Functional Analysis In Applied Mathematics And

Engineering First Edition Pedersen download

https://ebookbell.com/product/functional-analysis-in-applied-
mathematics-and-engineering-first-edition-pedersen-9953022

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Applied Functional Analysis 3rd Solutions Manual John Tinsley Oden

https://ebookbell.com/product/applied-functional-analysis-3rd-
solutions-manual-john-tinsley-oden-46249204

Functional Analysis And Applied Optimization In Banach Spaces

Applications To Nonconvex Variational Models 1st Edition Fabio Botelho
Auth

https://ebookbell.com/product/functional-analysis-and-applied-
optimization-in-banach-spaces-applications-to-nonconvex-variational-
models-1st-edition-fabio-botelho-auth-4930758

Recent Advances In Applied Nonlinear Dynamics With Numerical Analysis

Fractional Dynamics Network Dynamics Classical Dynamics And Fractal
Dynamics Changpin Li

https://ebookbell.com/product/recent-advances-in-applied-nonlinear-
dynamics-with-numerical-analysis-fractional-dynamics-network-dynamics-
classical-dynamics-and-fractal-dynamics-changpin-li-5147398

Solved Problems In Analysis As Applied To Gamma Beta Legendre And

Bessel Functions Reprint 1991 Orin J Farrell

https://ebookbell.com/product/solved-problems-in-analysis-as-applied-
to-gamma-beta-legendre-and-bessel-functions-reprint-1991-orin-j-
farrell-5423700
Functional Analysis In Interdisciplinary Applicationsii Icaam Lefkosa
Cyprus September 69 2018 1st Ed 2021 Allaberen Ashyralyev

https://ebookbell.com/product/functional-analysis-in-
interdisciplinary-applicationsii-icaam-lefkosa-cyprus-
september-69-2018-1st-ed-2021-allaberen-ashyralyev-51645986

Functional Analysis In Asymmetric Normed Spaces 2013th Edition Stefan

Cobzas

https://ebookbell.com/product/functional-analysis-in-asymmetric-
normed-spaces-2013th-edition-stefan-cobzas-4146754

Functional Analysis In Mechanics 2nd Ed L P Lebedev I I Vorovich

Michael J Cloud

https://ebookbell.com/product/functional-analysis-in-mechanics-2nd-ed-
l-p-lebedev-i-i-vorovich-michael-j-cloud-4158266

Functional Analysis In Clinical Treatment Peter Sturmey Eds

https://ebookbell.com/product/functional-analysis-in-clinical-
treatment-peter-sturmey-eds-4409270

Functional Analysis In Interdisciplinary Applications Astana

Kazakhstan October 2017 1st Edition Tynysbek Sh Kalmenov

https://ebookbell.com/product/functional-analysis-in-
interdisciplinary-applications-astana-kazakhstan-october-2017-1st-
edition-tynysbek-sh-kalmenov-6841718
Functional Analysis
in Applied
Mathematics
and Engineering
Studies in Advanced Mathematics
Series Editor
STEVEN G. KRANTZ
Washington University in St. Louis

Editorial Board

R. Michael Beals
Gerald B. Folland
Rutgers University
University o f Washington
Dennis de Turck
William Helton
University o f Pennsylvania
University o f California at San Diego
Ronald DeVore
Norberto Salinas
University o f South Carolina
University o f Kansas
iMwrence C. Evans
Michael E. Taylor
University o f California at Berkeley
University o f North Carolina

Titles Included in the Series

Steven R. Bell, The Cauchy Transform, Potential Theory, and Conformal Mapping
John J. Benedetto, Harmonic Analysis and Applications
John J. Benedetto and Michael W. Frazier, Wavelets: Mathematics and Applications
Albert Boggess, CR Manifolds and the Tangential Cauchy-Riemann Complex
Goong Chen andJianxin Zhou, Vibration and Damping in Distributed Systems, Voi. 1:
Analysis, Estimation, Attenuation, and Design. Voi. 2: WKB and Wave Methods,
Visualization, and Experimentation
Carl C. Cowen and Barbara D. MacCluer, Composition Operators on Spaces o f Analytic
Functions
John P. DAngelo, Several Complex Variables and the Geometry o f Real Hypersurfaces
Lawrence C. Evans and Ronald F. Gariepy, Measure Theory and Fine Properties o f
Functions
Gerald B. Folland, A Course in Abstract Harmonic Analysis
José García-Cuerva, Eugenio Hernández, Fernando Soria, and J osé-Luis Torrea, Fourier
Analysis and Partial Differential Equations
Peter B. Gilkey, Invariance Theory, the Heat Equation, and the Atiyah-Singer Index
Theorem, 2nd Edition
Alfred Gray, Modem Differential Geometry o f Curves and Surfaces with Mathematica, 2nd
Edition
Eugenio Hernández and Guido Weiss, A First Course on Wavelets
Steven G. Krantz, Partial Differential Equations and Complex Analysis
Steven G. Krantz, Real Analysis and Foundations
Kenneth L Kuttler, Modem Analysis
Clark Robinson, Dynamical Systems: Stability, Symbolic Dynamics, and Chaos, 2nd Edition
John Ryan, Clifford Algebras in Analysis and Related Topics
Xavier Saint Raymond, Elementary Introduction to the Theory o f Pseudodifferential
Operators
Robert Strichartz, A Guide to Distribution Theory and Fourier Transforms
André Unterberger and Harald Upmeier, Pseudodifferential Analysis on Symmetric Cones
James S. Walker, Fast Fourier Transforms, 2nd Edition
James S. Walker, Primer on Wavelets and their Scientific Applications
Gilbert G. Walter, Wavelets and Other Orthogonal Systems with Applications
Kehe Zhu, An Introduction to Operator Algebras
Functional Analysis
in Applied
Mathematics
and Engineering

MICHAEL PEDERSEN

CHAPMAN & HALL/CRC

Boca Raton London New York Washington, D.C.
 Library of Congress Cataloging-in-Publication Data

Pedersen, Michael.
Functional analysis in applied mathematics and engineering /
Michael Pedersen.
p. cm.— (Studies in advanced mathematics)
Includes bibliographical references and index.
ISBN 0-8493-7169-4 (alk. paper)
1. Functional analysis. I. Title II. Series.
Q A320.P394 1999
5 1 5 '.7 -^ c 2 1 99-37641
CIP

This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity o f all materials or for the consequences o f their use.

Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, microfilming, and recording, or by any information storage or
retrieval system, without prior permission in writing from the publisher.

The consent o f CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. Specific permission must be obtained in writing from CRC Press LLC
for such copying.

Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.

Tradem ark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation, without intent to infringe.

Visit the CRC Press Web site at www.crcpress.com

© 2000 by CRC Press LLC

No claim to original U.S. Government works

International Standard Book Number 0-8493-7169-4
Library o f Congress Card Number 99-37641
Printed in the United States of America 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
A ckn ow ledgm en ts

I would like to thank my colleagues, Prof.O.j0rsboe and Prof.L.MejIbro,

for correcting a number of errors and for collecting so many good exercises.
Also, a warm thanks to my former students. Dr.Roger Krishnaswamy, and
Dr.Michael Danielsen for the material on the Hilbert Uniqueness Method,
and Dr.Anda Binzer for assistance.

Michael Pedersen
P reface

Functional analysis is a ’’supermodel” for mathematical analysis that has

been developed in particular after the Second World War. The general idea
is to construct an abstract framework suited to deal with various problems
from mathematical analysis that perhaps at first glance seem to have noth­
ing in common, but when all ’’inessentials” are stripped away, appear to be
similar. The language from linear algebra is used to describe the setting of
the problems, and this notational simplification is both the power and dan­
ger of functional analysis: many formulations and formulas look extremely
simple but typically reflect highly delicate subjects.
The present book. Functional Analysis in Applied Mathematics and Engi­
neeringj is primarily written to cover the course in functional analysis given
at The Technical University of Denmark. This is a one-semester (14 weeks)
course, and in order to pass, the students will have to do 12 obligatory sets
of exercises and have 10 of them judged ’’passed”. Therefore, there is a
large number of exercises, many of them are folklore in the sense that they
apparently can be found in all basic functional analysis courses. Solutions
to all the exercises can be obtained by contacting the author by e-mail. The
course is recommended for students on third year or later having a solid
background in basic mathematical analysis or corresponding mathematical
“maturity”.
The aim of the book is to get as fast as possible to the operators on
Hilbert spaces with the spectral theorem as the final highlight, without
compromising on the mathematical rigor. The course will provide a solid
background for the understanding of the problems encountered in applied
mathematics and engineering, focusing in particular on the abstract for­
mulations of partial differential equations in a Hilbert space setting. A
lot of sophisticated theory that should be incorporated in a pure math,
functional analysis course is left out, such as measure theory and more ad­
vanced topological considerations. Instead, we typically argue by a density
or completion statement. We consider only separable Hilbert spaces since
nonseparable Hilbert spaces occur very rarely in applications.
 Chapters 1-6 are mandatory; they provide the theory that is necessary
in order to read and understand texts and papers in modern applied math­
ematics. The basic idea of these chapters is to ’’teach” the subject, and we
emphasize that in order to fully understand the theory, it is vital to cal­
culate a large number of the exercises. The exercises are ordered roughly
following the chapters, some of them are easy applications of the theory,
others are more interesting and it is highly recommended to do at least
50-60 of them. Therefore, it is also recommended that some of the time of
the regular lectures is used to discuss the aspects of the theory developed
in the exercises. The last chapters are introductions to different subjects
from functional analysis and can be read independently. The main purpose
of these chapters is to serve as an appetizer to a more specialized study
of functional analysis and control theory for partial differential equations.
There are no exercises to these chapters, and since the purpose is now
completely different from the first part of the book and the theory rather
advanced, a number of proofs are omitted and references are much more
common. In Chapters 12 and 13, however, detailed proofs of recent results
that so far only are published in scientific journals are incorporated. The
idea is to take one or two subjects from these chapters as examples of ad­
vanced applications as the conclusion of the regular 14-week course, but it
is also possible to use the material in a more advanced applied functional
analysis course.
There is a vast literature on functional analysis, the main part covering
what could be called general functional analysis in the sense that the aim
is to state and prove the theory in maximal generality, and the reader is
typically required to have a good knowledge of measure theory and gen­
eral topology. This kind of littérature is well suited for pure mathematics
students, where the requirements are essential also for many other subjects
during the period of study. For an engineering student who must be ed­
ucated in a number of other technical disciplines, it will typically not be
possible to meet the requirements of general functional analysis at the time
in the study where the mathematical problems encountered in the techni­
cal diciplines become so complicated that they actually require functional
analytic considerations. With the present book, we try to meet the back­
ground of typical engineering or applied mathematics students and give a
short but firm introduction to the subject. As mentioned above, we do not
compromise on the mathematical rigor and the ’’construction” of the book
is classical, with theorems and detailed proofs. The main part of the book
is focused on operators on Hilbert spaces, and closed unbounded operators
are introduced to provide the proper setting for the differential equations
from applied mathematics and engineering, some of which are presented in
the last part of the book.
Contents

1 Topological and Metric Spaces 1

1.1 Some Topology 1
1.2 Metric Spaces 2

2 Banach Spaces 11
201 Normed Vector Spaces 11
202 £P-spaces 0 0 0 0 0 0 0 12
203 Infinite Dimensional Spaces 19

3 Bounded Operators 23
301 Basic Properties 0 0 0 0 0 0 23
302 Bounded Linear Operators 0 26

4 Hilbert Spaces. 33
401 Inner Product Spaces 0 0 0 0 0 0 33
402 Hilbert Spaces 0 0 0 0 0 0 0 0 0 36
403 Construction of Hilbert Spaces 42
4.4 Orthogonal Projection and Complement 49
4o5 Weak Convergence 0 0 0 0 0 0 0 0 0 0 0 0 52

5 Operators on Hilbert Spaces 55

501 The Adjoint of a Bounded Operator 55
502 Compactness and Compact Operators 61
503 Closed Operators 0 0 0 0 0 0 0 0 0 0 0 0 65
5.4 The Adjoint of an Unbounded Operator 73

6 Spectral Theory 77
601 The Spectrum and the Resolvent 78
602 Operator-Valued Functions o 0 0 86
7 Integral Operators 91
7.1 Introduction . . . . . . . . . . . . . . 91
7.2 The Class of Hilbert-Schmidt Operators 92
7.3 Integral Equations . . . . . . . . . . . . 98

8 Semigroups of Evolution 101

8.1 Strongly Continuous Semigroups 101
8.2 The Resolvent . . . . . . . . . . . 108

9 Sobolev Spaces 111

9.1 Basic Definitions 111
9.2 Density Theorems . 115
9.3 Extension Theorems 116
9.4 Imbedding Theorems . 117
9.4.1 Example 0 •• 118
9.4.2 Applications 119
9.4.3 Example . 0 • 119
9.5 The Trace Theorem 120
9.6 Negative Sobolev Spaces and Duality . 124

10 Interpolation Spaces 127

10.1 Intermediate and Interpolation Spaces 127
10.1.1 The Operator L. . . 0 • • • 128
10.2 Intermediate Derivatives Theorem 129
10.2.1 An Example .. 0 • • • 131
10.2.2 A Continuity Property . 132
10.3 Interpolation Theorem . . . . 133
10.3.1 Reiteration Properties 135
10.3.2 Duality . . . . 0 • • • 136
10.4 Interpolation . . . . . 0 • • • 137
10.5 Interpolation with Hilbert Range 138
10.5.1 The Space LP(O, T; X) . 138
10.5.2 Distributions on ]0, T[ with Values in X. 139
10.5.3 Interpolation between LP(O, T; H) Spaces 139

11 Linear Elliptic Operators 141

11.1 Elliptic Operators . . . . . . . . . . . . . . 142
11.2 The Dirichlet Problem, Types of Solutions . 143
11.3 Boundary Operators . . . . . . . 145
11.3.1 Elliptic Equations . . . . . . . . . . 146
11.3.2 The Formal Adjoint of A . . . . . . 147
11.3.3 A Modified Version of Green's Formula 148
11.4 V-elliptic and V-coercive Forms . 149
11.5 The Boundary Conditions . . . . . . . . . . . . 153
 11.501 Realization of the Operator A 153
11.502 The V-space 0 0 0 0 0 0 0 0 0 0 155
11.6 Variational Formulation 0 0 0 0 0 0 0 0 156
110 7 Assumptions for the Variational Problem 158
11.8 A Classical Regularity Result 0 0 0 0 0 0 0 159
110801 A Regularity Theorem 0 0 0 0 0 0 0 159
11.802 An Abstract Regularity Theorem 0 168
11.9 Transposition 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 169
11.901 Transposition of Nonhomogeneous Boundary Value
Problems 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
o 171

12 Regularity of Hyperbolic Mixed Problems 173

1201 Solvability Results 0 0 0 0 0 0 0 0 0 0 174
1201.1 Classical Solvability Results 0 0 0 0 0 174
1201.2 Newer Solvability Theorems 0 0 0 0 0 178
1201.3 A Discussion of the Regularity Results 0 180
1202 Proofs of the Newer Regularity Theorems 0 182
120201 Proofs of Theorems from 1201.2 0 191
1203 Interpolation Results 0 0 0 0 0 0 0 0 0 0 198
12.4 Some Additional Regularity Theorems 200
1205 Systems with Variable Coefficients 201
120501 A Regularity Theorem 0 0 201

13 The Hilbert Uniqueness Method 209

1301 Dirichlet Boundary Control 0 210
1301.1 Application of HUM 0 0 0 210
1301.2 Preliminary Lemmas 0 0 0 213
130103 A Discussion of the Control Area 0 223
1301.4 Exact Controllability in Polygons and Polyhedra 224
1302 The Hilbert Uniqueness Method 0 0 0 0 0 0 0 227
130201 The Problem of Exact Controllability 228
130202 The Operator A 0 0 0 0 0 0 0 0 0 0 0 0 229
1303 The Variable Coefficients Case 0 0 0 0 0 0 0 0 233
130301 Notation and Properties of the Operator A 233
130302 Preliminary Lemmas 0 234
130303 Application of HUM 0 0 0 0 0 0 0 0 0 0 0 0 o 236

Exercises. 244

References 291
C h a p ter 1
Topological and M e tric Spaces

In this chapter we introduce the most fundamental spaces in functional

analysis, the topological spaces and the metric spaces. Since we will proceed
almost immediately to the metric spaces, the topologcal spaces are defined
for reference purposes only.

1.1 Some Topology

Perhaps one of the most basic features in analysis is the concept of open
and closed sets. More precisely, we define a topological space S in the
following way: 5 is a set in which we have a collection of subsets r satisfying

5 e r, 0 G T (1.1)
if ^ 1, ^ 2, ...Ak G r, then nf Aj G r ( 1.2)
if Aj G r for all j G /, then Uj^/ Aj G r. (1.3)

where / is a (not necessarily finite) index set. The sets in r are denoted
open sets, and a set is defined to be closed if its complement is open. Note
that sets can be both closed and open. The closure A of a, set A is the
smallest closed set containing A, and the interior of a set is the largest
open set contained in it. A neighborhood of a point x in 5 is an open set
containing x. We call (5 ,r) a Hausdorff Space and r a Hausdorff topology
if every two distinct points in S have disjoint neighborhoods; that is, the
topology separates points in S. We define continuity in the following way:

D E F IN IT IO N 1.1 Let (5, r) and (T^a) be topological spaces, and let

f :S T be a mapping. Then f is said to be continuous if f~^{A) is
open in S for all open sets A in T.
 1. TOPOLOGICAL AND METRIC SPACES

If f~^ is itself a mapping and also continuous, / is called a homeomor-

phism . (We use the notation f~^{A) for the set {x G S\f{x) G A}, even
when f is not assumed to be injective. )

E xam ple 1.1

Let r in the definition above be all subsets of 5. Then all maps from S to
T are continuous. D

1.2 M etric Spaces

The study of continuous mappings between topological spaces is denoted
general topology^ and is a rich and vital area of mathematics. We will,
however, in this context usually demand that the spaces we work in have
even more structure. We would like the space to have a natural way to
measure distances between elements, giving rise to the notion of metric
spaces.

D E F IN IT IO N 1.2 Let M he a set and d : M x M -> [0;oo[ a function.

Then (M^d) is denoted a metric space and d a metric if for all x, y, z in
M we have:

(i) 0<d{x, y )
(ii) d{x,y) = 0 if and only if x — y
(Hi) d{x,y) = d{y,x)
{iv) d{x, y) < d{x, z) -f d{z, y).

The condition (iv) above is denoted the triangle inequality.

E xam ple 1.2

The set of real numbers i? is a metric space when equipped with the metric
d{x,y) = \x —y\. This metric is called the natural metric on R. Another
metric, d', on ii is defined from this in the following way:

d(x,y)
d'{x,y) =
l + d{x,y)'

(See Exercise 6). D

1.2. METRIC SPACES

Notice that any metric space (M,d) is a topological space if equipped

with the topology where the topology is stemming from the open balls
B{xo,r) = {x e M \ d{xo,x) < r}. A set A C M is defined to be open if
any point in A is in an open ball contained in A. (See Exercise 2.)

E xam ple 1.3

If M = C{[a;b]), the continuous functions on the finite interval [a;6], we
have several interesting metrics:

doo{f,g)= sup \ f i t ) - g{ t ) \ (1.4)

te[a;b]
rb
d i { f , g ) = [ \f{t) - g{t)\ dt (1.5)
Ja

or in general,

d p (/,p )= ^ \ m - g { t ) f dt for p > 1. ( 1. 6)

It is not at all evident that dp is a metric - but we will see later that these
metrics in fact are induced by norms. D

D E F IN IT IO N 1.3 We say that a sequence {xn) C M converges to

X e M if d{xn,x) — 0 for n oo.
In this case we write Xn x or limnXn = x. The element x ^ M is
denoted the limit point for the sequence {xn)>

E xam ple 1.4

Consider in C([—1; 1]) the sequence (/n) of functions below :
We see that

1
d i(/n ,0) = / f ^ { t ) d t = ---- >0
J-i n
so /n 0 in (C([—1; l]),di). On the other hand, it is obvious that in
(C([-l;l]),doo) we have

c?cx)(/n,0) 1 for all n,

so (/n) does not converge to 0 in this space. D

The triangle inequality ensures that any metric is continuous.

 1. TOPOLOGICAL AND METRIC SPACES

F IG U R E 1.1
G rap h of (/„)

P R O P O S IT IO N 1.1
If Xn X and yn y in (M,d), then d{xn,yn) —t d{x,y).

PR O O F

d{x,y) <d{x,Xn) + d{xn,y)

<d{x, Xn) + d(Xn,yn) + d{yn, 2/)

so

d{x, y) - d{Xn, Vn) < d{x, Xn) + d{yn, y)-

Interchanging (x,y) and (Xn,yn) gives

d(xn,yn) - d{x, y) < d{x, Xn) + d{yn,y)

so

\d{x,y) - d{xn,yn)\ < d{x,Xn) + d{yn,y) -> 0.

P R O P O S IT IO N 1.2
A convergent sequence has only one limit point.

PRO OF Assume that (a;„) has two limit points, x\ and X2 -

Then d{x\,X2 ) < d{x\,Xn) + d{xn,X2 ) —>■ 0 so d{x\,X2 ) = 0; hence
X\ = X2- I
1.2. METRIC SPACES

Following the properties of the real numbers we define the concept of a

Cauchy sequence in a metric space

D E F IN IT IO N 1.4 A sequence (xn) in the metric space {M,d) is de­

noted a Cauchy sequence if d{xn^Xm) 0 for n,m —)• oo.

R E M A R K 1.1 In perhaps more familiar notation, this is written

Ve > 0 3N : n , m > N => d{xn,Xm) < s. (1.7)

I
In R with the metric d{x, y) — \x —y\ we know that any Cauchy sequence
is convergent. We shall see below that this is an exclusive property of a
metric space. On the other hand, all convergent sequences are Cauchy:
d{Xn,Xrn) < d{Xn,x) + d(x, Xm) 0 if limn = X.

E xam ple 1.5

Consider in the metric space (C([—1; 1]), di) the sequence ( / n ) shown below:

FIGURE 1.2
Graph of the nth element of (/n)

It is obvious that

di (/n?/m) — J \fn{t) fm{t)\ dt

< 0
min{n,m}
hence, (/n) is a Cauchy sequence. But it is also obvious that (/n) cannot
converge to any continuous function since the only possible candidate /
 1. TOPOLOGICAL AND METRIC SPACES

must satisfy

rifo r te]o,i]
~ [ Oi o T t e [-1,0[,

and it is not possible to extend / to a continuous function on [—1; 1]. D

Since the property that all Cauchy sequences are convergent is exclusive,
we have the following definition:

D E F IN IT IO N 1,5 A metric space {M,d) is complete if all Cauchy

sequences are convergent in

E xam ple 1.6

We know that (i?, d) with d{x,y) = \x —y\ is complete.
We saw that {C{[a; 6]),di) was not complete, but we know from elemen­
tary analysis that (C([a; 6]), doo) is complete. The spaces {C{[a; b]), dp) for
p > 1 are not complete, either (an argument similar to the case p = 1 will
show). D

There is a way to overcome the fact that some spaces fail to be complete
- we make them so! This is not the whole truth, but the following will
explain what one does. First we need the concept of a set being dense in
another.

D E F IN IT IO N 1.6 Let (M,d) be a metric space. A set A C M is said

to be dense in another set B C M if, for all b E B and all e > 0, there is
an a E A with d{a, b) < e.

R E M A R K 1.2 Another way of saying this is that given b E B, there

is a sequence (on) C A with lim^ On = b. I
E xam ple 1.7
In (R,d) from example 1.6, Q is dense in R, and ]0;1[ is dense in [0; 1].
D

E xam ple 1.8

Consider the set F of polynomials on a bounded interval [a; 6]. Obviously
P C C{[a;b]). Moreover, in the metric space (C([a; 6]), doo), P is in fact
dense in C([a;6]). This is the Weierstrass Approximation Theorem, and
let’s give a proof.
1.2. METRIC SPACES

First notice that we can assume that the continuous function / we want
to approximate is real since we can approximate real and imaginary parts
one at a time. Moreover, we assume that [a; b] = [0; 1]; this is just a scaling.
Define the sequence of polynomials (pn(/))-

n —k

k=0
These are the so-called “Bernstein polynomials”. Notice that / is uniformly
continuous (since the interval [0; 1] is closed), that is,
Ve > 0 36 > OVs, t e [0; 1] : |s - t| < 5 \f{s) - f{t)\ < e.
Let e > 0 be given and choose 6 > 0 according to this. For t E [0; 1] we
have:

\fit)-Pn{t)\ = m ( i +(1 - t)r - p j (pj Q i ' = ( i - i)”-"

k n —k
i E |(f{t) - / i*^(i - i )

n —k
+ E m - f i -n

k-0 ^ ^ 0 -~ y ^
“p— >1^
where M = sup^^[o;i] l/(^)l- Hence,

1/(0 - p „ (/HOI < i + “ E (< - ; ) ' ( ¡ ) < ‘ (> 0n —k

k= 0

6^ n
since the expression in the last summation is just the variance of the bino­
mial distribution.
Then
2M 1 1
doo{f,Pnif)) < £ +
6'^ n 4
M
< 2e for n > 7^-,
“ 6^e
 1. TOPOLOGICAL AND METRIC SPACES

and the result follows. D

Now, continuing on the original path, we need the concept of isometric

spaces.

D E F IN IT IO N 1.7 The metric spaces (M, d) and {M',d') are isometric

if there is a bijective map T : M M ' such that d'{Tx,Ty) = d{x^y) for
all x, y e M.

T H E O R E M 1.3 (C om pleteness)
Let (M,d) be a metric space. There is a complete metric space {M\d' ) and
a dense set M C M ' such that (M^d) and {M,d’) are isometric.
The space (M',d') is denoted the com pletion of {M,d).

We will omit the proof but mention that all completions of a metric space
are isometric; this is why we talk of the completion. Following this idea,
that is to identify isometric spaces, we usually consider the original space
as embedded in the completion. The completion of (Q,d) is (R,d) (where
d is the natural metric) and we usually think of Q itself as a dense subset
of R, even though by the construction of R it is an isometric image T{Q)
that is dense in R. The plot is that we are not able to distinguish between
Q and T{Q). Such identifications are made all the time in mathematics,
and this book follows this tradition.

P R O P O S IT IO N 1.4
(C([a; 6]), doo) the completion of the space (P, doo)*

PROOF
P is dense in C([a; 6]), which is complete with respect to the metric doo-
I
We will now prove the famous Banach fixed point theorem, and for this
purpose we must define a certain class of mappings between metric spaces.

D E F IN IT IO N 1.8 A mapping f from a metric space (M,d) into itself

is called a contraction if there is an a g]0; 1[ such that

d{f{x),f{y)) < ad{x,y)

for all x, y e M.
We say that x E M is a fixed point for f if f{x) x.
i.2. METRIC SPACES

T H E O R E M 1.5 (Banach fixed point theorem)

Let (M, d) be a complete metric space and let f be a contraction. Then f
has exactly one fixed point.

PRO OF Let a be the contraction constant.

Assume that x and y are fixed points. Then

d{x,y) = d{f {x) , f ( y) ) < ad{x,y)

SO d{x^ y) = 0 since a < 1. Hence / has at most one fixed point.

Now choose xq e M and define the sequence (xn) by Xn = / ” (xq). This
is a Cauchy sequence since

^ ^ i f m)j f (^m-1))
^ OidiXfYi, Xm—l)

< a^d{xi,xo)-
Then

d(X n-j-p j ^ n ) ^ diXfi^pj Xn-j-p—1 ) + d{Xfi-^i , Xn)

< (a^+^-' + *.. + a^)d(xi,xo)

^l-aP
= a d(xi,xo).
1 —a
Since 0 < a < 1, we have
Q
d{Xn+pyXn) < -------d(xi,Xo),
1 —a

hence (xn) is a Cauchy sequence and is convergent to some x G M. We

apply the triangle inequality to show that f{x) = x:

d (/(x), x) < d (/(x), Xn) -f- d{xn,x)

= d (/(x ),/(X n -l)) -f d(Xn,x)
< ad(x,Xn-l) -I- d{Xn,x) -> 0

so d (/(x),x) = 0. I

R E M A R K 1.3 For some important applications of the fixed point

theorem to integral equations, see Exercise 23. I
One can show that it is enough that f ^ is a contraction for some m for
/ to have a unique fixed point, see Exercise 21.
10 1. TOPOLOGICAL AND METRIC SPACES

It is also interesting in some applications to notice that the fixed point

theorem applies if / is a contraction on some closed, /-invariant subset
A C M { That A is /-invariant means that f{A) C A).

E xam ple 1.9

For a particularly simple application of the fixed point theorem, consider
the integral equation
nh
z(t) —p k{t,s)x{s)ds = u{t),
Ja
where u G C([a;6]), a < b, k ^ C([a; 6]^) are known functions, /x G C is a
parameter, and x is the unknown function to be determined. To apply the
fixed point theorem, define / : C{[a;b]) C([a;b]) by
nb
/(^ ) = k(',s)x{s)ds,
Ja
and notice that solving the equation is equivalent to finding the fixed point
for / in the complete metric space (C7([a; 6]), doo)-
We will investigate when / is a contraction in the given metric.
Let x, y e C{[a; 6]). For t G [a; b] we have

\f{x){t) - f{y){t)\ = ImII f

Ja
k{t,s){x{s) - y{s))ds\

< 1^1 / \k{t,s)\ds ' doo (ar,2/)

Ja
< \p\M{b - a) ■doo(x,y),
where M = |A:(i,s)|.
This shows that
dooif{x),f{y)) < \fj,\M{b- a)doo{x,y),
SO / is a contraction when

M{b-a)'
When this is the case, we can solve the integral equation by the iteration

Xn+l = f{x„) = U + fl k{-, $)Xn{s)ds,

Ja
Starting with any xq G C([a;6]), and converging to the unique, continuous
solution. D
C h a p ter 2
Banach Spaces

The fundamental objects to be studied in functional analysis are functions,

and since we can, in general, make linear combinations of functions, we
are led into the framework of vector spaces. Furthermore, we would like
to consider continuous maps between vector spaces, so these must have a
topology. The vector spaces considered in this book are complex unless
otherwise stated.

2.1 Norm ed Vector Spaces

D E F IN IT IO N 2.1 Let V he a vector space. A norm on V is a function

II • II -> [0; oo[ satisfying for all x ,y e V and a G C:

I|x + 2/ | | < M + |M1 (2 .1)

llaxll = |a|||a:|| (2.2)
||x|| = 0 X = 0. (2.3)

If the last condition above is omitted from the definition, || • || is called a

semi norm.
We notice that any normed vector space is a metric space when we use
the norm induced metric d{x,y) — ||x —y\\. Notice also that any norm is
continuous, since |||x|| —||y|l| < ||x —y||.

D E F IN IT IO N 2.2 A normed vector space that is complete in the met­

ric induced by the norm is called a Banach space.

11
12 2. BANACH SPACES

We have already noticed that C([o; b)] with the sup-norm

II/- 5 II00 =doo{f,g) (2.4)

is a Banach space.

2.2 X^-spaces
We will now construct some very important Banach spaces of functions,
namely the L^-spaces. These will be constructed from a suitable subspace
of the continuous functions by imposing certain norms, and then completing
the normed (hence metric) spaces.
First we need some definitions.

D E F IN IT IO N 2.3 A set S in a normed (or just topological) space V

is called compact if every sequence in S has a subsequence that converges
to an element of S.

Notice here that perhaps the simplest compact sets in R are the closed
and bounded intervals.

D E F IN IT IO N 2.4 The support of a function f : R C is the closure

of the set {x E R \ f{x) ^ 0}, usually denoted supp{f). The space Co{R)
is the space of continuous functions with bounded, hence compact, support.

REMARK 2.1 A continuous function defined on a closed, bounded

interval can be considered as a function with compact support by a trivial
extension. I
We now equip Cq{R) with the p-norms, that is, the norms that induce
the dp-metrics from the previous chapter.

P R O P O S IT IO N 2.1
The map || • ||p, , from Co{R) to R, p > 1, given by

\\f\\p = i i \ f { x r d x ) K
Jr
is a norm on Cq{R).
2.2. L^-SPACES 13

PROOF The only condition that is trivial is (2.2).

Let us first verify (2.3).
Assume that ||/||p = 0 and f ^ 0. Then there is an xq such that \f{xo)\ >
0. By the continuity of / , we can find a 5 > 0 such that, say,

\f{x) - f{xo)\ < ^\f{xo)\ for \x-xo\<S.

Then \f{x)\ > l\f{xo)\ for |x —xo| < ¿, so we have that

II/IIp = \ f { x) f dx) i > (26)^ • i|/(x o )| > 0,

a contradiction. I
The proof of (2.1), the triangle inequality, is rather long and consists of
several steps, each of independent interest. We will therefore state the steps
separately but keep the goal in mind. The case p = 1 is, of course, trivial.

L E M M A 2.2
Young’s inequality
Let p > l,q > 1 and - + - = 1, and let a > 0 and b > 0.
Then
ab < -oF + - 6«.
P Q

PROOF
We consider the graph oi y = x^~^:

F IG U R E 2.1
G rap h o i y = x^ L
14 2. BANACH SPACES

Notice that x = ^, hence the area of the rectangle is ab, the area of I
is and the area of II is The inequality follows. I

L E M M A 2.3
Holder’s inequality, continuous case
Let p > IjQ > 1 and ^ ^ = 1, and assume that f , g e Cq{R).
Then
W f g Wi < W f W p M U ^

PROOF The statement is obviously true for / = 0 or ^ = 0, so assume

that WfWp > 0 and ||^||g > 0. Then

\f{^)9i^)\ / 1 l / W r 1 |p(2:)|^
^ 11 « 11r» ~l Q ’
ii/iipiMi, - P w m q iipii?
hence
f l/(a;)5 (a;)| . / 1 . 1 ,
.IIp II, P Q

and the inequality follows. I

REMARK 2.2 The special case of Holder’s inequality where p — q = 2

is called Cauchy-Schwarz’ inequality. It will be stated precisely later. I
Now we are ready to prove (2.1), the triangle inequality

L E M M A 2.4
Minkowski’s inequality, continuous case
P^ 1 assume that Cq{R).
Then
11/ + slip < ll/llp + Hp IIp-

PROOF The case p = 1 is trivial. For p > 1, write \ ^ ~ \ ^.nd see

that

f \f(3:)+9 ix)\Pdx< f \fix)\\f(x)+g{x)\^ ^dx

Jr Jr
f \9 ix)\\f{x)+9 {x)r^dx
Jr
< Wf U i f + + \\9\\p\\if + II,,
2.2. L^-SPACES 15

according to Holder’s inequality. Since q{p — 1) = p, the last expression

equals
(Il/llp + ll5llp)(/ \f(x)+g{x)\^dxy--^.
Jr
Then, dividing both sides of the inequality with the last factor, one gets

( f l/W + g{x)\Pdx)p < ll/llp + Iliilip,

Jr
which is the inequality. (If the last factor is zero, the inequality is, of course,
trivial.) I
Now we have shown that Co{R) with the p-norms are normed spaces.

R E M A R K 2.3 Any complex sequence (xn) can be considered as a

function f : N C where we define /(n ) = Xn> The natural way to
translate the above into results about sequences is to let Cq{N) be the
space of sequences with only a finite number of elements different from
zero. The inequalities above are then valid, with the integrals substituted
with summations.
Notice also that a similar identification allows us, for a vector x G C7^,
to speak of its p-norm; in particular, the 2-norm of a vector is denoted the
Euclidean norm. I
As we saw in example (1.5), the spaces Cq{R) with the p-norms are not
complete. We will now complete them as metric spaces. This is not a trivial
construction, and we will comment on it after the following definition.

D E F IN IT IO N 2.5 The vector space L^{R); p > 1 is the completion of

Co{R) in the metric induced by the p-norm.

The spaces are now Banach spaces by construction. It is not evident

what these spaces really consist of. Take for example a continuous function
/ , satisfying \f{x)\P < oo. Then fn = f - X[-n;n] is in Co(R), for all n,
where X[-n;n] is the function with the graph shown in Figure 2.2. We see
that
p n— n poo
/ \f{x)-fn{x)\^dx < / |/( a ; ) P d x + / \f{x)\^dx^0 for n ^ 00,
Jr J —oo Jn

since \f{x)\^dx is convergent.

Hence / ^ -> / in p-norm and / is in L ^ { R ) . But if we consider the
discontinuous function /i constructed from / by changing the value in one
single point (see Figure 2.3),
16 2. BANACH SPACES

FIGURE 2.2
Figure of X[-n;n]

A A

FIGURE 2.3
Figures of / and /]

we see that
[ - fi{x)\Pdx = 0,
Jr
SO the convergent sequence (/n) converges also to /i, so this function must
also belong to L^{R) - but we have now the obvious problem that in this
space f —f I must be the zero function !! This is the price we have to pay
if we want our p-normed spaces to be complete.
Functions cannot in general be considered, pointwise (of course, for the
good old continuous ones there are no problems, according to Proposition
2.1). This can be put into a consistent framework using the Lebesgue the­
ory of integration, which is not within the scope of this book. Instead of
functions, one speaks of equivalence classes of functions, and two functions
are said to be equivalent if their difference is the zero function in L^{R). In
2.2. L^-SPACES 17

this case we will write

f = 9 ae, (2.5)

where a.e means almost everywhere. But the integral defining the norm
is now the so-called Lebesgue integral That allows us to integrate a much
larger class of functions than the Riemann integrable ones. For example,
1q , the characteristic function for Q that takes the value I li x £ Q and is
0 elsewhere, is Lebesgue integrable with integral 0, whereas it is not Rie­
mann integrable over any interval. But for the continuous functions there
are no problems; if f is continuous, then / belongs to L^{R) if and only if
fn We must confine ourselves to the fact that there are
many more p-integrable functions, but we can always rely on the dense­
ness of Cq{R) in L^{R), so we can approximate with nice, continuous, and
compactly supported functions. We will refer to [26] for a comprehensive
introduction to integration theory.
All this can of course be done when R is replaced by R^ and the spaces
L^{R^) are defined in the obvious way. Moreover, if R is replaced by any
interval /, the spaces L^{I) are defined similarly.
This gives us also the precise formulations of Holder’s, Minkowski’s, and
Cauchy-Schwarz’ inequalities:

T H E O R E M 2.5 (Holders inequality)

Letp > l,q > 1 and = 1, and assume that f G L^{R) and g G L^{R).
Then
ll/^lli < Il/llpll5ll,

T H E O R E M 2.6 (Cauchy-Schwarz’ inequality)

Assume that f,g € L‘^{R) .
Then
ii/ffiii < 11/ 112II5II2.

T H E O R E M 2.7 (Minkowski’s inequality)

Let p > 1 and assume that f,g ^ L^{R).
Then
ll/ + 3l|p < ll/llp+ ll^llp-
It will be vital for us to be able to approximate in L^, so we will find
some particularly simple spaces that are dense.

D E F IN IT IO N 2.6 The space of functions that are infinitely many

times differentiable (smooth) and compactly supported is called C^{R). The
18 2. BANACH SPACES

space Cq^(I), where I is an interval, is defined analogously. Moreover,

P{R) and P{I) denote the spaces of polynomials on R or I, respectively.

REM ARK 2.4 One could wonder if such -functions exist. Consider
the function ^ -1
^(x-a)(b-x) for xG ]a;6[;
0 otherwise.
This is a C^(ii)-function . I
T H E O R E M 2.8
(i) Cq^(R) is dense in Co{R), both in the sup-norm and the p-norm.
(a) C ^{R ) is dense in L^{R).
(Hi) C^ { I ) is dense in L^{I)-
(iv) If I is finite, P{I) is dense in
(v) C ^ {I) is dense in C(I) with respect to the p-norm.

PROOF (i): We define first a so-called mollifier ma as follows: For

each a > 0, rria is a smooth, nonnegative function with rrioXx) == 0 for
|x| > a and Jj^ma{x)dx = 1. (Just think of (fa,b above, with a = —a,
b = a, and adjusted with a constant in order to make the integral = 1.)
Now, for / G Cq{R) we define the function fa by

fa(x)= / f { x - y ) 7na{y)dy,
Jr
and notice that by a change of variable we have

fa{x)= / f {z) ma{ x - z)dz,

Jr
so fa is in C^ {R) since ma is.
Since f is compactly supported it is uniformly continuous, so given € > 0
we can find a > 0, such that

\ f { x - y ) - f{x)\ < e for \y\ < a.

Then

\fcXx) - f(x)\ = I [ ( f i x - y ) ~ f{x))ma{y)dy\ ( 2.6)

Jr
< [ \ f { x - y ) ~ fix)\ma{y)dy (2.7)
Jr
< e, (2.8)
2.3. INFINITE DIMENSIONAL SPACES 19

due to the properties of the mollifier rria.

Hence \\fa - /||oo < ^ and C^{R) is dense in Co{R) with the sup-norm.
But observe that since / and fa are compactly supported, we can find
a,b e R such that

\ \ u = [ \fa{x) - f { x ) f d x < f é ’dx = eP{6 - a),

Jr Ja

SO
Wfa - /lip < e(6 - o)5
showing that C ^{R ) is also dense in Co{R) with the p-norm.
(ii): It is obvious that if A is dense in B, and B dense in C, then A is
dense in C. From (i), Cq^(R) is dense in Cq{R), which (by definition) is
dense in L^{R), and the result follows.
The proof of (v) is omitted, but it is a variant of the proof of (i). Now
(in) follows directly from (v). Moreover, (iv) follows directly from the
Weierstrass’ approximation theorem. I
Following the point of view from Remark 2.2.3, it is natural to define the
analogue sequence spaces to the L^{R) spaces, denoted P.

D E F IN IT IO N 2.7 The space of sequences (real or complex) P , p > 1,

consists of sequences (xn) satisfying
CO

^ \XnY‘ < 00-

n=l

The norm in P is

ii(a:«)iip = ( E i ^ " i ^ ) '-

n=l

The proper translation of Theorem 2.8 is that the space of sequences

with only a finite number of elements different from zero is dense in P for
all p.

2.3 Infinite Dim ensional Spaces

The Minkowski inequality shows, among other things, that the L^{R)
spaces are in fact vector spaces, since they are stable under the vector
space operations. This justifies the terminology we have been using from
20 2. BANACH SPACES

the very beginning. But since they differ quite a lot from the well-known
finite dimensional ones, we will take a glance at how the basic definitions
look.
Let xi,X2, be elements in the vector space V. Recall that this set
of vectors is called linearly independent if the equation

aixi -f- a2X2 -h ... + akXk - 0

is only satisfied for a i — «2 = ... = == 0. If this is not the case, the

set is called linearly dependent. Notice also that any subset of a linearly
independent set of vectors is again linearly independent. Moreover, a vector
X G F is called a linear combination of the vectors xi, X2, ..., if there are
constants ai,Q 2, . . . , such that a\Xi -h a 2 X2 + ... + OLk^k — x.
Let Í7 be a subset of V. By span{U) we denote the vector space con­
sisting of all finite linear combinations of vectors in U. This is of course
a subspace oí V. If the vector space V contains a finite set of k linearly
independent vectors xi , X 2 , -^>,Xk, and moreover any set of A: -f 1 vectors is
linearly dependent, we will say that the dimension oí V is k, and write
dim{V) = k. It follows that any vector x e V is then a linear combination
X == aiXi -h 02 X2 + ... + Oik^k of sit most k linearly independent vectors. If
there is no such k £ N, we will say that V is infinite dimensional

P R O P O S IT IO N 2.9
Any vector space containing C ^{R ) is infinite dimensional

PROOF Consider the set of functions {en}nez C Cq^{R), where Cn is

defined in the following way:

_ j for X G]n;n -f 1[;

^ n (^ ) — ^ n ,n + l (^ )
0 otherwise.

This is clearly an infinite set with the property that any finite subset is
linearly independent, hence {en}nez is infinite dimensional. Then C^{R)
is also infinite dimensional, and so is any vector space containing it. I
Now the vector space operations allow us to define what is meant by a
series.

D E F IN IT IO N 2.8 Let xi,X2, ... belong to the normed vector space V.

We say that the series -f- 2:2 + ... converges, if there is an x G V, such
that the sequence {sn) 0/ partial sums Sn = ZlILi converges to x. When
this is the case, we write x = Xi.
2.3. INFINITE DIMENSIONAL SPACES 21

This opens up for the precise definition of a basis for a normed vector
space y . If the dimension of V is finite, and dim{V) = fc, then any vector
X e V is a. linear combination of at most k linearly independent vectors from
F, and we know from linear algebra that any set U of k linearly independent
vectors from F is a basis for F. In this case we have that span{U) = V.
Recall that the dimension of F equals the number of vectors in the basis.
If the dimension of F is infinite and we furthermore assume that F is a
Banach space, we will say that a set U of linearly independent vectors from
F is total in F if any x G F is the limit of a sequence (sn) of vectors
Sn G span{U). Then F = span(U)^ and we see that any vector x e V
can be expressed as a series x = limn Sn = linin Yl7=i ~
vectors Xi G i7, Oi G C. We will call the total set U a (Schauder) basis ,
if all X £ H have such a unique series expression. A normed vector space
is called separable if it contains a countable, dense subset, and we are only
concerned about separable spaces in this book.
C h a p ter 3
B ou n ded O p era to rs

A mapping from a normed space V into another normed space W is called

an operator. If the mapping is linear, that is, if

T{ax + py) = aT x -1- PTy^

for all a,/3 in C and x^y in we call T a linear operator. We will
sometimes consider operators defined only on a subset of F, called the
domain of T and usually denoted D{T). If so, in order for T to be linear,
we will also demand that D{T) is a subspace oi V. In linear algebra we
have studied the linear operators from into (7^, the matrices. The aim
is now to extend the setting to study continuity properties of operators
when the spaces are only assumed to be normed. Since normed spaces are
topological spaces with the topology defined by the norms, the definition
of continuity can now be written in a perhaps more intuitive manner.

3.1 Basic Properties

D E F IN IT IO N 3.1 An operator T from a normed space V into another

normed space W is continuous at a point x G D{T) if for any e > 0 there
is a 6 > 0, such that \\Tx —Ty\\ < e for all y G D{T) with \\x —y\\ < 6. T
is continuous if it is continuous at all points of D{T).
T is uniformly continuous on D{T) if for any e > 0 there is a S > 0 ,
depending only on e, such that \\Tx —T y \\< e whenever ||x —y\\ < S.
T is bounded if there is a positive constant M such that \\Tx\\ < M|lxl|
for all X G D{T).

This is the obvious extension of the concept of continuity from real anal­
ysis, saying that objects close together are mapped into objects close to­

23
24 3. BOUNDED OPERATORS

gether. Continuous real functions in the usual sense are continuous opera­
tors in the sense above. Notice how the norm symbol || • || is used, denoting
both the F-norm and the T^-norm. In applications it is important to keep
in mind in which space one is working, but it is a significant power of
abstract mathematics to be able to write simple formulas and definitions.

Now we will show that continuity for linear operators is equivalent to

boundedness.

P R O P O S IT IO N 3.1
Let T be a linear operator from a normed space V into a normed space W .
The following statements are equivalent:
(i): T is continuous in 0.
(a): T is continuous.
): T is hounded.

PROOF Assume that T is continuous in 0. Let e > 0 be given, then we

can find J > 0, such that ||z|l < 6 implies that \\Tz\\ < e. Hence ||x —2/|| < S
implies that \\Tx —Ty\\ = \\T{x —y)\\ < e, so T is (uniformly) continuous.
It is trivial that (ii) implies (i), so assume now that T is bounded.
Then \\Tx —Ty\\ = \\T{x —y)\\ < M\\x —y\\, so T is continuous.
On the other hand, assume that T is unbounded. Then we can find a
sequence (xn) in V, such that l|Txn|| > n||a:n|| for all n E N. Now define the
sequence {yn) by yn = n\fx ~\\ notice that ||Tyn|| > 1? in contradiction
to the fact that obviously 2/n 0, hence T is not continuous in 0. I
This shows that a linear operator is either everywhere continuous or
nowhere continuous.
For a linear operator T we define the kernel, denoted ker{T), of T as the
set {x G D{T) I T x = 0}. It is easy to see that this is a subspace of V, and
it follows that ker{T) is closed whenever T is bounded.
A matrix is a continuous operator (since it is obviously bounded in the
usual norms on C^), so linear algebra does not provide us with an example
of an unbounded operator. But here is one :

E xam ple 3.1

Let / G C'^(J), where / is a bounded interval. Consider the canonical ex­
ample of a linear operator, namely differentiation: D f = / '. D is obviously
linear from the space C^{I) into C^{I), both equipped with the sup-norms.
Let (/n) be the sequence of functions

fn{x) = sin(nx).
3.1. BASIC PROPERTIES 25

so that l l / n l l o o = 1 and \\Dfn\\oo = This is an example of a sequence in

the unit ball of that is mapped into an unbounded sequence, so D is
unbounded.
But let us now choose the £)-norm on , defined by

ll/llD-ll/llood-lli^/lloo.
Now ||jD /|| oo < II/IIdj and D is now a bounded operator between two
normed spaces. This is, of course, just an example of how to choose the
right topology in order to make things work. D

E xam ple 3.2

Another example of a continuous map is the norm itself. Let V be any
normed space, then the norm is a map from V into R that satisfies

hence :V R is continuous (but not linear, of course). D

Before we continue with the study of the linear operators in particular,

we will show how to extend a continuous operator.

D E F IN IT IO N 3.2 Let T he an operator with domain D{T), and Ti

an operator with domain D{T\). If D{T) C D{Ti) and T x = T\x for all
X G D{T), we call T\ an extension o fT , and T is a restriction ofT \.

There is only one way to extend uniformly continuous operators, as we

will now see.

T H E O R E M 3.2
Let T he a uniformly continuous operator from a normed space V into a
Banach space W , and assume that D{T) is dense in V. Then T has a
unique^ continuous extension T\ \V

PROOF Let x' e V and choose a sequence (xn) from D{T) that con­
verges to x'. Recall that (xn) is then a Cauchy sequence. We will show
that {Txn) is a Cauchy sequence and define its limit to be T ix '.
Because T is uniformly continuous, for any e > 0 there is a 5 > 0 such
that \\Txn -TxmW < ^ whenever \\xn —Xm|| < S. Because (xn) is a Cauchy
sequence, there is an no, such that |lxn —Xm\\ < 5 for n,m > no- Hence

\\T x n -T x m \\< ^ for n ,m > n o ,

26 3. BOUNDED OPERATORS

showing that (Txn) is a Cauchy sequence in the Banach space W and

therefore convergent, Txn -> u, say. We define Tix' = u. That is, for
each x' e V we choose a sequence (xn) in D{T) with Xn x' and define
T\x' = linin(Txn). Since there are many sequences converging to x', we
have to show that this defines T\ uniquely. We will first show that T\ is
uniformly continuous.
Take x ',2/' G V , and corresponding sequences (xn), [Vn) in D{T) so that
Xn —> x' and yn ^ y' • Then

lli’ix' - Tiy'W = II lim(Ta;„ - Tj/„)|| = lim ||Ta;„ - T?/„||.

71 n
Since T is uniformly continuous, given any e > 0 there is a i > 0 such that

||T x „ -r? /„ || < I when ||x„ - t/„|| < ¿.

There is an no such that ||a;„ —a:'|| < | and ||j/„ —2/'|| < | for n > no- Now,
assume that ||x' —t/'|| < then

Ikn - VnW < \\Xn ~ x'\\ + ||x' - 2/'|| + ||j/„ - y'\\ < S,

hence
||Tx„ - TynW < - for n > no-

This shows that ||Tix' —Tiy'\\ < e for ||x' —y'\\ < and Ti is uniformly
continuous.
To see that Ti is unique, assume that T2 is also a continuous extension
of T. For any x' G F, take a sequence (xn) in D{T) converging to x'. Then

Tix' = lim(TiXn) = lim(Txn) = lim(T2Xn) = T2X',

n n n

SO Ti = T2. I

3.2 Bounded Linear Operators

We will now return to the study of linear operators. Let V and W be two
normed spaces, and let 5 and T be linear, bounded, and hence continuous
operators from V into W. It is obvious that for a,/3 G C, the operator
a S + ¡3T is also bounded (since it is continuous), so the bounded operators
from V into W form a vector space. This space we will denote B{V,W ).
In the special case where V — W, we will write B{V) instead of B{V,V).
We will now equip B{V, W) with a norm, the so-called operator norm.
3.2. BOUNDED LINEAR OPERATORS 27

D E F IN IT IO N 3.3 Let V and W be normed vector spaces. For T G

B{V,W ), we define the norm o fT by
llT|i = s u p { ||rx ||M N |< i} .

Notice how the norm symbol || • || is used for the norm in three different
spaces. It is easy to see that B (V ,W ) with the operator norm in fact is a
normed space.
We observe that l|Tx|| < ||T||||x|| for all x e V, and that any M > 0
satisfying \\Tx\\ < M\\x\\ for all x G F must also be > HTH.
Now, since the bounded operators form a vector space, we can consider
sequences and series of bounded operators, and an important result is the
following theorem.

T H E O R E M 3.3
Let V be a normed vector space and W a Banach space. Then B{V, W) is
a Banach space.

PROOF Assume that (Tn) is a Cauchy sequence in B{V, W). For e > 0
there is an no such that \\Tn —Tm\\ < e for n, m > no. Then, for any x e V
we have
\\T„X - TmX\\ = 1|(T„ - Tm)x\\ < ||T„ - Tmllllxll,
showing that (Tnx) is a Cauchy sequence in W. So for any x E V we can
define an operator T by T x — limn TnX. This operator is obviously linear,
since for all a, /3 G C, x, G
T{ax + ^y) = lim(T„(ax + 0y))
n

= lim(aT„x + PTny)
n

= a lim TnX -f lim TnV

n n

= a T x -f (3Ty.
Now we must show that T is bounded, and that (Tn) converges to it.
Let n > no. From the inequality
lIlTxll - Ilr„x||| < \\Tx - Tnx\\ = II limT^x - T„x|| = lim \\TmX - T„x||,
m m

we see that
IITxll < ||T„x|| + lim ||T„x - T„x|| < (||T„|| + lim ||T^ - T„||)||x||.
m m

Since (Tn) is a Cauchy sequence, it is bounded, and since n > no, we have
that lim„i ||Tn -- Tm\\ < This shows that ||Tx|l < (AT + ^)lkll5 some
constant A" > 0, hence T is bounded.
28 3. BOUNDED OPERATORS

Since
||Ta;-r„a;l| = || limTmX-T„a:|| = lim ||rma;-T„a:|| < lim ||rTO-T„||||x|| < e||a;l|,
m m m

it is clear that \\T —Tn\\ < e for n > no, hence Tn T in B{V^ W). I
The special case when the Banach space W is C calls for some comments.
Here B{V^W) is then the Banach space of bounded, linear operators from
V into C. This is also called the space of bounded, linear functionals on V,
or the dual space of V, frequently denoted V*,

E xam ple 3.3

Assume that p,q > 1 and ^ ^ 1, and let / € L^{R). We can define a
linear, bounded functional Tf E L^{RY by

/ f{x)g{x)dx,
Jr
for g E L^{R). From Holders inequality we see that

\Tfg\ < f \f{x)g{x)\dx < ll/llpllsll,,

Jr
hence ||Ty|| < ||/||p.
This shows that any / E L^{R) in a natural way defines a bounded, linear
functional Tf on L^{R). (It can be shown that, in fact, L^{R) = L^(ii)*.)
0
If V ,W ,Z are normed spaces, and 5 E B (V ,W ),T E B{W ,Z), we can
define the linear operator T S by T S x = T{Sx). Then T S E B{V, Z), since
\\TSx\\ < ||T||||5x|| < ||r||||51|l|x||,
showing that \\TS\\ < ||T||||5||. If T E B{V), we can define the powers
T'^,n E N oiT ; these are also bounded since successive applications of the
inequality gives
||T”x|l < ||r" ||||x || < l|T in|x||,
which implies that ||T^|| < \\T\Y.
This fact gives us the very interesting next proposition.

P R O P O S IT IO N 3.4
Assume that V is a Banach space and let ip . C C be a complex function
given by a convergent power series

viz) = ^ a„z" for \z\ < p,

n=0
3.2. BOUNDED LINEAR OPERATORS 29

where p is the radius of convergence for the series and On £ C. I fT G B{V)

and 1|T|| < p, the series is convergent in B{V). We define in
this way the operator-valued function (p by
oo

n=0
Moreover, we have that ||<^(T)|| < |(^(||T||)|.

PROOF We use the convention = /, where / is the identity operator.

Define the sequence (5n) in B(V) by Sn = Ylk=o We only have to
show that (Sn) is a Cauchy sequence.
Let e > 0. Since the radius of convergence is p, the real series l^nlll^ir
is convergent for ||T|| < p, so there is a number no e N such that
oo
^ |o „ |||T |r < e for k > no-
n =k

For n,p E N , n > no we have

n+p
I15„+p-5„|| = l| a,T>^\\
k=n-\-l
n+p

k—Ti+1
oo

fc=n+l

< e,
showing that (Sn) is a Cauchy sequence. This also shows the last statement
in the proposition. I
This proposition has a number of nontrivial applications, many frequently
used in numerical analysis. The following corollary defines the so-called
Neumann series of an operator, and is very commonly used.

C O R O L L A R Y 3.5
Assume that T G B{V), where V is a Banach space. If ||T|| < 1, the
operator I —T has an inverse {I —T)~^ G B{V). Moreover, (/ —T)~^ is
given by the Neumann series;

(7 -T )-^ = E ^ " (3.1)

n=0
30 3. BOUNDED OPERATORS

PROOF
Prom proposition 3.4 it is obvious that the series is convergent in B(V),
so denote the limit S. Hence S = lim„5„ = hin„ But notice
that
(J - T)Sn = (/ - T ){I + T + + ... + T")
_ j _ jin+l
= (7 + T + + ... + T ")(/ - T)
= S n {I-T ),
where I — > I for n —> oo because

||7 - ( 7 - r " + i ) || < ||T"+i|l < ||T||”+ i 0 for n -4 oo.

Hence (7 —T )S = S{I —T) and the result follows. I

E xam ple 3.4

Another operator frequently met in applications is the exponential e^, de­
fined for all T € B {y), since the series for converges for all z e C. Notice
that for any s e R, the operator is also in B{V), so if x G F, we can
define a mapping u : R V hy

u{s) = e^^x.

In the special case where V — R^, the bounded, linear operators are just
the matrices, and the u above is the usual way to express the solution of
the ordinary differential matrix equation
du
^Tu
ds
u(0) = X,
and, without much effort, this can be generalized to infinite dimensional
Banach spaces also. D

E xam ple 3.5

Let us consider V = R? and the linear operator represented by the matrix

0 (9>
Te =
-e 0,

Simple calculations show that

nr\2n o\ rp2n-{-l _
0 V
0 1, -1 0 ;
3.2. BOUNDED LINEAR OPERATORS 31

hence
oo

n=0
oo -
1
^rJ” + E

= co s(9 )(^ “) + s i n w ( _ "

/ cos(0) sin(0) \
Y- sin{6) cos{6) J
This shows that is a rotation in with the angle of rotation -6.
This is of course just linear algebra, but here we have no problems with
convergence of the series Yl^=o
D

We will conclude this section with two famous theorems that will be
stated without proofs.

T H E O R E M 3.6 (Hahn-Banach Theorem)

Assume that s is a semi-norm on a vector space V , and assume that (po is
a linear functional defined on a subspace D{(po) C V, and

\To{y)\ < s{y), y € D{<fo)- (3.2)

Then there is a linear functional ip on V such that
l(^(a;)| < s ( x ) , xeV, (3.3)
and

v{y) = ^o{y) (3.4)

f or y € D{(po).

T H E O R E M 3.7 (Principle of Uniform Boundedness)

Suppose that {T^ | A G i?} ¿5 a family of bounded linear operators from a
Banach space V into a normed space W and
sup{||T ax|| \ X e R} < 00 (3.5)
32 3. BOUNDED OPERATORS

for all X e V . Then

sup{||T a|| I a G i?} < 00. (3.6)
C h a p ter 4
H ilb e rt Spaces.

One very basic fact about the vector space is that it is possible to define
the angle between two vectors, giving us the fundamental concept of two
vectors being orthogonal to each other. This gives the space a rich geometric
structure that is inherited by a special class of Banach spaces, namely the
Hilbert spaces to be defined in the following.

4.1 Inner Product Spaces

The first ingredient to be introduced is the inner product^ the generaliza­
tion of the scalar product from linear algebra.

D E F IN IT IO N 4.1 Let V be a vector space. An inner product is a

mapping (*, •) from V x V into C satisfying
(%): {x,y) = (y,x) for x , y £ V ,
(ii): (axi +!3x2,y) = a{xi,y) + j3{x2,y) for xi , X 2 ,y € V and a,/3 € C,
(Hi): {XjX) > 0 and (x^x) = 0 if and only if x = 0.

Notice that the conditions imply that (x, ayi + ^ 2/2) = '^{x, yi)-\~0{x, 2/2),
so the inner product is linear in the first argument and conjugate linear in
the second.

E xam ple 4.1

In Co{R) we can define an inner product by

if, 9 )= / f{x)g{x)m{x)dx, (4.1)

Jr

where m{x) is a positive, continuous function.

33
34 4. HILBERT SPACES.

In L‘^{R) we have the inner product

[ f{x)g{x)dx, (4.2)
Jr
which is well defined by Cauchy-Schwarz’ inequality. D

We notice from Example 4.1 that in L^{R) the norm is given by

ll/ll2 - ( / , / ) ^ (4.3)
and we will always assume that L^{R) is equipped with that particular
inner product, unless otherwise specified. This is an example of a normed
space, where the norm is given by an inner product.
In we can define an inner product by

(a:, 2/) = (4.4)

2=1
notice how the usual Euclidean norm in is given by this inner product.
It is no coincidence that the inner product induces a norm; it is always
true that a space equipped with an inner product is a normed space.

D E F IN IT IO N 4.2 Let V be a vector space and assume that (*, •) is an

inner product on V . The induced norm on V is defined by
Ikll = (x,a;)2. (4.5)

We have not shown that this defines a norm in general, but the only
thing that is not trivial is the triangle inequality. But this follows from the
following version of Cauchy-Schwarz’ inequality:

P R O P O S IT IO N 4.1
Let V be a vector space and assume that (•, •) is an inner product on V.
Then
1(2;,y)|^ < {x,x){y,y) for all x, y €V. (4.6)

PROOF If (2/, y) = 0, we have that y = 0, and the inequality is obvious.

If (2/, y) ^ 0, we have for any a £ C that
0 < {x - ay, x - ay) = (x, x) - a{y, x) - a(x, y) H- aa{y, y),
and if we take
(x,y)
a =
(y,y)
4.1. INNER PRODUCT SPACES 35

the result follows. I

Now the triangle inequality for the induced norm follows, since we have

\\x + y f = \\x\f \\y\f + (x,y) A {y,x)

< \ \ x f + \\y\f4-2\{x,y)\
<||x||2 + |MP + 2||x|||M|
= (INI + IMI)^.
Then we can write the Cauchy-Schwarz inequality as

I(x,2/)I<INIIMI, (4.7)
which is the version we met earlier.
Notice also how the calculations in the proof show that the = in the
inequality is valid only if x and y are linearly dependent; that is, x = ay
OT y = 0.
As a simple application of Cauchy-Schwarz’ inequality we will show that
the inner product is continuous.

P R O P O S IT IO N 4.2
Let (*, •) be an inner product on the vector space V, and let V be normed
by the induced norm. If Xn x and yn y, then

{Xn,yn) (x,y).

PROOF Since \\xn —x|| — 0 and ||i/n —y\\ 0, we have that

\{xn,yn) - (a:,2/)| = |(x„ - x,yn - y) + (x„ - x,y) + (x,y„ - y)\

< \(x„ - x , y n - J/)| + I(x„ - x ,2/)| + |(x,2/„ - y)|
< Ikn - x||l|y„ - Î/II + ||x„ - x||||y|| + ||x||||y„ - y\\ 0.

I
In the special case where the inner product (x,y) is a real number for
all X and y, we speak of a real inner product. Notice that in a real inner
product space, we can define the angle between two vectors by

cos{6) =
jx,y) (4.8)
iNiiiî/ir
since the right hand side lies between —1 and 1. The most useful notion for
our purposes is that of a right angle, implying that (x,y) = 0. This leads
36 4. HILBERT SPACES.

to the following definition, which applies to both real and complex inner
product spaces.

D E F IN IT IO N 4,3 Two vectors x^y in an inner product space are said

to be orthogonal if {x, y) = 0. We will write x A. y if x and y are orthogonal.

This gives us an abstract version of Pythagoras’ theorem from elementary

geometry.

P R O P O S IT IO N 4.3
I f x l y , then ||.t + j/||- = ||x|p + ||2/||^.

PR O O F

+ 7j\\^ = {x + y , x + y)
= (x,x) + {x,y) + (y,x) + {y,y)
= {x,x) + (y,y)
= II^IP + 112/11^

since (x, y) = 0. I

4.2 Hilbert Spaces

We see that inner product spaces have many properties in common with
the Euclidean vector spaces we have met in linear algebra, a fact that we will
elaborate on in the following. The Euclidean vector spaces are complete
spaces; if this is also the case for an inner product space it is a Banach
space, and it is called a Hilbert space.

D E F IN IT IO N 4,4 A vector space with an inner product that is a Ba­

nach space with respect to the induced norm is called a Hilbert space.

We see that the vector spaces are Hilbert spaces; these are finite
dimensional Hilbert spaces. For examples of infinite dimensional Hilbert
spaces, take L^{R), or Ir with the corresponding inner product

((a;„), (?/„)) = '^Xky^.. (4.9)

A'=l
4.2. HILBERT SPACES 37

As we shall see later, this is, in a sense, the prototype of a Hilbert space.
Now we will discuss the concept of a basis in the Hilbert space case. In
Chapter 2 we defined what we will call a basis in a normed space, and here
we emphasize that we only consider separable spaces. There are of course
examples of nonseparable Hilbert spaces, but they are rare in applications.

D E F IN IT IO N 4.5 A set of vectors {xk} in an inner product space is

called an orthogonal set if {xi. xj) — 0 for i ^ j, and xj ^ 0 for all j. If
also (xj^Xj) = 1, that is, if all vectors are unit vectors, the set is called
orthonormal.
If an orthonormal set is a basis, we will call the set an orthonormal basis.

We notice that any finite orthogonal set is necessarily linearly indepen­

dent. Since orthogonal sets (and othonormal sets in particular) are conve­
nient to use as a basis, it is nice to have an algorithm to orthonormalize a
given set of linearly independent vectors. The next theorem tells us how.

T H E O R E M 4.4 (G ram -Schm idt)

Let {¡jk) be a sequence of linearly independent vectors in a Hilbert space
H . There is an orthonormal sequence (xk) in H , such that for any n £ N
we have
span{xk]^^i = span{yk}k=i-

PROOF Let
Xi = yi (4.10)
112/1ir
y-2 - {y2,xi)xi
X-2 = (4.11)
II2/2 - (2/2, 3;i)a:i||’
and if we assume that xi,xo, ..-,Xj are defined, let
2/j+i - T,i=iiyj+uXk)xk
Xj-fi —
Il2/j+i - T,i=i{yj+uXk)xk\\
The sequence (xk) is orthonormal and satisfies the claim I
We will now proceed with the process of expanding a vector in a Hilbert
space as a series of basis vectors. This is in fact a generalization of the
Fourier expansion of a function, as we shall soon see.

P R O P O S IT IO N 4.5
Let {xk) be an orthonormal sequence in the Hilbert space H , and let {a^) be
a sequence of real or complex numbers. The series is convergent
Exploring the Variety of Random
Documents with Different Content
 § 2. It was prevented from overpowering him by his labor. This,
continual, and as tranquil in its course as a ploughman’s in the field, by
demanding an admirable humility and patience, averted the tragic
passion of youth. Full of stern sorrow and fixed purpose, the boy set
himself to his labor silently and meekly, like a workman’s child on its first
day at the cotton-mill. Without haste, but without relaxation,—accepting
all modes and means of progress, however painful or humiliating, he took
the burden on his shoulder and began his march. There was nothing so
little, but that he noticed it; nothing so great but he began preparations
to cope with it. For some time his work is, apparently, feelingless, so
patient and mechanical are the first essays. It gains gradually in power
and grasp; there is no perceptible aim at freedom, or at fineness, but the
force insensibly becomes swifter, and the touch finer. The color is always
dark or subdued.
 78. Quivi Trovammo.

§ 3. Of the first forty subjects which he exhibited at the Royal

Academy, thirty-one are architectural, and of these twenty-one are of
elaborate Gothic architecture (Peterborough cathedral, Lincoln cathedral,
Malmesbury abbey, Tintern abbey, &c.). I look upon the discipline given
to his hand by these formal drawings as of the highest importance. His
mind was also gradually led by them into a calmer pensiveness.2
Education amidst country possessing architectural remains of some noble
kind, I believe to be wholly essential to the progress of a landscape artist.
The first verses he ever attached to a picture were in 1798. They are
from Paradise Lost, and refer to a picture of Morning, on the Coniston
Fells:—

“Ye mists and exhalations, that now rise

From hill or streaming lake, dusky or gray,
Till the sun paints your fleecy skirts with gold,
In honor to the world’s great Author rise.”

By glancing over the verses, which in following years3 he quotes from

Milton, Thompson, and Mallet, it may be seen at once how his mind was
set, so far as natural scenes were concerned, on rendering atmospheric
effect;—and so far as emotion was to be expressed, how consistently it
was melancholy.

He paints, first of heroic or meditative subjects, the Fifth Plague of

Egypt; next, the Tenth Plague of Egypt. His first tribute to the memory of
Nelson is the “Battle of the Nile,” 1799. I presume an unimportant
picture, as his power was not then availably developed. His first classical
subject is Narcissus and Echo, in 1805:—

“So melts the youth and languishes away,

His beauty withers, and his limbs decay.”

The year following he summons his whole strength, and paints what we
might suppose would be a happier subject, the Garden of the Hesperides.
This being the most important picture of the first period, I will analyze it
completely.

§ 4. The fable of the Hesperides had, it seems to me, in the Greek

mind two distinct meanings; the first referring to natural phenomena, and
the second to moral. The natural meaning of it I believe to have been
this:—
 The Garden of the Hesperides was supposed to exist in the
westernmost part of the Cyrenaica; it was generally the expression for
the beauty and luxuriant vegetation of the coast of Africa in that district.
The centre of the Cyrenaica “is occupied by a moderately elevated table-
land, whose edge runs parallel to the coast, to which it sinks down in a
succession of terraces, clothed with verdure, intersected by mountain
streams running through ravines filled with the richest vegetation; well
watered by frequent rains, exposed to the cool sea breeze from the
north, and sheltered by the mass of the mountain from the sands and hot
winds of the Sahara.”4

The Greek colony of Cyrene itself was founded ten miles from the sea-
shore, “in a spot backed by the mountains on the south, and thus
sheltered from the fiery blasts of the desert; while at the height of about
1800 feet an inexhaustible spring bursts forth amidst luxuriant
vegetation, and pours its waters down to the Mediterranean through a
most beautiful ravine.”

The nymphs of the west, or Hesperides, are therefore, I believe, as

natural types, the representatives of the soft western winds and
sunshine, which were in this district most favorable to vegetation. In this
sense they are called daughters of Atlas and Hesperis, the western winds
being cooled by the snow of Atlas. The dragon, on the contrary, is the
representative of the Sahara wind, or Simoom, which blew over the
garden from among the hills on the south, and forbade all advance of
cultivation beyond their ridge. Whether this was the physical meaning of
the tradition in the Greek mind or not, there can be no doubt of its being
Turner’s first interpretation of it. A glance at the picture may determine
this: a clear fountain being made the principal object in the foreground,—
a bright and strong torrent in the distance,—while the dragon, wrapped
in flame and whirlwind, watches from the top of the cliff.

§ 5. But, both in the Greek mind and Turner’s, this natural meaning of
the legend was a completely subordinate one. The moral significance of it
lay far deeper. In the second, but principal sense, the Hesperides were
not daughters of Atlas, nor connected with the winds of the west, but
with its splendor. They are properly the nymphs of the sunset, and are
the daughters of night, having many brothers and sisters, of whom I shall
take Hesiod’s account.

§ 6. “And the Night begat Doom, and short-withering Fate, and Death.

“And begat Sleep, and the company of Dreams, and Censure, and
Sorrow.

“And the Hesperides, who keep the golden fruit beyond the mighty
Sea.

“And the Destinies, and the Spirits of merciless punishment.

“And Jealousy, and Deceit, and Wanton Love; and Old Age, that fades
away; and Strife, whose will endures.”

§ 7. We have not, I think, hitherto quite understood the Greek feeling

about those nymphs and their golden apples, coming as a light in the
midst of cloud; between Censure, and Sorrow,—and the Destinies. We
must look to the precise meaning of Hesiod’s words, in order to get the
force of the passage.

“The Night begat Doom;” that is to say, the doom of unforeseen

accident—doom essentially of darkness.

“And short-withering Fate.” Ill translated. I cannot do it better. It means

especially the sudden fate which brings untimely end to all purpose, and
cuts off youth and its promise; called, therefore (the epithet hardly ever
leaving it), “black Fate.”

“And Death.” This is the universal, inevitable death, opposed to the

interfering, untimely death. These three are named as the elder children.
Hesiod pauses, and repeats the word “begat” before going on to number
the others.
 “And begat Sleep, and the company of Dreams.”

“And Censure.” “Momus,” the Spirit of Blame—the spirit which desires

to blame rather than to praise; false, base, unhelpful, unholy judgment;—
ignorant and blind, child of the Night.

“And Sorrow.” Accurately, sorrow of mourning; the sorrow of the night,

when no man can work; of the night that falls when what was the light of
the eyes is taken from us; lamenting, sightless sorrow, without hope,—
child of Night.

“And the Hesperides.” We will come back to these.

“And the Destinies, and the Spirits of Merciless Punishment.” These are
the great Fates which have rule over conduct; the first fate spoken of
(short-withering) is that which has rule over occurrence. These great
Fates are Clotho, Lachesis, Atropos. Their three powers are—Clotho’s over
the clue, the thread, or connecting energy,—that is, the conduct of life;
Lachesis’ over the lot—that is to say, the chance which warps, entangles,
or bends the course of life. Atropos, inflexible, cuts the thread for ever.

“And Jealousy,” especially the jealousy of Fortune, in balancing all good

by evil. The Greeks had a peculiar dread of this form of fate.

“And Deceit, and sensual Love. And Old Age that fades, and Strife that
endures;” that is to say, old age, which, growing not in wisdom, is
marked only by its failing power—by the gradual gaining of darkness on
the faculties, and helplessness on the frame, such age is the forerunner
of true death—the child of Night. “And Strife,” the last and the mightiest,
the nearest to man of the Night-children—blind leader of the blind.

§ 8. Understanding thus whose sisters they are, let us consider of the

Hesperides themselves—spoken of commonly as the “Singing Nymphs.”
They are four.
 Their names are Æglé,—Brightness; Erytheia,—Blushing; Hestia,—the
(spirit of the) Hearth; Arethusa,—the Ministering.

O English reader! hast thou ever heard of these fair and true daughters
of Sunset, beyond the mighty sea?

And was it not well to trust to such keepers the guarding of the golden
fruit which the earth gave to Juno at her marriage? Not fruit only: fruit on
the tree, given by the earth, the great mother, to Juno (female power), at
her marriage with Jupiter, or ruling manly power (distinguished from the
tried and agonizing strength of Hercules). I call Juno, briefly, female
power. She is, especially, the goddess presiding over marriage, regarding
the woman as the mistress of a household. Vesta (the goddess of the
hearth5), with Ceres, and Venus, are variously dominant over marriage,
as the fulfilment of love; but Juno is pre-eminently the housewives’
goddess. She, therefore, represents, in her character, whatever good or
evil may result from female ambition, or desire of power: and, as to a
housewife, the earth presents its golden fruit to her, which she gives to
two kinds of guardians. The wealth of the earth, as the source of
household peace and plenty, is watched by the singing nymphs—the
Hesperides. But, as the source of household sorrow and desolation, it is
watched by the Dragon.

We must, therefore, see who the Dragon was, and what kind of
dragon.

§ 9. The reader will, perhaps, remember that we traced, in an earlier

chapter, the birth of the Gorgons, through Phorcys and Ceto, from
Nereus. The youngest child of Phorcys and Ceto is the Dragon of the
Hesperides; but this latest descent is not, as in Northern traditions, a sign
of fortunateness: on the contrary, the children of Nereus receive gradually
more and more terror and power, as they are later born, till this last of
the Nereids unites horror and power at their utmost. Observe the gradual
change. Nereus himself is said to have been perfectly true and gentle.
 This is Hesiod’s account of him:—

“And Pontus begat Nereus, simple and true, the oldest of children; but
they call him the aged man, in that he is errorless and kind; neither
forgets he what is right; but knows all just and gentle counsel.”

§ 10. Now the children of Nereus, like the Hesperides themselves, bear
a twofold typical character; one physical, the other moral. In his physical
symbolism, Nereus himself is the calm and gentle sea, from which rise, in
gradual increase of terror, the clouds and storms. In his moral character,
Nereus is the type of the deep, pure, rightly-tempered human mind, from
which, in gradual degeneracy, spring the troubling passions.

Keeping this double meaning in view, observe the whole line of descent
to the Hesperides’ Dragon. Nereus, by the earth, begets (1) Thaumas
(the wonderful), physically, the father of the Rainbow; morally, the type
of the enchantments and dangers of imagination. His grandchildren,
besides the Rainbow, are the Harpies. 2. Phorcys (Orcus?), physically, the
treachery or devouring spirit of the sea; morally, covetousness or
malignity of heart. 3. Ceto, physically, the deep places of the sea; morally,
secretness of heart, called “fair-cheeked,” because tranquil in outward
aspect. 4. Eurybia (wide strength), physically, the flowing, especially the
tidal power of the sea (she, by one of the sons of Heaven, becomes the
mother of three great Titans, one of whom, Astræus, and the Dawn, are
the parents of the four Winds); morally, the healthy passion of the heart.
Thus far the children of Nereus.

§ 11. Next, Phorcys and Ceto, in their physical characters (the grasping
or devouring of the sea, reaching out over the land and its depth), beget
the Clouds and Storms—namely, first, the Graiæ, or soft rain-clouds; then
the Gorgons, or storm-clouds; and youngest and last, the Hesperides’
Dragon—Volcanic or earth-storm, associated, in conception, with the
Simoom and fiery African winds.
 But, in its moral significance, the descent is this. Covetousness, or
malignity (Phorcys), and Secretness (Ceto), beget, first, the darkening
passions, whose hair is always gray; then the stormy and merciless
passions, brazen-winged (the Gorgons), of whom the dominant, Medusa,
is ice-cold, turning all who look on her to stone. And, lastly, the
consuming (poisonous and volcanic) passions—the “flame-backed
dragon,” uniting the powers of poison, and instant destruction. Now, the
reader may have heard, perhaps, in other books of Genesis than
Hesiod’s, of a dragon being busy about a tree which bore apples, and of
crushing the head of that dragon; but seeing how, in the Greek mind, this
serpent was descended from the sea, he may, perhaps, be surprised to
remember another verse, bearing also on the matter:—“Thou brakest the
heads of the dragons in the waters;” and yet more surprised, going on
with the Septuagint version, to find where he is being led: “Thou brakest
the head of the dragon, and gavest him to be meat to the Ethiopian
people. Thou didst tear asunder the strong fountains and the storm-
torrents; thou didst dry up the rivers of Etham, πηγὰς καὶ χειμάῤῥους,
the Pegasus fountains—Etham on the edge of the wilderness.”

§ 12. Returning then to Hesiod, we find he tells us of the Dragon

himself:—“He, in the secret places of the desert land, kept the all-golden
apples in his great knots” (coils of rope, or extremities of anything). With
which compare Euripides’ report of him:—“And Hercules came to the
Hesperian dome, to the singing maidens, plucking the apple fruit from
the golden petals; slaying the flame-backed dragon, who twined round
and round, kept guard in unapproachable spires” (spirals or whirls, as of
a whirlwind-vortex).

Farther, we hear from other scattered syllables of tradition, that this

dragon was sleepless, and that he was able to take various tones of
human voice.

And we find a later tradition than Hesiod’s calling him a child of Typhon
and Echidna. Now Typhon is volcanic storm, generally the evil spirit of
tumult.

Echidna (the adder) is a descendant of Medusa. She is a daughter of

Chrysaor (the lightning), by Calliröe (the fair flowing), a daughter of
Ocean;—that is to say, she joins the intense fatality of the lightning with
perfect gentleness. In form she is half-maiden, half-serpent; therefore
she is the spirit of all the fatalest evil, veiled in gentleness: or, in one
word, treachery;—having dominion over many gentle things;—and chiefly
over a kiss, given, indeed, in another garden than that of the Hesperides,
yet in relation to keeping of treasure also.

§ 13. Having got this farther clue, let us look who it is whom Dante
makes the typical Spirit of Treachery. The eighth or lowest pit of hell is
given to its keeping; at the edge of which pit, Virgil casts a rope down for
a signal; instantly there rises, as from the sea, “as one returns who hath
been down to loose some anchor,” “the fell monster with the deadly sting,
who passes mountains, breaks through fenced walls, and firm embattled
spears; and with his filth taints all the world.”

Think for an instant of another place:—“Sharp stones are under him, he

laugheth at the shaking of a spear.” We must yet keep to Dante, however.
Echidna, remember, is half-maiden, half-serpent;—hear what Dante’s
Fraud is like:—

“Forthwith that image vile of Fraud appear’d,

His head and upper part exposed on land,
But laid not on the shore his bestial train.
His face the semblance of a just man’s wore,
So kind and gracious was its outward cheer;
The rest was serpent all: two shaggy claws
Reach’d to the armpits; and the back and breast,
And either side, were painted o’er with nodes
And orbits. Colors variegated more
Nor Turks nor Tartars e’er on cloth of state
With interchangeable embroidery wove,
Nor spread Arachne o’er her curious loom.
 As oft-times a light skiff moor’d to the shore,
Stands part in water, part upon the land;
Or, as where dwells the greedy German boor,
The beaver settles, watching for his prey;
So on the rim, that fenced the sand with rock,
Sat perch’d the fiend of evil. In the void
Glancing, his tail upturn’d, its venomous fork
With sting like scorpion’s arm’d.”

§ 14. You observe throughout this description the leaning on the

character of the Sea Dragon; a little farther on, his way of flying is told
us:—

“As a small vessel backing out from land,

Her station quits; so thence the monster loos’d,
And, when he felt himself at large, turn’d round
There, where the breast had been, his fork’d tail.
Thus, like an eel, outstretch’d at length he steer’d,
Gathering the air up with retractile claws.”

And lastly, his name is told us: Geryon. Whereupon, looking back at
Hesiod, we find that Geryon is Echidna’s brother. Man-serpent, therefore,
in Dante, as Echidna is woman-serpent.

We find next that Geryon lived in the island of Erytheia, (blushing),

only another kind of blushing than that of the Hesperid Erytheia. But it is
on, also, a western island, and Geryon kept red oxen on it (said to be
near the red setting sun); and Hercules kills him, as he does the
Hesperian dragon: but in order to be able to reach him, a golden boat is
given to Hercules by the Sun, to cross the sea in.

§ 15. We will return to this part of the legend presently, having enough
of it now collected to get at the complete idea of the Hesperian dragon,
who is, in fine, the “Pluto il gran nemico” of Dante; the demon of all evil
passions connected with covetousness; that is to say, essentially of fraud,
rage, and gloom. Regarded as the demon of Fraud, he is said to be
descended from the viper Echidna, full of deadly cunning, in whirl on
whirl; as the demon of consuming Rage, from Phorcys; as the demon of
Gloom, from Ceto;—in his watching and melancholy, he is sleepless
(compare the Micyllus dialogue of Lucian); breathing whirlwind and fire,
he is the destroyer, descended from Typhon as well as Phorcys; having,
moreover, with all these, the irresistible strength of his ancestral sea.

§ 16. Now, look at him, as Turner has drawn him (p. 298). I cannot
reduce the creature to this scale without losing half his power; his length,
especially, seems to diminish more than it should in proportion to his
bulk. In the picture he is far in the distance, cresting the mountain; and
may be, perhaps, three-quarters of a mile long. The actual length on the
canvas is a foot and eight inches; so that it may be judged how much he
loses by the reduction, not to speak of my imperfect etching,6 and of the
loss which, however well he might have been engraved, he would still
have sustained, in the impossibility of expressing the lurid color of his
armor, alternate bronze and blue.

§ 17. Still, the main points of him are discernible enough; and among
all the wonderful things that Turner did in his day, I think this nearly the
most wonderful. How far he had really found out for himself the collateral
bearings of the Hesperid tradition I know not; but that he had got the
main clue of it, and knew who the Dragon was, there can be no doubt;
the strange thing is, that his conception of it throughout, down to the
minutest detail, fits every one of the circumstances of the Greek
traditions. There is, first, the Dragon’s descent from Medusa and Typhon,
indicated in the serpent-clouds floating from his head (compare my
sketch of the Medusa-cloud, Plate 71); then note the grovelling and
ponderous body, ending in a serpent, of which we do not see the end. He
drags the weight of it forward by his claws, not being able to lift himself
from the ground (“Mammon, the least erected spirit that fell”); then the
grip of the claws themselves as if they would clutch (rather than tear) the
rock itself into pieces; but chiefly, the designing of the body. Remember,
one of the essential characters of the creature, as descended from
Medusa, is its coldness and petrifying power; this, in the demon of
covetousness, must exist to the utmost; breathing fire, he is yet himself
of ice. Now, if I were merely to draw this dragon as white, instead of
dark, and take his claws away, his body would become a representation
of a greater glacier, so nearly perfect, that I know no published engraving
of glacier breaking over a rocky brow so like the truth as this dragon’s
shoulders would be, if they were thrown out in light; there being only this
difference, that they have the form, but not the fragility of the ice; they
are at once ice and iron. “His bones are like solid pieces of brass; his
bones are like bars of iron; by his neesings a light doth shine.”

§ 18. The strange unity of vertebrated action, and of a true bony

contour, infinitely varied in every vertebra, with this glacial outline;—
together with the adoption of the head of the Ganges crocodile, the fish-
eater, to show his sea descent (and this in the year 1806, when hardly a
single fossil saurian skeleton existed within Turner’s reach), renders the
whole conception one of the most curious exertions of the imaginative
intellect with which I am acquainted in the arts.

§ 19. Thus far, then, of the dragon; next, we have to examine the
conception of the Goddess of Discord. We must return for a moment to
the tradition about Geryon. I cannot yet decipher the meaning of his
oxen, said to be fed together with those of Hades; nor of the journey of
Hercules, in which, after slaying Geryon, he returns through Europe like a
border forager, driving these herds, and led into farther battle in
protection or recovery of them. But it seems to me the main drift of the
legend cannot be mistaken; viz., that Geryon is the evil spirit of wealth,
as arising from commerce; hence, placed as a guardian of isles in the
most distant sea, and reached in a golden boat; while the Hesperian
dragon is the evil spirit of wealth, as possessed in households; and
associated, therefore, with the true household guardians, or singing
nymphs. Hercules (manly labor), slaying both Geryon and Ladon,
presents oxen and apples to Juno, who is their proper mistress; but the
Goddess of Discord, contriving that one portion of this household wealth
shall be ill bestowed by Paris, he, according to Coleridge’s interpretation,
choosing pleasure instead of wisdom or power;—there issue from this evil
choice the catastrophe of the Trojan war, and the wanderings of Ulysses,
which are essentially, both in the Iliad and Odyssey, the troubling of
household peace; terminating with the restoration of this peace by
repentance and patience; Helen and Penelope seen at last sitting upon
their household thrones, in the Hesperian light of age.

§ 20. We have, therefore, to regard Discord, in the Hesperides garden,

eminently as the disturber of households, assuming a different aspect
from Homer’s wild and fierce discord of war. They are, nevertheless, one
and the same power; for she changes her aspect at will. I cannot get at
the root of her name, Eris. It seems to me as if it ought to have one in
common with Erinnys (Fury); but it means always contention, emulation,
or competition, either in mind or in words;—the final work of Eris is
essentially “division,” and she is herself always double-minded; shouts
two ways at once (in Iliad, xi. 6), and wears a mantle rent in half (Æneid,
viii. 702). Homer makes her loud-voiced, and insatiably covetous. This
last attribute is, with him, the source of her usual title. She is little when
she first is seen, then rises till her head touches heaven. By Virgil she is
called mad; and her hair is of serpents, bound with bloody garlands.

§ 21. This is the conception first adopted by Turner, but combined with
another which he found in Spenser; only note that there is some
confusion in the minds of English poets between Eris (Discord) and Até
(Error), who is a daughter of Discord, according to Hesiod. She is
properly—mischievous error, tender-footed; for she does not walk on the
earth, but on heads of men (Iliad, xix. 92); i.e. not on the solid ground,
but on human vain thoughts; therefore, her hair is glittering (Iliad, xix.
126). I think she is mainly the confusion of mind coming of pride, as Eris
comes of covetousness; therefore, Homer makes her a daughter of Jove.
Spenser, under the name of Até, describes Eris. I have referred to his
account of her in my notice of the Discord on the Ducal palace of Venice
(remember the inscription there, Discordia sum, discordans). But the
stanzas from which Turner derived his conception of her are these—

“Als, as she double spake, so heard she double,

With matchlesse eares deformed and distort,
Fild with false rumors and seditious trouble,
Bred in assemblies of the vulgar sort,
That still are led with every light report:
And as her eares, so eke her feet were odde,
And much unlike; th’ one long, the other short,
And both misplast; that, when th’ one forward yode,
The other backe retired and contrárie trode.

“Likewise unequall were her handës twaine;

That one did reach, the other pusht away;
That one did make the other mard againe,
And sought to bring all things unto decay;
Whereby great riches, gathered manie a day,
She in short space did often bring to nought,
An their possessours often did dismay:
For all her studie was, and all her thought
How she might overthrow the thing that Concord wrought.

“So much her malice did her might surpas,

That even th’ Almightie selfe she did maligne,
Because to man so mercifull He was,
And unto all His creatures so benigne,
Sith she herself was of his grace indigne:
For all this worlds faire workmanship she tride
Unto his last confusion to bring,
And that great golden chaine quite to divide,
With which it blessed Concord hath together tide.”
 All these circumstances of decrepitude and distortion Turner has
followed, through hand and limb, with patient care: he has added one
final touch of his own. The nymph who brings the apples to the goddess,
offers her one in each hand; and Eris, of the divided mind, cannot
choose.

§ 22. One farther circumstance must be noted, in order to complete

our understanding of the picture,—the gloom extending, not to the
dragon only, but also to the fountain and the tree of golden fruit. The
reason of this gloom may be found in two other passages of the authors
from which Turner had taken his conception of Eris—Virgil and Spenser.
For though the Hesperides in their own character, as the nymphs of
domestic joy, are entirely bright (and the garden always bright around
them), yet seen or remembered in sorrow, or in the presence of discord,
they deepen distress. Their entirely happy character is given by Euripides:
—“The fruit-planted shore of the Hesperides,—songstresses,—where the
ruler of the purple lake allows not any more to the sailor his way,
assigning the boundary of Heaven, which Atlas holds; where the
ambrosial fountains flow, and the fruitful and divine land increases the
happiness of the gods.”

But to the thoughts of Dido, in her despair, they recur under another
aspect; she remembers their priestess as a great enchantress; who feeds
the dragon and preserves the boughs of the tree; sprinkling moist honey
and drowsy poppy; who also has power over ghosts; “and the earth
shakes and the forests stoop from the hills at her bidding.”

§ 23. This passage Turner must have known well, from his continual
interest in Carthage: but his diminution of the splendor of the old Greek
garden was certainly caused chiefly by Spenser’s describing the
Hesperides fruit as growing first in the garden of Mammon:—

“There mournfull cypresse grew in greatest store;

And trees of bitter gall; and heben sad;
 Dead sleeping poppy; and black hellebore;
Cold coloquintida; and tetra mad
Mortal samnitis; and cicuta bad,
With which th’ uniust Atheniens made to dy
Wise Socrates, who, thereof quaffing glad,
Pourd out his life and last philosophy.

* * * *

“The gardin of Prosèrpina this hight:

And in the midst thereof a silver seat,
With a thick arber goodly over dight,
In which she often usd from open heat
Herselfe to shroud, and pleasures to entreat:
Next thereunto did grow a goodly tree,
With braunches broad dispredd and body great,
Clothed with leaves, that none the wood mote see,
And loaden all with fruit as thick as it might bee.

“Their fruit were golden apples glistring bright,

That goodly was their glory to behold;
On earth like never grew, ne living wight
Like ever saw, but they from hence were sold;
For those, which Hercules with conquest bold
Got from great Atlas daughters, hence began.

* * * *

“Here eke that famous golden apple grew,

The which emongst the gods false Até threw.”

There are two collateral evidences in the picture of Turner’s mind

having been partly influenced by this passage. The excessive darkness of
the stream,—though one of the Cyrene fountains—to remind us of
Cocytus; and the breaking of the bough of the tree by the weight of its
apples—not healthily, but as a diseased tree would break.
 § 24. Such then is our English painter’s first great religious picture; and
exponent of our English faith. A sad-colored work, not executed in
Angelico’s white and gold; nor in Perugino’s crimson and azure; but in a
sulphurous hue, as relating to a paradise of smoke. That power, it
appears, on the hill-top, is our British Madonna; whom, reverently, the
English devotional painter must paint, thus enthroned, with nimbus about
the gracious head. Our Madonna,—or our Jupiter on Olympus,—or,
perhaps more accurately still, our unknown god, sea-born, with the cliffs,
not of Cyrene, but of England, for his altar; and no chance of any Mars’
Hill proclamation concerning him, “whom therefore ye ignorantly
worship.”

§ 25. This is no irony. The fact is verily so. The greatest man of our
England, in the first half of the nineteenth century, in the strength and
hope of his youth, perceives this to be the thing he has to tell us of
utmost moment, connected with the spiritual world. In each city and
country of past time, the master-minds had to declare the chief worship
which lay at the nation’s heart; to define it; adorn it; show the range and
authority of it. Thus in Athens, we have the triumph of Pallas; and in
Venice the assumption of the Virgin; here, in England, is our great
spiritual fact for ever interpreted to us—the Assumption of the Dragon.
No St. George any more to be heard of; no more dragon-slaying possible:
this child, born on St. George’s Day, can only make manifest the Dragon,
not slay him, sea-serpent as he is; whom the English Andromeda, not
fearing, takes for her lord. The fairy English Queen once thought to
command the waves, but it is the sea-dragon now who commands her
valleys; of old the Angel of the Sea ministered to them, but now the
Serpent of the Sea; where once flowed their clear springs now spreads
the black Cocytus pool; and the fair blooming of the Hesperid meadows
fades into ashes beneath the Nereid’s Guard.

Yes, Albert of Nuremberg; the time has at last come. Another nation
has arisen in the strength of its Black anger; and another hand has
portrayed the spirit of its toil. Crowned with fire, and with the wings of
the bat.

1 Notes on the Turner Collection at Marlborough House. 1857. Catalogue of

the Sketches of J. M. V. Turner exhibited at Marlborough House. 1858.

2 The regret I expressed in the third volume at Turner’s not having been
educated under the influence of Gothic art was, therefore, mistaken; I had not
then had access to his earlier studies. He was educated under the influence of
Gothic architecture; but, in more advanced life, his mind was warped and
weakened by classical architecture. Why he left the one for the other, or how
far good influences were mingled with evil in the result of the change, I have
not yet been able to determine.

3 They may be referred to with ease in Boone’s Catalogue of Turner’s

Pictures, 1857.

4 Smith’s Dictionary of Greek and Roman Geography. Art. “Cyrenaica.”

5 Her name is also that of the Hesperid nymph; but I give the Hesperid her
Greek form of name, to distinguish her from the goddess. The Hesperid
Arethusa has the same subordinate relation to Ceres; and Erytheia, to Venus.
Æglé signifies especially the spirit of brightness or cheerfulness including even
the subordinate idea of household neatness or cleanliness.

6 It is merely a sketch on the steel, like the illustrations before given of

composition; but it marks the points needing note. Perhaps some day I may be
able to engrave it of the full size.

CHAPTER XI.
 THE HESPERID ÆGLÉ.

§ 1. Five years after the Hesperides were painted, another great

mythological subject appeared by Turner’s hand. Another dragon—this
time not triumphant, but in death-pang; the Python, slain by Apollo.

Not in a garden, this slaying, but in a hollow, among wildest rocks,

beside a stagnant pool. Yet, instead of the sombre coloring of the
Hesperid hills, strange gleams of blue and gold flit around the mountain
peaks, and color the clouds above them.

The picture is at once the type, and the first expression of a great
change which was passing in Turner’s mind. A change, which was not
clearly manifested in all its results until much later in his life; but in the
coloring of this picture are the first signs of it; and in the subject of this
picture, its symbol.

§ 2. Had Turner died early, the reputation he would have left, though
great and enduring, would have been strangely different from that which
ultimately must now attach to his name. He would have been
remembered as one of the severest of painters; his iron touch and
positive form would have been continually opposed to the delicacy of
Claude and richness of Titian; he would have been spoken of, popularly,
as a man who had no eye for color. Perhaps here and there a watchful
critic might have shown this popular idea to be false; but no conception
could have been formed by any one of the man’s real disposition or
capacity.

It was only after the year 1820 that these were determinable, and his
peculiar work discerned.

§ 3. He had begun by faithful declaration of the sorrow there was in

the world. It is now permitted him to see also its beauty. He becomes,
separately and without rival, the painter of the loveliness and light of the
creation.
 79. The Hesperid Æglé.

Of its loveliness: that which may be beloved in it, the tenderest,

kindest, most feminine of its aspects. Of its light: light not merely
diffused, but interpreted; light seen pre-eminently in color.

Claude and Cuyp had painted the sunshine, Turner alone the sun color.

Observe this accurately. Those easily understood effects of afternoon

light, gracious and sweet so far as they reach, are produced by the softly
warm or yellow rays of the sun falling through mist. They are low in tone,
even in nature, and disguise the colors of objects. They are imitable even
by persons who have little or no gift of color, if the tones of the picture
are kept low and in true harmony, and the reflected lights warm. But they
never could be painted by great colorists. The fact of blue and crimson
being effaced by yellow and gray, puts such effect at once out of the
notice or thought of a colorist, unless he has some special interest in the
motive of it. You might as well ask a musician to compose with only three
notes, as Titian to paint without crimson and blue. Accordingly the
colorists in general, feeling that no other than this yellow sunshine was
imitable, refused it, and painted in twilight, when the color was full.
Therefore, from the imperfect colorists,—from Cuyp, Claude, Both,
Wilson, we get deceptive effect of sunshine; never from the Venetians,
from Rubens, Reynolds or Velasquez. From these we get only
conventional substitutions for it, Rubens being especially daring1 in
frankness of symbol.

§ 4. Turner, however, as a landscape painter, had to represent sunshine

of one kind or another. He went steadily through the subdued golden
chord, and painted Cuyp’s favorite effect, “sun rising through vapor,” for
many a weary year. But this was not enough for him. He must paint the
sun in his strength, the sun rising not through vapor. If you glance at that
Apollo slaying the Python, you will see there is rose color and blue on the
clouds, as well as gold; and if then you turn to the Apollo in the Ulysses
and Polyphemus—his horses are rising beyond the horizon,—you see he
is not “rising through vapor,” but above it; gaining somewhat of a victory
over vapor, it appears.

The old Dutch brewer, with his yellow mist, was a great man and a
good guide, but he was not Apollo. He and his dray-horses led the way
through the flats, cheerily, for a little time; we have other horses now
flaming out “beyond the mighty sea.”

A victory over vapor of many kinds; Python-slaying in general. Look

how the Python’s jaws smoke as he falls back between the rocks:—a
vaporous serpent! We will see who he was, presently.

The public remonstrated loudly in the cause of Python: “He had been
so yellow, quiet, and pleasant a creature; what meant these azure-
shafted arrows, this sudden glare into darkness, this Iris message;
Thaumantian;—miracle-working; scattering our slumber down in
Cocytus?” It meant much, but that was not what they should have first
asked about it. They should have asked simply, was it a true message?
Were these Thaumantian things so, in the real universe?

It might have been known easily they were. One fair dawn or sunset,
obediently beheld, would have set them right; and shown that Turner
was indeed the only true speaker concerning such things that ever yet
had appeared in the world. They would neither look nor hear;—only
shouted continuously, “Perish Apollo. Bring us back Python.”

§ 5. We must understand the real meaning of this cry, for herein rests
not merely the question of the great right or wrong in Turner’s life, but
the question of the right or wrong of all painting. Nay, on this issue hangs
the nobleness of painting as an art altogether, for it is distinctively the art
of coloring, not of shaping or relating. Sculptors and poets can do these,
the painter’s own work is color.
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com