Greens Function and the Statement of Orthonormality

In terms of the standard differentiation operator D =

d
,
dx

the Sturm-Liouville operator L acting on

some function y(x) reads

L(y) = D(p(x)D(y))+q(x)y
Any second-order linear differential operator
a2 HxL D2 + a1 HxL D + a0 HxL

can be written in Sturm-Liouville form by multiplying the operator by the integrating factor
r HxL =

a1 HxL
a2 HxL

a2 HxL

Example
x3 y '' - x y ' + 2 y = 0
Divide by x3 :
x
2
y '' - 3 y ' + 3 y = 0
x

Multiply throughout by integrating factor

3
2
e -xx x = e -1x x = e1x
this gives
e1x y '' -

e1x

y'+

x2
1x

since

de
dx

2 e1x
x3

y =0

1x

-e

x2

we can put the equation in Sturm Liouville form

Ie1x y 'M' +

2 e1x
x3

y =0

Before we proceed, let us discuss Greens formula (or identity). The

following lemma can be verified.
Lemma (the preliminary Greens formula)
Let (a, b) be some finite interval, and let L be the operator given by
L[j] =

df
d
BpHxL F
dx
dx

+ q(x)j

where p and q are any suitably smooth and integrable functions on (a, b) . Then
a f L@gD x = p HxL f HxL
b

dg
dx

b
a

- a p
b

df dg
dx dx

x + a qfg x
b

whenever f and g are suitably smooth and integrable functions on (a, b) .

Now suppose we have two functions u and v on (a, b) (assumed suitably smooth and
integrable but, possibly, complex valued), and suppose we want to compare
a u* L@vD x
b

and

a L@uD* v x
b

If p and q are real-valued functions, then it is trivial to verify that

L@uD* = L@u* D

Using this and the above preliminary Green formula, we see that
a u* L@vD x - a L@uD* v x = a u* L@vD x - a v L@u* D x
b

= Bpu*

dv b

- p
b

d u* d v

x + qu* v xF
b

a u* L@vD x
b

and

C8_1.nb

a L@uD* v x
b

If p and q are real-valued functions, then it is trivial to verify that

L@uD* = L@u* D

Using this and the above preliminary Green formula, we see that
a u* L@vD x - a L@uD* v x = a u* L@vD x - a v L@u* D x
b

= Bpu*
- Bpv

dv b
dx a

d u* b
dx a

-a p
b

-a p
b

d u* d v
dx dx

d u* d v
dx dx

x + a qu* v xF
b

x + a qu* v xF
b

Nicely enough, most of the terms on the right cancel out, leaving us with:
Theorem (Green formula)
Let (a, b) be some interval, p and q any suitably smooth and integrable real-valued functions on (a,
b) , and L the operator given by
L[j] =
Then

df
d
BpHxL F
dx
dx

+ q(x)j

*
a u* L@vD x - a L@uD* v x = p[u
b

dv
dx

-v

d u*
E
dx

b
a

for any two suitably smooth and differentiable functions u and v .

a Hu LHvL - v LHuLL x = pHbL HuHbL vx HbL - vHbL ux HbLL - pHaL HuHaL vx HaL - vHaL ux HaLL
b

If we impose the regular Sturm-Liouville boundary conditions from previous session on the functions
u and v, then the right-hand term in the preceding Greens formula vanishes, and we obtain
a u LHvL x = a v LHuL x
b

When the preceding statement holds, we say the operator L, for the appropriate boundary conditions, is self-adjoint.
If we now write the Sturm-Liouville differential equation for the two different eigenfunctions fn (x) and
fm (x), we have
LHfn ) + ln wHxL fn = 0
and
LHfm ) + lm wHxL fm = 0
Rearranging these equations and integrating over the interval yields
a Hfn LHfm L - fm LHfn LL x = Hln - lm L Ja fn HxL fm HxL wHxL xN
b

If the eigenfunctions satisfy the regular Sturm-Liouville boundary conditions from previous session,
then Greens formula shows the left-hand side to be zero and, thus, the right-hand side is also zero.
This yields the statement of orthogonality of the eigenfunctions:
Hln - lm L Ja fn HxL fm HxL wHxL xN = 0
b

The integral here denotes the inner product of the two different eigenfunctions with respect to the
weight function w(x) over the interval I = {x |a < x < b}. This result shows that if the eigenvalues are
not equal - that is, ln lm - then the eigenfunctions, corresponding to the different eigenvalues, are

If the eigenfunctions satisfy the regular Sturm-Liouville boundary conditions from previous session,
then Greens formula shows the left-hand side to be zero and, thus, the right-hand side isC8_1.nb
also zero.3
This yields the statement of orthogonality of the eigenfunctions:
Hln - lm L Ja fn HxL fm HxL wHxL xN = 0
b

The integral here denotes the inner product of the two different eigenfunctions with respect to the
weight function w(x) over the interval I = {x |a < x < b}. This result shows that if the eigenvalues are
not equal - that is, ln lm - then the eigenfunctions, corresponding to the different eigenvalues, are
orthogonal with respect to the weight function w(x) over the interval I.

The Statement of Orthonormality

To normalize a vector, we simply divide the vector by its length or norm. Recall that the norm is the
square root of the inner product of the vector with itself with respect to the weight function w(x) that is,

norm =

2
a fn HxL wHxL x
b

If we normalize the eigenfunctions - that is, divide each by its length (norm) - then the preceding
condition of orthogonality yields an equivalent "statement of orthonormality," which, in terms of the
Kronecker delta function, reads
a fn HxL fm HxL wHxL x = d(n,m)
b

This statement indicates the two different eigenfunctions to be orthonormal (a combination of being
orthogonal and normalized) with respect to the weight function w(x) over the interval I. This result
will play a significant role in the development of the generalized Fourier series.

Types of Boundary Conditions

Before we consider different examples of Sturm-Liouville operators with various types of nonmixed
homogeneous boundary conditions, we first classify the regular boundary conditions that we will
look at in three types. Let point a be the boundary point.
1. Type 1 or Dirichlet condition (vanishing of the zero derivative at the boundary)
y(a) = 0
2. Type 2 or Neumann condition (vanishing of the first derivative at the boundary)
yx (a) = 0
3. Type 3 or Robin condition (linear dependence between zero and first derivative at boundary)
yx (a) + hy(a) = 0
The constant h in the Robin condition has the dimension of the reciprocal of displacement. When
we consider actual boundary value problems, we will be able to attach physical significance to each
of these condition.

Example:
We seek the eigenvalues and corresponding eigenfunctions for the Euler differential equation
shown:
d2
d x2

yHxL + l yHxL = 0

over the interval I = {x | 0<x<1} with the boundary conditions

yx (0) = 0

and

y(1) = 0

From our earlier classification of boundary conditions, we have a type 2 condition at the left and a
type 1 condition at the right.
By comparison, we see the Euler differential equation to be a special case of a Sturm-Liouville
differential equation

d2
4

d x2
C8_1.nb

yHxL + l yHxL = 0

over the interval I = {x | 0<x<1} with the boundary conditions

yx (0) = 0

and

y(1) = 0

From our earlier classification of boundary conditions, we have a type 2 condition at the left and a
type 1 condition at the right.
By comparison, we see the Euler differential equation to be a special case of a Sturm-Liouville
differential equation
L(y)+l w(x)y = 0
where L is the Sturm-Liouville operator
L(y) =

d
(p(x)D(y))+q(x)y
dx

For the Euler differential equation, we identify the coefficients p(x) = 1, q(x) = 0, and w(x) = 1.
Solution: To solve for the eigenvalues, we consider three possible cases for values of l. We first
consider the case for l < 0. We set l = - m2 . A set of basis vectors for this case is
y1 (x) = sinh(mx)
and
y2 (x) = cosh(mx)
and the general solution can be written in terms of this basis as
y HxL = C1 sinh(mx) + C2 cosh(mx)
Substituting the boundary conditions at x = 0 and at x = 1, we get
C1 m = 0
and
C2 msinh(1) = 0
The only solution to the preceding is the trivial solution C1 = 0 and C2 = 0.
We next consider the case for l = 0. A set of system basis vectors is
y1 (x) = 1
and
y2 (x) = x
and the general solution in terms of this basis is
y(x) = C1 + C2 x
Substituting the boundary condition at x = 0 and at x = 1 gives
C2 = 0
and
C1 + C2 = 0
The only solution to the preceding is the trivial solution C1 = 0 and C2 = 0. From a geometric standpoint, we expected this result because it is impossible to have a nontrivial line that will satisfy the
boundary conditions at both the left and right end points of the interval.
Finally, we consider the case for l > 0.We set l = m2 . A set of basis vectors for this case is
y1 (x) = sin(m x)
and
y2 (x) = cos(m x)
The general solution here is
y(x) = C1 sin(mx) + C2 cos(m x)
Substituting the boundary conditions at x = 0 and at x = 1, we get
C1 m = 0
and
C1 sin HmL + C2 cos(m) = 0

The only nontrivial solutions to the preceding are that C1 = 0, C2 is arbitrary, and m has values that

y2 (x) = cos(m x)
The general solution here is
y(x) = C1 sin(mx) + C2 cos(m x)

C8_1.nb

Substituting the boundary conditions at x = 0 and at x = 1, we get

C1 m = 0
and
C1 sin HmL + C2 cos(m) = 0

The only nontrivial solutions to the preceding are that C1 = 0, C2 is arbitrary, and m has values that
are the roots of the eigenvalue equation
cos(m) = 0
Thus, m takes on the special values
mn =

H2 n-1L p
2

for n = 1, 2, 3, . . . .
The allowed eigenvalues ln = mn 2 are
ln =

H2 n-1L2 p2
4

and the corresponding eigenfunctions are

H2 n-1L px
fn HxL = I
M
2

for n = 1, 2, 3, . . . .
These eigenfunctions are not normalized.We must first find the norm of each eigenfunction by
evaluating the square root of the inner product of the eigenfunctions with respect to the weight
function w(x) = 1 over the interval.
The norm is
norm =

0 cosI
1

H2 n-1L px 2
M
2

Evaluating this integral yields

norm =

2
2

Dividing each eigenfunction by its norm yields the orthonormal set of eigenfunctions
fn (x) =

2 cosI

H2 n1L px
M
2

for n = 1, 2, 3, . . . .

The statement of orthonormality for this set of eigenfunctions reads

0 2 cosI
1

H2 n-1L px
H2 m-1L px
M cosI
M x
2
2

for n,m = 1, 2, 3, . . . .

= d(n,m)

Generalised Fourier Series Expansion

Since the eigenfunctions form a complete set with respect to any piecewise smooth function f(x)
over the interval I = {x |a < x < b}, then we can expand f(x) as a generalized Fourier series. In terms
of the orthonormalized eigenfunctions, the generalized expansion formula for f(x) reads
f(x) =
n=0 FHnL fn HxL

where the F(n) are the generalized Fourier coefficients of the function f(x).We proceed to evaluate
these coefficients in a formal manner by taking advantage of the statement of orthonormality given
earlier.
If we take the inner product of both sides of the preceding equation, with respect to the weight
function w(x) over the interval, and we assume it is valid to interchange the order of the summation
and integration operators (see exercises on conditions for this), then we get
a f HxL fm HxL wHxL x =n=0 FHnLJa fn HxL fm HxL wHxL xN
b

Because of the statement of orthonormality, we can write the integral on the right in terms of the
Kronecker delta function as
a f HxL fm HxL wHxL x = n=0 FHnL dHn, mL
b

The sum on the right is easy to evaluate, since d(n,m) is 0 for n m and 1 for n = m.
Evaluating the sum (only one term survives), we get
F(m) = a f HxL fm HxL wHxL x
b

Thus, we see that the Fourier coefficients are evaluated from the inner product of f(x) and the
corresponding orthonormal eigenfunction, with respect to the weight function w(x), over the interval I.

Convergence of the Fourier Series

Before we provide some examples, we first consider some important aspects on the convergence
of a series.We consider two important questions dealing with the convergence of the Fourier series:
(1)Will the series converge? (2) To what value does the series converge?
Both of these questions are addressed by the following two theorems.
Theorem 1 (Pointwise convergence): Let f(x) be piecewise smooth on the closed interval [a, b], and
let the set 8fn (x)} be the eigenfunctions of a regular Sturm-Liouville problem over that interval. Then
for each value of x in the corresponding open interval (a, b), the Fourier series, relative to these
eigenfunctions, converges "pointwise"in accordance with

n=0 FHnL fn HxL =

f Hx+0L+f Hx-0L
2

Here, the F(n) are the properly evaluated Fourier coefficients over the interval, and the term on the
right is the average of the left- and right-hand limits of f(x) at the point. For piecewise smooth functions, the quality of convergence of the series depends on the value of x in the interval.
Theorem 2 (Uniform convergence): Let f(x) be defined and continuous with continuous first and
second derivatives on the closed interval [a, b], and let f(x) satisfy the same boundary conditions as
the eigenfunctions 8fn (x)}, which are the solutions to the regular Sturm-Liouville problem over that
interval. Then the Fourier series expansion of f(x), relative to these eigenfunctions, converges
"uniformly" to f(x) on the closed interval.

Here, the F(n) are the properly evaluated Fourier coefficients over the interval, and the term on the
right is the average of the left- and right-hand limits of f(x) at the point. For piecewise smooth funcC8_2.nb
tions, the quality of convergence of the series depends on the value of x in the interval.
Theorem 2 (Uniform convergence): Let f(x) be defined and continuous with continuous first and
second derivatives on the closed interval [a, b], and let f(x) satisfy the same boundary conditions as
the eigenfunctions 8fn (x)}, which are the solutions to the regular Sturm-Liouville problem over that
interval. Then the Fourier series expansion of f(x), relative to these eigenfunctions, converges
"uniformly" to f(x) on the closed interval.

Functions that satisfy the more rigid conditions for uniform convergence are more well behaved and
have Fourier series expansions that converge more rapidly and with better fidelity to the actual
function. Uniform convergence obviously implies pointwise convergence at all points on the closed
interval, and, in addition, the convergence is independent of the value of x on that interval.
We now provide illustrations of Fourier series expansions of three different functions in terms of
those eigenfunctions derived in the problem in Chapter10b. This exercise demonstrates how the
quality of convergence of the resulting series depends on the character of the function and how well
it adheres to the boundary conditions on the problem.

Example
We seek the generalized series expansion of a piecewise smooth function f(x) over the interval I =
{x | 0 < x < 1}, in terms of the orthonormal eigenfunctions evaluated in Chapter 10b.
In terms of the evaluated orthonormal eigenfunctions, the expansion reads
f(x ) =
n=1 FHnL cosI

H2 n-1L p x
M
2

From the preceding, the Fourier coefficients are determined from the integral
F(n) = 0 f HxL cosI
1

H2 n-1L p x
M
2

2 x

We evaluate the series for three different functions f(x), each having a different character over the
interval.
Case 1 We let f 1(x) = x. For this case, f 1(x) does not satisfy either of the boundary conditions at
the left or right end points of the interval. The Fourier coefficient integral reads
F(n) = 0 x cosI
1

H2 n-1L p x
M
2

2 x

Evaluation of this integral yields

FHnL = -

2 H2 pH-1Ln n-pH-1Ln +2L

H2 n-1L2 p2

Thus, the Fourier series expansion of f 1(x) over the interval

is

1 HxL =
n=1

4 H2 p nH-1Ln -pH-1Ln +2L cosJ

H2 n-1L2 p2

H2 n-1L p x
2

A plot of both f 1(x) and the first five terms of the series expansion is shown in Figure below

C8_2.nb

Because the function f 1(x) does not satisfy either of the boundary conditions, the series converges
pointwise. The converges very slowly, as an alternating

1
n

series, and the quality of convergence

appears to be poor.
Case 2 We let f 2(x) = 1 - x. For this case, f 2(x) does not satisfy the boundary condition at the left,
but it does satisfy the condition at the right end point of the interval. The Fourier coefficient integral
reads
FHnL = 0 H1 - xL
1

2 cosI

H2 n-1L p x
M x
2

Evaluation of this integral yields

F HnL =

H2 n-1L2 p2

Thus, the Fourier series expansion of f 2(x) over the interval is

f 2 HxL =
n=1

8 cosJ

H2 n-1L p x
2
2 2

H2 n-1L p

A plot of both f 2(x) and the first five terms of the series expansion is shown in Figure below

The function f 2 (x) satisfies the boundary condition at the right but not at the left end point, and the
series converges pointwise.The series converges very rapidly as a

1
n2

series, and, sincethe function

satisfies the boundary condition at one of the end points, the quality of convergence appears to be
much better than that for case 1.

Case 3 We let f 3 HxL = 1 - x2 . For this case, f 3 (x) satisfies both boundary conditons at the left and
right end points of the interval.The Fourier coefficient integral reads
FHnL = 0 H1 - xL
1

2 cosI

H2 n-1L p x
M x
2

Evaluation of this integral yields

F HnL =

16

2 H-1Ln+1

H2 n-1L3 p2

Thus, the Fourier series expansion of f 3 (x) over the interval is

f 3 HxL =
n=1

32 H-1Ln+1 cosJ

H2 n-1L p x

H2 n-1L p
3

FHnL = 0 H1 - xL
1

2 cosI

C8_2.nb

H2 n-1L p x
M x
2

Evaluation of this integral yields

F HnL =

16

2 H-1Ln+1

H2 n-1L3 p2

Thus, the Fourier series expansion of f 3 (x) over the interval is

3 HxL =
n=1

32 H-1Ln+1 cosJ

H2 n-1L p x

H2 n-1L3 p2

A plot of both f 3 (x) and the first five terms of the series expansion is shown in Figure below

Because the function f 3(x) satisfies both of the boundary conditions at the left and right end points,
the series converges uniformly to f(x). The series converges very rapidly as a

1
n3

series, and the

quality of convergence is considerably much better than either of the two cases already discussed.