Noether’s Theorem

September 15, 2014

We continue looking at consequences of Noether’s theorem, applying it to 2-dimensional rotations, scal-

ings, and time translations. Recal that the conserved quantity for an infinitesimal symmetry δxi = εi (x)
is
∂L (x (λ)) i
I= ε (x)
∂ ẋi

1 Rotational symmetry and conservation of angular momentum (2

dim)
Consider a 2-dimensional system with free-particle Lagrangian
1
m ẋ2 + ẏ 2


L (x, y) =
2
The rotation
x → x0 = x cos θ − y sin θ
y → y 0 = x sin θ + y cos θ
for any fixed value of θ leaves the action unchanged,
ˆ
S [x] = Ldt

invariant.
For an infinitesimal change, θ  1, the changes in x, y are
ε1 = δx
= x0 − x
= x cos θ − y sin θ − x
   
1 2 1 3
= x 1 − θ + ... − y θ − θ + ... − x
2! 3!
≈ −yθ
ε2 = δy
= y0 − y
≈ xθ
Therefore, from Noether’s theorem, we have the conserved quantity,
∂L i
ε (x) = mẋε1 + mẏε2
∂ ẋi
= mẋ (−yθ) + mẏ (xθ)
= θm (xẏ − y ẋ)

1
as long as x and y satisfy the equations of motion. Since θ is just an arbitrary constant to begin with, we
can identify the angular momentum,
J ≡ m (ẏx − ẋy)
as the conserved quantity.
It is worth noting that J is conjugate to a cyclic coordinate. If we rewrite the action in terms of polar
coordinates, (r, ϕ), it becomes
ˆt2
1
m ṙ2 + r2 ϕ̇2


S [r, ϕ] =
2
t1
so that ϕ is cyclic. The momentum conjugate to ϕ is
∂L
pϕ =
∂ ϕ̇
= mr2 ϕ̇
Since tan ϕ = xy ,
1 ẏ y ẋ
ϕ̇ = − 2
cos2 ϕ x x
xẏ − y ẋ
ϕ̇ = cos2 ϕ
x2
xẏ − y ẋ x2
 
=
x2 r2
giving the same result,
pϕ = mr2 ϕ̇ = m (xẏ − y ẋ) = J
We will generalize this result to 3-dimensions after a complete discussion of rotations.

2 Conservation of energy
Conservation of energy is related to time translation invariance. However, this invariance is more subtle
than simply replacing t → t + τ , which is simply a reparameterization of the action integral. Instead, the
conservation law holds whenever the Lagrangian does not depend explicitly on time so that
∂L
=0
∂t
The total time derivative of L then reduces to
dL X ∂L ∂L i ∂L
= i
ẋi + ẍ +
dt i
∂x ∂ ẋi ∂t
X ∂L ∂L i
= i
ẋi + ẍ
i
∂x ∂ ẋi

Using the Lagrange equations to replace

∂L d ∂L
i
=
∂x dt ∂ ẋi
in the first term, we get
dL X  d  ∂L  ∂L i

i
= ẋ + ẍ
dt i
dt ∂ ẋi ∂ ẋi
!
d X ∂L i
= ẋ
dt i
∂ ẋi

2
Bringing both terms to the same side, we have
!
d X ∂L
i
ẋ − L = 0
dt i
∂ ẋi

so that the quantity

X ∂L
E≡ i
ẋi − L
i
∂ ẋ
is conserved. The quantity E is called the energy.
For a single particle in a potential V (x), the conserved energy is
X ∂L
E ≡ ẋi−L
i
∂ ẋi
   
X ∂ 1 1
= ẋi i mẋ2 − V (x) − mẋ2 − V (x)
i
∂ ẋ 2 2
 
X 1
= mẋi ẋi − mẋ2 − V (x)
i
2
1
= mẋ2 + V (x)
2
For the velocity-dependent potential of the Lorentz force law,

ˆt2  
1
S [x] = mẋ2 − qφ + q ẋ · A dt
2
t1

so that
   
X 1 ∂ 1
E = mẋ2 − qφ + q ẋ · A −
ẋi mẋ2 − qφ + q ẋ · A
i
2 ∂ ẋi 2
 
X
i i i

1 2
= ẋ mẋ + qA − mẋ − qφ + q ẋ · A
i
2
1
= mẋ2 + q ẋ · A − (−qφ + q ẋ · A)
2
1
= mẋ2 + qφ
2
is conserved.

3 Scale Invariance
As we have noted, physical measurements are always relative to our choice of unit. The resulting dilatational
symmetry will be examined in detail when we study Hamiltonian dynamics. However, there are other forms
of rescaling a problem that lead to physical results. These results typically depend on the fact that the
Euler-Lagrange equation is unchanged by an overall constant, so that the actions
ˆ
S = Ldt
ˆ
S 0 = α Ldt

3
have the same extremal curves.
Now suppose we have a Lagrangian which depends on some constant parameters (a1 , . . . , an ) in addition
to the arbitrary coordinates,
L = L xi , ẋi , a1 , . . . , an , t

These parameters might include masses, lengths, spring constants and so on. Further, suppose that each of
these variables may be rescaled by some factor in such a way that S changes by only an overall factor. That
is, when we make the replacements

xi → αxi
t → βt
α i
ẋi → ẋ
β
ai → γi ai

for certain constants (α, β, γ1 , . . . , γn ) we find that

 
i α i
L αx , ẋ , γ1 a1 , . . . , γn an , βt = σL xi , ẋi , a1 , . . . , an , t


β

for some constant σ which

 depends on the scaling constants. Then the Euler-Lagrange equations for the
system described by L αxi , α i
β ẋ , γ1 a1 , . . . , γn an , βt are the same as for the original Lagrangian, and we
may make the replacements in the solution.
Consider the simple harmonic oscillator. The usual Lagrangian is
1 1
L= mẋ2 − kx2
2 2
If we rescale,

x̃ = αx
m̃ = βm
k̃ = γk
t̃ = δt

then
1 βα2 1
L̃ = mẋ2 − γα2 kx2
2 δ2 2
so as long as δβ2 = γ we have S̃ = γα2 S as a scaling symmetry. Scaling x doesn’t depend on the other scales,
so there’s no information there.
Now consider a system with unit mass and unit spring constant,

m0 = 1
k0 = 1

and suppose this system is periodic, with period T0 . Then rescaling, the mass, spring constant and period
become

m = βm0 = β
k = γk0 = γ
T = δT0

4
 q
β
and scale invariance tells us that a periodic solution also holds for the scaled m, k and T as long as δ = γ.
Therefore r
m
T = T0
k
q
k
and the frequency is proportional to m .