Solutions to David Tong’s mf1 Lagrangian

problems.

Peeter Joot
August 25, 2008. Last Revision: Date : 2008/10/0321 : 50 : 02

Contents
1 Intro. 1
1.1 Problem 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Problem 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Problem 4. Double pendulum. . . . . . . . . . . . . . . . . . . . 5
1.4.1 N coupled pendulums. . . . . . . . . . . . . . . . . . . . 8
1.5 Problem 5. Pendulum on a rotating wheel. . . . . . . . . . . . . 8
1.5.1 Digression. Reduction of the Lagragian. . . . . . . . . . 9
1.5.2 Back to the problem. . . . . . . . . . . . . . . . . . . . . 11
1.6 Problem 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.1 Cylindrical Polar Coordinates. . . . . . . . . . . . . . . . 14
1.6.2 Problem 6i. . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.6.3 Problem 6ii. . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.7 Problem 7. Two circular constrained paths. Masses connected
by a spring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.7.1 Spring Potential? . . . . . . . . . . . . . . . . . . . . . . 20
1.8 Problem 8. Masses on string, one dangling. . . . . . . . . . . . 20
1.8.1 i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.8.2 ii. Motion of dangling mass not restricted to straight
down. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1 Intro.
These are my solutions to David Tong’s mf1 [Tong(b)] problem set (Lagra-
gian problems) associated with his excellent and freely available online text
[Tong(a)].

1
1.1 Problem 1.
I found it helpful to clarify for myself what was meant by gab (qc ). This is a
function of all the generalized coordinates:

gab (qc ) = gab (q1 , q2 , . . . , q N ) = gab (q)

So I think that a vector parameter reminder is helpful.
1
2 ∑ bc
L= g (q)q̇b q̇c

∂L 1 ∂g (q)
= ∑ q̇b q̇c bc a (1)
∂q a 2 ∂q
Now, procede to calculate the generalize momentums:

∂L 1 ∂(q̇b q̇c )
a
= ∑ gbc (q)
∂q̇ 2 ∂q̇ a
1
= ∑ gac (q)q̇c + gba (q)q̇b
2
= ∑ gab (q)q̇b

For
d ∂L ∂gab d b
dt ∂q̇ a
= ∑ ∂qd
q̇ q̇ + gba q̈b (2)

Taking the difference of 1 and 2 we have:

1 ∂g ∂gab d b
0= ∑ 2 q̇b q̇c ∂qbca − ∂qd
q̇ q̇ − gba q̈b
 
1 ∂gbc ∂gab
= ∑ q̇b q̇c − − gba q̈b
2 ∂q a ∂qc
 
1 ∂gbc 1 ∂gab 1 ∂gab
= ∑ q̇b q̇c − + + + gba q̈b
2 ∂q a 2 ∂qc 2 ∂qc
 
1 ∂g ∂gab ∂gac
= ∑ q̇b q̇c − bc + + + gba q̈b
2 ∂q a ∂qc ∂qb

Here a split of the symmetric expression

∂gab 1
X= ∑ q̇b q̇c ∂q c
= ( X + X ),
2
was used, and then an interchange of dummy indexes b, c.
Now multiply this whole sum by gba , and sum to remove the metric term
from the generalized acceleration

2
  
1 ∂gbc ∂gab ∂gac
∑ g gba q̈ = − 2 ∑ q̇ q̇ g − ∂qa + ∂qc + ∂qb
da b b c da

∑ δd b q̈b =
q̈d =

Swapping a, and d indexes to get form stated in the problem we have

 
1 ∂gbc ∂gdb ∂gdc
2∑
0 = q̈ a + q̇ b c ad
q̇ g − + + (3)
∂qd ∂qc ∂qb
= q̈ a + ∑ q̇b q̇c Γ a bc (4)
 
1 ad ∂gbc ∂gdb ∂gdc
Γ a bc = g − d + + b (5)
2 ∂q ∂qc ∂q

1.2 Problem 2.
1 2 4
L= m ẋ + m ẋ2 V − V 2
12

∂L d ∂L
=
∂x dt ∂ẋ 
d 1 2 3
m ẋ2 Vx − 2VVx = m ẋ + 2m ẋV
dt 3

When taking the time derivative of V, dV/dt 6= 0, despite no explicit time

dependence. Take an example, such as V = mgx, where the positional param-
eter is dependent on time, so the chain rule is required:

dV dV dx
= = ẋVx .
dt dx dt
Perhaps that’s obvious, but I made that mistake first doing this problem
(which would have been harder to make if I had used an example potential) the
1 2 4
first time. I subsequently constructed an alternate Lagrangian (L = 12 m ẋ −
2 2
m ẋ V + V ) that worked when this mistake was made, and emailed the author
suggesting that I believed he had a sign typo in his problem set.
Anyways, continuing with the calculation:

3
 m ẋ2 Vx − 2VVx = m2 ẋ2 ẍ + 2m ẍV + 2m ẋ2 Vx
m ẋ2 Vx − 2VVx − 2m ẋ2 Vx = m ẍ (m ẋ2 + 2V )
−(2V + m ẋ2 )Vx =

Cancelling left and right common factors, which perhaps not cooinciden-
tally equal 2E = V + 21 mv2 we have:

m ẍ = −Vx
This is what we’d get for our standard kinentic and position dependenct
Lagrangian too:
1 2
L= m ẋ − V
2

∂L d ∂L
=
∂x dt ∂ ẋ
d(m ẋ )
−Vx =
dt
−Vx = m ẍ

1.3 Problem 3.
Derive the relativistic equations of motion for a point particle in a position
dependant potential:
p
L = −mc2 1 − v2 /c2 − V (r)
The first thing to observe here is that for |v| << c, this is our familiar kinetic
energy Lagrangian

1 1 1 1
L = −mc2 (1 − v2 /c2 + (v/c)4 + · · · ) − V (r)
2 2 −2 2!
1
≈ −mc2 + mv2 − V (r)
2

The constant term −mc2 won’t change the equations of motion and we can
perhaps think of this as an additional potential term (quite large as we see from
atomic fusion and fission). For small v we recover the newtonian Kinetic en-
ergy term, and therefore expect the results will be equivalent to the newtonian
equations in that limit.

4
 Moving on to the calculations we have:
∂L d ∂L
i
=
∂x dt ∂ ẋi
d ∂L
q
∂V
− i = −c2 m i 1 − ∑( ẋ j )2 /c2
∂x dt ∂ ẋ
2 d 1 1 ∂L  
= −c m √
dt 2 1 − v2 /c2 ∂ ẋi
1 − ∑ ( ẋ j 2 2
) /c

d 1 1
= − c2 m √ (−2) ẋi /c2
dt 2 1 − v2 /c2
d 1
= m√ ẋi
dt 1 − v2 /c2
d
= mγ ẋi
dt
=⇒
 
∂ d
− ∑ ei i V = mγ ∑ ei ẋi
∂x dt
d(mγv)
−∇V =
dt

For v << c, gamma ≈ 1, so we get our Newtonian result in the limiting

case.
Now, I found this result very impressive result, buried in a couple line prob-
lem statement. I subsequently used this as the starting point for guessing about
how to formulate the Lagrange equations in a proper time form, as well as a
proper velocity form for this Kinetic and potential term. Those turn out to
make it possible to express Maxwell’s law and the Lorentz force law together
in a particularily nice compact covariant form. This catches me a up a bit in
terms of my understanding and think that I’m now at least learning and redis-
overing things known since the early 1900s;)

1.4 Problem 4. Double pendulum.

First consider a single pendulum (fixed length l).

x = l exp(iθ )
ẋ = li θ̇ exp(iθ )
| ẋ |2 = l 2 θ̇ 2

Now, if θ = 0 represents the downwards position at rest, the height above

that rest point is h = l − l cos θ. Therefore the Lagrangian is:

5
 1 2
L= mv − mgh
2
1
= ml 2 θ̇ 2 − mgl (1 − cos θ )
2

The constant term can be dropped resulting in the equivalent Lagrangian:

1 2 2
L0 = ml θ̇ + mgl cos θ
2
This amounts to a difference in the reference point for the potential energy,
so instead of measuring the potential energy V = mgh from a reference posi-
tion below the mass, one could consider that the potential has a maximum of
zero at the highest position, and decreases from there as:

V 0 = 0 − mgl cos θ.
Moving back to the EOMs that result from either form of Lagrangian, we
have after taking our derivatives:

d 2
−mgl sin θ = ml θ̇ = ml 2 θ̈
dt
Or,

θ̈ = − g/l sin θ
This is consistent with our expectations, and recovers the familiar small
angle SHM equation:

θ̈ ≈ − g/lθ.
Now, move on to the double pendulum, and compute the Kinetic energies
of the two particles:

x1 = l1 exp(iθ1 )
ẋ1 = l1 i θ̇1 exp(iθ1 )
| ẋ1 |2 = l1 2 θ̇12

6
 x2 = x1 + l2 exp(iθ2 )
ẋ2 = ẋ1 + l2 i θ̇2 exp(iθ2 )
= l1 i θ̇1 exp(iθ1 ) + l2 i θ̇2 exp(iθ2 )
| ẋ2 | = (l1 θ̇1 )2 + (l2 θ̇2 )2 + l1 i θ̇1 exp(iθ1 )l2 (−i )θ̇2 exp(−iθ2 ) + l1 (−i )θ̇1 exp(−iθ1 )l2 i θ̇2 exp(iθ2 )
2

= (l1 θ̇1 )2 + (l2 θ̇2 )2 + l1 l2 θ̇1 θ̇2 (exp(i (θ1 − θ2 )) + exp(−i (θ1 − θ2 )))
= (l1 θ̇1 )2 + (l2 θ̇2 )2 + 2l1 l2 θ̇1 θ̇2 cos(θ1 − θ2 )

Now calculate the potential energies for the two masses. The first has po-
tential of

V1 = m1 gl1 (1 − cos θ1 )
and the potential energy of the second mass, relative to the position of the
first mass is:
V20 = m2 gl2 (1 − cos θ2 )
But that is the potential only if the first mass is at rest. The total difference
in height from the dual rest position is:

l1 (1 − cos θ1 ) + l2 (1 − cos θ2 )

So, the potential energy for the second mass is:

V2 = m2 g (l1 (1 − cos θ1 ) + l2 (1 − cos θ2 ))

Dropping constant terms the total Lagrangian for the system is:

1 1
L= m v 2 + m2 v2 2 + m1 gl1 cos θ1 + m2 g (l1 cos θ1 + l2 cos θ2 )
2 1 1 2
1 1  
= m1 l1 2 θ̇12 + m2 (l1 θ̇1 )2 + (l2 θ̇2 )2 + 2l1 l2 θ̇1 θ̇2 cos(θ1 − θ2 )
2 2
+ m1 gl1 cos θ1 + m2 g (l1 cos θ1 + l2 cos θ2 )

Again looking at the resulting Lagrangian, we see that it would have been
more natural to measure the potential energy from a reference point of zero
potential at the horizontal position, and measure downwards from there:

V10 = 0 − m1 gl1 cos θ1

V20 = 0 − m2 g (l1 cos θ1 + l2 cos θ2 )

7
1.4.1 N coupled pendulums.
Now, with just two masses it is not too messy to expand out those kinetic en-
ergy terms, but for more the trig gets too messy. With the K2 term of the La-
grangian in complex form we have:

1 1 2
m1 l1 2 θ̇12 + m2 l1 θ̇1 + l2 θ̇2 exp(i (θ2 − θ1 ))

L=
2 2
+ m1 gl1 cos θ1 + m2 g (l1 cos θ1 + l2 cos θ2 )
1  2 1  2
= m1 l1 θ̇1 exp(iθ1 ) + m2 l1 θ̇1 exp(iθ1 ) + l2 θ̇2 exp(iθ2 )
2 2
+ m1 gl1 cos θ1 + m2 g (l1 cos θ1 + l2 cos θ2 )

By inspection we can also write the lagrangian for the N particle variant:
 2
1 N  j  N N
L= ∑ m j  ∑ lk θ̇k exp(iθk ) + g ∑ l j cos θ j ∑ mk
 
2 j =1 k =1 j =1 k= j

Can this be used to derive the wave equation?

If each of the N masses is a fraction m j = ∆m = M/N of the total mass, and
the lengths are all uniformly divided into segments of length l j = ∆l = L/N,
then the Lagrangian becomes:

 2
∆l  j
N  N N
L= ∑  ∑ θ̇k exp(iθk ) + ∑ cos θ j ∑ 1
 
2g j =1 k =1 j =1 k= j
 2
j
∆l N   N

2g j∑ ∑ k ∑ cos θ j
= exp ( iθ ) + ( N − j + 1 )

 θ̇ k 
=1 k =1 j =1


FIXME: return to this later. Want to finish typing up the rest of my solutions
for this ps.

1.5 Problem 5. Pendulum on a rotating wheel.

Express the position of the pivot point on the wheel with:

q1 = Re−iωt .
The position of the mass is then:

q2 = Re−iωt − ileiθ .

8
 The velocity of the mass is then:

q̇2 = −i (ω̇t + ω ) Re−iωt + l θ̇eiθ .

Let ωt = α, we have a Kinetic energy of:

1 1  2
m|q̇2 |2 = m −i α̇Re−iωt + l θ̇eiθ 

2 2
1  
= m R2 α̇2 + l 2 θ̇ 2 + 2Rl α̇θ̇Re(−ie−iα−iθ )
2
1  2 2 
= m R α̇ + l 2 θ̇ 2 + 2Rl α̇θ̇ cos(−α − θ − π/2)
2
1  2 2 
= m R α̇ + l 2 θ̇ 2 − 2Rl α̇θ̇ sin(α + θ )
2

The potential energy in the Lagrangian does not depend on the position of
the pivot, only the angle from vertical, so it is thus:

V = mgl (1 − cos θ )
V 0 = 0 − mgl cos θ

Depending on whether one measures the potential up from the lowest po-
tential point, or measures decreasing potential from zero at the horizontal. Ei-
ther way, combining the kinetic and potential terms, and dividing through by
ml 2 we have the Lagrangian of:
1 
L= ( R/l )2 α̇2 + θ̇ 2 − 2( R/l )α̇θ̇ sin(α + θ ) + ( g/l ) cos θ
2

1.5.1 Digression. Reduction of the Lagragian.

Now, in Tong’s solutions for this problem (which he emailed me since I ques-
tioned problem 2), he had α̇ = ω = constant, which allows the Lagrangian
above to be expressed as:

1  d
L= ( R/l )2 ω 2 + θ̇ 2 + (( R/l ) cos(ωt + θ )) + ω ( R/l ) sin(ωt + θ ) + ( g/l ) cos θ
2 dt
and he made the suprising step of removing that cosine term completely,
with a statement that it would not effect the dyanamics because it was a time
derivative. That turns out to be a generalized result, but I had to prove it to
myself. I also asked around on PF about this, and it wasn’t any named property
of Lagrangians, but was a theorm in some texts.
First consider the simple example of a Lagrangian with such a cosine deriva-
tive term added to it:

9
 d
L0 = L +
A cos(ωt + θ )
dt
and compute the equations of motion from this:

∂L0 d ∂L0
 
0= −
∂θ dt ∂θ̇
∂L d ∂L
 
∂ d d ∂ d
= − + A cos(ωt + θ ) − A cos(ωt + θ )
∂θ dt ∂θ̇ ∂θ dt dt ∂θ̇ dt
∂L d ∂L
 
∂ d ∂
= − − Aθ̇ sin(ωt + θ ) + Aθ̇ sin(ωt + θ )
∂θ dt ∂θ̇ ∂θ dt ∂θ̇
∂L d ∂L
 
∂ d
= − − Aθ̇ sin(ωt + θ ) + A sin(ωt + θ )
∂θ dt ∂θ̇ ∂θ dt
∂L d ∂L
 
= − − Aθ̇ cos(ωt + θ ) + Aθ̇ cos(ωt + θ )
∂θ dt ∂θ̇
∂L d ∂L
 
= −
∂θ dt ∂θ̇

Now consider the general case, altering a Lagrangian by adding the time
derivative of a positional dependent function:

df
L0 = L +
dt
and compute the equations of motion from this more generally altered func-
tion:
∂L0 d ∂L0
 
0= −
∂qi dt ∂qi
∂L d ∂L
 
∂ df d ∂ df
= i− + i −
∂q dt ∂qi ∂q dt dt ∂qi dt

Now, if f (q j , q̇ j , t) = f (q j , t) we have:

df ∂f ∂f
dt
= ∑ ∂q j q̇ j + ∂t

10
 We want to see if the following sums to zero:
   
∂ df d ∂ df ∂ ∂f ∂f d ∂ ∂f j ∂f
∂qi dt
−
dt ∂qi dt
= ∑ ∂qi ∂q j q̇ j
+
∂t
−
dt ∂q̇i ∑ ∂q j
q̇ +
∂t
2
∂ f 2
∂ f d

∂f 2
∂ f

= ∑ i j q̇ j + i − ∑ δij j + i
∂q ∂q ∂q ∂t dt ∂q ∂ q̇ ∂t
∂2 f j ∂2 f d ∂f
=∑ j
q̇ + i −
i
∂q ∂q ∂q ∂t dt ∂qi
∂2 f j ∂2 f 2
j ∂ f ∂2 f
=∑ q̇ + − ∑ q̇ −
∂qi ∂q j ∂qi ∂t ∂q j ∂qi ∂t∂qi

Therefore provided the function is sufficiently continuous that all mixed

pairs of mixed partials are equal, this is zero, and the d f /dt addition does not
change the equations of motion that the Lagrangian generates.

1.5.2 Back to the problem.

Now, return to the Lagrangian for this problem, and compute the equations of
motion. Writing µ = R/l, we have:
1 2 2 
L= µ α̇ + θ̇ 2 − 2µα̇θ̇ sin(α + θ ) + ( g/l ) cos θ
2

d ∂− ∂L
0=
dt ∂θ̇ ∂θ
d

= θ̇ − µα̇ sin(α + θ ) + µα̇θ̇ cos(α + θ ) + ( g/l ) sin θ
dt
= θ̈ − µα̈ sin(α + θ ) − µα̇ cos(α + θ )(α̇ + θ̇ ) + µα̇θ̇ cos(α + θ ) + ( g/l ) sin θ

Sure enough we have a cancelation of terms for constant ω. In general we

are left with:

θ̈ = µα̈ sin(α + θ ) + µα̇2 cos(α + θ ) − ( g/l ) sin θ

Or,

θ̈ = µ(ω̈t + 2ω̇ ) sin(ωt + θ ) + µ(ω̇t + ω )2 cos(ωt + θ ) − ( g/l ) sin θ

For constant ω, this is just:

θ̈ = µω 2 cos(ωt + θ ) − ( g/l ) sin θ

11
1.6 Problem 6.
First part of the problem is to show that the Lagrangian:
1
L= mv2 − qφ + qv/c · A
2
is equivalent to the Lorentz force law.
When I first tried this problem I had trouble with it, and also had trou-
ble following the text for the same in Tong’s paper. Later I did the somewhat
harder problem of exactly this, but for the covariant form of the Lorentz force
law, so I thought I’d come back to this and try again.
First step that seemed natural was to put the equation into four vector form,
despite the fact that the proper time Lagrangian equations were not going to
be used to produce the equation of motion. For just the Lorentz part of the
Lagrangian we have:

L0 = −φ + v/c · A
= −φ ∑ vi /cA j σi · σj
| {z }
1 (γ
2 i0j0 + γ j0i0 )=− γi · γ j

1
= − φc + ∑ vi Ai γi 2
c
1
= − φcγ0 2 + ∑ vi Ai γi 2
c

Thus with v = cγ0 + ∑ vi γi = ∑ vµ γµ , and A = φγ0 + ∑ Ai γi = ∑ Aµ γµ ,

we can thus write the complete Lagrangian as:
1
L= mv2 − qA · v/c
2
As usual we recover our vector forms by wedging with the time basis vec-
tor:

A ∧ γ0 = ∑ Ai γi0 = ∑ Ai σi = A
and v ∧ γ0 = · · · = v.
Notice the sign in the potential term, which is negative, unlike the same
Lagrangian in relativistic (proper) form: L = 21 mv2 + qA · v/c. That difference
is required since the lack of the use of time as one of the generalized coordinates
will change the signs of some of the results.
Now, this doesn’t matter for this particular problem, but also observe that
this lagrangian is almost in its proper form. All we have to do is add a − 12 mc2
constant to it, which should not effect the equations of motion. Doing so yields:

12
  
1 1 2
L= m(−c2 + v2 ) − qA · v/c = − mv + qA · v/c
2 2
I didn’t notice that until writing this up. So we have the same Lagrangian in
both cases, which makes sense. Whether or not one gets the traditional Lorentz
force law from this or the equivalent covariant form depends only on whether
one treats time as one of the generalized coordinates or not (and if doing so, use
proper time in the place of the time deriviatives when applying the Lagrange
equations). Cool.
Anyways, now that we have a more symmetric form of the Lagrangian, lets
compute the equations of motion.

∂L d ∂L
i
=
∂x dt ∂ẋi 
d i ∂v
= mv − q/cA · i
dt ∂ ẋ
d  
= mvi − q/cA · γi
dt
d  i 
= mv + q/cAi
dt
∂Ai
= pi + q/c ∑ ẋ j j
∂x
∂A
q · v/c =
∂xi
=⇒
∂Ai
 
∂A
ṗi = −q/c i
· v − ∑ ẋ j j
∂x ∂x

∂Aµ ν i 
j ∂A
= −q/c ∑ ∂xi
v γ µ · γ ν − ∑ v
∂x j
∂A0 0 2 ∂A j j 2 i
 
j ∂A
= −q/c ∑ ∂xi
v γ0 + ∑ ∂xi j ∑ ∂x j
v γ − v

=⇒
∂A0 0 2 ∂A j j 2 ∂Ai
 
∑ σi ṗi = p = q/c ∑ σi ∂x i
−v γ0 −
∂x i
v γj − v j j
∂x
 j i 
∂A ∂A
= −q∇φ + ∑ σi v j − j
∂xi ∂x

Now, it’s not obvious by looking, but this last expression is v × (∇ × A).
Let’s verify this by going backwards:

13
 1
v × (∇ × A) = (v ∧ (∇ × A))
i
1
= (v(∇ × A) − (∇ × A)v)
2i  
1 1 1
= v (∇ ∧ A) − (∇ ∧ A)v
2i i i
1
= − (v(∇ ∧ A) − (∇ ∧ A)v)
2
= (∇ ∧ A) · v
∂A j
= ∑ vk σi (σj · σk ) − σj (σi · σk )
∂xi
∂A j
= ∑ vk ∂xi σi δjk − σj δik
∂A j ∂A j
= ∑ v j ∂xi σi − vi ∂xi σj
∂A j ∂Ai
= ∑ v j ∂xi σi − v j ∂x j σi
 j
∂Ai

∂A
= ∑ v σi ∂xi − ∂x j
j

Therefore the final result is our Lorentz force law, as expected:

p = −q∇φ + qv/c × (∇ × A)

1.6.1 Cylindrical Polar Coordinates.

The next two parts of question 6 require cylindrical polar coordinates. I found
a digression was useful (or at least interesting), to see if the gradient followed
from the Lagrangian as was the case with non-orthonormal constant frame
basis vectors.
The first step required for this calculation (and the later parts of the prob-
lem) is to express the KE in terms of the polar coordinates. We need the velocity
to do so:

r = e3 z + e1 reiθ
ṙ = e3 ż + e1 (ṙ + r θ̇i )eiθ
2
|ṙ| = ż2 + ṙ + r θ̇i 


= ż2 + ṙ2 + (r θ̇ )2

14
 Now, form the Lagrangian of a point particle with a non-velocity dependent
potential:
1  2
2 
L= m ż + ṙ2 + r θ̇ −φ
2
and calculate the equations of motion:

∂L ∂L 0
 
=
∂z ∂ż
∂φ
− = (mż)0
∂z

0
∂L ∂L

=
∂r ∂ṙ
∂φ
− + mr θ̇ 2 = (mṙ )0
∂r

∂L ∂L 0
 
=
∂θ ∂θ̇
∂φ  0
− = mr2 θ̇
∂θ

There are a few things to observe about these equations. One is that we can
assign physically significance to an expression such as mr2 θ̇. If the potential
has no θ dependence this is a conserved quantity (angular momentum).
The other thing to observe here is that the dimensions for the θ coordinate
equation result has got an extra length factor in the numerator. Thus we can’t
multiply these with our respective frame vectors and sum. We can however
scale that last equation by a factor of 1/r and then sum:

 1  
 ∂ ∂ 1 ∂
ẑ(mż)0 + r̂ (mṙ )0 − mr θ̇ 2 + (mr2 θ̇ )0 = − ẑ + r̂ + φ
r ∂z ∂r r ∂θ

For constant mass this is:

   
2 1 2 ∂ ∂ 1 ∂
m ẑz̈ + r̂(r̈ − r θ̇ ) + 2rṙ θ̇ + r θ̈ = − ẑ + r̂ + θ̂ φ
r ∂z ∂r r ∂θ

However, is such a construction have a meaningful physical quantity? One

can easily imagine more complex generalized coordinates where guessing scale
factors in this fashion would not be possible.

15
 Let’s compare this to a calcultion of acceleration in cylindrical coordinates.

r̈ = e3 z̈ + e1 (r̈ + r θ̈i + ṙ θ̇i + (ṙ + r θ̇i )i θ̇ )eiθ

= e3 z̈ + e1 (r̈ + r θ̈i + 2ṙ θ̇i − r θ̇ 2 )eiθ
= ẑz̈ + r̂(r̈ − r θ̇ 2 ) + θ̂(r θ̈ + 2ṙ θ̇ )

Sure enough, the ad-hoc vector that was constructed matches the accerla-
tion vector for the constant mass case, so the right hand side must also define
the gradient in cylindrical coordinates.

∂ ∂ 1 ∂
∇ = ẑ + r̂ + θ̂
∂z ∂r
 r ∂θ 
∂ ∂ i ∂
= ẑ + r̂ +
∂z ∂r r ∂θ

Very cool result. Seeing this I finally understand when and where state-
ments like ”angular momentum is conserved” is true. Specifically it requires a
potential that has no angular dependence (ie: like gravity acting between two
point masses.)
I never found that making such an angular momentum conservation ”law”
statement to be obvious, even once the acceleration was expressed in a ra-
dial decomposition. This is something that can be understood without the
Lagrangian formulation. To do so the missing factor is that before a conser-
vation statement like this can be claimed one has to first express the gradient
in cylindrical form, and then look at the coordinates with respect to the gen-
eralized frame vectors. Conservation of angular momentum depends on an
appropriately well behaved potential function! Intuitively, I understood that
something else was required to make this statement, but it took the form of an
unproven axiom in most elementary texts.
FIXME: generalize this and prove to myself that angular momentum is con-
served in a N-body problem and/or with a rigid body rotation constraint on
N − 1 of the masses.

1.6.2 Problem 6i.

The Lagrangian for this problem is:
1
mv2 − eA · v
L=
2
Given a cylindrical decomposition, our velocity is:

16
 r = zẑ + rr̂
ṙ = żẑ + r r̂˙ + ṙr̂
= żẑ + r̂(ṙ + r θ̇i )
= żẑ + ṙr̂ + r θ̇ θ̂

The specific potential for the problem, using (z, θ, r ) coordinates is:

f (r )
A = θ̂
r
Therefore the Lagrangian is:

1  2  f (r )
L= m ż + ṙ2 + (r θ̇ )2 − e r θ̇
2 r
so the equations of motion for the z, θ, and r coordinates (respectively) are:

(mż)0 = 0
(mr2 θ̇ − e f (r ))0 = 0
(mṙ )0 = mr θ̇ 2 − e f 0 (r )θ̇

From second of these equations we have:

mr2 θ̇ − e f (r ) = K
In particular this is true for r = r (t0 ) = r0 , so

mr0 2 θ̇0 − e f (r0 ) = K

Or,
 r 2 e
0
θ̇ (t) − θ̇ (t0 ) = ( f (r ) − f (r0 ))
r mr2
Now, the problem is to show that
e
θ̇ = ( f (r ) − f (r0 ))
mr2
I don’t see how that follows? Ah, I see, the velocity is in the (r, z) plane for
t = 0, so θ̇ (t0 ) = 0.

17
1.6.3 Problem 6ii.
The potential for this problem is

A = rg(z)θ̂
Therefore the Lagrangian is:

1  2 
L= m ż + ṙ2 + (r θ̇ )2 − er2 g(z)θ̇
2
Taking θ, r, z derivatives:
 0
0 = mr2 θ̇ − er2 g(z)

mr θ̇ 2 − 2erg(z)θ̇ = (mṙ )0
−er2 g0 θ̇ = (mż)0
One constant of motion is:

mr2 θ̇ − er2 g(z) = K

Looking at Tong’s solutions another is the Hamiltonian. (I’ve got to go back
and read that Hamiltionian stuff since this didn’t occur to me).

θ̇ = (e/m) g(z) + (r0 /r )2 θ̇0 − (e/m) g(z0 )

With θ̇0 = 2eg(z0 )/m this is:

  r 2 
e 0
θ̇ = g(z) + g ( z0 )
m r
FIXME: think through the remainder bits of this problem more carefully.

1.7 Problem 7. Two circular constrained paths. Masses con-

nected by a spring.
With i = e1 ∧ e2 , the paths, (squared) speeds and separation of the masses can
be written:

18
 q1 = e1 R1 eiθ
q2 = ce3 + e1 ( ai + R2 eiα )
|q̇1 |2 = ( R1 θ̇ )2
|q̇2 |2 = ( R2 α̇)2
d2 = ( q1 − q2 )2
 2
= c2 + ai + R2 eiα − R1 eiθ 
 
 
= c2 + a2 + R2 2 + R1 2 + ai R2 e−iα − R1 e−iθ − R2 eiα + R1 eiθ
− R1 R2 (eiα e−iθ + e−iα eiθ )
= c2 + a2 + R2 2 + R1 2 + 2a( R2 sin α − R1 sin θ ) − 2R1 R2 cos(α − θ )

With the given potential:

1 2 2
V=
ω d
2
We have the following Lagrangian (where the constant terms in the separa-
tion have been dropped) :

1 1
L= m ( R θ̇ )2 + m2 ( R2 α̇)2 + ω 2 ( a( R2 sin α − R1 sin θ ) − R1 R2 cos(α − θ ))
2 1 1 2
Last part of the problem was to show that there is an additional conserved
quantity when a = 0. The Lagrangian in that case is:
1 1
L= m ( R θ̇ )2 + m2 ( R2 α̇)2 − R1 R2 ω 2 cos(α − θ )
2 1 1 2
Evaluating the Lagrange equations, for this condition one has:

− R1 R2 ω 2 sin(α − θ ) = (m1 R1 2 θ̇ )0
R1 R2 ω 2 sin(α − θ ) = (m2 R2 2 α̇)0

Summing these one has:

(m1 R1 2 θ̇ )0 + (m2 R2 2 α̇)0 = 0

Therefore the additional conserved quantity is:

m1 R1 2 θ̇ + m2 R2 2 α̇ = K

FIXME: Is there a way to identify such a conserved quantity without eval-

uating the derivatives? Noethers?

19
1.7.1 Spring Potential?
Small digression. Let’s take the gradient of this spring potential and see if this
matches our expectations for a −kx spring force.

−∇d V = −ω 2 dd̂ = −ω 2 d
Okay, this works, ω 2 = k, which just expresses the positiveness of this
constant.

1.8 Problem 8. Masses on string, one dangling.

1.8.1 i
The second particle hangs straight down (also Goldstein problem 9, also exam-
ple 2.3 in Hestenes NFCM.) First mass m1 on the table, and second, hanging.
The kinetic term for the mass on the table was calculated above in problem
7, so adding that and the KE term for the dangling mass we have:
1   1
K= m1 ṙ2 + (r ψ̇)2 + m2 ṙ2
2 2
Our potential, measuring down is:

V = 0 − m2 g ( l − r )
Combining the KE and PE terms and dropping constant terms we have:
1   1
L= m1 ṙ2 + (r ψ̇)2 + m2 ṙ2 − m2 gr
2 2
The ignorable coordinate is ψ since it has only derivatives in the Lagrangian.
EOMs are:

0
0 = (m1 r2 ψ̇)
m1 r ψ̇2 − m2 g = ((m1 + m2 )ṙ )0 = Mr̈

The first equation here is a conservation of angular momentum statement,

whereas the second equation has all the force terms that lie along the string (ra-
dially above the table, and downwards below). We see the r ψ̇2 = rω 2 angular
acceleration component when calculating radial and non-radial component of
acceleration.
Goldstein asks here for the equations of motion as a second order equation,
and to integrate once. We can go all the way, but only implicitly, as we can
write t = t(r ), using ṙ as an integrating factor:

20
 m1 r2 ψ̇ = m1 r0 2 ω0
=⇒
 r 2
0
ψ̇ = ω0
r
=⇒
r0 4 2
m1 ω0 − m2 g = Mr̈
r3
=⇒
r0 4
m1 ṙ 3 ω0 2 − m2 gṙ = Mṙr̈
r
 0  0
1
−m1 r0 2 2 ω0 2 − m2 gṙ = M ṙ2
r
1
K − m1 r0 4 2 ω0 2 − m2 gr = Mṙ2
r
=⇒
K = m1 r0 2 ω0 2 + m2 gr0 + Mṙ02

r02
m 1 ω0 2 r 0 2 1 − + Mṙ02 − m2 g (r − r0 ) = Mṙ2
r2
=⇒
dr
Z
t= r  
m1 r02 m2
2 2
M ω0 r 0 1− r2
+ ṙ02 − M g (r − r0 )

We can also write ψ = ψ(r ), but that doesn’t look like it is any easier to
solve:

dψ dr
ψ̇ =
dr dt
=⇒
dψ dt  r0 2
= ω0
dr dr r
r0 2 ω0 dr
Z
ψ= r  
m1 2
r2 2 2 1 − r0 + ṙ02 − m2
− r0 )
M ω0 r 0 r2 M g (r

1.8.2 ii. Motion of dangling mass not restricted to straight down.

This part of the problem treats the dangling mass as a spherical pendulum. If
θ is the angle from the verticle and α is the angle in the horizontal plane of

21
motion, we can describe the coordinate of the dangler (pointing ẑ = ĝ down-
wards), as:

q2 = R(sin θ cos α, sin θ sin α, cos θ )

and the velocity as:

q̇2 = Ṙ(sin θ cos α, sin θ sin α, cos θ )

+ R(cos θ cos α, cos θ sin α, − sin θ )θ̇
+ R(− sin θ sin α, sin θ cos α, 0)α̇

and can then attempt to square this mess to get the squared speed that we
need for the kinetic energy term of the Lagrangian. Instead, lets choose an
alternate parameterization:

q2 = R cos θ ẑ + e1 R sin θeiα

q̇2 = Ṙ cos θ − R sin θ θ̇ ẑ + e1 eiα Ṙ sin θ + R cos θ θ̇ + R sin θi α̇




2
2
|q̇2 |2 = Ṙ cos θ − R sin θ θ̇ + Ṙ sin θ + R cos θ θ̇ + ( R sin θ α̇)2
= Ṙ2 + ( Rθ̇ )2 + ( R sin θ α̇)2

Our potential is

V = 0 − m2 g(l − r ) cos θ,
so, the Lagrangian is therefore:

1   1  
L= m2 ṙ2 + (l − r )2 (θ̇ 2 + sin θ α̇)2 + m1 ṙ2 + (r ψ̇)2 + m2 g(l − r ) cos θ
2 2

References
[Tong(a)] Dr. David Tong. Classical mechanics. ”[Link]
uk/user/tong/dynamics/[Link]”, a.

[Tong(b)] Dr. David Tong. Classical mechanics lagrangian problem set 1.

”[Link] b.

22