Physics 711. Jeffrey R.

Schmidt (Fall 2018 update)

Contents
1 Physics 711. Lecture 1 3
1.1 Decomposition theorems for angular momentum and kinetic energy . . . . . . . . . . . . . . . . . . . . 3
1.2 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Principle of virtual work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Physics 711. Lecture 2 7

2.1 Dissipative systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 A few examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Potentials in generalized Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Physics 711. Lecture 3 11

3.1 Least action principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Example. Metric spaces and minimal distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Physics 711. Lecture 4 16

4.1 Constraints and Lagrange multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

5 Lecture 5. Dynamics with constraints 22

5.1 Nonholonomic constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2 Virtual (infinitismal) work and external forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3 A note on the Jacobi integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

6 Lecture 6. Rolling and rotating bodies 27

7 Physics 711 Lecture 7 30

7.1 Constants of the motion. Noether’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.2 Some examples. Nonpropagating variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

8 Lecture 8. The central force problem 33

8.1 Center of mass separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.2 Relative motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.3 The equation for the orbital trajectory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

9 Lecture 9. Kepler problem in time 36

10 Physics 711 Lecture 10 38

10.1 Stable and closed orbits in general . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
10.2 Spiral orbits. A simple inverse problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
10.3 Orbital precession . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
10.4 Runge-Lenz vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

11 Lecture 11. Positive energy trajectories. Scattering 43

11.1 ”Hyperbolic” trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
11.2 The inverse scattering transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

12 Lecture 12. Scattering in the lab frame 47

12.1 The virial theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
12.2 Orthogonal tranformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
12.3 Lecture 13. Hamilton’s ”quaternions” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
12.4 Euler angles parameterization of rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

1
13 Lecture 13.1415926... Clifford and Lie algebras 56
13.1 Clifford algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

14 Lecture 15. Rotating coordinate systems 65

14.1 Motion at the earth’s surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
14.2 Foucault pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

15 Lecture 16. Torque-free rotation 69

15.1 Rolling cone description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

16 Lecture 17. Lagrange’s case 75

16.1 The Symmetric Top . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

17 index 75

2
1 Physics 711. Lecture 1
First we simply state Newton’s laws as a starting point, applied in an inertial frame
d
F = ṗ = (mv)
dt
so for no net force p is constant. For torques
d d
L = r × p, N =r×F =r×
(mv) = (mr × ṙ) = L̇
dt dt
so if the net torque vanishes the angular momentum is constant. Work and kinetic energy
Z f Z f Z f
d m 2 m 2
Wi,f = F · dr = m v̇ · vdt = ( v )dt = Tf − Ti , T = v
i i i dt 2 2
Theorem; If Wi,f independent of the path connecting rf to ri then F = −∇V for some V .
Proof. Define a function Φ(r, ri ) Z
Φ(r, ri ) = F (r 0 ) · dr 0
C(r,ri )

C(r, ri ) is a path connecting the points. Directional derivative of Φ in the v direction

Φ(r + vδt, ri ) − Φ(r, ri )

∇v Φ = lim
δt→0 δt
(r 0 ) · dr 0
R
C(r+vδt,r)
F
= lim
δt→0 δt
Invariance of the path lets us take r 0 = r + vt

F (r 0 ) · dr 0
R
C(r+vδt,r)
∇v Φ = lim
δt→0 δt
R δt
0
F (r + vt) · v dt
= lim
δt→0 δt
= F (r) · v

from which we find each cardinal directional derivative by selecting cardinal unit vectors for v;
∂Φ
∇i Φ = = F (r) · i = Fx
∂x
and so forth, in other words F = ∇Φ and we select Φ = −V defining a potential energy; for purely path independent
works Z f Z f
Tf − Ti = Wi,f = F · dr = − ∇V · dr = −(Vf − Vi ), Tf + Vf = Ti + Vi
i i

1.1 Decomposition theorems for angular momentum and kinetic energy

X
Fj,i + Fie = ṗi
j6=i

Sum over particles X X X XX X

ṗi = mi r̈i = Fie + Fj,i = Fie
i i i i j6=i i

Define the center of mass vector

P
X mi ri X
M= mi , R= i
, M R̈ = Ṗ = Fie
i
M i

3
The center of mass acceleration is given by externally applied forces only.
Strong law of action/reaction; Fi,j is parallel to rj − ri .
X X d X XX
ri × ṗi = (ri × pi ) = L̇ = ri × Fie + ri × Fj,i
i i
dt i i j6=i
X X X
= ri × Fie + (ri − rj ) × Fj,i = ri × Fie = N e
i pairs i,j i
Introduce relative coordinates
ri = R + ri0 , vi = Ṙ + ṙi0 = Ṙ + vi0
X
L= mi (R + ri0 ) × (Ṙ + vi0 )
i
X X X X
= mi R × Ṙ + mi ri0 × ṙi0 + ( mi ri0 ) × Ṙ + R × ( mi ṙi0 )
i i i i
0
P
but i mi ri = 0, so for angular momentum
X X X
L= mi R × Ṙ + mi ri0 × ṙi0 = R × P + ri0 × p0i
i i i
and for T
1X 1X 1X X 1 X1
T = mi (v + vi0 ) · (v + vi0 ) = mi v 2 + mi (vi0 )2 + v · mi vi0 = M v 2 + mi (vi0 )2
2 i 2 i 2 i i
2 i
2
X Z f X Z f XX Z f
Wi,f = Fi · dri = Fie · dri + Fj,i · dri
i i i i i j6=i i

XZ f XZ f
1
= mi v̇i · vi dt = d( mi vi2 ) = Tf − Ti
i i i i 2
ω
v
Example (qualifier problem). A rolling disk can barely ”jump” a step of
height h; conserve angular momentum about the corner it pivots on

Lz = m(R − h)v + ICOM ω = (ICOM + mR2 )ω 0

(with v = ωR) and energy after the collision

ω0
1
(ICOM + mR2 )(ω 0 )2 = mgh
2

Refine the two contributions in the case of conservative forces

XZ f XXZ f X 
Fie · dri = − ∇i Vae · dri = − e
Va,f e
− Va,i
i i i a i a

Call
∂
drij = dri − drj , ∇ij = ” ” = ∇i = −∇j , Vij = Vij (|ri − rj |)
∂rij
the gradient with respect to the rij vector.
XXZ f X Z f  X Z f
Fi,i · dri = − ∇i Vij · dri + ∇j Vij · drj = − ∇ij Vij · (dri − drj )
i j6=i i pairs, i,j i pairs, i,j i

X Z f X  
=− ∇ij Vij · drij = − Vij (f ) − Vij (i)
pairs, i,j i i,j, pairs

4
1.2 Constraints
Holonomic constraints are of the form
g(q1 , q2 , · · · , t) = 0
for example, a rigid body moves such that any two points in/on it obey

(ri − rj )2 = cij

Confining to a volume is non-holonomic, so is rolling of an axled wheel; spin rate φ̇ gives the velocity of the contact
point (let the axle make angle θ with the x-axis)

v = aφ̇, ẋ = v sin θ, ẏ = −v cos θ

or
dx − a sin θdφ = 0, dy + a cos θdφ = 0
∂g ∂g ∂g
for the first, think of dg(x, θ, φ) = 0 = ∂x dx + ∂θ dθ + ∂φ dφ = 0 with

∂g
gx = = 1, gθ = 0, gφ = −a sin θ
∂x
and try to find an integrating factor f sp we can get g;
∂(f gφ ) ∂(f gx ) ∂(f gθ ) ∂(f gx ) ∂(f gφ ) ∂(f gθ )
= , = , =
∂x ∂φ ∂x ∂θ ∂θ ∂φ
∂ ∂f ∂f ∂(−af sin θ)
(−af sin θ) = , 0= , 0=
∂x ∂φ ∂θ ∂θ
The second says f is independent of θ, and this together with the third says that f = 0.

1.3 Principle of virtual work

Let δri be a virtual displacement consistent with constraints and forces of constraint (as opposed to a arbitrary
displacement in which forces of constraint may change). At equilibrium
X X X
Fi · δri = 0, Fi · δri = Fia · δri + fi · δri = 0
i i i

breaking forces into applied and forces of constraint. Certainly

X
fi · δri = 0
i

for example a body constrained to move along a surface S, then ri is parallel to the surface but f is ⊥ to S so
fi · δri = 0. This implies X
Fia · δri = 0
i

for equilibrium. Can this be ”boosted” to cover arbitrary motions and cases?
Since X
Fi − ṗi = 0, (Fi − ṗi ) · δri = 0
i
or X X X
(Fia − ṗi ) · δri + fi · δri = 0, =⇒ (Fia − ṗi ) · δri = 0
i i i

d’Alembert’s principle of virtual work. Use generalized coordinates q and summation convention;
∂ri ∂ri
ri = ri (q, t), vi = ṙi = q̇k +
∂qk ∂t

5
and introduce generalized forces
∂ri
Fi δri = Fi δqk ≡ Qk δqk
∂qk

X X X ∂ri
ṗi · δri = mi r̈i · δri = mi r̈i · δqj
i i i,j
∂q j

X d ∂ri d  ∂ri 
= (mi ṙi · ) − mi ṙi δqj
i
dt ∂qj dt ∂qj
X d ∂ri  ∂ ṙ 
i
= (mi ṙi · ) − mi ṙi δqj
i
dt ∂q j ∂q j
X d ∂ri ∂  mi 
= (mi ṙi · )− ṙi · ṙi δqj
i
dt ∂qj ∂qj 2
X d ∂ ṙi ∂  mi 
= (mi ṙi · )− ṙi · ṙi δqj needs some ’splaining
i
dt ∂ q̇j ∂qj 2
 d ∂T ∂T 
= − δqj
dt ∂ q̇j ∂qj
X X
= Fj · δrj = Qj · δqj
j j

or  d  ∂T  ∂T   d  ∂T  ∂(T − V )  ∂V
− = Qj , − =0 Qj = −
dt ∂ q̇j ∂qj dt ∂ q̇j ∂qj ∂qj
If T = T (q̇) then define L = T − V
 d  ∂L  ∂L 
− =0
dt ∂ q̇j ∂qj

6
2 Physics 711. Lecture 2
If generalized forces can be written as
∂U d  ∂U  d  ∂L  ∂L
Qj = − + for some U , d’Alembert’s principle says − = 0, L=T −U
∂qj dt ∂ q̇j dt ∂ q̇j ∂qj

Electrodynamics; introduce scalar potential and vector potentials

d  dA
U = qΦ − qA · v, mr̈ = Q = −∇U + ∇v U = −q∇Φ + q∇(A · v) − q Ȧ = −q∇Φ + q∇(A · v) − q
dt dt
use ∇(A · v) = (A · ∇)v + (v · ∇)A + A × (∇ × v) + v × (∇ × A) = (v · ∇)A + v × B

dA
mr̈ = −q∇Φ + q(v · ∇)A + qv × B − q
dt
 ∂A 
= −q∇Φ + q(v · ∇)A + qv × B − q + (v · ∇)A = qE + qv × B
∂t

2.1 Dissipative systems

For dissipative forces such as Fx = −kvx (example, viscous fluid force on sphere of radius a F = −6πηav) construct
1X
F= kvi · vi , Fi = −∇vi F
2 i

work done by system against the friction

dW = −F · dr = −F · vdt = −2Fdt

and the generalized force calculation uses implicit time dependency again
∂ri ∂ri ∂ ṙi ∂F  d  ∂L  ∂L  ∂F
Qj = F · = −∇vi F · = −∇vi F · =− , so that − + =0
∂qj ∂qj ∂ q̇j ∂ q̇j dt ∂ q̇j ∂qj ∂ q̇j

2.2 A few examples

Nielsen form of the equations of motion. The qi are independent of each other (apriori, until EOMs are imposed)
but depend on t explicitly. ri depends on the independent qi and t implicitly, and q̇ is a new independent
variable, is a function of t not connected to q until EOMs are imposed, in other words
∂q d ∂ ∂  ∂ ∂ ∂q
= 0, = q̇(t) + , q̇(t) = q̇(t) + q = q̇(t) +0
∂t dt ∂q ∂t ∂q ∂t ∂q
Now watch the birdie...
∂r ∂r ∂ ṙ ∂r
ṙ = q̇(t) + =⇒ =
∂q ∂t ∂ q̇ ∂q
∂  ∂r ∂r  ∂  ∂r ∂r 
r̈ = q̇ q̇ + + q̇ +
∂q ∂q ∂t ∂t ∂q ∂t
∂2r 2 ∂2r ∂2r ∂r
= q̇ + 2 q̇ + 2 + q̈
∂q 2 ∂q∂t ∂t ∂q
∂ r̈  ∂2r ∂2r  ∂  ∂r ∂r 
= 2 q̇ + = 2 q̇ +
∂ q̇ ∂q 2 ∂q∂t ∂q ∂q ∂t
∂ ṙ
= 2
∂q

7
As a self-test that you dig these dependencies, show from
d ∂ ∂ ∂
= + ṙ + φ̇ , r = rr̂, r̂ = cos φi + sin φj, φ̂ = − sin φi + cos φj
dt ∂t ∂r ∂φ
that    
ṙ = ṙr̂ + rφ̇φ̂, r̈ = r̈ − rφ̇2 r̂ + rφ̈ + 2φ̇ṙ φ̂

Back to the original problem; start with

∂r ∂ ṙ ∂   ∂ r̈  ∂   ∂ ṙ 
mr̈ · δr = mr̈ · δq = mr̈ · δq = mr̈ · ṙ − m · ṙ δq = mr̈ · ṙ − 2m · ṙ δq
∂q ∂ q̇ ∂ q̇ ∂ q̇ ∂ q̇ ∂q
and we get the Nielsen form
∂ Ṫ ∂T
−2 =Q
∂ q̇ ∂q
Old qualifier problem. Consider a DC-LRC circuit with dynamical variable q that is the charge on a capacitor, q̇
the current(sorry for redundant use of L, deal)

1 2 q2 1 2
T = Lq̇ , V = , F= Rq̇
2 2C 2
 d  ∂L  ∂L  ∂F 1 2 q2
− + = 0, L= Lq̇ −
dt ∂ q̇ ∂q ∂ q̇ 2 2C
˙
The equation of motion is the loop equation with voltage source E = −Lq̈ = −LI;
q
Lq̈ + + Rq̇ = 0
C
Another. Johnson noise. In the last problem we had a quadratic Lagrangian and so will have a quadratic
Hamiltonian, the equipartition theorem says that if this system is placed in thermal equilibrium with a heat reservoir,
the average ”kinetic energy” will be
kB T Φ2
hT i = =h i
2 2L
where now T is reservoir temperature and Φ = Lq̇ is magnetic flux in the inductor, same for ”potential” energy

kB T q2
hV i = =h i
2 2C
Quadrupolar magnetic field. The rather useful connection between dynamics and optics will be made manifestly
clear in our discussion of the Hamilton-Jacobi formalism, but let’s do a particle physics example now that illustrates
the connection. A quadrupolar field is used in particle storage rings to focus the particle beam.
B0  
B= yi + xj
r0
then with q > 0, for incoming particles the equations of motion from U = −qA · v are
qB0 qB0 qB0  
mẍ = − xż, mÿ = y ż, mz̈ = − ẋx − ẏy
r0 r0 r0
Assume that the field is weak, or that the particles are fast in the z-direction and slow in x, y, we can approximately
3/2
solve the first two and substitute into the third to verify that ẍ, ÿ ∼ B0 and z̈ ∼ B0 << B0 (f 0 means df /dz)
r
0 0 mr0 v
x ≈ x0 cos(z/λ) + λx0 sin(z/λ), y ≈ y0 cosh(z/λ) + λy0 sinh(z/λ), z ≈ vt, λ =
qB0

8
How does this act to focus a charged particle beam? Suppose that the beam enters the weak field with x = x0
and ẋ = x˙0 << v = ż. Then z ≈ vt and we can change variables and examine the trajectory (ray) after traveling a
distance ` parallel to the z-axis;
x = x0 cos(z/λ) + λx00 sin(z/λ)
    
x cos(`/λ) λ sin(`/λ) x0
=
dx
dz
−λ−1 sin(`/λ) cos(`/λ) dx
dz 0

(Read the short doc on matrix ray optics). This factors

!
1 − fd d(2 − fd )
     
cos(`/λ) λ sin(`/λ) 1 d 1 0 1 d
= =
−λ−1 sin(`/λ) cos(`/λ) 0 1 −1/f 1 0 1 − f1 1 − fd

solve; f = λ/ sin(`/λ), cos(`/λ) = 1 − d` sin(`/λ), d = λ( 1−cos(`/λ)

sin(`/λ) ) = λ tan(`/2λ) which if we compare to the matrix
ray-tracing formulas, is the same result that we would get with a pair of simple empty regions separated by a thin
converging lens with a focal length fc = λ/ sin(`/λ). Unfortunately this ”lens” has aberration, look at the y-direction
focusing;     
y cosh(`/λ) λ sinh(`/λ) y0
= dy
dy
dz
λ−1 sinh(`/λ) cosh(`/λ) dz 0

which also factors

     
cosh(`/λ) λ sinh(`/λ) 1 λ tanh(`/2λ) 1 0 1 λ tanh(`/2λ)
=
λ−1 sinh(`/λ) cosh(`/λ) 0 1 λ−1 sinh(`/λ) 1 0 1

a pair of simple empty regions separated by a thin diverging lens with a focal length fd = −λ/ sinh(v`/λ).
You can create a net-positive (and aberation-free) focusing system by following this quadrupole magnet with another
in which the field is reversed. This will result in

x = x0 cosh(z/λ) + λx00 sinh(z/λ), y = y0 cos(z/λ) + λy00 sin(z/λ)

and if the second magnet is separated from the first by suitable distance ds
!
1 − dc +dfds +dd
     
1 0 1 dc 1 ds 1 dd 1 0 dc + ds + dd
=
−1/fc 1 0 1 0 1 0 1 −1/fd 1 −1/fnet 1 − dc +dfsc +dd

with
1 1 1 dc + ds + dd
= + − >0
fnet fc fd fd fc
no matter which order the magnet systems are in.
Time-dependent constraints. Another old qualifier problem. The
slotted disk rotates so that φ = 21 αt2 , starting at rest at t = 0. The bead
y of mass m slides in the slot with friction exerted by the walls of the disk,
which also supply the normal force and torque of constraint. In cylindrical
coordinates
x = r cos φ, y = r sin φ
ẋ = ṙ cos φ − rφ̇ sin φ, ẏ = ṙ sin φ + rφ̇ cos φ
Free particle
1  2  1  
x T = m ẋ + ẏ 2 = m ṙ2 + r2 φ̇2
2 2
but our particle is constrained by the channel to have φ = 12 α t2 , so introduce
a constraint (or a non-propagating degree of freedom) λ
1  2   1 
T = m ṙ + r2 φ̇2 − λ φ − α t2
2 2

9
for which
d  ∂T  ∂T d  d  ∂T  ∂T d 2  d  ∂T  ∂T 1
= , mṙ = mrφ̇2 , = , mr φ̇ = −λ, = , φ − α t2 = 0
dt ∂ ṙ ∂r dt dt ∂ φ̇ ∂φ dt dt ∂ λ̇ ∂λ 2

Solving we get
1 2  
φ= αt , mr̈ = mr(α t)2 , r 2m ṙ αt + mαr = −λ = −r N
2
λ is the torque of constraint that makes the mass travel in a circle (there are no other forces/potentials present in
L), corresponding to a normal force
N = 2m ṙ αt + mαr
exerted by the walls of the channel. If there is friction exerted by the walls as well, it influences the radial accelerations
 
mr̈ = mr(α t)2 − µN = mr(α t)2 − µ 2m ṙ αt + mαr

As the angular rate ω = αt is increased from 0, the mass does not slip (accelerate or move radially, r̈ = 0 = ṙ until
time t∗   µ √
0 = mr(α t∗ )2 − µ 0 + mαr , t2∗ = at which time ω∗ = αt = µα
α

2.3 Potentials in generalized Lagrangians

Given the components of a magnetic field, can we be guaranteed that we can actually perform this construction (a
prerequisite is finding a suitable vector potential)? That depends...
It may be possible that there are perfectly cromulant magnetic fields (which obey ∇ · B = 0) that are not curls of
some vector field A, in other words

∇·B =0 (B ∈ ker(div)) 6
=⇒ B = ∇ × A, (B ∈ Im(curl))

On the surface we can say with certainty at most that those vectors which are the curl of some vector potential lie
in the kernel of div, which is a subspace of R3

Im(curl) ⊆ ker(div) since ∇ · (∇ × A) = 0

If we are working in some region U ⊆ R3 call

H 2 (U ) = ker(div)/Im(curl)

the coset consisting of vectors annihilated by the divergence unique up to addition of a vector that is the curl of
another vector (something in Im(curl)). As long as U is star-shaped, this coset is empty;
Proof. Given some B such that ∇ · B = 0, construct
Z 1
A= B(tr) × tr dt
0

Then Z 1
  d2  d2 
∇ × B(tr) × tr = t B(tr) , and so ∇×A= t B(tr) dt = B
dt 0 dt
so for example consider a quadrupolar magnetic field

B0 1  2
Z
B0    B0  
B= xj + yi , A= t xzi − t2 yzj + t2 (y 2 − x2 )k dt = xzi − yzj + (y 2 − x2 )k
r0 r0 0 3r0

10
3 Physics 711. Lecture 3
D’Alembert’s principle is a technological step-up from Newton’s laws: the laws were incorporated into its derivation,
but the principle itself introduces a new and very powerful advancement, the ability to use generalized coordinates,
which first appear as parameterizations of the position vector r = r(q, t). A set of q 0 s for each particle become the
dynamical degrees of freedom in the equations of motion instead of the actual positions r.
The ”Least action principle” (in quotes because it is not THE least action principle (Hamilton’s), but instead the
Lagrange-Euler version) is another powerful reformulation at an even lower level. It also incorporates the generalized
coordinates concept, and adds another equally if not more powerful notion: parameterization invariance. I will use
this repeatedly today to point out its true power.

3.1 Least action principle

Newton’s laws should be regarded as a system of postulates relating cause to effect for the motion of a body. Is there
an underlying single postulate from which all three can be ”derived”? The notion that the ”correct” trajectory of
a body is the path that minimizes some natural function is called the Principle of Least Action, and as a theory of
dynamics it supersedes Newton’s laws. Its mathematical foundation is calculus of variations.

The major advantage of the Principle of Least action is that it breaks us free of vectors in coordinate representations,
and allows us to use arbitrary generalized coordinates that may not form vector spaces at all, but in terms of which
the equations of motion are simpler and more intuitive. The core notion that the correct particle path is a path of
extremal ”something” gives dynamics a very optical flavor, as I mentioned in class, I find the distinction between
dynamics and optics to be artificial.

The fundamental problem of calculus of variations is a minimax problem of the following nature; find a function
Rb
f (x) such that f (a) = A, f (b) = B such that the integral a L(f, f 0 , x) dx = I[f ] is extreme. We recognize that
there are an infinite number of functions f (x) such that f (a) = A and f (b) = B. The ”variations” in the title of
the topic appropriately labels the method of solution to the problem; we begin with an unspecified correct f (x) that
solves the problem, then any alteration g(x) = f (x) + δf (x) such that g(a) = A and g(b) = B, meaning δf (a) = 0
and δf (b) = 0, must increase the value of the integral if f (x) makes it minimal. This we can translate into a set of
equations that can be solved for f (x). To see this, expand the integrand around the minimum f (x);
∂L(f, f 0 , x) ∂L(f, f 0 , x)
L(g, g 0 , x) = L(f + δf, f 0 + (δf )0 , x) = L(f, f 0 , x) + δf + (δf )0 + · · · ,
∂f ∂f 0
and put it back into the integral to get
Z b Z b
0 ∂L(f, f 0 , x) ∂L(f, f 0 , x) 
L(g, g , x) dx = L(f, f 0 , x) + δf + 0
(δf )0 + · · · dx,
a a ∂f ∂f
or Z b Z b Z b
∂L(f, f 0 , x) ∂L(f, f 0 , x) 
δI[f ] = L(g, g 0 , x) dx − L(f, f 0 , x) dx = δf + (δf )0
+ · · · dx.
a a a ∂f ∂f 0
Integrate by parts on the last term, since (δf )0 = dxd
(δf );
 ∂L(f, f 0 , x)  Z b
b ∂L(f, f 0 , x) d ∂L(f, f 0 , x) 
δI[f ] = δf |a + δf − δf + · · · dx,
∂f 0 a ∂f dx ∂f 0
but the first term on the right hand side is zero since δf (a) = δf (b) = 0, and we can factor the remainder to get
Z b
∂L(f, f 0 , x) d ∂L(f, f 0 , x) 
δI[f ] = − δf dx.
a ∂f dx ∂f 0
Now comes the minimax condition; f (x) is the function extremizing the integral if in some sense the ”derivative” of
I with respect to f is zero. We define the functional derivative of another function to be
δf (x)
= δ(x − x0 ),
δf (x0 )

11
which enables us to differentiate in the space of functions. This is a specialized form of the Gateaux derivative of
the action with respect to τ (we don’t need this though, but you may run into it)
I[f + τ δf ] − I[f ]
(dG I)[f, δf ] = lim
τ →0 τ
Then our minimax condition is the vanishing of
Z b
δI[f (x)] ∂L(f, f 0 , x) d ∂L(f, f 0 , x)  δf (x)
= − dx
δf (x0 ) a ∂f dx ∂f 0 δf (x0 )
Z b
∂L(f, f 0 , x) d ∂L(f, f 0 , x)  ∂L(f (x0 ), f 0 (x0 ), x0 ) d ∂L(f (x0 ), f 0 (x0 ), x0 )
= − 0
δ(x − x0 ) dx = 0
− 0 .
a ∂f dx ∂f ∂f (x ) dx ∂f 0 (x0 )
and we obtain our familiar Euler-Lagrange equation
∂L(f (x0 ), f 0 (x0 ), x0 ) d ∂L(f (x0 ), f 0 (x0 ), x0 )
0= 0
− 0 .
∂f (x ) dx ∂f 0 (x0 )
The methods of functional calculus are a major part of modern quantum field theory, in particular the path integral
approach to quantization. Julian Schwinger was the major contributor and developer of these techniques in the
1940’s and fifties, although the fundamental formalism goes back to the Bernoullis and Euler.

Example. The brachistochrone problem. This is a traditional example. One of the first such problems ever
proposed was that of finding a path f (x) = y(x) between two points a = (0, 0) and b = (x0 , y0 ), y0 < 0, such that a
particle will slide from one to the other under gravity alone in minimal time. The functional in this case is
Z b Z b Z bp 2
ds dx + dy 2
I[f ] = I[y] = T = dt = = ,
a a v a v
where v is the velocity
√ gotten from conservation of energy for a body falling initially at rest from the origin 12 mv 2 =
−mg(−y), v = 2gy. We obtain
Z x0 s 2 Z x0 p 2
dx + dy 2 ẋ + ẏ 2
I[y] = T = = √ ds, s is any parameter
0 2gy 0 2gy

Since there is no explicit x dependence here, the Euler-Lagrange equations become simplified;
d ẋ  ẋ
p = 0, p =C
ds 2 2
2gy(ẋ + ẏ ) 2gy(ẋ2 + ẏ 2 )
and now select the parameter wisely (always remember that face-melting scene from Indiana Jones and the Holy
dy
dx
Grail and choose wisely), let s = x, then ẋ = dx = 1 and ẏ = dx = y 0 , and rename our constant;
p √
y(1 + (y 0 )2 ) = K

and now repeat the idea, select a parameterization this time of y 0

dy
dy dy dφ dφ
y = y(φ), x = x(φ), y0 = = = dx
,
dx dφ dx dφ

and in the present case it is handy to manipulate this to

dx 1 dy
= 0 ,
dφ y dφ

and to perform the trigonometric substitution y 0 = cot φ2 so that the Euler-Lagrange equation becomes
K K
y= 0 2
= (1 − cos φ),
1 + (y ) 2

12
and it only remains to find x;
dx 1 dy K φ φ φ φ φ K
= 0 = tan sin φ = K tan sin cos = K sin2 = (1 − cos φ).
dφ y dφ 2 2 2 2 2 2 2
Integrate to
K
(φ − sin φ),
x(φ) =
2
and you may recognize these as the coordinates of points on a cycloid as functions of parameter φ.

On
p a side note, take
√ our first integral (an equation of motion integrated once to obtain a constant of the motion)
y(1 + (y 0 )2 ) = K and rewrite it as
K2
(y 0 )2 − = −1
y
Interpret this as T + V = E for bound state (E < 0) one-dimensional motion in a gravitational field. This is why we
troube ourselves today with this example (see PS-1.8).

3.2 Example. Metric spaces and minimal distance

Calculus of variations is useful in geometry and geometrical physics areas such as General Relativity, in which we
find paths of minimal arc length (geodesics) between two points in space.
You studied a sphere in PS-1,which is a surface of constant positive curvature. AdS2 or the Poincare’ upper-half
plane (y > 0), is a space of constant negative curvature, which has metric function (for finding arc-lengths, or kinetic
energy metric if you wish to stick to the ideas of PS-1.6)

dx2 + dy 2
ds2 =
y2
then with p
ẋ2 + ẏ 2
Z
dx dy
L= , I[x, y] = L dξ, ẋ = , ẏ =
y dξ dξ
the x-equations is d √ ẋ
dξ ( y ẋ2 +ẏ 2 ) = 0, select ξ = x, you get (with y 0 = dy/dx)
Z
p hy dy
1 = hy 1 + (y 0 )2 , p = x − x0
1 − h2 y 2

Integrate
1p
1 − h2 y 2 = (x − x0 ),
− h−2 = y 2 + (x − x0 )2
h
and so one class of geodesics are semicircular arcs of radius equal to the reciprocal (and finite) ”energy” of the curve.
A second class of geodesics corresponds to h → 0;
p
(y 0 )2 1 + (y 0 )2 −1
0= p − , 0= p , y 0 = ∞, x = x0
y 1 + (y ) 0 2 y 1 + (y 0 )2

straight lines ⊥ to the x-axis.

Example. Post Newtonian GR. A good problem; Show that the proper action from which you obtain the
relativistically correct equations of motion for a free particle is the proper time (times −mc2 )
Z Z r Z
1
S = −mc2 ds = −mc2 dt2 − 2 (dx · dx) = −mc2
p
gµν dxµ dxν
c

with x0 = t, x1 = x, x2 = y, x3 = z and gµν = diag(1, −c−2 , −c−2 , −c−2 ), g µν = diag(1, −c2 , −c2 , −c2 ). General
relativity lets gµν be coordinate dependent. You will fiddle with such a metric space on PS-1.6 where you found

13
that d’Alembert’s principle (or see the end of this lecture), and therefore calculus of variations using s itself as
parameter leads to ”uniform motion” along paths that obey
d2 xµ µ dxα dxβ
= −Γ αβ
ds2 ds ds
dxj dxj
For nonrelativistic speeds with ds ≈ dt << c, the largest term on the right side (go to spherical coordinates) for
a radial geodesic will be
d2 r r dx0 dx0 1 rr  
≈ −Γ 00 ≈ − g 2∂t g0r − ∂r g00
dt2 dt dt 2
and assuming gravity (spacetime curvature) to be a weak perturbation, we figure that g00 = 1 + Φ where Φ → 0 far
from a gravitating body, and we assume as well that g rr ≈ −c2 , the Minkowski flat spacetime value. We want this
to reproduce Newtonian gravity;
d2 r 1 MG 2M G
≈ + (−c2 )∂r (1 + Φ) = − 2 , Φ=−
dt2 2 r c2 r
and you have justified the form of the (Post-Newtonian) Schwartzchild metric.
Example. Post Newtonian approximation of GR deflection of light. Planetary orbits and light ray trajec-
tories come from extremizing proper time of transit in an optical picture
s
2M G 1
ds = (1 − 2 )dt2 − 2 dr · dr
c r c (1 − 2M G
c2 r )

Despite appearances this is not the Schwartzchild metric, it may not even be a solution to the field equations. In the
post-Newtonian approximation we concern ourselves with describing what happens in places where space is very
flat, so 2GM
c2 << r. The trajectory of a light ray is a null geodesic ds = 0, and so we find that
 dr   2GM 
 =c 1− 2
 
dt c r
so it is as if the effects of curved space make light travel at a speed that departs from c; spacetime behaves spatially
flat with refractive index
1 Φ 2GM GM
n= ≈ 1 + 2, Φ= , = 1475m for the sun
1 − 2M2
c r
G c r c2
Next time: To get the path of a light ray we minimize the time of transit a’la Fermat’s principle, the standard
optics methodology
Z Z s
dx dx
δ nds = δ n · dξ = 0
dξ dξ
s
dx
d  n dξ  dx dx d  dx 
q = ∇n · , let ξ = s; n = ∇n
dξ dx dx
· dξ dξ ds ds
dξ dξ

Integrate this along a ray from s = −∞ to s = ∞ (remote past to distant future)

y
 dx ∞ Z ∞
n  = ∇n ds

δ ds −∞ −∞

Note that lims→±∞ n(s) = 1, dot this into j and

R realize that dy/ds = sin δ ≈ δ, the angle that the
rays make with the horizontal, so the left side is our
ray deflection for light grazing the sun
x
Z ∞
∂n
δ= ds
−∞ ∂y

14
Integrate along the undeviated path (approximate, clearly x changes), this is correct to lowest order in GM/c2 (note
that angles such as δ are measured CCW, a negative δ is a CW deflection)
Z ∞
2GM R 4GM
δ≈− 2 (x2 + R2 )3/2
dx = − 2 ≈ −1.749” for the sun
−∞ c c R

Details, details,...
Consider the following action, divorced from any physical motivation
Z q Z q
dx
I= gxx dx2 + gyy dy 2 = gxx ẋ2 + gyy ẏ 2 dξ, ẋ =
dξ
in which ξ is any parameter, and gxx , gyy are both functions of x, y. As far as we are concerned this is just an
action that we will find Lagrange-Euler equations of motion for. These equations of motion are
∂gxx 2 ∂g
yy 2
d gxx ẋ ∂x ẋ + ∂x ẏ

1
p = 2
p
dξ gxx ẋ2 + gyy ẏ 2 gxx ẋ2 + gyy ẏ 2
ds
p p
similarly for y. Let dξ = gxx dx2 + gyy dy 2 = ds, then dξ =1= gxx ẋ2 + gyy ẏ 2 , so we obtain

d   ∂g
xx 2 ∂gyy 2   ∂g
xx ∂gyy 2 
gxx ẋ = 12 ẋ + ẏ , gxx ẍ + ġxx ẋ = 1
2 ẋ2 + ẏ
ds ∂x ∂x ∂x ∂x
∂gxx ∂gxx
but gxx = gxx (x, y) so the Chain Rule says ġxx = ∂x ẋ + ∂y ẏ

∂gxx ∂gxx  ∂g
xx 2 ∂gyy 2 
gxx ẍ + ( ẋ + ẏ)ẋ = 21 ẋ + ẏ
∂x ∂y ∂x ∂x
finally we get the form written on the previous page in summation notation
 ∂g ∂gxx ∂gyy 2   ∂g ∂gxx ∂gyy 2 
xx xx
gxx ẍ = − 12 ẋ2 + 2 ẋẏ − ẏ , ẍ = − 21 (gxx )−1 ẋ2 + 2 ẋẏ − ẏ
∂x ∂y ∂x ∂x ∂y ∂x

15
4 Physics 711. Lecture 4
The dynamically correct trajectory of a particle in classical mechanics is the one that minimizes the
action S or I. Technically this is not the least action principle (that one is due to Hamilton) but we will refer to
this for now as the least action principle. This principle replaces Newton’s first and third laws as the foundation
of dynamics (we still need action/reaction) and d’Alembert’s principle. It gives the foundations an optical flavor
which is very powerful, you saw one example Monday, you will examine another in PS-2 (please, try to contain your
excitement).

The action is defined as Z t1

I(t0 , t1 ) = L(q, q̇, t) dt
t0

in which q is a generalized coordinate, such as x, and L (Lagrangian) for Newtonian dynamics is

L(q, q̇) = T (q̇) − V (q)

which contains the full dynamical specification. We extremize this in the usual way, by setting derivatives (or
differentials) to zero
Z t1
δS(t0 , t1 ) = δL(q, q̇, t) dt = 0
t0

Aside from parameterization invariance, one of the major advan-

tages of the least action principle is its minimalistic requirements
placed on the use of coordinates qi they can be virtually any-
thing, and need only be peripherally related to the Cartesian
coordinates of a particle. They need not form vectors, but do
s M label points of a manifold, a space that is more general than a
vector space (less restrictive).
Example (seen on qualifiers) Construct the Lagrangian for
the gadget illustrated here.
For the large block M , its position is x1 = s, y1 = 0, for the small
θ block x2 = s + ` sin θ, x˙2 = ṡ + ` cos θ θ̇, and y2 = −` cos θ, ẏ2 =
m −` sin θ θ̇. The kinetic energy is therefore
1 1 1 1
T = M (ẋ21 + ẏ12 ) + m(ẋ22 + ẏ22 ) = M ṡ2 + m(ṡ2 + `2 θ̇2 + 2` cos θ ṡθ̇)
2 2 2 2
and the potential is
1
V = k(s − s0 )2 − mg` cos θ
2
making the Lagrangian
1 1 1
L= M ṡ2 + m(ṡ2 + `2 θ̇2 + 2` cos θ ṡθ̇) − k(s − s0 )2 + mg` cos θ
2 2 2
which does not contain Cartesian coordinates of particles.

The Lagrange-Euler equations of motion

d ∂L ∂L
( )=
dt ∂ q̇ ∂q

16
using the action with this L for q = s, θ are
∂L ∂L
= M ṡ + m` cos θθ̇, = −k(s − s0 )
∂ ṡ ∂s
d

so that M ṡ + m` cos θ θ̇ = −k(s − s0 )
dt
and finally M s̈ + m` θ̈ cos θ − m` θ̇2 sin θ = −k(s − s0 )
∂L ∂L
= m`2 θ̇ + m`ṡ cos θ, = −mg` sin θ
∂ θ̇ ∂θ
d
m`2 θ̇ + m`ṡ cos θ


so that = −mg` sin θ
dt
m`2 θ̈ + m`s̈ cos θ − m`ṡ θ̇ sin θ = −mg` sin θ

Solution of these equations is another issue entirely. The equations are second order and nonlinear, so solving
analytically can be a real challenge. There are special cases that are much easier; solving for equilibrium points
and small amplitude normal motions (all coordinates oscillate at same frequency) or coupled oscillations about the
equilibria (topics of chapters in Goldstein). For example in the latter problem we assume all generalized displacements
such as s − s0 = u and θ are small, and similarly all generalized velocities and accelerations are small. This simplifies
the equations considerably, in this case to

M ü + m` θ̈ = −ku, m`2 θ̈ + m`ü = −mg` θ

which are solved with standard linear algebra methods of differential equations theory to give small amplitude
normal modes of coupled oscillations of this system.
The equations of motion resulting from application of the Least Action Principle are intrinsically second order. The
solution of second order differential equations is highly nontrivial and constitutes a large field in mathematics. With
this in mind a major goal in the development of classical physics is the reduction to quadrature of the equations
of motion. This means the reduction of the equations of motion, through the use of variable transformations, to
first order equations that can immediately be integrated. With this goal in mind we ask, what types of variable
transformations are consistent with classical mechanics as it is formulated?

4.1 Constraints and Lagrange multipliers

Consider the quite standard minimax problem of finding the extrema (xm , ym ) of a function f (x, y) = z, solved by
∂f  ∂f 
= 0, =0
∂x x=xm ,y=ym ∂y x=xm ,y=ym
 

but now impose a constraint, such as the maximum value on a circle x2 + y 2 = R2 . The conventional method is the
use of Lagrange multipliers. Minimization of f is
∂f ∂f
df = 0 = dx + dy
∂x ∂y
but with the constraint imposed, dx and dy are not independent, but are related by

dR2 = 0 = d(x2 + y 2 ) = 2x dx + 2y dy

so that the points we are seeking are not global extrema of f anymore. They are however global extrema of a different
function, to see this notice that
∂f ∂f ∂f ∂f
df = 0 = dx + dy implies − dx = dy
∂x ∂y ∂x ∂y
and the constraint implies
−2x dx = 2y dy

17
and we can divide to get (at x = xm , y = ym )
 1 ∂f   1 ∂f 
= = λ = λ(xm , ym )
 
2x ∂x x=xm ,y=ym 2y ∂y x=xm ,y=ym
 

introducing the Lagrange multiplier λ. Then the condition for an extremum is

∂f ∂f
0= − 2λx, 0= − 2λy
∂x ∂y
which is the extremum of a new function

F = f − λ(x2 + y 2 − R2 )

Example Find the highest point on f (x, y) = 4 (x − 1)2 + 4 y 2 that lies on x2 + y 2 = 4.

Continuing in the way we have outlined above, we require that

0 = 8(x − 1) − λx, 0 = 8y − λy

The second equation has solutions y = 0 or λ = 8. The solution λ = 8 should be tossed out since it inconsistent with
8
0 = 8(x − 1) − λx, so we keep y = 0, find that this requires x = 8−λ , and now we get λ by requiring that this point
8
( 8−λ , 0) satisfy the constraint equation

8 2
4=( ) + 02 , λ=4
8−λ
In Lagrangian mechanics, holonomic constraints such as

fi (q1 , q2 , · · · , qn ) = 0, i = 1, · · · , Nc

are incorporated directly into the Lagrangian with Lagrange multipliers, one for each constraint, with Nc total
constraint equations.
Z b X
I= (T − V − λi fi (q)) dt
a i

Then variations qj → qj + δqj produces

Z b N N X Nc
X ∂L d ∂L
X ∂fi

0= − δqj − λi δqj dt
a j=1
∂qj dt ∂ q̇j j=1 i=1
∂qj

which results in the modified Euler-Lagrange equations

c N c N
d ∂L ∂L X ∂fi X
− = λi = λi ai,j = Qj
dt ∂ q̇j ∂qj i=1
∂qj i=1

in which we interpret the new terms as representing forces of constraint Qj .

Nonholonomic constraints
fi (q1 , q2 , · · · , qn , q̇1 , q̇2 , · · · , q̇n ) = 0
can be handled the same way, with simple Lagrange multipliers which lead to
d ∂L ∂L X  ∂fi d  ∂fi 
− = Qj , Qj = λi − λi
dt ∂ q̇j ∂qj i
∂qj dt ∂ q̇j

We can incorporate non-conservative additional forces such as air resistance or kinetic friction with
c N c N
d ∂L ∂L X ∂fi X
− = λi = λi ai,j + Fjext
dt ∂ q̇j ∂qj i=1
∂qj i=1

18
 x2
Example. The simple inclined plane
m
1
L= m(ẋ21 + ẋ22 ) − mg(−x1 sin φ + x2 cos φ)
2
Consider the inclined plane and block. We have the con-
straint that x2 = 0 or x˙2 = 0, so there is one constraint
x1
with
2
X ∂f1
a1,1 δx1 + a1,2 δx2 = δxj
j=1
∂xj

such that a1,1 = 0 and a1,2 = 1. This leads to

φ

d ∂L ∂L d ∂L ∂L
− = λ1 a1,1 + F1 , − = λ1 a1,2 + F2
dt ∂ ẋ1 ∂x1 dt ∂ ẋ2 ∂x2
or
mẍ1 − mg sin φ = F1 , mẍ2 + mg cos φ = λ + F2
and we can identify λ1 with the normal force (a force of constraint). With this interpretation we could add a frictional
force F1 = −µk λ1 = −µk N . The constraint x˙2 = 0 implies x¨2 = 0 resulting in

λ1 = mg cos φ, ẍ1 = g sin φ − gµk cos φ

the expected normal force, and equation of motion if we do incorporate kinetic friction.
Example. Rolling motion Consider a body rolling down
an incline; friction causes it to roll without slipping.
1 1
L= mẋ2 + Iω 2 − mg(` − x) sin φ
2 2

in which ω = θ̇. The rolling constraint is non-holonomic

dx
dt = Rω = Rθ̇ but can be integrated to a holonomic
θ constraint
0 = dx − R dθ
x1 There is one constraint, therefore one Lagrange multiplier
λ1 , and the constraint equation

0 = a1,x dx + a1,θ dθ = dx − R dθ

means that a1,x = 1 and a1,θ = −R.

φ
The forces of constraint are now

Fx = λ1 a1,x = λ1 , Fθ = λ1 a1,θ = −Rλ1

and the Euler-Lagrange equations of motion become

d ∂L ∂L d ∂L ∂L
− = Fx , mẍ − mg sin φ = λ1 , − = Fθ , I θ̈ = −Rλ1
dt ∂ ẋ ∂x dt ∂ θ̇ ∂θ

These two equations together with the constraint ẋ = Rθ̇ or ẍ = Rθ̈ can now be solved; incorporate the constraint
ẍ
to eliminate θ̈; mẍ − mg sin φ = λ1 , IR = −Rλ1 and between these eliminate λ1 ;

I mg sin φ
mẍ − mg sin φ = λ1 = − ẍ, ẍ =
R2 m + RI2

19
Now we can go back and recover the static frictional force of constraint λ1 = − RI2 mg sin φ
m+ I
.
R2
Example. The Atwood machine Another good example is the standard Atwood machine, with constraint given
by a fixed rope length ` = x1 + x2 . The Lagrangian is
1 1
L= m1 ẋ21 + m2 ẋ22 + m1 gx1 + m2 gx2
2 2
and the single constraint requires one Lagrange multiplier representing one force of constraint, the rope tension;

dx1 + dx2 = 0 = a1,1 dx1 + aa,2 dx2 for which F1 = a1,1 λ1 = λ1 , F2 = a1,2 λ1 = λ1

Example. The bead on a rotating wire. Contemplate a wire pivoted at the origin, rotating at fixed rate ω,
with a bead sliding on it freely as it rotates.
The normal force exerted by the wire on the bead is the constraint
y
force that keeps the bead on the wire. As coordinates use r as the
distance from the origin, and n measured normally from wire to bead.
The constraint is then n = 0.
This problem is particularly nice in that its solution illustrates use
n of a moving coordinate system, the constraint results in a velocity-
dependent normal force, and the addition of contact friction results in
r
what would be a damped oscillator equation for r if the radial forces
were reversed.
Begin with φ = ωt = 0 and so the wire lies on the x-axis. The bead
coordinates become (x, y) = (r, n). Now rotate the wire resulting in
φ = ωt bead coordinates
x
    
x cos ωt − sin ωt r
=
y sin ωt cos ωt n

Compute derivatives
       
ẋ cos ωt − sin ωt ṙ − sin ωt − cos ωt r
= +ω
ẏ sin ωt cos ωt ṅ cos ωt − sin ωt n

Note that
       
− sin ωt − cos ωt 0 −ω cos ωt − sin ωt cos ωt − sin ωt 0 −ω
= =
cos ωt − sin ωt ω 0 sin ωt cos ωt sin ωt cos ωt ω 0

This gives us a kinetic contribution

ẋ2 + ẏ 2 = ṙ2 + ṅ2 + ω 2 (r2 + n2 ) + 2ω(rṅ − nṙ)

With the constraint our Lagrangian is

1  2 
L= m ṙ + ṅ2 + ω 2 (r2 + n2 ) + 2ω(rṅ − nṙ) − λn
2
and the equations of motion

m(n̈ + ω ṙ) = −mω ṙ + mω 2 n − λ, m(r̈ − ω ṅ) = mω ṅ + mω 2 r

Impose the constraint n = ṅ = 0 and we obtain

r̈ = ω 2 r, λ = −2mω ṙ

The normal force exerted by the wire on the bead is velocity dependent. We can readily solve for r

r = Aeωt + Be−ωt

20
and we could introduce a contact frictional force opposing ṙ and proportional to the normal force λ to get

r̈ = ω 2 r − µk (2mω ṙ)

bf I think you can see that simply inserting constraints and Lagrange multipliers into L and applying the least action
principle to get equations of motion is much easier than determining Fi0 s and a0i,j s.

Example (standard qualifier problem). The breaking of a

constraint.
A small object of mass m is placed at the top of a friction-
less hill of radius R. As it begins to slide down, it reaches
θ a point where it loses contact with the hill. Find the angle
at which that happens, and the speed of the object at that point.

We have a rather simple procedure; the first step is to write a suitable Lagrangian in which to implement the
constraint. In this case we want r = R, the particle remains on a circle, so we use polar coordinates

x = r sin θ, y = r cos θ, ẋ2 + ẏ 2 = ṙ2 + r2 θ̇2

and our Lagrangian with the constraint is

1  2 
L= m ṙ + r2 θ̇2 − mgr cos θ + λ(r − R)
2
and equations of motion are
 
mr̈ = mrθ̇2 − mg cos θ + λ, m 2rṙθ̇ + r2 θ̈ = mgr sin θ, r = R, ṙ = r̈ = 0

We find that the constraint and equation of motion are

λ = mg cos θ − mRθ̇2 , Rθ̈ = g sin θ

Once the mass loses contact with the hill, it no longer follows the constraint. How do we handle this? By setting
the constraint Lagrange multiplier (the force of constraint) to zero at that point. When we lose contact
with the hill, at θ = θc 
mg cos θc = mRθ̇2 

θ=θc

We now integrate the equation of motion by multiplying it by θ̇ (we can’t just integrate an equation containing λ
since it is in general a time-dependent and as yet unknown function)

R d  2 d 
Rθ̈θ̇ = θ̇ = g θ̇ sin θ = −g cos θ
2 dt dt
Integrate from t = 0 when we are at the top of the hill, to tc when θ = θc to get the Jacobi integral (a first integral)

R  2    
θ̇  − 02 = −g cos(θc ) − cos(0)
2 θ=θc

and solve the Jacobi integral and remaining equation of motion for the critical angle
R g    2 2Rg
cos(θc ) = −g cos(θc ) − 1 , cos(θc ) = , (Rθ̇)2 =
2 R 3 3

21
5 Lecture 5. Dynamics with constraints
Example. Stop me if you hear this one before: a ladder of length 2` and mass m has one end resting against a
vertical frictionless wall, the lower end rests on a frictionless orizontal surface. Initially the ladder makes and angle
θ = 0+ with the vertcal wall, so it will topple. Which vanishes first, the normal force exerted by the vertical wall or
by the horizontal surface?
We have two constraints that translate into equations for the COM (x, y)

x − ` sin θ = 0, y − ` cos θ = 0

and therefore
1 1
L= m(ẋ2 + ẏ 2 ) + Ic θ̇2 − mgy + λ1 (x − ` sin θ) + λ2 (y − ` cos θ)
2 2
This gives the equations of motion

mẍ = λ1 , mÿ = −mg + λ2 , Ic θ̈ = λ2 ` sin θ − λ1 ` cos θ, x = ` sin θ, y = ` cos θ

The approach is to eliminate x, yλ1 , λ2 and obtain a first integral for θ of the form θ̇2 = F (θ). Next we set the
relevent force of constraint to its broken value in the effective equations of motion for θ, gotten by eliminating
x, yλ1 , λ2 , which will be solvable for anothere curve θ̇2 = G(θ). If there is an intersection F (θc ) = G(θc ), this is the
angle (configuration) for which the constraint breaks.
First

λ2 = mg + m`(cos θ)·· = −m`(cos θ θ̇2 + sin θ θ̈) + mg, λ1 = m`(sin θ)·· = m`(− sin θ θ̇2 + cos θ θ̈)

Substitute these into the θ EOM and solve

   
Ic θ̈ = ` sin θ − m`(cos θ θ̇2 + sin θ θ̈) + mg − ` cos θ m`(− sin θ θ̇2 + cos θ θ̈)

or
4m`2
Ic θ̈ = mg` sin θ − m`2 θ̈, θ̈ = mg` sin θ
3
Multiply by θ̇ and both sides are total time derivatives
d 1  d 
Ic θ̈θ̇ = mg` sin θθ̇ − m`2 θ̈θ̇, (Ic + m`2 )θ̇2 = − mg` cos θ
dt 2 dt
1 3g
(Ic + m`2 )θ̇2 + mg` cos θ = h = mg`, θ̇2 = (1 − cos θ)
2 2`
using Ic = 13 m`2 . This is the first curve.
Next we break the constraint of our choice, for example set λ1 = 0 (normal force exerted by the wall), our EOMS
are
λ1 = 0 = m`(− sin θ θ̇2 + cos θ θ̈), (Ic + m`2 )θ̈ = mg` sin θ
and use 12 (Ic + m`2 )θ̇2 + mg` cos θ = h = mg`, we obtain
 3g  3g
0 = m` − sin θθ̇2 + cos θ( sin θ) , θ̇2 = cos θ
4` 4`

The two θ̇2 versus θ curves do cross, so λ1 → 0 at

3g 3g 2
θ̇2 = cos θc = (1 − cos θc ), cos θc =
4` 2` 3
What about λ2 ? Set it to zero

3g g (1 − 34 sin2 θ)
λ2 = 0 = −m`(cos θ θ̇2 + sin θ θ̈) + mg = −m`(cos θ θ̇2 + sin θ ( sin θ)) + mg, θ̇2 =
4` ` cos θ

22
so we now solve
g (1 − 34 sin2 θc ) 3g 1
θ̇2 = = (1 − cos θc ), (1 − 3 cos θc )2 = 0, cos θc =
` cos θc 2` 3
since cos−1 (2/3) = 0.841rad < cos−1 (1/3) = 1.23rad you can judge which will be broken first. Once the first con-
straint breaks, you need completely new EOMs to describe the subsequent motion.

y
Example; perennial qualifier problem. A stick rests in a
vertical position on a frictionless surface, and begins to topple.
θ At what angle does it lose contact with the surface (if ever it
does)? We need a constraint connected to y-motion

1 2 1 `
L= Ic θ̇ + m(ẋ2 + ẏ 2 ) + λ(y − cos θ) − mgy
x 2 2 2
` `
mẍ = 0, Ic θ̈ = λ sin θ, mÿ = −mg + λ, y = cos θ
2 2
Eliminate y;
` m`  
Ic θ̈ = λ sin θ, − sin θθ̈ + cos θθ̇2 = −mg + λ
2 2
The normal force λ vanishes when contact is lost at θ = θc
at θc ; λ = 0, θ̈ = 0, cos θc θ̇c2 = 2g/`
Eliminate λ from the EOMs to get a first integral
`
Ic θ̈θ̇ = λ sin θθ̇ = −λẏ, mÿ ẏ = −mg ẏ + λẏ
2
1 2 1 1 1 `
add and integrate; Ic θ̇ + mẏ 2 + mgy = Ic θ̇02 + mẏ02 + mgy0 = mg
2 2 2 2 2
Evaluate at the critical angle (and incorporate constraints)
1 2 1 ` 2 ` ` mg`(1 − cos θ)
Ic θ̇c + m sin θc θ̇c + mg cos θc = mg , θ̇2 = 2 2
2 2 2 2 2 Ic + m`
4 sin θ

10

6 and eliminate θ̇c to get a single equation for cos θc . There

` 2
2g θ̇

is no real solution if the lower end of the stick slides

4 freely. The red curve is cos θc θ̇c2 = 2g/` and the blue is
θ̇2 = mg`(1−cos
m`2
θ)
2
1
for Ic = 12 m`2 .
Ic + 4 sin θ

0
0 0.2 0.4 0.6 0.8 1
cos θ
Suppose that the lower end of the stick is horizontally fixed, by a small bump or stop or wall on the surface,
so that x = 2` sin θ. This will change the Lagrangian and EOMs
1 2 1 ` `
L= Ic θ̇ + m(ẋ2 + ẏ 2 ) + λ(y − cos θ) − mgy + µ(x − sin θ)
2 2 2 2

23
 ` ` ` `
mẍ = µ, Ic θ̈ = λ sin θ − µ cos θ, mÿ = −mg + λ, y= cos θ x= sin θ
2 2 2 2
and the first integral to (show this)
1 2 1 1 1  `2  2 ` `
Ic θ̇ + m(ẏ 2 + ẋ2 ) + mgy = Ic θ̇2 + m θ̇ + mg cos θ = mg
2 2 2 2 4 2 2
10

8 Now there will be insufficient centripetal force on the COM

to keep it moving in a circle as it speeds up, it is pulled off
of the ground. From the constraints and EOMs
6
2g (4Ic + m`2 cos2 θ)
` 2

2mg` sin θ
2g θ̇

θ̇2 = , θ̈ =
` cos θ(4Ic + m`2 ) 4Ic + m`2
4
4mgIc sin θ
µ=−
cos θ(4Ic + m`2 )
2
1 2
and for a stick with Ic = 12 m` the solution is cos θc = 1/3.

0
0 0.2 0.4 0.6 0.8 1
cos θ
Here are the details from lecture for future reference; first
` `  ` 
ẋ = θ̇ cos θ, ẍ = θ̈ cos θ − θ̇2 sin θ , µ = m θ̈ cos θ − θ̇2 sin θ
2 2 2
` 
λ = mg + mÿ = mg − θ̈ sin θ + θ̇2 cos θ
2
so when
2g/` − θ̇2 cos θ
λ=0 =⇒ θ̈ =
sin θ
and at that time as well
` `2  
Ic θ̈ = 0 − µ cos θ = −m cos θ θ̈ cos θ − θ̇2 sin θ
2 4
or
m`2
sin θ cos θθ̇2
λ=0 =⇒ θ̈ = 4 m`2
Ic + 4 cos2 θ
eliminate θ̈ between these two constraint equations to arrive at the final constraint equation to be compared to the
first integral
2g 2
(Ic + m` 2
4 cos θ)
θ̇2 = ` 2
cos θ(Ic + m `4 )
1
`
This is the red curve in the figure, the blue one is the first integral. For Ic = m 12 the curves intersect at
3g g (1 + 3 cos2 θc ) 1
θ̇2 = (1 − cos θc ) = , cos θc =
` 2` cos θc 3

5.1 Nonholonomic constraints

The incorporation of non-holonomic constraints using Lagrange multiplies λ can (but does not necessarily) lead to
equations of motion involving λ and λ̇, since non-holonomic constraints can’t be integrated until the equations of
motion are actually solved. There is an example in Goldstein;
1
T = m(ẋ2 + ẏ 2 ), V = −mgy, ẋẏ = ky
2

24
 =⇒ mẍ − λ̇ẏ − λÿ = 0, mÿ − λ̇ẋ − λẍ = mg + kλ, ẋẏ = ky
Even a simple rolling condition that really is holonomic can lead to such EOMs;
1 1
L= mẋ2 + I θ̇2 + λ(ẋ − aθ̇) =⇒ mẍ + λ̇ = 0, I θ̈ − aλ̇ = 0, ẋ = aθ̇
2 2
However non-holonomic constraints that are linear in velocities can be built-in in such a way that derivatives of
Lagrange multipliers (forces of constraint) do not appear in the EOMs, for example see the solution to PS-1.1.

5.2 Virtual (infinitismal) work and external forces

This example for handling externally applied tensions was not covered today, but it is useful for systems with
externally applied forces (open systems). Suppose that several external forces Fi,ext are applied to a dynamical
system with kinetic energy T and potentials V , that are nice functions of some set of generalized coordinates
{qj |j = 1, 2, · · · , n} and {q̇j |j = 1, 2, · · · , n}.Include the associated generalized forces from the virtual work for the
forces Fi,ext , applied to the ith mass, technically returning to d’Alembert’s principle
Z X Z X Z X
dri
Fi,ext · dri = Fi,ext · dqj = Qj dqj
i i,j
dqj j

(not that these forces do work on the system of particles) for which
X dri
Qj,ext = Fi,ext ·
i
dqj

and incorporate them into d’Alembert’s principle (sum on repeated indices)

∂ri d  ∂T  ∂T 
ṗi · δri = mi ṙi · δqj = − δqj = Qj δqj
∂qj dt ∂ q̇j ∂qj
∂V
along with internal conservative forces with associated generalized forces Qj,c = − ∂qj

d  ∂T  ∂T d  ∂L  ∂L
− = Qj,ext + Qj,c or − = Qj,ext
dt ∂ q̇j ∂qj dt ∂ q̇j ∂qj

Here is a classic example of a difficult to explain 201/207/247 problem. A

spool rests upon a surface and is pulled by a tension F , but it rolls because of
ri friction with the table. Find the force of friction. Let the contact point begin
at x = 0, and let the angle θ of rotation be measured counterclockwise.
We will have a constraint for rolling which is non-holonomic at first sight,
F so treat it that way despite it being integrable to a holonomic
ro ẋ + ro θ̇ = 0

and if we let x and θ increment virtually by dx and dθ, the tension F does work since the point on the string at
which it was applied to the body moves in space by dx + ri dθ, so the work done to displace this point is
 
dW = F dx + ri dθ

Here is our Lagrangian, including the constraint coupled to a non-propagating degree of freedom (the normal force)
1 1
L= mẋ2 + I θ̇2 + λ(x + ro θ) − Vext (F )
2 2

25
and our equations of motion
∂Vext ∂Vext
ẍ = −ro θ̈, mẍ + λ = = Qx , I θ̈ + λro = = Qθ
∂x ∂θ
Now we obtain the generalized forces by matching coefficients of independent differential displacements (as if they
were linearly independent, lets remember this!)
 
dW = F dx + ri dθ = Qx dx + Qθ dθ, Qx = F, Qθ = ri F

Now we solve the system

ẍ = −ro θ̈, mẍ + λ = F, I θ̈ + λro = ri F
resulting in
(ri − ro ) ro (ro − ri ) I + mro ri
θ̈ = F, ẍ = F, λ= F
I + mro2 I + mro2 I + mro2
from which you see that the disk rolls clockwise (negative counter-clockwise), note that θ is measured positive if
CCW (positive angle measured from the x-axis, so the axis of rotation is the z axis). The beauty of the example is
that the correct directions for the motion and forces result from doing the math.

5.3 A note on the Jacobi integral

Take careful note of how we computed a ”first integral” (a conserved quantity generated as a constant of integration)
in the case in which there are forces of constraint. We first use the constraint to obtain a system of equations linear
in the constraints, so that they may be eliminated from the system algebraically.
Example. Imagine a mass m confined to slide frictionlessly on the inside of a paraboloid z = as2 . In cylindrical
coordinates
1
L = m(ṡ2 + s2 φ̇2 + ż 2 ) − mgz + λ(z − as2 )
2
Equations of motion
d
z = as2 , (ms2 φ̇) = 0, mz̈ = −mg + λ, ms̈ = msφ̇2 − 2asλ
dt

To obtain a Jacobi integral eliminate φ̇ and λ using solutions to the equations of motion (work ”on-shell”) ` = ms2 φ̇,

`2
mṡs̈ = ṡ − 2asṡλ
2ms3
mż ż = −mg ż + λż add
d 2 2 d `2 d
m
2 dt (ṡ + ż ) = (−mgz − 2
) + λ (z − as2 )
dt ms dt
m 2 `2
2 (ṡ + ż 2 ) = −mgz − + E integrating, using z = as2 , finally
ms2
m 2 `2
2 (ṡ + (2asṡ)2 ) = −mgas2 − +E
ms2
Since E = ẋ ∂L
∂ ẋ − L reverses the signs of potential terms in L but not kinetic terms, be careful about incorporating
constants of the motion into kinetic terms in L to obtain effective actions for the remaining variables: these terms
will not get the sign reversal apropos of potentials in subsequent calculation of the Jacobi integral, so be careful here.
.

26
6 Lecture 6. Rolling and rotating bodies
A mass m at r on a rigid body that is executing rotational motion with angular velocity ω about a fixed point will
have kinetic energy
1 1    
T = mv · v = m ω × r · ω × r
2 2
1  2 2 
= m ω r − (ω · r)2
2
1 X   1 1
= m ωi ωj δij r2 − ri rj = ω · I · ω = ω · L
2 i,j
2 2

introducing the moment of inertia tensor and the angular momentum vector. The moment of inertia tensor for
collections of masses or for uniformly distributed mass is
X   Z  
2
I= mk δij rk − rk,i rk,j , I = d3 r ρ(r) δij r2 − ri rj
k

A very useful example is that of a disk of mass m and radius a; using body fixed polar coordinates x = ρ cos φ and
y = ρ sin φ, z = 0 for a disk in the xy-plane
  M a2 
ρ2 sin2 φ

Z a Z 2π −ρ2 sin φ cos φ 0 4 0 0
M M a2
I= ρdρ dφ  −ρ2 sin φ cos φ ρ2 cos2 φ 0 = 0 0 
 
πa2 0 0 2
4 2
0 0 ρ 0 0 Ma
2

the moment of inertia of a disk spun at ω about an axis perpendicular to the disk through its center is I = 12 M a2 ,
and rotation about a diameter has half the moment of inertia.
The computation of the components of the ω vector need to be performed in such a way that the most general
rotational motion is described. This is actually pretty easy to do using the Euler angles for a rotating body. These
are rotations about body-fixed axes, but I intend to work in an inertial frame (space-fixed) for the moment.
As I pointed out in class, this is a standard way to parameterize a rotation by using two angles φ, θ to orient a
moving spin axis, and a rotation ψ about the moving rotation axis. Parameterizations dont’ ”commute”, so different
parameterizations and order of operations lead to different but equivalent descriptions. We want this description to
agree with Euler’s which is presented in chapter 4.
z, zb

ωz We begin the with actual spin rate about the symmetry axis of the body,
with the object oriented in an inertial frame so that {xb , yb , zb } coincide with
inertial axes {x, y, z}. Give the body a spin rate φ̇ about its zb = z axis
yb when φ = 0, so that
φ
y ωSF = (0, 0, φ̇)T
let it rotate by φ, so xb points in the (cos φ, sin φ, 0)T space-fixed direction.
x xb
z,
z zb
zb
ωz yb The next stage is to give it a right-handed spin rate θ̇ about the xb body
axis, which results in a total angular velocity vector
θ
ωSF = (θ̇ cos φ, θ̇ sin φ, φ̇)T
y
(components are with respect to space-fixed axes. Now it’s normal points
in the (sin θ sin φ, − sin θ cos φ, cos θ)T space-fixed direction.
x xb
At the instant that the body is tipped at θ, a right-handed rotation about x = xb axis, give it a spin at ψ̇ about the
body-fixed zb axis at the instant ψ = 0, noting that this axis has is space-fixed (sin θ sin φ, − sin θ cos φ, cos θ), at rate

27
ψ̇. The total angular velocity vector is now

ωSF = (θ̇ cos φ + ψ̇ sin θ sin φ, θ̇ sin φ − ψ̇ sin θ cos φ, φ̇ + ψ̇ cos θ)T

in the space-fixed coordinate system. Let’s play again, but in the body-fixed system. I will use rotation matrices
alluded to in class to find the axes, then I will return to my more intuitive procedure, reversing the steps of virtual
angular displacements done for the BFF case.
First a spin rate φ̇ is given about the space-fixed z-axis, which has components
      
cos ψ sin ψ 0 1 0 0 cos φ sin φ 0 0 sin ψ sin θ
 − sin ψ cos ψ 0   0 cos θ sin θ   − sin φ cos φ 0   0  =  cos ψ sin θ 
0 0 1 0 − sin θ cos θ 0 0 1 1 cos θ
 
sin ψ sin θ
in the body-fixed frame so we are at ωBF = φ̇  cos ψ sin θ . Now we rotate about the body-fixed axis whose
cos θ
space-fixed components are (cos φ, sin φ, 0), so find the body-fixed components
      
cos ψ sin ψ 0 1 0 0 cos φ sin φ 0 cos φ cos ψ
 − sin ψ cos ψ 0   0 cos θ sin θ   − sin φ cos φ 0   sin φ  =  − sin ψ 
0 0 1 0 − sin θ cos θ 0 0 1 0 0
 
φ̇ sin ψ sin θ + θ̇ cos ψ
so we are at ωBF =  φ̇ cos ψ sin θ − θ̇ sin ψ . Now we rotate about the body-fixed z axis at rate ψ̇ to arrive at
φ̇ cos θ
 
φ̇ sin ψ sin θ + θ̇ cos ψ
ωBF =  φ̇ cos ψ sin θ − θ̇ sin ψ 
φ̇ cos θ + ψ̇

The more intuitive way from class; rotate by ψ̇ about zb = (0, 0, 1)T . Just before this is applied (along with right-
handed displacement by ψ about zb ), the xb axis was (cos ψ, − sin ψ, 0)T when we applied θ̇, before the resulting
displacement by θ about this axis, the zb axis was (sin θ sin ψ, sin θ cos ψ, cos θ)T when we applied φ̇, so
     
0 cos ψ sin θ sin ψ
ωBF = ψ̇  0  + θ̇  − sin ψ  + φ̇  sin θ cos ψ 
1 0 cos θ

The kinetic energy is a rotational (orthogonal) invariant, and we know the moment of inertia in the body-fixed frame
only, so we use body-fixed ω components to compute T

1  M a2  2  M a2  2 
T = θ̇ + φ̇2 sin2 θ + ψ̇ + φ̇ cos θ
2 4 2
Suppose that this disk is skidding freely on the ground, we would have a combined rotational plus center of mass
kinetic energy plus a holonomic constraint z = a sin θ (incorporated directly without a multiplier, by substitution)

1  M a2  2  M a2  2  1  
L= θ̇ + φ̇2 sin2 θ + ψ̇ + φ̇ cos θ + m ẋ2 + ẏ 2 + a2 θ̇2 cos2 θ − mga sin θ
2 4 2 2
Let’s double-check something; what is the components of the body’s z axis in the space-fixed frame? In the body
fixed it is (0, 0, 1)T , so apply the inverse transformation
      
cos φ − sin φ 0 1 0 0 cos ψ − sin ψ 0 0 sin φ sin θ
 sin φ cos φ 0   0 cos θ − sin θ   sin ψ cos ψ 0   0  =  cos φ sin θ 
0 0 1 0 sin θ cos θ 0 0 1 1 cos θ

28
Phew! that worked out as expected. We set things up so that the lowest point on the disk (before we spin it at ψ̇)
was body-fixed (0, −a, 0)T which in the space-fixed frame is at (disk center at (0, 0, 0) we wil raise it)

(xc , yc , zc )T = (a sin φ cos θ, −a cos φ cos θ, −a sin θ)T

relative to the disk center, which has not moved yet. Yes, this really works out to be the lowest (contact point) of
the disk, we explicitely showed this in class by applying the virtual displacements by φ and θ (the ones that re-orient
the disk) in class.

This contact point will have velocity relative to the disk center of
   

 i j k 
 aψ̇ cos φ + aφ̇ cos φ cos θ − aθ̇ sin φ sin θ
vc = ω × rc =  θ̇ cos φ + ψ̇ sin θ sin φ θ̇ sin φ − ψ̇ sin θ cos φ φ̇ + ψ̇ cos θ  =  aψ̇ sin φ + aθ̇ cos φ sin θ + aφ̇ sin φ cos θ 

 a sin φ cos θ −a cos φ cos θ −a sin θ  −aθ̇ cos θ

Add onto this the center of mass velocity and the constraint z = a sin θ and we get our rolling constraints, a
stationary point of contact
vcom + ωSF F × rc = 0
In component form

0 = ẋ + aψ̇ cos φ + aφ̇ cos φ cos θ − aθ̇ sin φ sin θ, 0 = ẏ + aψ̇ sin φ + aθ̇ cos φ sin θ + aφ̇ sin φ cos θ

which certainly look non-holonomic, I don’t even want to check, but taking linear coombinations we get something
less icky, if fact they look reasonable

0 = ẋ cos φ + ẏ sin φ + aψ̇ + aφ̇ cos θ, 0 = ẋ sin φ − ẏ cos φ − aθ̇ sin θ

Associate with each of these constraints a force (not a Lagrange multiplier, these will not appear in a Lagrangian)
that can do no work since they are ”conjugate” to generalized displacements that are in fact zero
   
dW = 0 = λ dx cos φ + dy sin φ + a dψ + a dφ cos θ + µ dx sin φ − dy cos φ − a dθ sin θ

= Qx dx + Qy dy + Qφ dφ + Qθ dθ + Qψ dψ

Qx = λ cos φ + µ sin φ
Qy = λ sin φ − µ cos φ
Qφ = λa cos θ
Qθ = −µa sin θ
Qψ = λa

all of which look almost expected; λ is a tangential force that is trying to torque the spin ψ̇ of the disk. Given the
initial complexity,
 all of this comes as a bit of a relief (or let-down). These generalized forces appear on the right
d ∂L ∂L
hand side of dt ∂ q̇j − ∂q j
= Qj , which is the proper way to incorporate non-holonomic constraints. In summary:
holonomic constraints are included in the Lagrangian using Lagrange multipliers, non-holonomic constraints are
handled by d’Alembert’s principle as external forces that can do no work given that they are associated with vanishing
generalized displacements.

29
7 Physics 711 Lecture 7
7.1 Constants of the motion. Noether’s theorem
First integral of the motion are usually conserved quantities, and such objects are always connected with transfor-
mations that leave the action invariant. Let’s have a look at such transformations. Conservation laws emerge when
the right side of the Lagrange-Euler equations of motion vanishes
d  ∂L  ∂L
= =0
dt ∂ q̇ ∂q
   
∂L ∂L
then ∂ q̇ = ∂ q̇ , and the variable q is called cyclic.
t0 t1
Example.
1 2  p 1 2  d  ∂L 
L= ẋ + ẏ 2 − V ( x2 + y 2 ) = ṙ + r2 φ̇2 − V (r), =0
2 2 dt ∂ φ̇
and so ` = mr2 φ̇ (angular momentum) is conserved for central potentials.

There is a deep connection between symmetry group operations that leave S invariant (or L too, in a simpler
scenario), and conserved quantities. Consider

g : x → x0 = x + δx, t → t0 = t + δt, y → y 0 = y + δy

in which each function δt, δx, δy can all be functions of t, x but are all infinitismal, and restrict attention to transforma-
tions that leave the action invariant (in the homework you will study the simpler version in which the transformations
leave L invariant or at worst are gauge transformations of L)
Z Z
L0 (x0 , ẋ0 )dt0 − L(x, ẋ)dt = 0
I0 I

In the first integral t0 is a dummy; Z Z

L0 (x0 , ẋ0 )dt − L(x, ẋ)dt = 0
I0 I
in this next line δx is not a variation in the path;
Z   Z Z t1 +δt1   Z t1
L(x, ẋ) + δL(x, ẋ) dt − L(x, ẋ)dt = 0, L(x, ẋ) + δL(x, ẋ) dt − L(x, ẋ)dt = 0
I0 I t0 +δt0 t0

to first order in infinitismals

Z t1 Z t1
d 
0 ≈ δt L(x, ẋ) dt + δL(x, ẋ)dt
t0 dt t0
Z t1 Z t1 
d  ∂L(x, ẋ) ∂L(x, ẋ) 
0 ≈ δt L(x, ẋ) dt + δx + δ ẋ dt
t0 dt t0 ∂x ∂ ẋ

as long as we stay on a solution curve of the Euler-Lagrange equations of motion;

Z t1  Z t1 
d  d ∂L(x, ẋ) ∂L(x, ẋ) 
0 ≈ δt L(x, ẋ) dt + δx + δ ẋ dt
t0 dt t0 dt ∂ ẋ ∂ ẋ
Z t1 
d ∂L(x, ẋ) 
0 ≈ δt L(x, ẋ) + δx dt
t0 dt ∂ ẋ

so that we can conclude that the objects

∂L(x, ẋ)
C = δt L(x, ẋ) + δx
∂ ẋ

30
are conserved quantities for any such transformations leaving I invariant. Note that δx is a potentially time-dependent
quantity, but δx is computed at a fixed time on the E-L path.
In this formula δ x̄ is the variation in x at time-slice t induced by our original coordinate and time translations,
subject to staying on the Lagrange-Euler solution curve. Some transformations will advance us along the E-L curve
to later times, and we must undo this to pull back to t.
Suppose we examine x → x0 = x and t → t0 = t + .

E-L path
At time t in the integrand, we are at x(t), and make the transfor-
t + 2 mations x → x0 = x and t → t0 = t + , which does not change x,
however it is equivalent to advancing us along the E-L trajectory to
x(t + ) x(t + ), so pull this back to the t time-slice (dotted) and compute
t+
δ x̄, undoing the part of the coordinate transformation induced by the
δ x̄ time transformation, to get the ”pure” coordinate transformation for
t a fixed time, the time at which the integrand is evaluated.
x(t) = x0 (t)
Example. Suppose that δt(t, x) = , a simple translation (global) in time. If there is no change in x other than
what is induced by this change in t (induced by having x be a solution to the E-L EOMs), then

0 = δx + (x(t + δt) − x(t)) = δx + ẋδt = δx + ẋ, δx = −ẋ

(change in x is change at fixed time plus the change due to variation of the time) and we find that
 ∂L(x, ẋ) 
C1 =  L − ẋ = −h
∂ ẋ
is a constant, and using L = T (ẋ) − V we deduce that E = T + V is a constant if the Lagrangian is invariant under
time-translations and if V is independent of particle velocities (oops, I gave away a problem solution). This is the
Jacobi integral.
Example. Suppose that δt(t, x) = 0, and the intrinsic change in x is δx = , a rigid translation, then

∂L(x, ẋ)
 =  p, C2 = p
∂ ẋ
the canonical momentum, is conserved under rigid translations. Rs dt dx
Example. Lorentz transformation. Consider an action I = s01 L(t, x, ṫ, ẋ) ds, ṫ = ds , ẋ = ds , that is left
invariant under the action of a group of transformations G
1
g ∈ G, g : x → x0 = γ(x − vt), t → t0 = γ(t − vx), s → s0 = s γ=√
1 − v2
2
Find the infinitisimal version of the transformation (i.e. v << 1). This is easy, since if v << 1, γ ≈ 1 + v2 + · · · ≈ 1
and so δt = −vx, δx = −vt, and δs = 0 as a given.
Construct a conserved quantity using Noether’s theorem. Of course s is an invariant but this is not the one that
emerges from Noether’s theorem. Note that δx = δ x̄ + x(s + δs) − x(s) = δ x̄, and δt = δ t̄ + t(s + δs) − t(s) = δ t̄, and
ẋ = dx dt
ds , ṫ = ds
∂L ∂L  ∂L ∂L 
C = Lδs + δ x̄ + δ t̄ =v t +x
∂ ẋ ∂ ṫ ∂ ẋ ∂ ṫ
p
2 2
Interpret this conserved quantity for L = ṫ − ẋ , what is its physical meaning? We have that
1  dx dt  1  dx 
C=q −t +x =q x−t = γ(x − tv)
dt 2
( ds ) − ( dx 2 ds ds 1 − ( dx 2 dt
ds ) dt )

which is a Lorent-boosted position: the particle position in the boosted frame is constant if there are no forces.

31
7.2 Some examples. Nonpropagating variables
Consider a coordinate transformation (a rotation) and introduce a non-propagating Lagrange multiplier that will
make equations of motion first order
x1 → x01 = cos θ x1 − sin θ x2 , x2 → x02 = cos θ x2 + sin θ x1
The infinitismal version (very small θ) is
x1 → x01 = x1 − θ x2 , x2 → x02 = x2 + θ x1 , δx1 = x01 − x1 = −θ x2 , δx2 = x02 − x2 = θ x1
which we usually write as X
δxi = ωij xj ≡ ωij xj , ω T = −ω
j
using the summation convention. To create a Lagrangian linear in ẋ, I introduce a ”non-propagating degree of
freedom” p (apriori independent of x, on equal footing as a dynamical DOF with x) such that the equations of
motion for a free particle become
d  ∂L  dp  ∂L  d  ∂L   ∂L 
= = = 0, =0= = ẋ − p
dt ∂ ẋ dt ∂x dt ∂ ṗ ∂p
and then apply our transformation and look for conserved quantities
Z Z 
1 
I = L dt = ẋj pj − pj pj dt
2
Z  Z 
d  1  1 
δI = 0 = (xj + ωjk xk ) pj − pj pj dt − ẋj pj − pj pj dt
dt 2 2
Z  Z
 d 
= ωjk ẋk pj dt = ωjk xk pj dt
dt
(the equations of motion have been used to get the last equation). Write
1 1 1
ωjk xk pj = ωjk x{k pj} + ωjk x[k pj] = ωjk x[k pj] , x{k pj} = xk pj + xj pk , x[k pj] = xk pj − xj pk
2 2 2
and we have a conservation law, because ω is an arbitrary antisymmetric matrix
Z
d  dJjk d 
0 = δS = ωjk x[k pj] dt, = x[k pj] = 0
dt dt dt
which we recognize as conservation of (orbital) angular momentum. You should demonstrate
P that angular momentum
is again conserved if you add in a potential function that is a function of xi xi ≡ i xi xi only. You can see the genesis
of the Hamiltonian approach in which no integration by parts is needed to construct the equivalent of the Lagrange-
Euler equations of motion. Instead we introduce the momentum variable and (to quote J.A.Wheeler) ”let it flap in
the breeze”, tying it to ẋ by seeking the configuration of x, p that minimizes I.
Example. This I did not do in lecture, but it is important and interesting. The Lagrangian that correctly produces
∇ · A = 0 and ∂µ ∂ µ Aν = 0, the equations describing charge-free space electric and magnetic fields, is
Z
1 ∂Aµ ∂Aν
L = − d3 x Fµν F µν , Fµν = −
4 ∂xν ∂xµ
in gauge with A0 = 0.
Z Z Z
1  1 2 
L= Ld3 x = − d3 x F0j F 0j + Fij F ij = d3 x E − B2
2 2
∂L(x0 ) ∂L(x0 )
= = πj (x)δ 3 (x − x0 ) = Ej (x)δ 3 (x − x0 ) (these are Gateaux or variational derivatives)
∂(∂0 Aj (x)) ∂F0j (x)
(momentum conjugate to Aj is Ej ), so that the Hamiltonian approach would start with
Z
1 2  1  1 
H = Ȧj E j − L = −Ej Ek g kj − E − B2 = E2 + B2 , H = d3 x E2 + B2
2 2 2
 
Note that E2 ± B2 are Lorentz invariant according to these constructions, and therefore so is E · B.

32
8 Lecture 8. The central force problem
We begin a study of the central force problem by reducing the two-body problem to an effective one-body problem.
The smart and stylish way to do this is by canonical transformation. I like to jump around the book a bit, but that
would be overdoing it at this point.

8.1 Center of mass separation

For our two particles m1 at r1 and m2 at r2 and define two new vectors
m1 r1 + m2 r2
R= , r = r12 = r1 − r2
m1 + m2
Then
m2 m1
r1 = R + r, r2 = R − r
m1 + m2 m1 + m2
and the total kinetic energy of the system is
1  1 1 m1 m2 2
m1 ṙ21 + m2 ṙ22 = (m1 + m2 )Ṙ2 + ṙ
2 2 2 m1 + m2
The first term is the kinetic energy of all of the mass m1 + m2 treated as though it were all concentrated at R, the
position of the center of mass. The second term is the kinetic energy of the reduced mass µ = mm11+m
m2
2
.
The Lagrangian separates into two parts, one for the center of mass motion, one for motion about the center of mass
1 1 µmT G
L= mT Ṙ2 + µṙ2 +
2 2 |r|

in which
m1 m2
mT = (m1 + m2 ), µ=µ=
m1 + m2
Using the Lagrange-Euler or even Newtonian procedure we obtain two sets of equations of motion
r
(m1 + m2 )R̈ = −∇R V = 0, µr̈ = −∇r V = −µmT G
|r|3

from which we see that the center of mass moves like a free particle, but the reduced mass does not.

8.2 Relative motion

In spherical coordinates
1  2 
L= µ ṙ + r2 θ̇2 + r2 sin2 θφ̇2 − V (r)
2
and we see that the motion is confined to a plane θ = π/2;
1  2 
L= µ ṙ + r2 φ̇2 − V (r)
2
and has a conserved canonical (angular) momentum

d  ∂L  d 2 
= µr φ̇ = 0, ` = pφ = µr2 φ̇, ṗφ = 0
dt ∂ φ̇ dt

which is a first integral, an integrated EOM. And since the area swept out by the position vector in time dt is
1 1 1 1
dA = |r × dr| = |rr̂ × (drr̂ + rdφφ̂)| = |r2 dφk̂| = r2 dφ
2 2 2 2

33
we have that the rate at which the position vector sweeps out area is constant by virtue of ` being constant
dA 1 dφ `
= r2 =
dt 2 dt 2µ
Take the radial equation
∂V
µr̈ − µrφ̇2 + =0
∂r
multiply by ṙ
∂V d 1 2 
µr̈ṙ − µrφ̇2 ṙ + ṙ = 0, rewrite µṙ + V − µrφ̇2 ṙ = 0
∂r dt 2
however from ` conservation,

d  µr2 φ̇2 
2µrṙφ̇ + µr2 φ̈ = 0, µrṙφ̇2 + µr2 φ̇φ̈ = −µrṙφ̇2 , = µrṙφ̇2 + µr2 φ̇φ̈ = −µrṙφ̇2
dt 2
we obtain a second first integral (total energy)

d  1 2 µr2 φ̇2  1 2 µr2 φ̇2

µṙ + + V = 0, E= µṙ + +V
dt 2 2 2 2
and this allows a full reduction to quadrature;
Z r
1 ` dr
E = µṙ2 + + V, t=
2µr2
q
2 2
−V − `2
µ (E 2µr 2 )
r0

(good luck with that) which you must still invert to get r(t). Once a form for r(t) is known, φ(t) follows from
t φ
µr2
Z Z
` dt
dt = dφ, = dφ0
` t0 µr2 (t) φ0

I will concentrate on properties of bound solutions E < 0, especially of the Kepler problem. The trajectory
equation r = r(φ) can be used for scattering with minor modifications. In lecture I solved the equations of motion
with a variable change (yes Neil, this is math, the chain rule)

dr dφ dr ` dr
r = r(t), φ = φ(t) =⇒ = = 2
dt dt dφ µr dφ

8.3 The equation for the orbital trajectory

Begin with

1 2 `2 µr2 df (r) df (r) dφ ` df (r)

E= µṙ + + V, and change variables dt = dφ, = = 2
2 2µr2 ` dt dφ dt µr dφ
letting f = r;
µ  ` dr 2 `2
E= 2
+ + V (r)
2 µr dφ 2µr2
1
and change again, u = r
`2  du 2 `2 2
E= − + u + V (u)
2µ dφ 2µ
which can be integrated directly
Z φ Z u
du
dφ0 = q
φ0 u0 2µE 2µV (u)
`2 − `2 − u2

34
which can be completed in trigonometric functions for V = αu−n−1 for n = −3, −2, 1 or elliptic functions for
n = −7, −5, −4, 0, 3, 5. For example n = −2 is the Kepler problem V = − kr = −ku. Let φ = 0 where u = ua , the
minimal u (maximal r), and let E < 0
Z u  2u − u − u 
du a p (ua + up ) (up − ua )
φ − φ0 = p = sin−1 + π/2, u= − cos(φ − φ0 )
ua −(ua − u)(up − u) u p − u a 2 2

which is an ellipse. In terms of semimajor axis and eccentricity rp = a(1 − ) and ra = a(1 + )

1 +  cos φ
u= , φ0 set so that u = up when φ = 0
a(1 − 2 )

Also note that

1 2µE 2µk 2
−ua up = − = 2 , = ua + up =
a2 (1 − 2 ) ` `2 a(1 − 2 )
Divide
k 2E`2
E=− , 1 − 2 = −
2a µk 2
The period we can get from the areal velocity
dA ` πab `
= , =
dt 2µ T 2µ
Use
4π 2 µ2 a4 (1 − 2 ) 1 − 2 2E 1
T2 = and =− 2 =+
`2 `2 µk µak
and we get Keplers’ third law
s r
4π 2 µa3 2π k GmT
period obeys T 2 = , or frequency ω= = =
k T µa3 a3

35
9 Lecture 9. Kepler problem in time
Return to V = − kr and a description of the motion in time.
From the previous discussion we see that if we can get r(t)
we can find φ. It is indicative of the superintgeability of
the Kepler problem that you can find r(t) to whatever a
accuracy that you wish by introducing this auxillary an-
gle, the eccentric anomaly (anomalies in QFT/HEP are
obstructions to conservation laws, i.e. current anomalies, ψ φ
conformal anomalies, anomalies in astrophysics are angles) a r cos φ

r = a(1 −  cos ψ), <1

Big right triangle;

a + r cos φ = a cos ψ
and since
1 1 +  cos φ a − a2 − r cos ψ − 
= , = r cos φ, cos φ =
r a(1 − 2 )  1 −  cos ψ
the equation r = a(1 −  cos ψ) follows. This last relation between φ, ψ (note both are zero at perihelion) should
look very familar (think about the angle at which light is emitted in a collision in the lab frame versus some frame
moving at v = βc). This angle is the trig substitution needed to integrate the radial r versus t equation
Z r r Z r
dr µ dr
t= q = q
2 `2 2 `2
E + kr − 2µr
µ (E − V − 2µr 2 )
r0 r 0 2

and we start at perihelion r0 = a(1 − ) where ψ = 0, with dr = a sin ψ dψ

r Z r
µ dr
t = q
2 r0 E + k − `2
r 2µr 2
r Z r
µ r dr
= q
2k r0 r − r2 − a (1 − 2 )
2a 2
r Z ψ
µ a2 (1 −  cos ψ) sin ψ dψ
= q
2k 0 a − a cos ψ − a2 (1 − 2 cos ψ + 2 cos2 ψ) − a
+ a2
2 2
r Z ψ 2
µ a (1 −  cos ψ) sin ψ dψ
= q
2k 0 a2 2
2 (1 − cos ψ)
r
µa3 ψ
Z
1
= (1 −  cos ψ) dψ = (ψ −  sin ψ)
k 0 ω
This last equation
ωt = ψ −  sin ψ
can be solved numerically for ψ given t by iteration to any desired accuracy, and r, φ plotted very nicely from ψ, or
it can be solved formally and the formal expansion can be used to compute ψ as a function of t, from which we get
r(t) and φ(t) directly. Both approaches are easy to apply to any desired precision, you will take one in the homework
and I will expand the other right now. Start with
ωt = θ = ψ −  sin ψ
and note that as t ranges over the orbital period [0, T ] then 0 ≤ θ ≤ 2π. and 0 ≤ ψ ≤ 2π. Expand in a Fourier series
∞
X
eiψ = an einθ
n=−∞

36
 Z 2π
1
invert an = eiψ e−inθ dθ
2π 0
Z 2π
1
Change variables an = eiψ e−inθ (1 −  cos ψ)dψ
2π 0
Z 2π
1
an = eiψ e−in(ψ− sin ψ) (1 −  cos ψ)dψ
2π 0
2π
−1
Z
d  −in(ψ− sin ψ) 
= e eiψ dψ
2πin 0 dψ
2π
−1 iψ −i(ψ− sin ψ) 2π
Z
1
= e e  + eiψ e−in(ψ− sin ψ) dψ
2πin 0 2πn 0
Now use some identities from the theory of Bessel functions
∞
X 2m
eiz sin ψ = Jm (z) eimψ , J−m (z) = (−1)m Jm (z), Jm (−z) = (−1)m Jm (z), Jm−1 (z)+Jm+1 (z) = Jm (z)
m=−∞
z
Z 2π Z 2π
1 iψ −in(ψ− sin ψ) 1 X Jn−1 (n)
an = e e dψ = e−i(n−1)ψ Jm (n)eimψ =
2πn 0 2πn 0 m
n

The n = 0 term must be evaluated carefully (use limn→0 sin(n)

n =  and J1 (0) = 0),
∞
X Jn−1 (n) inθ
eiψ = e
n=−∞
n
∞ 
X Jn−1 (n) J−n−1 (−n) −inθ 
= (n=0 term) + einθ + e
n=1
n −n
∞ 
X Jn−1 (n) (−1)n+n Jn+1 (n) −inθ 
= (J1 (0) + J−1 (0)) + einθ + e
n=1
n −n
∞ 
X Jn−1 (n) Jn+1 (n) 
Im(eiψ ) = sin ψ = sin(nθ) + sin(−nθ)
n=1
n −n
∞ 
X Jn−1 (n) Jn+1 (n) 
= + sin(nθ)
n=1
n n
∞
X 2Jn (n)
= sin(nθ)
n=1
n

J−1 (n)
The n = 0 term will be (after taking imaginary parts) limn→0 n sin(nθ) = 0, and we have a very rapidly
converging (for small ) expansion for ψ as a function of t
∞
X 2
ψ = ωt + Jn (n)) sin(nωt)
n=1
n

so to figure out exactly where on an orbit the planet is after t = 0 when ψ = 0 and r = rp = a(1 − ) we select t,
compute ψ, then get
cos ψ −  p
r = a(1 −  cos ψ), cos φ = or miraculously x = a(cos ψ − ), y = a 1 − 2 sin ψ
1 −  cos ψ
(keep in mind the sun is at the origin of the Cartesian coordinate system, an orbital focus).

A very useful special functions reference; N. N. Lebedev Special Functions (Dover), no graduate student should leave
home without it. A copy is in Canvas/Files for you.

37
10 Physics 711 Lecture 10
We wrap up E < 0 central force motion, focusing on planetary motion in the Newtonian gravitational field.

10.1 Stable and closed orbits in general

Circular orbits have both r̈ = 0 and ṙ = 0, requiring
 ∂V   `2    
µr̈ = 0 = − + , E= m
2 ṙ2 + r2 φ̇2 + V (r0 ) = m 2 2
2 r0 φ̇ + V (r0 )
∂r r0 µr3 r0 r0

The second equation gives the energy of the circular orbit. From the radial equation

∂V ∂V `2
µr̈ = − + µrφ̇2 = − + 3
∂r ∂r µr

we get the primary condition for the class of circular orbits (which are a class of closed orbits) at r = r0

∂V  `2
0=−  + 3
∂r r0 µr0

and these will be stable (perturbations in r will experience a restoring force) if the total radial potential

`2
V0 =V +
2µr2
is ”convex upwards” at r0
∂ 2 V 0  ∂ 2 V  3`2
 ≥ 0,  + 4 ≥0
∂r2 r0 2
∂r r0 µr0
Rewrite this as
  
2
∂ V  3`2 3 ∂V   dFr   3   r0 − dF r
dr  d ln Fr 
 ≥− 4 =−  , −  ≥− − Fr  ,
   0 ≥ −3,  ≥ −3
2
∂r r0 µr0 r0 ∂r r0 dr 0 r0 0 d ln r 0
− Fr 

0

n
and for force laws Fr = kr , ln Fr = C + n ln r this requires

n ≥ −3

so Kepler, Hooke, everything that you know and love will have stable orbits (bounded) as you knew all along. But
what about closed orbits? Small perturbations about circular orbits will exhibit SHM, and if the SHM frequency is
commensurate with the orbital frequency, the orbits will close and repeat after a certain number of cycles.
The orbital equation is
dV ` d  ` dr   ` 2 dV
µr̈ − µrφ̇2 + = 0, µ − µr + =0
dr µr2 dφ µr2 dφ µr2 dr

`2 2 d2 u `2 3 dV d2 u µ dV µ
− u − u − u2 = 0, +u=− 2 = − 2 2 F (u) ≡ J(u)
µ dφ2 µ du dφ2 ` du ` u
Stable circular orbits have u0 = J(u0 ), expand around this

J(u) = J(u0 ) + (u − u0 )J 0 (u0 ) + · · ·

call ∆ = u − u0 and examine this SHM at frequency β

d2 ∆
= −∆ + J 0 (u0 )∆ + · · · = −β 2 ∆, β 2 = 1 − J 0 (u0 )
dφ2

38
and if β = p/q is rational, then after q orbits the trajectory will close and repeat. Pretty nifty!

d  µ  2µ µ
J00 = − 2 2 F = 2 3 F0 − 2 2 F00
du ` u 0 ` u0 ` u0

but
µ u0 0 d ln F  d ln F 
u0 = J0 = J(u0 ) = − F0 , =⇒ J00 = −2 + F0 = −2 +  = −2 −
` u20
2 F0 d ln u 0 d ln r 0


so we get closed orbits (eventually) for rational β and attractive potentials with

d ln F 
β2 = 3 +  = 3 + n, for F = −krn
d ln r 0

10.2 Spiral orbits. A simple inverse problem

We did not do to this example, we skipped over it to address precession. This is another inverse problem: given u(φ)
find V (u) such that u(φ) is an exact solution.
In the r-equation (call M = mT , total mass)

` d  ` dr  `2 GM µ
µ 2 2
= 2
−
µr dφ µr dφ µr r2

the radial force Fr = − GM

r2
µ d
could be replaced by any central force Fr = − dr V (r) resulting in

` d  ` dr  `2 d
µ = + V (r)
µr2 dφ µr2 dφ µr2 dr

follow this with our change of variables u = 1r , d

dr V
d
(r) = −u2 du V (u) to get

d2 u µ d
2
= −u − 2 V (u)
dφ ` du
The orbital problem for an arbitrary potential could be difficult to solve in the forward direction, but beginning with
an orbit u = u(φ) finding a potential function that produces such an orbit is trivial.

Consider a spiral orbit r = αφ (I think this is called Archimedes’ spiral, try hyperbolic r = a/φ or equiangular
r = aebφ ). What potential has such paths as trajectories?

1 d2 u 2
u= , = = 2α2 u3
αφ dφ2 αφ3
and so
µ d `2 2 `2 α2 4
− V (u) = u + 2α2 u3 , V (u) = − u − u
`2 du 2µ 4µ
You might want to try Cayley’s sextic r = 4a cos3 (φ/3), or Bernoulli’s lemniscate r2 = a2 cos(2φ), or the Rhodonea
r = a sin(kφ) (try k integer, rational, irrational).

39
10.3 Orbital precession
The uniformly precessing orbit has form

rp (1 + )
r=
1 +  cos(Ωφ)

(equivalent to a pure Keplerian elliptical orbit in a uniformly rotating

frame). If we begin the orbit at perhelion r = rp at φ = 0, it returns to
perihelion at φ = 2π Ω , which is a retrograde precession if Ω > 1, meaning
that if the planet/satellite is orbiting with φ increasing counter-clockwise,
the aphelion point orbits clockwise. Aphelion occurs at φ = π/Ω and
(1+)
ra = rp (1−) , the figure is rp = 1 and  = 0.6.

The angular displacement of perihelion per orbit is

2π
∆φ = − 2π
Ω
and the precession frequency is

∆φ (1 − Ω)
ωp = = 2π
T ΩT

Any departure whatsoever from a pure inverse-square force law will cause an orbit to precess (non-uniformly), but
”pure precession” (uniform) as I have illustrated can easily be found to be the exact solution of a particular potential
problem, which I will solve as an inverse problem so you get an example of such a process (get V from u(φ))

d2 u Ω2  1 
= − cos(Ωφ) = −Ω2 u −
dφ2 rp (1 + ) rp (1 + )
d2 u Ω2 µ dV
+u = (1 − Ω2 )u + =− 2
dφ2 rp (1 + ) ` du
`2  (1 − Ω2 )u2 Ω2 u  `2  (1 − Ω2 ) Ω2 
V = − + =− +
µ 2 rp (1 + ) µ 2r2 rrp (1 + )

and we can relate rp ,  to E, `; at perihelion

1 2 `2 `2 `2  (1 − Ω2 ) Ω2 
E= µṙ + + V (r) → − +
2 2µr2 2µrp2 µ 2rp2 rp2 (1 + )
or
( − 1)`2 Ω2 ` 2 Ω2 `2 Ω2 (1 − )
E= = − = −
( + 1)2µrp2 2µra rp 2µrp2 (1 + )
so if we call
rp ` 2 Ω2 k k ka(1 − 2 )(1 − Ω2 )
a= , k=− , then E = − , V =− −
1− µrp (1 + ) 2a r 2Ω2 r2
2 2
` Ω
and since rp = − µk(1+) , eliminate rp from E

`2 Ω2 (1 − ) `2 Ω2 (1 − ) µk 2 (1 − 2 ) 2|E|`2 Ω2
E=− =− 2 =− , 2 = 1 −
2µrp2 (1 + ) 2`2 Ω2 µk 2

`2 Ω2
2µ µk(1+) (1 + )

are the requirements for pure uniform precession of an otherwise elliptical orbit.

40
10.4 Runge-Lenz vector
For the Kepler problem start with
kr
L = r × p, ṗ = −
r3
and examine time derivatives of L × p, which is a vector in the plane of the orbit but ⊥ to the orbital path
 kr  k k
(L × p)· = L̇ × p + L × ṗ = L × − = 3 r × (r × p) = 3 (r · pr − r2 p)
r3 r r
but this last form is exactly equal/opposite the time derivative of kµ rr , which points from focal point to the planet
on the orbit;
 r · kµṙ r · ṙ kp r·p
kµ = − kµr 3 = − kr 3
r r r r r
and so the sum is the conserved Runge-Lenz vector
r
A = −kµ −L×p
r
and because it is conserved we can compute it at any convenient point, such as perihelion;

p2 k k
p = pj r = rp i = a(1 − )i, L = prp k, p = pj, − =− =E
2µ r 2a
so
r = rp i  rp 
A = − µki − 2µk(1 − i) = µki
2a
has magnitude given by the eccentricity and points from focus to
perihelion.
Example. Orbital corrections. We did not do this in class, but it illustrates something neat that the R-L vector
can do for you.
Where are you on your orbit? If you know E, , ` and r, your distance from the sun, can the angle φ between r and
true perihelion be found?
Dot r into A,

A · r = µki · r = µk r cos φ = −µkr − (L × p) · r = −µkr − (p × r) · L = −µkr + `2

therefore your position on the orbit is

`2 − µk r  2µEa r + `2 
cos φ = =−
µkr 2µEa r

(note the φ = 0 if you are at the perihelion point). Suppose that you make an orbital alteration that leaves ` fixed
but changes E, a,  (fire rockets radially), then note that

`2 − µk r
 cos φ = = 0 cos φ0
µkr

because the middle quantity is unchanged. That means if you are initially in a circular orbit,  = 0, then cos φ0 = 0
so φ0 = π2 ! A purely radial rocket burst will have the following outcomes;

41
 ∆p
a

a ∆p

In this picture we begin at ”a” with a purely

In this picture we begin at ”a” with a purely vertical vertical velocity and go into a circular orbit for
velocity and go into a circular orbit for one period. one period. Passing through ”a” we give the
Passing through ”a” we give the satellite an outward satellite an inward sudden push. In both cases
sudden push. We begin moving outward in the orbit. the angle between our position vector at the time
of the push, and the point of closest approach is π/2.

∆p
a a
∆p

Begin at ”a” with a purely vertical velocity and go

Begin at ”a” with a purely vertical velocity and go
into a circular orbit for one period. Passing through
into a circular orbit for one period. Passing through
”a” we give the satellite a sudden ”de-boost”, a
”a” we give the satellite an upwards (tangential)
retrograde tangential rocket burst, slowing it in
push, increasing its orbital speed. We begin moving
its orbit. Now the boost-point where the rocket
outward in the orbit, but the boost-point becomes
deployed becomes aphelion.
perihelion.

42
11 Lecture 11. Positive energy trajectories. Scattering
The area A(θ) is called the differential cross section, and is an annular region within the confines of a certain impact
parameter b and b + db, so that A(θ) = 2πb db, which we have projected onto the targets.
Consider classical scattering of particles of mass µ, initial speed v0 and impact parameter b by a static central force
field. If all particles with impact parameter in the range b → b + db are scattered into solid angle Ω → Ω + dΩ.

F
v
r ψ ψ Θ
b

Scattering cross-section represents the size of the target that an incoming projectile must hit in order to be scattered
through an angle between θ and θ + dθ. For a given target then if a projectile hits the ring of area 2πb db in the
figure, it will scatter through such an angle. Suppose that an incoming particle beam has a flux of F particles per
second per unit area. The rate with which projectiles scatter into θ → θ + dθ is
dN dN
= F · 2πb, = F · 2πb db
dt db dt
We translate this from a function of b to a function of θ with
dN db
= F · 2πb dθ
dt dθ
and now find the number of particles per unit time scattered into the cone of outer angle θ + dθ and inner angle θ,
which subtends a solid angle dΩ = 2π sin θ dθ;

d2 N d2 N db 1 b db
= dΩ = F · b (2π sin θ dθ) = F · dΩ
dt dt dΩ dθ sin θ sin θ dθ
and finally we define that quantity much loved by particle physicists; the differential cross section

dσ(θ) 1 d2 N b db
= =
dΩ F dt dΩ sin θ dθ
The trajectory of the particle can be gotten from the energy expression
µ 2
E= (ṙ + r2 φ̇2 ) + V (r)
2
and again we eliminate t in favor of φ using the angular momentum equation
dφ
` = µr2
dt
d ` d
or by inverting dt = µr 2 dφ . We find that

`2 dr 2 `2 `dr
E= ( ) + + V (r), or dφ =
2µr4 dφ 2µr2
q
2 `2
µr2 µ (E − V (r) − 2µr 2 )

43
 1
As usual set u = r and
u1
−du
Z
φ1 − φ0 = q
2µE 2µV
u0
`2 − `2 − u2
If we let the initial and final values of r be ∞ and the point of closest approach be r0 = rp , then we find that φ1 = π,
φ0 = θ + ψ and π = 2ψ + θ and we arrive at a ”potential to scattering angle” transform
Z umax
bdu
θ =π−2 q
0 1 − VE (u)
− b2 u2

with the angular momentum given by the asymptotic value

p
` = µbv0 = b 2µE
Here is an example by integral transform, the (repulsive) Coulombic potential V (u) = ku (which I did
by the asymptote method in class). First compute the point of closest approach
2b2 Eu + k
Z
1 bdu
r0 = , q = sin−1 ( √ )
umax 1 − ku − b 2 u2 k 2 + 4b2 E 2
E

The radial speed is zero at this point and so ṙ = 0 and

√
l2 u2max −k + k 2 + 4b2 E 2
E = kumax + , umax =
2µ 2b2 E
We can now compute the integral for the scattering angle and obtain
k
θ = π − 2 sin−1 (1) + 2 sin−1 ( √ )
k2 + 4b2 E 2
which can be solved for b in terms of θ
k θ
b= cot( )
2E 2
which in turn can be inserted into the cross section formula (Rutherford formula)
dσ(θ) k2 θ
= csc4 ( )
dΩ 16E 2 2

11.1 ”Hyperbolic” trajectories

An alternative approach (my favorite) that does not involve integral transformation is to use the orbital equation to
find the angle between the asymptotes of the hyperbolic path taken by the projectile.
The case of V = rk2 is a particularly simple example, since the orbital equation integrates
r
`2 d2 u `2 2kµ
−µ 2 u2 2 = µ 2 u3 + 2ku3 to u = u0 cos( 1+ θ)
µ dθ µ `2
which has two asymptotes r → ∞, u → 0 at
π
θ0 = ± q
θ0 2( 1 + 2kµ
`2 )
Θ θ0
so that the scattering angle is

π (1 − Θπ)
2
Θ=π− q for which b2 = kE −1 Θ Θ
π (2 − π )
2kµ
1+ `2

44
11.2 The inverse scattering transform
This is the ”potential to angle” forward transform that I mentioned in class, for a unit mass
1 2  1
ṙ + r2 φ̇2 + V (r) = E = v02
2 2
(a rescaling by E we mean E/m and by V we mean V /m) and

r2 φ̇ = ` = bv0

we obtain
dr √ dr
±p = dt, dφ = ∓b E √
2(E − V ) − 2Eb2 r−2 r2 E − V − Eb2 r−2
Let φ0 be the angle at which r = Rb , the largest zero of the radical (where ṙ = 0) then
Z ∞
dr
π − φ0 = φ0 − Θ = √
Rb r
2 b − r − V E −1 b−2
−2 −2

or Z ∞ Z ub
dr du
Θ=π−2 √ =π−2 p
Rb r2 b−2 − r−2 − V E −1 b−2 0 b−2 (1 − V (u)E −1 ) − u2
We write this as an invertible integral transform: let’s factor the radical
s
p u2 p
b−2 (1 − V (u)E −1 ) − u2 = b−2 − −1
1 − V (u)E −1
(1 − V (u)E )

and call
u2
W (u) =
1 − V (u)E −1
Notice that W (u) is well-defined on the interval [0, ub ], and in fact W (ub ) = b−2 and W (0) = 0, so we can write
Z b−2 du
Γ(W ) dW dW Γ(W ) du
Θ − θ0 = 2 √ , Γ(W ) = p , √ dW =
0 b−2 − W 1−V (u)E −1 W u

Another variable change; let W = x−2 , and Γ̄(x) = x−3 Γ(x−2 ) and we obtain an invertible Abel transform;
Z ∞
1 ∞ dF
Z
f (x) x dx dy
F (y) = 2 p =⇒ f (x) = − p
y
2
x −y 2 π x dy y − x2 2

(Θ − θ0 )
(b) = 2A[Γ̄](b)
2b
This allows one to completely reconstruct the potential V (r) from the scattering angle Θ, an inverse-scattering
transformation, which is the ultimate goal of a scattering experiment (to learn about the potential). This is much
harder to do in quantum mechanics (the quantum inverse scattering transform (QIST) is called the Gel’fand-Levitan-
Marchenko equation), but in classical physics it is very simple and procedural.
For example, begin with the result of one of our previous examples

π (Θ − π) π
Step1. Θ − π = − √ , (η) = − p = F (η)
kE −1 b−2 + 1 2η 2 kE −1 + η 2

Differentiate, and insert into the inverse transform

Z ∞
dF πη π η dη 1 1
Step2. = , f (b) = 3 =
2(kE −1 + η 2 )3/2 2 b2 + kE −1
p
dη 2·π b (kE −1 + η 2 ) 2 η 2 − b2

45
and now undo all of the variable transformations
1 1 1
Step3. r−3 Γ(r−2 ) = , r= √ , Γ(W ) = √
2(r2 + kE −1 ) W 2 W (1 + kW E −1 )

We find W = W (u);

Γ(W ) du dW du cW cu2
Step4. √ dW = , −1
= , u2 = , W (u) =
W u 2W (1 + kW E ) u 1 + kE −1 W c − kE −1 u2

and the final step is to recover V (u)

u2  u2 
Step5 W (u) = , V (u) = E 1 − = kc−1 u2
1 − V (u)E −1 W (u)

which is the correct potential for this scattering data.

46
12 Lecture 12. Scattering in the lab frame
We have been working in what is called the ”single-particle frame”, using
1 1
L= mT V 2 + µv 2 + V (|r|), (mT V )• = 0
2 2
The coordinate r is the position of projectile m1 relative to the origin that is r2 .
Single-particle coordinates
COM coordinates
Lab frame coordinates m1 r1 + m2 r2
R= , r = r1 − r2 µ µ
m1 + m2 r10 = r1 − R = r, v10 = v
r1 , r2 m1 m1
µ µ
r1 = R + r, r2 = R − r µ µ
v1 = ṙ1 , v2 = ṙ2 m1 m2 r20 = r2 −R = − r, v20 = − v
m2 m2
V = Ṙ
µ
Since v10
= m1 v, v10 ||v
at all times, therefore in the COM frame the scattering angle θ10 = Θ, the scattering angle in
the single-particle frame, and θ20 = π − Θ. In other words solving the scattering problem in the single-particle frame
gives the COM frame scattering angle. So what is the scattering angle θ1 of the projectile in the lab frame?

p01,f

θ10 θ1
p01,i θ20 p02,i θ2

s
COM
LAB
p02,f

Here is an important fact: V is parallel to the initial value of v1 . Call v1,i = v1 (−∞) and v1,f = v1 (∞) and so
forth. Since in the lab frame we have v2,i = 0, and V is constant since the only forces are functions of r
m1 v1 + m2 v2 m1 v1,i µ
V = = = v1,i
m1 + m2 m1 + m2 m2
and
µ
v1 = v10 + V =⇒ 0
v1,f = v1,f +V, 0
v1,f = vf
m1
we have the diagram

0 V v1,f
v1,f From the figure 0
v1,f cos θ10 + V = v1,f cos θ1
law of sines
θ1
θ10 sin θ10 sin θ1 sin θ10 cos θ1 sin θ1 sin θ10
= 0 , 0 = 0 , tan θ1 =
v1,f v1,f V + v1,f cos θ10 v1,f cos θ10 + vV0
v1,i 1,f

sin Θ sin Θ sin Θ

tan θ1 = = µ = m v
cos Θ + vV0 cos Θ + m2 v1,i
µ
cos Θ + m12 v1,i
f
m1 vf
1,f

47
12.1 The virial theorem
We did not cover this in lecture, I usually reserve the virial theorem for 715 since it deals with averaging in an
ensemble, but it can be applied to an average over time and so is appropriate for 711 as well. Consider the quantity
X
Q= pj · rj
j

summed over the particles in a system. Assume that each particle (j) is subjected to external forces Fj,ext and
internal interparticle forces, exerted for example by particle k through potentials

V (rj − rk ) = a |rj − rk |ν

This is why Q is significant: X

Q̇ = 2 T − νV + Fj,ext · rj
j

Proof.
X X
Q̇ = ṗj · rj + pj · ṙj
j j
X X
= Fj · r j + pj · ṙj
j j
X X X
= Fj,int · rj + Fj,ext · rj + pj · ṙj
j j j

The last term is the kinetic energy

X X 1
pj · ṙj = pj · ṗj = 2 T
j j
mj

The first is the potential energy for the potentials

X
V (rj − rk ) = a |rj − rk |ν , Fj = −νa (rj − rk ) |rj − rk |ν−2
k6=j

then
X XX
Fj · r j = −νa (rj − rk ) · rj |rj − rk |ν−2
j j k6=j
XX
= −νa (rj − rk ) · (rj − rk ) |rj − rk |ν−2
j k<j
XX
= −ν V (rj − rk ) = −ν V
j k<j

Next we show that

Q̇ = 2 T − νV − 3Pext,total V ol
if the system is enclosed in a box and the walls are under external pressure. A particle on a wall will experience
Fj,ext = −Pext,1 Awall n, where Pext,1 is the pressure per particle exerted by the wall. Add up all of these contributions
RT
over particles (n outward-direct, origin at box center), and time average over a very long time, hf i = T1 0 f (t) dt
X X
Fj,ext · rj = −Pext,1 Awall n · rj,wall
j j on wall
= −Nwall (t)Pext,1 Awall · 6 · (L/2) = −3Nwall (t) Pext,1 V
X
h Fj,ext · rj i = −hNwall (t)P1 iV
j
= −3Pext,T V ol

48
So what is this used for? In molecular dynamics one keeps track of particle positions and momenta, so the virial
theorem can be used to compute pressures. The virial hypothesis is that because any particle in a system confined
by walls or interparticle forces will sample a very large phase space (set of r, p values) over time
Z T
1
Q̇ dt = 0, 2 T − νV − 3Pext,total V ol = 0
T 0

In a system confined only by interparticle forces we have

0 = 2hT i − νhV i

which is especially useful for circular orbits in central potentials, since for such cases
ν ν 2+ν
hT i = T = hV i = V, E =T +V = V
2 2 2
As I mentioned above, the averages here can be time averages, or position averages
N N
1 X 1 X
hXi = lim X(ti ), hXi = lim X(ri )
N →∞ N N →∞ N
i i

We will use such averages quite frequently to separate secular (oscillatory) and nonsecular (accumulating) perturba-
tions by averaging over a period (or by averaging over eccentric anomaly values)
Z T Z 2π Z 2π
1 1 1
hXi = X(t) dt = X(ψ) · (1 −  cos ψ) dψ = X(ψ) · (1 −  cos ψ) dψ
T 0 ωT 0 2π 0

once we introduce these ideas in PS-4.2, and later in our discussion of canonical perturbation theory.

49
Rigid body dynamics
The 3N coordinates of the N masses making up the body have 12 N (N − 1) constraints |rj − ri | = cij , which is much
greater than 3N since they are not all independent. Select three ”generic” points Ri (in the algebraic geometry
sense), and specify how far each point in the body is from these three, so there are no more than 9 degrees of freedom
to specify the position (hence motion) of the body. How many numbers doe it take to specify a triangle in 3 − D
space?
It takes 3 degrees of freedom to select R1 and only 3 more numbers to specify the lengths of the sides of the triangle
made from Ri , so six degrees of freedom should be enough. We specify a COM position, place a space-fixed coordinate
triad there, then fix a coincident body-fixed triad there as well, and rotate it relative to the spce-fixed system.
x3
x03
Construct unit vectors i, j, k and i0 , j 0 , k0 which are related by

i0 = Ai + Bj + Ck, A = i0 · i = |i||i0 | cos θ11 = cos θ11

x02 and so forth, and we get a whole set of relations in terms of nine angles
 0    
i cos θ11 cos θ12 cos θ13 i
x2  j0  =  cos θ21 cos θ22 cos θ23   j 
k0 cos θ31 cos θ32 cos θ33 k

x1 but θij 6= θji since i · j 0 6= i0 · j.

x01

Any vector in the body can then be written in two ways

r = xi + yj + zk = x0 i0 + y 0 j 0 + z 0 k0

with
x0 = r · i0 = (xi + yj + zk) · (cos θ11 i + cos θ12 j + cos θ13 k) = x cos θ11 + y cos θ12 + z cos θ13
and so forth. One can apply this to any vector, not just position vectors.
Call i0 = e01 , j 0 = e02 , k0 = e03 , and so forth; X
e0i = cos θij ej
j

and X X X
e0i · e0i = 1 = cos θij cos θik ek · ej = cos θij cos θik δjk = cos2 θij
j,k j,k j
X X X
e0i · e0j = 0 = cos θi` cos θjk ek · e` = cos θi` cos θjk δ`k = cos θik cos θjk
`,k `,k k

In other words X
cos θik cos θjk = δij
k

so out of the nine angles, we have six relations, only three are independent. The matrix of direction cosines is an
orthogonal matrix, and there are many ways to parameterize an orthogonal matrix using three degrees of freedom.
We will look at a few useful parameterizations.

12.2 Orthogonal tranformations

Any transformation r → r 0 = M · r that leaves |r| unchanged is orthogonal

rj → rj0 = Mjk rk assume summation on repeated indices

50
 |r0 |2 = rj0 rj0 = Mjk rk Mj` r` = r` r` =⇒ Mjk Mj` = (M T )kj Mj` = δk` , MT · M = 1
you see the reason for this name, rows and columns form a mutually orthogonal set of vectors.
The stabilizer of z = x3 is the set of transformations that leave z invariant, z → z
 
a b 0
M =  −b a 0 
0 0 1

with a2 + b2 = 1, so we can parameterize with a single angle a = cos φ.

x2
x02 The
 coordinates
 of a vector r in the primed system whose coordinates are
a
in the unprimed are
0

x01 x01 = a cos φ, x02 = −a sin φ

for this counter-clockwise rotation.

φ
r x1
Define the composition of two linear transformations via matrix multiplication rules

r 0 = A · r, ri0 = Aij rj , r” = B · r 0 , r”j = Bjk Aki ri

then composition is associative by design;

XX X X XX XX 
Cjk Bk` A`i = Cjk Bk` A`i = Cjk Bk` A`i ) = Cjk Bk` A`i = Cjk Bk` A`i = C·(B·A) = (C·B)·A
k ` k ` ` k ` k

however it is not true that A · B = B · A, so the algebraic structure is associative but not commutative. The relation
M T · M = 1 says that we have inverses for elements M , we have an associative composition law, and we have closure;

C = A · B, AT · A = 1 = B T · B, =⇒ C T · C = (B T · AT ) · (A · B) = B T · B = 1

so if A, B are orthogonal transformations, so is C = A·B. We therefore have a group; the collection of linear orthogo-
nal transformations form a (Lie) group called O(N ) in N -dimensions. If we restrict our attention to transformations
M with det M = 1 (orientation preserving) we have SO(N ).

Orthogonal transformations are not necessarily rotations. Let n be any unit vector, and consider

r 0 = R · r = r − 2n(n · r)

This transformation is orthogonal (it is a reflection in the plane ⊥ to n) and has det(R) = −1. It is not a rotation,
since if n = i it leaves j, k invariant but i0 = −i, so it turns a right-handed coordinate system into a lefty. Let’s
double-check that R is orthogonal

Rij = δij − 2ni nj , RT = R, (R2 )ik = (δij − 2ni nj )(δjk − 2nj nk ) = δik − 2ni nj − 2ni nj + 4ni nj = δik

We therefore restrict our attention to the positive-determinant sheet of the orthogonal transformations that represent
orientation-preserving rotations.

51
12.3 Lecture 13. Hamilton’s ”quaternions”
Hamilton, Cayley and Maxwell were trying to construct a mathematical formalism in which combining successive
rotations of a three dimensional body would be similar to adding vectors. The figure shows the difficulty of this
problem, two successive 90-degree rotations about perpendicular axes has the same effect as a 120-degree rotation
about an intermediate axis. Combining rotations is not a commutative operation like addition, so the best they were
able to do is to impose a product combination law representing the rotations in succession, this puts an algebraic
structure (a Lie algebra) on the space of rotation operations that can be beautifully represented using extensions of
the complex numbers called quaternions.

z z y

y x x

x −y z

Inspect the figure, showing a right-handed rotation of the cube by 90 degrees about the z-axis will be represented by
an operator R1 , and a right-handed rotation by 90 degrees in the same sense about the (new) x-axis by 90 degrees
by R2 (90). The result of both in succession is R2 R1 . From the figure this is equivalent to a 120 degree rotation
about an axis through the upper front corner (facing you) and the center of the cube! If this is to be represented by
R2 · R1 , what must these operators look like? Hamilton’s answer is; this let i, j ,k be ”numbers” with
i2 = j 2 = k2 = −1, ij + ji = ik + ki = kj + jk = 0
ij = k, jk = i, ki = j
which extends the concept of complex number i2 = −1. Of course you recognize the relations of the cross-product
buried in here. The linear vector space spanned by i, j, k with these product rules defines a Clifford algebra Q and
by a trick of numerology a Lie algebra simultaneously. More on both at the end of this handout.

A rotation is specified by four numbers α, β, γ, θ, the first three are direction cosines. α, β, γ are the angles between
the axis of rotation and the three coordinate axes x, y, z respectively. The angle θ is the angle of rotation about
the axis of rotation. In Hamilton’s quaternion or hypercomplex number extension, such a rotation is represented by
the operator
θ θ 
R = cos − sin i cos α + j cos β + k cos γ
2 2
Example. My orientation of the initial axes in this written example differs from what I did in class so that the final
rotation axis will point out of the paper. The rotations in the figure become
1 1
R1 = √ (1 − k), R2 = √ (1 − i)
2 2
and the product is
1  1 
R2 R1 = 1 − (i + k − ik) = 1 − (i + k + j)
2 2
since the angle of the combined rotation satisfies cos θ2 = 12 we find that θ = 120 and extracting the sine of this as a
multiplier gives
 1 
R2 R1 = R3 = cos(60) − sin(60) √ (i + k + j)
3

52
from which we can just read off the correct cosines of the rotation axis

Write Hamilton’s rotation operator as

θ θ
Rn (θ) = cos − q · n sin , q = (i, j, k)
2 2
where n is the vector of direction cosines of the axis of rotation. In Hamilton’s way of performing rotations,
we associate an object with a vector r, and rotation about axis n by angle θ transforms vector r into r0 ;
 
r0 · q = Rn (θ) · r · q · Rn−1 (θ)

The beauty of this approach is that if we rotate a body by angle θ about axis n, and apply a second rotation by φ
about axis n0 , this is equivalent to rotating just once by ψ about n00 , and these two items can be found from matrix
multiplication
Example. We know that a counterclockwise rotation by θ about the z axis transforms

r = (x, y, z) → r0 = (x0 , y 0 , z 0 ) = (x cos θ − y sin θ, y cos θ + x sin θ, z)

or in Hamilton’s language, with

θ
!
θ θ ei 2 0
Rk (θ) = cos + iσz sin = θ
2 2 0 e−i 2

θ
! θ
! 
z0 x0 − iy 0 e−i 2 ei 2 e−iθ (x − iy)
   
0 z x − iy 0 z
= =
x + iy 0
0
−z 0 0
θ
ei 2 x + iy −z 0 e−i 2
θ
eiθ (x + iy) −z
 
z (x cos θ − y sin θ) − i(y cos θ + x sin θ)
=
(x cos θ − y sin θ) + i(y cos θ + x sin θ) −z
It is not easy(iest) to find ω in Hamiltons’s formulation, so let me present a formulation of rotations in which this
task is quite easy.
The trace of a rotation matrix gives the angle of rotation;
  
r − (n · r)n 0 −nz ny x
(n · r)n n × r =  nz 0 −nx   y 
−ny nx 0 z

n2x
  
nx ny nx nz x
n×r
(r · n)n =  nx ny n2y ny nz   y 
nz nx nz ny n2z z
Create the vector r 0 by rotating r about n (unit) by ψ right-
n handed;
 
r 0 = R · r = (r · n)n + r − (r · n)n cos ψ + (n × r) sin ψ

Take traces

T r R = 3 cos ψ + (1 − cos ψ)(n2x + n2y + n2z ) = 1 + 2 cos ψ

The reason for computing the trace is this; to show that one eigenvalue of an orthogonal matrix (element of SO(3)
) is one in 3 − D.
The orientation in space of a rigid body with fixed center of mass is given by an orthogonal transformation M (t)
relating the body-fixed coordinate axes to the space fixed, so if these coincide at t = 0, we must have M (0) = 1
and M (t) must be a transformation that evolves from 1 in a continuous manner; the general displacement of a rigid

53
body with fixed COM is rotation about some axis. Which is very nifty; we need two angles θ and φ to specify the
axis and an angle ψ for rotation about it (the Euler angles). Since the axis n of the rotation is not affected by the
rotation about n, the theorem is proved by showing that given r 0 = M · r gives the coordinates of a vector r after the
evolution, there is a vector n that is not changed; n0 = M · n = n, so M must have an eigenvector with eigenvalue
1 for Euler’s theorem to be true.
We know that M T · M = M · M T = 1, or

M T · (M − 1) = 1 − M T , det M T det(M − 1) = det(1 − M T ) = det((1 − M )T ) = det(1 − M )

but det M T = det M = 1, and so det(1 − M ) = det(M − 1) = 0 and one of the eigenvectors is 1. This means that

λ1 + λ2 + λ3 = 1 + λ1 + λ2 = 1 + 2 cos ψ, 1 · λ1 λ2 = 1

λ1 + λ−1
1 = 2 cos ψ, λ1 = e±iψ , λ2 = λ∗1 = λ−1
1

12.4 Euler angles parameterization of rotation

We have used 4 parameters to describe any 3 − D rotation by by ψ about an axis n = (cos α, cos β, cos γ)T , and of
course cos2 α + cos2 β + cos2 γ = 1. So these three angles of axis-orientation are not independent, we might as well
specify n using the polar angles
n = (sin θ cos φ, sin θ sin φ, cos θ)T
Recall that if we rotate a vector r (go back and look at the figure), not the coordinate system, by a right-handed
(CCW) rotation by ψ about n
 
r 0 = R · r = (r · n)n + r − (r · n)n cos ψ + (n × r) sin ψ

which is the same as rotating the coordinate system left-handed (CW), as described in the text.
  
0 −nz ny x
0
r − r = dr = dψ n × r, n × r =  nz 0 −nx   y  = (M · n) · r
−ny nx 0 z
or      
0 0 0 0 0 1 0 −1 0
 0 0 −1  = M1 ,  0 0 0  = M2 ,  1 0 0  = M3
0 1 0 −1 0 0 0 0 0
dr
= ω(n · M ) · r = ω × r
dt
Compare with notations in the text; let ψ → dψ << 1
  
0 −nz ny x
r 0 = (1 + )r,  = ni Mi dψ =  nz 0 −nx   y  dψ
−ny nx 0 z

Instead RHR rotate the coordinate system, which is equivalent to reversing either ψ or n, then
  
0 nz −ny x
dr
= ω × r, n × r =  −nz 0 nx   y 
dt
ny −nx 0 z

which simply reverses ω. Why Goldstein chooses to describe Euler angles using RH coordinate
rotations and then switch to LH is anyone’s guess, so be careful comparing formulas in the first half
of the chapter with those in the second.
Iterate the infinitismal form calling dψ = ψ/N ;

r 0 = lim (1 + )N r = lim (1 + ni Mi ψ/N )N r = eψni Mi r

N →∞ N →∞

54
These gadgets generate the left-handed rotations of the coordinate system itself, we say that they are the
infinitismal generators of the rotations eφMi . The objects eφMi form one-parameter subgroups of the group of 3 − D
rotations
e0M1 = 1 identity, e−φM1 eφM1 = 1 inverses, eθM1 eφM1 = e(θ+φ)M1 closure
but
eθM1 eφM2 6= eφM2 eθM1 6= eθM1 +φM2
Mathematicians say that the generators {M1 , M2 , M3 } of the one-parameter subgroups of the group of three-D
rotations form a Lie algebra, using either the matrix representation or these coordinate realizations we can deduce
the structure constants of the algebra; the numbers on the right side of
X
[Mi , Mj ] = ck ij Mk
k

[M1 , M2 ] = M3 , [M2 , M3 ] = M1 , [M1 , M3 ] = M2

we see that this is the same as the Lie algebra product relations of 2i, 2j, 2k for the Lie algebra living in the
quaternions (there’s that mysterious 2 again), another deep connection between rotations and (hyper) complex
numbers and Clifford algebras.
A Lie algebra is a vector space, the elements of the Lie algebra generated by {M1 , M2 , M3 } are all of the form

z = aM1 + bM2 + cM3

Note that objects such as 1, M32 , M1 M2 are not in this Lie algebra (Let me call it spin(3)). However we can construct
a second Lie algebra (call it U (spin(3))) that has a basis consisting of all powers of the generators (basis) of spin(3)

1, M1 , M2 , M3 , M12 , M1 M2 , · · ·

dim(spin(3)) = 3, but dim(U (spin(3))) = ∞. This is called the universal enveloping algebra of spin(3) and is the
Lie algebra of the operators associated with measurements of angular momentum in quantum theory.

The following parameterization (Euler) of a right handed rotation of the coordinate system

M = e−ψn·M

is not at all convenenient for finding the velocity of a point r because of the noncommutative nature of the quaternions.
What is useful is the alternative parameterization

M = e−ψM3 e−θM1 e−φM3 = Q3 Q2 Q1

55
13 Lecture 13.1415926... Clifford and Lie algebras
The true beauty of the quaternions is not immediately obvious until we play with them using the tools of abstract
algebra. Let’s collect some of these tools together. I did not cover this stuff in lecture, it is FYI, and it is
worth reading and knowing even if ”it is not on the test”.

13.1 Clifford algebras

All algebras are vector spaces V upon which additional structure in the form of a bilinear product V × V → V has
been imposed. A linear map from V into anything is a map f that obeys
f (λu + µv) = λ f (u) + µ f (v)
This map could be a linear transformation, which maps an element of V to another element of V (sends vector
to vector),
f :V →V
or could map a vector to a scalar
f :V →R
We call such a map a linear functional or dual. Dual maps of vector spaces are also vector spaces. Remember that
line from Forest Gump (but I paraphrase) ”Vectors is as vectors does”, meaning that we identify all things according
to their properties, and linear functions f : V → R obey the axioms of vector spaces;
1. There is an identity functional;

f (u) + 0 · (u) = f (u)

˜
2. There is an inverse functional f for any f ∈ V ∗ ;
f (u) + f˜(u) = 0 · (u), f˜ = −f
3. We have closure and an abelian addition; for f, g ∈ V ∗ define (f + g) by
(f + g)(u) = f (u) + g(u)
which is clearly
4, associative. We can define scalar multiplication of functionals from linearity as well; given λ ∈ R and f ∈ V ∗
(λ f ) ∈ V ∗ , (λ f )(u) = f (λu) = λ f (u)
All of the requirements to be a vector space are there.

A bilinear map inputs a pair of vectors and outputs a real number, (call it h , i)
h, i : V ⊗ V → R
is sesquilinear if
hu, vi = hv, ui, hw, λu + µvi = λhw, ui + µhw, vi
This type of map is incredibly important. For one thing a distance function is precisely this type of mapping. A
metric space is a vector space V or manifold V together with a bilinear map d;
d:V ×V →R
that maps a pair of vectors to a real number, with the following properties;
d(u, v) ≥ 0
d(u, v) = 0 if u = v
d(u, v) = d(v, u)
d(u, w) ≤ d(u, v) + d(v, w)

56
Such a function d is called a distance function and is a measure of how far apart the points represented by the
vectors are when their tails coincide. Our usual language for the length of a vector is its norm. The mapping
u ∈ V → ||u|| ∈ R is called a norm iff

||u + v|| ≤ ||u|| + ||v||

||λ u|| = |λ| ||u||
||u|| = 0 if and only if u = 0

A norm induces a metric; given a norm on a vector space we can create a distance function d from

||u − v|| = d(u, v)

A vector space V with a norm || · || is called a normed vector space, and we will work almost exclusively with
such vector spaces.

The distance function is a good example of a bilinear form; a mapping that inputs two vectors and outputs a number
in some field C. The notation for this is

Q : V ⊗ V → C, Q(v1 , v2 ) ∈ C

v i ei is an arbitrary vector.
P
Consider the d-dimensional vector space V with a basis {e1 , e2 , · · · , ed }, such that v = i
Let’s define a very simple bilinear form;
p
X d
X
i i
Qp,d−p (v, u) = vu − v i ui
i=1 i=p+1

You can see that Qd,0 (v, u) = v · u is our usual dot product.

A Clifford algebra C(Qp,d−p ) is probably the most primitive algebraic structure that can be placed on a vector field
to make it into an algebra. It is generated by the basis {e1 , e2 , · · · , ed } of the vector space, which is orthogonal
with respect to Qp,d−p (meaning Qp,d−p (ei , ej ) = 0 if i 6= j) which defines an associative and distributive product
#:V ×V →V;

# : C(Qp,d−p ) × C(Qp,d−p ) → C(Qp,d−p ), ei #ej + ej #ei = 2Qp,d−p (ei , ej )

for which 1 is the unit. All associative algebras contain a unit, since one can be canonically attached.

Let’s construct a few Clifford algebras, starting with C(Q0,1 ). This requires a “vector” e, with Q0,1 (e, e) = −1, such
that
e#e + e#e = 2Q0,1 (e, e) = −2, e#e = −1
and so the only objects in this algebra will be 1 and e, since the complete multiplication table of C(Q0,1 ) closes with

1#1 = 1, 1#e = e#1 = e, e#e = −1

This algebra is isomorphic to the complex numbers C;

1 · 1 = 1, 1 · i = i · 1 = i, i · i = −1

We establish isomorphism by constructing a surjective linear mapping from one algebra onto the other; C = {a + b ·
i | a, b ∈ R}
φ : C(Q0,1 ) ↔ C, φ(1) = 1, φ(e) = i
that preserves all bilinear operations
φ(z#z 0 ) = φ(z)φ(z 0 )
and our φ-map does so, which you should check. We have C = ∼ C(Q0,1 ).

57
There is an intimate connection between complex numbers and rotations which I illustrated in the previous lecture;
take a vector r = (x, y)T and make an object r = x + iy = r · (1, i) that lives in C(Q0,1 ) from it. Create another such
gadget Rθ = cos θ + i sin θ, then in C(Q0,1 )

Rθ r = (cos θ + i sin θ)(x + iy) = (x cos θ − y sin θ) + i(y cos θ + x sin θ)

is a rotation of r by angle θ. Hamilton discovered a way to extend this construction to three dimensions.

You can show that C(Q1,0 ) ∼ = R ⊕ R is the disjoint or direct sum of two copies of the real numbers; it has a basis 1
and e with e#e = 1, but of course 1 6= e. There is a periodicity (of period 8) in the structure of Clifford algebras
of increasing dimension (as vector spaces) called the Bott periodicity. This plays a role in whether or not Majorana
fermions can exist in spaces of d or d + 1 dimensions; if the Clifford algebra corresponding to a given metric has real
representations then yes, if not, no.
For larger Clifford algebras, we discover that no numbers could possibly satisfy the algebraic identities imposed by
the product rule, but matrices can. In fact all Clifford algebras are isomorphic to matrix algebras. The product
# is then simply the product defined for matrices.
Consider the case of C(Q2,0 ), generated by e1 , e2 with

e1 #e1 = 1, e2 #e2 = 1, e1 #e2 + e2 #e1 = 0

It is clear that no numbers could satisfy the third relation. Therefore we turn to matrices, and interpret 1 as the
unit matrix in some dimension, and # simply becomes
 the
 normal matrix
 product. 
0 1 1 0
It is a short calculation to show that e1 = and e2 = satisfy all three relations with
1 0 0 −1
 
1 0
1 = . Technically it is not correct to say that ei equals a matrix, but rather that ei has a matrix
0 1
representation (denoted by ρ(ei )), but this technicality will be temporarily ignored, since (much to the chagrin of
mathematicians) physicist don’t distinguish between objects and their representations. You can show that we must
include e1 #e2 in the algebra in order to be able to complete the multiplication table and have closure under the #
operation. Then one finds the matrix of e1 #e2 . The most arbitrary element z ∈ C(Q2,0 ) is therefore of the form

z = a0 1 + a1 e1 + a2 e2 + a3 e1 #e2 , ai ∈ R, i = 0, 1, 2, 3

and so dim(C(Q2,0 )) = 4. The end result is that C(Q2,0 ) ∼

= R(2), the set of 2 × 2 real matrices.

What about quaternions? Examine C(Q0,2 ) starting with two vectors e1 , e2 and let the bilinear form be the Cartesian
inner product of these objects with signature (−1, −1)

ei ej + ej ei = −2δij , i, j ∈ {1, 2}

This Clifford algebra has basis

{1, e1 , e2 , e1 e2 } = {1, i, j, k}
which evidently cannot be realized numerically, so we construct a  representation
 in terms of matrices.
 Based
 on our
0 i 0 −1
exploration of C(Q2,0 ) it should be clear that we can use i = e1 = = iσ1 and j = e2 = = iσ2 ,
  i 0 1 0
i 0
then k = e1 e2 = = iσ3 . This makes it pretty clear to those people who find such things pretty clear
0 −i
that there is a connection between intrinsic quantum angular momentum and quaternions/Clifford algebras. Let me
try to deepen the extent of these connections.

You can now show that in C(Q0,2 ) we have relations

ii = jj = kk = −1, jk = j(ij) = −j(ji) = −(jj)i = i, ki = j and of course ij = k

The algebra is the linear span C(Q0,2 ) = {1, i, j, k}, by a happenstance of numerology the subset {i, j, k} form a Lie
algebra
2ij = 2k = ij − ji ≡ [i, j]

58
This algebra is the linear span of {i, j, k} (1 is not in this algebra, nor are any bilinears ij, all of these are in the
universal enveloping algebra of the Lie algebra) has a product [ , ]; [A, B] = AB − BA that closes

[i, j] = 2k, [j, k] = 2i, [k, i] = 2j

and this algebra generates three-dimensional rotations. We will find represetations of the Lie algebra defined
by the quaternions. relations such as i2 = −1 are not Lie algebra relations, so our representations will not obey
such relations.
Representations ρ : A → GL(n, F)), mappings of an algebra into n × n matrices over some field can be found by
various constructions (you have an option to use group character methods in PS-5 if you read the Clifford algebra
stuff and want to explore it further). I illustrate the adjoint representation ρ = Ad, in which the vector space of the
representation is the algebra itself  
ea eb = cab c ec , Ad(ea ) = cabc
cb
for any associative algebra on the Lie algebra embedded in the quaternions and construct the adjoint representation;
defined to be the matrix representation associated with regarding the product law as a linear mapping of the algebra
into itself. Call i = e1 , j = e2 , k = e3

ρAd (k)i ≡ [k, i] = 2j, ρAd (k)j = [k, j] = −2i, ρAd (k)k = [k, k] = 0
 
0 −2 0
(Ad(k))21 = 2, (Ad(k))12 = −2 ρAd (k) =  2 0 0 
0 0 0
 
0 0 0
ρAd (i)j ≡ [i, j] = 2k, ρAd (i)k = [i, k] = −2j, ρAd (i)i = [i, i] = 0, ρAd (i) =  0 0 −2 
0 2 0
 
0 0 2
ρAd (j)i ≡ [j, i] = −2k, ρAd (j)k = [j, k] = 2i, ρAd (j)j = [j, j] = 0, ρAd (j) =  0 0 0 
−2 0 0
By the way, the ”2” appearing above is very important. Define what we mean by the exponential of an opera-
tor/matrix as
M2
eM f = f + M f + f + ···
2
and apply it to the linear operator M = Ad(k)/2
     
0 −1 0 −1 0 0 1 0 0
M =  1 0 0  , M 2 =  0 −1 0  , M 3 = −M, M 2n = (−1)n  0 1 0  , M 2n+1 = (−1)n M
0 0 0 0 0 0 0 0 0

Finally with M 0 = 1 (unit 3 × 3 matrix)

X φn X φ2m X φ2m+1
R = eφM = Mn = 1 + M 2m + M 2m+1
n
n! m=1
(2m)! m=0
(2m + 1)!
   
1 0 0 X (−1)m φ2m 0 −1 0 X (−1)m φ2m+1
= 1+ 0 1 0  + 1 0 0 
(2m)! (2m + 1)!
0 0 0 m=1 0 0 0 m=0
       
1 0 0 1 0 0 0 −1 0 cos φ − sin φ 0
=  0 1 0  +  0 1 0  (cos φ − 1) +  1 0 0  sin φ =  sin φ cos φ 0 
0 0 1 0 0 0 0 0 0 0 0 1

and you can see the connection with orthogonal matrices and rotations (exponentiation of antisymmetric matrices
results in an orthogonal matrix).

59
Recall that the direction cosines of any axis must form a unit vector, so

cos2 α + cos2 β + cos2 γ = 1

Note that
(σ · n)2 = (i cos α + j cos β + k cos γ)2 = −1
and
(σ · n)2n = (−1)n , (σ · n)2n+1 = (−1)n−1 σ · n
so that
θ
X (−1)m (θ/2)2m X (−1)m (θ/2)2m+1
e 2 σ·n = 1+ +σ·n
m=1
(2m)! m=0
(2m + 1)!
= cos(θ/2) + sin(θ/2) σ · n

We have found a 2 × 2 matrix representation of the Clifford algebra, so consider a rotation with α = β = π/2 and
γ = 1 (about z-axis)  iθ/2 
 
θ iθ
1 0
e 0
e 2 q·n = e 2 0 −1 =
0 e−iθ/2
But we also found the adjoint 3 × 3 representation of the corresponding Lie algebra
 
cos θ sin θ 0
 0 2 0

θ −2 0 0
θ 2
e 2 q·n = e 0 0 0
=  − sin θ cos θ 0 
0 0 1

The vector space in which the 2 × 2 matrix representation is made is spanned by two vectors. The fundamental
objects upon which R acts as a linear transformation s = s0 = R · s in the 2 × 2 rep. are called the spinors. The
fundamental objects upon which R acts as a linear transformation v = v 0 = R · v in the 3 × 3 rep. are called the
vectors.

60
Physics 711. Euler angles parameterization of rotation
We have used 4 parameters to describe any 3 − D rotation by by ψ about an axis n = (cos α, cos β, cos γ)T , and of
course cos2 α + cos2 β + cos2 γ = 1. So these three angles of axis-orientation are not independent, we might as well
specify n using the polar angles
n = (sin θ cos φ, sin θ sin φ, cos θ)T
Recall that if we rotate a vector r (go back and look at the figure), not the coordinate system, by a right-handed
(CCW) rotation by ψ about n
 
r 0 = R · r = (r · n)n + r − (r · n)n cos ψ + (n × r) sin ψ

which is the same as rotating the coordinate system left-handed (CW), as described in the text.
  
0 −nz ny x
0
r − r = dr = dψ n × r, n × r =  nz 0 −nx   y  = (M · n) · r
−ny nx 0 z
or      
0 0 0 0 0 1 0 −1 0
 0 0 −1  = M1 ,  0 0 0  = M2 ,  1 0 0  = M3
0 1 0 −1 0 0 0 0 0
dr
= ω(n · M ) · r = ω × r
dt
Compare with notations in the text; let ψ → dψ << 1
  
0 −nz ny x
r 0 = (1 + )r,  = ni Mi dψ =  nz 0 −nx   y  dψ
−ny nx 0 z

Instead RHR rotate the coordinate system, which is equivalent to reversing either ψ or n, then
  
0 nz −ny x
dr
= ω × r, n × r =  −nz 0 nx   y 
dt
ny −nx 0 z

which simply reverses ω. Why Goldstein chooses to describe Euler angles using RH coordinate
rotations and then switch to LH is anyone’s guess, so be careful comparing formulas in the first half
of the chapter with those in the second.
Iterate the infinitismal form calling dψ = ψ/N ;

r 0 = lim (1 + )N r = lim (1 + ni Mi ψ/N )N r = eψni Mi r

N →∞ N →∞

These gadgets generate the left-handed rotations of the coordinate system itself, we say that they are the
infinitismal generators of the rotations eφMi . The objects eφMi form one-parameter subgroups of the group of 3 − D
rotations
e0M1 = 1 identity, e−φM1 eφM1 = 1 inverses, eθM1 eφM1 = e(θ+φ)M1 closure
but
eθM1 eφM2 6= eφM2 eθM1 6= eθM1 +φM2
Mathematicians say that the generators {M1 , M2 , M3 } of the one-parameter subgroups of the group of three-D
rotations form a Lie algebra, using either the matrix representation or these coordinate realizations we can deduce
the structure constants of the algebra; the numbers on the right side of
X
[Mi , Mj ] = ck ij Mk
k

61
 [M1 , M2 ] = M3 , [M2 , M3 ] = M1 , [M1 , M3 ] = M2
we see that this is the same as the Lie algebra product relations of 2i, 2j, 2k for the Lie algebra living in the
quaternions (there’s that mysterious 2 again), another deep connection between rotations and (hyper) complex
numbers and Clifford algebras.
A Lie algebra is a vector space, the elements of the Lie algebra generated by {M1 , M2 , M3 } are all of the form

z = aM1 + bM2 + cM3

Note that objects such as 1, M32 , M1 M2 are not in this Lie algebra (Let me call it spin(3)). However we can construct
a second Lie algebra (call it U (spin(3))) that has a basis consisting of all powers of the generators (basis) of spin(3)

1, M1 , M2 , M3 , M12 , M1 M2 , · · ·

dim(spin(3)) = 3, but dim(U (spin(3))) = ∞. This is called the universal enveloping algebra of spin(3) and is the
Lie algebra of the operators associated with measurements of angular momentum in quantum theory.

The following parameterization (Euler) of a right handed rotation of the coordinate system

M = e−ψn·M

is not at all convenenient for finding the velocity of a point r because of the noncommutative nature of the quaternions.
What is useful is the alternative parameterization

M = e−ψM3 e−θM1 e−φM3 = Q3 Q2 Q1

(The Euler angles ψ, θ, φ) decomposing into a product of three simpler transformations. Why is this so spiffy for
finding ω? Consider
r 0 = e−φM3 r = Q1 r
with φ = φ(t) and ṙ = 0. Then
  0 
0 φ̇ 0 x
ṙ 0 = Q̇1 r = Q̇1 Q−1
1 Q1 r = Q̇1 Q−1 0
1 r = Q̇1 QT 0
1 r =  − φ̇ 0 0   y0 
0 0 0 z0

and recall that for any RHR rotation of the coordinates by φ about n
 
0 φ̇nz −φ̇ny
dr
= (n × r)φ̇ ≡ ω × r =  −φ̇nz 0 φ̇nx  r
dt
φ̇ny −φ̇nx 0

so like magic we deduce that

 
0 φ̇nz −φ̇ny
Q̇i Q−1
i = Q̇i QTi =  −φ̇nz 0 φ̇nx  = ”ω × ”
φ̇ny −φ̇nx 0

is the appropriate contribution to the ω ”operator”/vector/gadget. The velocity of the object as seen from the
primed coordinate system is a CW circulation around the origin.
What if r has some intrinsic change as well as this change due to the rotation of the coordinates? Then

(r˙0 ) = Q̇1 r + Q1 ṙ = Q̇1 Q1−1 Q1 r + Q1 ṙ = Q̇1 Q−1 0 0 T 0

1 r + (ṙ) = Q̇1 Q1 r + (ṙ)
0

in other words
(r˙0 ) = ω 0 × r 0 + (ṙ)0

62
 r 0 = Q3 Q2 Q1 r
(r˙0 ) = Q̇3 Q2 Q1 r + Q3 Q̇2 Q1 r + Q3 Q2 Q̇1 r
= Q̇3 Q2 Q1 (QT1 QT2 QT3 )r 0 + Q3 Q̇2 Q1 (QT1 QT2 QT3 )r 0 + Q3 Q2 Q̇1 (QT1 QT2 QT3 )r 0
= Q̇3 QT3 r 0 + Q3 Q̇2 QT2 QT3 r 0 + Q3 Q2 Q̇1 QT1 (QT2 QT3 )r 0 = ω 0 × r 0

Compare each part of this with the statements in Goldstein p. 174, about application of parts of the three factors
making up the Euler rotation, to various parts of ω.
Evaluate it;
 
0 ψ̇ 0
Q̇3 QT3 =  −ψ̇ 0 0 
0 0 0
   
cos ψ sin ψ 0 0 0 0 cos ψ − sin ψ 0
Q3 (Q̇2 QT2 )QT3 =  − sin ψ cos ψ 0   0 0 θ̇   sin ψ cos ψ 0 
0 0 1 0 −θ̇ 0 0 0 1
 
0 0 θ̇ sin ψ
=  0 0 θ̇ cos ψ 
−θ̇ sin ψ −θ̇ cos ψ 0
   
1 0 0 0 φ̇ 0 1 0 0
Q3 Q2 Q̇1 QT1 (QT2 QT3 ) = Q3  0 cos θ sin θ   −φ̇ 0 0   0 cos θ − sin θ  QT3
0 − sin θ cos θ 0 0 0 0 sin θ cos θ
   
0 −φ̇ cos θ φ̇ sin θ 0 φ̇ cos θ −φ̇ sin θ cos ψ
= Q3  φ̇ cos θ 0 0  QT3 =  −φ̇ cos θ 0 φ̇ sin θ sin ψ 
−φ̇ sin θ 0 0 φ̇ sin θ cos ψ −φ̇ sin θ sin ψ 0

Add and compare

ωz0 −ωy0
   
0 ψ̇ + φ̇ cos θ θ̇ sin ψ − φ̇ sin θ cos ψ 0
”ω 0 × ” =  −(ψ̇ + φ̇ cos θ) 0 θ̇ cos ψ + φ̇ sin θ sin ψ  =  −ωz0 0 ωx0 
−θ̇ sin ψ + φ̇ sin θ cos ψ −θ̇ cos ψ − φ̇ sin θ sin ψ 0 ωy0 −ωx0 0

which agree nicely with the calculation that we did in Chapter 3

ωz0 = ψ̇ + φ̇ cos θ, ωy0 = −θ̇ sin ψ + φ̇ sin θ cos ψ, ωx0 = θ̇ cos ψ + φ̇ sin θ sin ψ

for the dynamical degrees of freedom of a rolling body. If you want ω in the original (space-fixed frame) compute
just apply QT1 QT2 QT3 .

r0 = Q3 Q2 Q1 r
0 •
(r ) = Q̇3 Q2 Q1 r + Q3 Q̇2 Q1 r + Q3 Q2 Q̇1 r
QT1 QT2 QT3 (r 0 )• = QT1 QT2 QT3 Q̇3 Q2 Q1 r + QT1 QT2 Q̇2 Q1 r + QT1 Q̇1 r
= ṙ = ω × r

matrix q1(3,3),q2(3,3),q3(3,3);
depend ph,x;
depend th,x;
depend ps,x;
q3:=mat((cos(ps),sin(ps),0),(-sin(ps),cos(ps),0),(0,0,1));
q1:=mat((cos(ph),sin(ph),0),(-sin(ph),cos(ph),0),(0,0,1));
q2:=mat((1,0,0),(0,cos(th),sin(th)),(0,-sin(th),cos(th)));
for all y let cos(y)^2+sin(y)^2=1;

63
df(q3,x)*tp(q3);
% v’...
df(q3,x)*tp(q3)+q3*df(q2,x)*tp(q2)*tp(q3)+q3*q2*df(q1,x)*tp(q1)*tp(q2)*tp(q3);
%v...
tp(q1)*df(q1,x)+tp(q1)*tp(q2)*df(q2,x)*q1+tp(q1)*tp(q2)*tp(q3)*df(q3,x)*q2*q1;

Note that r transforms as a vector, r 0 = Rr = Q3 Q2 Q1 r but ω transforms as a tensor (second rank, antisymmetric,
in other words as a pseudo-vector)

ω 0 × = ṘRT = ṘR−1 , ω× = RT Ṙ = RT (ω 0 ×)R

64
14 Lecture 15. Rotating coordinate systems
Consider any pair of coordinate systems in which one is a rotation of the other about the origin. Possibly the best
example is the case of a previously studied problem; let (x, y) be space-fixed (inertial frame) coordinates of a block
on a turntable, (x0 , y 0 ) be axes drawn on a turntable which we set into rotation;
y
0
y  0    
x cos(ωt) sin(ωt) x
r0 = = = e−ωtMz · r
y0 − sin(ωt) cos(ωt) y
r x0 x0
    
x cos(ωt) − sin(ωt)
r= = = eωtMz · r 0
y sin(ωt) cos(ωt) y0
ωt We do the physics in an inertial frame, mr̈ = F ,
x
 ··
F = mr̈ = m eωtMz r 0
 ··  ·
= m eωtMz r 0 + 2m eωtMz (r˙0 ) + meωtMz (r¨0 )
 
= etωMz mω 2 Mz2 r 0 + 2mωMz (ṙ 0 ) + m(r̈ 0 )
e−tωMz F = mω 2 Mz2 r 0 + 2mωMz (ṙ 0 ) + m(r̈ 0 )
= mω 0 × (ω 0 × r 0 ) + 2mω 0 × (ṙ 0 ) + m(r̈ 0 )

in which the left side is the collection of applied forces expressed with components in the rotating frame. The right
side has been evaluated in terms of the cross-producr with ω using the previous lecture results.

m(r̈ 0 ) = e−tωMz F − mω 0 × (ω 0 × r 0 ) − 2mω 0 × (ṙ 0 )

In the picture below, note that the force of gravity F on a mass placed above the earth’s surface has both y, z
components , but e−tωMz F has only z 0 components.
ω, k ω0
k0
14.1 Motion at the earth’s surface
θ Look at a tangent-plane frame on the surface at the earth at colatitude θ, in this
frame the earth has angular velocity vector

F j0 ω 0 = cos θk0 − sin θj 0

θ Write all of the vectors in the body-coordinates (primed, rotated co-
j ordinates) but dispense with the primes; A body is dropped from a height
h above the earth. Neglecting air resistance, determine the displacement of its
impact point away from the value predicted by neglecting the earth’s rotation.
Neglecting as well the very small centrifugal term (ω ∼ 7 × 10−5 rad/s)

mr̈ = mg + F − 2mω × ṙ

Write it all out in component form

mẍ = −2m(−ω sin θż − ẏω cos θ), mÿ = −2m(ẋω cos θ), mz̈ = −mg + 2m(ω sin θẋ)

and solve this iteratively;

Neglect everything but gravity

m(ẍ(0) , ÿ (0) , z̈ (0) ) = (0, 0, −mg)

65
so the lowest order solutions are
ẍ(0) = 0, ÿ (0) = 0, z̈ (0) = −g −→ ẋ(0) = 0, ẏ (0) = 0, ż (0) = −gt
using our initial conditions.
Keep the Coriolis term as a perturbation on the lowest order solution.

mẍ(1) = −2mωgt sin θ, mÿ (1) = 0, mż (1) = −mg

gotten by inserting the zeroth order velocities into the right hand side of the first order force equations. These can
be integrated twice over time to give
1 1
x(1) (t) = − ωgt3 sin θ, y (1) (t) = 0, z (1) (t) = h − gt2
3 2
q
2h
It takes time t = g to fall, insert this into the x-equation to get a (eastward) displacement of

1  2h  32
∆x = − ωg sin θ
3 g

14.2 Foucault pendulum

This is lovely demonstration of the rotation of the earth. UW Physics used to have an impressive 6-story Foucault
pendulum in Sterling Hall, outside rm. 1300 Sterling, sadly removed after the building renovation. The Coriolis
forces will influence motion parallel to the plane tangent to the surface of the earth, and so we construct equations of
motion for the pendulum of length ` in Cartesian coordinates using a potential, without the influence of the earth’s
rotation the pendulum is subjected to a gravitational potential V (x, y, z) = mgz and the string tension. The string
constrains the pendulum to move in such a way that it remains a distance ` from the point on the z-axis (0.0, `)
where the string is attached, so if the pendulum coordinates are (x, y, z) then the potential in terms of x and y for
a body that is essentially sliding on the surface of a spherical bowl of radius ` resting on the earth’s surface.
1 2 
`2 = (` − z)2 + x2 + y 2 , z≈ x + y2
2`
As long as the pendulum cord ` >> x, y we neglect terms that this replacement introduces into the kinetic energy.
Foucault pendula, pendula designed to illustrate the precession of the pendulum, usually have very long strings. We
obtain
mg  2 
V (x, y, z) ≈ x + y2 , ma = −∇V
2`
in the absence of any rotation. The equations of motion for x and y separate;
g g
ẍ = − x, ÿ = − y
` `
Now we introduce the Coriolis term present for motion of the pendulum on the surface of the earth at colatitude θ,
g g
ẍ = − x + 2ωz ẏ, ÿ = − y − 2ωz ẋ
` `
neglecting the much smaller vertical motion. To solve both together let
g
ξ = x + iy, then ξ¨ + 2iωz ξ˙ + ξ = 0, ξ = A er1 t + B er2 t
`
where r1,2 are roots of the algebraic equation
r
2 g g
r + 2iωz r + = 0, r1,2 = −iωz ± −ωz2 −
` `
g
Since ω sin θ << ` we expand to get
r
g  √g √g 
r1,2 ≈ ±i − iω sin θ and recover ξ = e−iωz t Aei ` t + Be−i ` t
`

66
Taking real and imaginary parts
 rg   rg   rg   rg 
x = A cos ( − ωz ) t + B cos ( + ωz ) t y = A sin ( − ωz ) t − B sin ( + ωz ) t
` ` ` `
1.5 Gnu ode code below. Ode is a command-line
program for numerically solving DE’s, that is part
1.0 of gnu-plotutils, an open-source graphics library.

0.5 g=9.8
L=3
w=0.2
y

0.0 x’=p
y’=q
p’=-(g/L)*x+2*w*q
−0.5
q’=-(g/L)*y-2*w*p
p=0
−1.0 q=0
x=1
y=1
−1.5 print x,y
−1.5 −1.0 −0.5 0.0 0.5 1.0 1.5
step 0,10
x
Consider an expansion of a vector v into unit vectors forming a local basis

v = v a ea

and compute the change in the vector

dv = (dv a )ea + v a dea
Let’s parameterize the changes in the basis vectors; let the changes be linear combinations of local basis vectors

dea = dω b a eb

and suppose that we have several parameters or quantities q c that could change,

dea = Γb ac dq c eb

(we currently only have one, t).

dv = (dv a )ea + v a Γb ac dq c eb
and for rotating frames we have (call t = q 0 )
Dv Dv a dv a dv a
dv = (dv a )ea + v a Γb a0 dt eb , = ea = ea + v b Γa b0 ea = ea + (ω × v)a ea
dt dt dt dt
ans so we see the extra part of the rate of change of the vector due to changes in the orientation of the coordinate
system as a kind of Christofel symbol, and this points to the proper way of thinking about what a Christofel symbol
is; it represents the change in a vector attributable to changes in the orientation of the local coordinate system. If the
coordinate system does not only change in time but also in space, then our parameters q c could be all of t, x, y, z = xµ ,
and we have
dv = (dv µ )eµ + v ν Γµ αν dxα eµ
or
(Dv µ )eµ = (dv µ )eµ + v ν Γµ αν dxα eµ , Dv µ = dv µ + v ν Γµ αν dxα
Divide by ds, where s is a parameter on which xµ depends along some path
Dv µ dv µ dxα
= + v ν Γµ αν
ds ds ds

67
This represents the total rate of change of a vector along some path, intrinsic plus the part due to changes in the
coordinate systems against which components are computed. If this is zero, we say that the vector is ”parallel
transported” along the path, experiencing no change. We can find the equations of curves in space along which the
a
tangent vectors to the path experience no net change from point to point; let v a = dx
ds ;

d2 xµ dxν µ dxα
0= + Γ αν
ds2 ds ds
and of course you have seen this before, I hope you see it in new light now.

68
15 Lecture 16. Torque-free rotation
Given Euler’s theorem the most general motion of a rigid body with one point fixed is a rotation about some axis, so
that dr
dt = ω × r for any point on the body, a mass m at r on a such a rigid body will have kinetic energy

1 1    
T = mv · v = m ω × r · ω × r
2 2
1  2 2 
= m ω r − (ω · r)2
2
1 X   1 1
= m ωi ωj δij r2 − ri rj = ω · I · ω = ω · L
2 i,j
2 2

introducing the moment of inertia tensor and the angular momentum vector, which we discussed earlier. The moment
of inertia tensor for collections of masses or for uniformly distributed mass is
X   Z  
I= mk δij rk2 − rk,i rk,j , I = d3 r ρ(r) δij r2 − ri rj
k

A previous example is that of a disk of mass m and radius a; using body fixed polar coordinates x = ρ cos φ and
y = ρ sin φ, z = 0 for a disk in the xy-plane
  M a2 
ρ2 sin2 φ

Z a Z 2π −ρ2 sin φ cos φ 0 0 0
M  4 M a2
I= ρdρ dφ  −ρ2 sin φ cos φ ρ2 cos2 φ 0 = 0 0 

2
πa 0 4
0 2 2
0 0 ρ 0 0 Ma
2

the moment of inertia of a disk spun at ω about an axis perpendicular to the disk through its center is I = 12 M a2 ,
and rotation about a diameter has half the moment of inertia. Computation of T should use both I and ω computed
in the same frame, either inertial or rotating, and it will not matter which since T is a scalar (rotational invariant),
so we pick body-fixed and use our previously deduced components

ωx0 = θ̇ cos ψ + φ̇ sin θ sin ψ, ωy0 = −θ̇ sin ψ + φ̇ sin θ cos ψ, ωz0 = ψ̇ + φ̇ cos θ

Let’s restrict attention for now to the symmetric top, an axially-symmetric body with I1 = Ix0 = I2 = Iy0
1 
T = I1 (ωx20 + ωy20 ) + I3 ωz20
2
In the absence of external forces
1
1 
T =L= I1 (ωx20 + ωy20 ) + I3 ωz20 = I1 (θ̇2 + sin2 θ φ̇2 ) + I3 (ψ̇ + cos θ φ̇)2
2 2
Let me introduce the canonical momenta conjugate to the Euler angles
∂L ∂L ∂L
pψ = = I3 (ψ̇ + cos θ φ̇), pφ = = (I1 sin2 θ + I3 cos2 θ)φ̇ + I3 cos θ ψ̇, pθ = = I1 θ̇
∂ ψ̇ ∂ φ̇ ∂ θ̇
These gadgets are independent dynamical degrees of freedom, but the usual Lx , Ly , Lz non-canonical angular mo-
menta are not all independent;
L := I · ω
which for the symmetric top are (in the rotating frame)

Lx0 = I1 (θ̇ cos ψ + φ̇ sin θ sin ψ), Ly0 = I1 (−θ̇ sin ψ + φ̇ sin θ cos ψ), Lz0 = I3 (ψ̇ + φ̇ cos θ)

and we can eliminate the velocities to get

sin ψ cos θ sin ψ cos ψ cos θ cos ψ
Lx0 = pφ − pψ + pθ cos ψ, Ly0 = pφ − pψ − pθ sin ψ, Lz0 = pψ
sin θ sin θ sin θ sin θ

69
 z0 L0
ω0
Consider as our body a disc of mass m, radius a. We will
rotate instantaneously about an axis making an angle α with
the symmetry axis. The body axes z 0 , y 0 and the ω 0 vector will
remain coplanar. Let the rotation vector in the body frame be
y0
ω 0 = (0, ω sin α, ω cos α) = ωx0 i0 + ωy0 j0 + ωz0 k0

x0
The angular momentum in the body-fixed frame is
1 1
L0 = (0, ma2 ω sin α, ma2 ω cos α)
4 2
angular momentum makes an angle with the body-fixed symmetry axis (k0 ) of

L0y I1
tan β = = tan α
L0z I3

Note that these three vectors, k0 , L0 , ω 0 are coplanar, and so in any frame gotten by a rigid rotation will still be
coplanar.

The non-canonical momenta obey Newton’s laws in an inertial frame

L̇ = N

which we can transform by rigid rotation into the frame of our choosing, and we select the primed frame.
 ·  
L̇ = eωtMn L0 = eωtMn ωMn L0 + L̇0 = N , ωMn L0 + L̇0 = ω 0 × L0 + L̇0 = e−ωtMn N = N 0

in the torque-free case

I − I  I − I  I − I 
3 1 3 1 1 1
ω̇x0 + ωz0 ωy0 = 0, ω̇y0 − ωz0 ωx0 = 0, ω̇z0 = 0 = ωx0 ωy0
I1 I1 I3

Since ωz0 is constant, we will call it ω30 (note ω30 = ω cos α), then with I1 = I2 = I we have
I − I 
3
0 = ω̇x0 + ω3 ωy0
I
0 = ω̇x0 + Ωωy0
0 = ω̇y0 − Ωωx0

The solutions are |ω⊥ | = ω sin α, ωz0 = ω cos α

I − I  I − I 
3 3
ωx0 = ω sin α sin Ωt, ωy0 = ω sin α cos Ωt, Ω = ω30 =ω cos α
I I
(comparison with the intial value of ω 0 shows that the magnitudes of the x0 , y 0 components are ω sin α). We see ω 0
precesses about the symmetry axis k0 , and for that matter so does the body-fixed angular momentum

L0 = I(ωx0 i0 + ωy0 j 0 ) + I3 ω30 k0

however the angular momentum in the space-fixed frame must be constant in magnitude and direction since there
are no torques.

70
15.1 Rolling cone description
L, k ω
The stable motion of our disk can be easily described by noting
that the vectors L, ω and k0 must all remain coplanar. Since
L is fixed in space (there are no torques, angular momentum
k0 is conserved in an inertial frame) ω and k0 rotate about L.
As ω moves about L it traces out a cone (the space-fixed
cone) whose axis is L and the k0 vector forms the axis of a
body-fixed cone of half-angle α. These cones meet along the
ω axis. In order that L be between ω and k0 , the motion
of the body is such that it is as if the body-cone rolls
without slipping on the space-cone. In this picture,
LON stands for line-of-nodes, the line along which the plane
perpendicular to the body-fixed cone at its apex intersects
the plane perpendicular to the space-fixed cone at its ver-
tex. Then since L, ω and k0 all remain coplanar, picking
L and k parallel means that the LON rotates in the dotted
plane ⊥ to L at φ̇, which is the rate at which k0 rotates around k.

LON

In this figure, the line of nodes points out.

Recall that in terms of Euler angles, the angle between k0 and
L ω k is θ. We will take the fixed L vector (no torques) to point
k0 in the k direction, and so β = θ. I will show that the motion
is described by two cones, the body-fixed of angle α rolling
without slipping on the space-fixed of angle β.
β How fast do ω and k0 precess around L (the space-fixed angular
α momentum vector)? We will show that the vector k0 moves on
a circle of radius sin β about L.
Notation; ω is ω, Ω is the rate with which ω precesses around
k0 , φ̇ is the rate with which ω precesses around L, which is
fixed in space.

Let’s find the values of θ, θ̇ and φ̇ in terms of ω, α and β at

t = 0. Remember that at t = 0, you can take ψ = 0 for an
axially symmetric body. At t = 0 then only the second parts of
ωx0 , ωy0 describe precession, so are nonzero

ωx 0 = θ̇ cos ψ + φ̇ sin θ sin ψ = ω sin α sin Ωt, =⇒ θ̇ = 0

ωy0 = −θ̇ sin ψ + φ̇ sin θ cos ψ = ω sin α cos Ωt,
q
0
(ωx0 )2 + (ωy0 )2 = |ω⊥ | = φ̇ sin θ = ω sin α
ωz0 = ψ̇ + φ̇ cos θ = ω cos α = ω30

If we can establish the direction of L at this time relative to either set of axes, since it is conserved we have it
established for all t. We’ve got φ̇ 6= 0, ψ̇6=0 but θ̇ = 0 so that k0 will precess in a circle at rate φ̇ about the space-fixed
k axis at Euler angle θ by definition of the Euler angles, and about L0 if we pass to a frame with L0 fixed (in other
words the space-fixed frame) at angle β from the calculations above, so θ = β and L is in the z-direction at t = 0.

71
Solve for φ̇, the rate with which ω precesses around L a fixed axis in space;
sin α
ω sin α = φ̇ sin β, φ̇ = ω the ”rolling cone condition”
sin β

The rate φ̇ is called the precession rate of the line of nodes.

Let’s show that φ̇ is the rate with which ω precesses around L, which we will take to be in the k direction. Write
the components in the SFF
   
θ̇ cos φ + ψ̇ sin θ sin φ ψ̇ sin β sin φ
ω =  θ̇ sin φ − ψ̇ sin θ cos φ  =  −ψ̇ sin β cos φ 
φ̇ + ψ̇ cos θ φ̇ + ψ̇ cos β

and so if φ̇ and ψ̇ are constant, this describes a vector precessing about the k axis at rate φ̇. Express the vector L
at the instant that it lies in the k0 j 0 plane

L = (0, I1 ωy0 , I3 ωz0 )T = (0, L sin β, L cos β)T

then
I1 ωy0 I1 ω sin α
sin β = q = q
(I1 ωy0 )2 + (I3 ωz0 )2 ω I12 sin2 α + I32 cos2 α
I1 ω sin α sin α
= p
2 2 2
=q 2 −I 2 )
2
ω I1 + (I3 − I1 ) cos α (I
1 + 3I 2 1 cos2 α
1

I3 ωz0 I3 cos α
cos β = q = q
0 2 0
(I1 ωy ) + (I3 ωz )2 I1 (I 2 −I 2 )
1 + 3I 2 1 cos2 α
1

then s s
ω sin α (I 2 − I 2 ) (I32 − I12 )
φ̇ = =ω 1 + 3 2 1 cos2 α = ω30 sec2 α +
sin β I1 I12

Now find ψ̇, the spin rate about the k0 axis, and φ̇ as well in terms of ω and α only.
tan α p p
ψ̇ = ω cos α−φ̇ cos β = ω cos α(1− ) = −ω cos α, ψ̇ = −ω cos α, and φ̇ = ω 1 + 3 cos2 α = ω30 sec2 α + 3
tan β
and
I3 − I1
ψ̇ = ω cos α − φ̇ cos β = −( )ω cos α
I1
which we could have predicted by comparing the ω 0 and ω column vectors directly.

We have not yet actually shown that the condition of rolling without slipping of the BFC on the SFC is equivalent
to the solutions of the Euler equations, which is all that we used to derive all of the various formulas for φ̇, ψ̇, θ and
Ω, so let’s address this now.

72
 I contend that the whole motion can be exhibited by making the outer
body-fixed cone in the picture roll without slipping on the inner space-fixed
cone (oblate body, I1 = I2 < I3 ), the line of contact being ω a generator of
each cone. Think of the body (disk) as attached to the outer body-fixed
ω cone, the horizontal plane (dotted) attached to the space-fixed. This you
k0 L can see from the no-slip condition for two disks one rolling within another,
which implies the rolling-cone condition;
k0 precesses around ω at Ω, always R sin α apart, and ω precesses at φ̇
around L (always diametrically opposed to k0 ) always R sin(α − β) from L.
The no-slip condition is gotten from writing the instantaneous speed of the
contact line/point two ways

Ω sin α = φ̇ sin(α − β)
I3 − I1 I3 cos α sin α I3 sin α cos α
( )ω3 sin α = φ̇(sin α q − cos α q ) = φ̇( − 1) q
I1 I1 1 + (I3 −I1 ) cos2 α
2 2 2 2
(I −I )
1 + 3 1 cos2 α I1 (I 2 −I 2 )
1 + 3 1 cos2 α
I12 I12 I12
 sin α 
ω30 sin α = ω cos α sin α = φ̇ cos α q , =⇒ ω sin α = φ̇ sin β
(I32 −I12 )
1+ I12
cos2 α

Example. A symmetric (I1 = I2 ) oblate body with I1 = 23 mk 2 and I3 = mk 2 (k is some radius of gyration) is given
ω = ω2 j + ω3 k at the instant represented by the left picture, t = 0, when primed/unprimed axes coincide. A short
time later (at ∆t) the object looks like the right figure. Both figures are views along the i (SF) axis.

k k

j j, k 0

Draw the BF and SF cones for both pictures.

Find a value of ω2 /ω3 for which the configuration of the right picture will occur given that k0 tips no further from
vertical.
Find ∆t, the time at which the right snapshot was taken, in terms of ω3 .

73
 k k

2β − α 2β − α
ω
L L
ω

α
α
j j, k 0

Solution:
ω2 I2 ω2 2
tan α = , tan β = = tan α
ω3 I3 ω3 3
The tip of k 0 is
√
ω2 3 13 π ω sin α 3
(2β − α) + α = π/2, β = π/4, = , ω= ω3 , ∆t = , φ̇ = = √ ω3
ω3 2 2 φ̇ sin β 2

74
16 Lecture 17. Lagrange’s case
Before we study the aspects of rigid body motion (Kowaleski’s case) that seeded some very important advances in
nonlinear dynamics, we perform the standard treatment of Lagrange’s case of the symmetric top with one point (not
the COM) fixed, in a gravitational field. Our previous example was Euler’s case (COM fixed, no torques). Both
Euler’s and Lagrange’s cases can be brought to full quadrature in terms of elliptic functions because of the presence
of several conserved quantities.

16.1 The Symmetric Top

The ”symmetric” in the desciption refers to I1 = I2 . We begin with the Lagrangian
I1 2 I3
L= (θ̇ + φ̇2 sin2 θ) + (ψ̇ + φ̇ cos θ)2 − M g` cos θ
2 2
for which there are several constants of the motion, ṗφ = 0 and ṗψ = 0;
∂H ∂H
ṗψ = − = 0, ṗφ = − =0
∂ψ ∂φ
since H = KE and there is no φ or ψ-dependence in the kinetic energy.
In addition Jacobi’s function (the total energy) is conserved
I1 2 I3
E= (θ̇ + φ̇2 sin2 θ) + (ψ̇ + φ̇ cos θ)2 + M g` cos θ
2 2
if there is no explicit time dependence written into the Lagrangian.
The constants of the motion can be used to eliminate several variables.
pφ − pψ cos θ pψ pψ pφ cos θ − pψ cos2 θ
φ̇ = 2 , and ψ̇ = − φ̇ cos θ = −( )
I1 sin θ I3 I3 I1 sin2 θ
Insert these into the Jacobi function to get
pφ pψ
I1 I1  I1 − I1 cos θ 2
E 0 = θ̇2 + + M gl cos θ
2 2 sin θ
Now let u = cos θ, −u̇ = θ̇ sin θ, and we arrive at
I1 2 1
E 0 (1 − u2 ) = u̇ + (pφ − pψ u)2 + M glu(1 − u2 )
2 2I1
solving for dt;
du
q = dt
2 0 1 2M gl
I1 E (1 − u2 ) − I12
(pφ − pψ u)2 − I1 u(1 − u2 )
which can be integrated
Z u(t)
du
q =t
2 0 1 2M gl
u(t0 )
I1 E (1 − u2 ) − I12
(pφ − pψ u)2 − I1 u(1 − u2 )

This is an elliptic integral, which are a few of my favorite things. The motion can in fact be very simply described.
Consider the function within the square root, it is a cubic that can be written as
2E 0 2M gl pψ pφ
f (u) = (1 − u2 )(α − βu) − (b − au)2 , with α = , β= , a= , b=
I1 I1 I1 I1
which can have up to three roots. This completes the full quadrature, since knowing θ(t) lets you get both φ, ψ by
integrating the canonical momenta equations.

17 index

75
Index
Abel transform, 45 Euler’s theorem, 54
angular momentum, 33 Euler-Lagrange equations, 12
areal velocity, 35 exact differential, 5
Atwood machine, 20
falling body, 66
bead on wire, 20 Fourier series, 36
bending of light, 13 friction, 18, 19
Bertrand’s theorem, 38
Bessel function, 37 Gateaux derivative, 12, 32
bilinear form, 57 General relativity, 14
body-fixed, 50 generalized force, 25
body-fixed cone, 71 generators of rotations, 55, 61
body-fixed frame, 28 geodesic, 13, 68
Bott periodicity, 58 gradient, 3
brachistochron, 12 group laws, 51
breaking constraint, 21
Hamiltonian, 32
calculus of variations, 11 Holonomic constraints, 5
canonical momenta, 69 holonomic constraints, 18
Center of mass separation, 33 homomorphism, 57
Christofel symbol, 68 hyperbolic trajectories, 44
Clifford algebra, 57
Clifford algebras, 56 incline, 19
closed orbits, 38 inclined plane, 19
closure, 51 integrating factor, 5
cohomology group, 10 isomorphism, 57
conserved quantity, 30 iterative solution, 66
constraint, 17
Jacobi integral, 21, 26
constraints, 9
Johnson noise, 8
Coriolis force, 66
coset, 10 KE decomposition, 3
critical angle, 21 Keplers’ third law, 35
cross-section, 43
curl, 10 Lagrange multipliers, 17
cycloid, 13 least action, 11
Lie algebra, 55, 59, 62
d’Alembert’s principle, 5 Lie group, 55, 61
DC-LRC circuit, 8 linear functional, 56
direction cosine, 50 Lorentz transformation, 31
directional derivative, 3
displacement, 66 magnetic field, 10
dissipative function, 7 magnetic focusing, 8
distance function, 56 magnetic lens, 9
dual, 56 matrix exponentiation, 59
matrix ray optics, 9
eccentric anomaly, 36 matrix representation, 58
eccentricity, 35 metric, 56
effective Lagrangian, 26 metric space, 13
electrodynamics, 7 moment of inertia, 27
elliptic functions, 35 momentum conservation, 31
Euler angle, 55, 61
Euler angles, 27, 55, 62 Nielsen form, 7

76
Noether’s theorem, 30 star-shaped region, 10
non-canonical momenta, 70 Strong law of action/reaction, 4
non-propagating degree of freedom, 25, 32 symmetry, 30
non-propagating DOF, 10
nonholonomic constraints, 18 toppling stick, 21
normal force, 19
normal mode, 17 universal enveloping algebra, 55, 62
normed vector space, 57
variational derivative, 12
ode, 67 vector, 60
orbit perturbation, 38 vector potential, 7
orbital correction, 41 vitual work, 25
orbital period, 35
orbits, 34
orthogonal transformation, 50

perihelion, 35, 36
Poincare’ upper half plane, 13
Post Newtonian approximation, 13
potential, 3
precession, 71
precession (orbit), 40
pseudo-vector, 64
punk bands, 36

quadrupolar field, 8, 10
quaternion, 52

reflection, 51
relative coordinates, 4
representation, 58
retrograde precession, 40
rolling, 19, 27
rolling cone description, 71
rolling disk, 4
rotation operator, 53
Runge-Lenz vector, 41

scalar potential, 7
scattering, 43
scattering angle, 44
scattering transform (backwards), 45
scattering transform (forward), 44
semimajor axi, 35
sesquilinearity, 56
slipping at contact point, 23
slotted disk, 9
small amplitude motion, 17
space-fixed cone, 71
space-fixed frame, 28
spin rate, 27
spin(3), 55, 62
spinor, 60
spiral orbit, 39
stable orbits, 38

77