Sturm-Liouville Problems

Theory and Numerical Implementation

Monographs and Research Notes in Mathematics

Series Editors:
John A. Burns, Thomas J. Tucker, Miklos Bona, Michael Ruzhansky

Actions and Invariants of Algebraic Groups, Second Edition

Walter Ricardo Ferrer Santos, Alvaro Rittatore

Lineability
The Search for Linearity in Mathematics
Richard M. Aron, Luis Bernal-Gonzalez, Daniel M. Pellegrino, Juan B. Seoane Sepulveda

Iterative Methods and Preconditioning for Large and Sparse Linear Systems with
Applications
Daniele Bertaccini, Fabio Durastante

Monomial Algebras, Second Edition

Rafael Villarreal

Matrix Inequalities and Their Extensions to Lie Groups

Tin-Yau Tam, Xuhua Liu

Elastic Waves
High Frequency Theory
Vassily Babich, Aleksei Kiselev

Difference Equations
Theory, Applications and Advanced Topics, Third Edition
Ronald E. Mickens

Sturm-Liouville Problems
Theory and Numerical Implementation
Ronald B. Guenther, John W. Lee

For more information about this series please visit: https://www.crcpress.com/Chapman–HallCRC-

Monographs-and-Research-Notes-in-Mathematics/book-series/CRCMONRESNOT
Sturm-Liouville Problems
Theory and Numerical Implementation

Ronald B. Guenther
John W. Lee
Department of Mathematics
Oregon State University Corvallis
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2019 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-138-34543-0 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made
to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all
materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all
material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not
been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized
in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying,
microfilming, and recording, or in any information storage or retrieval system, without written permission from
the publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.com (http:==
www.copyright.com=) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA
01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users.
For organizations that have been granted a photocopy license by the CCC, a separate system of payment has
been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for iden-
tification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data

Names: Guenther, Ronald B., author. | Lee, John W., 1942- author.
Title: Sturm Liouville problems : theory and numerical implementation / R.B. Guenther,
J.W. Lee (Department of Mathematics, Oregon State University, Corvallis, OR).
Description: Boca Raton, Florida : CRC Press, 2018. | Series: Monographs and research
notes in mathematics | Includes bibliographical references and index.
Identifiers: LCCN 2018035973| ISBN 9781138345430 (hardback : alk. paper) |
ISBN 9780429437878 (ebook)
Subjects: LCSH: Sturm-Liouville equation. | Differential equations. | Eigenvalues.
Classification: LCC QA372 .G84 2018 | DDC 515/.352--dc23
LC record available at https://lccn.loc.gov/2018035973

Visit the Taylor & Francis Web site at

http:=== www.taylorandfrancis.com

and the CRC Press Web site at

http:=== www.crcpress.com
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

1 Setting the Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Euler Buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Hanging Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Vibration Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.1 Vibrations of a String . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.2 Vibrations of a Circular Membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.3 Spherically Symmetric Vibrations in a Ball . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Diffusion Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5.1 Chemical Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.2 Heat Conduction in a Rod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.3 Heat Conduction in a Disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Steady State Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6.1 Heat Conduction in a Rectangular Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.2 Heat Conduction in a Circular Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 On Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Sturm-Liouville Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.9 Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.10 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.11 The Path Ahead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.11.1 Thread I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.11.2 Thread II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.11.3 Finding Eigenvalues and Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.12 Intrinsic Interest of Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.13 Real Versus Complex Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25
2.1 Euclidean Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Real Euclidean Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.2 Complex Euclidean Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1.3 Elements of Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.4 Upper Bounds and Sups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.5 Closed and Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Calculus and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Differential Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.3 Integral Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.4 Sequences and Series of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

v
vi Contents

2.3 Matrix and Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.3.1 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.2 Systems of Linear Algebraic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.3 Linear Dependence and Linear Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.4 Eigenvalues and Eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.5 Self-Adjoint and Symmetric Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3.6 Principal Axis Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.7 Matrices as Linear Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4 Interpolation and Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4.1 Tchebycheff Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.4.2 Total Positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.5 Linear Spaces and Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.5.1 Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.5.2 Normed Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.5.3 Inner Product Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.5.3.1 Gram-Schmidt Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.6 Completeness and Completion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.7 Compact Sets in C[a, b] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.8 Contraction Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.9 Bisection and Newton-Raphson Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.9.1 Bisection Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.9.2 Newton-Raphson Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.10 Maximum Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3 Integral Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .77

3.1 Integral Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2 More General Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.3 Eigenvalues of Operators and Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4 Self-Adjoint Operators and Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.4.1 Hilbert-Schmidt Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.4.2 Mercer’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.5 Nonnegative Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.5.1 Positive Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .101
3.5.2 Kernels Positive on the Open Diagonal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .104
3.5.3 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .108
3.6 Kellogg Kernels and Total Positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.6.1 Compound Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .109
3.6.2 Spectral Properties of Compound Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .113
3.6.3 Spectral Properties of Kellogg Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .116
3.7 Singular Kellogg Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.7.1 Compound Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .120
3.7.2 Spectral Properties of Compound Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .124
3.7.3 Spectral Properties of Kellogg Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .125

4 Regular Sturm-Liouville Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

4.1 Sturm-Liouville Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.2 Sturm-Liouville Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.3 Initial Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
 Contents vii

4.3.1 Basis of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .137

4.3.2 Variation of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .138
4.3.3 Continuous Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .139
4.4 BVPs and EVPs - Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5 BVPs and EVPs - Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.6 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.6.1 Separated Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .157
4.6.2 Mixed Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .167
4.7 Adjoint Operators and Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.7.1 Separated Adjoint Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .174
4.7.2 Mixed Adjoint Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .178
4.8 Eigenvalue Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.8.1 Recasting the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .183
4.8.2 Separated Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .185
4.8.2.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .186
4.8.2.2 Case 1: Weight Function r(x) = 1 for all x in [a, b] . . . . . . . . . . . .186
4.8.2.3 Case 2: r(x) is a General Weight Function . . . . . . . . . . . . . . . . . . . .192
4.8.3 Oscillation and Approximation Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .196
4.8.4 Rayleigh Quotient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .200
4.8.5 Mixed Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .201

5 Singular Sturm-Liouville Problems - I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

5.1 Properties of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
5.2 Initial Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
5.3 Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.4 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
5.5 Eigenvalue Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
5.5.1 Fundamental Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .234
5.5.2 Oscillation and Approximation Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .242
5.5.3 Rayleigh Quotient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .246
5.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

6 Singular Sturm-Liouville Problems - II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

6.1 Properties of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.2 Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
6.3 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
6.4 Eigenvalue Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
6.4.1 Fundamental Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .284
6.4.2 Oscillation and Approximation Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .291
6.4.3 Rayleigh Quotient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .296
6.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298

7 Approximation of Eigenvalues and Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . 299

7.1 Regular Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
7.1.1 The Shooting Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .302
7.1.2 Bisection Method and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .303
viii Contents

7.1.3 Newton’s Method and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .306

7.1.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .306
7.2 Singular Problems - I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
7.2.1 The Shooting Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .315
7.2.2 Bisection Method and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .319
7.2.3 Newton’s Method and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .321
7.2.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .321
7.2.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .329
7.3 Singular Problems - II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
7.3.1 The Shooting Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .332
7.3.2 Bisection Method and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .336
7.3.3 Newton’s Method and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .338
7.3.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .339
7.3.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .347

8 Concluding Examples and Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349

8.1 Hanging Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
8.2 Vibrating Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
8.3 Vibrating Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
8.3.1 Homogeneous Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .366
8.3.2 Inhomogeneous Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .368

A Mildly Singular Compound Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373

A.1 Mildly Singular Kernels of Type (i) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
A.2 Mildly Singular Kernels of Type (ii) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378

B Iteration of Mildly Singular Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383

B.1 Mildly Singular Behavior of Type (i) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
B.2 Mildly Singular Behavior of Type (ii) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
B.3 Iterated Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387

C The Kellogg Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389

C.1 Consequences of Conservation of Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
C.2 Consequences of H2 and H2* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
C.3 Consequences of H4 (H2 and H2*) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
Preface

This book on Sturm-Liouville problems is written for scientists and engineers, for applied
mathematicians, and for advanced undergraduate and graduate students in these fields. We
have endeavored to keep the level of mathematical precision at an accessible level for all read-
ers. A reader with the logical reasoning skills acquired in a course in Euclidean geometry, a
beginning course in matrix and linear algebra, and a good course in calculus can profitably
read this book.
Scientists and engineers may find this book most useful as a reasonably-comprehensive,
one-stop reference for the main results about Sturm-Liouville boundary value problems and
eigenvalue problems that are needed for real-world applications. The book also can serve as
a text for a capstone course in applied mathematics for advanced undergraduate and beginning
graduate students. It gives these students valuable insight into how abstract theorems of anal-
ysis and linear algebra, that are typically covered in isolation, were motivated and discovered
by the need to carefully analyze significant applied problems.
We have endeavored to choose topics that will be most useful to scientists, engineers, and
applied mathematicians who encounter Sturm-Liouville problems in their modeling work.
Some readers will want to use results from the book, such as existence theorems, continuous
dependence, convergence of eigenfunction expansions, and error analyses of numerical meth-
ods, but are not especially interested in the proofs. Other readers will want the proofs. The
book is organized to accommodate both groups, with some guidance given along the way.
Chapters 4, 5, 6, and 7 that cover regular Sturm-Liouville problems, two types of singular
problems, and the numerical approximation of eigenvalues and eigenfunctions by shooting
methods are written to substantially reduce the dependence of one of the chapters on its pre-
decessors. Thus, a reader primarily interested in a topic in Chapter 6, say, will find little need to
consult prior chapters for essential background material.
Sturm-Liouville problems arise naturally in engineering, physics, and, more recently, in
biology and the social sciences. These problems lead to eigenvalue problems for ordinary
and partial differential equations. This book addresses, in a unified way, the key issues that
must be faced in science and engineering applications when separation of variables, variational
methods, or other considerations lead to Sturm-Liouville eigenvalue problems and boundary
value problems. In addition, effective numerical procedures for the approximation of eigenval-
ues and eigenfunctions of both regular and singular Sturm-Liouville problems are presented.
Such procedures are essential because explicit evaluation of the eigenvalues and eigenfunctions
is rarely possible.
Both regular and singular problems are treated with a high level of rigor and an emphasis
on the types of problems that actually arise in mathematical modeling. Our treatment often
follows familiar lines but also contains new results, especially in the chapters dealing with sin-
gular problems and in the careful justification of shooting methods that can be used to find
accurate numerical approximations to eigenvalues and eigenfunctions of both regular and sin-
gular problems. A significant feature of the book is its treatment of singular problems: proper-
ties of solutions established for singular problems that involve Bessel functions or other special
functions are shown to hold for two classes of singular Sturm-Liouville problems that embrace
the special cases.

ix
x Preface

Regular and singular problems are treated with a high level of rigor for more than mathe-
matical reasons: it is an essential element of good mathematical modeling. The quality and
accuracy of predictions obtained from a mathematical model depend upon the precision
with which physical concepts are incorporated in the model, the effectiveness of those concepts
at capturing the principal physical attributes of the real-world situation, and a clear under-
standing of the likely impact of the simplifying assumptions made to derive the governing
equations in the model. Solid physical reasoning and rigorous mathematics lead to better mod-
els and better models to better predictions. A rigorous physical model solved by rigorous math-
ematical methods enables the user to determine under what conditions the model holds and to
have confidence in the predictions it makes. All of the models in Chapter 1 and later chapters
can be derived from a careful interplay of underlying physical laws and accompanying math-
ematical precision. This book concentrates on the properties (predictions, behavior) of solu-
tions to such models when Sturm-Liouville boundary value and eigenvalue problems arise.
For example, a model for the vibrations of a violin string or a drum head leads to such
Strum-Liouville problems. It must be proved that the model equations predict the oscillatory
behavior observed in the real world. Moreover, the mathematical analysis of the model should
lead to a better understanding of the physical situation being studied by adding precision to
our understanding of expected behaviors and, ideally, by predicting behavior not previously
observed. Without a rigorously derived mathematical theory to support the modeling none
of the foregoing can be accomplished.
During the 19th century and the first half of the 20th century a number of analytical tech-
niques and an extensive theory for dealing with Sturm-Liouville problems were developed. As
long as the equations had constant coefficients or were of a special form such as a Bessel or
hypergeometric equation, the problems could be dealt with by hand, in the sense that the
eigenfunctions could be expressed in terms of special functions and the eigenvalues could be
expressed in terms of the zeros of such functions. Asymptotic expansions, series representa-
tions, and other specialized methods made it possible to calculate numerical approximations
to the first few eigenvalues and eigenfunctions. Eigenfunction expansions, whose coefficients
involved integrals, were often hard to find explicitly and laborious to evaluate numerically.
Moreover, if the medium was not homogenous, explicit solutions could rarely be found and
one had to revert to approximations which were computationally intensive. All in all, numer-
ical calculations were quite challenging.
The availability of digital computers began to change the picture dramatically by the mid-
dle of the 20th century, with increasing effect in subsequent decades. The advent of the modern
computer seemed to be a quick and easy way out of all the difficulties mentioned above. One
could reduce the differential equation to a system of difference equations and let the computer
solve the resulting linear system of equations. The eigenvalues of a matrix eigenvalue problem
approximated the (first few) eigenvalues of the Sturm-Liouville problem. It turned out that
finding the eigenvalues in this way could be difficult, especially because after finding the first
two or three eigenvalues, the approximation of subsequent eigenvalues tended to decreases
noticeably in accuracy. Often this was the result of use of a numerical orthogonalization pro-
cedure. An alternative approach is to use the Ritz-Galerkin method. This technique yields an
approximation to the first eigenvalue and gives an approximation to a corresponding eigen-
function. Finding the second eigenvalue can already be rather painful. Fortunately, it is usually
the first eigenvalue that is of most interest.
We make the foregoing observations to contrast the more commonly used finite difference
schemes for approximation of some eigenvalue, eigenfunction pairs of a Sturm-Liouville prob-
lem with the shooting methods developed in Chapter 7. The shooting methods determine each
pair independently of the others, involve no orthogonalization procedure, and in principle can
be used to find any eigenvalue as accurately as desired.
A precis of each chapter follows.
 Preface xi

Chapter 1 Setting the Stage

Typical problems of applied mathematics that lead to Sturm-Liouville problems are formu-
lated to make it clear what types of differential equations and boundary conditions arise in real-
world applications and what the natural assumptions on the data are. The chapter starts with
the first pure eigenvalue problem to arise historically, Euler buckling. Separation of variables
in initial boundary value problems and steady-state problems for partial differential equations
lead to Sturm-Liouville eigenvalue and boundary value problems for ordinary differential
equations: examples arising from vibration problems (e.g., mechanical and acoustical waves),
diffusion problems (e.g., heat flow and chemical transport), and steady-state phenomena are
given. Connections with variational methods are discussed briefly. Green’s functions are intro-
duced and motivated by physical considerations. The chapter concludes with a discussion of
the mathematical problems that must be faced to establish that Sturm-Liouville eigenvalues
and eigenfunctions have the qualitative properties that physical intuition suggests and to fully
justify the formal solutions that arise from eigenfunction expansions.

Chapter 2 Preliminaries
This chapter is designed as a convenient reference for background material needed for a rig-
orous treatment of Sturm-Liouville problems. Putting the background material here has three
purposes. First, it enables us not to interrupt the treatment of Sturm-Liouville problems to
develop background material on the spot in the midst of other reasoning unique to the prob-
lems at hand. Second, it frees the reader already familiar with the background material from
an unnecessary distraction. Third, the chapter introduces most of the notational conventions
used throughout the book. We recommend that readers familiar with the results collected here
just skim the chapter to become familiar with the notation that is used later. We hope other
readers will find it convenient to have proofs of some essential background results available
in one place.

Chapter 3 Integral Equations

The unified approach to Sturm-Liouville eigenvalue and boundary value problems that we
follow emphasizes the interplay between the mathematical assumptions made and their con-
nection to the characteristic oscillation properties that physical systems leading to such prob-
lems exhibit. This is accomplished by recasting a problem expressed in terms of a differential
equation and related boundary conditions as a single integral equation. Of course, this leads us
to a presentation of the principal results of Hilbert-Schmidt theory.
The unified treatment also includes material on Kellogg kernels that has essentially van-
ished from the current literature on Sturm-Liouville problems. This is unfortunate because
use of Kellogg kernels adds considerable insight into the interplay between mathematical prop-
erties of eigenvalues and eigenfunctions of Sturm-Liouville problems and the well-documented
behavior of physical systems that lead to such problems. Our treatment of singular Sturm-
Liouville problems leads naturally to singular Kellogg kernels. We believe the material on sin-
gular Kellogg kernels is new.
The development we follow also involves extensions of properties of integral operators
with suitably positive kernels (due originally to Jentzsch) and some pioneering work of Schur
on the relationship between the eigenvalues and eigenfunctions of a kernel and those of its com-
pound kernels. The work of Jentzsch and especially of Schur is not as well known as it
should be.
The field of integral equations is vast. We have restricted the chapter to just those results
needed for a unified treatment of Sturm-Liouville problems.
xii Preface

Chapter 4 Regular Sturm-Liouville Problems

A Sturm-Liouville eigenvalue is regular on an interval a , x , b if the coefficients in the
differential equation −(p(x)y ′ )′ + q(x)y = λr(x)y are continuous on the corresponding closed
interval and p and r are positive there. Likewise for the Sturm-Liouville boundary value prob-
lems where λry is replaced by f, a given continuous function. We treat regular Sturm-
Liouville problems in depth when the associated boundary conditions are linear and separated
and also in considerable detail when they are mixed. Green’s functions are introduced and
characterized in a standard way. A glance at the table of contents indicates the scope of topics
treated.
Since p(x) is only assumed to be continuous, which may be the natural assumption corre-
sponding to the underlying physical situation, we look closely into what it means for y to be a
solution to such an equation and to the consequences of the definition adopted for a solution.
Many treatments of Sturm-Liouville differential equations assume further smoothness on p(x).
For that reason, the existence, uniqueness, and continuous dependence results established
here are more general than those often encountered in the literature.

Chapter 5 Singular Sturm-Liouville Problems - I

The treatment of the singular problems in this chapter and the next was motivated by
an appendix in [43] by Tychonoff and Samarski, that seems to have had much less influence
that it should have. We flesh out and extend the important developments of the appendix.
The appendix sets out the qualitative properties of bounded solutions to singular Sturm-
Liouville differential equations of the form (p(x)y ′ )′ + (λr(x) − q(x))y = 0 where p and/or q
are singular and a , x , b. The nature of the singularities are chosen to cover all the classical
Sturm-Liouville differential equations and special functions that arise from practical applica-
tions. In particular, the equations of Bessel, Legendre, Hermite, and Laguerre are included.
We assume in Chapter 5 that p, q, and r are continuous on a ≤ x ≤ b, p . 0 on a , x ≤ b,
and p has a simple zero at x = a. The weight function r(x) is positive on a , x ≤ b but can have
a zero of finite order at x = a. Bessel’s equation of order 0 is a prototype for such problems.
The conversion of a singular Sturm-Liouville problem to an integral equation is done via a
mildly singular Green’s function. (See the precis of Appendix A.) As far as we know, a careful
treatment of mildly singular Green’s functions and their compound kernels is new as is the
introduction of singular Kellogg kernels.
The topics covered in this setting are given in the table of contents for the chapter. The bot-
tom line is that all the familiar oscillation behavior exhibited by regular Sturm-Liouville prob-
lems extends to this setting.

Chapter 6 Singular Sturm-Liouville Problems - II

Just as for Chapter 5, the treatment of the singular problems in this chapter was motivated
by an appendix in [43] by Tychonoff and Samarski: see the precis for Chapter 5.
In Chapter 6, the coefficient q is allowed to have a simple pole at x = a in addition to the
singular behavior allowed in Chapter 5. This is the situation for Bessel’s equations of order
n . 0. To accommodate the additional singularity in q some additional smoothness is needed.
Just as in Chapter 5, a singular Sturm-Liouville problem is converted into an equivalent sin-
gular integral equation whose kernel (Green’s function) is mildly singular. As far as we know, a
careful treatment of the mildly singular kernels of Chapter 6 and their compound kernels is new.
The full scope of topics covered in Chapter 6 is given in the table of contents. The bottom
line is that all the familiar oscillation behavior exhibited by regular Sturm-Liouville problems
extends to this setting.
 Preface xiii

Chapter 7 Approximation of Eigenvalues and Eigenfunctions

Shooting methods are presented for the regular and singular Sturm-Liouville eigenvalue
problems studied in Chapters 4, 5, and 6. The shooting method used for the regular problems
of Chapter 4 extends naturally to the singular problems of Chapter 5 with the aid of the boun-
dary condition, (q(a) − λr(a))y(a) − p′ (a)y ′ (a) = 0, that is forced by the singularity at x = a
and by the fact that y(a) = 0 for any bounded nontrivial solution to the singular differential
equation in the eigenvalue problem. In sharp contrast, in Chapter 6 any bounded solution to
the singular differential equation satisfies y(a) = 0 and y ′ (a) may not exist. Consequently, a
completely different shooting method must be used. A careful convergence analysis is given
for all three methods. Many examples are given with output from each shooting method and
effective strategies for choosing initial guesses are discussed.

Chapter 8 Concluding Examples and Observations

The results of preceding chapters are applied to a few particularly important problems of
both historical and current importance. The main examples are the vibrations of a hanging
chain (or cable), the oscillations of a string, and the vibrations of a bar (beam). We also indi-
cate how some of the shooting methods in Chapter 7 can be extended to higher order, self-
adjoint problems.

Appendix A Mildly Singular Compound Kernels

The Green’s functions for the singular Sturm-Liouville problems of Chapters 5 and 6 are
mildly singular in different senses. This appendix develops the properties of two types of mildly
singular kernels and their compound kernels that include the Green’s functions encountered in
Chapters 5 and 6 and that are needed to establish the oscillation properties of singular Kellogg
kernels and the singular Sturm-Liouville eigenvalue problems in Chapters 5 and 6.

Appendix B Iteration of Mildly Singular Kernels

This appendix develops the properties of kernels that arise by iteration of mildly singular
kernels and of their compound kernels that are needed to establish the properties of the singu-
lar Sturm-Liouville eigenvalue problems in Chapters 5 and 6.

Appendix C The Kellogg Conditions

The defining conditions of a Kellogg kernel are shown to be essentially equivalent to
five simple physical properties of elastic strings. The treatment closely follows that given in
Gantmacher and Krein [16] and makes this important link between mathematical assumptions
and observed physical behavior available to a wider audience.
We extend our sincere thanks to the individuals of the acquisitions, editorial and publica-
tion staffs at CRC Press, Taylor & Francis Group, and Nova Techset who assisted in the pub-
lication of our book. We especially appreciated the friendly, helpful attitudes and quick
responses to our queries from Sarfraz Khan, Editor, Mathematics; Callum Fraser, Editorial
Assistant; Suzanne Lassandro, Production Editorial Manager; Teena Lawrence, Manager
(Project Management); and Jeanne Washington, Freelancer (proofreader) whose combined
efforts have resulted in a much-improved final product.
Finally, the authors would be grateful to receive feedback from readers about misprints,
mistakes, misconceptions, and so on that you find. Thank you. We can be reached at
guenth@math.orst.edu and jwlee@math.oregonstate.edu.
Chapter 1
Setting the Stage

There are many reasons for seeking eigenvalues and their corresponding eigenfunctions. Here
are a few of them. First, solutions to many problems modeled by ordinary and partial
differential equations can often be given explicitly in terms of eigenfunction expansions.
(For the problems we shall treat, such eigenfunction expansions are strictly analogous to
the representation of a vector in terms of its i, j, and k components in 3-space or the corre-
sponding representations in n-space.) Second, eigenvalues are often of independent interest.
They may tell us where bifurcations can occur in nonlinear models by looking at their
linearizations. Eigenvalues give sharp estimates about the rates of decay (or growth) of solu-
tions arising in heat conduction, concentration analyses, flow in porous media, and so on. In
vibration problems, they give fundamental frequencies and overtones of musical instru-
ments. Eigenvalues are important in determining the critical mass for nuclear reactions in
a given geometry. Finally, eigenvalues arise naturally in optimization and in the calculus
of variations.
Since most eigenvalue problems cannot be solved explicitly, we will take a hard look at the
qualitative behavior of both the eigenvalues and the eigenfunctions, and analyze both regular
and singular problems. For the same reason, we present an effective numerical technique for
the practical evaluation of eigenvalues and corresponding eigenfunctions.
To further motivate the types of Sturm-Liouville problems that are the subject of this
book, we present, without detailed derivations, a few important problems of mathematical
physics and the Sturm-Liouville eigenvalue and boundary value problems to which they
lead, usually via separation of variables in a partial differential equation. However, eigenvalue
problems arise in many contexts involving ordinary and partial differential equations as well as
in matrix theory and more general operator settings. It seems likely that Euler considered the
first eigenvalue problem when he discussed the buckling of a beam. We start our survey with
that problem.
In the survey of problems that follows, we assume that all functions are real-valued and all
constants are real numbers, which is natural for the scenarios presented. See the final section of
this chapter concerning complex-valued functions and data.

1.1 Euler Buckling

A straight elastic bar (beam, rod) of length l is positioned vertically upward and is
anchored at its base. Experimentally one can take a thick metal wire. A small compressive force
of magnitude K acts vertically downward on the free end of the bar as in Figure 1.1.
The equations governing the shape of the bar are


EIy ′′ = −Ky, 0,x,l
y(0) = 0, y(l) = 0

1
2 Sturm-Liouville Problems: Theory and Numerical Implementation

FIGURE 1.1: The Euler Beam

where y = y(x) is the transverse deflection of the midline of the bar from its vertical equilibrium
position. The physical constants E and I are determined by the elastic and geometric proper-
ties of the bar. The governing equations always have the solution y(x) = 0 for 0 ≤ x ≤ l. Exper-
iments confirm that this is the shape of the bar when K is small but that the bar will buckle
when K is increased to a critical value. Buckling means the bar will deflect from the vertical
into a new equilibrium shape. Do the governing equations predict buckling? Euler answered
this question in the affirmative in 1757.
Express the governing equations as

y ′′ + λy = 0, y(0) = y(l) = 0,

where λ = K /EI . 0. If buckling occurs, it must be possible to find a solution (or solutions)
to the governing equations different from the obvious solution y(x) = 0 for 0 ≤ x ≤ l, the
so-called trivial solution. Other solutions, if any exist, are called nontrivial. The differential
equation has general solution

√ √
y = A cos ( λx) + B sin ( λx).

Since y(0) = √0, A = 0. If nonzero deflections (solutions) are possible, we must have B ≠ 0 and
y(l) = B sin ( λl) = 0. Thus, nonzero solutions y exist if and only if λ = λn = (nπ/l)2 for n a
positive integer. The corresponding nontrivial solutions are y = yn (x) = Bn sin (nπx/l) with Bn
≠ 0. The values of λ (hence, K ) that permit nontrivial deflections are now called eigenvalues
and the corresponding nontrivial solutions are called eigenfunctions. The problem we have
just solved is called an eigenvalue problem.
Buckling in the Euler beam first occurs at λ1 = π 2 /l 2 ; that is when K = EI π 2 /l 2 and the
bent beam takes the new equilibrium state

y = B sin (πx/l)
 Setting the Stage 3

FIGURE 1.2: The Buckled Beam

for some B ≠ 0. Figure 1.2 illustrates the case with B . 0. Since λ1 = (K /EI )1/2 , the smallest
eigenvalue determines the minimum compressive force needed to buckle a beam of given flex-
ural rigidity EI.
The Euler model predicts that buckling can occur and does occur only at the eigenvalues λn
and that the corresponding buckled equilibrium states are multiples of sin (nπx/l). Actually,
once the bar has buckled, a new model is needed because the physical situation has become
nonlinear. Nevertheless, even in the nonlinear regime the linear problem above, which is the
linearization of an appropriate nonlinear model, still predicts the values of K at which buckling
can occur. Problems of this sort are called bifurcation (branching) problems because nonlinear
states branch from a stable linear state (usually y = 0) at certain critical values, the eigenvalues
of the linearized problem. The eigenfunction yn corresponding to the branch point determined
by the eigenvalue λn approximates the shape of the nonlinear buckled responses of small ampli-
tude that occur near the branch point.
The governing equations for Euler buckling,

y ′′ + λy = 0, y(0) = y(l) = 0,

where λ = K /EI . 0 comprise a regular Sturm-Liouville eigenvalue problem. Regular means

that the differential equation is regular. A detailed study of regular Sturm-Liouville boundary
value problems and eigenvalue problems is presented in Chapter 4.

1.2 Hanging Chain

In contrast to Euler buckling, the problem of determining the normal modes of a hanging
chain leads to a singular Sturm-Liouville differential equation and corresponding eigenvalue
problem. This problem was solved by Daniel Bernoulli (1700–1782) in 1732 and involved
the first use of a Bessel function. Small transverse displacements from equilibrium were
assumed. The problem was discussed further by Euler in 1781. F. W. Bessel (1784–1846)
investigated the functions that now bears his name.
In Bernoulli’s treatment the chain is a one-dimensional continuum that has constant
density. We formulate a slightly more general model that permits variable density. Suppose
the length of the chain is l. Set up coordinates so that the x-axis is directed vertically
upward with the origin at the free end of the chain when the chain hangs in its vertical
4 Sturm-Liouville Problems: Theory and Numerical Implementation

equilibrium position. Let ρ0 (x) be the density of the chain when it is hanging in equilibrium
and u(x, t) be the transverse displacement at time t of the point on the chain that is located
at position x when the chain hangs in equilibrium. The only external force acting on the chain
is gravity, with constant acceleration g, and the tension at a cross section of the chain acts
tangentially and is due to the part of the chain that lies below the cross section.
Under these assumptions the initial boundary value problem for the chain is
⎧
⎨ ρ0 (x)utt = (p(x)ux )x , 0 , x , l, t . 0,
|u(0, t)| , 1, u(l, t) = 0, t ≥ 0, (1.1)
⎩
u(x, 0) = f (x), ut (x, 0) = v(x) 0 ≤ x ≤ l,

where
 x
p(x) = g ρ0 (ξ) dξ
0

for 0 ≤ x ≤ l, f (x) specifies the initial shape of the chain, and v(x) is its initial velocity
profile. Observe that the differential equation is singular because p(0) = 0. Typically such
equations can have both bounded and unbounded solutions. Physically realistic solutions for
the displacement u(x, t) must be bounded. This leads to the boundary condition |u(0, t)| , 1
which means that the displacement is bounded for x . 0 and near 0 for all time t. It
follows that u(x, t) is bounded in space and time.
The normal modes of the chain are the motions where each point of the chain vibrates at
the same frequency. Such motions have the form u(x, t) = T (t)X(x), so-called separated
solutions of the partial differential equation. A separated solution will satisfy the wave
equation in (1.1) if and only if

ρ0 (x)T ′′ X = (p(x)TX ′ )′ ,
T ′′ (p(x)X ′ )′
= X = −λ,
T ρ0 (x)

where −λ is a separation constant. Thus,

T ′′ + λT = 0

and

(p(x)X ′ )′ + λρ0 (x)X = 0, |X(0)| , 1, X(l) = 0,

where the boundary conditions on X follow from those in (1.1). The normal modes
u(x, t) = T (t)X(x), apart from the trivial solution, are determined by those values of λ (eigen-
values) for which the X-problem has nontrivial solutions
√ (eigenfunctions). For such λ the
frequency of oscillation of each point on the chain, λ/2π, is determined by the T-equation.
In Bernoulli’s original problem ρ0 (x) = ρ0 a given positive constant, p(x) = gρ0 x, the
wave equation for the chain is

utt = g(xux )x ,

and the separated solutions u(x, t) = T (t)X(x) are determined by

T ′′ + λT = 0
 Setting the Stage 5

and
g(xX ′ )′ + λX = 0, |X(0)| , 1, X(l) = 0.
It turns out that the X-equation is reducible to a Bessel’s equation of order 0 and, hence, that
the spatial component of a normal mode is a multiple of a bounded solution of that equation.
We will discuss Bernoulli’s problem further in Chapter 8 together with numerical results. For
the moment, we mention that Bessel’s equation of order 0 is a prototype for the singular Sturm-
Liouville boundary value and eigenvalue problems that are the subject of Chapter 5.
If the density is ρ0 (x) = ρ0 x n where ρ0 is a positive constant and n ≥ 0, then the wave
equation is
 n+1
x
x n utt = g ux .
n+1 x

The normal modes u(x, t) = T (t)X(x) are determined from

g
x n T ′′ X = (x n+1 TX ′ )′ .
n+1

Hence,
T ′′ g 1
= (x n+1 X ′ )′ = −λ,
T (n + 1) x X
n

where λ is a separation constant,

T ′′ + λT = 0,

and
 ′
g
x n+1 X ′ + λx n X = 0, |X(0)| , 1, X(l) = 0.
n+1

The X-equation is reducible to a Bessel’s equation of order n and, hence, the spatial component
of a normal mode is expressible in terms of a bounded solution of that equation. We will
discuss this generalized Bernoulli problem further in Chapter 8 together with numerical
results. For the moment, we mention that Bessel’s equation of order n with n . 0 is a prototype
for the singular Sturm-Liouville boundary value and eigenvalue problems that are the subject
of Chapter 6.

1.3 Separation of Variables

The method of separation of variables seeks to solve a linear partial differential equation,
such as the wave equation or heat equation, together with given side conditions by ultimately
reducing the problem to finding solutions to a family (or families) of related ordinary differen-
tial equations with corresponding side conditions. The separated solutions are superposed
appropriately so as to satisfy any remaining conditions of the original problem. This solution
strategy leads naturally to many interesting and important eigenvalue problems. Most of the
physical situations described in this chapter involve either the homogeneous wave equation
butt = div(p∇u) − cu

or the homogeneous heat equation (also known as the diffusion equation)

but = div(p∇u) − cu
6 Sturm-Liouville Problems: Theory and Numerical Implementation

with u = u(x, t) where x varies in real Euclidean n-space and b(x) . 0, p(x) ≥ 0, and c(x) ≥ 0
are given functions of the spatial variables in a domain of interest. A natural first step in the
method of separation of variables is to seek separated solutions of the form u = w(x)T (t).
Such a separated solution satisfies the wave equation or the heat equation if and only if
T ′′ div(p∇w) − cw
=
T bw

or
T ′ div(p∇w) − cw
=
T bw

holds for all relevant times t and positions x. The left member of each equation must be cons-
tant in time because the right member does not change as time varies. Likewise, the right mem-
ber must be constant in space because the left member does not vary as the spatial variable
changes. Thus, both sides of each equation must be one and the same constant; that is,
T ′′ div(p∇w) − cw
= −λ and = −λ,
T bw

or
T′ div(p∇w) − cw
= −λ and = −λ,
T bw

for some separation constant, here −λ. In typical applications, λ is positive: for the wave
equation this is equivalent to separated solutions u(x, t) = T (t)w(x) being periodic in time
while for the heat equation it is equivalent to separated solutions that decay in time. For either
type of problem
div(p∇w) − cw + λbw = 0

in the interior of the spatial domain of interest. If p is constant, the differential equation has
the form
Δw − cw + λbw = 0,

where Δw is the Laplacian of w. For separated solutions to be useful they must satisfy some of
the homogeneous side conditions of the problem and they must be nontrivial, not identically
zero. This is how eigenvalue problems emerge.
Separation of variables was first used by Euler (1748) in an isolated case to find a solution of
the one-dimensional wave equation together with boundary and initial data and to determine
the fundamental frequencies of a vibrating violin or piano string. D’Alembert gave the general
solution to the one-dimensional wave equation in 1746. A heated controversy arose between
Euler and D’Alembert about the meaning of a solution to the wave equation. Lagrange sided
with Euler in the debate. Later Fourier (1805) significantly extended the method of separation
of variables in his pioneering studies on heat conduction. Dirichlet put Fourier’s method on a
firm foundation about 25 years later. These developments led to the boundary value problems
and eigenvalue problems now called Sturm-Liouville problems. In the 1820s and 1830s Sturm
and Liouville initiated the systematic study of such problems and in the process initiated the
study of qualitative properties of solutions to differential equations when explicit solutions
were not available.
We shall deal with problems that have one spatial dimension or can be reduced to one spa-
tial dimension. Such problems include higher-dimensional spatial situations where the geom-
etry and symmetry lead to an initial boundary value problem with only one spatial dimension.
 Setting the Stage 7

1.4 Vibration Problems

Problems of this kind typically lead to an initial boundary value problem for the wave
equation.

1.4.1 Vibrations of a String

Consider a taut, homogeneous string, such as a piano or violin string, stretched between
two posts. Pluck the string so that it experiences small transverse vibrations. If u = u(x, t)
is the deflection of the string from its rest (unplucked) position, then u satisfies the wave equa-
tion, initial conditions, and boundary conditions that follow:
⎧
⎨ utt = c2 uxx for 0 , x , l, t . 0,
ut (x, 0) = f (x), ut (x, 0) = g(x), for 0 ≤ x ≤ l, (1.2)
⎩
u(0, t) = 0, u(l, t) = 0, for t ≥ 0,
√
where l is the length of the string, c = τ/ρ is the speed of wave propagation, τ is the (cons-
tant) horizontal component of tension in the string, and ρ is the linear density of the string.
The functions f and g specify the initial displacement and velocity of the string.
If external forces act on the homogeneous string the wave equation becomes
utt = c2 uxx + F(x, t), where F(x, t) models the external transverse forces, such as those that
arise when the string is bowed as in Figure 1.3.
If the string is inhomogeneous, then ρ = ρ(x) . 0 and a careful derivation of the wave
equation shows that τ . 0 is either a constant or a function of time t. In the more general
case, the wave equation in the model becomes

ρ(x)utt = (τ(t)ux )x .

 is homogeneous so that ρ(x) = ρ0 and τ = τ0 where ρ0 and τ0 are constants

If the string
and c = τ0 /ρ0 , then

u(x, t) = X(x)T (t)

will satisfy the wave equation and the boundary conditions in (1.2) if X(x) and T (t) satisfy

X ′′ (x) + λX(x) = 0, X(0) = 0, X(l) = 0,

T ′′ (t) + λc2 T (t) = 0, (1.3)

where −λ is the separation constant. Furthermore, a separated solution u(x, t) = X(x)T (t)
will only be useful if it is not identically zero. The equation for X(x) always has the trivial

FIGURE 1.3: Vibrations of a String

8 Sturm-Liouville Problems: Theory and Numerical Implementation

solution X(x) = 0. Thus, separated solutions will only be useful if there are values of λ such
that the problem (1.3) has nontrivial solutions. Thus, separation of variables has led to an
eigenvalue problem for X. It is the same eigenvalue problem that we encountered in
Euler buckling.
If the string is inhomogeneous and the horizontal component of the tension is time depend-
ant, then (1.3) becomes
X ′′ (x) + λρ(x)X(x) = 0, X(0) = 0, X(l) = 0. (1.4)
and the temporal factor of a separated solution satisfies
T ′′ (t) + λτ(t)T = 0.
We will discuss the vibrations of a string more fully in Chapter 8.

1.4.2 Vibrations of a Circular Membrane

Consider the membrane of a circular drum, assumed homogeneous. Suppose the membrane
is set in motion by a displacement and velocity that is radially symmetric. The transverse
vibrations u that result are radially symmetric; that is, u = u(r, t). Use of the 2-D wave equa-
tion in polar coordinates with pole at the center of the membrane leads to the initial boundary
value problem
⎧ 1
⎪
⎨ utt = c2 (rur )r for 0 ≤ r , b, t . 0
r (1.5)
⎩ u(r, 0) = f (r), ut (r, 0) = g(r) for 0 ≤ r ≤ b,
⎪
u(b, t) = 0, for t ≥ 0,
where b is the radius of the membrane. If separated solutions u = R(r)T (t) are required to sat-
isfy the wave equation and the homogeneous boundary condition, the eigenvalue problem
which arises for R(r) is
1
R′′ + R′ + λR = 0, R(b) = 0.
r
The differential equation has a singularity at r = 0. Solutions to such equations can become
unbounded. Unbounded solutions are not physically meaningful for the vibrating membrane.
So there is an unstated (hidden) boundary condition that R must satisfy. It must be bounded.
This is often expressed by stating the eigenvalue problem for R as
1
R′′ + R′ + λR = 0, |R(0)| , 1, R(b) = 0, (1.6)
r
because any unbounded behavior of R must occur at the singularity r = 0.
Separation of variables succeeds in this example because polar coordinates were used. Sep-
aration of variables fails in rectangular coordinates because there are no nontrivial separated
solutions in that coordinate system. Use of polar coordinates facilitates the solution process but
brings in a singularity that is somewhat fake. The vibrating drum has no intrinsic physical sin-
gularity that would lead to the singular term, 1/r, in the mathematical model or its solution.
This singularity is an artifact of expressing the wave equation in polar coordinates: the trans-
formation x = r cos θ, y = r sin θ is not 1-1 at r = 0 and its Jacobian ∂(x, y)/∂(r, θ) = r is zero
at r = 0. This problem with the choice of coordinates is overcome by introducing the hidden
boundary condition |R(0)| , 1 in the eigenvalue problem. Similar situations arise when cylin-
drical or spherical polar coordinates are used.
The differential equation in this eigenvalue problem is a Bessel equation. One of the moti-
vations for investigating Bessel functions and other higher transcendental functions was to
solve problems arising in mathematical physics.
 Setting the Stage 9

1.4.3 Spherically Symmetric Vibrations in a Ball

If a ball with radius b experiences radially symmetric vibrations u = u(r, t), then u satisfies
the 3-dimensional wave equation
1 2
utt = c2 (r ur )r , 0 ≤ r , b, t . 0
r2
and the eigenvalue problem that arises from separation of variables is
2
R′′ (r) + R′ (r) + λR(r) = 0, |R(0)| , 1, R(b) = 0. (1.7)
r
Problem (1.4) is typical of a regular Sturm-Liouville eigenvalue problem: the coefficients in
the differential equation are smooth functions and no coefficient multiplying the highest order
derivative in the equation is zero at any point. Problems (1.6) and (1.7) are typical singular
Sturm-Liouville eigenvalue problems: at least one condition of a regular problem is violated.
The singular problems point out that the more general treatment of singular eigenvalue prob-
lems in Chapters 5 and 6 must allow singularities that behave like 1/r.

1.5 Diffusion Problems

Many diffusion problems in one spatial dimension are governed by a diffusion equation of
the form
ut = (p(x)ux )x + b(x)ux − c(x)u (1.8)

together with appropriate initial and boundary conditions. Here x is the spatial variable and t
is time, 0 ≤ x ≤ l, t . 0, p ≥ 0 is a diffusion coefficient and b and c ≥ 0 are given. The physical
meaning of b, c, and u = u(x, t) depends on the problem at hand. Depending on the field, the
names associated with (1.8) are Fourier, Fick, Darcy, and Nerust among others. In typical
applications p . 0, except possibly at x = 0 or x = l and the diffusion equation is satisfied for
0 , x , l and t . 0.
It is beneficial to express the diffusion equation, and other such equations that have the
term b(x)ux , in what is called formally self-adjoint form by means of the change of variable
u(x, t) = g(x)v(x, t) where
 x
g(x) = exp − b(ξ)/p(ξ) dξ .

It is easy to check that diffusion equation for v(x, t) has the form
g(x)vt = (p(x)g(x)vx )x − q(x)v,
so that the change of variables preserves p(x)g(x) . 0, except possibly at x = 0 or x = l.
Replacing p(x)g(x) by p(x), the transformed equation has the form
g(x)vt = (p(x)vx )x − q(x)v.
Separated solutions v(x, t) = X(x)T (t) of this partial differential equation must satisfy
the pair of ordinary differential equations
(p(x)X ′ (x))′ − q(x)X(x) + λg(x)X(x) = 0,
T ′ (t) + λT (t) = 0,
where the separation constant is −λ.
10 Sturm-Liouville Problems: Theory and Numerical Implementation

1.5.1 Chemical Transport

A simple model for the flow of a chemical through a partially saturated soil column in which
the ground water is at rest and scrubbers reduce the concentration of chemical to zero at the
ends of the column is
⎧
⎨ ut = (p(x)ux )x for 0 , x , l, t . 0,
u(x, 0) = f (x) for 0 ≤ x ≤ l, (1.9)
⎩
u(0, t) = 0, u(l, t) = 0 for t ≥ 0.

Here u(x, t) is the concentration of the chemical or pollutant at position x at time t and p ≥ 0 is
the diffusion coefficient. The coefficient b in the general diffusion equation is zero because the
ground water is at rest, and c also is zero when only the diffusion effect is modeled. The initial
concentration is given by the function f (x).
The eigenvalue problem for X that arises when separated solutions are required to satisfy
the boundary conditions is

(p(x)X ′ (x))′ + λX(x) = 0, X(0) = X(l) = 0.

1.5.2 Heat Conduction in a Rod

A laterally insulated rod of length l has an initial temperature distribution g(x) and is
surrounded by a medium held at temperature zero. Each end of the rod communicates heat
to its surroundings according to Newton’s law of cooling. A reasonable model in this situation
for the temperature u(x, t) in the rod at position x and time t is
⎧
⎪
⎪ ut = (p(x)ux )x − q(x)u, 0 , x , l, t . 0,
⎨
u(x, 0) = g(x), 0 ≤ x ≤ l,
(1.10)
⎪
⎪ αu(0, t) − βux (0, t) = 0, t ≥ 0,
⎩
γu(l, t) + δux (l, t) = 0, t ≥ 0,

In this model, q(x) ≥ 0 is a heat loss coefficient that allows for imperfect lateral insulation along
the lateral surface of the rod, and α, β, γ, and δ are positive constants determined by the char-
acteristics of the rod and Newton’s law of cooling.
The eigenvalue problem for X that arises when separated solutions are required to satisfy
the boundary conditions is

(p(x)X ′ (x))′ − q(x)X(x) + λX(x) = 0, 0 , x , l,

αX(0) − βX ′ (0) = 0, γX(l) + δX ′ (l) = 0.

Explicit solutions for the eigenvalues λ and the corresponding eigenfunctions X do not exist
except for a few simple but important choices of p(x), q(x), and the boundary conditions. How-
ever, the basic heat equation ut = auxx with a a positive constant and with homogeneous
Dirichlet boundary conditions u(0, t) = 0 and u(l, t) = 0 leads to the eigenvalue problem

X ′′ (x) + λX(x) = 0, 0 , x , l,
X(0) = 0, X(l) = 0,

whose solution yields eigenvalues λn = (nπ/l)2 and corresponding eigenfunctions Xn (x) =

sin nπx/l for n = 1, 2, . . . .
 Setting the Stage 11

1.5.3 Heat Conduction in a Disk

Consider a heat conducting disk of radius b with insulated top and bottom, constant ther-
mal conductivity 1, initial temperature distribution f (r, θ), and with its bounding circle held at
temperature zero. The temperature u(r, θ, t) is given by
⎧
⎨ ut = Δu for 0 ≤ r , b, 0 ≤ θ ≤ 2π, t . 0,
u(r, θ, 0) = f (r, θ) for 0 ≤ r ≤ b, 0 ≤ θ ≤ 2π,
⎩
u(b, θ, t) = 0 for 0 ≤ θ ≤ 2π, t . 0.

Separation of variables in space and time with separation constant −λ via u = T (t)v(r, θ)
leads to
T ′ + λT = 0 and Δv + λv = 0.

So T (t) is a multiple of e−λt . A second separation of variables via v = R(r)Θ(θ) yields

r 2 R′′ + rR′ + λr 2 R − μR = 0,
Θ′′ + μΘ = 0.

Since (r, θ) and (r, θ + 2π) mark the same point in the plate, Θ must be 2π periodic
Θ′′ + μΘ = 0, Θ(0) = Θ(2π), Θ′ (0) = Θ′ (2π).
The condition on the derivative follows from Fourier’s law of heat flow.
The eigenvalue problem for Θ has eigenvalues μ = μn = n 2 for n = 0, 1, 2, . . . and corre-
sponding eigenfunctions Θ = Θn = an cos nθ + bn sin nθ where an2 + b2n = 0. The eigenvalue
μn = n 2 has multiplicity 2 because two linearly independent eigenfunctions correspond to it,
unlike all the foregoing examples where each eigenvalue has multiplicity 1.
Since μ = μn = n 2 , the differential equation for R = Rn (r) becomes
r 2 R′′ + rR′ + (λr 2 − n 2 )R = 0,

which is Bessel’s equation of order n with parameter λ. Equivalently,

 2
′ ′ n
−(rR ) + − λr R = 0,
r

which reveals a singularity in the highest derivative term because r = 0 at the origin and a
singularity in the coefficient of R that becomes positively infinite as r  0. We will deal
with singular problems of this type in more generality in Chapter 6.
Each separated solution un (r, θ, t) = e−λt Θn (θ)Rn (r) will satisfy the condition that the
temperature on the boundary of the disk is 0 and be physically realistic (remain bounded) if
the separation constant λ can be chosen so that R = Rn (r) satisfies
 2
′ ′ n
−(rR ) + − λr R = 0, 0 , r , b,
r
|R(0)| , 1, R(b) = 0,

which is an eigenvalue problem for a Bessel’s equation for each n = 0, 1, 2 . . . .

1.6 Steady State Regimes

If a wave or diffusion phenomenon involves significant external forcing or sources (sinks),
the homogeneous wave and heat equations must be modified to include such effects. The
12 Sturm-Liouville Problems: Theory and Numerical Implementation

resulting equations are the inhomogeneous wave equation,

butt = div(p∇u) − cu − f ,
and the inhomogeneous heat equation,
but = div(p∇u) − cu − f ,
where u = u(x, t) and x varies in some domain in Euclidean n-space, t is time, b . 0, p ≥ 0, and
c ≥ 0 are given functions of the spatial variables in a domain of interest, and f describes the
external influences. If f is independent of t or f (x, t)  f (x), a time independent limit, and
the boundary conditions are time independent, then it is natural to expect that u(x, t) will con-
verge to a steady-state, that is, time independent, solution u(x) to the wave or heat equation.
In either case, the steady-state solution u = u(x) satisfies the differential equation
div(p∇u) − cu = f
or
pΔu + ∇p · ∇u − cu = f
and any prescribed boundary conditions. If p is a constant, the steady-state equation has
the form
div(∇u) − cu = f
or
Δu − cu = f .

If c = 0 the steady-state equation becomes

Δu = f ,
which is the Laplace equation when f = 0 and the Poisson equation when f ≠ 0.

1.6.1 Heat Conduction in a Rectangular Plate

If the plate with insulated top and bottom has dimensions 0 ≤ x ≤ l1 , 0 ≤ y ≤ l2 , its hori-
zontal side 0 ≤ x ≤ l1 , y = 0 is held at temperature f (x), and the other three sides are held at
temperature zero, the thermal diffusivity p = p(x) is independent of y, and only diffusion
effects are modeled, then the steady-state temperature u = u(x, y) satisfies
⎧
⎨ (pux )x + (puy )y = 0 for 0 , x , l1 , 0 , y , l2 ,
u(x, 0) = f (x) for 0 ≤ x ≤ l1 ,
⎩
u=0 on the other sides.
The initial data f (x) must satisfy the compatibility conditions f (0) = f (l1 ) = 0 because the
temperature on the boundary must be continuous.
Separated solutions u(x, y) = X(x)Y (y) that satisfy the differential equation and the
homogeneous boundary conditions lead to the eigenvalue problem
(p(x)X ′ )′ + λp(x)X = 0, X(0) = X(l1 ) = 0
and the companion problem
Y ′′ − λY = 0, Y (l2 ) = 0.
√
The companion problem has for solutions multiples of Y (y) = sinh λ(l2 − y). So once the
eigenvalues λn and eigenfunctions Xn (x) are determined the separated solutions are constant
multiples of

Xn (x) sinh λn (l2 − y).
 Setting the Stage 13

1.6.2 Heat Conduction in a Circular Plate

If the heat conducting body with insulated top and bottom is a circular disk with radius b,
it is natural to use polar coordinates r and θ with origin at the center of the disk. Assume the
thermal conductivity is independent of θ so that p = p(r) and the circular boundary of the disk
is held at temperature f (θ). In this situation, the steady-state temperature u = u(r, θ) satisfies
⎧ 
⎨ p(r) u + 1 u + 1 u ′
rr r θθ + p (r)ur = 0 for 0 ≤ r , b, 0 ≤ θ ≤ 2π,
r r2
⎩
u(b, θ) = f (θ) for 0 ≤ θ ≤ 2π

The initial data f (θ) must satisfy the compatibility condition f (0) = f (2π) because the temper-
ature cannot have two values at the same point.
Separated solutions u(r, θ) = R(r)Θ(θ) that satisfy the partial differential equation must
satisfy the ordinary differential equations

r 2 R′′ + r(1 + rp′ /p)R′ − λR = 0

and
Θ′′ + λΘ = 0.

Since Θ must be 2π periodic, λ = λn = n 2 for n = 0, 1, 2, . . . and

Θn (θ) = an cos nθ + bn sin nθ

with an2 + b2n = 0.

The differential equation and “hidden” boundary condition for R = Rn are

r 2 R′′ + r(1 + rp′ /p)R′ − n 2 R = 0, |R(0)| , 1.

The R problem reduces to

r 2 R′′ + rR′ − n 2 R = 0, |R(0)| , 1

when the thermal conductivity is constant. This is an Euler equation and its bounded solutions
are the constant multiples of
Rn (r) = r n .

Euler equations also are obtained if p(r) is proportional to any power of r.

1.7 On Models
The models presented earlier arise in other settings, for example in biological systems and in
acoustics. Diffusion of molecules in a fluid at rest satisfies the partial differential equation
  
∂C ∂ ∂C ∂ ∂C ∂ ∂C
= D + D + D
∂t ∂x ∂x ∂y ∂y ∂z ∂z

where C (x, y, z, t) is the concentration of the substance and D is the diffusion coefficient.
The equation modeling sound waves in the atmosphere,
 2
∂2 p 2 ∂ p ∂2 p ∂2 p
= c + + ,
∂t 2 ∂x 2 ∂y 2 ∂z 2
14 Sturm-Liouville Problems: Theory and Numerical Implementation

where p is the pressure and c is the speed of sound, is fundamental to acoustics theory.
The vibrations of a circular membrane also satisfies an equation of this type.
The fact that the same equations arise over and over in different contexts is one of the
fundamental strengths of mathematical modeling: consider

∂u ∂2 u
= 2,
∂t ∂x
a non-dimensionalized partial differential equation. But what does it mean? The meaning
depends on the context in which it was derived. One researcher might say that u is a concen-
tration of molecules and the equation describes their diffusion. Another might say u is temper-
ature and the equation describes heat conduction. A third might say u is pressure and the
equation describes flow through a porous medium. A fourth might say u is the signal in a fiber
optic cable when the leakage to ground is negligible, and so on. The same equation holds in all
cases. The time and length scales differ as do the interpretations of the solution, but, after
introducing dimensionless coordinates, the partial differential equation is the same.
From the mathematical standpoint this means that the solution techniques developed in
one field can be applied in another field in which the approach may not naturally suggest itself.
From a practical standpoint, intuition developed from the study of, say, heat conduction can
be applied to the study of molecular diffusion, Brownian motion, pressure waves, and so on as
long as the underlying partial differential is the same.

1.8 Sturm-Liouville Boundary Value Problems

Return to heat conduction in the laterally insulated rod and the initial boundary value
problem (1.10). If we assume that there are distributed heat sources and sinks along the
rod, then the heat equation contains a new term f (x, t) that describes the heat generation
due to the sources and sinks. The modified problem is
⎧
⎪
⎪ ut = (p(x)ux )x − q(x)u + f (x, t), 0 , x , l, t . 0,
⎨
u(x, 0) = g(x), 0 ≤ x ≤ l,
(1.11)
⎪
⎪ αu(0, t) − βux (0, t) = 0, t ≥ 0,
⎩
γu(l, t) + δux (l, t) = 0 t ≥ 0.

If f (x, t) is independent of t or tends to a time independent limit, say f (x), as t  1, then we

expect that the temperature will tend to a time independent limit, u = u(x), that will satisfy
the heat equation with ∂u/∂t = 0 and the (time independent) boundary conditions


−(p(x)u ′ )′ + q(x)u = f (x), 0 , x , l,
(1.12)
αu(0) − βu ′ (0) = 0, γu(l) + δu′ (l) = 0

A problem of this form consisting of a Sturm-Liouville differential equation and certain boun-
dary conditions is called a Sturm-Liouville boundary value problem.
We note as a matter of convenience that if we replace q(x) by q(x) − λr(x) in the differential
equation, the Sturm-Liouville boundary value problem becomes


−(p(x)u ′ )′ + (q(x) − λr(x))u = f (x) 0 , x , l,
αu(0) − βu ′ (0) = 0, γu(l) + δu′ (l) = 0

This problem reduces to the general Sturm-Liouville eigenvalue problem when f = 0 and to the
general Sturm-Liouville boundary value problem when λ = 0.
 Setting the Stage 15

1.9 Calculus of Variations

The calculus of variations is another source of Sturm-Liouville boundary value problems
and eigenvalue problems. For example, according to the principle of minimum mechanical
energy, a one-dimensional continuum, modeled by a curve y = y(x) that extends from
y(a) = ca to y(b) = cb assumes the shape that minimizes (more properly, makes stationary)
the integral
 b
1 1
I (y) = p(x)y ′2 + q(x)y 2 − yf (x) dx
a 2 2

among all continuously differentiable functions y that satisfy the boundary conditions. In the
context of an elastic string suspended between two posts, p(x) is the mass density along the
string, q(x) is a coefficient of elasticity, and f (x) is an external force. The first variation of
the integral is
b
d
δI (y)(ζ) = I (y + εζ) = (p(x)yζ′ + q(x)yζ − ζf (x))dx
dε ε=0 a

where ζ is any continuously differentiable function satisfying ζ(a) = 0 and ζ(b) = 0. The
conditions on ζ guarantee that y + εζ satisfies the boundary conditions and, hence, determines
a potential shape assumed by the continuum. If this derivative is zero, that is, if the first var-
iation δI (y)(ζ) = 0 for some function y and all ζ, then y has a continuous second derivative by
the Theorem of Du-Bois Reymond and the fundamental lemma of the calculus of variations
implies that
−(p(x)y ′ )′ + q(x)y = f (x).

Thus, the problem of “minimizing” I (y) is equivalent to solving the Sturm-Liouville boundary
value problem
−(p(x)y ′ )′ + q(x)y = f (x), y(a) = ca , y(b) = cb .

The problem of finding the eigenvalues and eigenfunctions of the Sturm-Liouville eigen-
value problem
−(p(x)y ′ )′ + q(x)y = λy, y(a) = 0, y(b) = 0

is equivalent to “minimizing” the integral

 b
1 1
J (y) = p(x)y ′ 2 + q(x)y 2 dx
a 2 2

over continuously differentiable functions y satisfying y(a) = 0 and y(b) = 0 subject to the
normalizing constraint
b
y 2 dx = 1.
a

In this case, the eigenvalue arises essentially as a Lagrange multiplier. There must be a
constant λ such that the “minimizing” y is also a stationary value of
 b
1 1 1
K (y) = p(x)y ′ 2 + q(x)y 2 − λy 2 dx.
a 2 2 2
16 Sturm-Liouville Problems: Theory and Numerical Implementation

1.10 Green’s Functions

Since the solution to a Sturm-Liouville boundary value problem involves two integrations,
it is natural to expect that there is a solution formula that involves integration. There is such a
formula when the problem has a unique solution. It expresses the solution to the boundary
value problem in terms of a Green’s function, or influence function, that is determined by
the Sturm-Liouville differential operator and the boundary conditions of the problem, and
which has an important physical interpretation.
In this section, we use physical arguments to motivate the existence of a Green’s function,
to point out some of its important properties, and to find the solution formula. A fully rigorous
mathematical treatment of these topics will be given in later chapters.
We carry out the physical reasoning in the context of steady-state heat flow in a rod, as
described in (1.11) and (1.12). It is convenient to denote the steady-state temperature by
y(x) instead of u(x) as in (1.12) so y(x) satisfies


−(py ′ )′ + q(x)y = f (x), 0 , x , l,
(1.13)
αy(0) − βy ′ (0) = 0, γy(l) + δy ′ (l) = 0.

Let Ly = −(py ′ )′ + q(x)y so the differential equation in (1.13) is Ly = f.

The solution formula for the Sturm-Liouville problem (1.13) is
 l
y(x) = g(x, s)f (s) ds, (1.14)
0

where g(x, s) is called the Green’s function or influence function for the Sturm-Liouville
differential operator Ly and the given boundary conditions. More precisely, g(x, s) is called a
Green’s function for (1.13) if it is continuous on 0 ≤ x, s ≤ l and uniquely solves the given
Sturm-Liouville problem for all continuous right-hand sides f (x).
We argue as follows to understand why the solution formula (1.14) is reasonable: the right
member f (x) is the given rate at which heat is generated per unit length per second by sources
and sinks along the rod. Let ε . 0 be fixed and fs (x) specify a unit rate of heating per second
concentrated near the point x = s in the rod. That is, fs (x) is continuous, zero outside the inter-
val s − ε , x , s + ε, and
 l
fs (x) dx = 1.
0

Let y = gε (x, s) be the steady-state temperature in the rod produced by the input fs (x).
Analytically, gε (x, s) is the solution to (1.13) when f = fs. It is plausible that, as ɛ tends to
zero, the temperature distribution gε (x, s) will converge to a limiting continuous temperature
distribution g(x, s) that corresponds to a heat source of unit intensity at the point s. Thus,
g(x, s) is the temperature at x due to a unit heat source applied at location s in the rod.
Now, let f (x) be an arbitrary continuous rate of heat generation per unit length per
unit time along the rod. Imagine the rod decomposed into n nonoverlapping segments each
of length Δs and centered at sk. In the kth segment, the heat input from the continuous
distribution f (s) is closely approximated by f (sk )Δs, with the approximation improving
as Δs  0. Consequently, the contribution to the temperature y(x) at point x due to
the heating in the kth segment is approximately g(x, sk )f (sk )Δs, and this approximation
should improve as Δs  0. Since the differential equation governing heat flow in the rod is
 Setting the Stage 17

linear, the temperature that arises in the rod due to the combined effect of all the inputs
f (sk )Δs is
n
g(x, sk )f (sk )Δs
k=1

and this should closely approximate the temperature y(x) in the rod produced by the contin-
uous distribution f (x), with the approximation improving as Δs  0. This suggests that
n l
y(x) = lim g(x, sk )f (sk )Δs = g(x, s)f (s) ds,
n1 0
k=1

which is just the solution formula (1.14).

The foregoing reasoning also leads to several important properties of the Green’s function.
The solution gε (x, s) to (1.13) with right member fs (x) satisfies

Lgε (x, s) = fs (x) = 0 for |x − s| . ε,

where the differential operator L acts on functions of x. Since gε (x, s) converges to g(x, s) as ɛ
tends to zero, this suggests that

Lg(x, s) = 0 for x = s,

where L acts on functions of x. Furthermore, each solution gε (x, s) satisfies the boundary con-
ditions of the problem so passing to the limit as ɛ tends to zero it follows that, as a function of x
for fixed s,

g(x, s) satisfies the boundary conditions of the problem.

Next, we look at the effect of the infusion of a unit of energy at the point x = s. This infusion
suggests that some type of singular behavior must occur in g(x, s) when x = s. Integrate
Lgε (x, s) = fs (x) from 0 to l to obtain
 s+ε
′ x=s+ε
−pgε (x, s) x=s−ε + q(x)gε (x, s) dx = 1,
s−ε

where the prime indicates differentiation with respect to x. As ɛ tends to 0, the integral tends to
zero because the temperature gε (x, s) must remain bounded. Thus,

−pg ′ (x, s)
x=s+
x=s−
= 1,
1
g ′ (x, s)
x=s+
x=s−
=−
p(s)

because p(x) is continuous; hence,

x=s+
∂g(x, s) 1
=− ;
∂x x=s− p(s)

that is, the derivative of the Green’s function with respect to x is discontinuous at x = s and has
a jump of −1/p(s) there. We will show later that the Green’s function is characterized by the
foregoing properties.
It is informative to think about the solution formula (1.14) in a slightly different way, in
terms of inverse processes. The Green’s function g(x, s), called a kernel in this context, defines
18 Sturm-Liouville Problems: Theory and Numerical Implementation

an integral operator G that transforms a continuous function f into another continuous func-
tion Gf defined by
 l
Gf (x) = g(x, s)f (s) ds.
0

Since (1.13) is uniquely solved by

 l
y(x) = g(x, s)f (s) ds = Gf (x),
0

the Sturm-Liouville differential operator L together with its boundary conditions and integral
operator G are related by

Ly = f if and only if y = Gf .

The integral operator G is the inverse of the differential operator L.

Finally, the Green’s function enables us to express a Sturm-Liouville eigenvalue problem as
an eigenvalue problem for an integral operator: simply replace f by λry in (1.13) and (1.14) to
find that the Sturm-Liouville eigenvalue problem Ly = λry can be expressed as
 l
y(x) = λ g(x, s)r(s)y(s) ds.
0

The eigenvalues λ of (1.13), that is, the values of λ for which (1.13) has a nontrivial solution
y, are also called the eigenvalues of the kernel g(x, s)r(s). This conversion to an integral equa-
tion eigenvalue problem will be our principal means for studying Sturm-Liouville eigenvalue
problems in Chapters 4, 5, and 6.

1.11 The Path Ahead

The theoretical results that stand behind applications of Sturm-Liouville boundary value
and eigenvalue problems divide roughly into two threads. The first thread concerns the exis-
tence and basic properties of eigenvalues, the orthogonality of the eigenfunctions, and the con-
vergence of eigenfunction expansions. The second thread concerns the oscillatory behavior of
the eigenfunctions, the nature of their zeros, and the approximation properties they possess
that are very much like those of ordinary polynomials. Both threads are approached most nat-
urally by expressing a Sturm-Liouville eigenvalue problem as a corresponding integral equa-
tion eigenvalue problem. Indeed, much of the theory of integral equations was motivated
and developed in order to analyze eigenvalues problems that arose in the realm of differential
equations. The conversion to an equivalent integral equation is advantageous for two primary
reasons – the integral operator defined by the Green’s function is better behaved than the dif-
ferential operator in the original problem and the boundary conditions are built into the
Green’s function, which avoids the need to deal separately with side conditions defined by
additional equations.
Once the theoretical results of both threads are established, questions of practical imple-
mentation arise. Effective numerical procedures for the approximation of eigenvalues and
eigenfunctions of both regular and singular Sturm-Liouville problems are needed because
explicit evaluation of the eigenvalues and eigenfunctions is rarely possible.
 Setting the Stage 19

We describe the main problems addressed by the two threads and state a few key results in
the sections, Thread I and Thread II, that follow. All of the issues raised will be addressed more
fully later in the book.

1.11.1 Thread I
The vast majority of eigenvalue problems that come up in practice are self-adjoint. For now
it is enough to know that the Green’s function g(x, s) associated with a self-adjoint problem
with all real-valued data is symmetric: g(x, s) = g(s, x). Sturm-Liouville eigenvalue problems
with real-valued data and separated boundary conditions are self-adjoint. (Separated means
that each boundary condition only involves the function and its first derivative at one end-
point.) Periodic boundary conditions also determine self-adjoint problems. The discussion
that follows is restricted to the case of separated boundary conditions because they occur
most frequently in applications.
Our overall approach is as follows. An eigenvalue problem will be reduced to Sturm-
Liouville form: a differential equation of the form
−(p(x)y ′ )′ + (q(x) − λr(x))y = 0
for a , x , b together with appropriate boundary conditions. In simple cases, the eigenvalues
and eigenfunctions can be found explicitly. In general, explicit solutions are not available and
to obtain theoretical properties of the eigenvalues and eigenfunctions, we replace the eigen-
value problem by an equivalent integral equation
b
y(x) = λ g(x, s)r(s)y(s) ds,
a

where g(x, s) is the Green’s function corresponding to the Sturm-Liouville differential operator
and its associated boundary conditions.
To make clearer the issues to be faced, we introduce them via the diffusion problem (1.10)
with f (x, t) = 0 (no sources or sinks along the rod), homogeneous Dirichlet boundary condi-
tions, and initial temperature distribution g(x) now relabeled f (x):
⎧
⎨ ut = (p(x)ux )x − q(x)u, 0 , x , l, t . 0,
u(x, 0) = f (x), 0 ≤ x ≤ l, (1.15)
⎩
u(0, t) = 0, u(l, t) = 0, t ≥ 0.
When separation of variables is used to seek nontrivial separated solutions u(x, t) = X(x)T (t)
that satisfy the diffusion equation and the homogeneous boundary conditions, one is led to the
eigenvalue problem


(p(x)X ′ )′ − q(x)X + λX = 0, 0 , x , l,
(1.16)
X(0) = 0, X(l) = 0;

and the companion equation T ′ + λT = 0 for the time factor. A key step in separation of var-
iables is to superpose the separated solutions with the aim of satisfying all the remaining con-
ditions in the initial boundary value problem at hand. For this to work, one almost always
needs an infinite superposition of the separated solutions. That is, the eigenvalue problem
must have an infinite number of eigenvalues and corresponding eigenfunctions. We will estab-
lish these properties for general regular and singular Sturm-Liouville eigenvalue problems. In
(1.15), as for most problems with physically realistic boundary conditions, the eigenvalues are
all real, simple,
0 , λ 1 , λ2 , · · · , λn , · · · and λn  1 as n  1.
20 Sturm-Liouville Problems: Theory and Numerical Implementation

The corresponding eigenfunctions ϕ1 (x), . . . , ϕn (x), . . . can be chosen real and orthonormal,
l
ϕn (x)ϕm (x) dx = δnm
0

where δnm is the Kronecker delta with value 1 if n = m and value 0 otherwise. For (1.15), the
corresponding functions T (t) in the separated solutions are multiples of Tn (t) = e−λn t .
Since any (finite) linear combination of the separated solutions will satisfy the diffusion
equation and the homogeneous boundary conditions, it is reasonable to expect that an infinite
superposition
1
u(x, t) = αn e−λ
n
nt
ϕn (x)
n=1

will too. This is true if the infinite series is suitably convergent. This is another of the issues we
must face. Finally we want the series to satisfy any remaining conditions imposed by the model.
Here we want
1
u(x, 0) = αn ϕn (x) = f (x).
n=1

That is we need to know that any reasonable function f (x) can be represented by an eigenfunc-
tion expansion. So two more issues emerge. What do we mean by a reasonable function?
In what sense does the series converge?
The questions raised above are addressed in the Hilbert-Schmidt theorem and its corollar-
ies, which are among the principal results of Chapter 3. Applications to Sturm-Liouville
problems are given in Chapters 4, 5, and 6.

1.11.2 Thread II
We continue to assume that the Sturm-Liouville eigenvalue problem has separated
boundary conditions, just as in Thread I. In this case, each eigenvalue has a uniquely deter-
mined corresponding eigenfunction up to nonzero constant multiples and the eigenvalues
can be listed as
λ0 , λ1 , · · · , λn , · · ·
with λn  1 as n  1. The corresponding eigenfunctions are denoted by
ϕ0 (x), ϕ1 (x), ϕ2 (x), . . . , ϕn (x), . . .
and are continuous and orthogonal in the underlying interval, say 0 ≤ x ≤ 1. The eigen-
functions ϕ0 (x), ϕ1 (x), ϕ2 (x), . . . , ϕn (x) have oscillatory and approximation properties analo-
gous to those possessed by ordinary polynomials of degree n. Several of these properties
are listed later in this section and established in later chapters. A unified approach to the
properties we have in mind began with O. D. Kellogg in 1916–1918, [26] and [27], when he
introduced what are now called Kellogg kernels. He wanted to determine what proper-
ties of the Green’s function of a Sturm-Liouville eigenvalue problem, or more generally of
a real-valued symmetric kernel, would imply all the familiar oscillatory properties of the
eigenfunctions. Kellogg discovered the properties from a purely mathematical perspective.
Later, beginning in the mid 1930s, Gantmacher and Krein [16] significantly extended
Kellogg’s pioneering work and added an important physical perspective: a few simple physical
properties of an elastic continuum imply that its influence function must be a Kellogg kernel
and, hence, have the properties Kellogg discovered. The investigations of Gantmacher and
Krein also extended Kellogg’s results to include certain nonsymmetric kernels. See Pincus
 Setting the Stage 21

[32] for a much deeper analysis of the contributions of Kellogg and of Gantmacher and Krein
than is given here. Just as in Thread I, the unified treatment we will give later of the oscillatory
and approximation properties of Sturm-Liouville eigenvalue problems is made possible by con-
verting the Sturm-Liouville eigenvalue problem into an equivalent integral equation eigen-
value problem.
Kellogg starts his 1916 paper with an example of three continuous, piecewise linear, orthog-
onal functions on [0, 1], say ψ 0 (x), ψ 1 (x), and ψ 2 (x) such that ψ 0 (x) has no zero in the interval
and ψ 1 (x) and ψ 2 (x) each have exactly one zero in [0, 1] that occurs at an interior point of the
interval. His point is that the familiar properties of the orthogonal eigenfunctions of a real,
symmetric kernel, say k(x, s), cannot all be a consequence only of their orthogonality. Kellogg
goes on to show that the familiar oscillatory and approximation properties of the eigenfunc-
tions ϕ0 (x), ϕ1 (x), ϕ2 (x), . . . hold if
det [ϕi (xj )]n×n . 0 (1.17)
for all 0 , x1 , · · · , xn , 1, i = 0, . . . , n − 1, and n = 1, 2, . . . . In particular, Kellogg showed
that the determinantal inequalities imply:

• Given n + 1 distinct
 points in (0, 1) and given n + 1 values, there is a unique ϕ-polyno-
mial of the form nk=0 ak ϕk (x) that takes on the given values at the given points.

• If a nonzero ϕ-polynomial nk=0 ak ϕk (x) vanishes at n distinct points, then it changes
sign at those points.
• ϕn (x) has exactly n zeros in (0, 1) and changes its sign at each of these zeros.

• The zeros of ϕn−1 (x) and ϕn (x) in (0, 1) strictly interlace.


• For m ≤ n, a nonzero ϕ-polynomial nk=m ak ϕk (x) changes sign at least m times and at
most n times on (0, 1).

Kellogg concluded his 1916 paper by stating that it would be desirable to find conditions
on the kernel k(x, s) that imply the inequalities (1.17). He did just that in his 1918 paper.
Kellogg’s conditions from 1918 are:
K1. det [k(xi , xj )]n×n . 0 for 0 , x1 , · · · , xn , 1,
0 ≤ x1 ≤ · · · ≤ xn ≤ 1,
K2. det [k(xi , sj )]n×n ≥ 0 for
0 ≤ s1 ≤ · · · ≤ sn ≤ 1,
for n = 1, 2, 3, . . . and all choices of x1, x2, . . . , xn and s1, s2, . . . , sn that satisfy the given con-
ditions. As noted above, Gantmacher and Krein significantly extended Kellogg’s work,
establishing the existence of an infinite sequence of positive eigenvalues and corresponding
eigenfunctions for nonsymmetric kernels that satisfy Kellogg’s conditions and explaining the
physical meaning of the Kellogg conditions. Reference [16], which concentrates on the sym-
metric case for ease of exposition, gives a rich account of the interplay between the Kellogg
conditions and the oscillatory behavior of discrete and continuous mechanical systems. See
Appendix C for a physical motivation of the Kellogg conditions.

1.11.3 Finding Eigenvalues and Eigenfunctions

We continue in the context of the initial boundary value problem (1.15) and the corre-
sponding eigenvalue problem (1.16)


(p(x)y ′ )′ − q(x)y + λy = 0, 0 , x , l,
y(0) = 0, y(l) = 0,
22 Sturm-Liouville Problems: Theory and Numerical Implementation

where X(x) has been replaced by y(x) for convenience. The eigenvalues and eigenfunctions for
this problem can only be found explicitly for simple choices of p(x) and q(x). So we must face a
basic question. How can we actually construct the eigenvalues λn and the eigenfunctions ϕn (x),
both theoretically and numerically?
We will use the following procedure to address both issues. Although the basic idea we are
about to describe is not new, the proofs needed to justify it for both regular and singular Sturm-
Liouville problems are new, as far as we know. The basic idea is this: To solve the eigenvalue
problem


(p(x)y ′ )′ − q(x)y + λy = 0, 0 , x , l,
y(0) = 0, y(l) = 0

consider the initial value problem

(p(x)u ′ )′ − q(x)u + λu = 0,
u(0) = 0, u ′ (0) = 1

and denote its solution by u(x, λ) for 0 ≤ x ≤ l. The solution u(x, λ) will be an eigenfunction of
the eigenvalue problem and λ will be its corresponding eigenvalue if

u(l, λ) = 0.

So more issues arise that we must face. How do we establish theoretically the global solvability
of the initial value problem? How do we know that u(l, λ) = 0 has an infinite number of solu-
tions? And, once these questions are answered, how do we compute accurate numerical
approximations of the eigenvalues and eigenfunctions? The answers to these questions involve
basic existence, uniqueness, and continuous dependence results for ordinary differential equa-
tions, the Newton-Raphson method, and initial value problems solvers for ordinary differential
equations. The existence, uniqueness, and continuous dependence results needed are standard
for regular Sturm-Liouville problems when p′ is continuous. They are either new or not well
known for singular Sturm-Liouville problems or when p is merely continuous for regular prob-
lems. All of these matters are addressed in Chapters 4, 5, 6, and 7.

1.12 Intrinsic Interest of Eigenvalues

Eigenvalues are often of interest in their own right. In the discussion of the heat conduction
model (1.10) we noted that the eigenvalues and eigenfunctions for the basic diffusion or heat
conduction model, ut = auxx with a . 0 a constant and with Dirichlet boundary conditions,
are λn = (nπ/l)2 and Xn (x) = sin nπx/l. For this problem, the series for the temperature
u(x, t) given in Section 1.11.1 is
1 
u(x, t) = cn exp (−aλn t) sin λn /l.
n=1

The first term in the expansion of the solution contains the factor exp (−aλ1 t) that determines
the overall rate at which the solution decays in time. In particular, it determines how soon the
transient effects due to the initial conditions can be neglected and when a steady state is
reached, or in the case of forcing, how soon only the forcing terms have an appreciable effect
on the solution.
 Setting the Stage 23
√
In the case of acoustics, or vibration problems, c λn , where c is the speed of propagation of
the disturbance, gives the frequency of the vibrations.
√ In the case of a piano string, those values
are basis
√ √for tuning the piano. The frequency c λ 1 is called the fundamental tone and the ratio,
c λn /c λ1 = n is an integer, a fact discovered empirically by the Pythagoreans.
In studying chain nuclear reactions, one is led to an equation of the form ut = aΔu + ku,
where u represents the number of neutrons per  unit volume.
 Separation
 of variables with
u = T (t)v(x) leads to the series solution u = 1 n=1 exp (k − aλ n )t v n (x) where λn and vn (x)
are the eigenvalues and eigenfunctions for −Δv together with appropriate boundary condi-
tions. The eigenvalues depend upon the geometry of the container. If k is greater than aλn
for some value of n, the result is a reaction out of control, if aλn is greater than k for all n,
the reaction damps out, and if aλ1 = k, the reaction is critical and one has a controlled reaction
which can be used to generate electric power.
In quantum mechanics, the eigenvalues of the Schrödinger equation yield the energy levels
of, say, electrons (see [43] or [45]).
Eigenvalues occur in many other contexts. We have seen a simple example in the case of
Euler buckling, but they arise also in more general buckling problems. They arise in mathemat-
ical biology, in particular in the study of populations.
In applied problems, it is often the case that only the first eigenvalue is of critical interest
because it determines how the system will behave for large values of the time. In such instances,
the first eigenvalue and eigenfunction or a small number of eigenvalues and eigenfunctions can
be used to accurately represent the solution as it evolves in time.

1.13 Real Versus Complex Solutions

There are occasions when it is desirable or necessary to deal with complex-valued solu-
tions to differential equations. Most applied problems lead to differential equations and side
conditions that involve only real data and the solutions of interest are real-valued or must
be real-valued. The catalog of problems just discussed are typical examples.
Perhaps the most important situation involving Sturm-Liouville problems and where
complex-valued functions enter is when separation of variables is used in the Schrödinger
equation

h̵ 2
ih̵ Ψt = − ΔΨ + V (x, t)Ψ,
2m

where i is the imaginary unit, h̵ is Planck’s constant divided by 2π, Ψ is a wave function, and
V is a real-valued potential energy function. Complex-valued solutions must be considered
in any situation in which the differential equation or side conditions involve complex-valued
data.
In the chapters that follow we often allow the coefficients in a differential equation to
be complex-valued and likewise any constants in the problem may be complex numbers.
The results obtained about solutions and their properties apply to any complex-valued or
real-valued solutions that may exist. Frequently, if the equations and data involve only
real quantities, it is natural to expect that the solutions must be real-valued. We establish
such results for initial value problems, for boundary value problems, and for Green’s func-
tions in sufficient generality to cover scenarios in typical applications. Corresponding results
are established for eigenfunctions of eigenvalue problems whose eigenvalues are known to
be real.
24 Sturm-Liouville Problems: Theory and Numerical Implementation

A final observation is in order. If a problem is expressed in terms of linear differential and

linear boundary condition equations any of which can be either homogeneous or inhomoge-
neous but involve only real-valued data, then the real part of any complex-valued solution
of the problem is a real-valued solution and the imaginary part is a real-valued solution of
the corresponding homogeneous problem.
Chapter 2
Preliminaries

We collect together in this chapter background results from calculus, analysis, and linear alge-
bra that play a prominent role later, as we study Sturm-Liouville boundary value and eigen-
value problems. The chapter serves as a convenient reference and avoids the obligation to
develop background material in the midst of arguments in later chapters that are focused on
differential and integral equations.
All readers should at least skim through the chapter to become familiar with the notation
that we use. The notation is standard, for the most part. Readers who are familiar with the
topics in the chapter can move on quickly to later chapters, perhaps never needing to refer
back. For other readers, we have endeavored to present the material as a focused, readable
introduction to essential background results that can be consulted as needed.
We emphasize that although solutions and other functions may sometimes assume complex
values, the domains of all solutions and other functions are either sets of real numbers or sets in
real n-dimensional space.

2.1 Euclidean Spaces

We use standard notation and denote the real numbers equipped with the usual algebraic
operations by R and the complex numbers equipped with the usual operations by C.

2.1.1 Real Euclidean Spaces

Real n-dimensional Euclidean space is denoted by Rn,
Rn = {x = (x1 , . . . , xn ) : xj a real number for each j}.
The elements x of Rn are called points or more often vectors, when we identify x with the posi-
tion vector from the origin to the point x. Vectors in Rn are added and multiplied by scalars
(real numbers) componentwise. The norm (length or magnitude) of a vector x is
 1/2


n
x = |xj |2
.
j=1

For any vectors x and y and scalar α, the norm satisfies

x ≥ 0 with equality only if x = 0,
αx = |α| x,
x + y ≤ x + y (triangle inequality).
The usual inner product in Rn enables us to define angles between vectors x and y in Rn. It is
defined by


n
kx, yl = xj yj
j=1

25
26 Sturm-Liouville Problems: Theory and Numerical Implementation

and related to the norm through


x = kx, xl

and the Cauchy-Schwarz inequality

|kx, yl| ≤ xy.

The usual inner product is linear in its first variable and is symmetric: for any vectors x, y, z and
scalars α and β
kαx + βy, zl = αkx, yl + βky, zl,
kx, yl = ky, xl.

Consequently, the inner product is linear in its second variable as well.

Fix a , b. We use the following notation for a simplex in Rn
Δn = {x = (x1 , x2 , . . . , xn ) [ Rn : a ≤ x1 ≤ x2 ≤ · · · ≤ xn ≤ b}.

So Δ1 is the closed interval [a, b] on the real line, Δ2 is the triangle with vertices (a, ca) , (a,
b) and (b, b) in the Euclidean plane, Δ3 is tetrahedron in 3-space. The set of points inside the
simplex, its interior, is
◦
Δn = {x = (x1 , x2 , . . . , xn ) [ Rn : a , x1 , x2 , · · · , xn , b}.

2.1.2 Complex Euclidean Spaces

Apart from simplices, corresponding language and results hold in complex Euclidean
spaces. Complex n-dimensional Euclidean space is denoted by Cn,

Cn = {z = (z1 , . . . , zn ) : zj a complex number for each j}.

If n = 1, each complex number c can be expressed as c = a + ib where a and b are real numbers.
The real number a is the real part of c and is denoted by Re(c). The real number b is the imag-
inary part of c and is denoted by Im(c). The complex
√  conjugate of c is c = a − ib. The absolute
value of a complex number is |c| = a 2 + b2 .
The elements z of Cn are called points or more often vectors, when we identify z with the
position vector from the origin to the point z. Vectors in Cn are added and multiplied by scalars
(complex numbers) componentwise. The norm (length or magnitude) of a vector z is
 1/2

n
z = |zj |2 .
j=1

For any vectors z and w and scalar α, the norm satisfies

z ≥ 0 with equality only if z = 0,

αz = |α|z,
z + w ≤ z + w (triangle inequality).

The usual inner product in Cn is defined by

n
kz, wl = j
zj w
j=1
 Preliminaries 27

and related to the norm through


z = kz, zl
and the Cauchy-Schwarz inequality
|kz, wl| ≤ z w.
The usual inner product is linear in its first variable and is complex symmetric: for any vectors
z, w, u and scalars α and β
kαz + βw, ul = αkz, ul + βkw, ul,
kz, wl = kw, zl.
Consequently, the inner product in Cn is conjugate linear in its second variable

 wl.
ku, zl + βku,
ku, αz + βwl = α

2.1.3 Elements of Convergence

A sequence {xn }1 n=1 in real or complex Euclidean space is an ordered list of points in
that space. Other notations for a sequence are {xn } or simply xn, with the range of the index
n understood from the context. A sequence {xn }1 n=1 converges to x, if xn − x  0 as
n  1; that is, given any ε . 0 there is an integer N, dependent on ε, such that
xn − x , ε whenever n . N.
A subsequence of a sequence {xn }1 1
n=1 is an ordered sub-list {xnk }k=1 where 1≤n1,n2,· · · .
If a sequence xn in a Euclidean space converges, say with limit x, then given ε . 0 there is
an integer N such that xn − x , ε/2 whenever n . N. Consequently,
m, n ≥ N implies that xm − xn  ≤ xm − x + x − xn  , ε.
A sequence with this property is called a Cauchy sequence. So convergent sequences in
Euclidean spaces are Cauchy (sequences). The converse is true: every Cauchy sequence in a
Euclidean space converges. This so-called Cauchy criterion for convergence gives a means
of establishing convergence in Euclidean spaces without knowing the limit in advance.

2.1.4 Upper Bounds and Sups

Let X be a nonempty set of real numbers. A real number b is an upper bound for X if x ≤ b
for all x in X, in which case X is said to be bounded above. A real number l is the least upper
bound of X, also called the supremum of X, if l is an upper bound for X and l ≤ b whenever b
is an upper bound for X. The supremum of X is usually denoted by sup X.
A fundamental property of the real number system, which is equivalent to the Cauchy
criterion in R, is:
every nonempty set in R that is bounded above has a supremum.
We will need the following results that follow directly for the definition of a supremum.

Lemma 1 Let X be a nonempty set of real numbers that is bounded above. A real number l is
the supremum of X if and only if l is an upper bound for X and given any ε . 0 there is an
element x in X such that l − ε , x ≤ l.

Corollary 2 If l = sup X then there is a sequence xn of elements in X that converge to l.

The terms lower bound, bounded below, and greatest lower bound (or infimum) are defined
by reversing the inequalities in the earlier definitions.
28 Sturm-Liouville Problems: Theory and Numerical Implementation

2.1.5 Closed and Compact Sets

There are several equivalent ways to define closed sets and compact sets in Euclidean
spaces. We prefer sequential definitions. A set S in Rn or Cn is closed if whenever a sequence
{xn } with elements in S converges its limit lies in S. A set S in Rn or Cn is compact if every
sequence of elements in S contains a convergent subsequence whose limit lies in S.

Theorem 3 (Heine-Borel) A set S in Rn or Cn is compact if and only if it is closed and

bounded.
A proof can be found in any advanced calculus book.

2.2 Calculus and Analysis

Sturm-Liouville problems involve analyzing solutions y(x) of a differential equation of
the form
−(p(x)y ′ (x))′ + q(x)y(x) = f (x) for a , x , b,
where y(x) is subject to certain boundary conditions at x = a and x = b or is subject to initial
conditions. Since y(x) is required to satisfy the differential equation, it is subject to certain
smoothness (differentiability) requirements on a , x , b. It is important for applications to
know if, and to what extent, that smoothness extends to the behavior of y(x) at x = a and
x = b. To answer this question among others, we will make use of standard concepts and results
from advanced calculus on continuity, differentiability, and integrability. The most important
of them are stated here for convenient reference. Occasionally, a proof is given, especially for
less familiar results. Missing arguments can be found in almost any book on advanced calculus.
Reliable references include [15], [29], [30], [34] and [39].

2.2.1 Continuity
A real or complex-valued function f defined on a set S in Euclidean space is continuous at x0
in S if given any ε . 0 there is a δ . 0, dependent on x0 and on ε, such that |f (x) − f (x0 )| , ε
whenever x ∈ S satisfies |x − x0 | , δ. A function f is continuous on S if it is continuous
at every point of S. A function f is uniformly continuous on S if given any ε . 0 there is
a δ . 0, dependent only on ε, such that |f (x) − f (x ′ )| , ε whenever x, x ′ [ S satisfy
|x − x ′ | , δ. Uniform continuity on a set S means that there is a single δ . 0 in the definition
of continuity of f at x0 that works simultaneously for all x0 in S.

Theorem 4 (Maximum Minimum Value Theorem) A real-valued continuous function

defined on a closed and bounded set in finite dimensional Euclidean space assumes its maxi-
mum and minimum values at points in the set.

Theorem 5 A real or complex-valued continuous function defined on a closed and bounded set
in finite dimensional Euclidean space is uniformly continuous on the set.

Theorem 6 (Intermediate Value Theorem) If f (x) is a real-valued continuous function on an

interval [a, b] and C is a value strictly between A = f (a) and B = f (b), then there is a point c
with a , c , b such that f (c) = C .

Proposition 7 If a function f (x) with values in a Euclidean space is uniformly continuous on

an open interval (a, b), then f (x) has a unique extension by continuity to a continuous function
on the closed interval [a, b].