Derive WKB PDF
Derive WKB PDF
Derive WKB PDF
This condition
means that (t) changes very little, where L(t) is some unknown function; let us now derive
i.e. d
dt t , during one period of oscillation, t 1/.
what this L(t) must be. Substituting Eq. (10) into Eq. (8)
This is the precise meaning of the condition that (t) should gives the following equation for L(t),
be slow-changing.
We have for simplicity assumed that 2 > 0 for all t; iL L2 + 2 = 0, (11)
in case 2 (t) becomes zero or negative for some values of
t, the basic formula is still valid (we need to treat (t) as and it follows that the function L(t) must be complex-
complex-valued function) but there are several subtleties valued. So we write L(t) = W (t) iB(t), where the choice
in how one glues the solutions in different intervals of t. I of the sign is the same as above, and W (t) and B(t) are
shall not discuss these issues here. some real-valued and (so far) unknown functions. Then we
substitute this into Eq. (11) and separate the real and the
imaginary parts of that equation. We find that
3 Derivation of Eq. (2)
B B 2 = 2 W 2 , (12)
Here is a simple derivation that involves a little guess-
2W B = 0.
W (13)
ing. We know that if = const then the solutions are
The second equation can be immediately solved,
1.5
W 1
B= , (14)
2W 0.5
-0.5
W 3W 2
= 2 W 2 . (15) -1
2W 4 W2 -1.5
pansion
0 10 20 30 40
0
-0.5
The actual value of must be = 1 since we do not really Here are some numerical examples of how well the WKB
wish to modify Eq. (17). However, now the smallness is approximation works, for some specific functions (t).
labeled by , and for instance any terms multiplied by 2 The first example is (t) = 1 + 0.6 tanh(t/2). The exact
will certainly be smaller than any terms with . So we can (numerically obtained) solution and the WKB approxima-
expand Eq. (19) in Taylor series in , e.g. tion are both plotted in Fig. 1. The worst agreement is about
20%. The function (t) is not really slow-changing since
!
W 3W 2
d
W + ..., (20) 0.5 2 for t = 0. (23)
2 2W 4 W2 dt
and also assume that W (t) is itself a power series in , This example shows the error of the WKB approximation
in an exaggerated way; the difference between the WKB
W (t) = + S1 (t) + 2 S2 (t) + ... (21) approximation and the exact solution would be invisible
on the graph for e.g. (t) = 5 + 2 tanh t.
Now we can compute the terms S1 , S2 , etc., one by one, The second example is (t) = 1 + 0.06 cos(2t). This func-
simply by substituting Eq. (21) into Eq. (20) and by collect- tion exhibits the so-called parametric resonance: the solution
ing terms with equal powers of . This task can be easily x(t) grows without bound with t. However, Fig. 2 shows
performed on a computer using an algebraic manipulation that the WKB approximation completely fails to track this
system such as Maple or Mathematica. The first terms of growth, even though the condition (4) is well satisfied for
the series (21) are this (t).
1
3 2
The last example shows that the WKB approximation is
W (t) = fundamentally limited in its precision. One cannot repro-
4 2 8 3
... duce the parametric resonance even if one uses the more
1 (4) 2 2 297 3
2 5 13 99 precise ansatz (22) and computes many terms of the se-
+ + + ...
16 4 8 5 32 5 32 6 128 7 ries (21).