Bohr-Sommerfeld quantization

in Quantum and Statistical Mechanics.

Victor Gurarie
 U(x)

ta tb x
Figure 1: Potential U (x). The particles moves with a large energy E, with turning points
ta and tb . Classically, the particle bounces back and forth between ta and tb .

1 Bohr-Sommerfeld quantization
We concentrate on one dimensional space. (In more than one dimension, we simply need
to consider each dimension separately). Consider a quantum particle moving in one di-
mensions in a potential U (x). It satisfies the Schrödinger equation

h̄2 d2 ψ
− + U (x)ψ = Eψ. (1.1)
2m dx2
Suppose the energy of the particle is large, as shown on Fig. 1 (the precise meaning of
“large” will be established later).
Under these conditions, we are going to solve the Schrodinger equation approximately,
using the so-called WKB approximation. Let us show that the solution to (1.1) approxi-
mately is given by
iZx
  √ q
ψ(x) = exp dy p(y) , where p(y) = 2m E − U (y). (1.2)
h̄ ta
Notice that p(y) is exactly equal to the classical value of the momentum of the particle at
the point y, or p2 (y)/(2m) + U (x) = E. Indeed, we find
dψ(x) i
= p(x)ψ(x). (1.3)
dx h̄
Then we find
 √ 
d2 ψ(x) p2 (x)
" #
i dp(x) 2m i m dU (x)
= − 2 + ψ(x) = − 2 (E − U (x)) − √ q  ψ(x).
dx2 h̄ h̄ dx h̄ h̄ 2 E − U (x) dx
(1.4)

1
Now the point is, at large enough energy E, the second term in the square brackets above
is always much smaller that the first term and can be safely neglected. (The exception of
this is when x is close to the turning points ta or tb where U (x) is close to E. These points
are subtle and will be briefly discussed later; for now we assume that we can ignore this
problem).
Thus we can write
d2 ψ(x) 2m
2
≈ − 2 (E − U (x)) ψ(x). (1.5)
dx h̄
From this it immediately follows that (1.1) is indeed satisfied.
We also observe that the second solution of the Schrödinger equation (1.1) is
iZx
 
ψ2 (x) = exp − dy p(y) . (1.6)
h̄ ta
Now the particle cannot move to the left of the point ta and to the right of tb (well, we
know that quantum mechanically it can, but if E is large, the particle almost does not go
to these regions at all). Thus the wave function must vanish for x < ta . The appropriate
linear combination of the wave functions ψ and ψ2 which vanishes at x = ta is (take their
difference and divide by 2i)
1Zx
 
ψ(x) = sin dy p(y) . (1.7)
h̄ ta
Now this wave function must also vanish at x = tb . This leads to the condition
1 Z tb
 
ψ(tb ) = sin dy p(y) = 0, (1.8)
h̄ ta
or equivalently
1 Z tb 1 Z tb √ q
dy p(y) = dy 2m E − U (y) = πn, (1.9)
h̄ ta h̄ ta
where n is an integer.
Solving (1.9) for E at a given n gives us the “allowed” values of energy, or energy levels
En .
(1.9) is called the Bohr-Sommerfeld quantization condition. It was proposed by
Bohr (Bohr’s allowed orbits in the hydrogen atom) and improved by Sommerfeld in 1910s,
before quantum mechanics and Schrödinger equations were invented. As is clear from its
derivation, it can give energy levels En only if En is sufficiently large. Thus (1.9) typically
gives the ground state energy badly, but quickly improves as n increases.
In the derivation above, we mentioned that at the turning points ta and tb , the ex-
pression for the wave function (1.7) generally breaks down. A more refined treatment,

2
p

Figure 2: The trajectory of a classical particle moving in a potential U (x). At each point
√ q
x the magnitude of its momentum is p(x) = 2m E − U (x). On the top half of the
trajectory the particle moves to the right, and on the bottom half, it moves to the left.

which carefully studies what happens at these points, gives an improved Bohr-Sommerfled
quantization condition

1 Z tb 1 Z tb √ q 1
 
dy p(y) = dy 2m E − U (y) = π n + . (1.10)
h̄ ta h̄ ta 2

This treatment goes beyond this text however. It is given, for example, in Landau and
Lifshitz, Quantum Mechanics, page 170 (section 48).
Nevertheless we observe that the difference between (1.9) and (1.10) when n is large
(and where these equations become good approximations in the first place) is insignificant,
so for our purposes this is not important.
Finally, we observe that the Bohr-Sommerfeld quantization condition can be given the
following graphical interpretation.
Z tb Z tb √ q
dy p(y) = dy 2m E − U (y) (1.11)
ta ta

is but exactly half the area surrounded by the classical trajectory of the particle in phase
space, as shown on Fig. 2. Let us denote the entire area inside the trajectory corresponding
to the energy E (which is twice the integral in (1.9)) as AE . Then we find

AE = 2πh̄n (1.12)

is the Bohr-Sommerfeld condition for the energy of n-th level.

3
2 Bohr-Sommerfeld quantization in statistical physics.
In statistical physics we often need to calculate sums of the form
En
X  
f . (2.13)
n T
Here f is some function. It could be exponential if we are computing a partition function,
or it could be something else if we are computing, for example, a grand canonical potential
for a Bose-Einstein or a Fermi-Dirac gas. In some cases it can even be one, f = 1, if we
apply microcanonical ensemble and simply would like to know the total number of states
Γ(E) with energy below given energy E.
It turns out that under the condition En+1 − En  T , and under the condition that
most terms in (2.13) correspond to highly excited energy levels, (2.13) is equal to
p2
!
En dxdp U (x)
X   Z
f ≈ f + . (2.14)
n T 2πh̄ 2mT T
(2.14) is the foundation which allows us to apply Bohr-Sommerfeld quantization to statis-
tical physics.
Let us show that this is correct. Consider a sequence of trajectories in the phase space,
one for the energy E0 (the ground state), another for the energy E1 (the first excited
state), yet another for the energy E2 (second excited state) and so on. Bohr-Sommerfeld
quantization condition says that the area of the phase space enclosed between n-th and
n + 1-th trajectories is exactly equal to 2πh̄.
Let us denote the domain enclosed between the n-th and n + 1-th trajectory as An .
Then it is easy to see that
En En Z dxdp
   
f =f . (2.15)
T T An 2πh̄
R
Here An denotes the integral over the domain An . Indeed, this relation is almost tautology,
R dxdp
since An 2πh̄ = 1 by construction. However, using the fact that En and En+1 are very
close (En+1 − En  T ) and that En is nothing but the value of the classical energy
p2 /(2m) + U (x) on a trajectory corresponding to the energy En , we can rewrite this as
p2
!
En En Z dxdp Z dxdp U (x)
   
f =f ≈ f + . (2.16)
T T An 2πh̄ An 2πh̄ 2mT T
Finally,
p2 p2
! !
En dxdp U (x) dxdp U (x)
X   XZ Z
f ≈ f + = f + . (2.17)
n T n An 2πh̄ 2mT T 2πh̄ 2mT T

4
p

trajectory 1
trajectory 2
trajectory 3

Figure 3: The trajectories of a classical particle moving in a potential U (x) (shown ex-
traordinary schematically). Each larger trajectory corresponds to bigger energy. The area
of the phase space enclosed between n-th and n + 1-th trajectories is equal to 2πh̄.

Here the last integral is over the entire phase space (not just over some domain in the
phase space). This concludes the proof of the Bohr-Sommerfeld relation in statistical
physics (2.14). Thus in statistical physics when we work under the conditions of the
applicability of classical physics we can replace
X Z
dxdp
→ , (2.18)
n 2πh̄

simultaneously replacing En with p2 /(2m) + U (x).