WKB Approximation

General WKB method

Ref: Bender & Orszag, Adv Math Methods for Scientists and Engineers

The WKB is a general method for obtaining an asymptotic approximation to the solution y(x)
of linear differential equation whose highest derivative is multiplied by a small parameter ,
dn y dn−1 y dn−2 y
 + a(x) + b(x) + · · · + z(x)y = 0. (1)
dxn dxn−1 dxn−2
(Here, we will exclusively confined ourselves in 1-dimension. Generalization to higher dimen-
sion, though not very difficult, requires use of ∇ operator and hence need little more effort.)
Solutions to such equations usually develop regions of rapid variation as  → 0. If the extent of
these regions remains finite as  → 0 in such cases WKB is used. The region is called breakdown
region. The approximate solution for such a differential equation is of plane-wave kind,

y(x) ∼ A(x) eS(x)/δ , δ → 0. (2)

The phase S(x) is assumed to be non-constant and slowly varying in the breakdown region.
When S(x) is real, there is a boundary layer of thickness δ; when S(x) is imaginary there is a
region of rapid oscillation characterized by waves having wavelength of order δ. When S(x) is
constant, the behavior of y(x) is expressed by slowly varying amplitude A(x). The exponential
form (2) is not in a form most suitable for deriving asymptotic approximations because A(x)
and S(x) depend implicitly on δ (however, Griffiths made use of this formula) and can be
expressed as asymptotic series expansion,
" ∞ #
1X n
y(x) ∼ exp δ Sn (x) δ → 0. (3)
δ
n=0

Substitution of the above ansatz in the differential equation (1) and canceling out the expo-
nential terms allows one to solve for an arbitrary number of terms Sn (x) in the expansion.
Keeping Schrödinger equation in mind, consider a second order homogeneous linear differential
equation with y 0 absent (henceforth we will mostly work with this specialized form),

2 y 00 = Q(x)y, with Q(x) 6= 0. (4)

" ∞ # " ∞ #
1 X 1 X
y0 ∼ δ n Sn0 exp δ n Sn
δ δ
n=0 n=0
 !
∞
!2 ∞
" ∞ #
1 X 1 X 1 X
y 00 ∼  2 δ n Sn0 + δ n Sn00  exp δ n Sn (5)
δ δ δ
n=0 n=0 n=0

Substituting (5) in (4) and dividing away the exponential factor,

 
2 1 0 0 2 0

2 1 00 00 2 00


 S + δS1 + δ S2 + · · · + S + δS1 + δ S2 + · · · = Q(x)
δ2 0 δ 0
 
2 1 02 0 0 2 0 0 2 02

1 00 00


or,  S + 2δS1 S0 + 2δ S0 S2 + δ S1 + · · · + S + δS1 + · · · = Q(x)
δ2 0 δ 0
2 02 2 0 0 2 00
or, S + 2 S S + S0 + · · · = Q(x). (6)
δ2 0 δ 0 1 δ

1
In the limit δ → 0, the largest term in the expression (6) is 2 /δ 2 S002 and must have the same
order as Q(x), implying δ ∝ . This is obvious because the ratio /δ being constant, the change
of order in Q(x) can only be absorbed in S00 (x). For simplicity let δ =  and then comparing
powers of  we have,

0 : S002 = Q(x) Eikonal equation (7)

1
: 2S00 S10 + S000 =0 Transport equation (8)
n−1
X
n : 2S00 Sn0 + Sn−1
00
+ Si0 Sn−i
0
=0 for n ≥ 2. (9)
i=1

The Eikonal equation (7) is the lowest i.e. leading order term and can be solved as,
 2 Z xp
dS0
= Q(x) ⇒ S0 (x) = ± Q(x0 ) dx0 . (10)
dx

The Transport equation (8) can be solved using the solution of Eikonal equation (10),

dS0 dS1 d dS0

2 + =0
dx dx dx dx
p dS1 dp
2 Q(x) + Q(x) = 0
dxp dx
1 d Q(x) 1 p 1
or, dS1 = − p ⇒ S1 (x) = − log Q(x) = − log Q(x). (11)
2 Q(x) 2 4

The next two terms are given below, whose derivation are left to the students as exercise,

Q00 5 Q02
Z x 
S2 (x) = ± − dx0 (12)
8Q3/2 32 Q5/2
Q00 5 Q02
S3 (x) = − + . (13)
16Q2 64 Q3
Combining S0 and S1 , the first order WKB approximation to a linear second-order Schrödinger-
like equation (4) is,
   
1 1
y(x) ∼ exp (S0 + δS1 ) = exp S0 + S1
δ δ
 Z xp 
1 0 0 −1/4
= exp ± Q(x ) dx + log Q(x)
δ
 Z 
−1/4 1 p 0 0
= Q(x) exp ± Q(x ) dx ,
δ
 Z   Z 
−1/4 1 p 0 0 −1/4 1 p 0 0
y(x) ∼ C1 Q(x) exp Q(x ) dx + C2 Q(x) exp − Q(x ) dx (14)
δ δ

where C1 and C2 are constant to be determined from boundary conditions. This is obviously
the leading order WKB approximation to the solution of (4); it differs from the exact solution
by terms of order of  in regions where Q(x) 6= 0.

2
Problem -- solve the following differential equations using 1st order WKB

1.  y 00 + y = 0 where y(0) = 0, y(1) = 1 as  → 0

√
sin x/ 
Ans : y(x) = √ ,  6= (nπ)−2
sin 1/ 
2. 2 y 00 = (1 + x2 )2 where y(0) = 0, y 0 (0) = 1
x3
  
 1
Ans : y(x) = √ sinh x+
1 + x2  3

Condition for validity of WKB

WKB theory is a singular perturbation theory which is evident from the 1/δ term in (3) and
also from the fact that the order of the differential equation changes rapidly when  = 0.
Unless S0 (x) = 0, the approximation ceases toP exist when δ = 0. Besides, the singular nature
also shows up in the fact that WKB series n
n δ Sn usually diverges! Still it gives good
approximation to y(x). Hence there is a criteria when WKB is useful. How small δ must be
for WKB series to approximate y(x) to some relative error requiresP to be addressed. In order
WKB to be valid on an interval, it is necessary that the series n δ n−1 Sn be asymptotic in δ
as δ → 0 uniformly for all x in the interval and this requires the asymptotic relations as δ → 0,
1
S1 (x)  S0 (x)
δ
δ S2 (x)  S1 (x)
δ n Sn+1 (x)  δ n−1 Sn (x) as δ → 0 (15)

However,
P WKB series appears in exponent and hence the above is not sufficient to ensure that
exp( δ n−1 Sn ) be a good approximation to y(x). Because even if the above condition (15)
holds true, the term δ n Sn+1 can still be large and upon exponentiation the contribution of the
term can be  1. Therefore, for WKB series terminated at the term δ N −1 SN to be a good
approximation, the next term should be small compared to 1,

δ N SN +1  1 as δ → 0. (16)

The relative error between y(x) and WKB is small,

y(x) − exp( δ n−1 Sn )

P
∼ δ N SN +1  1 as δ → 0. (17)
y(x)

For good approximation both (15) and (16) have to be satisfied.

Problem -- Solve the Airy equation y 00 = xy using WKB for x → ∞.

Here Q(x) = x,  = 1 and the terms in exponent are
2 1 5
s0 = ± x3/2 , S1 = − log x S2 = ± x−3/2
3 4 48
3/2
For  = 1 we have S2  S1  S0 /. For x > 0, the solution is x−1/4 e±2/3x .
Problem -- Solve the following equation to appropriate order for large x,
 2
ln x
y 00 (x) = y(x)
x

3
With  = 1 and Q(x) = x−2 (ln x)2 , the phases Si ’s are,
Z xp
1
S0 (x) = ± Q(x0 ) dx0 = ± (ln x)2
2
1 1 1
S1 (x) = − ln Q(x) = ln x − ln(ln x)
4  2 2
Q00 5 Q02
Z x
1 3
S2 (x) = ± 3/2
− 5/2
dx0 = ± ln(ln x) ± (ln x)−2
8Q 32 Q 8 16
Q00 5 Q02 3 1
S3 (x) = − 2
+ 3
= (ln x)−4 − (ln x)−2 .
16Q 64 Q 16 16
From the above expressions it is clear S2  S1  S0 as x → ∞ but S2  1 for
x → ∞. Hence, the condition (15) is satisfied but the condition (16) with N =
1 is violated. However, with N = 2 the condition (16) is recovered and the asymptotic
solution to y(x) is given by,

y(x) = eS0 +S1 +S2 .

The terminology that is often used in connection with WKB depending on the order of ap-
proximation, either N = 0 or N = 1 are,

y(x) = eS0 (x)/δ , δ→0 Geometrical optics approx (18)

S0 (x)/δ+S1
y(x) = e , δ→0 Physical optics approx (19)

Usually, the approximation of geometrical optics give the classical solution but does not consti-
tute an asymptotic approximation to y(x). The relative error between y(x) and approximation
of physical optics is of order δS2 , which vanishes uniformly with δ if S2 (x) is bounded.
Connecting WKB to Quantum Mechanics
The Schrödinger equation is evidently of the form (4),

~2 d2 ψ(x)
− + V (x) ψ(x) = E ψ(x) ⇒ −~2 ψ 00 (x) = 2m(E − V (x)) ψ (20)
2m dx2
where 2 = −~2 and Q(x) = 2m(E − V (x)), where E > V (x) the so-called classical region.
Clearly, for the case of constant potential,
p
V (x) = V0 = const ⇒ classical mom p0 = 2m(E − V0 ) → ψ ∼ eipx/~ . (21)

The momentum is classical because, being constant for all x, the commutation with x is zero.
The wavefunction is oscillatory with constant wavelength λ = h/p0 . Besides, constant or
not, there is no operator form hypothesized for momentum. For potentials varying slowly in
comparison to λ, then it is fair to assume the wavefunction remains essentially oscillatory,
except the wavelength and amplitude are changing slowly with x, where λ(x) = h/p(x). The
natural choice for the solution is (2) like,

ψ(x) ∼ A(x) eiS(x)/~ , (22)

where, in the leading order S(x) → S0 → p0 x, which is just the classical solution. This is the
reason why WKB is called semi-classical approximation. The ansatz (22) can be used to arrive

4
at the solution of (20). Here we attempt the solution in Griffith’s way (in class we used eqns.
(10, 11) directly),
i
ψ 0 = A0 eiS/~ +A S 0 eiS/~
~
 2
00 00 iS/~ i 0 0 iS/~ i i
ψ =A e +2 A S e ++ A S 02 eiS/~ + A S 00 eiS/~
~ ~ ~
2 p2 (x)
 
~ 2i i 1
⇒ − A00 + A0 S 0 + AS 00 − 2 AS 02 = A[E − V (x)] = A (23)
2m ~ ~ ~ 2m
The above expression (23) can be splitted up in imaginary and real parts. The imaginary part
leads to the determination of the amplitude A(x),
2A0 S 0 + AS 00 = 0 ⇒ d A2 S 0 = 0

C
or, A2 S 0 = real const = C 2 ⇒ A(x) = p (24)
S 0 (x)
The real part is used to solve for S(x),
1 p2 1
A00 − 02
A ⇒ A00 = 2 A S 02 − p2


2
AS = − 2
(25)
~ ~ ~
Using the assumption that A(x) varies slowly, we can choose A00 = 0 implying,
Z x
S 02 − p2 = 0 ⇒ S 0 (x) = ±p(x) ⇒ S(x) = ± p(x0 ) dx0 .


(26)

Putting (24) and (26) together,

 Z x 
C i 0 0
ψ(x) = p exp ± p(x ) dx . (27)
p(x) ~
Deriving the same expression above (27) from Eikonal equation (7) and Transport equation
(8), with δ = −i~, is also straight forward, this amounts to physical optics approximation but
the validity of WKB to this order depends on the condition (16) i.e. |δS2 |  1. Using the
expression for S2 in (12) and observing that
Q(x) = 2m(E − V (x)) = p2 (x) → Q0 = −2mV 0 and Q00 = −2mV 00
we have,
x
2mV 00 (−2mV 0 )2
Z 
5
S2 = − − dx0 (28)
8{2m(E − V )}3/2 32 {2m(E − V )}5/2
The first term on rhs of (28) can be written as,
V 00 V0 V 00 −2mV 02
 
d 3
→ = −
{2m(E − V )}3/2 dx {2m(E − V )}3/2 {2m(E − V )}3/2 2 {2m(E − V )}5/2
00 0 m2 V 02
 
mV d mV 3
or − = − + . (29)
4{2m(E − V )}3/2 dx 4{2m(E − V )}3/2 4 {2m(E − V )}5/2
Putting (29) in (28), we have
mV 0 m2 V 02
Z x   Z x 
d 0 3 5
S2 = − dx + − dx0
dx 4{2m(E − V )}3/2 4 8 {2m(E − V )}5/2
1 2 1 dV 2 0
Z x  
1 1 dV
= − m 3 + m 5 dx . (30)
4 p dx 8 p dx
The following points are to be noted,

5
 1. Obviously, for S2 (x) to be smaller, the dV /dx got to be smaller, i.e. slowly varying
potential.

2. E − V (x) must not be too small and certainly not zero i.e. Q(x) 6= 0, in which case
WKB fails completely.

3. In similar way, we can reason that S3 , S4 , . . . contain higher derivative of V (x) and
therefore can be considered small.

As for the condition |δS2 |  1, firstly,

" 2 #
Z x 
1 1 dV 1 2 1 dV i~ 1 dV
δS2 = −i~ − m 3 + m dx0 ≈ m 3 (31)
4 p dx 8 p5 dx 2 p dx

assuming that the maximum of the second term of the above expression (31) is the first term.
Now, since we can write

mV 0 mV 0
⇒ p0 = − p
p
p(x) = 2m(E − V (x)) =−
2m(E − V ) p

substituting this in (31) above,

i~ −mV 0 1 i~ p0
δS2 = − = −
2 p p2 2 p2
~ dp(x)
⇒ |δS2 (x)| = 2  1. (32)
p dx

The valid use of WKB for approximating solutions for quantum mechanical problem is given
by the condition in (32). However, this condition appears in textbooks in all sorts of forms
and all of those can be derived after some simple algebra from the well-known relation,

h ~ ~p0 k0
p= = ~k = ⇒ = = λ̄0
λ λ̄ p2 k2
Hence,
~ 0 k0
2
p 1 ≡  k ≡ λ̄0 λ̄  λ̄. (33)
p k

Problems -- (1) Infinite square well potential with bumpy bottom. Investigate
both flat bottom, flat bottom with a shelf of height V0 half-way across and electron
bound to H-nucleus by Coulomb force. (2) Scattering / Tunneling in a rectangular
barrier with bumpy top. Specialize in flat top of height V0 , V0 −λx and Gamow-Teller
alpha-decay problem.
Turning points and connection formulas
It has been mentioned before that the validity of WKB series in an interval is violated if Q(x)
vanishes in that interval. Plainly put, as  → 0, the condition S0 /  S1 ∝ ln Q(x) breaks
down at the zero of Q(x) simply because S1 becomes singular. The points where Q(x) = 0
are called turning points. The expression for turning points comes from Schrödinger equation
when E = V (x) implies Q(x) = 0. The classical orbit of a particle in potential V (x) is confined
to regions where E ≥ V (x). The particle moves until it reaches a point where E = V (x) and
then it turns around. The physical optics approximation is clearly invalid at a turning point.

6
However, if Q(x) analytic then the exact solutions of the differential equation are regular.
Asymptotic matching is done to resolve this puzzle.
p
E > V (x): In the classical region, p(x) = 2m(E − V (x)) is a real function and the most
general solution of Schrödinger equation is (27),
 Z x
i x
  Z 
B i 0 0 C 0 0
ψ(x)cl = p exp p(x ) dx + p exp − p(x ) dx . (34)
p(x) ~ p(x) ~
p
E < V (x): It is the non-classical, i.e. classically forbidden, region, p(x) = −2m(V (x) − E)
is imaginary and the general solution of Schrödinger equation is,
1 x
 Z   Z x 
D 0 0 F 1 0 0
ψ(x)ncl = p exp − |p(x )| dx + p exp |p(x )| dx . (35)
|p(x)| ~ |p(x)| ~
The breakdown happens where classical region meets non-classical region and that is where
E = V (x), i.e. p(x) = 0 = Q(x) and ψ(x) → ∞. This turning points cannot be simply ignored
because boundary conditions there determine the allowed energies and tunneling probability.
Suppose that x = a is a classical turning point with the given energy as shown in the figure,
where clearly x < a is the classical region and x > a is the non-classical region and the solutions
follow from (34) and (35),
 Z a
i a
  Z 
B i 0 0 C 0 0
x < a : ψ(x)cl = p exp p(x ) dx + p exp − p(x ) dx (36)
p(x) ~ x p(x) ~ x
1 x
 Z 
D 0 0
x > a : ψ(x)ncl = p exp − |p(x )| dx . (37)
|p(x)| ~ a

In the neighborhood of x = a, we approximate the potential by a straight line (a fair approxi-

mation since potential is varying slowly),
V (x) = V (a) + (x − a)V 0 (x)|x=a ⇒ V (x) = E + α(x − a) → V (x) − E ≈ α (x − a) ≡ α x̃,
(38)
where α is the slope at x = a. The neighborhood of x = a (shaded block) is called patch region.
The Schrödinger equation in the patch region is,
~2 d2 ψp d2 ψp 2mα
− + (E + αx̃) ψp = E ψp → = β 3 x̃ ψp where β 3 = . (39)
2m dx2 dx̃2 ~2
The β can be absorbed into an independent variable by,
d2 ψ 1 d2 ψ
 
dψ dψ dx̃ 1 dψ d dψ d 1 dψ dx̃
z = β x̃ ⇒ = = and = = = 2 2, (40)
dz dx̃ dz β dx̃ dz 2 dz dz dx̃ β dx̃ dz β dx̃

7
which when used in (39) yields Airy’s equation,

d2 ψp d2 ψp
= β 3 x̃ ψp ⇒ = z ψp . (41)
dx̃2 dz 2
The solutions of Airy’s equation are Airy’s functions Ai(z) and Bi(z),
 
1 2 3/2 π
Ai(z) ∼ √ sin (−z) + z0 (42)
π (−z)1/4 3 4
 
1 2 3/2
∼ √ exp − z z0 (43)
2 π z 1/4 3
 
1 2 π
Bi(z) ∼ √ 1/4
cos (−z)3/2 + z0 (44)
π (−z) 3 4
 
1 2 3/2
∼ √ 1/4 exp z z0 (45)
πz 3

Therefore, the patching function (the solution of Schrödinger equation in patching region) is,

ψ(x)p = M Ai(β x̃) + N Bi(β x̃). (46)

The patch function ψp in the neighborhood of turning point is to be matched with WKB
solutions (36) and (37) in the overlap regions, i.e. on the boundary of the shaded blocks, on
either side. The overlap regions are close enough to the turning point that the linear potential
approximation (38) is reasonably accurate so that ψp is a good approximation to the true
wavefunction and still far enough away from turning point that WKB is valid. In overlap of
patch and non-classical zone, therefore,
p √ √ √
p(x) = 2m(E − V (x)) = 2mα −x̃ = i~ β 3/2 x̃
Z x Z x̃ √
0 0 2
⇒ |p(x )| dx = ~β 3/2
x̃0 dx̃0 = ~(β x̃)3/2 . (47)
a 0 3
Hence, the WKB wavefunction in non-classical region is,
 
D 2 3/2
ψ(x)ncl = √ exp − (β x̃) . (48)
~ β 3/4 x̃1/4 3

The patch solution is,

   
M 2 3/2 N 2 3/2
ψp = √ exp − (β x̃) +√ exp (β x̃) . (49)
2 π (β x̃)1/4 3 π (β x̃)1/4 3

Comparing ψncl and ψp , we see that,

r
M D 4π
√ 1/4
=√ ⇒ M= D and N = 0. (50)
2 π (β x̃) ~ β x̃1/4
3/4 β~

The same procedure has to be repeated for the overlap of patch zone and classical zone. This
time though,
p √ √ √
p(x) = 2m(E − V (x)) = 2mα x̄ = ~ β 3/2 x̄, where x̄ = a − x
Z a Z 0√
0 0 2
⇒ p(x ) dx = ~β 3/2
x̄0 (−dx̄0 ) = ~(β x̄)3/2 . (51)
x x̄ 3

8
Hence, the WKB wavefunction in classical region is,
   
B 2i 3/2 C 2i 3/2
ψ(x)cl = √ exp (β x̄) +√ exp − (β x̄) . (52)
~ β 3/4 x̄1/4 3 ~ β 3/4 x̄1/4 3

Remembering N = 0 in (50), the patch solution is,

 
M 2 3/2 π
ψp = √ sin (β x̄) +
π (β x̄)1/4 3 4
    
M 1 iπ 2i 3/2 iπ 2i 3/2
= √ exp + (β x̄) − exp − − (β x̄) . (53)
π (β x̄)1/4 2i 4 3 4 3

Comparing ψcl and ψp above, we see that,

M eiπ/4 1
r
B iπ/4 π
√ = √ ⇒ Me = 2i B
~ (β x̄)1/4 β 1/2 π (β x̄)1/4 2i β~
M e−iπ/4
r
C 1 π
√ = −√ ⇒ M e−iπ/4 = −2i C. (54)
~ (β x̄)1/4 β 1/2 π (β x̄)1/4 2i β~

Therefore, the connection formulas are now obtained by joining WKB solutions at either side
of the turning point obtained by combining (50) and (54),

B = −ieiπ/4 D and C = ie−iπ/4 D. (55)

Thus the WKB solutions ψcl in classical and ψncl in non-classical zone got matched and we are
done with the patching function ψp and the solutions get expressed in terms of constant D,

1 x
 Z 
D 0 0
ψ(x)ncl = p exp − |p(x )| dx x>a (56)
|p(x)| ~ a
i a i a
  Z   Z 
D iπ 0 0 iπ 0 0
ψ(x)cl = p −i exp + p(x )dx + i exp − − p(x )dx
p(x) 4 ~ x 4 ~ x
i a i a
  Z   Z 
2D 1 iπ 0 0 iπ 0 0
= p exp + p(x )dx − exp − − p(x )dx
p(x) 2i 4 ~ x 4 ~ x
 Z a 
2D 1 0 0 π
= p sin p(x ) dx + x < a. (57)
p(x) ~ x 4

In exactly the same way we will get the classical and non-classical wavefunction at turning
point x = b, the only difference being the change in the sign of the slope at the turning point,

D0 1 b
 Z 
0 0
ψ(x)ncl = p exp − |p(x )| dx x<b (58)
|p(x)| ~ x
2D0
 Z x 
1 0 0 π
ψ(x)cl = p sin p(x ) dx + x > b. (59)
p(x) ~ b 4

The solutions for the turning points can be straight away used in potential well i.e. two turning
points problem, as shown in the figure below.

9
The region b < x < a is the classical region where E > V (x) and x < b, x > a are the
non-classical regions since there E < V (x). If bound states of the system exist then it should
be in the classical region. The figure showing two turning points is clearly a combination of
the two cases we already discussed and hence the solutions in various regions are known. Of
particular interest is the classical region b < x < a, where the wavefunction can either be (57)
or (59),
 Z a 
a 2D 1 0 0 π
ψ(x)cl = p sin p(x ) dx + x<a (60)
p(x) ~ x 4
2D0
 Z x 
b 1 0 0 π
ψ(x)cl = p sin p(x ) dx + x>b (61)
p(x) ~ b 4

Actually ψcla and ψ b must have the same functional form, but it is apparent that the argument
cl
of sine functions are not identical. To achieve the equality, we rewrite the integral in the
argument of sine in (60) as,
Z a Z a Z x  Z a
1 x
Z 
2D 1 0 0 0 0 π
= − ⇒ p sin p(x ) dx − p(x ) dx +
x b b p(x) ~ b ~ b 4
 Z x Z a 
−2D 1 1 π
= p sin p(x0 ) dx0 − p(x0 ) dx0 −
p(x) ~ b ~ b 4
 Z x Z a 
−2D 1 π 1 π
= p sin p(x0 ) dx0 + − p(x0 ) dx0 − . (62)
p(x) ~ b 4 ~ b 2

In order that this expression be functionally identical to that of (61), it is necessary that,

1 a
Z Z a  
π 1 h
p(x) dx + = nπ ⇒ p(x) dx = n − and D = (−1)n D0 , n = 1, 2, . . . (63)
~ b 2 b 2 2
This is essentially the quantization condition we encounter in quantum mechanics of bound
state problems. The above quantization condition (63) can be put in a well-known form if we
perform the above integral over full cycle i.e. over a complete period,
Z a I  
1
2× p(x) dx ≡ p(x) dx = n − h, (64)
b 2

which is the Wilson-Sommerfeld quantization rule plus the zero-point energy term h/2. Thus
WKB helps getting quantization rules requiring the knowledge of potential and boundary
conditions and involving only integration over the momentum without actually solving the
Schrödinger equation.

10
Problem -- (1) Simple harmonic oscillator, (2) quantum bouncer in gravitational
field, (3) harmonic oscillator with rigid wall, (4) Zettili’s problems 9.26, 9.28,
9.29
Tunneling problem
The WKB method can be used to analyze tunneling problem through an arbitrary shape
potential barrier. Consider a generic barrier as shown below.

Identification of classical and non-classical region is straight forward: in x < a and x > b
the classical WKB wavefunctions are ψ cl and ψcl b and in a < x < b the non-classical WKB

wavefunction is ψncl ,
1 h i Ra 0 0 i a
R 0 0
i
ψ(x)acl = p Ae ~ x p(x )dx + Be− ~ x p(x )dx x<a (65)
p(x)
1 h 1 Rx 0 0 1 x
R 0 0
i
ψ(x)ncl = p Ce ~ a |p(x )|dx + De− ~ a |p(x )|dx a<x<b (66)
|p(x)|
1 i x
R 0 0
ψ(x)bcl = p F e ~ b p(x )dx x>b (67)
p(x)
Consider the left slope of the potential barrier where the particle strikes it with energy E.
The momentum integration in x < a region (65) has already been calculated in (51) and the
corresponding WKB wavefunction obtained in (52),
1 h 2i 3/2 2i 3/2
i
ψ(x)acl = √ A e 3 (β x̄) + B e− 3 (β x̄) . (68)
h β 3/4 x̄1/4
The corresponding non-classical wavefunction at the left slope is given by (48), except that
this time both the terms survive,
1 h 2 3/2 2 3/2
i
ψ(x)ncl = √ C e 3 (β x̃) + D e− 3 (β x̃) . (69)
~ β 3/4 x̃1/4
The patch solution (49), however, remains unchanged,
 
1 2
(β x̃)3/2 N − 2 (β x̃)3/2
ψ(x)p = √ M e3 + e 3 . (70)
π (β x̃)1/4 2
Comparing the patch wavefunction ψp with the non-classical wavefunction ψncl for matching,
we obtain, r r
π 4π
M= C and N = D. (71)
β~ β~

11
Clear enough that in this case both the terms of patch wavefunction (70) survives. Comparison
of ψp with the classical solutions ψcla is rather lengthy because now both the terms are there,

therefore from (46), (42) and (44) we write ψp as,

    
1 2 3/2 π 2 3/2 π
ψ(x)p = √ M sin (β x̄) + + N cos (β x̄) +
π(β x̄)1/4 3 4 3 4

1 M  2i/3(β x̄)3/2 +iπ/4 3/2

= √ e − e−2i/3(β x̄) −iπ/4 +
π(β x̄)1/4 2i
N  2i/3(β x̄)3/2 +iπ/4 
−2i/3(β x̄)3/2 −iπ/4
e +e
2
1 h 2i
(β x̄)3/2 +iπ/4 − 2i (β x̄)3/2 −iπ/4
i
= √ (−iM + N ) e 3 + (iM + N ) e 3 (72)
2 π(β x̄)1/4
Comparing (72) and (68), we get
r   r  
β~ −iM + N β~ iM + N
A= eiπ/4 and B= e−iπ/4 (73)
π 2 π 2
Combining (71) and (73) the result is,
   
C C
A= D−i eiπ/4 and B = D+i e−iπ/4 . (74)
2 2
These are the connection formulas at the left slope at x = a.

Next consider right slope at x = b, the non-classical and classical WKB solutions here are,
1 h 1 Rx 0 0 1 x
R 0 0
i
ψ(x)ncl = p Ce ~ a |p(x )|dx + De− ~ a |p(x )|dx
|p(x)|
1 h 1 Rb 0 0 1 b
R 0 0 1 b
R 0 0 1 b
R 0 0
i
= p Ce ~ a |p(x )|dx − ~ x |p(x )|dx + De− ~ a |p(x )|dx + ~ x |p(x )|dx
|p(x)|
1 h 1 x
R 0 0 1 x
R 0 0
i
= p Ceγ e ~ b |p(x )|dx + De−γ e− ~ b |p(x )|dx
|p(x)|
1 h 2 3/2 2 3/2
i
= √ C 0 e 3 (β x̄) + D0 e− 3 (β x̄) (75)
~β 3/4 x̄1/4
1 b
Z
where γ= |p(x0 )| dx0 , C 0 = C eγ and D0 = D e−γ
~ a
1 2i 3/2
ψ(x)bcl = √ F e 3 (β x̃) , (76)
~β 3/4 x̃1/4
We arrive at the coefficients C 0 and D0 at x = b, by making use of the following patch solution,
 
1 2
(β x̄)3/2 N − 2 (β x̄)3/2
ψp = √ M e3 + e 3 , (77)
π(β x̄)1/4 2
r r
π 0 π 0
⇒ M= C and N = 2 D. (78)
β~ β~
The patch for classical WKB wavefunction at x = b is simply the expression (72) only x̃
replacing the x̄ and comparing that with (76), we get,
r
1 β~
F = (−iM + N ) eiπ/4 and 0 = (iM + N ) ⇒ N = −iM (79)
2 π

12
Therefore, the coefficients are,
r r
π −iπ/4 π −iπ/4
M =i e F, and N = e F
β~ β~
1 γ −iπ/4
C = C 0 e−γ = i e−γ e−iπ/4 F and D = D0 eγ = e e F (80)
2
Using (80) in (74), the coefficient A can be determined in terms of F ,
 
iC 1 γ 1
eiπ/4 = e − e−γ F = 1 − e−2γ eγ F.



A= D− (81)
2 2 2

Hence, the transmission coefficient is,

2
F 4 e−2γ
T = = ⇒ T ≈ e−2γ as γ → ∞. (82)
A (1 − e−2γ )2
Rb
The large value of γ = (1/~) a |p(x)|dx implies broad barrier.
Problems -- (1) Griffiths’ problems 8.12, 8.13, 8.14, example 8.2

13