Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

WKB

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

221A Lecture Notes

WKB Method

1 Hamilton–Jacobi Equation
We start from the Schrödinger equation for a single particle in a potential

h̄2 ~ 2
" #

ih̄ ψ(~x, t) = − ∇ + V (~x) ψ(~x, t). (1)
∂t 2m

We can rewrite this equation by using ψ(~x, t) = eiS(~x,t)/h̄ :


" #
∂S 1 ~ 2 ih̄ ~ 2
− ψ= (∇S) − (∇ S) + V ψ. (2)
∂t 2m 2m

Assuming ψ 6= 0, this leads to an equation


∂S 1 ~ 2 ih̄ ~ 2
− = (∇S) − (∇ S) + V. (3)
∂t 2m 2m
Now taking the formal limit h̄ → 0, it becomes the same as the classical
Hamilgon–Jacobi equation
∂S 1 ~ 2
− = (∇S) + V. (4)
∂t 2m
It is often said that the h̄ → 0 limit is indeed the classical mechanics because
it reduces to the Hamilton–Jacobi equation.

2 Classical Limit
The fact that the exponent S satisfies the Hamilton–Jacobi equation may
not mean the classical behavior to you right away. The way to see it is by
forming a wave packet, and ask the question how does the wave packet moves
as a whole. To simplify the discussion, let us consentrate on the case where
the Hamiltonian does not depend explicitly on time, and we can separate the
variable as
S(~x, t) = S̃(~x, E) − Et. (5)

1
The inverse Legendre transformation says
∂ S̃(~x, E)
t= . (6)
∂E
This condition can be viewed as an implicit equation to determine the posi-
tion of the particle ~x as a function of time t.
To form a wave packet, we need to put together waves of slighly different
energies. We can write
Z
ψ(~x, t) = dEg(E)ei(S̃(~x,E)−Et)/h̄ , (7)
The question here is where the wave packet is. At most places, the averaging
over E makes the phase oscillate very fast because of the small h̄, and there
is a strong cancellation. Only at special points where the phase factor is
stationary with respect to E, the wave function is sizable. Therefore, the
position of the wave packet is determined by the stationary condition
∂   ∂ S̃(~x, E)
S̃(~x, E) − Et = − t = 0. (8)
∂E ∂E
This is nothing but Eq. (6). In other words, the wave packet is sizable only
where the position satisfies the classical equation of motion. Therefore, the
wave packet follows the classical equation of motion.

3 h̄ Expansion
Now that we have rederived classical mechanics from quantum mechanics,
our interest now turns to quantum mechanics itself. We will use the fact
that h̄ is indeed small to find a useful approximation method in quantum
mechanics.
The Eq. (3) is exact as long as ψ 6= 0. One can use this equation, and
consider an expansion
S(~x, t) = S0 + h̄S1 + h̄2 S2 + · · · . (9)
This is an expansion in h̄, and hence called h̄-expansion or semi-classical
expansion. Plugging in the expansion into Eq. (3), we find
∂S0 1 ~
− = (∇S0 )2 + V, (10)
∂t 2m
∂S1 1 h ~2 ~ 0 )(∇S
i
~ 1) ,
− = −i∇ S0 + 2(∇S (11)
∂t 2m
2
and similarly for higher terms in h̄. The leading equation has only S0 , and
it is exactly the same as Hamilton–Jacobi equation. Once you solve these
equations starting from S0 , S1 , etc, you have solved the wave function ψ in
a systematic expansion in h̄.
There is one importance difference between the “classical limit” which
reduces to the Hamilton–Jacobi equation and the idea of expansion in h̄.
In the classical limit, S is of course real because it does not make sense
otherwise. However, in the sense of an expansion in h̄, we don’t know if S0
has to be real. In fact, in many interesting cases, S0 turns out to be complex .
We will see these examples below.

4 WKB Approximation
WKB Approximation, due to Wentzel, Kramers, and Brillouin, keeps terms
up to O(h̄) in the h̄ expansion. It is used mostly for the time-independent
case, or in other words, for an eigenstate of energy E. Then the wave function
has the ordinary time dependence e−iEt/h̄ . We also restrict ourselves to the
one-dimensional problem. In terms of S, it corresponds to

S(~x, t) = S(~x) − Et. (12)

Therefore only S0 has the time dependence S0 (x, t) = S0 (x) − Et, while
higher order terms Si = Si (x) for i 6= 0 do not depend on time.
The lowest order term S0 satisfies the Hamilton–Jacobi equation (see
Eq. (10)
1
E= (S 0 )2 + V (x). (13)
2m 0
The differential equation can be solve immediately as
Z x q Z x
S0 (x) = ± 2m(E − V (x0 ))dx0 = p(x0 )dx0 (14)

up to an integration constant which can be determined only after imposing


aqboundary condition on the wave function. We used the notation p(x) =
± 2m(E − V (x)) because it is nothing but the momentum of the particle
in the classical sense. Once we know S0 , we can also solve for S1 . Starting
from Eq. (11), and using ∂S1 /∂t = 0, we find

2S00 S10 = iS000 , (15)

3
which has a solution
Z x S000 (x0 ) 0 i
S1 (x) = i 0 0
dx = log p(x) + constant. (16)
2S0 (x ) 2
Therefore the general solution to the Schrödinger equation up to this order
is

ψ(x, t) = ei(S0 (x)+h̄S1 (x))/h̄ e−iEt/h̄


1 i Z xq
 
= c exp ± 2m(E − V (x ))dx e−iEt/h̄ ,
0 0
(17)
p(x)1/2 h̄
and the overall constant c is of course undetermined from this analysis.
This solution makes it immediately clear that this approximation breaks
down when p(x) goes to zero. Or in other words, the approximation is bad
where the classical particle stops and turns because of the potential. Such
points are called “classical turning points.”

4.1 Validity of the WKB Approximation


The approximation to stop with S1 in the h̄ expansion is valid only when S1
is much smaller than S0 . Or in other words, if the term with h̄ in Eq. (2) is
much smaller than the other terms. In particular, we require
~ 2 |  h̄|∇
|(∇S) ~ 2 S|. (18)

In the one-diimensional time-independent case discussed above, this is

p(x)2  h̄|p0 (x)|. (19)


q
Using the definition of p(x) = ± 2m(E − V (x)), we find

h̄dV (x)/dx
 1. (20)

2(E − V (x))p(x)

Again we find the same conclusion: the WKB approximation breaks down
close to the classical turning point V (x) = E (e.g., p(x) = 0).
For example, take a harmonic oscillator V (x) = 12 mω 2 x2 . The validity
condition Eq. (20) can be rewritten as
3
1

8 E − mω x  (h̄ω)2 mω 2 x2 .
2 2

(21)
2

4
This inequality is always satisfied exactly at the origin x = 0, but once away
from the origin, it is impossible to satisfy unless E  h̄ω. In this sense, we
are indeed in the classical regime. However, even for a large E  h̄ω, the
approximation is not valid close to the classical turning points E = 12 mω 2 x2 .
Here is the surprise. The validity condition Eq. (20) may be satisfied
even in the region where the particle cannot enter classicaly E < V (x). For
example with the harmonic q oscillator again, the validity condition is always
satisfied for large x  2E/mω 2 for any value of E. In other words, the
WKB approximation is good away from the classical turning points both
where a classical particle exists and where a classical particle cannot exist.
This is why the WKB approximation is not really a classical limit. It applies
also where physics is truly quantum mechanical.
In the classically forbidden region, the solution Eq. (17) needs to be mod-
ified to
ψ(x, t) = ei(S0 (x)+h̄S1 (x))/h̄ e−iEt/h̄
1 1 Z xq
 
=c exp ± 2m(V (x ) − E)dx e−iEt/h̄ ,(22)
0 0
(2m(V (x) − E))1/4 h̄
by following the same steps as in the classically allowed region.

4.2 Matching
WKB approximation can be good both in the region E > V (x) and the region
E < V (x) but cannot be good in between the regions close to the classical
turning point E = V (xc ). In order to utilize the WKB approximation to
work out wave functions, we need to somehow overcome this limitation. The
standard method is to expand around xc and solve for the wave function
“exactly.” Then you can match on to the WKB solutions away from xc to
determine the entire wave function.
The common method is to approximate the potential around the classical
turning point xc by a linear one:
V (x) = V (xc ) + V 0 (xc )(x − xc ) + O(x − xc )2 , (23)
and ignore the second order term. By definition V (xc ) = E. Schrödinger
equation around this point is therefore
h̄2 d2 h̄2 d2
! !
− + V (x) − E ψ = − + V 0 (xc )(x − xc ) ψ = 0. (24)
2m dx2 2m dx2

5
Using the new variable
!1/3
2m dV
u= (xc ) (x − xc ), (25)
h̄2 dx

the differential equation simplifes drastically to

d2
!
− u ψ = 0. (26)
du2

The solution to this equation is known as the Airy function


1Z∞ 1 3
 
Ai(u) = dt cos t + ut . (27)
π 0 3
This can be checked as follows. By acting the differential operator in Eq. (26)
on the definition of the Airy function, we find

d2
!
1Z∞ d 1 3
 
− u Ai(u) = − dt sin t + ut . (28)
du2 π 0 dt 3

The boundary term at t = 0 obviously vanishes. The behavior at t = ∞ is


trickier. The point is that the argument of the sin grows as t3 and oscillates
more and more rapidly as t → ∞. Therefore for any infinitesimal interval of
large t, the oscillation basically cancels the integrand except for a “left-over”
that goes down as ∼ 1/t2 . Therefore the boundary term for t → ∞ can also
be dropped.
It can be shown using the steepest descent method that the asymptotic
behavior of the Airy function smoothly matches to the WKB solutions. The
asymptotic behavior is
  1/2  
1 1
exp − 23 u3/2
√ u0


Ai(u) ∼  π u1/2
2
 √  . (29)
√1 cos 2
u −u + π
u  0


π −u 3 4

Note first that, for u  0, the asymptotic behavior is


 1/2
h̄(2mV 0 /h̄2 )1/3  1 Z xq π
 
Ai(u) =  q cos 2m(E − V )dx0 + . (30)
π 2m(E − V ) h̄ xc 4

6
Here weqused the linear expansion V (x) = E +V 0 (xc )(x−xc ) to relate powers
of u to 2m(E − V ). This expression is consistent with the WKB solution
for a particular choice of the overall constant. Similarly, for u  0,
 1/2
1 h̄(2mV 0 /h̄2 )1/3  1 Z xq
 
Ai(u) =  q exp − 2m(V (x0 ) − E)dx0 . (31)
2 π 2m(V − E) h̄ xc

4.3 Bound States


Consider the following situation. In the region I (x < a), E < V and it
is classically forbidden. In the region II (a < x < b), E > V and the
the particle is classically allowed. But again in the region III (x > b) it
is classically forbidden. In this case, we expect bound states with discrete
energy levels.
From the matching using the Airy function at x = b, we want the wave
function in the classicaly allowed region II to be as in Eq. (30),
 1/2
h̄(2mV 0 (b)/h̄2 )1/3  1 Z xq π
 
ψ(x) = cb  q cos 2m(E − V (x0 ))dx0 + .
π 2m(E − V ) h̄ b 4
(32)
On the other hand from the matching at x = a, the wave function in the
region II must be
 1/2
h̄(2mV 0 (a)/h̄2 )1/3  1 Z xq π
 
ψ(x) = ca  q cos 2m(E − V (x0 ))dx0 − .
π 2m(E − V ) h̄ a 4
(33)
cb and ca are normalization constants. The minus sign in fron of π/4 is
because the classically allowed region is to the right of the turning point and
hence we need to use Ai(-u) instead of Ai(u) for matching. The two behaviors
obtained from matching at both sides must be the same in order to have a
consistent wave function for the entire region II. Because we do not know the
coefficients ca,b , which can in particular differ in signs, we have to require
1 Z xq π 1 Z xq π
2m(E − V (x0 ))dx0 + = 2m(E − V (x0 ))dx0 − − nπ. (34)
h̄ b 4 h̄ a 4
Further simplifying the equation we obtain
Z b q 1
 
2m(E − V (x))dx = n + πh̄. (35)
a 2

7
This is reminiscent of Bohr–Sommerfeld quantization condition that ap-
peared at the early stage of quantum mechanics,
I
pdq = nh (36)

with h = 2πh̄ except that there is an extra contribution 1/2 here. Remember
that the Bohr–Sommerfeld quantization condition was an ad hoc requirement
which happened to give correct energy spectra for the hydrogen-like atoms.
Nonetheless, this is an important equation because it puts constraints on
the allowed values of E, and hence we find discrete energy levels. Because
the WKB approximation is better for larger E, the constraint Eq. (35) is
expected to give correct energy levels at least for highly excited states.
Let us apply this constraint Eq. (35) for a harmonic oscillator. The
condition then is
Z xmax √
1
 
2 2 2
2mE − m ω x dx = n + πh̄. (37)
xmin 2
2mE π
The l.h.s. is an elementary integral mω 2
and we find
1
 
E = n+ h̄ω. (38)
2
This is precisely the energy levels of a harmonic oscillator. Of course, ob-
taining the exact energy levels is an accident for a harmonic oscillator. In
general, we expect to get approximately correct energy levels for high n. In-
deed, if we apply the same constraint for the hydrogen-like atoms (for l = 0
and applying the WKB method to the radial wave function), we would ob-
tain the correct results if we drop 1/2 (this is what Bohr–Sommerfeld) did,
but don’t if we keep 1/2 as obtained by the WKB method.
What we obtain from the WKB method is not just the energy levels,
but more details such as the approximate wave function. To simplify the
notation, let us choose a unit system such that m = ω = h̄ = 1. Then the
wave function in the region II for the harmonic oscillator is
1
Z x √ 0 π

ψ∝ cos 02
2E − x dx − . (39)

(2E − x2 )1/4 − 2E 4
Doing the standard integral, we find
  s  
1 E sin−1 √
x π x x2  π 
ψ∝ cos + + √ 1 − − . (40)
(2E − x2 )1/4 2E 2 2E 2E 4

8
Because E = n + 1/2 in this unit, it further simplifies to
   q 
x2
 cos E sin−1 √x + √x 1− n even

1

2E 2E 2E
ψ∝   q  (41)
(2E − x2 )1/4  sin E sin−1

 √x + √x 1− x2
n odd
2E 2E 2E

4.4 Tunneling
To study a tunneling process, we again have three regions, a classically al-
lowed region I x < a, where the particle initially exist, a classically forbidden
region II a < x < b, and a classically allowed region III x > b where the
particle tunnels to. We follow the same matching procedure as in the bound
state example. A difference is that the matching at x = b uses a different
version of the Airy function which rises exponentially away from x = b. This
version of the Airy function Bi is explained in the note on steepest descent
method. The bottom line is that the matching at x = a is done as
 1/2
h̄(2mV 0 (a)/h̄2 )1/3  −1 Z a q π
 
ψ '  q cos 2m(E − V )dx0 + , (x < a)
π 2m(E − V ) h̄ x 4
 1/2
1  h̄(2mV 0 (a)/h̄2 )1/3  −1 Z x q
 
ψ ' q exp 2m(V − E)dx0 . (x > a)
2 π 2m(V − E) h̄ a

The matching at x = b is, on the other hand,


 1/2
i  h̄(2mV 0 (b)/h̄2 )1/3 
!
+1 Z b q
ψ ' C q exp 2m(V − E)dx0 , (x < b)
2 π 2m(V − E) h̄ x
 1/2
i h̄(2mV 0 (b)/h̄2 )1/3  1 Z xq π
 
ψ ' C  q sin 2m(E − V )dx0 + . (x > b)
2 π 2m(E − V ) h̄ b 4

The overall normalization factor C is determined by the requirement that


the behavior of the wave function is consistent for a < x < b between two
matching procedures. We therefore find
!1/6
V 0 (a)
!
1 Z bq
C = −i exp − 2m(V (x) − E)dx . (42)
V 0 (b) h̄ a

9
Comparing two classically allowed regions, and taking advantage of a further
normalization change, the matching reduces to
1 −1 Z a q π
 
0
ψ ' cos 2m(E − V )dx + , (x < a)
(2m(E − V ))1/4 h̄ x 4
!
1 Z bq
 Z xq
1 1 1 π

0
ψ ' exp − 2m(V (x) − E)dx sin 2m(E − V )dx + . (
2 h̄ a (2m(E − V ))1/4 h̄ b 4

In other words, the amplitude in the region x > b due to tunneling from the
region x < a is suppressed by
!
1 1 Z bq
exp − 2m(V (x) − E)dx . (43)
2 h̄ a

Usually people refer to Gamov’s transmission coefficient


!
1 Z bq
exp −2 2m(V (x) − E)dx (44)
h̄ a

as a suppression factor for the tunneling rate (square of the amplitude).

10

You might also like