Chapter 8

Sturm-Liouville boundary value problems

In this chapter, we consider a class of two-point boundary value problems. The so-called Sturm-Liouville
Problems. They are a class of eigenvalue problems, which include many of the previous problems as
special cases.
We let  
d dϕ
Lϕ = − p(x) + q(x)ϕ
dx dx
Suppose we have some Sturm-Liouville problem with differential equation

Lϕ = λrϕ for 0<x<l (8.1)

Here, the functions p, p′ , q, and r are continuous on [0, ℓ] with

p(x) ≥ 0 and r(x) > 0, 0≤x≤ℓ

We couple (8.1), with the following boundary conditions

α1 y(0) + α2 y ′ (0) = 0, β1 y(ℓ) + β2 y ′ (ℓ) = 0 (8.2)

We define the Sturm-Liouville eigenvalue problem (SL Problem) as:

Ly = λr(x)y
α1 y(0) + α2 y ′ (0) = 0
(8.3)
β1 y(ℓ) + β2 y ′ (ℓ) = 0
p(x) > 0, r(x) > 0

Remark 8.0.1.

• If in (8.3) we choose p(x) = 1, q(x) = 0, r(x) = 1, α1 = 1, α2 = 0, β1 = 1 and β2 = 0, then

we obtain the following problem

2
y ′′ + λy = 0
 
λn = nπ , n = 1,
2, . . .
=⇒ ℓ
y(0) = 0 = y(ℓ) yn (x) = sin nπx
ℓ

• If in (8.3) we choose p(x) = 1, q(x) = 0, r(x) = 1, α1 = 0, α2 = 1, β1 = 0 and β2 = 1, then

we obtain the following problem

y ′′ + λy = 0 nπ 2
 

λn = , n = 0, 1,

2, . . .
=⇒ ℓ
y (0) = 0 = y ′ (ℓ)
′ nπx
yn (x) = cos ℓ

167
168 CHAPTER 8. STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

• Notice that these boundary conditions are specified at separate endpoints and are called
separated boundary conditions. The periodic BC X(−ℓ) = X(ℓ) are not separated, so the
following problem is not technically an SL Problem.
y ′′ + λy = 0
y(−ℓ) = y(ℓ)
y ′ (−ℓ) = y ′ (ℓ)

• If p > 0 and r > 0 on a finite interval [0, ℓ], then the SL problem is said to be regular. If
either p(x) or r(x) is zero for some x, or if the domain is unbounded (e.g., [0, ∞) ), then
the problem is singular.

• If P0 , P1 , P2 , and R are continuous and P0 and R are positive on a closed interval [a, b],
then the general equation

P0 (x)y ′′ + P1 (x)y ′ + P2 (x)y + λR(x)y = 0

can be rewritten in the SL form. Indeed, if we divide the previous equation by P0 (x), we obtain

P1 (x) ′ P2 (x) R(x)

y ′′ + y + y+λ y=0
P0 (x) P0 (x) P0 (x)
R P1 (x)
dx
We multiply the equation by the integrating factor p̃(x) = e P0 (x) , we get:
 
′′ P1 (x) ′ P2 (x) R(x)
p̃(x)y + p̃(x)(x) y + p̃(x) +λ y=0
P0 (x) P0 (x) P0 (x)
The first two terms give:
 
d dy
p̃(x)
dx dx
Thus, the equation becomes:
 
d dy P2 (x) R(x)
p̃(x) + p̃(x) y = −λp̃(x) y
dx dx P0 (x) P0 (x)
Define the following functions:
R P1 (x)
dx
p(x) = −p̃(x) = −e P0 (x)
P2 (x)
q(x) = p̃(x)
P0 (x)
R(x)
r(x) = −p̃(x)
P0 (x)

Therefore,  
d dy
− p(x) + q(x)y = λr(x)y
dx dx
 169

Example 8.0.1.
1. Consider the boundary value problem
ϕ′′ + xϕ′ + λϕ = 0, 0<x<1
ϕ(0) = 0, ϕ(1) = 0
To write it into SL form, multiply by the integrating factor
2 /2
R
xdx
µ(x) = e = ex
Since P (x) = 1, Q(x) = x and R(x) = 1, we have:
2 /2 2 /2 2 /2
ex ϕ′′ + ex xϕ′ + λex ϕ=0
That is  ′
x2 /2 ′ 2 /2
− e ϕ = λex ϕ
Thus, the SL form is obtained with
2 /2 2 /2
p(x) = ex and r(x) = ex
2. Consider the boundary value problem
−y ′′ + x4 y ′ = λy
To convert into the SL form, we rewrite the equation as:
y ′′ − x4 y ′ = −λy
The integrating factor is
x4 dx 5 /5
R
µ(x) = e− = e−x
Multiplying the original equation by µ(x), we obtain:
5 /5 5 /5 5 /5
−e−x y ′′ + e−x x4 y ′ = λe−x y
That is  5 ′ 5
− e−x /5 y ′ = λe−x /5 y
Thus, the SL form is obtained with
5 /5 5 /5
p(x) = e−x and r(x) = e−x
3. Consider the eigenvalue problem
y ′′ + 3y ′ + (2 + λ)y = 0, y(0) = 0, y(1) = 0
The integrating factor is R
3dx
µ(x) = e = e3x
Multiplying the equation by µ(x) yields

′
− −e3x y ′ + 2e3x y + λe3x y = 0
So that the eigenvalue problem can be written as

′
− −e3x y ′ + 2e3x y = −λe3x y, y(0) = 0, y(1) = 0
Thus the SL form is obtained with
p(x) = −e3x , q(x) = 2e3x , and r(x) = −e3x
170 CHAPTER 8. STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

8.1 Properties of SL problems

Eigenvalues:
(a) The eigenvalues λ are all real.
(b) There are infinitely many eigenvalues λj satisfying

λ1 < λ2 < · · · < λj → ∞ as j → ∞

(c) Provided α1
α2
< 0, ββ12 > 0, and q(x) > 0, then λj > 0.

Eigenfunctions:
For each eigenvalue λj , there exists an eigenfunction ϕj (x) (unique up to a multiplicative constant)
such that:
(a) The eigenfunctions ϕj (x) are real and can be normalized to satisfy
Z ℓ
r(x)ϕ2j (x)dx = 1
0

(b) Eigenfunctions corresponding to distinct eigenvalues are orthogonal with respect to the weight
function r(x) :
Z ℓ
⟨ϕj , ϕk ⟩ := r(x)ϕj (x)ϕk (x)dx = 0, j ̸= k
0

(c) Each eigenfunction ϕj (x) has exactly j − 1 zeros in the interval (0, ℓ).

Expansion property: The eigenfunctions {ϕj (x)} form a complete set. Thus, if f (x) is piecewise
smooth, then we have

∞
X
f (x) = cn ϕn (x)
n=1
Rℓ
r(x)f (x)ϕn (x)dx
with cn = 0
Rℓ
0
r(x)ϕ2n (x)dx

Lagrange’s Identity For sufficiently differentiable functions u and v, Lagrange’s Identity states that
Z ℓ
ℓ
[vLu − uLv]dx = − p(x) (u′ v − uv ′ )|0
0

Proof. Write Z ℓ Z ℓ
′
v − (pu′ ) + qu dx
 
vLudx =
0 0
Z ℓ Z ℓ
′ ℓ ′ ′
= − vpu |0 + pu v dx + quvdx
0 0
Z ℓ
′ ℓ ′ ℓ ′
u − (pv ′ ) + qv dx
 
= − vpu |0 + upv |0 +
0
Z ℓ
ℓ ℓ
= − vpu′ |0 + upv ′ |0 + uLvdx
0

Hence,
8.1. PROPERTIES OF SL PROBLEMS 171

Z ℓ Z ℓ
ℓ
vLudx − uLvdx = − p(x) (u′ v − uv ′ )|0
0 0

If u and v satisfy the SL boundary conditions

α1 u(0) + α2 u′ (0) = 0, β1 u(ℓ) + β2 u′ (ℓ) = 0

α1 v(0) + α2 v ′ (0) = 0, β1 v(ℓ) + β2 v ′ (ℓ) = 0
then the boundary terms cancel and
Z ℓ Z ℓ
vLudx = uLvdx
0 0

Example 8.1.1.

1. We want to solve the eigenvalue problem

′ 2 1
(xy ′ ) + y = −λ y, x>0
x x
subject to the following boundary conditions

y ′ (1) = 0, y ′ (2) = 0
Note that r(x) = x1 . Expanding the derivative, we have

2 1
xy ′′ + y ′ + y = −λ y
x x
Multiply through by x to obtain the Cauchy-Euler type equation:

x2 y ′′ + xy ′ + (λ + 2)y = 0

The characteristic equation is

r2 + λ + 2 = 0
Case 1: λ + 2 < 0. p
We have two solutions r1,2 = ± −(λ + 2). The general solution is then,
√ √
y(x) = c1 x− −(λ+2) + c2 x −(λ+2)
The boundary conditions y ′ (1) = y ′ (2) = 0 implies that c1 = c2 = 0. Trivial solution.
Case 2: λ + 2 = 0.
We have a double solution r1 = r2 = 0. The general solution is then,

y(x) = c1 + c2 ln x
The boundary conditions y ′ (1) = 0 implies that c2 = 0 and so y(x) = c1 is a nontrivial solution.
Case 3: λ + 2 > 0.
Thus, the general solution is
√ √
y(x) = c1 cos( λ + 2 ln x) + c2 sin( λ + 2 ln x)
172 CHAPTER 8. STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

Next we apply the boundary conditions. y ′ (1) = 0 forces c2 = 0. This leaves

√
y(x) = c1 cos( λ + 2 ln x)
The second condition, y ′ (2) = 0, yields
√
sin( λ + 2 ln 2) = 0
This will give nontrivial solutions when
√
λ + 2 ln 2 = nπ, n = 1, 2, 3 . . .
In summary, the eigenfunctions for this eigenvalue problem are
 nπ 
yn (x) = cos ln x , 1 ≤ x ≤ 2
ln 2

2
and all (including λ = −2 ) the eigenvalues are λn = lnnπ2 − 2 for n = 0, 1, 2, . . .
We can check the orthogonality of eigenfunctions. We recall that
Z 2
⟨yn , ym ⟩ = yn (x)ym (x)r(x)dx
1

Let y = π ln x
ln 2
. Then, we have:
Z 2  nπ   mπ  dx
⟨yn , ym ⟩ = cos ln x cos ln x
1 ln 2 ln 2 x
Z π
ln 2
= cos(ny) cos(my)dy
π 0
ln 2
= δn,m
2
where
(
1, if n = m
δn,m =
0, ̸ m
if n =
is the so called Kronecker delta.
Now let us consider expanding a function f (x) in terms of a "Fourier Series" of these new eigen-
functions in the following form
∞ ∞  
X X nπ ln x
f (x) = cn ϕn (x) = cn cos
n=0 n=0
ln 2
In order to determine the coefficients cn we project the function f (x) onto the basis functions
ϕn (x) as follows:
Z 2   ∞ Z 2    
mπ ln x dx X mπ ln x nπ ln x dx
f (x) cos = cn cos cos
1 ln 2 x n=0 1 ln 2 ln 2 x
ln 2
= cm
2
Hence,
8.2. APPLICATION: SOLVING THE HEAT EQUATION WITH ROBIN BOUNDARY CONDITIONS173

Z 2  
2 nπ ln x dx
cn = f (x) cos
ln 2 1 ln 2 x

Note
• If the operator L and the boundary conditions satisfy

Z ℓ Z ℓ
vLudx = uLvdx
0 0
then L is said to be self adjoint.
Rℓ
• With the notation ⟨f, g⟩ = 0
f (x)g(x)dx, the selfadjoint property can be written as

⟨v, Lu⟩ = ⟨u, Lv⟩

8.2 Application: Solving the heat equation with Robin bound-

ary conditions
We consider the function u(x, t), which models the temperature distribution in a heat-conductive rod
of length L that is perfectly insulated along its sides. The left end of the rod is also perfectly insulated,
meaning no heat escapes or enters at x = 0. Meanwhile, the right end at x = L loses thermal energy
at a rate proportional to its temperature at that point. The initial temperature distribution along the
rod is given by the function f (x).

Example 8.2.1. The problem is stated as follows:

ut = α2 uxx , 0 < x < L, t > 0
ux (0, t) = 0, ux (L, t) + κu(L, t) = 0 (8.4)
u(x, 0) = f (x)

where κ > 0. We use the method of separation of variables to solve the problem (8.4). We first
seek separated solutions of the form u(x, t) = X(x)T (t) satisfying all of the homogeneous linear
requirements, i.e. the first three conditions. Substituting the separated solution into the PDE
yields
X ′′ T′
XT ′ = c2 X ′′ T ⇒ = 2 = λ (constant)
X cT
since the two sides of the latter equation are functions of distinct independent variables. This
gives us the pair of separated ODEs

X ′′ − λX = 0, T ′ − λc2 T = 0

From the first boundary condition we obtain

X ′ (0)T (t) = 0 ⇒ X ′ (0) = 0

since we do not want T ≡ 0. The second boundary condition requires that

174 CHAPTER 8. STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

X ′ (L)T (t) = −κX(L)T (t) ⇒ X ′ (L) = −κX(L)

We have thus obtained an ODE boundary value problem in X that requires a case by case analysis
of the possible values of k for which there are nontrivial (nonzero) solutions.
Case 1: λ = µ2 > 0.
In this situation the ODE for X becomes

X ′′ − µ2 X = 0

with characteristic equation

r 2 − µ2 = 0
whose roots are r = ±µ. The solutions are then given by

X = c1 eµx + c2 e−µx

The boundary conditions require that

µ c1 eµ·0 − c2 e−µ·0 = 0 ⇒ c1 − c2 = 0

µ c1 eµL − c2 e−µL = −κ c1 eµL + c2 e−µL ⇒ c1 (κ + µ)eµL + c2 (κ − µ)e−µL = 0

or in matrix form     
1 −1 c1 0
= .
(κ + µ)eµL
(κ − µ)e−µL c2 0
The determinant of the coefficient matrix is

(κ − µ)e−µL + (κ + µ)eµL = κ eµL + e−µL + µ eµL − e−µL > 0

which means that the only possibility is that c1 = c2 = 0, or in other words X ≡ 0.

Case 2: λ = 0.
Now the ODE in X simplifies to X ′′ = 0 which means that X = ax + b. The first boundary
condition immediately implies a = 0 and the second then becomes

0 = −κb ⇒ b = 0

which once again tells us that X ≡ 0. So we move on.

Case 3: λ = −µ2 < 0.
Things finally get interesting. The ODE becomes X ′′ + µ2 X = 0 whose characteristic equation has
roots ±iµ, so that X(x) = c1 cos(µx) + c2 sin(µx). From the first boundary condition we find

−c1 sin 0 + c2 cos 0 = 0 ⇒ c2 = 0

The second boundary condition is then

κ
−µc1 sin(µL) = −κc1 cos(µL) ⇒ tan µL =
µ
since we do not want c1 = 0 at this point. This equation has an increasing sequence of positive
solutions (We have proved it using a graphic), which we label

0 < µ1 < µ2 < µ3 < · · ·

8.2. APPLICATION: SOLVING THE HEAT EQUATION WITH ROBIN BOUNDARY CONDITIONS175

Finally, we obtain the nontrivial solutions

Xn = cos (µn x) , n∈N

Since −µ2 = λ, for each n ∈ N the ODE for T becomes

2
T ′ + (cµn )2 T = 0 ⇒ T ′ = − (cµn )2 T ⇒ T = Tn = cn e−(cµn ) t

We have finally obtained our separated solutions:

2
un (x, t) = cn e−(cµn ) t cos (µn x) , n∈N

where µn is the nth positive solution to tan µL = κ/µ.

Hence,
∞ ∞
X X 2
u(x, t) = un (x, t) = cn e−(cµn ) t cos (µn x)
n=1 n=1

Using the initial condition, we have

∞ ∞ ∞
−(cµn )2 ·0
X X X
f (x) = u(x, 0) = un (x, t) = cn e cos (µn x) = cn cos (µn x) (8.5)
n=1 n=1 n=1

Now, we want to show that the functions cos µn x are pairwise orthogonal on the interval [0, L].
Assume that for different eigenvalues µ2n and µ2m the corresponding eigenfunctions are given by

Xn (x) = cos (µn x) and Xm (x) = cos (µm x)

with µn ̸= µm .
Our goal is to show that these eigenfunctions are orthogonal:
Z L
Xn (x)Xm (x)dx = 0 for n ̸= m
0

We know that Xn (x) satisfies

Xn′′ (x) + µ2n Xn (x) = 0

If we multiply the latter identity by Xm (x) and integrate over [0, L] we obtain:
Z L Z L
′′ 2
Xm (x)Xn (x)dx + µn Xm (x)Xn (x)dx = 0
0 0
Let the first term in the above equation be:
Z L
I= Xm (x)Xn′′ (x)dx
0
An integration by parts gives:
Z L
L
I= [Xm (x)Xn′ (x)]0 − ′
Xm (x)Xn′ (x)dx
0
Using the boundary conditions, we get
176 CHAPTER 8. STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

Z L
′
I = −κXm (L)Xn (L) − Xm (x)Xn′ (x)dx
0
Returning to Substituting back we now have:
Z L Z L
′ ′ 2
−κXm (L)Xn (L) − Xm (x)Xn (x)dx + µn Xm (x)Xn (x)dx = 0
0 0
A second integration by parts along with the equation solved by Xm lead us to

L
Z
2 2
µn − µm Xn (x)Xm (x)dx = 0
0

Since µ2n ̸= µ2m for n ̸= m, we must have

Z L
Xn (x)Xm (x)dx = 0
0

Finally, we multiply equation (8.5) by cos (µm x) and integrate over [0, L]. Thanks to the orthog-
onality relation, we obtain
RL
f (x) cos (µn x) dx
cn = 0 R L
0
cos2 (µn x) dx

Bibliography
[1] Lectures notes of Prof. Anthony Peirce, https://personal.math.ubc.ca/~peirce/math257_316e14.
htm.
[2] Abada Nadjet. Equations différentielles du second ordre. Cours 4 ème année, 2019-2020.
[3] Boyce & DiPrima. Elementary Differential Equations and Boundary Value Problems.