Classroom Tips and Techniques:

Eigenvalue Problems for ODEs - Part 1

Robert J. Lopez
Emeritus Professor of Mathematics and Maple Fellow
Maplesoft

Initializations
O restart
O with LinearAlgebra :
with Student Calculus1
with RootFinding :

Introduction
Some boundary value problems for partial differential equations are amenable to analytic techniques.
For example, the constant-coefficient, second-order linear equations called the heat, wave, and
potential equations are solved with some type of Fourier series representation obtained from the
Sturm-Liouville eigenvalue problem that arises upon separating variables.
In this, and the next two articles in this series, we examine the role of Maple in the solution of such
boundary value problems. We will show efficient techniques for separating variables, then show
how to guide Maple through the solution of the resulting Sturm-Liouville eigenvalue problems. In
four previous columns we have demonstrated how to implement the Fourier series calculations in
Maple. Together, these articles represent the essence of the typical undergraduate course in boundary
value problems. In fact, the author's Advanced Engineering Mathematics electronic book includes a
complete course in the subject area.
In subsequent articles, we will explore separation of variables for Laplace's equation in cylindrical
coordinates, a process that leads to Bessel's equation as the ODE in the Sturm-Liouville eigenvalue
problem, which will now be singular and not regular as in this article. Finally, we will separate
variables for Laplace's equation in spherical coordinates so that the resulting ODE in the SturmLiouville eigenvalue problem will be Legendre's equation, again a singular problem. The difference
between these two problems is that for Bessel's equation, the resulting Bessel functions themselves
become eigenfunctions, whereas for Legendre's equation, it is far more difficult to pass from
Legendre functions in a general solution to the Legendre polynomials that are the eigenfunctions.

Separating Variables in a PDE

It is not difficult to implement separation of variables in Maple. For example, the one-dimensional
wave equation
O

v2
2

vt

u x, t = c2

v2
2

vx

u x, t

v2
v2
2
u
x,
t
=
c
u x, t
2
2
vt
vx
separates under the assumption that u x, t can be written as the product

(3.1)

O U := X x T t
U := X x T t

(3.2)

Substitution leads to
O (3.1)
u x, t = U
2

d
d
T t = c2
X x
2
dt
dx2
whereupon division by u x, t in its separated form yields
X x

O simplify

T t

(3.3)

(3.3)
U c2
d2
2

T t

dt
T t c2

d2

X x
2
dx
=
X x

(3.4)

Note that we have also divided through by c2, the wave speed. The Bernoulli separation constant is
now introduced, resulting in the two ordinary differential equations
O q1 d lhs (3.4) =
q2 d rhs (3.4) =
d2
T t
dt2
q1 :=
T t c2
d2
X x
dx2
q2 :=
X x
more commonly written as
O DE1 d numer normal lhs q2 Krhs q2

=0

DE2 d numer normal lhs q1 Krhs q1

=0

(3.5)

DE1 :=

d2
dx2

X x K X x = 0

d
2
T t K T t c = 0
(3.6)
2
dt
The first equation governing X x is a formally self-adjoint ODE that will become part of the SturmLiouville eigenvalue problems considered in the following section.
DE2 :=

Solving the Sturm-Liouville Eigenvalue Problem

Homogeneous Dirichlet Conditions
A Sturm-Liouville eigenvalue problem arises if we couple the differential equation
O DE1
d2
X x K X x = 0
dx2

(4.1.1)

with the homogeneous Dirichlet boundary conditions X 0 = X = 0. This is an example of

"separated" conditions for which we are guaranteed a complete set of orthogonal eigenfunctions.
One approach to solving this eigenvalue problem might be to invoke Maple's dsolve command,
with the result
O dsolve

DE1, X 0 = 0, X = 0 , X x
X x =0
Unfortunately, Maple returns the zero-solution, and does not even begin to solve the eigenvalue
problem. That is why we are writing this series of articles. Maple can actually be of great help in
solving such problems, but it must be guided by the user.
The characteristic equation for (4.1.1) is obtained by making the exponential guess
O EG := em x
mx

EG := e
Substitution and the obvious algebra then give
O

1
collect DE1
EG

, exp
X x = EG
m2 K = 0

It should be obvious that the characteristic roots are m =G , and that = 0 is the "repeated
root" case. Since the form of the general solution of (4.1.1) changes for this case, we consider it
first as a special case. The differential equation is then
O DE1a d DE1
=0

d2

DE1a :=

dx2

X x =0

and a solution of this equation with just the left-hand boundary condition is
O dsolve

DE1a, X 0 = 0 , X x
X x = _C1 x

(4.1.2)

Applying the right-hand boundary condition to this solution leads to the equation
O rhs (4.1.2)

=0
x=
_C1 = 0

from which it is clear that the multiplicative constant has to be zero. Since this leads to the trivial
solution X x = 0, we conclude that = 0 is not an eigenvalue.
We next consider the case s 0, and again use the strategy of applying just the left-hand
boundary condition. The resulting solution is
O X1 d rhs dsolve

DE1, X 0 = 0 , X x
X1 := K_C2 e

K x

C_C2 e

Because Maple could return this solution with _C1 as the arbitrary constant, we use the following
device to rewrite the solution in terms of the constant c. No matter what name Maple returns for
the arbitrary constant, our construction will always be in terms of c.
O param d remove has, indets X1 , , x
X2 d X1

1 :

param = c
X2 := Kc e

Cc eK

Applying the right-hand boundary condition leads to the algebraic equation

O X2

=0
x=
Kc e

Cc eK

=0

Without intervention, Maple will provide just the zero solution for , as we confirm with
O solve (4.1.3),
0
The requisite tool is the environment variable
O _EnvAllSolutions := true
_EnvAllSolutions := true
which, if set to the value true, enables Maple to return the solution

(4.1.3)

O solve (4.1.3),
2

K_Z1
(4.1.4)
Maple uses the notation _Zk, k = 1, 2,, to indicate an integer. Unfortunately, each time this
solve command is executed, the index k will increment by 1, so again, we cannot reference this
symbol in a fail-safe way. Thus, we resort to the following device
O temp d indets (4.1.4) 1 :
(4.1.4)
temp = n
2

Kn
to write the eigenvalues with a unique notation. The eigenvalues are the negative numbers
2
Kn , n = 1, 2,.
It is also possible to lead Maple through the steps resulting in this solution for the eigenvalues.
We begin by writing (4.1.3) in the form
O factor (4.1.3)
Kc e KeK = 0
Since c = 0 would result in the trivial solution X x = 0, we must have c s 0, so that

(4.1.5)

O map coeff, (4.1.5), c

Ke CeK = 0
This difference of exponentials suggests rewriting the left-hand side as

(4.1.6)

O convert (4.1.6), trig

K2 sinh

=0

(4.1.7)

from which we realize that if is real, the only solution of this equation will be = 0, again
leading to the trivial solution. This inspires the substitution
O simplify evalc (4.1.7)

assuming O 0
2

= K

K2 I sin = 0

(4.1.8)

from which it easily follows that is an integer, given in Maple by

O solve (4.1.8),
_Z2~

Homogeneous Neumann Conditions

The Sturm-Liouville eigenvalue problem that arises when the homogeneous (and separated)
Neumann boundary conditions X# 0 = X# = 0 are applied to the differential equation

O DE1
2

d
X x K X x = 0
2
dx
will have a different set of eigenvalues and eigenfunctions than we obtained with the Dirichlet
conditions. Since we already know that = 0 must be handled as a special case, we begin by
obtaining the general solution of the ODE
O DE1a
2

d
X x =0
2
dx
We again take precautions to write the solution with unique arbitrary constants, obtaining
O temp := rhs dsolve DE1a, X x

X0 d eval temp, _C1 = b, _C2 = a

X0 := b x Ca
To this general solution, we apply the two boundary conditions, resulting in the algebraic
equations
O X0 '

O X0 '

=0
x=0
b =0

=0

x=
b =0

and

Consequently, = 0 is an eigenvalue, and the corresponding eigenfunction is X0 x = a

Now, consider the case s 0. We again find the general solution, and write it with uniquely
named arbitrary constants.
O temp := rhs dsolve DE1, X x

X1 d eval temp, _C1 = c1, _C2 = c2

X1 := c1 e

Cc2 eK

To this general solution we apply the boundary conditions, obtaining the homogeneous algebraic
equations
O Eq1 d X1 '

O Eq2 d X1 '

=0
and

x=0
Eq1 := c1

Kc2

=0

=0
x=

Eq2 := c1

Kc2

eK

Maple's solve command cannot determine values of for which this system has nontrivial
solutions. We guide Maple by writing the system matrix as

=0

O GenerateMatrix

Eq1, Eq2 , c1, c2

1
K

(4.2.1)
e

K e

and setting its determinant to zero, the condition under which the system will have nontrivial
solutions for c1 and c2.
O Determinant (4.2.1) = 0
K

K e

C e

=0

(4.2.2)

Clearly, = 0 is a solution, but that belongs to the first case. "Canceling" and solving, gives
O solve

(4.2.2)
,

K_Z32

(4.2.3)

where again, we replace this notation by

O temp d indets (4.2.3) 1 :
(4.2.3)
temp = n
Kn2
so that the eigenvalues are more easily seen to be =Kn2, n = 1, 2,.
Having determined the eigenvalues, we return to the algebraic equations determining c1 and c2,
and make the obvious substitution for to obtain the equations
assuming n T posint;

O Eq3 d simplify Eq1

=Kn

=Kn

assuming n T posint

Eq4 d simplify Eq2

Eq3 := I c1 n KI c2 n = 0
Eq4 := I n K1

c1 Kc2 = 0

The solution of these equations is now

O solve

Eq3, Eq4 , c1, c2

c2 = c1, c1 = c1

so that the remaining eigenfunctions are

O temp := subs c1 = C, c2 = C, = Kn2, X1 :
simplify evalc temp

assuming n T posint
2 C cos n x

Of course, the constant C can be taken as

1
, or as
2

to obtain orthonormal eigenfunctions.

2

Example: An ODE that is Not Self-Adjoint

We conclude this article with an example of an eigenvalue problem in which the differential equation
is not self-adjoint. Such an example is not covered by the Sturm-Liouville theory, but it does
illustrate how Maple can be guided through some of the more difficult calculations that arise in
eigenvalue problems. In particular, note the more general separated (but still homogeneous)
boundary conditions in the following problem.
d2

d
y x
dx
dx
bc1 d 3 D y 0 C5 y 0 = 0

O deq :=

y x C2

bc2 d 2 D y

C y x = 0

Cy = 0
2

deq :=

d
d
y x C2
y x C y x = 0
2
dx
dx
bc1 := 3 D y 0 C5 y 0 = 0
bc2 := 2 D y

Cy = 0

To obtain solutions of the differential equation, we seek a fundamental set of linearly independent
members. Making the exponential guess leads to the characteristic equation
O

1
collect deq
, exp
EG
y x = EG
2

m C2 m C = 0

(5.1)

whose solutions are

O solve (5.1), m
K1 C

1 K , K1 K

1 K

(5.2)

From these characteristic roots, we can see that = 1 is the repeated-root case for which the solution
of the ODE assumes the special form
O temp := dsolve deq

,y x

=1
Y1 d unapply eval rhs temp , _C1 = r1, _C2 = s1 , x :
'Y1 ' x = Y1 x
Y1 x = r1 eKx Cs1 eKx x
For s 1, the general solution of the ODE can be written as
O temp := dsolve deq, y x

Y2 d unapply eval rhs temp , _C1 = r2, _C2 = s2 , x :

'Y2 ' x = Y2 x
Y2 x = r2 e

K1 C

K C 1 x

Cs2 e

K1 K

K C 1 x

Even though Maple defaults to an exponential form for the members of its fundamental set, we
should not assume that is characterized by anything other than the condition s 1. In fact, at this
point we don't even know if the eigenvalues are real. Consequently, we avoid separating the
calculation into the cases ! 1 and O 1 because that would imply that we knew was not a
complex number.
We let the boundary conditions determine by considering just the cases = 1 and s 1. In the first
case, we get the algebraic equations
O for k to 2 do
qk d bck
y = Y1
end do
q1 := 2 r1 C3 s1 = 0
K

q2 := Kr1 e Ks1 e

C2 s1 e

=0

and in the second,

O for k to 2 do
q2 C k := bck
y = Y2
end do
q3 := 3 r2 K1 C
q4 := 2 r2 K1 C
Cr2 e

K1 C

K C1
K C 1

K C1
e

K1 C

Cs2 e

C3 s2 K1 K

K C 1

K1 K

K C1

C2 s2 K1 K

K C 1

C5 r2 C5 s2 = 0

K C1

K1 K

K C 1

=0

The determinants of the system matrix in each case are then

O Q1 := Determinant GenerateMatrix

q1, q2 , r1, s1

Q1 := K2 eK C7 eK

(5.3)

and
O Q2 := collect Determinant GenerateMatrix
Q2 := 8 K7

K C1 K6 e

C6 eK 1 C

K1 C

K C 1

q3, q4 , r2, s2
C K8 K7

K C1

, exp
(5.4)

K C 1

respectively. Since the determinant in (5.3) is nonzero, we must have r1 = s1 = 0, so the corresponding

solution is X x h 0, and = 1 is therefore not an eigenvalue.

There are zeros of the determinant in (5.4), some of which are
O Analytic Q2, re = K10 ..10, im = K10 ..10
0.4558571180, 1.116356583, 4.243407308, 9.251574985, 1.000000060

(5.5)

The Analytic command in the RootFinding package finds all real and complex roots in a region of the
complex plane. We already know that = 1 is not an eigenvalue, so we have the real eigenvalues
O remove w/is w K1 ! 0.001 , (5.5)
0.4558571180, 1.116356583, 4.243407308, 9.251574985
(5.6)
While the Analytic command provides a guarantee that all the solutions in a fixed region have been
found, it is slow. If we accept (5.5) as evidence that the eigenvalues are positive and real, then we
can find them slightly faster by changing the form of the determinant in (5.4) to
O simplify evalc Q2
K2 I eK 7

assuming O 1

K1 cos

K1 K8 sin

K1 C6 sin

K1

(5.7)

and using the Roots command to find

O L := Roots

(5.7)

, = 0.1 ..26, numeric = true

K
Ie
L := 0.4558571181, 1.116356583, 4.243407309, 9.251574987, 16.25414949, 25.25529797
(5.8)
For any such eigenvalue, the equations q3 and q4 are redundant, so we can solve one of them for s2,
say, in terms of r2, obtaining
O S := solve q3, s2
S :=

r2 2 C3

K C1

K2 C3

K C1

For any eigenvalue , the corresponding eigenfunction is then given (up to a multiplicative constant)
by
O Yd

1
collect Y2 x
r2
Y := e

K1 C

, r2, exp
s2 = S
K C 1 x

2 C3

K C1
K2 C3

K1 K

K C 1 x

K C1

With O 1, the exponential terms will have trigonometric equivalents, so we write

O simplify evalc Y assuming O 1
1
K
2 eKx 9 cos K1 x K9 cos
K5 C9

K1 x C6

K1 sin

K1 x

(5.9)

K4 I sin

K1 x C6 I

K1 cos

K1 x

and then
O collect (5.9), sin, cos
Kx

2e

K1 K4 I sin
K5 C9

6
Kx

2e

K1 x

9 K9 C6 I K1
K5 C9

(5.10)
K1 x

cos

The real and imaginary parts are respectively

O Ya := evalc R (5.10)
Yb := evalc I (5.10)
Ya := K

12 eKx

assuming O 1;
assuming O 1
2 eKx 9 K9 cos K1 x
K
K5 C9

K1 sin
K1 x
K5 C9

8 eKx sin
K1 x
Yb :=
K5 C9

12 eKx

K1 cos K1 x
K5 C9

In fact, each of the real and imaginary parts is separately a solution. But they cannot be independent
solutions - they must be multiples of each other. This common multiple is
Ya
, x, 3 , polynom
Yb
3
f := K
K1
2
and we can verify that indeed, the real and imaginary parts of Y x are linearly dependent by
showing the linear combination f Yb KYa vanishes.
O f := simplify convert taylor

O simplify Yb f KYa
0
The common multiple can be found with
O g := Ya
x=0
g := K

2 9 K9
K5 C9

and a simplified form of the real part is then

O Yc := collect

Ya
, exp, cos, sin
g
Yc := cos

K1 x C

K1 sin
K1 x
9 K9

Using Yc, we obtain the first six eigenfunctions as

eKx

O unassign ''
1 := simplify expand convert Yc

, exp

assuming x O 0;

= L1
for k from 2 to nops L do
k d Yc
= Lk
end do
K0.2623395890 x
K1.737660411 x
1 := 0.0481208852 e
C0.9518791148 e
Kx

2 := cos 0.3411108075 x K1.954399128 sin 0.3411108075 x

3 := cos 1.800946226 x K0.3701757760 sin 1.800946226 x

4 := cos 2.872555480 x K0.2320813893 sin 2.872555480 x

eKx

5 := cos 3.905656090 x K0.1706926190 sin 3.905656090 x

6 := cos 4.924966799 x K0.1353647027 sin 4.924966799 x

Kx

Kx
Kx

A graph of the first six eigenfunctions is given in Figure 1, where our normalization has all
eigenfunctions passing through 0, 1 . The eigenfunctions k x , k = 1,, 6, are given in dotted
black, then solid black, blue, red, green, and magenta.
O plot

seq k, k = 1 ..6

, x = 0 .., color = black, black, blue, red, green,

magenta , linestyle = 2, 1$5

1.0

0.5

0
1

K0.5
Figure 1 Eigenfunctions k, k = 2. , 6 are in solid black, blue, red, green, and
magenta, respectively, whereas 0 is in dotted black
Since the differential equation is not self-adjoint, we are not assured of orthogonal eigenfunctions.
The inner products i, j , 1 % i, j % 6, are computed and displayed in Table 1. If these
eigenfunctions were orthogonal, the off-diagonal entries would be zero. That they are not is our
evidence the eigenfunctions are not orthogonal.

O Matrix 6, 6, i, j / i j dx
0

0.3099959961

0.2770259909

0.1891121866 0.1283061895 0.08823647716 0.06327843418

0.2770259909

0.2727275218

0.2129038605 0.1415643917 0.09585856523 0.06683519055

0.1891121866

0.2129038605

0.2557919318 0.2067341959 0.1341268143 0.08793179240

0.1283061895

0.1415643917

0.2067341959 0.2524088150 0.2037678526

0.1302398216

0.08823647716 0.09585856523 0.1341268143 0.2037678526 0.2511708133

0.2024635603

0.06327843418 0.06683519055 0.08793179240 0.1302398216 0.2024635603

0.2505873914

Table 1 The inner products i, j , 1 % i, j % 6

We expended some energy separating real and imaginary parts in the solution Y x , more in showing
that these parts were multiples of each other. A more direct way to find the simple trigonometric
form we used to generate the eigenfunctions is to obtain the general solution of the differential
equation under the assumption that O 1. This leads to
O temp := dsolve deq, y x assuming O 1 :
Y3 d unapply eval rhs temp , _C1 = r3, _C2 = s3 , x :
'Y3 ' x = Y3 x
Y3 x = r3 eKx sin

K1 x Cs3 eKx cos

K1 x

and, after applying the boundary conditions, to

O for k to 2 do
q4 C k := bck
y = Y3
end do
K1 C2 s3 = 0

q5 := 3 r3
q6 := Kr3 eK sin
K

K2 s3 e

sin

K1 C2 r3 eK cos
K1

K1

K1 Ks3 eK cos

K1

K1 = 0

If we now express r3, say, in terms of s3, as in

O isolate q5, r3
r3 = K

2
3

s3

(5.11)

K1

we can write this new general solution as

O Yd d

1
collect eval Y3 x , (5.11) , exp, s3
s3
Yd := K

2 sin
3

K1 x
K1

Ccos

K1 x

eKx

To show this is equivalent to our Yc x , we show the difference Yd KYc vanishes.

O simplify Yd KYc
0
Legal Notice: The copyright for this application is owned by the author(s). Neither Maplesoft nor the
author are responsible for any errors contained within and are not liable for any damages resulting
from the use of this material. This application is intended for non-commercial, non-profit use only.
Contact the author for permission if you wish to use this application in for-profit activities.