Jim Lambers

MAT 606
Spring Semester 2015-16
Lecture 20 Notes

These notes correspond to Section 6.4 in the text.

Properties of Sturm-Liouville Eigenfunctions and Eigenvalues

We continue our study of the Sturm-Liouville eigenvalue problem

L[v] = −[(p(x)u0 )0 + q(x)u] = λr(x)u, a < x < b, (1)

with boundary conditions

Ba [v] = Bb [v] = 0. (2)
In particular, we state and prove several useful properties of their solutions, which are eigenfunc-
tions, and their corresponding values of λ, which are eigenvalues.

Proposition 1 Let uλ and vλ be linearly independent solutions of (1) for the same value
of λ. Then λ is an eigenvalue of the Sturm-Liouville problem (1), (2) if and only if
 
Ba [uλ ] Ba [vλ ]
det = 0. (3)
Bb [uλ ] Bb [vλ ]

Proof: By assumption, the general solution of (1) is

w(x) = cuλ (x) + dvλ (x).

Let w be a nontrivial solution of (1). Then w is an eigenfunction of (1), (2) with eigenvalue λ if
and only if
        
Ba [w] cBa [uλ ] + dBa [vλ ] Ba [uλ ] Ba [vλ ] c 0
= = = .
Bb [w] cBb [uλ ] + dBb [vλ ] Bb [uλ ] Bb [vλ ] d 0

This system of linear equations has a nontrivial solution if and only if (3) is satisfied. 2

Example 1 Consider the Sturm-Liouville problem

v 00 + λv = 0, 0 < x < L,

v(0) = v 0 (L) = 0.
The general solution of this ODE is
√ √
v(x) = c cos( λx) + d sin( λx).

Therefore λ is an eigenvalue of this Sturm-Liouville problem if and only if

√ √
 
√ cos 0√ √ sin 0
√
det = λ cos( λL) = 0.
− λ sin( λL) λ cos( λL)

1
It follows that
√ 2n − 1
λL = π, n = 1, 2, . . . ,
2
which yields the eigenvalues
 2
(2n − 1)π
λn = , n = 1, 2, . . . .
2L
2

Symmetry
Let L be a regular Sturm-Liouville operator as defined in (1) and let u, v satisfy the boundary
conditions (2). We then have

uL[v] − vL[u] = u[−(p(x)v 0 )0 − q(x)v] − v[−(p(x)u0 )0 − q(x)u]

= −u(p(x)v 0 )0 − q(x)uv + v(p(x)u0 )0 + q(x)vu
= v(p(x)u0 )0 − u(p(x)v 0 )0

Using integration by parts, we obtain

Z b Z b
uL[v] − vL[u] dx = v(p(x)u0 )0 − u(p(x)v 0 )0 dx
a a
Z b
b
= p(x)(vu0 − uv 0 )a − v 0 p(x)u0 − u0 p(x)v 0 dx
a
0 0 b

= p(x)(vu − uv ) a ,

which is known as Green’s identity.

From the fact that u and v satisfy the boundary conditions,

αu(a) + βu0 (a) = 0, αv(a) + βv 0 (a) = 0,

γu(b) + δu0 (b) = 0, γv(b) + δv 0 (b) = 0.

Multiplying the boundary conditions for u by v and vice versa, we obtain

α[v(a)u(a) − u(a)v(a)] + β[v(a)u0 (a) − u(a)v 0 (a)] = β[v(a)u0 (a) − u(a)v 0 (a)] = 0,

γ[v(b)u(b) − u(b)v(b)] + δ[v(b)u0 (b) − u(b)v 0 (b)] = δ[v(b)u0 (b) − u(b)v 0 (b)] = 0,
and therefore
Z b
uL[v] − vL[u] dx = p(b)[v(b)u0 (b) − u(b)v 0 (b)] − p(a)[v(a)u0 (a) − u(a)v 0 (a)] = 0.
a

That is,
hu, L[v]i = hL[u], vi.
It follows that L is its own adjoint; we say that L is self-adjoint or symmetric; this is analogous to a
square matrix being symmetric. It can be shown that a Sturm-Liouville operator is also self-adjoint
in the case of periodic boundary conditions.

2
Real Eigenvalues
Just as a symmetric matrix has real eigenvalues, so does a (self-adjoint) Sturm-Liouville operator.

Proposition 2 The eigenvalues of a regular or periodic Sturm-Liouville problem are real.

Proof: As before, we consider the case of a regular Sturm-Liouville problem; the periodic case
is similar. Let v be an eigenfunction of the problem (1), (2) with eigenvalue λ. Then

hv, L[v]i = hv, λvi = λkvk2 .

Similarly,
hL[v], vi = λkvk2 .
However, by the symmetry of L, hv, L[v]i = hL[v], vi, which means λ = λ. We conclude that λ is
real. 2

Orthogonality
Just as a symmetric matrix has orthogonal eigenvectors, a (self-adjoint) Sturm-Liouville operator
has orthogonal eigenfunctions.

Proposition 3 Let v1 and v2 be eigenfunctions of a regular Sturm-Liouville operator (1)

with boundary conditions (2) corresponding to distinct eigenvalues λ1 , λ2 , respectively.
Then v1 and v2 are orthogonal with respect to the weight function r(x); that is,
Z b
hv1 , v2 ir = v1 (x)v2 (x)r(x) dx = 0.
a

This property also holds with periodic boundary conditions.

Proof: We consider the case of a regular Sturm-Liouville problem; the periodic case is similar.
From the relations
L[v1 ] = λ1 rv1 , L[v2 ] = λ2 rv2 ,
and the symmetry of L, we obtain

0 = hv1 , L[v2 ]i − hL[v1 ], v2 i

= hv1 , λ2 rv2 i − hλ1 rv1 , v2 i
= (λ2 − λ1 )hv1 , v2 ir ,

where we have used the fact that the eigenvalues are real. Because λ2 6= λ1 , we must have hv1 , v2 ir =
0. 2

Example 2 The regular Sturm-Liouville problem

v 00 + λv = 0, 0 < x < L,

with Dirichlet boundary conditions

v(0) = v(L) = 0

3
has eigenvalues and eigenfunctions
 nπ 2 nπx
λn = , vn (x) = sin , n = 1, 2, . . . .
L L
It can be verified using product-to-sum identities that for m, n = 1, 2, . . . ,
Z L 
nπx mπx 0 n 6= m
hvn , vm i = sin sin dx = L .
0 L L 2 n=m

We see that these eigenfunctions are orthogonal, and that the set
(r )∞
2 nπx
sin
L L
n=1

consists of orthonormal eigenfunctions. 2

Example 3 The regular Sturm-Liouville problem

v 00 + λv = 0, 0 < x < L,

with Neumann boundary conditions

v 0 (0) = v 0 (L) = 0

has eigenvalues and eigenfunctions

 nπ 2 nπx
λn = , vn (x) = cos , n = 0, 1, 2, . . . .
L L
It can be verified using product-to-sum identities that for m, n = 0, 1, 2, . . . ,

Z L
nπx mπx  0 n 6= m
L
hvn , vm i = cos cos dx = 2 n = m 6= 0 .
0 L L
L n=m=0


We see that these eigenfunctions are orthogonal, and that the set
(r ) (r )∞
1 2 nπx
∪ cos
L L L
n=1

consists of orthonormal eigenfunctions. 2

Example 4 The regular Sturm-Liouville problem

v 00 + λv = 0, 0 < x < L,

with periodic boundary conditions

v(0) = v(L), v 0 (0) = v 0 (L)

has eigenvalues and eigenfunctions

2nπ 2
 
2nπx
λn = , un (x) = cos , n = 0, 1, 2, . . . ,
L L

4
 2nπx
vn (x) = sin
, n = 1, 2, . . . .
L
It can be verified using product-to-sum identities that for m, n = 0, 1, 2, . . . ,

Z L
2nπx 2mπx  0 n 6= m
L
hun , um i = cos cos dx = 2 n = m 6= 0 ,
0 L L
L n=m=0


and for m, n = 1, 2, . . . ,,
Z L 
2nπx 2mπx 0 n 6= m
hvn , vm i = sin sin dx = L ,
0 L L 2 n=m

and for n = 0, 1, 2, . . ., m = 1, 2, . . . ,
Z L
2nπx 2mπx
hun , vm i = cos sin dx = 0.
0 L L

We see that these eigenfunctions are orthogonal, and that the set
(r ) (r )∞ (r )∞
1 2 2nπx 2 2nπx
∪ cos ∪ sin
L L L L L
n=1 n=1

consists of orthonormal eigenfunctions. 2

Real Eigenfunctions
The eigenfunctions of a Sturm-Liouville problem can be chosen to be real.

Proposition 4 Let λ be an eigenvalue of a regular or periodic Sturm-Liouville problem.

Then the subspace spanned by the eigenfunctions corresponding to λ admits an orthonor-
mal basis of real-valued functions.

Proof: The result is trivially true if λ is a simple eigenvalue. If λ has multiplicity 2, which
is the maximum possible since the Sturm-Liouville ODE is second-order, then λ has two linearly
independent eigenfunctions
v1 = a1 + ib1 , v2 = a2 + ib2 .
Because L has real coefficients, it can easily be shown that a1 , a2 , b1 , b2 are all eigenfunctions of L
corresponding to λ.
Suppose that from the set {a1 , a2 , b1 , b2 }, there are not two linearly independent functions. Then
all four functions are scalar multiples of one another, but then it follows that v1 is a scalar multiple
of v2 , which contradicts the assumption that v1 and v2 are linearly independent. Thus two functions
from {a1 , a2 , b1 , b2 } are linearly independent, and by applying Gram-Schmidt orthogonalization to
these two functions, two real-valued orthonormal eigenfunctions can be obtained. 2

5
Simple Eigenvalues
The following property regarding the multiplicity of eigenvalues greatly simplifies their numerical
computation.

Proposition 5 The eigenvalues of a regular Sturm-Liouville problem are simple.

Proof: Let v1 and v2 be eigenfunctions of the regular Sturm-Liouville problem (1), (2) with
eigenvalue λ. Then we have
v1 L[v2 ] − v2 L[v1 ] = −v1 [(p(x)v20 )0 + q(x)v2 ] + v2 [(p(x)v10 )0 + q(x)v1 ]
= v2 (p(x)v10 )0 − v1 (p(x)v20 )0
= v2 p0 (x)v10 + v2 p(x)v100 − v1 p0 (x)v20 − v1 p(x)v200
= p(x)[v2 v100 − v1 v200 ] + p0 (x)[v2 v10 − v1 v20 ]
= [p(x)(v2 v10 − v1 v20 )]0 .
However, we also have
v1 L[v2 ] − v2 L[v1 ] = v1 λv2 − v2 λv1 = λ(v1 v2 − v2 v1 ) = 0.
It follows that Q(x) = p(x)(v2 v10 − v1 v20 ) is a constant function. But because v1 and v2 both satisfy
the boundary conditions, we have Q(a) = Q(b) = 0. Therefore, Q(x) ≡ 0 and
W (v1 , v2 ) = v1 v20 − v2 v10 = 0.
We conclude that v1 and v2 are linearly dependent. 2
Note that this result only applies to regular Sturm-Liouville problems; for periodic problems, recall
that most eigenvalues have multiplicity 2.

Countably Infinite Eigenvalues

The following essential result characterizes the behavior of the entire set of eigenvalues of Sturm-
Liouville problems.

Proposition 6 The set of eigenvalues of a regular Sturm-Liouville problem is countably

infinite, and is a monotonically increasing sequence

λ0 < λ1 < λ2 < · · · < λn < λn+1 < · · ·

with limn→∞ λn = ∞. The same is true for a periodic Sturm-Liouville problem, except
that the sequence is monotonically nondecreasing.
The difference in behavior of the eigenvalues between the regular and periodic problems is due to
the fact that the eigenvalues of a regular problem are simple, whereas for the periodic case they
can have multiplicity 2.
The following result follows from the preceding proposition, as well as earlier results pertaining
to the eigenfunctions. Recall that Er (a, b) is the space of piecewise continuous functions on [a, b]
with inner product h, ir , where r(x) is the weight function from (1).

Corollary 1 A regular or periodic Sturm-Liouville problem admits an orthonormal se-

quence of real-valued eigenfunctions in Er (a, b). Furthermore, the sequence of eigenvalues
is not bounded above, but is bounded below.

6
Completeness
The eigenfunctions of a Sturm-Liouville problem can be used to describe piecewise continuous
functions, which is very useful for solving time-dependent PDE for which separation of variables
yields a Sturm-Liouville problem.

Proposition 7 The orthonormal set {vn }∞ n=0 of eigenfunctions of a regular or periodic

Sturm-Liouville problem is a basis for Er (a, b); that is, Er (a, b) is complete.

The expansion of a function v ∈ Er (a, b) in the orthonormal basis of eigenfunctions, given by

X∞ Z b
v= an vn , an = hvn , vir = vn (x)v(x)r(x) dx,
n=0 a

is called an eigenfunction expansion of v.

The eigenfunction expansion has these essential properties.

Proposition 8 Let {vn }∞ n=0 be an orthonormal set of eigenfunctions of a regular or pe-

riodic Sturm-Liouville problem.

1. If f is continuous and piecewise differentiable on [a, b] and satisfies the boundary

conditions of the Sturm-Liouville problem, then the eigenfunction expansion of f
converges uniformly to f on [a, b].

2. If f is piecewise differentiable on [a, b], then for x ∈ (a, b) the eigenfunction expan-
sion of f converges to [f (x+ ) + f (x− )]/2, where f (x+ ) and f (x− ) are the left- and
right-hand limits of f at x.

Example 5 We compute the expansion of f (x) = 1 in the orthonormal basis { 2/L sin(nπx/L)}∞
p
n=1 ,
which are eigenfunctions of the Sturm-Liouville problem
v 00 + λv = 0, 0 < x < L, v(0) = v(L) = 0.
We have r ∞
2X nπx
f (x) = 1 = bn sin ,
L L
n=1
where
*r +
2 nπx
bn = sin ,1
L L
r Z L
2 nπx
= sin dx
L 0 L
nπx L
r 
2 L
= − cos
L nπ L 0
√
2L
= − [(−1)n − 1]
√ nπ
2L
= [1 − (−1)n ]
(nπ
0√ n even
= 2 2L .
nπ n odd

7
This yields
r ∞ √ ∞
2 X 2 2L (2k − 1)πx 4X 1 (2k − 1)πx
f (x) = sin = sin .
L (2k − 1)π L π 2k − 1 L
k=1 k=1

This series expansion converges to f (x) = 1 on (0, L), but it does not converge uniformly on [0, L],
because the boundary conditions are not satisfied. Due to this compatibility between f (x) and the
boundary condition, truncated expansions exhibit oscillations at x = 0 and x = L characteristic of
Gibbs’ phenomenon. 2

Example 6 We compute the expansion of f (x) = x in the orthonormal basis { 1/L}∪{ 2/L cos(nπx/L)}∞
p p
n=1 ,
which are eigenvalues of the Sturm-Liouville problem

v 00 + λv = 0, 0 < x < L, v 0 (0) = v 0 (L) = 0.

We have r r ∞
1 2X nπx
f (x) = x = a0 + an cos
L L L
n=1

where *r + r Z r
1 1 L 1 L2 L3/2
a0 = ,x = x dx = = ,
L L 0 L 2 2
and, for n = 1, 2, . . . ,
*r +
2 nπx
an = cos ,x
L L
r Z L
2 nπx
= x cos dx
L 0 L
r  Z L 
2 L nπx L L nπx
= x sin  − sin dx
L nπ L 0 nπ 0 L
r L
2 L2 nπx 
= cos
L (nπ)2 L 0
r
2 L2
= [(−1)n − 1]
L (nπ)2
(
0 √ n even
= 2L 2L .
− (nπ)2 n odd

We conclude that
r r ∞ √ ∞
1 L3/2 2 X 2L 2L (2k − 1)πx L 4L X 1 (2k − 1)πx
f (x) = − 2
cos = − 2 2
cos .
L 2 L ((2k − 1)π) L 2 π (2k − 1) L
k=1 k=1

This expansion converges uniformly to x on [0, π], even though it does not satisfy the boundary
conditions. Truncated expansions do not exhibit Gibbs’ phenomenon. 2

8
Rayleigh Quotients
We now develop a useful technique for estimating eigenvalues, which is very useful for numerical
computation.

Definition 1 The principal eigenvalue, also known as the ground state energy, of
a Sturm-Liouville problem is the minimal eigenvalue λ0 . The principal eigenfunction
is the eigenfunction corresponding to the principal eigenvalue.

Definition 2 Let L be the differential operator from (1). The expression

Rb
hu, L[u]i uL[u] dx
R(u) = = Ra b
hu, uir 2
a u r dx

is called the Rayleigh quotient of u.

Proposition 9 The principal eigenvalue λ0 of a regular Sturm-Liouville problem (1), (2)

satisfies the variational principle, known as the Rayleigh-Ritz formula:

λ0 = inf R(u),
u∈V,u6=0

where V is the space of all twice continuously differentiable functions on [a, b] that satisfy
the boundary conditions (2).

Proof: Using the orthonormality and completeness of the eigenfunctions, as well as the mono-
tonicity of the eigenvalues, we obtain
hu, L[u]i
R(u) =
hu, ui
P∞ r
h m=0 am vm , L [ ∞
P
a v ]i
= P∞n=0 n n
h ∞
P
a v
m m , n=0 n vn ir
a
P∞ m=0 P∞
am an hvm , L[vn ]i
= Pm=0∞ Pn=0∞
am an hvm , vn ir
P∞m=0P∞n=0
a a hv , λ rv i
= m=0 P∞ m n 2m n n
n=0
|an |
P∞ P∞ n=0
a a λ hv , v i
= m=0 P∞ m n 2n m n r
n=0
n=0 |an |
P∞ 2
n=0 |an | λn
= ∞
|an |2
P
P∞n=0 2
n=0 |an | λ0
≥ P ∞ 2
n=0 |an |
∞
|an |2
P
≥ λ0 Pn=0
∞ 2
n=0 |an |
≥ λ0 .

9
If we choose u = v0 , then R(u) = λ0 . This proves the result. It is important to note that we have
used the fact that u is twice continuously differentiable to conclude that its eigenfunction expansion
converges uniformly on [a, b], which allows term-by-term integration and differentiation. 2
Using integration by parts, the Rayleigh-Ritz formula can be rewritten as follows:

λ0 = infR(u)
u∈V,u6=0
Rb 0 0
a −u(p(x)u ) − uq(x)u dx
= inf Rb
u∈V,u6=0 2
a u r dx
Rb 0 2 2 0 b
a p(x)(u ) − q(x)u dx − puu |a
= inf Rb .
u∈V,u6=0 2
a u r dx

This leads to the following result.

Corollary 2 If q ≤ 0 and puu0 |ba ≤ 0 for u ∈ V , then the eigenvalues of the Sturm-
Liouville problem are nonnegative. In particular, the eigenvalues are nonnegative for the
Dirichlet, Neumann and periodic Sturm-Liouville problems.

Example 7 Consider the Sturm-Liouville problem

v 00 + λv = 0, 0 < x < 1, v(0) = v(1) = 0.

The principal eigenfunction is v0 (x) = sin πx, with corresponding eigenvalue λ0 = π 2 . We can
estimate this eigenvalue using a test function u(x) = x − x2 , which, like sin πx, has roots at x = 0, 1
and is concave down on (0, 1). We have
R1
− (x − x2 )(−2) dx
R(u) = R0 1 = 10 ≥ π 2 ≈ 9.87.
(x − x 2 )2 dx
0

That is, the Rayleigh quotient yields a upper bound of the principal eigenvalue. 2

Zeros of Eigenfunctions

Proposition 10 The nth eigenfunction vn of a regular Sturm-Liouville problem has ex-

actly n roots on the interval (a, b).

10