Asymptotic Analysis and Perturbation Theory

Prof Thomas Neukirch

(with thanks to Dr Antonia Wilmot-Smith)
University of St Andrews

6 November 2024
Asymptotic & Analytical Methods: Lecture 5 – WKBJ
While we wait: Differentiate y(x) = eY (x) :

dy d2 y
= =
dx dx2
We will need these today.

Also: Which differential equation has oscillatory solutions and

which exponential solutions?

y 00 (x) + λ2 y(x) = 0 (1)

00 2
y (x) − λ y(x) = 0 (2)

(1) : y = Aeiλx + Be−iλx = C sin(λx) + D cos(λx).

(2) : y = Aeλx + Be−λx
Asymptotic & Analytical Methods: Lecture 5 – WKBJ
While we wait: Differentiate y(x) = eY (x) :
2
d2 y d2 Y Y (x)

dy dY Y (x) dY
= e ; = eY (x) + e
dx dx dx2 dx dx2

We will need these today.

Also: Which differential equation has oscillatory solutions and

which exponential solutions?

y 00 (x) + λ2 y(x) = 0
y 00 (x) − λ2 y(x) = 0

(1) : y = Aeiλx + Be−iλx = C sin(λx) + D cos(λx).

(2) : y = Aeλx + Be−λx
Introduction

WKB Method
Also known as WKBJ Method.
Introduction

WKB Method
Also known as WKBJ Method.
In the last two lectures we considered singular perturbation
problems with local boundary layer behaviour. These types of
singular perturbation problems are called locally singular .
Introduction

WKB Method
Also known as WKBJ Method.
In the last two lectures we considered singular perturbation
problems with local boundary layer behaviour. These types of
singular perturbation problems are called locally singular .

Here we look at a class of problems which are globally singular and

cannot be solved using matched expansions or related methods.
Introduction

WKB Method
Also known as WKBJ Method.
In the last two lectures we considered singular perturbation
problems with local boundary layer behaviour. These types of
singular perturbation problems are called locally singular .

Here we look at a class of problems which are globally singular and

cannot be solved using matched expansions or related methods.

We use what are called WKB methods after the physicists

Wentzel, Kramers and Brillouin (about 1927).
Actually the idea originated much earlier with Liouville (1820),
Green (1830), Rayleigh (1890) and Jeffreys (1906) but the label
WKB seems to have stuck.
Introduction

Consider the apparently simple singular perturbation problem

y 00 + y = 0, y(0) = 0, y(1) = 1.
Introduction

Consider the apparently simple singular perturbation problem

y 00 + y = 0, y(0) = 0, y(1) = 1. (3)

Suppose we did not know better and tried the straightforward

expansion
y = y0 (x) + y1 (x) + · · · (4)
Introduction

Consider the apparently simple singular perturbation problem

y 00 + y = 0, y(0) = 0, y(1) = 1. (3)

Suppose we did not know better and tried the straightforward

expansion
y = y0 (x) + y1 (x) + · · · (4)
Matching powers of  gives the trivial result

y0 = 0, y1 = 0, ··· , yn = 0.

So y ≡ 0.
y 00 + y = 0, y(0) = 0, y(1) = 1.
 y 00 + y = 0, y(0) = 0, y(1) = 1.
Since, the coefficients are constants, the solution can be obtained
analytically via auxiliary equation
i √ √
m2 + 1 = 0 so m = ± √ ⇒ y(x) = A sin(x/ ) + B cos(x/ ).

 y 00 + y = 0, y(0) = 0, y(1) = 1.
Since, the coefficients are constants, the solution can be obtained
analytically via auxiliary equation
i √ √
m2 + 1 = 0 so m = ± √ ⇒ y(x) = A sin(x/ ) + B cos(x/ ).


Matching the boundary conditions gives

 
sin √x 1
y(x) =  ,  6= 2 2 , (5)
sin √1 n π

and the solution exhibits rapid oscillations as  → 0+ .

y 00 + y = 0, y(0) = 0, y(1) = 1.


sin √x 1
y(x) =  ,  6= 2 2
sin √1 n π

 y 00 + y = 0, y(0) = 0, y(1) = 1.

sin √x 1
y(x) =  ,  6= 2 2
sin √1 n π


This is nothing like the straightforward expansion. The problem is

that the solution is globally singular. A different approach is
needed. The solution does not have a series solution in powers of 
as  → 0. Our choice of anzatz is wrong.
The WKB Method

Applies to second order, linear ordinary differential equations in

which there is a large parameter.
The WKB Method

Applies to second order, linear ordinary differential equations in

which there is a large parameter.

Consider the boundary value problem

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β, (6)

√
for λ ≡ 1/  → ∞ as  → 0.
The WKB Method

Applies to second order, linear ordinary differential equations in

which there is a large parameter.

Consider the boundary value problem

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β, (6)

√
for λ ≡ 1/  → ∞ as  → 0.

Can think of  as multiplying the highest derivative but, since there

is no first derivative term  = 0 implies equation collapses to
algebraic rather than differential equation.
The WKB Method

Applies to second order, linear ordinary differential equations in

which there is a large parameter.

Consider the boundary value problem

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β, (6)

√
for λ ≡ 1/  → ∞ as  → 0.

Can think of  as multiplying the highest derivative but, since there

is no first derivative term  = 0 implies equation collapses to
algebraic rather than differential equation.

For convenience we work with the large parameter λ rather than

the small parameter .
The WKB Method
We will consider

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β.

The WKB Method
We will consider

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β.

I If a linear equation has a first derivative present then we can

always eliminate the first derivative via a suitable
transformation:

y 00 + a(x)y 0 + b(x)y = 0,
The WKB Method
We will consider

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β.

I If a linear equation has a first derivative present then we can

always eliminate the first derivative via a suitable
transformation:

y 00 + a(x)y 0 + b(x)y = 0, y(x) = u(x)z(x),

u00 a(x)2
Z  
1
ln u = − a(x) dx, z 00 + − + b(x) z = 0.
2 x 2
The WKB Method
We will consider

y 00 + λ2 G0 (x)y = 0, y(a) = α, y(b) = β.

I If a linear equation has a first derivative present then we can

always eliminate the first derivative via a suitable
transformation:

y 00 + a(x)y 0 + b(x)y = 0, y(x) = u(x)z(x),

u00 a(x)2
Z  
1
ln u = − a(x) dx, z 00 + − + b(x) z = 0.
2 x 2
I Higher order linear equations can also tackled via the WKB
method.
The WKB Method

y 00 + λ2 G0 (x)y = 0,
The WKB Method

y 00 + λ2 G0 (x)y = 0,

Suppose that G0 (x) = G0 > 0 is a constant.

The WKB Method

y 00 + λ2 G0 (x)y = 0,

Suppose that G0 (x) = G0 > 0 is a constant.

Then the general solution is

√ √
y = Aeiλ G0 x
+ Be−iλ G0 x
, A, B constants.
The WKB Method

y 00 + λ2 G0 (x)y = 0,

Suppose that G0 (x) = G0 > 0 is a constant.

Then the general solution is

√ √
y = Aeiλ G0 x
+ Be−iλ G0 x
, A, B constants.

This suggests that the correct behaviour of y as λ → ∞ has a

rapid oscillatory behaviour.
The WKB Method

y 00 + λ2 G0 (x)y = 0,

Suppose that G0 (x) = G0 > 0 is a constant.

Then the general solution is

√ √
y = Aeiλ G0 x
+ Be−iλ G0 x
, A, B constants.

This suggests that the correct behaviour of y as λ → ∞ has a

rapid oscillatory behaviour.

Strategy: Change dependent variable from y to Y where

y = eλY (x;λ) . (7)

WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

2
y 0 = λY 0 eλY , y 00 = λ2 Y 0 eλY + λY 00 eλY .
WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

2
y 0 = λY 0 eλY , y 00 = λ2 Y 0 eλY + λY 00 eλY .
Substituting gives, on dividing by λ2 eλY ,
1 00
2
Y + Y 0 + G0 (x) = 0.
λ
WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

2
y 0 = λY 0 eλY , y 00 = λ2 Y 0 eλY + λY 00 eλY .
Substituting gives, on dividing by λ2 eλY ,
1 00
2
Y + Y 0 + G0 (x) = 0. (8)
λ

Expand Y in a power series in powers of λ−1 that is valid for

λ−1 → 0.
1
Y = Y0 + Y1 + · · · (9)
λ
WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

2
y 0 = λY 0 eλY , y 00 = λ2 Y 0 eλY + λY 00 eλY .
Substituting gives, on dividing by λ2 eλY ,
1 00
2
Y + Y 0 + G0 (x) = 0. (8)
λ

Expand Y in a power series in powers of λ−1 that is valid for

λ−1 → 0.
1
Y = Y0 + Y1 + · · · (9)
λ
 
1 00 1 00
Y0 + Y1 + · · ·
λ λ
WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

2
y 0 = λY 0 eλY , y 00 = λ2 Y 0 eλY + λY 00 eλY .
Substituting gives, on dividing by λ2 eλY ,
1 00
2
Y + Y 0 + G0 (x) = 0. (8)
λ

Expand Y in a power series in powers of λ−1 that is valid for

λ−1 → 0.
1
Y = Y0 + Y1 + · · · (9)
λ
   
1 00 1 00 0 2 2 0 0
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · ·
λ λ λ
WKB Idea

y 00 + λ2 G0 (x)y = 0, λ → ∞. y = eλY (x;λ) .

2
y 0 = λY 0 eλY , y 00 = λ2 Y 0 eλY + λY 00 eλY .
Substituting gives, on dividing by λ2 eλY ,
1 00
2
Y + Y 0 + G0 (x) = 0. (8)
λ

Expand Y in a power series in powers of λ−1 that is valid for

λ−1 → 0.
1
Y = Y0 + Y1 + · · · (9)
λ
   
1 00 1 00 0 2 2 0 0
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · + G0 (x) = 0.
λ λ λ
WKB Idea

So far:
   
1 00 1 00 0 2 2 0 0
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · + G0 = 0.
λ λ λ
WKB Idea

So far:
   
1 00 1 00 0 2 2 0 0
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · + G0 = 0.
λ λ λ
1
Compare powers of λ.
WKB Idea

So far:
   
1 00 1 00 0 2 2 0 0
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · + G0 = 0.
λ λ λ
1
Compare powers of λ. At O(1) we have

2
2
Y00 + G0 = 0 ⇔ Y00 = −G0 .
WKB Idea

So far:
   
1 00 1 00 0 2 2 0 0
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · + G0 = 0.
λ λ λ
1
Compare powers of λ. At O(1) we have

2
2
Y00 + G0 = 0 ⇔ Y00 = −G0 .

Solution will depend on sign of G0 :

1/2
Y00 = ±iG0 , G0 > 0, (10)
Y00 = ±(−G0 ) 1/2
, G0 < 0. (11)
WKB Idea
So far:
1/2
Y00 = ±iG0 (x), G0 (x) > 0,
Y00 = ±(−G0 )1/2 (x), G0 (x) < 0.
WKB Idea
So far:
1/2
Y00 = ±iG0 (x), G0 (x) > 0,
Y00 = ±(−G0 )1/2 (x), G0 (x) < 0.

I Notice two different forms depending on whether G0 is

positive or negative.
WKB Idea
So far:
1/2
Y00 = ±iG0 (x), G0 (x) > 0,
Y00 = ±(−G0 )1/2 (x), G0 (x) < 0.

I Notice two different forms depending on whether G0 is

positive or negative.
I Notice also that there are two different solutions for Y0 in
each case given by the plus and minus signs.
WKB Idea
So far:
1/2
Y00 = ±iG0 (x), G0 (x) > 0,
Y00 = ±(−G0 )1/2 (x), G0 (x) < 0.

I Notice two different forms depending on whether G0 is

positive or negative.
I Notice also that there are two different solutions for Y0 in
each case given by the plus and minus signs.
I Second order linear differential equations do have two linearly
independent solutions.
WKB Idea
So far:
1/2
Y00 = ±iG0 (x), G0 (x) > 0,
Y00 = ±(−G0 )1/2 (x), G0 (x) < 0.

I Notice two different forms depending on whether G0 is

positive or negative.
I Notice also that there are two different solutions for Y0 in
each case given by the plus and minus signs.
I Second order linear differential equations do have two linearly
independent solutions.
Integrating gives
R 1/2
±i G0 (x) dx,
 G0 (x) > 0,
Y0 = (12)
± (−G0 )1/2 (x) dx G0 (x) < 0.
 R
WKB Idea

So far:  R 1/2
 ±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
 R
WKB Idea

So far:  R 1/2
 ±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
 R

We can ignore arbitrary constants in the integration. Any constant

would simply change the arbitrary constants multiplying this
solution.
WKB Idea

So far:  R 1/2
 ±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
 R

We can ignore arbitrary constants in the integration. Any constant

would simply change the arbitrary constants multiplying this
solution.
 R 1/2   R 1/2   R 1/2 
y = AeλY0 = A eiλ G0 dx+C = A eiλ G0 dx eC = A∗ eiλ G0 dx .
WKB Idea

So far:  R 1/2
 ±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
 R

We can ignore arbitrary constants in the integration. Any constant

would simply change the arbitrary constants multiplying this
solution.
 R 1/2   R 1/2   R 1/2 
y = AeλY0 = A eiλ G0 dx+C = A eiλ G0 dx eC = A∗ eiλ G0 dx .

We can include a constant in the limit of the integration without

altering the final solution and take
Rx 1/2
y = ei a G0 du
.

Here a is the location of one of the boundary conditions.

WKB Idea
Continue with the expansion:
   
1 00 1 00 0 2 0 0
Y0 + Y1 + · · · + (Y0 )2 + Y0 Y1 + · · · + G0 = 0.
λ λ λ
WKB Idea
Continue with the expansion:
   
1 00 1 00 0 2 0 0
Y0 + Y1 + · · · + (Y0 )2 + Y0 Y1 + · · · + G0 = 0.
λ λ λ
Now consider terms in λ−1 .
WKB Idea
Continue with the expansion:
   
1 00 1 00 0 2 0 0
Y0 + Y1 + · · · + (Y0 )2 + Y0 Y1 + · · · + G0 = 0.
λ λ λ
Now consider terms in λ−1 .
Y000
Y000 + 2Y00 Y10 = 0 ⇒ Y10 = − .
2Y00
WKB Idea
Continue with the expansion:
   
1 00 1 00 0 2 0 0
Y0 + Y1 + · · · + (Y0 )2 + Y0 Y1 + · · · + G0 = 0.
λ λ λ
Now consider terms in λ−1 .
Y000
Y000 + 2Y00 Y10 = 0 ⇒ Y10 = − .
2Y00
Right hand side is an exact derivative and so
1  
Y1 = − log Y00 = log (Y00 )−1/2 .


2
WKB Idea
Continue with the expansion:
   
1 00 1 00 0 2 0 0
Y0 + Y1 + · · · + (Y0 )2 + Y0 Y1 + · · · + G0 = 0.
λ λ λ
Now consider terms in λ−1 .
Y000
Y000 + 2Y00 Y10 = 0 ⇒ Y10 = − .
2Y00
Right hand side is an exact derivative and so
1  
Y1 = − log Y00 = log (Y00 )−1/2 .


2
Substituting for the expression for Y00 in terms of G0 gives

 log G0 (x)−1/4 ,


G0 > 0,
Y1 =
 log (−G (x))−1/4 , G < 0.


0 0
WKB Idea
Continue with the expansion:
   
1 00 1 00 0 2 0 0
Y0 + Y1 + · · · + (Y0 )2 + Y0 Y1 + · · · + G0 = 0.
λ λ λ
Now consider terms in λ−1 .
Y000
Y000 + 2Y00 Y10 = 0 ⇒ Y10 = − .
2Y00
Right hand side is an exact derivative and so
1  
Y1 = − log Y00 = log (Y00 )−1/2 .


2
Substituting for the expression for Y00 in terms of G0 gives

 log G0 (x)−1/4 ,


G0 > 0,
Y1 = (13)
 log (−G (x))−1/4 , G < 0.


0 0

Any constants can be absorbed into the arbitrary constants.

WKB Idea

So far:

y 00 + λ2 G0 y = 0, λ → ∞. y = eλY (x;λ) = eλY0 +Y1 +··· .

( R 1/2
±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
R

 log G0 (x)−1/4 ,


G0 > 0,
Y1 =
 log (−G (x))−1/4 , G < 0.


0 0
WKB Idea

So far:

y 00 + λ2 G0 y = 0, λ → ∞. y = eλY (x;λ) = eλY0 +Y1 +··· .

( R 1/2
±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
R

 log G0 (x)−1/4 ,


G0 > 0,
Y1 =
 log (−G (x))−1/4 , G < 0.


0 0

y = eλY
WKB Idea

So far:

y 00 + λ2 G0 y = 0, λ → ∞. y = eλY (x;λ) = eλY0 +Y1 +··· .

( R 1/2
±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
R

 log G0 (x)−1/4 ,


G0 > 0,
Y1 =
 log (−G (x))−1/4 , G < 0.


0 0

y = eλY = eλ(Y0 + λ Y1 +O( λ2 ))

1 1
WKB Idea

So far:

y 00 + λ2 G0 y = 0, λ → ∞. y = eλY (x;λ) = eλY0 +Y1 +··· .

( R 1/2
±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
R

 log G0 (x)−1/4 ,


G0 > 0,
Y1 =
 log (−G (x))−1/4 , G < 0.


0 0

y = eλY = eλ(Y0 + λ Y1 +O( λ2 )) = eλY0 eY1 eO(1/λ)

1 1
WKB Idea

So far:

y 00 + λ2 G0 y = 0, λ → ∞. y = eλY (x;λ) = eλY0 +Y1 +··· .

( R 1/2
±i G0 dx, G0 > 0,
Y0 =
± (−G0 )1/2 dx G0 < 0.
R

 log G0 (x)−1/4 ,


G0 > 0,
Y1 =
 log (−G (x))−1/4 , G < 0.


0 0

  
1
y = eλY = eλ(Y0 + λ Y1 +O( λ2 )) = eλY0 eY1 eO(1/λ) = eλY0 eY1
1 1
1+O ,
λ

using the Taylor expansion eO(1/λ) = 1 + O(1/λ).

The WKB Approximation
For the case G0 > 0,
The WKB Approximation
For the case G0 > 0,
 Z   Z   
A 1/2 B 1/2 1
y= 1/4
exp iλ G0 dx + 1/4 exp −iλ G0 dx + O .
G0 G0 λ
The WKB Approximation
For the case G0 > 0,
 Z   Z   
A 1/2 B 1/2 1
y= 1/4
exp iλ G0 dx + 1/4 exp −iλ G0 dx + O .
G0 G0 λ

This can be expressed in terms of sines and cosines as

 Z   Z 
A 1/2 B 1/2
y= 1/4
cos λ G0 dx + 1/4 sin λ G0 dx + O(1/λ).
G0 G0
The WKB Approximation
For the case G0 > 0,
 Z   Z   
A 1/2 B 1/2 1
y= 1/4
exp iλ G0 dx + 1/4 exp −iλ G0 dx + O .
G0 G0 λ

This can be expressed in terms of sines and cosines as

 Z   Z 
A 1/2 B 1/2
y= 1/4
cos λ G0 dx + 1/4 sin λ G0 dx + O(1/λ).
G0 G0

For G0 < 0 the trigonometric functions are instead growing and

decaying exponentials of the form
 Z 
C 1/2
y = exp λ (−G0 ) dx
(−G0 )1/4
 Z   
D 1/2 1
+ exp −λ (−G0 ) dx + O .
(−G0 )1/4 λ
The WKB Approximation
For the case G0 > 0,
 Z   Z   
A 1/2 B 1/2 1
y= 1/4
exp iλ G0 dx + 1/4 exp −iλ G0 dx + O .
G0 G0 λ

This can be expressed in terms of sines and cosines as

 Z   Z 
A 1/2 B 1/2
y= 1/4
cos λ G0 dx + 1/4 sin λ G0 dx + O(1/λ).
G0 G0

For G0 < 0 the trigonometric functions are instead growing and

decaying exponentials of the form
 Z 
C 1/2
y = exp λ (−G0 ) dx
(−G0 )1/4
 Z   
D 1/2 1
+ exp −λ (−G0 ) dx + O .
(−G0 )1/4 λ

These solutions are called the WKB approximations.

The WKB Approximation
For the case G0 > 0,
 Z   Z   
A 1/2 B 1/2 1
y= 1/4
exp iλ G0 dx + 1/4 exp −iλ G0 dx + O .
G0 G0 λ

This can be expressed in terms of sines and cosines as

 Z   Z 
A 1/2 B 1/2
y= 1/4
cos λ G0 dx + 1/4 sin λ G0 dx + O(1/λ).
G0 G0

For G0 < 0 the trigonometric functions are instead growing and

decaying exponentials of the form
 Z 
C 1/2
y = exp λ (−G0 ) dx
(−G0 )1/4
 Z   
D 1/2 1
+ exp −λ (−G0 ) dx + O .
(−G0 )1/4 λ

These solutions are called the WKB approximations.

Comments on WKB Method: I

I The WKB idea can be extended to equations of the form

y 00 + λ2 G0 (x; λ)y = 0,

as long as G0 can be expanded in inverse powers of λ.

Comments on WKB Method: I

I The WKB idea can be extended to equations of the form

y 00 + λ2 G0 (x; λ)y = 0,

as long as G0 can be expanded in inverse powers of λ.

I The WKB method fails near a zero of G0 (x) due to the
(G0 (x))−1/4 factor in the approximation. Such a point is
called a transition point or turning point.
Comments on WKB Method: I

I The WKB idea can be extended to equations of the form

y 00 + λ2 G0 (x; λ)y = 0,

as long as G0 can be expanded in inverse powers of λ.

I The WKB method fails near a zero of G0 (x) due to the
(G0 (x))−1/4 factor in the approximation. Such a point is
called a transition point or turning point.
I We can use matched expansions to connect, via a boundary
layer, the regions where G0 (x) > 0 and G0 (x) < 0. Solutions
are oscillatory for G0 (x) > 0 and exponentially growing and
decaying for G0 (x) < 0. There is a change in character of the
solution at a transition point.
Comments on WKB Method: II

I Any constants of integration appearing in Y0 , Y1 etc. can be

absorbed into the arbitrary constants in the General Solution
and so we can ignore them.
Comments on WKB Method: II

I Any constants of integration appearing in Y0 , Y1 etc. can be

absorbed into the arbitrary constants in the General Solution
and so we can ignore them.
I The WKB procedure can be extended to higher order
equations. The WKB solutions must produce the same
number of independent solutions as the order of the equation.
Comments on WKB Method: II

I Any constants of integration appearing in Y0 , Y1 etc. can be

absorbed into the arbitrary constants in the General Solution
and so we can ignore them.
I The WKB procedure can be extended to higher order
equations. The WKB solutions must produce the same
number of independent solutions as the order of the equation.
I It is sometimes useful to write the General Solution in terms
R x 1/2
of definite integrals of the form a G0 dx.
Example 1
Find the WKB expansion for the boundary value problem

y 00 + λ2 x4 y = 0, y(1) = 1, y(2) = 1.
Example 1
Find the WKB expansion for the boundary value problem

y 00 + λ2 x4 y = 0, y(1) = 1, y(2) = 1.

Since G0 (x) = x4 > 0, we will have oscillatory solutions.

Example 1
Find the WKB expansion for the boundary value problem

y 00 + λ2 x4 y = 0, y(1) = 1, y(2) = 1.

Since G0 (x) = x4 > 0, we will have oscillatory solutions.

x3
Z Z
1
G0 (x)1/2 dx = x2 dx = , G0 (x)−1/4 = ,
3 x
Example 1
Find the WKB expansion for the boundary value problem

y 00 + λ2 x4 y = 0, y(1) = 1, y(2) = 1.

Since G0 (x) = x4 > 0, we will have oscillatory solutions.

x3
Z Z
1
G0 (x)1/2 dx = x2 dx = , G0 (x)−1/4 = ,
3 x
so the WKB expansion is

λx3 λx3
   
A B
y = cos + sin .
x 3 x 3
Example 1
Find the WKB expansion for the boundary value problem

y 00 + λ2 x4 y = 0, y(1) = 1, y(2) = 1.

Since G0 (x) = x4 > 0, we will have oscillatory solutions.

x3
Z Z
1
G0 (x)1/2 dx = x2 dx = , G0 (x)−1/4 = ,
3 x
so the WKB expansion is

λx3 λx3
   
A B
y = cos + sin .
x 3 x 3

Applying the boundary conditions gives the simultaneous equations

1 = A cos(λ/3) + B sin(λ/3),
A B
1 = cos(8λ/3) + sin(8λ/3).
2 2
Example 1

1 = A cos(λ/3) + B sin(λ/3),
A B
1 = cos(8λ/3) + sin(8λ/3).
2 2
Solving for A and B gives

sin(8λ/3) − 2 sin(λ/3) 2 cos(λ/3) − cos(8λ/3)

A= , B= .
sin(7λ/3) sin(7λ/3)
Example 1

Figure: Comparison between numerical solution and WKB approximation.

Example 2

Find the WKB expansion for the general solution to

y 00 − λ2 x4 + λx3 y = 0;


x > 0.
Example 2

Find the WKB expansion for the general solution to

y 00 − λ2 x4 + λx3 y = 0;


x > 0.

There is an extra term, λx3 , and so the simple WKB solution must
be modified. But follow same strategy and put

y = eλY = eλY0 (x)+Y1 (x)+··· .

Example 2

y 00 − λ2 x4 + λx3 y = 0, y = eλY = eλY0 (x)+Y1 (x)+··· .

Example 2

y 00 − λ2 x4 + λx3 y = 0, y = eλY = eλY0 (x)+Y1 (x)+··· .

The differential equation becomes

 
λ2 (Y 0 )2 eλY + λY 00 eλY − (λ2 x4 + λx3 )eλY = 0
Example 2

y 00 − λ2 x4 + λx3 y = 0, y = eλY = eλY0 (x)+Y1 (x)+··· .

The differential equation becomes

 
λ2 (Y 0 )2 eλY + λY 00 eλY − (λ2 x4 + λx3 )eλY = 0

which is  
1 00 0 2 4 1 3
Y + (Y ) − x + x = 0.
λ λ
Example 2

y 00 − λ2 x4 + λx3 y = 0, y = eλY = eλY0 (x)+Y1 (x)+··· .

The differential equation becomes

 
λ2 (Y 0 )2 eλY + λY 00 eλY − (λ2 x4 + λx3 )eλY = 0

which is  
1 00 0 2 4 1 3
Y + (Y ) − x + x = 0.
λ λ

Expanding gives
   
1 00 1 00 0 2 2 0 0 1
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ
Example 2

y 00 − λ2 x4 + λx3 y = 0, y = eλY = eλY0 (x)+Y1 (x)+··· .

The differential equation becomes

 
λ2 (Y 0 )2 eλY + λY 00 eλY − (λ2 x4 + λx3 )eλY = 0

which is  
1 00 0 2 4 1 3
Y + (Y ) − x + x = 0.
λ λ

Expanding gives
   
1 00 1 00 0 2 2 0 0 1
Y0 + Y1 + · · · + (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

Now we can compare orders.

Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 ,
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives

Y00 = ±x2 ,
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives

x3
Y00 = ±x2 , Y0 = ± .
3
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives

x3
Y00 = ±x2 , Y0 = ± .
3
Substitute this solution into the equation for Y10 gives
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives

x3
Y00 = ±x2 , Y0 = ± .
3
Substitute this solution into the equation for Y10 gives

±2x ± 2x2 Y10 − x3 = 0,

Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives

x3
Y00 = ±x2 , Y0 = ± .
3
Substitute this solution into the equation for Y10 gives

x 1
±2x ± 2x2 Y10 − x3 = 0, ⇒ Y10 = ± −
2 x
Example 2

   
1 1 2 0 0 1
Y000 + Y100 + · · · 0 2
+ (Y0 ) + Y0 Y1 + · · · − x4 − x3 = 0.
λ λ λ λ

2
λ0 : Y00 = x4 , λ−1 : Y000 + 2Y00 Y10 − x3 = 0.
Solving for Y00 gives

x3
Y00 = ±x2 , Y0 = ± .
3
Substitute this solution into the equation for Y10 gives

x 1 x2
±2x ± 2x2 Y10 − x3 = 0, ⇒ Y10 = ± − ⇒ Y1 = ± − ln x
2 x 4
Example 2

Take
x3 x2
Y0 = ± , Y1 = ± − ln x
3 4
and substitute into

y = eλY = eλY0 (x)+Y1 (x)+··· .

Example 2

Take
x3 x2
Y0 = ± , Y1 = ± − ln x
3 4
and substitute into

y = eλY = eλY0 (x)+Y1 (x)+··· .

The general WKB solution is

 
A 3 2 B 3 2 1
y = e+λx /3+x /4 + e−λx /3−x /4 + O .
x x λ
Example 2

Compare WKB and numerical solutions for

y 00 − λ2 x4 + λx3 y = 0,


y(1) = 1, y(2) = 0
Eigenvalue problems

The WKB method can be used to obtain approximate values for

eigenvalues in eigenvalue problems.
Eigenvalue problems

The WKB method can be used to obtain approximate values for

eigenvalues in eigenvalue problems.

Suppose we have the boundary value problem

y 00 + λ2 G0 (x)y = 0, 0 < x < 1, (14)

with
y(0) = y(1) = 0,
where G0 (x) > 0 (so that we have oscillatory solutions).
Eigenvalue problems

The WKB method can be used to obtain approximate values for

eigenvalues in eigenvalue problems.

Suppose we have the boundary value problem

y 00 + λ2 G0 (x)y = 0, 0 < x < 1, (14)

with
y(0) = y(1) = 0,
where G0 (x) > 0 (so that we have oscillatory solutions).

Non-trivial solutions only exist for particular values of λ, called

eigenvalues. This is called an eigenvalue problem, the associated
values of λ the eigenvalues and the solutions, y the eigenfunctions.
Example
The corresponding problem when G0 = 1 is

y 00 + λ2 y = 0, ⇒ y = A sin λx + B cos λx.

Example
The corresponding problem when G0 = 1 is

y 00 + λ2 y = 0, ⇒ y = A sin λx + B cos λx.

Boundary conditions are y(0) = y(1) = 0 so at x = 0 we need

0=A×0+B×1 ⇒ B = 0,
Example
The corresponding problem when G0 = 1 is

y 00 + λ2 y = 0, ⇒ y = A sin λx + B cos λx.

Boundary conditions are y(0) = y(1) = 0 so at x = 0 we need

0=A×0+B×1 ⇒ B = 0,

and at x = 1

0 = A sin λ = 0 ⇒ λ = nπ,

where n is an integer.
Example
The corresponding problem when G0 = 1 is

y 00 + λ2 y = 0, ⇒ y = A sin λx + B cos λx.

Boundary conditions are y(0) = y(1) = 0 so at x = 0 we need

0=A×0+B×1 ⇒ B = 0,

and at x = 1

0 = A sin λ = 0 ⇒ λ = nπ,

where n is an integer. Thus, the eigenvalues are

λ = nπ

and the eigenfunctions are

yn = A sin(nπx). A is an arbitrary constant.

Eigenvalue problems
In general

y 00 + λ2 G0 (x)y = 0, 0 < x < 1, y(0) = y(1) = 0

has an infinite number of eigenvalues, {λn }, called a spectrum,

with λn → ∞ as n → ∞.
Eigenvalue problems
In general

y 00 + λ2 G0 (x)y = 0, 0 < x < 1, y(0) = y(1) = 0

has an infinite number of eigenvalues, {λn }, called a spectrum,

with λn → ∞ as n → ∞.

We can use the WKB method to approximate λn as n → ∞.

Eigenvalue problems
In general

y 00 + λ2 G0 (x)y = 0, 0 < x < 1, y(0) = y(1) = 0

has an infinite number of eigenvalues, {λn }, called a spectrum,

with λn → ∞ as n → ∞.

We can use the WKB method to approximate λn as n → ∞.

For λ  1 and G0 (x) > 0, we have

  Z x   Z x 
1 1/2 1/2
y = 1/4 A cos λ G0 dx + B sin λ G0 dx .
G0 0 0

Error is O(1/λ).
Eigenvalue problems
In general

y 00 + λ2 G0 (x)y = 0, 0 < x < 1, y(0) = y(1) = 0

has an infinite number of eigenvalues, {λn }, called a spectrum,

with λn → ∞ as n → ∞.

We can use the WKB method to approximate λn as n → ∞.

For λ  1 and G0 (x) > 0, we have

  Z x   Z x 
1 1/2 1/2
y = 1/4 A cos λ G0 dx + B sin λ G0 dx .
G0 0 0

Error is O(1/λ).
Notice that we use the lower limit in the integral because one of
the boundary conditions is applied at x = 0.
Eigenvalue problems

y 00 + λ2 G0 (x)y = 0, y(0) = y(1) = 0

  Z x   Z x 
1 1/2 1/2
y= 1/4
A cos λ G0 dx + B sin λ G0 dx .
G0 0 0
Eigenvalue problems

y 00 + λ2 G0 (x)y = 0, y(0) = y(1) = 0

  Z x   Z x 
1 1/2 1/2
y= 1/4
A cos λ G0 dx + B sin λ G0 dx .
G0 0 0

Apply the boundary conditions: First

y(0) = 0 = A.
Eigenvalue problems

y 00 + λ2 G0 (x)y = 0, y(0) = y(1) = 0

  Z x   Z x 
1 1/2 1/2
y= 1/4
A cos λ G0 dx + B sin λ G0 dx .
G0 0 0

Apply the boundary conditions: First

y(0) = 0 = A.

Now at x = 1, with A = 0 we have

 Z 1 
B 1/2
y(1) = 0 = sin λ G0 (x) dx .
G0 (1)1/4 0
Eigenvalue problems

y 00 + λ2 G0 (x)y = 0, y(0) = y(1) = 0

  Z x   Z x 
1 1/2 1/2
y= 1/4
A cos λ G0 dx + B sin λ G0 dx .
G0 0 0

Apply the boundary conditions: First

y(0) = 0 = A.

Now at x = 1, with A = 0 we have

 Z 1 
B 1/2
y(1) = 0 = sin λ G0 (x) dx .
G0 (1)1/4 0

If B 6= 0, then we must have

Z 1
λ G0 (x)1/2 dx = nπ.
0
 Z 1
λ G0 (x)1/2 dx = nπ.
0
and so the eigenvalues are approximated by
 Z 1
λ G0 (x)1/2 dx = nπ.
0
and so the eigenvalues are approximated by
nπ
λn = R 1 .
0 G0 (x)1/2 dx
 Z 1
λ G0 (x)1/2 dx = nπ.
0
and so the eigenvalues are approximated by
nπ
λn = R 1 .
0 G0 (x)1/2 dx

The eigenfunctions are

Rx !
B nπ 0 G0 (u)1/2 du
yn = sin R1 .
G0 (x)1/4 1/2 dx
0 G0 (x)
Eigenvalue Problems: Example

Consider the eigenvalue problem

y 00 + λ2 (1 + x)2 y = 0, y(0) = y(1) = 0.

Eigenvalue Problems: Example

Consider the eigenvalue problem

y 00 + λ2 (1 + x)2 y = 0, y(0) = y(1) = 0.

G0 (x) = (1 + x)2 > 0. So the WKB solutions are oscillatory.

Eigenvalue Problems: Example

Consider the eigenvalue problem

y 00 + λ2 (1 + x)2 y = 0, y(0) = y(1) = 0.

G0 (x) = (1 + x)2 > 0. So the WKB solutions are oscillatory.

Thus, we need the following integral
Z x Z x
1/2 x2
G0 (u) du = (1 + u) du = x + .
0 0 2
Eigenvalue Problems: Example

Consider the eigenvalue problem

y 00 + λ2 (1 + x)2 y = 0, y(0) = y(1) = 0.

G0 (x) = (1 + x)2 > 0. So the WKB solutions are oscillatory.

Thus, we need the following integral
Z x Z x
1/2 x2
G0 (u) du = (1 + u) du = x + .
0 0 2

Hence, the WKB solution is

x2 x2
     
A B
y=√ cos λ x + +√ sin λ x + .
1+x 2 1+x 2
Eigenvalue Problems: Example

To satisfy the boundary condition at x = 0 we require A = 0

Eigenvalue Problems: Example

To satisfy the boundary condition at x = 0 we require A = 0 and

to satisfy the boundary condition at x = 1 we need
nπ 2nπ
λn = = .
1 + 1/2 3
Eigenvalue Problems: Example

To satisfy the boundary condition at x = 0 we require A = 0 and

to satisfy the boundary condition at x = 1 we need
nπ 2nπ
λn = = .
1 + 1/2 3

The eigenfunctions are

x2
  
Bn 2nπ
yn = √ sin x+ ,
1+x 3 2

where Bn is an arbitrary constant.

Eigenvalue Problems: Example
For this problem, can solve the differential equation in terms of
Bessel functions of order 1/4.
Eigenvalue Problems: Example
For this problem, can solve the differential equation in terms of
Bessel functions of order 1/4. Eigenvalues are obtained from a
transcendental equation and must be obtained numerically.
Eigenvalue Problems: Example
For this problem, can solve the differential equation in terms of
Bessel functions of order 1/4. Eigenvalues are obtained from a
transcendental equation and must be obtained numerically. A
comparison between the WKB estimates, λn , and exact values are
given in the table below.

Table: Comparison between the exact eigenvalues and the WKB

estimates.

n 1 2 3 4 5
λn (exact) 2.06035 4.16863 6.26905 8.36675 10.46322
λn (WKB) 2.09440 4.18879 6.28319 8.37758 10.47198
% abs. err. 3.404 2.016 1.413 1.083 0.876
Eigenvalue Problems: Example
For this problem, can solve the differential equation in terms of
Bessel functions of order 1/4. Eigenvalues are obtained from a
transcendental equation and must be obtained numerically. A
comparison between the WKB estimates, λn , and exact values are
given in the table below.

Table: Comparison between the exact eigenvalues and the WKB

estimates.

n 1 2 3 4 5
λn (exact) 2.06035 4.16863 6.26905 8.36675 10.46322
λn (WKB) 2.09440 4.18879 6.28319 8.37758 10.47198
% abs. err. 3.404 2.016 1.413 1.083 0.876

Note the approximations are more accurate for larger values of n,

(λ is larger).
Transition Points

WKB solutions break down at points where G0 (x) = 0, known as

transition points or turning points. Solution changes from
oscillatory when G0 > 0, to exponential when G0 < 0. The
approximate solutions to the ODE must somehow reflect this.
Transition Points

WKB solutions break down at points where G0 (x) = 0, known as

transition points or turning points. Solution changes from
oscillatory when G0 > 0, to exponential when G0 < 0. The
approximate solutions to the ODE must somehow reflect this.

Basic WKB solutions do not show this change.

To progress, we consider the simplest equation that has this

transitory behaviour.
Transition Points

WKB solutions break down at points where G0 (x) = 0, known as

transition points or turning points. Solution changes from
oscillatory when G0 > 0, to exponential when G0 < 0. The
approximate solutions to the ODE must somehow reflect this.

Basic WKB solutions do not show this change.

To progress, we consider the simplest equation that has this

transitory behaviour.

Consider the equation

y 00 − xy = 0. (15)
Airy’s Equation

y 00 − xy = 0.
This equation is known as Airy’s Equation and the solutions to it
are called Airy functions.
Airy’s Equation

y 00 − xy = 0.
This equation is known as Airy’s Equation and the solutions to it
are called Airy functions. There are two linearly independent
solutions, labelled by Ai(x) and Bi(x).
Airy’s Equation

y 00 − xy = 0.
This equation is known as Airy’s Equation and the solutions to it
are called Airy functions. There are two linearly independent
solutions, labelled by Ai(x) and Bi(x).

For x > 0, the solutions are exponential in character and behave as

1 3/2 /3 1 3/2 /3
y∼ e2x , and y ∼ e−2x .
x1/4 x1/4
Airy’s Equation

y 00 − xy = 0.
This equation is known as Airy’s Equation and the solutions to it
are called Airy functions. There are two linearly independent
solutions, labelled by Ai(x) and Bi(x).

For x > 0, the solutions are exponential in character and behave as

1 3/2 /3 1 3/2 /3
y∼ e2x , and y ∼ e−2x .
x1/4 x1/4

For x < 0 the solutions are oscillatory and behave as

   
1 2 3/2 1 2 3/2
y∼ cos (−x) , and y ∼ sin (−x) .
(−x)1/4 3 (−x)1/4 3
Airy’s Equation

1 2x3/2 /3 1 3/2
y∼ e , and y ∼ 1/4 e−2x /3 .
x1/4 x
   
1 2 3/2 1 2 3/2
y∼ cos (−x) , and y ∼ sin (−x) .
(−x)1/4 3 (−x)1/4 3
Airy’s Equation

1 2x3/2 /3 1 3/2
y∼ e , and y ∼ 1/4 e−2x /3 .
x1/4 x
   
1 2 3/2 1 2 3/2
y∼ cos (−x) , and y ∼ sin (−x) .
(−x)1/4 3 (−x)1/4 3
The key points to note here are that
1. Ai(x) is exponentially decaying for x > 0 and Bi(x) is
exponentially increasing.
Airy’s Equation

1 2x3/2 /3 1 3/2
y∼ e , and y ∼ 1/4 e−2x /3 .
x1/4 x
   
1 2 3/2 1 2 3/2
y∼ cos (−x) , and y ∼ sin (−x) .
(−x)1/4 3 (−x)1/4 3
The key points to note here are that
1. Ai(x) is exponentially decaying for x > 0 and Bi(x) is
exponentially increasing.
2. For x < 0 they are both oscillatory but out of phase with each
other.
Airy’s Equation

1 2x3/2 /3 1 3/2
y∼ e , and y ∼ 1/4 e−2x /3 .
x1/4 x
   
1 2 3/2 1 2 3/2
y∼ cos (−x) , and y ∼ sin (−x) .
(−x)1/4 3 (−x)1/4 3
The key points to note here are that
1. Ai(x) is exponentially decaying for x > 0 and Bi(x) is
exponentially increasing.
2. For x < 0 they are both oscillatory but out of phase with each
other.
3. The Airy functions can be expressed in terms of integrals and
the asymptotic behaviour of the integrals can be determined
by the methods of Laplace and Stationary Phase.
The question is how do these exponential solutions transform into
the oscillatory solutions as x passes from positive values, through
the transition point, to negative values.
The question is how do these exponential solutions transform into
the oscillatory solutions as x passes from positive values, through
the transition point, to negative values.

The limits of the Airy functions, for large positive and negative x,
are related to the WKB solutions.
Figure: Airy function, Ai(x), showing transition from oscillatory to
exponential decay.
Figure: Airy function, Bi(x), showing transition from oscillatory to
exponential growth.
Summary of Airy Functions
1 3/2
x→∞ Ai(x) ∼ √ 1/4 e−2x /3
2 πx
1 3/2
Bi(x) ∼ √ 1/4 e2x /3
πx
Summary of Airy Functions
1 3/2
x→∞ Ai(x) ∼ √ 1/4 e−2x /3
2 πx
1 3/2
Bi(x) ∼ √ 1/4 e2x /3
πx
1
x → −∞ Ai(x) ∼ √ sin (2(−x)3/2 /3 + π/4)
π(−x)1/4
1
Bi(x) ∼ √ cos (2(−x)3/2 /3 + π/4)
π(−x)1/4
Summary of Airy Functions
1 3/2
x→∞ Ai(x) ∼ √ 1/4 e−2x /3
2 πx
1 3/2
Bi(x) ∼ √ 1/4 e2x /3
πx
1
x → −∞ Ai(x) ∼ √ sin (2(−x)3/2 /3 + π/4)
π(−x)1/4
1
Bi(x) ∼ √ cos (2(−x)3/2 /3 + π/4)
π(−x)1/4
These expressions are valid away from x = 0.
Summary of Airy Functions
1 3/2
x→∞ Ai(x) ∼ √ 1/4 e−2x /3
2 πx
1 3/2
Bi(x) ∼ √ 1/4 e2x /3
πx
1
x → −∞ Ai(x) ∼ √ sin (2(−x)3/2 /3 + π/4)
π(−x)1/4
1
Bi(x) ∼ √ cos (2(−x)3/2 /3 + π/4)
π(−x)1/4
These expressions are valid away from x = 0. At x = 0 the Airy
functions have the specific values given by
1 1
Ai(0) = ≈ 0.355, Bi(0) = ≈ 0.6149.
32/3 Γ(2/3) 31/6 Γ(2/3)
(16)
where Z ∞
Γ(x + 1) = e−t tx dx
0
Example

Consider the solutions to the differential equation

y 00 + λ2 (1 − x3 )y = 0, (17)

for λ → ∞.
Example

Consider the solutions to the differential equation

y 00 + λ2 (1 − x3 )y = 0, (17)

for λ → ∞.

We will obtain WKB solutions that are valid for x > 1 and x < 1.
Example

Consider the solutions to the differential equation

y 00 + λ2 (1 − x3 )y = 0, (17)

for λ → ∞.

We will obtain WKB solutions that are valid for x > 1 and x < 1.

Then, match the two solutions using the Method of Matched

Asymptotics (involving the use of Airy functions as the inner
solution).
Example

Consider the solutions to the differential equation

y 00 + λ2 (1 − x3 )y = 0, (17)

for λ → ∞.

We will obtain WKB solutions that are valid for x > 1 and x < 1.

Then, match the two solutions using the Method of Matched

Asymptotics (involving the use of Airy functions as the inner
solution).

Can compare this with internal boundary layers in Matched

Asymptotic Expansions.
Outer Solutions

The outer solutions are oscillatory for x < 1 (denoted as y l ) and

exponential for x > 1 (denoted as y r ).
Outer Solutions

The outer solutions are oscillatory for x < 1 (denoted as y l ) and

exponential for x > 1 (denoted as y r ). Hence, for x < 1,
 R √   R √ 
1 1
c1 cos λ x 1 − v 3 dv + c2 sin λ x 1 − v 3 dv
yl ∼ . (18)
(1 − x3 )1/4
Outer Solutions

The outer solutions are oscillatory for x < 1 (denoted as y l ) and

exponential for x > 1 (denoted as y r ). Hence, for x < 1,
 R √   R √ 
1 1
c1 cos λ x 1 − v 3 dv + c2 sin λ x 1 − v 3 dv
yl ∼ . (18)
(1 − x3 )1/4

For x > 1,
 R √   Rx√ 
x
a1 exp λ 1 v 3 − 1dv + a2 exp −λ 1 v 3 − 1dv
yr ∼ .
(x3 − 1)1/4
(19)
Outer Solutions

The outer solutions are oscillatory for x < 1 (denoted as y l ) and

exponential for x > 1 (denoted as y r ). Hence, for x < 1,
 R √   R √ 
1 1
c1 cos λ x 1 − v 3 dv + c2 sin λ x 1 − v 3 dv
yl ∼ . (18)
(1 − x3 )1/4

For x > 1,
 R √   Rx√ 
x
a1 exp λ 1 v 3 − 1dv + a2 exp −λ 1 v 3 − 1dv
yr ∼ .
(x3 − 1)1/4
(19)
These solutions are valid away from x = 1.
Inner Solutions

y 00 + λ2 (1 − x3 )y = 0
Inner Solutions

y 00 + λ2 (1 − x3 )y = 0

Near x = 1, we transform from x to ξ by

ξ = (x − 1)λν .

ν must have a definite value.

Inner Solutions

y 00 + λ2 (1 − x3 )y = 0

Near x = 1, we transform from x to ξ by

ξ = (x − 1)λν .

ν must have a definite value. Also we see that x > 1 corresponds

to ξ > 0 and x < 1 to ξ < 0.
Inner Solutions

y 00 + λ2 (1 − x3 )y = 0

Near x = 1, we transform from x to ξ by

ξ = (x − 1)λν .

ν must have a definite value. Also we see that x > 1 corresponds

to ξ > 0 and x < 1 to ξ < 0. Then in terms of the inner solution
the ODE is Y (ξ),

d2 Y
λ2ν + λ2 1 − (1 + λ−ν ξ)3 Y = 0.


dξ 2
 d2 Y −ν 3
λ2ν 2


+ λ 1 − (1 + λ ξ) Y = 0.
dξ 2
 d2 Y −ν 3
λ2ν 2


+ λ 1 − (1 + λ ξ) Y = 0.
dξ 2
Expanding the cubic term

1 − (1 + λ−ν ξ)3 = 1 − 1 − 3λ−ν ξ − 3(λ−ν ξ)2 − (λ−ν ξ)3 .

 d2 Y −ν 3
λ2ν 2


+ λ 1 − (1 + λ ξ) Y = 0.
dξ 2
Expanding the cubic term

1 − (1 + λ−ν ξ)3 = 1 − 1 − 3λ−ν ξ − 3(λ−ν ξ)2 − (λ−ν ξ)3 .

and keeping the leading term (as λ → ∞), gives

d2 Y
− 3λ2−3ν ξY = 0.
dξ 2
 d2 Y −ν 3
λ2ν 2


+ λ 1 − (1 + λ ξ) Y = 0.
dξ 2
Expanding the cubic term

1 − (1 + λ−ν ξ)3 = 1 − 1 − 3λ−ν ξ − 3(λ−ν ξ)2 − (λ−ν ξ)3 .

and keeping the leading term (as λ → ∞), gives

d2 Y
− 3λ2−3ν ξY = 0.
dξ 2
For this we see that the correct value for ν is

ν = 2/3.
 d2 Y −ν 3
λ2ν 2


+ λ 1 − (1 + λ ξ) Y = 0.
dξ 2
Expanding the cubic term

1 − (1 + λ−ν ξ)3 = 1 − 1 − 3λ−ν ξ − 3(λ−ν ξ)2 − (λ−ν ξ)3 .

and keeping the leading term (as λ → ∞), gives

d2 Y
− 3λ2−3ν ξY = 0.
dξ 2
For this we see that the correct value for ν is

ν = 2/3.

The equation is now

d2 Y
− 3ξY = 0.
dξ 2
d2 Y
− 3ξY = 0.
dξ 2
 d2 Y
− 3ξY = 0.
dξ 2
Finally, we can remove the factor of 3 and reduce the equation to
Airy’s equation by a final change of variable.
 d2 Y
− 3ξY = 0.
dξ 2
Finally, we can remove the factor of 3 and reduce the equation to
Airy’s equation by a final change of variable. Set ξ = 3−1/3 X, so
that
x = 1 + 3−1/3 λ−2/3 X,
 d2 Y
− 3ξY = 0.
dξ 2
Finally, we can remove the factor of 3 and reduce the equation to
Airy’s equation by a final change of variable. Set ξ = 3−1/3 X, so
that
x = 1 + 3−1/3 λ−2/3 X,
and
d2 Y
− XY = 0.
dX 2
 d2 Y
− 3ξY = 0.
dξ 2
Finally, we can remove the factor of 3 and reduce the equation to
Airy’s equation by a final change of variable. Set ξ = 3−1/3 X, so
that
x = 1 + 3−1/3 λ−2/3 X,
and
d2 Y
− XY = 0.
dX 2
The inner solution is

Y = b1 Ai(X) + b2 Bi(X). (20)

 d2 Y
− 3ξY = 0.
dξ 2
Finally, we can remove the factor of 3 and reduce the equation to
Airy’s equation by a final change of variable. Set ξ = 3−1/3 X, so
that
x = 1 + 3−1/3 λ−2/3 X,
and
d2 Y
− XY = 0.
dX 2
The inner solution is

Y = b1 Ai(X) + b2 Bi(X). (20)

Now match the inner solutions to the outer solutions.

 d2 Y
− 3ξY = 0.
dξ 2
Finally, we can remove the factor of 3 and reduce the equation to
Airy’s equation by a final change of variable. Set ξ = 3−1/3 X, so
that
x = 1 + 3−1/3 λ−2/3 X,
and
d2 Y
− XY = 0.
dX 2
The inner solution is

Y = b1 Ai(X) + b2 Bi(X). (20)

Now match the inner solutions to the outer solutions. We use the
large X behaviour of the Airy functions in terms of x and match
to the outer solutions.
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 X 1/2 + O(λ−2/3 ).
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 X 1/2 + O(λ−2/3 ).

Use dx = 3−1/3 λ−2/3 dX,

Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 X 1/2 + O(λ−2/3 ).

Use dx = 3−1/3 λ−2/3 dX, so the integral in the WKB outer

solution for x > 1 is
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 X 1/2 + O(λ−2/3 ).

Use dx = 3−1/3 λ−2/3 dX, so the integral in the WKB outer

solution for x > 1 is
Z xp Z X
2
3
x − 1dx ∼ (31/3 λ−1/3 X 1/2 ) (3−1/3 λ−2/3 )dX = λ−1 X 3/2 .
1 0 3
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 X 1/2 + O(λ−2/3 ).

Use dx = 3−1/3 λ−2/3 dX, so the integral in the WKB outer

solution for x > 1 is
Z xp Z X
2
3
x − 1dx ∼ (31/3 λ−1/3 X 1/2 ) (3−1/3 λ−2/3 )dX = λ−1 X 3/2 .
1 0 3

Then the WKB outer solution for x > 1, expressed in terms of the
inner variable, is
Asymptotic Matching
The outer solutions are expressed in terms of the inner variable, X,
and expanded for large λ.
Consider x = 1 + 3−1/3 λ−2/3 X > 1,
p q q
x3 − 1 = (1 + 3−1/3 λ−2/3 X)3 − 1 = 32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 X 1/2 + O(λ−2/3 ).

Use dx = 3−1/3 λ−2/3 dX, so the integral in the WKB outer

solution for x > 1 is
Z xp Z X
2
3
x − 1dx ∼ (31/3 λ−1/3 X 1/2 ) (3−1/3 λ−2/3 )dX = λ−1 X 3/2 .
1 0 3

Then the WKB outer solution for x > 1, expressed in terms of the
inner variable, is

a1 exp 23 X 3/2 + a2 exp − 32 X 3/2

r
y = . (21)
31/6 λ−1/6 X 1/4
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
In this case, we have the oscillatory solutions.
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
In this case, we have the oscillatory solutions. Hence,
p q q
1 − x3 = 1 − (1 + 3−1/3 λ−2/3 X)3 = −32/3 λ−2/3 X + O(λ−4/3 )
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
In this case, we have the oscillatory solutions. Hence,
p q q
1 − x3 = 1 − (1 + 3−1/3 λ−2/3 X)3 = −32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 (−X)1/2 + O(λ−2/3 ).

dx = 3−1/3 λ−2/3 dX
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
In this case, we have the oscillatory solutions. Hence,
p q q
1 − x3 = 1 − (1 + 3−1/3 λ−2/3 X)3 = −32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 (−X)1/2 + O(λ−2/3 ).

dx = 3−1/3 λ−2/3 dX and the integral is

Z 1 Z 0
p 2 −1
1 − x3 dx ∼ (31/3 λ−1/3 (−X)1/2 )(3−1/3 λ−2/3 )dX = λ (−X)3/2 .
x X 3
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
In this case, we have the oscillatory solutions. Hence,
p q q
1 − x3 = 1 − (1 + 3−1/3 λ−2/3 X)3 = −32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 (−X)1/2 + O(λ−2/3 ).

dx = 3−1/3 λ−2/3 dX and the integral is

Z 1 Z 0
p 2 −1
1 − x3 dx ∼ (31/3 λ−1/3 (−X)1/2 )(3−1/3 λ−2/3 )dX = λ (−X)3/2 .
x X 3

Therefore, the outer WKB solution for x < 1, expressed in terms of the
inner variable, is
Asymptotic Matching
Now consider x = 1 + 3−1/3 λ−2/3 X < 1, or in other words X < 0.
In this case, we have the oscillatory solutions. Hence,
p q q
1 − x3 = 1 − (1 + 3−1/3 λ−2/3 X)3 = −32/3 λ−2/3 X + O(λ−4/3 )
= 31/3 λ−1/3 (−X)1/2 + O(λ−2/3 ).

dx = 3−1/3 λ−2/3 dX and the integral is

Z 1 Z 0
p 2 −1
1 − x3 dx ∼ (31/3 λ−1/3 (−X)1/2 )(3−1/3 λ−2/3 )dX = λ (−X)3/2 .
x X 3

Therefore, the outer WKB solution for x < 1, expressed in terms of the
inner variable, is

c1 cos 23 (−X)3/2 + c2 sin 23 (−X)3/2

l
y = (22)
31/6 λ−1/6 (−X)1/4
Finally, we take the large argument expansions of the inner
solutions.
Finally, we take the large argument expansions of the inner
solutions. Assume that b1 and b2 are our two arbitrary constants
for the second order differential equation.
Finally, we take the large argument expansions of the inner
solutions. Assume that b1 and b2 are our two arbitrary constants
for the second order differential equation. Consider first the x > 1
region:
Finally, we take the large argument expansions of the inner
solutions. Assume that b1 and b2 are our two arbitrary constants
for the second order differential equation. Consider first the x > 1
region:
2 3/2 2 3/2



a1 exp X + a2 exp − X
yr = 3 3
31/6 λ−1/6 X 1/4
Finally, we take the large argument expansions of the inner
solutions. Assume that b1 and b2 are our two arbitrary constants
for the second order differential equation. Consider first the x > 1
region:
2 3/2 2 3/2



a1 exp X + a2 exp − X
yr = 3 3
31/6 λ−1/6 X 1/4

As X → ∞ then Y = b1 Ai(X) + b2 Bi(X) is, asymptotically,

b1 2 3/2 b2 2 3/2
Y ∼ √ 1/4
e− 3 X + √ 1/4
e3X
2 πX πX
Finally, we take the large argument expansions of the inner
solutions. Assume that b1 and b2 are our two arbitrary constants
for the second order differential equation. Consider first the x > 1
region:
2 3/2 2 3/2



a1 exp X + a2 exp − X
yr = 3 3
31/6 λ−1/6 X 1/4

As X → ∞ then Y = b1 Ai(X) + b2 Bi(X) is, asymptotically,

b1 2 3/2 b2 2 3/2
Y ∼ √ 1/4
e− 3 X + √ 1/4
e3X
2 πX πX

So
31/6 b2 31/6 b1
a1 = , a2 = .
λ1/6 π 1/2 2λ1/6 π 1/2
For x < 1 region,
For x < 1 region,
2 3/2 + c sin 2 (−X)3/2



l c1 cos 3 (−X) 2 3
y = 1/6 −1/6
3 λ (−X)1/4
For x < 1 region,
2 3/2 + c sin 2 (−X)3/2



l c1 cos 3 (−X) 2 3
y = 1/6 −1/6
3 λ (−X)1/4

As X → −∞ then Y = b1 Ai(X) + b2 Bi(X) is, asymptotically,

For x < 1 region,
2 3/2 + c sin 2 (−X)3/2



l c1 cos 3 (−X) 2 3
y = 1/6 −1/6
3 λ (−X)1/4

As X → −∞ then Y = b1 Ai(X) + b2 Bi(X) is, asymptotically,

2(−X)3/2 2(−X)3/2
   
1 π 1 π
b1 √ sin + +b2 √ cos +
π(−X)1/4 3 4 π(−X)1/4 3 4
For x < 1 region,
2 3/2 + c sin 2 (−X)3/2



l c1 cos 3 (−X) 2 3
y = 1/6 −1/6
3 λ (−X)1/4

As X → −∞ then Y = b1 Ai(X) + b2 Bi(X) is, asymptotically,

2(−X)3/2 2(−X)3/2
   
1 π 1 π
b1 √ sin + +b2 √ cos +
π(−X)1/4 3 4 π(−X)1/4 3 4

Use
 π 1 1  π 1 1
sin A + = √ sin A+ √ cos A, cos A + = √ cos A− √ sin A
4 2 2 4 2 2
For x < 1 region,
2 3/2 + c sin 2 (−X)3/2



l c1 cos 3 (−X) 2 3
y = 1/6 −1/6
3 λ (−X)1/4

As X → −∞ then Y = b1 Ai(X) + b2 Bi(X) is, asymptotically,

2(−X)3/2 2(−X)3/2
   
1 π 1 π
b1 √ sin + +b2 √ cos +
π(−X)1/4 3 4 π(−X)1/4 3 4

Use
 π 1 1  π 1 1
sin A + = √ sin A+ √ cos A, cos A + = √ cos A− √ sin A
4 2 2 4 2 2

Then
31/6 (b1 + b2 ) 31/6 (b1 − b2 )
c1 = , c2 = .
21/2 π 1/2 λ1/6 21/2 π 1/2 λ1/6
Now the constants for the x < 1 solution, yl , (c1 and c2 ) and the
constants for the x > 1 solution, yr , (a1 and a2 ) are known in
terms of the constants for the inner solution (b1 and b2 ):
Now the constants for the x < 1 solution, yl , (c1 and c2 ) and the
constants for the x > 1 solution, yr , (a1 and a2 ) are known in
terms of the constants for the inner solution (b1 and b2 ):

31/6 b2 31/6 b1 31/6 (b1 + b2 ) 31/6 (b1 − b2 )

a1 = 1/6 1/2
, a2 = 1/6 1/2 , c1 = 1/2 1/2 1/6 , c2 = 1/2 1/2 1/6 .
λ π 2λ π 2 π λ 2 π λ
Now the constants for the x < 1 solution, yl , (c1 and c2 ) and the
constants for the x > 1 solution, yr , (a1 and a2 ) are known in
terms of the constants for the inner solution (b1 and b2 ):

31/6 b2 31/6 b1 31/6 (b1 + b2 ) 31/6 (b1 − b2 )

a1 = 1/6 1/2
, a2 = 1/6 1/2 , c1 = 1/2 1/2 1/6 , c2 = 1/2 1/2 1/6 .
λ π 2λ π 2 π λ 2 π λ

If we wish to only use the outer solution constants, we can consider

a1 and a2 as our constants and eliminate b1 and b2 to obtain
Now the constants for the x < 1 solution, yl , (c1 and c2 ) and the
constants for the x > 1 solution, yr , (a1 and a2 ) are known in
terms of the constants for the inner solution (b1 and b2 ):

31/6 b2 31/6 b1 31/6 (b1 + b2 ) 31/6 (b1 − b2 )

a1 = 1/6 1/2
, a2 = 1/6 1/2 , c1 = 1/2 1/2 1/6 , c2 = 1/2 1/2 1/6 .
λ π 2λ π 2 π λ 2 π λ

If we wish to only use the outer solution constants, we can consider

a1 and a2 as our constants and eliminate b1 and b2 to obtain
1 1
c1 = √ (2a2 + a1 ) and c2 = √ (2a2 − a1 )
2 2
Summary of Connection Formulae
The solution that behaves as
Rx√

a1 exp λ 1 v 3 − 1dv
, x > 1, (23)
(x3 − 1)1/4

behaves as
 R √ 
1
a1 cos λ x 1 − v 3 dv + π4
, x < 1. (24)
(1 − x3 )1/4

Similarly, the solution that behaves as

Rx√

a2 exp −λ 1 v 3 − 1dv
, x > 1, (25)
(x3 − 1)1/4

behaves as
 R √ 
1
2a2 sin λ x 1 − v 3 dv + π4
, x < 1. (26)
(1 − x3 )1/4
Summary of Connection Formulae
The solution that behaves as
Rx√

a1 exp λ 1 v 3 − 1dv
, x > 1, (23)
(x3 − 1)1/4

behaves as
 R √ 
1
a1 cos λ x 1 − v 3 dv + π4
, x < 1. (24)
(1 − x3 )1/4

Similarly, the solution that behaves as

Rx√

a2 exp −λ 1 v 3 − 1dv
, x > 1, (25)
(x3 − 1)1/4

behaves as
 R √ 
1
2a2 sin λ x 1 − v 3 dv + π4
, x < 1. (26)
(1 − x3 )1/4

A similar result is true for all G0 (x) with a single transition point.
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0.
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.

Match the Airy functions, valid near x = 1 to the oscillatory

solutions for x < 1.
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.

Match the Airy functions, valid near x = 1 to the oscillatory

solutions for x < 1.

Consider the boundary layer is at x = 1. Set

X = (x − 1)21/3 λ2/3 ,
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.

Match the Airy functions, valid near x = 1 to the oscillatory

solutions for x < 1.

Consider the boundary layer is at x = 1. Set

X = (x − 1)21/3 λ2/3 , x < 1, X → −∞ as λ → ∞,
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.

Match the Airy functions, valid near x = 1 to the oscillatory

solutions for x < 1.

Consider the boundary layer is at x = 1. Set

X = (x − 1)21/3 λ2/3 , x < 1, X → −∞ as λ → ∞,
and the inner solution is Y = b1 Ai(X) + b2 Bi(X).
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.

Match the Airy functions, valid near x = 1 to the oscillatory

solutions for x < 1.

Consider the boundary layer is at x = 1. Set

X = (x − 1)21/3 λ2/3 , x < 1, X → −∞ as λ → ∞,
and the inner solution is Y = b1 Ai(X) + b2 Bi(X).

The boundary condition at x = 1, transforms into Y (0) = 0 so

Y (X = 0) = b1 Ai(0) + b2 Bi(0) = 0.
Example

y 00 + λ2 (1 − x2 )y = 0, y(0) = y(1) = 0. (27)

There is a transition point at x = 1 at a boundary condition.

Match the Airy functions, valid near x = 1 to the oscillatory

solutions for x < 1.

Consider the boundary layer is at x = 1. Set

X = (x − 1)21/3 λ2/3 , x < 1, X → −∞ as λ → ∞,
and the inner solution is Y = b1 Ai(X) + b2 Bi(X).

The boundary condition at x = 1, transforms into Y (0) = 0 so

Y (X = 0) = b1 Ai(0) + b2 Bi(0) = 0.
√
From the result Bi(0) = 3Ai(0) get
√
b1 + 3b2 = 0. (28)
Outer solution for x < 1.
Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
√
From connection formulae and b1 + 3b2 = 0 find
√
21/6 (b1 + b2 ) b2 21/6 (− 3 + 1)
c1 = = ,
21/2 π 1/2 λ1/6 21/2 π 1/2 λ1/6
Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
√
From connection formulae and b1 + 3b2 = 0 find
√
21/6 (b1 + b2 ) b2 21/6 (− 3 + 1)
c1 = = ,
21/2 π 1/2 λ1/6 21/2 π 1/2
√ λ1/6
21/6 (b1 − b2 ) b2 21/6 ( 3 + 1)
c2 = = −
21/2 π 1/2 λ1/6 21/2 π 1/2 λ1/6
Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
√
From connection formulae and b1 + 3b2 = 0 find
√
21/6 (b1 + b2 ) b2 21/6 (− 3 + 1)
c1 = = ,
21/2 π 1/2 λ1/6 21/2 π 1/2
√ λ1/6 √
21/6 (b1 − b2 ) b2 21/6 ( 3 + 1) ( 3 + 1)
c2 = = − 1/2 1/2 1/6 = − √ c1 .
21/2 π 1/2 λ1/6 2 π λ (− 3 + 1)
Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
√
From connection formulae and b1 + 3b2 = 0 find
√
21/6 (b1 + b2 ) b2 21/6 (− 3 + 1)
c1 = = ,
21/2 π 1/2 λ1/6 21/2 π 1/2
√ λ1/6 √
21/6 (b1 − b2 ) b2 21/6 ( 3 + 1) ( 3 + 1)
c2 = = − 1/2 1/2 1/6 = − √ c1 .
21/2 π 1/2 λ1/6 2 π λ (− 3 + 1)
The condition at x = 1 links b1 and b2 .
Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
√
From connection formulae and b1 + 3b2 = 0 find
√
21/6 (b1 + b2 ) b2 21/6 (− 3 + 1)
c1 = = ,
21/2 π 1/2 λ1/6 21/2 π 1/2
√ λ1/6 √
21/6 (b1 − b2 ) b2 21/6 ( 3 + 1) ( 3 + 1)
c2 = = − 1/2 1/2 1/6 = − √ c1 .
21/2 π 1/2 λ1/6 2 π λ (− 3 + 1)
The condition at x = 1 links b1 and b2 .The matching to the outer
solution links these with c1 and c2 .

The condition y(0) = 0 restricts λ:

Outer solution for x < 1. This is the usual oscillatory WKB
solution
 R   R 
1 1
c1 cos λ x (1 − v 2 )1/2 dv + c2 sin λ x (1 − v 2 )1/2 dv
y= ,
(1 − x2 )1/4
√
From connection formulae and b1 + 3b2 = 0 find
√
21/6 (b1 + b2 ) b2 21/6 (− 3 + 1)
c1 = = ,
21/2 π 1/2 λ1/6 21/2 π 1/2
√ λ1/6 √
21/6 (b1 − b2 ) b2 21/6 ( 3 + 1) ( 3 + 1)
c2 = = − 1/2 1/2 1/6 = − √ c1 .
21/2 π 1/2 λ1/6 2 π λ (− 3 + 1)
The condition at x = 1 links b1 and b2 .The matching to the outer
solution links these with c1 and c2 .

The condition y(0) = 0 restricts λ:

 Z 1  √  Z 1 
2 1/2 3+1 2 1/2
c1 cos λ (1 − x ) dx − √ c1 sin λ (1 − x ) dx = 0.
0 1− 3 0
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4

and so √
√
 
λπ − 3+1
tan = √ = −2 + 3.
4 3+1
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4

and so √
√
 
λπ − 3+1
tan = √ = −2 + 3.
4 3+1
Therefore,
π √
λ = tan−1 (−2 + 3) + nπ
4
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4

and so √
√
 
λπ − 3+1
tan = √ = −2 + 3.
4 3+1
Therefore,
π √ 11π
λ = tan−1 (−2 + 3) + nπ = + nπ
4 12
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4

and so √
√
 
λπ − 3+1
tan = √ = −2 + 3.
4 3+1
Therefore,
√
 
π −1 11π 11
λ = tan (−2 + 3) + nπ = + nπ ⇒ λn = 4 n + .
4 12 12
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4

and so √
√
 
λπ − 3+1
tan = √ = −2 + 3.
4 3+1
Therefore,
√
 
π −1 11π 11
λ = tan (−2 + 3) + nπ = + nπ ⇒ λn = 4 n + .
4 12 12

If transition point ignored, would have obtained

 Z 1 
π
sin λ (1 − x2 )1/2 dx = 0 ⇒ λn = nπ.
0 4
To simplify the condition on λ,
Z 1 Z π/2
2 1/2 π
(1 − x ) dx = cos2 θ dθ = ,
0 0 4

and so √
√
 
λπ − 3+1
tan = √ = −2 + 3.
4 3+1
Therefore,
√
 
π −1 11π 11
λ = tan (−2 + 3) + nπ = + nπ ⇒ λn = 4 n + .
4 12 12

If transition point ignored, would have obtained

 Z 1 
π
sin λ (1 − x2 )1/2 dx = 0 ⇒ λn = nπ.
0 4

Thus, would have λn = 4n, missing the term in 11/12.

Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0.

Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0. (29)

f (x) must be positive for large positive and negative values of x

to give exponential behaviour as x → ±∞.
Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0. (29)

f (x) must be positive for large positive and negative values of x

to give exponential behaviour as x → ±∞. f (x) must be negative
in some range in x, to satisfy boundary conditions. Consider the
case where f (x) has two simple transition points at x = A and
x = B, with A < B.
Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0. (29)

f (x) must be positive for large positive and negative values of x

to give exponential behaviour as x → ±∞. f (x) must be negative
in some range in x, to satisfy boundary conditions. Consider the
case where f (x) has two simple transition points at x = A and
x = B, with A < B. Assume

f (x) > 0, x < A, exponential

Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0. (29)

f (x) must be positive for large positive and negative values of x

to give exponential behaviour as x → ±∞. f (x) must be negative
in some range in x, to satisfy boundary conditions. Consider the
case where f (x) has two simple transition points at x = A and
x = B, with A < B. Assume

f (x) > 0, x < A, exponential

f (x) < 0, A < x < B, oscillatory
Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0. (29)

f (x) must be positive for large positive and negative values of x

to give exponential behaviour as x → ±∞. f (x) must be negative
in some range in x, to satisfy boundary conditions. Consider the
case where f (x) has two simple transition points at x = A and
x = B, with A < B. Assume

f (x) > 0, x < A, exponential

f (x) < 0, A < x < B, oscillatory
f (x) > 0, x > B, exponential
Two transition Point Problems: Final Problem

Consider the homogeneous boundary value problem

y 00 − λ2 f (x)y = 0, y(+∞) = y(−∞) = 0. (29)

f (x) must be positive for large positive and negative values of x

to give exponential behaviour as x → ±∞. f (x) must be negative
in some range in x, to satisfy boundary conditions. Consider the
case where f (x) has two simple transition points at x = A and
x = B, with A < B. Assume

f (x) > 0, x < A, exponential

f (x) < 0, A < x < B, oscillatory
f (x) > 0, x > B, exponential

Write the solution in terms of four WKB solutions.

Two transition Point Problems: Final Problem

The WKB solution for x < A, satisfying y(−∞) = 0 is

Two transition Point Problems: Final Problem

The WKB solution for x < A, satisfying y(−∞) = 0 is

 Z Ap 
−1/4
yI = c1 (f (x)) exp −λ f (v)dv .
x
Two transition Point Problems: Final Problem

The WKB solution for x < A, satisfying y(−∞) = 0 is

 Z Ap 
−1/4
yI = c1 (f (x)) exp −λ f (v)dv . (30)
x

This matches through x = A to the solution valid in A < x < B.

Two transition Point Problems: Final Problem

The WKB solution for x < A, satisfying y(−∞) = 0 is

 Z Ap 
−1/4
yI = c1 (f (x)) exp −λ f (v)dv . (30)
x

This matches through x = A to the solution valid in A < x < B.

The connection formulae (from before) relate this solution to
 Z x 
π
yII = 2c1 (−f (x))−1/4 sin λ
p
−f (v)dv + . (31)
A 4
Two transition Point Problems: Final Problem

We now repeat the process for large positive values of x.

WKB solution for x > B satisfying y(+∞) = 0 is
Two transition Point Problems: Final Problem

We now repeat the process for large positive values of x.

WKB solution for x > B satisfying y(+∞) = 0 is
 Z xp 
−1/4
yIII = c2 (f (x)) exp −λ f (v)dv .
B
Two transition Point Problems: Final Problem

We now repeat the process for large positive values of x.

WKB solution for x > B satisfying y(+∞) = 0 is
 Z xp 
−1/4
yIII = c2 (f (x)) exp −λ f (v)dv . (32)
B

This matches through x = B to

 Z B 
−1/4
p π
yIV = 2c2 (−f (x)) sin λ −f (v)dv + ,
x 4
Two transition Point Problems: Final Problem

We now repeat the process for large positive values of x.

WKB solution for x > B satisfying y(+∞) = 0 is
 Z xp 
−1/4
yIII = c2 (f (x)) exp −λ f (v)dv . (32)
B

This matches through x = B to

 Z B 
−1/4
p π
yIV = 2c2 (−f (x)) sin λ −f (v)dv + , (33)
x 4

that is valid in A < x < B.

Two transition Point Problems: Final Problem

The two solutions valid in A < x < B must be equal

Two transition Point Problems: Final Problem

The two solutions valid in A < x < B must be equal and so

yII = yIV .
 Z x 
p π
yII = 2c1 (−f (x))−1/4 sin λ −f (v)dv +
A 4
!
Z B
−1/4
p π
yIV = 2c2 (−f (x)) sin λ −f (v)dv + ,
x 4

Rewrite the limits on the integrals to be the same,

Two transition Point Problems: Final Problem

The two solutions valid in A < x < B must be equal and so

yII = yIV .
 Z x 
p π
yII = 2c1 (−f (x))−1/4 sin λ −f (v)dv +
A 4
!
Z B
−1/4
p π
yIV = 2c2 (−f (x)) sin λ −f (v)dv + ,
x 4

Rewrite the limits on the integrals to be the same,

Z xp Z B p Z xp
−f (v)dv = −f (v)dv + −f (v)dv
A A B
Two transition Point Problems: Final Problem

The two solutions valid in A < x < B must be equal and so

yII = yIV .
 Z x 
p π
yII = 2c1 (−f (x))−1/4 sin λ −f (v)dv +
A 4
!
Z B
−1/4
p π
yIV = 2c2 (−f (x)) sin λ −f (v)dv + ,
x 4

Rewrite the limits on the integrals to be the same,

Z xp Z B p Z xp
−f (v)dv = −f (v)dv + −f (v)dv
A A B
Z B p Z B p
= −f (v)dv − −f (v)dv,
A x
Two transition Point Problems: Final Problem

Express yII (or equivalently use yIV ) as

!
Z B Z B
−1/4
p p π
yII = 2c1 (−f (x)) sin −λ −f (v)dv + λ −f (v)dv +
x A 4
Two transition Point Problems: Final Problem

Express yII (or equivalently use yIV ) as

!
Z B Z B
−1/4
p p π
yII = 2c1 (−f (x)) sin −λ −f (v)dv + λ −f (v)dv +
x A 4
Z B p π
= −2c1 (−f (x))−1/4 sin λ −f (v)dv + −
x 4
!!
Z B p π
λ −f (v)dv + .
A 2
Two transition Point Problems: Final Problem

Express yII (or equivalently use yIV ) as

!
Z B Z B
−1/4
p p π
yII = 2c1 (−f (x)) sin −λ −f (v)dv + λ −f (v)dv +
x A 4
Z B p π
= −2c1 (−f (x))−1/4 sin λ −f (v)dv + −
x 4
!!
Z B p π
λ −f (v)dv + .
A 2

Here we have taken the minus sign out of the sine function and used
−π/4 = +π/4 − π/2.
Two transition Point Problems: Final Problem

Express yII (or equivalently use yIV ) as

!
Z B Z B
−1/4
p p π
yII = 2c1 (−f (x)) sin −λ −f (v)dv + λ −f (v)dv +
x A 4
Z B p π
= −2c1 (−f (x))−1/4 sin λ −f (v)dv + −
x 4
!!
Z B p π
λ −f (v)dv + .
A 2

Here we have taken the minus sign out of the sine function and used
−π/4 = +π/4 − π/2. Compare with
Z Bp !
−1/4 π
yIV = 2c2 (−f (x)) sin λ −f (v)dv +
x 4
Two transition Point Problems: Final Problem

Hence, yII = yIV if

Z B  
p 1
λ −f (v)dv = n+ π
A 2
Two transition Point Problems: Final Problem

Hence, yII = yIV if

Z B  
p 1
λ −f (v)dv = n+ π (34)
A 2

and the constants c1 and c2 are related by

c1 = (−1)n c2 .
Two transition Point Problems: Final Problem

Hence, yII = yIV if

Z B  
p 1
λ −f (v)dv = n+ π (34)
A 2

and the constants c1 and c2 are related by

c1 = (−1)n c2 .

This is called the Bohr-Sommerfeld condition.

Two transition Point Problems: Final Problem

Hence, yII = yIV if

Z B  
p 1
λ −f (v)dv = n+ π (34)
A 2

and the constants c1 and c2 are related by

c1 = (−1)n c2 .

This is called the Bohr-Sommerfeld condition.