MATH3121 – Mathematical Methods and

PDEs
Topic 4: Sturm-Liouville Equations

Dr Chris Angstmann

School of Mathematics and Statistics

The University of New South Wales
Sydney, Australia

1
The Sturm-Liouville Equation
The Sturm-Liouville equation is an ODE of the form
 
d dφ
p(x) + (q(x) + λr (x))φ = 0,
dx dx

where p(x), q(x) and r (x) are continuous functions,

p(x), r (x) > 0 on an interval a ≤ x ≤ b, and λ is a parameter.

Why are we interested in this?

2
A Motivating Example
Recall that when using separation of variables to solve we will
need to solve ODE eigenvalue problems.
For example the heat equation,

∂u ∂2u
= κ 2,
∂t ∂x
defined over the domain {(x, t) : 0 < x < 1, t > 0}, with an
initial condition, u(x, 0) = g(x), and Dirichlet boundary
conditions, u(0, t) = u(1, t) = 0.
Here we assumed a solution of the form u(x, t) = X (x)T (t)
where,
1 d 2 X (x) 1 dT (t)
2
= = −λ
X (x) dx κT (t) dt

3
A Motivating Example
In the end we found a general solution of the form,
∞
X
u(x, t) = An sin(nπx) exp(−κn2 π 2 t).
n=1

where the coefficients An needed to be calculated from the

initial condition.

This could be achieved easily via integration because the

functions sin(nπx) formed an orthogonal spanning set so that,
Z 1 
0 if n 6= m
sin(nπx) sin(mπx) = 1
0 2 if n = m

4
A Motivating Example
The sine functions arose from the solution of the ODE,

d 2 X (x)
+ λX (x) = 0.
dx 2
So a natural question to ask is;

How can we ensure that the ODEs that arise in solving PDEs
via separation of variables will produce eigenfunctions that form
an orthogonal set?

A partial answer to this comes from considering Sturm-Liouville

problems, as the solution of Strum-Liouville problems will
always be orthogonal.

5
An Example of Sturm-Liouville Form
Show that the following ODE is a Sturm-Liouville equation,

d 2 X (x)
+ λX (x) = 0.
dx 2

The general form of the Sturm-Liouville equation is,

 
d dφ
p(x) + (q(x) + λr (x))φ = 0.
dx dx

So with p(x) = 1, q(x) = 0, and r (x) = 1 we recover the form

of the original ODE.

6
Sturm-Liouville BVPs
In general we will consider a Sturm-Liouville BVP with the ODE,
 
d dφ
p(x) + (q(x) + λr (x))φ = 0,
dx dx

considered over the one dimensional interval (a, b) subject to

Robin BCs,

α1 φ(a) + β1 φ0 (a) = 0,
α2 φ(b) + β2 φ0 (b) = 0.

Note that with appropriate choice of coefficients the Robin BCs

may be Dirichlet or Neuman conditions.

7
Orthogonality

A set of functions {φ1 , φ2 , φ3 , . . .} is orthogonal on the interval

(a, b) with respect to the weight function w(x) if
Z b
φn (x)φm (x)w(x)dx = 0, for all n, m such that n 6= m.
a

If w(x) = 1 the functions are then said to be orthogonal.

8
Orthogonality Example

∞
The functions sin nπx

a n=1
are orthogonal on the interval
(0, a) since
Z a  nπx   mπx 
sin sin dx = 0 when n 6= m.
0 a a

9
Sturm-Liouville Orthogonality

Theorem 1
The eigenfunctions of Sturm-Liouville BVP above satisfy the
integral relationship:
Z b
r (x)φn (x)φm (x)dx = 0, if m 6= n,
a

where φ1 , φ2 , φ3 , . . . are eigenfunctions and φn corresponds to

the eigenvalue λn .

Hence, the set of eigenfunctions for the Sturm-Liouville problem

are orthogonal on the interval of interest w.r.t the weight
function r (x).

10
Proof
Since φn and φm are eigenfunctions they must satisfy the ODE
 
d dφn
p(x) + q(x)φn = −λn r (x)φn (1)
dx dx
 
d dφm
p(x) + q(x)φm = −λm r (x)φm (2)
dx dx

Multiply (1) by φm and (2) by φn and subtract,

   
d dφn d dφm
φm p(x) − φn p(x) = (λm − λn )r (x)φn φm
dx dx dx dx

he left-hand-side can be expressed as,

  
d dφn dφm
p(x) φm − φn
dx dx dx
11
Proof
So
  
d dφn dφm
p(x) φm − φn = (λm − λn )r (x)φn φm
dx dx dx

Integrate both sides w.r.t x from a to b,

dφm b
   Z b
dφn
p(x) φm − φn = (λm − λn ) r (x)φn φm dx.
dx dx a a

The left hand side is

dφm b
  
dφn
p(x) φm − φn = p(b)[φm (b)φ0n (b) − φn (b)φ0m (b)]
dx dx a
− p(a)[φm (a)φ0n (a) − φn (a)φ0m (a)]
= 0,
using the boundary conditions above.
12
Proof
If λm 6= λn , (for m 6= n), then
Z b
(λm − λn ) r (x)φm φn dx = 0
a

and so, Z b
r (x)φm φn dx = 0,
a
and the proof is complete.

13
Repeated Eigenvalues
What about the case when φn and φm are distinct
eigenfunctions corresponding to the same eigenvalue.
So what will happen if λm = λn ?

It is possible to show that the eigenfunction corresponding to

any eigenvalue is unique within a multiplicative constant.

Thus this case does not arise in these types of problems.

14
Singular Sturm-Liouville problems
As part of the orthogonality proof we required

dφm b
  
dφn
p(x) φm − φn = 0.
dx dx a

If p(a) = 0 we would only need φ and φ0 to be finite at x = a.

In this case, it can be shown that the uniqueness of the

eigenfunction result still holds.

Problems where p(a) = 0 and/or p(b) = 0 are called singular

Sturm-Liouville problems.

15
Legendre’s Equation
Legendre’s equation is,
 
d dφ
(1 − x 2 ) + λφ = 0, −1 ≤ x ≤ 1.
dx dx

This arises in the separation of solution to the heat equation or

Laplace’s equation in spherical coordinates.
Note p(x) = 1 − x 2 = 0 at the boundary x = ±1. Typically, φ
and φ0 are finite at x = ±1.
Hence, the eigenfunctions will satisfy the integral relation
Z 1
φn φm dx = 0, n 6= m, (r (x) ≡ 1)
−1

It can be shown that Legendre’s equation has solutions that are

finite at the endpoints x = ±1 only if λ = n(n + 1) for
n = 0, 1, 2, . . .. So the boundary conditions uniquely determine
the eigenvalues.
16
Legendre Polynomials
The Legendre polynomials {Pn (x)}∞ n=0 are the only solutions of
Legendre’s differential equation
 
d 2 dφ
(1 − x ) + n(n + 1)φ = 0, n = 0, 1, 2, . . .
dx dx

that are finite at both x = ±1.

17
Legendre Polynomials
The first few Legendre polynomials are,

1.0

P0 (x) = 1
P1 (x) = x 0.5

1
P2 (x) = (3x 2 − 1) -1.0 -0.5 0.5 1.0

2
1 -0.5

P3 (x) = (5x 3 − 3x)

2 -1.0

1
P4 (x) = (35x 4 − 30x 2 + 3)
8
1
P5 (x) = (63x 5 − 70x 3 + 15x)
8

18
Recurrence Relation
P0 = 1 and P1 = x. Legendre polynomials of higher degree can
be obtained using the recurrence relation

nPn (x) = (2n − 1)xPn−1 (x) − (n − 1)Pn−2 (x), n ≥ 2.

n = 2,

2P2 (x) = 3xP1 (x) − P0 (x) = 3x 2 − 1

1
P2 (x) = (3x 2 − 1)
2

19
Orthogonality Example
Show that the first two Legendre polynomials, P0 (x) = 1 and
P1 (x) = x are orthogonal for x ∈ [−1, 1].
Determine constants α and β such that the function
h(x) = 1 + αx + βx 2 is orthogonal to both P0 and P1 for
x ∈ [−1, 1].
How does h(x) compare with P2 (x)?

20
Bessel’s Equation
The parametrised Bessel’s equation is

x 2 y 00 + xy 0 + (k 2 x 2 − m2 )y = 0, with order m and parameter k .

So what does the Sturm-Liouville version of the parametrised

Bessel equation look like?

Rearranging the parametrised Bessel equation

y0 m2
 
00 2
y + + k − 2 y =0
x x

Multiplying throughout by ρ

y0 m2
 
00 2
ρy + ρ + ρ k − 2 y = 0 (3)
x x

21
Bessel’s Equation
We want to write this as
d m2
(ρy 0 ) + ρ(k 2 − 2 )y = 0 (4)
dx x
which, if we expand the first term, is

m2
ρy 00 + ρ0 y 0 + ρ(k 2 − )y = 0 (5)
x2

Comparing (5) with (3), and equating the coefficients of y 0 , we

have
ρ dρ ρ
ρ0 = or = (6)
x dx x

22
Bessel’s Equation
Rearranging, and integrating we have
Z Z
dρ dx
=
ρ x
or
ln ρ = ln x + C
Choose C = 0, so ρ = x
Substituting ρ = x into the (3) we get
m2
   
d dy
x + − + k 2 x y = 0,
dx dx x
which, if we compare with the standard Sturm-Liouville form
 
d dy
p(x) + (q(x) + λr (x))y = 0
dx dx
gives
p(x) = x, λ = k 2, q(x) = −m2 /x, r (x) = x.
23
Bessel functions Jn and Yn
With k 2 = 1, the Bessel equation is

x 2 y 00 + xy 0 + (x 2 − m2 )y = 0.

The solutions to this equation define the Bessel functions Jm (x)

and Ym (x).
The functions Jm are often called Bessel functions of the first
kind. It is possible to define Jm by its Taylor series around x = 0
∞
X (−1)k  x 2k +m
Jm (x) =
k !Γ(k + m + 1) 2
k =0

It can also be obtained via an integral,

1 π
Z
Jm (x) = cos(ms − x sin(s))ds
π 0

24
Bessel functions Jn
Alternatively, Bessel functions of the first kind follow the
recurrence relation,
2n
Jn+1 (x) + Jn−1 (x) = Jn (x).
x
1.0

0.8

0.6

0.4

0.2

2 4 6 8 10
-0.2

-0.4

25
Bessel functions Yn
Bessel functions of the second kind follow the recurrence
relation,
dYn (x)
Yn+1 (x) − Yn−1 (x) = −2 .
dx

2 4 6 8 10

-1

-2

-3

26
Orthogonality of Bessel functions
We have shown that Bessel’s equation can be put into
Sturm-Liouville form
m2
(xφ0 )0 − φ = −λxφ
x
Here, p(x) = x, q(x) = −m2 /x, r (x) = x.
With boundary conditions of the form φ(a) = 0, and φ, φ0 finite
when x = 0, we obtain the eigenfunctions satisfying the
orthogonality relation
Z a
φj (x)φk (x)r (x)dx = 0, j 6= k .
0

27
Orthogonality of Bessel functions
Now r (x) = x,
q
φj (x) = Jm ( λj x), j = 1, 2, 3, . . . ,

p (m)
where λj = µj /a is the j-th root of Jm (x)/a.
Thus
 
a (m) (m)
!
Z µj µk
xJm  x  Jm x dx = 0, j 6= k
0 a a

(m)
   (m)

µj µk
i.e. the functions Jm a x and Jm a x are orthogonal on
0 ≤ x ≤ a with respect to the weight function x.

28
Chebyshev Equation
Another major type of orthogonal functions that often arise in
numerical approximation are Chebyshev polynomials.
The Chebyshev differential equation is,

(1 − x 2 )y 00 − xy 0 + n2 y = 0.

This can be put into the Sturm-Liouville form.

The eigenfunctions of this equation are Chebyshev polynomials
of the first kind, Tn .

29
Chebyshev Polynomials
The first few Chebyshev polynomials are,

1.0

T0 (x) = 1
0.5
T1 (x) = x
T2 (x) = 2x 2 − 1 -1.0 -0.5 0.5 1.0

3
T3 (x) = 4x − 3x -0.5

T4 (x) = 8x 4 − 8x 2 + 1 -1.0

T5 (x) = 16x 5 − 20x 3 + 5x

30
Chebyshev Polynomials
The Chebyshev polynomials can be defined indirectly via the
identity,
Tn (cos θ) = cos(nθ).
Again a recurrence relation exists for the polynomials,

Tn (x) = 2xTn−1 (x) − Tn−2 (x) n ≥ 2.

and T0 (x) = 1, and T1 (x) = x.

31
Orthogonality Example
Show that the first two Chebyshev polynomials, T0 (x) = 1 and
T1 (x) = x are orthogonal with respect to the weighting function
1
r (x) = (1 − x 2 )− 2 for x ∈ [−1, 1].
Determine constants α and β such that the function
h(x) = 1 + αx + βx 2 is orthogonal to both T0 and T1 with
respect to the weighting function r (x) for x ∈ [−1, 1].
How does h(x) compare with T2 (x)?

32