Ph106a Class Notes

Lim Soon Wei Daniel

1
Chapter 1

Week 1

1.1 Tuesday, 29 Sept 2015

1.1.1 Motivation and overview
Example 1 Consider a particle constrained to move along a curved surface in 3D. There will be three equations (3 compo-
nents) along with an additional constraining equation that the particle is along the surface. This will be simplified to
two equations (2 degrees of freedom). In analytical mechanics, it is not necessary to deal with the constraints directly.
Example 2 Rigid body motion: Constraints are such that the distance between particles does not change. Body has only
3 degrees of freedom.
Variational calculus Consider the curved surface problem. First consider the 4 dimensional space that describes the
motion of the particle (u, v, u̇, v̇). Newton’s laws will give the trajectory of the particle in the configuration space.
Variational mechanics allows us to pick out the physical path from the other possible other paths - pick based on taking
the extremum (usually minimum) of the action:

Z
A= L(u, v, u̇, v̇)dt (1.1)

For a curved surface, the physical path connecting any two points (under no other forces other than the constraint)
will be the geodesic of the surface.
Example 3 Consider coupled harmonic oscillators on a ring. The strategy is to convert the N coupled equations in N
decoupled normal modes.

Example 4 Consider central force (orbital) motion. We reduce the 3D equations into 2D ones by arguing that motion occurs
along a plane, the decouple the two dimensions into radial and angular components.
Example 5 But some systems cannot be separated. Consider a central body that is almost spherical (spherical with
bumps). Then the equation of motions of particles around this body will be the ideal equation with perturbations (such
as precession).
Chaotic systems Cannot be separated. Perturbations in initial conditions result in an exponential increase in phase space
occupancy so that in a finite amount of time the motion is difficult to predict.

1.1.2 Review of Newtonian Mechanics

Notation :

• Position: ~r = (x, y, z) ∼ xî + y ĵ + z k̂.

• Displacement: ∆~r = ~r2 − ~r1
• Velocity ~r˙ = ~v
• Acceleration ~v˙ = ~a.

2
Newton’s Laws (1st) In an inertial frame, an object will not experience an acceleration (will move with constant velocity)
in the absence of a net external force. (2nd) Relates force to acceleration. (3rd) f~12 = −f~21 . Leads to momentum
conservation:

p~ ≡ m~v (1.2)
d~
p
F~ = (1.3)
 dt 
X d~
ptot X XX 1 XX ~ XX 1 XX ~
p~tot = p~j =⇒ = F~j = f~ij =  fij + f~ij  = (fij + f~ji ) (1.4)
j
dt j j
2 j i
2 j
i6=j i6=j j6=i i6=j

where f~ij is the force from the ith particle to the jth particle. Now implement Newton’s 3rd law to ensure that the
last statement (1.4) becomes zero.
Problems with N3L Consider two charged particles moving orthogonal to each other. The magnetic forces do not appear
to satisfy N3L. Recall that the magnetic field is controlled by ~r × ~v . Resolve this by considering that the forces act
between the object and the field, not directly between objects. We need to introduce the field momentum.
Angular momentum Define:

~ ≡ ~r × p~
L (1.5)

The rate of change is given by:

~˙ = ~r˙ × p~ + ~r × p~˙ = ~r × F~
L (1.6)

and the first term vanishes because of the definition of momentum (momentum parallel to velocity).
Central force field is such that:

~˙ = 0
~r × F~ = 0 =⇒ L (1.7)

because the force is always oriented towards the origin, which is parallel to the position vector. Hence the central force
field conserves angular momentum. We can always draw the plane in configuration space that is orthogonal to the
conserved angular momentum vector. Since ~r × ~v = L ~ and L~ is a constant, we have that ~r ⊥ L
~ and ~v ⊥ L.
~ Hence the
~
~r, ~v vectors lie in the plane orthogonal to the L vector, and we have reduced the 3D problem into a 2D problem with
the constraint.
Origin displacement Angular momentum is defined with respect to a single point. It is possible to show that the definition
of torque in terms of the rate of change of angular momentum is independent of origin position.
Energy and work Define:

Z 2
W12 ≡ F~ · d~s (1.8)
1

for a particle moving from position 1 to position 2.

Z 2
W12 = m~v˙ · ~v dt (1.9)
1
1 2
= m [~v · ~v ]1 F.T.C (1.10)
2
1 1
= mv22 − mv12 (1.11)
2 2

and we define the kinetic energy T = 12 mv 2 accordingly.

3
Conservative forces are such that the work done between any two points does not depend on the path taken. Equivalently,

I
F~ · d~s = 0 (1.12)

Potential Since the conservative force work done is independent of path, we can always pick an arbitrary point to be the
origin and then integrate the work done along that path to define the potential:

Z ~
r
V (~r) ≡ − F~ · d~s (1.13)
O
W12 = W1→O + WO→2 = V (~r1 ) − V (~r2 ) (1.14)

This can be combined with the expression for kinetic energy to give:

W12 = T2 − T1 =⇒ T1 + V1 = T2 + V2 (1.15)

Force and Potential Energy Given a potential energy that is a single-valued function of position, we have:

F~ = −∇V (1.16)

This can be shown by considering the small displacement (x, y, z) → (x + ∆x, y, z):

Z (x+∆x,y,z)
∆V = − dwFx (x + w, y, z) (1.17)
(x,y,z)

where we parametrize the path using the variable w. Taking the limit as ∆x → 0, we obtain:

∂V
= −Fx (1.18)
∂x

Repeat for other directions.

Geometric interpretation The force points in the direction opposite to that of greatest increase in potential.

1.2 Thursday, 1 Oct 2015

1.2.1 Newtonian Gravity
Gravitational Potential

∇2 φ = 4πGρ (1.19)
Gρ(r~0 ) 3 ~0
Z
φ(~r) = d r (1.20)
|~r − r~0 |
V (~r) = mφ(~r) (1.21)
F~ = −m∇φ (1.22)
3
ρ(~r) = M δ (~r − ~r0 ) (1.23)

Note that the mass in equation (1.21) is called the passive gravitational mass in that it passively experiences the grav-
itational field. On the contrary, the mass in equation (1.23) is called the active gravitational mass because it generates
the gravitational field.

4
Near Earth’s Surface take the acceleration to be a constant so that:

φ = −~g · ~r + const (1.24)

Clearly, if ~g is a constant,

F~ = −∇φ = ∇(~g · ~r) = ∇(gx X + gy Y + gz Z) = (gx , gy , gz ) = ~g (1.25)

1.2.2 Example of constrained motion

Description Consider an object with mass m on an inclined surface of angle α.
Method I: Standard solution Resolve forces. Consider constraint explicitly: ẏ = −ẋ tan α.
Method II: Clever choice of coordinates We note that the system has only 1 degree of freedom. Hence we choose a
coordinate to represent motion along that degree of freedom. Resolve gravity.

Method III: Energy balance Claim that energy is conserved:

m 2
(ẋ + ẏ 2 ) + mgy = E0 (1.26)
2

Implement the constraint ẏ = −ẋ tan α. Combine to obtain ODE:

m
ẏ 2 + mgy = E0 (1.27)
2 sin2 α

Rearranging:

s
2(E0 − mgy) sin2 α
ẏ = ± (1.28)
m

Note that the plus minus sign depends on the sign of the initial velocity ẏ0 . Hence we can write:

s
2(E0 − mgy) sin2 α
ẏ = sgn(ẏ0 ) (1.29)
m

Integrating the ODE directly,

Z y(t) Z t
dy
q = sgn(ẏ0 )dt0 (1.30)
y0 2 0
2(E0 /m − gy) sin α
p p
2(E0 /m − gy) = 2(E0 /m − gy0 ) − sgn(ẏ0 )g sin αt (1.31)

To remove the sign function, square both sides and note that the cross term sgn(ẏ0 )|ẏ0 | = ẏ0 . The second term is
actually:

p
2(E0 /m − gy0 ) = |ẏ0 | (1.32)

5
General form Note that the 1D COE equation can be written in the form:

1 2
ẏ + V (y) = E (1.33)
2

where we normalize the mass. The ODE to solve is hence:

p
ẏ = ± 2(E − V (y)) (1.34)

This is considered solved. But there is a subtle issue when V (y) = E for some values of y. If the potential energy
crosses the total energy with nonzero slope in y (so that you reach the intersection in a finite amount of time), we can
always linearize around the intersection point:

E0 − V = E0 − V (y ∗ ) − V 0 (y ∗ )(y − y ∗ ) + O(y − y ∗ )2 = V 0 (y ∗ )(y ∗ − y) + O(y − y ∗ )2 (1.35)

Then we can use analytic integrals to evaluate the motion around that point, and we will obtain something like
y ∼ y ∗ − 21 V 0 (y ∗ )t2 .
Potential energy touching total energy tangentially In such a case, the time taken for the particle to reach the inter-
section point is infinite. In this case, the first order Taylor expansion is insufficient because it vanishes at the point as
well. Hence we need the second order:

1
E0 − V = − V 00 (y ∗ )(y − y ∗ )2 + O(y − y ∗ )3 (1.36)
2

Now we have a quadratic term in the square root of the denominator. This integral is not convergent because it
integrates to a logarithm. It takes an infinite amount of time for the particle to approach the turning point.

1.2.3 System of particles

Total momentum
X
P~ = p~j (1.37)
j

Under N3L, F~ij = −F~ji , then P~ = F~ ext . By definition:

  P !
X d~rj d X d j mj ~rj
P~ = mj = mj ~rj  = M P (1.38)
j
dt dt j
dt j mj

where M is the total mass

P ~ the mass weighted average of the positions:
mj . Define the center of mass R,
j

P !
j mj ~rj
~ =
R P (1.39)
j mj
dR~
=⇒ P~ = M (1.40)
dt
¨
~ ~
=⇒ M R = Rext (1.41)

Center of mass frame has an offset of R ~¨ (so it is generally not inertial). Converting
~ in position and has an acceleration R
to the CM frame coordinate r~0 j :

6
 ~ + r~0 j
~rj = R (1.42)

Also, by definition of the center of mass, the CM of the system is at the origin:

X X
~ =⇒
mj ~rj = M R mj r~0 j = 0 (1.43)
j j

Angular momentum Recall that:

X
~ =
L ~r × mj ~vj (1.44)
j

In the CM frame:

X
~ =
L (R ~ + v~0 j )
~ + r~0 j ) × mj (V (1.45)
j
X X
= ~ × mj V
R ~ + ~ × mj v~0 j + r~0 j × mj V
R ~ + r~0 j × mj v~0 j (1.46)
j j
X X X
~ × P~ + R
= MR ~× mj v~0 j + ~ +
mj r~0 j × V mj r~0 j × v~0 j (1.47)
j j j
X
~ × P~ +
= MR mj r~0 j × v~0 j (1.48)
j
~ CM + L
=L ~ rotation about CM (1.49)

So that the rate of change is:

~
dL X X
= ~rj × p~˙j + ~r˙j × p~j (1.50)
dt j j
X
= ~rj × p~˙j (1.51)
j
 
X X
= ~rj × F~jext + F~ij  (1.52)
j i6=j
XX
~ ext +
=N (~rj × F~ij ) (1.53)
j i6=j

~ ext + 1 XX
=N (~rj − ~ri ) × F~ij (1.54)
2 j
i6=j

Hence for the internal torques to vanish, we require not only that the internal forces are equal and opposite, but also
~ ~ ext as expected. If the reaction forces do not lie on the same
that they lie along the same line. Then we have ddtL = N
line, we can expect that the forces are the result of a field with non-zero angular momentum.

Considering the CM frame,

~˙ = R
L ~˙ 0
~ × P~˙ + L (1.55)
~˙ 0
X  X
=⇒ ~rj × F~jext − R~ × F~jext = r~0 j × F~jext = L (1.56)
j j

and we recognize the term on the LHS to be the torque with respect to the CM.

7
Energy of a many-particle system Write the force equation as:

X
F~j = −∇j Vj − ∇j Vij (1.57)
i6=j

and assume that the potential has the form:

Vij = Vij (|~ri − ~rj |) (1.58)

then the energy is conserved:

X X 1 XX
Tj + Vj + Vij = const (1.59)
j j
2 j
i6=j

8
Chapter 2

Week 2

2.1 Monday, 6 Oct 2015

2.1.1 Lagrangian Mechanics
Example 1: Block on inclined plane There are 6 equations of motion (2 vector equations and 2 constraint equations).
Let N1 be the normal force along the inclined plane and N2 be the normal force on the ground. Then:

~ 1 = m~am
m~g + N (2.1)
~1 + N
m~g − N ~ 2 = M~aM (2.2)
~aM · ĵ = 0 (2.3)
(~am − ~aM ) · n̂ = 0 (2.4)

Note that these 6 equations completely define the 6 unknowns (4x coordinates and 2x normal force magnitudes).
Example 1 using parametrized variables Define X and d to be the position of the inclined plane and block along the
inclined plane respectively. Call the (X, d) space the configuration space. Defining:

~aM = Ẍ î (2.5)
~am ¨ î cos α + ĵ sin α)
= Ẍ î + d( (2.6)

Inserting these equations into the 6 equations removes 2 equations (constraints are automatically satisfied) and removes
2 unknowns (4 variables parametrized by 2 variables). But we can do better.
Virtual displacement Consider the configuration space. Consider a displacement of the configuration by a small amount
(δX, δd). We want to know how much work is induced by this displacement (computed in the Euclidean space):

X
δW = F~i · δ~ri (2.7)
i

where we sum over all particles.

Under a dissipationless system, the constraint forces will not do any work. Hence:

X
δW = F~iext · δ~ri (2.8)
i

Now we also know that:

X
δW = p~˙i · δ~ri (2.9)
i

9
 The summations will result in a linear combination of perturbed variables δX, δd. Hence equating the two summations,
we note that δX and δd are independent, so their coefficients must match on both sides of the equation.

We may further simplify the equation using the chain rule and some manipulation:

X ∂~ri
δ~ri = δxj (2.10)
j
∂xj

where we sum over the parametrized variables in the configuration space instead of the number of particles.
Note further that for holonomic constraints:

∂~ri ∂~r˙i ∂~vi

= = (2.11)
∂xj ∂ ẋj ∂ ẋj

Substituting this into the expression for the momentum summation:

X X X ∂~r˙i
p~˙i · δ~ri = mi~r˙i · δxj (2.12)
i i j
∂ ẋj

We can write this in terms of derivatives of the total kinetic energy. See Lecture 3 notes if absolutely necessary. The
end result is:

 
d ∂T δW
= (2.13)
dt ∂ Ẋ δX
 
d ∂T δW
= (2.14)
dt ∂ d˙ δd

Example 2: Bead on rotating rod Simplifies to:

 
d ∂T ∂T δW
− = (2.15)
dt ∂ q̇ ∂q δq

The kinetic energy is:

m
q̇ + (q sin αω)2


T = (2.16)
2

where α is the angle of the rod from the vertical. Substituting into the virtual work expression, we have:

LHS = mq̈ − mω 2 sin2 αq (2.17)

the work done by the external force (gravity) is −mg cos α hence the equation of motion is:

mq̈ − mω 2 sin2 αq = −mg cos α (2.18)

Move into the rod non-inertial frame of reference. Then there is the vertical force of gravity, the centrifugal force
outwards, m(q sin α)ω 2 , the Coriolis force 2m~v × ω
~ . The Coriolis force is zero because ~v is orthogonal to the angular
velocity vector (vertical). Resolving forces along the q direction, we will obtain the same equation of motion.

10
 Note that even though we required the virtual work to be zero, the actual work is non-zero because the bead will move
outwards and speed up (runaway process). The actual work is:

dW = N ~ · d~r (2.19)
∂~r ∂~r
d~r = dq + dt (2.20)
∂q ∂t

Note that the virtual work does not consider the time term because we fix the time. Only the first term is relevant for
virtual work. Hence:

   
~ ∂~r ~ ∂~r
dW = N · dq + N · dt (2.21)
∂q ∂t
the first term is zero but the second term is non-zero.

2.2 Thursday, 8 Oct 2015

2.2.1 Notation
Holonomic system Let there be generalized coordinates q1 , q2 , . . . , qN with N independent degrees of freedom. Let there
be M particles described by 3M degrees of freedom ~r1 , . . . , ~rM . There is a map from the N degrees of freedom to the
3M coordinates in Euclidean space. Write this map as:

~rj (q1 , . . . , qN , t) = ~rj (qk , t) (2.22)

where k is understood to run from 1, . . . , N .

We need 3M − N constraint equations:

fl (~r1 , . . . , ~rM , t) = 0, l = 1, 2, . . . , 3M − N (2.23)

Using the identity derived using the rotating rod example from last lecture, we write the equation for each generalized
coordinate as:

 
d ∂T ∂T δW
− = (2.24)
dt ∂ q̇k ∂qk δqk

Derivation Start by writing the velocity of the jth object:

X ∂~rj ∂~rj
~rj (qk , t) =⇒ ~vj (qk , q̇k , t) = q̇k + (2.25)
∂qk ∂t
k

For a holonomic system:

∂~vj ∂~rj
= , ∀j, k (2.26)
∂ q̇k ∂qk

Also observe that:

!
∂~vj ∂ X ∂~rj ∂~rj X ∂ 2~rj ∂ 2~rj

d ∂~rj

= q̇l + = q̇l + = (2.27)
∂qk ∂qk ∂ql ∂t ∂qk ∂ql ∂qk ∂t dt ∂qk
l l

11
by the interchanging of partial derivatives.

Also consider the kinetic energy:

X mj
T = ~vj2 (2.28)
j
2

We first use d’Alembert’s principle:

X
p~˙j · δ~rj = δW (2.29)
j

We examine the LHS:

X X
p~˙j · δ~rj = mj ~v˙ j · δ~rj (2.30)
j j
X X ∂~rj
= mj ~v˙ j · δqk (2.31)
j
∂qk
k
M X
N  
X
˙ ∂~rj
= mj ~vj · δqk (2.32)
j=1 k=1
∂qk
 
N M
X X ∂~rj 
=  mj ~v˙ j · δqk exchanging order of summation (2.33)
j=1
∂q k
k=1
 
N
X X d ∂~rj d ∂~rj 
=  M mj ~vj · − mj ~vj δqk (2.34)
j=1
dt ∂q k dt ∂qk
k=1
N  
X d ∂T ∂T
= − δqk (2.35)
dt ∂ q̇k ∂qk
k=1

Now examine the RHS:

X
δW = F~j · δ~r (2.36)
j
N
!
X X ∂~
rj
= F~j · δqk (2.37)
j
∂qk
k=1
 
N
X X ∂~
r j
=  F~j · δqk (2.38)
j
∂q k
k=1

Define the generalized force:

 
X ∂~rj 
Fk =  F~j · (2.39)
j
∂qk

so that the RHS becomes:

N
X
δW = Fk δqk (2.40)
k=1

12
 If the force is conservative, we have that:

F~j = −∇j V (~r1 , . . . , ~rM ) (2.41)

Write the potential as a function of the generalized coordinates and time (because each of the Euclidean coordinates
can depend on time):

V (q1 , . . . , qN , t) (2.42)

We can simplify the generalized force accordingly using the chain rule:

M M
∂V X ∂V ∂~rj X ∂~rj
− = − · = −∇j V · = Fk (2.43)
∂qk j=1
∂~rj ∂qk j=1
∂qk

Hence we have:

d ∂T ∂T ∂V
− = (2.44)
dt ∂ q̇k ∂qk ∂qk

If we consider that V does not depend on the velocities, then we have:

d ∂L ∂L
= (2.45)
dt ∂ q̇k ∂qk
L=T −V (2.46)

which is the Euler-Lagrange equation.

Define the generalized momentum:

∂L
pk = (2.47)
∂ q̇k

so that:

∂L
ṗk = (2.48)
∂qk

Time derivative of Lagrangian Proceed by definition:

N  
X ∂L ∂L ∂L
L̇ = q̇k + q̈k + (2.49)
∂qk ∂ q̇k ∂t
k=1
N  
X ∂L ∂L
= q̇k + q̈k kinetic and potential energy not explicitly dependent on time (2.50)
∂qk ∂ q̇k
k=1
N
X
= (ṗk q̇k + pk q̈k ) (2.51)
k=1
N
X d
= (pk q̇k ) (2.52)
dt
k=1

13
Hence:

!
d X ∂L
pk q̇k − L =− =0 (2.53)
dt ∂t
k

Hence the LHS term is conserved wrt time. Define this as the Hamiltonian:

!
X
H= pk q̇k − L (2.54)
k

Note that it can be shown that:

X
pk q̇k = 2T (2.55)
k

so that H = 2T − (T − V ) = T + V , the total energy in the conservative force case.

14
Chapter 3

Week 3

3.1 Tuesday 13 Oct 2015

Invariance implies conservation Recall that if ∂L
P
∂t = 0, then we can use the Hamiltonian H = k pk q̇k − L to show that
dH dL
dt = 0. Note the distinction between the total derivative and the partial derivative. Generally, dt 6= 0.

When L does not depend on time explicitly, qk (t − t0 ) is also a solution when qk (t) is a solution (because the Euler-
Lagrange equation will only depend on qk and q̇k ).

d ∂L
Transational invariance Suppose L has a cyclic/ignorable coordinate q0 . Then ṗ0 = dt ∂ q̇0 = 0 using the Euler-Lagrange
equations. This means that any solution displaced in q0 remains a solution. We can also change variables to displace
the coordinates in q0 and the Lagrangian form remains identical.
Lagrangian T-V must be defined in inertial frame Consider a freely falling particle. Define a new accelerating frame
2
ỹ = y − at2 . In the inertial frame, the Lagrangian was just:

mẏ 2
L(y, ẏ, t) = − mgy (3.1)
2

But in the non-inertial frame, we make the coordinate substitutions:

at2
y = ỹ + (3.2)
2
ẏ = ỹ˙ + at (3.3)

so that the Lagrangian is now dependent on time explicitly:

at2
 
˙ t) = m ˙
2
L(ỹ, ỹ, ỹ + at − mg ỹ + (3.4)
2 2

Note that the kinetic and potential energies now pick up time-dependent terms, and hence are not simply the kinetic
and potential energies in the accelerated frame. In order to write the Lagrangian as T − V , we must perform the
measurements in an inertial frame.
Non-holonomic constraints: Example 1 (Differential but non-integrable constraint) Coin rolling without slipping
on a plane. Let the position of the coin along the plane be (x, y) and let its velocity vector be pointing at the angle
θ. The coin has an additional degree of freedom φ, the angle its plane makes with the vertical. There are two degrees
of freedom: θ and φ which characterize the motion of the coin. The constraints are that the motion is rolling without
slipping: dx = Rdφ cos θ, dy = Rdφ sin θ. Note that θ, φ are not sufficient to describe the state of the system. You need
to know the entire variation of θ, φ in time to do this.
Non-holonomic constraints: Example 2 (Constraint conditions are inequalities) Ball falling off larger ball.

15
 3.1.1 Calculus of variations
Consider the functional:

Z L
I[y] = F (y(x), y 0 (x), x)dx (3.5)
0

Only consider functions whose endpoints are fixed. We want to extremize I[y] for functions with the fixed endpoints.
Example: Least-time problem Consider trajectory y(x) under the influence of gravity. We require that y(0) = 0 and
y(L) = a. The differential time is dt = ds
v =
√ds .
2gy

Condition for extremum

I(y + δy) = I(y) + O(δy)2 (3.6)

We can introduce a perturbation  so that:

I(y + δy) = I(y) + O(2 ) (3.7)

Hence we examine:

Z
I(y + δy) = F (y(x) + δy(x), y 0 (x) + δy 0 (x), x)dx (3.8)
Z L Z L
∂F ∂F
= I(y) +  δy(x)dx +  δy 0 (x)dx + O(2 ) (3.9)
0 ∂y x 0 ∂y 0 x
Z L  L Z L  
∂F ∂F d ∂F
= I(y) +  δy(x)dx +  0 δy(x) −  δy(x)dx + O(2 ) (3.10)
0 ∂y x ∂y 0 0 dx ∂y 0
Z L   
∂F d ∂F
= I(y) +  − δy(x)dx + O(2 ) (3.11)
0 ∂y dx ∂y 0

Define the variational derivative:

 
δF ∂F d ∂F
≡ − (3.12)
δy(x) ∂y dx ∂y 0

so that we can write:

Z L
δF
δI = I(y + δy) − I(y) = δydx + O(δy 2 ) (3.13)
0 δy(x)

δI must vanish for all δy so that we have the extremum. This gives us the Euler equation:

 
∂F d ∂F
− =0 (3.14)
∂y dx ∂y 0

Application to Lagrangian mechanics Consider the action:

Z L
S[q(t)] = L(q(t), q̇(t), t)dt (3.15)
0

We want to extremize the action, fixing the end points.

16
3.2 Thursday 15 Oct 2015
Curved Plane Let u, v parametrize the plane. Then the differential distance is:

ds2 = g11 (u, v)du2 + 2g12 (u, v)dudv + g22 (u, v)dv 2 (3.16)

We want to compute the function giving the extremum distance between two points A and B. The distance functional
is:

Z L p Z L p
I= g11 du2 + 2g12 dudv + g22 dv 2 = dv g11 (u0 )2 + 2g12 u0 + g22 (3.17)
0 0

p
Hence the function inside the integral is F (u0 , u, v) = g11 (u0 )2 + 2g12 u0 + g22 , and we can use Euler’s equation to
solve for the equation of the curve. These curves are called geodesics.

Note that is a unique geodesic for two points close by. But when the distance is increased, you may reach a conjugate
point where uniqueness fails and there are many geodesics (think poles of the sphere). Conjugate points only occur in
surfaces with positive curvature (spheres).
m 2
Example Consider the Lagrangian L = 2 ẏ −V (y). The variational term is (after integrating by parts and Taylor-expanding
the potential):

Z t Z t 2

0 m 2 00 (δy)
δS = (−V − mÿ)δydt + (δ ẏ) − V dt + O[(δy)3 ] (3.18)
0 0 2 2

The first order term is just N2L, so it vanishes. Note that for the action to be a minimum, we require that the second
00
p m This is always fulfilled by V > 0. Also, we can estimate short time evolution by writing T δ ẏ ∼ δy so
term be positive.
that if T < V 00 , then the second order coefficient is still positive and S is minimized for short times. See homework
for details.
Change of variables Consider the change of variables from qk space to Qk space: Qk (qk , t), Q̇k (qk , q̇k , t). To operate in
the Qk space, we just replace the terms in the Lagrangian with the inverse map:

L(qk (Qk , t), q̇k (Qk , Q̇k , t), t) (3.19)

and use the Euler-Lagrange equations for Qk accordingly. You will obtain the same path in the configuration space.
Non-holonomic constraints Consider the coin rolling on the surface example. The parameters x, y, θ, φ are not indepen-
dent but are related by the constraints on their velocities:

ẋ = Rφ̇ cos θ (3.20)

ẏ = Rφ̇ sin θ (3.21)

We cannot use the Euler-Lagrange equation directly.

Review: Constrained optimization Suppose we want to extremize F (x1 , . . . , xn ) subject to gk (x1 , . . . , xn ) = 0, k =
1, 2, . . . , l. We want to find the infinitesimal displacement δ~x such that:

∇F · δ~x = 0 first order condition for extremum (3.22)

∇~g1 · δ~x = 0 constraint equations (3.23)
..
. (3.24)
∇~gl · δ~x = 0 (3.25)

17
 This implies that ∇F lies in the space spanned by ∇gk :

l
X
∇F = −λk ∇gk (3.26)
k=1

We may hence perform unconstrained optimization on the function:

X
F+ λk gk (3.27)
k

and require:

X
∇(F + λk gk ) = 0 (3.28)
k
g1 = 0, . . . , gl = 0 (3.29)

Example: Shape of hanging wire Consider a wire hung between points A and B. We want to find the configuration such
that the potential energy is minimized:

Z a
V = ρgyds (3.30)
−a

subject to the constraint that the total length of the wire is fixed:

Z a
l= ds (3.31)
−a

We hence consider the function:

V + λl (3.32)

and optimize it using the Euler-Lagrange equation.

Example: Pendulum The constraint for a fixed length pendulum is:

g(x, y) = x2 + y 2 − l2 = 0 (3.33)

and the Lagrangian is the usual L = T (ẋ, ẏ) − V (y). Since we require that the constraint be fulfilled at all times, we
consider the Lagrange multiplier problem:

 Z t  Z
0 0 0 0
δ S+ λ(t )g(x(t ), y(t ))dt = 0 =⇒ δ (L + λg)dt = 0 (3.34)
0

from which we have the Euler-Lagrange equations:

∂L d ∂L ∂g
− +λ =0 (3.35)
∂x dt ∂ ẋ ∂x
and the same for y. Substituting, we obtain the system:

−mẍ + 2λx = 0 (3.36)

−mg − mÿ + 2λy = 0 (3.37)
2 2 2
x +y =l (3.38)

which is 3 equations for 3 unknowns. We can solve this by linearizing around the bottom of the pendulum swing.

18
General holonomic constraints on coordinates and not velocities. Write:

X
L0 = L + λj gj (3.39)
j

and the new Euler-Lagrange equations are:

∂L d ∂L X ∂gj
− =− λj (3.40)
∂qk dt ∂ q̇k j
∂qk

Non-holonomic constraints and Lagrangians Consider the rolling coin on a plane again. Suppose the plane is angled
at α with respect to the horizontal. Let x be horizontal and let y point along the plane towards the lower side of the
plane. The Lagrangian is:

3 1
L= mR2 (φ̇)2 + mR2 (θ̇)2 + mgy sin α (3.41)
4 8

where there are two forms of rotational motion with different moments of inertia for each. We include the constraints
using the Lagrange multiplier method, using a Lagrange multiplier that is only dependent on time. See Lecture 6 notes
for a full calculation.

19
Chapter 4

Week 4

4.1 Tuesday, 20 Oct 2015

Midterm Week 1-3 material. No Mathematica.
Review of previous weeks Recall that we noted that for the physical path, the action was stationary:

X δL
δS = 0 ⇐⇒ δqk = 0 (4.1)
δqk
k

Given further constraints on qk , we wrote these constraints as Gj (qk ) = 0 so that:

X ∂Gj
δqk = 0 (4.2)
∂qk
k

For some non-holonomic constraints, the constraints were written as:

(j)
X
fk δqk = 0 (4.3)
k

In the holonomic case, we defined the new Lagrangian as satisfying:

δL X ∂Gj
− λj =0 (4.4)
δqk j
∂qk

and for the non-holonomic special case, the equation was:

δL X (j)
− λj fk = 0 (4.5)
δqk j

4.1.1 Simple Harmonic Oscillator

Consider the simple pendulum:

1 2 2
L= ml θ̇ + mgl cos θ (4.6)
2
The equation of motion is:

ml2 θ̈ = −mgl sin θ (4.7)

20
Definition of equilibrium θ(t) = θ0 . Clearly, two such constant solutions are θ(t) = 0, θ(t) = π. We can perform a Taylor
expansion of V (θ) around θ0 :

¨ = ±mgl∆θ
ml2 ∆θ (4.8)

where the positive sign corresponds to θ0 = π and the negative corresponds to θ0 = 0. Clearly, the θ0 = 0 case gives
stable oscillations parametrized by sin ωt, cos ωt. The θ0 = π case gives unstable motion that goes as eλt , e−λt .
Hamiltonian picture Recall that H = T − V . Then:

1 2 2
H= ml θ̇ − mgl cos θ (4.9)
2

and since the Lagrangian does not depend explicitly on time, the Hamiltonian is conserved. Plot the total energy as a
function of θ to obtain the classically allowed region as a function of the initial energy.
General equilibrium Given a Lagrangian and Hamiltonian:

1 2
L= q̇ − V (q) (4.10)
2
1
H = q̇ 2 + V (q) (4.11)
2

then the equilibrium condition is given by:

∂V
=0 (4.12)
∂q

Stable and unstable equilibrium evaluated at the equilibrium point can be written as:

∂2V
> 0 =⇒ stable equilibrium (4.13)
∂q 2
∂2V
< 0 =⇒ unstable equilibrium (4.14)
∂q 2

Consider the Taylor expansion of the Lagrangian L(q, q̇). The most general second order expansion about (q, q̇) = (0, 0)
is:

L = A + Bq + C q̇ + Dq 2 + Eq q̇ + F q̇ 2 + . . . (4.15)

But we know that any time derivative can be added to the Lagrangian without changing the physical motion. Hence
d 2
the C q̇ and Eq q̇ = E2 dt q terms are not important. Then the Lagrangian looks like the sum of a kinetic energy (F q̇ 2 )
and potential energy A + Bq + Dq 2 . The Euler Lagrange equation gives the equation of motion:

2F q̈ = 2Dq + B (4.16)

For an equilibrium, we require B = 0 because we want ∂V ∂q = 0 at the equilibrium. Hence the stability of an equilibria
depends on the relative signs of F and D. If they are the same sign, the system is unstable about that point because
the solution is an exponential. If they have opposite sign, the system is stable about that point because the solutions
are sinusoids. We rewrite the equation of motion:

D
q̈ = q (4.17)
F

21
 To rescale the equation, define τ = βt. Then we have the equation:

d2 q D
= β2 q (4.18)
dτ 2 F
q
F
We hence want to choose β = D so that we can write:

d2 q
= ±q (4.19)
dτ 2

where the sign depends on the stability of the equilibrium point. The action can be written in non-dimensionalized
form as:

Z Z
S= Ldt = β Ldτ (4.20)

R
Now we know that the Lagrangian is invariant under scaling, hence the action can be written as S = Ldτ . We can
also neglect the A term in the Lagrangian because it does not affect the physical system. The action can hence be
written as:

 2 !
|D|
Z
dq
S= F + Dq 2 dτ (4.21)
|F | dτ

Rescaling this equation further,

Z  2 !
F dq D 2
S= + q dτ (4.22)
|F | dτ |D|

Hence the rescaled action is given by:

Z " 2 #
1 dq 2
S= ±q dτ (4.23)
2 dτ

where the plus represents the unstable equilibrium and the negative represents stable equilibrium.
Damped harmonic oscillator
q̇
q̈ + q = − (4.24)
Q

Free evolution of the DHO The characteristic equation is:

1
−α2 + iα+1=0 (4.25)
Q
r
i 1
=⇒ α = ± 1− (4.26)
2Q 4Q2

Root locus for DHO For large Q, roots are located at ±1 on the real axis. As Q decreases, the roots move towards the
positive imaginary axis and touches when Q = 12 . As Q decreases further, the roots move further apart on the imaginary
axis. One root goes to zero and the other goes to infinity.

22
Green’s function approach to forced damped harmonic oscillator First solve the impulse response of the oscillator.
That is, we want to find G such that:

Ġ(t, t0 )
G̈(t, t0 ) + + G(t, t0 ) = δ(t − t0 ) (4.27)
Q

Clearly, G(t, t0 ) = 0 for t < t0 . For later time, we write the solution as a linear combination of the two orthogonal
functions.

Note that G has to be continuous at t = t0 . Otherwise, Ġ contains a delta function, and G̈ has a derivative of the delta
function. This is not the case. Hence G must be continuous at t = t0 . We integrate the ODE about the point t = t0 :

t0 +
" #
Ġ(t, t0 )
Z
0 0
G̈(t, t ) + + G(t, t ) dt = 1 (4.28)
t0 − Q

This simplifies to:

t0 + t0 + Z t0 +
G
Ġ + + Gdt = 1 (4.29)
t0 − Q t0 − t0 −

Now G is continuous, hence the second and third term vanishes when  → 0. Hence we have the condition:

G(t0 − 0) = G(t0 + 0) (4.30)

0 0
Ġ(t + 0) = Ġ(t − 0) + 1 (4.31)

This provides the initial velocity of the system, which we can use to find G by plugging it into the normal solution. For
instance, in the underdamped solution:

sin ω 0 (t−t0 ) −(t−t0 )/2Q

(
0 ω0 e , t > t0
G(t, t ) = 0
(4.32)
0, t≤t

q
where ω 0 = 1− 1
4Q2 .

We want to use the Green’s function to construct the particular solution to the ODE. This can be shown to be:

Z ∞ Z t
qp (t) = G(t − t0 )F (t0 )dt0 = G(t − t0 )F (t0 )dt0 (4.33)
−∞ t0

Observe that this is the steady state solution to the problem because the homogeneous solution decays away in time.
If t − t0 >> 2Q, the transient is not important and we can extend the integral to minus infinity:

Z t
qp (t) = dt0 G(t − t0 )F (t0 ) (4.34)
−∞

Note that we have decomposed the solution into the complementary portion (solution to the homogeneous equation
that satisfies the initial conditions q0 , q̇0 ) and the particular solution (solution that takes into account the RHS and has
initial conditions that are zero).

23
4.2 Thursday 22 Oct 2015
Review Recall that with suitable scaling of time, we can always write the Lagrangian in the form:

1 2 q
L= q̇ − (4.35)
2 2

Fourier transform method Write:

Z ∞
dω −iωt
q(t) = e q̃(ω) (4.36)
−∞ 2π

and taking the Fourier transform of both sides of the ODE:

Z ∞   Z ∞
dω iω dω
−ω 2 + + 1 q̃(ω)eiωt = F̃ (ω)eiωt (4.37)
−∞ 2π Q −∞ 2π

so we can obtain the Fourier transform of the solution:

F̃ (ω)
q̃(ω) = (4.38)
−ω 2 + iω/Q + 1
=⇒ q̃(ω) = G̃(ω)F̃ (ω) (4.39)

because the Fourier transform of δ(t) is a constant.

Resonance Take ω = 1. Then the steady state solution can be written as:

Z t Z t Z t
q(t) = sin(t − t0 )F (t0 )dt = sin t cos t0 F (t0 )dt0 − cos t sin t0 F (t0 )dt0 (4.40)
0 0 0

Observe that if F (t0 ) contains a component that is exactly in phase with any of the sin t, cos t components, there will
be an integrated component that increases in time, indicating resonance. In steady state, we write:

F = F0 eiωt (4.41)
iωt
q = q0 e (4.42)

which allows us to obtain q0 by substitution into the ODE:

F0
q0 = (4.43)
−ω 2 + iω/Q + 1

The denominator does not vanish for finite Q. First consider Q >> 1. The total energy goes as |q0 |2 . This can be
approximated near resonance ω ≈ 1:

|F0 |2 |F0 |2
|q0 |2 = ≈ (4.44)
(ω 2 2 2
− 1) + ω /Q 2 4(ω − 1)2 + 1/Q2

1
This curve shape is called a Lorentzian. The half-maximum occurs when ω = 1 ± 2Q so that the full frequency width
1
at half-maximum is Q . With units:

∆ωF W HM 1
= (4.45)
ω Q

24
 The phase of q0 can be written as:

 
1
φ = arg = arg(1 − ω 2 − iω/Q) (4.46)
−ω 2 + iω/Q + 1
Observe that for small ω, the argument is zero. As ω increases, the argument decreases below zero, reaches − π2 upon
resonance, then asymptotes to −π as ω → ∞. Call −φ the phase lag, because q(t) goes as cos(t + φ).
Theory of small vibrations Consider the two coupled pendulums. The Lagrangian can be parametrized by the two angles:

kl2
   
1 2 2 2 mgl 2 2 2
L= ml (θ̇1 + θ̇2 ) − (θ1 + θ2 ) + (θ2 − θ1 ) (4.47)
2 2 2
The EL equations gives:

g k
θ̈1 + θ1 + (θ1 − θ2 ) = 0 (4.48)
l 4m
g k
θ̈2 + θ2 + (θ2 − θ1 ) = 0 (4.49)
l 4m
Non-dimensionalizing the equation, we define:

r
g
ω0 = (4.50)
l
kl
η= (4.51)
4mg
so we have the equations:

θ̈1 + ω02 θ1 + ω02 η(θ1 − θ2 ) = 0 (4.52)

θ̈2 + ω02 θ2 + ω02 η(θ2 − θ1 ) = 0 (4.53)

We hence have the matrix equation:

d2
    
θ1 1+η −η θ1
+ ω02 =0 (4.54)
dt2 θ2 −η 1+η θ2
We hence want to obtain the normal modes that satisfy:

   
θ1 A1
= eiωt (4.55)
θ2 A2
which oscillates with a single frequency. Substitution into the matrix differential equation gives:

   
2 A1 A1
−ω I + ω02 M =0 (4.56)
A2 A2

which indicates that ω 2 is an eigenvalue of ω02 M and the A vector is an eigenvector of ω02 M . Define ω 2 = λω02 , which
gives the eigenvalues:

λ1 = 1 λ2 = 1 + 2η (4.57)

and the eigenvectors are:

 
1
v1 = (4.58)
1
 
1
v2 = (4.59)
−1

25
4.3 Lagrangian of stretched string
Consider a small segment of string.

1
dT = λdxẏ 2 (4.60)
s 2 
 2
dy 1
dV ≈ F (ds − dx) = F  1 + − 1 dx ≈ F y 02 dx (4.61)
dx 2

So the action is given by:

Z t Z xb 
1 1
S= dx( λẏ 2 − F y 02 ) (4.62)
0 xa 2 2

Note that the Lagrangian is given by y, ẏ, y 0 , t, which is different from the usual Lagrangian. The same action minimizing
principle in δy applies. Consider the Lagrangian density:

Z  
∂L ∂L ∂L
δS = dtdx δy + 0 δy 0 + dẏ (4.63)
∂y ∂y ∂ ẏ

We integrate the second term by parts and eliminate the boundary term. Also integrate the third term by parts and
cancel the boundary term. The final equation is:

   
∂L ∂ ∂L ∂ ∂L
− − =0 (4.64)
∂y ∂x ∂y 0 ∂t ∂ ẏ

Applying this to the string, we obtain the equation of motion:

∂2y ∂2y
F = λ (4.65)
∂x2 ∂t2

26
Chapter 5

Week 5

5.1 Tuesday 27 Oct 2015

Theory of small vibrations - General Write the equation of motion in terms of a matrix DE:

d2
x + Mx = 0 (5.1)
dt2
where M = M T is symmetric. The eigenvalues are the resonant frequencies ω and the eigenvectors are the normal
modes A. The general solution is:

 
XN
x(t) = <  Cj Aj eiωj t  (5.2)
j=1

where Cj ∈ C.
Kinetic and Potential energies Let the system have kinetic and potential energies of the form:

X
T = φ̇j Tj,k φ̇k (5.3)
j,k
X
V = φj Vjk φk (5.4)
j,k

which can be obtained using Taylor expansion and keeping the second order terms alone. The matrix terms are:

1 ∂2T
Tj,k = (5.5)
2 ∂ φ̇j ∂ φ̇k
1 ∂2V
Vj,k = (5.6)
2 ∂φj ∂φk
Note that we can write these energies in the matrix quadratic form:

 
φ1
 φ2 
φ= (5.7)
 
.. 
 . 
φN
T
φ = (φ1 , . . . , φN ) (5.8)
T = {Tj,k }N
j,k=1 =T T
symmetric (5.9)
V = {Vj,k }N
j,k=1 =V T
symmetric (5.10)
T
T = φ̇ T φ̇ (5.11)
T
V =φ Vφ (5.12)

27
Euler-Lagrange equations in matrix form for Small Vibrations Let l = 1, 2, . . . , N .

 
d ∂L d ∂  X
= Tj,k φ̇j φ̇k  (5.13)
dt ∂ φ̇l dt ∂ φ̇l
j,k

d X  
= Tj,k δkl φ̇j + φ̇k δjl (5.14)
dt
j,k
 
d  X X
= Tlj φ̇j + Tkl φ̇k  (5.15)
dt j
k
 
d  X X
= Tlj φ̇j + Tlk φ̇k  , T is symmetric (5.16)
dt j
k
X
=2 Tlj φ̈j (5.17)
j
d ∂L
=⇒ = 2T φ̈ (5.18)
dt ∂ φ̇

Proceed similarly for the potential to obtain:

∂L ∂V X
=− = −2 Vlj φj (5.19)
∂φl ∂φl j
∂L
=⇒ = −2V φ (5.20)
∂φ

Implementing the Euler-Lagrange equation:

T φ̈ + V φ = 0 (5.21)

where T , V are N × N symmetric, positive definite, real-valued matrices:

T = TT (5.22)
T
V =V (5.23)
T T
φ T φ ≥ 0, φ Vφ≥0 (5.24)

where equality holds only when φ = 0. Note that while the kinetic energy is always positive anyway, the potential
positive definite condition requires that the potential increase in all directions away from the equilibrium point. This
is valid for small oscillations.
Solution to E-L equations using Normal Mode Approach Look for solutions in the form:

φ = Aeiωt (5.25)
2
=⇒ −ω T A + V A = 0 (5.26)
2
=⇒ (−ω T + V )A = 0 (5.27)

Results from Linear Algebra Linear Algebra says that two positive definite matrices can be simultaneously diagonalized.
That is, there exists a set of vectors {Φj }, j = 1, 2, . . . , N such that:

ΦTj T Φk = δjk (5.28)

ΦTj V Φk = vj δjk (5.29)

28
 where vj has units of potential energy divided by kinetic energy, or inverse time squared. The vectors have units of
inverse square roots of energy. Note that ΦTj T Φk can be defined as the inner product of (ΦTj , Φk ) such that the vectors
are orthonormal. Note that the Φ are chosen so that the “eigenvalues” under the transformation are all unity and
dimensionless. Substituting these normal modes into the E-L matrix equation,

ΦTj (−ω 2 T + V )Φk = 0 ∀j, k (5.30)

2
=⇒ (−ω + vj )δjk = 0 ∀j (5.31)
=⇒ ωk2 = vk (5.32)

Note further that without degeneracy, rescaled normal modes associated with different frequencies are orthogonal using
the definition of the inner product as similar to that above:

(Φj , Φk ) ≡ ΦTj T Φk = δkj (5.33)

and we can obtain the frequencies using the rescaled normal modes:

ΦTk (−ωk2 T + V )Φk = −ωk2 + ΦTk V Φk (5.34)

=⇒ ωk2 = ΦTk V Φk (5.35)

Example: Triatomic molecule Let two springs of constant k connect a central molecule M to two side molecules of mass
m. Let the displacements of the molecules be x1 , x2 , x3 from one side to other respectively. Then we write the energies:

m 2 M 2
T = (ẋ1 + ẋ23 ) + ẋ (5.36)
2 2 2
k
(x3 − x2 )2 + (x2 − x1 )2

V = (5.37)
2

Writing down the matrix elements explicitly:


1 0 0
m
T = 0 r 0  (5.38)
2
0 0 1
 
1 −1 0
k
V =  −1 −2 −1  (5.39)
2
0 −1 1
M
r≡ (5.40)
m

We require:

(−ω 2 T + V )Φ = 0 (5.41)

mω 2
Non-dimensionalizing, we define λ = k and look for the values of λ such that (−λT + V ) has zero determinant. This
gives λ = 0, 1, 1 + 2r .

5.2 Thursday 29 Oct 2015

5.2.1 Review of Linear Algebra
Congruent transformation Start with a symmetric matrix A and define the congruent transformation:

29
 A → P T AP (5.42)
A can always be transformed into a matrix with +1s or -1s or 0s along the diagonal. A positive definite matrix A can be
transformed into the identity matrix (all 1s along the diagonal). The eigenvalues of A are not conserved under a congruent
transformation.
Small oscillation Lagrangian Recall that we could write the Lagrangian in the quadratic form as (tilde means trans-
pose):

˙
L = φ̃tφ̇ + φ̃vφ (5.43)
where φ is a column vector of generalized coordinates.
The Euler Lagrange equation requires that:

tφ̈ + vφ = 0 (5.44)
which has solutions that look like:

φ = Φeiωt (5.45)
and the matrix equation for the normal modes Φ is:

(−ω 2 t + v)Φ = 0 (5.46)

and for a non-trivial Φ, we required that:

det(−ω 2 t + v) = 0 (5.47)
which is an N th order polynomial in ω 2 . The solutions (indexed by the associated eigenvalues ωk2 ) can be scaled to satisfy
the orthonormal relations:

Φ̃k tΦj = δjk (5.48)

Φ̃k vΦj = vk δjk (5.49)
where we define the inner product:

(Φk , Φj ) = Φ̃k tΦj (5.50)

and the rescaling condition:

Φk
Φk → p (5.51)
(Φk , Φk )

Initial value condition Suppose we know the initial positions and velocities φ0 , φ̇0 . Then the time evolution of the
initial condition can be written in terms of a complex superposition of normal modes:

" N
#
X
iωk t
φ(t) = < A k Φk e (5.52)
k=1

where Ak ∈ C. At t = 0:

" N
#
X X
φ(0) = < Ak Φ k = (<Ak )Φk (5.53)
k=1 k
X
φ̇(0) = − ωk (=Ak )Φk (5.54)
k

30
 where we note that the normal modes are real-valued. This gives a system of 2N linear relations so that we can obtain the
real and imaginary parts of Ak to obtain the time evolution of the oscillator. We use the distributivity of the inner product
and orthogonality condition to obtain the real and imaginary parts of Ak :

!
X X X
(Φj , φ0 ) = Φj , (<Ak )Φk = (<Ak )(Φj , Φk ) = (<Ak )δjk = <Aj (5.55)
k k k

(Φj , φ̇0 ) = −ωj (=Aj ) (5.56)

New generalized coordinates: Normal Coordinates Note that we can write any oscillation as a linear combination
of normal modes:

X
φ= zj Φj ⇐⇒ zj = (Φj , φ) (5.57)
j
X
φ̇ = żj Φj ⇐⇒ żj = (Φj , φ̇) (5.58)
j

so zj and żj can be seen as a set of new generalized coordinates and velocities. Call these Normal Coordinates. But
this means that:

  !
X X X X
T = φ̃tφ =  żj Φ̃j  t żk Φk = żj żk δjk = żj2 (5.59)
j k j,k j

and similarly,

X
V = vj zj2 (5.60)
j

Hence in this generalized coordinate space, the Lagrangian is simple:

X
żj2 − vj zj2


L= (5.61)
j

and the Euler-Lagrange equations require that:

z̈j = −vj zj (5.62)

Natural Unit System The systematic approach is to find quantities that have different dimensions and are independent
(cannot achieve a quantity that is dimensionless by taking products of the units).

Coupled Oscillators Consider N masses on a string clamped down between fixed ends. Let the tension in the string
be τ and let the separation between beads be d. The kinetic and potential energies are calculated accordingly in the small
amplitude approximation:

mX 2
T = ẏ (5.63)
2 j j
N N
X hp i τ X
V = τ d2 + (yk+1 − yk )2 − d ≈ (yk+1 − yk )2 (5.64)
2d
k=0 k=0

where we define y0 = yN +1 = 0 as the zero displacement clamps. In matrix form, the t matrix is diagonal, but the v
matrix is not. In fact, the v matrix has 2s along the diagonal and -1s along the super and sub diagonal (multiplied by a
scaling constant).

We change the units by setting m = τ = d = 1. Applying the Euler-Lagrange equation:

31
 ÿk = yk−1 − 2yk + yk+1 , k = 1, 2, . . . , N (5.65)

We can make a guess as to the form of the mode. We make a standing wave discrete Fourier transform ansatz:

N
X πnk −iωn t
yk = an sin e (5.66)
n=1
N +1

so that when k = 0 or k = N + 1, then yk vanishes. We want to ensure that each of the basis terms satisfies the Euler
Lagrange equation. Write a single term as:

πnk πn
yk = e−iωn t sin = e−iωn t sin γn k, γn ≡ (5.67)
N +1 N +1
Substituting into the E-L equations, we want:

−ωn2 eiωn t sin γn k = 2 sin(γn k)(cos γn − 2)eiωn t (5.68)

−ωn2
= 2(cos γn − 1) (5.69)
γn nπ
=⇒ ωn = ±2 sin = ±2 sin (5.70)
2 2(N + 1)

Plotting ωn against n, we get a discrete function that increases like a sine from n = 1 to n = N , and peaks at a maximum
value of 2. The phase and group velocities are as follows:

ω
vp = (5.71)
γ
dω
vg = (5.72)
dγ
Massive object on weighted string Substitute a massive object of mass M and connected to the ground with spring
constant k instead of one of the walls of the coupled masses on the string. Let the displacement of the big mass be z. Let
the other end of the string remain connected to a stationary wall. Then the equation of motion of the big mass is:

τ
M z̈ + kz = (y1 − z) (5.73)
d
while the other masses still obey the usual equation of motion. Note that in the limit as d → 0:

y1 − z ∂y
→ (5.74)
d ∂x
Hence we want to solve the wave equation for the other masses with the boundary condition:

y(x = 0) = z(t) (5.75)

∂y
M z̈ + kz = τ (5.76)
∂z
2 2
1 ∂ y ∂ y
= (5.77)
v 2 ∂t2 ∂x2
We want the solution as a combination of waves moving left and right:

y = yL + yR = yL (x + vt) + yR (x − vt) (5.78)

We hence have the two differential equations (to check how to obtain):

32
 1 ∂yL ∂yL
− =0 (5.79)
v ∂t ∂x
1 ∂yR ∂yR
+ =0 (5.80)
v ∂t ∂x
Now we are only interested in wave propagating to the right (do not want to drive the mass with left-moving waves).
Then we have that (noting that the string is connected to the mass and hence must have the same velocity at x = 0):

∂y 1 ∂y 1 dz
=− =− (5.81)
∂x v ∂t v dt
τ dz
=⇒ M z̈ + kz = − (5.82)
v dt
τ
=⇒ M z̈ + ż + kz = 0 (5.83)
v
and we observe that the big mass is damped by the string. This is one way to model friction and the flow of energy away
from the mass.

33
Chapter 6

Week 6

6.1 Tuesday, 3 Nov 2015

6.1.1 1D Motion
1D Lagrangian We can always rescale the coordinates to write the Lagrangian as:

q̇ 2
L= − V (q) (6.1)
2
and the Hamiltonian will be:

q̇ 2
E= + V (q) (6.2)
2
Example: 1D Pendulum with unit length and unit mass.

θ̇2
E= + (1 − cos θ) = E (6.3)
2

Phase space Plot θ̇ against θ. The trajectory must have constant E. Hence the phase space trajectories are level sets
2 2
of E. For small energies, we approximate cos θ by the quadratic approximation, which gives θ̇2 + θ2 = E, which are circles
around the origin.

Note that when the energy reaches a critical value, there is a crossing point at θ = ±π that the system does not actually
achieve in finite time. The crossing point is called a separatrix, which separates the phase space into distinct regions (bound
and unbound motion).

Consider a pendulum that is bound so that it achieves a maximum angle θmax . Then by energy conservation:

34
 θ̇2
+ (1 − cos θ) = 1 − cos θmax (6.4)
2
dθ
=⇒ p = dt (6.5)
2(cos θ − cos θmax )
dθ
=⇒ q = 2dt (6.6)
sin2 θmax
2 − sin2 θ
2

Re-parametrizing, define α so that:

θ θmax
sin = sin sin α (6.7)
2 2
since we expect that θ ≤ θmax . The integral becomes:

Z π/2
1
T =4 q dα (6.8)
1 − sin2 θmax
sin2 α


0
2

This is an elliptic integral. The elliptic integral of the first type is:

Z φ
1
F (φ|m) = p dθ (6.9)
0 1 − m sin2 θ
If φ = π/2, then the integral is called a complete elliptic integral of the first type.

For small θmax , we can Taylor expand the integrand to get an approximation:

π/2 Z π/2 
θ2 2
πθ2
Z     
1 1 θmax
T =4 1 + sin2 max sin2 α dα ≈ 4 1+ sin2 α dα = 2π 1 + max + · (6.10)
0 2 2 0 2 4 16
where we note that the leading order term is the first order approximation if we had made the expansion sin θ ≈ θ.
The second order correction term is positive, implying that the period increases as θmax increases, which is consistent with
intuition since the cosine potential falls off more rapidly as θ increases than the quadratic potential approximation.

6.1.2 Central Force

3D motion to 2D motion First consider the full Lagrangian:

M1 (~r˙1 )2 + M2 (~r˙2 )2
L= − V (|~r1 − ~r2 |) (6.11)
2
Note that the system exhibits displacement symmetry. That is, if we displace the coordinates by a fixed constant, the
Lagrangian remains the same:

∂ L̃
L̃(~rj , ~r˙j , ~a) = L(~r1 + ~a, ~r2 + ~a, ~r˙1 , ~r˙2 ) =⇒ =0 (6.12)
∂~a
∂L ∂L
=⇒ + =0 (6.13)
∂~r1 ∂~r2
d
=⇒ (~p1 + p~2 ) = 0 (6.14)
dt
Hence there is an ignorable coordinate. We define the center of mass coordinate:

~ = M1~r1 + M2~r2 ,
R ~r = ~r1 − ~r2 (6.15)
M1 + M2

35
 and the Lagrangian becomes:

M ~˙ 2 µ ˙ 2
L= (R) + (~r) − V (r) (6.16)
2 2
Clearly, R is an ignorable coordinate. Proceed by use of a Routhian. Then we just need to solve for a one-body motion
in terms of the coordinate r.

Proof that motion is in a plane Move into spherical coordinates. Then the Lagrangian for the CM frame (removing
the constant effect of the CM motion) is:

µ 2 
L= ṙ + r2 θ̇2 + r2 sin2 θ + φ̇2 − V (r) (6.17)
2
Clearly, φ is ignorable, hence pφ is a constant. But this means that:

∂L
pφ = = µr2 sin2 θφ̇ (6.18)
∂ φ̇
is conserved. But this is precisely the angular momentum with respect to the z-axis. Note that the angular momentum
can be written as:

~ AM = ~r × (µ~r˙ )
L (6.19)

and since it is a constant, both ~r and ~r˙ must lie in the plane orthogonal to L
~ AM . Moving into this plane, we realize that
π
θ can just be taken to be 2 so that the Lagrangian becomes:

µ 2 
L= ṙ + r2 φ̇2 − V (r) (6.20)
2
The total energy is given by:

µṙ2 l2
E= + + V (r) = constant (6.21)
2 2µr2
l
Hence the strategy is to obtain r(t) by solving the 1D motion, then obtaining φ̇ = µr(t)2 , and integrating to obtain φ(t)
to fully describe the system.

6.1.3 Keplerian central force motion

−k
Consider V = r . Conservation of energy gives:

µṙ2 l2 k
E= + 2
− (6.22)
2 2µr r
Consider the phase space plot of this system, ṙ against r.

36
 The separatrix corresponds to the level set that first extends to infinity.

The equation of motion in 1D can be re-written as:

s
l2
 
2
ṙ = E−V − (6.23)
µ 2µr2
l
φ̇ = (6.24)
µr2
and taking the ratio of these two, we obtain:

φ̇ dφ l 1
= = 2r  (6.25)
ṙ dr µr 2 l2

µ E−V − 2µr 2

Now further that if we define u = 1r , the energy of the system can be written as:

2
l2 l2 2

du
+ u − ku = E (6.26)
2µ dφ 2µ
which is a harmonic oscillator conservation of energy equation (somewhat)! Taking the derivative of this equation with
respect to φ,

d2 u µk
2
+u= 2 (6.27)
dφ l
1 µk
=⇒ u = = 2 + A cos φ, A∈R (6.28)
r l
l2
Define p = µk , which has a unit of distance. Also define pA = . Then we have:

P
r= (6.29)
1 +  cos φ
which is the equation of a conic section. If  = 0, it is a circle, if  < 1, we have an ellipse, if  = 1 we have a parabola, if
 > 1 we have a hyperbola.

The energy of the system can be parametrized in terms of :

k
E=− (1 − 2 ) (6.30)
2P

37
6.2 Thursday 5 Nov 2015
Keplerian motion: Elliptic case Recall the parametrization of the ellipse:

P
r= (6.31)
1 +  cos φ
with parameters:

l2
P = (6.32)
µk
k
E=− (1 − 2 ) (6.33)
2P
The periapsis and apoapsis are given by:

P
rperi = (6.34)
1+
P
rapo = (6.35)
1−
and the semi-major and semi-minor axes are:

P
a= (6.36)
1 − 2
P
b= √ (6.37)
1 − 2
The focal length c is the distance between the focus and the origin, and is given by:

P
c= (6.38)
1 − 2
The relation between a, b and c is:

b2 + c2 = a2 (6.39)

which can be proven by noting that the sum of the distances of any point on the ellipse to the two foci is equal and will
have value 2a.
Note that the area swept up by the radial vector is:

1 2 dA 1 l
dA = r dφ =⇒ = r2 φ̇ = (6.40)
2 dt 2 2µ
Hence angular momentum conservation implies Kepler’s second law. We can also use this to calculate the period of the
orbit by knowing the total area of the ellipse:

r
l µa3
πab = T =⇒ T = 2π (6.41)
2µ k
Explicit time dependence We parametrize the ellipse:

X = a cos ψ (6.42)
Y = b sin ψ (6.43)

where ψ is measured from the origin. The angle φ can be defined similarly with reference to the focus. By convention, φ
is called the true anomaly and ψ is called the eccentric anomaly.

38
 The radial coordinate can be related to ψ using:

r = a(1 −  cos ψ) (6.44)

We note that we can relate the time and the radial coordinate by rearranging the conservation of energy equation:

r Z
µ dr
t= q (6.45)
2 E+ k
− l2
r 2µr 2

and inserting the expression for r(ψ):

r r
µa3 µa3
Z
t= dψ(1 −  cos ψ) = (ψ −  sin ψ) (6.46)
k k
q
µa3
This implicitly defines ψ(t). Clearly, as ψ → ψ + 2π, we re-obtain the expression for the period: T = 2π k .

Hyperbolic Motion We parametrize the left branch of the hyperbola:

X = −a cosh ψ (6.47)
Y = b sinh ψ (6.48)

where ψ is measured from the origin. The corresponding implicit definition of ψ(t) is given by:

r
µa3
T (ψ) = ( sinh ψ − ψ) (6.49)
k
Scattering problem Consider a repulsive central force problem k < 0. We then define:

l2
P =− (6.50)
µk
which gives the polar representation of the orbit:

P
r= (6.51)
 cos φ − 1
where φ is measured from the focus. This corresponds to the other branch of the hyperbola that turns away from the focus.

Closed Orbits Define ∆φ to be the angular difference between two instances where r = rmax that is separated by one
instance of r = rmin . Note that in the Keplerian motion, ∆φ = 2π so that the orbit is closed. We can consider a perturbation
to the Kepler potential so that ∆φ = 2π + δ, where δ is small. Hence for each orbit, the perihelion precesses by δ.

Potential and closed orbits Now we want the condition on the potential so that the orbit is closed. Note that the
condition for closed orbits is for ∆ to be a rational multiple of π:

m
∆φ = π, m, n ∈ Z (6.52)
n
Note that ∆φ corresponds to the radial motion undergoing one periodic cycle. We want to find the period of the angular
motion.

Consider a perturbation on r. Define the equilibrium radius r0 that satisfies:

l2
− + V 0 (r0 ) = 0 (6.53)
µr03

Then expand E(r) around r0 . The first derivative must vanish because r0 is an equilibrium point.

39
 ṙ2 1 3l2
 
00
+ + V (r0 ) (r − r0 )2 = E (6.54)
2 2 µ2 r04

This gives a Harmonic oscillator potential which has an angular frequency given by:

s r
3l2 l r0 V 00
Ω= + V 00 (r0 ) = 2 3+ (6.55)
µ2 r04 µr0 V0

so that the radial motion is given by:

r = r0 + δ cos Ωt (6.56)

The angular motion is given by the angular momentum conservation equation:

 
l l l 2δ cos Ωt
φ̇ = 2 = 2 δ
≈ 2 1− (6.57)
µr µr0 (1 + r0 cos Ωt)2 µr 0 r0

2π
Integrating over 1 cycle of radial motion t : 0 → Ω , we obtain that:

l 2π
∆φ = + O(δ 2 ) (6.58)
µr02 Ω

Considering the first order terms:

2π 2π
∆φ = q 0
=q (6.59)
r dV d log V 0
3+ dr 3+ d log r
V0
0
Hence for ∆φ to be a rational multiple of π at all times and all positions, we require that ddlog V
log r be a constant. But this
means that V has to be a power-law function of r. Then we have the necessary condition for closed orbits:

V = αrn (6.60)
p √
To obtain the sufficient condition, we need to ensure that 3 + (n − 1) = 2 + n is rational. Clearly, n = −1 for the
Keplerian motion works. Also, the Harmonic potential n = 2 works.

6.2.1 Scattering Problem

Let particles with energy E move in the x direction towards a target particle at the origin. Depending on the distance of
the particles from the x-axis, the scattering effect will be different. Define the impact parameter s to be the vertical distance
between the horizontal incoming particle path and the x-axis. The impact parameter is then related to the scattering angle.

Note that the solid angle around the target particle can be written in terms of the angle made with the scattered particles:

dΩ = 2π sin θdθ (6.61)

The particles that pass through this solid angle can be traced back to a ring of incoming particles with impact parameters:

dN = R2πsds (6.62)

where R is the incoming flux. Then we have the equality:

1 dN sds
= (6.63)
R dΩ sin θdθ

40
 This gives us the differential cross section on the RHS. This cross section gives us the relation between θ and s. Hence
by measuring the number of particles per unit solid angle and divided by the incident rate, we can calculate the differential
cross section.

For the Keplerian problem, the relation between the impact parameter and the exit angle is:

k θ
s= cot (6.64)
2E 2
so that the differential cross section is:

k2 1
σ(θ) = (6.65)
16E 2 sin4 (θ/2)

6.2.2 Virial theorem

Define:

X
G= p~j · ~rj (6.66)
j

Calculating the total time derivative:

dG X ~ X X
= Fj · ~rj + p~j · ~r˙j = F~j · ~rj + 2Ttot (6.67)
dt j j j

If the potential has the form:

1X
Vtot = V (|~rj − ~rk |) (6.68)
2
j6=k

where the individual potentials are power law types V = krn , then:

X
F~j · ~rj = −nVtot (6.69)
j

Hence:

dG
2Ttot − nVtot = (6.70)
dt
dG
For a system in steady state, we can take the average of dt over time, which should vanish. Then the time average of
the kinetic and potential energies give:

2T̄tot − nV̄tot = 0 (6.71)

41
Chapter 7

Week 7

7.1 Tuesday, 10 Nov 2015

7.1.1 Noether’s Theorem
Example: Azimuthal symmetry Consider a frame that is rotated about the z axis at an angle θ to the x-axis. We
know that the Lagrangian will be invariant under this coordinate transform. We proceed in the Cartesian coordinate system,
pretending that we are unable to construct a coordinate system with the ignorable φ coordinate.

To have a symmetry, we require that the functional form (they are the same function!) of the Lagrangian be identical in
both the transformed and untransformed coordinates:

L0 (x0 , y 0 , z 0 , ẋ0 , ẏ 0 , ż 0 , t) = L(x, y, z, ẋ, ẏ, ż, t)

Hence L must be independent of the parameter θ. Hence we can take the partial derivative of L(x0 , y 0 , . . .) with respect to
θ around θ = 0 and require that it be zero. We will get that (along the trajectory so that we can apply the Euler-Lagrange
equations):

d
[px y − py x] = 0
dt
and hence the angular momentum in the z-direction is zero.

General proof of Noether’s Theorem Suppose we have a set of coordinates qk , where k = 1, 2, . . . , N . Suppose we
have a map parametrized by s such that:

qk → Ql (qk , s) (7.1)
Ql (qk , 0) = ql (7.2)

Define the derivative:

∂Ql
γl (Qk ) = (7.3)
∂s (qk ,s)

Note that γl does not depend on s! That is, the transformation is linear in s. Any infinitesimal transformation can be
written in terms of a first order linear transformation.

A symmetry of the system means that:

L(Ql (qk , s), Q̇l (qk , q̇k , s), t) (7.4)

is independent of s. Note further that:

42
 X ∂Ql
Q̇l = q̇k (7.5)
∂qk
k

Taking the derivative of the Lagrangian with respect to s,

∂L X ∂L ∂Qk X ∂L ∂ Q̇k
= + =0 (7.6)
∂s ∂qk ∂s ∂ q̇k ∂s
k k

But by the equality of mixed partial derivatives:

∂ Q̇k X  ∂ ∂Qk  d ∂Qk

= q̇m = (7.7)
∂s m
∂qm ∂s dt ∂s
Hence we may combine this with the derivative of the Lagrangian to get:
" #
d X ∂Qk ∂L
pk = 0, pk = (7.8)
dt ∂s ∂ q̇k
k
X
=⇒ pk γk = constant (7.9)
k

Back to rotations Note that the transformed coordinate vector can be written as:

~ = ~x + s~n × ~x
X (7.10)
where ~n is a constant vector that we are rotating around. In terms of components:

X
Xj = xj + s jkl nk xl (7.11)
k,l

so that the components of the gamma vector are:

X
γj = jkl nk xl (7.12)
k,l

The cross product was written in terms of the Levi-Civita symbol:

X
~ × B)
(A ~ j= jkl Ak Bl (7.13)
k,l

Then the constant of motion is:

X
I= pj γj (7.14)
j
X
= pj jkl nk xl (7.15)
j,k,l
X X
= nk jkl xl pj (7.16)
k j,l
X X
= nk klj xl pj (7.17)
k j,l

= ~n · (~x × p~) (7.18)

and we recognize this to be the n̂ component of the angular momentum. Hence the invariance of the Lagrangian under a
rotation about ~n corresponds to the conservation of the momentum about that axis.

Actually the easy way to do this without the Levi-Civita symbol is to use the vector triple product:

I = p~ · ~γ = p~ · (~n × ~x) = ~n · (~x × p~) (7.19)

43
7.1.2 Legendre Transform
Recall from Thermodynamics Recall that the first law of thermodynamics can be written as:

dU (S, V ) = T dS − P dV (7.20)

where the parameters are related by:

   
∂U ∂U
= T, = −P (7.21)
∂S V ∂V S

If we want to use T and V to parametrize the system, we use the Helmholtz free energy F = U − T S, the Legendre
transform of the internal energy U , so that dF = −SdT − pdV is written in differentials of T and V .

Mathematical definition Suppose we have parameters y, z and two functions A(y), B(z). Suppose that:

dA(y)
z= (7.22)
dy

Then define B(z) = yz − A(y) to be the Legendre transform of A(y). Note that in the definition of B(z), the values of y
are obtained from z by inverting z = dA(y)
dy .

The total derivative of B(z) is:

 
dA dA
dB = ydz + zdy − dA = ydz + zdy − dy = ydz + dy z − = ydz (7.23)
dy dy
dB(z)
=⇒ =y (7.24)
dz
The following properties hold:

A + B = yz (7.25)
dA = zdy (7.26)
dB = ydz (7.27)
=⇒ dA + dB = zdy + ydz (7.28)

In order to solve for the transform in a unique fashion, we require that A is convex:

d2 A
>0 (7.29)
dy 2

The geometric interpretation is that if A(y) is plotted against y, then the slope is given by z, and the linear approximation
at a point has a vertical axis intercept given by −B.

Relation to Lagrangian and Hamiltonian dynamics Recall that:

∂L
p= (7.30)
∂ q̇
so z ↔ p, y ↔ q̇. Hence the Legendre transformation of the Lagrangian is:

pq̇ − L = H (7.31)

so that the partial derivative of H with respect to p and q will be simple. Note that q̇ is represented by inverting the
definition of the canonical momentum so that it is expressed in terms of p, q.

Derivatives of the Hamiltonian

44
 ∂L ∂L ∂L ∂L ∂L
dH = q̇dp + pdq̇ −
dq − d(q̇) − dt = q̇dp − dq − dt (7.32)
∂q ∂ q̇ ∂t ∂q ∂t
∂H ∂H ∂L d ∂L
=⇒ = q̇, =− =− = −ṗ (7.33)
∂p ∂q ∂q dt ∂ q̇
dH ∂L
=− (7.34)
dt ∂t
Total derivative of the Hamiltonian

dH ∂H ∂H ∂H ∂H
= q̇ + ṗ + = (7.35)
dt ∂q ∂p ∂t ∂t
Hence the total time derivative of the Hamiltonian is equal to the partial time derivative as well! Hence:

dH ∂H ∂L
= =− (7.36)
dt ∂t ∂t

7.2 Thursday, 12 Nov 2015

Review of Noether’s Theroem Consider a continuous map parametrized by s: Qk (ql , s) so that Qk (ql , 0) = qk and
∂Qk
∂s = γk (Ql ), where γk is independent of s. Defining the canonical momenta:

∂L
pk = (7.37)
∂ q̇k Ql ,Q̇l ,t

The following quantity is then conserved in time:

X
I= pk γk (7.38)
k

Example 1: SHO Consider a Lagrangian:

1 2 1 2
mq̇ − kq
L= (7.39)
2 2
∂L
p= = mq̇ (7.40)
∂ q̇
p
=⇒ q̇ = (7.41)
m
p2 p2 1 p2 1
H = pq̇ − L = − + kq 2 = + kq 2 (7.42)
m 2m 2 2m 2
The Hamilton equations of motion are hence:

∂H p
q̇ = = (7.43)
∂p m
∂H
ṗ = − = −kq (7.44)
∂q
Example 2: Coupled oscillators

1X 1X 1 1
L= q̇k tkl q̇l − qk vkl ql = q̇ T tq̇ − q T vq (7.45)
2 2 2 2
k,l k,l
∂L
p= = tq̇ (7.46)
∂ q̇
1 1 1 1
H = p · q̇ − L = pT q̇ − L = (q̇ T t)q̇ − q̇ T tq̇ + q T vq = q̇ T tq̇ + q T vq (7.47)
2 2 2 2
1 −1 T −1 1 T 1 T −1 T −1 1 T
=⇒ H = (t p) t(t p) + q vq = p (t ) tt p + q vq (7.48)
2 2 2 2
1 T −1 T 1 T 1 T −1 1 T
=⇒ H = p (t ) p + q vq = p t p + q vq (7.49)
2 2 2 2

45
 where we note that t−1 = (t−1 )T because t and t−1 are both symmetric. The canonical equations are:

∂H
q̇ = = t−1 p (7.50)
∂p
∂H
ṗ = − = −vq (7.51)
∂q
Example 3: Spherical coordinates

m 2 
L= ṙ + r2 θ̇2 + r2 sin2 θφ˙2 (7.52)
2
∂L
pr = = mṙ (7.53)
∂ ṙ
pθ = mr2 θ̇ (7.54)
2 2
pφ = mr sin θφ̇ (7.55)
m 2 
H = pr ṙ + pθ θ̇ + pφ φ̇ − L = ṙ + r2 θ̇2 + r2 sin2 θφ˙2 (7.56)
2
2
p2r p2r p2θ p2φ

m 2  pθ 2 m 2 2 pφ
=⇒ H = + r + r sin θ = + + (7.57)
2m 2 mr2 2 mr2 sin2 θ 2m 2mr2 2mr2 sin2 θ
The sum of the second and third terms is the total angular momentum squared.

Liouville’s Theorem Consider a phase space density ρ(p, q, t) corresponding to the number of particle trajectories per
unit phase space. Draw a box from (q, q + ∆q) × (p, p + ∆p). We want to find the rate of change of the number of particles
contained in that box. In a small time interval ∆t:

Z p+∆p Z p+∆p
0 0 0
∆N = dp ∆tq̇(q, p , t)ρ(q, p , t) − dp0 ∆tq̇(q + ∆q, p0 , t)ρ(q + ∆q, p0 , t) (7.58)
p p
Z q+∆q Zq+∆t
+ dq 0 ∆tṗ(q 0 , p, t)ρ(q 0 , p, t) − dq 0 ∆tṗ(q 0 , p + ∆p, t)ρ(q 0 , p + ∆p, t) (7.59)
q q
   
∂ ∂
=⇒ ∆N ≈ − (q̇ρ) ∆q∆p∆t + − (ṗρ) ∆q∆p∆t (7.60)
∂q ∂p
∆ρ ∂ ∂
=⇒ = − (q̇ρ) − (ṗρ) (7.61)
∆t ∂q ∂p
In the limit where ∆t → 0:

 
∂ρ ∂ q̇ ∂ ṗ ∂ρ ∂ρ
= −ρ + − q̇ − ṗ (7.62)
∂t ∂q ∂p ∂q ∂p
This is a continuity equation. Consider the fluid analogue:

∂ρ ∂ρ
+ ∇ · ~j = 0 =⇒ + ∇ · (ρ~v ) = 0 (7.63)
∂t ∂t
∂ρ X ∂ρ
=⇒ + vj + ρ(∇ · ~v ) = 0 (7.64)
∂t j
∂xj

∂ρ P ∂ρ
The term ∂t + j vj ∂xj
refers to the change in ρ following a particle trajectory. Write it as a total derivative. Then:

dρ
+ ρ(∇ · ~v ) = 0 (7.65)
dt
Similarly, we express the phase space conservation law accordingly:

 
dρ ∂ q̇ ∂ ṗ
+ρ + =0 (7.66)
dt ∂q ∂p

46
 We can generalize this result to a large number of dimensions. Then:

 
dρ X ∂ q̇k ∂ ṗk
+ρ + =0 (7.67)
dt ∂qk ∂pk
k

Using the canonical equations of Hamiltonian dynamics:

∂H
∂ −∂H
!
dρ X ∂ ∂pk ∂qk
+ρ + =0 (7.68)
dt ∂qk ∂pk
k
dρ X  ∂2H ∂2H

=⇒ +ρ − =0 (7.69)
dt ∂qk ∂pk ∂pk ∂qk
k
dρ
=⇒ =0 (7.70)
dt
by the equality of mixed particle derivatives.

Consider an initial condition where all the particle trajectories are initially contained in a volume Ω, so that:
(
ρ0 , inside Ω
ρ= (7.71)
0, outside
by the conservation of particle number:
Z
ρdpdq = N = ρ0 V (Ω) (7.72)

Now let the system evolve. Let the trajectories be contained in a new volume Ω0 after time t. By the statement that
dρ
dt = 0 earlier, we know that the density of phase space cannot change along the trajectories. Then:

(
ρ0 , inside Ω0
ρ(t) = (7.73)
0, outside
Implementing the conservation of particle number,
Z
ρdpdq = N = ρ0 V (Ω0 ) (7.74)

But this means that V (Ω0 ) = V (Ω). Hence the volume occupied by the particle trajectories does not change in time.

Poincare Recurrence Theorem Consider a system bound in some finite region of phase space. Consider a Hamiltonian
that is independent of time. Weak version: For any point of phase space P, in any neighbourhood U of P, all points of U
either stay in U forever, or there exists 1 point x ∈ U that leaves U and then returns.

Proof of weak form Suppose a point Q leaves U and reaches R. Pick a region around Q called Ω0 such that Ω0 ∈ U .
Then Ω0 evolves to Ω1 6∈ U by making Ω0 small enough. We can repeat this analysis to get a sequence of regions {Ωi }. But
since the phase space volume is finite, there will exist k < n such that Ωk ∩ Ωn 6= 0, and there is some overlap between the
regions occupied by the phase volume at two times. But this also means that Ωk−1 and Ωn−1 must have overlapped as well,
because they have the same volume at all times. We can trace this backwards to declare that Ω0 must have overlapped with
Ωn−k .

Formulation of action in the Hamiltonian picture

Z t2
S[p, q] = (pq̇ − H)dt (7.75)
t1
Z t2  
∂H ∂H
=⇒ δS[p, q] = q̇δp + pδ q̇ − δp − δq dt (7.76)
t1 ∂p ∂q
Z t2  
d ∂H ∂H
= q̇δp + (pδq) − δp − δq dt (7.77)
t1 dt ∂p ∂q

47
where we only consider trajectories that start at a fixed q1 and end at a fixed q2 . That is, δq(t1 ) = δ(t2 ) = 0. Rearranging,

Z t2     
∂H ∂H
q̇ − δp − ṗ + δq dt = 0 (7.78)
t1 ∂p ∂q

which simply gives us the canonical Hamilton equations.

48
Chapter 8

Week 8

8.1 Tuesday 17 Nov 2015

8.1.1 Canonical transformations
Coordinate transforms in Lagrangian dynamics Recall that we can write Q(q, t) so that the velocity Q̇(q, q̇, t) =
∂Q(q,t) ∂Q(q,t)
∂q q̇ + ∂t is given automatically. Then the new Lagrangian L̄ must satisfy:

L̄(Q(q, t), Q̇(q, q̇, t), t) = L(q, q̇, t) (8.1)

Canonical transformation We want to find the maps Q(q, p, t) and P (q, p, t) that satisfy Hamilton’s equations. Note
that defining Q does not automatically define P , unlike the case of the point transformation Q(q, t) earlier (since Q̇ is auto-
matically defined from Q).

Constraints on the canonical transformation The basis of the constraint is Liouville’s theorem (volume of phase
space remains constant). The area of the image of the phase space volume under the transformations P, Q must remain
constant. This can be written as the Jacobian:

∂(P, Q)
= constant
∂(p, q)
0
The constant turns out to be AA , the ratio of the volume under P, Q to the volume without the map. We hence write the
condition for a canonical transformation:

  ∂Q  
∂Q
 
∂q ∂p
det   p,t  q,t  = C
∂P ∂P
∂q ∂p
p,t q,t

Suppose we want the area under the mapping to be the same. That is, C = 1. This can always be done by rescaling P, Q
and H so that the determinant is ±1. More precisely, the rescaling will take C → λ2 C. Switching P ↔ Q will change the
sign of the constant C → −C. We can just choose the constant to be +1 by convention.

Poisson Bracket Define:

∂A ∂B ∂A ∂B
{A(q, p, t), B(q, p, t)}q,p ≡ −
∂q ∂p ∂p ∂q
By comparison to the condition on the Jacobian determinant, we have the necessary condition of the canonical transfor-
mation:

{Q, P }q,p = 1 (8.2)

Generating Function First construct a class of maps from q, p → Q, P . We want to find H̃ so that:

49
 Z t2 h i
δ P Q̇ − H̃(Q, P, t) dt (8.3)
t1

is stationary if:

Z t2
δ [pq̇ − H(q, p, t)] dt = 0 (8.4)
t1

where we have taken δQ(t1 ) = δQ(t2 ) = 0.

We can write this condition as:

dF (q, Q, t)
P Q̇ − H̃(Q, P, t) = pq̇ − H(q, p, t) − (8.5)
dt
∂F ∂F ∂F
=⇒ P Q̇ − H̃(Q, P, t) = pq̇ − H(q, p, t) − q̇ − Q̇ − (8.6)
∂q ∂Q ∂t

Note that we disallowed F to depend on q̇ and Q̇. Comparing coefficients, we note that if construct F such that:

∂F
P =− (8.7)
∂Q
∂F
p= (8.8)
∂q
then we obtain that:

∂F (q, Q, t)
H̃(Q, P, t) = H(q, p, t) + (8.9)
∂t
Note that it is possible to invert the equations for P, p to complete determine P, Q in terms of p, q.
Given the stationary action condition:

Z t2 h i Z t2
δ P Q̇ − H̃(Q, P, t) dt = δ [pq̇ − H(q, p, t)] dt − δF (q2 , Q2 , t2 ) + δF (q1 , Q1 , t1 ) (8.10)
t1 t1

It can be shown that if p, q satisfy the canonical equations under H, then under the Hamiltonian H̃(Q, P, t), Hamilton’s
equations for Q, P also hold:

∂ H̃
Q̇ = (8.11)
∂P
∂ H̃
Ṗ = − (8.12)
∂Q
Constructing the actual generating function Given the conditions on the generating function:

∂F
P =− (8.13)
∂Q
∂F
p= (8.14)
∂q
We can define the function F :

Z q0 ,Q0
F (q0 , Q0 ) = −P (dQ) + pdq (8.15)
O

where O is a reference point. Note that this integral definition is valid only if the integral is path-independent. This path
independence implies that the closed loop integral must vanish:

50
 I
(−P dQ + pdq) = 0 (8.16)

But the loop integral is simply the difference in phase space volume in the two maps. Since we required that the phase
space volume remain constant under the mapping, the loop integral is zero, and hence F (q, Q) is defined and satisfies the
requirements of the generating function.

Examples using the canonical transformation Consider the generating function F = qQ. Then:

∂F
P =− = −q (8.17)
∂Q
∂F
p= =Q (8.18)
∂q
Note that this transformation exchanges the role of position and momentum (up to a sign change).

Harmonic Oscillator using Canonical transformations Consider the Hamiltonian:

p2 + ω 2 q 2
H= (8.19)
2
Define the generating function:

1 2
F (q, Q) = ωq cot(2πQ) (8.20)
2
∂F
p= = ωq cot(2πQ) (8.21)
∂q
∂F πωq 2
P =− = (8.22)
∂Q sin2 2πQ
Solving these two equations simultaneously,

ω2 2
p2 + ω 2 q 2 = P (8.23)
π2
Hence the new Hamiltonian is:

ω
H̃(Q, P, t) = P (8.24)
2π
Implementing the Hamilton’s equations,

∂ H̃ ω ω
Q̇ = = =⇒ Q = Q0 + t (8.25)
∂P 2π 2π
∂ H̃
Ṗ = − = 0 =⇒ P = P0 (8.26)
∂Q
Inverting the equations for P, Q,

r
P
q= sin(2πQ) (8.27)
πω
r
ωP
p= cos(2πQ) (8.28)
π
Hence the amplitude of the oscillation is a constant, and the time dependence is contained inside the trigonometric
function.
Generating other canonical transformations

51
 Recall that the generating function was previously obtained by:

dF1 (q, Q, t)
P Q̇ − H̃(Q, P, t) = pq̇ − H(q, p, t) − (8.29)
dt
We can rearrange the terms to write:

d
−Ṗ Q − H̃ = pq̇ − H − (F1 + P Q) (8.30)
dt
Define F2 (q, P, t) = F1 + P Q. Then we can pick the derivatives of F2 to cancel the other terms:

∂F2
=p (8.31)
∂q
∂F2
=Q (8.32)
∂P
∂F2
=⇒ H̃ = H + (8.33)
∂t
Why introduce more transformations? F2 (q, P, t) is useful because we can choose transformations that have
F2 (q, P, t) = f (q)P so that the generating function conditions become:

∂F2 df
p= = P (8.34)
∂q dq
∂F2
Q= = f (q) (8.35)
∂P
This allows us to represent the new position Q as an arbitrary function of the old position q. Observe that if we pick
f (q) = q, then we obtain the identity. This means that the generating function F2 = qP achieves the identity transformation.

Infinitesimal transformation Consider the canonical transformation:

F2 = qP + G(q, P ) (8.36)

Note that as  → 0, the transformation becomes the identity transformation. Then the generating function equations
give:

∂F2 ∂G(q, P )
p= =P + (8.37)
∂P ∂q
∂F2 ∂G(q, P )
Q= =q+ (8.38)
∂P ∂P
Note that since p is close to P by the first equation, we can approximate G(q, P ) = G(q, p) + O(), so that we obtain the
approximate relations (note that P,Q are on the left and p,q are on the right):

∂G(q, p)
P =p− (8.39)
∂q
∂G(q, p)
Q=q+ (8.40)
∂p
to first order.

But this looks like Hamilton’s equations! G then tells how we should update p, q to obtain P, Q. Hence H(q, p) is related
to G(q, p), and the infinitesimal canonical transformation F2 = qP + H.

52
Chapter 9

Week 9

9.1 Tuesday, 24 Nov 2015

9.1.1 Parametric Resonance
Define the vector
 
q1
 q2

..
 
 
  .
~x = 
  (9.1)
 p1 

 p2 
 
..
.

Then Hamilton’s equations result in a set of first order differential equations that can be written:

~x˙ = A~x (9.2)

where A is a 2n × 2n matrix that does not depend on time.

To include parametric effects, we need to let A depend on time:

~x˙ = A(t)~x (9.3)

and let A be periodic with period T .

Define ~ej to be the solution to the matrix differential equation that has initial conditions that is ~x0 = (0, . . . , 1, 0, . . .)
where the 1 occurs in the ith position. Then the matrix where ~ei are in the columns is the fundamental solution:

~x(t) = M (t)~x(0) (9.4)

The fundamental solution satisfies the matrix differential equation as well:

M 0 (t) = A(t)M (9.5)

Note further that M (0) = I, the identity matrix.

Floquet Theorem Statement:

M (t + T ) = M (t)M (T ) (9.6)

For multiple periods,

53
 M (t + jT ) = M (t)M j (T ) (9.7)

To examine stability, we hence want to examine the powers of M (T ). The best way to raise matrices to power is to move
into the diagonal basis. Note that:

M j (T ) = (U −1 DU )j = U −1 D j U

We want to examine the eigenvalues of M (T ), which are the diagonal elements of D.

54
Chapter 10

Week 10

10.1 Tuesday 1 Dec 2015

10.1.1 Equivalent statements for determining if a transformation is canonical
1. Phase space area is preserved Note that we require that the Jacobian determinant is unity for the volume in phase
space to be the same:
∂Q ∂P ∂Q ∂P
− =1
∂q ∂p ∂p ∂q
But the LHS is the Poisson bracket. Hence we require that:

[Q, P ]q,p = 1

Once the Poisson bracket of the coordinates is unity, we immediately have that the Poisson brackets in each of the
coordinate systems is invariant:

[A, B]p,q = [A, B]P,Q

2. Canonical transformations are generated with a generating function

First consider a generating function of the first type. Then F1 (q, Q, t) satisfies:

∂F1 (q, Q, t)
p= , invert to get Q(q, p, t)
∂q
∂F1 (q, Q, t)
P =− , substitute previous inversion to get P (q, p, t)
∂Q
Then the new Hamiltonian is:

∂F1
K=H+
∂t
Generating function for infinitesimal canonical transformation Consider F2 (q, P ) = qP + G(q, P ). Recall that
F2 (q, P ) = qP is the identity transformation. The second term corresponds to a small perturbation. The generating function
produces the transformation:

∂G(q, P )
P =p−
∂q
∂G(q, P )
Q=q+
∂P
We can replace P with p and group the quadratic and higher order corrections:

55
 ∂G(q, p)
P =p− + O(2 )
∂q
∂G(q, p)
Q=q+ + O(2 )
∂p

Hamilton-Jacobi theory Suppose we can find a generating function F2 (q, P, t) so that the new Hamiltonian is zero.
Then in the new coordinate system, P, Q are constants. The time-dependent evolution is hence obtained from the map from
P, Q → p, q from the generating function F2 .
Now we want:

∂F2 (q, P, t)
K(q, P, t) = H(q, p, t) + =0 (10.1)
∂t
We want to write the equation using exactly two variables in {q, p, Q, P }. We choose to write the equation in terms of q, P .

Recall that the transformation equations for F2 are:

∂F2 (q, P, t)
p= (10.2)
∂q
∂F2 (q, P, t)
Q= (10.3)
∂P
We can use the first equation to replace all instances of p. Then the new Hamiltonian is:

 
∂F2 (q, P, t) ∂F2 (q, P, t)
K(q, P, t) = H q, ,t + (10.4)
∂q ∂t

We now write:

S(q, t) = F2 (q, P, t) (10.5)

for any fixed P (it is going to be a constant anyway). Then the equation we want to solve is:

 
∂S(q, t) ∂S(q, t)
H q, ,t + =0 (10.6)
∂q ∂t

The previous equation is known as the Hamilton-Jacobi equation. If we can solve this equation for S(q, t) we will obtain
a whole family of possible functions corresponding to different values of P due to the constants of integration. To indicate
that P is a constant, we write it as P = α. Then:

∂S(q, α, t)
p= (10.7)
∂q
Hamilton-Jacobi Equation in multiple dimensions

 
∂S ∂S
H qk , ,t + =0 (10.8)
∂qk ∂t

which has N + 1 constants of integration. One of these constants will correspond to an additive constant to S. For the
other N constants of integration, we denote them by αk , k = 1, 2, . . . , N and express the solution in terms of them:

S(qk , αk , t) (10.9)

Time-independent Hamiltonian If H is not dependent on time, its partial derivative with respect to time is zero.
Then we can introduce:

56
 S = −Et + W (qk , α1 , . . . , αN −1 ) (10.10)

which gives us the simplified Hamilton-Jacobi equation with N − 1 constants of integration:

 
∂W ∂W
H q1 , . . . , q N , ,..., −E =0 (10.11)
∂q1 ∂qN −1

The canonical transformation equations generated by the solution to the H-J theory is:

∂S(qk , Pk , t)
Ql = (10.12)
∂Pl
∂S(qk , Pk , t)
pl = (10.13)
∂ql
Example of HJ Theory Consider 1D motion:

p2
H= + V (q) (10.14)
2
First write down the H-J equation for time-independent Hamiltonians:

 2
1 ∂S
+ V (q) − E = 0 (10.15)
2 ∂q

Rearranging,

Z q p
W (q) = 2E − 2V (x)dx (10.16)
q0

where q0 is a constant of integration and isn’t important. Hence we have the formal solution:

Z q p
S(q, P, t) = 2(P − V (x))dx − P t (10.17)
q0

where we noted that the only relevant constant of integration is E, which we can write as P = E. Imposing the
transformation equations,

Z q
∂S(q, P, t) 1
Q= = p dx − t (10.18)
∂P q0 2(P − V (x))
dx
which indeed is the formal solution to 1D motion because the integrand is the inverse velocity and v = dt. The other
coordinate is:

∂S(q, P, t) p
p= = 2(P − V (x)) (10.19)
∂q
which indicates that Q represents time while p represents the usual momentum.
Multi-dimensional system using Hamilton-Jacobi theory Consider the Hamiltonian in spherical coordinates with
a special form for the potential:

!
1 p2 p2φ Uθ (θ) Uφ (φ)
H= p2r + 2θ + 2 2 + Ur (r) + + 2 2 (10.20)
2 r r sin θ r2 r sin θ

The solutions are separable. That is, we can write:

57
 S = W − Et = Wr (r) + Wθ (θ) + Wφ (φ) − Et (10.21)

Substituting this ansatz into the H-J equation:

 2 "  2 # "  2 #
1 dWr 1 1 dWθ 1 1 ∂Wφ
+ Ur (r) + 2 + Uθ (θ) + 2 2 + Uφ (φ) = E (10.22)
2 dr r 2 dθ r sin θ 2 dφ

Now E is already a constant of integration. We need two other constants of integration. We note that the φ terms have
to be a constant:

 2
1 ∂Wφ
+ Uφ (φ) = αφ (10.23)
2 dφ

We pick the other integration constant to be:

 
1 dWθ αφ
+ Uθ (θ) + = αθ (10.24)
2 dθ sin2 θ
Then we want to solve the equation:

 2
1 dWr αθ
+ Ur (r) + =E (10.25)
2 dr r2

The solution to these differential equations fully specifies S.

10.1.2 Connection to QM
In operator terms, the QM Hamiltonian is:

P̂ 2
Ĥ + V (q) (10.26)
2
and the momentum operator is:

P̂ = −i~∂q (10.27)

The wavefunction can be written as an exponential:

ψ ≈ AeiS(q,t)/~ (10.28)

we demand that the phase S(q, t) changes rapidly with respect to time so that it has a high frequency, but that the
frequency changes very slowly compared to the timescale of the frequency itself, so that we can write:

i ∂S iS/~
∂t ψ ≈ A e (10.29)
~ ∂t
The spatial derivative is also approximately:

i ∂S iS/~
∂q ψ ≈ A e (10.30)
~ ∂q
∂S
The second derivative of space is approximately (throwing out the derivative of ∂q since it is assumed to change very
slowly):

58
  2
i ∂S
∂q2 ψ ≈A eiS/~ (10.31)
~ ∂q

This is known as the WKB approximation. The Schrodinger equation is hence:

∂S
i~∂t ψ = −A eiS/~ (10.32)
∂t
 2
~2 2 A ∂S
− ∂q ψ ≈ eiS/~ (10.33)
2 2 ∂q
 2
1 ∂S
=⇒ + V (q) + ∂t S = 0 (10.34)
2 ∂q

but this is the Hamilton-Jacobi equation! Hence in the limit where the quantum mechanical de-Broglie wavelength goes
to zero, the principal function S gives the phase of the wavefunction. This gives us a method of approximately solving the
wave equation using the Hamilton-Jacobi principal solution.

To construct the full wavefunction, we need to superimpose plane wave solutions:

Z
ψ= dEf (E)eiS(q,E,t)/~ (10.35)

where f (E) is a weighting function that is peaked at E0 with a phase that may vary linearly:

f (E) = g(E − E0 )e−i(E−E0 )β/~ (10.36)

where g is a function that is peaked at zero. We use the method of steepest descent/method of stationary phase.
We note that the phase of the integrand is given by:

i
[S(q, E, t) − Eβ] (10.37)
~
Hence for the values of q for which the wavefunction has non-zero value, we require that the phase is first order stationary
with respect to E so that they add constructively when integrated across E:

∂S(q, E, t)
=β (10.38)
∂E
But this is the same relation as the canonical transformation:

∂S
Q= (10.39)
∂P
because P = E is a constant of integration in the principal solution.
Relation to the Path Integral formulation of QM Recall that the propagator can be written as:

X
U (t) ∝ eiS/~ (10.40)
all paths

This suggests that the classical action is related to the principal function. Recall that the classical action is given by:

Z t2
Scl = Ldt (10.41)
t1

Consider the time derivative of the principal function:

59
 dS ∂S ∂S ∂S
= q̇ + Ṗ + (10.42)
dt ∂q ∂P ∂t

Note that Ṗ = 0 because the new coordinates are constants. Also note that since:

∂S
H+ =0 (10.43)
∂t
∂S
=p (10.44)
∂q
the time derivative is given by:

dS
= pq̇ − H = L (10.45)
dt
Hence the principal function is actually the classical action! Hence the generating function that converts the new Hamil-
tonian to zero is actually the classical action.

10.1.3 Action-Angle Variables

Consider motion that is periodic. We want to find the new variables I, ψ (action and action-angle respectively) so that the
new Hamiltonian is a function of I alone and is not a function of ψ. Hence I and ψ are constants along the motion. The
geometric meaning of I is the area enclosed by a closed trajectory in pq phase space (divided by 2π) and ψ varies from 0 to
2π as the system moves along a closed trajectory.

10.2 Thursday, 3 Dec 2015

10.2.1 Parametric resonance
Recall the Mathieu equation:

θ̈ + (a − 2q cos 2t)θ = 0 (10.46)

The time dependent coefficient is essentially an external force feeding energy into the system. If the frequency of the
external force matches the natural frequency of the system, the external force will feed energy into the system continually so
that the system will become unstable.

10.2.2 More on Action-Angle Variables

Consider a time-independent Hamiltonian. Consider periodic motion. We want to find a canonical transformation (q, p) →
(ψ, I) so that I is constant along the classical trajectory. Then we require that the Hamiltonian satisfy:

∂H
I˙ = − =0 (10.47)
∂ψ
so that the Hamiltonian is just a function of I alone. Also, we require that along the trajectory:

I I I
pdq = Idψ = I dψ (10.48)

since the canonical transformations preserve the volume of phase space. For a closed periodic path, we can define I to be
the area enclosed by the trajectory:

I
1
pdq = I(q, p) (10.49)
2π
Hence ψ ranges from 0 to 2π over an orbit:

60
 I
dψ = 2π (10.50)

We define that the system when ψ increases by 2π should be the same system. Then we can wrap the plot of I against
ψ onto a cylinder to match the 0 and 2π points.

We want to find a generating function F2 (q, I) that performs this transformation.

Now by the transformation equations:

∂F2 (q, I)
p(q, I) = (10.51)
∂q
One such generating function can be obtained by directly integrating that transformation equation:

Z q
F2 (q, I) = p(x, I)dx (10.52)
q0

The other transformation equation is:

Z q
∂F2 ∂p(x, I)
ψ= = dx (10.53)
∂I q0 ∂I
H R
We now want to verify that dψ = ψ̇dt = 2π. The time derivative is:

∂p(q, I)
ψ̇ = q̇ (10.54)
Z Z ∂I Z
∂p(q, I) ∂ d
=⇒ ψ̇dt = dq = pdq = (2πI) = 2π (10.55)
∂I ∂I dI
Observe further that:

∂H(I)
ψ̇ = (10.56)
∂I
which is the constant oscillation frequency:

∂H(I)
ω= = ψ̇ (10.57)
∂I
Adiabatic invariance Consider a Hamiltonian parametrized by a single variable α. Then we write:

H(q, p, α) (10.58)

For a fixed α, I(α) will be constant and ψ = ω(α)t. Then the canonical transformation gives us:

q(ψ, I, α), p(ψ, I, α) (10.59)

Let the canonical transformation come from the generating function F2 (q, I, α).
Now suppose α has time-dependence. We can still use the same generating function because the transformation does not
depend on derivatives of α. The new Hamiltonian is now equal to:

∂F2
H̃ = H(q(ψ, I, α), p(ψ, I, α), α) + α̇ (10.60)
∂α
because F2 now has explicit time-dependence through α. Now note that the old Hamiltonian does not depend on ψ
because I˙ = − ∂H
∂ψ = 0. Hence we can write the new Hamiltonian as:

61
 ∂F2
H̃ = H(I, α) + α̇ (10.61)
∂α
By Hamilton’s equations,

 
∂∂F2
I˙ = − α̇ (10.62)
∂ψ∂α
 
∂ ∂F2
ψ̇ = ω(I, α) + α̇ (10.63)
∂I ∂α

Now the functional dependence of the generating function is F2 (q(I, ψ, α), I, α). F2 is hence a periodic function in ψ.
This means that I˙ is a derivative of a periodic function in ψ. Hence when integrated across a period, I will be constant.
Then we can write:

∂
I˙ = f (ψ, I, α)α̇ (10.64)
∂ψ
ψ̇ = ω(I, α) + g(ψ, I, α)α̇ (10.65)

Now consider time evolution when ψ increases by 2π. Let ∆I be the change in I over that increase. We want to show
that ∆I 2
2π is of small order, that is, it is of order α̇ and α̈. Then:

dI I˙ 1 ∂f
= = α̇ + O(α̇2 ) (10.66)
dψ ψ̇ ω0 ∂ψ

Now we can also approximate:

f (ψ, I, α) ≈ f (ψ, I0 , α0 ) + O(α̇) (10.67)

and hence we can throw the terms into the higher order collection terms:

dI ∂f (ψ, I0 , α0 ) 1
= α̇0 + O(α̇2 , α̈) (10.68)
dψ ∂ψ ω0
Integrating both sides from ψ = 0 to ψ = 2π, we hence have:

Z 2π
dI
dψ = O(α̇2 , α̈) (10.69)
0 dψ

because f (ψ, I0 , α0 ) is purely periodic in ψ. This is the concept of adiabatic invariance.

10.2.3 Review of term

62