Physics 210 Homework #4

Hanwen Qin

December 8, 2016

1 Simple Harmonic Oscillator, Hamilton-Jacobi Approach

The difficulty of checking a solution goes like the reciprocal of finding the solution. Let me compute up
front the total derivative of S(q, α, t),

q cos(ωt) − α α cos(ωt) − q mω 2 2qα cos(ωt) − (q 2 + α2 )

dS = mω dq + mω dα + dt
sin(ωt) sin(ωt) 2 sin(ωt)2

As with any generating function, the differential action

p dq − Hdt = β dα − Kdt + dS

must be preserved. The dq term tells me, implicitly, how to transform the coordinate α,

∂S q cos(ωt) − α
p= = mω
∂q sin(ωt)

the dα terms tells me how to transform the momentum β,

∂S α cos(ωt) − q
β=− = −mω (1.1)
∂α sin(ωt)

and the dt term tells me how to transform the Hamiltonian K,

 
∂S ∂S
K = H q, p = +
∂q ∂t
q cos(ωt) − α 2 mω 2 2 mω 2 2qα cos(ωt) − (q 2 + α2 )
 
1
= mω + q +
2m sin(ωt) 2 2 sin(ωt)2

When I expand the square, all terms cancel out, leaving the new Hamiltonian identically zero. So
S(q, α, t) indeed solves the Hamilton-Jacobi equation. Moreover, both α and β and conserved, so I can
invert the definition of β (1.1) to find the physical coordinate

β
q(t) = α cos(ωt) + sin(ωt)
mω

which I recognize as the general solution to the simple harmonic oscillator. In particular, α = q(0) is the
initial position and β = mq̇(0) is the initial momentum.

2 Damped Harmonic Oscillator, Canonical Approach

Part (a) The Hamiltonian Newton’s Second Law directly yields the equation of motion,

−V 0 (q) − 2mγ q̇ = mq̈ (2.1)

1
Now I take the Lagrangian
mq̇ 2
 
2γt
L(q, q̇, t) = e − V (q)
2
and compute the conjugate momentum
∂L
p= = e2γt mq̇ (2.2)
∂ q̇
its rate of change
d ∂L
ṗ = = e2γt (2γmq̇ + mq̈)
dt ∂ q̇
and the generalized force
∂L
= −e2γt V 0 (q)
∂q
Clearly the resulting Euler-Lagrange equation

V 0 (q)
q̈ = − − 2γ q̇
m

is equivalent to Newton’s Second Law (2.1), so the proposed Lagrangian is valid. Next, to derive the
Hamiltonian, I solve equation (2.2) for the velocity q̇ = (p/m) e−2γt , and take the Legendre transform,
 2 
2γt mq̇
H = pq̇ − L = e + V (q)
2
p2
H(q, p, t) = e−2γt + e2γt V (q)
2m

Part (b) A Constant of Motion Given F2 (q, P, t) = eγt qP , I compute its total derivative,

dF2 = eγt P dq + eγt q dP + γeγt qP dt

and require the differential action

p dq − Hdt = P dQ − Kdt + dF2

to be preserved. The caveat is F2 has a dP rather than a dQ, so I must subtract the total differen-
tial d(P Q) = P dQ + QdP from the right-hand side—that amounts to an integration by parts. Then I
collect differentials,

(p − eγt P ) dq + (Q − eγt q) dP + (K − H − γeγt qP ) dt = 0

and set each term to zero because q, P , t are independent variables. The dq term tells me the new
momentum P = e−γt p, the dP terms tells me the new coordinate Q = eγt q, and the dt term tells me
the new Hamiltonian,

p2
K = H + γeγt qP = e−2γt + e2γt V (q) + γeγt qP
2m
P2
K(Q, P, t) = + e2γt V (q = Q e−γt ) + γQP
2m

2
In the case of a harmonic oscillator potential

mω 2 q 2 mω 2 −2γt 2
V (q) = −→ V (q = Q e−γt ) = e Q
2 2
the transformed Hamiltonian becomes independent of time,

P2 mω 2 2
K(Q, P ) = + Q + γQP (2.3)
2m 2

Whenever the Hamiltonian has no explicit time dependence, it is a constant of motion.

Part (c) The Solution The Hamiltonian produces two canonical equations of motion,

∂K P
Q̇ = = + γQ (2.4)
∂P m
∂K
Ṗ = − = −mω 2 Q − γP (2.5)
∂Q

Taken together with the statement that K is conserved, one out of three equations is redundant. I am
tempted to solve (2.4) and (2.5) since they are first-order linear equations with straightforward initial
conditions, (
Q(0) = q(0) = x0
(2.6)
P (0) = p(0) = mq̇(0) = mv0
but I am told to use the constant of motion (2.3), so here it goes. I solve for P in terms of Q and K,
using the quadratic formula,
r
2K
P = −mγQ ± m − (ω 2 − γ 2 )Q2
m
Two things to notice: first, I don’t know yet which sign to pick. Second, in order for P to be real, the
quantity under the square root must be non-negative. An underdamped oscillator with γ < ω is therefore
subject to the constraint s
2K
|Q| ≤
m(ω 2 − γ 2 )
The maximum allowed Q corresponds to the oscillator’s amplitude, which decays exponentially. Now I
substitute this expression for P into the equation of motion for Q (2.4), making it separable,

Q̇
q = ±1
2K
m − (ω 2 − γ 2 )Q2

Integrating the right-hand side over (0, t) just gives me ±t. The system evolves forward in time, so it’s
plausible that I’ll have to choose the positive
p sign. Meanwhile,
p integrating the left-hand side from Q(0)
to Q(t) calls for the change of variable ω 2 − γ 2 Q = 2K/m sin φ.
Q(t) φ(t)
φ(t) − φ(0)
Z Z
dQ 1
q =p dφ = p
Q(0) 2K
− (ω 2 − γ 2 )Q2 ω2 − γ 2 φ(0) ω2 − γ 2
m

3
 p
So I obtain φ(t) = φ(0) + ω 2 − γ 2 t, which looks reassuringly like the phase of oscillation. Changing
back to Q,
s
2K
Q(t) = sin φ(t)
m(ω 2 − γ 2 )
s
2K h p p
2 − γ 2 t) + cos φ(0) sin( ω 2 − γ 2 t)
i
= sin φ(0) cos( ω
m(ω 2 − γ 2 )

The initial phase and energy are of course determined by the initial conditions (2.6).

P (0)2 mω 2 m
K= + Q(0)2 + γQ(0)P (0) = (v02 + ω 2 x20 ) + γmx0 v0
r2m 2 r 2
2 2
m(ω − γ ) 2
m(ω − γ )2
sin φ(0) = Q(0) = x0
2K r 2K
m(ω 2 − γ 2 )x20
q
cos φ(0) = ± 1 − sin2 φ(0) = ± 1 −
2K
Those ugly square roots all cancel out, leaving me in peace.
p v0 + γx0 p
Q(t) = x0 cos( ω 2 − γ 2 t) ± p sin( ω 2 − γ 2 t)
ω2 − γ 2

Finally, the physical coordinate q is related to the canonical coordinate Q by q = Q e−γt ,

" #
p v0 + γx0 p
q(t) = e−γt x0 cos( ω 2 − γ 2 t) + p sin( ω 2 − γ 2 t)
ω2 − γ 2

I’ve decided on the plus sign because it gives the correct initial condition for q̇. I’ve also checked that
this solution indeed satisfies the equation of motion given by Newton’s Second Law, q̈ = −ω 2 q − 2γ q̇. I
would much prefer to have solved that equation from the get-go.

3 Anharmonic Oscillator, Action-Angle Approach

The potential is periodic with infinitely tall barriers, which a classical particle can’t tunnel through, so
I’ll only consider the piece 0 < x < πx0 . I can immediately write down the Hamiltonian,
p2 a
H(x, p) = +
2m sin(x/x0 )2
It is completely integrable since there’s once one spatial dimension. The conservation of energy H = E
allows me to write the momentum as a function of position,
s  
a
p(x) = ± 2m E −
sin(x/x0 )2

where the sign depends on whether the particle is moving right (+) or left (−). The turning points are
where momentum approaches zero,
p p
x1 = x0 arcsin a/E x2 = x0 (π − arcsin a/E)

4
There’s always two turning points for E > a. E = a puts the particle at rest in equilibrium; E < a is
not allowed. The action variable is defined as the contour integral
I Z x2 Z x1
J= p dx = (+p) dx + (−p) dx
H=E x1 x2
Z x2 s  
a
=2 dx 2m E −
x1 sin(x/x0 )2

I will not attempt to evaluate this integral. If I did, I would then solve for E as a function of J, and
calculate the orbital frequency ν = dE/dJ. But I’m more interested in the period of oscillation,
r
2m x2
Z
1 dJ dx
T = = = q
ν dE E x1 1 − a/E sin(x/x0 )2

This integral is quite hard to solve. By trial and error I found that the substitution
p
1 − a/E sin φ = − cos(x/x0 ) (3.1)
p
1 − a/E cos φ dφ = sin(x/x0 ) dx/x0

maps the interval of integration x : (x1 , x2 ) 7→ φ : (−π/2, π/2), and makes the integrand trivial:
r r
2m π/2
Z
2m
T = x0 dφ = πx0 (3.2)
E −π/2 E

Interestingly the period of oscillation depends on energy. More energetic, and thus faster, particles have
shorter period. I may also calculate the angular frequency,
s
2π 2E
ω= = (3.3)
T mx20

I came up with two ways to check this result. First, one-dimensional motion can always be reduced
to quadrature. The canonical equation for the coordinate x,
dx ∂H p
= =
dt ∂p m
can be converted to a directly integrable equation for time,
dt m
=
dx p(x)
If I integrate from the left turning point to the right, I get half the period, so the full period is twice that
integral, r r
Z x2
2m x2
Z
m dx 2m
T =2 dx = q = πx0
x1 |p(x)| E x1 1 − a/E E
sin(x/x0 )2

Same integral, same answer. In fact, I may go one step further and solve the motion completely. Just
relax the upper limit of integration and make the same substitution as above,
Z x
m φ
r Z r
m m
t= dx = x0 dφ = x0 (φ + π/2)
x1 |p(x)| 2E −π/2 2E

5
and surprisingly enough, φ turns out to be the phase of oscillation,
s
2E
φ=t − π/2 = ωt − π/2
mx20

I then invoke the substitution (3.1) again to find the displacement,

hp i
x(t) = x0 arccos 1 − a/E cos(ωt)

under the initial condition that the particle starts at rest with total energy E.
The second check is to consider small oscillations δx near the equilibrium x = πx0 /2. If I Taylor-
expand the potential V (x) about the equilibrium,
 πx   πx   πx  1  πx0  2
0 0 0
V + δx = V +V0 δx + V 00 δx + O(δx3 )
2 2 2 2 2
The first term is the constant a. The second term is zero. The third term is a harmonic oscillator
potential with “force constant” V 00 (πx0 /2) = 2a/x20 . So the frequency of small oscillations is
r s
V 00 2a
ω= =
m mx20

which is the result (3.3) in the limit E → a.

4 Canonical Transformation
The invariance of Poisson brackets is a necessary and sufficient condition for a canonical transformation.
It is also easy to check; taking Q1 = q12 , Q2 = q1 + q2 ,
1. [Q1 , Q1 ] = [Q2 , Q2 ] = [P1 , P1 ] = [P2 , P2 ] = 0 is automatically satisfied.
2. [Q1 , Q2 ] = 0 because there’s no p dependence;
3. [Q1 , P1 ] = 2q1 ∂P
∂p1 = 1 −→
1 ∂P1
∂p1 = 1
2q1

4. [Q1 , P2 ] = 2q1 ∂P
∂p1 = 0 −→
2 ∂P2
∂p1 =0
∂P1 ∂P1 ∂P1
5. [Q2 , P1 ] = ∂p1 + ∂p2 = 0 −→ ∂p2 = − ∂P 1 1
∂p1 = − 2q1
∂P2 ∂P2 ∂P2 ∂P2
6. [Q2 , P2 ] = ∂p1 + ∂p2 = 1 −→ ∂p2 =1− ∂p1 =1
From the partial derivatives I can glimpse the general form

P1 = p1 − p2 + f (q1 , q2 )

2q1
P2 = p2 + g(q1 , q2 )


where f , g are undetermined functions of q1 , q2 . The last Poisson bracket,

∂P1 ∂P2 ∂P1 ∂P2 ∂P1 ∂P2 ∂P1 ∂P2
[P1 , P2 ] = − + −
∂q1 ∂p1 ∂p1 ∂q1 ∂q2 ∂p2 ∂p2 ∂q2
1 ∂g ∂f 1 ∂g
=− + + =0
2q1 ∂q1 ∂q2 2q1 ∂q2

6
is the only constraint on f and g. Now consider the Hamiltonian
 2
p1 − p2
H(q1 , q2 , p1 , p2 ) = + p2 + (q1 + q2 )2
2q1

I think f = 0, g = (q1 + q2 )2 is a good choice, and it satisfies the last Poisson bracket condition as well.
So my canonical transformation is
Q1 = q12




Q2 = q1 + q2


p1 − p2 (4.1)
P1 =



 2q1
P2 = p2 + (q1 + q2 )2


and my new Hamiltonian is

H(P1 , P2 ) = P12 + P2
Well, Q1 and Q2 are ignorable, so P1 and P2 are conserved. The other two canonical equations are
trivially solved, 
∂H
Q̇1 = = 2P1

 (
∂P1 Q1 (t) = 2P1 t + Q1 (0)
−→
 ∂H Q2 (t) = t + Q2 (0)
Q̇2 =
 =1
∂P2
To find the initial conditions for the new variables, I evaluate the transformations (4.1) at t = 0, which
is the same as adding a subscript 0. Then I iteratively solve for the original variables,
 r
p p10 − p20 2
q1 = Q1 = t + q10





 2q 10
 r
q = Q − q = q + q + t − p10 − p20 t + q 2



2 2 1 10 20 10
2q10
p2 = P2 − (q1 + q2 )2 = p20 + (q10 + q20 )2 − (q10 + q20 + t)2





 r
p10 − p20 p10 − p20


p1 = 2P1 q1 + p2 =

 2 + p + (q + q )2 − (q + q + t)2
t + q10 20 10 20 10 20
q10 q10

I’ve double-checked that the initial conditions are self-consistent. I have no idea what physical system
this describes.