LECTURE 14

Green’s Functions

Consider the following non-homogeneous Sturm-Liouville problem

 
d dy
(1) L [y] = p (x) − q (x) y = f (x) ∀ x ∈ [a, b]
dx dx

a1 y(a) + a2 y 0 (a) = 0
(2)
b1 y(b) + b2 y 0 (b) = 0 .
Our goal is to determine a function G(x, ξ) so that the general solution of (1) and (2) can be written
Z b
(3) y(x) = G(x, s)f (s)ds .
a
Such a function will be called a Green’s function. It can be of great utility as it reduces the problem of
solving equations (1) , (2) to the task of computing a single integral.

To find such a Green’s functionn, we shall first look for a function G(x, s) with the following properties:
   
d d
(4) p (s) − q (s) G (x, s) = 0 ∀ s ∈ [a, b] except s = x
ds ds
∂
(5) a1 G(x, a) + a2 G (x, s) =0
∂s s=a
∂
(6) b1 G(x, b) + b2 G (x, s) =0
∂s s=b
(7) lim G(x, s) − lim G(x, s) = 0
s→x+ s→x−
∂ ∂ 1
(8) lim G(x, s) − lim G(x, s) = −
x→x+ ∂s s→x− ∂s p(x)
We’ll then show that plugging a solution of (4)-(8) into equation (3) solves (1) and (2).

To construct a function satisfying (4), (5) and (6) is relatively easy. From the general theory of linear
ODE’s, we know that the general solution of the homogenous problem
 
d dy
(9) p(s) − q(s)y = 0
ds ds
corresponding to (1) will have the form
y(s) = C1 y1 (s) + C2 y2 (s) .
If we impose one boundary condition we will reduce the space of solutions by one dimension. Thus, the
solutions of (1) satisfying the boundary condition a1 y(a) + a2 y 0 (a) = 0 will be 1-dimensional and so of the
form
y(s) = c1 u1 (s)
with
u1 (s) = a2 (y2 (a) + y20 (a)) y1 (s) − a1 (y1 (a) + y10 (a)) y2 (s).
Similarly, the general solution of (1) satisfying the boundary condition b1 y (b) + b2 y 0 (b) will have the form
y(s) = c2 u2 (s)
1
 14. GREEN’S FUNCTIONS 2

where
u2 (s) = b2 (y2 (b) + y20 (b)) y1 (s) − b1 (y1 (b) + y10 (b)) y2 (s) .
Let us regard the c1 and c2 as functions of a parameter x (but still independent of s). Let us now set

c1 (x)u1 (s) , a<s<x
(10) G(x, s) =
c2 (x)u2 (s) , x<s<b
and impose the conditions (7) and (8). Plugging (10) into (7) and (8), one obtains
c1 (x)u1 (s) − c2 (x)u2 (s) = 0
(11) −1
c1 (x)u01 (s) − c2 (x)u02 (s) = p(x)
But equations (11) now furnish us with a system of 2 equations which we can solve to express the coefficients
c1 and c2 as explicit functions of s. Solving (11) for c1 (s) and c2 (s) yields
u2 (x)
c1 (x) = p(x)W [u1 ,u2 ](x)
u1 (x)
c2 (x) = p(x)W [u1 ,u2 ](x) .

Thus,
( u1 (s)u2 (x)
p(x)W [u1 ,u2 ](x) a≤s<x≤b
(12) G (x, s) ≡ u1 (x)u2 (s)
p(x)W [u1 ,u2 ](x) a≤x<s≤b
will satisfy each of the conditions (4) – (8).

We can simplify the solution (12) even more. For since both u1 and u2 satisfy (9), hence
   
d d du2 d du1
(p(x)W [u1 , u2 ] (x)) = p(x)u1 (x) − p(x)u2 (x)
dx dx dx dx dx

 
d du2
= p(x)u01 (x)u02 (x) + u1 (x)p(x) (x)
dx dx
 
d du1
− p(x)u02 (x)u01 (x) − u2 (x) p(x) (x)
dx dx

= −u1 (x)q(x)u2 (x) + u2 (x)q(x)u1 (x)

=0
Thus,
(13) p(x)W [u1 , u2 ] (x) = A , a constant .

Notice also that if G (x, s) is defined by

(
u1 (s)u2 (x)
A , a≤s<x≤b
G(x, s) = u1 (x)u2 (s)
A , a≤x<s≤b
then (
u1 (x)u2 (s)
A , a≤x<s≤b
G (s, x) = u1 (s)u2 (x) = G (x, s)
A , a≤s<x≤b
So G (x, s) is symmetric under the interchange x ←→ s.

We will now show that the function G (x, s) defined by equation (12) then
Z b
y (x) = G (x, s) f (s) ds
a
 14. GREEN’S FUNCTIONS 3

is indeed the solution of (1) and (2).

Inserting the expression (14) for G(x, s) into (3) we get

Z b Z x Z b
y(x) = G (x, s) f (s) ds = G (x, s) f (s) ds + G (x, s) f (s) ds
a a x
Z x Z b
1 1
(15) = u1 (s) u2 (x) f (s) ds + u1 (x) u2 (s) f (s) ds
A a A x

Differentiating (15) using the rule

Z x Z x Z x
d ∂F ∂F
F (x, s) ds = F (x, s)|s=x + (x, s) ds = F (x, x) + (x, s) ds
dx c c ∂x c ∂x

we obtain
s=x− Z x
1 1
y 0 (x) = u1 (s)u2 (x)f (s) + u1 (s)u02 (x)f (s)ds
A A a
Z b
1 1
+ u1 (x)u2 (s)f (s) + u1 (x)0 u2 (s)f (s)ds
A s=x+ A x
1 1
= u1 (x)u2 (x)f (x) − u1 (x)u2 (x)f (x)
A A
1 x 1 b
Z Z
+ u1 (s)u02 (x)f (s)ds + u1 (x)0 u2 (s)f (s)ds
A a A x
or
Z x Z b
0 1 1
(16) y = u1 (s)u02 (x)f (s)ds + u1 (x)0 u2 (s)f (s)ds
A a A x

Differentiating again yields

− s=x
1 1
y 00 (x) = u1 (s)u02 (x)f (s) + u01 (x)u2 (x)f (s)
A A s=x+
00 Z x 00 Z b
u (x) u (x)
+ 2 u1 (s)f (s)ds + 1 u2 (s)f (s)ds
A a A x
1
= (u1 (x) u02 (x) − u01 (x) u2 (x)) f (x)
A
u002 (x) x u001 (x) b
Z Z
+ u1 (s)f (s)ds + u2 (s)f (s)ds
A a A x
u00 (x) x u00 (x) b
Z Z
W [u1 , u2 ] (x)
= f (x) + 2 u1 (s)f (s)ds + 1 u2 (s)f (s)ds
A A a A x

or, using (13),

x b
f (x) u002 (x) u001 (x)
Z Z
(17) y 00 (x) = + u1 (s)f (s)ds + u2 (s)f (s)ds
p(x) A a A x

If we now apply the differential operator

 
d d
L= p (x) − q (x)
dx dx
d2 d
= p(x) 2 + p0 (x) − q(x)
dx dx
 14. GREEN’S FUNCTIONS 4

to the right hand side of (15) and use the results (16) and (17) we get
L [y] = p (x) y 00 (x) + p0 (x) y 0 (x) − q (x) y (x)
p (x) u002 (x) x p (x) u001 (x) b
Z Z
= f (x) + u1 (s)f (s)ds + u2 (s)f (s)ds
A a A x
p0 (x) x p0 (x) b
Z Z
+ u1 (s)u02 (x)f (s)ds + u1 (x)0 u2 (s)f (s)ds
A a A x
(x) x q (x) b
Z Z
− u1 (s) u2 (x) f (s) ds + u1 (x) u2 (s) f (s) ds
A a A x
Z x
1
= f (x) + [p (x) u002 (x) + p0 (x) u02 (x) − q (x) u2 (x)] u1 (s) f (s) ds
A a
Z b
1
+ [p (x) u001 (x) + p0 (x) u01 (x) − q (x) u1 (x)] u2 (s) f (s) ds
A x
L [u2 ] (x) x L [u1 ] (x) b
Z Z
= u1 (s) f (s) dx + u2 (s) f (s) dx + f (x)
A a A x

= 0 + 0 + f (x)

= f (x)
Thus, the function y(x) defined by equations (3) and (4) - (8) (or equivalently, by (3) and (10)), is a (in
fact, the) solution to the original nonhomogeneous Sturm-Liouville problem (1) and (2).

Summary:

We sought to determine a Green’s function for a general Sturm-Liouville problem with homogeneous bound-
ary conditions; i.e., an integral kernel G(x, ξ) which would allow us to immediately express the solution of
(18) L [y] + f (x) = 0
(19) a1 y (a) + a2 y 0 (a) = 0
(20) b1 y (b) + b2 y (b) = 0
as
Z b
(21) y(x) = G(x, ξ)f (ξ) dξ .
a

We found we could construct such a Green’s function by settingTo accomplish this we first showed that the
conditions
( u1 (s)u2 (x)
p(ξ)W [u1 ,u2 ](x) , 0<x<ξ
(22) G(x, s) = u1 (x)u2 (s)
p(x)W [u1 ,u2 ](x) , ξ<x<b
where u1 (s) and u2 (s) are (any) two linearly independent solutions of,
 
d dy
p (s) − q (s) y = 0 ,
ds ds
respectively, satisfying the boundary conditions
(23) a1 u1 (a) + a2 u01 (a) = 0
(24) b1 u2 (b) + b2 u02 (b) = 0 .