Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.

org/terms-privacy

OT164_TREFETHEN_FM_V5.indd 1
Extended Edition
Approximation Theory
and Approximation Practice

10/14/2019 9:20:55 AM
Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

The cover photo shows the Thames River and Port Meadow in Oxford, United
Kingdom. This scene has been largely unchanged for the past thousand years.

OT164_TREFETHEN_FM_V5.indd 2 10/14/2019 9:20:56 AM

Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Approximation Theory
and Approximation Practice
Extended Edition

Lloyd N. Trefethen
University of Oxford
Oxford, United Kingdom

Society for Industrial and Applied Mathematics

Philadelphia

OT164_TREFETHEN_FM_V5.indd 3 10/14/2019 9:20:57 AM

Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Copyright © 2020 by the Society for Industrial and Applied Mathematics

10 9 8 7 6 5 4 3 2 1

All rights reserved. Printed in the United States of America. No part of this book may
be reproduced, stored, or transmitted in any manner without the written permission
of the publisher. For information, write to the Society for Industrial and Applied
Mathematics, 3600 Market Street, 6th Floor, Philadelphia, PA 19104-2688 USA.

MATLAB is a registered trademark of The MathWorks, Inc. For MATLAB product

information, please contact The MathWorks, Inc., 3 Apple Hill Drive, Natick, MA
01760-2098 USA, 508-647-7000, Fax: 508-647-7001, [email protected],
www.mathworks.com.

Publications Director Kivmars H. Bowling

Executive Editor Elizabeth Greenspan
Developmental Editor Mellisa Pascale
Managing Editor Kelly Thomas
Production Editor Louis R. Primus
Copy Editor Susan Fleshman
Production Manager Donna Witzleben
Production Coordinator Cally A. Shrader
Graphic Designer Doug Smock

Library of Congress Cataloging-in-Publication Data

Names: Trefethen, Lloyd N. (Lloyd Nicholas), author.
Title: Approximation theory and approximation practice / Lloyd N.
Trefethen.
Description: Extended edition. | Philadelphia : Society for Industrial and
Applied Mathematics, [2019] | Includes bibliographical references and
index. | Summary: “Including two new appendices, this extended edition
of Trefethen’s textbook covers classical polynomial and rational
approximation theory for the twenty-first century”-- Provided by
publisher.
Identifiers: LCCN 2019027107 (print) | LCCN 2019027108 (ebook) | ISBN
9781611975932 (hardback) | ISBN 9781611975949 (ebook)
Subjects: LCSH: Approximation theory.
Classification: LCC QA221 .T73 2019 (print) | LCC QA221 (ebook) | DDC
511/.4--dc23
LC record available at https://lccn.loc.gov/2019027107
LC ebook record available at https://lccn.loc.gov/2019027108

is a registered trademark.

OT164_TREFETHEN_FM_V5.indd 4 10/14/2019 9:20:57 AM

 Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

OT164_TREFETHEN_FM_V5.indd 5
For Kate

10/14/2019 9:20:57 AM
 Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

OT164_TREFETHEN_FM_V5.indd 6
10/14/2019 9:20:57 AM
Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Contents

Preface to the Extended Edition ix

1. Introduction 1
2. Chebyshev Points and Interpolants 7
3. Chebyshev Polynomials and Series 13
4. Interpolants, Projections, and Aliasing 25
5. Barycentric Interpolation Formula 33
6. Weierstrass Approximation Theorem 43
7. Convergence for Differentiable Functions 49
8. Convergence for Analytic Functions 55
9. Gibbs Phenomenon 63
10. Best Approximation 73
11. Hermite Integral Formula 81
12. Potential Theory and Approximation 89
13. Equispaced Points, Runge Phenomenon 95
14. Discussion of High-Order Interpolation 103
15. Lebesgue Constants 107
16. Best and Near-Best 117
17. Orthogonal Polynomials 123
18. Polynomial Roots and Colleague Matrices 133
19. Clenshaw–Curtis and Gauss Quadrature 143
20. Carathéodory–Fejér Approximation 155
21. Spectral Methods 165
22. Linear Approximation: Beyond Polynomials 177
23. Nonlinear Approximation: Why Rational Functions? 189
24. Rational Best Approximation 199
25. Two Famous Problems 209
26. Rational Interpolation and Linearized Least-Squares 221
27. Padé Approximation 235
28. Analytic Continuation and Convergence Acceleration 251
Appendix A: Six Myths of Polynomial Interpolation and Quadrature 263
Appendix B: “ATAP for Periodic Functions” (SISC 2015) 273
Appendix C: “ATAP for the Unit Disk” (SINUM 2014) 297
References 329
Index 357

vii
Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy
Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Preface to the Extended Edition

Approximation Theory and Approximation Practice (“ATAP”), originally published

in 2013, concerns approximation of nonperiodic functions on the interval [−1, 1], the
Chebyshev setting of constructive analysis. But this is just one of three essentially
equivalent situations:

Chebyshev, for nonperiodic functions of x ∈ [−1, 1],

Fourier, for periodic functions of θ ∈ [−π, π],
Laurent/Taylor, for functions of z on the unit circle |z| = 1.

As discussed on p. 14, my original plan had been to give equal treatment to all three
settings, but it became clear that a book written to that plan would be ponderous.
So Chebyshev became the subject, since it was the most important of the three for
numerical computation and also the least understood.
In this extended edition, we have found a way to give a kind of summary of the
parallel mathematics of the Fourier and Laurent/Taylor cases without changing the
main text. We have added appendices reprinting two papers published recently in
SIAM journals:

Appendix B. Fourier.
G. B. Wright, M. Javed, H. Montanelli and L. N. Trefethen
(2015), Extension of Chebfun to periodic functions, SIAM J. Sci. Comput.
37, C554–C573.
Appendix C. Laurent/Taylor.
A. P. Austin, P. Kravanja and L. N. Trefethen (2014), Numerical
algorithms based on analytic function values at roots of unity, SIAM J.
Numer. Anal. 52, 1795–1821.

ix
 x Approximation Theory and Approximation Practice

In the first of these papers, readers will see a list of 10 parallels between Fourier
and Chebyshev spelled out in Table B.1, followed by a discussion of 10 respects in
Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

which the two cases differ. These differences are not deep mathematically (there
are no deep mathematical differences between Fourier and Chebyshev!), but they
involve conceptual twists. Readers may also enjoy Figure B.9, highlighting the
continuity from Runge in 1904 to his descendants in 2015. One may think of Wright
et al. as a kind of “ATAP for periodic functions.” Specifically, it is noted on p. 275
that Section B.2 of the paper corresponds to Chapter 3 of this book, Section B.3 to
Chapters 2, 4, and 5, and Section B.4 to Chapters 6, 7, 8, 10, and 19.
In the second paper, a notable feature of the presentation is a sequence of 13
short MATLAB code segments (summarized on pp. 320–321), many just one line
long, for interpolation, differentiation, integration, and zero- and polefinding from
data values at roots of unity. The links to Chapters 2–5, 18, and 19 of this book are
very close. The history of this mathematics is collected in Figure C.1 of the paper,
and Figure C.4 shows a striking image related to the Jentzsch, Walsh, and Blatt–
Saff theorems discussed here on pp. 140–141. Austin et al. is a kind of “ATAP for
the unit disk.”
Since the first edition of ATAP was published, the use of Chebfun has grown
around the world. The code was completely rewritten during 2013–14 by a team
led by Nick Hale, and version 5 was released in June 2014. By the combined
efforts of Alex Townsend, Behnam Hashemi, Grady Wright, Heather Wilber, and
Nicolas Boullé, Chebfun has been expanded to compute with functions not only on
intervals but also on rectangles, boxes, spheres, disks, and balls. Multidimensional
approximation is not treated in this book, but it is a hot topic in the era of big data,
and these new capabilities of Chebfun make use of low-rank compression ideas that
are familiar to today’s engineers and data scientists.
I have taken the opportunity of the new edition to correct a number of errors
pointed out by readers and colleagues, of whom I would like in particular to acknowl-
edge Folkmar Bornemann, Behnam Hashemi, Mohsin Javed, Yuji Nakatsukasa,
Grady Wright, and Kuan Xu. The corrections have included updates to Chebfun
syntax such as the replacement of interp1, chebpolyplot, and chebellipseplot
by chebfun.interp1, plotcoeffs, and plotregion, respectively.
Although this is not a thoroughgoing revision of ATAP, about a dozen new
references have been added. Let me mention two areas of particularly interest-
ing developments since 2013. One is Chebyshev spectral discretization of ordinary
differential equations, a subject reflected in the new SIAM book Exploring ODEs
[Trefethen, Birkisson & Driscoll 2018]. ATAP and Exploring ODEs are siblings,
two mathematical textbooks based on Chebfun, and a PDF file of the latter is
freely available at http://people.maths.ox.ac.uk/trefethen/ExplODE. Cheb-
fun now makes systematic use of the rectangular spectral discretizations introduced
by Driscoll and Hale [2015]; a leader of this effort was Ásgeir Birkisson. Details can
be found in [Aurentz & Trefethen 2017a] and in Appendix A of Exploring ODEs.
The other area to highlight is rational approximation. As described in the last
six chapters of this book, rational approximations are particularly effective for func-
tions that have singularities since they can cluster poles and zeros near them. On
the other hand, they are traditionally hard to compute. A few years after this book
was published, a flexible new method for computing rational approximations on
arbitrary real or complex domains, the AAA algorithm, was introduced by Nakat-
 Preface to the Extended Edition xi

sukasa, Sète, and myself [2018]. The AAA algorithm gets its power from combining
a barycentric rational representation with an adaptive choice of support points.
Downloaded 03/23/22 to 2.207.81.236 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy

Most recently, attempts to extend AAA have led to the introduction of lightning
Laplace solvers, which exploit rational functions to solve certain partial differen-
tial equations with remarkable speed and accuracy on regions with corners, where
solutions almost invariably have singularities [Gopal & Trefethen 2019a,b]. The
approximation theory underlying these new methods is built on generalizations for
rational functions of the Hermite integral formula and the potential theory presented
in Chapters 11–12 for polynomials. Together, these encouraging developments in
the numerical use of rational functions suggest, as we like to say (see the references
labeled 1885a and 1964), that “Runge” is joining forces with “Newman.”