Bertrand’s Theorem

October 8, 2012

Circular orbits
The effective potential,
L2ϕ
Vef f = + V (r)
2µr2
has a minimum or maximum at r0 if and only if

dVef f 
0 =
dr r0
L2ϕ

dV 
= − 3+
µr0 dr r0
L2ϕ
= − − f (r0 )
µr03

so we must have
L2ϕ
f (r0 ) = −
µr03
At this radius, there is no net radial force, so that circular orbits are possible. Such orbits may be stable or
unstable, depending on the sign of the second derivative of the effective potential. Stable orbits occur only
if
d2 Vef f 

0 <
dr2 r0
3L2ϕ d2 V 

= +
µr04 dr2 r0
3L2ϕ df
= 4 − (r0 )
µr0 dr
L2
and since f (r0 ) = − µrϕ3 , we have
0
df 3
(r0 ) < − f (r0 )
dr r0

General case
Lϕ
Eliminating ϕ̇ = µr 2 , the remaining, radial equation of motion is

L2ϕ
µr̈ − − f (r) = 0
µr3

1
Write this as an orbit equation, using
dϕ dr
ṙ =
dt dϕ
Lϕ dr
=
µr2 dϕ
 
dϕ d Lϕ dr
r̈ =
dt dϕ µr2 dϕ
 2 !
Lϕ 2Lϕ dr Lϕ d2 r
= − 3 + 2 2
µr2 µr dϕ µr dϕ
2  2 2
2Lϕ dr Lϕ d2 r
= − 2 5 + 2 4 2
µ r dϕ µ r dϕ
Substituting,
L2ϕ
0 = µr̈ − − f (r)
µr3
2L2ϕ dr 2 L2ϕ d2 r L2ϕ
 
= − 5 + 4 2 − 3 − f (r)
µr dϕ µr dϕ µr
1
Now let r = u

dr 1 du
= −
dϕ u2 dϕ
d2 r
 
d 1 du
= − 2
dϕ2 dϕ u dϕ
 2
2 du 1 d2 u
= −
u3 dϕ u2 dϕ2
The equation of motion now becomes
2L2ϕ dr 2 L2ϕ d2 r L2ϕ
 
0 = − 5 + 4 2 − 3 − f (r)
µr dϕ µr dϕ µr
2  2 !
2 5  2 4
2Lϕ u 1 du Lϕ u 2 du 1 d2 u L2ϕ
= − − 2 + − − − f (r)
µ u dϕ µ u3 dϕ u2 dϕ2 µr3
2L2ϕ u du 2 2L2ϕ u du 2 L2ϕ u2 d2 u L2ϕ u3
   
= − + − 2
− − f (r)
µ dϕ µ dϕ µ dϕ µ
L2ϕ u2 d2 u L2ϕ u3
= − − − f (r)
µ dϕ2 µ
or
d2 u
 
µ 1
2
+ u = − 2 2f
dϕ Lϕ u u
Finally, write the force in terms of the potential,
 
1 dV
f = −
u dr
du dV
= −
dr du
dV
= u2
du

2
Now we have simply
d2 u µ dV
+u=− 2
dϕ2 Lϕ du
This is the equation of a driven harmonic oscillator. This striking set of transformations is what happens
when people spend 300 years working on a problem.
d2 u
Now consider circular orbits. This means that u does not change with ϕ at all, so we have dϕ 2 = 0 and

therefore, if we set
µ dV
h (u) = − 2
Lϕ du
then
u0 = h (u0 )
Next, expand u and the force for small perturbations about u0 ,

u = u0 + η
µ dV
− 2 = h (u)
Lϕ du
1
= h (u0 ) + h0 (u0 ) η + h00 (u0 ) η 2 + · · ·
2
Substituting these into the equation of motion,

d2 η 1
+ u0 + η = h (u0 ) + h0 (u0 ) η + h00 (u0 ) η 2 + · · ·
dϕ2 2
0 1 00
= u0 + h (u0 ) η + h (u0 ) η 2 + · · ·
2
d2 η 1 00
+ [1 − h0 (u0 )] η = h (u0 ) η 2 + · · ·
dϕ2 2

If 1 − h0 (u0 ) < 0, then the equation has exponential instead of oscillatory solutions, and the circular orbits
are not stable. For stable orbits, we therefore set

λ2 = 1 − h0 (u0 ) > 0

If we neglect the quadratic and higher terms on the right side of the equation, with initial condition η = 0
when ϕ = 0, we have solutions
η = A sin λϕ
In order to reproduce the initial conditions after some integer number, q, of complete orbits, we require

η (0) = η (2πq)
0 = A sin 2πqλ

so that
qλ = p
with p another integer. We see that λ must be rational,
p
λ=
q

3
Assuming the force is a continuous function of position, h0 is continuous and λ is also continuous. Therefore,
λ = pq for all nearly circular orbits. Returning to the definition of λ, and using the constancy of λ,

λ2 = 1 − h0 (u0 )
  
d µ 1
= 1− − 2 2f
du Lϕ u u
   
2µ 1 µ 0 1
λ2 − 1 = − 2 3f + 2 2f
Lϕ u0 u0 Lϕ u u0

Since the circular orbit satisfies

u0 = h (u0 )
 
µ 1
= − 2 2f
Lϕ u0 u0
L2ϕ 3
 
1
f = − u
u0 µ 0

so that
!
L2ϕ 3
 
2 2µ µ 0 1
λ −1 = − 2 3 − u0 + 2 2 f
Lϕ u0 µ Lϕ u u0
 2 
Lϕ 3
µ  − µ u0  0 1
 
λ2 − 1 = 2 + 2 2    f
Lϕ u0 f u10 u0
 
u0 1
λ2 − 3 = −  f0
f u10 u0

This must hold regardless of the value of u0 , so we may drop the subscript and integrate
df
du
− = λ2 − 3
f u
 
df 1
λ2 − 3 − 2 rdr


− =
f r
df dr
λ2 − 3


=
f r
2
−3
f (r) = Arλ

and in order to have stable, perturbatively closed orbits, the force law must be a rational power law. This
give h as
µ A
h =
L2ϕ u2 uλ2 −3
µA 1−λ2
= u
L2ϕ

Now return to the full equation of motion

d2 η 1 1
+ u0 + η = u0 + h0 (u0 ) η + h00 (u0 ) η 2 + h000 (u0 ) η 3 + · · ·
dϕ2 2 6

4
and expand in a Fourier series,

η (ϕ) = η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·

We also need powers of η. Keeping up to third order,

2
(η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·) = η02 + η0 η1 cos λϕ + η0 η2 cos 2λϕ + η0 η3 cos 3λϕ
+η0 η1 cos λϕ + η12 cos2 λϕ + η2 η1 cos λϕ cos 2λϕ
+η0 η2 cos 2λϕ + η1 η2 cos 2λϕ cos λϕ + η0 η3 cos 3λϕ
3
(η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·) = η03 η0 + η02 η1 cos λϕ + η02 η2 cos 2λϕ + η02 η3 cos 3λϕ + · · ·
+η02 η1 cos λϕ + η0 η12 cos2 λϕ + η2 η0 η1 cos λϕ cos 2λϕ
+η0 η02 η2 cos 2λϕ + η1 η0 η2 cos 2λϕ cos λϕ
+η02 η3 cos 3λϕ
+η02 η1 cos λϕ + η0 η12 cos2 λϕ + η0 η1 η2 cos λϕ cos 2λϕ
+η0 η12 cos2 λϕ + η13 cos3 λϕ
+η0 η2 η1 cos λϕ cos 2λϕ
+η02 η2 cos 2λϕ + η0 η1 η2 cos 2λϕ cos λϕ
+η0 η1 η2 cos 2λϕ cos λϕ
+η02 η3 cos 3λϕ

and using addition formulas

1
cos2 λϕ = (1 + cos 2λϕ)
2
1
cos λϕ cos 2λϕ = [cos (λϕ + 2λϕ) + cos (λϕ − 2λϕ)]
2
1
= [cos (3λϕ) + cos (λϕ)]
2
1
cos3 λϕ = cos λϕ (1 + cos 2λϕ)
2
1 1 1
= cos λϕ + cos (3λϕ) + cos (λϕ)
2 4 4
3 1
= cos λϕ + cos (3λϕ)
4 4
Therefore,
2 1
(η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·) = η02 + η12 + (2η0 η1 + η1 η2 ) cos (λϕ)
 2 
1 2
+ 2η0 η2 + η1 cos 2λϕ + (2η0 η3 + η1 η2 ) cos 3λϕ
2
3
η0 + 3η0 η1 cos λϕ + 3η02 η2 cos 2λϕ + 3η02 η3 cos 3λϕ
3 2




(η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·) =
+3η0 η12 cos2 λϕ + 6η0 η1 η2 cos 2λϕ cos λϕ + η13 cos3 λϕ
η03 + 3η02 η1 cos λϕ + 3η02 η2 cos 2λϕ + 3η02 η3 cos 3λϕ




=
1 1
+3η0 η12 (1 + cos 2λϕ) + 6η0 η1 η2 [cos (3λϕ) + cos (λϕ)]
2 2
3 1
+η13 cos λϕ + η13 cos (3λϕ)
4 4 
3 3 2 2 3 3
= η0 + η0 η1 + 3η0 η1 + 3η0 η1 η2 + η1 cos λϕ
2 4

5
    
3 1
+ 3η02 η2 + η0 η12 cos 2λϕ + 3η02 η3 + η13 + 3η0 η1 η2 cos 3λϕ
2 4
Substituting and expanding,
d2 η 1 1
+ u0 + η = u0 + h0 (u0 ) η + h00 (u0 ) η 2 + h000 (u0 ) η 3 + · · ·
dϕ2 2 6
0
−η1 λ2 cos λϕ − 4λ2 η2 cos 2λϕ − 9λ2 η3 cos 3λϕ + · · · + η


= h (u0 ) (η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·)
1 2
+ h00 (u0 ) (η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·)
2
1 3
+ h000 (u0 ) (η0 + η1 cos λϕ + η2 cos 2λϕ + η3 cos 3λϕ + · · ·)
6
Equating like terms,
1 1 1 1 1 3
η0= h0 (u0 ) η0 + h00 (u0 ) η02 + h00 (u0 ) η12 + h000 (u0 ) η03 + h000 (u0 ) η0 η12
 2 2 2 6  6 2 
2

0 1 00 1 000 2 3 3
−η1 λ + η1 cos λϕ = h (u0 ) η1 + h (u0 ) (2η0 η1 + η1 η2 ) + h (u0 ) 3η0 η1 + 3η0 η1 η2 + η1 cos λϕ
2 6 4
    
1 1 1 3
−4λ2 η2 cos 2λϕ + η2 cos 2λϕ = h0 (u0 ) η2 + h00 (u0 ) 2η0 η2 + η12 + h000 (u0 ) 3η02 η2 + η0 η12 cos 2λϕ
2 2 6 2
  
1 1 1
−9λ2 η3 cos 3λϕ + η3 cos 3λϕ = h0 (u0 ) η3 + h00 (u0 ) (2η0 η3 + η1 η2 ) + h000 (u0 ) 3η02 η3 + η13 + 3η0 η1 η2 cos 3λϕ
2 6 4
Now we keep only the lowest order terms from each equation, i.e., we have η1  η1 η0 and so on. We set
h0 (u0 ) = 1 − λ2 and solve
 
2 1 00 2 1 000 3 1 000 1 00
0 = −λ η0 + h (u0 ) η0 + h (u0 ) η0 + h (u0 ) η0 + h (u0 ) η12
2 6 4 4
2 1 00 2
0 = −λ η0 + h (u0 ) η1
4
1 00
η0 = h (u0 ) η12
4λ2
1 00 1
0 = h (u0 ) (2η0 η1 + η1 η2 ) + η13 h000 (u0 )
2  8 
1 00 1 00 1 00 1
= h (u0 ) 3
h (u0 ) η1 − h (u0 ) η1 + η13 h000 (u0 )
3
2 2λ2 12λ2 8
1  2

= η13 2
5 (h00 (u0 )) + 3λ2 h000 (u0 )
24λ
1 00
η2 = − h (u0 ) η12
12λ2 
1 1 00 1 3 000
η3 = − 2 h (u0 ) η1 η2 + η1 h (u0 )
8λ 2 24
or, summarizing,
1 00
η0 = h (u0 ) η12
4λ2
1 3 00 2 2 000

0 = η 1 5 (h (u 0 )) + 3λ h (u0 )
24λ2
1 00
η2 = − h (u0 ) η12
12λ2 
1 1 00 1 3 000
η3 = − 2 h (u0 ) η1 η2 + η1 h (u0 )
8λ 2 24

6
where η1 is constant. Now, using
µA 1−λ2
h= u
L2ϕ
we may find the derivatives,

µA −λ2
1 − λ2 = h0 (u0 ) = 1 − λ2 u
L2ϕ 0
µA −λ2
u = 1
L2ϕ 0

µA −λ2 −1
h00 (u0 ) = −λ2 1 − λ2 u
L2ϕ 0


λ2 1 − λ2
= −
u0



1 + λ2 λ2 1 − λ2
h000 (u0 ) =
u20

so we may solve the equality

1  00 2 2 000

0 = η13 5 (h (u0 )) + 3λ h (u 0 )
24λ2

2
2
!
1 5λ4 1 − λ2 2 1+λ
2
λ 1 − λ2
0 = + 3λ
λ2 u20 u20

2
0 = 5λ2 1 − λ2 + 3λ2 1 + λ2 1 − λ2

= 8λ2 − 10λ4 + 2λ6

2λ2 1 − λ2 4 − λ2



=
2
−3
and we see that the only candidate power laws, f (r) = Arλ , for stable, closed orbits are:

f (r) = Ar−3
f (r) = Ar−2
f (r) = Ar

The first of these yields only perfectly circular orbits, so the only nontrivial cases are the inverse square law
and Hookes’ law.
This condition only shows that these power laws are necessary. That they are sufficient to produce closed
orbits requires solving for their orbits exactly.