Chapter 5

Small Oscillations

5.1 Small oscillations about stable equilibrium

Consider a situation with N unconstrained generalized coordinates qi de-
scribed by a mass matrix Mij ({qk }) and a potential U ({qi }), and suppose
that U has a local minimum at some point in configuration space, qi = qi0 .
Then this point is a stable equilibrium point, for the generalized force at that
point is zero, and if the system is placed nearly at rest near that point, it will
not have enough energy to move far away from that point. We may study
the behavior of such motions by expanding the potential1 in Taylor’s series
expansion in the deviations ηi = qi − qi0 ,

X ∂U 1 X ∂2U
U (q1 , . . . , qN ) = U (qi0 ) + ηi + ηi ηj + ... .
i ∂qi 0 2 ij ∂qi ∂qj 0

The constant U (qi0 ) is of no interest, as only changes in potential matter,

so we may as well set it to zero. In the second term, − ∂U/∂qi |0 is the
generalized force at the equilibrium point, so it is zero. Thus the leading
term in the expansion is the quadratic one, and we may approximate

1X ∂2U
U ({qi }) = Aij ηi ηj , with Aij = . (5.1)
2 ij ∂qi ∂qj 0

Note that A is a constant symmetric real matrix.

1
assumed to have continuous second derivatives.

123
124 CHAPTER 5. SMALL OSCILLATIONS
P
The kinetic energy T = 12 Mij η̇i η̇j is already second order in the small
variations from equilibrium, so we may evaluate Mij , which in general can
depend on the coordinates qi , at the equilibrium point, ignoring any higher
order changes. Thus Mij is a constant. Thus both the kinetic and potential
energies are quadratic forms in the displacement η, which we think of as a
vector in N -dimensional space. Thus we can write the energies in matrix
form
1 1
T = η̇ T · M · η̇, U = η T · A · η. (5.2)
2 2
A and M are real symmetric matrices, and because any displacement corre-
sponds to positive kinetic and nonnegative potential energies, they are pos-
itive (semi)definite matrices, meaning that all their eigenvalues are greater
than zero, except that A may also have eigenvalues equal to zero (these
are directions in which the stability is neutral to lowest order, but may be
determined by higher order terms in the displacement).
Lagrange’s equation of motion
d ∂L ∂L d
0= − = M · η̇ + A · η = M · η̈ + A · η (5.3)
dt ∂ η̇i ∂ηi dt
is not necessarily diagonal in the coordinate η. We shall use the fact that
any real symmetric matrix can be diagonalized by a similarity transforma-
tion with an orthogonal matrix to reduce the problem to a set of independant
harmonic oscillators. While both M and A can be diagonalized by an orthog-
onal transformation, they can not necessarily be diagonalized by the same
one, so our procedure will be in steps:

1. Diagonalize M with an orthogonal transformation O1 , transforming the

coordinates to a new set x = O1 · η.

2. Scale the x coordinates to reduce the mass matrix to the identity ma-
trix. The new coordinates will be called y.

3. Diagonalize the new potential energy matrix with another orthogonal

matrix O2 , giving the final set of coordinates, ξ = O2 · y. Note this
transformation leaves the kinetic energy matrix diagonal because the
identity matrix is unaffected by similarity transformations.

The ξ are normal modes, modes of oscillation which are independent in

the sense that they do not affect each other.
5.1. SMALL OSCILLATIONS ABOUT STABLE EQUILIBRIUM 125

Let us do this in more detail. We are starting with the coordinates η and
the real symmetric matrices A and M , and we want to solve the equations
M · η̈ + A · η = 0. In our first step, we use the matrix O1 , which linear
algebra guarantees exists, that makes m = O1 · M · O1−1 diagonal. Note O1 is
P P
time-independent, so defining xi = j O1 ij ηj also gives ẋi = j O1 ij η̇j , and
1 T
T = η̇ · M · η̇
2
1 T  −1 
= η̇ · O1 · m · O1 · η̇
2
1 T 
= η̇ · O1T · m · (O1 · η̇)
2
1
= (O1 · η̇)T · m · (O1 · η̇)
2
1 T
= ẋ · m · ẋ.
2
Similarly the potential energy becomes U = 12 xT · O1 · A · O1−1 · x. We know
that the matrix m is diagonal, and the diagonal elements mii are all strictly
√
positive. To begin the second step, define the diagonal matrix Sij = mii δij
P
and new coordinates yi = Sii xi = j Sij xj , or y = S·x. Now m = S 2 = S T ·S,
so T = 12 ẋT · m · ẋ = 12 ẋT · S T · S · ẋ = 12 (S · ẋ)T · S · ẋ = 12 ẏ T · ẏ. In terms of
y, the potential energy is U = 12 y T · B · y, where

B = S −1 · O1 · A · O1−1 · S −1

is still a symmetric matrix2 .

Finally, let O2 be an orthogonal matrix which diagonalizes B, so C =
O2 · B · O2−1 is diagonal, and let ξ = O2 · y. Just as in the first step,
1 1
U = ξ T · O2 · B · O2−1 · ξ = ξ T · C · ξ,
2 2
while the kinetic energy
1 1 1
T = ẏ T · ẏ = ẏ T · O2T · O2 · ẏ = ξ˙T · ξ˙
2 2 2
is still diagonal. Because the potential energy must still be nonnegative,
√ all
the diagonal elements Cii are nonnegative, and we will call them ωi := Cii .
2
O1 · A · O1−1 is symmetric because A is and O1 is orthogonal, so O1−1 = O1T .
126 CHAPTER 5. SMALL OSCILLATIONS

Then
1 X ˙2 1X 2 2
T = ξ , U= ω ξ , ξ¨j + ωj2 ξj = 0,
2 j j 2 j j j

so we have N independent harmonic oscillators with the solutions

ξj = Re aj eiωj t ,

with some arbitrary complex numbers aj .

To find what the solution looks like in terms of the original coordinates
qi , we need to undo all these transformations. As ξ = O2 · y = O2 · S · x =
O2 · S · O1 · η, we have

q = q0 + O1−1 · S −1 · O2−1 · ξ.

We have completely solved this very general problem in small oscilla-

tions, at least in the sense that we have reduced it to a solvable problem of
diagonalizing symmetric real matrices. What we have done may appear ab-
stract and formal and devoid of physical insight, but it is a general algorithm
which will work on a very wide class of problems of small oscillations about
equilibrium. In fact, because diagonalizing matrices is something for which
computer programs are available, this is even a practical method for solving
such systems, even if there are dozens of interacting particles.

5.1.1 Molecular Vibrations

Consider a molecule made up of n atoms. We need to choose the right level
of description to understand low energy excitations. We do not want to
describe the molecule in terms of quarks, gluons, and leptons. Nor do we
need to consider all the electronic motion, which is governed by quantum
mechanics. The description we will use, called the Born-Oppenheimer
approximation, is to model the nuclei as classical particles. The electrons,
which are much lighter, move around much more quickly and cannot be
treated classically; we assume that for any given configuration of the nuclei,
the electrons will almost instantaneously find a quantum-mechanical ground
state, which will have an energy which depends on the current positions of
the nuclei. This is then a potential energy when considering the nuclear
motion. The nuclei themselves will be considered point particles, and we
5.1. SMALL OSCILLATIONS ABOUT STABLE EQUILIBRIUM 127

will ignore internal quantum-mechanical degrees of freedom such as nuclear

spins. So we are considering n point particles moving in three dimensions,
with some potential about which we know only qualitative features. There
are 3n degrees of freedom. Of these, 3 are the center of mass motion, which,
as there are no external forces, is simply motion at constant velocity. Some
of the degrees of freedom describe rotational modes, i.e. motions that the
molecule could have as a rigid body. For a generic molecule this would be
three degrees of freedom, but if the equilibrium configuration of the molecule
is linear, rotation about that line is not a degree of freedom, and so only two
of the degrees of freedom are rotations in that case. The remaining degrees
of freedom, 3n − 6 for noncollinear and 3n − 5 for collinear molecules, are
vibrations.

2
2 2
Figure 5.1: Some simple molecules in their equilibrium positions.

For a collinear molecule, it makes sense to divide the vibrations into

transverse and longitudinal ones. Considering motion in one dimension only,
the nuclei have n degrees of freedom, one of which is a center-of-mass motion,
leaving n − 1 longitudinal vibrations. So the remaining (3n − 5) − (n − 1) =
2(n − 2) vibrational degrees of freedom are transverse vibrational modes.
There are no such modes for a diatomic molecule.

Example: CO2
Consider first the CO2 molecule. As it is a molecule, there must be a position
of stable equilibrium, and empirically we know it to be collinear and sym-
metric, which one might have guessed. We will first consider only collinear
motions of the molecule. If the oxygens have coordinates q1 and q2 , and the
carbon q3 , the potential depends on q1 − q3 and q2 − q3 in the same way, so
128 CHAPTER 5. SMALL OSCILLATIONS

the equilibrium positions have q2 − q3 = −(q1 − q3 ) = b. Assuming no direct

force between the two oxygen molecules, the one dimensional motion may be
described near equilibrium by

1 1
U = k(q3 − q1 − b)2 + k(q2 − q3 − b)2
2 2
1 1 1
T = mO q̇1 + mO q̇2 + mC q̇32 .
2 2
2 2 2
We gave our formal solution in terms of displacements from the equilibrium
position, but we now have a situation in which there is no single equilibrium
position, as the problem is translationally invariant, and while equilibrium
has constraints on the differences of q’s, there is no constraint on the center
of mass. We can treat this in two different ways:

1. Explicitly fix the center of mass, eliminating one of the degrees of free-
dom.

2. Pick arbitrarily an equilibrium position. While the deviations of the

center-of-mass position from the equilibrium is not confined to small
excursions, the quadratic approximation is still exact.

First we follow the first method. We can always work in a frame where
the center of mass is at rest, at the origin. Then mO (q1 + q2 ) + mC q3 = 0
is a constraint, which we must eliminate. We can do so by dropping q3
as an independant degree of freedom, and we have, in terms of the two
displacements from equilibrium η1 = q1 + b and η2 = q2 − b, q3 = −(η1 +
η2 )mO /mC , and
 
1 1 1 mO
T = mO (η̇12 + η̇22 ) + mC η̇32 = mO η̇12 + η̇22 + (η̇1 + η̇2 )2
2 2 2 mC
  
1 m2O 1 + mC /mO 1 η̇1
= ( η̇1 η̇2 ) .
2 mC 1 1 + mC /mO η̇2

Now T is not diagonal, or more precisely M isn’t. We must find the orthog-
onal matrix O1 such that O1 · M · O1−1 is diagonal. We may assume it to be
a rotation, which can only be
 
cos θ − sin θ
O=
sin θ cos θ
5.1. SMALL OSCILLATIONS ABOUT STABLE EQUILIBRIUM 129

for some value of θ. It is worthwhile to derive a formula for diagonalizing a

general real symmetric 2 × 2 matrix and then plug in our particular form.
Let
   
a b c −s
M= , and O = ,
b d s c
where we have abbreviated s = sin θ, c = cos θ. We will require the matrix
element m12 = (O · M · O−1 )12 = 0, because m is diagonal. This determines
θ:
   
−1 c −s a b c s
O·M ·O =
s c b d −s c
    
c −s · as + bc · acs + bc2 − bs2 − scd
= =
· · · bs + cd · ·
where we have placed a · in place of matrix elements we don’t need to cal-
culate. Thus the condition on θ is
1
(a − d) sin θ cos θ + b(cos2 θ − sin2 θ) = 0 = (a − d) sin 2θ + b cos 2θ,
2
or
−2b
tan 2θ = .
a−d
Notice this determines 2θ only modulo π, and therefore θ modulo 90◦ , which
ought to be expected, as a rotation through 90◦ only interchanges axes and
reverses directions, both of which leave a diagonal matrix diagonal.
In our case a = d, so tan 2θ = ∞, and θ = π/4. As x = O1 η,
      
x1 cos π/4 − sin π/4 η1 1 η 1 − η2
= =√ ,
x2 sin π/4 cos π/4 η2 2 η1 + η2
and inversely
   
η1 1 x1 + x2
=√ .
η2 2 −x1 + x2
Then
" #
1 (ẋ1 + ẋ2 )2 (ẋ1 − ẋ2 )2 mO √
T = mO + + ( 2ẋ2 )2
2 2 2 mC
130 CHAPTER 5. SMALL OSCILLATIONS
 
1 1 2mO 2
= mO ẋ21 + mO 1 + ẋ2
2 2 mC
1 1
U = k(q3 − q1 − b)2 + k(q2 − q3 − b)2
2 " 2
 2  2 #
1 mO mO
= k η1 + (η1 + η2 ) + η2 + (η1 + η2 )
2 mC mC
" #
1 2 2 2m2O 2 2mO 2
= k η1 + η2 + 2 (η1 + η2 ) + (η1 + η2 )
2 mC mC
" #
1 4m O
= k x21 + x22 + 2 (mO + mC )x22
2 mC
 
1 2 1 mC + 2mO 2 2
= kx1 + k x2 .
2 2 mC

Thus U is already diagonal and we don’t need to go through steps 2 and 3,

the scaling and second orthogonalization, except to note that if we skip the
scaling the angular frequencies are given by ωi2 = coefficient
q in U / coefficient
in T . Thus we have one normal mode, x1 , with ω1 = k/mO , with x2 = 0,
η1 = −η2 , q3 = 0, in which the two oxygens vibrate in and out together,
symmetrically about the carbon, which doesn’t move. We also have another
mode, x2 , with
v s
u
u k(mC + 2mO )2 /m2C k(mC + 2mO )
ω2 = t = ,
mO (1 + 2mO /mC ) mO mC

with x1 = 0, η1 = η2 , in which the two oxygens move right or left together,

with the carbon moving in the opposite direction.
We have successfully solved for the longitudinal vibrations by eliminating
one of the degrees of freedom. Let us now try the second method, in which
we choose an arbitrary equilibrium position q1 = −b, q2 = b, q3 = 0. Then

1 1
T = mO (η̇12 + η̇22 ) + mC η̇32
2 2
1 h i
U = k (η1 − η3 ) + (η2 − η3 )2 .
2
2
T is already diagonal, so O1 = 1I, x = η. In the second step S is the diagonal
√ √ √
matrix with S11 = S22 = mO , S33 = mC , and yi = mO ηi for i = 1, 2,
5.1. SMALL OSCILLATIONS ABOUT STABLE EQUILIBRIUM 131

√
and y3 = mC η3 . Then
 !2 !2 
1  y1 y3 y2 y3 
U = k √ −√ + √ −√
2 mO mC mO mC
1 k h √ i
= mC y12 + mC y22 + 2mO y32 − 2 mO mC (y1 + y2 )y3 .
2 mO mC
Thus the matrix B is
 √ 
mC 0 − mO mC
 √
B= 0 m − mO mC  ,
√ √ C
− mO mC − mO mC 2mO
√ √ √
which is singular, as it annihilates the vector y T = ( mO , mO , mC ),
which corresponds to η T = (1, 1, 1), i.e. all the nuclei are moving by the same
amount, or the molecule is translating rigidly. Thus this vector corresponds
to a zero eigenvalue of U , and a harmonic oscillation of zero frequency. This is
free motion3 , η = η0 +vt. The other two modes can be found by diagonalizing
the matrix, and will be as we found by the other method.

Transverse motion
What about the transverse motion? Consider the equilibrium position of
the molecule to lie in the x direction, and consider small deviations in the z
direction. The kinetic energy
1 1 1
T = mO ż1 + mO ż22 + mC ż32 .
2 2 2
is already diagonal, just as for
the longitudinal modes in the sec-
ond method. Any potential en-
ergy must be due to a resistance
to bending, so to second order,
U ∝ (ψ − θ)2 ∼ (tan ψ − tan θ)2 = θ 2
ψ
[(z2 −z3 )/b+(z1 −z3 )/b]2 = b−2 (z1 +
z2 − 2z3 )2 . 1
Note that the potential energy is proportional to the square of a single linear
3
To see that linear motion is a limiting case of harmonic motion as ω → 0, we need to
choose the complex coefficient to be a function of ω, A(ω) = x0 − iv0 /ω, with x0 and v0
real. Then x(t) = limω→0 Re A(ω)eiωt = x0 + v0 limω→0 sin(ωt)/ω = x0 + v0 t
132 CHAPTER 5. SMALL OSCILLATIONS

combination of the displacements, or to the square of one component (with

respect to a particular direction) of the displacement. Therefore there is no
contribution of the two orthogonal directions, and there are two zero modes,
or two degrees of freedom with no restoring force. One of these is the center of
mass motion, z1 = z2 = z3 , and the other is the third direction in the abstract
space of possible displacements, z T = (1, −1, 0), with z1 = −z2 , z3 = 0, which
we see is a rotation. Thus there remains only one true transverse vibrational
mode in the z direction, and also one in the y direction, which together with
the two longitudinal ones we found earlier, make up the 4 vibrational modes
we expected from the general formula 2(n − 2) for a collinear molecule.
You might ask whether these oscillations we have discussed are in any
way observable. Quantum mechanically, a harmonic oscillator can only be in
states with excitation energy E = nh̄ω, where n ∈ Z is an integer and 2πh̄
is Planck’s constant. When molecules are in an excited state, they can emit
a photon while changing to a lower energy state. The energy of the photon,
which is the amount lost by the molecule, is proportional to the frequency,
∆E = 2πh̄f , so by measuring the wavelength of the emitted light, we can
determine the vibrational frequencies of the molecules. So the calculations
we have done, and many others for which we have built the apparatus, are
in fact very practical tools for molecular physics.

5.1.2 An Alternative Approach

The step by step diagonalization we just gave is not the easiest approach to
solving the linear differential equation (5.3). Solutions to linear differential
equations are subject to superposition, and equations with coefficients inde-
pendent of time are simplified by Fourier transform, so we can express the
N dimensional vector of functions ηj (t) as
Z ∞
ηj (t) = dωψj (ω)e−iωt .
−∞

Then the Lagrange equations become

Z ∞ X 
dω Aij − ω 2 Mij ψj (ω)e−iωt = 0 for all t.
−∞ j

But e−iωt are linearly independent functions of t ∈ R, so

X 
Aij − ω 2 Mij ψj (ω) = 0.
j
5.2. OTHER INTERACTIONS 133

This implies ψj (ω) = 0 except when the matrix Aij − ω 2 Mij is singular,
det (Aij − ω 2 Mij ) = 0, which gives a discrete set of angular frequencies
ω1 . . . ωN , and for each ωj an eigenvector ψj .

5.2 Other interactions

In our treatment we assumed a Lagrangian formulation with a kinetic term
purely quadratic in ~q˙, together with a velocity independent potential. There
is a wider scope of small oscillation problems which might include dissipative
forces like friction, or external time-dependent forces, or perhaps terms in
the Lagrangian linear in the velocities. An example of the latter occurs
in rotating reference frames, from the Coriolus force, and is important in
the question of whether there is a gravitationally stable location for small
objects caught between the Earth and the moon at the “L5” point4 . Each of
these complications introduces terms, even in the linear approximation to the
equations of motion, which cannot be diagonalized away, because there is not
significant freedom of diagonalization left, in general, after having simplified
T and U . Thus the approach of section 5.1 does not generalize well, but the
approach of section 5.1.2 can be applied.
For example, we might consider adding a generalized force Qi on ηi , con-
P
sisting of a dissipative force j Rij η̇j and a driving force Fi . We will assume
R is a symmetric matrix, as might be a result of a Rayleigh dissipation func-
tion (see Section 2.7 or Ref. [6]). We will consider the motion to first order
in F , so any coordinate dependence of R or F is replaced, as it was for M
and A, by their values at the equilibrium position. Thus the equations of
motion become X
(Mij η̈j + Rij η̇j + Aij ηj ) − Fi = 0.
j

Again making the ansatz that

Z ∞
ηj (t) = dωψj (ω)e−iωt
−∞

and expressing Fi (t) in terms of its fourier transform

Z ∞
Fj (t) = dω f˜i (ω)e−iωt
−∞

4
See problem 5.3.
134 CHAPTER 5. SMALL OSCILLATIONS

we find X 
−ω 2 Mij − iωRij + Aij ψj = f˜i .
j

Except for at most 2N values of ω the matrix multiplying ψj will have a non-
zero determinant and will be invertible, allowing us to find the response ψj
to the fourier component of the driving force, f˜i . Those values of ω for which
the determinant vanishes, and the vector ψj which the matrix annihilates,
correspond to damped modes that we would see if the driving force were
removed.

5.2.1 Forced Harmonic Oscillations

In this section we will explore the effects of driving forces on oscillators. For
simplicity let us consider a damped oscillator with one degree of freedom,
with a driving force F (t):

mẍ(t) + Rẋ(t) + kx(t) = F (t).

For thisZ linear oscillator, we can solve by Fourier transform. Writing

∞
x(t) = x̃(ω)e−iωt dω, we find (−mω 2 − iRω + k) x̃(ω) = F̃ (ω), where the
−∞
1 Z∞
Fourier transformed force is F̃ (ω) := F (t)eiωt dt.
2π −∞
Without
q any forces, we have solutions for ω 2 + 2iρω − ω02 = 0, (where
ω0 := k/m, ρ = R/2m), so the solutions are at ω = −iρ ± ω̄, with ω̄ =
q
ω02 − ρ2 . Due to the negative imaginary part of either of these ωs, the
unforced oscillations will decay with time. If we do have a forcing function,
however, we have an inhomogeneous solution (with f˜ = F̃ /m)

f˜(ω)
x̃(ω) = .
ω02 − ω 2 − 2iρω
As x(t) and f (t) are real-valued functions of time, the fourier transforms
must satisfy x̃∗ (ω) = x̃(−ω), f˜∗ (ω) = f˜(−ω), and the steady state solution
is
Z ∞   Z ∞
−iωt ∗ iωt
x(t) = dω x̃(ω)e + x̃ (ω)e = 2 Re dωx̃(ω)e−iωt
0 0
Z ∞ f˜(ω)
= 2 Re dω e−iωt . (5.4)
0 ω02 − ω 2 − 2iρω
5.2. OTHER INTERACTIONS 135

If we consider a forcing function of only one positive frequency, say f˜(ω) =

aδ(ω − ωex ) for ω ≥ 0, we have
a
x(t) = 2 Re e−iωex t ,
ω02 − ωex − 2iρωex
2

with amplitude

2a
A =
ω02 − ωex
2 − 2iρω
ex
2|a|
= q .
2
(ω02 − ωex
2 ) + 4ρ2 ω 2
ex ρ

We see that the response in the fre-

quency domain is proportional to the force,
with a frequency dependence which is
sharply peaked if the damping coefficient ρ 1

is small compared to the natural frequency, 1 ω/ω0 2

ρ  ω0 .
If we ask in the temporal domain, what is the effect on x(t) of a force
f (t0 ), we have
Z
1 Z∞ 0
∞ f (t0 ) 0
x(t) = dω dt 2 eiω(t−t )
−∞ 2π −∞ ω0 − ω − 2iρω
2
Z ∞
f (t0 )
= dt0 G(t − t0 ) , (5.5)
−∞ m
where
1 Z∞
0
0 e−iω(t−t )
G(t − t ) := dω 2 (5.6)
2π −∞ ω0 − ω 2 − 2iρω
is the temporal Green’s function. Note that it would appear from (5.5) that
effects could preceed causes, as the integral is over all t0 , including times
after t, but in fact G vanishes there. For t0 > t we may evaluate (5.6) by
closing the integration contour in the upper half plane, for the exponential
will vanish for large Im ω > 0 when t < t0 . As the integrand is analytic
in the upper half plane, the contour integral vanishes, and G(t − t0 ) = 0 for
t0 > t. On the other hand, for ∆t = t − t0 > 0, the integration contour may
be closed in the lower half plane, picking up the residues from the poles at
136 CHAPTER 5. SMALL OSCILLATIONS

ω = −iρ ± ω̄. The residue there is ∓e−ρ ∆t e∓iω̄∆t /ω̄, so

1 −ρ ∆t
G(∆t) = e sin ω̄∆t.
ω̄

Weakly nonlinear oscillating systems

The oscilator we just considered could be solved exactly because it is a linear
system. The equation of motion is a linear operator (including time deriva-
tive operators) acting on the dynamical variable x(t), set equal to a forcing
term which is a given function of time. Most systems, however, are not ex-
actly linear. If the equation of motion is close to linear, we might imagine a
perturbative calculation in which we bring the difference from linearity, con-
sidered small, to the right hand side, evaluate it in the linear approximation,
and consider it a forcing term. For example, we are quite used to the idea
that a pendulum may be approximated by a harmonic oscillator. A forced,
linearly damped pendulum has an equation of motion

m`2 θ̈ + Rθ̇ + mg` sin θ = F (t),

which in the approximation sin θ ≈ θ reduces to the harmonic oscillator we

just considered. More precisely, we can write

θ̈ + 2ρθ̇ + ω02 θ = f (t) − ω02 (sin θ − θ),

q
where ρ = R/2m`2 , ω0 = g/` and f (t) = F (t)/m`2 . If the forcing function
f (t) and the oscillations are small (θ  π), we can imagine a sequence of
approximations, first evaluating θ(t) dropping the (sin θ − θ) term, and then
evaluating the n+1’st approximation to θ(t) by using the n’th approximation
to evaluate (sin θ − θ)(t) as a forcing term.
We will return to this issue, discussing both how this works and why it
may not be the ideal way to do a perturbative expansion, in Chapter 7.

5.3 String dynamics

In this section we consider two closely related problems, transverse oscilla-
tions of a stretched loaded string, and of a stretched heavy string. The latter
is is a limiting case of the former. This will provide an introduction to field
theory, in which the dynamical degrees of freedom are not a discrete set but
5.3. STRING DYNAMICS 137

are defined at each point in space. In Chapter 8 we will discuss more in-
teresting and involved cases such as the electromagnetic field, where at each
point in space we have E ~ and B
~ as degrees of freedom, though not without
constraints.
The loaded string we will consider is a light string under tension τ stretched
between two fixed points a distance ` apart, say at x = 0 and x = `. On
the string, at points x = a, 2a, 3a, . . . , na, are fixed n particles each of mass
m, with the first and last a distance a away from the fixed ends. Thus
` = (n + 1)a. We will consider only small transverse motion of these masses,
using yi as the transverse displacement of the i’th mass, which is at x = ia.
We assume all excursions from the equilibrium positions yi = 0 are small, and
in particular that the difference in successive displacements yi+1 − yi  a.
Thus we are assuming that the angle made by each segment of the string,
θi = tan−1 [(yi+1 − yi )/a]  1. Working to first order in the θ’s in the equa-
tions of motion, and second order for the Lagrangian, we see that restricting
our attention to transverse motions and requiring no horizontal motion forces
taking the tension τ to be constant along the string. The transverse force on
the i’th mass is thus
yi+1 − yi yi−1 − yi τ
Fi = τ +τ = (yi+1 − 2yi + yi−1 ).
a a a
The potential energy U (y1 , . . . , yn ) then satisfies
∂U τ
= − (yi+1 − 2yi + yi−1 )
∂yi a
so
U (y1 , . . . , yi , . . . , yn )
Z yi
τ
= dyi (2yi − yi+1 − yi−1 ) + F (y1 , . . . , yi−1 , yi+1 , . . . , yn )
0 a
τ 2 
= yi − (yi+1 + yi−1 )yi + F (y1 , . . . , yi−1 , yi+1 , . . . , yn )
a  
τ
= (yi+1 − yi )2 + (yi − yi−1 )2 + F 0 (y1 , . . . , yi−1 , yi+1 , . . . , yn )
2a
Xn
τ
= (yi+1 − yi )2 + constant.
i=0 2a
The F and F 0 are unspecified functions of all the yj ’s except yi . In the last
expression we satisfied the condition for all i, and we have used the convenient
definition y0 = yn+1 = 0. We can and will drop the arbitrary constant.
138 CHAPTER 5. SMALL OSCILLATIONS
P
The kinetic energy is T = 12 m n1 ẏi2 .
Before we continue with the analysis of this problem, let us note that
another physical setup also leads to the same Lagrangian. Consider a one
dimensional lattice of identical atoms with a stable equilibrium in which they
are evenly spaced, with interactions between nearest neighbors. Let ηi be the
longitudinal displacement of the i’th atom from its equilibrium position. The
P
kinetic energy is simply T = 12 m n1 η̇i2 . As the interatomic distance differs
from its equilibrium position by ηi+1 − ηi , the interaction potential of atoms
i and i + 1 can be approximated by U (ηi+1 , ηi ) ≈ 12 k(ηi+1 − ηi )2 . We have
in effect atoms separated by springs of spring constant k, and we see that
if k = τ /a, we get the same Lagrangian for longitudinal oscillations of this
lattice as we had for the transverse oscillations of the loaded string.
P
As the kinetic energy is simply T = 12 m n1 ẏi2 , the mass matrix is already
proportional to the identity matrix and we do not need to go through the
first two steps of our general process. The potential energy U = 12 y T · A · y
has a non-diagonal n × n matrix
 
−2 1 0 0 ··· 0 0
 1 −2 1 0 · · · 0 0 
 
 
τ

0 1 −2 1 · · · 0 0 
A = −  .. .. .. .. . . .. ..  .
a . . . . . . . 

 
 0 0 0 0 · · · −2 1 
0 0 0 0 · · · 1 −2

Diagonalizing even a 3 × 3 matrix is work, so an n × n matrix might seem

out of the question, without some hints from the physics of the situation. In
this case the hint comes in a roundabout fashion — we will first consider a
limit in which n → ∞, the continuum limit, which leads to an interesting
physical situation in its own right.
Suppose we consider the loaded string problem in the limit that the spac-
ing a becomes very small, but the number of masses m becomes large, keeping
the total length ` of the string fixed. If at the same time we adjust the in-
dividual masses so that the mass per unit length, ρ, is fixed, our bumpy
string gets smoothed out in the limit, and we might expect that in this limit
we reproduce the physical problem of transverse modes of a uniformly dense
stretched string, like a violin string. Thus we wish to consider the limit

a → 0, n → ∞, ` = (n + 1)a fixed, m → 0, ρ = m/a fixed.

5.3. STRING DYNAMICS 139

It is natural to think of the degrees of freedom as associated with the label x

rather than i, so we redefine the dynamical functions {yj (t)} as y(x, t), with
y(ja, t) = yj (t). While this only defines the function at discrete points in x,
these are closely spaced for small a and become dense as a → 0. We will
assume that the function y(x) is twice differentiable in the continuum limit,
though we shall see that this is not the case for all possible motions of the
discrete system.
What happens to the kinetic and potential energies in this limit? For the
kinetic energy,
1 X 2 1 X 2 1 X 1 Z`
T = m ẏi = ρ aẏ (xi ) = ρ ∆xẏ (xi ) → ρ dx ẏ 2 (x),
2
2 i 2 i 2 i 2 0
where the next to last expression is just the definition of a Riemann integral.
For the potential energy,
 2 !2
τ X τX yi+1 − yi τZ` ∂y
U= (yi+1 − yi )2 = ∆x → dx .
2a i 2 i ∆x 2 0 ∂x
The equation of motion for yi is
∂L ∂U τ
mÿi = =− = [(yi+1 − yi ) − (yi − yi−1 )],
∂yi ∂yi a
or
τ
ρaÿ(x) = ([y(x + a) − y(x)] − [y(x) − y( x − a)]).
a
We need to be careful about taking the limit
y(x + a) − y(x) ∂y
→
a ∂x
because we are subtracting two such expressions evaluated at nearby points,
and because we will need to divide by a again to get an equation between
finite quantities. Thus we note that
y(x + a) − y(x) ∂y
= + O(a2 ),
a ∂x x+a/2
so
!
τ y(x + a) − y(x) y(x) − y( x − a)
ρÿ(x) = −
a a a
 
τ  ∂y ∂y →τ
∂2y
≈ − ,
a ∂x x+a/2
∂x x−a/2
∂x2
140 CHAPTER 5. SMALL OSCILLATIONS

and we wind up with the wave equation for transverse waves on a massive
string
∂ 2y 2
2∂ y
− c = 0,
∂t2 ∂x2
where s
τ
c= .
ρ
Solving this wave equation is very simple. For the fixed boundary condi-
tions y(x) = 0 at x = 0 and x = `, the solution is a fourier expansion
∞
X
y(x, t) = Re Bp eickp t sin kp x,
p=1

where kp ` = pπ. Each p represents one normal mode, and there are an
infinite number as we would expect because in the continuum limit there are
an infinite number of degrees of freedom.
We have certainly not shown that y(x) = B sin kx is a normal mode for
the problem with finite n, but it is worth checking it out. This corresponds
to a mode with yj = B sin kaj, on which we apply the matrix A
X τ
(A · y)i = Aij yj = − (yi+1 − 2yi + yi−1 )
j a
τ
= − B (sin(kai + ka) − 2 sin(kai) + sin(kai − ka))
a
τ
= − B(sin(kai) cos(ka) + cos(kai) sin(ka) − 2 sin(kai)
a
+ sin(kai) cos(ka) − cos(kai) sin(ka))
τ
= B (2 − 2 cos(ka)) sin(kai)
a
2τ
= (1 − cos(ka)) yi .
a
So we see that it is a normal mode, although the frequency of oscillation
s s
2τ τ sin(ka/2)
ω= (1 − cos(ka)) = 2
am ρ a
q
differs from k τ /ρ except in the limit a → 0 for fixed k.
The wave numbers k which index the normal modes are restricted by
the fixed ends to the discrete set k = pπ/` = pπ/(n + 1)a, for p ∈ Z, i.e. p is
5.4. FIELD THEORY 141

an integer. This is still too many (∞) for a system with a finite number of
degrees of freedom. The resolution of this paradox is that not all different k’s
correspond to different modes. For example, if p0 = p + 2m(n + 1) for some
integer m, then k 0 = k + 2πm/a, and sin(k 0 aj) = sin(kaj + 2mπ) = sin(kaj),
so k and k 0 represent the same normal mode. Also, if p0 = 2(n + 1) − p,
k 0 = (2π/a)−k, sin(k 0 aj) = sin(2π −kaj) = − sin(kaj), so k and k 0 represent
the same normal mode, with opposite phase. Finally p = n + 1, k = π/a
gives yj = B sin(kaj) = 0 for all j and is not a normal mode. This leaves as
independent only p = 1, ..., n, the right number of normal modes for a system
with n degrees of freedom.
The angular frequency of the p’th normal mode
r
τ pπ
ωp = 2 sin
ma 2(n + 1)

in plotted in Fig. 5.3. For fixed values of p and ρ, as n → ∞,

s s
τ1 paπ τ pπ
ωp = 2 sin →2 = ckp ,
ρa 2` ρ 2`

as we have in the continuum limit.

But if we consider modes with a
fixed ratio of p/n as n → ∞, we
do not have a smooth limit y(x),
and such nodes are not appropri-
ate for the continuum limit. In the
physics of crystals, the former kind
of modes are known as accous-
tic modes, while the later modes,
in particular those for n − p fixed,
which depend on the discrete na- 1 2 3 4 5 6 7 8 9 10 11 12

ture of the crystal, are called opti- Fig. 5.3. Frequencies of oscillation
cal modes. of the loaded string.

5.4 Field theory

We now examine how to formulate the continuum limit directly.
142 CHAPTER 5. SMALL OSCILLATIONS

5.4.1 Lagrangian density

We saw in the last section that the kinetic and potential energies in the
continuum limit can be written as integrals over x of densities, and so we
may also write the Lagrangian as the integral of a Lagrangian density
L(x),
 !2 
Z L 1 1 ∂y(x, t)
L=T −U = dxL(x), L(x) =  ρẏ 2 (x, t) − τ .
0 2 2 ∂x

This Lagrangian, however, will not be of much use until we figure out what is
meant by varying it with respect to each dynamical degree of freedom or its
corresponding velocity. In the discrete case we have the canonical momenta
Pi = ∂L/∂ ẏi , where the derivative requires holding all ẏj fixed, for j 6= i, as
P
well as all yk fixed. This extracts one term from the sum 12 ρ aẏi2 , and this
would appear to vanish in the limit a → 0. Instead, we define the canonical
momentum as a density, Pi → aP (x = ia), so

1 ∂ X
P (x = ia) = lim a L(y(x), ẏ(x), x)|x=ai .
a ∂ ẏi i

We may think of the last part of this limit,

X Z
lim a L(y(x), ẏ(x), x)|x=ai = dx L(y(x), ẏ(x), x),
a→0
i

if we also define a limiting operation

1 ∂ δ
lim → ,
a→0 a ∂ ẏi δ ẏ(x)
1 ∂
and similarly for a ∂yi
, which act on functionals of y(x) and ẏ(x) by

δy(x1 ) δ ẏ(x1 ) δy(x1 ) δ ẏ(x1 )

= δ(x1 − x2 ), = = 0, = δ(x1 − x2 ).
δy(x2 ) δy(x2 ) δ ẏ(x2 ) δ ẏ(x2 )

Here δ(x0 − x) is the Dirac delta function, defined by its integral,

Z x2
f (x0 )δ(x0 − x)dx0 = f (x)
x1
5.4. FIELD THEORY 143

for any function f (x), provided x ∈ (x1 , x2 ). Thus

δ Z ` 01 2 0 Z `
P (x) = dx ρẏ (x , t) = dx0 ρẏ(x0 , t)δ(x0 − x) = ρẏ(x, t).
δ ẏ(x) 0 2 0

We also need to evaluate

!2
δ δ Z ` 0 −τ ∂y
L= dx .
δy(x) δy(x) 0 2 ∂x x=x0

For this we need

δ ∂y(x0 ) ∂
0
= 0
δ(x0 − x) := δ 0 (x0 − x),
δy(x) ∂x ∂x
which is again defined by its integral,
Z x2 Z x2 ∂
f (x0 )δ 0 (x0 − x)dx0 = f (x0 ) δ(x0 − x)dx0
x1 x1 ∂x0
Z x2
0 0 x ∂f
= f (x )δ(x − x)|x21 − dx0 δ(x0 − x)
x1 ∂x0
∂f
= (x),
∂x
where after integration by parts the surface term is dropped because δ(x −
x0 ) = 0 for x 6= x0 , which it is for x0 = x1 , x2 if x ∈ (x1 , x2 ). Thus
Z `
δ ∂y ∂2y
L = − dx0 τ (x0 )δ 0 (x0 − x) = τ 2 ,
δy(x) 0 ∂x ∂x
and Lagrange’s equations give the wave equation
∂2y
ρÿ(x, t) − τ = 0. (5.7)
∂x2
We have derived the wave equation for small transverse deformations of
a stretched string by considering the continuum limit of a loaded string,
in the process demonstating how to formulate Lagrangian mechanics for a
continuum system. Of course it is more usual, and simpler, to derive it
directly by considering Newton’s law on an infinitesimal element of the string.
Let’s include gravity for good measure. If the string point initially at x
144 CHAPTER 5. SMALL OSCILLATIONS

has a transverse displacement y(x)

and a longitudinal displacement
η(x), both considered small, the τ x ∆x
θ x θ x ∆x
slope of the string dy/dx is also
small. The segment [x, x + ∆x] has ∆x
τx
a mass ρ∆x, where as before ρ is x x+∆x
the mass per unit length, and the
forces on it are
in x direction: τ (x + ∆x) cos θ(x + ∆x) − τ (x) cos θ(x) = ρ∆xη̈
in y direction: τ (x + ∆x) sin θ(x + ∆x) − τ (x) sin θ(x) − ρg∆x = ρ∆xÿ

As θ << 1, we can replace cos θ by 1 and sin θ with tan θ = ∂y/∂x, and
then from the first equation we see that ∂τ /∂x is already small, so we can
consider τ as constant in the second equation, which gives
!
∂y ∂y
τ − − ρg∆x = ρ∆xÿ,
∂x x+∆x
∂x x

or
∂2y
τ − ρg = ρÿ.
∂x2
This agrees with Eq. 5.7 if we drop the gravity term, which we had not
included in our discussion of the loaded string.

5.4.2 Three dimensional continua

Could we do the same kind of analysis on a three dimensional solid object?
We might label each piece of the object with an equilibrium or reference
position ~x, and consider the dynamics of possible displacements ~η (~x). We
will assume this displacement is small and smooth function of ~x and t, in
fact twice differentiable. Consider the dynamics of an infinitesimal volume
element ∆V . The acceleration of each volume element will be determined by
the ratio of the net force on that volume to its mass, ρ∆V , where ρ is now the
density, mass per unit volume, and is also assumed to be a smooth function,
though not necessarily constant. The forces we will consider will be of two
types. There may be external forces which will be taken to be extensive,
that is, proportional to the volume, called volume forces. One example is
gravity near the Earth’s surface, with F~ = −ρg∆V êz . If the material under
5.4. FIELD THEORY 145

discussion had an electric charge density ρE (~x) in an external electric field

~ x), there would also be a volume force ρE (~x)E(~
E(~ ~ x) ∆V . In addition to the
volume forces, there are also surface forces which the rest of the object
exerts on the element under consideration. We will assume that all such
forces are local, due to the material on the other side of the surface, and
continuously varying, so the force across an infinitesimal element of surface
dS will be proportional to its area, at least if we keep the direction fixed.

In fact, we can show that the force

~ is linear
across an infinitesimal surface dS
~
in the vector dS even when the direction
changes. Consider two elements dS ~1 and
dS~2 , shown as rectangles, and the third 1
side dS ~3 , which is the opposite of their
sum in the limit that size shrinks to zero.
Together with the two parallel triangular
pieces, these bound an infinitesimal vol-
ume. Let us scale the whole picture by a
factor λ. The force on each side is propor-
tional to λ2 , but the mass of the volume
2
is proportional to λ3 , so as λ → 0, the
coefficient of λ2 in the sum of the forces must vanish. The triangular pieces
cancel each other, so the sum of the forces through dS ~1 and dS ~2 cancels
~
the force through dS3 . That is, the force is linear (additive) in the surface
elements dS.~
But the force is not necessarily in the same direction as dS. ~ This would
be true for the pressure in a gas, or in a nonviscous or static fluid, in which no
tangential forces could be exerted along the boundary. But more generally,
a force linear in dS~ will be specified by a matrix, and the force exerted on
dV across dS ~ will be Fi = − Pj Pij dSj , where P is known as the stress
tensor5 .

5
P
To be clear: jPij dSj is the force exerted by the back side of the surface element on
R
~ is an outward normal, the force on the volume is − P Pij dSj ,
the front side, so if dS S j
and a pressure corresponds to P = +pδij . This agrees with Symon ([17]) but has a reversed
sign from Taylor’s ([18]) Σ = −P.
146 CHAPTER 5. SMALL OSCILLATIONS

Though P is not a scalar or di-

agonal in general, there is one con-
straint on the stress tensor — it is
symmetric. To see this, consider z
the prism shown, and the torque in y
the y direction. The forces across
the two faces perpendicular to z λ
are of order λ, and are equal and
opposite, so they provide a torque λ x
−λ2 hPxz in the y direction. Sim-
ilarly the two faces perpendicular to x provide a torque +λ2 hPzx in that
direction. The equal forces on other two faces have a moment arm parallel
to y and therefore provide no torque in that direction. But the moment of
inertia about the y axis is of order λ2 dV = λ4 h. So if the angular acceleration
is to remain finite as λ → 0, we must have Pzx − Pxz = 0, and P must be a
symmetric matrix.
We expect that the stress forces the material on one side of a boundary
exerts on the other is due to some distortion of the material. Near any value
of x, we may expand the displacement as
X ∂ηi
ηi (x + ∆x) = ηi (x) + ∆xj + ...
j ∂xj

Moving the entire object as a whole, ~η (x) = constant, or rotating it as a

P
rigid body about an axis ω ~ , with ∂ηi /∂xj = k ijk ωk , will not produce any
stress, and so we will not consider such displacements to be part of the strain
tensor, which we therefore define to be the symmetric part of the derivative
matrix: !
1 ∂ηi ∂ηj
Sij = + .
2 ∂xj ∂xi
In general, the properties of the material will determine how the stress tensor
is related to the strain tensor, though for small displacements we expect it
to depend linearly.
Even linear dependence could be quite complex, but if the material prop-
erties are rotationally symmetric, things are fairly simple. Of course in a crys-
tal we might not satisfy that condition, but if we do assume the functional
dependence of the stress on the strain is rotationally invariant, we may find
the most general possibilities by decomposing the tensors into pieces which
5.4. FIELD THEORY 147

behave suitably under rotations. Here we are generalizing the idea that a
vector cannot be defined in terms of pure scalars, and a scalar can depend on
vectors only through a scalar product. A symmetric tensor consists of a piece,
its trace, which behaves like a scalar, and a traceless piece, called the de-
viatoric part, which behaves differently, as an irreducible representation6 .
The only possible linear relationships are thus
 
1 1
Tr P = −α Tr S; Pij − δij Tr P = −β Sij − δij Tr S . (5.8)
3 3
These are known as the generalized Hooke’s law for an elastic solid.
The tensor stress and strain we have described here are perhaps not as fa-
miliar as some other relations met in more elementary courses. First consider
the bulk modulus B, the inverse of the ratio of the fractional decrease in
volume to the isotropic pressure which causes it. Here the stress and strain
tensors are both multiples of the identity, P = +pδij and d~η = −cd~x, so
S = −cδij and c = p/α. For a linear contraction ~x → ~x − c~x the volume will
contract by ∆V = −3cV . Therefore the
p p α
bulk modulus B := = = .
−dV /V 3c 3

Next, consider a shear, in which the displacement might be ~η = cyêx

produced by forces ±Fx on the
horizontal faces shown, and ±Fy Fx
on the vertical faces. To have
no rotation we need wFx = LFy . Fy w Fy
The shear modulus G is defined by Fx
−Pxy = Fx /A = Gdηx /dy = Gc, L
where A is the area of the top face.
As !
1 ∂ηx ∂ηy 1 c
Sxy = + = (c + 0) =
2 ∂y ∂x 2 2
6
Representations of a symmetry group are defined as vector spaces which are invariant
under the action of the symmetry, and irreducible ones are those for which no proper
subspace is closed in that fashion. For more on this, see any book on group theory for
physicists. But for representations of the rotation group a course in quantum mechanics
may be better. The traceless part of the symmetric tensor transforms like a state with
angular momentum 2.
148 CHAPTER 5. SMALL OSCILLATIONS

and all other components are zero, we can set

Pxy
β=− = 2G.
Sxy

Finally, consider a rod being pulled by a force F stretching a distance ∆L

over a length L. Hooke’s constant is k = F/∆L and Young’s modulus
Y is defined by
F ∆L
=Y so Y = kL/A.
A L
The strain S11 = ∆L /L, and the stress has −P11 = Y S11 , with all other
components of the stress zero. But there may be displacement in the trans-
verse directions. If the rod is axially symmetric we may assume S22 = S33 ,
so

− Tr P = −P11 = Y S11 = α Tr S = α (S11 + 2S22 ) ,

   
1 Y 1 β
− P22 − Tr P = 0 − S11 = β S22 − Tr S = (S22 − S11 )
3 3 3 3
Thus solving the two equations

Y S11 = α (S11 + 2S22 )

Y β
− S11 = (S22 − S11 )
3 3
gives the value of Young’s modulus
3αβ
Y =
2α + β
and the contraction of the transverse dimensions,
β−α
S22 = S11 .
2α + β

The Equation of Motion

Now that the generalized Hooke’s law provides the forces for a solid in a given
configuration, we can write down the equations of motion. The infinitesimal
volume
R
originally at the reference point ~r is at position ~r + ~η (~r, t). Its mass
is V ρdV , and the force on it is the sum of the volume force and the surface
5.4. FIELD THEORY 149
R
force. We will write the volume force as F~vol = V E(~
~ r)dV , where E could be
−ρgêz for gravity or some other intensive external force. The surface force is
Z X Z
Fisurf = − Pij (~r)dSj or F~ surf = − ~
P(~r) · dS.
S j S

In this vector form we imply that the first index of P is matched to that
of F~ surf , while the second index is paired with that of dS ~ and summed
over. Gauss’s law tells us that this is the integral over the volume V of the
divergence, but we should take care that this divergence dots the derivative
with the second index, that is
Z X
∂
Fisurf = − Pij (~r)dV.
V j ∂xj

However, as P is symmetric, we can get away with writing

Z
F~ surf = − ~ · P(~r)dV.
∇
V

Writing Hooke’s law as

1 1
P = −βS + 1I(Tr P + β Tr S) = −βS − 1I(α − β) Tr S,
3 3
(where 1I is the identity matrix 1Iij = δij ), Newton’s second law gives
∂ 2 ~η (~r) ~ r) − ∇
~ · P(~r)
ρ(~r) = E(~
∂t2
= E(~ ~ · S(~r) + α − β ∇
~ r) + β ∇ ~ Tr S(~r)
3
where in the last term we note that the divergence contracted into the 1I
gives an ordinary gradient on the scalar function Tr S. As the strain tensor
is already given in terms of derivatives of ~η , we have
! !
X ∂ 1 ∂ηi ∂ηj 1 ∂ ~
~ · S(~r)]j =
[∇ + = ∇ · ~η + ∇2 ηj ,
i ∂xi 2 ∂xj ∂xi 2 ∂xj
~ · S(~r) = 1 ∇(
~ ∇~ · ~η ) + 1 ∇2 ~η . Also Tr S = P ~ · ~η , so we find
or ∇ 2 2 i ∂ηi /∂xi = ∇
the equations of motion
!
∂ 2 ~η (~r) ~ r) + α + β ∇(
~ ∇~ · ~η ) + β ∇2 ~η .
ρ(~r) 2
= E(~ (5.9)
∂t 3 6 2
150 CHAPTER 5. SMALL OSCILLATIONS

This equation is called the Navier equation. We can rewrite this in terms of
the shear modulus G and the bulk modulus B:
 
∂ 2 ~η (~r) ~ r) + B + G ∇(
~ ∇~ · ~η ) + G∇2 ~η .
ρ(~r) 2
= E(~
∂t 3

Fluids
In discussing the motion of pieces of a solid, we specified which piece of the
material was under consideration by its “original” or “reference” position ~r,
from which it might be displaced by a small amount ~η (~r). So ~r is actually a
label for a particular hunk of material. This is called the material descrip-
tion. It is not very useful for a fluid, however, as any element of the fluid
may flow arbitrarily far from some initial position. It is more appropriate to
consider ~r as a particular point of space, and ρ(~r, t) or ~v (~r, t) or T (~r, t) as
the density or velocity or temperature of whatever material happens to be
at point ~r at the time t. This is called the spatial description.
If we wish to examine how some physical property of the material is chang-
ing with time, however, the physical processes which cause change do so on a
particular hunk of material. For example, the concentration of a radioactive
substance in a hunk of fluid might change due to its decay rate or due to its
diffusion, understandable physical processes, while the concentration at the
point ~r may change just because new fluid is at the point in question. In
describing the physical processes, we will need to consider the rate of change
for a given hunk of fluid. Thus we need the stream derivative, which involves
the difference of the property (say c) at the new position ~r 0 = ~r + ~v ∆t at
time t + ∆t and that at the old ~r, t. Thus

dc c(~r + ~v ∆t, t + ∆t) − c(~r, t) ~ + ∂c .

(~r, t) = lim = ~v · ∇c
dt ∆t→0 ∆t ∂t
In particular, Newton’s law refers to the acceleration of a hunk of material,
so it is the stream derivative of the velocity which will be changed by the
forces acting on the fluid:
!
d~v ~ v (~r, t) + ∂~v (~r, t)
ρ(~r)∆V = ρ(~r)∆V ~v · ∇~ = F~ surf + F~ vol .
dt ∂t

The forces on a fluid are different from that in a solid. The volume force
is of the same nature, the most common being F~ vol = −ρgêz dV , and the
5.4. FIELD THEORY 151

pressure piece of the stress, Pp = +p1I is also the same. Thus we can expect
part of the force of the form F~ = (−ρgêz − ∇ ~ · 1Ip)dV = dV (−ρgêz − ∇p).
~
A static fluid can not experience a shear force. So there will be no shear
component of the stress due to a deviatoric part of the strain. But there can
be stress due to the velocity of the fluid. Of course a uniformly moving fluid
will not be stressed, but if the velocity varies from point to point, stress could
be produced. Considering first derivatives, the nine components of ∂vi /∂xj
have a scalar piece ∇~ · ~v , an antisymmetric piece, and a traceless symmetric
piece, each transforming differently under rotations. Thus for an isotropic
fluid the stress may have a piece
!
∂vi ∂vj ~ · ~v 1I
Pij = −µ + − ν∇
∂xj ∂xi
in addition to the scalar, pressure, piece p1I. The coefficient µ is called the
viscosity. The piece proportional to ∇ ~ · ~v may be hard to see relative to the
pressure term, and is not usually included7
~ v , is in fact just the fractional rate of
The scalar component of ∂vi /∂xj , ∇·~
change of the volume. To see that, consider the surface S which bounds the
material in question. If a small piece of that surface is moving with velocity
~v , it is adding volume to the material at a rate ~v · dS,~ so
I Z
dV ~ ~ · ~v dV.
= ~v · dS = ∇
dt S V

As the mass of the material in question is constant, d(ρV )/dt = 0, so

dρ ~ · ~v = 0.
+ ρ∇
dt
This is known as the equation of continuity.
With !
∂vi ∂vj X ∂vj
Pij = p1I − µ + − ν1I
∂xj ∂xi j ∂xj

the surface force is

I !
XI ∂vi ∂vj X I ∂vj
F~isurf = −pdSi + µ + dSj + ν dSi
S j S ∂xj ∂xi j S ∂xj

7
Tietjens ([19]), following Stokes, assumes the trace of P is independent of the “velocity
of dilatation” ∇~ · ~v , which requires ν = −2µ/3. But Prandtl and Tietjens [12] drop the
~ ~
∇(∇ · ~v ) term in (5.10) entirely, equivalent to taking ν = −µ.
152 CHAPTER 5. SMALL OSCILLATIONS
 
Z X ∂ 2 vi X ∂ 2 vj
−
∂p  dV
= +µ 2
+ (µ + ν)
V ∂xi j ∂x j j ∂x i x j

where the last equality is by Gauss’ law. This can be rewritten in vector
form: Z  
~
F surf = ~ + µ∇2~v + (µ + ν)∇(
−∇p ~ ∇
~ · ~v dV
V

Adding in F~ vol = −ρgêz dV and setting this equal to ρ dV d~v /dt, we find
d~v ∂~v (~r, t) ~ v (~r, t)
= + ~v · ∇~ (5.10)
dt ∂t
1~ µ µ + ν ~ ~ 
= −gêz − ∇p(~ r, t) + ∇2~v (~r, t) + ∇ ∇ · ~v (~r, t) .
ρ ρ ρ
This is the Navier-Stokes equation for a viscous fluid. For an inviscid fluid,
one with a negligible viscosity (µ = ν = 0), this reduces to the simpler Euler’s
equation
∂~v (~r, t) ~ v (~r, t) = −gêz − 1 ∇p(~
~ r, t).
+ ~v · ∇~ (5.11)
∂t ρ
If we assume the fluid is inviscid and incompressible, so ρ is constant,
and also make the further simplifying assumption that we are looking at a
steady-state flow, for which ~v and p at a fixed point do not change, the partial
derivatives ∂/∂t vanish, and ∇ ~ · ~v = 0. Then Euler’s equation becomes

~ v (~r) = −gêz − 1 ∇p(~

~v · ∇~ ~ r). (5.12)
ρ
In a steady state situation, any function f (~r) has a stream derivative
d
f = ~v · ∇f,
dt
~
so the first term in (5.12) is d~v /dt, and the second term is −∇(gz). Dotting
the equation in this form into ρ~v , we have
 
d~v ~ = 0 = d 1 ρv 2 + ρgz + p
ρ~v · + ρ~v · ∇(gz) + ~v · ∇p
dt dt 2
which implies Bernoulli’s equation:
1 2
ρv + ρgz + p = constant along a streamline
2
where the restriction is because a streamline is the set of points in the flow
which are traversed by an element of the fluid as time goes by.
5.4. FIELD THEORY 153

Exercises
5.1 Three springs connect two masses to each other and to immobile walls, as
shown. Find the normal modes and frequencies of oscillation, assuming the system
remains along the line shown.

2
5.2 Consider the motion, in a fixed vertical plane, of a double pendulum consist-
ing
of two masses attached to each other and to a fixed
point by inextensible strings of length L. The upper
mass has mass m1 and the lower mass m2 . This is all in
a laboratory with the ordinary gravitational forces near
the surface of the Earth.

a) Set up the Lagrangian for the motion, assuming the

strings stay taut.
b) Simplify the system under the approximation that the
motion involves only small deviations from equilibrium.
1
Put the problem in matrix form appropriate for the pro-
cedure discussed in (5.1).
c) Find the frequencies of the normal modes of oscilla-
tion. [Hint: following exactly the steps given in class will
be complex, but the analogous procedure reversing the
order of U and T will work easily.] 2

5.3 (a) Show that if three mutually gravitating point masses are at the vertices
of an equilateral triangle which is rotating about an axis normal to the plane of
the triangle and through the center of mass, at a suitable angular velocity ω, this
motion satisfies the equations of motion. Thus this configuration is an equilibrium
in the rotating coordinate system. Do not assume the masses are equal.
(b) Suppose that two stars of masses M1 and M2 are rotating in circular orbits
about their common center of mass. Consider a small mass m which is approx-
imately in the equilibrium position described above (which is known as the L5
point). The mass is small enough that you can ignore its effect on the two stars.
154 CHAPTER 5. SMALL OSCILLATIONS

Analyze the motion, considering specifically the stability of the equilibrium point
as a function of the ratio of the masses of the stars.

5.4 In considering the limit of a loaded string we found that in the limit a →
0, n → ∞ with ` fixed, the modes with fixed integer p became a smooth excitation
y(x, t) with finite wavenumber k and frequency ω = ck.
Now consider the limit with q := n+1−p fixed as n → ∞. Calculate the expression
for yj in that limit. This will not have a smooth limit, but there is nonetheless a
sense in which it can be described by a finite wavelength. Explain what this is,
and give the expression for yj in terms of this wavelength.

5.5 Consider the Navier equation ignoring the volume force, and show that
a) a uniform elastic material can support longitudinal waves. At what speed do
they travel?
b) an uniform elastic material can support transverse waves. At what speed do
they travel?
c) Granite has a density of 2700 kg/m3 , a bulk modulus of 4 × 1010 N/m2 and a
shear modulus of 2.5 × 1010 N/m2 . If a short spike of transverse oscillations arrives
25 seconds after a similar burst of longitudinal oscillations, how far away was the
explosion that caused these waves?