Advanced Classical Mechanics

Hannah Bossi, hannah.bossi@yale.edu

December 2018

1 Preamble
The document which follows is a study guide which I compiled in December of 2018 for the final
exam of the Advanced Classical Mechanics course at Yale University. All mistakes or typos present
in this document are my own. If you would like to correct any typos or add to the document, send
me an email and I’d be happy to provide you with the raw files. Feel free to distribute this guide as
you see fit. Use it well. - Hannah

2 Newtonian Dynamics and Kinematics

2.1 Newton’s Laws and Their Galilean Invariance
Newton’s Laws (which are invariant under Galilean Transformations)
1. A body remains at rest or moves with a uniform velocity unless acted upon bu an outside
force.
2. F~ = m~a = d~
p
dt where p~ = m~v

3. Action-Reaction Law: F~12 = −F~21

2.2 Single Particle Conservation Law

• Linear Momentum: If F~ = 0, then p~˙ = 0 (linear momentum is conserved).

~, N
• Angular Momentum: If torque,N ~˙ = 0 (angular momentum is conserved).
~ = 0 then L

• Energy: If the force is conservative, E = T + V = const

DerivationR For the Work-Kinetic Energy Theorem: The work done between points R 2 1 and 2 is defined
2 R2
as W12 = 1 F · d~s = 1 m~v˙ · d~s. Applying that d~s = ~v dt, we see this becomes 1 m~v˙ (~v dt). This can
R2
be alternatively written as 1 12 m dt
d
(~v 2 )dt = 21 mv 2 = KE

2.3 System of Particles: Conservation Laws

In general for a system of particles, the center of mass moves as if the mass of the system is at the
center of mass.
• Linear Momentum: If external force F~e = 0, then p~ = constant. (F~e = p~˙)

~˙ = 0. (N~ e = L)
• Angular Momentum: If external torque,N~ e , N~ e = 0 then L ~˙

1
 • Energy: If both internal and external forces are conservative, then E = constant.
The general procedure for doing these sorts of problems is as follows
1. Decompose the force on particle i into internal and external forces.
2. Make cancellations/assumptions regarding the internal forces that results in a conservation
law.
Conditions for Conservative Force
• W12 between any two points is path independent.
• Force can be written as the gradient of some scalar function of position. (F = −∇V (r))

2.4 Accelerated Coordinate Systems

All Newton’s laws are valid only in intertial frames - frames which are neither accelerating nor
rotating. All rotations have a common fixed point which is the origin. We can think of two successive
rotations as a combined rotation vector where the individual ones add. This also gives us the idea
that infinitessimal rotations commute. (It doesn’t matter which order they are performed in, the
result is still the same.)
~ ~ ~ for a generic vector A
~
• ( ddtA )space = ( ddtA )body + ω
~ ×A
d~
ω
• dt ω×ω
is the same in both frames as (~ ~) = 0

ω × ~vbody + ( d~
• ~ainertial = ~abody + 2~ ω
dt )body × ~ ~ × (~
r+ω ω × ~r)

• ~v = ω
~ × ~r
• mr̈ = F + 2mṙ × ω + m(ω × r) × ω = F + Fcor + Fcf

2.4.1 Motion on the Earth

To first order, say ~r = r0 + r1 where r0 is the zeroth order contributions (where ω
~ = 0) and r1 is the
1st order contributions.

• r¨0 = ~g (solve for r0 (t))

• r~0 (t) = − 21 gt2 ẑ (now plug this in)
• to first order in ω
ω × r~˙0
• r¨1 = −2~

• r¨1 = 2gt~
ω × ẑ
• r1 = 13 gt3 ωSin(θ)ŷ

2
2.5 Particle on a Scale
~e = ~ae = 0. We know that the normal force will match the force of gravity such that N
V ~ = −mg.
Recall the general force equation above

0 = F + 2mṙ × ω + m(ω × r) × ω

Here our F = −GmMe r̂ ~ and our Coriolis force is zero resulting in

+N
r3
~

−GmMe r̂ ~ + m(ω × r) × ω = −GmMe r̂ − m~g + m(ω × r) × ω

0= +N
~r3 ~r3
This results in an effective acceleration of gravity as
−GMe r̂
~g = + (ω × r) × ω (1)
~r3
where the second term depends on the colatitude.

2.5.1 Holonomic Constraints

Constraints which can be written as equations fj (X1 , X2 , ....X3N , t) = 0 j = 1,..... k. The net work
done by the forces of constraint is zero. There are many difficulties with constraints are...
1. Not all constraints are independent
• To address this, use generalized coordinates.

2. Forces of constraint are unknown (want a mechanism to get rid of forces of constraint)
• eliminate forces of constraint from the formalism

3 Lagrangian Dynamics
Virtual displacements are defined as infinitesimal instantaneous displacement of the coordinates
consistent with the constraints. We freeze the constraints at time t and consider the displacement
δxi consistent with these constraints at time t.
A very useful theorem is D’Lambert’s Principle. This principle says that The net virtual work
P P (a) (a)
done by the forces of constraint is zero. fi δxi = (Fi − p˙i ) = 0. Where Fi represents the
applied force.The latter form of expression is superior as it gives the advantage that the forces of
constraint have disappeared from the problem. This characterizes how the forces of constraint act
during the motion of the particle. I will now provide an overview of how to arrive at the Euler-
Lagrange equations from this principle. We begin with
X (a)
(Fi − p˙i ) = 0

If all xi are independent → Fi = p˙i . We can therefore write

X X (a) ∂xi
[ (Fi − p˙i ) ]δqλ = 0
∂qλ
We will define Qλ as a generalized force conjugate to qλ by
X ∂xi
Qλ = Fi
∂qλ

3
We can make our equation simpler in terms of this generalized force Qλ .
X
[Qλ − Rλ ]δqλ

But what is Rλ ? Let’s figure this out

X ∂xi X ∂xi X d ∂xi
p˙i = mi ẍi = mi ( ẋi ) (2)
∂qλ ∂qλ dt ∂qλ
With the chain rule we see that this is equivalent to
X d ∂xi X d ∂xi
mi (ẋi )− mi ẋi ( )
dt ∂qλ dt ∂qλ
∂T d ∂T
Algebra shows that the second term is equivalent to ∂qλ
, the first term is equivalent to dt ( q˙λ ).
Plugging this in for Rλ we see that
X X d ∂T ∂T
[Qλ − Rλ ]δqλ = [Qλ − ( ( )− )]δqλ
dt q˙λ ∂qλ
Now, using the fact that all qλ are independent
d ∂T ∂T
( ( )− ) = 0.
dt q˙λ ∂qλ
If the forces are conservative we see that
−∂V
Qλ =
∂qλ
Since V is velocity independent, this becomes the Euler-Lagrange equations which are given by
d ∂L ∂L
− =0
dt ∂ q˙σ ∂qσ
where L = T − V .

3.1 Example for Lagrangian Mechanics

3.2 Calculus of Variations
A single variational principle which implies the whole set of Lagrange equations. Let x be an
independent variable defined over the interval [x1 , x2 ] and let y(x) be some differentiable function of
dy
x defined on this interval, such that y 0 (x) = dx . Define φ = φ[y(x), y 0 (x), x]. In a two dimensional
space
ds = [(dx)2 + (dy)2 ] = [1 + (y 0 )2 ]1/2 dx
define
φ = [1 + (y 0 )2 ]1/2
since φ has no explicit dependence on y, the Euler-Lagrange equation takes the form
d y0
=0
dx [1 + (y 0 )2 ]1/2
A trivial integration yields the equation for the straight line y = mx + b.We can also use the calculus
of variations in order to prove Lagrange’s equations from the calculus of variations. Begin with the
action which we will call I.
Z Z Z X
∂L ∂L
δI = δ L(q, q̇, t) = δL(q, q̇, t) = [ δqλ + δ q˙λ ]dt
∂qλ ∂ q˙λ

4
 d ∂L
R
Now in an aside we note that δ q˙λ = dt δqλ . Now, we use integration by parts to evaluate ∂ q˙λ δ q˙λ .
We have Z Z Z
∂L d ∂L ∂L d ∂L
δ q˙λ = − ( )δqλ dt + ( )δqλ |tt21 = − ( )δqλ dt
∂ q˙λ dt ∂ q˙λ ∂ q˙λ dt ∂ q˙λ
From Hamilton’s principle of least action we know that δI = 0. Plugging this into our overall
equation from above we see that
Z X
∂L d ∂L
δI = [ − ( )]δqλ dt = 0
∂qλ dt ∂ q˙λ
This gives us the Euler-Lagrange Equations as
∂L d ∂L
− ( )=0
∂qλ dt ∂ q˙λ

3.3 Hamilton’s Principle of Least Action

Hamilton’s Principle of Least action says that the motion between t1 and t2 is such that the action
is stationary for any δq(t) with δq(t1 ) = δq(t2 ) = 0. In other words δI = 0 for any δq(t) such that
δq(t1 ) = δq(t2 ) = 0.

3.4 Method of Lagrange Multipliers

Lagrange multipliers are a particularly useful way of solving problems with forces of constraint. The
book defines the general method of this as...
• Suppose not all qλ are independent
• We have k constraints fj (q1 , ...., qn , t) = 0. Find an extremum of a function under the con-
straint.
• f(x,y,z) with constraint g(x,y,z) = 0 (method of Lagrange multipliers). We introduce a con-
straint λ and consider the extremum of f − λg.
Trevor also came up with a step-by-step guide of this which is particularly useful.
1. Write down L in terms of necessary coordinates and unnecessary ones that will include the
constraint. L(q1 ...., qN , q˙1 ...qN
˙ ).
2. Write down the constraint equation: i.e. q2 = a
3. Write down the derivative of that δq2 = 0 (rearrange this to equal zero if needed)
Rt
4. Write δI + t12 λ(whateverequalszero)dt = 0. Do this for however many constraint equations.
d ∂L ∂L
5. Then write down dt ( ∂ q˙i ) − ∂qi = (whatever multiplies δqi = 0, 0 if it is not there)
• These are the constraint equations.
• This could also be viewed as the constraint force is the λ times (whatever equals 0) term
written with δqi
6. Then solve for whatever the problem is.
7. Consider a virtual displacement in the direction of the constraint, then write δW = Qdirection δdirection
(i.e. Qr δr)
• Remember sign, this often includes a negative.

5
3.4.1 One Cylinder Rolling on Another
See page 75 in Fetter and Walecka for this example. One uniform cylinder of mass m and radius R2
slipping on another cylinder of radius R1 . We can write the kinetic energy as
1 2 1 1
T = m(ṙ2 + r2 θ˙1 ) + ( mR22 )θ˙22
2 2 2
Now consider the potential
V = mgrcos(θ1 )
We now consider our equations of constraint, we have two of them
r = R1 + R2
the second is a little bit more complicated
R1 θ1 = R2 (θ2 − θ1 )
Now take the variation of the first constraint and multiply by λ1 and the variation of the second
constraint and multiply that by λ2 . This yields
λ1 δr = 0
λ1 [(R1 + R2 )δθ1 − R2 δθ2 ] = 0
Now use the return the Lagrangian, which is
1 2 1 1
L= m(ṙ2 + r2 θ˙1 ) + ( mR22 )θ˙22 − mgrcos(θ1 )
2 2 2
We then solve Lagrange’s equations for the variables r, θ1 , θ2 and we get the following
−mrθ˙12 + mgcos(θ1 ) = λ1
d
mr2 θ˙1 − mgrsinθ1 = (R1 + R2 )λ2
dt
1 2¨
R θ2 = −R2 λ2
2 2
Apply the constraints to the first of our Lagrange equations to get
−m(R1 + R2 )θ˙12 + mgcos(θ1 ) = λ1
similarly, applying our constraint equations to the third Euler-Lagrange equations yields
1
− m(R1 + R2 )θ¨1 = λ2
2
Now take the second and third Lagrange equations and set them equal using λ2 .
1
m(R1 + R2 )2 θ¨1 − mg(R1 + R2 )sinθ1 = (R1 + R2 )λ2 = − m(R1 + R2 )2 θ¨1
2
Simplifying this yields
3
(R1 + R2 )θ¨1 − gsinθ1 = 0
2
Multiply both sides by θ˙1 and integrating we get
4
(R1 + R2 )θ˙12 = g(1 − cosθ1 )
3
Now, plug back into earlier equation to get
1
N (θ1 ) = mg(7cosθ1 − 4)
3

6
3.5 Is h the Energy?
P
Define h as h = pi q˙i − L. This is numerically equivalent to the Hamiltonian, but it is not exactly
the same. h = h(q, q̇, t). Is h the energy like the Hamiltonian? We will discuss this. The kinetic
energy is given
X1
T = mi ẋ2
2
where xi = xi (q, t). So we see that ẋ is given as
X ∂xi ∂xi
ẋi = q˙λ +
∂qλ ∂t
using the above, we can write
X1 X ∂xi ∂xi X ∂xi ∂xi
T = mi ( q˙λ + )( q˙µ + )
2 ∂qλ ∂t ∂qµ ∂t
We see that this will result in T = T2 + T1 + T0 where Tn is a homogenous equation of order n in q̇.
These terms are given by
1 ∂xi ∂xi
T2 = mi q˙λ q˙µ
2 ∂qλ ∂qµ
∂xi
T1 = mi q˙λ ẋi
∂qλ
1 ∂xi 2
T0 = mi ( )
2 ∂t
The idea here is that we can write the Lagrangian in terms of these as L = L2 + L1 +
PL0 .∂f
We also have Euler’s Theorem which says if f is homogenous of degree n, then xi ∂xi = nf .
We can apply this to these scenarios. Return to our definition of h.
X
h= pi q˙i − L
Using Hamilton’s equations to simplify the first term
X X ∂L
pi q˙i = q˙i
∂ q˙i
use now that L = L2 + L1 + L0 to split this up.
X X ∂L2 X ∂L1 X ∂L0
pi q˙i = q˙i + q˙i + q˙i
∂ q˙i ∂ q˙i ∂ q˙i
now use Euler’s theorem. X
pi q˙i = 2L2 + L1 + 0L0
now returning to h we see
h = 2L2 + L1 − (L2 + L1 + L0 ) = L2 − L0
So now the question that we have to ask is when is this equal to E? We see that if the transformations
do not depend on time, xi = xi (q1 .q2 ...) then only T2 exists and the potential energy is velocity
independent, then L2 = T2 = T . Since the potential is velocity independent, L0 = −V . Therefore,
h = L2 − L0 = T + V = E
In conclusion, we have the following statements.
1. If the Lagrangian does not depend on time, then the hamiltonian is a constant of the motion.
∂H
∂t = 0.

2. If there are only time independent potentials with time independent constraints, then the
hamiltonian is not only a constant of the motion, but is also the total energy.

7
3.6 Symmetry Principles and Conserved Quantities: Noether’s Theorem
If qσ is a cyclic coordinate, then pσ is constant.
• Noether’s Theorem: There is a constant of the motion associated with a continuous symmetry.
We can prove this in the Lagrangian formalism. Consider a transformation near the identity.
0
qi = qi + G

take the time derivative of this which is

X ∂G
q˙i = q˙i + 
0
q˙j
∂qj

so we know that
X ∂G
δ q˙j =  q˙j
∂qj
Now consider the variation in the Lagrangian.
X ∂L X ∂L
δL = δqj + δ q˙j
∂qj ∂ q˙j

plug in δ q˙j from above.

X ∂L X ∂L ∂G
δL = G +  q˙j
∂qj ∂ q˙j ∂qj
Substitute to bring momenta into your equation
X ∂G X X d
δL = p˙j G + pj 
q˙j =  (pj Gj )
∂qj dt
d
P
For a conserved quantity, δL = 0, so dt (pj Gj ) = 0, so pj Gj is conserved.

4 Two-body central force problem

4.1 The differential equation for the orbit
l2
• Vef f = V (r) + 2µr 2

We must make the following two assumptions about V.

1
1. V falls slower than r2 when r → inf
1
2. V diverges slower than r2 when r → 0
3. Take V to be attractive.
We can classify orbits as the following.
1. For E = Vef f (rmin ) the orbit is a circle at r = rmin .
2. For Vef f (rmin ) ≤ E ≤ 0, the motion is bounded between the turning points r1 and r2 .
3. For E > 0 only one turning point r3 and the motion is unbounded.
Differential Equation for the Orbit

8
 Figure 1: plot of effective potential

d2 u
• dθ 2 + u(θ) = − l2 uµ2 (θ) F (r)

We can derive this relatively simply. First, consider r = r(φ). Taking the time derivative we see
that
∂r ∂r l
ṙ = φ̇ =
∂φ ∂φ µr2
Now consider the equation for the energy, plug our ṙ into this
1 2 l 1 ∂r l 2 l
E= µṙ + + V = µ( ) + +V
2 2µr2 2 ∂φ µr2 2µr2
Now define a new quantity U such that U = 1r . Now
du 1 dr
=− 2
dφ r dφ
so the energy equation becomes
1 l2 du 2 l2 2
E= ( ) + u +V
2 µ dφ 2µ
Differentiate both sides with respect to u
1 l2 du du 2 dφ l2 dV
0= 2 ( ) + u+
2 µ dφ dφ du µ du
which gives the orbit equation
d2 u µ dV
2
+u=− 2 (3)
dφ l du
Note that the solution to the differential orbit equation u(φ) has the property that if u(φ) is a
du
solution, then u(−φ) will also be a solution. At a turning point dφ = 0. So u(φ = 0) = u(−φ = 0).
Both solutions have the same initial conditions so the solutions are the same as each other (2nd order
differential equation with the same initial conditions will have the same solution). so u(φ) = u(−φ).
In conclusion the orbit is symmetric about the turning point. Therefore, we choose the polar axis
to go through the turning point.
Bertrand’s Theorem:The only central forces that result in closed orbits for all bound particles
are the inverse square law and Hooke’s Law.

9
4.2 Kepler Problem
−k
• v= r = −ku
d2 u µk
• dθ 2 + u(θ) = l2

• You end up getting the following equation for r which is the equation of a conic section with
eccentricity e.
1
• r= µk
(1+ecos(θ))
l2

Four possible cases for the eccentricity e.

1. For E > 0, e > 1 → hyperbola (scattering trajectory)
2. For E = 0, e = 1 → parabola (scattering trajectory)
−mk2
3. For 2l2 < E < 0, 0 < e < 1 → ellipse (bound orbit)
−mk2
4. For E = 2l2 , e=0 → circle (bound orbit)
The Kepler Problem has 7 constants of the motion, 5 are independent. They are listed below.
1. Energy
2. Angular Momentum
3. Runge-Lenz Vector A (dependent on 2, AL̇ = 0
4. Center or Mass motion (3 different coordinates)
5. Motion is confined to a plane. (dependent on 4)

4.3 Derivation of the Virial Theorem

The fundamental equations of motion are

p˙i = Fi

We are interested in the quantity X

Gi = pi · ri
Take the time derivative of this
dG X X
= p˙i · ri + r˙i · pi
dt
substituting our fundamental equations of motion we have
dG X X
= Fi · ri + r˙i · mr˙i
dt
Simplifying the second term we see
X X
r˙i · mr˙i = mvi2 = 2T

this results in
dG X
= Fi · ri + 2T
dt

10
now, we need to integrate. Multiply both sides by dt.
Z Z X
dG = ( Fi · ri + 2T )dt

Choose the integration bounds on the G term to represent one period. Therefore, the left hand side
of this equation goes to zero.
X
0= Fi · ri + 2T
Then we can solve to get the Virial Theorem
1X
T =− Fi · ri
2
To get the more well known version of the virial theorem, substitute − ∂V
∂r for the force. Assume the
potential follows V = rn+1 choose n = −2 to get the well known result. This is because so many
forces are inverse square law forces.

4.4 Runge-Lenz Vector

The main important thing about the Runge-Lenz Vector is that it is the fifth conserved quantity in
the Kepler problem. We will show that it is conserved here. By Newton’s second law we have
k
F~ = p~˙ = − 2 r̂
r
Now consider
d ~ = p~˙ × L ~˙
~ + p~ × L
p × L)
(~
dt
however the second term goes to zero as L ~ is conserved. Now plug in − k2 r̂ for p~˙ and L
~ = ~r × p~ =
r
˙
~r × µ~r. These simplifications yield
d ~ = − k r̂ × (~r × µ~r˙ ) = − µk r̂ × (~r × ~r˙ )
p × L)
(~
dt r2 r2
Use the vector identity
~ × (B
A ~ × C)
~ = (A
~ · C)
~ B~ − (A
~ · B)
~ C~

to simplify this to be

d ~ = − µk [(r̂ · ~r˙ )~r − (r̂ · ~r)~r˙ ] = − µk [ṙ~r − r~r˙ ] = µk d [ ~r ]

p × L)
(~
dt r2 r2 dt r
we see that
d ~ − µk ~r ) = 0
p×L
(~
dt r
Therefore the Runge-Lenz vector, given by

~ − µk ~r
~ = p~ × L
A
r

is a constant of the motion. What is the direction of A? ~ We choose this to be the polar axis.
Therefore it is in the direction of the turning point with magnitude A = µke. Also worth noting is
that this is an over-integrable system.

11
4.5 Classical Scattering
Assume we have a beam with intensity I that is colliding with the target. We want to ask the
question: How many particles are scattered in each direction? This is calculating a differential cross
section. In general this is given by
number of particles
I=
(unit area perpendicular to beam)(unit time)

In general, this follows the image in the figure below. The impact parameter, which we will refer
to as b (but is given by s in the figure), measures the distance from the axis. The particles are
scattered into a solid spatial angle dΩ with change of impact parameter db. The general formula for

the differential cross section is given as

number of particles scattered into dΩ
σ(Ω)dΩ =
(unit time)(I)

The full explicit equation for the differential cross section goes as

b db
σ=
sinθ dθ
√
Some useful relations are θ = θ(E, b),l = µvb = b 2mE (as E = 21 mv 2 ). In general the process for
solving these problems is as follows.
1. Given θ = θ(E, b)

2. Invert to find b = b(θ, E)

db
3. This lets you find dθ

However, sometimes it is not that easy to find θ = θ(E, b). If you are not given this, use the following
formula. Z ∞
dr
θ(b) = π − q
rm r 2 ( 2mE 2mV 1
l2 − l2 − r 2 )

A cool thing about scattering is that the procedures to describe scattering are the same in classical
mechanics as quantum mechanics.

12
5 Small Oscillations
We will make the following assumptions about our system
1. Assume that we have n independent generalized coordinates (q1 , ...qN ).
2. Assume that we have a conservative and time-independent potential V = V (q1 , ...qN )
3. Assume that the constraints do not have any explicit time dependence.

4. Assume that the transformation equations are time-independent.

d ∂T ∂T ∂V
Lagrangian is time-independent so dt ( ∂ q˙λ ) − ∂qλ
= Qλ = − ∂q λ
. We see here that the potential
determines the generalized forces. Static equilibrium is characterized by the condition
∂V
q˙λ = q¨λ = 0, =0
∂qλ
So we are looking for the stationary points of V vs. all qλ . For stable equilibrium, we are looking
for local mins of V vs. all qλ at the same time. At an equilibrium point, expand V around qλ0 . Note
that ηλ = qλ − qλ0 . Expanding this we see
X ∂V 1 X ∂2V
V = V (qλ ) + |0 ηλ + |0 ηλ ηµ + ...
∂r 2 ∂qµ ∂qλ

But this is a minimum so the first derivative term will be zero.

1 X ∂2V
V = V (qλ ) + |0 ηλ ηµ + ...
2 ∂qµ ∂qλ

From our definition of ηλ we see that η˙λ = q˙λ . We can use this to construct the kinetic energy of
the system
1X ∂xi ∂xi 1
T = mi ( )q˙λ q˙µ = mλ,µ q˙λ q˙µ
2 ∂qλ ∂qµ 2
P ∂xi ∂xi
where mλ,µ = mi ( ∂q λ ∂qµ
). Using η˙λ = q˙λ we see

1
T = mλ,µ η˙λ η˙µ
2
Now we can write out the Lagrangian as we have the kinetic and potential energies.
1 T 1
L=T −V = η̇ M η̇ − η T V η
2 2
These are n coupled second order linear differential equations. We need to uncouple these equations.
We do this by making a linear transformation to a set of coordinates where there are uncoupled.
η → ζ in general assume η = Rζ where R is a matrix of constants. Let’s see how the Lagrangian
transforms in these equations first, η̇ = Rζ̇, η T = ζ T RT , η˙T = ζ˙T RT . Using these, we redefine the
Lagrangian as
1 1
L = ζ˙T (RT M R)ζ̇ − ζ T (RT V R)ζ
2 2
Call Ṽ = (RT V R) and M̃ = RT M R. Then the Lagrangian becomes
1 ˙T 1
L= ζ M̃ ζ̇ − ζ T Ṽ ζ
2 2

13
Theorem: Assume that any M, V are real and symmetric matrices and M is positive definite (all
eigenvalues are positive). Then there is a congruence transformation that diagonalizes M and V
simultaneously i.e. there is a matrix R such that RT M R = I and RT V R is a diagonal matrix where
the diagonal entries are the eigenvalues. √
1
Proof of this Theorem: Since M is positive definite, so is M −1 . Therefore, M − 2 = M −1 is
a real matrix. There is matrix M such that M = A (diagonal matrix of γ’s)A−1 where all γ’s are
1
positive. Therefore, we can also create a matrix M − 2 = A(diagonal matrix of γ 1/2 ’s)A−1 . Now
apply M −1/2 as a congruence transformation. We get
0 1 1 1 1
M = (M − 2 )T M (M − 2 ) = M − 2 M M − 2 = I
0 1 1 0
Then V = M − 2 V M − 2 where V is real and symmetric. This can be diagonalized by an orthogonal
0
transformation O. Such that OT V O is diagonal. OT IO = O−1 O = I. We will now define R =
M −1/2 O, RT = OT (M 1/2 )T .
The basic method for solving these problems is as follows.
1. Construct M(mass matrix), V(potential matrix)

2. Find the frequencies from det(V − ω 2 M ) = 0.

3. For each of the frequenciesωj , find the column vector rj from (V − ω 2 M )rj . You can then use
these column vectors to construct the R matrix for η = ζR.
4. Transform to ζ coordinates using ζ = R−1 η. (Note that when you do this varies from problem
to problem. Usually there will be an indicator of when to do this.)
5. Diagonalize V and M using R.
6. Transform back to the original coordinates using η = ζR.

6 Special Relativity
6.1 The Lorentz Transformation
Galilean Transformations take the form of
0 0
~r = ~r − ~v t, t = t

In special relativity, we consider Lorentz transformations which take the form

0 ~v
~r = ~r + (~v · ~r) (γ − 1) − γ~v t
v2
0 ~v · ~r
t = γ(t − )
c2
Note that the postulates of special relativity require that speed of light and Maxwell’s equations are
the same in all reference frames. The quantity c2 dt2 − (dr2 ) = c2 dτ 2 is also invariant under Lorentz
transformations. (Where τ is the proper time.) We can define the mminkowski metric as follows.
 
1 0 0 0
0 −1 0 0
η= 0 0 −1 0 


0 0 0 −1

14
Be very careful, as this is defined differently in gravitational studies. The Minkowski metric is
invariant under Lorentz transformations.
η = ΛT ηΛ
The matrix form of the Lorentz transformations is given as
−γ vc 0 0
 
γ
−γ v γ 0 0
Λ= c 
 0 0 1 0
0 0 0 1
We can also relate the time to the proper time which is the time measured in the particle’s rest
frame by dt = γdτ , where γ = √ 1 2 and β = vc . Now that we have four dimensions (3 spatial and
1−β
time) we need to create a new quantity for the velocity which we will call the 4 velocity. This is
given as
dxν dt dxi dt
Uν = = (c , ) = (γc, γv i )
dτ dτ dt dτ
where the four momentum is given as pν = muν . Note that if u2 = c2 > 0 the vector is said to be
timelike.

6.2 Relativisic Dynamics

Newtonian dynamics are not invariant under Lorentz transformations, so we must also express our
force differently. The relativistic force is defined as
dpν
fν =
dτ
where the relativistic force is related to the usual force by
fi
Fi =
γ
Our energy and kinetic energy must also change. The Energy is defined as E = γmc2 which also
obeys
E 2 = p2 c2 + m2 c4
where the kinetic energy is T = (γ − 1)mc2 .

6.3 The Lagrangain Formulation

This clearly requires that we redefine our lagrangian, which in special relativity is given as
r
2 v2
L = −mc 1 − 2 − V
c
in order to have h be the total energy under the conditions described earlier. We also have the
covariant description (meaning it will hold in any Lorentz frame) which begins by having defining
Lagrange’s functions which are given by
1
Λ= muν uν
2
We can now define the Euler-Lagrange equations as
∂ ∂Λ ∂Λ
( )− =0
∂τ ∂ x˙ν ∂xν

15
7 Rigid-Body Motion
A rigid body is a body for which |r~ij | = cij = const for any two points on the rigid body. There
are 6 degrees of freedom here, 3 for the center of mass and 3 for the angles of orientation of the
rigid body. We will now derive a few important equations in rigid body theory. The first being the
kinetic energy. Begin by using the normal definition of kinetic energy while also using the fact that
~v = w~ × ~r.
1X 1X
T = m(ω~λ × r~λ )2 = m(ω~λ × r~λ ) · (ω~λ × r~λ )
2 2
Now we will employ the identity
~ × B)
(A ~ · (C
~ × D)
~ = (A
~ · C)
~ · (B
~ · D)
~ − (A
~ · D)
~ · (B
~ · C)
~

This simplifies the kinetic energy to

1X
T = m[ωλ2 rλ2 − (ω~λ · r~λ ) · (ω~λ · r~λ )]
2
introduce another sum over i, j
1 XX
T = [mλ ω 2 rλ2 δij − mλ Xλi Xλj )ωi ωj
2
which simplifies using the definition of the moment of inertia tensor
1X 1
T = Iij ωi ωj = ω~ ·I ·ω
~
2 2
Now moving onto the angular momentum. We begin by defining this in the usual way
X X
~ =
L ~r × p~ = mλ (r~λ × v~λ )

now use v = ω × r X
~ =
L mλ (r~λ × (ω × r~λ ))
using the BAC-CAB vector identity this becomes
XX X
Li = mλ [rλ2 δij − xλi xλj ]ωj = Iij ωj

~ =I ·ω
so L ~.

7.1 Inertia Tensor, Principal Axis

Note that the moment of inertia is a second rank symmetric tensor that is constant in the body
frame. We will now derive the parallel axis theorem. Consider a transformation of coordinates from
~r = r~0 − ~a. As we want the moment of inertia in the original coordinates. Using our definition of
the moment of inertia tensor we write
Z
Iij = d~rρ(~r)(r2 δij − xi xj )

0
where r is the center of mass. Transform this to the center of mass coordinates
Z
0 0 0 0
Iij = d~r ρ(~r )[(r~0 − ~a)2 δij − (xi − a)(xj − a)]

16
expanding this we see
Z Z Z
0 0 0 0 0 0 0 0 0 0
2 2
Iij = d~r ρ(~r )(r δij − xi xj ) + d~r ρ(a δij − ai aj ) + d~r ρ(−2~r · ~a + ai xj + aj xi )

The last term goes to zero and we are left with the parallel axis theorem

Iij = ICM + M (a2 δij − ai aj )

We can also define a principal frame which is a frame in which the moment of inertia tensor is
diagonal.

7.2 Euler’s Equations

~ ~ We begin by
Euler’s equations combine the relation ddtL = ~τ e and ( dL dL
~ × L.
dt )space = ( dt )body + ω
taking the principal frame to be the body frame noting that Li = Ii ωi where ωi are the components
of ω in the body frame.
dLi ~ i = τie
( ω × L)
)body + (~
dt
substitute in Li = Ii ωi . Use i = 1 as an example.
dω1
I1 + ω2 L3 − ω3 L2 = τ1e
dt
now substitute L3 = I3 ω3 to get Euler’s equations
dω1
I1 = ω2 ω3 (I2 − I3 ) + τ1e
dt
the other two coordinates have Euler equations of
dω2
I2 = ω3 ω1 (I3 − I1 ) + τ2e
dt
dω3
I3 = ω2 ω1 (I1 − I2 ) + τ3e
dt
The orientation of the body frame is specified by a set of three angles these are Euler angles. eˆ0i
(intertial frame) → eˆi (body frame). This transformations is broken up into the product of three
rotations which in summary are given by
1. Rotation by α around z.
2. Rotation by β around new y.
3. Rotation by γ around new Z.

8 Hamiltonian Dynamics and Transformation Theory

It is possible to recast the equations of motion so that they are invariant (in form) under a larger
class of transformations. We will define the Hamiltonian as H = H(q, p, t) where
X
H= pi q˙i − L

Where we now have twice the number of equations of motion. These are given by
∂H
q˙i =
∂pi

17
 ∂L ∂H
p˙i = =−
∂qi ∂qi
We can derive these from Hamilton’s principle. Recall the original form of this was
Z
δI = δ Ldt = 0

Now substituting in our definition for the the for the Hamiltonian
Z X
δ ( pi q˙i − H) = 0

Now vary the term in parentheses

Z X
∂H ∂H
δI = (q˙i δpi + pi δ q˙i − δqi − δpi )dt = 0
∂qi ∂pi

but we want the second term to have a p˙i in it. Use integration by parts (with term canceling based
on definition of t1 and t2 .
Z Z X Z X
X d
dt pi δ q˙i = pi (δqi )dt = − p˙i δqi dt
dt
Now plugging this into the above equation we see that
Z X
∂H ∂H
δI = (q˙i δpi − p˙i δqi − δqi − δpi )dt = 0
∂qi ∂pi
under this formulation qi and pi are independent so we see that we can split these up.
Z X
∂H
(q˙i δpi − δpi )dt
∂pi
Z X
∂H
(−pi δ q˙i − δqi )dt = 0
∂qi
Hamilton’s equations naturally fall out of these.
∂H
q˙i =
∂pi
∂H
p˙i = −
∂qi

8.1 Canonical (Symplectic) Transformations

Define a matrix η such that  
q
η=
p
Also define J such that is it a block matrix with 0’s on the diagonal block and 1’s and -1’s on the
upper and lower off diagonal blocks respectively. This matrix has the following properties
1. J 2 = −I
2. JJ T = 1
3. J T = −J

18
Hamilton’s equations can be expressed in this formalism as
∂H
η̇ = J
∂η
We will transform to a new set of coordinates such that Q = Q(q, p, t) and P = P (q, p, t). These
will keep the form of Hamilton’s equations, but there will be a new Hamiltonian K such that
∂K
Q̇ =
∂P
∂K
Ṗ = −
∂Q
Where the new Hamiltonian K is defined by
X X ∂F
pi q˙i − H = Pi Q̇i − K +
∂t
where F is referred to as the generating function of the canonical transformation. We will discuss this
further in the following section. We also have a matrix method to determine if the transformation
is canonical. first consider the matrix M, which is defined as
∂ζi
Mij =
∂ηj
this is also called the Jacobi Matrix. We see that if the following equation holds, the transformation
is canonical.
M JM T = J
This is derived by
∂H ∂H
ζ̇ = M η̇ = M J = M JM T
∂η ∂ζ

8.2 Generating Functions

There are four different types of generating functions. Consider F1 such that F1 = F1 (q, Q̇, t). The
above equation then becomes
X X X ∂F X ∂F ∂F
pi q˙i − H = Pi Q̇i − K + q˙i + Q̇i +
∂qi ∂Qi ∂t
Hence in order for this to hold we must have
∂F1
pi =
∂qi
∂F1
Pi = −
∂Qi
Leaving
∂F1
K=H+
∂t
Consider F2 = F2 (q, P, t)
∂F2
pi =
∂qi
∂F2
Qi =
∂Pi

19
Now for F3 = F3 (p, Q, t)
∂F3
Pi = −
∂Qi
∂F3
qi = −
∂pi
The fourth type is given by F4 = F4 (p, P, t). A canonical transformation in the vicinity of the
identity generated by G is given by
∂G
δη = J
∂η

8.3 Poisson Brackets

The Poisson bracket is defined as
n o X ∂ω ∂λ ∂ω ∂λ
ω, λ = ( − )
∂qi ∂pi ∂pi qi
This can be written more compactly in the symplectic notation as
n o δA δB
A, B = ( )T J
δη δη
n o
The poisson bracket is invariant under a canonical transformation. Also if A, H , then A is a
n o
constant of the motion. We can also prove that if A and B are constants of the motion then A, B
is also a constant of the motion. This is proven using the Jacobi identity which says
n n oo n n oo n n oo
A, B, C + B, C, A + C, A, B =0

Substituting this in we see that the proof is relatively simple

n n oo n n oo n n oo n n oo n n oo
H, A, B + A, B, H + B, A, H = H, A, B + 0 + 0 = H, A, B =0

We now move on to Liouville’s Theorem which says that the classical phase space distribution
function is constant along the classical trajectory. Mathematically this is stated as
∂ρ n o
= H, ρ
∂t
Proof of Liouville’s Theorem: First, the volume in phase space is invariant under a canonical
transformation.
dζ
dζ = det( )dη = det(M )dη = ±dη
dη
If the volume is preserved (and mass is preserved due to conservation of mass), then ρ will be
constant. We also know that time evolution is a canonical transformation generated by H. We know
for cannonical transformations that
∂G
δη = J
∂η
For q and p we have
∂G
δqi = 
∂pi
∂G
δpi = −
∂qi

20
Hamilton’s equations give how p and q evolve with time.
∂H
q˙i =
∂pi
∂H
p˙i = −
∂qi
Upon comparing these equations, it is perfectly clear that time evolution is a canonical transfor-
mation generated by H where  = dt. After we have shown that time evolution is canonical, we
know ρ(q(0), p(0), 0) = ρ(q(t), p(t), t). The use Ehrenfest’s theorem to show that this translates to
Liouville’s theorem.

8.4 Symmetries and Conserved Quantities

Canonical transformations are given by
n o
δA =  A, G

For a continuous symmetry, δH = 0.

n o n o
δH =  H, G = 0 = H, G

so the generator of a continuous symmetry is a constant of the motion.

8.5 Hamilton-Jacobi Theory

Consider a canonical transformation of the second type F2 = F2 (q, P, t) such that the new Hamilto-
nian K = 0. This means the new variables are all cyclic. This is called Hamilton’s principal function
defined by F2 = S(q, P, t). Hamilton-Jacobi Equations are given by

∂S ∂S
H(q, p = , t) + =0
∂q ∂t
If the Hamiltonian does not depend on time explicitly, we can write S as

S = w − αt

where W = W (q, α) is Hamilton’s characteristic function. Therefore, we can write

∂W
H(q, p = ) = α1
∂q
where α1 is usually the energy. The other part of the transformation equations take the form of a
new constant β. (Remember all of the new coordinates are cyclic.)

∂S ∂W
=β= −t
∂α ∂α

8.6 Action-Angle Variables

We will assume a system that is separable and periodic such that w(q1 , ....qN , α1 , ...αN ) = W1 (q1 , α1 , ...αN )+
W2 (q2 , α1 , ...αN )..... + WN (qN , α1 , ...αN ). We also assume periodicity in each pair (q,p). However,
this periodicity takes two forms

21
 1. Libration: qi , pi come back to the same point (ex: Harmonic Oscillator)
2. Rotation: pi is a periodic function of qi
We can then ask what are the frequencies νi ? In order to determine this we will define action
variables, the number of which is equivalent to the number of degrees of freedom. These are definted
by I
Ji = pi dqi

These action variables yield Ji = Ji (α1 , ...αN ). These can then be inverted in order to yield αi =
αi (J1 , ...JN ). These allow us to express the Hamiltonian in terms of action variables. In this way,
we are replacing our momentum with J. The action variables are the new integration constant. We
will now consider W as the generating function where the angle variables are the new coordinates
in this transformation.
∂W
θi =
∂Ji
Then (θi , Ji ) are the canonical coordinates and momenta generated by the generating function w.
We can also define the frequencies as
∂H
νi =
∂J
these are such that βi = θi − νi t. The proof to show that these are frequencies is relatively straight-
forward. Consider the system to undergo ni periods τi such that ∆t = ni τi for each i. Therefore we
have θi = νi ∆t = νi ni τi we then have
X ∂θi
dθi = dqj (4)
∂qj
∂W
recall that θi = ∂Ji where θi = θi (q, J). The above equation then becomes

∂ X
dθi = ( Pj dqj )
∂J
Integrating this we see
Z Z X
∂ ∂ X
∆θi = dθi = ( Pj dqj ) = ( nj Jj ) = nj δij = ni
∂Ji ∂Ji
this implies νi τi = 1. So indeed νi are frequencies.

9 Completely Integrable Systems

There exists a n
complete
o set ofnconstants
o ofn the motion:
o I1 , ...., In that are involution with each other,
meaning that Ii , Ij = 0 = H, Ij = H, Ii . We can define action-angle variables
I X
Ji = Pj dqj
γi

where the γi ’s form a set of n-independent paths. We see that phase space is composed of invariant
n-dimensional torii. Each torus is characterized by the given values of the constant I1 , ....IN . If
n1 ν1 + n2 ν2 = 0 with n1 , n2 being integers, then the torus is a resonating torus. Otherwise the
trajectory densely fills the torus. The question is can we look at completely integrable systems which
are not separable? To check if a system is completely integrable, must find the second constant.
One example of such a system is the Toda Hamiltonian.

22
10 Regular and Chaotic Motion of Hamiltonian Systems
In general most systems will not be completely integrable. The central question is how do we
characterize the irregualrity of the dynamics? We are interested in the long term behaviour of the
system, but we will do this by taking ”snapshots” of the system called a Poincare Surface of Section.
This would be the plane in the figure below. Call G the function that evolves your system with time.

G(ηi ) = ηi+1
where G is symplectic. A fixed point η ∗ is a point that is mapped to itself G(η ∗ ) = η ∗ . A n-cycle is
a fixed point of Gn such that
G(η 0 ) = η1 , ....G(ηn−1 ) = η 0
we can now define a matrix K such that
∂G
K=( )
∂η
Finding the eigenvalues and eigenvectors of this matrix will tell us how the system evolves. we see
the following cases for the eigenvalues
1. If λ > 1 the system diverges from a fixed point η ∗ .
2. If λ < 1 the system converges to a fixed point η ∗
The figure below characterizes a plot that we can make in the angle-variable space which then moves
to a torus. Think of this like PacMan. (If you go off one side of the board, you will reappear on the
other.) The rotation number is the ratio of the frequencies. ννni . If the rotation number is a rational

number, then every point on the invariant curve is an n-cycle. When we add a perturbation to the
integrable system, then usually only two n-cycles survive - one is elliptic and the other is hyperbolic.
We also have the KAM Theorem which states that there exists a slightly deformed torus with the
same rotation number if the rotation number is ”far enough” away from a rational number. This is
defined as
m
α− > n−5/2
n
Where  is a small number that depends on the perturbation. This means that for sufficiently small
, most of the torii survive. How do we characterize chaotic vs. non-chaotic trajectories? This is
done through separation of orbits. we can write this in terms of the action-angle variables as
0
Ji = Ji

23
 0 0 0
θi − θi = (νi − νi )t + βi − βi
The general idea is that for regular trajectories, the distance is linear in time. However, for chaotic
trajectories, the distance is exponential in time or

D(t) ∼ eλt

where λ is the Lyapronov exponent. We have two cases of λ resulting in different trajectories.

1. When λ = 0 the trajectory is regular.

2. When λ > 0 the trajectory is chaotic

24