The WKB Approximation

Marcus A. M. de Aguiar
August 30, 2017

1
Contents
1 Introduction 3

2 General form of the WKB wave-function 7

3 The WKB wave-function for bound states 9

4 Solution near a turning point with V 0 > 0 10

4.1 Statement of the problem and solution . . . . . . . . . . . . . 10
4.2 Solving the equation . . . . . . . . . . . . . . . . . . . . . . . 11

5 Solution near a turning point with V 0 < 0 15

6 Semiclassical quantization 17

7 The WKB wave-function for scattering states 20

8 The other Airy function 21

8.1 The Airy equation . . . . . . . . . . . . . . . . . . . . . . . . 21
8.2 The function Bi . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.3 Asymptotic form of Bi . . . . . . . . . . . . . . . . . . . . . . 22

9 Semiclassical tunneling 24

2
1 Introduction
The Schroedinger equation for a particle in a potential can be solved explic-
itly only if the potential has a sufficiently simple form, otherwise numerical
methods have to be used. In many cases it is preferable to have approx-
imate explicit formulas, that can be further used to calculate other quan-
tities, than to have exact numerical solutions. The WKB approximation,
developed by Wentzel, Kramers and Brillouin, is a semiclassical method to
solve Schroedinger’s equation that does not require the potential to be a
perturbation of a solvable problem. Instead, it only assumes that certain
classical quantities having the dimension of action (energy×time) are much
larger than the Planck’s constant. In this paper I will present the method in
a simple but complete way, paying particular attention to the famous con-
nection formulas, that are the main source of difficulty in the presentation.
To make things clear from the beginning, let me be very specific about
the problems I have in mind. I want to find stationary solutions for a particle
moving in a one-dimensional potential. I will consider two situations: (a) the
potential has bound states, as illustrated in figure 1, or; (b) the potential has
scattering states and is different from zero only in a small region around the
origin, as illustrated in figure 2. In the first case I am particularly interested in
computing the discrete energy levels of the system, whereas in the second case
I want to compute the tunneling coefficient. In both cases the Schroedinger
equation reads
~2 ∂ 2 ψ ∂ψ
− 2
+ V (x)ψ(x, t) = i~ . (1)
2m ∂x ∂t
The wave-function also satisfies the continuity equation
∂ρ ∂j
+ =0 (2)
∂t ∂x
where
ρ(x, t) = |ψ(x, t)|2 (3)
is the probability density and
 
~ ∗ ∂ψ
j(x, t) = Im ψ (4)
m ∂x
is the probability current. The continuity equation can be derived by multi-
plying eq.(1) by ψ ∗ and subtracting the resulting equation from its complex
conjugate.

3
 V(q)

I II III

A B q

Figure 1: A particle with energy E in a bound potential. The turning points

A and B, where V = E, separate the classically allowed region II from the
forbidden regions I and III.

V(q)

E
I II III

A B q

Figure 2: A particle with energy E in a scattering potential. In our calcu-

lations I will assume that the particle is incident from the left side of the
barrier and that E < V0 .

4
 In developing the WKB approximation it is important to write the com-
plex wave-function ψ in terms of its modulus and phase as
p i
ψ(x, t) = ρ(x, t)e ~ S(x,t) (5)
where the explicit dependence on ~ in the phase is introduced by convenience.
In terms of ρ and S the probability current becomes
ρ ∂S
j= . (6)
m ∂x
The Schroedinger equation, on the other hand, reads
"  2  2 #
~2 1 ∂ 2ρ 1 ∂ρ i ∂ρ ∂S i √ ∂ 2S 1√ ∂S
− √ − √ + √ + ρ − ρ
2m 2 ρ ∂x2 4ρ ρ ∂x ~ ρ ∂x ∂x ~ ∂x2 ~2 ∂x
 
√ 1 ∂ρ i √ ∂S
+V (x) ρ = i~ √ + ρ (7)
2 ρ ∂t ~ ∂t
where I have already canceled the global factor exp (iS/~). In terms of ex-
plicit powers of ~ this equation has three types of factors, involving ~0 , ~1
and ~2 . I am going to show that the factors in ~1 vanish exactly because of
the continuity equation. Indeed, the two terms of this type on the left side
of the equation are
i ∂ρ ∂S i √ ∂ 2S
− √ − ρ 2 (8)
2m ρ ∂x ∂x 2m ∂x
which can be re-organized as
 
i ∂ ρ ∂S i ∂j i ∂ρ
− √ =− √ = √ (9)
2 ρ ∂x m ∂x 2 ρ ∂x 2 ρ ∂t
and exactly cancel the term in ~ on the right side.
√
Dividing the remaining terms in eq.(7) by ρ I obtain
 2
1 ∂S ∂S
+ VQ (x) = − (10)
2m ∂x ∂t
where I have defined the quantum potential
2
~2 ∂ 2 ρ ~2

∂ρ
VQ (x) = V (x) − +
4mρ ∂x2 8mρ2 ∂x
(11)
2
 
~ ∂ 1 ∂ρ
= V (x) − √ √ .
4m ρ ∂x ρ ∂x

5
 Equation (10) is exact and, together with the continuity equation, it
is completely equivalent to Schroedinger’s equation. If it were not for the
terms in ~2 in the quantum potential, eq.(10) would be the Hamilton-Jacobi
equation of classical mechanics, whose solution is given by the action func-
tion (see below). The WKB approximation consists exactly in discard these
terms. David Bohm proposed that we could solve these equations as they
are, and find the trajectories of the particle in the quantum potential. No-
tice, however, that VQ depends on the wave-function itself. Given ψ(x, t0 ) we
would calculate VQ and propagate the wave-function to t0 + dt. This would
allows us to re-calculate VQ and proceed with the integration. The difficulty
is that VQ is singular at the nodes of the wave-function, where ψ(x, t) = 0.
In either case equation (10) is a partial differential equation that can be
solved for S(x, t) given S(x, 0). However, our interest here is to calculate
stationary states, whose time dependence is of the form
iEt
ψ(x, t) = φ(x)e− ~ . (12)

Comparing with equation (5) we see that this amounts to impose

ρ = ρ(x) S(x, t) = W (x) − Et (13)

so that iW (x)
p iEt
ψ(x, t) = ρ(x)e ~ e− ~ .
In terms of W and E equation (10) becomes
 2
1 ∂W
+ VQ (x) = E. (14)
2m ∂x

6
2 General form of the WKB wave-function
In the limit where certain classical quantities (to be more clearly specified
below) become much larger than ~, it is legitimate to discard the quantum
corrections to the potential VQ and replace it by the classical potential. In
this case eq.(10) reduces to the classical Hamilton-Jacobi equation
 2
1 ∂S ∂S
+ V (x) = − . (15)
2m ∂x ∂t
For stationary states, where

ρ = ρ(x) and S(x, t) = W (x) − Et

it reduces to  2
1 ∂W
+ V (x) = E. (16)
2m ∂x
The function S(x, t) is called Hamilton’s principal function, whereas W (x)
is Hamilton’s characteristic function. Equation (16) can be solved for W as
Z xp
W (x) = ± 2m(E − V (x0 ))dx0 (17)

where the integrand is readily recognized as the momentum p(x, E) of a

particle with energy E at position x, since
p2 p
+ V (x) = E → p(x, E) = 2m(E − V (x)). (18)
2m
Comparing the classical Hamilton-Jacobi equation (16), which is inde-
pendent of ~, with its fully quantum version (14), we can see that quantum
corrections to W are proportional to ~2 . The same is true for ’classical’ ρ,
that we compute below using the continuity equation.
Since ρ is time independent, the current must be constant, since ∂j/∂x =
0 in this case. Using eqs.(6), (13) and (17) we see that
ρ
j= p(x, E) = const.
m
and, therefore,
const
ρ(x) = . (19)
p(x, E)

7
 Putting all this information together we obtain the general form of the
time independent, stationary, wave function φ(x) as
c i x
R 0 0
φ(x) = p e± ~ p(x ,E)dx . (20)
p(x, E)

In the next sections I will now apply these ideas to the specific prob-
lems illustrated in figures 1 and 2. In both cases, for a given total energy
E the x-axis is divided into three regions by the turning points A and B.
In the classically forbidden regions V (x) > E and p(x, E) becomes purely
imaginary. In these regions it is convenient to define the real function
p
q(x, E) = 2m(V (x) − E) = |p(x, E)| (21)

and re-write the general form of the wave-function as

c 1 x
R 0 0
φ(x) = p e± ~ q(x ,E)dx . (22)
q(x, E)

The lower limit of integration in equations (20) and (22) will be fixed later.

8
3 The WKB wave-function for bound states
I will now consider the specific problem of bound states illustrated in figure
1. In the classically forbidden regions I and III the WKB wave-function
involves real exponentials and we need to be careful not to let it diverge.
In the classically allowed region II, on the other hand, both solutions (with
plus or minus sign in the exponent) are well behaved and should be included.
Taking this considerations into account we write

D1 1 R x q(x0 ,E)dx0

 √ e~ x∈I
2 q







  Z x 
 C 1 0 0
φ(x) = √ cos p(x , E)dx + ξ x ∈ II (23)

 p ~





 D2 − 1 R x q(x0 ,E)dx0
 √ e ~ x ∈ III


2 q

where I replaced the combination of the two exponentials in region II by a

cosine with a phase.
We now get to the hard part of the WKB procedure, which is to match
the constants D1 , C, D2 and ξ. These constants cannot be arbitrary, since
φ(x) needs to be an approximation for the true wave-function in all space
and all these constants must be related to the single undetermined constant
that we could find later by normalization.
The problem is that we cannot simply match the different parts of the
wave-function φ(x) because it diverges right at the matching points x = A
and x = B! At these points p(x, E) = q(x, E) = 0. The trick to overcome
this difficulty is to go back to the original Schroedinger equation and to
solve it in the neighborhood of these points by approximating the potential
by a straight line. The next two sections are dedicated to that. Once we
have these solutions we can compare them with our WKB forms and relate
the coefficients to each other. We will find that the matching cannot be
always performed and the condition for it to work is the Bohr-Sommerfeld
quantization rule that discretizes the energy levels.

9
4 Solution near a turning point with V 0 > 0
4.1 Statement of the problem and solution
Consider a turning point x0 with V 0 (x0 ) > 0, where the prime means differ-
entiation with respect to x. This is the case of x = B for bound states (see
fig.1) or x = A for scattering states (see fig.2). In the vicinity of x0 we can
expand the potential as

V (x) ≈ V (x0 ) + V 0 (x0 )(x − x0 ) = E + V 0 (x0 )(x − x0 ). (24)

The time-independent Schroedinger equation becomes, in operator form,

p2
|φi + [E + V 0 (x0 )(x − x0 )]|φi = E|φi (25)
2m
or
p2
|φi + V 0 (x0 )x|φi = V 0 (x0 )x0 |φi. (26)
2m
I am going to solve this equation step by step in this section. The final
result will be written as
Z +∞
φ(x) = c cos (t3 /3 − ty)dt (27)
0

where c is a normalization constant and

1/3
2mV 0 (x0 )

y= (x0 − x). (28)
~2

This is exact, but not very helpful. The variable y is zero at the turning point
x = x0 . If we assume that the ratio ~2 /mV 0 (x0 ) is so small that y becomes
very large as we move away from x0 , without leaving the region where the
linear approximation of the potential is valid, we can derive approximate
formulas for this integral that directly relate to our WKB wave-functions. In
this case I will show that
  R 
√ c 1 x 0 0
 cos ~ x0
p(x , E)dx + π/4 x < x0
 p(x,E)

φ(x) ≈ (29)
1 x 0 ,E)dx0
R

 √ c − q(x
e ~ x0 x > x0 .

2 q(x,E)

10
4.2 Solving the equation
In the momentum representation hp|φi = φ(p) equation (26) reads

p2 ∂φ(p)
φ(p) + i~V 0 (x0 ) = V 0 (x0 )x0 φ(p). (30)
2m ∂p
This equation has a simple solution

φ(p) = A exp (iαp3 − ipx0 /~) (31)

with
1
α= (32)
6m~V 0 (x0 )
as the reader can check by substitution. The corresponding wave-function in
position representation is obtained by Fourier transforming:
Z +∞
A 3
φ(x) = √ e(iαp +ip(x−x0 )/~) dp
2π~ −∞
which can be easily re-arranged as
Z +∞
0
φ(x) = A cos (αp3 + p(x − x0 )/~)dp.
0

Finally, changing variables to t = p(3α)1/3 and defining

1/3
2mV 0 (x0 )

y= (x0 − x) (33)
~2

we obtain Z +∞
φ(x) = c cos (t3 /3 − ty)dt ≡ cAi(−y) (34)
0

where Ai(z) is the Airy function.

The solution (34) should be compared with the WKB solution derived in
the previous section. However, as it is, the comparison is not immediate. In
order to see that equation (34) indeed resembles the WKB forms we need to
do a little more work.
I am going to assume that the ratio ~2 /mV 0 (x0 ) is so small that, for
values of x where the linear approximation is still good, y can become much

11
larger than 1. In this case the cosine in the Airy function oscillates very fast
between +1 and −1 as t changes, and the integral seems to tend to zero.
However, there are points along the t axis where the phase of the cosine
becomes nearly constant, and in the vicinity of these points the cosine stops
oscillating and contributes some finite value to the integral. All we need to
do is to find these points and expand the cosine around them.
In order to do that I will re-write the Airy function as an integral from
−∞ to +∞ as
1 +∞ iγ(t,y)
Z
Ai(−y) = e dt (35)
2 −∞
where
γ(t, y) = t3 /3 − ty. (36)
We need to find the points where ∂γ/∂t = 0. This condition defines the
points where γ becomes stationary. We find
√
t± = ± y. (37)
Since the important contributions to the integral come from the neighbor-
hood of these points, we may expand γ to second order around them:
00
γ(t, y) ≈ γ(t± , y) + 21 γ (t± , y)(t − t± )2
(38)
√
= ∓ 23 y 3/2 ± 2
y(t − t± ) .
(the primes now denote differentiation with respect to t and remember that
γ 0 (t± , y) = 0). This approximation takes us to Gaussian integrals around
each stationary point that we know how to do:
1 ∓i 2 y3/2 +∞ ±i√y(t−t± )2
Z
Ai(−y) ≈ e 3 e dt (39)
2 −∞

However we must be careful: y > 0 only if x < x0 . This corresponds to

the classically allowed region (both in the case of bound states and scattering
√
states). In this case y is real, there is no risk of divergence in the exponents
and both stationary points contribute to the integral and we must sum their
contributions. We get
q 2 3/2
q 2 3/2
Ai(−y) ≈ 12 √iπy e−i 3 y + 21 −iπ √ ei 3 y
y

q (40)
√π 2 3/2


= y
cos 3
y − π/4 .

12
 √ √
For x > x0 we find that y < 0 and y = i −y. In this case only the
stationary point t+ contributes, since t− would lead to a diverging integral.
We get
2 3/2 R +∞ −√−y(t−t )2
Ai(−y) ≈ 12 e− 3 (−y) −∞
e +
dt
q (41)
2 3/2
1 π − (−y)
= 2 √−y e 3 .

The asymptotic analysis presented above is not rigorous, since we have not
shown that the contour of integration can be deformed so as to pass through
the stationary points. In any case, it gives an idea of the general procedure.
To complete the analysis we only need to relate y with p(x, E). We start
from
p p p
p(x, E) = 2m(E − V (x)) ≈ −2mV 0 (x0 )(x − x0 ) = 2mV 0 (x0 )(x0 − x)

and
Z x Z x
0 0
p p 2p
p(x , E)dx ≈ 2mV 0 (x
0) (x0 − x0 )dx0 = − 2mV 0 (x0 )(x0 −x)3/2 .
x0 x0 3

Therefore,
Z x
2 3/2 2p 1
y = 2mV 0 (x0 )(x0 − x)3/2 = − p(x0 , E)dx0
3 3~ ~ x0

and also
p(x, E)
y 1/2 = .
[2m~V 0 (x0 )]1/3
Therefore, equation (40) can be re-writen as
 Z x 
c 1 0 0
Ai(−y) ≈ p cos p(x , E)dx + π/4 . (42)
p(x, E) ~ x0
p
for x < x0 , where c = π[2mV 0 (x0 )~]1/3 and we used that cos (−θ) = cos (θ).
Similarly, for x > x0 ,
p
q(x, E) ≈ 2mV 0 (x0 )(x − x0 ),
Z x
2p
q(x0 , E)dx0 ≈ 2mV 0 (x0 )(x − x0 )3/2 ,
x0 3

13
 Z x
2 1
(−y)3/2 ≈ q(x0 , E)dx0
3 ~ x0

and
q(x, E)
(−y)1/2 = .
[2m~V 0 (x0 )]1/3
This gives
c − 1 x q(x0 ,E)dx0
R
Ai(−y) ≈ p e ~ x0 . (43)
q(x, E)

These results are summarized in equations (27) and (29) at the beginning
of this section.

14
5 Solution near a turning point with V 0 < 0
This is very similar to what we did in the previous section, but with some
important sign changes. The time-independent Schroedinger’s equation is
the same
p2
|φi + V 0 (x0 )x|φi = V 0 (x0 )x0 |φi (44)
2m
and the solution in the p representation is now written as

φ(p) = A exp (−iαp3 − ipx0 /~) (45)

with
1 1
α=− = . (46)
6m~V 0 (x0 ) 6m~|V 0 (x0 )|
Following the same steps of the last section we can write φ(x) as
Z +∞
φ(x) = c cos (t3 /3 − ty)dt ≡ cAi(−y) (47)
0

where we defined t = p(3α)1/3 and

1/3
2m|V 0 (x0 )|

y= (x − x0 ). (48)
~2
The asymptotic form of this expression for large y is identical to the ones we
derived in section 4: for x > x0 , in the classically allowed region, we get
 
π 2 3/2
r
Ai(−y) ≈ √ cos y − π/4 . (49)
y 3

For x < x0 , in the classically forbidden region, we find

r
1 π 2 3/2
Ai(−y) ≈ √ e− 3 (−y) . (50)
2 −y

Finally we need to relate y with p(x, E). We find

p
p(x, E) ≈ 2m|V 0 (x0 )|(x − x0 ),
Z x
2p
p(x0 , E)dx0 ≈ 2m|V 0 (x0 )|(x − x0 )3/2 ,
x0 3

15
 Z x
2 3/2 1
y ≈ p(x0 , E)dx0
3 ~ x0

and
p(x, E)
y 1/2 = .
[2m~|V 0 (x0 )|]1/3
Therefore, the asymptotic solutions can be re-writen as
  R 
√ c 1 x 0 0
 cos ~ x0 p(x , E)dx − π/4 x > x0
 p(x,E)

φ(x) ≈ (51)
1 x
q(x0 ,E)dx0
 R
 √c

e ~ x0 x < x0 .
2 q(x,E)

16
6 Semiclassical quantization
We are now in a position to match the different parts of the wave-function
for bound states, as given by equation (23) that we repeat here:

D1 1 R x q(x0 ,E)dx0

 √ e~ x∈I
2 q







  Z x 
 C 1 0 0
φ(x) = √ cos p(x , E)dx + ξ x ∈ II

 p ~




 D2 − 1 R x q(x0 ,E)dx0

 √ e ~ x ∈ III


2 q

We start by connecting regions II and III, where the turning point B satisfies
V 0 (B) > 0. Comparing with the approximate solutions near the turning
point, equation (29), also repeated below,
  R 
√ c 1 x 0 0

 p(x,E) cos ~ x0
p(x , E)dx + π/4 x < x0

φ(x) ≈
− 1 x q(x0 ,E)dx0
 R
 √c

e ~ x0 x > x0 .
2 q(x,E)

we find that D2 = C, the lower limit of the integrals should be fixed as B

and ξ = π/4.
On the other hand, regions I and II, connected by the turning point A
where V 0 (A) < 0, must match equations (51):
  R 
√ c 1 x 0 0
 cos ~ x0
p(x , E)dx − π/4 x > x0
 p(x,E)

φ(x) ≈
1 x
q(x0 ,E)dx0
 R
 √c

e ~ x0 x < x0 .
2 q(x,E)

In this case we find that D1 = C, the lower limit of the integrals should be
fixed as A and ξ = −π/4.

17
 Putting all this information together we may write
C 0 1 R x q(x0 ,E)dx0


 √ e~ A x∈I
2 q






  Z x  Z x
 C0

1 π

C 1 π

0 0 0 0
φ(x) = √ cos p(x , E)dx − = √ cos p(x , E)dx + x ∈ II

 p ~ A 4 p ~ B 4




 C 1 x
R 0 0
 √ e− ~ B q(x ,E)dx

x ∈ III


2 q

The two forms in region II are written in order to connect with regions I
and III respectively. But of course they must be representations of the same
function and, therefore, identical. In order to assure their identity, we re-
write the argument of the first cosine in terms of the argument of the second
cosine:
1 x 1 B π 1 x
Z Z Z
0 0 π 0 0 π
p(x , E)dx − = p(x , E)dx − + p(x0 , E)dx0 + .
~ A 4 ~ A 2 ~ B 4
We now impose that
Z B
1 π
p(x0 , E)dx0 − = nπ (52)
~ A 2
so that the arguments of the two cosine become related by
 Z x   Z x 
1 0 0 π n 1 0 0 π
cos p(x , E)dx − = (−1) cos p(x , E)dx + .
~ A 4 ~ B 4

This last sign, (−1)n is finally absorbed by imposing C 0 = (−1)n C.

The final result is
(−1)n 1 R x q(x0 ,E)dx0


 √ e~ A x∈I
2 q







  Z x 
 1 1 π
0 0
φ(x) = C √ cos p(x , E)dx + x ∈ II (53)

 p ~ B 4




 1 − 1 R x q(x0 ,E)dx0

 √ e ~ B x ∈ III


2 q

18
where C is related to the normalization of φ(x). This global matching is
possible only if the quantization condition (52) is satisfied. This can be
re-writen as Z B  
1
p(x, E)dx = n + π~ (54)
A 2
which is the well known Bohr-Sommerfeld quantization condition of the old
quantum theory.

19
7 The WKB wave-function for scattering states
Let us now consider the problem of scattering states illustrated in figure 1,
where a particle (or beam of particles) is incident from the left with E smaller
than the height of the barrier. The general form of the WKB wave-function
for this setting is

A i R x p(x0 ,E)dx0 R − i R x p(x0 ,E)dx0

 √ e ~ + √ e ~ x∈I
p p








 C 1 Rx 0 0 D 1 Rx 0 0
φ(x) = √ e ~ q(x ,E)dx + √ e− ~ q(x ,E)dx x ∈ II . (55)

 q q





 T i R x p(x0 ,E)dx0
 √ e~

 x ∈ III
p

Notice that the forbidden region is now region II, which is finite. There-
fore the exponentials with both + and − signs are allowed. As we shall see,
this introduces a complication in the matching of the different parts of the
wave-function. We will deal with that in the next section.
Once the matching of the coefficients has been completed, we will calcu-
late the transmission and reflection coefficients, given by

|T |2
t=
|A|2

and
|R|2
r= .
|A|2

20
8 The other Airy function
8.1 The Airy equation
Let us return to the Schroedinger equation in the vicinity of a turning point
x0 , but this time in the x representation. Starting from

p2
|φi + V 0 (x0 )x|φi = V 0 (x0 )x0 |φi. (56)
2m
we find
~2 ∂ 2 φ
− + V 0 (x0 )(x − x0 )φ(x) = 0. (57)
2m ∂x2
In terms of 1/3
2mV 0 (x0 )

y= (x0 − x) (58)
~2
we obtain
∂ 2φ
+ yφ = 0 (59)
∂y 2
which is the Airy equation. Let us check that
Z +∞
φ(x) = cos (t3 /3 − ty)dt = Ai(−y) (60)
0

is indeed a solution. We differentiate twice with respect to y and manipulate

the result as follows:
00 R +∞
φ = − 0 t2 cos (t3 /3 − ty)dt
R +∞  ∂
=− 0 ∂t
[sin (t3 /3 − ty)] − y cos (t3 /3 − ty) dt (61)

= sin (t3 /3 − ty)|∞

0 − yφ = −yφ.

The boundary term is a bit delicate: at t = 0 it is clearly zero, but at t = ∞

it is not well defined. The argument is that it oscillates so fast for large t
that any average of this term will be zero, and therefore we set it to zero.

21
8.2 The function Bi
The Airy equation is a second order equation and must have two independent
solutions. Looking at Ai(−y) we may guess that the other solution has the
same form with the cosine replaced by a sine. That, however, does not work.
The other solution is given by
Z +∞ h i
−t3 /3−yt 3
φ(x) = e + sin (t /3 − ty) dt ≡ Bi(−y). (62)
0

The need of the extra exponential becomes evident when we calculate the
second derivative of Bi with respect to y:
00 R +∞ h 2 −t3 /3−yt 2 3
i
φ = 0 te − t sin (t /3 − ty) dt

R +∞ n ∂ −t3 /3−yt ∂ −t3 /3−yt

o
= 0 − ∂t [e ]+ ∂t
[cos (t3 /3 − ty)] − ye 3
− y sin (t /3 − ty) dt

3 /3−yt
= −e−t |∞ 3 ∞
0 + cos (t /3 − ty)|0 − yφ = −yφ.
(63)
If it were not for the extra exponential term, the boundary term resulting
from the cosine at t = 0 would leave a factor 1, which is canceled by a similar
factor coming from that exponential.

8.3 Asymptotic form of Bi

Far from the turning point x0 we can derive simplified expressions for the
Bi functions just as we did with Ai. I will not do that here and will only
present the results. We find that, for V 0 (x0 ) > 0,
  R 
√ c 1 x 0 0
 cos ~ x0
p(x , E)dx − π/4 x < x0
 p(x,E)

φ(x) ≈ (64)
1 x 0 ,E)dx0
R

 √ c q(x
e ~ x0 x > x0

q(x,E)

and, for V 0 (x0 ) < 0,

  R 
√ c 1 x 0 0
 cos ~ x0
p(x , E)dx + π/4 x > x0
 p(x,E)

φ(x) ≈ (65)
1 x 0 0
R
 √ c e− ~ x0 q(x ,E)dx

x < x0 .

q(x,E)

22
Notice that both forms diverge as x → ±∞ in the forbidden regions. This is
why the function Bi does not appear in the calculation of bound states, where
the forbidden regions extend to infinity. However, in scattering problems, the
forbidden region is finite, no divergence occurs and Bi has to be included.

23
9 Semiclassical tunneling
Using the results from sections 5 and 8 we can match the different parts of
the wave-function for scattering states, as given by equation (55) that we
repeat here:

A i R x p(x0 ,E)dx0 R − i R x p(x0 ,E)dx0

 √ e ~ + √ e ~ x∈I
p p








 C 1 Rx 0 0 D 1 Rx 0 0
φ(x) = √ e ~ q(x ,E)dx + √ e− ~ q(x ,E)dx x ∈ II

 q q




 T i R x p(x0 ,E)dx0

 √ e~

 x ∈ III
p

We start with region III, where the solution represents a particle moving
to the right only. Near the turning point B, where V 0 (B) < 0 we can write
the solution as a combination of Ai and Bi as

φ(x) ≈ F Ai(−y) + GBi(−y).

In the allowed region III these functions have asymptotic forms given by
equations (51) and (65):
Rx Rx
φ(x) ≈ √ F cos ~1 B p(x0 , E)dx0 − π/4 + √ G cos ~1 B p(x0 , E)dx0 + π/4



p(x,E) p(x,E)

Rx Rx
=√ 1 1
p(x0 , E)dx0 − π/4 + G sin 1
p(x0 , E)dx0 − π/4



F cos ~ B ~ B
p(x,E)

The desired form for region III can be recovered with the choice G = iF . We
get  Z x 
F i 0 0
φ(x) ≈ p exp p(x , E)dx − π/4
p(x, E) ~ x0
which gives our first coefficient T = F e−iπ/4 .
This same solution, with G = iF , must work in region II, where

φ(x) ≈ F [Ai(−y) + iBi(−y)]

 
1
Rx 0 0 1
Rx 0 0
≈F √1 e~ B q(x ,E)dx + √ i e− ~ B q(x ,E)dx .
2 q(x,E) q(x,E)

24
Therefore, fixing the lower limit of integration in region II as x0 = B, the
coefficients must be C = F/2 = T /2eiπ/4 and D = iT eiπ/4 .

We now repeat the matching between regions I and II, where x0 = A and
0
V (A) > 0. In the vicinity of x = A we write

φ(x) ≈ f Ai(−y) + gBi(−y).

In region II this becomes

1
Rx 1
Rx
q(x0 ,E)dx0 q(x0 ,E)dx0
φ(x) ≈ √ f e− ~ A +√ g
e~ A
2 q(x,E) q(x,E)

h i
f e−∆/~ − ~1
Rx 1
Rx
q(x0 ,E)dx0 q(x0 ,E)dx0
=√ 1
2
e B + ge∆/~ e ~ B
q(x,E)

where we have defined Z B

∆= q(x, E)dx.
A
Comparing with the form in region II we see that
f e−∆/~
C = 2
≡ T2 eiπ/4

D = ge∆/~ ≡ iT eiπ/4
or π
f = T e∆/~+i 4
π
g = iT e−∆/~+i 4
Finally, we look at region I to determine A and R as a function of T . In
this region we have

φ(x) ≈ f Ai(−y) + gBi(−y)

f
Rx g
Rx
≈√ 1
p(x0 , E)dx0 + π/4 + √ 1
p(x0 , E)dx0 − π/4 .



cos ~ A
cos ~ A
p(x,E) p(x,E)

25
Writing the cosines as exponentials and comparing with the general form in
region I we see that
iπ/4 −iπ/4
A = f e +ge 2

T eiπ/2+∆/~ +iT e−∆/~

= 2

= iT cosh (∆/~)
and
f e−iπ/4 +geiπ/4
R = 2

T e∆/~ +iT e−∆/~+iπ/2

= 2

= T sinh (∆/~).

The final result is


i cosh (∆/~) i R x p(x0 ,E)dx0 sinh (∆/~) − i R x p(x0 ,E)dx0

 √ e~ A + √ e ~ A x∈I
p p








 1/2 R x
i 1 0 0 i3/2 1 R x 0 0
φ(x) = T √ e ~ B q(x ,E)dx + √ e− ~ B q(x ,E)dx x ∈ II .

 2 q q





 1 i Rx 0 0
 √ e ~ B p(x ,E)dx x ∈ III


p
(66)
The transmission and reflection coefficients are
1
t= 2 ≈ 4e−2∆/~ (67)
cosh (∆/~)

and
r = tanh2 (∆/~) ≈ 1 − 4e−2∆/~ . (68)

26