

8 Just for the reader’s reference, quantum mechanics confirms the same behavior, near each quantum transition
frequency, for a charged particle confined in a potential well of arbitrary shape – not only of the quadratic shape
corresponding to a harmonic oscillator – see, e.g., QM Problem 6.18.
9 Exploring the physics of this force is the subject of the next problem.
10 See, e.g., P. Jones et al., Optical Tweezers: Principles and Applications, Cambridge U. Press, 2016.

Problems with Solutions Page 255

Essential Graduate Physics EM: Classical Electrodynamics

(i) Directing the z-axis along the wave’s propagation direction, and the x-axis along its electric
field, we may use Eqs. (7.6)-(7.8) and (7.11) of the lecture notes to express the wave’s fields as

 
Er, t   n x Re E e i kz t  , Br, t    0 H r, t    0
nz nx
Z0

Re E e i kz t  
ny
c
 
Re E e i kz t  , 
so the Lorentz force (5.10) exerted by the wave on a particle with charge q, located at r = {x, y, z}, is


Fr, t   qE  v  B  qn x Re E e i kz t   q  vny
c
 
Re E e i kz t  . (*)

The electric force, expressed by the first term on the right-hand side, clearly has no time average,
while the second term does. In the non-relativistic limit, when v/c << 1, the magnetic force expressed by
the second term is much smaller than the electric one, so we may first calculate v from the 2nd Newton
law in which the magnetic force is ignored:11


mv  qn x Re E e i kz t  . 
Integrating both parts of this simple equation, we get

v  nx
q
m

Re iE e i kz t  , 
where we have neglected the possible free motion of the particle by inertia, which is unrelated to the
wave’s effect. (In any realistic system, such ballistic motion eventually stops because of the energy loss
due to some inevitable interaction of the particle with its environment.)
Now plugging this result into the second term of Eq. (*), we may calculate the magnetic force:

Fm r, t  
q2
mc
  
n x  n y Re iE e i kz t  Re E e i kz t  

q2
mc 2
1

n z iE e i kz t   c.c.
1
2
 
E e i kz t   c.c. . 
The multiplication of the square brackets gives four terms, but only two of them (both proportional to
EE*) have their time dependences canceled, thus potentially contributing to the average force:

F r   Fm r  
q2
4mc

n z E E* ie ikz e ikz  c.c 
q2

4mc
n z E E* i  c.c.  0 .

Hence, in this case, the full Lorentz force has no average. This result is in accordance with the
solution of the previous problem, in which the average interaction energy is constant (and hence has no
gradient) if EE* is constant – as it is in a traveling plane wave.
(ii) The situation changes if the wave is a standing one. Taking its fields, for example, in the
form given by Eqs. (7.61)-(7.62) of the lecture notes, with the same linear polarization as above,

11It is straightforward (and hence left for the reader as an additional exercise) to use the last term of Eq. (*) to
show that the magnetic force leads to much smaller z-oscillations of the particle with frequency 2.

Problems with Solutions Page 256

Essential Graduate Physics EM: Classical Electrodynamics

 
Er, t   2n x Re iE e it sin kz, Br, t    0 Hr, t   2
c
ny
 
Re E e it cos kz ,

we see that the time-averaged electric component of the Lorentz force equals zero again. Using the same
approach as in the previous task to find the magnetic component of the force, we get

 
mv  2qn x Re iE e it sin kz, giving v  n x
2q
m

Re E e it sin kz , 
so
Fm r, t   qv  B  
4q 2
m c
  
n x  n y Re E e it sin kz Re E e it cos kz 

4q 2 1

n z E e it  c.c.
m c 2
1
2
  
E e it  c.c sin kz cos kz.

Here again, only two of the four terms resulting from the parentheses multiplication have their time
dependences canceled and hence contribute to the average force, but now they add up rather than
subtract:
F r   
q2
mc
 
n z E E*  E* E sin kz cos kz  
q2
m c
n z E E* sin 2kz . (**)

This force vanishes only at all points where 2kz = n (with n = 0, –1, –2, …, because our result is valid
only for z  0 – see Fig. 7.8), but only at such points with even n, in which the electric field vanishes, it
leads to a stable equilibrium.12
Let us compare this result with that following from Eq. (**) of the model solution of the
previous problem:
q2
U E E* .
4 m 2
Now we should recall that to obtain this result, we took E in the form nxRe(E e–it), so to comply with
our current assumptions, we have to make the following replacement: E  2iE sinkz, getting
q2 q2
U  U r   2iE sin kz 2iE sin kz   2 E E* sin 2 kz .
*
4m 2
m
This average potential energy, depending only on z, yields an average force with just one Cartesian
component:
F r   U r   n z
q2
m 2
E E
 
* d
dz
sin2
kz  n z
q2
mc
E E* sin 2kz,

i.e. exactly the same result as Eq. (**) of this solution.

Hence, we may conclude that the average force exerted on a charged particle by a wave is
essentially the average magnetic component of the Lorentz force, but it would not arise without the
particle’s motion induced by the (much larger) electric component of the force.

12The last statement may be readily verified by exploring the equation of motion along axis z – an additional
useful exercise for the reader.

Problems with Solutions Page 257

Essential Graduate Physics EM: Classical Electrodynamics


Problem 7.6. Use the first of Eqs. (7.54) of the lecture notes to relate the integral  "  d
0
to the plasma frequency for the Lorentz oscillator model of a system of non-interacting particles.
Solution: As Eq. (7.33) shows, in the Lorentz oscillator model with low damping (j << j), the
energy loss function ”() is nonvanishing only in very small vicinities of the resonant frequencies j.
Hence, if the wave frequency  is well above them all, the first of the dispersion relations (7.54) is
reduced to

2
' ( )   0  2  " (Ω)d, , for    j .
 0

But as was noted in Sec. 7.2 of the lecture notes, the condition  >> j reduces the same Eq. (7.33), with
the sum rule
 f j  1, j
(*)

to Eq. (7.36):
  p2   p2
 
 ( )  ' ( )   0 1  2   0   0 2 .
   
 
Comparing these two expressions for ’(), we see that they coincide if


 " d  2  
2
0 p
.
0

Since this result is conditioned by Eq. (*), it is also sometimes called the sum rule. Quantum
mechanics shows13 that its validity extends well beyond the classical Lorentz model.

Problem 7.7. Prove that Eq. (6.42) of the lecture notes cannot be correct for all frequencies, and
suggest its correction making the result compatible with both the causality principle and the physical
model (6.39).
Solution: Let us assume that Eq. (6.41),14
q 2n ,
j    E , with     i (*)
m
is valid for all frequencies, and apply it to the case then the electric field is an ultimately short pulse
applied at t = 0:
E t   c t ,

where c is a constant. Then, by expanding this expression into a Fourier series,


E t    E e
it
d ,


13See, e.g., QM Chapter 6.

14Since here we are not discussing magnetic field effects, we may ignore the vector character of j and E – say,
by considering the Cartesian components of these vectors in the field’s direction.

Problems with Solutions Page 258

Essential Graduate Physics EM: Classical Electrodynamics

using the reciprocal transform to calculate E,

 
1 it c it c
E 
2  E t e dt 

2 
  t e dt 
2
,

plugging the result into Eq. (*),

q 2n q 2n c
j  i E  i ,
m m 2
and calculating the current density as a function of time,15 we get

cq 2 n e it d
   
it cq 2 n cos t d cq 2 n sin t d
j t    j e 2m   2m  2m 
d  i  i 

 
(**)
sin  d cq 2 n   , for t  0,
 
cq 2 n sin t d cq 2 n
 
2m     2m    , for t  0.
 0  sgn t
 2m  
So, Eq. (*) predicts a nonvanishing current to flow before the electric field’s pulse has been
applied – an evident nonsense from the causality point of view, in any physical situation when the
electric current is caused by the electric field.
The Kramers-Kronig dispersion relations offer an easy and natural fix for the situation. Indeed, if
we take the complex function f() in Eq. (7.52) equal to () rather than () – 0, then instead of the
second of Eqs. (7.53), we get

1 d
"     P  '   .
   
Evidently, if we take
q2n
'       , (***)
m
we immediately get
q2n
"    ,
m
i.e. the same imaginary part of the complex conductivity as in Eq. (*). However, its additional real part
(***) gives the following additional contribution j to the current:
q2n c q2n
j  '  E      c   ,
m 2 2m
 
 i t q 2n i  t q 2n
j t     j  e d  c  
2m 
  e d   c ,

2m
so the full current becomes
cq 2 n 1, for t  0,
j  t   j t   j t   
m 0, for t  0,

15 The last step of this calculation uses MA Eq. (6.12).

Problems with Solutions Page 259

Essential Graduate Physics EM: Classical Electrodynamics

thus fixing the causality problem without changing the imaginary part of (), and hence the free-
particle model (6.39) behind it. Indeed, Eq. (***) has a clear physical sense, describing the infinite final
velocity of the particles, and hence the infinite current they carry, for a zero-frequency (dc) electric field
applied to a free particle described by this simple model.

Problem 7.8. Calculate, sketch, and discuss the dispersion relation for electromagnetic waves
propagating in a medium described by the Lorentz oscillator model (7.32), for the case of negligible
damping.
Solution: The given assumption, 0 = 0, reduces Eq. (7.32) to a simpler form:

q2 1   p2 
 ( )   0  n   
0 1  , (*)
m ( 02   2 ) 
 0  
2 2

where 0 is the resonance frequency of the oscillators, and p is the plasma frequency defined by Eq.
(7.36) of the lecture notes. (For similar oscillators, that expression is reduced to Eq. (7.37) with e  q:
nq 2
 p2  ;
 0m
note that the parameters p and 0 are completely independent because p is proportional to the
oscillator density n, while 0 is a property of a single oscillator.) By plugging Eq. (*) into the general
dispersion relation (7.28) for a uniform linear medium, assuming that our medium is not magnetically
active: ()  0, and recalling that 00  1/c2, we get
1/ 2 2
   p2
k    1 2  .
c   0   
2
 p /  0  0 .3
1.5
The figure on the right shows a typical
  ck
1/ 2
plot of this function (for a particular ratio p/0).    p2 
1  
It reminds the standard level-anticrossing  0 1  2 
 0 
diagram16 but coincides with it only in the limit   ck
p << 0, i.e. for very low density of the 0.5
oscillators. For a non-zero ratio p/0, there is a
substantial frequency gap,17

 0     0' , with  0'   02   p2  ,

0
1/ 2 0 0.5 1 1.5 2
k /( 0 / c )
in which () < 0, and hence the electromagnetic
wave propagation is impossible. At frequencies far from (both below and above) the gap, the dispersion
relation is close to that of waves in free space, k  /c, testifying that here the coupling between the
wave and the medium oscillators is weak. However, at both approaches to the forbidden frequency gap,
the dispersion relation plots strongly bend, indicating substantial energy sharing between the wave and
the oscillators. In quantum mechanics, the elementary excitations of such a strongly coupled field are

16 See, e.g., CM Sec. 6.1 and/or QM Secs. 2.7 and 5.1.

17 Note that according to Eq. (*), the ratio 0’/0 has a very simple physical sense: (0’/0)2 = (0)/0  (0)/().

Problems with Solutions Page 260

Essential Graduate Physics EM: Classical Electrodynamics

called polaritons, because of the electric-polarization origin of Eq. (7.32). (Note that the term
“polaritons” is currently used in a much broader context than the Lorentz oscillator model – for virtually
any excitations resulting from the interaction of traveling electromagnetic waves with a set of resonant
dipoles.)

Problem 7.9. As was briefly discussed in Sec. 7.2 of the lecture notes,18 a wave pulse of a finite
but relatively large spatial extension z >>   2/k may be formed as a wave packet – a sum of
sinusoidal waves with wave vectors k within a relatively narrow interval. Consider an electromagnetic
plane wave packet of this type, with the electric field distribution

E(r, t )  Re  E k e

i kz   k t dk , with k   k   k   k  ,
1/ 2



propagating along the z-axis in an isotropic, linear, and dissipation-free (but not necessarily dispersion-
free) medium. Express the full energy of the packet (per unit area of the wave’s front) via the complex
amplitudes Ek, and discuss its dependence on time.
Solution: For an isotropic linear medium we may use the first of Eqs. (6.113):19
ED HB
 u
,
2 2
so the total energy of the pulse (per unit area of the wave’s front) may be calculated as
  
U 1 1
  udz   E  Ddz   H  Bdz . (*)
A  2  2 
Just as it was done in Sec. 7.2 of the lecture notes for the derivation of Eq. (7.42), let us use the Fourier
expansions of all these fields over real k, at arbitrary time t:
1 
 
i k i
E(r, t )  Re  E k e dk    E k e k dk  c.c., with the total phases  k  kz   k t ,

2  
1  1   
 
i k ' i k ' i
D(r, t )  Re  D k ' e dk'    D k ' e dk'  c.c.      k'  E k' e k ' dk'  c.c.,

2    2  
and absolutely similarly for H and B. Plugging these expressions into the first integral of Eq. (*), and
changing the integration order, we get
 i  k   k'  i  k   k'  
    
1
 E  Ddz    *
  *
4     
dk dk'   k' E k  E k ' e dz    k' E k  E k ' e dz  c.c. . (**)
   
By using MA Eq. (14.4), we may readily work out both internal integrals, which do not depend on Ek:

18 For even more detail, see CM Sec. 5.3 and especially QM Sec. 2.2.
19 Note that the second of Eqs. (6.13) is valid only for dispersion-free media.

Problems with Solutions Page 261

Essential Graduate Physics EM: Classical Electrodynamics



i  k   k' dz  exp i   k' t  e


i k  k' z dz
e

k

exp 2i k t k  k' , for the upper sign,
 2 exp i k   k' t k  k'   2  
  k  k' , for the lower sign,
so the integral over k’ in Eq. (**) is also elementary, and that equality is reduced to

 E  Ddz  2    E 
 

k k  E  k exp 2i k t   *  k E k E k*  c.c. dk ,
 

because due to our definition of the partial wave frequencies, –k = k, and hence (–k) = (k).
According to the discussion in Sec. 7.2 of the lecture notes, in a lossless medium, k is real, and (k) is
also a real, even function of frequency – and hence of k, so *(k) = (k). As a result, by spelling out
the complex conjugate term, we get

    E 
 

 E  Ddz  k k  E  k exp 2i k t  E*k  E* k exp 2i k t  2 E k E k* dk .

2 

Up to this point, the calculations are valid for any linear superposition of sinusoidal waves. If the
packet is narrow as mentioned in the assignment, the first two terms in the square brackets are rapidly
oscillating functions of time, with nearly zero averages, for all essential values of k. Neglecting these
terms, we get
 

 E  D dz      k E k E k dk .
*
 

Carrying out an absolutely similar calculation for the magnetic energy, we get
 

 H  B dz      H
 
k k H k*dk .

Now taking into account that Eqs. (7.6) and (7.7), generalized in accordance with Eq. (7.28):
   k  
1/ 2
n  Ek
Hk  z sgn( k ), with Z  k     ,
Z  k     k  
we get
E k E k*
  k H k H k*    k     k E k E k* ,
Z  k  2

so both contributions to the energy are equal and Eq. (*) yields

U
 2    k E k E k* dk .
A 0

This expression shows that the packet’s energy does not depend on time, even though its shape
may be deformed (most typically, spreads out with time) very significantly by the frequency dispersion

Problems with Solutions Page 262

Essential Graduate Physics EM: Classical Electrodynamics

of the medium.20 This result certainly fits our intuitive notion of the wave packet’s motion in the
absence of attenuation.

Problem 7.10. Prove the Lorentz reciprocity relation (6.121) for a linear isotropic medium.
Solution: As was stated in Sec. 6.8 of the lecture notes, this formula relates the complex
amplitudes (rather than the instantaneous values!) of two separate monochromatic electromagnetic fields
of the same frequency , which are induced in a linear medium by stand-alone currents with complex
amplitudes, respectively, j1(r) and j2(r). As was discussed in Sec. 7.1, in such media, the complex
amplitudes of the variables obey all the relations pertinent to the instantaneous values of the fields, with
the replacement21

 i.
t
In particular, the Maxwell equations (6.99a) take, for such amplitudes, the following form:
  E  iB  0,   H  iD  j.
Let us scalar-multiply all terms of the first of these equations, written for the fields labeled 1, by H2, and
those of the second equation, by E2:
H 2    E1  iH 2  B1  0, E 2    H 1  iE 2  D1  E 2  j1 .
Flipping the field indices and the signs of all terms, we get
 H 1    E 2  iH 1  B 2  0,  E1    H 2  iE1  D 2  E1  j 2 .
Now let us add up each side of all these four equations. Grouping similar terms, we get

H 2    E1  E1    H 2   E 2    H1  H1    E 2 
(*)
 i  H 2  B1  H 1  B 2   i E 2  D1  E1  D 2   E 2  j1  E1  j 2 .
But for an isotropic linear medium, the complex amplitudes of each field are related, at each
spatial point, as B1,2 = ()H1,2 and D1,2 = ()E1,2, so the last two parentheses on the left-hand side of
this equation vanish – even for nonuniform systems where () and () depend on coordinates.22 Next,
according to the vector calculus,23 each of the first two parentheses on the left-hand side of Eq. (*) is
just the divergence of the vector product of the corresponding fields, so the equation reduces to

20 Again, see QM Sec. 2.2 for a detailed discussion and examples of this effect.
21 Here, as everywhere in my series, I am following the “physical” convention, representing a single complex
component of a sinusoidal function of time as fexp{–it}, with the negative sign in the exponent. The transfer to
the “engineering” convention, with the positive sign in the exponent, is obviously equivalent to the change of the
sign before . Since Eq. (6.121) that we want to prove does not include  as all, the choice of the convention
makes no difference for it.
22 Moreover, even the reader whose tensor calculus is somewhat rusty can easily see why these cancellations (and

hence the final result of this calculation) are valid even for an anisotropic medium with symmetric tensors jj’()
and jj’(). In this case, for example, E2D1  j,j’jj’()(E2)j(E1)j’ =j,j’j’j()(E2)j(E1)j’  E1D2.
23 See, e,g., MA Eq. (11.7), read backward.

Problems with Solutions Page 263

Essential Graduate Physics EM: Classical Electrodynamics

  E1  H 2  E 2    H 1   E 2  j1  E1  j 2 .

By integrating this expression over the volume V inside an arbitrary closed surface S and using the
divergence theorem,24 we get

 E
S
1  H 2  E 2  H 1 n d 2 r   E 2  j1  E1  j 2  d 3 r ,
V
(**)

i.e. the Lorentz reciprocity relation (6.121). As will be shown in the next chapter, this relation may be
further simplified if both field sources are space-localized.
Let me leave it to the reader to prove that in the quasistatic case, Eq. (**) may be used to derive
the Rayleigh-Lorentz-Carson reciprocity relation, whose alternative proof was the subject of Problem
4.3.

Problem 7.11.* A plane wave of frequency  is normally incident, from free space, on a plane
surface of a collision-free plasma with the electron density growing slowly and linearly with the
distance from the surface: n = z for z  0, where  > 0 is a small constant. Calculate the functional form
of the resulting standing wave’s “tail” inside the plasma.
Solution: Reviewing the derivation of the first of Eqs. (7.3) of the lecture notes at the beginning
of Sec. 7.1, we see that if the medium’s properties are spatially dependent, these equations acquire
additional terms, which couple them:
 2 2  H  2 2  E
    2  E     ,     2  H     . (*)
 t  t  t  t
However, if the parameters  and  depend only on the distance from the surface, and their changes on
one wavelength of the normally incident radiation are relatively small, the right-side terms of these
equations only produce effects of the second order in small  and ,25 and may be neglected in
comparison with the much more dramatic effects provided by the spatial dependence of the product 
on their left-hand sides.
Hence, if the plasma is magnetically inactive and its density is a slow function of z, we may still
describe the propagation of a normally incident wave in it by the usual 1D wave equation
 2 2 
 2   2  E  0 ,
 z t 
where  is a slow function of z, while  = 0 = const. According to Eqs. (7.36)-(7.37) and their
discussion, in a collision-free plasma we may take
  p2   2

     0 1  2    0 1  ne .
     m 2 
   0 e 
For our current problem where

24If you still do not remember it, see, e.g., MA Eq. (12.2).
25For example, at   0 and  = 0, the first-order correction due to the right-hand side of the second of Eqs.
(*) results only in a proportionally small change of H, which has no way to sneak back into the equation for E.

Problems with Solutions Page 264

Essential Graduate Physics EM: Classical Electrodynamics

 0, for z  0,
n
z , for 0  z ,
this gives the following ordinary differential equation for the complex amplitude of a plane
monochromatic wave propagating along the z-axis,
1, for z  0,
 d2  2 
 2  k  z  E   0 , where k  z   2  1  e
2
2 2
(**)
 dz  c z , for 0  z.
  m 2
 0 e

As was discussed in Sec. 7.3 of the lecture notes, its solution for negative z is the usual standing
wave of a constant amplitude
E  2 E 0 sin k 0  z     ,

where E0 is the amplitude of the incident wave, k0  (00)1/2  /c is the wave number in free space,
and – is a phase shift describing the wave’s penetration into a surface layer of the plasma – see, e.g.,
Fig. 7.4 of the lecture notes. The solution of Eq. (**) for z > 0 and an arbitrary positive  is proportional
to one of the so-called Airy functions, namely Ai(), where in our current case,
1/ 3
 e 2   0 me  2 .
    z    , with   
  0 c me e 2
2

(As Eq. (**) shows, + has the simple physical meaning of the z-point where the propagation constant
k2(z) changes its sign.) However, since in this series, the Airy functions are discussed, for the first time,
only in the QM course (Sec. 2.4), and our task is limited to sufficiently small , the problem may be
solved without using them.
For that, let us notice that if z > +, then k2(z) is negative, so k(z) = i(z), with
e 2
 2 z   z      0 . (***)
 0 me c 2
As we know from Eqs. (7.64) and (7.68), if k2 < 0 of a medium is constant, the wave inside it decreases
as exp{–z}. Hence, we may expect that if  does depend on z but slowly enough, then at (z)(z – +)
>> 1, z z
df
E  z   exp f  z , with    z     z , i.e. f  z      z'  dz '     z' dz ',
dz
where the correction (z) would be relatively small:  (z)  << (z). Indeed, plugging the assumed
solution into Eq. (**), we see that it is indeed satisfied if, in the first approximation in (z),
d 1 d
2 
dz
 0, i.e.  
d
 ln  1 / 2 ,
2 dz dz
 
so our solution becomes26
 z   z 
E  z   exp f  z   exp     z' dz '  ln  1 / 2    1
exp     z' dz '.
   1/ 2  

26This is the particular case (for k2 < 0) of the famous WKB approximation. This method is one of the main
theoretical tools of quantum mechanics and hence will be discussed, in much more detail, in QM Sec. 2.4.

Problems with Solutions Page 265

Essential Graduate Physics EM: Classical Electrodynamics

In our particular case (***), this solution reads

 2  e 2 
1/ 2

E  z  
1
exp    z    3 / 2   0, at  z      .
z    1 / 4  3   0 me c
2
 
This is indeed the correct asymptotic behavior of the Airy function Ai(). Note that this solution
depends on the wave’s frequency only via the position +  2 of the effective reflection boundary.

Problem 7.12.* Analyze the effect of a time-independent uniform magnetic field B0, parallel to
the direction n of an electromagnetic wave propagation, on the wave’s dispersion in plasma, within the
same simple model that was used in Sec. 7.2 of the lecture notes for the derivation of Eq. (7.38). (Limit
your analysis to relatively weak waves, whose magnetic field is much smaller than B0.)
Hint: You may like to represent the incident wave as a linear superposition of two circularly
polarized waves, with opposite polarization directions.
Solution: For clarity, let us first repeat the derivation of Eq. (7.38), valid for B0 = 0, for a plasma
of free noninteracting particles, now spelling out the spatial orientations of the field and of the field-
induced particle motion. For a free particle (0 = 0, 0 = 0), Eq. (7.30) may be rewritten as
mr  qEt  , (*)
so for a monochromatic, linearly polarized wave with E(t) = nxRe[E exp{–it}] (where nx is the
polarization direction), we may look for the solution in a similar form, r(t) = nxRe[r exp{–it}].
Plugging these expressions into Eq. (*), we get the following equation for the complex amplitude of the
particle’s forced oscillations:
q
 m 2 r  qE , giving r   E .
m 2
Now calculating the complex amplitude,
nq 2
P  np  nqr   E ,
m 2
of the resulting electric polarization of the plasma, P(t) = nxRe[P exp{–it}], we may use Eq. (3.33) to
calculate its frequency-dependent permittivity,
D P   p2 
     0    0 1  2 ,
E E   
 
with the plasma frequency p expressed similarly to Eq. (7.37):
nq 2
 p2  ,
 0m
i.e. reproduce Eq. (7.36) of the lecture notes. As was discussed in Sec. 7.2 of the lecture notes, this
permittivity immediately leads to the plasma dispersion law (7.38),

Problems with Solutions Page 266

Essential Graduate Physics EM: Classical Electrodynamics

1/ 2
   p2 
k        0  1 2 
1/ 2
 ,
c   

which, in particular, forbids the propagation of waves with frequencies  < p in the plasma.
In the case of nonvanishing dc field B0, we have to generalize Eq. (*), taking into account the
full Lorentz force (5.10) acting on the particle:
mr  q  Et   r  B 0  . (**)

This equation is still linear in r(t), and hence the plasma’s response still may be characterized by a
(frequency-dependent) permittivity. However, Eq. (**) shows that if the field B0 is directed along the
wave’s propagation (say, the z-axis), i.e. B0 = nz B0, any motion in the direction x of the wave’s electric
field causes a magnetic force in the y-direction, and vice versa, so the electric polarization acquires not
only the x-, but also the y-component. This means that the plasma’s response becomes anisotropic, and
its permittivity for linearly polarized waves has to be described with a 22 tensor rather than a scalar.
However, the calculation may be simplified using the provided Hint, i.e. representing the
incident wave, with an arbitrary polarization, as a sum of two circularly polarized ones – see the
discussion at the end of Sec. 7.1 of the lecture notes. For example, a wave linearly polarized in the x-
direction may be represented as
 E E 
Et   n x ReE exp it   Ren x  exp it  in y  exp it
  2 2  (***)

  Re n x  in y  E exp it ,
 
with E E
2

.


Indeed, looking for the solution of Eq. (**) in a similar form, i.e. as a sum of two simultaneous circular
motions but with not necessarily equal complex amplitudes:


rt    Re n x  in y r exp it , 

we see that since
n x  in y  B 0  n x  in y  n z B0  n x  n z  in y  n z B0   n y  in x B0  iB0 n x  in y  ,

Eq. (**) naturally separates into two independent and different relations for their complex amplitudes:


 m 2 r  q E  ir  iB0  , 
giving
q q
r   E   E ,
m  qB
2
m    c 
where c  –qB0/m is the well-known cyclotron frequency of a particle’s motion in the magnetic field.27
This result is very natural: it shows that the longitudinal magnetic field speeds up the particle’s rotation

27 For a detailed discussion of this motion (in both the non-relativistic and relativistic cases), see Sec. 9.6 of the
lecture notes.

Problems with Solutions Page 267

Essential Graduate Physics EM: Classical Electrodynamics

in one transverse direction and slows its rotation in the opposite direction (induced by the wave’s
electric field) but does not entangle these two circular motion modes.
Now calculating the resulting electric permittivity of the plasma exactly as was done above, we
see that for each circularly polarized wave, it may be still characterized by a scalar function of
frequency, but for the waves with opposite polarizations, these functions are different:
P   p2 
     0     0 1 

,
E      c  
and so are their dispersion relations:
1/ 2
  p2 

k        0
 
 1/ 2
 1  
c      c 
. (****)

The last relation shows, in particular, that the magnetic field increases the cut-off frequency min
for wave propagation, at which the wave vector vanishes, for one circular polarization (depending on the
sign of the particle’s charge) and decreases it for the counterpart polarization – though never fully
suppresses this threshold:
 2  c2 
1/ 2
c   c ,
 
   p     for  c /  p   .
4 
min
 2  p /  c ,
2

Moreover, even for the frequencies above both values of min, the phase and group velocity for
each of the two modes is different, so an incident wave of any polarization (even a linear one), upon
entering the plasma, splits into two independently propagating circularly polarized waves. (For example,
for the Earth ionosphere in the planet’s own magnetic field, p is typically below 108 s-1, while  c  
107 s-1, so the wave splitting effects may be quite noticeable.) In the case of small splitting, when
 c p2

k  k  k  
 k  , for  c   ,  p ,
c 2

its effect may be conveniently represented as a slow rotation of the linear polarization plane by an angle
  z. Indeed, after propagation by distance z, we may write kz = kz  , where k is the wave vector in
the absence of the magnetic field and   (k) z/2, so the initially x-polarized wave (***) becomes

   
E
2
 
E z, t    Re n x  in y   exp i k  z  t    Re n x  in y   e i exp ikz  t  .
E
2
     
Representing this wave as an explicit sum of two Cartesian components,
E
    
E z, t   Re  n x e i  e i  in y e i  e i exp i kz  t , 
 2 
we see that the ratio of their complex amplitudes is
Ey


i e i  e  i   tan  ,
Ex e i  e i

Problems with Solutions Page 268

Essential Graduate Physics EM: Classical Electrodynamics

i.e. remains real. According to Eq. (7.17) of the lecture notes, this means that the wave remains linearly
polarized at each point, but the polarization plane slowly rotates by the angle  proportional both to the
traveled distance z and to k  c, i.e. to the applied magnetic field. This is the Faraday effect (or
“Faraday rotation”) – one of several magneto-optical phenomena.
It is also an example of the so-called non-reciprocal effects, which violate the Lorentz
reciprocity relation (6.121). Indeed, as was discussed at the end of Sec. 6.8 of the lecture notes, that
relation predicts, in particular, that the electromagnetic waves propagate through a medium similarly in
both directions. In the case of the Faraday rotation, however, at the wave’s direction reversal (say, at its
reflection from a mirror), the direction of the magnetic field “as seen by the wave” changes. As a result,
the sign of c in Eq. (****) changes, so the direction of the rotation of the wave’s quasi-linear
polarization reverses from the wave’s “point of view”, and hence remains the same as observed from the
lab frame. In other words, in the lab frame, the rotation goes in the same direction regardless of the
wave’s direction, accumulating at the wave’s roundtrip, so with the appropriate choice of the plasma
layer’s thickness, the returning wave may have the alternative (y-) linear polarization, and may be
readily separated from the incident one by using a polarization filter.
This effect is used for the implementation of very important non-reciprocal
microwave and optical devices – circulators, operating as shown in the figure on
the right. (Typically they use dc-magnetic-field-biased ferromagnetic materials
rather than gaseous plasmas.) The circulators allow, for example, to stabilize the
operation of reflection amplifiers based on the effective negative resistance of
nonlinear active electron devices.

Problem 7.13 A monochromatic plane electromagnetic wave is normally incident, from free
space, on a uniform slab with electric permittivity  and magnetic permeability , with the slab’s
thickness d comparable with the wavelength.
(i) Calculate the power transmission coefficient T, i.e. the fraction of the incident wave’s power,
that is transmitted through the slab.
(ii) Assuming that  and  are frequency-independent and positive, analyze in detail the
frequency dependence of T. In particular, how does the function T () depend on the slab’s thickness d
and the wave impedance Z  (/)1/2 of its material?
Solutions:
 0 , 0 ,   0 , 0
(i) Since the wave may be partly reflected not only from the
front surface but also from the back surface of the slab (see the 1 T T0
figure on the right), we have to look for the solution of the 1D
wave equation in the form of five traveling waves: 0 d z
e ik0 z ik0 z R0
 R0 e , for z  0, R


E  E e it   Te ikz  R e ikz , for 0  z  d ,
 ik z
 T0 e 0 , for d  z.


Problems with Solutions Page 269

Essential Graduate Physics EM: Classical Electrodynamics

The four (generally, complex) coefficients R, R0, T, and T0 participating in these expressions may
be found by plugging them into the four boundary conditions that express the continuity of the electric
field E and the magnetic field H = E/Z at both surfaces (z = 0 and z = d). These conditions give us the
following system of four linear equations:
1  R0 T  R
1  R0  T  R ,  ,
Z0 Z
ik d
ik0d Te ikd  Re ikd T0 e 0
Te ikd  Re ikd  T0 e ,  .
Z Z0
Solving this system, we get, in particular,
4 ZZ 0
T0  expi k  k 0 d , (*)
( Z  Z 0 )  ( Z  Z 0 ) 2 exp{2ikd }
2

and since, for our system, the wave impedances in the regions in
front of the slab and behind it are equal, the power transmission Im
coefficient may be now calculated just as T =  T0 2.
(ii) The simplest way to analyze the dependence of T0 and denominator
T on the system’s parameters (at constant and positive  and , ( Z  Z 0 ) 2 2kd
and hence constant and positive Z and Z0), is to consider the
diagram in the figure on the right, which shows the denominator 0 (Z  Z 0 ) 2 Re
in Eq. (*), a complex-variable function, as the difference of two
2D vectors, of lengths (Z + Z0)2 and (Z – Z0)2 < (Z + Z0)2,
respectively, with the angle 2kd between them.
It is clear, first of all, that the denominator,
and hence T0 and T, oscillate as functions of both the 1.3
wave frequency  and the slab’s thickness d, with the
period (kd) =  (see the figure on the right), with the 0.8
0.7
transmitted power having equal maxima at28
T
kd  m, m  1, 2, ... . 0.6 0.5
This is the well-known constructive interference
0.4
condition, fulfilled when the slab’s thickness d equals
an integer number of /k, i.e. of the de Broglie half- Z / Z0  3
wavelengths /2.29 0.2

The fact that would be harder to predict is that

in each maximum, the transmission is perfect: 0
0 0.5 1 1.5 2
kd / 

28 Formally, the value m = 0 should be in this list as well, but it corresponds to physically trivial cases of either d
= 0 (no slab at all), or k = 0, i.e.  = 0 (no wave at all).
29 A similar effect in quantum mechanics is called resonant tunneling – see, e.g., QM Sec. 2.5.

Problems with Solutions Page 270

Essential Graduate Physics EM: Classical Electrodynamics

4 ZZ 0
T0  T0 kd m
  1,
max
(Z  Z 0 ) 2  (Z  Z 0 ) 2
regardless of the Z/Z0 ratio. This ratio, however, does affect the sharpness of these resonances and the
depth of the transmission minima: the farther Z/Z0 from 1, the lower the transmission minimum at such
destructive interference,
4ZZ 0 2( Z / Z 0 ) 4( Z / Z 0 ) 2
T0  T0   , so T min   1.
min kd  ( m 1 / 2 )
(Z  Z 0 )  (Z  Z 0 )
2 2
(Z / Z 0 ) 2  1 (Z / Z 0 ) 1
2

2

Problem 7.14. A plane electromagnetic wave with a free-space wave number k0 is normally
incident on a planar conducting film of thickness d ~ s << 1/k0. Calculate the power transmission
coefficient of the system and analyze the result in the limits of small and large values of the ratio d/s.
Solution: This problem may be solved by re-deriving (or just using:-) the result of the previous
problem, which may be rewritten as
4ZZ 0
exp ik 0 d  .
2
T  T0 , where T0  (*)
( Z  Z 0 ) exp{ikd }  ( Z  Z 0 ) 2 exp{ikd }
2

By reviewing the calculations that have led to this result, we may see that it is valid even if the wave
impedance Z and the wave number k inside the film are complex numbers evaluated at the wave’s
frequency  – see Eqs. (7.28) of the lecture notes:
1/ 2
  ( ) 
Z  Z     k  k      ( )  ( )
1/ 2
 , .
  ( ) 
The problem’s assignment implies that the film is non-magnetic, so () = 0, and that its
conductance is Ohmic and non-dispersive, so we may account for it by using Eq. (7.46) with real () =
 = const:

     opt    i .

Since, per Eq. (6.33), the skin depth s equals (2/0)1/2, the given condition s << 1/k0  1/(00)1/2
means that / >> 0, so unless we are in very close vicinity of some optical resonance (which should
have been specified in the assignment), / >> opt() as well, and we may neglect the first contribution
to the complex permittivity, taking30, 31
1/ 2
   1 i
    i , so Z   0  , and k  i 0 
1/ 2
 , (**)
  i  s

30 Actually, this means that for the thin film description, we are using the quasistatic approximation (see Sec. 6.3
of the lecture notes), i.e. are neglecting the displacement currents in the film in comparison with the actual
(Ohmic) currents in it. However, using Eq. (*) for T, which follows from the full system of Maxwell equations,
means that we are still keeping the due account of the displacement currents in the space around the film.
31 You may notice that the last expression for k differs from Eq. (6.30) for  only by the multiplication by the

imaginary unity. (The difference is due to a different definition of . – see Eq. (6.29).)

Problems with Solutions Page 271

Essential Graduate Physics EM: Classical Electrodynamics

  / i 
1/ 2

   2  k 0 s
1/ 2
Z 1
 0    0  0 1 / 2    .
Z0  0 /  0 1/ 2
2i 1/ 2

 0   1 i

Plugging these expressions into Eq. (*) and taking into account that the condition k0s << 1 means, in
particular, that Z << Z0, and that since k0d << 1 as well, i.e. exp{–k0d}  1, we get
1
4 ZZ 0  1 i 
T0  2  cos kd  sin kd  .
  
Z 0  2 ZZ 0 exp{ikd }  Z 0  2 ZZ 0 exp{ikd } 
2
 2k 0 s 
Calculating the power transparency T from the last (innocently looking :-) expression, we have
to be careful because according to the last of Eqs. (**), the argument kd of the involved trigonometric
functions is a complex number. After the separation of their real and imaginary parts,32 we get
1
  d  1 i  d 
T0  cos 1  i    sin 1  i   
   s  2k 0 s   s 
 d  d 1 d  1 d d
 cosh  cos  sin   sinh cos 
 s   s 2k 0 s  s  2k 0 s s s 
1
 d  d 1 d  1 d d  
 i sinh  sin  cos   cosh sin   ,
  s   s 2 k 
0 s  s  2k 0 s s  s  

so the final result is given by a somewhat bulky formula

2
 d  d 1 d  1 d d
2

T  T0  cosh  cos  sin   sinh cos 

 s   s 2k 0 s  s  2k 0 s s s 
1
 d  d 1 d  1 d d
2

 sinh  sin  cos   cosh sin   .
 s   s 2k 0 s s  2k 0 s s s  
It may be readily simplified in the limits of relatively large and small values of the film’s thickness:
 ( k 0 s ) 2  2d  1 d
 exp   0, for 2 ln  ,
 2  s  k 0 s s
T  2 (***)
 1  d  , for
d
 1 .
   s
  k 0 s 
2

Plots in the figure below show the calculated T as a function of the d/s ratio for several values of
the parameter k0s, typical for the transmission of microwave radiation through thin metallic films. The
crossover between the asymptotic behaviors (***), taking place at d ~ s, is clearly visible.

32 See, e.g., MA Eq. (3.5).

Problems with Solutions Page 272

Essential Graduate Physics EM: Classical Electrodynamics

0.1

0.01
10 2
3
110
10 3
4
110
T
k 0 s  10 4
5
110

6
110

7
110

8
110
4 3
110 110 0.01 0.1 1 10
d /s

The plots also show that it makes sense to break out the second of the asymptotes, for d/s << 1,
into two sub-limits:
  k  2 2 d
  0 s   1, for k 0 s   1,
 d  
T  s

 1  2d  1, for d  k   1.
 k 0 s2 s
0 s

The first of these functions describes the quasi-linear part of the log-log plots in the figure above;
the smaller k0s, the broader this region. The second of them may be rewritten in a different form by
using Eq. (6.33) for the skin depth, and Eq. (7.8) for the free-space wave impedance:
d 0 Z0
T  1  1 , for Z 0 /R  1 , (****)
k0 R
(where R  1/d),33 and derived in a much simpler way. Indeed, if R >> Z0, the film almost does not
disturb the incident wave (so in the crudest approximation, T  1), and the electric field E applied to the
film is just that of the incident wave – at the normal incidence we are exploring now, it is parallel to the
film. According to the first form of Eq. (4.39), this field leads to the following Joule power loss per unit
area of the film:
dP E2
 pd  E 2 d  .
dA R

33As was already mentioned several times, starting from the model solution of Problem 4.12, the R so defined is
called the sheet resistance (or the “resistance per square”) of the thin film.

Problems with Solutions Page 273

Essential Graduate Physics EM: Classical Electrodynamics

On the other hand, according to Eq. (7.9), the power of the incident wave (also per unit area of its front)
is S = E2/Z0, so the lost fraction of the power is (dP/dA)/S = Z0/R , immediately leading to Eq. (****)
for the transmitted fraction of the power.
Note that the second of Eqs. (***) may be also rewritten in terms of the Z0/R ratio (in this more
general case, not limited to its small values):
1
T  , for d   s .
1  Z 0 / 2 R 2
The derivation of this expression from scratch, utilizing the smallness of d in comparison with s from
the very beginning, will be the subject of the next problem.

Problem 7.15. One of the results of the previous problem’s solution was the following expression
for the coefficient of power transmission of a plane electromagnetic wave through a thin conducting film
of thickness d << s, , at normal incidence:
1
T  ,
1  Z 0 / 2 R 2
where R  1/d is the sheet resistance (“resistance per square”) of the film. Derive this formula in a
simpler way, utilizing the smallness of d from the very beginning. Also, calculate the power reflection
coefficient R, compare it with T, and comment.
Solution: Just as it was done in Sec. 7.4 of the lecture notes, after placing the origin of the z-axis
(directed along the wave’s propagation and hence normal to the film) at the film’s location, we may look
for the complex amplitudes of the fields (all proportional to e-it) in the following form,34

E  E 0  
 
 e ikz  R e ikz , for z  0,
H 

E 0  e  R e

ikz ikz

, for z  0,
Te ikz , for 0  z , Z 0 Te ikz , for 0  z.

These expressions satisfy the Maxwell equations outside the film, so in order to find the
coefficients R and T, we should only formulate and use appropriate boundary conditions. The first of
them is, as before, the continuity of the only (tangential) component of the electric field at z = 0, giving
1 R  T . (*)
However, the magnetic field is not continuous because of the Ohmic current induced, by the wave, in
the film. Repeating the Ampère-law arguments that give, in particular, Eq. (6.38) of the lecture notes,
for our current case, we get
H z 0  H z 0  J ,

where J is the linear density of the current. Due to condition d << s, the current is uniformly distributed
across the film’s thickness, and may be calculated just as

34Note that by writing these expressions we have already used the condition d << , by neglecting the distance
between the two regions, i.e. film’s thickness d, on the scale of wavelength   2/k.

Problems with Solutions Page 274

Essential Graduate Physics EM: Classical Electrodynamics

E z 0
J  jd  E z 0 d ,
R
resulting in the following second boundary condition:
1 R T T
  . (**)
Z0 Z0 R
Now readily solving the simple system of linear equations (*) and (**), we get
1 Z 0 / 2R
T , R ,
1  Z 0 / 2R 1  Z 0 / 2R
so for the power transmission coefficient, T   T 2, we indeed get the formula cited in the assignment,
while the power reflection coefficient is:35
2 Z 0 / 2 R 2
R  R  .
1  Z 0 / 2 R 2
Most noticeably, a very resistive film with R >> Z0 affects the power transmission more significantly
than its reflection:
2
Z  Z  Z0
T  1 0 , R   0  , for  1 .
R  2R  R

One may wonder why the same ac current J(t) flowing in the film does not radiate equally to
both sides; the reason is that it acts on the background of the incident wave’s field E(t). Since both
processes, and hence the incident wave and the wave induced by the current have the same frequency,
and are in a firm phase relation (coherent with each other), their powers do not add up – see a more
detailed discussion of this issue in Chapter 8.

Problem 7.16. A plane wave of frequency  is normally

 0 , 0 ,  ', μ'
incident, from free space, on a plane surface of a material with real
electric permittivity ’ and magnetic permeability ’. To minimize 1 T T'
the wave’s reflection from the surface, it may be covered with a
layer, of thickness d, of another transparent material – see the
figure on the right. Calculate the optimal values of , , and d. 0 d z
Solution: This problem is an evident generalization of R0 R
Problem 13, and may be approached very similarly, by considering
five plane waves with the complex amplitudes indicated in the
figure. The system of four linear equations relating these amplitudes, resulting from the macroscopic
boundary conditions on the two interfaces, is also a straightforward generalization of that derived and
solved in Problem 13:

35Note that for any finite Z0/Rs ratio, the sum (T + R) is below 1, with the balance corresponding to the power
absorbed in the film due to the Joule dissipation – see Eq. (4.39) of the lecture notes.

Problems with Solutions Page 275

Essential Graduate Physics EM: Classical Electrodynamics

1  R0 T  R
1  R0  T  R ,  ,
Z0 Z
Te ikd  Re ikd T 'e ik'd
Te ikd  Re ikd  T 'e ik 'd ,  ,
Z Z'
where Z0, Z, and Z’ are the wave impedances (7.7) of the corresponding materials, and k = ()1/2.
However, now the equations have more parameters, so their solutions are more bulky, and may be
harder to analyze. The simplest way toward such analysis is to solve the system for the reflected wave’s
amplitude R0, eliminating the variables T and R first, and then the product T’eik’d treated as one
variable.36 The result is
R0 

Z'  Z 0  Z cos kd  i Z 2  Z 0 Z' sin kd
.
 (*)

Z'  Z 0  Z cos kd  i Z 2  Z 0 Z' sin kd 
As the simplest sanity check, for Z = Z0, i.e. a free-space “layer”, this result reduces to
Z'  Z 0 2ikd
R0 Z  Z  e ,
0 Z'  Z 0
i.e. to the expression that differs from the first of Eqs. (7.68), with the proper notation replacements Z–
 Z0 and Z+ Z’, only by the reflected wave’s additional phase shift 2kd – just as we should have for
its 2d-long roundtrip inside the layer. As another check, for the constructive interference case, i.e. kd =
m with m = 0, 1, 2,…, i.e. for sinkd = 0 (in particular for d = 0, i.e. for the surface with no coating at
all), Eq. (*) reduces to Eq. (7.68) for any Z.
The result we are seeking may be obtained in the middle between those resonance values of d,
i.e. at the destructive interference of the waves reflected from the interfaces. In this case, kd = (m + ½),
so coskd = 0, and Eq. (*) yields
Z 2  Z 0 Z'
R0  2 .
kd 2 m  ½  Z  Z Z'
0
Now, by selecting
Z  Z opt  Z 0 Z'  ,
1/ 2
(**)

we get R0 = 0, i.e. the system does not reflect – at least at the discrete frequencies providing the above
resonant condition of kd. It is easy to use Eq. (*) to verify that the frequency bandwidth around such
value, within which the reflection is relatively low,37 is the largest for m = 0, i.e. at
  
kd  , i.e. d   ,
2 2k 4
where  = 2/k is the wavelength inside the “coating” layer.

36 Another possible approach to this problem is to use the transfer matrix method. For the solution of this simple
problem, its introduction and usage would take more space than the given solution, but for more complex cases,
such as multilayer systems, the method is invaluable – not only in electromagnetism but also in quantum
mechanics and statistical physics – see, e.g., QM Sec. 2.5 and SM 4.5.
37 Even larger bandwidths may be reached using multilayer coatings, which are becoming more and more popular
as their fabrication technologies improve.

Problems with Solutions Page 276

Essential Graduate Physics EM: Classical Electrodynamics

Nowadays, such quarter-wavelength antireflective coating is broadly used in optics – from

billion-dollar telescopes all the way down to consumer glasses. Note that most optical materials are not
magnetic,   ’  0, so Z  1/2  n, Eq. (**) is more frequently represented as38

n  nopt  n'  ,
1/ 2

where n is the index of refraction – see Eq. (7.84). However, this form of the full result (**) should not
obscure the important physical fact that it relates the wave impedances of the materials, and has nothing
to do with their wave propagation speed, which is most frequently associated with the refractive index in
many undergraduate (and even some graduate) textbooks.

Problem 7.17. A monochromatic plane wave is incident from inside a medium with  > 00 on
its planar surface, at an incidence angle  larger than the critical angle c = sin–1(00/)1/2. Calculate
the depth  of the evanescent wave penetration into the free space, and analyze its dependence on .
Does the result depend on the wave’s polarization?
Solution: Selecting the x- and z-axes within the plane of ,   0 , 0
incidence (see the figure on the right), we can claim that in our problem k
/y = 0, so the 3D wave equation that describes the electric and
magnetic fields at z > 0 (see, e.g, Eq. (7.3) of the lecture notes) is    c  ?
reduced to its 2D form z
 2 2 1 2 
 2  2  2 2  f  0,
 x z c t 

where f is any field’s component, and c = 1/(00)1/2 is the speed of light

in free space. Let us look for the solution of this differential equation in x
a natural 2D-wave form
f  f  exp i k x x  k z z  t  . (*)

In order to satisfy the boundary conditions at z = 0 for an arbitrary t, the transverse component kx of the
evanescent wave vector has to match those of the incident and reflected waves:

k x  k sin      sin  ,

1/ 2
(**)
– cf. Eq. (7.80) of the lecture notes. On the other hand, in the free-space region, the wave vector’s
magnitude should be equal to k0  /c, so
k x2  k z2  k 02 . (***)

At  > c  sin-1(00/)1/2 = sin-1(k0/k), we have kx > k0, so Eq. (***) may be only satisfied by a purely
imaginary kz = i/. Plugging this expression into Eq. (*), we get

38 For a typical optical glass, n’  1.5, so the optimal coating material would have nopt = (n’)1/2  1.22. However,
there are no convenient solid transparent materials with such refractive index, so the most popular practical choice
is magnesium fluoride (Mg2F) having n  1.38 and hence giving R02  0.015 instead of R02  0.04 for uncoated
glass.

Problems with Solutions Page 277

Essential Graduate Physics EM: Classical Electrodynamics

f  f  expi k x x  t exp z /  ,
so  is exactly the penetration depth we need. As a result, Eqs. (**) and (***) yield
1 1 c sin  c 0 sin  c
    ,
k 2
x k 0
2 1/ 2
  sin    0  0 
2 1/ 2
 sin   sin  c 
2 2 1/ 2
2 sin   sin 2  c 1 / 2
2

where 0 = 2/k0 is the free-space wavelength.

The result shows that  is the smallest at the “grazing-angle” incidence,   /2:
0 sin  c 
  0 tan  c ,
2 1  sin  c 
2 1/ 2
2

and diverges as  approaches c. Finally, note that this result for  is a result of “kinematic” analysis,
and hence (just like the Snell and reflection laws) does not depend on the incident wave’s polarization.

Problem 7.18. Calculate the critical angle c for a wave of frequency , incident from free space
upon a planar surface of a plasma with electron density n, and discuss the implications of the result for
ultraviolet and X-ray optics.
Solution: The critical angle of a plane interface between two media may be calculated from Eq.
(7.85) of the lecture notes,
1/ 2
    
sin  c    ,
    
with the indices  referring to the media, respectively, behind and in front of the interface – see Fig.
7.10. For our particular case, we should use Eqs. (7.36)-(7.37) for +, while taking – = 0 and + = – =
0. The result is
1/ 2
  p2  ne 2 .
sin  c  1  2  , with  p2 
    0 me
 
For the analysis, it is helpful to rewrite this relation in the equivalent form
p
cos  c  .

It shows that the real angle c exists only for the frequencies  above the plasma frequency p. (At  <
p the wave is totally reflected from the plasma at any angle of incidence because it cannot propagate in
it at all – see Fig. 7.6 and its discussion in Sec. 7.2 of the lecture notes.) As the wave’s frequency grows,
cosc decreases, i.e. the angle c grows. Eventually, at  >> c the angle becomes very close to /2.
Expanding cosc near this point into the Taylor series: cosc = /2 – c + …, and keeping only the two
leading terms of the expansion, we get
 p
c   , for    p .
2 

Problems with Solutions Page 278

Essential Graduate Physics EM: Classical Electrodynamics

This result means that the total reflection of waves from the plasma may be obtained even at very
high frequencies, at the so-called grazing angles on incidence. This effect enables the fabrication of
reflective components of ultraviolet and X-ray optical systems (most importantly, of EUV lithography
tools) from metals, whose p is barely above the visible light range – see the estimate in Sec. 7.2.

Problem 7.19. Analyze the possibility of propagation of surface electromagnetic waves along a
planar boundary between plasma and free space. In particular, calculate and analyze the dispersion
relation of the waves.
Hint: Assume that the magnetic field of the wave is parallel to the boundary and perpendicular to
the wave’s propagation direction. (After solving the problem, justify this
z
choice.)
E z E
Solution: With the suggested assumption, the problem may be
readily solved for a planar interface between two linear, isotropic media   ,  H Ex
with even arbitrary frequency dispersion of both () and ().39 Indeed,
with the coordinate frame selection shown in the figure on the right, where   ,   x
Ez
the x-axis is directed along the wave propagation, the fields of a E
monochromatic surface wave may be represented in the form

 
E  r, t   Re E x n x  E y n y  E z n z expi kx  t     z , 
(*)
H r, t   ReH 

n y expi kx  t     z .

Here the x-component k of the wave vector is taken the same for both media to satisfy the boundary
conditions (see below) simultaneously at all points of the interface, while its z-components are assumed
imaginary to ensure that the wave fields decay into the media bulk.40 Plugging Eqs. (*) into the source-
free (homogeneous) macroscopic Maxwell equations (7.2), we see that they indeed represent a valid
solution, provided that the following relations are satisfied:

 2  k 2   2        , (**)
 k
E x   H , E y  0, E z   H. (***)
i       
These relations show, in particular, that if both media are lossless (i.e. if  and  are both real), the
exponents  are real as well, and the phases of oscillations of the Cartesian components Ex and Ez are
shifted by /2. The last fact means that the vector E = {Ex, 0, Ez} in each medium rotates, its end
following an ellipse with the semi-axes proportional to the coefficients  and k, respectively.
In order to find these coefficients, Eq. (**) alone is insufficient. An additional equation may be
obtained from applying the boundary conditions (3.37), (3.56), and (5.117):

39 As was discussed in Sec. 7.2 of the lecture notes, any non-zero Ohmic conductivity  () of a medium may be
incorporated into its complex electric permittivity () – see Eq. (7.46).
40 This means that in contrast to the situation at the total internal reflection (see its discussion in Sec. 7.4), the
interface wave is evanescent in both media rather than in just one of them.

Problems with Solutions Page 279

Essential Graduate Physics EM: Classical Electrodynamics

E x  E x  E x ,    E z     E z , H  H.

These homogeneous linear equations41 are compatible with Eqs. (***) only if
   
  . (****)
    
This is the key relation for the interface/surface waves. It shows that such waves cannot
propagate on the interface between two “usual” wave media, with real and positive dielectric constants.
However, as was discussed in detail in Sec. 7.2, dilute gases feature frequency ranges with negative
electric permittivity immediately above resonance (in quantum mechanics, quantum-transition)
frequencies – see, e.g., Fig. 7.5.42 Moreover, according to Eq. (7.36), in a collision-free plasma, the
function () is real and negative at all frequencies below the plasma frequency p given by Eq. (7.37).
For the waves on the surface of such a plasma with – = 0, Eqs. (**), (****), and (7.36) yield
1/ 2
   p   
2 2
2
k , for    p .
c   p2  2 2 
This dispersion relation is shown with the red line 1.5

in the figure on the right, while the blue line shows the   ck
relation (also shown in Fig. 7.6 in the lecture notes) for 
the usual (“3D”) waves in the same plasma. 1
p
  p / 2
At low frequencies ( << p), the surface waves43
have almost the same properties as the usual 0.5
electromagnetic waves in free space: k  /c, vgr  vph  c.
(This is natural, because in this case, according to Eq.
(****) with +() = 0 and -()  –0p2/2  –, + << 0
–, i.e. the wave propagates mostly is free space, only 0 0.5 1 1.5 2

weakly “guided” by the plasma surface.) As the frequency k /  p / c 

grows, the wave vector diverges at   p/2, so the
surface plasmons cannot propagate at higher frequencies. In this limit, –()  –0 = –+(), so +  –,
i.e. the wave is equally shared by the plasma and free space.
Note that the conceptual importance of such plasmons extends beyond plasma physics and
engineering. For example, let us return to the simplest (and very common) situation shown in Fig. 2.26
of the lecture notes, and ask ourselves: if a charge q has been brought close to a metallic surface from
afar very fast, how long would it take for the familiar stationary picture of the charge image to become
quantitatively valid? There is of course the relativistic limitation t >> d/c, but the limitation imposed
by surface plasmon propagation, t >> 1/p, where is the plasma frequency of the conduction electrons

41 Actually, one of these relations is redundant.

42 Note that if the interface waves can propagate, i.e. if +() and -() have different signs, the signs of z-
components of the electric field in the two bordering media are opposite, as shown in the figure above. As a
result, the electric field vectors in the media rotate in opposite directions – see, e.g., the nice animation available
online at http://en.wikipedia.org/wiki/Surface_plasmon_polariton#Dispersion_relation.
43 The quantized excitations of these waves are called either surface plasmons or (due to their similarity to the
polaritons discussed in Problem 7.8) surface plasmon polaritons.

Problems with Solutions Page 280

Essential Graduate Physics EM: Classical Electrodynamics

in the metal, may be much more significant. As was mentioned in Sec. 7.2 of the lecture notes, for good
metals, p ~ 1016 s–1, so the time needed for the image charge formation (read: for the surface charges to
approach their stationary distribution) is of the order of a femtosecond.
Finally, let us revisit the initial assumption Hz = 0. Due to the symmetry of the homogeneous
Maxwell equations (6.100) to the simultaneous swap E  H, D  –B, it is evident that similar surface
waves with Hz  0 but Ez = 0, are also possible, but only if () < 0 in one of the media, which is not
common. Note also that conceptually similar (though quantitatively different) mechanical surface waves
may propagate at sharp interfaces between different elastic media. 44

Problem 7.20. Light from a very distant source arrives to an observer z

through a plane layer of nonuniform medium with a certain 1D gradient of its
refraction index n(z), at angle 0 – see the figure on the right. What is the genuine i
direction i to the source, if n(z)  1 at z  ? (This problem is evidently
n z  0
important for high-precision astronomical measurements at the Earth’s surface.)
Solution: If the layer thickness’ scale d is much larger than light's
wavelength , we may represent it locally (on a distance z such that  << z <<
d) as a plane wave – see, e.g., Eq. (7.79) of the lecture notes. Hence, if n changes only in the z-direction,
we may repeat the arguments at the beginning of Sec. 7.4 to conclude that the wave vector’s component
in the direction perpendicular to this axis, equal to k sin  n sin, cannot change at the wave's
refraction between any two elementary sub-layers. Hence, the product n(z) sin(z) has to stay constant
throughout the whole layer. Applying this conclusion to the initial direction i (at z >> d, where n(z) 
1) and the observer's location (z = 0, where n(z) = n(0)  n0), we finally get a very simple answer:
sin  i  n0 sin  0 , i.e.  i  sin 1 n0 sin  0  , (*)
i.e. just the Snell law, regardless of the details of the function n(z).
As Table 3.1 of the lecture note shows, for the air on the Earth's surface, n0  1/2  1.0003, so
the angular corrections R  i – 0 to the actual positions of stars and planets (called atmospheric
refraction) may be rather substantial – about one angular minute at 0 ~ /4. Still, it is typically much
smaller than 0, so expanding sini in the Taylor series in small R and keeping only the two leading
terms, we may approximate Eq. (*) as
sin  0  R cos  0  n0 sin  0 , giving R  n0  1 tan 0 (**)
– the form suggested by W. Snellius himself.
Note, however, that not only Eq. (**) but even the more general Eq. (*) should be applied to
practical astronomical observations at angles 0  /2 with care because here the Earth’s curvature,
ignored in the derivation of that result, becomes important. With the account of these effects, the
atmospheric refraction at the horizon is as large as ~35 angular minutes. i.e. is of the same order as the
visible diameters of the Sun and the Moon.

44 See, e.g., CM Sec. 7.7, and references therein.

Problems with Solutions Page 281

Essential Graduate Physics EM: Classical Electrodynamics

Problem 7.21. Calculate the TEM impedance ZW of uniform transmission lines with well-
conducting electrodes and the cross-sections shown in the figure below:
(i) (ii) (iii)
w  d
2R
d1
d  R d d2
2R w  d1, 2

(i) two parallel round wires separated by distance d >> R,

(ii) a microstrip line of width w >> d,
(iii) a stripline with w >> d1 ~ d2,
in all cases using the coarse-grain boundary conditions on conductor surfaces. Assume that the
conductors are embedded into a linear dielectric with constant  and .
Solutions:
(i) From the similar limit of the solution of Problem 2.13, with R1 = R2 = R, generalized to the
dielectric filling by replacing 0 with , the mutual capacitance per unit length is

C0  .
ln (d / R)
The inductance L0 per unit length may be found either from magnetostatics or (even easier) from the
general relation (7.114):45
  d
L0   ln .
C0  R
Combining these expressions, we get
1/ 2 1/ 2
L  Z d 
ZW   0   ln , where Z    .
 C0   R  
This formula is close in structure to that for the coaxial cable – see Eq. (7.120) of the lecture notes, with
an extra factor of 2 due to two (rather than one) thin conductors, and with the distance d between the
wires playing the role similar to the diameter of the outer conductor of the cable.
As was mentioned in Sec. 7.6 of the lecture notes, such wire pairs, lighter and cheaper than the
coaxial cable, have been the basis of all early electric communication technologies (such as telegraphy),
and are still used, in the form of twisted pairs, in landline telephony at frequencies up to a few hundred
kHz, and in relatively short Ethernet cables at frequencies up to a few hundred MHz.
(ii) Neglecting the fringe field effects (whose relative contribution scales as d/w << 1), we may
use Eq. (3.55) for the line’s capacitance per unit length:
w
C0  .
d

45Note again that this relation is valid only at the coarse-grain boundary condition (6.38), acceptable when the
magnetic field penetration into the conducting electrodes is negligible (in our current case, when s << R).

Problems with Solutions Page 282

Essential Graduate Physics EM: Classical Electrodynamics

The inductance per unit length may be found, again, either from magnetostatics46 or from the same Eq.
(7.114):
 d
L0   .
C0 w
Combining these expressions, we get
d
Z W  Z  Z .
w
This expression shows that the microstrip lines may have a very low impedance. Such low-
impedance lines are broadly used in electronics, e.g., on printed circuit boards. Due to the low fringe
fields, they may ensure low mutual interference (“crosstalk”) even if spaced rather closely – at distances
of the order of w >> d.
(iii) The stripline geometry, which is broadly used in multi-layer printed circuit boards and for
on-chip interconnects in microelectronics, provides even lower crosstalk between adjacent transmission
lines. Its parameters may be readily found by calculating C0 as a parallel connection of two plane
capacitors, and L0 from that result and Eq. (7.114):
1 1/ 2 1
1 1    1
1  L  1 1 1 
C 0  w   , L0      , ZW   0   Z    .
 d1 d 2  C 0 w  d1 d 2   C0  w  d1 d 2 
If one of the gaps is much narrower than the other one (say, d1 << d2), the electromagnetic field is
concentrated in it, and the result is reduced to that for the microstrip line.

Problem 7.22. Modify the solution of Task (ii) of the previous problem for a superconductor
microstrip line, taking into account the magnetic field’s penetration into both the strip and the ground
plane.
Solution: In this case, the expression for L0 has to be replaced with the solution of Problem 6.20:

L0  d  2 L  ,
w
while the capacitance C0 per unit length remains the same as in the solution of the previous problem,
because the electric field penetrates into conductors (including superconductors) by a much smaller,
typically negligible screening length – see Sec. 2.1 of the lecture notes. As a result, the TEM wave
propagation speed in such a line is lower than that (v) of plane waves:47

46 As was discussed in the solution of Problem 6.20 in the context of superconducting electrodes, if the width d of
the gap between the conductors and the depth of the field penetration into the line electrodes (say, the skin-effect
depth s) are much smaller than w, the field in the gap is uniform, B = I/w. If, in addition, the field penetration
depth is also much less than d, then the magnetic field is localized in the gap, and the magnetic flux (per unit
length) is /l = Bd = Id/w, immediately giving the above formula for L0.
47 This difference is especially large (v’/v ~ 0.1) in the so-called long (or “distributed”) Josephson junctions with

d ~ 1 nm and L ~ 100 nm, enabling very interesting (pseudo-relativistic) effects of interactions between the slow
TEM waves and magnetic-field-induced “waves” of supercurrent – see, e.g., Section 6.4 of the monograph by M.
Tinkham (see the last section of References).

Problems with Solutions Page 283

Essential Graduate Physics EM: Classical Electrodynamics

1/ 2
1 1  d  v
v'       v.
( L0 C 0 ) 1/ 2
 1 / 2  d  2 L  1  2 L / d 1 / 2
Note that in this case, the “general” relation (7.114) is not satisfied because its derivation is
based on the assumption of similar boundary conditions for the magnetic and electric fields on the
conductor surfaces.

Problem 7.23.* What lumped ac circuit would be equivalent to the TEM-line system shown in
Fig. 7.19 of the lecture notes, with an incident wave’s power Pi? Assume that the wave reflected from
the lumped load does not return to it.
Solution: It is intuitively clear that the circuit has to include:
(i) the lumped load with the impedance ZL(),
(ii) a lumped circuit element representing passive properties of the transmission line, with the ac
impedance equal to the line’s impedance ZW(), and
(iii) some ac generator representing the incident wave.
Following these arguments, let us try to use the circuit
V t 
shown in the figure on the right, where V(t) is an e.m.f.
representing the incident wave. An elementary calculation of the
Z ( ) I t  Z L ( )
complex amplitude of the current I(t) in the load yields I = W
V/[ZL() +ZW()], so the voltage drop across the load is V = I
ZL() = V ZL()/[ZL() +ZW()]. Now the calculation similar to that carried out at the derivation of Eq.
(7.42) yields the following expression for the average power absorbed in the load:
2
1 1 V
PL  V I *  Re Z L ( ) . (*)
2 2 Z L ( )  Z W ( ) 2

On the other hand, we can calculate the same power in the real system (Fig. 7.19) by subtracting,
from the incident power Pi, the reflected wave’s power Pr = PiR2, where R is given by Eq. (7.118):

   
2 2 2
2  Z L ( )  Z W ( )   P Z L ( )  Z W ( )  Z L ( )  Z W ( ) .
PL  Pi 1  R  Pi 1 
 Z L ( )  Z W ( )
2  i
Z L ( )  Z W ( )
2
 
In the most important case of a loss-free line, its impedance ZW is real, and this expression reduces to
4 Z W  
PL  Pi Re Z L ( ) (**)
Z L ( )  Z W  
2

Comparing Eqs. (*) and (**), we see that they give similar results (and hence the lumped circuit is
equivalent to the wave system shown in Fig. 7.19), if the effective ac e.m.f. has the amplitude

V  8Pi Z W  1 / 2 .
(Note the somewhat counter-intuitive numerical coefficient.)

Problems with Solutions Page 284

Essential Graduate Physics EM: Classical Electrodynamics

Such representation of a transmission line, carrying an incident wave, by its impedance and an
e.m.f. is valid even if the lumped load is nonlinear and/or time-dependent, and is broadly used for the
analysis of not only linear but also nonlinear and parametric microwave devices. Note, however, that
this result is only valid under the assumption mentioned in the assignment, that the wave reflected from
the load does not return to it. This condition (generally beneficent for applications, e.g., for the amplifier
stability) may be satisfied using such non-reciprocal devices as circulators – see the discussion in the
model solution of Problem 12.

Problem 7.24. Find the lumped ac circuit equivalent to a

 
loss-free TEM transmission line of length l ~ , with a small cross-
section area A << 2, as “seen” (measured) from one end, if the
line’s conductors are galvanically connected (“shortened”) at the l~
other end – see the figure on the right. Discuss the result’s dependence on the signal frequency.
Solution: Neglecting losses in the transmission line, we can describe any “global” variable, e.g.,
the voltage V(z, t) or the current I(z, t) in it, by a sum of two sinusoidal waves: one traveling to the
shortened end, and another one reflected from it. Taking the shortened end’s position for z = 0, for the
complex amplitudes of these variables we can write


V ( z  0)  V0 e ikz  Re ikz ,  V
I  ( z  0)  0 e ikz  Re ikz .
ZW
 
– cf. Eqs. (7.64)-(7.65) for plane waves. Now requiring the ratio V/I to vanish at the shorted end (at z
= 0), we get R = –1, so at the other end of the line (at z = –l)
V e ikl e ikl
Z ( )   ZW  iZ W tan kl . (*)
I  z  l e ikl e ikl
This result shows that in contrast to the TEM line impedance ZW = (L0/C0)1/2, which does not
depend on frequency, the effective ac impedance Z() of the shorted line segment is a strong function of
. In particular, at low frequencies  << v/l (i.e. at kl << 1) this expression reduces to
1/ 2
L 
Z ( )  iZ W kl  i  0   L0 C 0 1 / 2 l  iL0 l  iL ,
 C0 
where L = L0l is the segment’s inductance. This is a well-known expression for the complex impedance
of a lumped inductance. (The negative sign is due to the fact that we are using the “physics” convention
exp{–it} for the time dependence of all variables, vs the exp{+it} common in electrical engineering.)
However, as kl is increased, the impedance given by Eq. (*) grows faster than that of a lumped
inductance and diverges at kl = /2, i.e. at wavelength l = /2k  /4. This divergence means that in the
absence of power losses, a nonvanishing voltage V may be sustained by vanishing current. Such
resonant behavior is similar to that of an ac circuit consisting of a lumped inductance and a lumped
capacitance connected in parallel; it repeats at all frequencies where

knl   n , (**)
2
i.e. (n +1/2)n/2= l. These so-called parallel resonances are interleaved, at wave number values

Problems with Solutions Page 285

Essential Graduate Physics EM: Classical Electrodynamics

k' n l  n , (***)

with resonances of a different kind, where Z() = V/I vanishes, and hence a finite ac current may be
sustained with little, if any, ac voltage. These are so-called series resonances, similar to those in lumped
in-series LC circuits.
Such distributed resonators are broadly used in electrical engineering at high frequencies where
the implementation of genuinely lumped circuits may lead to unacceptably high energy losses.

Problem 7.25. Represent the fundamental H10 wave in a rectangular waveguide (see Fig. 7.22 of
the lecture notes) as a sum of two plane waves, and discuss the physics behind such a representation.
Solution: Let us use the last of Eqs. (7.138) of the lecture notes to write the following expression
for the instant electric field of this wave:
 ka x 
E(r, t )  Re i ZH l sin expi k z z  t  n y
  a 
       
 Re  E expi  x  k z z  t   E expi   x  k z z  t  n y ,
  a   a  
with E  (ka/2)ZHl. The last displayed expression shows that inside the waveguide, the mode is
nothing more than the sum of two plane waves, linearly polarized along the y-axis, with equal
amplitudes E, and wave vectors k with the Cartesian components kx = /a, ky = 0, and the kz defined
by Eqs. (7.122) and (7.132) (the latter one, with n = 1 and m = 0), so
2 kz a
k x2  k y2  k z2  k 2  k 2  2
,
v
k k
– see the left panel of the figure on the right.
This figure shows that the plane wave      
propagation angle  satisfies the following v v
relation:
kx  0 / 2  0  kx
sin     ,  
k ka a a a
where 0 = 2/k is the TEM wavelength. This angle asymptotically approaches zero at very high
frequencies ( << c) but grows as the frequency is reduced toward the cutoff value c, so at   c, 
 /2, i.e. kz  0. Note that the cutoff frequency corresponds to the half-wavelength of the plane wave
being exactly equal to the wide side of the waveguide, so longer waves cannot propagate in it.
The fact that we are dealing with two usual plane waves may be confirmed by the
electromagnetic field amplitude ratio calculation. Indeed, the similar decomposition of the H10 mode’s
magnetic field, given by Eqs. (7.131) and (7.137), also yields two plane waves:
1  k a      k a     
H (r, t ) H l Re   z n x  n z  expi  x  k z z  t    z n x  n z  expi   x  k z z  t  ,
2     a      a  
each with the amplitude

Problems with Solutions Page 286

Essential Graduate Physics EM: Classical Electrodynamics

1/ 2
H  k z a  2  ka 1
H  l    1  Hl  E ,
2     2 Z
whose ratio to E is given by the plane wave’s impedance Z = (/)1/2, regardless of the ratio /c.
Now, let us discuss why this simple decomposition of waves in the waveguide into plane waves
is possible. If these two plane waves propagated in an unlimited isotropic medium, we could notice, first
of all, that an insertion of conducting walls normal to the y-axis at any locations (say, at y = 0 and y = b,
see Fig. 7.22) would not perturb the field between them. Indeed, as we know from Chapter 2, the electric
field normal to the conductor’s surface is screened from penetration into its bulk by surface charges with
density (x,y) = Ez(x,y), without any perturbation of the applied field. Similarly, the magnetic fields of
these two plane waves are tangential to the wall surfaces and are shielded from penetration into their
bulk by the skin-effect currents of the linear density (6.38), without any field’s perturbation outside of
the walls.
The situation with the other two walls (at x = 0 and x = a, see Fig. 7.22) is more involved since at
an arbitrary location of such a wall, the sum of two plane waves would have some tangential component
of the electric field, and some normal component of the magnetic field, which cannot interact with the
wall without incident wave’s perturbation. However, as the above formulas for E and H show, at the
specific positions xn = n/kx, with any integer n, the component sums vanish, and the wall insertion
leaves the field between the walls intact.
These results may be also interpreted by saying that the fundamental H10 wave is formed by a
plane wave repeatedly reflected from the side walls of the waveguide – see the right panel of the figure
above.

Problem 7.26.* For the coaxial cable (see, e.g., Fig. 7.20 of the lecture notes), find the lowest
non-TEM mode and calculate its cutoff frequency.
Solution: The analysis may repeat that of the hollow circular waveguide (Fig. 7.23a) in Sec. 7.7
of the lecture notes, up to the derivation of the Bessel equation (7.140) for the radial factor R() of the
longitudinal field f – depending on the mode, either Ez or Hz. However, in the coaxial cable, the axial
points with  = 0 are inaccessible for the field, and hence, instead of the simple solution given by Eq.
(7.141), we have to look for its radial part in the form of a linear superposition of the Bessel functions of
the first and the second kind:
f nm  c1 J n (k nm  )  c 2Yn (k nm  ) cos n   0 , with n  0, 1, 2,... ,

where the eigenvalues knm of the transverse wave number kt have to be chosen to satisfy the boundary
conditions at both  = a and  = b.
For the E-modes, the boundary condition Eq. (7.124) yields the following system of two
equations for the constants c1,2:
c1 J n (k nm a )  c 2Yn (k nm a )  0,
c1 J n (k nm b)  c 2Yn (k nm b)  0 .

Problems with Solutions Page 287

Essential Graduate Physics EM: Classical Electrodynamics

These two linear, homogeneous equations are compatible if the denominator of the system equals zero.
This requirement yields
J n (k nm a)Yn (k nm b)  J n (k nm b)Yn (k nm a) ,

where the integer index m (taking the values 1, 2, 3,…) numbers the roots of this characteristic equation,
for each fixed angular index n (taking the values 0, 1, 2,…). Introducing the dimensionless variable nm
 knma, we may rewrite this characteristic equation as
 b  b
J n  nm Yn   nm   J n   nm  Yn  nm  . (*) n=0 n=1 n=0 n=1
 a  a b/a m=1 m=1 m=2 m=2
The table on the right shows the results for the 4 3.073 3.336 6.243 6.403
product (b/a – 1)nm  knm(b – a), obtained by numerical 2 3.123 3.197 6.273 6.312
solution of this equation for a few values of the ratio 1    2  2
b/a, and a few lowest numbers n and m.48 It shows that
the product knm(b – a) changes very slowly with that ratio, staying close to the asymptotic value m
(reached at b/a  1), for all reasonable values of b/a. This result may be readily understood physically:
at the cutoff frequency c = knmv (when kt = k = 2/), nearly m TEM half-wavelengths fit the distance
between the external and internal conductors – almost independently of n, provided that the latter
number is not too high. (Still, as could be expected, the axially-symmetric distribution with n = 0 gives
the lowest kt, and hence the lowest cutoff frequency.)
For the H-modes, we need to use another boundary condition, Eq. (7.126), that gives, instead of
Eq. (*), a different characteristic equation:
 b  b
J n'  nm Yn'   nm   J n'   nm  Yn'  nm  , (**) n=1 n=1
 a  a b/a m=1 m=2
where the prime sign means the derivative of the Bessel function over its 4 2.055 3.760
whole argument. The table on the right gives results for a different 2 2.031 4.023
dimensionless combination, (1 + b/a)nm  knm(a + b), which is more  1  2 4
relevant (virtually parameter-independent) in this case, for the two lowest
modes with n = 1.49 It shows that for any realistic b/a ratio, this combination is close to 2m. Physically,
this means that at the cutoff frequency, m TEM wavelengths  = 2/kt fit the average circumference p =
(a + b) of the cable’s cross-section.
Now we can compare the approximate values of the lowest knm for the E- and H-modes:
 2
for E  ,
ba 01 b  a for H11

48H. Dwight, J. Math. Phys. 27, 84 (1949).

49 For n = 0, we can use the recurrence relations (2.143), which in particular yield J0’ = –J1, Y0’= –Y1, to rewrite
Eq. (**) in a form similar to Eq. (*) with n = 1:
 b  b
J 1  0 m Y1   0 m   J 1   0 m  Y1  0 m  .
 a  a
This comparison shows that values k0m of the H-modes are equal to those of k1m of the E-modes (for the same m),
and hence are higher than the values k0m of the latter modes, so an H-mode with n = 0 cannot be the lowest one.

Problems with Solutions Page 288

Essential Graduate Physics EM: Classical Electrodynamics

so (just as in the circular waveguide analyzed in Sec. 7.7 of the lecture notes), the lowest non-TEM
mode is H11, with the cutoff frequency c  2v/(a + b), i.e. the TEM wavelength max  (a + b). This
result (which, at a/b  0, is close to Eq. (7.145) for the single-hole circular waveguide) is important
because it imposes a practical limit,  > (a + b) for using coaxial cables as TEM transmission lines, in
order to avoid unintentional excitation of non-TEM modes on unavoidable small inhomogeneities.
Note, however, that in some practical systems with long cables, the wavelength may be restricted
even more severely by the wave attenuation effects – see Sec. 7.9 of the lecture notes.

Problem 7.27. Two coaxial cable sections are connected

coaxially – see the figure on the right, which shows the system’s cut b a' b'
along its symmetry axis. Eqs. (7.118) and (7.120) of the lecture a
notes seem to imply that if the ratios b/a of these sections are equal,
their impedance matching is perfect, i.e. a TEM wave incident from
one side on the connection would pass it without any reflection at
all: R = 0. Is this statement correct?
Solution: The derivations of both formulas mentioned in the assignment were based on the
assumption that the wave fields in both sections are strictly of the TEM type, with the field vectors E
and H exactly normal to the wave vector k, i.e. to the system’s axis. This cannot be true close to the
connection point; for example, in the connection plane, this would lead to the vector E being parallel to
the “quasi-lid” surfaces of both conductors, thus contradicting the first of the boundary conditions
(7.104). Thus there is always some field perturbation, and as a result, some wave reflection from such a
sharp jump of the cable’s parameters. Moreover, according to the solution of the previous problem, if
the wavelength is smaller than c  (a + b), waves of other types may propagate at least in the wider
section and may be effectively excited by the TEM wave incident on the step.50
However, if the wavelength  is much longer than this c (as it is in most practical applications
of coaxial cables), such propagation is impossible, and the field perturbations may be only local,
spreading at distances of the order of a, b from the connection plane. Since the telegrapher’s equations
(7.110)-(7.111), whose corollary Eq. (7.118) is, are derived by averaging the field effects at distances
much larger than the cross-section’s dimensions (in our case, a and b), the effect of these perturbations
scales as c/ << 1. Thus if the matching condition a’/b’ = a/b is fulfilled, the reflection coefficient R is
not exactly zero but scales as the (small) c/ ratio.

Problem 7.28. Calculate the cutoff frequencies c of the

fundamental mode and the next lowest mode in the so-called ridge a w a
waveguide with the cross-section shown in the figure on the right, in the b
limit t << a, b, w. Briefly discuss possible advantages and drawbacks of t

50 Note that some (even very popular) textbooks describe over-simplified calculations of the higher mode
excitation at waveguide connections, based on expansions of the perturbed fields in series over all possible
traveling-wave modes. Such calculations, neglecting the field perturbations localized at the connection, are not
strictly valid at frequencies below the highest c of the connected parts; in this case, their results may be used as
estimates at best.

Problems with Solutions Page 289

Essential Graduate Physics EM: Classical Electrodynamics

such waveguides for signal transfer and physical experiment.

Solution: As was discussed in Sec. 7.5 of the lecture notes, at the cutoff frequency, kz = 0, i.e. the
field distribution is two-dimensional: both E and H at a point are uniquely defined by its position at the
waveguide’s cross-section. Hence our task is reduced to finding the oscillation frequency of such a 2D
standing wave.
Next, looking at Fig. 7.32 of the lecture notes for the rectangular waveguide, it is easy to
understand how its fundamental mode is deformed by an increasing ridge: the electric field becomes
more and more concentrated in the gap (of thickness t) between the ridge and the bottom wall. In our
limit t << a, b, w, this gap is essentially a plane capacitor with a very large capacitance given by Eq.
(2.28):
 wl
C 0 ,
t
where l >> a, b, w is the length of the waveguide.
In this mode, the surface charges of all other surfaces and the electric field outside the gap are
negligibly small. Hence, the linear density J of the currents flowing on the waveguide’s wall surfaces as
a result of the periodic recharging of the capacitance C is also constant outside the gap. As a result, each
of the two long, rectangular side cylinders with the cross-section’s area A = ab, separated by the ridge,
work as lumped single-wire-turn solenoids. According to Eq. (5.75) with N = 1, the inductance of each
of them is
 A  ab
L1  0  0 .
l l
Since these two inductances are connected to the capacitance C in parallel, the effective
inductance L equals L1/2, and the oscillation frequency of the resulting LC circuit (and hence the cutoff
frequency of the fundamental mode of the ridge waveguide) is
1/ 2 1/ 2 1/ 2
 2   
  c 
1 2t 2t 
 c LC     
  .
LC 1/ 2
 L1C   0  0 wab   wab 
(As the easiest sanity check, the frequency does not depend on l.) The corresponding transverse wave
number (7.123) is
1/ 2

k t LC  1  c LC   2t  .
c  wab 
Since we have assumed that t << a, b, w, this wave number is much smaller than any reciprocal
dimension of the cross-section. On the other hand, the next lowest mode is evidently the H10 wave inside
each of the side ab-cylinders;51 according to Eq. (7.133), its transverse wave number is
 c
k t 10  , so that  c 10  c k t 10    c 10 . (*)
max [ a, b] max [ a , b ]
Hence, the ridge waveguide has a very large gap between the cutoff frequencies of the
fundamental LC mode and all higher modes – the property to some extent similar to that of the coaxial

51 Actually, the gap with t  0 couples these modes in two side volumes, splitting their cutoff frequency into two
different ones, one of them lower than the value given by Eq. (*). However, at t << a, b, w, this shift is very small.

Problems with Solutions Page 290

Essential Graduate Physics EM: Classical Electrodynamics

cable.52 According to the general Eq. (7.122), within most of this frequency gap, i.e. at frequencies
(c)LC <<  < (c)10, such a waveguide has only one mode, with low dispersion, making it convenient
for signal transfer. In addition, the uniformity of the electric field inside the t-gap and of the magnetic
field outside it makes such waveguides very convenient for some experiments. On the other hand, the
wave attenuation of the LC mode is somewhat larger than that of the fundamental mode of a simple
rectangular waveguide – with the same external dimensions, at the same frequency.53
Finally, note a close similarity between this problem and Problem 7.41 below.

Problem 7.29. Prove that TEM-like waves may propagate, in the radial 0
direction, in the free space between two coaxial, round, well-conducting cones – see
the figure on the right. Can this system be characterized by a certain transmission line
impedance ZW, as defined by Eq. (7.115) of the lecture notes?
Solution: The term "radial direction" means that the wave vector k has to be
directed along the radius r (with the origin at the point where the cones meet). Any "TEM-like" wave
should have its vectors E and H directed normally to the vector k, and to each other. Due to the axial
symmetry of the system, this means that the complex amplitudes of the (monochromatic) wave fields,
similar to those defined by Eq. (7.98), may have only one of the following two spatial structures: either
E  r   n  E r ,  , and H  r   n  H r ,   , (*)
or, vice versa,
E  r   n  E r , , and H  r   n  H r ,  , (**)
where r, , and  are the usual spherical coordinates – see the figure on the right,
showing the axial cross-section of the structure. 0 
However, only the wave (*) can satisfy the coarse-grain boundary conditions
(7.104) at the cone surfaces. For its complex amplitudes, the four homogeneous r
Maxwell equations (7.2) written in the spherical coordinates,54 with  = 0, and  = 0,
are reduced to
1  rE  1  rH 
 i 0 H  0,   i 0 E  0, (***)
r r r r
1  E sin  
 0, 0  0, (****)
r sin  
the last (trivial) relation meaning that the Maxwell equation H = 0 is automatically satisfied for any
vector function H(r) = nH(r, ). The nontrivial relation of Eqs. (****) immediately says that e  Esin
may be a function of r alone. Since Eqs. (***) do not involve the polar angle , they may be identically
satisfied for all  only if H follows the same dependence on the polar angle, i.e. if

52 Note, however, that in contrast with the coaxial cable, (c)LC  0, and the corresponding fundamental mode is
not a TEM one – just as it should be for any waveguide with singly-connected walls.
53 The calculation of this attenuation is a simple additional exercise, highly recommended to the reader.
54 See, e.g., MA Eqs. (10.10) and (10.11) with / = 0.

Problems with Solutions Page 291

Essential Graduate Physics EM: Classical Electrodynamics

er  hr 
E r ,    , H r ,    .
sin  sin 
Now Eqs. (***), multiplied by r, may be rewritten as
d re  d rh 
 i0 rh   0,   i 0 re  0 .
dr dr
But this means that the products re  E(r,) rsin and rh  H(r,) rsin are related exactly as the
complex wave amplitudes in the usual plane waves in free space, i.e. have r-independent amplitudes,
may propagate along the radius r (in any direction) with velocity c  1/(00)1/2 and have an r- and -
independent ratio (7.8):
re E r ,  
1/ 2
 0 
  Z 0     const .
rh H r ,    0 
Let us check whether the wave impedance ZW of such a "transmission line", defined by the first
of Eqs. (7.115), depends on r. The complex amplitude V of the voltage between two points of the
opposite cones, located as the same distance r from the center of the system, may be calculated as
  0  0  /2   / 2   / 2
d d cos   d
V   E   dr   Erd  2re    2re    2re  
0 sin   0 1  cos   0 1  
2 2
 0 0
  / 2
 1 1  1  cos  0
 re      d  re  ln .
 0  1   1    1  cos  0

Similarly, let us calculate the complex amplitude I of the current flowing in each cone at a distance r
from the center, where the cone's cross-section radius is (r) = rsin0. Applying the general Eq. (6.38) to
the relation between the linear density of the cone’s surface current (flowing in the radial direction) and
the azimuthal magnetic field (*) near the surface, we get
I   2 J   2r sin  0 H r ,  0   2 rh  ,
so the impedance
V 1 re 1  cos 0 Z 0 1  cos 0
ZW   ln  ln
I  2 rh  1  cos 0 2 1  cos 0
is also independent of r, i.e. is a constant parameter of the conical structure. (Moreover, the expression
for ZW is close in structure to Eq. (7.120) of the lecture notes for the usual coaxial cable, in particular
featuring a similar weak (logarithmic) divergence at 0  0, i.e. at the vanishing cone’s "thickness"
c(r)/r = sin0.)
These relations are the basis for the design of the so-called conical antennas, which may provide
wideband coupling of free-space waves to small-size (lumped) sources and detectors of electromagnetic
radiation.

Problem 7.30. Use the recipe outlined in Sec. 7.7 of the lecture notes to prove the characteristic
equation (7.161) for the HE and EH waves in step-index optical fibers with a round cross-section.

Problems with Solutions Page 292

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Let us start with expressing the constant coefficient in Eq. (7.160) via the amplitude fl
defined by Eq. (7.156), from the requirement that at the core-to-cladding interface ( = R), both Ez and
Hz have to be continuous (as the components tangential to the interface), i.e. f+ = f–. This relation,
combined with Eq. (7.156) for f–, enables us to make Eq. (7.160) more specific:
J n (k t R)
f  fl K n ( t  ) cos n   0  .
K n ( t R)
Since we are now dealing with a linear superposition of two longitudinal fields, it is easier to operate
with the complex-exponential form of this expression and Eq. (7.156):
J n (k t R)  E , for f   E  ,
f  fl K n ( t  )e in , f   f l J n ( t  )e in , where f l   l
K n ( t R)  H l , for f   H  ,
where the constant phases 0 (which may be different for Ez and Hz) are incorporated into the complex
amplitudes fl.55 Now plugging these expressions into the first of Eqs. (7.121), and using the general
vector-algebra expressions for the cylindrical coordinates,56
f 1 f 1 f f
 t f   ,  t f   , n z   t f   , n z   t f   ,
     
we get the following formulas for the transverse components of the electric field:
i  in  in
E  
k t2 k z k t J n' (k t  ) El  k  Z   J n (k t  ) H l  e ,
 
i  in  in
E 
k t2 k z  J n (k t  ) El  k  Z  k t J n' (k t  ) H l  e ,
 
i J n (k t R)  in 
E     k z  t K n' ( t  ) El  k  Z  K n ( t  ) H l  e in ,
 t K n ( t R) 
2
 
i J n (k t R)  in 
E    k z K n ( t  ) El  k  Z   t K n' ( t  ) H l  e in .
 t K n ( t R)  
2

where the prime signs denote the differentiation of each Bessel function over its whole argument.
Now we should write the boundary conditions for the transverse components of the electric field:
  E     E  , E  E , at   R .

Requiring these conditions to be satisfied at all angles , we get two linear equations for the complex
constants El and Hl. At – = + = 0, the boundary conditions for the transverse components of the
magnetic field give an equivalent system. The same equality for  ensures that the ratio +/– may be
represented as k+2/k–2, and that k+Z+ = k–Z– = 0, so the system of two equations reduces to

55 This means that here we are working with the complex amplitudes of the fields, defined as in Eq. (7.98), for
brevity implying the index .
56 See, e.g., MA Eq. (10.2) with /z = 0, so the 3D operator  reduces to the 2D operator  .
t

Problems with Solutions Page 293

Essential Graduate Physics EM: Classical Electrodynamics

k2 J ' k2 K '  in k2 k2 

k z   n   n  El   0  2  2  H l  0,
 kt J n  t K n  R  kt  t 
in  1 1   1 J n' 1 K n' 
kz  2  2  El   0    H l  0,
R  kt  t   kt J n  t K n 
where the Bessel functions and their derivatives are taken at  = R. Now requiring the determinant of
this system to be zero, so the equations would be consistent, we indeed arrive at Eq. (7.161):
 k 2 J n' k 2 K n'  1 J n' 1 K n'  n 2  k 2 k 2  1 1 
      2  2  2  2  2  .
 k t J n  t K n  k t J n  t K n  R  k t  t  k t  t 

Problem 7.31. Derive an approximate equation describing spatial variations of the complex
amplitude of a general monochromatic paraxial beam propagating in a uniform medium, for the case
when these variations are sufficiently slow. Is the Gaussian beam described by Eq. (7.181) one of the
possible solutions of this equation? Give your interpretation of the last result.
Solution: As was discussed in 7.8 of the lecture notes (in the context of resonant cavities), for a
monochromatic wave of frequency , Eqs. (7.3) (similar for both fields, E and H) are reduced to the
same 3D Helmholtz equation (7.204) for their complex amplitudes R(r):57
2
 2

 k 2 R   0,   2 .
with k 2 
v2
Looking for its solution in the form R(r) = f(r)exp{ikz}, and assuming that the medium is uniform, so v
(and hence k) are spatially-independent, we get
 2  2 
  t  2ik  2  f  0, (*)
 z z 
where, as in Secs. 7.5-7.6, t is the del operator acting only in the directions normal to the z-axis.
So far, this is an exact equation. However, in many practical situations, especially in optics, the
function f changes slowly on the wavelength scale:
f  kf .

In a less formal language, this means that our wave propagates predominantly along the z-axis, with its
complex amplitude changing slowly on the wavelength’s scale. (The paraxial beams discussed in Sec.
7.7 is a typical but not the only possible example of such a situation.) In this case, the second derivative
2f/z2 in Eq. (*) is negligible in comparison with kfz, and this equation reduces to the so-called
paraxial (or “parabolic”) wave equation
 2 
  t  2ik  f  0, (**)
 z 
which is, for many problems, more convenient than the initial Helmholtz equation.58

57In the quasi-plane-wave situation we are going to discuss, the amplitudes of both fields are proportional to each
other, and hence to the same scalar function R(r).

Problems with Solutions Page 294

Essential Graduate Physics EM: Classical Electrodynamics

Now, plugging Eq. (7.181), which describes the Gaussian beam of width a,
 2 
f r   f 0 exp  2 , where  2  x 2  y 2 , (***)
 2 a 
into the paraxial wave equation, we may readily see that if a is a z-independent parameter, then Eq.
(***) is not a solution of Eq. (**). As will be discussed in Chapter 8, this is a result of the beam’s
diffraction, leading to its gradual broadening at distances z ~ ka2.59 The reason why we have received
Eq. (***) with a = const in Sec. 7.7 is that the radial -dependence of the dielectric constant in a graded
fiber, which is not described by the paraxial equation in its simplest form (**), may provide effective
beam focusing that continuously compensates for its diffraction.

z
Problem 7.32. Calculate the lowest resonance frequencies and the
corresponding field distributions of standing electromagnetic waves inside a l
round cylindrical cavity with well-conducting walls (see the figure on the
right), neglecting the skin depth s in comparison with l and R. 0
Solution: Due to the simple geometry of the system, and hence the R
simple structure of the fields in it, we can use the first approach to the
eigenfrequency calculation, described in Sec. 7.8 of the lecture notes, by employing the analysis of
circular metallic waveguides in Sec. 7.6. In particular, for the H-modes we may use Eqs. (7.121), and an
evident generalization of Eq. (7.144):
 ' 
H z  H l J n   nm  cos n   0  .
 R
Without calculating the transverse fields Et and Ht explicitly from Eq. (7.121) with Ez = 0, we may use
that formula to notice that Et has the same phase as Hz, which does not depend on the point’s position on
the cross-section z = const. Hence if a couple of well-conducting walls, normal to the z-axis, are inserted
into the waveguide at any of the distances z = p(z/2) = p/kz (where p = 1, 2, ..) between them, they do
not disturb the field distribution – besides turning the traveling wave into a standing one.
Now by requiring this z to be equal l, and using the general relation (7.122) in the form
2
k  2
2
 k t2  k z2 ,
v
and Eq. (7.143) for the eigenvalues of the transverse component kt of the wave vector, we get the
following eigenfrequency spectrum of the H-modes:
  ' 2
  p 
2 
 2
nmp  v k  v  nm
2 2 2
    . (*)
 R   l  

58 Conceptually, the parabolic equation is very close to the van der Pol approximation in the classical theory of
oscillations (called RWA in quantum mechanics) – see, e.g., CM Secs. 5.3-5.5 and QM Secs. 6.5 and 9.4.
59 Just for the reader’s reference: a gradually broadening Gaussian beam with the following z-dependent width: a
= [a02 + (z – z0)2/k2a02]1/2 (where a0 and z0 are arbitrary constants), and certain z-dependences of its amplitude and
phase (see Chapter 8 for details) does satisfy the paraxial equation (**).

Problems with Solutions Page 295

Essential Graduate Physics EM: Classical Electrodynamics

The sets of possible integer numbers m and p start with 1, so according to Table 7.1, the lowest of these
frequencies is
 ' 2   2   1.841  2    2 
2   11 
 011  v        v 2 
2
   . (**)
 R   l    R   l  
(As a reminder, v is the plane wave’s velocity in the dielectric filling of the resonator: v2 = 1/.)
However, this is not necessarily the lowest frequency of the resonator. Indeed, using an
absolutely similar analysis of the E-waves, whose longitudinal electric field is described by the natural
modification of Eq. (7.144) of the lecture notes:60
 
E z  El J n   nm  cos n   0  ,
 R
for the corresponding modes of the resonator, we get an expression very similar to Eq. (*):
  nm  2   p  2 
 nmp
2
 v 2     .
 R   l  
However, for the E modes, the boundary conditions at the lid surfaces (Et = 0, Hz/z = 0) allow
non-zero fields even if p = 0. (In this case, all components of the fields E or H are z-independent.)
Hence the lowest frequency of these modes is independent of l:
2 2
   2.405 
 2
010  v  01   v 2 
2
 . (***)
 R   R 
Note that the field distribution in this E010 mode is very similar to that in the fundamental mode of the
rectangular resonator – see Fig. 7.30, where b should be replaced with l, and a with 2R.
A direct comparison of Eqs. (**) and (***) shows that the latter frequency is lower, and hence
E010 with the eigenfrequency (***) is the fundamental mode of the resonator if l < R/(201 – ’211)1/2 
2.03R. In the opposite case, H011, with the eigenfrequency given by Eq. (**), is the fundamental mode.61
Just for the reader’s reference, let me note that the E-modes with p = 0 (i.e. with the fields
independent of the z-coordinate directed along the cylinder’s axis), and high indices n >> 1, and low
indices m ~ 1, are called the whispering gallery modes. This name62 is due to the fact that (as, for
example, Fig. 2.18 implies) the fields of such modes are localized mostly at the cavity’s walls (at   R) – just
as acoustic waves may be virtually localized at the walls of a round hall or gallery.

Problem 7.33. Analyze electromagnetic waves that may propagate inside a relatively narrow gap
between two well-conducting concentric spherical shells of radii R and R + d, in the limit d << R.

60 Note the change from the roots ’nm to the roots nm, due to the different relevant boundary conditions on the
side wall of the resonator (at  = R): for the E waves, Ez = 0, instead of the dHz/d = 0 for the H waves.
61 A similar crossover in the rectangular resonator, disguised by the similarity of field distribution along its sides,
takes place at l = min [a, b], i.e. essentially at the same condition as in the circular resonator, if we parallel its
diameter 2R with the smallest lateral size of the rectangular resonator.
62 It is due to Lord Rayleigh who was first (in 1878) to notice this effect – for human whispers in the circular
gallery of London’s St. Paul Cathedral.

Problems with Solutions Page 296

Essential Graduate Physics EM: Classical Electrodynamics

(i) Within the coarse-grain approximation, derive the 2D equation describing such waves with
relatively large wavelengths  ~ R >> d.
(ii) Calculate the lowest resonance frequencies of the system.
Solutions:
(i) On the scale of the distance d << R, both spherical surfaces are locally flat, so the differential
equation describing the wave propagation between them is the same as for the gap between two parallel
plane surfaces of good conductors. In the coarse-grain approximation, the wave’s fields at the exterior
sides of these conductors have to satisfy the boundary conditions (7.104):
E  0, Hn  0 .

In such a system, the requirement  >> d may be satisfied only by TEM waves with their electric field
normal to these surfaces and the magnetic field tangential to them, both independent of the coordinate
normal to the surfaces. For these fields, the derivatives 2/z2 vanish, so the general 3D Helmholtz
equation (7.204) for their complex amplitude R(r) is reduced to a 2D form:

 2
t 
 k 2 R ρ   0,

where  is the 2D radius vector in the plane of the conductors’ surfaces, and t is the del operator acting
only in this plane.
Now returning to the “global” (spherical) geometry of our system, in the usual spherical
coordinates, we may spell out this equation as63
 1     1 2 2
  sin    kR  R  ,    0 . (*)
 sin      sin   
2 2

(ii) We may apply, to this equation, the variable separation method as this was done with the
Laplace equation in Sec. 2.8, besides that now the solution has to be r-independent:
R  ,    c k P k cos  Fk   , (**)
k

where, so far, k is just a symbol standing for a certain set of eigenvalues. Now plugging this expansion
into Eq. (*), we see that the functions P and F satisfy, respectively, Eqs. (2.164) and (2.165). As was
discussed in Sec. 2.8, their eigenfunctions are, respectively, the associated Legendre functions Pln(cos)
and sinusoidal functions F(). Obviously, all solutions of our current problem, and hence all functions
F(), have to be 2-periodic in the azimuthal angle , so the index  has to be an integer. As a result,
taking into account Eq. (2.178), Eq. (**) may be spelled out as
 l
R  ,    cl ,nP l n cos  Fn   .
l 0 n 0

The partial products in its right-hand part are the real spherical harmonics Yln(, ). The only property of
these harmonics needed for this solution is that the eigenvalues of Eq. (*) do not depend on the

63 See, e.g., MA Eq. (10.9) with r = R = const.

Problems with Solutions Page 297

Essential Graduate Physics EM: Classical Electrodynamics

azimuthal index n.64 So, calculating them by plugging the partial solutions PlnFn into this equation, we
may take n = 0, and use the Legendre equation (2.168) for the functions Pl0  Pl. The result is very
simple:

 l l  1  kR  R  0 ,
2

so we may label the eigenvalues of the wave number k with the index l alone:
l l  1
kl  , with l  0,1, 2,...
R
From here and Eq. (7.28), the lowest resonant frequencies of the system are
l l  1
 l  vl k l  vl where vl    l   l  .
1/ 2
,
R
Note that the (mathematically acceptable) eigenvalue l = 0 gives 0 = 0 and describes not a wave but
just a possible stationary electric field between the spherical surfaces.
This system, with vl = c  3.0108 m/s and R  6.4106 m, may be used as a model for the so-
called Schumann resonances in the layer between the Earth’s surface and the lower boundary of its
ionosphere. Its frequency spectrum l may be experimentally measured, for example, by observing
resonant peaks in the spectral density of the random electromagnetic noise generated by lightning
strikes. Very surprisingly for such a crude model, which completely ignores the Earth’s surface terrain
and (even more importantly) local variations of the ionosphere, it overestimates the lowest resonance
frequencies by only ~20%. The lowest and the strongest of the observed resonances, for l = 1, is at the
cyclic frequency 1/2  8 Hz.
Finally, note a very similar CM Problem 8.14 on water waves on the Earth’s ocean surface.

Problem 7.34. A molecule with an electric polarizability  is placed inside an otherwise empty
macroscopic cavity with well-conducting walls. Express the resulting shifts of its resonance frequencies
via the unperturbed field distribution in the corresponding mode.
Hint: You may like to use the adiabatic theorem of classical mechanics in application to a
harmonic oscillator.65
Solution: Since the effect of a single molecule on a macroscopic cavity is very small, and its size
is much smaller than that of the cavity, we may calculate the average energy of its interaction with the
electric field from Eqs. (3.15b) and (3.48) of the lecture notes,
1 
U int   p  Er, t    E 2 r, t  ,
2 2
by using the unperturbed value of the field at the point r of the molecule’s location. At free oscillations
of a jth resonant mode, the field changes sinusoidally, so the time averaging yields

U int   E 2j r  ,
4

64 This fact follows from the discussion in Sec. 2.8 of the lecture notes but, admittedly, was not emphasized there.
65 See, e.g., CM Sec. 10.2.

Problems with Solutions Page 298

Essential Graduate Physics EM: Classical Electrodynamics

where Ej is the real amplitude of the field.

On the other hand, the unperturbed energy of the field in the cavity may be calculated from Eq.
(6.113) with E = Ej and B = 0, because the magnetic field of any standing wave vanishes at the moments
when the electric field takes its maximum (amplitude) value:
0
E r  d
2 
U fielsd  2
j
3
r.
V

(In the course-grain approximation, valid when the skin depth s is negligible in comparison with the
cavity’s linear dimensions, V may be taken equal to just its internal volume.) Hence, the relative change
of the total field’s energy due to the molecule’s insertion may be calculated as
U j U  E 2j r 
 int   .
Uj U fielsd 2 0  E j r  d r
2 3

V
However, according to the adiabatic theorem of classical mechanics (where the word
“mechanics” is understood in its broad sense),66 for any harmonic oscillator, this relative change has to
be equal to that of the oscillation frequency, so
 j U j  E 2j r 
  .
j Uj 2 0  E 2j r  d 3 r
V

Problem 7.35. A plane monochromatic wave propagates through a medium with an Ohmic
conductivity  and negligible electric and magnetic polarization effects. Calculate the wave’s
attenuation and relate the result to a certain calculation carried out in Chapter 6 of the lecture notes.
Solution: In a plane wave, the only energy losses are those in the propagation medium, so the
attenuation coefficient is completely described by Eq. (7.218) of the lecture notes:

  2 Im 1 / 2   1 / 2  . (*)

For the medium described in the problem’s assignment, Eq. (7.46) takes the form

     0  i ,

so Eq. (*), with () = 0, yields
1/ 2
     
1/ 2 1/ 2
 
1/ 2
 i  1 
  2 Im   0  i   01 / 2   2 Im 1    2 1  2 2   1 ,
 (**)
   c   r  c    r 
 
  
where r  0/ is the charge relaxation time defined by Eq. (4.10), for the particular case  = 1.
This result may be simplified in two limiting cases. At relatively high frequencies (or at
relatively small conductivity), the formula is reduced to

66 Note that according to the Bohr-Wilson-Sommerfeld quantization rules (see, e.g., QM Sec. 2.4), the same result
is valid for a broad range of quantum systems as well.

Problems with Solutions Page 299

Essential Graduate Physics EM: Classical Electrodynamics

 1 1 
 2    Z 0  k 0 , for  r  1, i.e. for   k 0 ,
c 2  
2 2 1/ 2
r
c r c 0
where Z0  (0/0)1/2 and k0 = /c are, respectively, the wave impedance and the wave number in free
space. In this case, the wave decays at a distance (ld  1/) much longer than its wavelength 0  2/k0.
In the opposite limit of low frequencies (or high conductivity), the result yields
1/ 2
 1    
 2  2 0   k 0 , for  r  1, i.e.   k 0 .
c  r 1 / 2  2 
In its reciprocal ld  2/, i.e. the wave decay length, we may readily recognize the skin depth s (6.33),
which was derived in Sec. 6.3 without the account of the displacement currents. Indeed, in this limit, the
decay length ld is much smaller than the free-space wavelength 0 at this frequency, so the quasistatic
approximation is fully justified – see Sec. 6.8. Note that in this limit, it is hardly possible to speak about
the wave propagation: upon entering the medium, the wave decays almost immediately – physically,
because the high Ohmic conductivity results in very high Joule losses (4.39).

Problem 7.36. Generalize the telegrapher’s equations (7.110)-(7.111) by accounting for small
energy losses:
(i) in the transmission line’s conductors, and
(ii) in the medium separating the conductors,
using their simplest (Ohmic) models. Formulate the conditions of validity of the resulting equations.
Solutions:
(i) Physically, Eq. (7.111) is just a balance between Faraday’s e.m.f. induced on a unit-length
segment of the transmission line (described by the right-hand side of that equation) and its drop on the
inductance of that segment. If the line’s conductors have the (total) resistance R0 per unit length, it
makes an additional contribution R0I to the balance, so the equation becomes
I V
L0   R0 I . (*)
t z
(ii) Similarly, Eq. (7.110) has a clear physical sense of the charge conservation law: its left-hand
side is the rate of change of the charge at the capacitance of a unit length of the transmission line, while
the right-hand side is the balance of the currents flowing in and out of such a unit-length segment. A
nonvanishing Ohmic conductance G0 (per unit length) of the media separating the wires evidently brings
into this balance an additional leakage current G0V, so the equation takes the following form:
V I
C0    G0V . (**)
t z
The correctness of the signs of the additional terms in these generalized telegrapher’s equations
may be double-checked from the natural requirement that if R0 and G0 are positive (dissipative), they
lead to the wave’s attenuation. In order to calculate the attenuation, we may plug into Eqs. (*) and (**)
the standard monochromatic traveling-wave solution I = Re[I ei(kz – t)], V = Re[V ei(kz – t)], getting the
following system of two linear equations for their complex amplitudes:

Problems with Solutions Page 300

Essential Graduate Physics EM: Classical Electrodynamics

 iL0 I   ikV  R0 I  ,  iC 0V  ikI   G0V

– cf. Eqs. (7.113) of the lecture notes. The condition of their consistency,
 iL0  R0 ik
0,
ik  i C 0  G 0

immediately yields the dispersion relation

k  L0  iR0 C 0  iG0  ,
1/ 2

which is a generalization of Eq. (7.114) to the case of nonvanishing Ohmic losses. In particular, in the
most important case when the losses are relatively low (Re k  k’ >> k”  Im k),67 the relation reduces to
1 R 
with k'   L0 C0  , k"   0  G0 Z W  ,
1/ 2
k  k'  ik" ,
2  ZW 
1/2
where ZW  (L0/C0) is the lossless transmission line’s impedance (7.115). This result shows that,
indeed, positive R0 and G0 do give positive contributions to the attenuation constant (7.216),   2k”.
Since the above derivation of Eqs. (*)-(**) was based on such global (or “instant”)
characteristics of the transmission line as R0 and G0, the necessary condition of validity of these
equations is the smallness of the linear scale a of the transmission line’s cross-section in comparison
with wave’s decay length:
1 Z 1 
a  l d   min  W , .
2k"  R0 G 0 Z W 

Problem 7.37. Calculate the skin-effect contribution to the attenuation constant  of a TEM wave
in the microstrip line discussed in Problem 21 (ii).
Solution: As was discussed in the model solution of Problem 21 (ii), at d, s w  d
<< w, both the electric and the magnetic field within the gap between the
conductors of the microstrip line (see the figure on the right) are uniform while
being negligibly small everywhere outside the gap, besides small fringe regions of d
width ~ d, s near the strip’s edges. In addition, if the skin depth is much smaller
than d as well,68 the fields are localized within the area wd of the gap’s cross-
section, so the transmitted power may be calculated directly from Eq. (7.9) of the lecture notes, as
1/ 2

P  Swd    H 2 wd .
 
On the other hand, the wave power loss at an elementary length dz of the line, due to the skin
effect, may be calculated by multiplying the result given by Eq. (6.36) by 2wdz – the area of contact of
the magnetic field and the two conductor surfaces:

67If they are not, the wave decays so fast that we cannot really speak about its propagation in the line.
68 According to Eq. (7.78), which may be used for crude estimates of the skin effect losses in any skin-effect
system, this relation is necessary to keep the attenuation low,  << k, which is, in turn, the necessary condition of
neglecting the losses at the forthcoming calculation of H and P.

Problems with Solutions Page 301

Essential Graduate Physics EM: Classical Electrodynamics

 s
dPloss  H 2
2 wdz .
4
Plugging these expressions into the definition (7.214) of the attenuation constant, we get a result
independent of the microstrip’s width w:
dPloss / dz skin  s k s
 skin    
1/ 2
 .
P 2d 2d
This result may be compared with Eq. (7.225) for another TEM transmission line, the coaxial
cable. It also confirms that in order to keep the attenuation low ( << k), the skip depth s has to be much
smaller than the gap width d.

Problem 7.38. Calculate the skin-effect contribution to the attenuation coefficient  defined by
Eq. (7.214) of the lecture notes, for the fundamental (H10) mode propagating in a waveguide well-
conducting walls, of a rectangular cross-section – see Fig. 7.22. Use the results to evaluate the wave
decay length ld  1/ of a 10 GHz wave in the standard X-band waveguide WR-90 (with copper walls, a
= 23 mm, b = 10 mm, and no dielectric filling), at room temperature. Compare the result with that
(obtained in Sec. 7.9 of the lecture notes) for the standard TV coaxial cable, at the same frequency.
Solution: As was discussed in Sec. 7.6 of the lecture notes, in the H10 wave, the electric field has
just one Cartesian component, with the complex amplitude
ka x
E y ( x)  i ZH l sin ,
 a
while its magnetic field has two components, with the following complex amplitudes:
x kza x
H z ( x)  H l cos , H x ( x)  i H l sin , (*)
a  a
all in the notations of Fig. 7.22. Of those two components, only Hx contributes to the longitudinal (z-)
component of the time-averaged Poynting vector
E x H *y  E y H *x kk z a 2 2 x
Sz   Z H l sin 2 ,
2 2 2
a
which gives the following total power flow along the waveguide:
a b
kk z a 3 b
P   dx  dy S z 
2
Z Hl , (**)
0 0 4 2

giving us the denominator on the right-hand side of Eq. (7.214).

In order to calculate the numerator of that fraction, i.e. the power dissipation per unit length, we
may use Eqs. (*) to integrate the skin-effect losses per unit area, given by Eq. (7.219), over the cross-
section’s perimeter:
dPloss  0 s  0 s 
  
a b b

H 2  H x ( x)  H z ( x) dx   H z (0) dy   H z (b) dy 
2 2 2 2 2
 0 ( x) dl 
dz 4 C
4  0 0 0 

Problems with Solutions Page 302

Essential Graduate Physics EM: Classical Electrodynamics

 0 s  a  k a  2 x      k a  2  
2 x
2  
2 2
 Hl z
 sin  cos 2  dx  2b   0 s H l 
z
  1 a  2b .
4  0    a a   4     
Per the wave’s dispersion relation given by Eqs. (7.122) and (7.133),

k z2  k t2  k 2   2  , with k t 10 
,
a
the expression in the last square brackets is just (ka/)2, so using Eq. (**) for P, we finally get

 s  ka  2b   ka  2 2b 
2
1 dPloss s
  2
          .
P dz Zkk z a 2 b    a  (k z a /  )ab    a 

Note that  scales approximately as s/A, where A  ab is the waveguide’s cross-section area.
More particularly,  diverges at b  0, because the transmitted power vanishes while the losses in the
broader walls of the waveguide remain constant (at a fixed field amplitude). Another interesting fact is
that the dependence of the attenuation on frequency is non-monotonic:  diverges at   c where kz
 0, but also grows (as ks  1/2) at    where kz  k >> a–1, b–1. As a result, the lowest
attenuation is reached at a frequency ~30% above the threshold, i.e. when the higher modes (H11, H20,
and E11) still cannot propagate in the waveguide.
For a 10 GHz wave in the WR-90 waveguide we get  = 0  30 mm, ka/  2a/  1.53, kza/
= [(ka/)2 – 1]1/2  1.16, while for copper at room temperature and that frequency, s  6.510–7 m; as a
result, our final formula yields   0.025 m-1 (i.e. ~0.1 dB/m), i.e. ld  1/  40 m. Hence the waveguide
may provide an attenuation well below that (  0.16 m-1, see Sec. 7.9) of the standard TV coaxial cable
RG-6/U at the same frequency. (Admittedly, the difference is mostly due to the waveguide’s larger
cross-section rather than to its different wave mode.)

Problem 7.39. Calculate the skin-effect contribution to the attenuation coefficient  of

(i) the fundamental (H11) wave, and
(ii) the H01 wave,
in a conductor-wall waveguide with the circular cross-section (see Fig. 7.23a of the lecture notes), and
analyze the low-frequency (  c) and high-frequency ( >> c) behaviors of  for each of these
modes.
Solutions:
(i) For the H11 mode, the longitudinal component Hz of the magnetic field (or rather its complex
amplitude) is described by Eq. (7.144) of the lecture notes:
 
H z  ,    H l J 1  11'  cos  , with 11'  1.841 ,
 R
where the constant 0, which does not affect  due to the axial symmetry of the system, was taken for 0.
Plugging this result into the second of Eqs. (7.121) with Ez = 0, and using the general vector-algebra
equalities69

69 See, e.g., MA Eq. (10.2) with /z = 0 and/or the model solution of Problem 30.

Problems with Solutions Page 303

Essential Graduate Physics EM: Classical Electrodynamics

f 1 f 1 f f
 t f   ,  t f   , n z   t f   , n z   t f   ,
     

we get the following formulas for transverse components of the electric and magnetic fields:

ikZ 1   ikZ  
E   ,    H l J 1  11'  sin  , E  ,    H l J 1'  11'  cos  ,
kt2
  R kt  R
k   ik 1  
H   ,    i z H l J 1'  11'  cos  , H   ,     2z H l J 1  11'  sin  ,
kt  R kt   R
where k  ()1/2, Z  (/)1/2, and kt = ’11/R, while kz should be found from the general Eq. (7.102):

k z  k 2  k t2  .
1/ 2

From here and Eq. (7.219), the time-averaged areal density of the energy loss rate due to the skin
effect is
dPloss  s
dA

4

H   R,    H z R,   
2 2  s
4
2 2

'  kz

2

 kt R

H l J 1 11   4 2 sin 2   cos 2   .

Integrating this expression over the cross-section’s perimeter, we can find the full power loss per unit
length of the waveguide:
dPloss 2 dPloss   1 k z2 
  Rd  J 12 11'   s R H l
2
  1.
 k t R  k t
2 2
dz 0
dA 4 
By the definition of ’11, J1(’11) is just the first maximum of Bessel function J1(), which is close to
0.5819 (see, e.g., Fig. 2.18), while (ktR)2 = (’11)2  3.389.
The average longitudinal component of the Poynting vector is
E  H *  E H * Z kk  1     
Sz   H l2 2z  2 2 J 12  11'  sin 2   J 1' 2  11'  cos 2   .
2 2 kt  kt   R  R 
Integrating it over the waveguide’s cross-section, for the average propagating power we get
2 R
kk z R  1 2 '   ' 2  '   I
P   d   d S z  Z H 2   J 1  l
2
11   J 1  11   d  ' 4 Z H l kk z R ,
2 4

0 0
2 kt 0  kt 
2 2
 R  R  11
where I is a numerical constant:
11' '
 J 2    1 11

I    1 2  J 1'2    d   J 02    J 22    d  0.5822 . 
0    2 0

From here, taking into account that /kZ = 1, the attenuation constant is


1 dPloss J 12 11' 11' 2  s

   k z2  
 2  11' 2   0.4929 s 3
 k z2 
 2  3.389  .
P dz 4I kz R3 k  kz R k 
 t   t 

Problems with Solutions Page 304

Essential Graduate Physics EM: Classical Electrodynamics

The attenuation’s dependence on the wave frequency , following from this formula, is very
similar to that for the rectangular waveguide (see the solution of the previous problem):  diverges
when  is reduced to the cutoff frequency c = kt/()1/2 because at that point kz  0, and also increases
as ks  1/2 at high frequencies  >> c, where kz  k >> kt. (The minimum of  is reached between
these two extremes, at   4c.)
(ii) For the H01 mode, the longitudinal field is described by Eqs. (7.141) and (7.143) with n = 0
and m = 1:
 ' 
H z  ,    H l J 0   01 ,  01'  3.831 ,
 R
i.e. is independent of the azimuthal angle. Because of that independence, the field calculations,
absolutely similar to those carried out in Task (i), yield simpler results:
k  '  k  ' 
E   ,    0, E  ,    i 2
ZH l J 0'   01   i ZH l J 1   01  ,
k  R kt  R
kz  '  k  ' 
H   ,    i H l J 0'   01   i z H l J 1   01 , H   ,    0 .
kt  R kt  R
where kt = ’01/R. As a result of the vanishing azimuthal magnetic field, the skin-effect losses are
determined only by the longitudinal component of H:
2
dPloss  s  s 
H l J 02  01 ,  R  loss d  J 02  01 
dPloss P
H z R,   
2 2 2
 '
so '
 s R H l .
dA 4 4 dz 0
A 2
The longitudinal component of the Poynting vector is also contributed by only one component product:
E H * Z 2 kk z 2  '  
Sz    H l 2 J 1   01  ,
2 2 kt  R
so the total average power flow is
'
2  01
R
kk z R 2  '   kk
P   d   d S z  Z H 2  J 1   01   d   I'Z H l2 2z R 2 , with I'   J   d  0.6384 .
2 2
l 1
0 0 kt 0  R kt 0

As a result, the expression for the attenuation constant,

1 Ploss J 02  01
'
  s k t2 R  1.865  s ,
 
P dz 2 I' kz kz R3
has a rather different frequency dependence: while still diverging at   c (where kz  0), it is
proportional to s/k   –1/2  0 at /c  , where kz  k >> kt.
This property, which makes the H01 mode (as well as higher H0m modes) attractive for long-
distance microwave energy transfer, is due to the fact that the transverse component of the magnetic
field vanishes at the wall surface ( = R). As a result, the skin-effect losses are due only to the
longitudinal component of the magnetic field, which decreases (at a fixed power of the wave) as its
frequency is increased.

Problems with Solutions Page 305

Essential Graduate Physics EM: Classical Electrodynamics

Problem 7.40. For a rectangular cavity of dimensions abl, with b  a, l, calculate the Q-factor
of the fundamental oscillation mode, due to the skin-effect losses in its conducting walls. Evaluate the
factor for a 232310 mm3 cavity with copper walls, at room
temperature. y
b
Solution: Selecting the Cartesian coordinates as in Fig. 7.29
of the lecture notes (partly reproduced on the right) and using Eq. x
(7.137) with kz = l/, for the fundamental (H101) mode discussed in
0 a
Sec. 7.8, we may write the magnetic field distribution as follows:
z
a x z x z l
H x  H l sin cos , H y  0, H z  H l cos sin .
l a l a l
The electromagnetic field energy in the cavity may be calculated, for example, as the largest energy of
the magnetic field:
0 a b l
0 a l
a2 x z x z 
 dx  dy  dz H x  H z   H l b  dx  dz  2 sin 2 cos 2
2
U 2 2
 cos 2 sin 2 
2 0 0 0
2 0 0 l a l a l 
0 2  a2 
 H l abl  2  1 ,
8 l 
where it assumed that the wall’s conductivity is sufficiently high to have s << a, b, l, In this case, the
time-averaged power losses due to the skin effect may be calculated just as it was for a waveguide in the
model solution of Problem 38, i.e. by integrating Eq. (7.219) over the area of all the walls:

 0 s  a 
 
l a l

2b   H x ( x,0) dx   H z (0, z ) dz   2 dx  dz H x ( x, z )  H z ( x, z )

2 2 2 2
Ploss 
4   0 0  0 0 
 0 s 2  a  a  a2 
3
 H l l b  3  1   2  1 .
4  l  2 l 
As a result, the Q-factor defined by Eq. (7.227) is

U 1   a2    a3  a  a2 
Q   ab  2  1 b  3  1   2  1 .
Ploss 2 s   l   l  2 l 
As could be expected from Eq. (7.78), Q scales as the ratio of some effective size of the cavity
(at l << a, b ~ , tending to l), to the skin depth s. According to Eq. (7.206) of the lecture notes, the
resonance frequency of the cavity specified in the assignment is   5.71010 s-1 (f  9.2 GHz), so the
skin depth s for copper walls is close to 6.810–7 m, and the above formula yields Q  7.9103. For a
microwave metallic cavity of a practicable size, operating at its fundamental mode, this is almost as high
a quality factor as you can get at room temperature. Note that for these numbers, the condition Q >> 1 is
well satisfied, so our approximate method of calculation of the Q-factor is indeed legitimate.

z
* r
Problem 7.41. Calculate the lowest eigenfrequency and the Q-
factor (due to the skin effect) of the axially symmetric toroidal cavity with d
0 R

Problems with Solutions Page 306

Essential Graduate Physics EM: Classical Electrodynamics

well-conducting walls and the interior’s cross-section shown in the figure on the right, for the case d <<
r, R. 70
Solution: Looking at the field distribution in the fundamental, H101 mode of a rectangular cavity
(Fig. 7.30 of the lecture notes), and the virtually similar distribution in the E010 mode of the circular
cylindrical cavity (see the model solution of Problem 32), it is easy to understand that the lowest mode
of the toroidal cavity has the electric field concentrated almost exclusively in its central part, with the
distance d between the parallel planar walls, while the magnetic field is virtually limited to the toroidal
(“doughnut”) part of radii r and R. Due to this field separation, the resonance frequency may be
calculated just as for the lumped LC circuit,
1
 ,
LC 1 / 2
with the capacitance equal to that of the plane capacitor of thickness d and area A = (R – r)2,
  R  r 2
C  ,
d
and the inductance calculated as in Problem 5.20, but with N = 1:

 
L   R  R 2  r 2 )1 / 2 , 
so, finally,
d
2 

 R  r  R  R 2  r 2 
2 1/ 2
. (*)

The TEM wavelength corresponding to the frequency is


2

2 R  r  
 2 3 / 2 1 / 2 R  R 2  r 2 )1 / 2  .
1/ 2

k   
1/ 2
d
This formula shows that at r ~ R, the wavelength scales as (R3/d)1/2>>R,r>>d, so our quasistationary
treatment of the electric and magnetic fields in indeed valid.
To calculate the Q factor, we need to evaluate both the numerator and denominator of the right-
hand side of Eq. (7.227) of the lecture notes. Assuming that Q >> 1, this may be done neglecting the
effect of losses on the field distribution. In this approximation, the average energy of oscillations may be
found as the maximum energy of the electric field:
2 2
Q Q d
U  , (**)
2 R  r 
2
2C
where Q is the complex amplitude of the total electric charge of each planar “lid” of the cavity.
In order to calculate the energy loss rate due to the skin effect, we can use Eq. (7.219). With the
account of the fundamental Eq. (6.38), it may be rewritten as

70 Such resonators are broadly used in particle accelerators and also in vacuum electron devices for high-power
microwave amplification and generation (e.g., the so-called klystrons), where the electric field has to be
concentrated in the region of charged particle passage – typically, along the symmetry axis (in the figure above,
the z-axis), through a pair of small holes in the cavity’s walls, which do not affect the field distribution
substantially.

Problems with Solutions Page 307

Essential Graduate Physics EM: Classical Electrodynamics

dPloss  s
J    ,
2
 (***)
dA 4
where J() is the complex amplitude of the linear density of the surface currents at the point of the
cavity wall surface, that is separated from the z-axis by distance . Let us express this density via its
value J0  J (R – r) in the cavity’s “neck”, i.e. at the connection of its planar and toroidal parts, because
J0 may be readily related to Q by using the charge conservation law. With the account of the sinusoidal
law of time evolution of all the variables (/t  –i), this relation yields
1
 iQ  2 R  r J 0 , i.e. J 0  i Q . (****)
2 R  r 
The current density in the toroidal part of the cavity, with negligible displacement currents, may
be calculated from the conservation of the total surface current, 2J() = 2 (R – r)J0, giving
Rr
J  ( )  J 0 , for   R  r .

However, in the planar part of the device (with  < R – r), the displacement currents are substantial,
because of the smallness of d. The easiest way to calculate J here is to write the charge conservation
law similar to that written above for J0, –idQ()/dt = –2J(), where Q () = Q [/(R – r)]2 is the
part of the full charge Q, that is located within a circle of radius .71 As a result, we get
2
  
J  ( )  J 0   , for   R  r .
Rr
As the sanity check, J (R – r) = J0, and J (0) = 0.
Now we have everything ready for the energy loss calculation using Eq. (***):

 s  s  Rr    4 

Rr
2

 J    2    2 d  2  
2 2
Ploss  d r
2
J0  2 rd  ,
4 over all 4  0  R  r  0
  
surfaces
where  is the angle pointing to the considered point of the torus from the center of its cross-section, so
 = R + r cos. An easy integration72 yields
 s  2  R  r 
2

 R  r   4 r 2
2 2
Ploss  J0 2
.
4  3 R  r 2 1 / 2 
Finally, combining this expression with Eqs. (*), (**), and (****), we get

Q 
U


R  R2  r2 
1/ 2
 1

r 
.
 2 1/ 2 
Ploss s 12 2( R  r ) 
2

This formula shows that in the quasistatic approximation, the Q-factor does not depend on d and
that at r ~ R,73 it scales as ~R/s, i.e. is much lower than the value ~/s that we could get from the

71 This particular calculation is similar to the one carried out in the model solution of Problem 6.30.
72 In the second case, by using MA Eq. (6.3c).

Problems with Solutions Page 308

Essential Graduate Physics EM: Classical Electrodynamics

“usual” resonance cavity, with all three linear dimensions comparable. This is the price we have to pay
for the system’s convenience for particle-beam applications. This result explains why in such unique
applications as particle accelerators where refrigeration costs are not the primary concern, accelerating
rf cavities are fabricated from a superconducting material (usually, niobium with the critical temperature
of 9.2 K) and are kept at “helium” temperatures of a few K, to keep Q sufficiently high.

Problem 7.42. Express the contribution to the damping coefficient (the reciprocal Q-factor) of a
resonant cavity by small energy losses in the dielectric that fills it, via the complex functions () and
() of the material.
Solution: For the dispersion-free and loss-free case, Eq. (7.202) of the lecture notes has the
partial variable-separated solutions (7.203) whose temporal factors obey the ordinary differential
equation similar to the second of Eqs. (7.201):
k2
T   2T  0, where   k v 
2 2 2
, (*)

and k is the corresponding eigenvalue of the Helmholtz equation (7.204). For a lossless medium,  and 
are real, so T describes persisting sinusoidal oscillations of frequency :

 
T (t )  Re T  e it  Re T  cos t  ImT  sin t .

Acting in analogy with the derivation of Eq. (7.28), we may generalize Eq. (*) to the case of a
dispersive medium as
k2
2  . (**)
    
If the dielectric is lossy, then () = ’() + i”(), and () = ’() + i”(),74 and the solution of
this equation is complex,  = ’ + i”. This means that the solution of Eq. (**) is not exactly
sinusoidal, but decays with time:

    
T (t )  Re T  e it  Re T  e i (' i" )t  e"t Re T  e i't , 
so the oscillations’ energy decays as
U (t )  U (0)e 2"t .
Comparing this expression with Eq. (7.228), we see that the contribution of the complex  to the
damping coefficient 1/Q is
1 2"
   .
 Q  dielectric '

73 For fixed R and s, Q reaches its maximum at r  0.8R, while vanishing at both r  0 and r  R. This fact
justifies our efforts to carry out the calculations for arbitrary ratio r/R.
74 For most dielectrics, () is very close to  , so ”() is negligible, and in this solution, that component is kept
0
mostly for the sake of generality.

Problems with Solutions Page 309

Essential Graduate Physics EM: Classical Electrodynamics

In the most important case when the losses are small,  ”()  << ’() and  ”()  << ’(), so
” << ’, and the characteristic equation (**) rewritten as

'  i" 2 '  i" '  i"   k 2 , (***)

may be solved by successive approximations. In the 0th approximation, ” = 0, while ’ is just the
unperturbed real frequency 0 of the cavity, which has to be found self-consistently from the equation
02 = k2/’(0)’(0). In the 1st approximation, we may plug this result, ’ = 0, into Eq. (***)
linearized with respect to small ”, ”, and ”:

2'"''  ' 2 '"  "'   0 .

Its solution is
' '"  "'  '  " " 
"       ,
2'' 2  ' ' 
so, finally, we get a very simple and natural expression:
1 " " Im  ( 0 ) Im  ( 0 )
      .
 Q  dielectric ' ' Re  ( 0 ) Re  ( 0 )

Problem 7.43. For the dielectric Fabry-Pérot resonator (Fig. 7.31 of the lecture notes) with the
normal wave incidence, calculate the Q-factor due to radiation losses, in the limit of a strong impedance
mismatch (Z >> Z0), by using two approaches:
(i) from the energy balance, by using Eq. (7.227), and
(ii) from the frequency dependence of the power transmission coefficient, by using Eq. (7.229).
Compare the results.
Solutions:
(i) The calculation of the field distribution in such a system, 1 T T0
with a unit-amplitude external wave incident from one side, was the
subject of Problem 13. The last two equations of the full system for
the traveling wave amplitudes T, T0, R, and R0, normalized to the 0 d z
incident wave amplitude E (for notation, see the figure on the R0 R
right), may be readily solved to give
Z0  Z Z0  Z
Te ikd  T0 , Re ikd  T0 .
2Z 0 2Z 0
At the exact resonance (kd = kmd = m, where m = 1, 2, 3, …), both exponential factors are equal
m
to (–1) , and from the solution of Problem 13, we know that T0 = 1, so at Z >> Z0, R  –T, with
2
2 2  Z 
T  R     1 .
 2Z 0 
Thus the field inside the resonator (0  z  d) may be approximated with a standing wave:

Problems with Solutions Page 310

Essential Graduate Physics EM: Classical Electrodynamics


E ( z )  E Te ikz  R e ikz   E  Z
2Z 0

e ikz  e ikz  iE
Z
Z0
sin 
mz
d
.

Note that the standing wave’s amplitude is much higher than that (E) of the incident wave. (This is the
well-known effect of energy accumulation at resonance.). Also notable is the /2 phase shift between the
standing wave and the incident wave (represented by the coefficient –i  exp{–i/2}), which is also
typical for the exact resonance.75
Now we can use Eq. (3.79) to calculate the full (time-independent) oscillation energy as the
time-maximum electric field energy in the resonator (per unit area):
2d 2
U    Z  mz   Z  d
 
d
  max t E 2 ( z , t ) dz  E  sin
2 2
  2
dz  E   .
A 20 2  Z0  0
d 2  Z0  2
The resonator’s energy loss (per unit time and unit area), due to the radiation through its
interfaces, is the sum of the time-averaged Poynting vectors of the two outcoming waves, whose
amplitudes T0 and R0 are (in our approximation) equal to each other:
Ploss 1 2
 2S T  T0 E ,
A Z0
so at the resonance (T0 = 1), Eq. (7.227) yields
2 2
U d  Z  k d  Z  m Z
Q  m  m Z 0    m Z 0    . (*)
Ploss 4  Z0  4Z  Z0  4 Z0

This formula is only valid if Q >> 1, as it is in our case Z >> Z0.

(ii) The final result of the solution of Problem 13 was the following expression for the
(amplitude) transmission coefficient:
4 ZZ 0 N
T0  exp{i (k  k 0 )d }  ,
( Z  Z 0 )  ( Z  Z 0 ) exp{2ikd }
2 2
D
whose denominator D (a complex number) may be conveniently represented by the 2D vector diagrams
reproduced in the figure below. (The top-right panel of that figure confirms, in particular, that the
resonance transmission (T0 = 1) is reached at kmd = m.)
Im Im
kd  k m d kd  k m d  m

D D
2kd (Z  Z 0 ) 2

Re 0 Re
(Z  Z 0 ) 2 (Z  Z 0 ) 2 (Z  Z 0 ) 2
.

75 See, e.g., CM Sec. 5.1, in particular, the right panel of Fig. 5.1.

Problems with Solutions Page 311

Essential Graduate Physics EM: Classical Electrodynamics

Im
kd  k m d

(Z  Z 0 ) 2 D
2(k )d
0 (Z  Z 0 ) 2 Re

The diagrams indicate that at a strong impedance mismatch (Z >> Z0), even a small deviation k
 k – km from the exact resonance leads to a fast growth of the denominator D and hence to a fast
suppression of T. As a result, in order to find the power transmission coefficient T2 near the
resonance, we may use the following approximation (see the bottom panel of the figure above):

D
2

 (Z  Z 0 ) 2  (Z  Z 0 ) 2   (Z  Z
2
0 ) 2 2(k )d   4ZZ   Z
2
0
2 2
2
2(k )d .

The denominator increases to twice its resonance value (and hence the transmitted power drops to one-
half of its resonance value) when the second term in this expression becomes equal to the first one, i.e.
when
2Z
(k ) 1 / 2 d   0 .
Z
According to Eq. (7.229),76 the Q factor of the resonator may be determined from the “full-width
half-maximum” (FWHM) bandwidth:
m k d m  mZ
Q  m   ,
 (k )d 2 k 1 / 2 d 4Z 0

i.e. exactly the same result as given by Eq. (*).

76 Again, the (quite simple) derivation of that formula may be found, for example, in CM Sec. 5.1.

Problems with Solutions Page 312

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 8. Radiation, Scattering, Interference, and Diffraction

Problem 8.1. Equation (8.8) of the lecture notes obviously has standing-wave solutions (r, t) =
Re [Csinkr exp{–it}], turning the scalar potential  = /r into a finite constant at r = 0 and into zero at
kr = n, with n = 1, 2, 3,… This fact seems to imply that a cavity of radius R, carved inside a good
conductor, has resonant modes with a purely radial electric field E(r) = nrE(r) and that the lowest
nonvanishing of them, with k = /R, gives the lowest (fundamental) frequency   vk = (v/R) of the
cavity. Is this conclusion correct?
Solution: The described electric field distributions would indeed satisfy the wave equation (7.3)
inside the cavity (0  r  R) and the coarse-grain boundary conditions (7.104)
E  E  0,

on its surface. However, as it follows from simple symmetry arguments77 (and as MA Eq. (10.11)
confirms), the curl of such a purely radial field equals zero at all points, so according to the first of the
Maxwell equations (6.100a), the corresponding B/t vanishes everywhere in the cavity, making the
standing electromagnetic wave impossible.
Just for the reader’s reference, the actual fundamental frequency of the cavity is even a bit lower:
11 = 11(v/R), where 11  2.744 <  is the first root of the following transcendental equation:
d sin  cos 
 j1    0, where j1     .
d  2

This j1() is one of the so-called spherical Bessel functions jl().78 (The scalar potential’s distribution
suggested in the assignment is proportional to another function of the same set:   j0() = sin/, with
 = kr.)

Problem 8.2. Simplify the Lorentz reciprocity theorem (6.121) for space-localized field sources.
Then find out what it says about the fields of two compact, well-separated sources of electric-dipole
radiation.
Solution: Let both field sources j1(r) and j2(r), participating in the Lorentz theorem,

 E
S
1  H 2  E 2  H 1 n d 2 r   E 2  j1  E1  j 2  d 3 r ,
V
(*)

be localized in space, and select the surface S in the form of a much larger sphere, with the center at
these sources. As we know from Sec. 8.1 of the lecture notes, the fields induced by any localized source
are generally a combination of a radiated wave decreasing with distance as 1/r and near-zone fields
decreasing even faster. Since the sphere’s surface grows with its radius as r2, the contribution of the

77 Indeed, if the vector E was not equal to zero, where would it be directed without violating the spherical
symmetry? (By the curl’s definition, it cannot be radial.)
78 These functions, already mentioned in Sec. 2.7, will be discussed in the QM part of this series.

Problems with Solutions Page 313

Essential Graduate Physics EM: Classical Electrodynamics

latter fields to the integral on the left-hand side of Eq. (*) tends to zero at r  0. On the other hand, the
wave fields are, on the sphere’s surface, quasi-planar, i.e. obey Eq. (7.6),
E1  Z H 1  n, E 2  ZH 2  n ,
where in this case, n  r/r, so the expression under the surface integral is proportional to
H 1  n   H 2  H 2  n   H 1  H 1  H 2  n   H1  H 2  n  .
Applying to each of these double vector products the bac minus cab rule,79 we get

H 2 H1  n   n H1  H 2   H1 H 2  n   n H 2  H1   H 2 H 1  n   H1 H 2  n  .

But the fields of these transverse waves are normal to the direction n of their propagation, so both terms
in the last form of this expression, and hence the whole left-hand side of Eq. (*) vanish, and the Lorentz
reciprocity relation (as a reminder, valid for complex amplitudes of ac fields) is reduced to simply

 E r   j r  d r   E 2 r   j1 r  d 3 r ,
3
1 2 (**)
V V

i.e. becomes very similar to the reciprocal relation for dc fields in electrostatics,

  r  r  d r    2 r 1 r  d 3 r , (***)
3
1 2
V V

which was proved in Chapter 1.

Formula (**) is an important tool for electromagnetic system design. In particular, it means that
the angular distribution of the radiation emitted by an antenna is similar to its receiving pattern.
Similarly, applied to a system that may either emit or absorb radiation (e.g., an atom), it implies that if
the system is subjected to the external radiation handcrafted to reproduce its own emission pattern, the
radiation may be fully absorbed, without any reflection – the effect sometimes called coherent perfect
absorption.80
Now note that in Eq. (**), just as in Eq. (***), each integration may be restricted to the volume
where the corresponding source density, either j(r) or (r), is different from zero. Hence if we consider
two sources of dipole radiation, each of a size small in comparison not only with the distance between
them but also with the emitted wavelength, then both electric field vectors in Eq. (**) may be taken out
of the integrals:
E1 r2    j 2 r  d 3 r  E 2 r1    j1 r  d 3 r .
V V

But according to Eq. (8.22) of the lecture notes, such integrals of the instant current densities j(r, t) are
just the time derivatives of the corresponding dipole moments. Since for the monochromatic fields we
are discussing, the time derivative is equivalent to the multiplication by a constant coefficient (–i),
these coefficients on both sides of the reciprocity relation cancel, and it becomes simply

E1 r2   p 2  E 2 r1   p1 .

79 See, e.g., MA Eq. (7.5).

80 See, e.g., the experiments by W. Wan et al., Science 331, 889 (2011).

Problems with Solutions Page 314

Essential Graduate Physics EM: Classical Electrodynamics

Problem 8.3. In the electric-dipole approximation, calculate the angular distribution and the total
power of electromagnetic radiation by the hydrogen atom within the following classical model: an
electron rotates, at a constant distance R, about a much heavier proton. Use the result to calculate the
law of a gradual reduction of R in time. Finally, evaluate the classical lifetime of the atom by borrowing
the initial value of R from quantum mechanics: R(0) = rB  0.5310–10 m.
Solution: For the radial component of the instantaneous Poynting vector, we can use Eq. (8.26)
of the lecture notes:
Z
Sr  p 2 sin 2 .
(4 vr ) 2

In our current problem, with the dipole moment p = qR = –eR (where R z n

is the electron’s radius vector, with R << r) rotating, the angle  is a 
function of time – see the figure on the right, where the z-axis is taken to
be normal to the rotation plane. The time average of sin2 may be t  y
calculated in several equivalent ways, for example by returning to Eq.  t 
(8.24) to write 0
2
p 2 sin 2   n  p
 , (t )
p x
and calculating this vector product in some immobile reference frame – for example, that shown in the
figure above, with the x- and y-axes, both within the plane of particle’s rotation, chosen so that the
observation point r is within the plane [x, z], i.e. for that point, y = 0. As the figure shows, in this frame,

n  n x sin   n z cos  ,   eR 2 n x cos   n y sin  ,

p with   t  const,

where  is the angular velocity of the electron’s rotation. Now the vector multiplication yields
nx ny nz
(t )  eR 2 sin 
np 0 cos   eR 2  n x cos  sin   n y cos  cos   n z sin  sin   .
cos  sin  0

Averaging the square of this vector over the period of rotation, i.e. over the interval  = 2, we get

 p sin  2  q 2 R 2 4 cos 2  sin 2   cos 2  cos 2   sin 2  sin 2    e 2 R 2 4  cos 2   1 sin 2   .
 2 
This formula shows that the angular distribution of the average radiated power is indeed different
from that produced by an oscillating dipole of a fixed orientation: the radiation is strongest at  = 0 and
 = , i.e. along the direction normal to the charge rotation plane (in our notation, axis z), but is also
nonvanishing along any other direction. The total average power of the radiation is
2 2
  2 eR   1    2 eR  4 Z  4e2 R 2
P   S r r dΩ  Z 0 
2
 2   cos 2   sin 2   sin d  Z 0   2  0 .
4  4 c  0  2   4 c  3 6 c 2

Comparing this result with Eq. (8.29) for a fixed-direction dipole, we see that the radiation
power by a rotating charge equals a sum of those from its two oscillating components px(t) and py(t),
calculated independently. The physical reason for this independence (which is not immediately apparent

Problems with Solutions Page 315

Essential Graduate Physics EM: Classical Electrodynamics

and requires a proof, for example, the one given above) is that the polarization directions of the partial
waves radiated by the dipole components are mutually perpendicular.
Proceeding to the calculation of the classical atom’s lifetime, elementary classical mechanics
says that at a circular motion of a non-relativistic particle in the Coulomb attractive field (with the
potential energy U = –e2/40R < 0), its kinetic energy T  mv2/2 = –U/2 > 0, so the full energy E  T +
U equals U/2. In our case, this means that
e2 e2 e2 v 2 2T / me e2
U  , E  , T , so  2    .
4 0 R 8 0 R 8 0 R R2 R2 4 0 me R 3
Weak radiation of power P << E causes a relatively slow energy reduction: dE/dt = –P, where P may
be calculated (as this was done above) for a circular orbit. In our particular case, this equation is
dE e2 d  1  Z 0 4 e 2 R 2 Z 0e6 1
     0.
8 0 dt  R  6c 6c 4 0  me R
2 2 4
dt 2 2

This equation, describing a monotonic increase of (1/R), i.e., the electron’s fall on the point-like nucleus
(proton), may be rewritten in the dimensionless form
1 dt
    , i.e. 3 2 d   ,
3 2

where   R(t)/R(0), and  is the time scale of the decay process:
4 π 2 m e2 R 3 ( 0 )
  .
e4Z00
This dimensionless differential equation may be readily integrated to give
1/ 3 1/ 3
 t  t
 (t )  1   , i.e. Rt   R01   ,
   
showing that R(t) vanishes (i.e. the electron drops on the proton) exactly at t = . For an electron (e  –
1.6010–19 C, me  0.9110–30 kg), initially rotating at the Bohr radius R(0) = rB  0.5310–10 m, the
above formula gives   1.5610–11 s.
Such atomic collapse is very fast on the human scale of events (so let us thank quantum
mechanics for preventing this disaster!), but since for our parameters  ~ 41016 s–1, this process is slow
on the rotation period scale:  ~ 106 >> 1. This strong relation justifies this gradual approach to the
problem, which neglects the loss of energy (and hence of R) during one rotation period, at the radiated
power calculation stage. Another necessary sanity check is that the radiation wavelength scale,  =
2c/ ~ 510–8 m, is much larger than R ~ 10–10 m, so the radiating charge could indeed be treated as an
electric dipole – as it was.

Problem 8.4. A non-relativistic particle of mass m, with electric charge q, is placed into a time-
independent uniform magnetic field B. Derive the law of decrease of the particle’s kinetic energy due to
its electromagnetic radiation at the cyclotron frequency c = qB/m. Evaluate the rate of such radiation

Problems with Solutions Page 316

Essential Graduate Physics EM: Classical Electrodynamics

cooling for electrons in a magnetic field of 1 T, and estimate the electron energy interval in which this
result is quantitatively correct.
Hint: The cyclotron motion will be discussed in detail (for arbitrary particle velocities) in Sec.
9.6 of the lecture notes, but I hope that the reader already knows that in the non-relativistic case (v <<
c), the above formula for c may be readily obtained by combining the 2nd Newton law mv2/R = qvB
for the particle’s circular rotation under the effect of the magnetic component of the Lorentz force
(5.10), and the geometric relation v = Rc. (Here v is the particle’s velocity in the plane normal to the
vector B.)
Solution: Let us assume that the particle’s energy loss because of the radiation during one
revolution is much smaller than the energy itself. (This condition has to be verified a posteriori for the
application of the calculation result to any particular system.) Then its orbit in the plane normal to the
field is approximately circular, and we may reuse the following intermediate result of the solution of the
previous problem: the power of the electric-dipole radiation into the free space by a particle of charge q,
moving on a circular orbit of radius R with angular velocity , averaged over the rotation period, is
Z 0 q 2 4 R 2 Z 0 q 2 2
P   R 2 .
6c 2
6c 2

For our current case,  = c = qB/m, and R = v, so81

2
Z 0 q 2  qB  2 Z 0 q 4 B 2
P   v  T, (*)
6c 2  m  3c 2 m 3
where T  mv2/2 is the kinetic energy of the particle’s motion in the plane normal to the magnetic
field’s direction. The magnetic field’s force qvB cannot do any work on the particle, i.e. does not
change its kinetic energy, so we may represent the left-hand side of Eq. (*) as the kinetic energy’s
decrease rate, –dT/dt. As a result, we get a very simple equation for the energy’s evolution:
dT T 3c 2 m 3
 , where   ,
dt  Z0q4 B2
with the evident solution T(t) = T(0) exp{–t/}, so the constant  defined above gives a fair scale of the
energy decay. (Note that in contrast with the previous problem, the decay time is independent of the
initial energy of the particle.)
For an electron (q = –e  1.6010-19 C, m = me  0.9110-30 kg) in the field B = 1 T, the above
expression yields   2.5 s, while c  1.81011s–1. This means that the relative energy loss T/T 
2/c during one rotation is indeed very small (~10-11), so the applied approximation of constant
energy at each revolution is indeed valid.
Another condition of validity of our result is imposed by the applicability of the electric dipole
approximation: kcR << 1, where kc  c/c is the radiation’s wave vector. Since R = v/c, this condition
is reduced to v << c, i.e. that the particle is non-relativistic, meaning that energy T has to be much lower
than mc2. (For an electron, mc2  0.5 MeV.)

81 The generalization of this relation to the relativistic case (when this effect is referred to as synchrotron
radiation) will be discussed in Sec. 10.3 of the lecture notes.

Problems with Solutions Page 317

Essential Graduate Physics EM: Classical Electrodynamics

Finally, we have treated the particle as a classical one. In the magnetic field, this treatment is
only valid if T is much higher than the distance c between the so-called Landau levels,82 where  
1.05510-34 Js is the Planck constant. For the above example, c  0.12 meV. Thus, the above
evaluation is valid within at least 9 orders of the electron’s energy.

Problem 8.5. A particle with mass m, electric charge q, and an initial kinetic energy T << mc2
collides head-on with a much more massive particle of charge Zq, in free space. Calculate the total
energy of electromagnetic radiation during this collision, assuming it to be much lower than T.
Solution: The last condition in the assignment means that we may calculate the particle’s
velocity and acceleration neglecting the energy losses to radiation. Also, due to the large mass of the
heavier particle, we may consider it as an immobile center providing a repulsive Coulomb force
described by Eq. (1.3). In our case of a head-on collision, the force has just one Cartesian component:
Zq 2
F ,
4 0 r 2
where r is the distance between the particles, so according to the 2nd Newton equation of the 1D motion
of the lighter particle, its acceleration is
F
r  
Zq 2
. (*)
m 4 0 mr 2
Hence, according to the Larmor formula (8.26) with p = qr, the instantaneous power of the electric-
dipole radiation (which dominates the electromagnetic radiation of this non-relativistic system) is
2
Z Z  Zq 3 
P  0 2 p 2  0 2   .
6 c 6 c  4 mr 2 
 0 
Due to the time symmetry of the collision process with respect to the nearest-approach point r0 =
r(t0), the total radiated energy may be calculated as
 2 2 
Z  Zq 3  dt Z0  Zq 3  dr dr
Erad  2  Pdt  2 0 2   t r 4  2 6 c 2    , where u  . (**)
6 c  4 0 m   4 0 m  r u r  r
4
t dt
0 0 0

The function u(r), needed for working out this integral, may be obtained from the energy conservation
law – essentially the first integral of Eq. (*):
1/ 2
mu 2 Zq 2  2T  r  Zq 2
  T, giving u   1  0  , where r0  .
2 4 0 r m r  4 0T
so Eq. (**) becomes
2 1/ 2 
Z  Zq 3   m  1 d
Erad  2 02     I, where I   .
6c  4 0 m   2T  r03 1 1  1 /  1 / 2  4

82 See, e.g., QM Sec. 3.2.

Problems with Solutions Page 318

Essential Graduate Physics EM: Classical Electrodynamics

I may be reduced to a table integral83 by using the variable substitution   1 + 2:

 
d d 16
I    1     ,
1 1  1/   
1/ 2 4

2 7/2 15
so we finally get
2 1/ 2 3 3/ 2
Z  Zq 3   m   4 0T  16 64 1  T 
Erad  2 02      2 
   2 T.
6c  4 0 m   2T   Zq  15 45 2 Z  mc 
This result (curiously, independent of q!) is important because it gives the scale (and, moreover,
the largest value) of the radiation energy at the Coulomb scattering with an arbitrary impact parameter b.
In particular, it shows that for non-relativistic particles, with T << mc2, the initial assumption that the
particle’s energy loss to radiation is always relatively small, Erad << T, is indeed justified.

z
Problem 8.6. Solve the dipole antenna radiation problem discussed l/2
in Sec. 8.2 of the lecture notes (see Fig. 8.3, partly reproduced on the
right) for the optimal value l = /2 of its length, assuming84 that the I  ( 0)
P in
current distribution in each of its arms is sinusoidal: I(z, t) = I0cos(z/l)
0
cost.
Solution: Since the main condition of the dipole approximation
condition, kl << 1, is not satisfied for the antenna that long (kl  (2/)l = l/2
), we cannot use the formulas derived in Sec. 8.2, and need to return to
the general expressions (8.17) for the retarded potentials. However, we may use the experience of Sec.
8.2, indicating that the fields in the far-field zone (kr >> 1) may be more readily calculated from the
vector potential than from the scalar potential. Integrating Eq. (8.17b) over the antenna wire’s cross-
section, we get the following expression for the vector potential in the free space around the antenna: 85
0 I 0 l / 2
 z'   R  dz'  I  l / 2  z'   R  dz' 
Ar, t   n z  cos cos   t    n z 0 0 Re   cos exp i  t   , (*)
4 l / 2
l   c  R 4  l / 2 l   c  R 

where R = [2 +(z – z’)2]1/2, with  and z being the cylindrical coordinates of the observation point r in
the reference frame shown in the figure above. At large distances, r  (2 + z2)1/2 >> l, the product in the
complex exponent may be approximated as
 R
 t 
 c
2

  t  k    z  z' 
2 1/ 2

 t  k  2  z 2  2 zz'   t  kr 1  2   t  kr   kz' cos  ,
1/ 2  zz' 
 r 
where k = /c is the wave number and   cos (z/r) is the angle between the direction toward the
–1

observation point and the z-axis. In the same limit, R in the denominator of Eq. (*) may be approximated

83 See, e.g., MA Eq. (6.5d) with n = 3.

84 As was emphasized in the lecture notes, this is a reasonable guess rather than a controllable approximation. The
exact (rather involved!) theory shows that this assumption gives errors ~5%, depending on the wire’s diameter.
85 Just as in Sec. 8.2, we ignore the possible distortions of the field by the generator and/or transmission line used

to create the current I(0, t) in the antenna. Since the TEM line’s cross-section may be much less than , in many
practical cases this is a very good approximation.

Problems with Solutions Page 319

Essential Graduate Physics EM: Classical Electrodynamics

with r, and moved out of the integral over z’.86 These simplifications enable us to reduce Eq. (*), in the
far-field zone, to
 I  l / 2
z' 
Ar, t   n z 0 0 Re expi kr  t   cos exp ikz ' cos dz' , at kr , kl  1, (**)
4 r  l / 2
l 
and to work out the remaining integral analytically:
l / 2  / 2 k  / 2k
z' 1
 cos exp ikz' cos dz'   cos kz' exp ikz' cos dz'   expikz'  1  cos dz'
l / 2 l  / 2k 2  / 2 k
     
sin   cos   cos cos  
1 expikz'  1  cos   kz'   / 2 2 2 2
  ik  1  cos  kz'   2 .
2    / 2  k  1  cos   k sin 
2

Since this result is purely real, the operator in Eq. (**) is redundant and the vector potential is
   
cos cos   cos cos  
 I
A r, t   n z 0 0 2  cost  kr   n  0 I (0, t  r / c) 2 .
2 2kr
z
sin 
2
kr sin 
2

The structure of this expression is exactly the same as that of Eq. (8.23), with the following replacement:
 
cos cos  
2 I 0, t  2 ,
p t   n z
k sin 
2

and we may repeat all the steps from that formula to Eq. (8.26) to get the following radiation power
density, i.e. the Poynting vector’ radial component:

 2 I0, t   Z I 2 cos 2 t  kr 

2 1
Z0
Sr    f    0 0 f   ,
4 cr 2  k  4 2 r 2
0.75
where
cos 2  / 2  cos 
f    . sin 2  f 
sin 2  0.5

This expression shows that the angular distribution

of the radiation, shown in the figure on the right, is 0.25
somewhat (though not too much) different from that
(sin2) of a radiating dipole – in particular, of a short
dipole antenna analyzed in Sec. 8.2. Because of that, its 0
0 0.25 0.5 0.75 1
solid-angle integral, /

J  f d  2  f sin d ,
4 0

86 Note that these are essentially the same approximations as used at the derivation of Eq. (8.78), which expresses
the Huygens principle, so the reader who needs more detailed explanations may be referred to that part of the
lecture notes.

Problems with Solutions Page 320

Essential Graduate Physics EM: Classical Electrodynamics

is also slightly different – it equals approximately 7.658 instead of 8/3  8.378 for sin2. As a result,
the total average radiation power is
J I2
P  r 2  S r d  Z 0 2 0 .
4 4 2
Now, just as was done in Sec. 8.2 for the short antenna, we may recalculate this result into the antenna’s
impedance as “seen” by the generator (or transmission line) feeding it:
J 7.658
Re Z A  Z0  Z 0  73.1  .
4 2
4 2
This impedance is very close to one of the coaxial cable standard values (75 ), making its
matching with the antenna straightforward. (If this statement is not clear, please revisit the discussion at
the end of Sec. 7.5 of the lecture notes.) Again, all these results are only valid with an accuracy of a few
percent, because they are based on the exactly sinusoidal (rather than self-consistently calculated)
distribution of the current along the antenna's length.

Problem 8.7. A plane wave is scattered by a localized object in free space. Relate the differential
cross-section of the wave’s scattering to the average force it exerts on the object. Use this general
relation to calculate the force exerted by a plane monochromatic wave on a free non-relativistic particle
and compare the result with those obtained in Problems 7.4 and 7.5.
Solution: According to Eq. (6.115) of the lecture notes, a plane wave with a Poynting vector S
carries linear momentum equal to cg = S/c per unit time per unit front area. In the problem we are
considering, this expression is applicable not only to the incident wave but also to the scattered wave if
its intensity S is measured (as it is accepted in virtually all theories of scattering) at a sufficiently large
distance from the scatterer where the wave is locally-planar.
Per the 2nd Newton law, the force F exerted on an object has to be equal to the change of its
linear moment per unit time, and according to the 3rd Newton law, this change has to be equal and
opposite to the change of the net momentum (also per unit time) of the agents exerting the force – in our
current case, of the electromagnetic waves. By using the definitions of the full and differential cross-
sections, for the values averaged over the wave’s period,87 this balance may be represented as
S incident S S  d 
F    scattered r 2 d  incident  n 0   n
 d  ,

(*)
c 4
c c  4  d  
where n0 is the direction of the incident wave and n is that of the scattered one – over which the
integration has to be carried out.88
Superficially, Eq. (*) contradicts the results obtained in the solution of Problems 7.4 and 7.5.
Indeed, in the first of these solutions, we have concluded that a traveling wave does not exert any
average force on a free point charge. This is not what Eq. (*) predicts. Indeed, according to Eqs. (8.27)
and (8.36), the charge’s differential cross-section is proportional to sin2, where  is the angle between

87 Such averaging allows us not to worry about the (possibly, different) time delays between the waves’
interaction with the object and the measurement of their intensity.
88 Alternatively, this relation may be derived from the Maxwell stress tensor, to be discussed in Sec. 9.8.

Problems with Solutions Page 321

Essential Graduate Physics EM: Classical Electrodynamics

the electric field vector E of the incident wave and the direction n toward the observer. Taking the
direction of E for the z-axis (so that  would coincide with the usual polar angle ), we see that all
Cartesian components of the second term in Eq. (*) vanish:
 2
d
n d   n sin 2 d   sin d  d n x sin  cos   n y sin  sin   n z cos  sin 2 
4
d 4 0 0
 1
 
 2n z  sin d cos  sin   2n z  1  cos 2  cos  d cos    0.
2

0 1

Hence we are left with the first term; according to Eqs. (7.9b) and (8.40), it gives a non-zero force
directed along the propagation of the incident wave:
S incident E 2 Z 02 q 4
F  n0  n0  0 , (**)
c Z 0 c 6c 2 m 2
where q is the particle’s charge and m is its mass.
In order to reveal the nature of this contradiction, let us compare the magnitude of the force
given by Eq. (**) with the maximum value of the force exerted by a standing plane wave of frequency
, which was calculated in Problem 7.5 using the same approach as in Problem 7.4:
q2 q2
F'  E E*  2E 2 ,
max
m c mc
where k = /c is the wave number. The comparison yields a ratio independent of the wave’s amplitude:
F Z0 q4 2q 2 kr 2r
  c  c,
F' 6c 3 m 2 m c 3 
where  = 2/k is the wavelength and rc  q2/40mc2 is the classical radius of the particle (for an
electron, ~310–15 m – see Eq. (8.41) and its discussion), so the above ratio is extremely small for all
frequencies where classical electrodynamics makes sense. As a result, the expressions obtained in the
solutions of Problems 7.4 and 7.5 are sufficient for virtually all practical purposes. (The reason why they
are not exact is that at their derivation, the contribution of the scattered wave into the ac electric field
applied to the particle has been neglected.89)
One more remark: as it follows from the above derivation of Eq. (*), if the object not only
scatters but also partly absorbs the incident wave, the corresponding total cross-section a (see Problem
15 below) has to be added to its first term, while the second term remains unaltered.

Problem 8.8. Use the Lorentz oscillator model of a bound charge, described by Eq. (7.30) of the
lecture notes, to explore the transition between the two scattering limits discussed in Sec. 8.3 of the
lecture notes and, in particular, the resonant scattering taking place at   0. In the last context,
discuss the contribution of the scattering to the oscillator’s damping.
Solution: The differential equation Eq. (7.30) describing the charge dynamics is linear; hence the
forced oscillations x(t) of the charge’s displacement and its dipole moment p(t) = qx(t), induced by a

89 Such problems are common for classical electrodynamics and will be further discussed in Sec. 10.6.

Problems with Solutions Page 322

Essential Graduate Physics EM: Classical Electrodynamics

monochromatic incident wave with frequency , are sinusoidal, with the same frequency. The average
power of the wave radiated (in the case of scattering, re-radiated, i.e. scattered) by such a sinusoidally
oscillating dipole into free space is given by Eq. (8.28) with Z = Z0 and v = c:
Z 0 4 2 Z 0 4 2 2
P  p  q x , (*)
12 c 2 12 c 2
where x is the complex amplitude of the charge’s oscillations. Using Eq. (7.31) to express this
amplitude via that of the electric field of the incident wave, we get
2
Z  4q4 E
P  0 2 2 2 ,

12 m c  0   2 2  2 0 2 
so Eq. (8.39) yields the following total cross-section of scattering:
Z 02 q 4 4
 . (**)

6 m 2 c 2  02   2 2  2 0 2
In the high-frequency limit, /0  , this expression reduces to the Thomson scattering
formula (8.40)-(8.41), while in the opposite limit, /0  0, it leads to Eq. (8.45) with
qx q2
  0  ,
E m 02
and hence to the Rayleigh scattering. However, Eq. (**) shows that on the way between these two
limits, the scattering cross-section may have a very high peak at   0, with a height limited only by
the oscillator’s damping:
2
Z 2q4  0 
  0  0 2 2    Q 2   ,
6m c  2 0 
where Q is the Q-factor of the resonance, which may be much larger than 1. This is resonant scattering.
The damping may be contributed not only by some internal energy losses inside the bound
charge system but also by the scattering itself. Indeed, the average rate of the energy loss (i.e. the
dissipation power) at forced oscillations of a harmonic oscillator is90
2
Ploss  2m x 2 0  m 2 x  0 .
If the electromagnetic wave’s (re-) radiation is the only energy loss mechanism of the oscillator, this
expression has to be equal to the one given at   0 by Eq. (*). As a result,

2 0Z 0 q 2 2     q
2
 2
Q 1
   Z 0 c 0      krc ,
 3
 6c 2 m 3  c   4 0 mc
2


90 This expression may be readily obtained by interpreting the second term on the left-hand side of Eq. (7.30) as a
viscous friction force proportional to the charge’s velocity: Fv = –2mv0, giving the instant dissipation power Ploss
= –Fvv = 2mv20.

Problems with Solutions Page 323

Essential Graduate Physics EM: Classical Electrodynamics

where k = /c is the wave number, and rc is the classical radius defined by Eq. (8.41) of the lecture
notes. For an electron, rc ~ 310-15 m, so the product krc is very small at all frequencies up to hard 
rays. Hence, the radiation-defined Q is very large, so the resonant scattering may be much stronger than
the Thomson and Rayleigh scattering.
Quantum mechanics91 not only confirms the above formulas for the harmonic oscillator, but also
shows that the resonant scattering takes place for any system with discrete energy levels En, as soon as
the wave frequency  approaches any of the quantum transition frequencies nn’ = (En – En’)/, with n 
n’, provided that the corresponding matrix element of the electric dipole moment’s operator is different
from zero and that at least one of the involved energy levels is occupied.

Problem 8.9.* A sphere of radius R, made of a material with a uniform permanent electric
polarization P0 and a constant mass density , is free to rotate about its center. Calculate its average total
cross-section for scattering of a linearly polarized plane electromagnetic wave with frequency  << c/R,
incident from free space, in the weak-wave limit, assuming that the initial orientation of the polarization
vector is random.
Solution: As we know from the solution of Problem 3.13, the electric field induced by a
polarized sphere is identical to that of a point electric dipole, placed in the sphere’s center, with the
moment given by a very simple and natural formula
4 3
p  VP0  R P0 .
3
At the given condition  << c/R, the wave’s electric field E is virtually uniform on the distances of the
order of R. Per Eq. (3.17) of the lecture notes, such a uniform electric field E exerts on such a dipole
(i.e. on our sphere) the following mechanical torque:
τ  pE.
As we know from the basic classical mechanics,92 under the effect of such a torque, the mechanical
angular momentum L of the sphere evolves as
L  τ .
Classical mechanics also says93 that the angular momentum L of a spherically-symmetric body
(frequently called the spherical top) equals I, where  is the instantaneous angular velocity vector,
and I is the (only) principal moment of inertia of such a body (with the mass M = V = (4/3)R3): 94
2 8
I MR 2  R 5 .
5 15
Combining the above relations, we get the following equation of time evolution of the vector :
  pE.
IΩ (*)

91 See, e.g., QM Chapters 6 and 9.

92 See, e.g., CM Eq. (1.33).
93 See, e.g., CM Secs. 4.1-4.2, in particular, Eqs. (4.31)-(4.32).
94 See, e.g., the model solution of CM Problem 4.1.

Problems with Solutions Page 324

Essential Graduate Physics EM: Classical Electrodynamics

One more relation between the vectors  and p follows from the kinematics of rotation of any
vector of constant length, in particular, p:95
p  Ω  p . (**)

A joint solution of the two vector equations (*) and (**), with a time-dependent (oscillating)
vector E, is not a trivial task but it may be simplified if the incident wave’s amplitude is relatively weak.
To see this, let us differentiate Eq. (**) over time, and then plug Eq. (*) and (again) Eq. (**) into the
resulting right-hand side:
  Ω
p   p  Ω  p  1 p  E   p  Ω  Ω  p  . (***)
I
Since p = const, the first term in the last expression contains a component oscillating with the wave’s
frequency , and proportional to its amplitude E. On the other hand, Eq. (*) shows that the amplitude
of such oscillations of the vector  is also proportional to E, so at E  0, the amplitude of the fast
oscillations of the term (p) is proportional to  E 2, i.e. is much smaller than the first term. (This
reasoning is strict in the absence of the initial rotation of the sphere, but even if such a rotation is present,
its angular velocity 0 has to be as high as  to affect the approximation used below.)
Dropping, on this grounds, the second term on the right-hand side of Eq. (***) (but only for a
while – see below), we get
1
   p  p  E  .
p
I
According to this expression, if the incident wave is linearly polarized, i.e. the p
vector E is rapidly oscillating along a fixed direction, the resulting vector p
 also E 
oscillates, with the same frequency, along a line normal to the vector p (see the  /2
figure on the right), with the following magnitude:
pE
p  p  E 
2
p E
p   sin  ,
I
where  is the angle between the vectors E and p. Plugging this expression into the general Eq. (8.27),
for the average power of radiation into free space (with Z = Z0 and v = c), we get a frequency-
independent result:
2 2
Z0  p2  2 Z0  p2 
 E t sin  
2
P  
2 
2

2 
 E sin 2  .
6 c  I  12 c  I 

Here the power has been averaged over the period of the wave frequency . Now returning to
Eq. (***), we have to notice that the second term on its right-hand side, neglected in the above
calculation, also has a certain small (proportional to  E 2) but nonvanishing average over the period.
This average component of the second derivative of the vector p results in a rather complex slow
evolution of the direction of the vector, and hence to a change of the angle . Though slow on the scale
of the wave frequency , this evolution may be noticeable on the scale of the scattering measurement
experiment and, generally speaking, should be accounted for in theory. However, if the initial direction

95 See, e.g., CM Sec. 4.1, in particular Eq. (4.9).

Problems with Solutions Page 325

Essential Graduate Physics EM: Classical Electrodynamics

of the sphere’s polarization vector P0 is random, the evolution details are not important for the statistical
average96 of the radiation power, which may be calculated as
 2 
1 1 1 Z0  p2 
4   sin
2
P  P d  2 P sin d    E 2
 sin d .
4 4 0
2 12c 2  I  0

By using the standard variable substitution   cos, it is easy to see that the last integral equals 4/3, so,
finally,
2
1 Z0  p2  2 4 Z0 p4 2
P    E
2 
 E .
2 12c  I  3 18c I 2 2

For the similarly averaged cross-section (8.39), this result yields

P Z 02 p 4
   .
E / 2Z 0 9c I
2 2 2

Plugging in the above expressions of p and I for a uniform sphere, we get

4 2 2
Z 2  4 3   8  100  P02 /  0 
  02  R P0   R 5     R 2 . (****)
9c  3   15  81  c 2 

It is curious that the scattering cross-section is proportional to the physical cross-section 0 
R of the sphere; however, for any non-exotic matter, it is much smaller. Indeed, as was mentioned in
2

Sec. 3.3 of the lecture notes, the highest values of the remnant polarization P0 in ferroelectrics do not
exceed ~1 C/m2, while their density  is of the order of 104 kg/m3. Plugging these values (both in the SI
units) into our result, we see that the ratio /0 can hardly be higher than ~10-20.97
This smallness is not occasional; indeed, the largest polarization of usual condensed matter
corresponds to the displacement of a few outer-shell electrons by the distance of the order of interatomic
distances a, which are of the order of the Bohr radius rB, so the largest dipole moment per atom is p0 ~
erB, and the largest polarization is P0 ~ p0/rB3 ~ e/rB2. Plugging in the quantum-mechanical expression
for the Bohr radius,98 rB = 2/me(e2/40), and the natural (though crude) estimate of the material’s
density,  ~ Amp/rB3, where A is the atomic number, we see that the fraction in the last form of Eq.
(****) is of the order of (me/Amp)2 << 1, where   e2/40c  1/137 is the fine-structure constant,
which characterizes the strength (or rather the weakness :-) of electromagnetic interactions on the
quantum-relativistic scale – see also the model solutions of Problems 5.18 and 9.14.

96 Such an average is all we need if we measure the simultaneous, independent scattering of a wave by many
similar spheres with random polarization directions – just as was discussed in Sec. 8.3 for the Rayleigh scattering.
(Please note the statistical averaging sign  , which differs from the temporal averaging sign ¯; it will be used
very often in the QM and SM parts of this series.)
97 This extreme smallness of the calculated effect makes it hard to detect on the masking background of other
effects, not accounted for in our simple model, such as small but nonvanishing differential (“dynamic”)
polarizability dyn  dp0/dE of the ferroelectric’s atoms.
98 It was already mentioned in Sec. 2.1 of the lecture notes, and in the model solutions of Problems 5.14 and 5.15,
and will be discussed in detail in the QM part of this series.

Problems with Solutions Page 326

Essential Graduate Physics EM: Classical Electrodynamics

Problem 8.10. Use Eq. (8.56) of the lecture notes to analyze the
interference/diffraction pattern produced by a plane wave’s scattering on a set of N
similar equidistant small objects located on a straight line normal to the direction of the k0 d
N
incident wave’s propagation – see the figure on the right. Discuss the trend(s) of the d
pattern in the limit N  .
Solution: As was discussed in Sec. 8.4 of the lecture notes, in the case of similar
small scatterers located at points rj, the scattered wave intensity is proportional (besides the smooth
polarization-related function sin2) to the scattering function F =  I(q) 2, where I(q) is the phase sum
(8.57): x
I q    exp iq  r j ,
j  2d
and q  k – k0 is the wave vector’s change at scattering. k
Introducing the Cartesian coordinates just as in Fig. 8.6 (see the  d qx q
figure on the right), we get rj = xjnx, with xj = jd, so for the 
observation point within the plane of the drawing (common for the 0 z
k0
line of the elementary scatterers and the wave vector k0), where qx
d
= ksin, we get
N N
 2d
I q   exp ikjd sin    exp ij  ,
  (*)
j 1 j 1

where   kdsin, so
*
 N  N  N
F    I q   I q I q     exp ij   exp ij'  
*
 expi j'  j  .
2

 j 1  j' 1  j , j' 1

The three most significant features of the function F(), describing an interplay between the
interference of the waves scattered from the adjacent objects and the diffraction of the incident wave on
the system as a whole, are immediately apparent from this formula:
(i) the function is even because every term of the last sum is invariant with respect to the
simultaneous replacements {  –, j  j’} and the sums over j and j’ are similar;
(ii) the function is 2-periodic because each term of the sum is; and
(iii) the function reaches its major maxima,99 Fmax = N2, at  = 0 (and hence at all  = 2n, with
any integer n) because at these points, each term of the sum reaches its maximum equal to 1.
These features are clearly visible on the numerical plots of the function for several values of N,
shown in the figure below. (Its right panel is a zoom on any of the major peaks.) The plots show that as
the number N of scatterers is increased, the width  of the function’s major peals decreases as ~ /N,100
and that at N  , the function rapidly approaches a periodic set of the Fraunhofer diffraction patterns –
see Fig. 8.8 of the lecture notes:

N   2n  k  Nd sin 
2
 sin  
Fn     N   N 2 sinc 2 , with     Nn, for N  1 .
   2 2

99 As the figure below shows, the function has (N – 1) additional, lower maxima at each period.
100 The exact coefficient in this relation depends on the accepted quantitative definition of the peak’s width.

Problems with Solutions Page 327

Essential Graduate Physics EM: Classical Electrodynamics

This is natural because, at N  , the sum in Eq. (*) tends to the phase integral defined by Eq. (8.63)
and given, for our case, by Eq. (8.64) with the replacement V  N.

N 5
20 20
N 5 4
F   F  

10 10 3

2
2
0 0
4 2 0 2 4 1  0.5 0 0.5 1
 /  /
Note, however, that in terms of the observation angle  = sin (/kd), the transition N   looks
-1

rather different in the two ultimate cases.

(i) If the total length a  Nd of the scattering system is kept fixed (i.e. the distance d between the
adjacent scatterers is decreased as 1/N), then the width
2 N 
 sin       
ka ka ka
of each diffraction peak does not depend on N, while the intervals between the peaks,
ka N
  sin      N ,
2 2
grow as N. As a result, since the largest magnitude of sin is 1, in the case ka ~ 1, none of the
diffraction peaks but one (with n = 0, i.e.   0) exist. This is the case of diffraction, which was
discussed in Sec. 8.4 of the lecture notes.
(ii) In the opposite case when the period d of the scattering structure is kept fixed, so the increase
of N leads to the proportional increase of its total length a = Nd, the intervals
2 N 2
sin       
ka ka kd
between the major peaks remain constant, so they show up at the same values of the observation angle .
In contrast, the width  of each peak shrinks:
 1
 sin    0,
kd N
i.e. the diffraction peaks become very sharp, so the difference between their positions for slightly
different wavelengths  = 2/k may be used for high-resolution spectroscopy. This is exactly the limit
pursued in diffraction gratings – see Sec. 8.8 of the lecture notes for a brief discussion of these important
devices.

Problems with Solutions Page 328

Essential Graduate Physics EM: Classical Electrodynamics

Finally, note that if the observation point is off the common plane shown in the first figure
above, the interference/diffraction pattern is quantitatively different because the vector q also has an off-
plane component, so the scattering angle  is now different from (/2 – ) – see, e.g., Fig. 8.5 in the
lecture notes. However, in the most interesting case N >> 1 and kd ~ 1, when the most important
features of the pattern pertain to small scattering angles, they remain qualitatively the same.

Problem 8.11. Use the Born approximation to calculate the differential cross-section of a plane
wave’s scattering by a uniform dielectric sphere of an arbitrary radius R. In the limits kR << 1 and 1 <<
kR (where k is the wave number), analyze the angular dependence of the differential cross-section and
calculate the total cross-section of scattering.
Solution: According to Eqs. (8.62)-(8.63) of the lecture notes, the differential cross-section may
be calculated as
2
d  k 2  2
   (  1) 2 I (q) sin 2 ,
d  4 
with
I (q)   exp iq  rd 3 r , (*)
V

where V is the scatterer’s volume, and q  k – k0 is the wave vector’s change due to scattering. Since, at
the elastic scattering we are discussing, k = k0, the magnitude of q is related to the scattering angle 
(i.e. the angle between the vectors k and k0) as q = 2ksin(/2).
For a sphere, it is natural to work out the phase integral (*) by taking the direction of the vector q
for the polar axis, so qr = qrcos, and
R  R 1
I (q)  2  r 2 dr  exp{iqr cos  } sin d  2  r 2 dr  exp{iqr cos  } d (cos  )
0 0 0 1

 
4 4
R R
1
 2  r 2 dr e iqr  e iqr   rdr sin qr  3 sin qR  qR cos qR .
0
 iqr q 0 q
As a result, the differential cross-section is
1
d
dΩ
 2

 k 2 R 3 (  1) 2 sin 2  f (qR), (**) 1 / 9
0.1

where the function f, defined as

0.01  4
f ( ) 
sin    cos   2
, 3
6 1 10

4
f ( )
behaves as the red line in the figure on the right shows. 1 10

As Fig. 8.5 (reproduced, with additions, below) shows, 5

the polarization angle  participating in Eq. (**), i.e. the 1 10

angle between the vectors E and k, is related to the 6

1 10
scattering angle  and the azimuthal angle  as follows: 0 2 4 6 8 10


Problems with Solutions Page 329

Essential Graduate Physics EM: Classical Electrodynamics

k y2  k z2
sin  
2
 sin 2  sin 2   cos 2  . (***)
k2
For a relatively large sphere, kR >> 1, the k x  k sin  cos 
function f(qR) oscillates already at relatively small
values of the scattering angle . As Eq. (***) shows,
at such angles,   /2, so d/d is virtually k
independent of the azimuthal angle , so Eq. (**) E q  k  k0
describes a round-ring diffraction pattern. Per this nr
formula, as the number n of the intensity maximum is  

increased, its position rapidly approaches the k sin 
following asymptotic value: 0 k0
k z  k cos 
 1
 n   n   ,
 2 k y  k sin  sin 
i.e. the following scattering angle:
n  1 
n   n    1 .
kR  2  kR
As qR is increased, the envelope of the intensity oscillations decreases very fast, as (qR)–4 – see
the blue dashed line in the first figure above. As a result, the integral (8.50), which gives the total cross-
section of scattering, converges already at small scattering angles  << 1, where we may take sin2  1,
q  k, and sin  , so
1 / k  1 
d
 1 / k  1 d  2k R (  1)
4 6 2
0 f (kR )d  2k R (  1) 0 f ( )d .
2 4 2

4
d
The last integral is equal to ¼,101 so, finally,
 1
 (kR 2 ) 2 (  1) 2   0 (kR) 2 (  1) 2 , for kR  1 ,
2 2
where 0  R2 is the geometric cross-section of the sphere, as “viewed” by the incident wave. Since kR
>> 1, it is tempting to conclude in this limit, that  may be larger than 0. However, a straightforward
analysis of the validity of the Born approximation for extended uniform objects102 yields the condition 
<< 0, so the above result is only valid if ( – 1) << 1/kR.
In the opposite limit, kR << 1, we can use Eq. (**) with qR << 1 and f(qR)  f(0) = 1/9 for all
possible directions of the vector k (because qR  2kR), so
2
d  k 2 R 3 
   (  1) 2 sin 2 .
dΩ  3 
This result coincides with the general Eq. (8.53) with V = (4/3)R3 and shows that in this limit, there is
no diffraction pattern to speak about. As we have already seen in the derivation of Eq. (8.27), the solid-
angle integral of sin2 equals 8/3, so the total cross-section in this limit is

101 See, e.g., MA Eq. (6.3b).

102 See the model solution of Problem 14 below.

Problems with Solutions Page 330

Essential Graduate Physics EM: Classical Electrodynamics

8 4 6 8
 k R (  1) 2   0 (kR) 4 (  1) 2 , for kR  1 . (****)
27 27
The last expression clearly shows that in this limit,  is also much smaller than the geometric
cross-section 0 of the sphere.

Problem 8.12. A sphere of radius R is made of a uniform dielectric material, with an arbitrary
dielectric constant. Calculate its total cross-section of scattering a linearly-polarized low-frequency (k
<< 1/R) wave and compare the result with the solution of the previous problem.
Solution: In the low-frequency limit kR << 1, i.e. when the wavelength  = 2/k is much larger
than the sphere’s radius R, the incident wave’s field is essentially uniform at distances of the order of R.
As the analysis in Sec. 3.4 has shown, in this case, the electric field created by the sphere’s polarization
outside it is exactly the same as that of a point electric dipole with the dipole moment (3.64):103
 1
p   4 0 R 3 E . (*)
 2
This means, in particular, that the power re-radiated by the sphere is given by Eq. (8.28) of the lecture
notes, with p given by Eq. (*). Now using Eq. (8.39), we get
2
8
 kR 4    1  R 2 .
3   2
Comparing this expression with Eq. (****) in the solution of the previous problem (which was
obtained in the Born approximation) in the same frequency limit, we see that they coincide only in the
limit   1, when ( + 2)2  9. On the other hand, if the dielectric constant is increased, the cross-
section gradually approaches its largest possible value
8  max 8
   max  kR 4 R 2 , so that  kR  ,
4
for   1 .
3 R 2
3
The last expression clearly shows that in the low-frequency limit (kR << 1) even this largest
value is much smaller than the geometric cross-section 0 R2 of the sphere.

Problem 8.13. Use the Born approximation to calculate the differential cross-section of a plane
wave’s scattering on a right circular cylinder of length l and radius R, for an arbitrary angle of incidence.
Solution: In order to spell out Eq. (8.62) of the lecture notes, we have to calculate the phase
integral (8.63):
l / 2 R 2
I (q)   exp iq  r' d 3 r'   dz' exp iq z z'  'd'  d' exp iq ' cos ',
V l / 2 0 0

where qz is the component of the scattering vector q = k – k0 along the cylinder’s axis (taken for the z-
axis), q is the magnitude of its component in the plane of the cylinder’s cross-section, and ’ is the

103Here and below, for any material with frequency-dependent complex electric permittivity (), the dielectric
constant  has to be understood as the ratio ()/0.

Problems with Solutions Page 331

Essential Graduate Physics EM: Classical Electrodynamics

angle between that component and the vector ’  r’ – nzz’. The first integral is simple – see, e.g.,
Eqs.(8.64)-(8.66) and Fig. 8.7:
l / 2
iq z' ql

l / 2
e z dz'  l sinc z ,
2
while the second (double) integral may be expressed via a Bessel function of the first kind, by using its
integral representation104 and then the recurrence relation (2.143) for n = 1:
R 2 q R
2 2 q R 2R
R
iq ' cos'
 'd'  d' e  2  J 0 q ' 'd'  2
q  J  d  q J  0  q J q R  .
0 2 1

1 
0 0 0 0  

With these substitutions, Eqs. (8.62)-(8.63) yield

 J q R  
2 2
d  k 2 lR 2  ql
    12  1   sin 2  sinc 2 z .
dΩ  4   q R / 2  2
Here the factors inside the square brackets have 1
been grouped to emphasize that this fraction is
very much similar to the sinc function, and in
particular tends to 1 when its argument (in our
0.5 J 1  
case, qR) tends to zero – see the figure on the
right.  /2
In this form, it is evident that for a very 0
small cylinder with kl, kR << 1, our result
reduces to Eq. (8.53) with V  R2l. However, sinc
for relatively large cylinders, our apparently
 0.5
simple analytical result describes a relatively 0 5 10 15 20 25
complex angular pattern, because the relation 
between qz, q, and  depends on the direction of the vectors E, k, and k0 relative to the cylinder’s axis
– see Fig. 8.5 of the lecture notes.

Problem 8.14. Formulate the quantitative condition of the Born approximation’s validity for a
uniform dielectric scatterer, with all linear dimensions of the order of the same scale a.
Solution: As was mentioned at the beginning of Sec. 8.3 of the lecture notes, the general
condition of the Born approximation’s validity is the weakness of the scattered wave in comparison with
the incident wave Ein inside the scatterer. For a linear-dielectric, non-magnetic object that scatters a
plane, monochromatic incident wave with a wave vector k0, this means that its electric polarization P at
point r’ is given by Eq. (3.45) with E(r’) = Ein(r’) = Re[(E)inexp{ik0r’}]:


P r'     1 0 E in r'     1 0 Re E  in e
ik 0 r'
.

104 See, e.g., MA Eq. (6.15a) with n = 0.

Problems with Solutions Page 332

Essential Graduate Physics EM: Classical Electrodynamics

Using this expression to calculate the dipole moment dp(r’) = P(r’) d3r’ of an elementary volume of the
scatterer, we may use Eq. (8.17b) to find the contribution of this volume to the complex amplitude of the
vector potential of the scattered wave at a certain observation point r:
 0 j r'  ik R 3   idp   ikR  1 ik r'
dA  r   e d r'  0 e  i 0   1 0 E  e ikR e 0 d 3 r' ,
2 R 2 R 2 R
where R  r – r’ is the vector connecting the source and observation points. Summing up these
contributions over all the scatterer’s volume V, and taking into account that according to Eq. (6.7), the
vector potential’s amplitude in the (plane and monochromatic) incident wave is (A)in =
(E)inexp{ik0r}/(–i), we get105

A  scat  0  0 2
  1 e ik 0 r  e ikR e ik 0 r' d r'  k   1 e ik 0 r  e ikr  r' e ik 0 r' d r'
3 2 3

A  in 2 V
R 2 V
R
(*)
k2
  1 e ik  k 0 r
3 2 3
ik r' d r' k
 e ik r' e 0    1  e iqr' d r' , with q  k  k 0 .
2 V
R 2 V
r  r'

The Born approximation is valid if the right-hand side of this equality is much less than 1 at any
point of the scatterer. For a sample much smaller than the wavelength, we can take qr’ << 1 at any point,
so the integral in the last form of Eq. (*) is of the order of a2,106 and the approximation’s validity
condition is
(ka) 2
  1  1, for ka  1 . (**)
2
Comparing this expression with Eq. (8.53) of the lecture notes, which gives the total cross-section
2 2
d  k 2V  4  k 2V 
 d     1 4
2
sin d     1 ,
4
d  4  3  4 
with the obvious estimates V ~ a3 for the scatterer’s volume and 0 ~ a2 for its physical cross-section, we
see that Eq. (**) may be rewritten just as
   0 .

On the other hand, for an extended object, with a >> , the function under the integral in Eq. (*)
rapidly oscillates in the direction of the vector q, so the contribution of this direction to the integral is
accumulated only on distances ~ ~ 1/k rather than a. As a result, a fairer estimate of the integral is ka,
so the Born approximation is valid if107
ka
  1  1, for 1  ka . (***)
2

105 For distant observation points (R, r >> a), this integral is just I(q)/r, where I(q) is the phase integral (8.63), but
for our current task, we need to consider points inside the scatterer: r ~ a.
106 Let me remind the reader that the 3D singularity 1/r is integrable: d3r/r = 2r2  0 at r  0.
107 For the analog of the conditions (**) and (***) for the Born approximation in quantum mechanics, see QM
Eq. (3.77).

Problems with Solutions Page 333

Essential Graduate Physics EM: Classical Electrodynamics

However, we should remember that the estimates (**) and (***) are only valid if all linear
dimensions of the scatterer are comparable, and should be modified, for example, for strongly elongated
objects – say, long/thin rods. In such objects, for certain parameters, the Born approximation may work
well for some scattering directions, and give wrong results for others.

Problem 8.15. If a scatterer absorbs some part of the incident wave’s power, it may be
characterized by an absorption cross-section a defined similarly to Eq. (8.39) for the scattering cross-
section:
Pa
a  2
,
E / 2 Z 0

where the numerator is the time-averaged absorbed power. Use two different approaches to calculate a
for a small sphere of radius R << k–1, s, made of a nonmagnetic material with an Ohmic conductivity 
and the high-frequency permittivity opt = 0. Can a of such a sphere be larger than its geometric cross-
section R2?
Solution:
Approach 1. Due to the condition R << k–1, the displacement currents are negligible, and the
external field is virtually uniform on the spatial scale of R, while due to the condition R << s, the
magnetic field of the eddy currents is also negligible. In this case, we may calculate the electric field
inside the sphere from the first of Eqs. (3.65) of the lecture notes, which may be rewritten as
3 0
E int  E0 ,
  2 0
with the replacements discussed in Sec. 7.2: Eint  (Eint), E0 E,   (). For an Ohmic conductor
in question, the complex electric permittivity may be calculated using Eq. (7.46), with opt = 0.
Combining these replacements, we get
3 0 3 0 1
E int   E  E  E ,
    2 0  0  i /    2 0 1  i / 3 r
where r  0/ is the charge relaxation time of the sphere’s material,108 which was discussed in Sec. 4.2
– see the second of Eqs. (4.10) with  = 1.
Now calculating the temporal average of the Joule power loss (4.40) exactly as this was done for
the Poynting vector in Eqs. (7.41)-(7.42) of the lecture notes, we get
2

Pa    E d r   ReEint  e 
it 1 4 3 1 1 4 3
V   Eint 
2 2
2 3
R  E R
2 1  i / 3 r
int
V
2 3 3
1 1 2 4 3
 E R ,
2 1  1 / 3 r 2 3

108Let me hope that my usage, in this solution, of the Greek letter  in two different meanings would not result in
confusion. (It denotes the cross-section only when used with the index “a”.)

Problems with Solutions Page 334

Essential Graduate Physics EM: Classical Electrodynamics

so the absorption cross-section is

P 4 Z 0R 3
a  a
. (*)
2
E / 2 Z 0 3 1  1 / 3 r 2

Approach 2. The initial step of another way to express the power absorbed in the sphere may be
used for any object of a size well below the wavelength. Integrating Eq. (4.38) over its volume, we get

Pa   E  j d 3 r .
V

If at all points, we take E equal to the unperturbed field E0,109 which may only change on the spatial
scale much larger than the object’s size, we may take the electric field out of the integral. Representing
the remaining integral of j in the form of a sum over all charges of the object, just as it was done at the
derivation of Eq. (8.133), we may express it as the time derivative of its dipole moment p:
drk d dp
 jd r   qk v k   q k   q k rk 
3
.
V k k dt dt k dt

If p is a linear function of E0, it changes with the frequency  of the wave, so the time-averaged
absorbed power becomes

Pa  Re  E  e it 
d
dt

Re p  e it 
i
4
   

E   p *  E*  p    Im E  p * .
2

Assuming that the polarization is isotopic: p  E, and defining the complex polarizability ()  ’()
+ i”() similarly to Eq. (3.48), i.e. by the relation p = ()E, we get

Pa  "   E2 , .
2
giving a very useful general relation
 a  Z 0"   .
For our particular case of a sphere, we may take p from another result of the same electrostatic
solution, namely Eq. (3.64), giving
 1
  4R 3 0.
 2
Replacing  in this formula with ()/0, with () from the same Eq. (7.46), we arrive at the same Eq.
(*) as in Approach 1.
This result shows that in the limit of low frequencies, now in the sense  << 1/r = /0, the
absorption cross-section is proportional to the material’s resistivity 1/:
4 Z 0R 3 Z 0  02 R 3 2
a   12Z 0R   r  12
3 2 2
, for  r  1 ,
3 1 / 3 r 2 

109 This is a nontrivial step of this approach. It may be strictly justified in the dipole approximation, similarly to
the derivation of Eq. (3.15) in Sec. 3.1 of the lecture notes – a simple exercise, highly recommended to the reader.

Problems with Solutions Page 335

Essential Graduate Physics EM: Classical Electrodynamics

and vanishes at   0. This is natural because the low-frequency field is effectively screened out of the
conductor (Eint  r  0).
As the wave’s frequency is increased, the a given by Eq. (*) grows, but then saturates at a
frequency-independent value proportional to the sphere material’s conductivity :
4 a 4
a  Z 0R 3 , so that  Z 0 R , for  r  1 .
3 R 2
3
(Note that, as the elementary Eq. (4.21) shows, the right-hand side of the last relation, besides a
numerical factor of the order of 1, is equal to the ratio of the free-space wave impedance Z0  377  to
sphere’s dc resistance.110) Plugging into this result the above high-frequency condition in the form  <<
0, and the first condition accepted at this solution, R << 1/k  c/, we get
a c
~ Z 0R  Z 0  0   Z 0  0 c  1 ,
R 2

so our result is valid only if the ratio is much less than one. The second condition, R << s  (2/0)1/2,
also together with the high-frequency condition  << 0, leads to a similar restriction:
1/ 2 1/ 2
a      
~ Z 0R  Z 0    Z 0  0   1.
R   0    0 
2

This means that a, within the specified restrictions R << k–1, s, cannot be larger than the
geometric cross-section R2 of the sphere.

Problem 8.16. Use the Huygens principle to calculate the wave’s intensity at the symmetry plane
of the slit diffraction experiment (x = 0 in Fig. 8.12 of the lecture notes), for an arbitrary ratio z/ka2.
Solution: The general solution of the problem is given by Eqs. (8.105)-(8.107) of the lecture
notes. For the particular case x = 0, Eq. (8.107) becomes
a / 2 a / 2 1/ 2 
 kx' 2 
 cos    i sin  d
ikx' 2 kx' 2  2z 
I x   exp dx'  2   cos  i sin  dx'  2  
2 2

a / 2
2z 0  2z 2z   k  0
1/ 2 1/ 2 1/ 2
 z   k    a2 
 2  C    iS  , where     a    ,
 k   8z   4 z 
and C() and S() are the Fresnel integrals defined by Eqs. (8.112). Combining this relation with Eqs.
(8.105) and (8.106), for the relative wave intensity we get

S 0, z  f  0, z 
2

S0

f0

 2 C 2    S 2   . 

110 Assuming that the resistance is measured between two electrodes reasonably distributed on the opposite sides
of a sphere’s diameter.

Problems with Solutions Page 336

Essential Graduate Physics EM: Classical Electrodynamics

This result is shown with the solid red curve in the 2

figure on the right. As the distance z of the observation
point from the slit is increased (i.e. the argument  of the 1.5
Fresnel integrals is decreased), the intensity first
oscillates, with an increasing amplitude, about the average S 0, z 
value equal to S0, but just beyond the Fresnel-to- S0 ) 1
Fraunhofer diffraction crossover, the intensity starts a
monotonic drop (see the dashed line in the same figure):
0.5
S 0, z  4 2 a 2 a2
    1, at z  .
S0  z  0
0 0.5 1 1.5
Qualitatively, this behavior is similar to that for the
diffraction at a circular orifice (see the next problem), but
z / a2 /   
the intensity oscillations with z are less pronounced (with the maximum intensity never reaching 2S0,
and the minimal intensity not approaching zero), and its drop at large distances is not so fast – all
because in our current case, the diffraction takes place is only one dimension.

Problem 8.17. A plane wave with wavelength  is

normally incident on an opaque planar screen with a round S ( z)  ?
orifice
orifice of radius R >> . Use the Huygens principle to 2R
calculate the passing wave’s intensity on the system’s
S0 0 z
symmetry axis, at distances z >> R from the screen (see the
figure on the right), and analyze the result.
Solution: According to Eq. (8.91) of the lecture notes,
the complex amplitude of the wave is

f ( z)  f 0
k
R 2
 'd'  d'
 
exp ik ' 2  z 2  1/ 2
.
2 i 0 0 ' 2
z 
2 1/ 2

Using the axial symmetry of the problem, and the conditions  << R << z, this expression may be
simplified, just as it has been done in Sec. 8.5 – see, e.g., Eq. (8.86):
R kR / 2 z 2
k  k' 2  k ikz z ikz   kR 2  
f  ( z )  f 0 e ikz  expi   'd '  f 0 e  exp  i  d   f 0 e  exp i   1 ,
iz 0  2z  iz k 0   2 z  
so the wave’s intensity is
2
 kR 2   kR 2   R 2 
S  z   S 0 expi   1  2 S 
0 1  cos   2S 0 1  cos  .
 2z   2z   z 
The result shows that the intensity oscillates, reaching zero at points z = zn where cos(R2/zn) =
1, i.e. at R2/zn = 2n (with n = 1, 2,…), so
R2
zn  .
2n

Problems with Solutions Page 337

Essential Graduate Physics EM: Classical Electrodynamics

Between any two adjacent minima, the intensity reaches a maximum (twice larger than the incident
wave’s intensity!) at points z = z’n where cos(R2/z’n) = –1, i.e. R2/z’n = 2n – , so111
R2
z' n  .
2 n  1 / 2 
Note that the first few of these maxima and minima (the ones most distant from the screen)
correspond to the border between the Fresnel and Fraunhofer diffraction limits: z/R ~ R/, and that in the
latter limit, the intensity does not oscillate but rather gradually vanishes at z  :
2
 R 2  1 R2
S z   S 0    2 , at z  . (*)
 z  z 
This decrease may be readily explained as the wave’s taking a spherical form – the behavior natural for
very large distances, from which the orifice looks like a point source.

Problem 8.18. A plane monochromatic wave is S ( z)  ?

disk
now normally incident on an opaque circular disk of 2R
radius R >> . Use the Huygens principle to calculate
S0 0 z
the wave’s intensity at distances z >> R behind the
disk’s center – see the figure on the right. Discuss the
result.
Solution: According to Eq. (8.91) of the lecture notes, the complex amplitude of the wave on the
system’s symmetry axis is

f ( z)  f 0
k


 'd'  d'
2
 
exp ik ' 2  z 2 
1/ 2
.
2 i R 0 ' 2
z 
2 1/ 2

Using the axial symmetry of the problem, and the Huygens principle conditions (8.83), z >> R >> , this
expression may be simplified just as this has been done in Sec. 8.5 – see, e.g., Eq. (8.86):
 
k ikz  k' 2  k ikz z 
f ( z)  f 0 e  expi  'd '  f 0 e  exp i d   f 0 e ikz exp i  2 .
iz R  2 z  iz k 2
kR / 2 z
kR / 2 z

Since exp{i}  cos + i sin is an oscillating rather than a decaying function, the result of the upper-
limit substitution may not be immediately apparent. However, since this is an analytical function, we
can always replace i with (i – ), where  is a small real positive number. As a result, the upper-limit
substitution disappears, and now we can make, in the lower-limit substitution, the limit   0, getting
 kR 2 
f  ( z )  f 0 e expi .
ikz

 2z 
This result leads to a somewhat counter-intuitive answer

111Note that these results are quantitatively valid only at z >> R, i.e. for n << R/; however, since that last ratio
has to be large for the Huygens principle to be valid, the prediction of many maxima and minima is a sound one,
and may be readily verified numerically or experimentally.

Problems with Solutions Page 338

Essential Graduate Physics EM: Classical Electrodynamics

S  S0 .

Curiously, this result was first obtained in 1818 by Poisson who used it as an argument against
the theory of diffraction that had been developed by Fresnel – and against the wave theory of light as a
whole. The following optical experiments have shown, however, that the pattern of light diffraction on a
disk with R << z indeed features a light spot at the center.
Note also that in the Fraunhofer diffraction limit (i.e. at z >> kR2), this result may be obtained
directly from Eq. (*) in the model solution of the previous problem, and the Babinet principle (8.129).

Problem 8.19. Use the Huygens principle to analyze the Fraunhofer diffraction of a plane wave
normally incident on a square-shaped hole, of size aa, in an opaque screen. Sketch the diffraction
pattern you would observe at a sufficiently large distance, and quantify the expression “sufficiently
large” for this case.
Solution: As was discussed in Sec. 8.8 of the lecture notes, in the Fraunhofer limit, the
diffraction pattern is just a 2D Fourier transform of the initial wave amplitude distribution. In our case,
this means the complex amplitude of the scattered wave may be calculated as
a / 2 a / 2
f ( x, y )  const   dx'
a / 2
 dy' exp{i(
a / 2
x x'   y y' )} ,

with x = k (x/z), y = k (y/z), k = 2/. This is merely a product of

two similar 1D elementary integrals of complex-exponential (i.e.
sinusoidal) functions, giving finally
sin( x a / 2) sin( y a / 2) y
f ( x, y )  const 
(k x a / 2) (k y a / 2)
 kax   kay 
 const  sinc  sinc .
 2z   2z 
The figure on the right shows lines of equal intensity
x
(proportional to f 2), drawn at each 1% of its largest value. Note
that though the secondary maxima, centered at points {kx = n/a, ky = m/a} do exist for all integer
numbers n and m, they are extremely weak unless one of these numbers equals 1, while the other one is
not much larger than 1.
This result is valid for “sufficiently large” distances z in the sense that they are well beyond the
crossover from the Fresnel to Fraunhofer diffraction (see Sec. 8.6 of the lecture notes): z >> a2/ ~ ka2.

Problem 8.20. Use the Huygens principle to analyze the propagation of a monochromatic
Gaussian beam, described by Eq. (7.181) of the lecture notes, with the initial characteristic width a0 >>
, in a uniform isotropic medium. Use the result for a semi-quantitative derivation of the so-called Abbe
limit for the spatial resolution of an optical system,

wmin  ,
2 sin 

Problems with Solutions Page 339

Essential Graduate Physics EM: Classical Electrodynamics

where  is the half-angle of the wave cone propagating from the object and captured by the system.
Solution: Plugging Eq. (7.181) into Eq. (8.91), and using the Cartesian coordinates similar to
those shown in Fig. 8.11 of the lecture notes (see the figure below), we get

f  r  
k  ' 2  e ikR 2
f 0  exp
k
 
 
  x' 2  y' 2  e ikR
2i  
 d r'  f dx' dy' exp   , (*)
2i z '0
0
 2  R  2  R

where R = [(x – x’)2 + (y – y’)2 + z2]1/2, and   1/a02.

“source” point x' observation point

x
r’ = {x’, y’, 0} R  r - r' r = {x, y, z}

0 z

y' y

Since the Huygens principle, in its simple form (8.91), is only valid if z2 >> x2, y2, x’2, y’2 ~ a2,
we may make the approximations used at the derivation of Eq. (8.87) and reduce Eq. (*) to
k e ikz
f  r   f0 IxIy,
2i z
where Ix and Iy are two similar integrals, e.g.,

 x' 2 ik 2
I x   exp   x  x'  dx' .
  2 2z 
This is a typical Gaussian integral;112 it may be readily worked out by representing the exponent’s
argument as a full square of the sum (x’ + const), plus another term independent of x’:
x' 2
ik x' 2 ik 2 ik ik 2
  x  x'   
2
  x  xx'  x'
2 2z 2 2z z 2z
2
  ik 1 / 2 ik   ik 
1 / 2
  ik  k  2   ik  1 
     x'     x           x 2 .
 2 2 z  2z  2 2z    2 z  2 z   2 2 z  
Now we may argue that though the coefficient before x’ is a complex number, since its real part is
positive and the function under the integral is analytical, it may be worked out exactly as that of a real
argument:113

112 Such integrals are very frequently met in various fields of physics (and other qualitative sciences). For
example, they naturally arise at the analysis of the dispersion of 1D wave packets – see, e.g., QM Sec. 2.2.
113 Since the integration limits are infinite, the constant offset of x’, expressed by the second term inside the first
square brackets, does not affect the integral.

Problems with Solutions Page 340

Essential Graduate Physics EM: Classical Electrodynamics

 ik  k  2   ik  1  2       ik 1 / 2  
1 / 2 2
ik   ik 
I x  exp        x   exp    x'     x  dx'
 2 z  2 z   2 2 z       2 2 z  2z  2 2z   
 ik  k  2   ik  1  2   ik  1 / 2  
 exp        x     exp   2 d .  
 2 z  2 z   2 2 z    2 2 z  

The last integral114 equals 1/2, so forming the absolutely similar integral Iy, we finally get115

e ikz   ik 
1
 ik  k  2   ik  1  2 
   exp        x  y 
k 2 
f  r   f0
2i z  2 2z   2 z  2 z   2 2 z   
1
ikz  i z   i 2 z   2   2 z 2 
 f 0 e 1   exp    1  2 .
 k   2k 2  k 
Hence, the wave’s intensity is proportional to
1
  2z2     2z2 
f  r   f 0
2 2
1  2  exp  2 1  2 
 k    k 
1 2
(**)
2
2
   2  z 2
 a    2
1  2 4   f 0
z 2
 f 0 1  2 4  exp 2   exp 2 ,
 k a0   a 0  k a0   a0   a 
where a is the beam’s effective radius at the distance z from the initial plane:116
z2 z 2 2
a 2  a 02   a 2
0  . (***)
k 2 a 02 4 2 a 02
So, the beam retains its Gaussian shape but gradually broadens. The most interesting feature of
this effect is that the narrower the initial beam, the faster it broadens.117 Actually, this is not quite
surprising, and is just the standard feature of the Fraunhofer diffraction, as discussed in Sec. 8.6 of the
lecture notes. Indeed, the finite width a0 of the initial beam plays the same role as the finite width of the
orifice (or slit) at the usual diffraction – see Fig. 8.12. At relatively small distances, z/a0 << a0/, the
beam is the subject of the Fresnel diffraction. As we know from the lecture notes, this diffraction is
limited to small vicinities of sharp edges of the wave distribution. The Gaussian beam does not have
such edges and hence remains virtually intact. On the other hand, at large distances, where z/a0 >> a0/,
the beam is affected by the Fraunhofer diffraction, at which the effective half-angle   a/z of the
pattern’s divergence is independent of z – see, e.g., Eq. (8.110). Similarly, in this limit, Eq. (***) yields
a = z/2a0, i.e.

114 See, e.g., MA Eq. (6.9b).

115 Alternatively, this result may be obtained from the paraxial (or “parabolic”) wave equation f/z =
(1/2ik)2f, asymptotically valid in the limit f/kf  0, which was discussed in Chapter 7.
116 Note also the pre-exponential factor in Eq. (**), which ensures that the total power of the beam, proportional

to  f 2a2, is independent of z – as it has to be in a lossless medium.

117 A similar behavior, described by a very similar math, is pertinent to the 1D wave packets propagating in
dispersive media – see, e.g., QM Sec. 2.2.

Problems with Solutions Page 341

Essential Graduate Physics EM: Classical Electrodynamics

a 
  .
z 2a 0

Note that the smaller a0 the larger this angle – though the Huygens approximation used in this solution is
valid only if a0 >> , i.e. if the diffraction angle is small,  << 1.
Now rewriting the same result as

a min  a 0 
,
4
we see that at  << 1 (when our calculation is valid), it has the same functional form as the Abbe limit’s
formula.118 (Some difference in the numerical coefficients is due to the common definition of w, which
is larger than a.)
T
Problem 8.21. Within the Fraunhofer 1
approximation, analyze the pattern produced by a
diffraction grating with the 1D-periodic transparency
profile shown in the figure on the right, for the
normal incidence of a monochromatic plane wave. d 0 w d x
Solution: As was discussed in Sec. 8.8 of the lecture notes, within the Fraunhofer approximation
(valid when the observer’s distance z from the lattice is much larger than d2/), the diffracted wave’s
amplitude f() is just the 2D-spatial Fourier transform of the incident wave’s amplitude f (’) just
behind the lattice, i. e. of its transparency T(’)  f (’) /f0. In our 1D case, we have
 
 kx 
f  (ρ)  f  ( x)  const   T ( x' ) expi t x' dx'  const   T ( x' ) expi x' dx' .
   z 
Since in our case, the transparency is periodic: T(x’ + d) = T(x’), the diffracted wave’s intensity is
nonvanishing only for a discrete set of equidistant values
2 z 2 z
n  n, i.e. x n  n  n, with n  0,  1,  2,...
d k kd
which show up on a screen as a set of thin “interference stripes” (or “fringes”) parallel to those on the
diffraction lattice. Their intensities are
d w
 x'   x' 
f n  const   T ( x' ) exp2 i ndx'   exp2 i ndx'
0  d  0  d 
x'  w
d   x'   d   w  
  exp2 i n    exp2 i n  1 .
2 in   d   x' 0 2 in   d  
For analysis, it is convenient to represent this complex amplitude as a product of a real function
by a phase factor (which does not affect the wave’s intensity):

118Note that usually the Abbe limit is represented as wmin = 0/2nsin, where n is the refraction index defined by
Eq. (7.84), and 0 = n is the free-space wavelength.

Problems with Solutions Page 342

Essential Graduate Physics EM: Classical Electrodynamics

d wn  w  wn  w 
fn  sin expi n   sinc expi n .
n d  d  d  d 
Thus the stripe intensity follows the square of the sinc function:
2 wn
fn  sinc 2 ,
d
– see the solid red line in Fig. 8.8 of the lecture notes. Note that if d is a multiple of w, some interference
stripes disappear; a convenient choice is d = 2w, leading to the suppression of every other interference
fringe and thus doubling the wavelength range available for unperturbed spectroscopic analysis of the
diffracting wave.

q
Problem 8.22. N equal point charges are attached, at equal intervals, to a
circle rotating with a constant angular velocity about its center – see the figure on R
the right. For what values of N the system emits:
(i) the electric dipole radiation?
 2 / N
(ii) the magnetic dipole radiation?
(iii) the electric quadrupole radiation?
Solution: Let us number the charges sequentially by index k = 1, 2, …N;
then the angular position of the kth charge is
2
k  k  0 , where  0  t  const ,
N
so that in the reference frame with the origin in the circle’s center, and the z-axis normal to its plane, the
Cartesian coordinates of the charges are
 2   2 
x k  R cos  k  R cos k   0 , y k  R sin  k  R sin  k   0 , z k  0 . (*)
 N   N 
With this analytical representation, we are ready to address all particular tasks of the problem.
(i) Per the definition (3.6) of the electric dipole moment p, its Cartesian components, for this
system, are as follows:
N
 2 
p x  qR  cos k   0   qR c  N  cos  0   s  N sin  0 ,
k 1  N 
N
 2 
p y  qR  sin  k   0   qR s  N  cos  0   c  N sin  0 , p z  0,
k 1  N 
where
N
2 N
2
 c  N    cos k,  s  N    sin k. (**)
k 1 N k 1 N

Problems with Solutions Page 343

Essential Graduate Physics EM: Classical Electrodynamics

Mathematics tells us119 that both these sums equal 0 for any integer N > 0, with the sole
exception of c(1) = cos2 = 1. Hence only the system with one rotating charge can emit the electric
dipole radiation. (Indeed, its calculation was the task of Problem 3.) This result may be interpreted as a
destructive interference of coherent radiation by the elementary dipole moments qrk of N > 1 charges.
(ii) For a system of N point charges q moving with velocities vk, the definition (5.91) of the
magnetic dipole moment takes the form
q N
m   rk  v k .
2 k 1
For our system, the velocity vector vk of each charge is normal to both its radius vector rk and the z-axis,
so all vector products rkvk, and hence the net moment m, are directed along that axis:
q N qN 2
m  mn z , with m  
2 k 1
rk v k 
2
R .

Though this moment is not equal to zero for any N > 0, its time derivative is, so according to
Eqs. (8.137)-(8.139), the system does not emit the magnetic-dipole radiation at any N. (Note that the
answer might be different if the angular velocity  depended on time.)
(iii) Plugging Eqs. (*) into Eq. (8.142) for the electric quadrupole tensor, with all qk = q, we get

 3 cos 2  k  1 3 cos  k sin  k 0

N 
Q  qR 2   3 sin  k cos  k 3 sin 2  k  1 0
k 1 
 0 0  1
 1  3 cos 2 k  / 2 3 / 2sin 2 k 0

N 
 qR   3 / 2sin 2 k
2
1  3 cos 2 k  / 2 0 .
k 1 
 0 0  1
Besides time-independent terms (which, according to Eqs. (8.143) and (8.147), do not provide any
radiation), the tensor elements are proportional to the following sums:
N N

 cos 2 k   'c N cos 2 0   's N sin 2 0 ,

k 1
 sin 2
k 1
k   's N  cos 2 0   'c N sin 2 0 ,

where the trigonometric function sums differ from those defined by Eq. (**) by an additional factor 2
before the functions’ argument:
N
4 N
4
 c'  N    cos k,  's N    sin k.
k 1 N k 1 N
Due to this difference, the first sum does not vanish not only for N = 1, but also for N = 2:

119 See, e.g., MA Eq. (3.6) with n = N and  = 2/N. A simple alternative way to prove this property is to
represent the vectors Rk connecting the adjacent vertices of a regular N-side polygon (so Rk = const), by complex
numbers Rkexp{i(2k/N + const)}, and plug this representation into the condition that the polygon is a closed line:
kRk = 0. (Evidently, this trick does not work for the special case N = 1.)

Problems with Solutions Page 344

Essential Graduate Physics EM: Classical Electrodynamics

 'c 1  cos 4  1,  'c 2   cos 2  cos 4  2 ,

so the electric-quadrupole radiation is emitted not only by a single rotating charge (for which it is much
weaker than the electric-dipole radiation) but also by a pair of equal charges on the ends of the same
circle’s diameter (which does not emit the electric-dipole radiation).
Completing the calculation of the electric-quadrupole radiation for the latter case, by using the
above expression for the tensor Q together with Eqs. (8.143) and (8.147) of the lecture notes is a simple
but useful additional exercise, highly recommended to the reader.

Problem 8.23. What general statements can you make about:

(i) the electric dipole radiation, and
(ii) the magnetic dipole radiation,
due to a collision of an arbitrary number of similar non-relativistic classical particles?
Solutions:
(i) According to Eq. (8.26) of the lecture notes, the electric dipole radiation power is
proportional to the square of the second time derivative of the net electric dipole moment of the system,
p   q k rk .
k

But for similar particles with equal masses and charges, the first derivative of the moment is
proportional to the total linear momentum of the system:
q
p  q  rk   mrk .
k m k
Classical mechanics says120 that the momentum cannot change in time as a result of internal interactions
between the components of the system, and hence d2p/dt2 = 0 as well, so the collision cannot induce any
electric dipole radiation.
(ii) Very similarly, according to Eq. (8.139), the magnetic dipole radiation power is proportional
to the square of the second derivative of the net magnetic dipole moment m of the system. But according
to Eq. (5.95), for a system of particles with the same masses and electric charges, m is proportional to
the angular momentum L, and according to classical mechanics,121 cannot change in time as a result of
internal interactions, i.e. dm/dt = 0, and hence d2m/dt2 = 0, so the collision cannot induce any magnetic
dipole radiation either.
Note that at such conditions, the electric quadrupole radiation can still take place – see, e.g., the
two-particle example given at the end of Sec. 8.9 of the lecture notes.

Problem 8.24. Calculate the angular distribution and the total power radiated by a small planar
loop antenna of radius R, fed with ac current with frequency  and amplitude I0, into free space.

120 See, e.g., CM Eq. (1.10).

121 See, e.g., CM Eq. (1.38).

Problems with Solutions Page 345

Essential Graduate Physics EM: Classical Electrodynamics

Solution: If the antenna’s radius is sufficiently small, R <<   2/k = 2c/, the multipole
expansion of the loop’s magnetic field is dominated by the effect of its dipole moment m(t), directed
normally to the loop’s plane, with the magnitude given by Eq. (5.97) of the lecture notes:

mt   R 2 I t   R 2 I 0 cost   0 ,  t   R 2 It    2 R 2 I 0 cost   0  ,

so m

where 0 is a constant. Hence the angular distribution of the instant power of radiation into free space
may be calculated using Eq. (8.139) with Z = Z0 and v = c:
dP
 r 2 Sr 
Z0
 2 t sin 2  
m
Z0
 2 R 2 I 0  cos 2 t   0  sin 2  ,
2

d 4c 
2 2
4c 
2 2

where  is the angle between the direction of the vector m  (in our case, the symmetry axis of the
antenna’s loop) and the direction toward the observation point.
Averaging this expression over time, we get the following average power per unit solid angle:
dP
 r 2Sr 
1 Z0
 2
2
R 2 I 0 sin 2  
1
Z 0 I 02 kR  sin 2  ,
4

d 2 4c 2  2
32
and an easy integration (similar to that carried out for the electric dipole radiation in Sec. 8.2) yields the
total average power
1 1 8 
P  Z 0 I 02 kR 4  sin 2 d  Z 0 I 02 kR 4  Z 0 I 02 kR  .
4

32 4
32 3 12
Just as in the electric-dipole case analyzed in Sec. 8.2 of the lecture notes, this expression may be
used to calculate the effective impedance of the antenna:
P 
ZA   Z 0 kR   Z 0 .
4
2
I0 / 2 6
The impedance grows very fast (even faster than in the electric-dipole case) with the loop’s radius R
until that becomes comparable with the wavelength  = 2/k of the emitted radiation.

Problem 8.25. The orientation of a magnetic dipole, with a constant magnitude m of its moment,
is rotating about a certain axis with an angular velocity , with the angle  between them staying
constant. Calculate the angular distribution and the average power of its radiation into free space.
Solution: The rotating dipole moment vector m(t) may be represented as a sum of two
components: the time-independent component mz = mcos along the z
rotation axis, and the perpendicular component mz n
m (t )
m  t   m sin  n x cos   n y sin  , with   t  const, 

rotating about this axis (see the figure on the right), so y
 t 
m   t    m 2 sin  n x cos   n y sin  .
 t   m
0
m  (t ) x
This expression is absolutely similar to the formula for p t  , used in 

Problems with Solutions Page 346

Essential Graduate Physics EM: Classical Electrodynamics

the model solution of Problem 3, with the following replacement:

eR   m sin  .
Now we may use the fact that Eq. (8.139) for the magnetic dipole radiation is similar to Eq.
(8.26) for the electric dipole radiation, with the following replacement:
1
  m
p  .
c
Hence, the angular distribution of the average radiation power in these two problems is also similar:
1
S r  cos 2   sin 2  ,
2
where  is the angle between the axis of the moment’s rotation and the direction n toward the
observation point – see the figure above. Performing both these replacements in the expression for the
total average power obtained in the solution of Problem 3, we get
Z 0 4 m 2 sin 2 
Pmagnetic  . (*)
6c 4
Note that such a rotation of a magnetic dipole moment may be caused, in particular, by a
constant external magnetic field Bext directed along the z-axis, due to the torque  = mBext it exerts on
the magnetic dipole – see Eq. (5.101) and a discussion of the torque-induced precession.122 In this case,
the radiation characterized above leads to a gradual reduction of the potential energy (5.100), U = –
mBext = –mBextcos of the dipole-field interaction via a gradual increase of the angle , eventually
leading to the dipole’s alignment with the field (  , sin  0) and, according to Eq. (*), the
corresponding expiration of the radiation’s emission.

Problem 8.26. Solve Problem 12 (also in the low-frequency limit kR << 1), for the case when the
sphere’s material has a frequency-independent Ohmic conductivity , and opt = 0, in two limits:
(i) of a very large skin depth (s >> R), and
(ii) of a very small skin depth (s << R).
Solutions:
(i) In the limit s >> R, the eddy currents, which are responsible for the skin effect, may be
neglected – and hence their magnetic moment as well. For the electric dipole moment, we may use Eq.
(3.64) of the lecture notes with the replacements p  p, E0  E, and   ()/0:
   /  0  1
p   4 0 R 3 E .
   /  0  2
For an Ohmic conductor, the complex electric permittivity () may be calculated using Eq. (7.46), so
with opt = 0, we get123

122 See, e.g., CM Sec. 4.5.

123 See also the solution of Problem 15.

Problems with Solutions Page 347

Essential Graduate Physics EM: Classical Electrodynamics

 0  i /   /  0  1 1
p   4 0 R 3 E  4 0 R 3 E ,
 0  i /   /  0  2  1  3i r
where r  0/ is the charge relaxation time constant of the material – see Eq. (4.10) of the lecture
notes.124 Now Eqs. (8.28) and (8.39) of the lecture notes yield

8 kR 
2
8
4
1
 kR 4 R2  0, where  0  R 2 .
3 1  3i r 3 1  9 2 r2

This result shows that if r << 1, i.e. if the conductivity of the sphere’s material is relatively
high, the total cross-section of scattering is the same as max in the solution of Problem 12. It first grows
fast (as 4) with the wave’s frequency, but as  reaches the reciprocal relaxation time, this increase is
tempered by the factor 2 in the denominator.
(ii) In this limit, the skin effect does not allow the electromagnetic wave’s fields to penetrate into
the sphere’s bulk. As we know from the discussions in Secs. 3.4 and 5.6 of the lecture notes (which are
applicable to our current problem because of the low-frequency condition kR << 1), in this case, the
electric and magnetic fields induced by the sphere outside it coincide exactly with those of the point
electric and magnetic dipoles. Their moments are given, respectively, by Eq. (3.11):
p  4 0 R 3 E ,
and Eq. (5.125) in the ideal-diamagnetic limit   0:
m  2R 3 H ,
where E and H are the unperturbed fields of the incident wave.125 Since these fields change in time
coherently (see Eqs. (7.6)-(7.8) of the lecture notes),
Et    Z 0 n 0  H t  ,

so do the induced electric and magnetic moments, and their radiation fields – see Eqs. (8.24) and
(8.137), both with  = 0,  = 0, and hence v = c:
0     0 4 R 3 Z n  n  H
    0 R n  n  H
3
 ,
Be      0 4 0 R 3n  E
np
4rc 4rc 4rc
0 0 0 0
c2r
0 0     1  0 R n  n  H
3
Bm  n  n  m
    2R 3
n  n  H   .
4rc 2 4rc 2 2 c2r
(Here n0  k0/k and n  k/k are the unit vectors directed along the propagation of, respectively, the
incident and scattered waves, and the retarded time argument is implied.) These expressions show that
the induced fields are not only coherent but also comparable in magnitude, so to calculate the total
radiation power correctly, we have to add up the fields rather than their partial powers:

124 The Ohmic conductivity  participating in this expression should not be confused with the total cross-section
of scattering. Let me hope that these notions may be readily distinguished from the context.
125 As it follows from the derivation of these formulas in Secs. 3.4 and 5.6 of the lecture notes, their difference by
the factor (-1/2) is due to the different conditions for the electric and magnetic fields outside a field-free object, on
its boundary: E = 0 vs. Hn = 0 – see also Eq. (7.104).

Problems with Solutions Page 348

Essential Graduate Physics EM: Classical Electrodynamics

0 R3 
B 2 n  n 0  H
   1 n  n  H
  .
 (*)
c r  2 
To spell out the expression in the square brackets, let us use the x
coordinate system shown in the figure on the right, selected so that the
unit vector n0 is directed along the z-axis (so the vector H  , normal to n

H
n0, is within the [x, y] plane), and the vector n is within the [x, z] plane.
In these coordinates, each of the vectors has one zero component:


n  sin  , 0, cos  , n 0   0, 0,1,  cos  , sin  , 0 ,
  H
H 0 n0 z
so the result of the (straightforward) calculation of the double vector y
products participating in Eq. (*) is not too cumbersome:

 
  1 n  n  H
n  n0  H    cos  cos  1  1 cos  , sin   1  cos  , sin  cos  1  1 cos   ,
  H 
2   2  2   2 
giving
2
 2 2


n  n     1 n  n  H
H    H 2 1  1 cos   cos 2  cos 2   sin 2  sin 2     1  cos   sin 2   .
 0
2 
 2  2  
Now we may use Eq. (7.9b) to calculate the instant total power of the radiation:
2 2
 B  Z R6
4 n  n 0  H   2 n  n  H  d .
 1 
P  r  S r d  Z 0 r  
2 2
 d  0 4
4

4  0  c
 ,
Here the integration is nominally over all directions of the vector n, at a fixed direction of the vector H
but due to the spherical symmetry of the scatterer, it is equivalent to the usual integration over the
spherical angles  and  (see the figure above):
Z 0 R 6  2   1
2 2 2

P H 0 sin d 
0 d
1  cos 

 cos 2
 
cos 2
  sin 2
 sin 2
 
1
  cos 
 sin  
2

c4  2  2  
Z R 6  2   1  
2 2
 1 Z R6 10
 04 H   1  cos      cos    sin d  04 2 H 2 .
c  2
0   2   c r 3

Finally, the time average of this power (at the sinusoidal time dependence of the incident field’s H) is

 2   H 2 ,
4
H 
2
so by using the definition Eq. (8.39) of the total cross-section (with E = H/Z0), we get
10 R 6 4 10 4 6 10
   k R  kR   0 , where  0  R 2 .
4
4
3 c 3 3
This result should be compared with that of Problem 12 for a non-conducting dielectric sphere,
in the corresponding limit    (when the sphere’s interior is also fully screened from the penetration
of the incident field):

Problems with Solutions Page 349

Essential Graduate Physics EM: Classical Electrodynamics

8
  kR 4  0 .
3
The comparison shows that in contrast to electrostatics, where a conductor may be adequately described
as a dielectric with an infinite dielectric constant, in dynamics, the ac eddy currents flowing on its
surface add an extra 25% to the total cross-section of wave scattering. Note also that (as Eq. (*) clearly
shows), the polarization of the scattered wave’s component due to these currents is also different from
its component originating from the purely electric polarization.

Problem 8.27. Complete the solution of the problem started in q q

Sec. 8.9 of the lecture notes, by calculating the full power of radiation of
the system of two charges oscillating in antiphase along the same straight z
line – see Fig. 8.16, partly reproduced on the right. Also, calculate the d (t ) d (t )
average radiation power for the particular case of harmonic oscillations, d(t) = acost, compare it with
the case of a single charge performing similar oscillations, and interpret the difference.
Solution: Per Eq. (8.147), the full power of the electric-quadrupole radiation into free space is

 Q 
Z0 3
2
P jj ' .
1440 c 4 j , j '1

By using Eqs. (8.144) for the components of the quadrupole tensor in our case, we get
  d 3 2 2
 Z q2 2
2   d3 2    d3 2 
P
Z0
1440 c 4
2
  3 
 2 qd    3 4 qd   
0
  3 d t  . (*)
  60c
4
  dt   dt  dt 
For the particular case of harmonic oscillations, d(t) = acost, the time derivative is
d3 2 a2 d 3
3
d t   3
1  cos 2t   4a 2 3 sin 2t , (**)
dt 2 dt
so the average radiation power is
Z0q2 2 Z 0 q 2 a 4 6
P 
60 c 4
4 a 
2 3 2
 sin 2
2t 
15c 4
.

For a similar motion of a single charge q, the electric dipole radiation dominates. For it, we may
use Eq. (8.29) of the lecture notes:
Z 0 q 2 a 2 4
P1  .
12 c 2
The ratio of these two values of power is
P 8  a  2
   .
P1 5  c 
The fraction in the parentheses is just the ratio vmax/c, where vmax is the amplitude of velocity
oscillations. Now we should recall that all expressions of Chapter 8, starting from Sec. 8.2, and hence
the above results, are only valid for non-relativistic particles moving with velocity v << c. In this case,
the calculated electric quadrupole radiation power is much smaller than that of the electric-dipole
radiation.

Problems with Solutions Page 350

Essential Graduate Physics EM: Classical Electrodynamics

This difference is natural, because the two charges moving coherently, in antiphase, kill the
dipole component of each other’s radiation by destructive interference, and the quadrupole radiation is
just the main part of what remains of the total radiation after this cancellation. Note also that according
to Eq. (**), the quadrupole system of two charges radiates at frequency 2, while the dipole radiation of
each charge at the basic frequency  is completely canceled by the interference.

Problem 8.28. The system of four alternating charges located at the angles of a square,
considered in Problem 3.3(i), is now being rotated about the axis normal to their plane and passing
through the square’s center, with a constant angular frequency  << y
v/a. Calculate the time-averaged angular distribution and the total 2
power of the resulting radiation.
Solution: Let us consider an arbitrary instant t of the square’s 1
rotation, placing the coordinate origin into its center and directing the a R
x-axis to the initial position of one of the positive charges – see the t
figure on the right. Then basic trigonometry gives 0 x
x1  R cos t , y1  R sin t ,
x 2   R sin t , y 2  R cos t , 3
(*)
x3   R cos t , y 3   R sin t ,
4
x 4  R sin t , y 4   R cos t ,
where R = a/2, with all zk = 0. From Eq. (*), the Cartesian components of the electric dipole moment p
of the system are
4 4 4
p x   q k x k  0, p y   q k y k  0, p z   qk z k  0 ,
k 1 k 1 k 1

so p = 0, confirming the result obtained in the solution Problem 3.3(i) – or rather generalizing it for an
arbitrary rotation angle.
The same is true for the magnetic dipole moment m of the system. Indeed, at the rotation of our
rigid system about the z-axis, the individual charge velocities are126
v k  ω  rk , with ω  n z  ,

so by applying the bac minus cab rule127 to the discrete-charge version of Eq. (5.93), and taking into
account that in our system rk    rk  nz = 0, we get

m
1

2 k
1 1
 ω
q k rk v k   q k rk  ω  rk    q k ωrk2  rk rk  ω    q k rk2 .
2 k 2 k 2 k

Since in our system, all rk2 are equal (from Eqs. (*), rk2  xk2 + yk2+ zk2 = R2 = a2/2), and the sum of all qk
vanishes, the last expression yields m = 0.

126 See, e.g., CM Eq. (4.9). Alternatively, in our simple case this result may be readily obtained by differentiating
Eqs. (*) over time.
127 If you still need a reminder, see MA Eq. (7.5).

Problems with Solutions Page 351

Essential Graduate Physics EM: Classical Electrodynamics

Hence, the system’s radiation is dominated by its electric quadrupole component. Indeed, as was
shown in the solution of Problem 3.3(i), some components of the electric quadrupole tensor were
different from zero even for the special position considered there (corresponding to our current t = /4
+ 2n). Using Eqs. (*) to generalize that result to an arbitrary rotation angle, i.e. arbitrary moment of
time, we get
Qxx   q k 3 x k2  rk2   3a 2 q cos 2t , Qyy   q k 3 y k2  rk2   3a 2 q cos 2t ,
4 4

k 1 k 1
4
(**)
Qxy  Qyx  3 q k x k y k  3a q sin 2t ,
2
Qxz  Qzx  Qyz  Qzy  Qzz  0.
k 1

Now, defining the direction toward the radiation’s observer by the usual spherical angles  and ,
r
n  n x sin  cos   n y sin  sin   n z cos  , (***)
r
we may readily calculate the vector Q defined by the second of Eqs. (8.140), whose Cartesian
components are related to Qjj’ by Eq. (8.141):
3
Qj   Qjj' n j' .
j '1

A straightforward calculation of these components yields128

Q  3a 2 q sin   n x cos 2t cos   sin 2t sin    n y sin 2t cos   cos 2t sin  

 3a 2 q sin  n x cos2t     n y sin 2t    , 
so
Q  24a 2 q 3 sin   n x sin 2t     n y cos2t   .

Plugging this expression, together with Eq. (***), into Eq. (8.143) of the lecture notes, we may
calculate the magnetic field Hq  Bq/ of the quadrupole radiation at a distance r >> a from its source:
nx ny nz
1 qa  2 3
H q r, t   Qt r   n  sin  sin 2t r     cos2t r    0
24 rv 2
 rv 2
sin  cos  sin  sin  cos 
qa 
 
2 3
 sin   n x cos  cos2t r     n y cos  sin 2t r     n z sin  cos2t r  2  ,
 rv 2
where tr  t – r/v is the retarded time. This expression, accompanied by Eq. (7.6), Eq = ZHqn for this
locally-plane wave, determines all its characteristics. (In particular, note that the radiation frequency is
2 rather than , and its polarization is generally elliptical.) However, for our task of calculating the
time-averaged radiation power, we need just the sum of squares of the field’s Cartesian component
amplitudes:

128In hindsight, it is very plausible that the vector Q of this symmetric system should be a function of just one
azimuthal angle, 2t – . If we foresaw this fact in advance, we could simplify this calculation a bit by referring
time to the moment when this angle equals 0.

Problems with Solutions Page 352

Essential Graduate Physics EM: Classical Electrodynamics

 
2
 qa 2 3 
S r  Z H  Z 
2
q sin   cos 2  cos 2 2t r     cos 2  sin 2 2t r     sin 2  cos 2 2t r  2 
  rv
2

2 2
 qa 2 3  cos 2   cos 2   sin 2   qa 2 3  1  cos 4 
 Z  2 
 sin 2
  Z  2 
 .
  rv  2   rv  2
So, as we could expect, the average radiation power is axially symmetric and vanishes at  = 0,
i.e. on the z-axis normal to the square’s plane. Its total power is
 2 2
 qa 2 3  1  cos 4  8  qa 2 3 
Pq  r  S r d  Z 
2
2 
 2  sin  d   Z   ,
4   v  0
2 5  v 2

growing rapidly with both a and (especially) . As a sanity check, the last expression may be also
calculated from the general Eq. (8.147):
 
3
Z
4 
2
Pq  Qjj' ,
720v j , j' 1
which, with our Eqs. (**) for the tensor elements, immediately gives a time-independent value, because

 Q   Q   Q   Q   Q 

3
2 2 2 2 2
jj' xx yy xy yx
j , j' 1


 24a 2 q 3   sin
2 2
  2
2t  sin 2 2t  cos 2 2t  cos 2 2t  2 24a 2 q 3 ,
and hence does not require further temporal averaging.

Problems with Solutions Page 353

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 9. Special Relativity

Problem 9.1. Use the pre-relativistic picture of light propagation with velocity c in a Sun-
bound aether to derive Eq. (9.4) of the lecture notes.
Solution: As observed in the reference frame bound to a Michelson interferometer, in its
longitudinal arm (i.e. the arm directed along the aether’s velocity v) of length ll, light moves with еру
speed (c – v) in one direction and (c + v) in the opposite direction, so the roundtrip travel time is
ll l 2l 1
tl   l  l .
cv cv c 1 v2 / c2
The calculation of the roundtrip time tt in the transverse arm, of
length lt, is easier in the aether-bound reference frame. In this frame, y'
during a time interval tt, the interferometer passes distance (–vtt) in the lt
longitudinal direction, so the total length of the light’s roundtrip (see
the figure on the right) is
1/ 2
 2  vt t 
l  2 l t    .  vtt  vtt / 2 0 x'
  2 
In this frame, by the aether’s definition, the light’s velocity equals c, so we have to require that l = ctt.
Solving the resulting equation, we get
2l 1
tt  t .
 
c 1  v / c 2 1/ 2
2

For the roundtrip time difference t  tl – tt, these expressions yield the first form of Eq. (9.4):

2  ll lt 
.
t  

c 1 v / c
2 2

1 v / c2
2

1/ 2 

Now by Taylor-expanding both fractions on the right-hand side in small parameter (v/c)2 ~ 10-8,
in the linear approximation, we get

2   v2 v 2  2(l l  lt ) (2l l  lt )  v 
2
 
t  ll 1  2   lt 1  2      .
c  c   2c  c c c
Though at ll  lt the first term may be much larger than the second one, it does not change as a result of
the interferometer’s rotation relative to the aether’s motion direction and may be readily filtered out
from the data. On the contrary, the second term, which does not vanish even at ll = lt, describes the effect
we are looking for. For a nearly-symmetric instrument with ll = lt  l, it yields the second form of Eq.
(9.4):
2
l v
t    .
cc

Problems with Solutions Page 354

Essential Graduate Physics EM: Classical Electrodynamics

Problem 9.2. Show that two successive Lorentz space/time transforms, with velocities u’ and v in
the same direction, are equivalent to the single transform with the velocity u given by Eq. (9.25) of the
lecture notes.
Solution: Let us denote the normalized velocity u’/c as 1 and v/c as  2, with the corresponding
indices of the Lorentz factor  – see Eq. (9.17). Then per Eq. (9.19b), the first transform gives
x1   1  x   1ct , ct1   1 ct  1 x  ,

and for the second transform, we get:

x 2   2  x1   2 ct1    1 2  x   1ct    2 ct   1 x    1 2 1  1  2 x   1   2  ct ,
(*)
ct 2   2 ct1   2 x1    1 2 ct  1 x    2  x   1ct    1 2 1   1  2 ct   1   2  x  .

Let us spell out the product 12 involved in this result:

1 1 1 1
 1 2    . (**)
1    1   
1
2 1/ 2 2 1/ 2
2 1   1   
1
2 2
2
1/ 2
1   1
2
 22  12   22 
1/ 2

On the other hand, in this notation, Eq. (9.25) takes the form
1   2
 ,
1  1  2
so the corresponding Lorentz factor is129
1 1 1  1  2
    ,
1    2 1/ 2
  1   2 2 
1/ 2
1   1
2
 22   12   22 
1/ 2

1  2 
 1   1  2  
and by using Eq. (**), we may reduce the last relation to
   1 2 1   1  2  .

As a result, the direct transform from {x, t} to {x2, t2}, using these values of  and , is
   2 
x 2    x   ct    1 2 1   1  2  x  1 ct ,
 1   1  2 
   2 
ct 2   ct   x    1 2 1   1  2  ct  1 x ,
 1   
1 2 
giving the same results as Eqs. (*) that follow from two sequential transforms.

129Note that  is very different from 12, so the length contraction and time dilation formulas (9.20) and (9.21)
cannot be applied sequentially! (Explain why.)

Problems with Solutions Page 355

Essential Graduate Physics EM: Classical Electrodynamics

Problem 9.3. N + 1 reference frames numbered by index n (taking values 0, 1, …, N) move in the
same direction as a particle. Express the particle’s velocity in the frame number 0 via its velocity uN in
the frame number N and the velocities vn of the frame number n relative to the frame number (n – 1).
Solution: According to Eq. (9.25) of the lecture notes, applied to reference frames number (n – 1)
and number n:130
u n  vn   βn u v
u n 1  , i.e.  n 1  n , where  n  n , while β n  n , (*)
1  u n vn / c 2
1   nβ n c c

where un is the particle’s velocity measured in the nth reference frame. A sequential application of N
such formulas, starting from n = N all the way down to n = 1, would give us the required answer for u0,
but at large N the formulas would be rather bulky. However, let us use the definition of rapidity ,
mentioned in Sec. 9.1:
tanh    , (**)
to rewrite Eq. (*) as
tanh  n  tanh  n un v
tanh  n 1  , where tanh  n   n  , while tanh  n  β n  n .
1  tanh  n tanh  n c c
But the first of these relations is just the algebraic identity for tanh(n + n), so we may write simply
 n 1   n   n .
(This is why the notion of rapidity is so useful: these parameters add up as non-relativistic velocities
even in the relativistic case.) Now applying this formula N times, starting from n = N, we get
 N 1   N   N ,
 N  2   N 1   N 1   N   N    N 1   N   N   N 1 , ,
N
 0   N   N   N 1  ...  1    N    n .
n 1

What remains is to use Eq. (**) to rewrite the last result in terms of the usual velocities:
 u N
v 
u 0  c tanh tanh 1 N   tanh 1 n  . (***)
 c n 1 c 
For relatively low (non-relativistic) velocities we may use the fact that in this limit, both
involved functions, tanh and tanh–1, are well approximated by their arguments, so our general result is
reduced to the obvious equality
N
u 0  u N   vn .
n 1

In the opposite limit, when either one or several of the involved velocities approach c, the
corresponding tanh–1 functions become very large and the external tanh in Eq. (***) tends to 1, so we
get a natural answer: u0  c.

130 Let me hope that the difference in fonts used for n and n is sufficient to distinguish these two parameters.

Problems with Solutions Page 356

Essential Graduate Physics EM: Classical Electrodynamics

Problem 9.4. A spaceship moving with a constant velocity v directly from the Earth sends back
brief flashes of light with a period tS – as measured by the spaceship's clock. Calculate the period with
which an Earth-based observer may receive these signals – as measured by their clock.
Solution: Due to the time dilation effect, described by Eq. (9.21) of the lecture notes, the period
between the light flash emitting events, as measured by the Earth-bound clock, is tE = tS, where  
(1 – v2/c2)1/2  1. During such a time interval, the spaceship moves farther from the Earth by the space
interval x = vtE = vtS – again, as measured in the Earth's reference frame. Now, by the Earth’s clock
again, it will take the next signal, propagating with the speed of light, an additional time tA = x/c =
vtS/c  tS (where   v/c) to return to the point where the previous flash was emitted, so the total
interval between the received signals is
1/ 2
1  
t R  t E  t A  t S  t S  1    t S    t S .
1  
Note that the ratio tR/tS is expressed by the fraction reciprocal to that describing the Doppler
effect for frequencies – see Eq. (9.44) of the lecture notes, with the relevant (upper) sign, corresponding
to the motion of the wave source. This is natural because 1/t is just the (cyclic) frequency of the
periodic flashes. (In such a dispersion-free “medium” as free space, the group velocity of signal
propagation coincides with the phase velocity of phase propagation.) In the non-relativistic limit, v << c,
the ratio tR/tS approaches (1 + )  1, while at v  c, the ratio grows as 2  .

Problem 9.5. From the point of view of observers in a reference frame y y'
0', a straight thin rod, parallel to the x'-axis, is moving without rotation with a u'
constant velocity u' directed along the y'-axis. The reference frame 0' is itself
moving relative to another ("lab") reference frame 0 with a constant velocity v
along the x-axis, also without rotation – see the figure on the right. Calculate: v
0 x 0' x'
(i) the direction of the rod's velocity, and
(ii) the orientation of the rod on the [x, y] plane,
– both as observed from the lab reference frame. Is the velocity, in this frame, perpendicular to the rod?
Solutions:
(i) To calculate the components of the rod's velocity u in frame 0, we may use Eqs. (9.23) of the
lecture notes, for the particular case u'x = 0 and u'y = u'. The result is y n
u' 1 n
u x  v, uy  , where   , u
 1  v 2
/c 
2 1/ 2  /2
u
so the velocity's direction may be characterized by the angle u (see the figure
on the right) with 0 x
u y u'
tan u   . (*)
u x v
(ii) The simplest way to calculate the rod's orientation in frame 0 is to select the space and time
origins of both frames to coincide with each other and with the middle of the rod, at t = t' = 0. Then we
may use Eqs. (9.19) of the lecture notes, with {x', y' = u't'} being the Cartesian coordinates of another

Problems with Solutions Page 357

Essential Graduate Physics EM: Classical Electrodynamics

point of the rod, displaced by x' from its center, as measured in the "moving" frame 0'. To find the rod's
direction in the "lab" frame 0, we need to calculate the coordinates {x, y} of that point in this frame at a
certain fixed "lab" time t, for example, t = 0. With this substitution, Eqs. (9.19a) take the form
v
x    x'   ct' , y  u't' , 0  ct'   x' ,
with   ,
c
so the elementary substitution of the last result rewritten in the form t' = –x’/c into the first two
relations yields
x   x'   2 x'   ,
x' u'
y    x' .
 c
From this result, the angle describing the direction of the vector n normal to the rod (see the figure
above) obeys the relation
x c c2
tan  n     . (**)
y  u'  u'v
A comparison of Eqs. (*) and (**) shows that, generally, the angles u and n are different even
for non-relativistic velocities, and coincide (so the directions of the vectors u and n are the same) only
asymptotically, in the limits when either v  0 (so tanu  tann   and hence u  n  /2), or
u'  c (so tanu  tann  c/v  1/).

Problem 9.6. Starting from the rest at t = 0, a spaceship moves directly from the Earth, with a
constant acceleration as measured in its instantaneous rest frame. Find its displacement x(t) from the
Earth, as measured in the Earth’s reference frame, and interpret the result.
Hint: The instantaneous rest frame of a moving particle is the inertial reference frame that, at the
considered moment of time, has the same velocity as the particle.
Solution: Let us apply Eq. (9.25) of the lecture notes,
u'  v
u , (*)
1  u'v / c 2
to the velocities of the spaceship at the moment (t + dt), as measured, respectively, in the Earth frame (u)
and in the instantaneous rest frame of the spaceship (u’), if that frame’s velocity, as measured in the lab
frame at the moment t, is v. By the instantaneous frame’s definition, at dt = 0, u = v, i.e. u’ = 0, so for a
non-zero but small dt, u’  du’ is also small and we may Taylor-expand both parts of Eq. (*) with
respect to small deviations proportional to dt, at constant v = u(t). Keeping only the two leading terms of
each expansion, we get
du'  u t  1  du' / u t    1 u t    u 2 t  
u t   du   u t   u t 1  du'  
 u (t ) c 2    u t   du' 1  2 ,
1  du'u t  / c 2 1  du'u t  / c 2      c 
i.e.
 u 2 t  du'
du  du' 1  2   2 , (**)
 c  

where  is the Lorentz factor (9.17) of the spaceship at moment t. By definition, the acceleration a’
measured in the instantaneous frame of the spaceship (frequently called the proper acceleration) is

Problems with Solutions Page 358

Essential Graduate Physics EM: Classical Electrodynamics

equal to du’/d, where d = dt/ is the proper time interval corresponding to dt – see Eq. (9.60) of the
lecture notes. On the other hand, the acceleration a measured in the Earth’s frame is evidently du/dt, so
Eq. (**) yields
3/ 2
du 1 du' 1 du' 1  u2 
a     
a'  1  2  a' .
dt  2 dt  2 d  3  c 
Now we may integrate the resulting equation for the function u(t),
3/ 2
du  u 2  d (u / c) a'
 1   a' , i.e.  dt ,
dt  c 2  1  u 2
/c 
2 1/ 2 c
with a’ = const, over time, in a straightforward way:
d u / c  a'
 1  u 2
/c 
2 3/ 2

c 
dt  const .

Using the substitution u/c  sin , so d(u/c) = d(sin)  cos  d and (1 – u2/c2)3/2 = cos3, for our initial
condition u(0) = 0, we get
sin 1 u / c 
d  1 u  a't
0 cos 2   tan sin c   c ,
giving us the following equation,
dx  a't  a't
u  c sin  tan 1  ,
dt  c  1  a't / c 2 1 / 2  
for the distance x(t) we are seeking. From here, with our initial condition, x(0) = 0, we get
t a't / c  1/ 2

  a't / c c   a't  
2
d
c2 c2
 
2
tdt 1/ 2 
x  a'    1     1  2  1      1. (***)
0 1  a't / c  
2 1/ 2 a' 0
2 1/ 2 a'  0 a'   c 
  

As a sanity check, expanding the right-hand side of this formula in the Taylor series in small
time t << c/a’ (so u << c) and keeping only the leading term, we get the well-known non-relativistic
result:
2
c 2  a't  a't 2
x    .
2a'  c  2
For arbitrary t, Eq. (***) may be represented in a simple space-time symmetric form,131
c2
x  x0 2  ct 2  x02 , with x0  , (****)
a'

131 This result is frequently derived in a different, less direct way, by first integrating Eq. (**) multiplied by 2 
1/(1 – u2/c2), namely du/(1 – u2/c2) = du’  a’d, to get u = (c/a’)sinh(a’/c). Now this expression may be plugged
into the definition of , giving  = cosh(a’/c). Finally, the expressions for u and  may be plugged into the
relation dx/d  (dx/dt)(dt/d)  u, giving an easy differential equation for x as a function of . Its solution yields
x = (c2/a’)cosh(a’/c) – x0, so the identity cosh2 – sinh2 = 1, when applied to   a’/c, coincides with Eq.
(****). A useful byproduct of this alternative approach is an explicit relation between x and .

Problems with Solutions Page 359

Essential Graduate Physics EM: Classical Electrodynamics

which describes, on the [x, ct] plane, a hyperbola with /4-sloped x x  ct

asymptotes, offset by x0 from the origin – see the figure on the right. From
this plot, the constant x0 may be clearly interpreted as the travel distance
“lost” (at a fixed, large travel time t >> x0/c) because of the final value of
acceleration. For the Earth-surface gravity acceleration a’ = g  9.8 m/s2
(that would create the best comfort for a human astronaut :-), x0 = c2/g  0 ct
16
10 m – almost exactly one light-year. Hence, for the hypothetical multi-  x 0
light-year space missions, much higher accelerations would not create any
substantial distance benefits, and just exacerbate the hardest problem of future interstellar travel – that
of the initial mass of the rocket fuel, which is extremely challenging even for non-relativistic
velocities.132

Problem 9.7. Analyze the twin paradox for the simplest case of 1D travel with a piecewise-
constant acceleration.
Hint: You may use an intermediate result of the solution of the previous problem.
Solution: Let us send one of the twins to the space roundtrip consisting of the following four
equal-length stages:
– the first stage, starting from the rest at Mother Earth, of some duration t’ with a certain
constant acceleration a’, both as measured in the instantaneous rest frame of the ship,
– the second one, of the same duration, with equal but oppositely directed acceleration, bringing
the spaceship to a halt,
– the third stage of the same duration with the same acceleration as at the second stage
(essentially extending it by accelerating the ship toward the Earth), and
– the final deceleration stage, with the same t’ and  a’, to stop the ship at its return to the Earth.
Starting with the first stage, let us use the result for the ship’s velocity u as a function of time t,
both as measured in the “lab” (Earth’s) reference frame, that was derived in the model solution of the
previous problem:
a't
u .

1  a't / c 
2 1/ 2

The corresponding Lorentz factor is
1/ 2
1   a't  2 
   1    .

1  u / c 
2 1/ 2
  c  
Now we may use Eq. (9.60) of the lecture notes to calculate the proper time interval d of the spaceship,
i.e. the interval dt’ of the time measured in its instantaneous rest frame:
dt dt
dt'   .
 1  a't / c  2 1/ 2

Integrating this relation over the first stage’s duration, we get

132 See, e.g., the solution of CM Problem 1.11.

Problems with Solutions Page 360

Essential Graduate Physics EM: Classical Electrodynamics

Δt' t a't / c
dt c d c a't
t'   dt'  
0 0 1  a't / c 
2 1/ 2

a' 0 1    2 1/ 2

a'
sinh 1
c
,

so, finally, the stage’s duration as measured in the Earth’s reference frame is
c a't'
t  sinh .
a' c
Now we may use the time-symmetry of the Lorentz transform, which is the origin of the key
relation (9.60), to argue that the relation between t and t’ at the second stage of the trip is similar, and
the similarity of its two last stages with the first two ones to conclude that the full lengths of the
roundtrip, in both t and t’, are just four times those of the first stage. As a result, we may write
4c a't' full t full t' full 4c
t full  sinh , i.e.  sinh , where   .
a' 4c   a'
The last form of the result shows especially clearly that, first, the roundtrip’s time tfull measured
by the Earth-bound clock is always longer than (t’full) measured by the ship-bound clock, so the
traveling twin is always older at the reunion. Second, the difference between the twins’ ages is
substantial only if the time is much longer than the scale . At the acceleration perhaps most comfortable
for human travel, a’ = g  9.81 m/s2, this time scale is close to 1.22108 s  3.87 years. As a result if, for
example, t’full = 5 years, then the difference between the twin’s ages is marginal: tfull  6.5 years.
However, if the traveling twin spends t’full = 25 years (by their clock) for the roundtrip, then the
difference is rather dramatic: tfull  1,230 years.

Problem 9.8. Suggest a natural definition of the 4-vector of acceleration (commonly called the 4-
acceleration) of a point and calculate its components for a relativistic point moving with velocity u =
u(t).
Solution: In non-relativistic mechanics, the linear acceleration of a point is defined via its
velocity exactly as the latter variable is defined via the point’s radius vector:
dr du
, u
a .
dt dt
As was discussed in Sec. 9.3 of the lecture notes, the relativistic 4-velocity of a point is formed from its
4-component radius vector (9.48) similarly, “only” with the replacement of the non-relativistic time t
with the (reference-frame-independent) proper time  defined by Eqs. (9.59)-(9.60):133
dx  dt
u  , where x    cdt , dr, and dτ  , so u    c, u.
d 
Hence it is natural to define the 4-acceleration of the point as134

133Here, for certainty, all 4-vectors are in their contravariant form; as was discussed in Sec. 9.4, the covariant
forms differ only by the opposite sign before all their 3D-vector components.
134Note that with this definition, the key equation (9.145) of particle motion in an electromagnetic field takes the
form ma = qFu, very reminiscent of the non-relativistic 2nd Newton law.

Problems with Solutions Page 361

Essential Graduate Physics EM: Classical Electrodynamics

du  d
   c, u.
a  (*)
d dt
Here we should be careful. If we spoke about observations from two inertial reference frames
moving with a constant relative velocity v, as we did in the first two sections of Chapter 9, the Lorentz
factor  defined by Eq. (9.17) would be a function of that velocity (or rather its square) and hence a
time-independent parameter. However, at the introduction of the proper time of a moving point in Sec.
9.3, we essentially used the instantaneous frame – an inertial reference frame moving, at the considered
time instant, with the same velocity as the point itself, so the  participating in Eq. (9.60) and hence in
Eq. (*) is a function of u:
1
  .
1  u / c 2 1 / 2
2

Hence the differentiation in Eq. (*) should take this dependence into account. Representing a as {a0, a}
(and hence a as {a0, –a}),135 we get
d d 1 1 1 d 4
a 0   c   c  c u  u   u  u,
dt 
dt 1  u  u / c 2 
1/ 2

2 1 u2 / c2 
3/ 2
c 2 dt c
(**)
d d u  u d 1  4
a   u     u 1/ 2 
  2
u
  u u  u ,
dt 
dt 1  u  u / c 2 
1/ 2

 1  u 2 / c 2 
1/ 2

dt 1  u  u / c 2   c2
where
du 1 du
u   .
dt  d
In the non-relativistic limit case when u/c  0 (and hence   1), the second of Eqs. (**) is
naturally reduced to
du du
a  u   (***)
dt d
One may wonder whether the last result may be valid in the reference frame that is bound to the point in
question because, in such a frame, the point’s acceleration has to vanish. The answer is: since all special
relativity relations are only valid in inertial reference frames, Eq. (9.6) and hence Eq. (***) are only
valid in the instantaneous rest frame, i.e. the inertial frame whose time-independent velocity coincides
with the point’s time-dependent velocity u at the considered moment of time. Relatively to such a frame,
the point may indeed have a non-vanishing acceleration.
In the simplest particular case of a 1D motion when the vectors u and a are aligned: u = un, a =
an, Eqs. (**) give
4 du 4 u du 4 du du
a0  uu  u 3 , a   u  2
u 2 u   4 u   4 3 .
c c dt c d c 2
dt d

135Note that in the relativistic case, a so defined is not the point’s acceleration du/dt as observed from the rest
plane, just as the spatial component of the 4-velocity u is not the corresponding velocity u – see also the model
solution of Problem 6.

Problems with Solutions Page 362

Essential Graduate Physics EM: Classical Electrodynamics

Problem 9.9. Calculate the first relativistic correction to the frequency of a harmonic oscillator as
a function of its amplitude.
Solution: Combining the potential energy of the oscillator,
m 02 2
U x  x , 2

2 2
with the Taylor expansion of its kinetic energy in small u2/c2, and keeping just two first terms (see Eq.
(9.74) of the lecture notes), we would describe the usual harmonic oscillator with frequency 0
independent of the oscillation amplitude A. To get the first relativistic correction to this result, we need
to extend that expansion by one more term:
mc 2 mu 2 3mu 4
 mc 
2
 ,
1  u 2
/ c2 
1/ 2
2 8c 2

so (ignoring the motion-independent term mc2) we may approximate the full oscillator’s energy E as
mu 2 3mu 4 m 02 x 2
E   .
2 8c 2 2
This energy is an integral of motion,136 and hence is always equal to its value at maximum
deviations x = A from the equilibrium, when u  dx/dt = 0:
m 02 A 2
E .
2
Equating these two expressions for E and dividing all terms by m/2, we get the following relation
between u and x:
 3u 2 
u 2 1  2    02 A 2  x 2 .  
 4c 
With our assumption of the almost-non-relativistic motion, the second term in the first
parentheses is small, so we may plug into it the velocity calculated in the non-relativistic limit: (u2)0 =
02(A2 – x2), and get the 1st relativistic approximation:
 02 A 2  x 2 
u2  .
1  3 02 4c 2 A 2  x 2 
Plugging this approximation into the well-known expression for the period of oscillations in a
symmetric confining potential,137

136 This (almost evident) fact may be strictly proved using the full Lagrangian function L of the oscillator, which
may be formed by subtracting U from the free-particle’s Lagrangian given by Eq. (9.68). According to analytical
mechanics (see, e.g., CM Sec. 2.3), since the total L does not depend on time explicitly (L/t = 0), the
Hamiltonian function H of the system is an integral of motion: dH /dt = 0. Now calculating H exactly as it was
done at the derivation of Eq. (9.72) of the lecture notes, we may see that in such a system (even for an arbitrary
ratio u/c, and any time-independent U), H and E coincide, so dE /dt = 0 as well.
137 See, e.g., CM Sec. 3.3.

Problems with Solutions Page 363

Essential Graduate Physics EM: Classical Electrodynamics

A
dx dx dx
T   dt      4 ,
all period all period
dx / dt all period u 0
u
and then introducing a new, dimensionless integration variable   x/A, we get

4
A
1  3 2
4c 2 A 2  x 2   1/ 2
4
1
1  3 2
A 2 4c 2 1   2  1/ 2

T   dx   d
0 0

0 0 A 2
 x2 
1/ 2
0 0 1    2 1/ 2

1
 3 02 A 2  1
d 4 3 02 A 2
1

4

 0 0 1   2  
1
1
/ 2 1 2
 1   2  d 
4
0  1    2 1/ 2

 0 8c 2  1   
2 1/ 2
d .
8c  0 0

Both integrals may be readily worked out by the same substitution   sin (with d = cosd,
(1 –  ) = cos), giving
2 1/2

 /2  /2  /2
d  
1 1

 1   
1
 1  cos 2  d  4
2 1/ 2
 1      d  d   cos  d 
2
,
2 1/ 2 2 2
0 0 0 0 0

so, finally, the oscillation period is

2 2 3 02 A 2 2  3 02 A 2 
T    1  ,
0  0 16c 2  0  16c 2 
and the oscillation frequency
1
2  3 02 A 2   3 02 A 2 
   0 1     0 1  . (*)
T  16c 2   16c 2 
This formula (which is quantitatively correct only if the second term in parentheses is much
smaller than 1) shows that the oscillation frequency reduces with A – a natural result of the growth of
the particle’s relativistic mass M = m/(1 – u2/c2)1/2 with the increase of its velocity amplitude umax  0A.
Note that alternatively, Eq. (*) may be obtained by using the Lagrangian function (9.183) of the
system,
mc 2 m 02 x 2 mu 2 mu 4 m 02 x 2
L    mc 2   2  ,
 2 2 8c 2
to derive its (approximate) Lagrange equation of motion,
d  mu 3 
 mu  2   m 02 x  0,
dt 
i.e. x   02 x  
1 d 3
2
 
3
x   2 x 2 x , (**)
2c  2c dt 2c
with the right-hand side considered small, and then solving this equation using the standard van der Pol
method (alternatively called the “rotating-wave approximation” – RWA).138 The result is exactly the
same;139 however, for our particular case, when the right-hand side of Eq. (**) is a function of u  dx/dt
alone, the first approach used above is a bit simpler.

138 See, e.g., CM Sec. 5.2.

139 See, e.g., the solution of CM Problem 5.7.

Problems with Solutions Page 364

Essential Graduate Physics EM: Classical Electrodynamics

Problem 9.10. An atom with an initial rest mass m has been excited to an internal state with an
additional energy E, while still being at rest. Next, it returns to its initial state, emitting a photon.
Calculate the photon’s frequency, taking into account the relativistic recoil of the atom.
Hint: In this problem, and also in Problems 13-15 below, you may treat photons as classical
ultra-relativistic point particles with zero rest mass, energy E = , and momentum p = k.
Solution: Combining the energy conservation law,


mc 2  ΔE    (mc 2 ) 2  p 2 c 2 
1/ 2
,
and the momentum conservation law

p 0
c
(both in the lab reference frame, which, in this problem, is also the center-of-mass frame), we readily get

ΔE 1  ΔE / 2mc 2 ΔE
  .
 1  E / mc 2 
For real atoms (m ~ 10-26 kg, E ~ 1 eV), E << mc2, so the above expression is reduced to
ΔE  ΔE 
 1  .
  2mc 2 
Hence the recoil correction of frequency is typically relatively small (~10-10), but still may be
measurable for quantum transitions with small natural linewidth. This correction is dramatically reduced
at the Mössbauer effect,140 in which the role of m is played by the mass of the whole crystal lattice. This
effect serves as the basis of the high-resolution Mössbauer spectroscopy capable of detecting minute
changes in the chemical environment of the nucleus.

Problem 9.11. A particle of mass m, initially at rest, decays into two particles with rest masses
m1 and m2. Calculate the total energy of the first product particle, in the c.o.m. reference frame.
Solution: In the rest frame of the initial particle, before the decay, its momentum p was zero, and
energy was mc2. Hence, due to the energy and momentum conservation, in this frame (which is the
c.o.m. frame of the system), the 4-momenta of the decay products may be represented as
E   mc 2  E1 
p1   1 , p 1 ,
p 2   ,  p1 .
c   c 
Due to the Lorentz invariance of the scalar products of 4-vectors, the norm of each of these 4-
vectors has to be equal to its value in the corresponding rest frame, i.e. to (m1c)2 and (m2c)2, respectively
– see Eq. (9.78) of the lecture notes. This gives us a system of two equations for E1 and p1:

E1 2
 p  m1c  ,
2 2 mc 2
 E1 
2

 p12  m2 c  .
2
1
c2 c 2

140 See, e.g., H. Fraunfelder, The Mössbauer Effect, W. A. Benjamin, 1963.

Problems with Solutions Page 365

Essential Graduate Physics EM: Classical Electrodynamics

Now an elementary elimination of p1 yields

m 2  m12  m22 2
E1  c .
2m
As a sanity check, if both product particles are similar (m1 = m2), this result is reduced to E1 =
mc /2 – very naturally, because in this case, the initial energy mc2 is equally shared by the products.
2

Also, Eq. (*) is duly symmetric with respect to the index swap 1  2.

Problem 9.12. A relativistic particle with a rest mass m, moving with velocity u, decays into two
particles with zero rest mass.
(i) Calculate the smallest possible angle between the decay product velocities (in the lab frame,
in that the velocity u is measured).
(ii) What is the largest possible energy of a product particle?
Solutions:
(i) In the reference frame moving with the velocity of the initial particle (i.e. that of the center of
mass of the system), that particle does not move, so the total momentum conservation requires the decay
products to move in opposite directions, along the same straight line. Since per the basic Eq. (9.78), the
speed of a massless particle with nonzero energy has to be equal to y
c, the Cartesian components of the product particle velocities, as 1
measured in this c.o.m. frame, depend on just one free parameter, c sin 
say the angle  shown in the figure on the right. (Here the plane of 
the drawing is the common plane of the initial and final particle  c cos  0
velocities, and the coordinate axes are oriented so that u = unx.) c cos  x

Now we may use Eqs. (9.23) of the lecture notes, with v = u, 2  c sin 
to Lorentz-transform these velocity components into the lab frame:
c cos   u 1 c sin 
v x1  , v y1  ,
1  u cos  / c  1  u cos  / c
 c cos   u 1 c sin 
vx2  , v y2   .
1  u cos  / c  1  u cos  / c
Since the speed of both particles in the lab system is still c, the trigonometric functions of the angles 1,2
of their propagation in that frame (relative to the x-axis) may be calculated as cos1 = vx1/c, etc., giving
cos    1 sin   cos    1 sin 
cos 1  , sin 1  ; cos  2  , sin  2   , (*)
1   cos   1   cos  1   cos   1   cos 
where   u/c. Now we can calculate the angle –  1 – 2 between the two particle velocities in the lab
frame:
 2  cos 2  sin 2 
cos    cos1   2   cos 1 cos  2  sin 1 sin  2  
1   2 cos 2   2 1   2 cos 2  
 2 sin 2   1   2    1  2 sin 2 
  , where   .
 2 sin 2   1   2    1 1  2

Problems with Solutions Page 366

Essential Graduate Physics EM: Classical Electrodynamics

Since the function ( – 1)/( + 1) grows monotonically with  at all its possible (non-negative) values,
cos– is the largest, and hence – (if defined on the segment [0, ]) is the smallest, at sin2 = 1, i.e., at
the product particles’ propagation along the y-axis, as observed in the c.o.m. frame. In this case, we have

 2  1   2 
 cos 1 2 2  1  2 cos 1   2 cos 1
u
cos  max   2  2  1,   min .
 2  1   2  c
So, at low velocities of the initial particle, the smallest angle – tends to , i.e. to the value this
angle has in the c.o.m. frame, while at u  c, it gradually approaches zero.
(ii) Evidently, the largest difference between the product energies is achieved at  = 0, when 1 =
0 and 2 = , i.e. when the particles move along the same line (in the figure above, the x-axis), in the
opposite directions.141 In this case, we may employ the conservation of energy and of the x-component
of the system’s momentum in the lab frame, by using the basic Eqs. (9.70) and (9.73), with p = E/c for
each of the massless product particles:
E1 E2
mu   , mc 2  E1  E2 .
c c
Solving this simple system of equations, we get142
1/ 2
mc 2  u  mc 2 1   mc 2  1   
E1, 2  1       .
2  c 2 1   2 1/ 2 2  1  
 
The result shows that if the initial particle is non-relativistic (  0), its energy (in this limit,
close to the rest energy mc2) is always equally divided between those of the decay products, but if it is
ultra-relativistic (  1), almost all its energy may go to just one product particle.

Problem 9.13. A relativistic particle flying in free space with velocity u decays into two
photons.143 Calculate the angular dependence of the photon detection probability, as measured in the lab
frame.
Solution: Let  be the angle between the direction of propagation of the detected photon and the
vector of the velocity of the initial particle, as measured in the lab frame, while ’ be the similar angle
measured in the reference frame moving with the initial particle, i.e. with the velocity v = u. Then they
are related by the first of Eqs. (9.27) of the lecture notes, with v = u, ux = c cos, and u’x = c cos’:144
c cos  '  u cos  '   u 1
c cos   , i.e. cos   , with   ,   .
1  uc cos  ' / c 2 1   cos  ' c 1   2 1 / 2

141 On the contrary, as it follows from Eq. (*), at  = /2, i.e. at cos = 0, and sin = 1, the product particles
propagate at equal angles to the initial direction (1 = –2), and hence have equal energies.
142 Note a substantial similarity between the last expression and Eq. (9.44) describing the longitudinal Doppler
effect.
143 Such a decay may happen, for example, with a neutral pion.
144 Actually, this relation, in a different notation, is just the first of Eqs. (*) of the solution of the previous
problem, devoted to another aspect of a similar process.

Problems with Solutions Page 367

Essential Graduate Physics EM: Classical Electrodynamics

because the photon’s speed equals c in any reference frame. The reciprocal relation is
cos   
cos  '  . (*)
1   cos 
Next, in the reference frame moving with the initial particle, it was at rest, so in the absence of
any external polarizing field, the angular distribution of the resulting photons has to be isotropic. This
means that of the total N >> 1 photons resulting from a large set (the statistical ensemble145) of similar
but independent experiments, the number of photons going into any small solid angle interval d’ =
sin’d’d’ is the same: dN = N(d’/4)  N(sin’d’d’/4). From the point of view of the lab frame,
these dN photons go into the interval d = sindd, with  = ’ due to the axial symmetry of the
problem. Hence the probability density of the photon propagation at angle  is
1 dN 1 sin θ'dθ' 1 d cos θ' 
w     .
N d 4 sin θdθ 4 d cos θ 
This relation is general (for axially-symmetric systems); now using the particular Eq. (*) to calculate the
involved derivative, we finally get
1 1  2
w   .
4 1   cos  2
(As a sanity check, the integral of this function over the full solid angle, i.e. the total probability of
propagation of a photon in some direction, equals 1.)
This result shows that the photon detection probability is the largest in the direction of the initial
particle’s propagation,  = 0:
1 1 
w0  ,
4 1  
and that for ultra-relativistic particles ( >> 1), this enhancement may be very significant, scaling as 2.

p
Problem 9.14. A photon with wavelength  is scattered by an  me
electron, initially at rest. Calculate the wavelength ’ of the
scattered photon as a function of the scattering angle  – see the 
figure on the right.146 '
Solution: The problem may be most simply solved by writing
the laws of conservation of the total relativistic momentum (9.70) and the relativistic energy (9.73), both
in the lab reference frame:
k  k'  p, (*)


  me c 2  '  me c 2 
2
 p 2c 2 
1/ 2
, (**)

145 This notion is central not only for statistical physics, but for quantum mechanics as well, and is discussed in
detail in the QM and SM parts of this series.
146 This is the famous Compton scattering effect, whose discovery in 1923 was one of the major motivations for
the development of quantum mechanics – see, e.g. QM Sec. 1.1.

Problems with Solutions Page 368

Essential Graduate Physics EM: Classical Electrodynamics

where p is the electron’s momentum after the scattering event (see the figure above), and k = /c =
2/. From Eq. (*), p = (k – k’), so147
p 2   2 (k  k' ) 2   2 k 2  k' 2  2kk' cos   .

Plugging this expression into Eq. (**), and solving it for ’  2/k’, we get
2
'    1  cos     .
me c
The only fundamental constant, with the dimension of length, participating in this result,
2
 2.426  10 12 m ,
me c
is called the Compton wavelength.148 Note the relation between this length and the classical radius rc of
the same particle, given by Eq. (8.41):
rc e2 1 me c 1  1 e2  
    ~ 10 3 ,
2 / me c 4 0 me c 2 2 2  4 0 c  2
where
1 e2 1
 
4 0 c 137
is the fine structure constant,149 which characterizes the strength (or rather the weakness :-) of
electromagnetic interactions relative to the quantum and relativistic effects.

Problem 9.15. Calculate the threshold energy of a -photon for the reaction
γ  p  p  π0 ,
if the proton was initially at rest.
Hint: For protons, mpc2  938 MeV, while for neutral pions, mc2  135 MeV.
Solution: Similarly to the analysis of the p  p  p  p  p  p reaction in Sec. 9.4 of the lecture
notes, we may represent the conservation of the net 4-momentum of the system as150

(γ  p) (γ  p)  ( p  π 0 ) ( p  π 0 ) , (*)

where the particle names typeset in italics are used to denote the corresponding 4-momenta. After
opening the parentheses, we can evaluate the Lorentz-invariant scalar product p p on the left-hand side
of Eq. (*) as mp2c2, and the similar product for the -photon as zero because photons do not have rest

147 Note this useful (and frequently used) approach of avoiding the introduction of the direction of the vector p.
148 This historic name is rather misleading: no particle in the problem has this wavelength, and (to the best of my
knowledge :-) neither had Dr. A. Compton.
149 See also the solutions of Problems 5.18 and 8.9.
150 Here, just as in Eqs. (9.88)-(9.89) of the lecture notes, I am using the Italic font to denote the components of
the 4-momentum of each particle, leaving the Roman font to denote the particle as such in the reaction’s
description.

Problems with Solutions Page 369

Essential Graduate Physics EM: Classical Electrodynamics

masses. Similarly, the right-hand side of Eq. (*) is equal to (mp + m)2c2 because, at the reaction’s
threshold, its products have zero velocity in the c.o.m. frame. As a result, Eq. (*) becomes

2  p   mp2 c 2  mp  m π  c 2 ,
2

giving 2  p   mp  m π   mp2 c 2 .
2

In the lab system, in which the initial proton rests, its 4-vector p has only one nonvanishing component,
p0 = mpc, while the corresponding component 0 for the photon is just E/c. As a result, we get

E 
1
2m p
  
mp  mπ 2  mp2 c 2  mπ c 2 1  mπ

.

 2mp 
The second term in the parentheses represents the energy price we have to pay for not having the
center of mass of the whole initial system at rest. With the given rest masses of the involved particles,
this correction is close to 7% (i.e., quite noticeable!), and we finally get E  145 MeV.

Problem 9.16. Calculate the largest possible velocity of the electrons emitted by (initially,
resting) neurons at their -decays:
n  p  e  e .

Hint: Electron neutrinos e and antineutrinose are virtually massless (on the energy scale of
this problem); the rest energies E  mc2 of the other involved particles are as follows: 939.565 MeV for
the neutron, 938.272 MeV for the proton, 0.511 MeV for the electron.
Solution: The initial energy of the system, mnc2, is shared by all decay products, and only that of
the neutrino may be zero, due to its zero rest mass. Because of that, the largest energy (and hence
velocity) of the electron corresponds to the neutrino’s energy approaching zero. As a result, for the
purposes of this problem, we may ignore the contributions of this particle to the energy and momentum
of the finite system, and write their conservation laws as follows:151

En  Ep2   p p c 2  
 Ee2   p e c  
1/ 2 2 1/ 2
, 0  pp  pe .

The latter of these relations immediately yields pp2 = pe2. Denoting this magnitude as p2, and plugging it
into the former of the formulas, we get a single equation for (pc)2:

En  Ep2   pc 2  
 Ee2   pc  
1/ 2 2 1/ 2
.

This equation may be readily solved (as usual, by squaring both sides, and then singling out and
squaring the remaining square root product), giving
1/ 2
1  En  Ep  Ee  
2 2 2 2

pc    Ep2Ee2  . (*)
En  2 
 
 
In order to recalculate pc into the electron’s velocity, we may use the fundamental Eq. (9.70) of
the lecture notes:

151In this solution, for brevity, symbol E is used for the rest energies mc2 of the particles, rather than their full
energies. (High energy practitioners use the particle rest masses m for this purpose, implying units with c = 1.)

Problems with Solutions Page 370

Essential Graduate Physics EM: Classical Electrodynamics

u 
p  me u  me , i.e. pc  Ee .
1  u 2
/c 
2 1/ 2
1    2 1/ 2

Solving it for the reduced velocity   u/c, we get

pc / Ee
 . (**)
1   pc / E   e
2 1/ 2

Plugging the numbers provided in the Hint into Eqs. (*) and (**), we finally get pc  1.187 MeV,  
0.918.

Problem 9.17. A relativistic particle with a rest mass m and an energy E collides with a similar
particle, initially at rest in the laboratory reference frame. Calculate:
(i) the final velocity of the center of mass of the system, as measured in the lab frame,
(ii) the total energy of the system, in the center-of-mass frame, and
(iii) the final velocities of both particles (in the lab frame), provided that they move along the
same direction.
Solutions:
(i) Before the collision, the total momentum ps of the system in the lab frame equals that (p) of
the only moving particle, which may be calculated from the basic Eq. (9.78) of the lecture notes:
1/ 2
 E 2 
ps  p      (mc) 2  ,
  c  
while the total energy of the system is evidently
Es  E  mc 2 .
The center of mass of a system may be interpreted as an imaginary particle having the same total
momentum ps and the total mass Ms as the whole system. (In relativity,152 the latter is the dynamic mass
Ms = Es/c2.) Applying, to that particle, the general formula p = Mu, giving  = p/Mc = pc/E, we get

 com 
 E / c 2
 (mc) 2 1/ 2
c  E  mc 2 
  
1/ 2

, so  com 
1  E  mc 2
 


1/ 2

. (*)
2 
E  mc 2  E  mc  1  β 
2
com
1/ 2
 2mc
2

Note that although the particles are similar, com  /2 only in the non-relativistic limit; in the
opposite, ultra-relativistic limit (v  c,   1, E >> mc2), our result yields com 1, i.e. the center-of-
mass velocity is also close to c, while com  (E/2mc2)1/2 >> 1.
(ii) Applying the Lorentz transform formula (see, e.g., Eqs. (9.50)-(9.51)), valid for any 4-vector,
to the 0th component of the total system’s 4-momentum ps = {Es/c, ps, 0, 0}, we obtain

152 See the footnote at the end of Sec. 9.3 of the lecture notes.

Problems with Solutions Page 371

Essential Graduate Physics EM: Classical Electrodynamics

Es com E  mc 2  E  2  
1/ 2
 Es  
  com    com ps    com    com    mc    .
2

c c   c  c   
Plugging in the expressions for com and com from Eq. (*), we get153

Es com 
 2 com mc 2  2mc 2 (E  mc 2 )  1/ 2
. (**)

Note that for ultra-relativistic particles, E >> mc2, this energy (which is the parameter crucial for
new particle generation) grows only as the square root of that in the lab frame; this is why particle
colliders are more popular nowadays than the resting-target accelerators. (The caveat is that the colliders
should have a very high density of particle beams to ensure a sufficiently high collision rate.)
(iii) By the definition of the center-of-mass reference frame, the velocities of two similar
particles, observed from the frame, should be equal and opposite, and since if the total kinetic energy is
conserved at collision, their magnitudes ’ should be the same both before and after it. If the particles,
after the collision, move along the same direction as before it, this means that they just exchange their
velocities – see the figure below.
 '  '
before:
 '  '
after:

But this particular fact (of the particles exchanging velocities) should be true in any reference
frame because the Lorentz transform of a velocity does not depend on either the particle’s number or
whether the transformed velocity is the one before or after the collision. Hence, in the lab system, the
final velocities of the particles are also the same as the initial ones:
2 1/ 2
  mc 2  
   
cp 1
 
 E 2  (mc 2 ) 2 1 / 2  1     ,    0.
E E   E  

Problem 9.18. A “primed” reference frame moves, relative to the “lab” frame, with a reduced
velocity   v/c = nx. Use Eq. (9.109) of the lecture notes to express the elements T’00 and T’0j (with j =
1, 2, 3) of an arbitrary contravariant 4-tensor T via its elements in the lab frame.
Solution: For the relative motion of the frames along the x-axis, the Lorentz transform equations
have their traditional form (9.19), so the mixed Lorentz tensor has the form (9.98):

153 Another (even simpler) way to derive Eq. (**) is to require the norm of the total 4-momentum of the system,
(ps)(ps), to be Lorentz invariant:
Es com / c 2   ps com  Es  / c 2  ps2 ,
2 2 2

and then notice that the net momentum in the center-of-mass frame, (ps)com, is zero.

Problems with Solutions Page 372

Essential Graduate Physics EM: Classical Electrodynamics

    0 0
 
   

x'  0 0 1
 L   , where   .
x  0 0 1 0 1    2 1/ 2
 
 0 0 0 1 

Due to this abundance of zero elements, of the 16 terms in each implicit sum over the indices  and  in
the second of Eqs. (9.109), giving the requested transform, there are only 4 non-zero terms in
expressions for each of T’00 and T’01:
3
x' 0 x' 0  x' 0 x' 0 00 x' 0 x' 0 01 x' 0 x' 0 10 x' 0 x' 0 11
T' 00    x  T  x 0 x 0 T  x 0 x1 T  x1 x 0 T  x1 x1 T
 ,  0 x


  T 00     T 01     T 01      T 11   2 T 00   T 01  T 10    2T 11 , 
3
x' 0 x' 1  x' 0 x' 1 00 x' 0 x' 1 01 x' 0 x' 1 10 x' 0 x' 1 11
T' 01    x  T  x 0 x 0 T  x 0 x1 T  x1 x 0 T  x1 x1 T
 ,  0 x


    T 00   T 01      T 10     T 11   2 T 01   T 00  T 11    2T 10 , 
and only 2 nonvanishing terms for each of T’02 and T’03:

x' 0 x' 2  x' 0 x' 2 02 x' 0 x' 2 12

 
3
T' 02
   
T  0 T  1 T  T 02    T 12   T 02  T 12 ,
 ,  0 x x x x x x
2 2

and absolutely similarly, with the replacement 2  3:


T' 03   T 03  T 13 . 
As the simplest sanity check, at   0 (when   1), each tensor element is the same in both
reference frames.

Problem 9.19. Prove that quantities E2 – c2B2 and EB are Lorentz-invariant.
Solution: Both facts may be proved in a straightforward way, by writing these expressions in the
Cartesian components,

  
E 2  c 2 B 2  E x2  E y2  E z2  c 2 B x2  B y2  B z2 , E  B  E x Bx  E y B y  E z Bz ,

and then using Eqs. (9.134) of the lecture notes for the Lorentz transformation of each component.
A much shorter way to reach the same goal is to notice that according to Eqs. (9.125) and
(9.131) of the lecture notes,154
F F    2 E 2  c 2 B 2 ,
2 4
F G   E  B,
c c

154 Actually, the first of these facts was encountered in the lecture notes – see Eq. (9.217)

Problems with Solutions Page 373

Essential Graduate Physics EM: Classical Electrodynamics

so both considered combinations are double scalar products (multiplied by constants) and hence are
Lorentz-invariant.
Note that the first of these facts means that if  E  >  cB  in some reference frame, the same
relation holds in any other frame, while according to the second fact, if the fields E and B are mutually
perpendicular (as they are, e.g., in a plane electromagnetic wave) in some reference frame, they are also
perpendicular in any other frame.

Problem 9.20. Consider the situation when static fields E and B are uniform but arbitrary (both
in magnitude and in direction). What should be the velocity of an inertial reference frame in order to
have the vectors E’ and B’, observed from that frame, parallel? Is this solution unique?
Solution: The only special direction n defined by both vectors E and B is the one perpendicular
to both vectors:
EB
n . (*)
EB

It is natural to guess that at least one of the reference frames 0’, in which E B, should move along that
direction: v = nv. According to Eq. (*), for that direction, in the lab frame, E = 0, E = E, and B = 0, B
= B, so per Eq. (9.135) of the lecture notes, the fields measured in a moving frame are
E'    E   0, B'    B   0,

E'    E  v  B , 
B'    B  v  E / c 2 . 
The requirement to have the vectors E’ and B’ parallel may be represented as E’  B’ = 0, so for the
system’s velocity v, we get the following equation:

E  v  B  B  v  E / c 2   0 . (**)
Opening the parentheses, we can notice that all four operands of the 2
result are vectors directed along the vector n. Also, E(vE) = vE2n 
because the vector vE is perpendicular to E; similarly, (vB)B = –vB2n.
Finally, (vB)(vE) = v2BE because the angle between the vectors vB
1
and vE is equal to that between B and E. As a result, Eq. (**) is reduced to
a quadratic equation for v. This equation looks especially simple for the 
reduced velocity   v/c:
2( E / c )  B 0
 2  2 1   1  0, where κn  . )
( E / c) 2  B 2
This quadratic equation has two solutions: 
    1    2  1 .
1/ 2
1

For any values and directions of the vectors E and B, the parameter , by its
definition, is confined between –1 and +1, so one of the roots  (plotted as 
functions of  in the figure on the right) is always between –1 and + 1, and 2
1 0 1
hence a physically acceptable reference frame with    < 1, providing the 

Problems with Solutions Page 374

Essential Graduate Physics EM: Classical Electrodynamics

required property E’  B’, exists for any pair of fields.

Now we may argue that since E’  B’ in that frame, then according to the Lorentz transform
(9.135), in any other reference frame 0” moving relative to the frame 0’ with a velocity parallel to these
two vectors (again, as measured in the frame 0’, rather than in the lab frame!), the transverse fields
would not appear, i.e. any such frame also is a solution to our problem.
q1 u
Problem 9.21. Two charged particles moving with equal constant
velocities u are offset by a constant vector R = {a, b} (see the figure on the right), b
q u
as measured in the lab frame. Calculate the force of interaction between the 2
particles – also in the lab frame. a
Solution: Let us first calculate the electric and magnetic fields created by
charge q2 at the location of particle 1, as measured in the lab frame. Since these fields do not depend on
whether particle 1 moves or not, this task is equivalent to the problem solved at the end of Sec. 9.5 of
the lecture notes (see Fig. 9.11a), with the evident replacements ut  –a and q y
 q2. With these replacements, and with the same Cartesian coordinate q1 u
choice (see the figure on the right), the final forms of Eqs. (9.139)-(9.140) b
become
q2 u
q2 a q2 b
Ex  , Ey  , E z  0, a 0 x
4 0  2 a 2  b 2 3 / 2 4 0  2 a 2  b 2 3 / 2
u q2 b u 1
B x  0, B y  0, Bz   2 Ey , where   .
c 4 0  a  b 
2 2 2 2 3/ 2
c 1 - u 2
/c 2 
1/ 2

Now we may calculate both components of the Lorentz force (5.10), F12 = q1(E + uB), exerted
by these fields on particle 1. Since at our choice of coordinates, the calculated magnetic field B has only
one Cartesian component, Bz, and the velocity u also has only one component, ux = u, so the vector
product uB also has just one component, (uB)y = –uBz = –(u2/c2)Ey, the full Lorentz force has just two
nonvanishing components:
q1 q 2 a q1 q 2 b  u2  qq b
Fx  , Fy  1  2   1 2 .

4 0  a  b 2
2 2

3/ 2

4 0  a  b 2
2 2

3/ 2
 c 
 4 0   a  b 2
2 2
 3/ 2

The reciprocal force F21 may be calculated absolutely similarly, with the replacements a  –a
and b  –b, giving the same result as above but with the opposite signs of both Cartesian components –
thus complying with the 3rd Newton law. Note, however, that the ratio Fx/Fy equals 2a/b, i.e. coincides
with the ratio a/b only in the non-relativistic limit, so generally (in the lab frame), the forces between the
charges are not directed along the line connecting them. This is a natural result of the finite speed of
electromagnetic interactions’ propagation: one particle “does not know” where exactly the other particle
is in this particular instant.

Problem 9.22. Each of two thin, long, parallel particle beams of the same velocity u, separated
by distance d, carries electric charges with a constant density  per unit length, as measured in the
reference frame moving with the particles.

Problems with Solutions Page 375

Essential Graduate Physics EM: Classical Electrodynamics

(i) Calculate the distribution of the electric and magnetic fields in the system (outside the
beams), as measured in the lab reference frame.
(ii) Calculate the interaction force between the beams (per particle) and the resulting
acceleration, both in the lab reference frame and in the frame moving with the particles.
(iii) Compare the results and give a brief discussion of the comparison.
Solutions:
(i) In the reference frame moving with the particles, they are static, so there is no magnetic field:
B’ = 0. The electric field observed in that frame may be represented as a sum of two fields (each created
by one beam), y
E '  E'   E'  ,
d ρ

and each of these components may be readily calculated, for example, from 2 observation
the Gauss theorem applied to a round cylinder of radius  (see the figure on point
the right, in which the beams propagate along the x-axis, i.e. normally to the
0 z
plane of the drawing), with the corresponding beam taken for the cylinder’s ρ
axis: d
ρ  
E'   , 2
2 0  2
i.e., with the coordinate choice shown in the figure above:
  y  d / 2 z
E'  x  0, E'  y  E'  z 

2 0  y  d / 2   z 2
2
 ,

2 0  y  d / 2   z 2
2
.
155
Now using the Lorentz transform (9.134), in the lab system we also can write E = E+ + E–, with
  y  d / 2 z
E  x  0, E  y  E'  y   , E  z  E'  z  

2 0  y  d / 2   z 2
2
 
2 0  y  d / 2   z 2
2
,
u u z u u   y  d / 2
B x  0, B y   E'  z   , B z  E'  y 
c 2 2

c 2 0  y  d / 2   z 2 2
 c 2 2

c 2 0  y  d / 2 2  z 2 .
(ii) The Lorenz force exerted on one beam (say, the one located at y = +d/2 in the figure above)
comes only from the fields created by the other beam (located at y’ = –d/2).156 As a result, in the frame
moving with the particles,
q
F' y  qE'  y y   d / 2, z 0  , F' z  0 ,
2 0 d
while in the lab frame, the magnetic field contributes to the force as well:
F  q (E   u  B  ) y   d / 2 , z  0 ,
so we get

155 The Lorentz transform does not change the lengths perpendicular to the relative velocity, so y’ = y, z’ = z.
156 Of course, charged particles within each beam do interact (repulse each other), but the forces of those
interactions are directed along the beam. Also, these forces mutually cancel at their summation over the particles.

Problems with Solutions Page 376

Essential Graduate Physics EM: Classical Electrodynamics

  u 2  1 q
Fy  q ( E  y  uB z )  q 1    , Fz  q( E  z  uB y )  0 .
2 0 d  c 2   2 0 d
The resulting vertical acceleration, as measured in the moving frame (where M’ = m), is
F' y q
a' y   ,
m 2 0 dm
while in the lab frame, where M = m, it is
Fy 1 q a' y
ay    .
M  2 2 0 dm  2
(iii) So at u  c, the results for Fy and ay, obtained to the two reference frames, are very much
different. However. they are actually consistent because time runs differently in the two frames. For
example, if the acceleration produces a beam shift y << d, with the corresponding transverse velocity
still much below c, we may write
a' y a' y ay a' y
Δy'  Δt' 2  Δ 2 , Δy  Δt 2  Δt 2 .
2 2 2 2 2

Since, per Eq. (9.21) of the lecture notes, the “proper” time interval t’ =  and the lab frame interval
t are related as t =  (describing the relativistic time dilation), we get that y’ = y – as it should
be due to the Lorentz relation y’ = y between these transverse coordinates.
Note that alternatively, this problem may be solved by the summation (actually, integration) of
the result of the previous problem over all particles of a beam, with the same b = d but different a – the
additional exercise highly recommended to the reader.

Problem 9.23.
(i) Spell out the Lorentz transform of the Cartesian components of the scalar potential and the
vector potential of an arbitrary electromagnetic field.
(ii) Use this general result to calculate the potentials of the field created by a point charge q
moving with a constant velocity u, as measured in the lab reference frame.
Solutions:
(i) As was discussed in Sec. 9.5 of the lecture notes, the scalar potential  (divided by c) and the
three Cartesian components of the vector potential A form a legitimate 4-vector (9.116), whose
contravariant version is
    
A   , A    , Ax , Ay , Az  .
c  c 
Hence the Lorentz transform of the elements of this tensor follow the regular rule (9.91):
A  L A'  ,
(with the usual implied summation over all four values of the repeated index ), where L is the mixed
Lorentz tensor. If the x-axis is directed along the velocity v of the “primed” (i.e. “moving”) frame – as
shown, for example in Fig. 9.1 of the lecture notes, then the Lorentz tensor has the standard form (9.92):

Problems with Solutions Page 377

Essential Graduate Physics EM: Classical Electrodynamics

  0 0
 
    0 0 v 1
L   , with   , and   .
0 0 1 0 c 1   2 1 / 2
 
 0 0 0 1 

Performing the usual matrix-by-vector multiplication, we get an explicit form of the required transform:
 ' '
  A' x , Ax    A' x , Ay  A' y , Az  A' z .
c c c
Frequently, it is more convenient to use these relations rewritten in the semi-vector form similar
to Eqs. (9.135) for the fields:
 ' '
  A'   , A   β  A'   , A   A'  , (*)
c c c
where the indices  and  mark the vector components that are, respectively, parallel and perpendicular
to the vector v.157
(ii) Now let us use Eq. (*) to calculate  and A of a point charge q moving with velocity u
relative to the “lab” frame in that the potentials are measured. In the “primed” frame moving together
with the charge in its origin, i.e. with velocity u=v relative to the lab frame, the charge’s field is purely
electrostatic – see Eq. (1.35):
q
'  , A'  0 ,
4 0 r'
where r’ is the observation point (as measured in the moving reference frame), so Eqs. (*) yield
1 q ' v 1 q
  '   , A   β  , A  0 . (**)
4 0 r' c c 2 4 0 r'
In a certain sense, this is already the required solution, but for its applications, we may need to
spell out the potentials as explicit functions of the observation point’s position r and time t, as measured
in the lab frame. For that, we may decompose the radius vector r’ into the parallel and normal
components as well: r’ = r’ + r’, and then use the first three of Eqs. (9.19b), which may be rewritten
in a more compact semi-vector form similar to that used in Eq. (*):
r'     r   βct    r   ut , r'   r ,

so, with r = nxx and r = nyy +nzz, for the length r’ of the vector r’ we get


r'  r'  r'    r' 2
2
1/ 2

  2  x  ut   y 2  z 2
2

1/ 2
. (***)

If we select the coordinate origin just as it was done in Sec. 9.5 of the lecture notes for the
derivation of expressions (9.139)-(9.140) for the electric and magnetic fields in the same situation,
namely to provide x = 0 for the observation point (for example, as in Fig. 9.11a), then Eq. (***)
simplifies, and with its substitution, Eq. (**) yields

157 The reciprocal transform may be obtained either by solving the system of equations (*) for the primed
potentials, or by using Eq. (9.97b) with the reciprocal Lorentz matrix (9.98), or just by reversing the velocity sign.

Problems with Solutions Page 378

Essential Graduate Physics EM: Classical Electrodynamics

1 q u 1 q
  , A    , A  0 ,

4 0  2 u 2 t 2  b 
2 1/ 2

c 4 0  2 u 2 t 2  b 2
2

1/ 2

where b  (y2 + z2)1/2 is the “impact parameter”, i.e. the closest distance of the observation point from the
charge’s trajectory.
Comparing this result with Eqs. (9.139)-(9.140), we see that they involve the same denominator,
but the expressions for the potentials are substantially simpler than
those for the fields; in particular, they do not depend on the exact y y'
choice of the y- and z-axes within the plane normal to the vector nx b r'
(and hence u). However, we would need to make such a choice (for q x
example, taking y = b, z = 0, as in the figure on the right) in order to 0
0' v  u x'
derive the field formulas (9.139)-(9.140) from our result.158 ut'

Problem 9.24. Calculate the scalar and vector potentials created by a time-independent electric
dipole p, as measured in a reference frame that moves relative to the dipole with a constant velocity v,
with the shortest distance (“impact parameter”) equal to b.
Solution: In the lab frame, in whose origin the dipole rests, its scalar potential is given by Eq.
(3.7) of the lecture notes, while its magnetic field (and hence the vector potential) equals zero:
1 p r
 , A 0.
4 0 r 3
Applying to these potentials the Lorenz transform reciprocal to Eq. (*) of the previous problem’s
solution,
'  
    A  , A'     β  A  , A'   A  ,
c c c
we get
1 p r  v 1 p r
'     , A'     β   2 , A'   0 . (*)
4 0 r 3
c c 4 0 r 3
Just as Eq. (**) in the previous problem’s solution, this is already an answer, but it is natural to
express the potentials explicitly via the time t’ and the observation point’s position r’, as measured in
the moving system – especially since the vector r in the above formulas is an implicit function of time.
For that, we may use the standard coordinate choice shown in Fig. 9.1 of the lecture notes, and hence the
coordinate transform given by Eq. (9.19a):
x    x'  vt' , y  y' , z  z' .
From here,
r 2  x 2  y 2  z 2   2  x'  vt'   y' 2  z' 2 , p  r  p x x  p y y  p z z  p x   x'  vt'   p y y'  p z z' ,
2

158 For that, we would need to plug Eqs. (**), with r’ given by Eq. (***), into Eqs. (9.121) of the lecture notes,
spell them out in Cartesian components, perform the partial differentiation over x, y, z, and t, and only after that
set x = 0, y = b, and z = 0. This calculation is more math than physics, but it may still be recommended to the
reader as an additional exercise.

Problems with Solutions Page 379

Essential Graduate Physics EM: Classical Electrodynamics

so at the observation point with x’ = 0,159 y’ = b, and z’ = 0 (see Fig. 9.11 of the lecture notes), we get
1 p x vt'  p y b  v 1 p x vt'  p y b
'   , A' x   β   , A' y  A' z  0 .

4 0  2 v 2 t' 2  b 2 
3/ 2
c  
c 2 4 0  2 v 2 t' 2  b 2 3 / 2

Note that the Cartesian components px and py of the dipole moment, which participate in this expression,
are still as measured in the lab frame rather than in the moving frame. (Remember that in contrast to the
electric charge, the dipole moment is not Lorentz-invariant!)
One may run into claims that in the limit v << c when   1, this result is reduced to the sum of
the statically calculated potentials of the initial electric dipole p and a co-located magnetic dipole m =
vp. However, this claim is incorrect. This may be readily proved by applying the bac minus cab rule of
the vector algebra160 to the double vector product (vp)r that would participate in that case in Eq.
(5.90) of the lecture notes:
 m  r  0 v  p   r  v  p  r   p r  v 
A' m  A m  0   0 .
4 r 3
4 r 3
4 r3
Comparing this result with the two last Eqs. (*) which, in this limit, may be represented in a single
vector form:
v 1 p r  v  p  r
A'   2  0 , (**)
c 4 0 r 3
4 r 3
we see that A’  A’m besides the particular nearest-passage moment t = t’ = 0 when r  v.
Note, however, that the magnetic field B’ of the dipole in the moving reference frame, even in
this linear approximation in v/c << 1, is different from that statically calculated from the potential A’
given by Eq. (**).161 Such a calculation (a very good additional exercise, highly recommended to the
reader) shows162 that the resulting B’(r’, t’) may be decomposed into a sum of the statically calculated
field of a magnetic dipole with the moment m’ = vp/2 (i.e. twice smaller than the m mentioned above)
with a the displacement-current contribution of the field of a co-located electric quadrupole moment and
higher-multipole terms.

Problem 9.25. Solve the previous problem, in the limit v << c, for a time-independent magnetic
dipole m.
Solution: In the lab frame, in whose origin the magnetic dipole rests, its electric field and hence
the scalar potential equal zero, while its vector potential is given by Eq. (5.90) of the lecture notes:
0 m  r
  0, A . (*)
4 r 3

159 As in Sec. 9.5 of the lecture notes and in the previous problem (see the figure in its model solution), this
choice of x’ implies such a setup of the moving clock that at t’ = 0, the distance between the observation point and
the dipole is the smallest (equal to b).
160 See, e.g., MA Eq. (7.5).
161 Performing the calculation, one has to use the basic Eq. (6.7) in the form B’(r’, t’) = ’A’(r’, t’), i.e. take
into account the spatial dependence of the potential A’ at an arbitrary observation point r’ = {x’, y’, z’}.
162 See, e.g., V. Hnizdo, Am. J. Phys. 80, 645 (2012) and references therein.

Problems with Solutions Page 380

Essential Graduate Physics EM: Classical Electrodynamics

In the leading, linear approximation in  << 1, the Lorenz factor  = 1/(1 – v2/c2)1/2 may be taken for 1,
so the Lorentz transform relations used in the model solution of the previous problem,
'  
   A  , A'     β  A  , A'   A  ,
c c c
are reduced to just
'  
   A  , A'     β  A  , A'   A  . (**)
c c c
Since in our current problem,  = 0, the two last Eqs. (**) may be summarized as A’ = A, i.e. the
vector potential is described by the second of Eqs. (*) even in the moving system. On the contrary, as
the first of Eqs. (**) shows, the scalar potential ’ does not vanish even in the linear approximation in v:
 m  r  
'  cA   v 0 .
4 r3
Let us use the fact that the velocity vector v has only one (longitudinal) component, to rewrite the last
expression as
 v  m  r 
'   0 ,
4 r3
and then use the operand rotation rule of vector algebra163 to transform it to
 0 r  v  m  1  v / c 2  m  r
'     .
4 r3 4 0 r3
Comparing this expression with that for an electric dipole, in the same linear approximation in v << c:
1 p r
' e   e  ,
4 0 r 3
we see that they coincide if we take
vm
p . (***)
c2
Hence, in the low-velocity approximation, the electric potential of a static magnetic dipole m
equals, for a moving observer, to that of a co-located electric dipole with the moment (***). As was
discussed in the model solution of the previous problem, the reciprocal statement for an electric dipole
potential is incorrect. However, as in that problem, the electric field of a magnetic dipole, observed in
the moving reference frame (which should be calculated as E’ = –’’ – A’/t’, i.e. includes the
Faraday-induced component) is more complex and may be decomposed into a sum of the statically
calculated field of a twice smaller electric dipole p’ = p/2 and an electric-quadrupole field.164

Problem 9.26. Review the solution of Problem 6.23 (on the hypothetical magnetic monopole
passing through a superconducting ring) for the case when this particle moves with an arbitrary constant
velocity u.

163 See, e.g., MA Eq. (7.6).

164 See. e.g., http://kirkmcd.princeton.edu/examples/movingdipole.pdf.

Problems with Solutions Page 381

Essential Graduate Physics EM: Classical Electrodynamics

Solution: As was discussed in Sec. 5.6 of the lecture notes (and in the model solution of Problem
6.23), the magnetic point charge qm is usually defined so that if it is at rest at the origin, its magnetic
field at point r in the surrounding free space is
q
B  m3 r . (*)
4r
If the monopole moves with a constant velocity u, then Eq. (*), with the replacements B  B’ and r 
r’, is valid in the “primed” reference frame moving with it. Now let us see how the monopole’s field
looks from the lab frame. Directing the x-axis along the vector u, and the y-axis in such a way that the
observation point has coordinates {0, b, 0}, where b is its distance from the monopole’s trajectory (see,
e.g., either Fig. 9.11a in the lecture notes, or its reproduction in the model solution of Problem 6.23), and
using the Lorentz transform (9.135), we get the following analog of Eqs. (9.139):
qm u t q b
Bx   , By  m , Bz  0 , (**)
4 u  t  b 
2 2 2 2 3 / 2
4 u  t  b 2 3 / 2
2 2 2

where the moment t = 0 corresponds to the closest approach of the monopole to the observation point.
Now we may use the first of these results to calculate the monopole-induced magnetic flux
through a circular area with b  R:
bR
d (u 2  2 t 2  b 2 )
R R
qm bdb qm
 ext t   2  B x bdb   u t    ut  2 2 2 2 3/ 2
0 u  t  b  b  0 u  t  b 
2 2 2 2 2 3/ 2 4
0

qm 1 bR qm  ut 
 ut   2 22  sgn t  .
2 
u 2 2t 2  b 2 
1/ 2
b0 2 
 u  t  R 2 
1/ 2

This is the same result as was obtained in the  ext
solution of Problem 6.23, with the proper notation  qm / 2
replacement z’  ut (see the schematic plot in the figure qm
on the right), creating the same contradiction (with the
same Dirac-suggested resolution) as was discussed in that 0 t
 qm / 2
solution. Besides the instantaneous jump at t = 0, the lab-
frame time scale of the changes of the monopole’s flux (and the induced supercurrent) is t = R/u,
reflecting the relativistic time dilation (9.21b) by the factor   1/(1 – u2/c2)1/2 > 1. For a typical size
(~10 cm) of the superconducting loops used in the 1980s for the magnetic monopole search, and u ~ c,
this time scale is well below a nanosecond, so the experiments were looking for practically instant step-
like changes of the supercurrent induced in the loop, not trying to time-resolve their dynamics.

Problem 9.27. Re-derive Eq. (9.161) of the lecture notes for the simplest case p(0) = 0, by using
the 4-vector form (9.145) of the equation of motion and the notion of rapidity   tanh-1 that was
briefly discussed in Sec. 9.2.
Solution: To use Eq. (9.145):
dp 
 qF  u  , (*)
d

Problems with Solutions Page 382

Essential Graduate Physics EM: Classical Electrodynamics

let us direct the x-axis along the electric field, so E = Enx; then we may use the particular form (9.125a)
of the tensor F, with only two non-zero elements:
 0 E /c 0 0
 
 E / c 0 0 0
F   .
0 0 0 0
 
 0 0 0 0 

For the two non-zero components of the 4-momentum p = m{u0, ux, 0, 0}, Eq. (*) gives:165
du 0 E du x E
m  q ux , m  q u0 . (**)
d c d c
Admittedly, this system of two linear differential equations may be readily solved by elementary
means, but let us do this by using the rapidity  – just to illustrate some elegant (and useful) math this
notion enables.166 Since, by definition,
 1
tanh     and  2  , so  2     1,
2
,
 1  2

we can immediately see that

  cosh  ,   sinh  .
As a result, the two equations resulting from plugging the definitions of the 4-velocity components, u0 
c and ux  u  c(), into Eqs. (**),
d qE
  , d    qE  ,
d mc d mc
are equivalent to just one simple equation for the rapidity:
d qE
 ,
d mc
which may be very useful for the solution of problems with time-dependent electric fields. In our simple
case of a time-independent field and the motion starting from the rest (for  = 0, u = 0 and hence  = 0,
i.e.  = 0), the integration of this equation is elementary:
qE qE qE qE
 , so   cosh ,   sinh ,   tanh .
mc mc mc mc
Now we may use the above result for  to calculate the particle’s coordinate x as a function of
the proper time . Taking its initial position for the origin, we get
  
qE' mc  qE 
x    u x '  d'  c    d'  c  sinh d'  c  cosh  1 . (***)
0 0 0 mc qE  mc 

165In the second of Eqs. (**), the negative sign in the relation F10 = –E/c is compensated by that in the covariant
form of the 4-velocity u = {c, –ux, 0, 0}.
166 Also note the model solution of Problem 3 where the rapidity is used for analyzing a series of several
(possibly, many) sequential Lorentz transforms.

Problems with Solutions Page 383

Essential Graduate Physics EM: Classical Electrodynamics

What remains is to express x via the lab-frame time t. This may be done by integrating the basic Eq.
(9.60), dt = d, for our particular function ():
t  
qE' mc qE qE qEt
t   dt'    '  d'   cosh d'  sinh , i.e. sinh  ,
0 0 0 mc qE mc mc mc
and hence
1/ 2
qE  qEt  
2

cosh     1 .
mc  mc  
With this substitution, Eq. (***) gives the final result:
 qEt 2  1 / 2 
mc 2
  
x    1  1.
qE
 mc   
With the proper notation replacement x  z, it coincides with Eq. (9.161), provided that the initial
energy E0 is equal to mc2, as it has to be for a particle at rest.

Problem 9.28.* Calculate the trajectory of a relativistic particle in a uniform electrostatic field E,
for an arbitrary direction of its initial velocity u(0), by using two different ways – at least one of them
different from the approach described in Sec. 9.6 of the lecture notes for the case u(0)  E.
Solution: An elegant way to solve this problem is to integrate the 4-vector equation (9.145),
dp 
 qF  u  ,
d
directly, by taking the proper time  of the particle (rather than the lab-frame time t) for the argument.
For the three non-zero components of the 4-velocity u,167 this gives us three equations, which may be
conveniently rewritten as follows:
d c  d u x  d u z 
 u z ,  0,  c ,
d   d   d  
where   qE/mc is a constant parameter with the dimensionality of the reciprocal time (s–1). The
integration of the middle equation is elementary, and yields
cu x 0
u x  const  u x  C.
 0
c 2
 u x2 0   u z2 0  1/ 2

The remaining two equations may be combined (by an additional differentiation of any of them over 
and plugging the complementary equation into the result) to give two similar second-order linear
differential equations
d2 d2
c    c ,
2
u z    2 u z ,
d 2
d 2

with the solutions

167 Iam using the same coordinate system choice as in Sec. 9.6 of the lecture notes, with the z-axis directed along
the electric field, and the x-axis within the plane of the motion, so uy = 0 for any .

Problems with Solutions Page 384

Essential Graduate Physics EM: Classical Electrodynamics

 c  A cosh (   0 ),  u z  A sinh (   0 ), (*)

where 0 is the proper time at that uz = 0, i.e. the particle reaches the lowest value of z.
The constants A and 0 participating in these solutions may be found from the initial conditions

  0c    0u z 0 

2 2

 0c  A cosh  0 ,  0u z 0   A sinh  0 , so      1,

(**)
 A   A 
giving
1/ 2
 c 2  u z2 (0)   ( 0) 1 1 u (0)
A  c 2  c , with  z (0)  ;  0   tanh -1 z .
 c  u x ( 0)  u z ( 0) 
2 2
 z ( 0) 1  u z2 (0) / c 2 1 / 2
 c

From here, we can calculate the tangent to the particle’s trajectory:

dz u z A
  sinh (   0 ) ,
dx u x C
but to integrate this equation, we still need to express its right-hand side as a function of either x or z.
For example, we can find z() by integrating the second of Eqs. (*):
dz dz
u z     A sinh (   0 ) ,
dt d
yielding

A
z  A  sinh (   0 )d  cosh (   0 )  cosh  0 .
0

From this relation and Eqs. (**):
1/ 2
z z  (0)c z  z 
2

cosh (   0 )   cosh  0      z (0), sinh (   0 )     z (0)   1 ,
A A A A  A  
and now we can finally express the trajectory’s tangent via z:
dz A
 sinh (   0 ) 
dx C
1
C

z  A z (0) 2  A 2  1/ 2
,

and then find the trajectory itself:

z
dz C  z  
x C  cosh 1    z (0)   cosh 1 z (0) . (***)
0 z  A z (0)   A
2

2 1/ 2

  A  
This is a natural generalization of the result derived in Sec. 9.6 for the particular case uz(0) = 0.
Indeed, in that case, z(0) = 1, so the above expressions yield A = c(0), A/ = (0)mc2/qE  E(0)/qE, C/
= p(0)c/qE, and Eq. (***) is reduced to the form

p0c  1  qEz 
x cosh   1 ,
qE   E0 
equivalent to Eq. (9.165) of the lecture notes.

Problems with Solutions Page 385

Essential Graduate Physics EM: Classical Electrodynamics

An alternative way to solve this problem is to notice that Eq. (9.165),

E 0  qEx 
z  cosh  1 ,
qE  cp0  
obtained for the case u(0)  E, may be used in the general case as well, z
provided that we shift the origins of x, z, and t (see the figure on the E
z0 u(t 0 )
right),
E (0)  qE ( x  x0 ) 
z  z0   cosh  1 ,
qE  cp(0)  0 u(0) x0 x

and the “only” remaining thing to do is to express the parameters E(0), p(0), and z0 via the components
ux(t0) and uz(t0) of the new initial velocity, which in this notation replace ux(0) and uz(0) However, the
recalculation (leading to the same result as above) makes this approach not much easier than the first,
more elegant way. Still, it is recommended to the reader as a useful additional exercise.

Problem 9.29. A charged relativistic particle with the rest mass m performs a planar cyclotron
rotation, with velocity u, in a uniform external magnetic field of magnitude B. How much would the
velocity and the orbit’s radius change at a slow change of the field to a new magnitude B' ?
Solution: The initial orbit's radius may be found from Eq. (9.153) of the lecture notes:
p mu m u
R   .
qB qB qB 1  u / c 2 1 / 2
2

Thus, the initial magnetic field flux through the orbit's area A = R2 is
m2 u2
  R 2 B   .
q2B 1 u2 / c2
For the new field value B' , the flux is described by a similar formula:
m2 u' 2
'  R' 2 B'   ,
q 2 B' 1  u' 2 / c 2
where u’ is the new velocity of the particle.
As was argued in Sec. 9.7 (in the context of slow spatial variations of the magnetic field), the
magnetic flux through the orbit is an adiabatic invariant of the particle’s motion. These arguments
remain valid for sufficiently slow temporal variations of the field as well, so we may write ' = ,
giving us the following equation for finding u':
1 u' 2 1 u2
 .
B' 1  u' 2 / c 2 B 1  u 2 / c 2
This is a linear equation for u' 2, easy to solve. The result,
1 / 2
 B u2  B 
u'  u   2 1   ,
 B' c  B' 

Problems with Solutions Page 386

Essential Graduate Physics EM: Classical Electrodynamics

becomes simpler in both the non-relativistic limit: u'  u(B'/B)1/2, and the ultra-relativistic limit: u'  u 
c. At arbitrary u, the field’s increase results in an increase of the particle’s velocity. Physically, this
change of the particle’s velocity, and hence of its energy, is the result of Faraday’s induction e.m.f. (6.2)
appearing during the field’s change.
For the orbit's radius, the same adiabatic invariance of the flux, ' = , provides the equation
B'R' 2  BR 2 ,
which gives a simpler result independent of the particle's velocity:
1/ 2
B
R'  R  .
 B' 
It shows that a field’s increase leads to a shrinkage of the orbit.

Problem 9.30.* Analyze the motion of a relativistic particle in uniform, mutually perpendicular
fields E and B, for the particular case when E is exactly equal to cB.
Solution: Evidently, for this particular case, both velocities defined by Eqs. (9.168) and (9.174)
equal c, so the introduction of a reference frame moving with this velocity
(the trick used in Sec. 9.6(iii) of the lecture notes) is not too helpful, so we y
better start from scratch.
E
Directing the coordinate axes as in Fig. 9.12 of the lecture notes
(partly reproduced on the right), so that Ey = E and Bz = B = E/c are the only
non-zero field components, for an arbitrary velocity u of the particle we get B 0 x
uB = nxuyE/c – nyuxE/c, so Eq. (9.144), decomposed into three scalar z
equations, gives
dp x uy dp y  u  dp z
 q E,  qE 1  x ,  0, (*)
dt c dt  c  dt
while Eq. (9.148) gives the following equation for energy E:
dE
 qEu y . (**)
dt
Combining the first of Eqs. (*) with Eq. (**), we get

E  cp x  const  a, (***)
while the last of Eqs. (*) yields pz = const as well, so we may form from it another (positive) constant of
motion, (mc2)2 + c2pz2  b  0. Plugging these two constants (a and b) into the basic Eq. (9.78) for the
relativistic energy of the particle, rewritten as

E 2  c 2 p x2  E  cp x E  cp x   (mc 2 ) 2  c 2 p y2  c 2 p z2 ,
we get
a E  cp x   c 2 p y2  b . (****)

Problems with Solutions Page 387

Essential Graduate Physics EM: Classical Electrodynamics

Now considering Eqs. (***) and (****) a system of two linear equations for E and px, we may
express these variables via py (and the constants of motion):
c 2 p y2  b a c 2 p y2  b a
E  , cp x   .
2a 2 2a 2
The first of these relations may be plugged into the product of E by the derivative dpy/dt, expressed from
the differential equation (*) for py, and then transformed by using the universal relativistic relation p =
(E/c2)u, which was mentioned in Sec. 9.3:
dp y
 u   u 
 E qE 1  x   qE  E  E x   qE E  cp x   qEa .
E (*****)
dt  c   c 
The result of this substitution,
 c 2 p y2  b a  dp y d  c2  b a 
     p 3y     p y   qEa .
 2a 
2  dt dt  6a  2a 2  

may be readily integrated over time to obtain the time dependence of py in an implicit form:
c2 3  b a 
py     p y  qEa t  t 0  ,
6a  2a 2 
where t0 is the time moment (determined by initial conditions) when py = 0.
Even without solving this cubic equation, we may use the relation dt/E = dpy/qEa (which follows
from Eq. (*****) above) to express the particle’s coordinates as functions of py:
c  c p y  b a 
2 2
c 2 px c c  c2 3  b a  
x   u x dt  
qEa  qEa   2a
dt  cp x dp y   dp y   py     p y   const,
E 2  qEa  6a  2a 2  
c2 py c c2
y   u y dt  
qEa 
dt  cp y dp y  p y2  const,
E 2qEa
c2 pz c c 2 pz
z   u z dt  
qEa 
dt  cp z dp y  p y  const .
E qEa
Since, according to the last formula, z is a linear function of py, the first two of these relations
yield the particle’s trajectory in the form x(z), y(z). Note that at large times, the magnitude of py (and
hence of z) grows at (t – t0)1/3, the coordinate y grows faster, as (t – t0)2/3, while the motion along the x-
axis (i.e. in the direction perpendicular to both vectors E and B) is the fastest one:
c3
x p 3y  c(t  t 0 ) .
6qEa 2
This is essentially the drift that was discussed in Sec. 9.6(iii), whose velocity formally obeys
both Eqs. (9.168) and (9.174), but since in the reference frame moving with this velocity (v = c), both
the electric and magnetic fields disappear, the frequency (9.172) vanishes, so the particle does not
perform the simultaneous cyclotron motion.

Problems with Solutions Page 388

Essential Graduate Physics EM: Classical Electrodynamics

Problem 9.31. Find the law of motion of a relativistic particle in uniform static fields E and B
parallel to each other.
Solution: If E  B, then even for an arbitrary direction of the initial momentum p(0) of the
particle, we can always select such a reference frame that
E  0, 0, E, B  0, 0, B, p(0)  p x (0), 0, p z (0).
Then the 4-vector equation (9.145) of the particle’s motion,
du  q
 F  u  ,
d m
where F is given by Eq. (9.125a) of the lecture notes, may be separated into four scalar components as

d (c) d (u x ) d (u y ) d (u z )

 (u z ),   (u y ),   ( u u x ),  (c) , (*)
d d d d
where   qE/mc, and   qB/m.168 The first and the last Eqs. (*) form an independent system, which
may be readily reduced to one second-order equation (by the usual trick of differentiation of one of the
initial equations over , and then plugging the counterpart equation into the result).169 As a result, we get
two similar equations,
d2 d2
c    2
c , u z    2 u z ,
d 2
d 2

which may be readily integrated, with integration constants determined by the initial values of  and uz.
The result is
 (0) cosh   sinh 
 ( )   (0)cosh    z (0) sinh  , u z ( )  c z , (**)
cosh    z (0) sinh 
where (0) = [1 – x2(0) – z2(0)]–1/2; it describes an accelerated motion of the particle in the z-direction.
The second and third of Eqs. (*) form a similar system, in which the electric field does not
participate. It is formally the same as for the non-relativistic cyclotron motion and also may be readily
integrated. With the appropriate choice of origins of the proper time  and of the x-axis (the last one in
the plane of the initial velocity), we get
p x ( 0) p x ( 0)
u x  cos  , u y   sin  .
m m
Since ux  dx/dt = dx/d (and similarly for y), these equations may be readily integrated again, giving
a simple cyclotron orbit170
x  R sin   const, y   R cos   const , (***)

168 Note that the  so defined is not the usual relativistic cyclotron frequency (9.151), c = qB/M  qB/m. Indeed,
the denominator in the definition of  does not include the Lorentz factor  = (1 – 2)–1/2, as it would if we dealt
not with the proper time  of the particle, but with the lab time t.
169 Note that since this system of two equations does not depend on the magnetic field, it gives an alternative way
of solving the problem discussed in Sec. 9.6(ii) of the lecture notes.
170 Note that if qB > 0, i.e.  > 0, the particle rotates in the “negative” (clockwise) direction.
z

Problems with Solutions Page 389

Essential Graduate Physics EM: Classical Electrodynamics

whose radius R = px(0)/qB is not affected by either the initial momentum or the acceleration of the
particle in the z-direction.
Note, however, that at E  0, the proper time  is not proportional to the lab time t. Indeed, their
relation may be found by plugging the first of Eqs. (**) into the fundamental relation dt = d, and then
integrating it. With the natural choice t = 0 at  = 0, this simple integration gives
 0
t sinh    z (0)cosh   1 .

As a result, as viewed from the lab frame, the frequency of the particle’s rotation is not constant. In
particular, at large times ( >> 1/), the proper time grows only as a logarithm of the lab time, so the
cyclotron motion gradually slows down. Physically, this happens due to the gradual increase of the
relativistic mass M = m of the particle – because of a continuing increase of its energy by the electric
field.

Problem 9.32. An external Lorentz force F is exerted on a relativistic particle with an electric
charge q and a rest mass m, moving with velocity u, as observed from some inertial “lab” frame.
Calculate its acceleration as observed from that frame.
Solution: According to Eqs. (9.70) and (9.144) of the lecture notes, we may write
d
u   F, where F  qE  u  B  .
m (*)
dt
Let us spell out the derivative on the left-hand side of this equation:
 β d 2 
d
u   c d β   c d β
dt 1   2 1 / 2
 c  
β
  .
 1   2  21   2  dt
1/ 2 3/ 2
dt dt 
For what follows, it is convenient to differentiate  2 in the last term as the scalar product , getting
  
d
u   c  β2 1 / 2  β
    
2β  β   c 3 1   2 β  β β  β ,
 1    21   2 
3 / 2
dt 
so the first of Eqs. (*) becomes
   
mc 3 1   2 β  β β  β  F . (**)

To solve this equation for β , let us scalar-multiply both sides by , getting

    
mc 3 1   2 β  β   2 β  β  mc 3 β  β  β  F, so β  β 
βF .
mc 3
(***)

Plugging the last relation back into Eq. (**), we get a simple equation,
 βF 

mc 3  1   2 β  β 3
 mc β  ββ  F   F ,
 mc 
which yields an explicit formula for the acceleration:

Problems with Solutions Page 390

Essential Graduate Physics EM: Classical Electrodynamics

F  ββ  F  F  ββ  F 
β  , i.e. a  u  cβ  . (****)
mc m
In the non-relativistic limit   0, the second term in the numerator is negligible and   1, so
this result is reduced (as it should) to the usual 2nd Newton law a = F/m, with the rest mass m. In the
relativistic case, the acceleration depends on the applied force’s direction: if it is normal to the particle’s
velocity (F  ), then (F) = 0, so a = F/m  F/M, where M is the relativistic mass m. However, for
aligned vectors  and F, Eq. (****) yields the result a = (1 – 2)F/m  F/m3  F/M2, showing that due
to the Lorentz transform, in the ultra-relativistic case  >> 1, the longitudinal acceleration (as observed
from the lab frame) is even much lower.

Problem 9.33. Neglecting relativistic kinetic effects, calculate the lowest voltage V that has to be
applied between the anode and cathode of a magnetron (see Fig. 9.13 and its discussion in Sec. 9.6 of
the lecture notes) to enable electrons to reach the anode, at negligible electron-electron interactions
(including the space charge effects) and collisions with the residual gas molecules. You may:
(i) model the cathode and anode as two coaxial round cylinders, of radii R1 and R2, respectively;
(ii) assume that the magnetic field B is uniform and directed along their common axis; and
(iii) neglect the initial velocity of the electrons emitted by the cathode.
After the solution, estimate the validity of the last assumption, and of the non-relativistic approximation,
for reasonable values of parameters.
Solution: In the non-relativistic limit, we may restrict the Taylor expansion
1/ 2
1  u2  u2
 1  2   1  2  ...
  c  2c
to the two leading terms spelled up above, and hence approximate Eq. (9.183) of the lecture notes as
m 2
L  mc 2  u  q  qu  A. (*)
2
In our current problem, in the cylindrical coordinates r = {, , z} (with the z-axis coinciding
with the common axis of the two cylindrical electrodes), the magnetic field has only a z-component, B =
nzB, while the electric field has only a radial component: E = –nd()/d. Since the emitted electrons
have negligible initial velocities, neither component of the Lorentz force (5.10) drives them in the z-
direction at any time. As a result, each electron moves only in a plane normal to the common axis, and
its velocity has only two components:
u  n    n   ,
so
u 2   2   2 2 .
Selecting the vector potential’s gauge so that it has the natural axially-symmetric form A =
nA(), and using the well-known expression for the curl of such a vector in the cylindrical

Problems with Solutions Page 391

Essential Graduate Physics EM: Classical Electrodynamics

coordinates,171 A = nz[(A)/]/, we see that the potential’s definition, A = B, with B = nzB =
const, is satisfied if A() = B/2. So the scalar product participating in the Lagrangian function (*) is
B
u  A   ,
2
and the whole function (excluding its inconsequential rest-energy part mc2) becomes
B
L
2

m 2
   2 2   q    q
2
 const ,

with m = me and q = –e. Now using this function in a regular way172 to derive the Lagrange equations of
motion of two generalized coordinates,  and , we get

m   m 2  q d  qB  0, d  B 2

d 
 m   q
2
   0 , (**)
dt d dt  2 
with the function () determined from the system’s electrostatics.173 (In our simple case of
unconstrained motion, Eqs. (**) may be also obtained directly from the non-relativistic version of Eq.
(9.144):
d
mu   qE  u  B  ,
dt
spelled out in the polar coordinates, but technically, this pedestrian way is a bit more cumbersome.)
The second of Eqs. (**) is the simplest of the two to solve, immediately giving us the following
first integral of motion:174
B 2
m 2  q  const
2
The constant on the right-hand side of this relation may be evaluated from the condition of negligible
velocity (including its azimuthal component) on the cathode’s surface, i.e. at  = R1. The result may be
represented as
qB  R12 
   1  . (***)
2m   2 
This equation may be combined with the law of conservation of the electron’s energy:175
m 2
2
u  q    
m 2
2
 
   2 2  q    const .

171 See, e.g., MA Eq. (10.5) with f = A, A = Az = 0, and / = /z = 0.

172 If you need a reminder (I hope you do not :-), please see CM Eq. (2.19).
173 See, e.g., Eq. (2.45) of the lecture notes – which we (somewhat counter-intuitively) will not need in this
solution.
174 This integral might also be derived by applying Eq. (9.201) to a circular contour of radius , pierced by a

constant, normal magnetic field B, so  = 2B.

175 It may be also readily obtained by the integration of the first of Eqs. (**) over , and then plugging into the
result the second of these equations, integrated over t.

Problems with Solutions Page 392

Essential Graduate Physics EM: Classical Electrodynamics

Applying this relation to points  = R1 (where both velocity components are negligible) and  = R2, and
taking (R1) = 0 and (R2) = V, where V is the voltage between the anode and the cathode, we get
m 2
2
 
   2 2   R2  qV  0 .

Plugging Eq. (***), written for  = R2, into this equality, we may readily solve it for the voltage V:

m 2 R12  
2 2
2  qB  
V       R  R2   1   . (****)
2q  2  2m   R22  
 
(For electrons, with their negative charge q = –e, this voltage is positive.)
If V is lower than some threshold value Vt, the particles move along
loop trajectories not reaching the anode – see the (very schematic) dashed line
in the figure on the right. In this case, the radial velocity  is vanishing at B R1
some point, most distant from the axis but still smaller than R2. The threshold R2
voltage Vt we are looking for may be defined as the value of V at which this 0
most distant point of the trajectory exactly equals R2 – see the solid line in the
same figure. For this value, Eq. (****), with q = –e, and m = me, yields
2
eB 2 2  R12 
Vt  R2 1  2  .
8m e  R2 
For typical magnetrons (say, those used in kitchen microwave ovens) with R2 ~ 1 cm, R1 ~ R2/2,
and B ~ 0.1 T,176 this expression yields realistic values of Vt of the order of 10 kV. Note that on the 10-
keV scale of the energy provided by this field, the random thermal energies of the order of kBT ~ 0.1 eV
of the electrons emitted by heated cathodes (with temperature T ~ 103 K) are indeed negligible. On the
other hand, such kinetic energy is still much lower than the rest energy mec2  0.5 MeV of an electron,
so the kinetic177 relativistic effects are negligible as well.

Problem 9.34. A charged relativistic particle has been injected into a region with a uniform
electric field whose magnitude oscillates in time with frequency . Calculate the time dependence of the
particle’s velocity, as observed from the lab reference frame.
Solution: Directing the x-axis along the electric field, and the y-axis normally to it but within the
plane of the initial velocity of the particle (so that z  0), we can rewrite the relativistic equation
(9.144) of its motion, in the lab frame, as
d ( x ) d ( y )
  cos t ,  0,
d (t ) d (t )

176 This value is chosen so that the average frequency    of the electron motion, following from Eq. (***),
would be close to the own frequency of the resonating cavities of the device (see Fig. 9.13 of the lecture notes),
and hence to the frequency of the microwaves it generates. This frequency is government-mandated to be within a
narrow range around 2.45 GHz, to avoid microwave spectrum contamination.
177 This qualifier is intended to remind the reader that the magnetic field (which cannot be neglected in this
problem) is itself an essentially relativistic effect – see, e.g., the discussion in Sec. 5.1 of the lecture notes.

Problems with Solutions Page 393

Essential Graduate Physics EM: Classical Electrodynamics

with the dimensionless parameter:

qE
 ,
mc
E being the electric field’s amplitude. These equations may be readily integrated, giving
 x   0  x 0   sin t ,  y   0  y 0 . (*)
Since
i.e.  2 1   2   1,
1
2  , so  2  1  ( ) 2  1  ( x ) 2  ( y ) 2 ,
1  2
we may use Eqs. (*) to calculate the Lorentz factor  as an explicit function of time:

  1  ( x ) 2  ( y ) 2 
1/ 2

 1   0  x 0   sin t    0  y 0 
2

2 1/ 2
. (**)

Now we can combine Eqs. (*) and (**) to represent both reduced velocities as functions of time:
 0  x 0   sin t  0  y0
x  , y  ,
1   0  x 0   sin t 2   0  y 0 2 
1/ 2
1   0  x 0   sin t 2   0  y 0 2 
1/ 2

  0  x 0   sin t 2   0  y 0 2 
1/ 2

   x2   
2 1/ 2
  .
1   0  x 0   sin t    0  y 0  
y 2 2

For the simplest particular case of no initial velocity (x0 = y0 = 0), the last expression is reduced to
 sin t
 . (***)
1   2
sin 2 t 
1/ 2

The results show that if the field is relatively weak,

  1, i.e. qE a  mc 2 ,
where a  c/ is the relativistic scale of the particle’s oscillation amplitude, its oscillations are virtually
sinusoidal – natural for a free non-relativistic particle driven by a sinusoidal force.178 However, in the
opposite limit  >> 1, Eq. (***) shows that   1 (i.e. u  c) during most of the oscillation period,
except for very narrow time intervals during which the velocity “switches” from +c to –c and back.

Problem 9.35.* A linearly-polarized plane electromagnetic wave of frequency  is incident on an

otherwise free relativistic particle with electric charge q. Analyze the dynamics of the particle’s
momentum and compare the result with those of the previous problem and Problem 7.5.
Solution: Directing the z-axis along the direction of the wave’s propagation and the x-axis along
its electric field’s polarization, and per of Eq. (7.6), we may represent the field distribution is
E  n x E 0 cos , B  n y E 0 / c  cos , where   kz  t  const,

178 A useful sanity check, in this non-relativistic limit, the velocity’s amplitude umax  maxc = c  qE/m
coincides with the corresponding intermediate result of the model solution of Problem 7.5.

Problems with Solutions Page 394

Essential Graduate Physics EM: Classical Electrodynamics

where the last constant may be always made zero by the appropriate choice of the time’s origin. Let us
use these expressions to spell out the Cartesian components of the basic equation of the particle’s
motion – see Eq. (9.144) of the lecture notes:
dp x  u  dp y dp z u
 qE 0 cos  1  z ,  0,  qE 0 cos  x . (*)
dt  c  dt dt c
The second of these equations may be immediately integrated, giving py(t) = py(0). The
integration of the first equation may be simplified by noticing that the expression in the parentheses on
its right-hand side may be simply expressed via the full derivative of the total phase  of the wave,
taking into account the particle’s motion along the z-axis:
d dz u 
 k    ku z      z  1 ,
dt dt  c 
so the equation takes the simple form
dp x qE d qE 0 d sin  
  0 cos   ,
dt  dt  dt
and now may be readily integrated over time:
qE 0 qE 0
px  sin kz  t   const ,
sin   const  (**)
 
where the constant may be always selected to equal zero, by a proper choice of the z-axis’ origin.
Next, let us by notice that for our problem, the scalar product uE is merely uxE0cos, so Eq.
(9.148) of the lecture notes is reduced to
dE
 qE 0 cos  u x ,
dt
where E is the energy (9.78) of the particle (not including its interaction with the wave’s field). As a
result, the last of Eqs. (*) takes the form dpz/dt = (dE/dt)/c and may be also readily integrated, giving

 C  mc   p x2  p y2 0   p z2 
E  2 1/ 2
pz  C, (***)
c  

where px is given by Eq. (**) and the constant C may be found from the initial condition:
E 0
 p z 0  mc   p x2 0  p y2 0   p z2 0   p z 0 .
 2 1/ 2
C
c  

The algebraic equation (***) may be simplified by bringing the term C over to the left-hand side and
squaring the resulting relation. After the cancellation of the terms pz2, we get an apparently explicit
expression for this component:
2
p x2 1  qE0 
pz   C'    sin 2 kz  t   C' , (****)
2C 2C   

Problems with Solutions Page 395

Essential Graduate Physics EM: Classical Electrodynamics

where C’ is another constant:179

mc2  p y2 0  C 2
C'  .
2C
The simplicity of our final formulas (**) and (****) is somewhat deceiving: they express the
particle momentum components not explicitly via z and t, but only via a linear combination of these
independent arguments. In order to find their relation z(t), i.e. the z-component of the particle’s motion
r(t), we need to integrate an explicit relation for the corresponding Cartesian component uz of the
particle’s velocity u = p/(u)m. Since the Lorentz factor  depends on all three components of u, even
obtaining such an explicit expression from our results runs into some prohibitive algebra.
In the non-relativistic limit, which requires, in particular, the wave’s field to be sufficiently
weak:
qE0
px   mc,
max

Eq. (**) reduces to the results of both the previous problem (formally corresponding to the low-
frequency limit k  0) and that of Problem 7.5, describing sinusoidal oscillations of the particle in the
direction of the wave’s electric field. In this limit, Eq. (****) shows that the particle’s oscillations along
the z-axis, i.e. the direction of the wave’s propagation, are much smaller, and their fundamental
frequency is 2 (rather than  as for the x-motion).180

Problem 9.36. Analyze the motion of a non-relativistic particle in a region where the electric and
magnetic fields are both uniform and constant in time, but not necessarily parallel or perpendicular to
each other. y
Solution: Let us select the direction of the vector B for the z-axis, and direct Ey
the y-axis so that the vector E resides in the [y, z] plane – see the figure on the right. E
Then in the non-relativistic case, Eq. (9.144) of the lecture notes (with u = 1, i.e. p
= mu) has the following Cartesian components: B
z 0 x
mx  qy B, my  q E y  xB , mz  qE z . (*) Ez
The last equation is independent of its counterparts and may be easily integrated to
give the usual “free fall” motion along the z-axis, with the constant acceleration az = qEz/m. The first
two equations may be merged by the introduction of the rotational 2D velocity
Ey
u  n x  x  u d   n y y , with u d  .
B
Differentiating this vector over time, and then using Eqs. (*), we get

179The constants C and C’ are especially simple in the particular case when the particle is at rest at least at one
(say, initial) point: py(0) = pz(0) = 0; then C = (mc)2 and C’ = 0.
180Note also that according to Eq. (****), the motion in the z-direction may also include some average
drift, though the exact calculation of its velocity also requires the function z(t).

Problems with Solutions Page 396

Essential Graduate Physics EM: Classical Electrodynamics

( E y  xB) qB
u  n x x  n y y  n x
qy B
m
ny
m

m

n x y  n y  x  u d    c n x u y  n y ux  , 
i.e. a first-order differential equation similar to Eq. (9.150) of the lecture notes:
u   c u  n z ,

with the cyclotron frequency (9.151): c = qB/m. This means that the particle’s projection on the [x, y]-
plane performs a circular motion around a point that moves along the x-axis with a constant drift
velocity ud = Ey/B, i.e. essentially the same motion (along a trochoidal trajectory) as for the case of
mutually perpendicular fields (Ey = E, Ez = 0), which was analyzed in the lecture notes – see Fig. 9.12
and its discussion.
Thus, the only essentially new feature resulting from the arbitrary angle between the fields is an
accelerated motion of the particle along the z-direction of the vector B. This is qualitatively true for a
relativistic particle as well but quantitatively, in that case, the motion is more complex because all
velocity components become interrelated via their common Lorentz factor u.

Problem 9.37. A static distribution of electric charge in otherwise free space has created a time-
independent distribution E(r) of the electric field. Use two different approaches to express the field
energy density u’ and the Poynting vector S’, as observed from a reference frame moving with a
constant velocity v, via the Cartesian components of the vector E. In particular, is S’ equal to (–vu’)?
Solution:
Approach 1. In the lab frame where the charges rest, their magnetic field equals zero, so the
Lorentz transform formulas (9.135) are reduced to
E'    E   , B'    0,
v  E
E'   E  , , B'   
c2
where the indices  and  mark the field components parallel and normal to the velocity vector v. As a
result, the field energy density (6.113),181 as observed in the moving frame, is

E'   2 B'   2 E 
2
0 1 0 1 0 1  vE  
u'  E' 
2
B'  2 2
 E' 2

2
 B' 2

2
 E 2 2
   
2 2 0 2 
0
 
2 0  c 2 
(*)
0  1 2 2 2 0  2 1  2 2 
 E 2   0    E  E  E   ,
 2    1   2

 2 2 0 c
2
2  
while the corresponding Poynting vector (6.114) is

S' 
0
1
E'  B' 
0
1
E' 
 E '   B'    B'    1
0
E  

 E     0  
v  E 
c2 




  0  E  v  E    E   v  E  .
 

181 See also Eq. (9.225).

Problems with Solutions Page 397

Essential Graduate Physics EM: Classical Electrodynamics

Applying the bac minus cab rule182 to both double vector products, and then using the mutual
parallelism of the vectors v and E, and their normality to the vector E, we may reduce the last
expression to a simpler one:


S'   0  v E    E    E  v  E     v E   E    E  E   v    0  E  vE   vE 2 . (**)

Approach 2. Let us use the standard choice of coordinates (see, e.g., Fig. 9.1 of the lecture
notes), so E  nxE = nxEx, and E = nyEy +nzEz, i.e. E2 = Ey2 + Ez2. According to Eqs. (9.225)-(9.229)
of the lecture notes, in our case, the symmetric 4-tensor (9.221) of the energy-momentum of the field in
the lab reference frame (in which the charges are at rest and hence B = 0) is
 E 2  E 2 0 0 0 
 
0  0  E 2  E 2  2 E  E y  2 E  E z 
  .
2 0  2 E  E y  E y2  E 2  E z2  2E y E z 
 0  2 E  E z  2E y E z  E z2  E 2  E y2 

Applying to this tensor the Lorentz transform (9.109), just as was spelled out in the model solution of
Problem 18, we get

 
  
u'   ' 00   2  00    01   10    2 11   2 0 E 2  E 2    0   2  E 2  E 2
2
 

1 0
1  2 2
 
1   2 E 2  1   2 E 2 
– the expression equivalent to Eq. (*) that was obtained using Approach 1. Similarly, the same Lorentz
transform yields
S' x
c

  ' 01   2  01    00   11    2 10 

    
  2 0 0   E 2  E 2   E 2  E 2   2  0   0  2  E 2 ,
2

S' y 
c
 
 2

  ' 02    02   12   0   0  2 E  E y    0  E  E y ,
 

S' z
c
 
 
  ' 03    03   13   0   0  2 E  E z    0  E  E z .
2
 
The set of these three scalar relations is obviously equivalent to the vector equality (**),
obtained using Approach 1. These formulas show that in the general case when E  0, the Poynting
vector S’ has not only a component directed along the velocity vector v, as the apparent result (–vu’)
would give, but also a component directed along the vector E, i.e. normally to the vector v.

Problem 9.38. A traveling plane wave of frequency  and intensity S is normally incident on a
perfect mirror moving with velocity v in the same direction as the wave.

182 See, e.g., MA Eq. (7.5).

Problems with Solutions Page 398

Essential Graduate Physics EM: Classical Electrodynamics

(i) Calculate the reflected wave’s frequency, and

(ii) use the Lorentz transform of the fields to calculate the reflected wave’s intensity
– both as observed from the lab reference frame.
Solutions:
(i) Per Eq. (9.44) of the lecture notes, the incident radiation frequency, as perceived in the
reference frame moving with the mirror, is
1/ 2
1   v
'     , with   .
1   c
From the point of view of a mirror-bound observer, the frequency does not change upon reflection, so
the mirror may be treated as a source of reflected radiation of the frequency ’. From the point of view
of the lab-frame observer, this source moves from the lab-frame observer with the velocity v. Hence, we
may apply Eq. (9.44) again to calculate the reflected wave’s frequency as perceived by the observer:
1/ 2
1   1  cv
 r  '     . (*)
1   1  cv
(ii) According to Eqs. (7.6) and (7.8), in the incident plane wave propagating in free space along
some direction n,
nE nE .
B  0 H  0  (**)
Z0 c
Now applying Eqs. (9.135) to this transverse wave (with E = 0, B = 0, E = E, and B = B), we may
calculate its fields as perceived in the reference frame moving with the mirror:

n  n  E 
1/ 2
  v 1  
E'   E  v  B    E  v   E  1    E   ,
 c  c 1  
1/ 2
 vE n  E v  E  n  E  v  n  E 1  
B'    B  2      2   1      ,
 c   c c  c  c c  1   
so the relation between the vectors E’ and B’ is also given by Eq. (**) – as could be expected.
Calculating the Poynting vector (6.114), we get
E'  B ' E B 1  1  1  cv
S'  E'  H'    E H S S .
0 0 1   1  1  cv
the last factor giving the relation between the wave intensities in the two reference frames.
As was discussed in Sec. 7.4, for the observer moving with the mirror, at a wave’s reflection
from a perfect mirror, its intensity does not change, so the last relation (with the opposite sign of S
because of the change of the wave’s direction) is also valid for the reflected wave – as observed from
the moving reference frame. Now repeating the above calculation, we can find the intensity of the
reflected wave as observed from the lab frame:183

183This result may be also obtained by a completely different approach, using Eq. (9.237) of the lecture notes –
see the next problem.

Problems with Solutions Page 399

Essential Graduate Physics EM: Classical Electrodynamics

2
cv cv
S r  S'  S  . (***)
cv cv
Note that all the relations in this solution, including the final results (*) and (***), are valid for
any sign of the mirror’s velocity v. In particular, Eq. (***) shows that the wave’s intensity is decreased
if the mirror moves in the same direction as the incident wave (“trying to run away” from it) and is
increased in the opposite case. Since the intensity increase in the latter case (the mirror moving toward
the wave’s source) is frequency-insensitive, it may be interpreted as an amplification of the power of an
arbitrary waveform. However, this system is rather different from the usual amplifiers, because the
power gain is accompanied by an upward shift of all frequency components of the signal – see Eq. (*).

Problem 9.39. Perform the second task of the previous problem by using general relations
between the wave’s energy, power, and momentum.
Hint: As a byproduct, this approach should also give you the pressure exerted by the wave on the
moving mirror.
x
Solution: Let us direct the x-axis along the incident
wave’s propagation, and plot the positions of several vt t
representative fronts (i.e. the planes of equal total phase) of that
wave and the reflected wave, as well as of the mirror, as ct
x xr
functions of time – see the figure on the right. As the plots ct
show, during an arbitrary time interval t, the mirror is reached
by the incident wave that was initially spread over the following
volume:
V  Ax  A ct  vt   A c  v  t , t

where A is the wave front’s area. On the other hand, by the end of the period, the reflected wave is
spread over a different volume:
Vr  Ax r  A ct  vt   A c  v t .
(Though the figure above is drawn for v > 0, it is straightforward to verify that the above formulas are
valid for any mirror’s velocity with v  c.)
Let u be the energy density of the incident wave, and ur be that of the reflected wave – both as
measured in the lab frame. Then the initial wave’s energy within the volume V is uV = uA(c – v)t,
and that of the reflected wave, within volume Vr, is urVr = urA(c + v)t. Due to the energy
conservation, the difference between these two energies has to be equal to the work, Axm = Avt,
performed by the wave’s pressure force A at the mirror’s displacement xm = vt – see the figure above
again. This balance gives us, after the cancellation of A and t, the following equation for ur and :

u c  v   u r c  v   v . (*)

Another equation for the same two quantities may be obtained from the conservation of the total
momentum of the system: the force impulse Ft = Atnx given to the mirror by the wave, i.e. the
change of the mirror’s momentum during a time interval t, should be equal and opposite to the change
of the wave’s momentum, resulting from its reflection, during the same time interval. But according to

Problems with Solutions Page 400

Essential Graduate Physics EM: Classical Electrodynamics

Eq. (9.237), and (7.9) of the lecture notes (the latter one with v = c), for a plane wave propagating in free
space along an axis n, the momentum per unit volume is just
S u x
g  2  n. F
c c v
Hence, for our system’s geometry (see the figure on the right), the momenta balance
yields S
u u
At n x  n x V  r  n x Vr . Sr
c c
Now plugging in the above expressions for V and Vr, and canceling nx, A, and t, we get
u c  v   u r c  v 
 . (**)
c
Solving the simple system of two equations (*) and (**), we get
2 2
cv cv
ur    u, so that S r    S,
cv cv
i.e. the same result as by the Lorentz-transform approach used in the model solution of the previous
problem. Besides that, the same system of equations yields the following expression for the wave’s
pressure:
c  v 2S c  v
  2u  .
cv c cv
As a sanity check, for an immobile mirror (v = 0), this formula, with the account of Eq. (7.9b),
gives the result (9.245)-(9.246).

Problem 9.40. For the simple model of capacitor charging by a lumped

source of current I(t), shown in the figure on the right, prove that the
momentum given by a uniform, stationary, external magnetic field B to the   B

I (t )
current-carrying conductor is equal and opposite to the momentum of the
electromagnetic field that this current builds up in the capacitor. (You may
  
assume that the capacitor is planar and very broad, and hence neglect the
fringe field effects.)
Solution: According to Eq. (5.8) of the lecture notes, in the geometry shown in the figure above,
with a positive current I(t), the total magnetic force exerted on the wire is directed to the left and has the
magnitude F(t) = I(t)Bd, where d is the wire’s length, i.e. the distance between the interior surfaces of
the capacitor’s plates. Integrating this expression over time, for the momentum transferred from the
field to the wire, we get
p wire   F (t )dt  Bd  I (t )dt  Bd Q, (*)

where Q  I(t)dt is the capacitor’s charge built up by the current.

Problems with Solutions Page 401

Essential Graduate Physics EM: Classical Electrodynamics

Since I(t) > 0, the charge Q is also positive, and the electric field it creates is directed, in our
figure, from the top plate down, so the electromagnetic field’s momentum density vector (9.237) is
directed to the right. The total field momentum has the magnitude
S EH  (V / d )( B /  0 )  A (**)
p field    gd 3 r  gAd  2
Ad  2
Ad  2
Ad  0 VBd ,
c c c d
where A is the capacitor’s area, and V = Ed is the final change of the voltage between its plates. But
the fraction in the last form of Eq. (**) is just the capacitance C, so its product by V equals the same
Q that participates in Eq. (*). Thus the two momenta, pwire and pfield, are indeed equal and opposite, so
the total momentum of the whole system (capacitor + field) is conserved.

Problem 9.41. Consider an electromagnetic plane wave packet propagating in free space, with its
electric field represented as the Fourier integral

ik
E(r, t )  Re  E k e dk , with k  kz   k t , and  k  c k .


Express the full linear momentum (per unit area of wave’s front) of the packet via the complex
amplitudes Ek of its Fourier components. Does the momentum depend on time? (In contrast with
Problem 7.9, the wave packet is not necessarily narrow.)
Solution: Per Eq. (9.237) of the lecture notes, we need to calculate the time dependence of
  
p 1 1
  g z ( z , t )dz  2
A  c S z z, t dz  c 2  Ez, t   Hz, t  dz ,

z (*)

where A is the wave front’s area. According to Eqs. (7.6) and (7.7), the magnetic field of the packet may
be represented as
1 
 
i i
H ( z , t )  Re  H k' e k' dk'    H k' e k' dk'  c.c. ,

2  
1/ 2
n  Ek  
with H k  z sgn( k ), and Z 0   0  .
Z0  0 
Plugging the expressions for E and H into Eq. (*), taking into account that 1/cZ0  0, and changing the
order of integration, we get184

i   ' 

   ei  ' dz  c.c. .
   
p 0
  dk
A 4c    dk' sgn k  E
 k   n z  E 
k'  e dz  E k  n z  E*k ' (**)
   
Now we may readily work out both internal integrals, which do not depend on Ek:185

184 Cf. the derivation of Eq. (7.42) in Sec. 7.2.

185 See, e.g. MA Eq. (14.4).

Problems with Solutions Page 402

Essential Graduate Physics EM: Classical Electrodynamics



i k  k' dz  exp i 
  k' t   e
  dz
i k  k' z
e

k

exp 2i k t k  k' , for the upper sign,
 2 exp i k   k' t k  k'   2  
  k  k' , for the lower sign,
so the integral over k’ in Eq. (**) is also elementary, and that formula reduces to

  E   n   
p  0

 k z  E  k exp 2i k t  E k  n z  E*k  c.c. sgn k dk .
A 2c 

Taking also into account that in a plane wave propagating along the z-axis, both vectors Ek and E–k are
perpendicular to the vector nz,186 we may rewrite this expression in a simpler form:

 E 
 
 0 

 E 
p
 0 k  E  k exp 2i k t  c.c. sgn k dk  k  E*k  c.c. sgn k dk . (***)
A 2c 
2c 

Since with our definition of k, it is an even function of k,187 the first term on the right-hand side
is an integral, in symmetric limits, of an odd function of k, and hence is equal to zero. For the second
term, this is not necessarily true.188 In that term, the scalar product EkEk* equals just EkEk*  Ek2,189 so
taking into account the (equal) complex conjugate term, we finally get

p  0

 E k E k* sgn k dk .
A c 

Note that in contrast with the wave packet’s energy (whose calculation was the subject of
Problem 7.9), the cancellation of the rapidly oscillating terms in Eq. (***) is exact for the packet of any
form.

Problem 9.42. Calculate the forces exerted on well-conducting walls of a waveguide with a
rectangular (ab) cross-section, by a wave propagating along it in the fundamental (H10) mode. Give an
interpretation of the results.
Solution: According to Eq. (7.131) of the lecture notes (with n = 1 and m = 0), and Eqs. (7.137) –
(7.138), the broader walls of the waveguide (in Fig. 7.22, with y = 0, b) are exposed to both the electric
and magnetic fields of the wave, with complex amplitudes

186 A word of caution: the directions of the vectors Ek and E-k (in the plane normal to nz) do not necessarily
coincide (see, e.g., the last part of Sec. 7.1 of the lecture notes), so using their scalar product is important.
187 An additional question for the reader: what would the initial expression for E(r, t) describe if we took  = ck,
k
without the modulus sign?
188 For example, if the wave packet is narrow in the momentum space, and propagates in the direction of larger z,

the only significant amplitudes Ek correspond to values of k that are close to a certain positive value k0.
189 If there is any doubt in this statement: since the complex-number vector E is normal to the propagation axis z,
k
we may represent it as nxEx + nyEy = nxAxexp{ix} + nyAyexp{iy}, where Ex,y may be still complex but Ax,y and x,y
are real. Now scalar-multiplying this vector by its complex conjugate, we get Ax2 + Ay2, i.e. the same result if we
calculated Ek2  Ex2 + Ey2.

Problems with Solutions Page 403

Essential Graduate Physics EM: Classical Electrodynamics

ka x kza x x
E x  0, E y  i ZH l sin , E z  0; H x  i H l sin , H y  0, H z  H l cos ;
 a  a a
while at the narrower walls (x = 0, a), only the longitudinal magnetic field is nonvanishing, with the
complex amplitude equal to Hl. The wave vector components participating in these formulas are
related by Eqs. (7.132); for the H10 mode (with kx = /a, ky = 0), they may be rewritten as
2 2
 ka   k z a 
     1.
    
Plugging these relations into Eqs. (9.239) and (9.241), for the average pressure on the walls in terms of
the longitudinal field’s amplitude Hl, we get

dF  2 x x  dF  2
y  0,b  0 H l  cos 2  sin 2 , x  0, a  0 Hl ,
dA 4  a a dA 4
where the extra factors ½ came from the averaging over time.
So, the pressure on the narrower walls is always positive, due to the magnetic field’s effect,
while that on the broader walls is negative in their middle parts (for example, at x = a/2 where sin2(x/a)
= 1, cos2(x/a) = 0), due to the electric field effect (attraction). On average over x, these two pressures
on the broader wall compensate each other, so the net forces exerted on them are equal to zero.
These results may be readily explained in terms of the fundamental relation between the flux cg
= S/c of a plane wave’s momentum and its pressure on the mirror that reflects it. Indeed, as was
discussed in the model solution of Problem 7.25, the H10 wave may be represented as a sum of two plane
waves with wave vectors k = {/a, 0, kz}, which are reflected from the narrower walls of the
waveguide. These reflections create an outward pressure on the walls, similar to that at the normal
incidence – see Eq. (9.245) of the lecture notes and its discussion. On the other hand, since the H10
waves have ky = 0, they are not reflected from the broader walls of the waveguide, and as a result, do not
exert a net force on these walls.

Problems with Solutions Page 404

Essential Graduate Physics EM: Classical Electrodynamics

Chapter 10. Radiation by Relativistic Charges

Problem 10.1. Derive Eqs. (10.10) of the lecture notes from Eqs. (10.1) by a direct (but careful!)
integration.
Solution: Let us first recognize the error that led to the wrong results (10.4). A direct substitution
of Eq. (10.1b) into Eq. (10.1a) yields
q 1
 r, t   d
3
r"  r"  r'  . (*)
4 0 r  r"

The naïve integration over r”, giving the first of Eqs. (10.4), ignores the fact that according to Eq.
(10.1), the particle’s position r’ in the integral has to be taken at the retarded time t – R/c  t – r – r’/c,
i.e. at the point r’ that depends on the integration variable r” as well.
In order to overcome this difficulty, let us rewrite the first of Eqs. (10.1a) in a different but
equivalent form by introducing an additional integration over the auxiliary time argument t”, and
compensating it with a delta function at the appropriate point t” = t – r – r”/c:

1  r" , t"   r  r" 

 r, t    d r"  dt"
3
  t"  t  .

4 0  r  r"
c 
Now changing the integration order, and plugging in the first of Eqs. (10.1b), we get

q   r"  r' t"   r  r" 

 r, t    dt"  d r"
3
  t"  t  .

4 0 r  r"  c 
The internal integral may be limited to the vicinity of the point r’(t”), where it includes only a standard
3D delta function (multiplied by a smooth190 function of r”), and may be taken as usual, leaving us with
q 1  r  r' t"  
 r, t    dt"   t"  t  . (**)
4 0 r  r' t"   c 

This is still not a standard integral of a delta function but at least it is now one-dimensional. To
work it out, we may use the following property of the 1D delta function: if a monotonic scalar function
f() at some point 0 has a non-zero derivative df/d, then
    0 
  f  f0   , where f 0  f ( 0 ) . (***)
df / d  
0

The proof of this relation is easy: it is sufficient to integrate the right-hand side of Eq. (***) over any
interval of f, by using the standard chain rule:

190The function is singular only if r = r’, i.e. if the field observation point exactly coincides with the charge’s
position – the case we are not interested in.

Problems with Solutions Page 405

Essential Graduate Physics EM: Classical Electrodynamics

    0      0  df
 df / d  
df  
df / d   d
d .
0 0

The last form of this expression shows that if the interval of integration over  contains the point 0 (i.e.
the corresponding interval over f contains the point f0), then the integral equals 1,191 while if the interval
does not include this point, then the integral is equal to zero. Since this property is essentially the
definition of the delta function (f – f0),192 it proves Eq. (***).
Let us apply this identity to our case, taking   t”, f()  f (t”)  t” – t + r – r’(t”)/c, and the
point 0  (t”) 0 to correspond to the retarded time tret, i.e. to the physically-meaningful root (with tret < t)
of Eq. (10.5):
r - r' t ret 
t ret  t   0.
c
By our definition of the function f(t”), at this point, it turns to zero, i.e. f0 = 0. The derivative df/dt” at
this point may be calculated exactly as was done in Eqs. (10.11)-(10.16) of the lecture notes (just
replacing the notation tret with t”, and returning to tret after the calculation), giving

df 1
  1  β  n ret  0 .
dt" t" t t ret / t
ret

Plugging Eq. (***), with these specifications, into Eq. (**), we get

q  t"  t  r  r' t"  / c  q   f  f0 

 r, t    dt"   dt"
4 0 r  r' t"  4 0 r  r' t" 
q  t"  t ret  q  t"  t ret 

4 0 df / dt" t" t  dt" r  r' t" 

4 0 1  β  n ret 
dt"
r  r' t" 
.
ret

Since tret does not depend on t”, this is a standard integral of a delta function multiplied by a smooth
function, 193 so it may be evaluated in a regular way, giving
q 1 q 1
 r, t    .
4 0 1  β  n ret r  r' t ret  4 0 1  β  n ret Rret

This is Eq. (10.10a) of the lecture notes; the derivation of Eq. (10.10b) may be carried out
absolutely similarly.

Problem 10.2. Derive the radiation-related parts of Eqs. (10.19)-(10.20) of the lecture notes from
the Liénard-Wiechert potentials (10.10) by direct differentiation.

191 The sign of the derivative df/d at the point 0 does not affect the result, because the delta function behaves as
a symmetric one (i.e. may be considered as an approximation of a very short symmetric pulse).
192 See, e.g., MA Eq. (14.1).
193 Actually, the same approach might be applied directly to Eq. (*), but due to the 3D character of the involved
delta function, the necessary calculations are somewhat longer.

Problems with Solutions Page 406

Essential Graduate Physics EM: Classical Electrodynamics

Solution: We need to plug Eqs. (10.10),

q  1   0 qc  β 
 r, t   , Ar, t   ,
4 0  R(1  β  n)  ret
 4  R(1  β  n)  ret


into the general Eqs. (6.7)

A
E   , B  A, (*)
t
and perform these temporal and spatial differentiations. However, since we need to calculate only the
radiative parts of the fields, which dominate at R  , at these operations we may skip the derivatives
of the (1/Rret) factor. Indeed, let us consider, for example, the derivative194

 r, t  q   1 
 .
t 4 0 t  R(1  β  n)  ret


As was discussed in Sec. 10.1 of the lecture notes, at the fixed radius vector r of the observation point,
the retarded time tret is a unique function of t, so for this partial differentiation we may use the standard
chain rule:
 q   1  t ret
 , (**)
t 4 0 t ret  R (1  β  n)  ret t


where the last derivative is given by Eq. (10.16) of the lecture notes:
t ret  1 
   .
t  1  β  n  ret
Now, by taking the first derivative in Eq. (**) exactly, at the second step using Eq. (10.13), we get

  1   (1  β  n) ret / t ret R / t ret  β  n cβ  n 

t ret  R(1  β  n)       2 ret
     . (***)
  ret R(1  β  n) ret
2
R (1  β  n) ret  R(1  β  n)
2
R 2 (1  β  n)  ret

This expression shows that at r  , i.e. at Rret  r – r’(tret)  , namely at

βn
Rret  c 1  β  n  ~ c ,
β  n ret

where  is the time scale of the particle’s acceleration in the observer’s direction,195 the second term in
Eq. (**), which was the result of differentiation of (1/Rret), may be neglected. Similar estimates are valid
for all partial derivatives contributing to the fields (*). So, at the differentiations required for the
radiation field calculation, the fraction 1/Rret may be indeed treated as a constant.

194 Per Eq. (*), we do not need this particular derivative for the field calculation, but it is the simplest of all the
derivatives of this type while being conceptually similar to them.
195 Note that this condition is never fulfilled at zero acceleration. This conclusion is in accordance with the result
of the analysis, in Sec. 9.5 of the lecture notes, of the fields induced by a uniformly moving particle, which are
essentially its Lorentz-transformed Coulomb field rather than radiation – see Eqs. (9.139)-(9.140).

Problems with Solutions Page 407

Essential Graduate Physics EM: Classical Electrodynamics

Upon this simplification, the required differentiation of Cartesian components of the Liénard-
Wiechert potentials,
q  1   0 qc  βj  q  βj 
 , Aj       , with j  1, 2, 3 ,

4 0  R(1  β  n)  ret 4  R(1  β  n)  ret 4 0 c  R(1  β  n)  ret
becomes easy. For example,

A j q    j ret / t ret  j ret  (1  β  n) ret / t ret  t ret q   j  j β  n   

      .
t

4 0 c  R (1  β  n)ret 
R(1  β  n) 2 ret 
 t 4 0 c  R(1  β  n) 2 R(1  β  n) 3 
ret

Similar spatial differentiations are enabled by the relation similar to Eq. (10.16), cited in a
footnote in Sec. 10.1 of the lecture notes, whose jth Cartesian component is
t ret  nj 
   .
r j  c1  β  n   ret
For example,


q   1  t ret

q  n j β  n   
  .
r j 4 0 t ret  R(1  β  n)  ret r j 4 0 c  R(1  β  n) 3 
ret

From these results, for the jth component of the electric field, we get

Ej  
A j



q   j n j   j  β  n  q  
 n  β  j β  n   j (1  β  n) 
      .
t r j 4 0 c  R (1  β  n) 2 R (1  β  n) 3  4 0 c  R (1  β  n) 3  ret
ret

But this is exactly the result given by the radiative part of Eq. (10.19):

E
q 
 n  n  β   β  
  .
4 0 c  R(1  β  n) 3  ret
This fact may be verified either by the Cartesian components or (even simpler) by using, for the
numerator of the last fraction, the bac minus cab rule,196 and then taking into account that n is a unit
vector, so nn = 1:
     
n  n  β   β  n  β  n  β  β n  n  β   n  β  β  n  β 1  β  n  .

Evidently, the jth component of this vector is exactly the numerator in our result for Ej.
An absolutely similar calculation of the derivatives Aj/rj’ and then the Cartesian components of
the vector B = A,
 A j A j' 
B j"      jj'j" ,
 r r 
 j' j 

196 If you still do not remember it, please consult MA Eq. (7.5) again.

Problems with Solutions Page 408

Essential Graduate Physics EM: Classical Electrodynamics

shows that the radiative part of the magnetic field is indeed related to the similar part of the electric field
by Eq. (10.20): B = nretE/c, i.e. by the same equality as in a plane wave.
βt ret 
Problem 10.3. A point charge q that was in a stationary position on a q
circle of radius R is carried over, along the circle, to the opposite position on nt ret 
the same diameter (see the figure on the right) as fast as only physically
possible, and then is kept steady at this new position. Calculate and sketch the R
time dependence of the electric field E at the center of the circle. n fin r0 n ini
Solution: In this simple case, the distance R between the field
observation point (which we may take for r = 0) and the charge’s position r’ remains constant (equal to
the circle’s radius), so the delay between the retarded time tret (defined by Eq. (10.5) of the lecture notes)
and the field observation time t remains constant:
R
t  t ret  .
c
Let us take the retarded time of the beginning of the charge’s move for zero (with this and all other
times as measured in the lab reference frame). The length of the semi-circle’s arc is R, and the highest
possible speed of charge motion is u  c; hence the time of its arrival at the final point is R/u  R/c.
The corresponding interval of the field observation times t is
R R R R
t ini  t  t fin , with t ini  , t fin      1 . (*)
c c c c
Since the charge does not move at tret < 0 and at tret > R/c, the field observed outside of the
interval (*) may be calculated just from the Coulomb law:
q n , for t  t ini ,
E0, t     ini (**)
4 0 R n fin ,
2
for t  t fin ,

where n  R/R is the unit vector pointing from the charge’s position to the observation point – see the
figure above. Within the interval (*), E(0, t) has to be found from the retarded-field formula (10.19):197

Er, t  
q  nβ


n  (n  β)  β 
  , (***)
4 0   2 (1  β  n) 3 R 2 (1  β  n) 3 cR  ret

where the index “ret” marks the variable values at the retarded time tret = t – R/c. Since the Lorentz
factor   (1 – 2)–1/2 diverges at   1, in our case, the first term in the square brackets is negligible. In
the second term, the scalar product n in the denominator also vanishes, because at any instant tret of the
circular motion of the charge, the vectors (tret) and n(tret) are mutually perpendicular – see the figure
above. To calculate the numerator of that term, we may use the well-known formula for the acceleration
at a uniform circular motion:
u nu 2 / R nc
β    .
c c R

197 This formula, with   0 and  = 1, of course, yields Eqs. (**) as well.

Problems with Solutions Page 409

Essential Graduate Physics EM: Classical Electrodynamics

With this, and taking into account (twice) that nn = 0, and also that nn = 1, the double vector product
in the second term may be simplified as follows:


n  n  β   β   c
R
c c
n   n  β   n  n   β   n   β ,
R R
so Eq. (***) is reduced to
q q  R
E0, t    βt ret    β t  , for t ini  t  t fin . (****)
4 0 R 2
4 0 R 
2
c
Since in our approximation u = c, the length of the vector
Etfin  0
 equals 1, the magnitude of this field stays constant, equal to
that of the static Coulomb values (**), but its direction rotates in
time as the figure on the right shows. The instant leaps of the
field vector at moments tini and tfin, shown by dashed arrows, are Et  tini 
Et  tfin 
not quite surprising, taking into account that the problem 0
assignment implies an instant (and hence not quite physical)
acceleration/deceleration of the charge at these moments.
Etini  t  tfin 
Note also that while Eq. (****) results from the second
Etini  0
(“radiative”) term of Eq. (***), the field it gives in this particular
case, turns out to be proportional to 1/R2 – in contrast to the real radiation fields, which are proportional
to 1/R. This is also not quite surprising, because according to the discussion in Sec. 10.3 of the lecture
notes (see, e.g., Fig. 10.7a), the radiation of an ultra-relativistic particle at its circular motion is
concentrated within a very narrow cone centered to the vector (tret), and is always directed out of the
circle – not toward its center.

Problem 10.4. Express the instantaneous power of electromagnetic radiation by a relativistic

particle with electric charge q and rest mass m, moving with velocity u, via the Lorentz force F
providing its acceleration.
Solution: In the solution of Problem 9.32, we derived the following two formulas relating F and
the reduced velocity   u/c:
βF F  β β  F 
β  β  , β  .
mc 3
mc
Plugging them into the last form of the Liénard extension (10.37), we get:198

Z q2  4 
P 0 F  β β  F 2

1
β  F  2
 
Z0q2   
 
2



 F 2   2 β  F 2  2F  ββ  F  

6 mc 2  2  6  mc   1   2  β  F 2 
(*)
 
2
Z q2   
 F  β  F  .
2
 0 
2

6  mc 
This result may be expressed directly via the fields driving the particle. Since the magnetic part
of the Lorentz force is always normal to the vector   u/c, it drops out of the second term, and we get

198 An additional exercise: obtain this result by transforming Eq. (10.39) of the lecture notes.

Problems with Solutions Page 410

Essential Graduate Physics EM: Classical Electrodynamics

E  u  B  
2
Z0q 4   
 u  E / c  .
2 2
P  
6  mc 
Alternatively, this expression may be obtained by plugging Eqs. (9.125) and (9.145) into the first
(4-vector) form of Eq. (10.36) – a useful exercise, highly recommended to the reader.

Problem 10.5. A relativistic particle with rest mass m and electric charge q, initially at rest, is
accelerated by a constant force F until it reaches a certain velocity u and then is left to move by inertia.
Calculate the total energy radiated during the acceleration.
Solution: Evidently, the particle so accelerated moves along a straight line. The electromagnetic
wave power radiated at this linear acceleration is given by Eq. (10.40) of the lecture notes:
2
Z q2  dp 
P  02 2   ,
6m c  dt ret 
where the particle’s momentum p and the radiation emission time tret are both measured in the lab
reference frame. According to Eq. (9.144), dp/dtret = F, and the power may be expressed simply as
Z0q 2 F 2
P ,
6m 2 c 2
so it is time-independent through the acceleration process if F remains constant.199 Hence the total
radiated energy is
Z q2F 2
Erad  0 2 2 t , (*)
6m c
where t = pmax/F is the acceleration time. The maximum value pmax of the particle’s momentum may
be expressed via the given final value u of its velocity by using Eq. (9.70)
u
p max  mu  m ,
1  u / c 2 1/ 2
2

so Eq. (*) yields

Z q2F u
Erad  0 2 .
6mc 1  u / c 2 1 / 2
2

This result shows that at a fixed final velocity, the radiated energy increases proportionally to the
applied force F, i.e. a faster acceleration yields more radiation.

Problem 10.6. A charged relativistic particle with an initial momentum p0 flies ballistically from
a free-space region into a region of a constant, uniform electric field E, whose force is directed opposite
to p0. Calculate the energy radiated by the particle during its motion in the field, assuming that it is
small in comparison with the particle’s initial kinetic energy.

199
The same result readily follows from the last form of Eq. (*) in the previous problem’s solution because for the
1D motion, F2 – (F)2 = (1 – 2)F2  F2/2.

Problems with Solutions Page 411

Essential Graduate Physics EM: Classical Electrodynamics

Solution: Neglecting the radiation energy loss, we may use Eq. (9.144) of the lecture notes to
write the following scalar equation of motion for the only non-zero Cartesian component of the particle's
momentum (in the lab reference frame):
dp
  qE  const ,
dt
giving p(t) = p0 – qE(t – t0). From here, the time interval during which the initial momentum p0 is
exactly reversed is
2 p0
t  .
qE

Since the motion of the particle is symmetric with respect to the momentum-reversal instant, the particle
leaves the field-filled region of space at the end of this interval. Now using the last form of Eq. (10.40)
for the radiation power, we get
2
Z q 2 dp Z q2
P  0 2 2    0 2 2 qE 2 .
6m c  dt  6m c
(Alternatively, this expression follows from the last formula of the model solution of Problem 4, with B
= 0 and uE.)
Due to the constancy of this expression in time, we may calculate the full radiated energy (i.e.
the particle's energy loss) just by multiplying the power by the interval t:
Z0q2 2 2 p0 Z 0 q 2 p 0 qE
 E  P t  qE   .
6m 2 c 2 qE 3m 2 c 2
As in all situations with linearly-accelerated particles, for all realistic values of parameters, this
energy loss is much smaller than the initial energy of the particle – see the discussion at the end of Sec.
10.2 of the lecture notes.
Note that as a matter of principle, this problem could be solved using the Liénard extension
(10.37). However, since we are discussing a relativistic particle, the factors  and  participating in that
formula are somewhat involved functions of time – see Sec. 9.6(ii) of the lecture notes. So, we are lucky
that Eqs. (9.144) and (10.40) both involve the relativistic momentum p in such a simple form.

Problem 10.7. Calculate

(i) the instantaneous power, and
(ii)* the power spectrum
of the radiation emitted, into a unit solid angle, by a relativistic particle with charge q, performing 1D
harmonic oscillations with frequency 0 and displacement amplitude a.
Solutions:
(i) For this task, we may directly use Eq. (10.30) of the lecture notes:

dP 
Z 0 q 2 n  (n  β)  β
  2

 ,
dΩ (4 ) 2 (1  n  β) 5

Problems with Solutions Page 412

Essential Graduate Physics EM: Classical Electrodynamics

where all variables have to be evaluated at the retarded time tret. Selecting x n
the reference frame as shown in Fig. 10.4a of the lecture notes (reproduced
on the right), we may write

n  sin  , 0, cos  , r' t ret    0, 0, a cos  0 t ret ,
0 β β z
βt ret    0, 0,  0 sin  0 t ret , β t ret    0, 0,   0 0 cos  0 t ret ,

where 0 is the oscillation amplitude of the particle’s reduced velocity,

a 0
0  .
c
(Note that a and 0 have to satisfy the condition 0  a0/c  1.) Calculating the needed vector products,
n  β    0 cos  sin  0 t ret , n  β   β  n  β  0,  0 0 sin  cos  0 t ret , 0 ,
n  n  β   β    0 0 sin  cos  0 t ret  cos  , 0, sin  ,


n  n  β   β  2
  0 0 sin  cos  0 t ret  ,
2

we get
Z q 2  0 0 sin  cos  0 t ret 
2
dP
 0 2 .
dΩ (4 ) (1   0 cos  sin  0 t ret ) 5
For a non-relativistic oscillator (0 << 1), the denominator of the last fraction is close to 1, and
our result, after integration over the full solid angle, is reduced to the Larmor law – see Eq. (8.27) of the
lecture notes, with Z = Z0, v = c, and
p  qr'  qu  qc  qc  cos  t . 0 0 0

In particular, the radiation intensity is proportional to sin , and its time dependence (P  cos20t)
2

shows that the fields of the radiated wave are sinusoidal functions of the observation time t, with the
frequency 0. (Indeed, in the non-relativistic limit, the difference (10.5) between the times t and tret is
virtually a constant.)
However, in case 0  1, our result describes periodic radiation peaks (two per the oscillation
period, at sin0tret = 1), very narrow both in time and space (i.e. in angle ), emitted at the moments
when the particle’s speed approaches the speed of light. The angular structure of such a peak is the same
as at the linear acceleration (see Sec. 10.2 of the lecture notes, in particular, Fig. 10.4b), while its exact
temporal shape depends on whether we speak of the intensity as a function of tret or t, because exactly at
their emission moments, the difference between these times is especially significant – see Eq. (10.16).
Hence, this difference needs to be taken into account in the analysis of the radiation spectrum – see
below.
(ii) Since the radiation is periodic, its spectrum may only consist of discrete harmonics of the
motion frequency 0. Their amplitudes may be recovered from Eq. (10.61) of the lecture notes (which
has been derived with the continuous spectrum in mind) by careful juggling delta functions,200 but it is
more prudent (and instructive) to recast all basic equations of the spectral theory outlined in Sec. 10.3

200 Following this way may be recommended to the reader as an additional exercise.

Problems with Solutions Page 413

Essential Graduate Physics EM: Classical Electrodynamics

for a periodic time dependence of the radiated field. In particular, Eq. (10.51) needs to be replaced with
a sum over integer numbers (say, k – not to be confused with the wave number, which does not
explicitly participate in this solution):

Et   E k exp ik 0 t, with E  k  E*k .
k  

The magnetic field of the radiated wave (which, at a sufficiently large distance from the source, is
locally planar) may be expanded into a similar series,

H t   H k' exp ik' 0 t,
k ' 

with the complex amplitudes Hk’ = nEk’/Z0, so at the time averaging over the oscillation period T =
2/0, the Poynting vector falls apart into a sum of independent harmonic components:
  T
1 1 1
S n  E (t ) H (t )   E k E k ' exp i k  k'  0 t   E k E k '  exp i k  k'  0 tdt
Z0 k , k '  Z0 k , k '  T 0
   
1 1 1 2
E E E *
E
2
 k E k ' k  k' , 0  k Ek  k Ek  k .
Z0 k , k '  Z0 k   Z0 k   Z0 k 1

Thus instead of Eq. (10.54) of the lecture notes, we can write the following expression for the
average radiation power per unit solid angle:

dP dP d Pk 2R 2
 k,
2
with  R2 Sn  Ek . (*)
dΩ k 1 dΩ dΩ Z0
The Fourier amplitudes Ek should be calculated from the reciprocal Fourier transform, for example over
a symmetric time interval:
T / 2  / 0
1 
Ek   E(t ) expik 0 tdt  0  E(t ) expik tdt ,
2
0
T T / 2  / 0

so 2
 /  0
d Pk R 2 02

dΩ 2 2 Z 0  E(t ) expik tdt
 / 0
0 .

Plugging in the E(t) given by the second (radiative) term of Eq. (10.19), and repeating the
transformation from integration over t to that over tret discussed in Sec. 10.3 of the lecture notes, instead
of Eq. (10.61), we get
2
 /  0
d Pk Z 0 q 2 k 2 04    n  r' 
  n  n  β  expik 0  t   dt ret . (**)
dΩ 32 4  / 0    c  ret

As a sanity check, let us use this result to calculate the particle’s energy loss per one period of
motion, due to the radiation into a unit solid angle:

Problems with Solutions Page 414

Essential Graduate Physics EM: Classical Electrodynamics

dE d P 2 d P 2 
d Pk

d
T  
d Ω  0 dΩ  0
 dΩ .
k 1

We may use this formula for a non-periodic motion as well, by following the limit T  , i.e. 0  0,
while keeping the observation frequency   k0 finite. In this limit, the sum over k = /0 may be
replaced with an integral, and we get

dE 2

d Pk 2
 d P /     2 d P / 0
  0 dΩ dk   0 
0
d     I  d , where I    .
d 0 0
dΩ  0  0  02 dΩ
By using Eq. (**), for the function I() defined just like in the first of Eqs. (10.54), we get the
expression,
2
2 d P / 0 Z 0 q 2 2
 /  0
   n  r' 
I    2 n  n  β  expi  t 
16 3  / 0 
  dt ret ,
 0 dΩ   c  ret

Which, at 0  0, duly tends to Eq. (10.61).

Now plugging into Eq. (**) the expressions for n and  discussed in Task (i) of this solution,
performing the vector multiplications, and introducing the dimensionless integration variable   0tret,
we get
d Pk Z q 2  02 k 2  02 sin 2  2
 0 Ik ,
dΩ 32 4

where Ik is the following dimensionless integral:


Ik   sin  expik    cos  d ,

with    0 cos   1 .

Transforming this integral as

 
 
Ik 
1

2i 
 
e i  e i e ik e ik cos d 
1

2i 
e i ( k 1)  e i ( k 1) e ik cos d ,

we may notice that the integrals of the imaginary parts of exp{i(k  1)}  cos(k  1) + i sin(k  1), in
our symmetric limits for , vanish, so

1
cos(k  1)  cos(k  1)  e ik cos d
2i 
Ik 

 
1 1
  cos(k  1) e ik cos d   cos(k  1) e ik cos d .
i 0 i 0
Apart from constant coefficients, these integrals are the standard integral representations of the
Bessel functions of the first kind,201 and we get

201 See, e.g., MA Eq. (6.15a).

Problems with Solutions Page 415

Essential Graduate Physics EM: Classical Electrodynamics


Ik 
i
i k 1

J k 1 k   i k 1 J k 1 k   i k  J k 1 k   J k 1 k   i k 2
k

J k k ,

where, at the last step, the recurrence relation (2.142a) has been used. So, we get
d Pk Z 0 q 2 02 2 2
 k J k k 0 cos   tan 2  , for k  1 . (***)
dΩ 8 2
The figure on the right shows, on the 10
appropriate semi-log scale, the dependence of the k  30
product k2Jk2(k0) on the particle’s reduced velocity k J ( k 0 )
2 2
k
amplitude 0 = a0/c, for several representative 1 k  10
harmonics. As it follows from Eq. (***) and the
plots, in the non-relativistic limit (0  0), the first k 3
harmonic (k = 1) dominates, i.e. the radiation takes
0.1
place only at the frequency of the particle’s motion,
and our result for it,
k 1
d P1 Z 0 q 2 02 2
 J 1  0 cos   tan 2  0.01
dΩ 8 2

2
Z q 2 2   cos   Z q 2 2
 0 2 0  0  tan 2   0 2 0  02 sin 2  , 110
3
8  2  32 0 0.2 0.4 0.6 0.8
0
coincides with the time-averaged result of the first
part of the problem, and hence complies with the non-relativistic Larmor formula (8.28).
On the other hand, when the maximum value 0 of the normalized velocity of the particle
approaches 1, i.e. the particle becomes ultra-relativistic during two short parts of each period of its
motion, the intensity of high harmonics going in the direction of motion (cos  1) grows rapidly,
reflecting the narrow radiation cones generated during these small time intervals – see Fig. 10.4 and its
discussion. (The additional factor tan2 in Eq. (***) shows that these cones are hollow, just as in the
constant-acceleration case.)
A good additional exercise: review these results for the case when the force driving the particle
oscillation (rather than the particle displacement as above) is sinusoidal. The results should be different
only in the relativistic case, due to the increase of the effective mass (9.71) of the particle at the periods
of its fastest motion, and the corresponding reduction of its acceleration – cf. Problem 9.34.

Problem 10.8. Calculate and analyze the time dependence of the energy of a charged relativistic
particle rotating in a constant and uniform magnetic field B and as a result, emitting the synchrotron
radiation. Qualitatively, what is the particle’s trajectory?
Hint: You may assume that the energy loss is relatively slow (–dE/dt << cE) but should spell
out the condition of validity of this assumption.
Solution: With Eq. (9.153) of the lecture notes for the cyclotron orbit’s radius R,

Problems with Solutions Page 416

Essential Graduate Physics EM: Classical Electrodynamics

m u mc
R   , (*)
qB qB
and Eq. (9.73) for the particle’s energy, Eq. (10.45) for the radiative energy loss yields


dE Z 0 q 4 B 2 2 2 Z 0 q 4 B 2 2
      1 

E 2  mc 2
  
2

, where  
3m 3 c 2
. (**)
dt 6m 2 6m 2 2mc 2 Z0q4 B2
(Note that this constant  was already introduced and discussed in the model solution of Problem 8.4.)
Separating the variables of this differential equation,
d dt E
 , where    1, (***)
 1
2
2 mc 2
we can readily integrate it:202
1  1 t
ln   const .
2  1 2
Solving this simple equation for , we finally get

E  mc 2 coth 
t
 const  . (****)
 2 
This expression shows that the kinetic energy T  E – mc2 of the particle is gradually lost on the
time scale given by the constant . (As was already discussed in the model solution of Problem 8.4, for
an electron in a typical magnetic field of 1 T, this constant is close to 3 seconds.) At the finite stage of
this process, at t >> , the argument of coth becomes large, and this function may be approximated as
e   e   1  e 2 
coth    
  2
 1  2e  2 , for   1 ,
e e 1 e
so Eq. (****) is reduced to the exponential decay law obtained in the solution of Problem 8.4:
  t    t
T  E  mc 2  mc 2 coth  const   1  const  exp  .
  2    
If the initial value of T is much smaller than mc2, i.e. the particle is non-relativistic to start with, this
exponential law is obeyed at all times.
In the opposite limit when the particle is initially ultra-relativistic, the decay starts from a law
that differs from the exponential one. Indeed, in this limit, the function coth in Eq. (***) has to be much
larger than 1, so it may be approximated as
cosh  1
coth    , for   1 ,
sinh  
and the solution is reduced to

202As a math reminder, the integral on the left-hand side of Eq. (***) may be reduced to two integrals of the type
d/ = ln, by representing the involved fraction as the following sum:
1 1 1
  .
  1 2(  1) 2(  1)
2

Problems with Solutions Page 417

Essential Graduate Physics EM: Classical Electrodynamics

2
E t   mc 2 .
t  const
The constant participating in this expression may be readily related to the initial energy E(0) of the
particle, finally giving
E 0  E 0  t 
E t    E 0  1  , for E t   mc 2 .
1  E 0t / 2mc 2
 2mc  
2

This result describes an even faster initial decrease of energy, on the time scale
2mc 2
τ'     , for E 0   mc 2 .
E 0
The above simple theory, which assumes that each orbit is approximately circular, is only valid
if c >> 1. By using one of the forms of Eq. (9.151),
qB
c  ,
m
we may represent the validity condition as
m2c 2
  .
Z0q3B
For electrons in the field of 1 T, the right-hand side of this relation is of the order of 1012 (for heavier
particles, it is even larger), so in all practical systems, this condition is well satisfied.
As the kinetic energy of the particle goes down due to the radiative losses, so does the parameter
, so, as Eq. (*) shows, the cyclotron radius R decreases as well. Hence the particle’s trajectory is a
slowly converging spiral.

Problem 10.9. Analyze the polarization of the synchrotron radiation propagating within the
particle’s rotation plane.
Solution: The second term of Eq. (10.19) of the lecture notes shows that the instant electric field
E of the radiated wave is oriented as the vector
 dβ 
n  n  β    (*)
 dt 
at the retarded point, i.e. at the charge’s position. At the synchrotron radiation, the vectors  and d/dt
are always within the particle’s rotation plane, and perpendicular to each other. This is why if the
observer and hence the vector n are also in that plane, the inner product’s vector (n – )d/dt in Eq. (*)
is normal to that plane, so the outer product vector (and hence the vector E) is within the rotation plane
again.203

203 E is of course also perpendicular to vector n, i.e. to the direction of the wave’s propagation.

Problems with Solutions Page 418

Essential Graduate Physics EM: Classical Electrodynamics

In hindsight, this conclusion could be also made from the evident mirror symmetry of the
problem with respect to the rotation plane: if E had an out-of-the-plane component, what side would it
be directed to?

Problem 10.10. Analyze the polarization and the spectral contents of the synchrotron radiation
propagating in the direction normal to the particle’s rotation plane. How do the results change if not one,
but N > 1 similar particles move around the circle, at equal angular distances?
Solution: It is instrumental here to represent the double product in the basic Eq. (10.19),

 
n  n  β   β ,
describing the wave’s electric field, as the sum of two terms:

   
n  n  β  n  β  β , (*)
because in our current case, the second term vanishes. (Indeed, both vectors β and β rotate within the
particle’s rotation plane, so the inner vector product β  β is normal to the plane, just as the vector n, i.e.
the two vectors are parallel to each other, and their vector product equals zero.)
The double vector product in the remaining first term of Eq. (*) is opposite in direction to the
instant vector β , i.e. it rotates, within the plane, with the angular velocity c of the particle’s motion.
So, the vector E rotates (about the wave’s propagation direction n) with the same angular velocity, i.e.
the synchrotron radiation in the direction out of the rotation plane is circularly polarized, in the angular
direction of the particle’s rotation.
Now note that in our case, a different double vector product, n×(n×), also rotates similarly,
with the same angular velocity c, while the scalar product nr’ vanishes. According to Eq. (10.61) of
the lecture notes, this means that the out-of-plane radiation has only one frequency component (c) –
even in the ultra-relativistic case when the in-plane radiation has ~ 3 >> 1 harmonics of this basic
frequency!
If we have N > 1 similar particles moving around the same circle at equal angular distances  =
2/N, the net electric field is the vector sum of N vectors with equal magnitudes and the same angular
distances . As was already discussed in the model solution of Problem 8.22,204 such a sum equals
zero, so the system’s radiation in the direction normal to the particle rotation plane vanishes. (This is not
true for the in-plane radiation.)

Problem 10.11.* The basic quantum theory of radiation shows that the electric dipole radiation
by a particle is allowed only if the change of its angular momentum’s magnitude L at the transition is of
the order of Plank’s constant .
(i) Estimate the change of L of an ultra-relativistic particle due to its emission of a typical single
photon of the synchrotron radiation.
(ii) Do you think quantum mechanics forbid such radiation? If not, why?

204 And as follows, for example, from MA Eq. (3.6).

Problems with Solutions Page 419

Essential Graduate Physics EM: Classical Electrodynamics

Solutions:
(i) According to Eq. (9.70) of the lecture notes, the magnitude of the linear momentum of a
particle is p = mu, so the orbital angular momentum due to its circular motion with radius R is L = pR =
muR. From the basic geometry, the cyclotron orbit radius is R = u/c, so
mu 2
L ,
c
where the z-axis is normal to the rotation plane. On the other hand, according to the basic Eq. (9.73), the
particle’s energy is E = mc2, so in the ultra-relativistic limit (u  c), we may write
E  L c . (*)
This means that at the angular momentum change allowed by the usual quantum-mechanical selection
rule,  L  = , the change of the particle’s energy is
E   c .

On the other hand, Eq. (*) means that the most typical frequency (10.72) of the synchrotron
radiation, giving photons with energy
3
E   max ~  c ,
2
corresponds to a much larger angular momentum change:
E 3
L  ~    . (**)
c 2
(ii) The above result means that the selection rule  L  = , which is sometimes perceived as a
universal law, is invalid. This apparent violation of quantum mechanics is only superficial because the
standard selection rule is valid only in a non-relativistic spinless particle in the electric-dipole
approximation. As we know from Sec. 8.2, this approximation (and, as a result, the Larmor formula
(8.27) as well) is only valid if the size scale of the radiating system (e.g., a in Fig. 8.1) is much smaller
than the radiation wavelength  = 2/k = 2c/ = 2c/ E . However, for the synchrotron radiation
by an ultra-relativistic particle, the effective system’s size scale is R = c/c, and the relation is opposite:
2c 2c 2 2c 4
~  ~ 3  3 R  R ,
 max E  c 
so the naïve selection rule is invalid.

Problem 10.12. A relativistic particle moves along the z-axis, with velocity uz, through an
undulator – a system of permanent magnets providing (in the simplest model) a perpendicular magnetic
field, whose distribution near the axis is sinusoidal:205

205As the Maxwell equation for H shows, this stationary field distribution cannot be created in any
nonvanishing volume of free space. However, it may be created on a line – e.g., on the particle’s trajectory.

Problems with Solutions Page 420

Essential Graduate Physics EM: Classical Electrodynamics

B  n y B0 cos k 0 z .

Assuming that the field is so weak that it causes negligible deviations of the particle’s trajectory from
the straight line, calculate the angular distribution of the resulting radiation. What condition does the
above assumption impose on the system’s parameters?
Solution: Under the assumption of a nearly linear trajectory of the particle, u  nzuz, the Lorentz
force acting on the particle has only one Cartesian component:

F  qu  B  n z  n y  qu z B0 cos k 0 z  n x qu z B0 cos k 0 z ,

so the equations of its motion (with all variables measured in the lab frame) are
mu x
p x  qu z B0 cos k 0 z, with p x  mu x  ,
1  u 2
x  u z2  / c 2 1/ 2

mu z
p z  0, with p z  mu z  .
1  u 2
x  u z2  / c 2 
1/ 2

In the first (linear) approximation in small ux, the equations are simplified:
1
mu x  qu z B0 cos k 0 z , mu z n z  0, now with    const,
1  u 2
z / c2 
1/ 2

so, with an appropriate choice of time origin, z  uzt, and hence

qu z B0 q z B0
u x   cosk 0 u z t , i.e. β  n x cosk 0 u z t , and β  n z  z . (*)
m m
Hence the mutual orientation of the vectors β and β is similar to that at the synchrotron radiation – see
Fig. 10.5 and Eq. (10.46) of the lecture notes. This means that for the instantaneous power of radiation
we may reuse Eq. (10.47):
dP 2Z 0 q 2  2 6
 β  f  ,  , (**)
dΩ 2
where the angular distribution function,
1  sin 2  cos 2  
f  ,    1  , (***)
8 6 (1   cos  ) 3   2 (1   cos  ) 2 

has been discussed in Sec. 10.3 of the notes and plotted (for the ultra-relativistic case) in Fig. 10.6. The
only (but very substantial!) difference of our current problem is that the particle’s acceleration’s
magnitude is not constant in time (as it is at the synchrotron radiation) but oscillates sinusoidally with
frequency  = k0uz. Time-averaging Eq. (**) and using the second of Eqs. (*), we get
dP Z q4B2
 0 2 2 0  z2  4 f  ,   .
dΩ  m
It may look like this approximate analysis is valid if the amplitude

Problems with Solutions Page 421

Essential Graduate Physics EM: Classical Electrodynamics

qu z B0 qB0
u0  
m mk 0

of the transverse velocity ux(t) following from the above calculation is much lower than uz. Using Eq.
(9.151) for the cyclotron frequency, this condition may be recast in a very simple and natural form c
<<   k0uz. However, since the angular distribution (***) has a width scale of the order of 1/, the
actual condition of the theory’s validity is more strict u0 << uz/, i.e.

 c  .

In modern undulators, this condition limits the magnet period  = 2/k0 to a few centimeters – for
typical numbers, see Sec. 10.3 of the lecture notes, in particular, Fig. 10.10.
Note, however, that the above solution, based on Eq. (10.30) of the lecture notes, neglects the
possible coherent addition of the radiation induced by different periods of the undulator’s magnetic field
– see the next problem.

Problem 10.13. Discuss possible effects of the interference of the undulator radiation from
different periods of its static field distribution. In particular, calculate the angular positions of the power
density maxima.
Solution: As was discussed in detail in Sec. 8.4 of the
d cos 
lecture notes, the constructive interference of the radiation from
two coherent, in-phase sources takes place when the difference t 
of times of propagation of the emitted waves toward the observer u
(in the figure on the right, t = (dcos)/c) equals the integer d d
number of the wave periods T = /c. However, in the case of the undulator with a period d (equal to
2/k0 of the previous problem), with a particle propagating along it with velocity u, the radiation
induced by adjacent periods of the system has an additional time shift d/u, due to the difference of the
particle arrival times, so the constructive interference condition becomes
d d cos  
 n , (*)
u c c
with any integer n. On the other hand, the wavelength  of the kth harmonic of the undulator’s radiation
(where k = 1, 2,…) equals cTk = cT1/k = c(d/u)/k. Combining these expressions, we see that the condition
for the angles of the radiation maxima does not include d:
1  n u
cos   1  , where   . (**)
 k c
Since 1/ cannot be smaller than 1, real angles  with –1  cos  +1 may satisfy this condition only for
certain positive values of the integer number n.
In particular, for ultra-relativistic particles with  >> 1 used in standard undulators, the
similarity between the undulator radiation and the synchrotron radiation (see the model solution of the
previous problem) allows us to use Eq. (10.72) of the lecture notes to conclude that the most intense

Problems with Solutions Page 422

Essential Graduate Physics EM: Classical Electrodynamics

radiation harmonics have numbers k ~ 3/2 >> 1. Since for such particles, 1/  1/(1 – 1/2)1/2  1 +1/22,
the condition (**) becomes
 1  2n  1  4n 
cos  n  1  2 1  3   1  2   1 .
 2    2   
Using the Taylor expansion cos  1 –  2/2, valid for small angles , we see that the lowest-order
maxima correspond to the angles
1/ 2
1  4n  2 
 n    1  n  nmin 1 / 2  1, where nmin   1 ,
   3/ 2
4
with a very small angular distance,
d n 1 2 1
 n   n 1   n n~ n  n~ nmin  3 / 2 1 / 2  2  ,
min dn  nmin  
between the adjacent maxima, completely masked by the relativistic width  ~ 1/ of the radiation –
again, see the model solution of the previous problem. In this case, all results of that solution, in which
the interference has been neglected, are quantitatively valid.
On the other hand, for non- (or nearly-) relativistic particles, especially in “soft” (sinusoidal or
nearly-sinusoidal) undulator field profiles, such as the sinusoidal profile of the previous problem, the
radiation is nearly sinusoidal, i.e. the largest power corresponds to k = 1. In this case, according to Eq.
(**), the distance between the adjacent interference maxima may be quite substantial:
1
cos  n k 1  ~ 1,

and they may be readily resolved. A quantitative calculation of the radiation power distribution in this
case (a good additional exercise for the reader :-) should start from Eq. (10.19), rather than (10.30), of
the lecture notes.
If the radiation is produced by a short bunch of particles with a length of the order of d, rather
than by a single particle, the power spectrum of the emitted radiation depends on the bunch’s shape.
This fact enables using this effect for the experimental non-destructive measurements of the shape of
very short (in the time domain, femtosecond) electron bunches. It is q u
interesting that for such experiments the undulator is not really necessary:
the needed space-periodic transverse force may be produced just by the
flying charge’s image in a periodic metallic structure – see the figure on q
the right. The standard diffraction gratings, with d ~ 1 m, are quite d d
convenient for this purpose because, with  ~ 1 (the condition which is
easy to reach with electrons), the resulting radiation wavelengths  ~ d/ are in the visible light range,
for which sensitive radiation detectors are readily available.206

206 This is exactly the arrangement that was used by S. Smith and E. Purcell in 1953 for the first observation of
this Smith-Purcell effect (which may be also considered as a specific form of the transition radiation discussed at
the end of Sec. 10.5 of the lecture notes). Historically that experiment was the first step toward modern undulators
and free-electron lasers, which use much stronger periodic fields and hence produce much more intense radiation.

Problems with Solutions Page 423

Essential Graduate Physics EM: Classical Electrodynamics

Problem 10.14. An electron launched directly toward a plane surface of a perfect conductor is
instantly absorbed by it at the impact. Calculate the angular distribution and the frequency spectrum of
the electromagnetic waves radiated at this event, provided that the initial kinetic energy T of the particle
is much larger than the conductor’s workfunction .207 Is your result valid near the conductor’s surface?
Solution: As was discussed in Sec. 10.5 of the lecture notes in E r
the context of transition radiation, at not very high frequencies ( << 
p),208 the field outside of the conductor may be viewed as a result of q n q
a head-on collision of the particle of the given charge q with its mirror 0 u
u z
image with the charge (–q), moving toward each other – see the figure
on the right.
Due to the condition  << T, the change of the particle’s
velocity, due to its Coulomb interaction with the image, may be neglected. Hence for the pre-impact
velocities of the particle and its image, we may write
u u
β charge  n z , β image   n z .
c c
At collision, both the particle’s charge and that of its image, disappear – in our approximation, instantly.
Hence, for the calculation of the emitted radiation, we may use the low-frequency approximation
described by Eq. (10.76), in which the term marked “fin” equals zero. Since the motion of these two
charges is correlated, we have to sum up their electric fields rather than radiated powers – in Eq. (10.76),
inside the modulus sign, taking into account the difference of their charge signs:

n  n  β charge  n  n  β image 
2
n  n  β 
2
1 1 1 1
I    k q k 1  β  nk  q charge  qimage
4 2 c 4 0 k 4 2 c 4 0 1  β charge  n 1  β image  n

q 2  2 n  n  n z  n  n  n z 
2
1 1 q2  2 sin 2 
   .
4 2 c 4 0 1   n z  n 1   n z  n 
 2 c 4 0 1   2 cos 2  2

So, the spectral density of the radiation is frequency-independent, corresponding to a short

(delta-function-like) pulse radiated at the collision, though in reality its time duration is spread by ~1/p,
i.e. the frequency spectrum is limited to frequencies below ~p. Concerning the angular distribution, for
non-relativistic particles, it follows the usual sin2 pattern, but in the ultra-relativistic case is squeezed
to a hollow cone of width  ~ 1/, qualitatively similar to that illustrated by Fig. 10.4b. Note that our
result is only valid outside of the conductor (into which the waves do not penetrate), so the cone
corresponds to the angles   .
Since the calculated function I() does not vanish (and in the non-relativistic case, even reaches
its maximum) at  = /2, i.e. in directions along the conductor’s surface, one may wonder whether this
is also an artifact of some approximation. Actually (again, at frequencies  << p) it is not. Indeed, as
we already know, the vector E of the electric field of the wave generated at any linear motion of the
charge lies within the common plane of the observer and the line of the charge’s motion – see, e.g., Eq.

207See Sec. 2.9, in particular Fig. 2.27a.

208Taking the (admittedly, too vague :-) characterization of the conductor as “perfect” in the assignment literally,
we may assume that its free carrier density n is infinite, and so is the plasma frequency p – see Eq. (7.37).

Problems with Solutions Page 424

Essential Graduate Physics EM: Classical Electrodynamics

(10.19). On the other hand, it is perpendicular to the vector n of wave propagation – see the figure
above. Hence at  = /2, the wave’s electric field is normal to the conductor’s surface (while its
magnetic field is parallel to the surface). As we know from the discussion in Sec. 7.6, this is fully
consistent with the boundary conditions (7.104) on a perfect conductor’s surface. Note that this means
that the propagation of the electromagnetic pulse from the collision point is accompanied by ring-shaped
pulses of the surface electric charge and the radially-directed current, propagating from the collision
point on the conductor’s surface.

Problem 10.15. A relativistic particle, with a rest mass m and an electric charge q, flies
ballistically with velocity u by an immobile point charge q’, with an impact parameter b so large that the
deviations of its trajectory from the straight line are negligible. Calculate the total energy loss due to the
electromagnetic radiation during the passage. Quantify the conditions of validity of your result.
Solution: The easiest way to solve the problem is to use the formula obtained in the solution of
Problem 4:
Z q4 
2
 uE 
2

P  0   E  u  B 2    . (*)
6  mc    c  
In the lab frame where the charge q’ rests, it produces only the Coulomb electrostatic field,
q' r ut
E ,
4 0 r r
2

q u
directed along the vector r that connects the charge positions, and hence is an 
implicit function of time – see the figure on the right. According to the b r
conditions of our current problem, the velocity u of the charge q may be taken
for a constant, so from this sketch, we get
q'
r  r  b  u t 
2 2 1/ 2 u r ut ut
2
,   c cos  , cos    ,
c r r b  u 2 t 2 1 / 2
2

where t is the time referred to the instant of the closest approach of the particles. As a result, the
expression in the square brackets of Eq. (*) is reduced to
2 2 2
uE  q'   q'   
 1   2 cos 2    
1 u 2t 2
E      1   2 2 
2

 4 0 r  4 0  b  u 2 t 2  b  u 2t 2
2 2
 c  
2
 
Integrating the power (*) over the whole duration of the particle passage, we get the full energy loss:
 2 2
Z q4     q'  ,
 E   Pdt  0     I

6  mc   4 0 
where
2   2 d 
  
1   2u 2t 2  d ut
I  b 1  2  dt  3     0 1   2 3  , with  
2
.

2
u t 2 2 2
  b u t
2 2
 ub  0 1   2 2  b

Problems with Solutions Page 425

Essential Graduate Physics EM: Classical Electrodynamics

The first of these dimensionless integrals is well-known end equal to /4, while the second one may be
readily reduced to a sum of the same integral and one more integral of the same type:209

 2 d 
1     1d  d  d
2  
 3 
 1  
0
2 3

 
0 1     1     1   
2 3
0
2 2
0
2 3

4
  ,
16 16
so
2  2     2 
I      1   ,
ub 3  4 16  2ub 3  4 
and, finally, the radiative energy loss is210
2 2 2
Z q4     q'    2  1 Z 0 q 4  q'  1  2 / 4
 E  0     1      .
  1   
2 3 3 
6  mc   4 0  2ub
3
 4  12 m b c  4 0
2

This result has been obtained by neglecting the change of the particle’s trajectory and velocity
(and hence its energy) due to its interaction with the Coulomb center. These assumptions are valid if the
scales of the potential energy of interaction and the calculated energy loss are much smaller than the
particle’s kinetic energy:
qq'
 E ,  mc 2   1 .
4 0 b

209
See, e.g., MA Eq. (6.5b) with, respectively, n = 2 and n = 3.
210 Additionaltask for the reader: compare this result semi-qualitatively with the one that would follow from a
combination of Eq. (10.79) with an interpretation of Eq. (10.94), that would be suitable for our current problem.

Problems with Solutions Page 426