METHODS IN NONLINEAR INTEGRAL EQUATIONS

Methods in Nonlinear
Integral Equations

by

Radu Precup
Department ofApplied Mathematics,
Babe§-Bolyai University,
Cluj, Romania

Springer-Science+Business Media, B.Y.

A C.I.P. Catalogue record for this book is available from the Library of Congress.

Printed on acidjree paper

ISBN 978-90-481-6114-0 ISBN 978-94-015-9986-3 (eBook)

DOT 10.1007/978-94-015-9986-3
All Rights Reserved
© 2002 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2002.
Softcover reprint of the hardcover 1st edition 2002

No part of this work may be reproduced, stored in a retrieval system, or transmitted

in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
To Rodica and Barbu with love
Contents

Preface xi

Notation xiii

o Overview 1

I FIXED POINT METHODS 11

1 Compactness in Metric Spaces 13

1.1 Hausdorff's Theorem. . . . . 13
1.2 The Ascoli-Arzela Theorem . 15
1.3 The Frechet-Kolmogorov Theorem 18

2 Completely Continuous Operators on Banach Spaces 25

2.1 Completely Continuous Operators 25
2.2 Brouwer's Fixed Point Theorem. 28
2.3 Schauder's Fixed Point Theorem . 32

3 Continuous Solutions of Integral Equations via Schauder's

Theorem 35
3.1 The Fredholm Integral Operator 35
3.2 The Volterra Integral Operator . 37
3.3 An Integral Operator with Delay 39

4 The Leray-Schauder Principle and Applications 43

4.1 The Leray-Schauder Principle . . . . . . . . . . . . 43
4.2 Existence Results for Fredholm Integral Equations 45
4.3 Existence Results for Volterra Integral Equations 50

Vll
viii Contents

4.4 The Cauchy Problem for an Integral Equation with Delay 53

4.5 Periodic Solutions of an Integral Equation with Delay . . 55

5 Existence Theory in £P Spaces 61

5.1 The Nemytskii Operator . . . 61
5.2 The Fredholm Linear Integral Operator 64
5.3 The Hammerstein Integral Operator .. 68
5.4 Hammerstein Integral Equations . . . . 69
5.5 Volterra-Hammerstein Integral Equations 72

References: Part I 77

II VARIATIONAL METHODS 83

6 Positive Self-Adjoint Operators in Hilbert Spaces 85

6.1 Adjoint Operators . . . . . . . . . . . . . . . . . . . 85
6.2 The Square Root of a Positive Self-Adjoint Operator 88
6.3 Splitting of Linear Operators in £P Spaces . . . . . . 90

7 The Fnkhet Derivative and Critical Points of Extremum 97

7.1 The Frechet Derivative. Examples . . . . . . . 97
7.2 Minima of Lower Semicontinuous Functionals . 103
7.3 Application to Hammerstein Integral Equations 107

8 The Mountain Pass Theorem and Critical Points of

Saddle Type 111
8.1 The Ambrosetti-Rabinowitz Theorem . . . . . . . . . . 111
8.2 Flows and Generalized Pseudo-Gradients 117
8.3 Schechter's Bounded Mountain Pass Theorem 121

9 Nontrivial Solutions of Abstract Hammerstein Equations 129

9.1 Nontrivial Solvability of Abstract Hammerstein Equations 129
9.2 Nontrivial Solutions of Hammerstein Integral Equations 132
9.3 A Localization Result for Nontrivial Solutions. . . . . .. 137

References: Part II 145

 ix

III ITERATIVE METHODS 149

10 The Discrete Continuation Principle 151
10.1 Perov's Theorem . . . . . . . . . . . . . . . . . . . . 151
10.2 The Continuation Principle for Contractive Maps on
Generalized Metric Spaces. . . . . . . . . . . . . . . 154
10.3 Hammerstein Integral Equations with Matrix Kernels 159

11 Monotone Iterative Methods 163

11.1 Ordered Banach Spaces . . . . . . . . . . . . . . . . 163
11.2 Fixed Point Theorems for Monotone Operators . . . 167
11.3 Monotone Iterative Technique for Fredholm Integral
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
11.4 Minimal and Maximal Solutions of a Delay Integral Equation 177
11.5 Methods of Upper and Lower Solutions for Equations of
Hammerstein Type . .. . . . . . . . . . . . . . . . . . . . . . 184

12 Quadratically Convergent Methods 195

12.1 Newton's Method . . . . . . . . . . 196
12.2 Generalized Quasilinearization for an Integral Equation with
Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

References: Part III 211

Index 217
Preface
Nonlinear integral equations represent the major source of nonlinear op-
erators and, by this, equally, the motor and the testing ground of a vast
domain of nonlinear analysis, namely the theory of nonlinear operators.
This book is intended to present some of the most useful tools of nonlin-
ear analysis in studying systems of integral equations (integral equations in
Rn).
The book is divided into three parts: fixed point methods, variational
methods, and iterative methods. Each part contains a number of chap-
ters of theory and applications. The methods we deal with are: Schauder's
fixed point theorem; the Leray-Schauder principle; direct variational meth-
ods; the mountain pass theorem; the discrete continuation principle; mono-
tone iterative techniques; methods of upper and lower solutions; Newton's
method; and the generalized quasilinearization method.
We note that the selected topics reflect the particular interests of the
author and that many contributions to the subject, as well as some other
important treatment methods of the nonlinear integral equations have not
been included for the sake of brevity.
The lists of references, one for each part, include only referenced titles;
they are just intended to guide the reader through the enormous existing
literature and not to provide a complete bibliography on the subject.
The presentation is essentially self-contained and leads the reader from
basic concepts and results to current ideas and methods of nonlinear anal-
ysis.
We hope the book will be of interest to graduate students, and theoret-
ical and applied mathematicians in nonlinear functional analysis, integral
equations, ordinary and partial differential equations and related fields.

Radu Precup
Notation

R+ set of all nonnegative real numbers

Rn set of all n-tuples x = (Xl, X2, ... , x n )
R+ set of all x E Rn with Xi ~ 0 for all i

Ixl Euclidian norm of xERn, Ixl = (

n
i~ x; ) 1/2

n
(x, y) Euclidian inner product in Rn, (x,y) = 2: XiYi,
i=l
also denoted by X· Y
x5,y natural order relation in R n : Xi 5, Yi for all i
x<y strict order relation in R n : Xi < Yi for all i
x5,a for x E R n and a E R : Xi 5, a for all i
J-L Lebesgue measure in Rn
X,Y,Z metric spaces, real Banach or Hilbert spaces
X* dual space of X
1·1 norm in X, also denoted by 1.lx
(., .) inner product in a Hilbert space; also for u E X
and u* E X*, (u*,u) is the value of u* at u, u* (u)
Br (u;X) open ball {v EX: lu - vi < r} ( Br (u) for short)
Br (u; X) closed ball {v E X; lu - vi 5, r} (B r (u) for short)
U, intU closure of U, interior of U
au boundary of U : au = U\ int U
convA convex hull of A
convA closed convex hull of A
c k (O;Rn) set of k- times continuously differentiable functions
u: 0 --t Rn (0 C RN open)
c k (0; Rn) space of all functions u E C k (0; Rn), u = (U1, ... , un)
such that Da Ui admits a continuous extension to 0 for
N
all i and a = (a1, ... , aN) with lal = 2: aj 5, k. Here
j=l
Da = alai / axr1 ... axc;.r
xiv

lul oo ma~ lu (x)1 (0 C RN bounded open, u E C (0; Rn))

xEn
Ck(O) stands for C k (0; R)
Ck [a,b] stands for C k ([a, b])
V(O;Rn) space of all measurable functions u : 0 --+ R n with
In lu (x)IP dx < 00 (0 C RN open, 1:::; p < 00)
1·lp norm in V (0; Rn), lul p = Un lu (x)IP dx)l/P
(., ·)2 inner product in L2 (0; Rn) ,
(u, v)2 = In (u (x) , v (x)) dx
Loo (0; Rn) space of all measurable functions u : 0 --+ Rn for which
there is a constant c with lu (x)1 :::; c for a.e. x E 0
1·100 norm in L oo (0; Rn),
lul oo = inf {c: lu (x)1 :::; c a.e. on O}
V(O) stands for V (0; R) (1:::; p :::; 00)
V (a, b; Rn) stands for V (0; Rn) with 0 = (a, b)
V (a, b) stands for V (a, b; R)
V(O;X) space of all stongly measurable functions u : 0 --+ X
with lulx E V (0)

Basic inequality:

lin u (x) dxl :::; in lu (x)1 dx, u E L1 (0; Rn) ,

where 0 C RN is open, 1.1 is the Euclidian norm in Rn, and

in u (x) dx = (in udx) dx, ... , in Un (x) dX) .

Here u = (U1, U2, ... , un).

 Chapter 0

Overview

As the title suggests, this book presents several methods of nonlinear anal-
ysis for the treatment of nonlinear integral equations. To this end the book
is developed on two levels which interfere. The first level is devoted to
the abstract results of nonlinear analysis: compactness criteria, fixed point
theorems, critical point results, and general principles of iterative approx-
imation. The second level is devoted to the applications for systems of
nonlinear integral equations (nonlinear integral equations in Rn). Here we
present existence, uniqueness, localization, and approximation results for
Fredholm, Volterra, and Hammerstein integral equations, and an integral
equation with delay.
Although most of the methods are classical, in several cases new points
of view, extensions and new applications are presented. Thus the selected
topics include: a unified existence theory for continuous and integrable so-
lutions of integral equations; Schechter's bounded mountain pass theorem;
a nontrivial solvability theory for Hammerstein integral equations; a local-
ization result for a superlinear elliptic problem; the discrete continuation
principle for generalized contractions; new fixed point results for increas-
ing and decreasing operators in ordered Banach spaces; monotone iterative
techniques; and the quasilinearization method for a delay integral equation
from biomathematics. Of course, the selective topics reflect the particular
interests of the author.
Let us now briefly describe the contents of this book without insisting
on the exact meaning of the notation and notions which are used.
Chapter 1 presents basic concepts and results of compactness in general
metric spaces, and in particular, in the spaces C (K; Rn) and LP (0; Rn) ,
where K is a compact metric space, 0 eRN is bounded open and

1
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
2 Chapter 0

1 :::; p < 00. We state and prove Hausdorff's theorem of characterization

of the relatively compact subsets of a complete metric space, in terms of
finite or relatively compact E-nets. Then we prove the Ascoli-Arzela com-
pactness criterion in C (K; Rn), and the Frechet-Kolmogorov theorem of
characterization of relatively compact subsets of V (0; Rn) (1:::; P < (0) .
We conclude this chapter by the following unified compactness criterion for
both C (0; Rn) and V (0; Rn).

Theorem 0.1 Let 0 C RN be bounded open and p E [1,00]. Let Y c

V (0; Rn) for 1 :::; p < 00, and Y c C (0; Rn) for p = 00. Assume that
there exists a function v E V (0) such that

lu (x)1 :::; v (x)

for almost every x E 0 and all u E Y. Then Y is relatively compact in
V (0; Rn) if and only if

sup ITh (u) - ulLP(o .Rn)

uEY h,
-+ 0 as h -+ o.
Here Oh = 0 n (0 - h) and Th (u)(x) = u (x + h) for all x E 0h·

Chapter 2 is devoted to the concept of a completely continuous operator.

We prove the theorem of representation of the completely continuous oper-
ators as uniform limits of sequences of continuous operators of finite rank.
Also we present the proof of one of the fundamental results of nonlinear
analysis, namely Schauder's fixed point theorem:

Theorem 0.2 Let X be a Banach space, D c X a nonempty convex

bounded closed set and let T : D -+ D be a completely continuous operator.
Then T has at least one fixed point.

In Chapter 3 we present three examples of completely continuous oper-

ators arising from the theory of integral equations: the Fredholm integral
operator T : C (0; Rn) -+ C (0; Rn) given by

T (u) (x) = in h (x, y, u (y)) dy (x E 0) ,

where 0 c RN is bounded open and h : 0 2 X R n -+ Rn is continuous; the

Volterra integral operator T : C ([a, b] ; Rn) -+ C ([a, b] ; Rn) ,

T(u)(t) = lth(t,s,u(s))ds (tE[a,b]),

Overview 3

with h : [a, bl 2
X Rn ~ Rn continuous, and the delay integral operator
T : D (T) c C ([0, tIl; Rn) ~ C ([0, tIl; Rn) defined by

T(u)(t) = l~Tf(s,u(s))ds (tE[O,tl]).

°
Here T> and cp E C ([-T, 0l ; Rn) are given, D (T) = {u E C ([0, tIl; Rn) :
u (0) = cp (O)}, u (t) = cp (t) for t E [-T, 0], U (t) = u (t) for t E [0, tl], and
fEe ([-T, tIl x Rn; Rn) satisfies

cp (0) = I: f (s, cp (s)) ds.

The complete continuity of these operators is proved via Ascoli-Arzela the-

orem (equivalently, via Theorem 0.1) and is used in order to deduce from
Theorem 0.2, the existence of continuous solutions to the corresponding in-
tegral equation u = T (u) .
Chapter 4 is devoted to the Leray-Schauder principle and its applications
to integral equations. In Section 4.1 we present an elementary proof (based
upon Urysohn's lemma and Schauder's fixed point theorem) of the following
version of the Leray-Schauder principle:

Theorem 0.3 Let X be a Banach space, K c X a closed convex subset,

U c K a bounded set, open in K and Uo E U a fixed element. Assume that
the operator T : U ~ K is completely continuous and satisfies the boundary
condition

u =I- (1 - A) Uo + AT (u) for all u E aU, A E (0,1).

Then T has at least one fixed point in U.

Using Theorem 0.3 in Section 4.2 we establish an existence principle for

the Fredholm integral equation which, in particular, yields existence results
for Hammerstein integral equations and two-point boundary value problems
for second order differential equations. Similar results for Volterra integral
equations and, in particular, for Volterra-Hammerstein equations and the
Cauchy problem for a first order differential system are established in Section
4.3. In Sections 4.4 and 4.5 we focus on the delay integral equation

u (t) = l~T f (s, u (s )) ds (0.1)

4 Chapter 0

which comes from biomathematics. We discuss the existence of solutions to

the initial value problem and of periodic solutions.
In Chapter 5 we first examine the Nemytskii superposition operator, the
Fredholm linear integral operator and the Hammerstein integral operator as
mappings acting in V spaces. Then using again the Leray-Schauder prin-
ciple we present an existence theory for the Hammerstein integral equation
in R n
u(x)= in"'(X,Y)f(y,U(Y))dY a.e. on 0, (0.2)

in a such way that the two cases of continuous solutions and of LP solutions
(1 ::; p < 00) are treated together by considering 1 ::; P ::; 00. The main
result is the following existence principle:

Theorem 0.4 Let 0 C RN be a bounded open set, '" : 0 2 ----) Rand

f : 0 x R n ----) Rn. Assume that there exists p E [1,00] and q E [1,00) such
that the following conditions are satisfied:

if 1 ::; p < 00 then", E}l' (0; U (0)) ;

{
ifp=oo then "'EC(O;U(O))

(here l/q + l/r = 1) and

if 1 ::; p < 00 then If (x, z)1 ::; g (x) + c Izi P/ q for a.e. x E 0,
{ all z E R and some g E Lq (O;R+), c E R+;
n

if P = 00 then for every R> 0 there is a gR E Lq (0) with

If (x, z)1 ::; gR (x) for a.e. x E 0 and all z E Rn with Izl::; R.

In addition assume the existence of a bounded open set U C V (0; Rn)

containing the null function such that

if u E U, >. E (0,1) and

{ u(x)=>.fn",(x,y)f(y,u(y))dy a.e. on 0, (0.3)
then u E U.

Then (0.2) has at least one solution u in V (0; Rn) with u E U. Moreover,
for p = 00, u E C (O;Rn).

In Section 5.5 we give sufficient conditions so that (0.3) holds in the case
of the Volterra-Hammerstein equation in Rn

u (t) =It '" (t, s) f (s, u (s)) ds a.e. on (a, b) .

Overview 5

In particular, we discuss the existence of global solutions for the Cauchy

problem
u' (t) = f (t, u (t) ) a.e. on (a, b),
{
u(a)=O,
where f: [a, b] x Rn --* Rn.
In the first two sections of Chapter 6 we review basic properties of self-
adjoint linear operators in Hilbert spaces. These are used in Section 6.3
where we present the Krasnoselskii's result concerning the splitting of a
bounded linear operator A : Lq (0; Rn) --* £p (0; Rn) (lIp + 11q = 1) in
the form
A=HH*,
where H : L2 (0; Rn) --* £p (0; Rn) and H* is the adjoint of H. Clearly
solving (0.2) in LP (0; Rn) is equivalent to solving the operator equation

u = ANf (u), u E LP (0; R n) , (0.4)

where

N f : LP (0; Rn) --* U (0; R n ), Nf (u) (x) = f (x, u (x))

and

A: U (0; Rn) --* LP (0; R n ), A (u) (x) = 10 /'l, (x, y) u (y) dy.

We finish Section 6.3 by showing that if A splits as above, then we may

replace the operator equation (0.4) by the equation

v = H* NfH (v), v E L2 (0; Rn) (0.5)

in the Hilbert space L2 (0; Rn) . More exactly we have:

Proposition 0.1 If u E £p (0; Rn) solves (0.4), then v = H* N f (u) is

a solution of (0.5). Conversely, if v E L2 (0; Rn) solves (0.5), then u =
H (v) is a solution of (0.4). Moreover, there is a one-to-one correspondence
between the solutions of (0.4) and the solutions of (0.5) .

In Chapter 7 we build the Golomb functional E : L2 (0; Rn) --* R,

E (v)
1 2
= "2lvl2 r
- in F (x, H (v) (x)) dx,
6 Chapter 0

where F is the potential of f. The Fnkhet derivative E' of E is the

operator
I - H* NjH : L2 (0; Rn) -+ L2 (0; Rn) ,
and so the equation (0.5) can be written in the variational form
E'(v) =0, VEL2(0;Rn),
and its solutions appear as critical points of E. In Section 7.2 we prove
the infinite-dimensional version of the classical Fermat's theorem about the
connection between extremum points and critical points, and we give suf-
ficient conditions for that a functional admits minimizers. In Section 7.3
the abstract results are applied to establish the existence of V solutions to
(0.4) .
Chapter 8 is devoted to the mountain pass principle. We first present the
Ambrosetti-Rabinowitz mountain pass theorem and its proof based on Eke-
land's variational principle. Then we state and prove Schechter's bounded
mountain pass theorem which guarantees the existence of a critical point in
a given ball of the space:

Theorem 0.5 Let X be a Hilbert space, R E (0,00]' BR = {u EX: lui <

R} and E E C l (BR) . Assume that for some Vo > 0, (E' (u) ,u) 2': -Vo for
all U E 8BR, and that there are Uo, UI E BR and r with Iuol < r < lUll
such that
max{E (uo) ,E (UI)} < inf {E (u) : U E BR, lui = r}.
Let rR = bE C([0,1];BR): 1'(0) =Uo, 1'(1) =ut} and
cR = inf max E
-yEr R tE[O,I]
b (t)) .
Then either there is a sequence of elements Uk E B R with
E (Uk) -+ cR, E' (Uk) -+ 0,

or there is a sequence of elements Uk E 8B R such that

E (Uk ) -+ cR, E '(Uk ) - (E'(Uk)

R2 ,Uk) Uk -+ 0, (
E '
(Uk),Uk
) :::; o.
If in addition E satisfies the (PS)R condition and
E' (u) + J.LU 1= 0, U E 8BR, J.L > 0, (0.6)
then there exists an element U E BR \ {uo, ut} with
E (u) = CR, E' (u) = 0.
Overview 7

We point out that for R < 00, (0.6) is the Leray-Schauder boundary
condition for the operator I - E', i.e., it is equivalent to

u=/=>.(I-E') (u), uEoER , >'E (0,1).

In Chapter 9 we apply Schechter's theorem to establish the nontrivial
solvability of the abstract operator equation of Hammerstein type

u = AN (u), u E Y, (0.7)
where Y is a Banach space, N : Y --'> Y*, and A : y* --'> Y is a linear
operator. In Section 9.1 we establish the following existence and localization
principle:

Theorem 0.6 Assume that A splits into A = H H*, where H: X --'> Y is

a bounded linear operator, X is a Hilbert space, H* : y* --'> X is the
adjoint of H, and N = J' for some functional J E Cl (Y; R) satisfying
J (0) = o. In addition assume that N (0) = 0 and the functional (N (.) ,.)
sends bounded sets into upper bounded sets and that there are VI EX \ {O},
r E (0, IVll) and R ~ IVll such that the following conditions are satisfied:

max {O, ~ IVll~ - JH (VI)} < inf {~Ivl~ - JH (V) : Ivlx = r},
V =/= >'H* NH (V) for Ivlx = R, >. E (0,1),
E satisfies the (PS)R condition.
Then there exists a V E X \ {O} with Ivlx :S R such that u = H (v) is a
non-zero solution of (0.7) .

We note that in our setting the equation (0.7) in Y can be converted

into the equation V = H* NH (v) in X, which is equivalent to the critical
point problem E' (v) = 0 for the energy functional
1 2
E: X --'> R, E(v) = "2lvlx - JH(v).

In Section 9.2 Theorem 0.6 is specialized for Hammerstein integral equa-

tions. Finally, in Section 9.3 we establish a localization result in a shell of
L2 (n) for the superlinear Dirichlet problem

-~u = lul p - 2 u in n,
{ (O.S)
u=o on on.
More exactly, we prove:
8 Chapter 0

Theorem 0.7 Let 0 C RN be a bounded open set with C 2-boundary and

let p E (2, 2N/ (N - 2)) if N ~ 3, and p E (2,00) if N = 1 or N = 2.
Then the problem (0.8) has a solution u with

IHI- 1 [(p -1) Ap _l]1/(P-2) ::; iH-1 (u)i2 ::; IHI- p /(P-2).
Here H is the square root of the operator (_~)-l considered on L2 (0) ,
IHI = sup{IH (v)lp : v E L2 (0), Ivl2 = I} and

Ap-l
. { In lul
= mf
p
-
2 1'VuI 2 dx
2/ : u E C 1 (-)
0 \ {O}, u = 0 on aO } .
(In lu1 2 dX)
P2 / p

Chapter 10 presents the discrete continuation principle for contractions

on spaces endowed with vector-valued metrics and one application to Ham-
merstein integral equations in Rn with matrix kernels.
Chapter 11 is devoted to the monotone iterative methods. The basic
notion in this chapter is that of an ordered Banach space. We try to localize
solutions of an operator equation in an ordered Banach space X, say u =
T (u), in an order interval [uo, vol. In addition we look for solutions which
are limits of increasing or decreasing sequences of elements of X. The basic
property of the operator T is the monotonicity. This, combined with certain
properties of the ordered Banach space X, guarantees the convergence of
monotone sequences. Thus this chapter explores the contribution of the
monotonicity to compactness. In Section 11.1 we review some notions and
properties of the theory of ordered Banach spaces, paying special attention
to the conditions which guarantee the convergence of bounded monotone
sequences. Section 11.2 presents some fixed point theorems for monotone
operators in general ordered Banach spaces. The method is then used in
Sections 11.3, 11.4 to discuss the existence, uniqueness, localization, and
monotone iterative approximation of solutions for the Fredholm integral
equation and the delay integral equation already considered in Chapters 3
and 4. In Section 11.5 monotone iterative methods are used to localize and
approximate solutions of the abstract Hammerstein equation in Rn

u (x) = ANf (u) (x) a.e. on O. (0.9)

Here Nf is Nemytskii's superposition operator associated to a given func-
tion f : 0 x Rn --t Rn (0 C RN bounded open), and A is a bounded linear
operator from Lq (0; Rn) to LP (0; R). We seek solutions u in an order in-
terval [uo, vol of LP (0; Rn) , where Uo is a lower solution of (0.9) and Vo is
Overview 9

an upper solution of (0.9). Of course, we assume that N f maps [uo, vol into
Lq (0; Rn). The main assumption is the mono tonicity of f in its second
argument. The advantage of working in the space V (0; Rn) (1 :S p < (0)
ordered by the regular cone of all positive functions is that we do not need
the complete continuity of the operator T = ANf . Additional properties of
the operator A in connexion with its spectrum, allow us to build upper and
lower solutions for (0.9) and, consequently, to obtain extremal solutions of
(0.9).
Finally, Chapter 12 discusses the problem of the convergence order of the
iterative methods. After explaining that the convergence of the methods de-
scribed in Chapters 10 and 11 is at most linear, this chapter briefly presents
Newton's method and its version for differential equations, the quasilin-
earization method. The last one is addapted for the quadratic bilateral
monotone approximation of the solution to the initial value problem for the
delay integral equation (0.1).
 Part I

FIXED POINT METHODS

 Chapter 1

Compactness
in Metric Spaces

In this chapter we first define the notions of a compact metric space and
of a relatively compact subset of a metric space. Then we state and prove
Hausdorff's theorem of the characterization of the relatively compact subsets
of a complete metric space in terms of finite and relatively compact E-
nets. Furthermore, we prove the Ascoli-Arzela and Fn3chet-Kolmogorov
theorems of characterization of the relatively compact subsets of C (K; Rn)
and LP (0; Rn), respectively. Here K is a compact metric space, 0 C RN
is a bounded open set and 1 ::; p < 00.

1.1 Hausdorff's Theorem

Proposition 1.1 Let (X, d) be a metric space. The following statements
are equivalent:
(a) Every sequence of elements of X has a convergent subsequence in
X.
(b) The space X is complete and for each E > 0 it admits a finite
covering by open balls of radius E.

Proof. (a) =} (b) : The completeness of X follows from (a) since every
Cauchy sequence which has a convergent subsequence is convergent. Assume
that for some E > 0 the space X does not admit a finite covering by balls
of radius E. Then, for any fixed element Ul E X there exists U2 E X with
d(Ul,U2) 2:: E. Furthermore, there exists U3 E X such that d(Ul,U3) 2:: E
and d (U2, U3) 2:: E, and so on. Thus we find a sequence (Uk) of elements of

13
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
14 Chapter 1

X such that
d(Uk,Uj) ~ c
for k i= j. This shows that (Uk) has no convergent subsequences, which
contradicts (a) .
(b) =} (a) : Let (Uk) be an arbitrary sequence of elements of X and
let (ck) be any decreasing sequence of positive numbers converging to zero.
According to (b) there exists VI E X such that the open ball Bel (VI) (of
center VI and radius cI) contains infinitely many terms of the sequence
(Uk). Similarly there exists V2 E X such that Be2 (V2) contains infinitely
many terms from those contained by Bel (VI)' and so on. Thus we find
a subsequence (Ukj) of (Uk) such that Ukl E Bel (VI), Uk2 E Bel (VI) n
Be2 (V2) and generally

Ukj E nj

i=1
Bei (Vi)

for j = 1,2, .... Since Uk i , Uk j E Be, (Vi) for i < j, we have

d (Uk"Ukj ) ~ d(Uki,Vi) + d (Vi,Ukj) < 2ci.

It follows that (Ukj ) is a Cauchy sequence. Since X is complete this sub-
sequence is convergent. _

Definition 1.1 A metric space X is said to be compact if it satisfies con-

dition (a) (equivalently (b)) in Proposition 1.1.
A subset Y of a metric space X is said to be relatively compact if its
closure Y is compact (as a metric subspace of X).

Definition 1.2 Let (X, d) be a metric space, Y a subset of X and c > O.

A subset ReX is said to be an c-net for Y if for every U E Y there
exists a V E R such that d (u, v) < c.

Notice that a metric space is compact if and only if it is complete and

for every c> 0 it admits a finite c-net. Consequently every compact metric
space is complete and bounded (in the sense that its metric takes values in
a bounded real interval).

Theorem 1.1 (Hausdorff) Let (X, d) be a complete metric space and

Y C X be a subset. The following statements are equivalent:
(a) Y is relatively compact.
( b) For every c > 0 there exists in X a finite c -net for Y.
(c) For every c > 0 there exists in X a relatively compact c-net for Y.
Compactness in Metric Spaces 15

Proof. (a):::}(b): Assume that Y is relatively compact. Then Y is

compact, and so for every c > 0 there exists in Y a finite c-net 'R, for Y.
Clearly, 'R, is a finite c-net (in X) for Y.
(b):::} (a): Assume (b) holds. Then for a given c > 0 there exists in X
a finite c/3-net 'R, for Y. For every u E 'R, we choose an element Vu E Y
such that d (u, vu ) < c/3 (if such an element exists). It is easy to see that
the set
'R,' := {v u : u E 'R,}

is a finite 2c/3-net for Y and also an c-net for Y. Hence Y is compact.

(b):::}(c): This is trivial, since every finite set is compact.
(c):::}(b): Assume (c) holds and take any c> o. Then there exists in X
a relatively compact c/2-net 'R, for Y. Furthermore, for the compact set
'R, there exists a finite c/2-net 'R,'. It is easy to see that 'R,' is an c-net (in
X) for Y. Hence (b) is true. _
We conclude this section by the following compactness property of the
uniform limit of a family of relatively compact subsets of a complete metric
space.

Proposition 1.2 Let (X, d) be a complete metric space and let {YA : A> O}
be a family of relatively compact subsets of X. Assume that Y C X is the
uniform limit of Y A as A ~ 0, that is for each c > 0 there is Ae > 0
such that for every u E Y and A E (0, Ae) there exists U A E Y A with
d (u, u A ) < c. Then Y is relatively compact.

The proof can be immediately given in terms of finite c-nets. We leave

the details to the reader.
In the next two sections, as applications of Hausdorff's theorem, we
shall prove the Ascoli-Arzela and Fnkhet-Kolmogorov theorems of charac-
terization of the relatively compact subsets in C (K; Rn) and LP (0; Rn),
respectively.

1.2 The Ascoli-Arzela Theorem

Let (K, d) be a compact metric space and C (K; Rn) be the Banach space
of all continuous functions from K to Rn, under the sup-norm 1.100.

Theorem 1.2 (Ascoli-Arzela) A subset Y of C (K; Rn) is relatively

compact if and only if the following conditions are satisfied:
16 Chapter 1

(i) Y is bounded, i. e., there exists a constant c > 0 such that

lu (x)1 sc
for all x E K and u E Y.
(ii) Y is equicontinuous, i. e., for every c > 0 there exists a 8 > 0 such
that for all u E Y,
lu (x) - u (x') I < c
whenever x, x' E K and d (x, x') < 8.

Proof. Assume that Y is relatively compact in C (K; Rn). Then,

obviously, it is bounded and according to Hausdorff's theorem, for every
c> 0 there exists in C (K; Rn) a finite c/3-net R for Y. Let

R = {Ul' U2, ... , urn}.

The functions Uk being uniformly continuous on the compact K, there

exists a 8> 0 such that for all k E {I, 2, ... , m},

lud x)- ud X' ) 1< c/3

whenever x, x' E K and d (x, x') < 8. Now for any u E Y there exists a
k E {I, 2, ... , m} with
lu - ukloo < c/3,
so if d (x, x') < 8, we have

Iu (x) -
u (x') I
< lu (x) - Uk (x)1 + IUk (x) - uk (x') 1+ IUk (x') - u (x') I
< c/3+c/3+c/3=c.
This shows that Y is equicontinuous.
Conversely, assume that (i) and (ii) hold. The space C (K; Rn) being
a closed subspace of the Banach space B (K; Rn) of all bounded functions
from K to R n, it is sufficient to prove that Y is relatively compact in
B (K; Rn). For this, according to Hausdorff's theorem, we have to show
that for every c > 0 there exists in B (K; Rn) a relatively compact c-net
for Y. For a fixed c > 0 we consider 8 > 0 given by condition (ii). Since
Compactness in Metric Spaces 17

K is compact, there is a finite o/2-net {Xl, X2, ... , xm} for K. Let

Kl (Xl) ,
= B 8/ 2
K2 = B8/2 (X2) \ K l ,

m-l
Km = B8/2 (Xm) \ UK j .
j=l

One has
m

K = UK j , Ki n K j = 0 for i -=1= j,
j=l
and
d (x, x') < 0 for all x, x' E Kj, j = 1, 2, ... , m.
Let 'Pj be the characteristic function of K j , i.e.,

I for X E K j
'Pj (x) = { 0 for X E K\Kj .

We consider the following set of bounded functions from K to R n ,

R = {t'PjAj: Aj ERn, IAjl:::; c, j = 1,2, ...

J=l
,m}.
Here c is the constant in (i). It is easy to see that R is relatively compact
in B (K; Rn) since the convergence of a sequence of functions in R reduces
to the convergence of the corresponding sequences of vectors Aj. Thus it
remains to prove that R is an c-net for Y For this let us choose an element
Xj E Kj for every j E {I, 2, ... , m}. For each U E Y we build the function
m

v(x)=L'Pj(x)u(Xj) (xEK).
j=l

Obviously v E R since Iu (Xj) I :::; c. In addition, for any X E K there exists

a j with X E K j , and so v (x) = U (Xj). Hence by (ii)

Iu(x) - v(x)1 = Iu(x) - u(Xj)1 < c

since d (x, Xj) < O. Consequently Iu - vl oo < c, which shows that R is an
c-net for Y •
18 Chapter 1

Let 1.lk 00 denote the norm of C k (0; Rn) (k E N\ {O}, 0 C RN bounded

open), '
lulk,oo = max {Iul oo , lu'l oo ' ... , luCk) loo} .
Corollary 1.1 Let 0 be a bounded open subset of RN. Every bounded sub-
set of the space (C 1 (0; Rn) ,1.iI,00) is relatively compact in (C (0; Rn) ,1.1 00 ),

Proof. Let Y be a bounded subset of (C1 (0; Rn) ,1.11,00) . Then Y is

also bounded in (C (0; Rn) ,1.100)' In addition, all functions in Y are Lip-
schitz with the same Lipschitz constant, and therefore Y is equicontinuous.
Now the Ascoli-Arzela theorem guarantees that Y is relatively compact in
(C (IT;Rn) , 1.100)' •
Remark 1.1 Let 0 be a bounded open subset of RN and k E N\ {O} . Every
bounded subset of the space (C
k (IT; Rn) ,I.koo) is relatively compact in

(C k -1 (0; Rn) , 1.lk-1,00) .

1.3 The Frechet-Kolmogorov Theorem

In this section we present some basic properties of the spaces V (0; Rn)
and the LP version of the Ascoli-Arzela theorem.
Let 0 C RN be open and pER, 1 :s: p < 00. Let V (0; Rn) be the
space of all measurable functions u : 0 ~ Rn such that lul P is Lebesgue
integrable on O. The space V (0; Rn) is endowed with the norm

lul p = (In lu (x)IP dX) lip

For p = 00 we let Loo (0; Rn) be the space of all measurable functions
u : 0 ~ R n for which there exists a constant c with lu(x)1 :s: c for
almost every (a.e. for short) x E O. The norm on LOO (0; Rn) is given by
lul oo = sup ess lu (x)l, i.e.,
lul oo = inf {c: lu (x)1 :s: c for a.e. x E O}.
Under the norm 1.lp ' LP (0; Rn) is a Banach space for each p E [1,00]. The
space L2 (0; Rn) is a Hilbert space with the inner-product

(u,vh = In (u(x),v(x))dx.
Compactness in Metric Spaces 19

In some places we shall denote the norm 1.l p by 1.ILP(O;Rn) in order to avoid
any confusion.
The space LP (0; R) is denoted by LP (0). Also, for 0 = (a, b) C R
we write V (a, b; Rn) instead of V ((a, b) ; Rn) and V (a, b) instead of
LP ((a, b)).
For 1 ~ p ~ 00 we let p' be the conjugate exponent of p, i.e., lip +
lip' = 1. Recall the well known Holder's inequality: if u E LP (0; Rn) and
v E Vi (0; Rn) then lullvl E L1 (0) and

Llu (x)llv (x)1 dx ~ lul p Ivl p l •

This inequality implies that if 0 is bounded then for 1 ~ p < q ~ 00, the
following inclusions hold
C (0; Rn) C Lq (0; Rn) C V (0; Rn).
We also recall that the dual of V (0; Rn) (1 ~ p < (0) is the space
Vi (0; R n), where lip + lip' = 1, whilst the dual of Loo (0; Rn) strictly
includes £1 (0; Rn) .
The following density result will be useful: the space Co (0; Rn) of all
continuous functions from 0 to Rn with compact support in 0 is dense
in V (0; Rn) for every p E [1, (0).
For details and further information on the V spaces, we refer the reader
to Brezis [6].
For the rest of this section we assume that 0 C RN is a bounded open
set. For a given number r > 0 and any locally integrable (integrable on
each compact subset) function u : RN - 7 Rn we define the average function
of u with respect to radius r, mr (u) : RN - 7 Rn, by

mr (u) (x) =~
Wr
r
JBr(x)
u (y) dy (x ERN) . (1.1)

Here Wr = fL (B r (0)), the measure of the ball of radius r of RN. For a

function u E V (0; Rn) we define the average function mr (u) in the same
way assuming that u (y) = 0 for y ~ O.
Lemma 1.1 For every u E LP (0; Rn) (1 ~ p < (0) and every r > 0, the
function mr (u) is continuous on RN and satisfies
Imr (u) (x)1 ~ w;1/p IUb(O;Rn), x ERN, (1.2)
and
Imr (u)b(O;Rn) ~ lub(O;Rn) . (1.3)
20 Chapter 1

Proof. We have

Imr (u) (x) - mr (u) (x') I w;ll r 1Br (x)

u(y)dy- r
1Br (x')
u(y)dy

< W;l r
1Br(x,x')
[U (y)[ dy,

where
Br (x, x') = (Br (x) U Br (x')) \ Br (x) n Br (x') .
Holder's inequality applied to u and to the constant function 1 yields

Imr (u) (x) - mr (u) (x') I :s W;lf-L (Br (x,x,))llp'[ub(O;Rn)'

Letting x' ----> x we deduce that

mr (u) (x') ----> mr (u) (x)

since f-L (Br (x, x')) ----> O. Hence mr (u) is continuous at any point x E RN.
Notice that (1.2) immediately follows from (1.1) if we apply Holder's
inequality to the functions 1 and u.
Finally, (1.3) is obtained if we apply Holder's inequality and we inter-
change the order of integration as show the following inequalities:

[m r (u)[p (In [m (u) (x)[P dX) lip

r

< (w;!!, (In,(,) lu (y)I dY) dX) P l/p

lip
w;llp r r [u(x+y)[Pdydx
( )
101Br(O)
lip
w;llp ( r
1 Br(O) 10
r [u (x + y)[P dxdy )
< [u[p.
This concludes the proof. •
For any hE RN and any u E LP (n; Rn) we define the translation of u
by h, to be the function Th (u) from RN to R n, given by

u (x + h) if x + hEn
Th(U)(X)= { 0 ifx+hERN\n.
Compactness in Metric Spaces 21

Theorem 1.3 (Fnkhet-Kolmogorov) Let 0 C RN be bounded open

and 1 ::; p < 00. A subset Y of LP (0; Rn) is relatively compact if and
only if the following conditions are satisfied:
(i) Y is bounded, i. e., there exists a c > 0 such that

Iul p ::; c
for all U E Y;
(ii) Th (u) -- U in LP (0; Rn) as h -- 0, uniformly for U E Y, i.e.,

sup ITh (u) - ulp -- 0 as h -- O.

uEY

Proof. Assume that Y is relatively compact in LP (0; Rn). Then

clearly (i) holds. Also, for a given c > 0 there exists in LP (0; Rn) a
finite c/3-net for Y. Since C (0; Rn) is dense in LP (0; Rn) (recall that
o is bounded), we may assume that the elements of the net belong to
C (0; Rn). Let these elements be Uj, j = 1,2, ... , m. Since Uj is uniform
continuous on the compact set 0, there exists a 8> 0 such that

IUj (x) - Th (Uj) (x)1 IUj (x) - Uj (x + h)1

c
<
(2· 3Pf-t (O))l/p

for all h E RN with Ihl < 8, x E Oh = 0 n (0 - h) and j = 1,2, ... , m. It

follows that
r IUj (x) -
Jflh Th (Uj) (x)IP dx < 2. 3P'
cp

In addition, by (i) we may assume that for each j we have

1fl\flh
IUj (x)IP dx ::; - - .
2· 3P
cp

Consequently
c
ITh (Uj) - ujlp < 3
for Ihl < 8 and j = 1,2, ... , m. Now, for a given U E Y there is a j E
{1,2, ... ,m} with Iu - ujlp < c/3. On the other hand, we have

Th (u) - U = Th (u - Uj) - (u - Uj) + (Th (Uj) - Uj) .

As a result, for Ihl < 8 one has

ITh (u) - ulp ::; ITh (u - uj)lp + Iu - ujlp + ITh (Uj) - ujlp < c.
22 Chapter 1

Thus (ii) holds.

Conversely, assume (i) and (ii) are satisfied. For any u E Y one has

Imr (u) (x) - mr (u) (x') I

w;ll { u(y)dy- ( u(y)dy

JBr(x) JBr(x')
< w;l { lu (y) - Tx,-x (u) (y)1 dy
JBr(x)
< w;l/p lu - Tx,-x (u)lp.

This inequality together with (ii) shows that the set

mr (Y) = {mr (u) : u E Y}

is equicontinuous. In addition, according to (1.2) and (i) the set mr (Y)

is bounded in C (D; Rn). Now the Ascoli-Arzela theorem guarantees that
mr (Y) is a relatively compact subset of C (IT; Rn), and so of £P (D; Rn) .
On the other hand, from

mr (u) (x) - u (x) W;l { (u (y) - u (x)) dy

JBr(x)
w;l { (Ty (u) (x) - u (x)) dy
JBr(O)
we deduce
Imr (u) - ul p ::; sup ITy (u) - ul p .
iyi:<:;r
This shows that Y is the uniform limit (in £P (D; Rn)) of mr (Y) as r ----t o.
Now Proposition 1.2 guarantees that Y is relatively compact in LP (D; Rn) .
• Notice we can slightly modify the Fnkhet-Kolmogorov theorem in order
that a unified criterion for compactness in both C (D; Rn) and £P (D; Rn)
(1 ::; p < 00) be possible. More exactly, we have the following theorem.

Theorem 1.4 Let D C RN be bounded open and p E [1,00]. Let Y c

£P (D; Rn) for 1 ::; p < 00 and Y C C (IT; Rn) for p = 00. Assume that
there exists a function lJ E £P (D) such that

lu (x)1 ::; lJ (x) (1.4)

Compactness in Metric Spaces 23

for a.e. x E 0 and all u E Y. Then Y is relatively compact in LP (0; Rn)

if and only if

sup
uEY
ITh (u) - ub(l1 .Rn) -+ 0
h,
as h -+ o. (1.5)

Here Oh = 0 n (0 - h).

Proof. 1) Assume p = 00. Then (1.4) is equivalent to the boundedness

of Y in C ([2; Rn) , whilst (1.5) means exactly the equicontinuity of Y.
2) Let 1 :S p < 00. Then (1.4) implies that Y is bounded in LP (0; Rn) .
Hence condition (i) in Theorem 1.3 holds. Now observe that (1.4) and (1.5)
guarantee condition (ii) in Theorem 1.3. Indeed, one has

ITh (u) - ul~p(l1;Rn) = ITh (u) - ul~p(l1h;Rn) + lu i,p(l1\Oh;Rn) .

1

Then using (1.4) we obtain

ITh (u) - ul~p(l1;Rn) :S ITh (u) - ul~p(l1h;Rn) + Iv1i,p(l1\Oh) .

This inequality together with (1.5) guarantees (ii) in Theorem 1.3, since

Iv1i,p(l1\Oh) -+ 0 as h -+ o.
Thus Theorem 1.3 applies and the proof is complete. •
 Chapter 2

Completely Continuous
Operators on Banach Spaces

In this chapter we define the notion of a completely continuous operator

from a Banach space to another Banach space and we present some simple
properties of the completely continuous operators. Next we prove Brouwer's
fixed point theorem. Finally, we prove the famous Schauder fixed point the-
orem which, like Banach's contraction principle, represents a fundamental
result in nonlinear analysis.

2.1 Completely Continuous Operators

Definition 2.1 Let X, Y be Banach spaces and T : Dc X ---> Y.
(a) The operator T is said to be bounded if it maps any bounded subset
of D into a bounded subset of Y.
(b) The operator T is said to be completely continuous if it is continuous
and maps any bounded subset of D into a relatively compact subset of Y.
(c) The operator T is said to be of finite rank if T (D) lies in a finite-
dimensional subspace of Y.

It is clear that a continuous operator T : D c X ---> Y is completely

continuous if and only if for every bounded sequence (Uk) with Uk E D,
the sequence (T (Uk)) has a convergent subsequence.
Notice that any completely continuous operator is a bounded operator.

Theorem 2.1 (1) If the operators T I , T2 : D c X ---> Yare bounded

(completely continuous) then for every a, f3 E R the operator aTl + f3T2 is
bounded (respectively, completely continuous).

25
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
26 Cbapter 2

(2) Let X, Y, Z be Banach spaces and T 1 , T2 be two operators which

are defined as follows:

-
DI TI TI (Dt) C D2 T2
- Z, DI C X, D2 C Y.

If both operators T 1 , T2 are bounded then the composite operator T2TI is

also bounded. If one of the operators T 1 , T2 is bounded continuous and the
other one is completely continuous, then T2TI is completely continuous.

Proof. Both statements (1) and (2) are simple consequences of Defini-
tion 2.1. For (2) also use the fact that a continuous operator maps relatively
compact sets into relatively compact sets . •

Theorem 2.2 (1) If the operators Tk : D -+ Y, D c X, k = 1,2, ... are

completely continuous and T: D -+ Y is such that
T (u) = lim Tk (u) (2.1)
k-+oo

uniformly on any bounded subset of D, then T is completely continuous

too.
(2) Let D C X be a bounded closed set and T : D -+ Y a completely
continuous operator. Then there exists a sequence of continuous operators
Tk : D -+ Y of finite rank such that (2.1) holds, uniformly on D, and
TdD) C conv (T (D))
for every k.

Proof. (1) We first note that T is continuous at any point Uo ED. This
follows, using (2.1) and the continuity of Tk, from the following inequality

IT (u) - T (uo)ly ::; IT (u) - Tk (u)ly + ITk (u) - n (uo)ly

+ ITk (uo) - T (uo)ly.
Let M c D be a bounded set. Since Tk is completely continuous we
have that Tk (M) is relatively compact. In addition, from (2.1) we have
that T (M) is the uniform limit of Tk (M) as k -+ 00. By Proposition 1.2,
T (M) is relatively compact. Hence T is completely continuous.
(2) Since T is completely continuous and D is bounded, the set T (D)
is relatively compact. Consequently for every c > 0 there exist finitely
many elements Vj E T (D), j = 1,2, ... , me, such that
m.
T (D) c U Be(vj).
j=l
Completely Continuous Operators 27

Let (cpj) be the partition of unity with respect to this covering of T (D).
Hence
me
CPj E C (T(D); [0,1]), sUPPCPj C Be (Vj), 2.:CPj (v) = 1
j=l

for all v E T (D). We define the operator Te : D -+ Y by

me
Te (u) = 2.: CPj (T (u)) Vj (u ED).
j=l

By its construction Te is a continuous operator of finite rank and

Te (D) c conv (T (D)) .

In addition, we have

me
IT (u) - Te (u)ly 12.: CPj (T (u)) (Vj - T (u))
j=l
IY
me
< 2.: CPj (T (u)) IVj - T (u)ly
j=l
me
< E: 2.:CPj (T(u)) = E:
j=l

since T (u) E Be (Vj) whenever CPj (T (u)) > o. Hence

T (u) = lim Te (u) ,

e->O

uniformly for u E D. Finally, take Tk := Te with E: = 11k . •

Here is a notion whose definition involves completely continuous opera-
tors.

Definition 2.2 Let (X, 1.lx) and (Y,I.ly) be Banach spaces such that
Y C X. We say that the embedding Y C X is continuous (respectively,
completely continuous) if the injection map j : Y -+ X, j (u) = u (u E Y)
is continuous, respectively, completely continuous.

We note that in the literature, the term 'compact embedding' is fre-

quently used instead of 'completely continuous embedding'.
28 Chapter 2

It is clear that the embedding Y c X is continuous if and only if there

is a constant c > 0 such that

JuJx ::; cJuJy, u E Y.

For example, according to Remark 1.1 the embedding Ck (0; Rn) c

C k - 1 (0; Rn) is completely continuous. Here 0 C RN is bounded open
and kEN \ {O} . Also, if 1 ::; p < q ::; 00 and 0 is bounded open then
the embedding Lq (0; Rn) c V' (0; Rn) is continuous since by Holder's
inequality
JuJp ::; /-L (0) = pq JuJ q , u E Lq (0; Rn) .

Here (q - p) I (pq) = lip if q = 00.

Note that if the embedding Y C X is continuous (respectively, com-
pletely continuous), then the embedding X* C y* is continuous (respec-
tively, completely continuous).

2.2 Brouwer's Fixed Point Theorem

The basic topological tool in Part I will be Schauder's fixed point theorem.
In this section we prove the finite-dimensional version of Schauder's theorem,
namely Brouwer's fixed point theorem [8].

Theorem 2.3 (Brouwer) Let D C RN be a nonempty convex compact

set and let T : D ----> D be a continuous mapping. Then there exists at least
one u E D with T (u) = u.

For the proof we need the following lemmas:

Lemma 2.1 Let F : R n +1 ----> Rn be a c= function. Consider the deter-

minant
A = det [DoF , ... , Di-1F , Di+1F , .'" DnF ]
whose columns are

Di F = (fJFl/OXi+l, ... , fJFnlfJxi+1)'

Then
n
L (_l)i DiA = 0, (2.2)
i=O
where DiA = fJAi/fJxi+1'
Completely Continuous Operators 29

Proof. Let i, j E {O, 1, ... , n} , i -::f. j. For i < j we define

Cij = det [DijF, DoF, ... , Di-1F, Di+1F, ... , Dj-1F, Dj+1F, ... , DnF]
where
DijF = ({)2 Fl/{)Xi+1{)Xj+l, ... , ()2 Fn/{)Xi+l{)Xj+l) .
For i > j we let C ij = C ji . We can easily check that
n
DiAi = L (-l)j C ij + L (_l)j-l C ij = L (-l)j (J (i,j) C ij
j<i j>i j=O
where
{ 1 if j < i,
(J(i,j) = 0 ifj=i,
-1 if j > i.
Hence
n n
L (_l)i DiA = L (_l)i+j (J (i,j) C ij .
i=O i,j=O
Since (J(i,j) = -(J(j,i) and C ij = C ji , we have

L L
n n
(_l)i+ j (J (i,j) C ij (_l)i+ j (J(j,i)Cji
i,j=O i,j=O

-L
n
(_l)i+ j (J (i,j) C ij ,
i,j=O

which proves (2.2). •

The next lemma is Brouwer's theorem for mappings of class Coo.

Lemma 2.2 Let T : B ----t B, where B = Bl (0; Rn), be a Coo function.

Then T has at least one fixed point.

Proof. Assume that T has no fixed points. Then for any u E B the
line generating by u and T (u) :

v = u + ex (u - T (u)) , ex E R,

cuts the sphere ()B in two points. Let ex (u) be the nonnegative value of ex
which corresponds to one of these two points. It is clear that ex (u) is the
biggest root of the equation

lu+ex(u-T(u))1 = 1,
30 Chapter 2

which can be rewritten in the equivalent form

0:21u - T (U)12 + 20: (u, U - T (u)) + lul 2 = 1.

Hence

0: (u) = lu - T (u)I- 2 {(u, T (u) - u)

+ ((u, u - T (u))2 + (1 -luI2) lu - T (u)1 2/12}.

This defines a C= function from B to R. In addition,

0: (u) = ° for lui = 1.

Let us now define the mapping F: [0, 1J x B ----) B by

F (t, u) = u + to: (u)( u - T (u)) .

We have that F is of class C= and satisfies
DtF(t,u) =0 for lui = 1,
{ F (0, u) = u, (2.3)
IF (1, u)1 = l.
Here DtF (t, u) = 8F (t, u) j8t. Let

Ao (t, u) = det [D1F (t, u) , ... , DnF (t, u)J

and
I(t)= LAO (t,u)du.

We have Ao (0, u) = 1 and 1(0) = f-1 (B) > 0. Also Ao (1, u) == 0. Indeed,
the vectors D1F (1, u) , ... , DnF (1, u) belong to the tangent hyperplane to
8B at F (1, u). Thus they are linearly dependent and so Ao (1, u) == 0, as
we claimed. Consequently 1(1) = 0. We obtain a contradiction once we
prove that I (t) is constant; more exactly, that I' (t) == 0. To this end we
differentiate under the integral sign and we use (2.2). We then obtain

I' (t) = 1B
DtAo (t, u) du = t (-1)i+11
i=l B
DiA (t, u) du,

where

Ai (t, u) = det [DtF, D1F, ... , Di-1F, Di+1F, ... , DnFJ.

Completely Continuous Operators 31

Let Bi = B n {u ERn : Ui = o}. For each U E Bi we define

<pi (u) = u + (0, ... , Ii (u) , ... , 0)

and
<pi (u) = u + (0, ... , -,i (u), ... ,0)
where Ii (u) = (1 - E u~)1/2.
J
'J.'
Then
Jr'

k DiAi (t, u) du = 11 1'i(U)

DiAi (t, u) du

ki
Bi -1'i(U)

= [Ai (t, <pi (u)) - Adt , <pi (u))]

XdUl ... dUi-ldui+l ... dUn
= 0

for i = 1,2, ... ,n. Indeed, since l<pf(u)1 = 1, according to (2.3), one has
DtF (t, <pf (u)) = O. Hence the first column ofthe determinant Ai (t, <pf (u))
is null, and so this determinant equals zero. Therefore l' (t) == O. •
Proof of Theorem 2.3. Step 1. Assume that D = B = Bl (0; Rn) .
Let
T (u) = (tl (u) , t2 (u) , ... , tn (u))
where ti : B ---+ [-1,1], i = 1,2, ... , n, are continuous functions. According
to Weierstrass' approximation theorem, each function ti is the uniform limit
of a sequence (tn k~l of functions tf : B ---+ [-1, 1] of class Coo satisfying

Itf (u)1 ~ Iti (u)1

for all u E B, i = 1,2, ... , nand k = 1,2, .... Define Tk : B ---+ B by

Tk (u) = (tT (u) , t~ (u) , ... , t~ (u)) .

Then
T (u) = lim Tk (u)
k-+oo

uniformly on B. By Lemma 2.2, for each k there exists a Uk E B with

Tk (Uk) = uk·
32 Chapter 2

Since B is compact, (Uk) has a subsequence (UkJ j 2:1 convergent to some

element Uo E B. Then

IT(uo)-uol ~ IT(uo)-T(UkJI+IT(UkJ-Tkj(UkJI
+ IUkj - Uo I ---t 0 as j ---t 00.

Therefore T (uo) = Uo.

Step 2. The case D = Br (0; Rn) can be reduced to the previous one
by the change of variables v = r-1u and by considering the operator T' :
B ---t B,
T' (v) = r-1T (rv) (v E B).
Step 3. We now consider the general case. We first show that there is
a continuous extension T of T to the whole space Rn with T (Rn) c D.
Indeed, D being compact in R n , we may choose a countable, dense subset
{Uj: j E N\ {O}} of D. For any U 1'. D and any j we let

'ljJj (u) = max { 2 - IU-Ujl

d(u,D) ,
o}.
Here
d (u, D) = inf {Iu - vi : v E D}.
Then the mapping T: Rn ---t D given by

T (u) if U E D,
T (u) = { (L 2- j 'ljJj (u))-l L 2- j 'ljJj (u) T (Uj) if U 1'. D
j2:1 j2:1

is the desired continuous extension of T. Now choose any r > 0 large

enough so that Dc Br (0; Rn) and define

S : Br (0; Rn) ---t D c Br (0; R n ), S (u) = T (u).

We are with S in the case considered at Step 2. Hence there exists a
U E Br (0; Rn) with S (u) = u, that is T (u) = u. Since T (Rn) c D, we
have U E D and so T (u) = u. •
For another proof of Brouwer's theorem see Kantorovitch-Akilov [29].

2.3 Schauder's Fixed Point Theorem

The main result in this chapter is the famous fixed point theorem of Schauder
[49].
Completely Continuous Operators 33

Theorem 2.4 (Schauder) Let X be a Banach space, K c X a nonempty

convex compact set and let T : K ---t K be a continuous operator. Then T
has at least one fixed point.

Proof. Obviously T is completely continuous. Consequently by The-

orem 2.2 there exists a sequence of continuous operators Tj : K ---t K of
finite rank such that
T (u) = .lim Tj (u)
)->00

uniformly on K. Let n = n (j) be the dimension of the subspace Xn

generated by Tj (K). We have

Tj : KnXn ---t KnXn .

Consequently by Brouwer's theorem there exists Uj E K n Xn with

Tj (Uj) = Uj.

Since K is compact there exists a subsequence of (Uj) convergent to some

element U E K. As in Step 1 in the proof of Brouwer's theorem we can
conclude that T (u) = u . •
The following variant of Schauder's theorem is most useful in applica-
tions.

Theorem 2.5 (Schauder) Let X be a Banach space, Dc X a nonempty

convex bounded closed set and let T : D ---t D be a completely continuous
operator. Then T has at least one fixed point.

We can derive Theorem 2.5 from Theorem 2.4 via the following result.

Lemma 2.3 (Mazur) The convex hull of any relatively compact subset of
a Banach space is relatively compact.

Proof. Let Y be a relatively compact subset of a Banach space X.

Then, given € > 0 we find a finite number of elements of X, say Ul, U2, ... , Urn
such that
rn

YCUBE(Ui). (2.4)
i=l

Let
R = conv {Ul' U2, ... , urn}.
34 Chapter 2

Our goal is to prove that n is a relatively compact €-net for conv (Y) .
Once this is proved, we may say that conv (Y) is a relatively compact set
in base of Hausdorff's theorem. To this end consider an arbitrary element
U E conv (Y). We have

n n
U=LAjVj, Aj>O, LAj=l, VjEY.
j=l j=l

According to (2.4), for each Vj there is an i j E {I, 2, ... , m} with Vj E

Be (Uij). Then

n n
Iu - LAjUijl IL Aj (Vj - Uij) I
j=l j=l
n
< L Aj IVj - Uij I
j=l
< €

n
and L AjUij E n. This shows that n is an €-net for conv (Y). Finally,
j=l
the compactness of n follows from the representation:

n= {f j=l
AjUj : 0::; Aj ::; 1, f
j=l
Aj = I} .
Thus the proof is complete. •
Proof of Theorem 2.5. Since T is completely continuous and D is
bounded the set T (D) is relatively compact. Mazur's lemma then implies
that the set K = conv (T (D)) is compact (and obviously convex). From
T (D) c D and D closed convex it follows that KeD. Now we apply
Theorem 2.4 to the operator T: K -+ K . •
 Chapter 3

Continuous Solutions
of Integral Equations
via Schauder's Theorem

This chapter presents three examples of nonlinear integral operators which

are completely continuous on some spaces of continuous functions: the Fred-
holm integral operator, the Volterra integral operator, and a particular in-
tegral operator with delay. Simultaneously, by means of Schauder's fixed
point theorem we prove existence theorems for continuous solutions of the
integral equations associated to these operators.

3.1 The Fredholm Integral Operator

Let 0 C RN be a bounded open set.

Theorem 3.1 Let h : 0 2 X R n ---) R n be a continuous mapping. Then the

Fredholm operator associated to h, T: C (0; Rn) ---) C (0; Rn) given by

T (u) (x) = in h (x, y, U (y)) dy (x E 0) (3.1)

is completely continuous.

Proof. We first prove that T is continuous. Let Uo E C (0; Rn) and

choose any number R > luo 100 . Let c > o. Since h is uniformly continuous
-2 -
on the compact set 0 x BR (0; Rn), there exists a Oc > 0 such that for

35
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
36 Chapter 3

every u E C(O;Rn) satisfying lu-uoloo:S be one has u(y) E BR(O;Rn)

and
Ih (x, y, u (y)) - h (x, y, Uo (y))1 :S €
for all x, y E O. Then

IT (u) (x) - T (uo) (x) I :S 10 Ih (x, y, u (y)) - h (x, y, Uo (y)) Idy

< € IL (0)

for every x E O. Hence

IT (u) - T (uo) 100 :S € IL (0)

whenever lu - uol oo :S be. Therefore T is continuous at uo·
Next, given a bounded subset Y of C (n; Rn), we shall prove that
T (Y) is relatively compact in C (0; Rn). According to the Ascoli-Arzela
theorem, we have to show that T (Y) is bounded and equicontinuous.
Indeed, since Y is bounded there exists a constant c > 0 such that

lul oo :S c for all u E Y.

It follows that for any u E Y we have

IT (u)loo :S MIL (0),

where
M = max Ih (x, y, z) I .
0 2 xBc(O;Rn)
Hence the set T (Y) is bounded in C (n; Rn) .
On the other hand, using the uniform continuity of h on the compact
o x -Be (0; Rn), for each € > 0 there exists a be > 0 such that
-2

Ih (x, y, u (y)) - h (Xl, y, U (y) ) I :S €

for all x, Xl, yEn with Ix - xii :S be and u E Y. This immediately yields

IT (u) (x) - T (u) (Xl) I :S € IL (0) ,

for all x, Xl E 0 satisfying Ix - xii :S be and u E Y. Thus T (Y) is equicon-

tinuous. •

The next result is a local version of Theorem 3.1.

Continuous Solutions of Integral Equations 37

-2 -
Theorem 3.2 Let R> 0 and h : 0 x BR (0; Rn) ---+ Rn be a continuous
mapping. Then the operator T : BR (0; C (0; Rn)) ---+ C (0; Rn) given by
(3.1) is completely continuous.

Proof. Essentially the same reasoning as in the proof of Theorem 3.1

establishes the result. •
Using Theorem 3.2 and Schauder's fixed point theorem we shall prove
the existence of continuous solutions in a ball of C (0; Rn) , to the Fredholm
integral equation in R n

u(x) = in h(x,y,u(y))dy, x E O. (3.2)

-2 -
Theorem 3.3 Let R> 0 and h : 0 x BR (0; Rn) ---+ Rn be a continuous
mapping. Assume
M JL(O) :::; R, (3.3)
where
M = max Ih (x,y, z)l.
TI 2 xBR(O;Rn)
Then (3.2) has at least one solution u E C (0; Rn) with lul oo :::; R.

Proof. According to Theorem 3.2, the operator T: BR (0; C (0; Rn)) ---+
C (0; Rn) given by (3.1) is completely continuous. On the other hand, (3.3)
guarantees that

T (BR (O;C (0; Rn))) C BR (O;C (0; Rn)) .

Thus the conclusion follows from Theorem 2.5. •

3.2 The Volterra Integral Operator

Essentially the same reasoning as in the proof of Theorem 3.1 establishes
the following two results:

Theorem 3.4 Let h : [a, b]2 X Rn ---+ Rn be continuous. Then the Volterra
operator associated to h, T: C ([a, b] ; Rn) ---+ C ([a, b] ; Rn) given by

T(u) (t) = it h(t,s,u(s))ds (t E [a,b]) (3.4)

is completely continuous.
38 Chapter 3

2 -
Theorem 3.5 Let R > 0 and let h: [a, b] x BR (0; Rn) - t Rn be contin-
uous. Then the operator T: BR (0; C ([a, b]; Rn)) - t C ([a, b] ; Rn) given by
(3.4) is completely continuous.

As an application we present an existence theorem for the Volterra in-

tegral equation in R n

U(t) = ith(t,s,u(s))ds+V(t), tE[a,b]. (3.5)

Theorem 3.6 Let h : [a, b]2 X Rn -7 Rn be continuous and let v E

C ([a, b] ; Rn). Assume that there exists constants 0:, (3 E R+ such that

Ih (t, s, z)1 :s 0: Izl + (3 (3.6)

for all t, s E [a, b], z ERn. Then (3.5) has at least one solution U E
C ([a, b] ; Rn) .

Proof. Let T : C ([a, b] ; Rn) -7 C ([a, b]; Rn) be given by

T (u)(t) = it h (t, s, U (s)) ds + v (t) .

According to Theorem 3.4, T is completely continuous. We now show that
T is a self-mapping of a closed ball of the space C ([a, b] ; Rn) endowed
with a suitable norm, equivalent to the sup-norm 1.100' Indeed, for any
given number 0 > 0 we have

IT (u)(t) I lit h(t,s,u(s))ds+V(t)1

< it Ih (t, s, u (s))1 ds + Ivloo

< 0: it lu (s)1 ds + (3 (b - a) + Ivloo

0: i t lu (s)1 e-OCs-a)eOCs-a)ds + (3 (b - a) + Ivloo

< 0: lu (.) e- OC .- a)100 it eOCs-a)ds + (3 (b - a) + Ivl oo

0: 0- 1 lu (.) e- OC .- a) 100 (e OCt - a) - 1) + (3 (b - a) + Ivl oo
< 0: 0- 1 lu (.) e- OC .- a) 100 eOCt - a) + (3 (b - a) + Ivl oo
Continuous Solutions of Integral Equations 39

Since e-O(t-a) :s 1 on [a, b], we deduce

IT (u)(t) Ie-O(t-a) :s aO- 1 lu (.) e-O(.-at>o +,6 (b - a) + Ivloo

and so

IT (u) (.) e-O(.-a) 100 :s a 0-1 lu (.) e-O(.-a) 100 +,6 (b - a) + Ivl oo . (3.7)

Now fix any 0> a. Then we can find R> 0 such that

aO- 1 R +,6 (b - a) + Ivloo :s R. (3.8)

Consider a new norm on C ([a, b] ; Rn), namely

Ilull = lu (.) e-O(.-a) 100 .

It is clear that the norm 1.100 and 11.11 are equivalent (thus T is also com-
pletely continuous with respect to 11.11). On the other hand, (3.7) and (3.8)
show that T maps the closed ball of center 0 and radius R of the space
(C ([a, b] ; Rn), 11.11), into itself. Now the conclusion follows from Theorem
2.5 . •
The reader could try to give variants of Theorem 3.6 to the case where
condition (3.6) is replaced by an inequality of the form

Ih (t, s, z)1 :s V> (Izl)

with other types of functions V> : R+ ---. R+.

3.3 An Integral Operator with Delay

The following delay integral equation

u (t) = l~T f (s, U (s)) ds

can be interpreted as a model for the spread of certain infectious diseases

with a contact rate that varies seasonally. In this equation u (t) is the
proportion of infectives in a population at time t, T is the length of time
an individual remains infectious, and f (t, u (t)) is the proportion of new
infectives per unit time.
40 Chapter 3

In this section we study the existence of continuous solutions on a given

interval of time [0, tIl, for the initial value problem

{
u (t) = fL-r f (s, u (s)) ds, O:S t :S tI,
(3.9)
U (t) = cp (t) , -T :S t :S o.
We assume

f E C ([-T, tIl x Rn; R n ), cp E C ([-T, 01; Rn)

i:
and the following sewing condition holds

cp (0) = f (s, cp (s)) ds. (3.10)

By a solution of (3.9) we mean a function u E C ([-T, tIl; Rn) with u (t) =

cp (t) for all t E [-T,Ol.
The initial value problem (3.9) was studied for the first time in Precup
[441·

Theorem 3.7 Assume f E C ([-T, tIl x Rn; Rn), cp E C ([-T, 01; Rn)
and that (3.10) holds. Then the delay integral operator T : D (T) --+
C ([0, tIl; Rn) given by

T (u) (t) = 1~-r f (s, u(s)) ds (t E [0, tID,

where
D (T) = {u E C ([0, tIl; Rn) : u (0) = cp (O)}
and
u(t) = { cp (t) for t E [-T, 01 ,
u (t) for t E [0, tIl,
is completely continuous.

Proof. Use the Ascoli-Arzela theorem and follow the same steps as in
the proof of Theorem 3.1. We omit the details. _

Theorem 3.8 Assume f E C ([-T, tIl x Rn; Rn), cp E C ([-T, 01; Rn) and
that (3.10) holds. In addition assume that there exist a, /3 E R+ such that

If(t,z)l:s a Izl +/3 (3.11)

for all z E Rn and t E [0, tIl. Then (3.9) has a solution u E C ([-T, tIl; Rn) .
 Continuous Solutions of Integral Equations 41

Proof. The proof closely resembles the proof of Theorem 3.6. Let

"I = max
tE[-T,O]
If (t, 'P (t))I·
We have

IT (u)(t) I ~ T"I+ fotlf(s,U(s))ldS

< T"I+a fo tlu (s)ldS+,Bt1

T"I+,Bt1+ a fotlu(s)le-OSeOSdS

< T"I + ,Btl + a lu (s) e-osloo fot eOsds

< T"I+,Bt1+a lu(s)e-OSlooe-1eOt.

It follows that

IT (u) (t) e-Otl oo ~ T"I +,Bh + ae- 1 lu (s) e-osloo'

Now choose e> a and a number R> a with
T"I + ,B t1 + a e- 1R ~ R.

Then T (B) c B, where

B = {u E D (T) : lu (t) e-Otl ~ R for all t E [0, t1l} .

It is clear that B is a nonempty convex bounded closed subset of C ([0, t1l ; Rn) .
The conclusion is now immediate from Theorem 2.5 . •
As in the previous section, the reader could try to obtain existence results
for (3.9) assuming instead of (3.11) a condition of the form

If (t, z)1 ~ 'lj; (Izl)

with different types of function 'lj; : R+ -+ R+.

 Chapter 4

The Leray-Schauder
Principle
and Applications

In applications one of the drawbacks of Schauder's fixed point theorem is the

invariance condition T (D) cD which has to be guaranteed for a bounded
closed convex subset D of a Banach space. The Leray-Schauder principle
[32] makes it possible to avoid such a condition and requires instead that a
'boundary condition' is satisfied. In this chapter we shall prove the Leray-
Schauder principle and we shall apply it in order to obtain existence results
for continuous solutions of integral equations. In particular, we give results
on the existence of continuous solutions of initial value and two-point bound-
ary value problems for nonlinear ordinary differential equations in R n. The
results will be better than those established by means of Schauder's theorem.

4.1 The Leray-Schauder Principle

Theorem 4.1 (Leray-Schauder) Let X be a Banach space, K c X a
closed convex subset, U c K a bounded set, open in K and Uo E U a fixed
element. Assume that the operator T : U -7 K is completely continuous
and satisfies the boundary condition
U -=J- (1 - A) Uo + AT (u) for all U E aU, A E (0,1) . (4.1)
Then T has at least one fixed point in U.
Proof. For the proof we use Granas' fixed point approach (see Dugundji-
Granas [19]' Granas [23]' O'Regan-Precup [37] and Zeidler [53]). Notice the

43
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
44 Cbapter 4

property of U of being open as well as the boundary aU are understood

with respect to the topology of K. We may assume that (4.1) holds on aU
for all A E [0,1]. Indeed, this is obvious for A = 0 since Uo E U, whilst if
(4.1) does not hold for A = 1, then the theorem is proved.
Let

s = {u E U : u = (1 - A) Uo + AT (u) for some A E [0, I]} .

Obviously S is nonempty (since Uo E S), closed, and S n aU = 0. By

Urysohn's lemma (see Dugundji-Granas [19]), there exists a function cp E
C (U; [0,1]) such that

cp (u) = { 0 for u E aU,

1 for u E S.

We now define the operator T: K ---; K by

T(u) = {(I-CP(u))uo+CP(U)T(U) for uE U,

~ ~uEK\U

It is clear that T is continuous and

T(K) c conv ({uo}UT(U)).

Since T is completely continuous, T (U) is relatively compact. Hence by
Mazur's lemma the following subset of K,

D= conv ({uo} uT (U)),

is convex and compact. In addition

T(D) cD.

Hence Theorems 2.4 applies and guarantees the existence of a u E D with

T (u) = u. By the definition of T, u must lie in U. Then
u = (1 - cp (u)) Uo + cp (u) T (u) .
This shows that u E S and so cp (u) = 1. As a result, u = T (u) . •
Notice that the essential idea of the Leray-Schauder principle consists
in joining the operator T to the constant operator Uo by means of the
homotopy H: U x [0,1] ---; K,

H (u, A) = (1 - A) Uo + AT (u)
The Leray-Schauder Principle and Applications 45

in a such way that the unique fixed point of H (., 0), namely uo, can be
'continued' in a fixed point of H (., A) for each A E [0, 1J, and, in particular,
in a fixed point of H (., 1) = T. This continuation process is possible if all
operators H (., A) for A E [O,lJ are fixed point free on the boundary of U.
In applications the Leray-Schauder principle is usually used together
with the so called 'a priori bounds technique':
Suppose we wish to solve the operator equation

u = T (u), u E K, (4.2)
where K is a closed, convex subset of a Banach space (X, 1.lx), and T :
K ----t K is completely continuous. Then we look at the set of all solutions
to the one-parameter family of equations

u = (1 - A) Uo + AT (u), u E K, (4.3)

°
when A E (0,1). Here Uo E K is fixed (in most cases Uo = 0). If this set
is bounded, i.e., there exists R> such that

lu-uolx<R
(strictly!) whenever u solves (4.3) for some A E (0,1), then we let U be
the intersection of K with the open ball B R (uo; X). Thus Theorem 4.1
applies and guarantees the existence of a solution to (4.2).
For other results of Leray-Schauder type we refer the reader to the mono-
graph O'Regan-Precup [37J.

4.2 Existence Results for Fredholm Integral

Equations
In this section we present general existence theorems for the Fredholm inte-
gral equation in R n

u(x)= kh(x,y,U(Y))dY, xEn. (4.4)

Here n c RN is a bounded open set. We seek continuous solutions with

values in a given ball

B = {z E Rn : Izl ::; R},

i.e., u E C (n;B).
Theorem 4.1 yields the following existence principle which can be sum-
marized as follows: 'boundedness yields existence'.
46 Chapter 4

-2
Theorem 4.2 Let h : 0 x B -4 Rn be continuous. Assume that

lul oo < R (4.5)

for any solution u E C (TI; B) to

u (x) = A in h (x, y, u (y)) dy, x E 0, (4.6)

for each A E (0,1). Then (4.4) has a solution in C (0; B) .

Proof. Let K = X = C (0; Rn) with norm 1.1 00 ,

U = {u E C (TI; Rn) : lul oo < R},

Uo be the null function and T: U -4 C (TI; Rn) be given by

T (u) (x) = in h (x, y, u (y)) dy (x EO).

The result follows from Theorem 4.1. •

Notice Theorem 3.3 is a corollary of Theorem 4.2. Indeed, if u E
C (0; B) is any solution of (4.6) for some A E (0,1), then for any x E 0,
using (3.3) we obtain

lu(x)1 Alin h(X,y,U(Y))dyl

< A in Ih (x, y, u (y))1 dy

~ AM 1-"(0)
< AR < R.
Hence u satisfies (4.5).
An immediate consequence of Theorem 4.2 is the following existence
result for the Hammerstein integral equation in Rn

u(x)= in/'i,(X,Y)f(y,U(Y))dY , xEO. (4.7)

-2 -
Corollary 4.1 Let /'i, 0 -4 Rand f O x B -4 R n be continuous
functions. Assume
1/'i,(x,Y)I~l (4.8)
The Leray-Schauder Principle and Applications 47

for all x, y E 0, and that there exists a continuous nondecreasing function

'lj; : [0, RJ ---7 R+ with 'lj; (R) > 0, and ¢ E C (0; R+), such that

If (y, z)1 ~ ¢ (y) 'lj; (Izl) (4.9)

for all y E 0, z E B, and

R
1¢ILlCr!) ~ 'lj; (R) . (4.10)

Then (4.7) has a solution u E C (0; Rn) with lul oo ~ R.

Proof. The result follows from Theorem 4.2, where

h (x, y, z) = "" (x, y) f (y, z) ,

once we show that (4.5) holds. For this, let u E C (0; B) be any solution
of (4.6) for some>. E (0,1). Then using (4.8)-(4.10) we obtain

lu(x)1 ~ >. kl,,"(x,y)I¢(y)'lj;(lu(y)l)dY

< >''lj;(R)k¢(Y)dY

>.'lj; (R) I¢iucr!)

< >'R<R.

Hence lul oo < R. Therefore Theorem 4.2 applies . •

In particular, if "" is the Green's function of a certain differential operator
with respect to some boundary conditions, we can guarantee (4.5) by some
other types of conditions on f.
Here are some examples for 0 = [0, 1J, that is for the Hammerstein
equation in Rn

u (x) = 1 1
,," (x, y) f (y, u (y)) dy, x E [0, 1J . (4.11)

Theorem 4.3 Let "" be the Green's function of the differential operator
-u" + u with respect to one set of the homogeneous Dirichlet, Neumann, or
periodic boundary conditions

u (0) = u (1) = 0 (Dirichlet), (4.12)

u' (0) = u' (1) = 0 (Neumann), (4.13)

~ ~~~4

u (0) - u (1) = u' (0) - u' (1) = ° (periodic conditions) . (4.14)

Let f : [O,lJ x B ---t Rn be continuous. Assume that for every z E B with
Izl = R, one has

(z, f (y, z)) ~ R2 for all y E [0, 1J . (4.15)

Then (4.11) has a solution u E C ([0, 1J; Rn) with lul oo ~ R.

Proof. To apply Theorem 4.2 we need to prove (4.5) for any solution
u E C ([0, 1J ; B) of (4.6). Let u E C ([0, 1]; B) be a solution of (4.6) for
some A E (0,1). Clearly, lul oo ~ R. Assume

lul oo = R.
Then there exists a point Xo E [O,lJ such that

lu(xo)1 = R.
On the other hand, since K, is a Green's function u E C 2 ([0, 1J ; Rn) and
solves the problem

{ -u" (x) + u (x) = Af (x, u (x)), x E [0, 1J , (4.16)

u E T3.

Here T3 stands for either homogeneous Dirichlet, Neumann, or periodic

boundary conditions. We shall derive a contradiction to (4.15) once we
show that
(u (xo) , f (xo, u (xo))) > R2. (4.17)
If Xo E (0,1) then since lul 2 achieves its maximum at Xo, we must have

(luI2)" (xo) ~ 0.
Direct computation shows that

(luI2)' = (u, u)' = 2 (u, u')

and
(luI2)" = 21u'12 + 2 (u,u"). (4.18)
Consequently
(u(xo) ,u" (xo)) ~ 0.
The Leray-Schauder Principle and Applications 49

This together with (4.16) yields

o > (u(xo),U(XO)-Aj(XO,U(xo)))
R2 - A (u (xo) ,j (xo, u (xo)))
> A(R2_(u(xO),j(xo,u(xo)))).
Thus (4.17) holds.
Next we prove (4.17) in case that Xo = O. This case can not hold under
conditions (4.12).
Case 1. Assume B means (4.13). Assume

(u(O),j(O,u(O))) ~R2.

Then

R2 > AR2
> (u (0) ,Aj (0, u (0)))
(u (0) ,-u" (0) + u (0))
R2 - (u (0) ,u" (0)) .

It follows that (u(O),u"(O)) > o. Now (4.18) implies (luI2)" (0) > 0 and
so (luI2)' is strictly increasing near o. Then

(luI2)' (x) > (luI2)' (0) = 2 (u (0), u' (0)) = 0

for x > 0 near O. So lul 2 is strictly increasing near 0, which is impossible

since lul 2 achieves its maximum at o. Thus (4.17) holds.
Case 2. Assume B means (4.14). If

(u(O),u'(O)) = (u(l),u'(l)) #0,

it follows that lul 2 can not achieve its maximum at Xo = 0 (equivalently,
at 1). Hence
(u(O),u'(O)) = (u(l),u'(l)) =0
and we make the same reasoning as at Case 1.
Finally, we prove (4.17) similarly if Xo = 1.
Therefore in all cases (4.17) holds, yielding a contradiction. •
50 Chapter 4

Remark 4.1 Theorem 4.3 is in fact an existence result for the two-point
boundary value problem

-u"(x)+u(x)=f(x,u(x)), XE[O,l]
{
uEB.

We refer the reader to Granas-Guenther-Lee [24] and O'Regan-Precup

[37] for numerous results on boundary value problems via the Leray-Schauder
principle.

4.3 Existence Results for Volterra Integral

Equations
This section presents general existence theorems for the Volterra integral
equation in R n

u(t)= ith(t,s,U(S))ds, tE[a,b]. (4.19)

We seek continuous solutions with values in the ball

B = {z E Rn : Izl :S R},
i.e., u E C ([a, b] ; B) .
First we state the analog of Theorem 4.2 for Volterra equations, another
direct consequence of Theorem 4.1.

Theorem 4.4 Let h : [a, b]2 x B ----+ Rn be continuous. Assume that

lul < R
CXl
(4.20)

for any solution u E C ([a, b] ; B) to

u (t) = oX it h (t, s, u (s)) ds, t E [a, b] (4.21)

for each oX E (0,1). Then (4.19) has a solution in C ([a, b] ; B).

Next we give a sufficient condition for (4.20) in the case of the equation
in R n
t (4.22)
u(t)= i "'(t,s)f(s,u(s))ds, tE [a,b].
The Leray-Schauder Principle and Applications 51

Corollary 4.2 Let K : [a, b]2 ~ Rand f : [a, b] x B ~ Rn be continuous

functions. Assume that
IK(t,s)l~l (4.23)

for all t, s E [a, b], there exists a continuous nondecreasing function 1/J
(0, R] ~ (0,00) and a <P E C ([a, b]; R+) such that

If (s, z)1 ~ <P (s) 1/J (Izj) (4.24)

for all s E [a,b] , z E B, and

1<PI£l[a,b} ~ l0
R 1
1/J-((J) d(J. (4.25)

Then (4.22) has a solution u E C ([a, b] ; Rn) with lul oo ~ R.

Proof. Let u E C ([a, b] ; B) be any solution of (4.21) for some A E

(0,1). Here
h(t,s,z) = K(t,s)f(s,z).
Then

lu (t)1 ~A it IK (t, s) f (s, u (s))1 ds ~A it <P (s) 1/J (Iu (s)l) ds (4.26)

for all t E [a, b]. Here we have understand that 1/J (0) = lim 1/J (t). Let
t!O

c (t) = min { R, A it <p (s) 1/J (Iu (s)1) dS} .

Clearly c is nondecreasing. We claim that c (b) < R. Assume the contrary.
Then, since c(a) = 0, there exists a subinterval [a',b'] c [a,b] with

c (a') = 0, c (b') = Rand c (t) E (0, R) for t E (a', b') .

Since by (4.26),
lu(t)1 ~ c(t) ~ R on [a,b] ,
and 1/J is nondecreasing on [0, R], we have

c' (s) = A <p (s) 1/J (Iu (s)l) ~ A <p (s)-1/J (c (s))
52 Chapter 4

for all s E [a', b']. Now integration from a' to b' yields

f b' _c_'('-'s)--ds rR 1
a' 'IjJ (c (s)) Jo 'IjJ (cr) dcr

< A 1, b'
¢ (s) ds

< A lb ¢ (s) ds

< lb ¢ (s) ds,

a contradiction. Notice we may assume 1¢1£1[a,bj > since otherwise we

have nothing to prove. Hence c(b) < R and so, by (4.26), lu (t)1 < R for
°
all t E [a, b]. Therefore lul oo < R and Theorem 4.4 applies. •

Remark 4.2 Notice Corollary 4.2 can be directly derived from Schauder's
fixed point theorem if we observe that T (D) c D, where
D {u E C ([a, b]; Rn) : lu (t)1 ::; I' (t) on [a, b]},
I' (t) [-1 (it ¢ (s) dS) ,
[(t)
rt 1
Jo 'IjJ (cr) dcr,
and T is the Volterra integral operator associated to the right hand side of
the equation (4.22). We note that I' (t) ::; R for all t E [a, b] because of
(4.25) .

Corollary 4.2 yields a global existence result for the initial value problem
u' (t) = f (t, u (t)) , t E [0, t1],
{ (4.27)
u (0) = 0.
Theorem 4.5 Let f E C ([0, t1] X Rn; Rn). Assume that there exists a
continuous nondecreasing function 'IjJ : (0,00) ----t (0,00) such that
If (s, z)1 ::; 'IjJ (Izi)
for all s E [0, t1] , z ERn, and
roo 1
t1 < Jo 'IjJ(cr)dcr.

Then (4.27) has a solution u E C 1 ([0, t1] ; R n) .

The Leray-Schauder Principle and Applications 53

°
Proof. Problem (4.27) is equivalent to (4.22). Here [a, bl = [0, tIl and
/'\, == 1. Choose any R >such that

[R 1
tl ::; Jo 'IjJ ((J") d(J".

Then (4.25) holds and Corollary 4.2 applies. _

4.4 The Cauchy Problem for an Integral Equation

with Delay
In this section we turn back to the initial value problem (3.9). The technique
we use here is based upon the Leray-Schauder principle. In addition we seek
positive solutions u with u (t) ~ a for all t E [0, tIl, where a E R+. Here,
for z, z' ERn, by Z ~ z' we mean Zi ~ z~, i = 1,2, ... ,n. Also Z ~ a
stands for Zi ~ a, i = 1,2, ... , n.
Our assumptions are as follows:
(i) fEe ([-T, tIl x [a, 00 )n; R+) ;
(ii) <p E C ([-T, 01; Rn), satisfies (3.10) and <p (t) ~ a on [-T, 01 ;
(iii) there exists a function gEe ([-T, tIl; Rn) such that

f(t,z)~g(t)

for all t E [-T, tIl, Z E [a, oo)n, and

l~T g (s) ds ~ a
for all t E [0, tIl;
(iv) there exists a continuous nondecreasing function 'IjJ (a, 00) --t

(0,00) such that

If (t, z)1 ::; 'IjJ (Izl)
for all t E [0, tIl, z E [a,oo)n, and
[00 1
tl < Jb 'IjJ((J")d(J",

where
b= I: If (s, <p (s))1 ds.
54 Chapter 4

Theorem 4.6 Assume that the assumptions (i)-(iv) are satisfied. Then
(3.9) has at least one solution u E C ([-r, tIl; Rn) with u (t) ~ a for all
t E [-r, tIl.

Proof. We first note that b ~ Icp (0)1. Indeed,

b= 1: If (s, cp (s))1 ds ~ 11: f (s, cp (s)) dsl = Icp (0)1·

We apply Theorem 4.1 with X = C ([0, tIl; Rn),

K = {u EX: u (0) = cp (0), u (t) ~ a on [0, tIl}

and T: K --t K given by

T (u) (t) = l~T f (s, u (s)) ds (t E [0, tIl) ,

where
u(t) = { cp (t), t E [-r,Ol
u (t), t E [0, tIl.
According to (i)-(iii), the operator T is well defined. In addition, T is
completely continuous (see Theorem 3.7). In what follows we shall establish
the boundedness of all solutions to

u = (1- 'x)uo + ,XT(u) (4.28)

for ,X E (0,1), where Uo is the constant function cp (0). To this end, let u
be any solution of (4.28) for some ,X E (0,1). Then

lu (t)1 ::; (1 -,X) b +,X l~T If (s, u (s))1 ds

Let c: [0, tIl --t R be given by

c(t) = (1-'x)b+'x l~Tlf(s,u(s))lds.

Notice c (0) = b. Furthermore, for t E [0, tIl one has

c'(t) = ,X(lf(t,u(t))I-lf(t-r,u(t-r))I)
::; 'xlf(t,u(t))1
< 'x'ljJ (Iu (t)1) ,
The Leray-Schauder Principle and Applications 55

and, since lu (t)l:s c(t) and 'IjJ is nondecreasing,

c' (t) :S )...'IjJ (c (t)).

Therefore
c' (s)
'IjJ (c(s)) :S)... < l.
Integration from 0 to t yields

re(t) I rt c' (s)

Jb 'IjJ(a)da= Jo 'IjJ(c(s))ds<t l.

Hence c (t) < R, where R is any fixed number satisfying

tl :S
rR 'IjJ (a)
Jb
I
da.

Therefore lu (t)1 < R for all t E [0, tl] . •

4.5 Periodic Solutions of an Integral Equation with

Delay
In this section we establish the existence of periodic solutions of a given
period w > 0 for the integral equation in R n

u (t) = l~T f (s, u (s )) ds. (4.29)

The ideas in this section were adapted from the paper Precup [46].
Our assumptions are:
(hI) fEe (R x R+;R+);
(h2) f(t+w,z) = f(t,z) for every t E R, z E R+;
(h3) there exists a number a > 0 and an w-periodic function 9 E
C (R; Rn) such that
f(t,z)2:g(t)
for all t E R, z E [a,oo)n, and

l~T 9 (s) ds 2: a
for all t E R;
56 Chapter 4

(h4) there exist two numbers b, R with

avn < b < R,

and a function 'IjJ E C ([a, R]; R+) with 'IjJ (t) > ° on [b, R], such that

If (t, z)1 ::; 'IjJ (Izl)

for all t E Rand z E [a, oo)n with Izl ::; R,

w::;
rR 'IjJ 1(0-) dO"
Jb (4.30)

and
b
If(t,z)l::; - (4.31)
T

for all t E Rand z E [a, oo)n satisfying b ::; Izl ::; R.

Theorem 4.7 Assume that (hl)-(h4) are satisfied. Then (4.29) has an w-
periodic continuous solution u such that

u (t) ~ a on [0, w] ,

min lu(t)1 < b and max lu(t)l::; R.

~~~ ~~~

Proof. We shall apply the Leray-Schauder principle. Here X is the

space of all continuous w-periodic functions u with the norm lul oo =
max lu (t)l,
tE[O,wj
K={UEX: u(t)~a on [O,w]}
and
u= {u E K: min lu (t)1 <
tE[O,wj
b, lul oo < R} .
Also, Uo is the constant vector-valued function (a, a, ... , a) simply denoted
by a, and T : K ~ K is given by

T (u) (t) = l~T f (s, u (s)) ds (t E R) .

It is easy to show that (h1)-(h3) guarantee that T is well defined and

completely continuous. We now claim that the Leray-Schauder boundary
The Leray-Schauder Principle and Applications 57

condition (4.1) is satisfied. Assume by contradiction that there are u E au

and A E (0,1) such that

u = (1 - A) uo+ AT (u) ,
that is
u(t)=(1-A)a+Al~Tf(S,u(S))dS, tER. (4.32)

Since u E au we have either

lul oo = R and min lu (t)1 < b, (4.33)

tE[O,w)

or
lul oo :::; R and min lu (t)1 = b. (4.34)
tE[O,w]

First assume (4.33). By differentiating (4.32) we obtain

u' (t) = Af (t,u (t)) - Af (t - T,U(t - T)).

Then

(u (t) , u' (t)) = A (u (t) , f (t, u (t) )) - A (u (t) , f (t - T, U (t - T))) .

Since all the components of the vectors u (t) and f (t - T, u (t - T)) are
nonnegative, we have that

(U(t),f(t-T,U(t-T))) 2 0.

Hence
(u (t) , u' (t)) :::; A (u (t) , f (t, u (t))) .
Furthermore, using (h4) we obtain

(u (t), u' (t)) :::; A lu (t)llf (t, u (t))1 :::; A lu (t)1 'lj; (Iu (t)I)·

Since
lui' = (R)' = (u,u')
lui
we obtain
lu (t)I' :::; A'lj; (Iu (t)1) . (4.35)
Let to E [0, w] be such that

lu (to) I = tE[O,w]
min lu (t) I
58 Chapter 4

and let t1 E (to, to+w] be such that lu (t1)1 = R. From (4.35), by integration
from to to t1, we obtain

1. lU (t dl

lu(to)1
1
~() dCT ~ ). (t1 - to)
'f/ CT
< w.

This, according to (4.30), is impossible since lu (to)1 < band lu (t1)1 = R.

Thus (4.33) can not hold.
Now assume (4.34). Let to E [0, w] be such that

lu (to)1 = tE[O,wj
min lu (t)1 = b.

Then from (4.32) and (4.31) we obtain

b= lu (to)1 ~ (1 - ).) avn + )'b < b,

a contradiction.
Thus the Leray-Schauder boundary condition holds and Theorem 4.1
applies . •

Remark 4.3 Assume (h1)-(h3) hold and that instead of (h4) , the following
condition is satisfied:
(h4*) If(t,z)1 ~ ~ for all t E [O,w] and z E [a,oo)n with Izl ~ R.
Then T (KR) c KR, where KR = {u E K: lu (t)1 ~ R on [O,w]}, and
the existence of a continuous w-periodic solution to (4.29) is guaranteed by
Schauder's fixed point theorem.

Example 4.1 Let n = 1, T =W = 1 and let f (t, z) = 1/;1 (z) (z E R+),

where
5z, zE [0,1]
-4z + 9, z E [1,2]
1/;r(z) = < 1, zE [2,3]
3z-8, zE [3,5]
z+ 2, zE [5,00).
The assumptions (h1)-(h4) are satisfied with a = 1, b = 2, R = 3, 9 (t) = 1
°
and 1/; (CT) = 1/;1 (CT) • However, for any R > there is no a < R such that
(h1)-(h3) and (h4*) are satisfied.

Remark 4.4 For a given function f satisfying (h1)-(h2) there could exist
several intervals [a, R] such that (h3) and (h4) hold. If these intervals are
disjoint, then the corresponding solutions guaranteed by Theorem 4.7 are
distinct.
The Leray-Schauder Principle and Applications 59

Example 4.2 Let n, T and 'lj; be as in Example 4.1, and let

f(t,z) = gl (t)'lj;k (z), t E R, z E [4(k -1) ,4k],

k = 1,2, ... , where

'lj;k (z) = 4 (k - 1) + 'lj;d z - 4 (k - 1))

and gl is any nonnegative continuous function with a period w > 0, such

i
that
t gl (s)ds::::: I, t E [O,w].
t-l

It is easy to see that if

w max gl (t) :s; (4 (k - 1) + 1)-1 ,

tE[O,w]

then all the assumptions of Theorem 4.7 are satisfied for

a=4(k-1)+l, b=4(k-1)+2, R=4(k-1)+3,

9 (t) (4(k -1) + l)gl (t),

'lj; (z) 'lj;k (z) max gl (t) .
tE[O,w]

Therefore for each kEN \ {O} satisfying

4 (k - 1) + 1 :s; (w max gl (t) )

tE[O,w]
-1 ,

the equation (4.29) has at least one continuous w-periodic solution Uk such
that
4 (k - 1) + 1 :s; inf udt) < 4 (k - 1) + 2
tER

and
sup udt) :s; 4 (k - 1) + 3.
tER

In particular, when gl (t) == 1 such solutions are the constant functions

9
udt) = 4 (k - 1) + "5
for k = 1,2, ....
60 Chapter 4

Other results on the integral equation (4.29) can be found in Precup-

Kirr [47] and Trif [50]. The same equation in a general Banach space is
studied in Guo- Lakshmikantham- Liu [27].
For other applications of the Leray-Schauder principle to integral, dif-
ferential, integra-differential and partial differential equations we refer the
reader to Bobisud [5], Constantin [13]' Frigon [21], Gilbarg-Trudinger [22],
Granas-Guenther-Lee [24], [25], Guenther-Lee [26]' Gupta [28], Lan-Webb
[31]' Mawhin [33], Ntouyas-Tsamatos [34], O'Regan-Meehan [35]' Pach-
patte [40] and Precup [45].
 Chapter 5

Existence Theory
in LP Spaces

In this chapter we present three examples of continuous operators acting

in LP spaces, namely: the Nemytskii superposition operator; the Fredholm
linear integral operator; and the Hammerstein nonlinear integral operator.
As applications we shall prove via the Leray-Schauder principle several ex-
istence results in LP for Hammerstein and Volterra-Hammerstein integral
equations in Rn. We show that these results immediately yield existence
theorems of weak solutions (in Sobolev spaces) to the initial value and two-
point boundary value problems for ordinary differential equations in Rn,
under some more general conditions than the continuity. Notice the weak
solutions are functions which satisfy the differential equations almost every-
where (a.e., that is, except a set of measure zero).
In addition we show that the two cases of continuous solutions and re-
spectively, of LP solutions (1 :s; p < 00), can be treated together by consid-
ering 1 :s; p :s; 00.

5.1 The N emytskii Operator

Let n c RN be an open set and f : n x R m ~ Rn be a given function.
The Nemytskii operator N f associated to f assigns to each function u :
n ~ Rm, the function Nf (u) : n ~ Rn, defined by
N f (u) (x) = f (x, u (x)) (x E n). (5.1)
Suitable conditions on f guarantee that N f has desired properties. Here we
present such conditions in order that N f maps LP (n; Rm) into Lq (n; Rn) .

61
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
62 Chapter 5

Definition 5.1 One says that a function f : 0 x Rm ~ Rn satisfies the

Caratheodory conditions if:
(i) f(.,y): 0 ~ Rn is measurable for every y E Rm;
(ii) f (x,.) : Rm ~ Rn is continuous for a.e. x E O.

Lemma 5.1 If f satisfies the Caratheodory conditions then Nf maps mea-

surable functions into measurable functions.

Proof. Let U : 0 ~ Rm be measurable. Then, there exists a sequence

of finitely-valued functions (Uk) with

Uk (x) ~ u(x) as k ~ 00 for a.e. x E O.

This together with (ii) implies

Nf (Uk) (x) = f (x, Uk (x)) ~ f (x, U (x)) = N f (U) (x) (5.2)

as k ~ 00 for a.e. x E O. On the other hand, Uk being finitely-valued

admits a representation as
Pk

Uk (x) = L Xkj (x) Ykj

j=l

where Ykj E Rm and Xkj is the characteristic function of some subset

Okj cO, with

U
Pk

0= Okj, Okj n Oki = 0 for j -1= i.

j=l

It follows that
Nf (Uk) lOki = f (., Ykj)
which, by (i), is a measurable function on Okj. As a result, Nf (Uk) is
measurable on O. Now according to (5.2) the function N f (u) is measurable,
as limit of a sequence of measurable functions. •

Definition 5.2 Let p E [l,ooJ and q E [1,00). A function f : 0 x Rm ~

Rn is said to be (p, q)-Caratheodory if the following condition is satisfied:

(a) if 1 ::; p < 00 then If (x, z)1 ::; 9 (x) + c Izi P/ q

{ for a.e. x E 0, all z E Rmand some 9 E Lq (0; R+), C E R+; (5.3)
(b) if p = 00 then for every R > Othere is a 9R E Lq (0) with
If (x, z)1 ::; 9R (x) for a.e. x E 0 and all z E Rmwith Izl::; R.
Existence Theory in £P Spaces 63

Theorem 5.1 Let p E [1,00] and q E [1,00). Assume that the function
f : n x R m --+ Rn is (p, q)-Caratheodory. Then the Nemytskii operator
N f : £P (n; Rm) --+ Lq (n; Rn) associated to f, given by (5.1) is well defined,
continuous and satisfies

(a) for 1:S; p < 00: INf (u)ILq(O;Rn):S; IgILq(O) + clul%'(O;R"')

{ for all u E £P (n; Rm) ; (5.4)
(b) for p = 00: INf (u)ILq(O;R"') :s; 19Rb(o)
for all U E LOO (n; Rm) with Iuloo :s; R and every R > o.

Here g, gR and c are those from Definition 5.2.

Proof. Step 1: Nf maps £P (n; Rm) into Lq (n; Rn) and satisfies
(5.4). To prove this, let us consider any function u E LP (n; Rm). By the
definition of the space £P (n; Rm), u is measurable. Lemma 5.1 implies
that N f (u) is measurable too. Furthermore, from (5.3), by means of the
norm inequality, we obtain for 1 :s; p, q < 00:

( lof INf (u) (xW dx )l/q :s; {flo (Ig (x)1 + clu (x)I P/ q)
q
dx
}l/q

< IgILq(O) + clul%'(O;R"') .

The case p = 00 is left to the reader.
Step 2: N f is continuous from £P (n; Rm) to Lq (n; Rn). The proof is
based on the well known Vitali's theorem (for a proof, see Dunford-Schwartz
[20], or Pascali-Sburlan [42]):

Lemma 5.2 (Vitali) Let (Uk) be a sequence of functions Uk E £P (n; Rn)

(1 :s; p < 00) such that Uk (x) --+ u (x) as k --+ 00 for a.e. x E n. Then
u E £P (n; Rn) and Uk --+ U in £P (n; Rn) as k --+ 00 if and only if:
(i) for each E > 0, there exists a 15 > 0 such that

L IUk (xWdx <E

for all k and every measurable subset A c n with f-l (A) < 15;
(ii) for each E > 0, there exists a subset Den such that f-l (D) < 00
and
f IUk(XWdx<E
low
for all k.
64 Chapter 5

Notice condition (ii) in Vitali's theorem is redundant if 0 is bounded.

Let (Uk) be any sequence of functions Uk E LP (0; Rm) such that

Uk . . . . . U

III £P (0; Rm) as k ......... 00. Without loss of generality we may assume that
Uk (x) ......... U (x) for a.e. x E O. Clearly

f (x, udx)) ......... f (x, U (x)),

i.e.,
N f (Uk) (x) . . . . . N f (U) (x) ,
for a.e. x E O. Now, for 1 ~ p < 00, conditions (i) and (ii) in Vitali's
theorem hold for (Uk) (the Necessity Part). Then, from (5.4), we can see
that conditions (i), (ii) in Lemma 5.2, with q instead of p, also holds for
(Nf (Uk)). Thus from Vitali's theorem (the Sufficiency Part) we obtain that

N f (Uk) . . . . . N f (u)

in Lq (0; Rn). Hence N f is continuous. •

Remark 5.1 The inequality in (5.4) shows that N f is a bounded operator

from £P (0; Rm) to Lq (0; Rn) .

5.2 The Fredholm Linear Integral Operator

Let X be a Banach space and 0 C RN an open set. A function U : 0 ......... X
is said to be finitely-valued if it is constant on each of a finite number of
disjoint measurable sets OJ C 0 with f-L (OJ) < 00 and equals zero on
o \ UOJ. A function U : 0 ......... X is said to strongly measurable if there
j
exists a sequence of finitely-valued functions convergent in X to U (x) for
a.e. x E O. For any p E [1,00], we let £P (0; X) be the set of all strongly
measurable functions U : 0 . . . . . X with Iulx E £P (0).
For strongly measurable functions Egorov's theorem is still true.

Lemma 5.3 (Egorov) Let X be a Banach space and 0 C RN be bounded

open. If (Uk) is a sequence of strongly measurable functions from 0 to X,
that strongly converges in X a. e. on 0 to a strongly measurable function
U : 0 . . . . . X, then for each E > 0, there exists a subset Of c 0 such that
f-L (0 \ Of) < E and on Of the convergence of Uk (x) to U (x) is uniform.
Existence Theory in £P Spaces 65

Proof. First we note that if v : 0 --t X is strongly measurable then

Ivlx is measurable. Indeed, from the definition of a strongly measurable
function it follows that there is a sequence (Vk) of finitely-valued functions
such that Vk (x) --t v (x) a.e. on O. Let

L Xki (x) Vki,

mk

Vk (x) =
i=l
where Vki E X and Xki is the characteristic function of a measurable subset
mk
Oki cO, with fl kj n Oki = 0 for j -I- i and 0 = U Oki. Since
i=l
mk

IVk (x)lx = L Xki (x) IVkilx,

i=l
we have
mk

Iv (x)lx - L xkdx) IVkilx IIV (x)lx -IVk (x)lxl

i=l
< Iv (x) - Vk (x)lx --t 0 as k --t 00.

Hence Ivlx is the limit of a sequence of measurable real functions. Thus,

Ivlx is measurable.
Of course we may assume that Uk (x) --t U (x) for every x E O. By the
above remark, since U - Uk is strongly measurable, Iu - Uk Ix is measurable
and so the set {x EO: Iu (x) - Uk (x)lx < E} is measurable. As a result
the set
OJ=
00

n
k=j+1
{xEO: lu(x)-udx)lx<E}

is measurable too. In addition OJ C Oi if j < i, and since Uk (x) --t u (x)

00

on 0, one has 0 = U OJ. Now

j=l

JL (0) JL (0 1 U (0 2 \ 0 1 ) U (03 \ O2 ) U ... )

JL (0 1 ) + JL (0 2 \ fl 1 ) + JL (0 3 \ O2 ) + ...
lim JL (OJ) .
J->OO

Hence lim JL (0 \ OJ) = 0, and therefore for each T/ > 0 there is a jo such
J->OO
that
JL(O\Oj) < T/, j ~jo.
66 Chapter 5

Thus for any positive integer k there exist a set O~ C 0 and an index Nk
such that

It (OD ~ ;k;
lu (x) - Uj (x)lx < 21k for j > Nk, x E 0 \ O~.
00

Let 0' = 0 \ U O~. Then

k=l

1
L It (O~) ~ L
00 00

It (0 \ 0') ~ € 2k =€
k=l k=l

and Uj (x) converges uniformly on 0'. •

Using Egorov's theorem we immediately obtain that if 0 c RN is
bounded open then for each U E V (0; X) one has

11Th (u) - ulxb(!1h) ---+ 0 as h ---+ O. (5.5)

Here Th (u) is the translation of U by h; also recall Oh = n n (0 - h).

For more information about the theory of functions with values in a
general Banach space see Guo-Lakshmikantham-Liu [27J and Yosida [52J.
Let /'i, : 0 2 ---+ R. We say that /'i, E V (0; £T (0)) if /'i, (x,.) E LT (0) for
a.e. x E 0, and the map x I----? /'i, (x,.) E LT (0) belongs to V (0; £T (0)) .
The notation /'i, E C (0; £T (0)) has a similar meaning.

Theorem 5.2 Let 0 C RN be bounded open, p, q E [1,00], r E [1,ooJ be

the conjugate of q, that is 1/q + l/r = 1, and let /'i, : 0 2 ---+ R. Assume

(a) if 1 ~ p < 00 then /'i, E V (0; LT (0));

(5.6)
(b) if p= 00 then /'i, E C (O;LT (0)).

Then the Fredholm linear integral operator A : Lq (0; Rn) ---+ V (0; Rn)
with kernel /'i, given by

A (u)(x) = 10 /'i, (x, y) U (y) dy (x E 0)

is well defined and completely continuous. Moreover, for p = 00 we have

A (U (0; Rn)) C C (0; Rn).

Existence Theory in V Spaces 67

Proof. 1) If u E Lq (0; Rn) then A (u) is a measurable function. We

omit the details.
2) A is well defined and bounded (hence continuous). Indeed, if u E
Lq (0; Rn) then by Holder's inequality we have

IA(u) (x)1 ~ (inll>:(X,YWdy)l/r (inIU(yWdy)l/q.

Hence

( in IA(u) (xWdx )
l/p
~ lulq
{(
in in II>: (x,y)l r dy
)p/r
dx
}l/P
,

that is
IA (u)lp ~ III>: (x, ·)Irlp lul q ·
This shows that A is well defined and bounded. Of course, some changes
have to be made in the proof if r = 00, q = 00, or p = 00.
3) Let p = 00 and let u E Lq (0; Rn). We show that A (u) E C (0; Rn) .
Indeed, for every x, x' EO, we have

IA (u) (x) - A (u) (x') I ~ in II>: (x, y) - I>: (x', y) Ilu (y)1 dy
< lul q II>: (x,.) - I>: (x',.) Ir ----> 0
as Ix -
x'I ----> O. Hence A (u) is a continuous function.
4) A is completely continuous. To prove this, let X c Lq (0; Rn) be a
bounded set. Assume that

lul q~ m for all u E X.

We have to show that Y := A (X) is relatively compact in LP (0; Rn) . For

this we use Theorem 1.4. Firstly,

IA (u) (x)1 ~ in II>: (x, y)llu (y)1 dy ~ mIk (x, ·)Ir =: v (x)
for a.e. x E 0, and v E LP (0), so (1.4) holds. Secondly,

ITh (A (u)) (x) - A (u) (x)1 ~ in IT~ (I>:) (x, y) - I>: (x, y)llu (y)1 dy
< mIT~ (I>:)(x, .) - I>: (x,.) Ir .
68 Chapter 5

Here T~ (/1:) is the translation of /1: by h with respect to the first variable,
i.e.,
T~(/1:)(X,y)= {/1:(X+h,y) if x+hEO,
o if x + h tf- O.
It follows that

ITh (A (u)) - A (u)ILP(rlh;Rn) :::; m IIT~ (/1:) (x,.) - /1: (x, .)lrILP(rlh)'

But from (5.5) we have

IIT~ (/1:) (x,.) - /1: (x,.) Ir ILP(rlh) ----+ 0 as h ----+ O.

Hence (1.5) also holds. Consequently Y is relatively compact in LP (0; Rn) .

•
Remark 5.2 Notice if /1: : 02 ----+ R is measurable, /1: (x,.) E U (0) for
a.e. x E 0, and the map x f-------t 1/1: (x, .)Ir belongs to LP (0), then the map
x f-------t /1: (x,.) E U (0) is strongly measurable, and so /1: E LP (0; U (0)).

5.3 The Hammerstein Integral Operator

Consider the Hammerstein integral operator given by

T (u) (x) = L /1: (x, y) f (y, u (y)) dy (x E 0) . (5.7)

This operator appears as the composition of the Fredholm linear integral

operator A of kernel /1: with the Nemytskii operator N f associated to f,
that is
T = ANf .

Theorem 5.3 Let 0 C RN be open, /1: : 02 ----+ Rand f : 0 X Rn ----+

Rn. Let p E [1,00]' q E [1,00) and let r E (1,00] be the conjugate of
q. Assume that the Fredholm linear integral operator A : Lq (0; Rn) ----+
LP (0; Rn) of kernel /1: is well defined and completely continuous. In addition
assume that f is a (p, q)-Caratheodory function. Then the Hammerstein
integral operator T: LP (0; Rn) ----+ LP (0; Rn) given by (5.7) is well defined
and completely continuous.

Proof. From Theorem 5.1, the Nemytskii operator N f is well defined,

bounded and continuous from LP (0; Rn) into Lq (0; Rn) . Now the conclu-
sion follows from Theorem 2.1, since T = ANf . •
Existence Theory in V Spaces 69

Remark 5.3 According to Theorem 5.2, if 0 is bounded and (5.6) holds

for "', then A is well defined and completely continuous; moreover, in this
case, for p = 00,
T (L OO (0; Rn)) c C ('0; Rn) ,
and so any fixed point of T belongs to C (0; Rn) .

5.4 Hammerstein Integral Equations

This section presents a very general existence principle for both LP solutions
and continuous solutions to the Hammerstein integral equation in R n

u(x) = 10 ",(x,y)f(y,u(y))dy a.e. on O. (5.8)

Here 0 C RN is an open set. For a given p E [1,00], we seek a weak

solution, that is a function u E V (0; Rn) which satisfies (5.8) for almost
every x E O. In addition, we look for solutions in U, where U is a bounded
open set of V (0; Rn) containing the null function.

Theorem 5.4 Let 0 C RN be open, "': 0 2 -> Rand f : 0 x R n -> Rn.

Assume that there exist p E [1,00]' q E [1, (0) such that the Fredholm
linear integral operator A : Lq (0; Rn) -> V (0; Rn) with kernel", is well
defined and completely continuous and f is (p, q)-Caratheodory. In addition
assume that there exists a bounded open set U C V (0; Rn) containing the
null function, such that
uEU (5.9)
for any solution u E U to

u(x) =A 10 ",(x,y)f(y,u(y))dy a.e. on 0, (5.10)

for each A E (0,1). Then (5.8) has a solution in V (0; Rn) with u E U.

Proof. Apply Theorem 4.1 to K = X = V (0; Rn) with norm 1.lp ' Uo
the null function, and T: U -> LP (0; Rn) given by

T (u)(x) = 10 '" (x, y) f (y, u (y)) dy (x E 0) .

This operator is completely cor..tinuous by Theorem 5.3. •

70 Chapter 5

Remark 5.4 If 0 is bounded and (5.6) holds for 1'\" then the Fredholm lin-
ear integral operator A of kernel I'\, is well defined and completely continuous
from Lq (0; Rn) to £P (0; Rn) ; in this case, for p = 00 the solutions of (5.8)
are continuous functions on O.

Corollary 5.1 Let 0 C RN be a bounded open set, I'\, : 0 2 -----> Rand

f: 0 x Rn -----> Rn. Assume that there exists p E [1,00]' q E [1,(0) such that
I'\, E LP (0; U (0)) (1/r + l/q = 1), f is (p, q)-Caratheodory and

{
(a) ifl~p<oo then (lglq+cRP/q)lll'\,(x,.)lrlp~R; (5.11)
(b) if p = 00 then 19R1q III'\, (x, .)Irloo ~ R.

Here g, gR and c are those from Definition 5.2. In addition, if p =

00 assume that the Fredholm linear integral operator A : Lq (0; Rn) ----->
LOO (0; Rn) with kernel I'\, is completely continuous. Then (5.8) has a solu-
tion u E £P (0; Rn) with lul p ~ R.

Proof. Let
u = {u E LP (0; Rn) : lulp < R}
and B = U. Let u E B be any solution of (5.10) for some ,\ E (0,1). Then

lu (x)1 ~ ,\ LII'\, (x, y)llf (y, u (Y))1 dy

< ,\ LII'\, (x, y)1 (g (y) + clu (y)I P/ q) dy

< ,\ (inr II'\, (x, yW dy) l/r ( inr (g (y) + clu (y)I P/ q) q dy) l/q
< ,\ (Igl q + clul~/q) II'\, (x, ·)Ir·

It follows that
lulp ~ ,\ (Igl q + clul~/q) III'\, (x, ·)Irlp· (5.12)

Since lul p ~ Rand ,\ < 1, (5.11) and (5.12) guarantee lul p < R, i.e., (5.9).
Similar estimations can be obtained if p = 00 . •
We note that in fact (5.11) implies T (B) c B. So Corollary 5.1 can be
directly derived from Schauder's fixed point theorem.
As in Section 4.2, a better result holds when I'\, is the Green's function
of a certain differential operator, with respect to some boundary conditions.
Existence Theory in V Spaces 71

For instance, for the equation in Rn

u(x) = 11 "'(x,y)f(y,u(y))dy a.e. on (0,1), (5.13)

the following existence theorem is true.

Theorem 5.5 Let '" be the Green's function of the differential operator
-u" + u with respect to one set of the homogeneous Dirichlet, Neumann, or
periodic boundary conditions

u (0) ° (Dirichlet),
= u (1) =
u' (0) u' (1) ° (Neumann),
= =

u (0) - u (1) u' (0) - u' (1) ° (periodic conditions) .

= =

Assume that f : (0,1) x Rn ----) R n is a

In addition, assume that there exists R > ° such that for
q)-Caratheodory function for
(00,
some q [1,
E 00 ).
every z E Rn with Izl = R, one has

(z, f (y, z)) :; R2 for a.e. y E (0,1).

Then (5.13) has a solution u E w 2,q (0, 1; Rn) with lul oo :; R.

For any real number 1 :; q :; 00 the Sobolev spaces wm,q (a, b; Rn)
(m E N\ {o}) are inductively defined as follows. A function u belongs
to w 1 ,q (a, b; Rn) if it is continuous on [a, b] and there exists a function
v E Lq (a, b; Rn) such that

u(t) = u(a)+ ltV(S)ds, tE [a,b].

It is clear that if u E w 1 ,q (a, b; Rn) then u is absolutely continuous on

[a, b] (note that any absolutely continuous function u : [a, b] ----) Rn is
differentiable almost everywhere on [a, b] and u' E £1 (a, b; Rn)) , u' E
Lq (a " b' Rn) , and

u(t) = u(a)+ ltu'(s)ds, tE [a,b].

Furthermore, for any integer m > 1

u E wm,q (a , b', Rn) if u , u' E w m- 1 ,q (a "

b· Rn)
.
72 Chapter 5

If u E Wm,q (a, b; Rn), then u E C m- 1 ([a, b]; Rn), u(m-l) is absolutely

continuous on [a, b] ,

u(m) : = (u(m-l»)' E Lq (a, b; Rn) ,

and
u(m-l) (t) = u(m-l) (a) + i t u(m) (s) ds, t E [a, b].

The space wm,q (a, b; Rn) is a Banach space with the norm

lulm,q = max {luU)lq; j = 0,1, ... , m}.

For more information about the theory of Sobolev spaces we refer the reader
to Brezis [6].

5.5 Volterra-Hammerstein Integral Equations

We conclude this section by some results on equation (5.8) in the Volterra
case. More exactly, we consider the Volterra-Hammerstein equation in Rn

u(t)= itK,(t,S)f(s,U(S))dS a.e. on (a,b). (5.14)

This equation can viewed as a special case of equation (5.8), where n = (a, b)
and
K,(t,s) = 0 for t < s.

The existence of solutions to (5.14) can be established without condition

(5.11).

Theorem 5.6 Let (a, b) be a bounded real interval, K, : (a, b)2 ---+ Rand
f: (a, b) x Rn ---+ Rn. Assume that there exists p E [1,00) and q E [p, 00) n
(1,00) such that K, E £P(a,b;Lr(a,b)) (1/q+1/r=1) and f is (p,q)-
Caratheodory. In addition assume that if q = p, then there exists an r' > r
such that
K, (t,.) E U' (a, b) for a.e. t E (a, b), and
the mapti-----'f 1K,(t,.)lr' belongs to £P(a,b).
Then (5.14) has a solution u E £P (a, b; Rn).
Existence Theory in V Spaces 73

Proof. First assume p < q. Then there exists R> 0 with

(lglq+cRP/q) 11K;(t,.)lrlp::; R
and the conclusion follows from Corollary 5.1.
Next, assume p = q > 1. We shall apply Theorem 5.4 to

u = {u E LP (a,b;Rn): Ilull < R},

any number R > 0 and a suitable equivalent norm 11.11 on V (a,b;Rn).
Let u E V (a, b; R n) be any solution of

u (t) = >-it K; (t, s) f (s, u (s)) ds a.e. on (a, b), (5.15)

for some >- E (0,1). For any e> 0 we have

Iu (t) I ::; >-it IK; (t, s) I (g (s) + cIu (s )I) ds

< >-itlK;(t,s)lee(s-a) (g(s)+clu(s)le-e(s-a»)ds.

Define an equivalent norm 11.11 on V (a, b; Rn), by

Ilull= (Ja (lu(s)le-e(s-a)Yds

b ) l/p

Since
1 r' - r 1
-+--+-=1
r' rr' p ,
Holder's inequality applied to three functions yields

b ) l/r' (Jt ) (r'-r)/(rr')

lu(t)1 ::; >- (Ja IK; (t, s)( ds a eerr'(s-a)/(r'-r)ds

X (Ja (g(s)+clu(s)le-B(s-a)Yds
b ) l/p

< >- (Igl p + cIlull) (r'err'- r) (r'-r)/(rr') eB(t-a) IK; (t, ·)Ir'·
It follows that
' ) (r'-r)/(rr')
Ilull ::; >- (Igl p + cIlull) ( re:r,r 11K; (t, ·)lr,lp·
74 Chapter 5

Now we choose 0 > ° such that

(Igl p + cR) (r'Orr'- r) (r'-r)/(rr')

11K: (t, .) Ir,lp ~ R

Then, since A < 1, Ilull < R and Theorem 5.4 applies. _

For p = 00 a better result can be established.

Theorem 5.7 Let (a, b) be a bounded real interval, K: : (a, b)2 --+ Rand
f : (a, b) x R n --+ R n. Assume that f satisfies the Caratheodory conditions,
there exist a, {3 E (1,00], 'Y E [1,00) with a' < {3 (l/a + l/a' = 1) and
1/a+1/{3+1h = 1, ¢ E try. (a, b), and a continuous nondecreasing function
'ljJ : (0,00) --+ (0,00) such that

K: E L 00 (a, b; L(3 (a, b)) ,

If (s, z)1 ~ ¢ (s) 'ljJ (lzl'Y) ,
for a. e. s E ( a, b) and all z ERn, and

[Ilk (t, .)1(3100 1¢la r (b - a) ~ 1000 U}_\1'Y dcr. (5.16)

In addition assume that the Fredholm linear integral operator A with ker-
nel K: is is well defined and completely continuous from L(3' (a, b; Rn) to
C([a,b] ;Rn) (1/{3+1/{3' = 1). Then (5.14) has a solution u E C([a,b] ;Rn).

Proof. We shall apply Theorem 5.4 with p = 00, any finite q satisfying
{3' < q ~ a (1/{3 + 1/{3' = 1), and

U = {u E L oo (a, b; Rn) : lul oo < R}

with a suitable radius R> 0. Let us show that such an R can be found so
that condition (5.9) holds.
Let u E Loo (a, b; Rn) be any solution of (5.15) for some A E (0,1).
Then u E C ([a, b]; Rn) since A has values in C ([a, b] ; Rn) . Using Holder's
inequality, we obtain

lu(t)1 ~ A lot 1K:(t,s)I¢(s)'ljJ(lu(s)I'Y)ds

< AIK:(t,·)I(3I¢la ( 1t
'ljJ(lu(s)I'Y)'Yds
)lh .
Existence Theory in I? Spaces 75

Hence
lu (t)I'Y ::; c(t) ,
where
c(t) = [A 11K; (t,.)l i3 loo 1¢lar (it ~(lu(sW)'Y dS).
From
c' (s) = [A 11K; (t, .)1 13 100 1¢lar ~ (Iu (sW)'Y
::; [AIIK;(t,.)li3lool¢lar~(c(s))'Y,
by integration from a to t, we obtain

ret)
Jo
1
~ (0")"1 dO"
i a
t c' (s)
~ (c (s ))'Y ds
< [AIIK;(t,.)l i3 lool¢lar(b-a)
< [11K;(t,.)l i3 lool¢lar(b-a).
This together with (5.16) shows that there exists an R> 0 independent of
A, with
c(t) < R'Y for all t E [a,bj.
Thus lu (t)1 < R on [a, bj . •
As a consequence we have the following global existence result for the
Cauchy problem
u' = f(t,u) a.e. on (a, b) ,
{ (5.17)
u (a) = O.

Theorem 5.8 Let (a, b) be a bounded real interval and f : (a, b) x Rn ---;
R n. Assume that f satisfies the Caratheodory conditions, there exists ex E
(1,00]' 'Y E [1,00) with l/ex + 1h = 1, ¢ E La (a,b), and a nondecreasing
continuous function ~ : (000) ---; (0,00) such that

If (s, z)1 ::; ¢(s) ~ (lzl'Y)

for a.e. s E (a, b) and all z ERn, and

I¢I~ (b - a) ::; 100 [~-(~-'-')j;N'YdO". (5.18)

Then (5.14) has a solution u E w1,a (a, b; Rn) .

76 Chapter 5

Proof. Apply Theorem 5.7 with j3 = 00 and Ii (t, 8) = 1 for 8 :S t,

Ii (t, 8) = 0 for t <
8. Notice the Fredholm linear integral operator A is now

it
given by
A (u) (t) = u (8) d8,

and is completely continuous from L1 (a, b; Rn) to C ([a, b] ; Rn) . •

The results in Sections 5.4-5.5 are presented here for the first time. We
have adapted here some ideas from the papers of O'Regan-Precup [36]' [38],
[39]. For similar results and extensions to equations and inclusions in Banach
spaces, see Couchouron-Precup [17].
For other methods, results and applications in nonlinear integral and
integro-differential equations, see the papers Appell-DePascale-Nguyeii-
Zabreiko [2], Appell-Vignoli-Zabreiko [3], Brezis-Browder [7], Browder [9],
Bugajewski-Szufla [10]' Burton [12], Couchouron-Kamenski [16], Papageor-
giou [41], Petru§el [43]' and the monographs Agarwal-O'Regan [1], Barbu
[4], Burton [11], Corduneanu [14], [15]' Deimling [18], Guo-Lakshmikantham-
Liu [27], Krasnoselskii [30]' Rus [48] and Viith [51].
 Part II

VARIATIONAL METHODS
 Chapter 6

Positive Self-Adjoint
Operators in Hilbert Spaces

In this part we present some variational methods with applications to the

existence of LP solutions of the Hammerstein integral equation in Rn

u(x) = in K,(x,y)f(y,u(y))dy a.e. on O. (6.1)

We show that this equation can be put under the variational form
E' (u) = o.
The construction of the functional E associated to equation (6.1) is based
on the theory of positive self-adjoint linear operators in Hilbert spaces.
The purpose of this introductory chapter is to present some definitions
and results of this theory that will be required in what follows. We define
the notions of adjoint operator, self-adjoint operator, positive operator, and
square root of a positive self-adjoint linear operator. As an application
we shall deal with bounded linear operators A : Lq (0; Rn) ----t LP (0; Rn) ,
where 1/p+1/q = 1. The goal is to split such an operator in the form HH*,
where H : L2 (0; Rn) ----t LP (0; Rn) and H* is the adjoint of H. When A
is the Fredholm linear integral operator of kernel K" such a representation
is useful in order to transfer the operator equation equivalent to (6.1) from
the Banach space LP (0; Rn) to the Hilbert space L2 (0; Rn) .

6.1 Adjoint Operators

Let X be a Hilbert space, Y a Banach space, and y* the dual of Y. We
shall denote by (.,.) both inner product of X and duality between Y and

85
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
86 Chapter 6

Y*, i.e., the bilinear functional on Y* x Y, defined as (v*, v) = v* (v) for

all v E Y and v* E Y*.

Proposition 6.1 For each bounded linear operator A : X -> Y there exists
a unique bounded linear operator A * : Y* -> X such that

(v*, A (u)) = (A* (v*), u) (6.2)

for all u E X and v* E Y*.

Proof. For any fixed v* E y* define a bounded linear functional f :

X -> R, by f (u) = (v*, A (u)). According to Riesz' representation theorem
there exists a unique wE X such that f (u) = (w, u) for all u E X. Define
A* (v*) = w. From (6.2), since A is linear and continuous, it immediately
follows that A * is a continuous linear operator. _

Definition 6.1 The operator A * given by Proposition 6.1 is called the

adjoint operator of A.

Definition 6.2 A bounded linear operator A : X -> X is said to be self-

adjoint if A = A *, i.e.,

(v, A (u)) = (A (v), u)

for all u, vEX.

Proposition 6.2 If A : X -> X is a self-adjoint operator, then

IAI=sup{I(A(u),u)l: lul=l}. (6.3)

Proof. By definition, the norm of the operator A is given by

IAI = sup {IA (u)1 : lui ~ I}.

For any u E X with lui ~ 1, we have

I(A (u) ,u)1 ~ IA (u)llul ~ IAI·

Then, if we denote 'Y = sup{I(A(u),u)l: lui = I}, we have 'Y ~ IAI. To

prove the converse inequality we first note that for every u E X one has

I(A (u), u)1 ~ 'Y lul 2 .

Positive Self-Adjoint Operators 87

Then since

(A (u + v) ,u + v) - (A (u - v) ,u - v) = 4 (A (u), v)
we obtain

I(A(u),v)l::; ~ (lu+vI2+lu-vI2) = ~ (luI 2+lvI2).

If we choose
lui
v = IA (u)I A (u) (for A (u) -I- 0),
we deduce that IA (u)1 ::; 'Y lui, which is also true if A (u) o. Hence
IAI ::; 'Y, and so IAI = 'Y. -
Definition 6.3 A linear operator A : X --) X is said to be positive (A 2 0)
if (A (u) , u) 2 0 for all u E X.

Proposition 6.3 If A : X --) X is a positive self-adjoint operator, then

I(A (u) ,v)1 2 ::; (A (u) ,u) (A (v), v)

for all u, VEX.

Proof. A being positive, one has (A (u + tv), u + tv) 20, that is

t 2 (A (v) , v) + t [( A (u) , v) + (A (v) , u) 1+ (A (u) , u) 2 0

for all t E R. Since A is self-adjoint this yields

t 2 (A (v) , v) + 2t (A (u) , v) + (A (u) , u) 2 0

for all t E R. Consequently

(A (u) ,v)2 - (A (u) ,u) (A (v) ,v) ::; 0

and the proof is complete. _

If A and B are two linear operators from X into X we say that
A ::; B if B - A is a positive operator. From (6.3) we deduce that if A, B
are self-adjoint then

o ::; A ::; B implies IAI::; IBI·

88 Chapter 6

Proposition 6.4 Let (Ak) be a sequence of positive self-adjoint operators

Ak : X ----> X and let B : X ----> X be a bounded linear operator. Assume
that
Ak :S Ak+1 :S B
for all kEN. Then for every u E X the sequence (Ak (u)) is convergent
to some element A (u), and the operator A : X ----> X defined in this way
is positive, self-adjoint, and satisfies

Ak :S A :S B, kEN.

Proof. For every kEN, we have

o :S (Ad u) , u) :S (Ak+ d u) , u) :S (B (u) , u) .

Hence the sequence of real numbers (Ak (u) ,u) is convergent. We shall
prove that (Ak (u)) is a Cauchy sequence. Indeed, for i < j one has
Aj - Ai 2: 0, and from Proposition 6.3 we obtain

I(Aj - A) (u)14 ((Aj - Ai) (u) , (Aj - Ai) (u))2

< ((Aj - Ai) (u) , u) ((Aj - A)2 (u), (Aj - Ai) (u))
< I(Aj - Ai) (u)12IAj - Ail [(Aj (u) ,u) - (Ai (u) ,u)].

From 0 :S Aj - Ai :S B, we have IAj - AI :S IBI. Then

IAj (u) - Ai (u)12 :S IBI [(Aj (u) , u) - (Ai (u), u)].

Hence the sequence (Ak (u)) is Cauchy and so convergent. The rest of the
proof is left to the reader. •

6.2 The Square Root of a Positive Self-Adjoint

Operator
Definition 6.4 Let A : X ----> X be a positive self-adjoint operator. A
positive self-adjoint operator B : X ----> X is said to be the square root of A
if B2 = A.

Proposition 6.5 For each positive self-adjoint operator A the square root
B exists and is unique. Moreover, if A is completely continuous B is
completely continuous too.
Positive Self-Adjoint Operators 89

Proof. 1) Existence. We may assume without loss of generality that

JAJ = 1. Then A :S 1, where 1 is the identity map of X. We define the
operators Bk : X ---) X for k 2': 2 by B2 == 0, and

Bk+ 1 = Bk + ~ (A - B~), (6.4)

for k = 2,3, .... We easily check that Bk is self-adjoint,

BjBi BiBj ,
1
1 - B k +1 - (1 - B k )
2
+ -1 (1 - A)
2 2
and
1
Bk+1 - Bk = "2 [(1 - Bk-d + (1 - Bk)] (Bk - Bk- 1 ) .
It follows that
Bk :S 1, Bk:S Bk+1
for all k E {2, 3, ... }. In particular, Bk 2': 0 since B2 = o. By Proposition
6.4, there exists a positive self-adjoint operator B with

B (u) = lim Bk (u), u E X.

k-+oo

Letting k ---) 00 in (6.4), we obtain

B=B+ ~ (A - B2),

whence B2 = A.
2) Uniqueness. First note if Bo is positive, self-adjoint, and B5 =
A, then ABo = BoA. By induction, using (6.4) we obtain that BkBO =
BoBk, whence letting k ---) 00, BBo = BoB. Now let u E X and v =
(B - Bo) (u). We have

(B (v) , v) + (Bo (v) , v) ((B + Bo) (v) , v)

((B + Bo) (B - Bo) (u) , v)
((B2 - Bg) (u) ,v)
o.
Since B, Bo 2': 0, we may infer that

(B (v) , v) = (Bo (v) , v) = O.

90 Chapter 6

On the other hand, since B is positive and self-adjoint there exists a positive
self-adjoint operator Bl with B = Br. From

IBl (v)12 = (Br (v) ,v) = (B (v) ,v) = 0,

it follows that Bl (v) 0 and so B (v) Bl (Bl (v)) O. Similarly

Bo (v) = O. Hence

IB (u) - Bo (u) 12 = (( B - Bo) 2 (u) , u) = (( B - Bo)( v) , u) = O.

Thus B (u) = Bo (u) for any u E X. Therefore B = Bo.

We shall denote by A 1/2 the square root of A.
3) Complete continuity. Assume A is completely continuous and let (Uk)
be any bounded sequence of elements of X. Then there exists a subsequence
(UkJ of (Uk) such that (A (UkJ) is convergent. Using the identity

IA1/2 (u) - A 1/ 2 (v) 12 = (u - v, A (u) - A (v)),

we deduce that the sequence (A 1/2 (Uk j )) is Cauchy, hence convergent.

Hence A 1/2 is completely continuous. •

6.3 Splitting of Linear Operators in I.J' Spaces

In this section we establish the possibility of representing certain linear op-
erators A: Lq (0; Rn) --t £P (0; Rn) (lip + 1/q = 1) in the form

A=HH*,

where H: L2 (0; Rn) --t £P (0; Rn) and H* is the adjoint of H.

The presentation in this section was adapted from the book of Kras-
noselskii [19].
Assume 0 C RN is a bounded open set. Recall that for 2 p < Po, :s:
one has

LPO (0; Rn) c £P (0; Rn) c L2 (0; Rn) c U (0; Rn) c LqO (0; Rn)

where q, qo are the conjugate exponents of p and Po, respectively.

Theorem 6.1 Let 2 < Po < 00, 11po + 1/qo = 1 and A : LqO (0; Rn) --t
LPO (0; Rn) a bounded linear operator. Assume that the restriction of A
to L2 (0; Rn), i.e., A : L2 (0; Rn) --t L2 (0; Rn) is a positive self-adjoint
Positive Self-Adjoint Operators 91

operator. Then for each p E [2,po) the operator A1/2 L2 (0; Rn) --+
L2 (0; Rn) satisfies

A 1/ 2 (L2 (0; Rn)) cV (0; Rn)

and the operator

H : L2 (0; Rn) --+ LP (0; Rn), H (u) = A1/2 (u) ,

is bounded (equivalently, continuous). If in addition, A : L2 (0; Rn) --+

L2 (0; Rn) is completely continuous then H is also completely continuous.

Proof. 1) Let u E L2 (0; Rn). Let us consider the sets

00 = {XEO: IA1/2(u)(x)l:Sl};

OJ = {x EO: 2j - 1 < IAI/2 (u) (x)1 :s 2j }

for j = 1,2, .... Let w j : 0 --+ Rn be given by

wI (x) = { ~gn (Al/2 (u) (x))i if x E OJ,

if x rf. OJ,

i = 1,2, ... ,n. We have

2j - 1 /-L(Oj) :s In (A1/2 (u) (x) ,wj (x)) dx

(Al/2 (u) , w j \

(u,A 1/2 (w j ))2

< lu1 2 1A1 / 2 (w j ) 12
. .) 1/2
lul 2 (A ( wJ ) ,wJ 2 .

Also

(A (w j ) ,w j )2:S IA (wj)lpo Iwjlqo:S IAlo Iwjl!o:S nIAlo/-L(Oj)2/qo.

Here IAlo stands for the norm of A as an operator from LqO (0; Rn) to
LPO (0; Rn). It follows that

2j - 1 /-L(Oj):S vinlub IAI~/2 /-L(Oj)1/ qO

92 Chapter 6

whence
. (
f..L(nj)5:c2-JPO, where c= 2Vnlul21Alol~)~ .

In
Then
IAl/2 (u) (x)I Pdx 5: c2 j (p-po)
J

(j = 1,2, ... ) . For j = 0, since p 2: 2 one has

r
Jn o
IAl/2 (u) (x)r dx < r
Jn o
IAl/2 (u) (x)1 2 dx

< In IAl/2 (u) (x)1 2 dx

(A (u), u)2
< IAllul~·
Consequently

In IAl/2 (u) (x)IP dx ~ 10j IA 1/2 (u) (x) IP dx

L 2j (p-po)
00

< IAllul~ + c
j=l

IAllul~ + c 2P- PO
1- :
IAllul~ + c
< 00.

Hence A l / 2 (u) E V' (0,; Rn) and

l/p
IH (u)lp 5: IAllul~ + 2Po-~ _
(
1
)
'

which shows that H is a bounded operator.

2) Assume A is completely continuous as operator from L2 (0,; Rn) to
L2 (0,; Rn). Then by Proposition 6.5 Al/2 is also completely continuous
from L2 (0,; Rn) to L2 (0,; Rn). The complete continuity of H now follows
from the next lemma.

Lemma 6.1 Let p > 2 and let Y c V' (0,; Rn) be relatively compact in
L2 (0,; Rn). If for each E> 0 there exists a bE: > 0 such that

llu (x)JP dx < EP (6.5)

Positive Self-Adjoint Operators 93

for all u E Y and D c 0 with IL (D) < be, then Y is relatively compact
in £P (0; Rn) .

Indeed, if Z c L2 (0; Rn) is any bounded set then Y = H (Z) c

£P (0; Rn), is relatively compact in L2 (0; Rn), and for every u E Z we
have

in IH(u) (xWdx = j Dn (nount u ... un jo )

IH (u)(xW dx

+ f
j=jo+1
j Dnnj
IH(u) (xWdx

L
00

< 2 joP IL (D) + CO 2j(p-po) ,

j=jo+1

where
CO = sup { (2y1n luI2IAI~/2rO : u E Z} .
For jo large enough we have
00
< -€p
L
j=jo+1
2 j (p-po)
- 2co

and setting be = €p 2- 1- jop , we see that Y satisfies (6.5). Hence by Lemma

6.1 H (Z) is relatively compact in £P (0; Rn) and the proof of Theorem
6.1 is complete .•
Proof of Lemma 6.1. By Hausdorff's theorem it sufficies to show
that for each € > 0 there exists in £P (0; Rn) a relatively compact €-
net for Y. To this end, for each a > 0, we consider the set of functions
Ya = {ua : u E Y}, where

u(x) if lu (x)1 < a,

{
Ua (x) = a lu (x)I- 1 u (x) if lu (x)1 2 a.

a) For each € > 0 there exists a sufficiently large a > 0 such that Ya
is an €-net in £P (0; Rn) for Y. Indeed, let be be such that (6.5) holds for
all u E Y and Dc 0 with IL(D) < be' For any u E Y we let

Ou = {x EO: lu (x)1 2 a}.

94 Chapter 6

Then, if f.l (Ou) < 6£, we have

Iu-ual p = (inJU(X)-IU~X)IU(X)IP dx)l/P (6.6)

< (inu Iu (x)IP dX) l/p

< c.
Let r be such that Iub::; r for all u E Y. Then

1/2 ( ) 1/2
af.l(Ou)1/2::; ( inu1u(x)12dX ) ::; in1u(x)12dX ::;r.

Hence f.l(Ou) ::; (r /a)2. Now choose a> 0 such that (r/a)2 < 6£. Then
f.l (Ou) < 6£, and so (6.6) holds.
b) Now we show that Ya is relatively compact in LP (0; Rn). Since Y
is assumed to be relatively compact in L2 (0; Rn), for a given c> 0 there
exists in L2(0;Rn) a finite c-netfor Y, say {u 1,u2, ... ,um }. We claim that
the set
1 , u2 m , Ua}
{ ua a, ...
is a (2a)1-2/ p c 2 / p -net in V (0; Rn) for Ya. Indeed,

IUa - u~lp 2a {in (2~ IUa - u~I)P dx riP

< 2a { in (21a Iu - ujl)
2
dx
}VP
< (2a)1-2/ p c 2 / p ,

for any j E {1, 2, ... , m} with lu - u j 12 < c. This completes the proof of
Lemma 6.1. •

Theorem 6.2 Under the assumptions of Theorem 6.1, for each p E [2,po),
the operator A : Lq (0; Rn) ----t LP (0; Rn) (l/p + l/q = 1) can be repre-
sented in the form A = HH*, where H: L2(0;Rn) ----t V (O;Rn) is a
bounded linear operator and H* is the adjoint of H.

Proof. Observe that the restriction of H* to L2 (0; Rn) coincides with

A1/2. Hence
HH*IL2(D.;Rn) = A1/2 Al/2 = AIL2(D.;Rn).
Positive Self-Adjoint Operators 95

Finally, use the density of L2 (0; Rn) in Lq (0; Rn) . •

Now let us return to the Hammerstein integral equation (6.1) and assume
that the the operators

Nj : LP (0; Rn) ----t Lq (0; R n), N j (u) (x) =f (x, u (x))

and

A:U(O;Rn)----tLP(O;Rn), A(u) (x) = L",(x,Y)U(Y)dY,

where l/p+ l/q = 1, are well defined and A admits a representation in the
form A = HH*, where H: L2(0;Rn) ----t LP (0; Rn) is a bounded linear
operator and H* is the adjoint of H.
Solving (6.1) in £P (0; Rn) is equivalent to solving the operator equation

u = ANf (u), u E £P (0; R n) . (6.7)

The next proposition allows us to replace the operator equation (6.7) in the
Banach space £P (0; Rn) by the operator equation

v = H* NfH (v), v E L2 (0; Rn) (6.8)

in the Hilbert space L2 (0; Rn) .

Proposition 6.6 If u E £P (0; Rn) solves (6.7) then v = H* N j (u) is

a solution of (6.8). Conversely, if v E L2 (0; Rn) solves (6.8) then u =
H (v) is a solution of (6.7). Moreover, there is a one-to-one correspondence
between the solutions of (6.7) and the solutions of (6.8).

Proof. 1) Assume that u E LP (0; Rn) solves (6.7) and let v

H* N j (u). Then

H* NfH (v) = H* NfHH* N f (u) = H* NjANj (u) = H*Nj (u) = v.

Hence v is a solution of (6.8).
2) Assume now that v E L2 (0; Rn) solves (6.8) and let u = H (v).
Then
ANf (u) = ANjH (v) = HH* NjH (v) = H (v) = u.
Hence u is a solution of (6.7).
3) Let S1, S2 be the sets of all solutions of (6.7) and (6.8), respectively.
For any u E S1, one has

HH* N f (u) = ANf (u) = u,

96 Chapter 6

and for any v E S2,

H* NfH (v) = v.
These show that the map H: S2 ----t S1 is invertible and its inverse is H* N f .
Thus there is a one-to-one correspondence between the solutions of (6.7)
and the solutions of (6.8) . •
In what follows we deal with the operator equation (6.8) in £2 (0; Rn) .
The main idea is to build a functional E : £2 (0; Rn) ----t R whose Fnkhet
derivative E' is 1- H* NfH. Thus the equation (6.8) can be written in the
variational form
E'(v)=O
and its solutions appear as critical points of E. Bya variational method for
(6.8) we mean any technique for proving the existence of critical points of
E.
 Chapter 7

The Frechet Derivative and

Critical Points of Extremum

In this chapter we present the notion of Frechet derivative of a functional

and we illustrate it by some examples. Then we build a functional E :
L2 (0; Rn) ----) R whose Frechet derivative is the operator
1- H* NfH : L2 (0; Rn) ----) L2 (0; Rn)

associated to (6.8). We prove the infinite-dimensional version of the classi-

cal Fermat's theorem about the connection between extremum points and
critical points, and we give sufficient conditions for that a functional admits
minimizers. The abstract results are then applied to establish the existence
of LP solutions for Hammerstein integral equations in Rn.

7.1 The Frechet Derivative. Examples

Let X be a Banach space, U c X an open set, E : U ----) R a functional,
and u E U a given point.

Definition 7.1 1) The derivative of E in direction v E X at u is defined

as
lim C 1 (E (u + tv) - E (u))
t-+O+
if this limit exists.
2) E is said to be Gateaux differentiable at u ifthere exists an E' (u) E
X* such that

(E' (u) , v) = lim C 1 (E (u + tv) - E (u))

t-+O+

97
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
98 Chapter 7

for all VEX. The element E' (u) is called the Gateaux derivative of E at
u.
3) E is said to be Frlichet differentiable at u if there exists an E' (u) E
X* such that

E (u + v) - E (u) = (E' (u) , v) + w (u, v) (7.1)

and
( ) · w(u,v)
w u,v = 0 (I v I) ,I.e, Ivl ---- 0, (7.2)

as v ---- 0. The element E' (u) is called the Fnichet derivative of E at u.

Remark 7.1 More generally, one can define the Frechet derivative at u of
a map E from an open subset containing u of a Banach space X to the
Banach space Y as a bounded linear operator E' (u) from X to Y for
which (7.1) and (7.2) hold as v ---- 0.

We note that if E is Frechet differentiable at a point u then E is

continuous at u. This immediately follows since

IE (u + v) - E (u)1 ~ IE' (u)llvl + Iw (u, v)1 ~ (IE' (u)1 + 1) Ivl (7.3)

for Ivl < 6. Here 6> ° is such that (7.1) and
Iw (u, v)1
Ivl ~ 1
hold for all v E X with Ivl < 6.
We say that the functional E is Gateaux (Frechet) differentiable in a
subset D of U, if E is Gateaux (respectively, Frechet) differentiable at any
point u E D.

Proposition 7.1 If E is Gateaux differentiable in an open neighborhood V

of u and E' : V ---- X* is continuous at u, then E is Frechet differentiable
at u and the two derivatives at u coincide.

°
Proof. Let r E (0,1] be such that Br (u) C V. Since E' : V ---- X* is
continuous at u, for every E> there exists a 6 E (0, r] such that

IE' (u + tv) - E' (u)1 < E for Ivl < 6, It I ~ 1. (7.4)

On the other hand, for each v E Br (0) the function

t E [0,1] f----+ (E' (u + tv) , v)

The Fnkhet Derivative 99

is the derivative of the function

g(t)=E(u+tv) (tE[O,l]).

Consequently

11 (E' (u + tv) , v) dt = 9 (1) - 9 (0) = E (u + v) - E (u) .

Hence

E (u + v) - E (u) - (E' (u) ,v) = 11 (E' (u + tv) - E' (u) , v) dt. (7.5)

11
Let
w(u,v) = (E'(u+tv) -E'(u),v)dt.

From (7.4), for all v E BE, (0) we have

Iw(u,v)1 :S 1 1
I E '(u+tv)-E'(u)llv l dt

< clvl·
Hence w (u, v) = 0 (Ivl) as v ---+ o. Then, (7.5) shows that E' (u) is the
Frechet derivative of E at u. •
We say that E E C 1 (U) if E is Frechet differentiable in U and its
Frechet derivative E' : U ---+ X* is continuous. Equivalently, in view of
Proposition 7.1 E E C 1 (U) if E is Gateaux differentiable in U and its
Gateaux derivative E' : U ---+ X* is continuous.
We say that E E C 1 (U) if E E C1 (U) and E, E' can be extended
continuously to U.

Example 7.1 Let X = R N , n c RN open, and E : n ---+ R. If E is

differentiable in a neighborhood of a point x E n in the classical sense, and
its partial derivatives DE jDxi' i = 1,2, ... , N, are continuous at x, then E
is Frechet differentiable at x and its Frechet derivative coincides with its
gradient, that is

E' (x) = \l E (x) = (~UX1E (x), ~E

UX2
(x), ... , ~E
UXN
(x)).
100 Chapter 7

Example 7.2 Assume X is a Hilbert space and

E (u) = 2"1 lUi 2 (U E X).

Then E E C 1 (X) and

(E' (u) ,v) = (u,v) , v E X.

Indeed, for every u, v E X we have

1 2 1 2 t2 2
E(u+tv)-E(u)=2"lu+ tv l -2"l u l =t(u,v)+2 Ivl .

Hence
lim C 1 (E (u + tv) - E (u)) = (u, v) .
t->O+

Example 7.3 Let X be a Hilbert space, Y a Banach space, H : X -7 Y a

bounded linear operator, and J : Y -7 R Fnkhet differentiable in Y. Then
the functional J H : X -7 R is Frechet differentiable in X and

(JH)' = H* J'H.

Here again H* is the adjoint of H. Indeed, for every u, v E X, since J is

Frechet differentiable in Y, one has

JH(u+v) - JH(u) J (H (u) + H (v)) - J (H (u))

(J' (H (u)) , H (v)) + w (H (u) , H (v))
(H* J' H (u) , v) + w (H (u) , H (v))

and
w(H(u),H(v))
-70,
IH (v)1
for IH (v)1 -7 O. Now the conclusion follows since H (v) -70 as v -7 0 and

Iw (H (u) ,H (v))1 Iw (H (u), H (v))1 IH (v)1

Ivl IH (v)1 Ivl
IHI Iw (H (u) ,H (v))1
< IH (v)1 .

Example 7.4 Following the ideas from Krasnoselskii [19] and using Theo-
rem 5.1 we can prove:
The Fhkhet Derivative 101

Theorem 7.1 Let n c RN be an open set, p E (1,00), q the conjugate

exponent of p, and F : n x Rn ----+ R a function with F (x, 0) = 0 on
n. Assume that F (., z) is measurable for all z E Rn and F (x,.) is
continuously differentiable for a.e. x E n. Let f : n x Rn ----+ Rn be defined
as
f (x,.) = \l F (x,.) (7.6)

and assume f is (p, q)-Caratheodory. Then the functional J: LP (n; Rn) ----+
R given by

J(u) = in F(x,u(x))dx

belongs to C 1 (LP (n; Rn)) and

J' = Nj ,

that is

(J' (u) , v) = in (v (x) , N j (u) (x)) dx = in (v (x) , f (x, u (x))) dx

for all v E LP (n; Rn) .

Proof. For every u, v E LP (n; Rn) define the measurable function

Bu,v : n ----+ [0,1], by

Bu,v(x) inf{t E [0,1] : F (x, v (x) + u (x))

-F (x, v (x)) (u (x), f (x, v (x) + tu (x)))}.

Then since F (x, 0) = 0,

IJ(u)1 :::; inlF(x,u(x))ldX (7.7)

< in lu (x)llf (x, Bu,o (x) u (x))1 dx

< lul p INj (Bu,o u)lq < 00.
102 Chapter 7

Thus J is well defined. On the other hand,

J(u+v)-J(u) 10 [F (x, u (x) + v (x)) - F (x, u (x))] dx

10 (v (x), N j (u + Bv,uv ) (x)) dx
10 (v (x) ,Nj (u) (x)) dx
+ 10 (v (x), N j (u + Bv,uv ) (x) - N j (u) (x)) dx
(Nj (u), v) + w (u, v).

Here
w (u, v) = 10 (v (x) ,Nj (u + Bv,uv) (x) - N j (u) (x)) dx.
We have
Iw (u, v)1 ~ Ivlp INj (u + Bv,uv) - Nj (u)lq,
whence
w (u, v)
Ivl p ---+ 0 as v ---+ 0,

since by Theorem 5.1 N j is continuous. Thus J' (u) = N j (u) . •

Definition 7.2 A function f : 0 x Rn ---+ Rn for which there exists F :

o x Rn ---+ R such that F (x, 0) = 0 a.e. on 0, F (., z) is measurable for
each z E Rn, F (x, .) is continuously differentiable and (7.6) holds for a.e.
x E 0, is said to be of potential type. In this case F is called the potential
of f.

Remark 7.2 For n = 1, each function f : 0 x R ---+ R which satisfies the

Caratheodory conditions is of potential type with

F(x,z) = lZ f(x,y)dy,

that is, F is the primitive of f in the second variable which vanishes at

zero.

Combining Examples 7.2-7.4 we obtain the following result.

The Frechet Derivative 103

Theorem 7.2 Assume that all the assumptions of Theorem 7.1 hold. In
addition assume that H : L2 (0; Rn) - t LP (0; Rn) is a bounded linear
operator. Then the functional E: L2 (0; Rn) - t R, given by

E (u) =
1
21ul2
2
- inr F (x, H (u) (x)) dx
is Frechet differentiable and

E' =1 -H*NfH

that is

(E'(u) ,v) = (u-H*NfH(u) , v)2 (u,v E L2(0;Rn)).

The functional E in Theorem 7.2 is known as the Golomb functional

generated by Hand f.

7.2 Minima of Lower Semicontinuous Functionals

Now we come back to the general functional E : U - t R, where U c X
is open. By definition u E U is a critical point of E if E is Fn§Chet
differentiable at u and E' (u) = O. The simplest example of critical points
is that of the points of local extremum.

Proposition 7.2 If Uo E U is a point of local extremum of E and E is

Frechet differentiable at Uo, then E' (uo) = o.

Proof. Assume that Uo is a point of local minimum, i.e.,

E (uo) ::; E (u)

for all u in a neighborhood of Uo. Since E is Frechet differentiable at uo,
we have
o ::; E (uo + v) - E (uo) = (E' (uo) , v) + w (Uo, v)
and
w(uo,v)
Ivl -t 0

as v -t O. Setting v = tw, where t> 0 and w EX, Iwl = 1, dividing by

t and then letting t -t 0+ we obtain

(E' (uo) , w) 2: O.
104 Chapter 7

Similarly, replacing w by -w we have

(E' (uo), -w) ~ o.

Hence
(E' (uo) , w) = 0
for all wE X, Iwl = 1. As a result E' (uo) = O.A similar argument applies
to the points of local maximum. _
Thus a first idea in finding critical points is to look for extremal values
of the functional.

Definition 7.3 A functional E : D c X -+ R is said to be lower semicon-

tinuous, l.s.c. for short, on D if for every r E R, the level set

(E 5, r) = {u ED: E (u) 5, r}

is closed in D. E is said to be weakly lower semicontinuous on D if for

every r E R, the level set (E 5, r) is closed in D with respect to the weak
topology of X.

For weak topologies and additional properties of linear functional anal-

ysis we refer the reader to Brezis [5J and Yosida [44J.

Proposition 7.3 Assume X is a reflexive Banach space, D c X is a

bounded closed convex nonempty set and E : D -+ R is weakly l.s.c. on D.
Then E is bounded from below and attains its infimum.

Proof. Let (Uk) be a minimizing sequence of E, i.e., Uk ED and

E (Uk) -+ ro = inf E E R U {-oo}.

D

Since X is reflexive and D is bounded, we may assume, passing eventually

to a subsequence, that Uk ->. Uo weakly as k -+ 00, where Uo E X. D being
closed convex we have Uo ED. Assume ro < E (uo). Then there exists an
E > 0 and a kc EN such that E (Uk) 5, E (uo) - E, that is
Uk E (E 5, E (uo) - E) ,
for all k ~ kc . This, since E is weakly l.s.c. on D, implies

Uo E (E 5, E (uo) - E)

too. Thus we have obtained E (uo) 5, E (uo) - E, a contradiction. Hence

E (uo) = roo _
The Frechet Derivative 105

Definition 7.4 A functional E : D c X ~ R defined on an unbounded

set D, is said to be coercive if E (u) ~ 00 as lui ~ 00.
The analogue result of Proposition 7.3 for an unbounded set D is the
following

Proposition 7.4 Assume X is a reflexive Banach space, D C X is an

unbounded closed convex set and E : D ~ R is coercive and weakly l.s.c.
on D. Then E is bounded from below and attains its infimum.

Proof. Fix any element Ul in D. Since E is coercive there exists an

R> 0 such that E (u) > E (Ul) for all u E D with lui> R. Consequently

infE= inf E
D DnBR(O)

and the conclusion follows from Proposition 7.3. •

Theorem 7.3 Let X be a reflexive Banach space and E : X ~ R be

coercive, weakly l.s.c. on X and Frechet differentiable in X. Then there
exists Uo E X with

E (uo) = inf E, E' (uo) = O.

X

Proof. Apply Propositions 7.4 and 7.2 . •

The next result is a criterium of weak lower semicontinuity. It is stated
in terms of convex functionals.

Definition 7.5 Let D be a convex subset of the Banach space X. A func-

tional E : D ~ R is said to be convex if

E (u + t (v - u)) ~ E (u) + t (E (v) - E (u)) (7.8)

for all u,v E D, u f. v and t E (0,1). The functional E is said to be

strictly convex if strict inequality occurs in (7.8). The functional E is said
to be concave if - E is convex.

Proposition 7.5 Assume X is a Banach space, D C X is closed convex

and E : D ~ R is convex and l.s.c. on D. Then E is weakly l.s.c. on D.

Proof. Notice each level set (E ~ r) is closed convex and so is weakly

closed. •
106 Chapter 7

Theorem 7.4 Let X be a reflexive Banach space and E : X ----t R be

convex, coercive, and Frechet differentiable in X. Then there exists Uo E X
with
E (uo) = inf E, E' (uo) = O.
X

If in addition E is strictly convex then E has a unique critical point.

Proof. Since E is Frechet differentiable, E is continuous and so l.s.c.

on X. Now the existence part follows from Proposition 7.5 and Theorem
7.3.
Assume E is strictly convex and suppose that Ul is a critical point of
E with Ul -=f. uo. Then for sufficiently small t> 0, we have

E (uo + t (Ul - uo)) - E (uo) t (E' (uo) ,Ul - uo) + w (uo, t (Ul - uo))
W (uo, t (Ul - uo))

and
w (uo, t (Ul - uo)) = 0 (t) .

On the other hand, since E is strictly convex, one has

E (uo + t (Ul - uo)) < E (uo) + t (E (Ul) - E (uo)) (7.9)

for all t E (0,1). Hence

w (uo, t (Ul - uo)) < t (E (ut) - E (uo))

for all t E (0,1) small enough. Dividing by t and letting t ----t 0+, we obtain
o :S E (ut) - E (uo). Similarly O:S E (uo) - E (ut). Hence
E (ut) = E (uo) = inf E,
X

and from (7.9) we deduce for t E (0,1) ,

E (uo + t (Ul - uo)) < inf E,

X

a contradiction. •
The Frechet Derivative 107

7.3 Application to Hammerstein Integral

Equations
We now use Theorems 7.3 and 7.4 to obtain existence and uniqueness results
for the Hammerstein integral equation in R n

u(x)= kK,(x,y)f(y,u(y))dy a.e.xEO. (7.10)

Theorem 7.5 Let 0 C RN be bounded open, 2:::; p < Po < 00, 1/p+1/q =
1, 1/po + 1/qo = 1, K,: 0 2 ----t Rand f : 0 x R n ----t Rn. Assume that the
following conditions are satisfied:
(i) the operator A: LqO (0; Rn) ----t Vo (0; Rn) given by

A (u) (x) = k K, (x, y) u (y) dy

is bounded and its restriction A: L2 (0; Rn) ----t L2 (0; Rn) is positive, self-
adjoint and completely continuous;
. (ii) f is (p, q)-Caratheodory of potential type;
(iii) the potential F of f satisfies the growth condition

IF (x, z)1 :::; ~ Izl2 + b (x) IzI 2-'Y + c(x) (7.11)

for a.e. x E 0 and all z ERn, where 0 < I < 2, b E L 2 h (O;R+),

cELl (0; R+), a E R+ and alAI < 1 (A as operator from L2 (0; Rn) to
L2 (0; Rn)).
Then the Hammerstein equation (7.10) has at least one solution u in
V (0; Rn) .
Proof. According to Proposition 6.6, (7.10) is equivalent to the operator
equation
v = H*NfH (v), v E L2 (O;Rn). (7.12)
Furthermore, this equation can be written as

E' (v) = 0,
where the functional E: L2 (0; Rn) ----t R is given by

E (v) ~ Ivl~ -
2 inr F (x, H (v) (x)) dx
1
21vl~ - JH (v).
108 Chapter 7

1) E is coercive. Indeed, by (7.11) we have

1 2
E (v) 21vl2 - JH (v)

> 1 2 - -IH
-IVI2
2
a (V)b2 -
2
1 n
b (x) IH (v) (X)I 2-'Y dx - Icll .
We have

IH (v)l~ = (Al/2 (v), A l / 2 (v)) = (v, A (v))

< IAllvl~
and

Inf b(x) IH (v) (x)1 2-'Y dx ~ ( Inf b (x)2h dx )'Y/2 ( Inf IH (v) (x)1 2dx ) 1-'Y/2
b IH (v)I~-'Y ~ b IAI1-'Y/2Ivl~-'Y,

where b = Ibl2h. Denote c= Icll. Then

E (v) ~ ~ Ivl~ (1 - alAI) - b IAI1-'Y/2Ivl~-'Y - c.

Since 1 - alAI> 0 and 2 - "I < 2 we may infer that

E (v) ----) 00 as Ivl2 ----) 00.

Thus E is coercive.
2) E is weakly l.s.c. on L2 (0; Rn). To this end let r> 0 and let (Vk)
be a sequence with

E (Vk) ~ rand Vk -' v weakly as k ----) 00.

Then
(Vk -v,u)2 ----) 0 as k ----) 00

for every u E L2 (0; Rn). Since the operator A acting in L2 (0; Rn) is
completely continuous, we may assume, passing eventually to a subsequence,
that A (Vk) ----) W in L2 (0; Rn). We have

IAI/2 (Vk) - A l / 2 (v)l: = (Vk - v, A (Vk) - A (v)h

= (Vk - v, A (Vk) - wh + (Vk - v, W - A (v))2 ----) 0

The Frechet Derivative 109

as k ----+ 00, since the sequence (Vk - v) is bounded, A (Vk) - w ----+ 0, and
(Vk - v,w - A(v))2 ----+ o. It follows that

Al/2 (Vk) ----+ A 1/ 2 (v)

strongly. On the other hand, H being completely continuous (see Theorem

6.1), we may assume (passing eventually to a new subsequence) that H (Vk)
converges in V (0; Rn). Since H (Vk) = Al/2 (Vk) the limit of H (Vk) must
be H (v) = A 1/ 2 (v). Hence

H (Vk) ----+ H (v)

in V (0; Rn). Consequently

JH (Vk) ----+ JH (v)

in R. In addition, it is well known that if Vk ----' v weakly then IvI2 <

lim inf IVk 12. Then since E (Vk) ::; r

1 2
21vkl2 - r::; JH (Vk)

and letting k ----+ 00 we obtain

1 2 .. 1 2
21vI2 - r::; hmmf 21vkl2 - r::; JH (V).

Hence E (v) ::; r. Thus, the level set (E::; r) is weakly closed. Therefore
E is weakly l.s.c. on L2 (0; Rn) .
Now the conclusion follows from Theorem 7.3. •
If instead of the complete continuity of A as an operator acting in
L2 (0; Rn) we ask for convexity of E, by Theorem 7.4 we immediately
obtain the following existence and uniqueness result.

Theorem 7.6 Let 0 C RN be bounded open, 2::; p < Po < 00, l/p+l/q =
1, l/po + Ijqo = 1, "': 0 2 ----+ Rand f : 0 x Rn ----+ Rn. Assume that the
following conditions are satisfied:
(i) the operator A: LqO (0; Rn) ----+ vo (0; Rn) given by

A (u) (x) = k'" (x, y) u (y) dy

is bounded and its restriction A : L2 (0; Rn) ----+ L2 (0; Rn) is positive and
self-adjoint;
110 Chapter 7

(ii) f is (p, q)-Caratheodory of potential type;

(iii) the potential F of f is concave in the second variable and satisfies
the growth condition

IF (x, z)1 ::; ~ IZl2 + b (x) IzI 2-" + c(x)

for a.e. x E [1 and all z ERn, where 0 < 'Y < 2, b E L 2h ([1; R+) ,
c E L1 ([1; R+) , a E R+ and alAI < 1 (A as operator from L2 ([1; Rn) to
L2 ([1; Rn)).
Then the equation (7.10) has a unique solution U E V ([1; Rn) .

Proof. The functional Ivl~ /2 is strictly convex. Also, since F (x, .) is

assumed concave the functional

-L F(x,H(v)(x))dx

is convex. Hence E is strictly convex and Theorem 7.4 applies. For

uniqueness assume that U1, U2 E LP ([1; Rn) are solutions to (7.10), that
is Ui = ANf (Ui) , i = 1,2. Let Vi = H* N f (Ui). Then Vi solve (7.12).
Since (7.12) has a unique solution (the unique critical point of E) we have
V1 = V2· Hence H* N f (U1) = H* N f (U2). Then

HH* N f (U1) = HH* N f (U2).

But HH* = A, so ANf (U1) = ANf (U2). Thus U1 = U2 . •

We conclude this chapter by some conditions on kernel K, which guarantee
(i) in Theorems 7.5:
1) K, E LPO ([1 x 0) ,
2) K, is symmetric, i.e., K,(x,Y)=K,(Y,x),
3) K, is positive semidefinite, i.e.,

{ K, (x, y) (u (x) , U (y)) dx dy 2: 0

inxn
for every U E L2 ([1; Rn) .
 Chapter 8

The Mountain Pass Theorem

and Critical Points of Saddle
Type

In Chapter 9 we shall continue the investigation of the LP solutions of the

Hammerstein integral equations under the assumption that f (x, 0) = 0,
that is, the null function is a solution. We are now interested in non-null
solutions. The technique we use is based on the so called mountain pass
theorem of Ambrosetti-Rabinowitz [3]. By this method one can establish
the existence of a critical point u of the functional E which in general is not
an extremum point of E, and has the property that in any neighborhood
of u there are points v and w with E (v) < E (u) < E (w). Such a critical
point is said to be a saddle point of E.
In this chapter we first present the Ambrosetti-Rabinowitz theorem with
a proof based on Ekeland's principle. Then we present Schechter's version
[40] of the mountain pass theorem, which guarantees the existence of a
critical point in a bounded region of the space. The proof is based on
deformation arguments and uses the notion of flow associated with a locally
Lipschitz map and a generalized pseudo-gradient lemma.

8.1 The Ambrosetti-Rabinowitz Theorem

Definition 8.1 A functional E E C 1 (X) is said to satisfy the Palais-

Smale condition, for short the (PS) condition, if any sequence of elements

111
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
112 Chapter 8

Uk E X for which

E (Uk) -t JL E R, E' (Uk) -t 0 (8.1)

as k - t 00, has a convergent subsequence.

We note that the (PS) condition is a compactness property. It asks that

for any sequence (Uk)k2:1 satisfying (8.1), the set {Uk: k 2: 1} is relatively
compact.

Theorem 8.1 (Amhrosetti-Rahinowitz) Let X be a Banach space and

E E C l (X). Assume that there exist Uo, UI E X and r with Iuol < r < lUll
such that

max { E (uo) , E (uI)} < inf {E (u) : U EX, Iu I = r} .

Let
r = bE C ([0, 1]; X) : I' (0) = Uo, I' (1) = ud (8.2)

and
c = inf max E
,Er tE[O,I]
h (t)) . (8.3)

Then there exists a sequence of elements Uk E X such that

E (Uk) -t c, E' (Uk) -t 0 as k - t 00.

If, in addition, E satisfies the (PS) condition then there exists an element
u E X \ {uo,ud with

E (u) = c, E' (u) = o. (8.4)

Remark 8.1 Notice r is the set of all continuous paths joining Uo and
UI. Roughly speaking, the mountain pass theorem says that if we are at
the point Uo of altitude E (uo) located in a cauldron surrounded by high
mountains, and we wish to reach to a point UI of altitude E (uI), over
there the mountains, we can find a path going from Uo to UI, through a
mountain pass. To find a mountain pass we have to choose a path which
mounts the least.

The proof that we present here relies on Ekeland's variational principle

[12].
 The Mountain Pass Theorem 113

Lemma 8.1 (Ekeland) Let (Z, d) be a complete metric space and let 'IjJ :
Z ~ R be a lower semicontinuous function bounded from below. Then given
c > 0 and Uo E Z there exists a point u E Z such that

'IjJ(v) - 'IjJ (u) +cd(u,v) 20 for all v E Z, (8.5)

'IjJ (u) :s; 'IjJ (uo) - cd (u, uo) . (8.6)

Proof. Since cd is an equivalent metric on Z, we may assume that

c = 1.
For each U E Z we consider the set

Z(U)={vEZ: 'IjJ(v)-'ljJ(u)+d(u,v):S;O}.

Clearly U E Z (u) and it is easily seen that if v E Z (u) then Z (v) c Z (u).
Let (ck) be a sequence of real numbers Ck > 0 with Ck ~ 0 as k ~ 00,
and let (Uk) be a sequence with

Uk+1 E Z (Uk)' 'IjJ (Uk+l):S; inf 'IjJ + ck+l

Z(Uk)

for k = 0,1, .... Since Uk+l E Z (Uk) we have

'IjJ (Uk+d - 'IjJ (Uk) + d (Uk+1' Uk) :s; O. (8.7)

Hence the sequence ('IjJ (Uk)) is decreasing. It is also bounded since 'IjJ is
bounded from below. Thus ('IjJ (Uk)) is convergent. On the other hand, from
(8.7) we obtain

'IjJ (u m ) - 'IjJ (Uk) + d (Urn, Uk) :s; 0 for k < m.

It follows that (Uk) is a Cauchy sequence. Let U E Z be its limit. Since

Uk E Z (uo) we have

'IjJ (Uk) - 'IjJ (uo) + d (Uk, uo) :s; 0,

and letting k ~ 00 we obtain (8.6). To prove (8.5) let v E Z be an arbitrary
element of Z. Two cases are possible:
1) v E nZ (Uk). Then

'IjJ (Uk+l):S; inf 'IjJ + Ck+l :s; 'IjJ (v) + Ck+l

Z(Uk)

for all k. From this, using the lower semicontinuity of 'IjJ we deduce 'IjJ (u) :s;
'IjJ (v), and so (8.5).
114 Chapter 8

2) v ~ nZ (Uk). Then there exists an m such that v ~ Z (Uk) for all

k :::: m. Consequently

'lj;(v) - 'lj; (Uk) + d(V,Uk) > 0, k:::: m.

Letting k -+ 00 we obtain (8.5). •
Here are two useful consequences of Ekeland's principle.

Corollary 8.1 Under the assumptions of Lemma 8.1, for each € > 0 there
exists an element U E Z such that (8.5) holds and

'lj; (u) :S i~f 'lj; + €. (8.8)

Proof. Apply Lemma 8.1 to an element Uo E Z with 'lj; (uo) :S i~f 'lj; + €.
Then (8.8) immediately follows from (8.6). •

Corollary 8.2 Under the assumptions of Lemma 8.1, if Z is a Banach

space with norm 1.1, and 'lj; is a C 1 functional there exists a sequence (Uk)
with
'lj;(Uk) -+ inf'lj;
z
(i. e., (Uk) is a minimizant sequence of 'lj;) and

'lj;' (Uk) -+ O.

Proof. For € = 11k, by Corollary 8.1 there is an element Uk E Z such

that
1
'lj; (v) - 'lj; (Uk) + k IUk - vi :::: 0, v E Z, (8.9)

'lj; (Uk) :S i~f 'lj; + l·

Take an arbitrary wE Z and apply (8.9) to v = Uk +tw, t E R. We obtain
1
'lj; (Uk + tw) - 'lj; (Uk) + k Itllwl :::: O.
It follows that for It I small enough we have
t ('lj;' (Uk) ,w) +o(ltl) + lltllwl:::: O.
For t> 0, t -+ 0 we deduce

('lj;' (Uk) ,w) :::: -llwl,

The Mountain Pass Theorem 115

whilst for t < 0, t -+ ° we obtain

(?jJ' (Uk) ,w) ::; ~ Iwl.

Hence
1(?jJ'(Uk),w)l::; ~Iwl, WEZ.

Therefore
1?jJ' (Uk)1 ::; ~
°
whence ?jJ' (Uk) -+ as k -+ 00. •

°
If X is a Banach space and v E X* then it is a simple consequence of
the definition of the norm of v, that for each c > there exists U E X with

lUi::; 1 and (v,u) > lvi-c.

The next lemma guarantees that if v depends continuously on a parame-
ter t then the corresponding element U can be choosen so that it depends
continuously on t as well.

Lemma 8.2 Let X be a Banach space and fEe ([0,1] ; X*). Then there
exists a function 'P E C ([0, 1] ; X) such that
I'P (t)1 ::; 1, (J (t) ,'P (t)) > If (t)l- c
for all t E [0,1].

Proof. Let to E [0,1]. According to the above remark there exists

Uo EX with Iuol ::; 1 and (J (to), uo) > If (to)l- c. Let
U (to) = {t E [0, 1] : (J (t) , uo) > If (t) I - c} .
Clearly to E U (to) and since f is continuous U (to) is open in [0,1]. Since
[0,1] = UtE[O,lj U (t), there exists a finite open covering

{U (tl), U (t2) , ... , U (tn)}

of [0,1]. Let Ui, i = 1,2, ... , n be the corresponding elements, i.e.,

U (ti) = {t E [0, 1] : (J (t) , Ui) > If (t) I - c} .

If pdt) =dist (t, [0, 1] \ U (ti)) define

(i (t) = ! i (t)
L Pj (t)
j=l
116 Chapter 8

Notice (i : [0,1] ----t [0,1] is continuous, (i (t) i- a if and only if t E U (ti)'

n
and I: (i (t) = 1 for all t E [0,1]. Now define
i=l
n
<.p (t) = l:= (i (t) Ui·
i=l

Clarly <.p E C ([0, 1] ; X) and

n n
l<.p (t)1 ::; l:= (i (t) IUil ::; l:= (i (t) = 1.
i=l i=l

Also it follows easily that <.p satisfies (f (t), <.p (t)) > If (t)l- E on [0,1] . •
Proof of Theorem 8.1. We apply Corollary 8.1 to Z = r given by
(8.2) and to the functional 'ljJ : r ----t R defined by

'ljJ b) = max E b (t)) .

tE[O,l]

This functional is lower semicontinuous and bounded from below by c (given

by (8.3)). It follows that for every E > a there exists a '1£ E r with

'ljJ('f]) - 'ljJb£) + Ed ('f], '1£) ~ 0, 'f] E r, (8.10)

c ::; 'ljJ b£) ::; inf'ljJ

r
+ E = C + E.
Let
A£={tE[O,l]: Eb£(t))='ljJb£)}.
For concluding the proof it is sufficient to show that there exists a t£ E A£
with IE' b£ (t£))1 < 2E. To this end we apply Lemma 8.2 to the function
f E C ([0,1]; X*) ,
f (t) = E' b£ (t)) .
Hence there exists a function <.p E C ([0, 1] ; X) such that I<.p (t) I ::; 1 and

(E'b£(t)),<.p(t)) > IE'b£(t))I-E

on [0,1]. In (8.10) take 'f] = '1£ - AW with A> a and
w(t) = ((t)<.p(t),

where ( : [0,1] ----t [0,1] is continuous, ((t) = 1 on A£, and ((0) = ((1) =
O. We have
d ('f], '1£) = Alwl ::; A,
The Mountain Pass Theorem 117

'ljJ (17) = max E (17 (t)) = E (17 (t A ))

tE[O,l]

for some tA E [0, 1J. Hence

E (17 (tA)) - max E

tE[O,l]
be (t)) + E.\ 2 O.

Since

E (17 (tA)) - E be: (tA)) = -.\ (E' be: (te)) ,w (t A)) + 0 (.\)
we deduce that

- (E' be: (t A)) ,w (t A)) + E + ±o (.\) 2 O.

We may assume that tA -+ te: E Ae: as .\ -+ O. Then

- (E' be: (te)) , cP (te:)) + E 2 o.

Thus
IE' be: (te:))I- E < (E' be (te:)) , cp (te:)) ::; E,

whence IE' be: (te:)) I < 2E . •

Notice that Theorem 8.1 does not give any information about the lo-
calization of the critical point u satisfying (8.4) (of saddle type). This
justifies the following question: what additional hypotheses will guaran-
tee the existence of a critical point of saddle type in a given region of the
space. Some answers to this question can be found in Pucci-Serrin [35J,
Guo-Sun-Qi [16]' Ghoussoub-Preiss [14], Schechter [39]' [40]' Schechter-
Tintarev [41], Ma [22], Frigon [13J and Liu-Sun [21]. In Section 8.3 we
shall present Schechter's bounded mountain pass theorem in a somewhat
particular case sufficient for our purposes.

8.2 Flows and Generalized Pseudo-Gradients

In this section we state and prove the lemmas that will be used in Section
8.3, in the proof of Schechter's mountain pass theorem.

Lemma 8.3 Let X be a Banach space and W X -+ X be a locally

Lipschitz map such that

IW (u)1 ::; M, u E X.
118 Chapter 8

Then for each u E X there exists a unique function a (u, .) E C I (R; X)

such that
dCT~~,t) = W (a (u, t)), t E R,
{ (8.11)
a (u, 0) = u.
In addition, a E C (X x R; X) and
a (u, t + s) = a (a (u, s) ,t)
for all u E X and t, s E R.

Proof. Let u E X be fixed and let r > be such that W is Lipschitz

in Br (u) with the Lipschitz constant L> 0. Let
°
r
tl < rL + IW (u)1
and
K={vEC([O,tIl;X): Iv(t)-ul:Sr on [O,tI]}'
Let A: K ---+ K be given by

A (v) (t) = u+ lot W (v (s)) ds.

From

IA (v) (t) - ul :S lot IW (v (s))1 ds

< lot (IW (v (s)) - W (u)1 + IW (u)1) ds
< (rL+ IW(U)l)tI
< r,
(t E [0, tID we have that A is well defined. In addition we can easily see
that A is a contraction with Lipschitz constant Ltl < 1. By Banach's
contraction principle, A has a unique fixed point a (u,.) E C ([0, tIl; X) .
Clearly a (u,.) E C I ([0, tIl; X) and satisfies (8.11) for t E [0, tIl. Let
t+ E (O,ool be the supremum of all numbers tl such that (8.11) has a
solution on [0, tIl. If t+ < 00 then we take any sequence tk < t+, with
tk ---+ t+ and we obtain

la (u, tj) - a (u, tk)1 = 11: W (a (u, s)) dsl

i

< M Itj - tkl.

The Mountain Pass Theorem 119

Hence the sequence (J (u, tk) is Cauchy and so convergent to some w. Now
again by the local existence theorem we can uniquely solve

d(J (u, t) = W ((J (u, t)) , (J(u,t+) =W

dt
in a certain interval [t+, t+ + 8). This will contradict the definition of t+.
Thus t+ = 00. A similar argument can be used for the interval (-00,0] . •

Definition 8.2 The map (J given by Lemma 8.3 is called the flow generated
by the locally Lipschitz map W.

A source of locally Lipschitz maps are the so called pseudo-gradients.

We now present an example of pseudo-gradient owed to Schechter [40].

Lemma 8.4 Let X be a Hilbert space and let a, e be real numbers with
o < a < 1 - e and e 2: 0. Then for any elements u, v EX \ {OJ satisfying
(u,v):S e lullvl
there exists an element hEX such that

alullhl, (v,h) < O.

(u,h) 2:

Proof. We may assume that lui = Ivl = 1. We look for h in the form
h =u - j3v with j3 2: O. Then

Ihl:S 1+j3, (v,h):S e-j3, (u,h) 2:1-j3e.

Since 0 < a < 1 - e, we may take a j3 > e such that a (1 + j3) :S 1 - j3e;
for instance, set j3 = (1 - a) / (a + e). This yields the desired result . •

Lemma 8.5 Let X be a Hilbert space. Let F, G : D C X -7 X be con-

tinuous, fj = {u ED: G (u) =1= O} and let Do C fj be closed. Assume
that
F(u) =1= 0, u E Do,
and that there is a e E [0, 1) such that

(F (u) , G(u)) :S e IF (u) II G(u) I, u E Do.

Then for each a with 0 < a < 1 - e, there exists a locally Lipschitz map
H : fj -7 X such that

IH (u)1 :S 1, (G (u) ,H (u)) 2: a IG (u)l, u E fj,

and
(F (u), H (u)) < 0, u E Do.
120 Chapter 8

Proof. Let a' be such that a < a' < 1 - e. Consider a map h : fj --+ X
with Ih (u)1 = 1 for all u E fj and with the following properties:

h (u) = IG (u)I- 1 G (u), u E fj \ Do

and
(G (u), h (u)) ~ a' IG (u)l, (F (u), h (u)) < 0, u E Do.
Such a map exists by Lemma 8.4. Notice that one has

(G (u) , h (u)) ~ a' IG (u) I , u E fj.

Since F and G are continuous, for each u E fj, there is an open neighbor-
hood V (u) of u such that

(G (v) , h (u)) ~ a IG (v) I, (F (v) , h ( u)) < 0, u E Do

for all VEV(u). For uEfj\Do, take V(u) so small that V(u)nDo=0
(recall Do is a closed set) and

(G (v) ,h (u)) ~ a IG (v)1 , v E V (u).

The collection {V (u) : u E fj} is an open covering of fj. Since fj is
paracompact, this covering has a locally finite refinement {Vr }. Let {'¢r}
be a locally Lipschitz partition of unity subordinate to this refinement, and
for each T, let U r E fj be an element for which Vr C V (u r ). Now let
H : fj --+ X be given by

H (v) = L '¢r (v) h ( r ) .

U
r

Clearly H is locally Lipschitz. Also, for every v E fj we have IH (v)1 S; 1

and
(G (v), H (v)) L '¢r (v) (G (v) , h ( r )) U
r
> a L '¢r (v) IG (v)1
r
a IG (v)1 ,
whilst for v E Do, one has

(F (v), H (v)) L '¢r (v) (F (v), h (u r ))

r
< 0,
which concludes the proof. •
The Mountain Pass Theorem 121

8.3 Schechter's Bounded Mountain Pass Theorem

Throughout this section X will be a Hilbert space and for any R E (0,00],
BR will be the open set {u EX: lui < R}. Clearly, for R = 00 one has
B R = B R = X and 8 B R = 0.

Definition 8.3 Let R E (0,00] and E E C l (BR). For R < 00, we say
that E satisfies Schechter's Palais-Smale condition, (PS)R condition for
short, if any sequence of elements Uk E BR \ {O} for which

E (Uk ) -; I-t E R, E '(Uk ) - (E'(Uk)2,Uk) Uk -; 0, (E '(Uk), uk ) -;

IUkl
1/ ~ °
(8.12)
as k -; 00, has a convergent subsequence. For R = 00 we let (PS)oo =(PS).

Remark 8.2 If E satisfies the (PS)R condition, then E satisfies the Palais-
Smale condition on B R , i.e., any sequence of elements Uk E BR satisfying

°
(8.1) has a convergent subsequence. For R = 00 this is true by definition.
Assume R < 00. Then since (Uk) is bounded and E' (Uk) -; we have
(E' (Uk), Uk) -; 0, so 1/ = 0. Also,

Uk) uk I ~ IE' (Uk) I -; 0.

I(E' (Uk)2'
IUkl
Consequently
E '( Uk ) - (E'(Uk) 2,Uk) uk -; . °
IUkl
Thus (Uk) satisfies (8.12) ,and so by the (PS)R condition it has a convergent
subsequence.

Theorem 8.2 (Schechter) Let X be a Hilbert space, R E (0,00]' and

E E C l (BR). Assume that for some 1/0> 0,

(E' (u) ,u) 2:: -1/0, U E 8BR (8.13)

and that there are Uo, Ul E BR and r with Iuol < r < lUll such that

max{E(uo) ,E(Ul)} < inf{E(u): U E BR, lui = r}. (8.14)

Let
fR = bEe ([0, 1]; BR) : I' (0) = Uo, I' (1) = ud
122 Chapter 8

and
CR = inf max E b (t)) .
,ErR tE[O,l]
Then either there is a sequence of elements Uk E B R with

E (Uk) -; CR, E' (Uk) -; 0, (8.15)

or there is a sequence of elements Uk E 8BR such that

E (Uk ) -; CR, E , (Uk ) - (E' (Uk)

n'J
, Uk)
Uk -; 0, (
E , (Uk) ,Uk
)
:s O. (8.16)

If in addition E satisfies the (PS)R condition and

E' (u) + J-lU i- 0, U E 8BR, J-l > 0, (8.17)

then there exists an element U E BR \ {uo, Ul} with

E (u) = CR, E' (u) = o.

Remark 8.3 For R = 00, rR = r, CR = c, conditions (8.13) and (8.17)
are trivial since 8BR = 0, and Theorem 8.2 reduces to Theorem 8.1.
Remark 8.4 For R < 00, (8.17) is the Leray-Schauder boundary condition
for the operator I - E', i.e., it is equivalent to

U i- ).. (I - E') (u), U E 8BR, ).. E (0,1) .

The proof of Theorem 8.2 is based on the following lemma [40j.

Lemma 8.6 Assume all the assumptions of Theorem 8.2 hold. In addition
assume that there are constants (), 6 with 0 :s () < 1, 6 > 0 such that

(E' (u) , u) + () R IE' (u) I ~ 0 (8.18)

for all U E 8BR satisfying

IE (u) - cRI :s 6. (8.19)

Then there exists a sequence of elements Uk E B R such that

E (Uk) -; cR, E' (Uk) -; O. (8.20)

The Mountain Pass Theorem 123

Proof. Assume there are no sequences satisfying (8.20). Then there

would be constants 8, m > 0 such that

IE' (u)1 "2 m (8.21 )

for all u in
Q = {u E BR: IE (u) - cRI ~ 38}.
Clearly, we may assume 38 < CR - max {E (uo) ,E (Ul)} and that (8.18)
holds in Q = Q n aBR. Denote

Qo {u E BR: IE (u) - cRI ~ 28} ,

Ql {u E BR: IE (u) - cRI ~ 8} ,
Q2 BR \ Qo,
d (u, Q2)
TJ (u)
d(U,Ql) +d(U,Q2)'

We have

TJ (u) = 1 in Ql' TJ (u) = 0 in Q2, 0 < TJ (u) < 1 otherwise.

We no~ apply Lemma 8.5 to F (u) = -u, G (u) = E' (u), D = BR and
Do = Q. It follows that for each positive a < 1 - (j, there exists a locally
Lipschitz map H : fj --+ X (here fj means the set {u E B R : E' (u) oF O} )
such that

IH (u)1 ~ 1, a IE' (u)1 ~ (E' (u), H (u)) , u E fj (8.22)

and
(u, H (u)) > 0, u E Q. (8.23)

Define W : BR --+ X by

W (u) ={ -TJ (u) H (u) for u E fj (8.24)

o for u E B R \ fj.

This map is locally Lipschitz and can be extended to a locally Lipschitz map
on the whole of X, by setting

W(u) =W C~lu) for lui> R.

124 Chapter 8

Let 0' be the flow generated by W as shows Lemma 8.3. Note 0' (u,.) does
not exit BR for every u E BR. Indeed, we have

d 10' ~~' t)12 = 2 (dO' ~~' t), 0' (u, t») = 2 (W (0') , 0'). (8.25)

Assume 10' (u, t)1 ::; R for all t E [0, to) and Ul = 0' (u, to) E fJBR, for some
to E R+. If Ul E Q then by (8.23) and (8.24) we see that

(W(O'(u,t»,O'(u,t» < °
for t in a neighborhood of to. If Ul t/: Q then
ry(O'(u,t» =0
for t in a neighborhood of to. Hence the right hand side of (8.25) is non-
positive in a neighborhood of to. Thus 10' (u, t) I is nonincreasing on some
interval [to, to + E). Therefore 0' (u,.) does not exit BR for t E R+.
Let us denote by EA the level set (E::; A), i.e.,

EA = {u E BR : E (u) ::; A} .
Now we look at the composite map E (0' (u, .». By (8.21) and (8.22) we
have
dE (0' (u, t»
(E' (0' (u, t» , dO' ~~' t») (8.26)
dt
(u, t» (E' (0' (u, t» , H (0' (u, t) ))
- ry ( 0'
< -ry (0' (u,t»am.
Let tl > 26/ (a m), and let u be any element of E cR + 8 . If there is a
toE [O,tl] with O'(u,to) t/: Ql, then

E(O'(U,tl» ::; E(O'(u,to»

< CR - 6.
Hence 0' (u, tl) E E CR -8. Otherwise, 0' (u, t) E Ql for all t E [0, tl], and so
== 1. Then (8.26) implies
ry (0' (u, t»

E(0' (u,tt» ::; E(u)-amtl

< CR + 6 - 26 = CR - 6.
Thus
0' (ECR +8 ' tl) C E CR - 8' (8.27)
The Mountain Pass Theorem 125

Now by the definition of CR, there is a "/ E rR with

"/ (t) E ECRH ' t E [0,1]. (8.28)

We define a new path "/1 joining Uo and U1 by

"/dt) = (T ("((t) , t1), t E [0,1].

Since TJ vanishes in the neighborhood of Uo and U1, we have (T (uo, t) == Uo

and (T (U1' t) == U1. Hence "/1 (0) = uo, "/1 (1) = U1 and so "/1 ErR. On the
other hand, from (8.27) and (8.28), we have

E(,,(dt))~CR-8, tE[O,I],

which contradicts the definition of CR. This finishes the proof of Lemma
8.6 .•

°
Proof of Theorem 8.2. Assume a sequence satisfying (8.16) does not
exist. Then there are constants m, 8> such that

IE' (u) - R- 2 (E' (u) ,u) ul ~ m

whenever
IE (u) - cRI ~ 8, (E' (u), u) ~ 0, u E aBR. (8.29)
Let 0 > ° be such that

0< 0- 2 - 1 < m 2 R2vo 2.

Then if u satisfies (8.29) we have

(E' (u), u)2 (0- 2 - 1) ~ R2IE' (u) - R- 2 (E' (u), u) ul 2 (E' (u), u)2v02.
Since (E' (u) ,u)2vo2 ~ 1 this yields

0- 2 (E' (u), u)2 ~ (E' (u) ,u)2 + R2IE' (u) - R- 2 (E' (u) ,u) ul 2
R2 IE' (u) 12 .

Hence u satisfies (8.18). Thus (8.18) holds for all u satisfying (8.29). It
also holds trivially for all u with (E' (u) , u) > 0. Hence (8.18) holds for
all u E 8BR satisfying (8.19). Therefore Lemma 8.6 applies and guarantees
the existence of a sequence (Uk) satisfying (8.20).
126 Chapter 8

Finally, assume that E satisfies the (PS)R condition. Then we may

assume, passing if necessary to a subsequence, that Uk ----t U for some u E
BR. Then
E (u) = CR, E' (u) = 0, (8.30)
or
lui = R, E (u) = CR, E' (u) - R- 2 (E' (u), u) u = 0, (E' (u), u) ::; 0,
(8.31 )
respectively. If in (8.31) (E' (u), u) = 0 then u also satisfies (8.30). Assume
(8.31) holds with (E' (u), u) < 0 and let f-L = _R-2 (E' (u), u) . Then f-L> 0
and E' (u) + f-LU = 0, which contradicts (8.17).
Notice that u i= Uo and u i= Ul since by (8.14),
E (u) = CR ~ inf {E (w) : Iwi = r} > max {E (uo) , E (Ul)} .
The proof is complete. •
Remark 8.5 Theorem 8.2 remains true if in (8.14) one has equality. More-
over, if
inf{E(w): wEBR, Iwl=r}=cR
then a sequence (Uk) like that in Theorem 8.2 can be found such that
d(Uk,aBr ) ----t o.
Consequently we obtain a critical point u with lui = r (see Schechter [40],
pp 103-104).
Remark 8.6 For R = 00, the proof of Theorem 8.2 represents another
proof of Theorem 8.1 based on deformation arguments.
The next result is also owed to Schechter (see [40]' Corollary 5.4.4) and
guarantees the existence of at least two distinct critical points.
Theorem 8.3 (Schechter) Assume all the assumptions of Theorem 8.2
hold and that E satisfies the (PS)R condition. In addition assume
mR = irg E (u) > -00.
uEBR

Then E has at least two different critical points u, Urn such that
uEBR\{UO,UI}, E(U)=CR
and
Urn E BR, E (urn) = mR·
Moreover, if E (Ul) ::; E (uo) then we may assume that Urn i= Uo·
The Mountain Pass Theorem 127

For much more information on critical point theory we refer the reader
to Ambrosetti [2], Benci [4], Brezis-Nirenberg [6]' Kavian [IS]' Mawhin-
Willem [23], Rabinowitz [36] and Struwe [42]. For nonsmooth critical point
theory, see Canino-Degiovanni [7], Chang [S], Corvellec [1o], Frigon [13] and
Motreanu-Varga [27].
 Chapter 9

Nontrivial Solutions
of Abstract
Hammerstein Equations

This chapter deals with nontrivial solvability in balls of abstract Hammer-

stein equations and Hammerstein integral equations in R n by a variational
approach. The variational method for treating Hammerstein integral equa-
tions goes back to the papers of Hammerstein [17] and Golomb [15]. For fur-
ther contributions see the monographs by Krasnoselskii [19], Krasnoselskii-
Zabreiko-Pustylnik-Sobolevskii [20] and Vainberg [43]. For more recent
results see the papers by Moroz-Vignoli-Zabreiko [24] and Moroz-Zabreiko
[25], [26]. The results presented in this chapter were adapted from Precup
[31]-[34].

9.1 Nontrivial Solvability of Abstract

Hammerstein Equations
Here we discuss the abstract Hammerstein equation

u = AN (u), U E Y, (9.1)
where Y is a Banach space, N Y --+ y* and A y* --+ Y is linear.
Assume that A splits into

A = HH* with H: X --+ Y and H* : y* --+ X, (9.2)

{
where X is a Hilbert space.

129
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
130 Chapter 9

Then (9.1) can be converted into an equation in X, namely

v = H* N H (v), v E X. (9.3)

Indeed, if u solves (9.1) then v = H* N (u) is a solution of (9.3), and

conversely if v solves (9.3) then u = H (v) is a solution of (9.1). Moreover,
H realizes an one-to-one correspondence between the solution sets of the
two equations (use a similar argument as in the proof of Proposition 6.6).
If, in addition, we assume

N = J' for some J E C 1 (Y; R), J (0) = 0, (9.4)

and
H is bounded linear and H* is the adjoint of H, (9.5)
then (9.3) is equivalent to the critical point problem

E' (v) = 0, v EX
for the energy (or Golomb) functional

1 2
E: X ---t R, E (v) = 21vlx - JH (v).

Here 1.lx stands for the norm of X. Notice E E Cl (X; R) and

E' (v) = v - H* N H (v), v E X.

We now state an existence principle for (9.1) in a ball of X.

Theorem 9.1 Assume (9.2), (9.4) and (9.5) . Assume that N (0) = and
the functional (N (.),.) sends bounded sets into upper bounded sets. In
°
addition assume that there are VI E X \ {O}, r E (O,IVll) and R 2: IVll
such that the following conditions are satisfied:

max { 0, 21vl1x
1 2 - JH (VI) }<.mf {121vlx -
2 JH (v) : Ivlx = r } , (9.6)

v -::J )"H*NH (v) for Ivlx = R, ).. E (0,1), (9.7)

E satisfies the (PS)R condition. (9.8)
Then there exists a v E X \ {O} with Ivlx ::; R such that u = H (v) is a
non-zero solution of (9.1).
Nontrivial Solutions of Abstract Hammerstein Equations 131

Proof. We apply Theorem 8.2. It is clear that (9.6) guarantees (8.14),

where Uo = 0, UI = VI, and (9.7) is equivalent to (8.17). Also (8.13) holds.
Indeed, for v E 8B R we have

(E' (v) , v) (v - H* N H (v) , v)

R2 - (N H (v) , H (v)) ,

and since H is bounded and (N (.) ,.) sends bounded sets into upper
bounded sets, there exists a lIO > 0 such that

(N H (v) , H (v)) :S R2 + lIo, v E 8 B R,

equivalently, (8.13) holds. Thus, also taking into account (9.8), all the as-
sumptions of Theorem 8.2 are satisfied. •
From Theorem 8.3 we can derive the existence of two solutions for (9.1).

Theorem 9.2 Assume all the assumptions of Theorem 9.1 hold. In addi-
tion assume that J sends bounded sets into upper bounded sets and

1 2
"2IVIlx - JH (VI) :S O.

Then (9.1) has at least two non-zero solutions U E Y of the form u = H (v)
with Ivlx :S R.

Proof. Apply Theorem 8.3. •

Remark 9.1 (The energy space) Assume (9.2), (9.4), (9.5) hold and H:
X ----> Y is injective. The space

XH:= H(X) cY
is a Hilbert space under the scalar product (., .)H defined by

(H (VI)' H (V2))H = (VI, V2) (VI, v2 EX),

and is called the ener9Y space generated by H. The corresponding energy

norm is given by
IH (v)IH = Ivl (v E X).
Since H: X ----> Y is bounded linear,

IH (v)ly :S IHllvi x = IHIIH (v)IH

132 Chapter 9

for all vEX. Hence

the embedding X H C Y is continuous. (9.9)

If u E Y is a solution of (9.1), then u = H (v) for v = H* N (u) E X, so

u E XH. Thus the solutions of (9.1) belong to the energy space XH. In
addition, u E X H solves (9.1) if and only if u = H (v) and v = H* N (u)
solves (9.3) in X, that is

(vJ,)=(H*NH(v),~), ~EX,

or, equivalently,

(H(v),H(~))H=(NH(v),H(~)), ~EX.

Hence u E XH is a solution of (9.1) if and only if u satisfies the variational

identity
(U,W)H = (N (u) ,w), wE XH.
Now define the functional E : XH ~ R, by

1 2
E(u) = "2lulH - J(u), u E XH.

Since J E C1 (Y; R), by (9.9), J E C1 (XH; R) and has the same derivative
N. Hence

(E' (u) ,w) = (u,w)H - (N(u) ,w) (u,w E XH).

Thus u E XH is a solution of (9.1) if and only if u is a critical point of E.

Therefore, a direct variational approach to (9.1) is possible in the energy
space XH.

9.2 Nontrivial Solutions of Hammerstein Integral

Equations
In this section we shall specialize Theorems 9.1 and 9.2 for the Hammerstein
integral equation in R n

u (x) = In /'i, (x, y) f (y, u (y)) dy a.e. on n. (9.10)

Nontrivial Solutions of Abstract Hammerstein Equations 133

Theorem 9.3 Let 0 C RN be bounded open, 2::; p < Po < 00, l/p+l/q =
1, l/po + l/qo = 1, /'i,: 0 2 ~ Rand f : 0 x Rn ~ Rn. Assume that the
following conditions are satisfied:
(i) the operator A: LqO (0; Rn) ~ LPO (0; Rn) given by

A(u) (x) = in /'i,(x,y)u(y)dy

is bounded and its restriction A: L2 (0; Rn) ~ L2 (0; Rn) is positive, self-
adjoint, and completely continuous;
(ii) f is (p, q)-Caratheodory of potential F and f (x, 0) = 0 a.e. on 0;
(iii) there are VI E L2 (O;Rn) \ {O} and r E (0, IV112) such that

max { 0, ~ IVII~ -in F (y, H (VI)(Y)) dY} (9.11)

< inf{~lvl~-inF(y,H(V)(Y))dY: VEL2(0;Rn), Ivb=r};

(iv) there is R 2: IVI12 such that

H (v) f= A in /'i, (., y) f (y, H (v) (y)) dy (9.12)

for every v E L2 (0; Rn) with Ivb = R and all A E (0,1).

Then the Hammerstein equation (9.10) has at least one non-zero solution
u in LP (0; Rn) of the form u = H (v) with v E L2 (0; Rn), Ivl2 ::; R.

Proof. Apply Theorem 9.1 to N = Nf and J given by

J (u) = in F (y, u (y)) dy (u E LP (0; Rn)) .

Since Nf is a bounded operator, the map (Nf (.),.) sends bounded sets
into bounded sets. By (iii), (iv), conditions (9.6) and (9.7) hold trivially. It
remains to show that the attached functional E satisfies the (PS)R condition.
To do this let (Vk) be any sequence of elements in L2 (0; Rn) with

0< IVkl2 ::; R

satisfying
E (Vk) ~ /-l E R, (E' (Vk) ,Vk)2 ~ l/ ::; 0
134 Chapter 9

and
E ' (Vk ) - (E' (Vk) 2, Vk)2 Vk --t
0
. (9.13)
IVkl2
We may assume that IVkl2 --t a, for some a E [0, R]. If a = 0 we have
finished. Assume a E (0, R]. Then

(E' (Vk) , Vk)2

~-'--~2;;---
--t ~
a2
E (-00,0].
IVkl2
On the other hand, (Vk) being bounded and H being completely continuous
(see Theorem 6.1), we may assume, passing eventually to a subsequence,
that (H (Vk)) converges. Then since N f and H* are continuous we deduce
that (H* NfH (Vk)) converges too. Now from (9.13) which can be written as

f k - (E'(Vk),Vkh
Vk - H *NH(v) 2 Vk --t
0
IVkb
we infer that the corresponding subsequence of (Vk) is convergent. Thus E
satisfies the (PS)R condition. •
Theorem 9.2 yields the following result.

Theorem 9.4 Assume all the assumptions of Theorem 9.3 hold. In addi-
tion assume that
r F (y, H (VI) (y)) dy ::; O.
"21 IVIb2 - in (9.14)

Then (9.1) has at least two non-zero solutions u E LP (0; Rn) of the form
u = H (v) with IvI 2 ::; R.

Proof. Notice J maps bounded sets into bounded sets as follows from
(7.7). The result is now a simple consequence of Theorem 9.2. •

Remark 9.2 If p> 2, a sufficient condition for (iii) is the following one:
(iii*) there is a VI E L2 (0; Rn) \ {O} such that (9.14) holds,

If (x, z) I ::; a ( 1 + IZ IP - I ) (9.15)

for a.e. x E 0 and all z ERn, where a E R+, and

lim If (x, z)1 (9.16)

z_Q Izl = 0
Nontrivial Solutions of Abstract Hammerstein Equations 135

uniformly for a.e. x E O.

Indeed, (9.16) implies that for any given c > 0 there is a 6 = 6 (c) > 0
such that
If (x, z) I :S c Izl for a.e. x E 0 and Izl < 6.
For Izl 2: 6, from (9.15) we deduce

If (x, z)1 :S a (';1 + IzIP-I)

< a (C;lr + I I Z P- 1 )

c(E) Izlp-I .
Here c (E) = a (6 1- p + 1) 2: o. Hence
If (x, z)1 :S 10 Izl + c (c) Izl p- I
for a.e. x E 0 and all z ERn. Now using (7.7) we obtain

1
E (v) 21vl~ - JH (v)

> ~ Ivl~ - k IH (v) (x)llf (x, BH(v),o (x) H (v)(x)) I dx

> ~ Ivl~ - k IH (v)1 (10 IH (v)1 + c (c) IH (v)I P - I ) dx

1 2 2
21vl2 - 10 IH (V)12 - c (E) IH (v)I~·

Since

IH (v)l~ (H (v) ,H (V))2 = (H* H (v), Vh

< IH*IIHllvl~
and
IH (v)lp :S c Ivb
because H is bounded, we deduce
1
E (v) 2: 21vl~ - 10 IH*IIHllvl~ - c (c) Ivl~
(~- cIH*IIHI- C(E) Ivl~-2) Ivl~.
136 Chapter 9

N ow we choose any c: > 0 such that

1
"2 - c: IH* IIHI > 0
and an r > 0 small enough so that
1
r < IVlI2, "2 - c: IH*IIHI- c(c:) r P- 2 > 0
(recall p > 2). Then
E(v) > 0 for Ivl2 = r.
This together with (9.14) guarantees (9.11).

Remark 9.3 Assume H : £2 (0; Rn) -7 £P (0; Rn) is bijective. Then,

by the theorem on the continuity of the inverse operator (see Brezis [5],
Corollary 11.6) there are constants a, (3 > 0 such that

a Ivl2 ::; IH (v)lp ::; (3lvI2' v E £2 (0; Rn).

Now a sufficient condition for (iv) is to exist an R 2': IVll2 with

IAlq,p (Igl q + C (3p-l RP-l) ::; aR.

Here IAlq,p is norm of A as operator from £q (0; Rn) to £P (0; Rn) , whilst
g E £q (0; R+) and c E R+ come from the (p, q)-Caratheodory property of
f, i.e.,
If (x, z)1 ::; g (x) + C Izl p- 1
for a.e. x E 0 and all z ERn.
Indeed, if v E £2 (0; Rn) satisfies Ivl 2 = R then

IANfH (v)lp ::; IAlq,p INfH (v)lq

< IAlq,p (Igl q + C IH (v)I~-l)
< IAI q,p (Igl q + C (3p-l RP-l)
::; aR
< IH (v)lp·
This guarantees (9.12) for all A E (0,1) .

Similar results for the existence of solutions in a ball of £p (0; Rn) are
presented in Precup [31], [32] by means of a mountain pass theorem on closed
convex sets owed to Guo-Sun-Qi [16].
Nontrivial Solutions of Abstract Hammerstein Equations 137

9.3 A Localization Result for Nontrivial Solutions

The results presented in the previous two sections can also be viewed as
theorems of localization of the nontrivial solutions in balls. In this section
we try to localize solutions in a shell. The main idea for the localization of
a solution to equation (9.1) in a shell of X consists in finding two numbers
Ro and R with 0 < Ro < R such that all the assumptions of Theorem 9.1
hold in the ball of X of radious Rand (9.3) has no solutions v satisfying
o < Ivlx < Ro. Then, of course, (9.1) will have a solution U E Y of the
form U = H (v), with v E X and

Ro :S Ivl x :S R.
Here we present just an example of a problem for which such a localization is
possible. The ideas in this section were adapted from Precup [34]. Naturally,
an open problem is to give extensions for other nonlinear problems.
The problem we deal with is the semi-linear Dirichlet problem

-flu = lul p- 2 u in 0,
{ (9.17)
u=0 on 80.

It is known (see Struwe [42], Theorem II.6.6) that if 0 c RN is a smooth

bounded open set, N 2: 3 and 2 < p < 2* = 2N/ (N - 2), then (9.17)
admits an unbounded sequence (Uk) of solutions Uk E W~,2 (0). Here we
shall only localize one of the solutions, in a shell.
In context we shall obtain an upper bound for the quantity

>"p-1 = inf {In lul p

-
2
lVul 2 dx: U E c 1 (0) \ {O}, } (9.18)
(In lu1P2/2 dX) 2/p U = 0 on 80 .

For p = 2, >"1 is the first eigenvalue of the Laplacian -fl under the
Dirichlet boundary condition, and 1/>"1 represents the best constant in
the Poincare's inequality (see Precup [30], Theorem 3.59). For p > 2 and
N = 1 this quantity arises in the study of compactness properties of integral
operators on spaces of functions with values in a Banach space (see Precup
[33]).

The topics in this section require a previous knowledge of linear partial

differential equations.
138 Chapter 9

We now recall some well known results from the theory of linear elliptic
boundary value problems.
(PI) Let 0 C RN be a bounded open set with C 2-boundary. The Lapla-
cian -~ is a self-adjoint operator on L2 (0) with domain W 2,2 (0) n
W~,2 (0) (see Precup [30]' Theorem 3.33). It can be regarded as a con-
tinuous operator from w 2,q (0) n w~,q (0) to Lq (0) for each q E (1,00) .
Moreover, -~ is invertible and

A:= (_~)-1

is a bounded linear operator from Lq (0) to w 2,q (0) (see Brezis [5], The-
orem 9.32). Also, A as an operator from L2 (0) to L2 (0) is positive,
self-adjoint, and completely continuous.
(P2) (Sobolev embedding theorem) Let 0 C RN be a bounded open set
with Lipschitz boundary, kEN, 1 :S q :S 00. Then the following holds:
(1°) If kq < N we have

wk,q (0) C U (0) (9.19)

and the embedding is continuous for r E [I, N q/ (N - kq)]; the embedding

is compact if rE [l,Nq/(N-kq)).
(2°) If kq = N, then (9.19) holds with compact embedding for r E
[1,00).
(3°) If 0 :S m < k - N/q < m +1 we have
wk,q (0) C Cm,a (0) (9.20)

and the embedding is continuous for 0 :S a :S k - m - N / q; the embedding

is compact if a < k - m - N/q.
The above results are valid for w;,q (0) spaces on arbitrary bounded
domains 0 (see Adams [1], Struwe [42] p 213, or Brezis [5] pp 168-169).
(P3) Let 0 C RN be a bounded open set with C2-boundary. Let Po =
2N/ (N - 2) if N::::: 3 and Po be any number of (2,00) if N = 1 or N = 2.
Let qo be the conjugate of Po. Clearly, Po E (2,00) and qo E (1,2). From
(PI), (P2) we have that A has the following properties:
(a) A: Lq (0) ---) LP (0) for every q E [qo, 2], l/p + l/q = 1;
(b) A is continuous from Lq (0) to £P (0) for every q E [qo,2] , l/p +
l/q = 1;
(c) as a map from L2 (0) to L2 (0), A is a positive self-adjoint oper-
ator.
Nontrivial Solutions of Abstract Hammerstein Equations 139

Indeed, A is continuous from Lq (0) to w 2 ,q (0). On the other hand

w 2 ,q (0) c LP (0)

with continuous embedding. This is clear if q 2: N /2. For q < N /2 and

l/p + l/q = 1, observe that

2N Nq
P < - - '¢=}p<--
- N - 2 - N - 2q

According to Theorem 6.2 the properties (a )-( c) are sufficient for that the
operator A considered from Lq (0) to £P (0), where p E (2,po) and
1/p + 1/ q = 1, admits a representation in the form
A=HH*

where
H : L2 (0) ----t LP (0), H (v) = Al/2 (v)
and
H* : L q (0) ----t L 2 (0)
is the adjoint of H. Notice H is injective since A is positive defined.
In what follows by IHI we shall mean

IHI = sup{IH (v)lp : v E L2 (0), Ivl2 = 1}.

Our first result gives a lower bound for all nontrivial solutions of (9.17).

Theorem 9.5 Let 0 be a bounded open set of RN with C 2 -boundary and

let p E (2,2N / (N - 2)) if N 2: 3, p E (2,00) if N = 1 or N = 2, and let
q be the conjugate of p. If u E w 2 ,q (0) n W5,q (0) is a non-zero solution
of (9.17) , then the function

v = H* (luIP-2 u) = H- 1 (u)
satisfies
Ivl2 2: IHI- 1 [(p - 1) Ap _l]1/(p-2) . (9.21 )

Proof. Let us first prove that any solution of (9.17) belongs to C 1 (0) .
For N = 1 this follows from (9.20) (choose a = 0, m = 1 and k = 2).
Assume N 2: 2 and fix any number qo > N (p - 1). If q 2: N /2 then
(P2) guarantees u E LqO (0) . Assume q < N /2 and denote ql = q. Since
u E W 2 ,ql (0) and ql < N/2, from (9.19) we have u E Lqi (0), where
140 Chapter 9

qi = N qd (N - 2qr) . Then lul p- 2 u E Lqi!(p-1) (0). Let q2 = qil (p - 1) .

Since u = A (luIP-2 u) and lul p- 2 u E Lq2 (0), from (PI) we have that
u E w2m (0). If q2 ~ N12, as above u E LqO (0) ; otherwise we continue
this way. At the step j we find that

~ 1
qJ- * Nqj-1
u E w 2,qj (0), qj = p _ l' %-1 = N _ 2qj-1 ' (9.22)

where q1,q2, ... ,qj-1 < NI2 (j ~ 2). We claim that there exists a j with
qj ~ N12. To prove this, assume the contrary, that is qj < NI2 for every
j ~ 1. Using p < 2NI (N - 2) we can show by induction that the sequence
(qj) is increasing. Consequently, qj ~ if E [q, N12] as j ~ 00. Next, from
(9.22) we obtain
qj (N - 2qj-1) (p - 1) = Nqj-1.
Letting j ~ 00 this yields if (N - 2if) (p - 1) = N if and so

N (p - 2)
if
2 (p - 1)
p
> q=-.
p-l

This implies p ~ 2NI (N - 2), a contradiction. Thus our claim is proved.

Therefore u E LqO (0). Furthermore, lul p - 2 u E Lqo/(p-1) (0), and since
u A (luIP-2 u) we have u E w 2,qo/(p-1) (0). Since qol (p - 1) > N, by
=

(9.20) one has w 2,qo/(p-1) (0) C C1 (0) (choose cy = 0, k = 2, m = 1).

Hence u E C1 (0) as claimed.
Let u = A (luIP-1), where u is a non-zero solution of (9.17) and let
v= (u). Clearly, like u,
H- 1 uE C1 (0) and u = 0 on 80. By the weak
maximum principle we have

lui :s; u on o. (9.23)

Hence
-!j.u = lul p- 1 :s; lul p - 2 u.
If we 'multiply' by uP- 1 and 'integrate' on 0, we obtain

(p - 1) in u p- 2 lVu l2 dx:S; in lul p- 2uP dx. (9.24)

Nontrivial Solutions of Abstract Hammerstein Equations 141

Now, Holder's inequality yields

r
in lul p- 2 v;Pdx:s
(inru p2 / 2dx )2/p ( inr lulPdx )(p-2)/p (9.25)

IH (v)I~-2 (In uP2/2dX) 2/p

Since IH (v)lp :S IHllvb and by (9.23) one has u # 0, from (9.24) and
(9.25) we deduce that

(p - 1) Ap-l :S IHIP-2Ivl~-2 ,
that is, (9.21) . •
The main result in this section is the following existence and localization
theorem.

Theorem 9.6 Let 0 be a bounded open set of RN with C 2-boundary and

letpE(2,2Nj(N-2)) ifN23, andpE(2,oo) ifN=1 orN=2.
Then the problem (9.17) has a solution u with

IHI- 1 [(p -1) Ap_l]1/(P-2) :S IH- 1(u)12 :S IHI- p/(p-2). (9.26)

Proof. First notice the left inequality in (9.26) is true for all non-zero
solutions of (9.17) according to Theorem 9.5. Next we prove that for each

R >r = IHI- p/(p-2) ,

(9.17) has a solution u such that

IH- 1(u)12 :S R. (9.27)

The equation (9.17) can be put under the form (9.1), where N = N f and

f (x, z) = Izlp-2 z (x E 0, Z E R).

Also
1
F (x, z) = -
p
IZ IP .
Thus the Golomb functional E: L2 (0) R is given by

1 11 IH
--t

E (v) 21vl22 - - (v)IP dx

p n
1 2 1
21 vb - PIH (v)I~·
142 Chapter 9

As in the proof of Theorem 9.3 we can show that E satisfies the (PS)R
condition.
If for this R the Leray-Schauder boundary condition (9.7) does not
hold, then there are v E £2 (0) and ,\ E (0,1) such that Ivl2 = Rand

v = '\H* N H (v) .

It is easy to see that the function v= ,\1/(p-2)v satisfies

v= H*NH(v),

i.e., v is a critical point of E. Clearly

o < Ivl2 < Ivl2 = R.

Then u = H (v) is a solution of (9.17) and satisfies (9.27). Assume now
that (9.7) holds for this R. We claim that the energy inequality (9.6) also
holds with a suitable Vl E £2 (0) satisfying r < IVlb :S R. To prove this
we fix any ,\ E (r, R] and we choose a p> 0 sufficiently small so that
1
¢ (,\) + -,\P P < ¢ (r) , (9.28)
p

where
¢ ((T) = ~(T2 - t(TP IHI P , (T E R+.

Notice r is the maximum point of ¢, ¢ is increasing on [0, r] and decreasing

on [r,oo). Now we choose a function V2 E £2 (0) with IV212 = 1 and

IH (v2)1~ ~ IHI P - p.

Then (9.6) holds for Vl = '\V2. Indeed

1 2 1
E (Vl) E ('\V2) = -,\
2
- p-).P IH (V2)IP
P
(9.29)

1 2 1 P 1 1
< -,\ --,\PIHI +-,\Pp=¢(,\)+-,\Pp.
2 p p p

Also, for every v E £2 (0) with IvI2 = r we have

E (v) 1 2-
-r 1 P IH (r- 1v) IP
-r (9.30)
2 p P
1 1
> _r2 - -rP IHI P = ¢(r).
2 p
Nontrivial Solutions of Abstract Hammerstein Equations 143

Now (9.29), (9.30) and (9.28) guarantee (9.6). Thus all the assumptions
of Theorem 9.1 hold and so there exists a non-zero critical point v of E
with Ivl2 ~ R. Therefore for each R > r (9.17) has a solution U E LP (0)
satisfying (9.27).
Finally, for each positive integer k we put R = r + 11k to obtain a
solution Uk with
IH- 1 (uk)12 ~ r + ~,
and the result will follow via a limit argument. •

Remark 9.4 From (9.26) we have the inequality:

1
Ap-1 ~ (p _ 1) IHI 2 .

The next inequality of Poincare type shows that

Ap-1 > 0

for p E [2, 2NI (N - 2)] if N 2': 3 and for p E [2, (0) if N = 2. Moreover,
its proof connects Ap-1 to the embedding constant of W~,2 (0) into V (0) .

Theorem 9.7 Let 0 C RN be bounded open and let p E [2, 2NI (N - 2)]
if N 2': 3, and p E [2,(0) if N = 2. Then there exists a constant c > 0
depending only on p and 0, such that

(in lu1P2/2 dX) 2/p in

~ c lulp- 21V'uI 2dx (9.31 )

for all u E c 1 (0) with u = 0 on 80.

Proof. According to (P2), we have W~,2 (0) c LP (0) with continuous

embedding. Hence there exists a constant Co > 0 such that

Ivlp ~ Co Ivlw~,2(n) for all v E W~,2 (0) .

in lV'vl2
Here
1/2
Ivlw~,2(n) = (
dx
)

Since Co (0) is dense in W~,2 (0) we may assume that

Co = sup {Ivlp : v E Co (0) , Ivlw~,2(n) = I} .

144 Chapter 9

The space Co (D) is also dense in {u E C1 (D) : u = 0 on 8D}, and so

Ap-1

(,up { (10 luIP' / ' dx

After substituting v
t P
: u E cQ'(n),

= (2/p) lul P/ 2 we obtain

10 lulP-'IVul' dx = I} r 1

Ap-1 = (P2)2 (sup {Ivl; : v E Co (D), Ivlw~,2(n) = 1})

-1

(p~o)2
Thus (9.31) holds with the smallest constant

c = _1_ = (PCO)2
Ap-1 2 '

and the proof is complete. •

For other applications of variational techniques to integral equations and
boundary value problems for various types of equations, see Conti-Merizzi-
Terracini [9], Dindi-Jebelean-Mawhin [11], Mawhin-Willem [23]' Moroz-
Vignoli-Zabreiko [24], Moroz-Zabreiko [25], [26], Pascali [28], Precup [29],
Rabinowitz [36], Radulescu [37]' Ricceri [38], Struwe [42] and Zeidler [45].
 Part III

ITERATIVE METHODS
 Chapter 10

The Discrete Continuation

Principle

The Banach contraction principle was generalized by Perov (see Perov-

Kibenko [40]) for contractive maps on spaces endowed with vector-valued
metrics. Also, Granas [25] proved that the property of having a fixed point
is invariant under homotopy for contractions on complete metric spaces.
This result was completed in Precup [50] (see also O'Regan-Precup [38] and
Precup [51], [52]) by an iterative procedure of discrete continuation along
the fixed points curve. This chapter presents a variant for contractive maps
on spaces with vector-valued metrics, first given in Precup [54].

10.1 Perov's Theorem

Let X be a nonempty set. By a vector-valued metric on X we mean a map
d: X x X ----> Rn with the following properties:
(i) d(u,v) 2 0 for all u, v E X; if d(u,v) = 0 then u = v;
(ii) d (u, v) = d (v, u) for all u, v EX;
(iii) d(u,v) ~ d(u,w) + d(w,v) for all u, v, w E X.
Here, if x, Y ERn, x = (Xl, X2, ... , Xn), Y = (Yr, Y2, ... , Yn) , and c E R,
by X ~ Y we mean that Xi ~ Yi for i = 1,2, ... , n, whilst by X ~ c we
mean that Xi ~ c for all i. Also X < Y stands for Xi < Yi, i = 1,2, ... , n,
and X < c for Xi < C, i = 1,2, ... , n. For a vector X = (Xl, X2, ... , Xn) we
write X = 00 if Xi = 00 for all i.
A set X endowed with a vector-valued metric d is said to be a gener-
alized metric space. For the generalized metric spaces the notions of a con-

151
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
152 Chapter 10

vergent sequence, Cauchy sequence, completeness, open subset, and closed

subset are similar to those for usual metric spaces.

Definition 10.1 Let (X, d) be a generalized metric space. A map T :

X -+ X is said to be contractive if there exists a matrix M E Mnxn (R+)
such that
Mk -+ 0 as k -+ 00 (10.1)

and
d(T(u),T(v)):::; Md(u,v) (10.2)

for all u, VEX. A matrix M which satisfies (10.1) is said to be convergent

to zero.

Lemma 10.1 Let M E Mnxn (R+). The matrix M is convergent to zero

if and only if 1 - M is non-singular and

(1 - M)-l = 1 + M + M2 + .... (10.3)

Here 1 is the unit matrix of order n.

Proof. Assume first that M is convergent to zero. To show that 1 - M

is non-singular it suffices to prove that the linear system

(I - M) z = 0

has only the null solution. Assume z E R n is a solution of this system.

Then
z = M z = M2 Z = ... = Mk z = ...

and letting k -+ 00 we deduce z = O. Hence 1 - M is non-singular.

Furthermore, (10.3) follows from the identity

1- (I - M) (1 + M + M2 + ... + Mk) = Mk+l. (10.4)

Conversely, if 1 - M is non-singular and (10.3) holds, then from the con-

vergence of the series (10.3) we conclude that Mk -+ 0 as k -+ 00 . •

Remark 10.1 Property (10.1) is also equivalent to the property of the

eigenvalues of M that they are all located inside the unit disc of the complex
plane (see Rus [55], Theorem 4.1.1).
The Discrete Continuation Principle 153

Remark 10.2 From (10.3) we have that

z ~ (1 - M)-l z

for every z E R"t. In particular, the set

PM = {z E R n : z > 0 and (1 - M) z > O}

is nonempty and coincides with the set

{ (1 - M) -1 Zo : Zo ERn, Zo > 0 } .

Indeed, if Zo E Rn and Zo > 0 then

Z := (I - M)-l Zo 2: Zo,

and so Z > o. In addition (1 - M) Z = ZOo Hence (1 - M) Z > 0 and

so Z E PM. Conversely, if Z E PM then Zo := (I - M) z > 0 and z =
(1 - M)-l ZOo

Remark 10.3 Assume that M is a matrix convergent to zero. A subset

U c X is open if and only if for each u E U there is apE PM such that
v E U whenever d (v, u) ~ p (take p = cPO for any Po E PM and c > 0
sufficiently small).

Theorem 10.1 (Perov) Let (X, d) be a complete generalized metric space

and T : X -> X be contractive with the Lipschitz matrix M. Then T has
a unique fixed point u*, and for each Uo E X one has

d (Tk (uo) , u*) ~ Mk (I - M)-l d (uo, T (uo)) (10.5)

for all kEN.

Proof. Let Uo be any element of X. Define a sequence (Uk) by

Uk+1 = T(uk), kEN. (10.6)

Then from (10.2)

d(Uk,Uk+1) ~ M kd(uo,U1)
and, as a consequence,

d (Uk, Uk+m) ~ (Mk + Mk+1 + ... + M k+m- 1) d (uo, U1).

154 Chapter 10

Using (10.3) we deduce that

d (Uk, Uk+m) ::; Mk (I - M)-1 d (uo, Ul). (10.7)

Hence (Uk) is a Cauchy sequence. Let U* be its limit. Then from (10.6) we
have u* = T (u*) , whilst from (10.7) and letting m ----) 00 we obtain (10.5).
For uniqueness let Ul, U2 be two fixed points of T, then

d (Ul' U2) = d (Tk (Ul) ,Tk (U2)) ::; Mkd (Ul' U2) .

Since Mk ----) 0 as k ----) 00, this implies d (Ul, U2) = 0, so Ul = U2. •

For some applications of Perov's theorem to systems of integral equa-
tions, see Avramescu [3], Perov-Kibenko [40] and Rus [55].

10.2 The Continuation Principle for Contractive

Maps on Generalized Metric Spaces
The main abstract result of this chapter is the following theorem from Precup
[54].

Theorem 10.2 Let (X, d) be a complete generalized metric space where d:

X x X ----) Rn and let U be an open set of X. Let H: U x [0,1] ----) X and
assume that the following conditions are satisfied:
(a1) there is a matrix M E Mnxn (R+) convergent to zero such that

d(H(u,A),H(v,A))::; Md(u,v) (10.8)

for all u, v E U and A E [0, 1] ;

(a2) H (u, A) =1= u for all u E aU and A E [0,1];
(a3) H is continuous in A, uniformly for u E U, i.e., for each E E (0,00)
and A E [0,1] there is apE (0,00) such that

d (H (u, A) ,H (u, J-L)) < E

whenever u E U and IA - J-LI < p.

In addition assume that Ho := H (., 0) has a fixed point. Then for
each A E [0,1] there exists a unique fixed point u (A) of H>.. := H (., A) .
Moreover, u (A) depends continuously on A and there exists r E PM U { 00 } ,
integers m, kl' k2, ... , k m - 1 and numbers

o < Al < A2 < ... < Am-l < Am = 1

The Discrete Continuation Principle 155

such that for any Uo E X satisfying d (uo, U (0)) :S r, the sequences (Uj,ik::o,
j = 1,2, ... , m, defined by

Ul,O = Uo
Uj,i+l = HAj (Uj,i), i = 0, I, .. .
Uj+l,O = Uj,kj , j = 1,2, ... ,m-1

are well defined and satisfy

d(Uj,i, u(A.j)) :S Mi (I - M)-l d(uj,o, HAj(uj,o)), i E N. (10.9)

Proof. Step 1. First we prove that for each A. E [0,1] HA has a fixed
point. Let

A = {>.. E [0,1]; H (u, A.) = U for some U E U}.

We have 0 E A by the assumption that Ho has a fixed point. Hence A is
nonempty. We will show that A is both closed and open in [0, 1J and so,
by the connectedness of [0,1], A = [0,1].
To prove that A is closed let A.k E A with A.k ---; A. as k ---; 00. Since
A.k E A there is Uk E U so that H (Uk, A.k) = Uk. Then by (10.8) we obtain

d(Uk' Uj) d(H (Uk, A.k), H(Uj, A.j))

< d (H (Uk, A.k), H (Uk, A.))
+d(H (Uk, A.) ,H(Uj, A.)) + d(H(Uj, A.), H(Uj, A.j))
< d (H (Uk, A.k) ,H (Uk, A.)) + M d(uk, Uj)
+d(H(Uj, A.), H(Uj, A.j)).
It follows that

d(Uk, Uj) :S (I - M)-l [d (H (Uk, A.k), H (Uk, A.)) + d(H(Uj, A.), H(Uj, A.j))].
This shows by (a3) that (Uk) is a Cauchy sequence. Then since X is com-
plete there is a U E X with d(uk,u) ---; 0 as k ---; 00. Clearly U E U.
Then
d (Uk, H (u, A.)) ---; d (u, H (u, A.))
and, from (10.8) and (a3), we have

d (Uk, H (u, A.)) d (H (Uk, A.k) ,H (U, A.))

:S d (H (Uk, A.k), H (U, A.k)) + d (H (U, A.k), H (U, A.))
< Md(Uk'U) +d(H(u,A.k) ,H(u,A.)) ---;0.
156 Chapter 10

Hence d (u, H (u, >..)) = 0, that is H (u, >..) = u. By (a2) u E U, and so

>.. E A.
To prove that A is open in [0, 1] let f-L E A and w E U such that
H (w, f-L) = w. Since U is open there exists apE PM (see Remark 10.3)
such that
d (u, w) -::; p implies u E U.
Also, by (a3) there is TJ = TJ (p) E (0,00) such that
d (w, H (w, >..)) = d (H (w, f-L) , H (w, >..)) -::; (1 - M) p (10.10)
for I>" - f-LI -::; TJ· Consequently
d(w,H(u,>..)) -::; d(w,H(w,>..))+d(H(w,>..),H(u,>..))
-::; (I - M) p + M d (w, u) -::; p
whenever d (w, u) -::; p and I>" - f-LI -::; TJ· This shows that for I>" - f-LI -::; TJ,
H>.. sends B into itself, where B = {u EX: d (w, u) -::; pl. Notice (B, d)
is a complete generalized metric space. Hence for I>" - f-LI -::; TJ, by Theorem
10.1 H>.. has a fixed point u (>..) E B c U. This shows that f-L is an interior
point of A and hence A is open in [0,1]. Notice for every u E Band
I>" - f-LI -::; TJ, we also have by Theorem 10.1 that the sequence (H~ (u)k~o
is well defined and
d(Hf (u), u (>..)) -::; Mk (1 - M)-l d (u, H>.. (u)), kEN.
Step 2. The uniqueness of u (>..) is a consequence of (10.8). Indeed, if
u, v E U are fixed points of H>.. then since the elements of M are nonneg-
ative, we have
d (u, v) d (H (u, >..), H (v, >..))
-::; Md(u,v)
< M 2 d(u,v)

< Mkd(u,v) ~ 0 as k ~ 00.

Hence d(u,v) = 0, and so u = v.

Step 3. u (>..) is continuous on [0,1]. Indeed,
d (u (>..) , u (f-L) )
d (H (u (>..) , >..) , H (u (f-L) , f-L) )
-::; d (H (u (>..) , >..) , H (u (f-L) , >..)) + d (H (u (f-L) , >..) , H (u (f-L) , f-L) )
< M d (u (>..) , u (f-L)) + d (H (u (f-L) , >..) , H (u (f-L) , f-L)) .
 The Discrete Continuation Principle 157

By (a3) this implies

d (u (A), u (~)) :S (I - M)-l d (H (u (~), A) ,H (u (~), ~)) ~ 0

as A ~~.
Step 4. Finding r. For any ~ E [0,1] and each i E {I, 2, ... , n} let

ri(~)=inf{ddu,u(~)): UEX\U}.

Here d = (d 1 , d2, ... , dn ). Since u (~) E U and U is open, ri (~) > O. We

claim that
inf {rd~): ~ E [0, In > O. (10.11)
To prove this let us assume the contrary. There are then ~k E [0,1] such
that ri (~k) ~ 0 as k ~ 00. Clearly we may assume that ~k ~ ~ for some
~ E [0,1]. Then from the continuity of u (A) we have

1
di (u (~k), u (~)) < '2ri (~), k 2 k1 · (10.12)

On the other hand, since "ri (~k) ~ 0 as k ~ 00,

1
ri (~k) < '2ri (~) , k 2 k 2. (10.13)

Let ko = max {kl' k 2 }. By (10.13) and the definition of ri (~ko) as infimum,

there is u E X \ U with

1
di(U,U(~ko)) < '2ri(~)' (10.14)

Then, from (10.12) and (10.14), we obtain

di (u, U (~)) :S ddu, u (~ko)) + ddu (~ko)' u (~))

1
< 2· '2rd~) = ri (~) ,

a contradiction. Thus (10.11) holds as claimed. Now we choose any ri > 0

less than the infimum in (10.11), with the convention that ri = 00 if the
infimum equals infinity; this happens when U = X and so all the ri '8 are
finite or all equal infinity. Then take r = (rl' r2, ... , rn). If r E (O,oot we
may assume that r E PM (otherwise, replace it by the vector E (I - M)-l r
with E> 0 small enough so that E (I - M)-I r :S r).
158 Chapter 10

Step 5. Finding of m and 0 < A1 < A2 < ... , Am-1 < 1. If r = 00 take
h = 1; if r E PM let h = TJ (r), where TJ (r) is chosen as in (10.10). Then
by what was shown at the end of Step 1 for each f-L E [0,1] ,

d (u, U (f-L)) ::; rand IA - f-LI ::; h imply

{ (H~ (u)k::o is well defined and (10.15)
d(H~ (u), U (A)) ::; Mk (I - M)-l d (u, H>. (u)), kEN.

Now we choose any partition

o = AO < A1 < ... < Am-1 < Am = 1

of [0,1] such that Aj+1 - Aj ::; h, j = 0,1, ... , m -1 (it is clear that m = 1
when r = 00).
Step 6. Finding of the integers k1, k2, ... , km- 1. From

d (U1,0, U (0)) = d (uo, u (0)) ::; r, A1 - AO ::; h,

by (10.15) we have that (U1,i\>0 is well defined and satisfies (10.9). By

(10.9), we may choose k1 EN such that

d(U1,kl'U(A1))::; r.

Now
d(U2,0,U(A1)) = d(U1,kl,U(A1))::; r, A2 - A1::; h,
and we repeat the above argument in order to show that (U2,i)i>0 is well
defined and satisfies (10.9). In general, at step j (1 ::; j ::; m - 1) we choose
kj E N such that
d( Uj,k j , u( Aj)) ::; r.
Then since

d(Uj+1,o,U(Aj)) = d(Uj,kj,U(Aj))::; r, Aj+1- Aj::; h,

by (10.15) we have that the sequence (uj+1,ik~0 is well defined and satisfies
(10.9) . •
The above proof yields the following algorithm for the approximation of
U (1) under the assumptions of Theorem 10.2:

Suppose we know rand h and we wish to obtain an approximation Itt

of u(l) with
d (U1, U (1)) ::; E:
The Discrete Continuation Principle 159

for some c E (0,00). Then we choose any partition

o = '\0 < '\1 < '\2 < ... < ,\m-l < '\m = 1
of [0,1] with ,\)+1 -,\j ~ h, j = O,l, ... ,m -I, any element Uo with
d (uo, U (0)) ~ r and we follow the next:
Iterative procedure:
Set ko := 0 and UO,ko := Un;
For j := 1 to m - 1 do
Uj,O := Uj-l,kj_1
i:= 0
While Mi (1 - A)-l d(uj,o, H>.)uj,o)) i r
Uj,i+1 := H>'j(uj,i)
i := i +1
kj = i
Set i := 0
While Mi (I - M)-l d (um,o, HI (um,o)) i c
Um,i+1 = HI (Um,i)
i := i + 1
Finally take Ul = Um,i'
Notice for n = I, Theorem 10.2 reduces to Corollary 2.5 in Precup [50].
For other continuation results involving generalized contractions, see
Agarwal-O'Regan [1], Frigon [23] and Precup [53]. For some related topics
and applications, see Rus [56], ~erban [58] and Tsachev-Angelov [59].

10.3 Hammerstein Integral Equations with

Matrix Kernels
In this section we give an application of Theorem 10.2 to the Hammerstein
integral equation in R n

U(x) = k~(x,y)f(y,u(y))dy, x E 0, (10.16)

in the case that the kernel ~ has matrix-values, i.e.,

~ : 02 --t Mnxn (R), ~ = [~ij].

The usual Hammerstein equation in R n with a scalar kernel appears as

a particular case of (10.16) when ~ij = 0 for i -1= j and ~ii = ~jj for all
i,j E {l,2, ... ,n}.
160 Chapter 10

The simplest examples of problems which allow us to systems of the form

(10.16) with matrix kernels are the boundary value problems for differential
equations of order 2': 2. For instance, the problem

u"=g(x,u,u'), xE [0,1]
{
u (0) = 0, u' (1) = 0

can be put in the form (10.16) if we let n = 2, Ul = u, U2 = u',

/1;11 (x,y) = { 0,1, Y~x

y>x ' /1;22 (x,y) =
{O,-1, y ~X
y> x

/1;12 = /1;21 = 0,
and
II (x, ul, U2) = u2, h (x, ul, U2) = g (x, Ul, U2) .
Before we state the main result we introduce the following notations.
For an element Z E Rn, Z = (ZI' Z2, ... , zn) we let

IIZII = (I Z11, IZ21, ... , IZnl)·

Also, for a function u E V (0; Rn) (0 C RN open, 1 ~ p ~ 00), u
(Ul' U2, ... , un) we let

Ilulip= (IU1Ip, IU2Ip' ... , lunlp).

Clearly 11.11 , 11.llpare vector-valued norms in Rn and V (0; Rn), respectively.
Endowed with the vector-valued metric

dp (u,v) = Ilu - vllp'

V (0; Rn) is a complete generalized metric space. Similarly, if 0 c RN

is bounded open then (C (IT; R n) , doo ) is a complete generalized metric
space.
We now state and prove a general existence and uniqueness principle for
(10.16).

Theorem 10.3 Let 0 C RN be a bounded open set, /1;: 02~ Mnxn (R)
measurable and f : 0 x R n ~ Rn. Assume that there are p E [1,00],
The Discrete Continuation Principle 161

q E [1,(0), p 2 q, and an open subset U of (V(O;Rn),II.ll p ) containing

the origin, such that the following conditions are satisfied:

(a) if 1 ::; p < 00 then K,ij (x,.) E U (0) for a.e. x E 0, and
{ the map x f--+ lK,ij (x, .)Ir belongs to LP (0) (ljq + 1jr = 1) ;
(b) ifp = 00 then K,ij (x,.) E U (0) for every x E 0, and
the map x f--+ K,ij (x,.) is continuous from 0 to U (0) ;
(10.17)
f satisfies the Caratheodory conditions, f (.,0) E Lq (0; Rn) and
{ Ilf(x,zl) - f(x,z2)11::; L(x) IiZl- z211 for a.e. x E 0,
all Zl, Z2 ERn, and some L E Vq/(p-q) (0; Mnxn (R+)).
(10.18)
Let M = [MikJ ,
n
Mik = L /1K,ij (x, ·)Ir/p ILjklpq/(p_q)
j=l
and assume M is convergent to zero. In addition assume that

uEU (10.19)

for any solution u E U to

u(x) =A 10 K,(x,y)f(y,u(y))dy a.e. on 0, (10.20)

for each A E [0, 1J. Then (10.16) has a unique solution u E U C V (0; Rn) .
Moreover, u E C (0; Rn) for p = 00.

Proof. We note that by pqj (p - q) we mean 00 if p = q, and q if

p = 00. We apply Theorem 10.2 to X = LP (0; Rn) with vector-valued norm
11.ll p and H: U x [O,lJ ----t V (0; Rn) given by

H (u, A)(X) =A In K, (x, y) f (y, u (y)) dy (x E 0).

From (10.18) we have

Ilf (x, z) II ::; Ilf (x, 0) II + L (x) Ilzll·

Hence n
Ifd x , z)1 ::; Ifdx, 0)1 +L Lij (x) IZjl· (10.21)
j=l
162 Chapter 10

By Young's inequality

Lij (x) IZjl ::; p - q Lij (X)p/(p-q) + !J.IZjIP/q.

p p

Since fi (.,0) , Lij (.)p/(p-q) E Lq (0), from (10.21) we obtain that

If (x, z) I ::; g (x) + c Izi P/ q

for some g E Lq (0) and c ~ o. Hence f is (p, q)-Caratheodory, and so
the Nemytskii operator associated to f maps LP (0; Rn) into Lq (0; Rn) .
From (10.17) we see that the Fredholm linear integral operators of kernels fLij
maps Lq (0; Rn) into LP (0; Rn). Hence H is well defined. Furthermore,

IHi (u, >.) (x) - Hi (v, >.) (x)1

< 1 t
n j=l
IfLij (x,y)llfj (y,u(y)) - Ii (y,v (Y))1 dy

< 1 t
n j=l
IfLij (x,y)1 tLjdy) IUk (y)
k=l
- vdy) I dy
n n
< I: I: IfLij (x, ·)Ir ILjklpq/(p_q) IUk - vkl p ·
k=lj=l
Consequently
n n
IHdU, >.) - Hi (v, >')Ip ::; I: I: IlfLij (x,.) Ir Ip ILjklpq/(p_q) IUk - vklp
k=lj=l
n
I: Mik IUk - vkl p ·
k=l
Thus
IIH (u, >.) - H (v, >')lIp ::; M Ilu - vll p ·
Now the conclusion follows from Theorem 10.2. The cases p = 00 and p = q
are left to the reader. •
Notice that for Volterra-Hammerstein integral equations with matrix
kernels one can use a similar argument to that in the proof of Theorem 5.6
in order to apply directly Perov's theorem.
For an extension of Theorem 10.2 to spaces with two vector-valued met-
rics and some applications to systems of abstract Hammerstein integral equa-
tions, see O'Regan-Precup [39].
 Chapter 11

Monotone Iterative Methods

The basic notion in this chapter is that of an ordered Banach space. We

try to localize solutions of an operator equation u = T (u) in an ordered
interval [uo, vol of an ordered Banach space X. In addition we look for
solutions which are limits of increasing or decreasing sequences of elements
of X. The basic property of the operator T is monotonicity. This com-
bined with certain properties of the ordered Banach space X guarantees
the convergence of monotone sequences. Thus we may say that this chapter
explores the contribution of monotonicity to compactness.

11.1 Ordered Banach Spaces

There is a vast literature in ordered linear spaces and in ordered Banach
spaces, in particular. We refer the reader to the books of Cristescu [17] and
to Chapter 6 in Deimling [20]. Our aim for this section is just to introduce
the reader to the theory of ordered Banach spaces and of monotone (in
the sense of order) nonlinear operators. We shall restrict ourselves to those
notions and results which will be used in the next sections.

Definition 11.1 Let X be a linear space. By a cone K of X we un-

derstand a convex subset of X such that >"K c K for all >.. E R+ and
K n ( - K) = {O} .

The proof of the following proposition is immediate and is left to the

reader.

Proposition 11.1 Let X be a linear space and K c X be a cone. The

163
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
164 Chapter 11

relation '5:.K on X defined by

U '5:.K v if and only if v - U E K,

is an order (reflexive, antisymmetric, and transitive) relation on X (called

the order relation induced by K), compatible with the linear structure of X,
i.e., whenever ui, Vi E X, Ui '5:.K Vi, i = 1,2, and A E R+, we have

Ul + U2 '5:.K VI + v2, AUI '5:.K AVI.

Conversely, if '5:. is an order relation on X compatible with the linear struc-

ture of X, then the set

K+={UEX: o '5:. u}
is a cone (called the positive cone) and '5:. = '5:.K+ .

The above proposition says that there is a one-to-one correspondence

between the cones of X and the order relations on X which are compatible
with the linear structure of X.
Throughout, we shall denote '5:.K by '5:. for short. Also sometimes we
shall write U < V to indicate that V - U E int K.

Definition 11.2 A linear space endowed with a cone (equivalently, with

an order relation compatible with its linear structure) is called an ordered
linear space.

Let (X, K) be an ordered linear space. For any u, v E X we define the

order interval [u, v] by

[U,v] (u+K)n(v-K)
{WEX:u'5:.w'5:.v}.

Clearly [u, v] is a convex set.

Definition 11.3 A Banach space endowed with a closed cone is called an

ordered Banach space.

In an ordered Banach space any order interval [u, v] is a closed convex

subset. Moreover,

o '5:. Uk, Uk ---+ U as k ---+ 00 implies 0 '5:. U.

Monotone Iterative Methods 165

Consequently

Uk :S Vk, uk --t U, vk --t v implies u:S v.

Thus in an ordered Banach space we may always pass to the limit in inequal-
ities. All these statements follow immediately from the property of the cone
of being closed. To have more we need additional properties of the cone.

Definition 11.4 Let (X, K) be an ordered Banach space and let :S be the
order relation induced by K. We say that:
1) K is a normal cone if

o :S Uk :S Vk, Vk --t 0 implies Uk --t O.

2) K is a regular cone if every increasing sequence which is bounded

from above is already convergent, i.e.

Uk :S Uk+l :S v for kEN implies (Uk) convergent.

We have the following characterization theorem of normal cones.

Proposition 11.2 Let (X,K) be an ordered Banach space with norm 1.1.
The following statements are equivalent:
(a) K is a normal cone.
( b) The norm 1.1 is semi-monotone, i. e., there exists a 'Y > 0 such that

o :S U :S v implies lul:S 'Y Ivl·

(c) inf {Iu + vi : u, v E K, lui = Ivl = I} > O.
Proof. (a)*(b): Since K is a normal cone there is a number 7] > 0
such that
o :S U :S v, Ivl = 7] implies lul:S 1. (11.1)
Indeed, otherwise we would find two sequences (Uk) and (Vk) with

1
o :S Uk :S vk, IVkl = k' IUkl > 1.
But this is contrary to the definition of a normal cone. Let 'Y = Ih. If
o :S U :S v then
o :S 7] Ivl- 1 U :S 7] Ivl- 1 v, 17] Ivl- 1 vi = 7].
166 Chapter 11

By (11.1) this implies

IT] Ivl- 1 ul ::; 1.

Hence lui::; T]- l lv l = 'Y Ivl· Thus 1.1 is semi-monotone.

(b)=}(a): Immediate.
(b)=}(c): For every u,v E K with lui = Ivl = 1, we have 0::; u::; u+v.
Since 1.1 is semi-monotone, this yields

1 = lui::; 'Y lu + vi·

Hence lu + vi 2 1h. Thus

inf{lu+vl: u,v E K, lui = Ivl = I} 2 ~ > o.

'Y
(c)=}(b): Assume (b) is not true. Then there are two sequences (Uk),
(Vk) with
0::; Uk ::; vk, IUkl > k IVkl·
Let
'ih = IUkl- 1 Uk, -vk = Ivk 1- 1 vk·
Also define
Wk = Ik -1-vk - -uk 1- 1 (k- 1-vk - -uk ) .

It is easy to check that

Wk E K, IWkl = IUkl = 1, IUk + wkl ----t O.

This shows that (c) does not hold. •

Proposition 11.3 Any regular cone K is a normal cone.

Proof. Assume the cone K is regular but not normal. Then statement
(c) in Proposition 11.2 does not hold. Hence we may find two sequences
(Uk) , (Vk) such that

1
Uk, Vk E K, IUkl = IVkl = 1, IUk + vkl ::; 2k .

Define the sequence (Wk) by

k
Wk = LUj.
j=l
Monotone Iterative Methods 167

Clearly we have Wk E K and Wk :S Wk+1' In addition

00

Wk :S I)Uj + Vj) E K.
j=1

Since the cone K is regular (Wk) is convergent. But

I W k+1 - wkl = IUk+11 = 1.

This contradiction proves that K is a normal cone. _

°
Remark 11.1 In an ordered Banach space with normal cone, any order
interval [u, v] is a bounded set. Indeed, if W E [u, v] then :S w-u :S v-u.
The norm being semi-monotone we have

IW - ul :S 'Y Iv - ul .
Example 11.1 Let Dc RN be bounded open. In (C (D;R n ) , 1.1(0)' the

°
set
K = {u E C (D; Rn) : u (x) ;?: on n}
is a normal cone. However, it is not regular; for example, take D = (0,1) ,
n = 1, and
kx, for x E [0, kJ
Uk ()
x - {
- 1, for x E [i,lJ .
Clearly Uk :S Uk+1 :S 1 but (Uk) does not converge in C [0, 1].

Example 11.2 Let Dc RN be open and let 1 :S p < 00. The set

K={UE£P(D;Rn): u(x);?:O a.e. on D}

is a regular cone of LP (D; Rn). For the proof use Beppo Levi's theorem
(see Brezis [7], p 55).

11. 2 Fixed Point Theorems for Monotone

Operators
Definition 11.5 Let X be a set endowed with an order relation :S, and
let T : D c X -) X be a map. We say that T is increasing (or isotone) if
T (u) :S T (v) for u, v E D with U :S v. The map T is said to be decreasing
(or anti-isotone) if T (v) :S T (u) for u, v E D with U:S v.
168 Chapter 11

We now state and prove the monotone iterative principle for increasing
operators in ordered Banach spaces.

Theorem 11.1 Let (X, K) be an ordered Banach space, [uo, vol c X an

order interval and T: [uo, vol ----t [uo, vol an increasing continuous operator.
Assume one of the following conditions:
(i) K is a regular cone.
(ii) K is a normal cone and T is completely continuous.
Then there exist u*,v* E [uo, vol such that u* ~ v*, T(u*) = u*,
T (v*) = v*, (Tk (uo)) is increasing and convergent to u*, and (Tk (vo))
is decreasing and convergent to v*. Moreover, any fixed point of T lies
between u* and v*.

Proof. Notice the increasing operator T maps [uo, vol into itself if and
only if
uo ~ T(uo) , T(vo) ~ Vo·
From
uo ~ T (uo) ~ T (vo) ~ Vo
we find
uo ~ T (uo) ~ T2 (uo) ~ T2 (vo) ~ T (vo) ~ vo,
and finally

uo ~ T(uo) (11.2)

< Tk (uo) ~ Tk+l (uo)

< Tk+l (vo) ~ Tk (vo)

< T (vo) ~ Vo·

Assume (i). Then (11.2) implies the convergence of the sequences (Tk (uo))
and (Tk (vo)). Let u*, v* be their limits. Then from

Tk+ 1 (uo) = T ( Tk (uo) )

we deduce that u* = T (u*) since T is assumed to be continuous. Similarly,

v*=T(v*).
Monotone Iterative Methods 169

Assume (ii). The order interval [uo, vol being bounded (see Remark
11.1) and T being completely continuous, we have that T ([uo, vo]) is rel-
atively compact. Consequently (Tk (uO))k>l has a convergent subsequence
">1' Let u* be its limit. Clearly u* E [uo, vol. We claim that
(Tkj (uo)) 1_
the whole sequence (Tk (uO))k>l converges to u*. Indeed, for each c > 0

1:
there is a jc such that -

ITkj< (uo) - u*1 :::; 'Y'

Here the constant 'Y > 0 comes from the semi-monotonicity of the norm.
On the other hand, for i 2: kj< we have
Tkj< (uo) :::; Ti (uo) :::; u*,
that is
0:::; Ti (uo) - Tkj< (uo):::; u* - Tkj< (uo).
Then
ITi (uo) - Tk je (uo)1 :::; 'Y lu* - Tk je (uo)l·
Hence
ITi (uo) - u*1 < ITi (uo) - Tkje (uo)1 + ITkje (uo) - u*1
< ('Y + 1) ITkje (uo) - u* I
< c.
Thus Ti (uo) ----t u* as claimed. Use similar arguments to prove the conver-
gence of (Tk (vo)).
Now assume that W E [uo, vol is a fixed point of T. Then
Uo :::; T (uo) :::; T (w) = w :::; T (vo) :::; vo
and in general
Tk (uo) :::; w :::; Tk (vo) .
Letting k ----t 00 we obtain u* :::; w :::; v* . •
We shall use the following terminology:
1) An element Uo for which Uo :::; T (uo) is said to be a lower solution
of the operator equation u = T (u) .
2) An element vo satisfying T (vo) :::; vo is said to be an upper solution
of the equation u = T (u) .
3) The fixed point u* given by Theorem 11.1 is the minimal solution of
the equation u = T(u) in [uo, vol , whilst v* is the maximal solution in
[uo,vol·
170 Chapter 11

Remark 11.2 Under the assumptions of Theorem 11.1, if u* = v* then

Tk (w) ---> u* for all wE [uo,vo].

The next result is an existence and uniqueness theorem.

°: : ;
Theorem 11.2 Let (X, K) be an ordered Banach space and let [uo, vol C
X be an order interval with Uo ::::; Vo such that there is a number A
with
A > 0, AVO::::; un. (11.3)
Assume T : [0, vol ---> X is increasing on [0, vo], continuous on [uo, vol ,
T ([uo, va]) C [uo, vol and that there is a function ¢ : (0,1) ---> R such that

¢ (A) > A, ¢ (A) T (u) ::::; T (AU) (11.4)

for all A E (0,1), u E [uo, vol. In addition assume one of the following
conditions:
(i) K is a regular cone.
(ii) K is a normal cone and T is completely continuous.
Then T has a unique fixed point u* in [uo, vo], (Tk (uo)) zs mcreasing
and convergent to u*, (Tk (vo)) is decreasing and convergent to u*, and

Tk (w) ---> u* for all w E [uo, vol .

Proof. Let u*, V* be the fixed point of T given by Theorem 11.1. By

(11.3),
AV* ::::; AVO::::; Uo ::::; u*.

Hence the set {). E (0,1] : AV* ::::; u*} is nonempty. Let

A* = sup {). E (0,1] : AV* ::::; u*}.

Clearly, A*V* ::::; u*. If A* = 1 we have finished. Assume A* < 1. Then

using (11.4) and the monotonicity of T on [0, vo], we obtain

¢(A*)V* ¢(A*)T(v*)
< T(A*V*)::::; T(u*) = u*.

Since ¢ (A *) > A*, this is in contradiction with the definition of A*. •

A similar existence and uniqueness result holds for decreasing operators.
Monotone Iterative Methods 171

Theorem 11.3 Let (X, K) be an ordered Banach space and let [uo, vol C
X be an order interval with 0 :::; uo :::; Vo such that there is a number A
with
A > 0, AVO:::; Uo. (11.5)
Assume T : [0, vol -+ X is decreasing on [0, vol, continuous on [uo, vol,
T ([uo, vo]) C [uo, vol and that there is a function X : (0,1) -+ (0,00) such
that
X(A) <:x'
1
T(AU):::; X(A)T(u) (11.6)

for all A E (0,1), u E [uo, vol. In addition assume one of the following
conditions:
(i) K is a regular cone.
(ii) K is a normal cone and T is completely continuous.
Then T has a unique fixed point u* in [uo, vol. Moreover, the sequences
(T2k (uo)) and (T 2k+1 (vo)) are increasing and convergent to u*, whilst
(T 2k+ 1 (uo)) and (T2k (vo)) are decreasing and convergent to u*. Also

Tk (w) -+ u* for all wE [uo, vol.

Proof. Since T is decreasing and T ([uo, vo]) C [uo, vol, we have
Uo :::; T (vo) :::; T (uo) :::; vo·
This implies

Uo :::; T (vo) :::; T2 (uo) :::; T2 (vo) :::; T (uo) :::; vo.
Finally

Uo :::; T(vo)

< T2k (uo) :::; T 2k+1 (vo)

< T 2k+ 1 (uo) :::; T2k (vo)

< T(uo):::; vo·

As in the proof of Theorem 11.1,

T2k (u)
o u*,
-+ T 2k +1 (u0
) -+ v*
, Uo :::; u* :::; v* :::; vo.
172 Chapter 11

We have
T(u*) = v*, T(v*) = u*.
Let
A* = sup {A E (0,1] : AV* :S u*}.
Then A*V* :S u*. We show that A* = 1. Assume A* < 1. Then using (11.6)
we deduce that

V* = T(u*)
< T (A*V*) :S X (A*) T (v*)
X(A*)U*.

Since X(A*) < I/A*, this contradicts the definition of A*. Hence A* = 1
and so u* = V*. Now, from

T2k (uo) :S T 2k +1 (vo) :S T2k (vo) :S T 2k - 1 (uo)

letting k ----t 00 we obtain

T2k+1 (vo) ----t u*, T2k (vo) ----t u*.

Finally, it is easy to conclude that Tk (w) ----t u* for every w E [uo, vol. •
The results presented in this section are more or less in connection
with those established by Amman [2], Guo-Lakshmikantham [27], Heikkilii-
Lakshmikantham [30] and Krasnoselskii [33].

11.3 Monotone Iterative Technique for Fredholm

Integral Equations
Consider the equation

u(x)= inh(x,y,U(Y))dY, xEO (11.7)

where 0 c R N is bounded open and h : -2

0 x Rn ----t Rn.

Theorem 11.4 Let Uo, Vo E C (0; Rn) be such that Uo (x) :S Vo (x) and

Uo (x) :S in h (x, y, Uo (y)) dy, in h (x, y, Vo (y)) dy :S Vo (x)

Monotone Iterative Methods 173

for all x E O. Assume h E C (0 2 X Rn; Rn) and

h (x, y, z) :s; h (x, y, z') (11.8)

for all z,z' ERn satisfying uo(x) :s; z:S; z':s; vo(x) , and all x,y E O.
Then the sequence (ukh~o given by

Uk+l (x) = in h (x, y, Uk (y)) dy, kEN

is increasing and converges to a solution u* E C (0; Rn) of (11.7) . Also the

sequence (Vk)k~O defined as

Vk+1 (x) = in h (x, y, Vk (y)) dy, kEN

is decreasing and converges to a solution v* E C (0; Rn) of (11.7). In

addition, Uo :s; u* :s; v* :s; Vo and any solution wEe (0; Rn) with Uo :s;
w :s; Vo satisfies u* :s; w :s; v*.

Proof. Apply Theorem 11.1 to X = C (0; Rn) endowed with the

positive cone K (see Example 11.1) and to the operator T: C (0; Rn) - t
C (0; Rn) given by

T (u)(x) = in h (x, y, u (y)) dy.

Here K is a normal cone and T is completely continuous. •

For a function u : D - t Rn, notation u> 0 stands for u (x) > 0 on D,
i.e., Ui (x) > 0 on D for i = 1,2, ... , n.

Theorem 11.5 Assume all the assumptions of Theorem 11.4 are satisfied.
In addition assume that Uo > 0, (11.8) holds for all z, z, E Rn satisfying
o :s; z :s; z' :s; Vo (x) and all x, y E 0, and there is a function ¢ : (0,1) - t R
such that ¢ ()..) >).. and

¢ ()..) h (x, y, z) :s; h (x, y, )..z) (11.9)

for).. E (0,1), z E R n with Uo (x) :s; z :s; Vo (x), and all x, y E O. Then
(11.7) has a unique solution u* in [uo, vol and

Tk (w) -t u* for all wE [uo, vol,

where (Tk (uo)) is increasing and (Tk (vo)) is decreasing.

174 Chapter 11

Proof. Apply Theorem 11.2. Condition (11.3) holds for

\
A = mIn
. {UOi
VOi
(x) : x E H,
(x) n' {
~ E 1,2, ... , n
}}
,

where Uo = (UOl' U02, ... , uOn ) and Vo = (VOl, V02, ... , VOn) . •
A better result can be established when Uo and Vo are constants. For
a real number a, we shall also denote by a the vector (a, a, ... , a) E Rn
and the corresponding constant vector-valued function. Also the notation
[a, b] stands for the order interval in Rn and in any space of functions with
values in Rn, ordered by the usual positive cone.

Theorem 11.6 Let a, bE R, a < a < band

a:S k h (x, y, a) dy, k h (x, y, b) dy :S b

for all x E n. Assume hE C (0'2 x [a, b]; Rn) and

h (x, y, z) :S h (x, y, z') (11.10)

for all z, z' E Rn satisfying a :S z :S z' :S b and all x, y E n. In addition

assume that there is a function ¢ : [alb, 1) - t R such that ¢ (,\) > ,\ and

¢ (,\) h (x, y, z) :S h (x, y, '\z) (11.11)

for,\ E [a/b,I), z E [a, b] with '\z 2: a and all x, y E n. Then (11.7) has a
unique solution u* in [a, b] ,

Tk (w) -t u* for all wE [a, b] ,

where (Tk (a)) is increasing and (Tk (b)) is decreasing.

Proof. Let u*, V* be the minimal and maximal solutions of (11.7) in

[a, b] guaranteed by Theorem 11.1. Let

{ v7(x):
u~(x) - }
'\=min xEn, iE{I,2, ... ,n} .

Clearly, ,\ E [alb, 1]. If ,\ = 1, then u* = V* and we are finished. Assume

,\ < 1. Define u = (Ul, U2, ... , un) by

Udx) = max {a, '\v; (x)} = '\max {~, v; (x)}.

Monotone Iterative Methods 175

Let v = (VI, V2, ... , v n ) be defined by

Vi (x) = max{~, V; (x)}.

We have a:S Ui (x) :s
u; (x). Hence U E [a,u*]. Also V E [v*,b]. Using
the monotonicity of T on [a, b] and (11.11), we obtain

¢(A)V* ¢(A)T(v*)
:s ¢ (A) T (v) :s T (AV) = T (u)
< T(u*)=u*.
This is in contradiction with the definition of A since ¢ (A) > A. •
Example 11.3 The equation

u(x) = 10 K,(x,y)u(y)""dy, x E n (11.12)

has a unique solution u E C (n) satisfying a :s u (x) :s b on n, where

0< a < b, if K, E C (n2; R+) , a E (0,1) and

al-a:s 10 K, (x, y) dy :s b
l- a, X E n.

Here ¢ (A) = Aa .

Similar results can be stated for the case when T is decreasing. We

leave their proofs to the reader.

Theorem 11.7 Let ua, Va E C (n; Rn) be such that 0 < Ua :s Va and

ua (x) :s 10 h (x, y, Va (y)) dy, 10 h (x, y, Ua (y)) dy :s Va (x)

for all x E n. Assume hE C (n2 x Rn; Rn) and

h(x,y,z'):s h(x,y,z) (11.13)

for all z,z' E Rn satisfying 0 :s z :s z' :s Va (x), and all x,y E n.

In addition assume that there is a function X : (0,1) -+ (0,00) such that
X (A) < 1/ A and
h (x, y, A2') :s X (A) h (x, y, z) (11.14)
176 Chapter 11

for A E (0,1), Z E Rn with uo(x):S z:S vo(x), and all x,y E n. Then
(11.7) has a unique solution u* in [uo, vol and

Tk (w) ---) u* for all w E [uo, vol.

Moreover, the sequences (T2k (uo)) , (T 2k +l (vo)) are increasing and the
sequences (T 2k +1 (uo)) , (T2k (vo)) are decreasing.

Theorem 11.8 Let a, b E R, °< a < band

a:S 10 h (x, y, b) dy, 10 h (x, y, a) dy :S b

for all x E n. Assume h E C (n2 x [a, b] ; Rn) and

h (x, y, z') :S h (x, y, z) (11.15)

for all z, z' E Rn satisfying a :S z :S z' :S b and all x, y E n. In addition

assume that there is a function X: [a/b, 1) ---) (0,00) such that X(A) < l/A
and
h (x, y, AZ) :S X (A) h (x, y, z) (11.16)

for A E [a/b,l), Z E [a,b] with AZ 2 a, and x,y En. Then (11.7) has a
unique solution u* in [a, b] ,

Tk (w) ---) u* for all wE [a, b] ,

the sequences (T2k (uo)) , (T2k+l (vo)) are increasing, and the sequences
(T 2k+1 (uo)) , (T2k (vo)) are decreasing.

Example 11.4 The equation (11.12) has a unique solution u E C (n)

satisfying a :S u (x) :S b on n, where °< a < b, if K; E C (n2; R+) ,
ex E (-1,0) and

ab-O'. :S In K; (x, y) dy :S a-O'.b, x En.

Here X (A) = AO'..

Monotone Iterative Methods 177

11.4 Minimal and Maximal Solutions of a Delay

Integral Equation
In this section we present several existence, uniqueness, and approximation
results for the initial valued problem and for the periodic problem concerning
the delay integral equation in R n

u (t) = l~T f (s, u (s )) ds. (11.17)

1. The Initial Value Problem

Here we are interested in continuous solutions u of (11.17), with u (t) 2
a on [0, tl] , such that

u (t) = 'P (t), t E [-T,O]. (11.18)

We assume that 'P E C ([-T, 0] ; Rn) and

'P (0) = I: f (s, 'P (s)) ds. (11.19)

The assumptions are as follows:

(al) fEe ([-T, tl] x [a, oo)n; R+) ;
(a2) 'P E C ([-T, 0] ; Rn), satisfies (11.19) and 'P (t) 2 a on [-T, 0];
(a3) there exists a function gEe ([-T, tl] ; Rn) such that

f(t,z)2g(t)

for all t E [-T, tl]' Z E [a,oo)n, and

l~T g (s ) ds 2 a
for all t E [0, tl] ;
(a4) there exists a continuous nondecreasing function 'IjJ (a, 00) -t

(0,00) such that

If (t, z)1 :S 'IjJ (Izl)
for all t E [0, tl], z E [a, oo)n, and

('X) 1
tl < Jb 'IjJ(<J)d<J,
178 Chapter 11

where

Let Ro E R+ be such that

b= I: If (s, tp (s)) I ds.

tl = i
b
RO 1
'ljJ ((}) d(}. (11.20)

Let X = C ([0, tl] ; Rn) be endowed with the usual positive cone, let
K={uEX:u2a}
and let T: K -> C ([0, tl] ; Rn) be given by

T (u) (t) = l~T f (s, u(s)) ds, (11.21)

where u(t) = tp (t) for t E [-T, 0] , U (t) = u (t) for t E [0, tl]'

Theorem 11.9 Assume (a1)-(a4) are satisfied. In addition assume that

f (t, z) is increasing in z on [a, Ro]. Then Tk (a) -> u* uniformly on
[0, tl], u* is the minimal solution of (11.17)-(11.18) in K, lu* (t)1 :S Ro
on [0, tl], and
a :S T (a) :S ... :S Tk (a) :S ... :S u*.

Proof. In this situation a is a lower solution of the equation u = T (u) .

By Theorem 4.6, (11.17)-(11.18) has at least one solution v E K. In addition
any solution v satisfies

Iv(t)I:SRo on [O,tl]'
Consequently, v :S Ro. Now we apply Theorem 11.1 by setting Uo = a and
Vo= v, where v E K is any solution of the initial value problem. •
The next result deals with the existence and approximation of the max-
imal solution in K of (11.17)-(11.18).

Theorem 11.10 Assume (a1)-(a4) are satisfied. In addition assume that

there is R 2 Ro such that

f (t,R) :S T-IR, t E [-T,tl] (11.22)

(i.e., f(t,tp(t)):S T-IR for t E [-T,O] and f(t,R):S T-IR for t E [O,tl]),
and f (t, z) is increasing in z on [a, R]. Then Tk (R) -> v* uniformly on
[0, tl], v* is the maximal solution in K of (11.17)-(11.18), and
a :S v* :S ... :S T k+ l (R) :S Tk (R) :S ... :S T (R) :S R.
Monotone Iterative Methods 179

Proof. By (11.22), we have T (R) :S R. Further the proof is analog with

that of Theorem 11.1. Here we choose as upper solution the constant func-
tion Vo = R and as lower solution un, any solution in K to our problem .
•
°
Theorem 11.11 Assume the assumptions of Theorem 11.9 are satisfied.
In addition assume a > qnd that there is a function ¢ : [a/ R o, 1) ~ R
such that for all A E [a/R o,I), t E [O,tl] and z E Rn with z E [a,R o] and
AZ ~ a, one has

¢ (A) > A, ¢ (A) f (t, z) :S f (t, AZ) . (11.23)

Then (11.17)-(11.18) has a unique solution in K.

Proof. Let u* be the solution given by Theorem 11.9 and let u E K

be any solution of (11.17)-(11.18). We will show that u = u*. Let

A=min{::g?: tE[O,tl], iE{I,2, ... ,n}}.

Since a :S u* :S u :S Ro, we have A E [a/ R o, 1]. As in the proof of Theorem

11.2, we deduce that A=1. Hence u=u* . •
For ¢ (A) = AU, a E (0,1), and n = 1, Theorem 11.11 becomes The-
orem 4 in Precup [47]. Another example of function ¢ satisfying (11.23),
IS
¢ (A) = log (1 + aA)
log(l+a)
for f (t, z) of the form q (t) log (1 + Izl) (see Dads-Ezzinbi-Arino [18]).

Corollary 11.1 Assume that all the assumptions of Theorems 11.10 and
11.11 are satisfied. Then (11.17)-(11.18) has a unique solution u* in K,
and for any wEe ([0, tl]; Rn) with w E [a, R] one has Tk (w) ~ u*
uniformly on [0, tl]'

The next result refers to functions f (t, z) which are decreasing in z.

Theorem 11.12 Assume (al)-(a4) are satisfied. Denote

R= max {Ro, max IT (a) (t)l}

tE[O,tlJ
180 Chapter 11

and assume f (t, z) is decreasing in z for 0 < a :::; z :::; R. Also assume
that there is a function X: [aj R, 1) --) (0,00) such that for all oX E [aj R, 1),
t E [0, tIl and z E Rn with z E [a, R] and oXz ~ a, one has
1
X (oX) <:.\, f (t, oXz) :::; X (oX) f (t, z) . (11.24)

Then (11.17)-(11.18) has a unique solution u* in K, Tk (w) --) u* uniformly

on [O,tl] for every wE [a,R] , the sequences (T2k(a)) , (T2k+l(R)) are
increasing, and the sequences (T 2k +I (a)), (T2k (R)) are decreasing.

Proof. Observe that T ([a, RJ) c [a, R] . •

For x (oX) =.xc" a E (-1,0), and n = 1, Theorem 11.12 becomes Theorem
6 in Precup [47].

2. Periodic Solutions
We now present similar results for the periodic solutions of the equation
(11.17).
We are interested in periodic continuous solutions u such that

a :::; u (t) :::; R for all t E R.

The hypotheses are those from Section 4.5:

(hI) f E C (R x [a,oo)n;R+).
(h2) There is w > 0 such that f (t + w, z) = f (t, z), t E R, z E [a,oo)n.
(h3) There exists an w-periodic function g E C (R; Rn) such that

f(t,z)~g(t)

for all t E R, z E [a,oo)n, and

l~T g (s ) ds ~ a
for all t E R.
(h4) there is a number b with avn
< b < R, and a function 'IjJ E
C ([a, R] ; R+) with 'IjJ (t) > 0 on [b, R], such that

If (t, z)1 :::; 'IjJ (Izl)

for all t E Rand z E [a,oo)n with Izl :::; R,

w:::; l b
R 1
'IjJ(a)da (11.25)
Monotone Iterative Methods 181

and
If(t,z)l::; ~T (11.26)

for all t E Rand z E [a, oo)n with b ::; Izl ::; R.

Let X be the Banach space of all continuous w-periodic functions u :
R R n, ordered by the usual positive cone, and let
---+

K = {u EX: u (t) 2 a on [0, w]}.

Under assumptions (h1)-(h4), the operator T: [a, R] ---+ K, given by

T (u)( t) = l~r f (s, u (s ) ) ds

is well defined and completely continuous.

Theorem 11.13 Assume (h1)-(M) are satisfied. In addition assume that

a> 0, f(t, z) is decreasing in z on [a, R], and that there exists a function
X : [aiR, 1) ---+ R satisfying (11.24) for all t E [O,w], A E [aIR,l) and
z E Rn with z E [a,R] and AZ 2 a. If

T2 (R) ::; R, (11.27)

then (11.17) has a unique w-periodic solution u* E [a, R]. Moreover,

Tk (R) ---+ u*

uniformly on [0, w], with (T 2k +1 (R)) increasing and (T2k (R)) decreasing.

Proof. By Theorem 4.7, there exists at least one w-periodic solution in

[a,R]. Let u E [a,R] be any w-periodic solution of (11.17). Since f(t,z)
is decreasing in z on [a, R], from a::; u ::; R, we obtain

a::; T (R) ::; T (u) = u.

Then
a::; T (R) ::; u ::; T 2 (R).

By (11.27) this yields

a ::; T (R) ::; T3 (R) ::; u ::; T2 (R) ::; R.

182 Chapter 11

We successively obtain

a < T (R) ~ T3 (R) (11.28)

< T2k+1 (R)

< u

< T2k (R)

< T4 (R) ~ T2 (R) ~ R.

Since T is completely continuous, there are two subsequences of (T2k+l (R))

and (T2k (R)) uniformly convergent to some u* E [a, R] and v* E [a, R] ,
respectively. By (11.28) we see that the entire sequences (T 2k+l (R)) and
(T2k (R)) converge uniformly to u*, respectively to v*, and

a ~ u* ~ u ~ v* ~ R.

Obviously
u* = T (v*) , v* = T (u*) .
Finally, use a similar argument as in the proof of Theorem 11.3 to conclude
that u* = u = v*. •

Theorem 11.14 Assume the assumptions of Theorem 11.13 hold with

T(a) ~ R (11.29)

instead of (11.27). Then (11.17) has a unique w-periodic solution u* E [a, R]

and Tk (w) ----+ u* uniformly on [O,w] , for any wE [a,R].

Proof. Notice (11.29) implies (11.27). Indeed, from a ~ T (R) ~ T (a) ,

we obtain
T2 (a) ~ T2 (R) ~ T(a) ~ R,
whence (11.27). Thus, Theorem 11.13 applies.
Further, for any wE [a, R] we have

a ~ T (R) ~ T (w) ~ T (a) ~ R.

Monotone Iterative Methods 183

This successively yields

a < T (R) :::; T3 (R)

:::; T 2[(k-l)/2]+1 (R)

:::; Tk (w)
< T2[k/2] (R)

< T4 (R) :::; T2 (R) :::; R.

Since Tk (R) ----+ u*, it follows that Tk (w) ----+ u*, as we wished . •

Remark 11.3 A sufficient condition for (11.29) is that

f (t, a) :::; T- 1R, t E R.

For the next result let us replace (h4) by the following condition used in
Guo-Lakshmikantham [28]:

(h4') f (t, z) :::; T- 1 R for all t E [0, w], Z E [a, R].

The next theorems complement the results established in [28].

Theorem 11.15 Assume (h1)-(h3) and (h4') are satisfied. In addition

assume a> 0, f (t, z) is increasing in z on [a, R], and there is a function
'P : [aiR, 1) ----+ R satisfying (11.23) for all t E [O,w], A E [aiR, 1) and
z E Rn with z E [a, R] and AZ ?: a. Then (11.17) has a unique w-periodic
solution u* E [a, R] and Tk (w) ----+ u* uniformly on [0, w], for any w E
[a,R].

Theorem 11.16 Assume (h1)-(h3) and (h4') are satisfied. In addition

assume a> 0, f (t, z) is decreasing in Z on [a, R], and there is a function
X : [aiR, 1) ----+ (0,00) satisfying (11.24) for all t E [0, w], A E [aiR, 1)
and Z E Rn with Z E [a, R] and AZ ?: a. Then (11.17) has a unique w-
periodic solution u* E [a, R] and Tk (w) ----+ u* uniformly on [0, w], for any
wE [a,R].

The proofs are similar to that of Theorem 11.13, so we omit the details.
184 Chapter 11

The monotone iterative method for periodic solutions to equation (11.17),

for the case when f (t, z) is increasing in z was discussed by Guo-Lakshmi-
kant ham [28]. See also Canada [11], Canada-Zertiti [12] and Guo-Lakshmi-
kantham-Liu [29].
For X(>') = A''', a E (-1,0), and n = 1, Theorems 11.13 and 11.15
were established in Precup [46], where several examples can also be found.

11.5 Methods of Upper and Lower Solutions for

Equations of Hammerstein Type
In this section monotone iterative methods are used to localize and approx-
imate solutions of the abstract Hammerstein equation in Rn

u (x) = ANf (u) (x) a.e. on O. (11.30)

Here N f is Nemytskii's superposition operator associated to a given function

f : 0 x R n ~ Rn (0 C RN open), and A is a bounded linear operator from
Lq (0; Rn) to £P (0; R). We seek solutions u in an order interval [uo, vol
of LP (0; Rn), where Uo is a lower solution of (11.30) and Vo is a upper
solution of (11.30). Of course, we shall assume that N f maps [uo, vol into
Lq (0; Rn). The main assumption is the monotonicity of f in its second
argument.
The advantage of working in the space £P (0; Rn) (1 :::; p < 00) ordered
by the regular cone of positive functions is that we do not need the complete
continuity of the operator T = ANf .

Theorem 11.17 Let p, q E [1,00), A: Lq (0; Rn) ~ £P (0; Rn) a bounded

linear operator and f : 0 x R n ~ Rn a (p, q)-Caratheodory function. Let
Uo, Vo E £P (0; R n ), Uo :::; va, be such that
Uo :::; ANf (uo) , ANf (vo) :::; va· (11.31)

Assume that
f(x,z):::;f(x,z') (11.32)
for all z,z' E Rn with Uo (x) :::; z:::; z':::; Vo (x), a.e. on 0, and that A is
increasing, i. e.,
A (v) 20 for all v 2 o. (11.33)
Then there are solutions u*, v* E [uo, vol to (11.30), minimal, respectively,
maximal in [uo, vol. Moreover, the sequences (Tk (uo)) , (Tk (vo)) converge
monotonically to u* and v*, respectively.
Monotone Iterative Methods 185

Proof. According to Theorem 5.1 N j is well defined, bounded, and

continuous from £P (0; Rn) to Lq (0; Rn). From (11.32) it follows that
N j is increasing on [uo, vol. This together with (11.33) guarantees that
T = ANj is increasing on [uo, vol. In addition T is continuous and

T ([uo, vo]) C [uo, vol

by (11.31). The conclusion now follows from Theorem 11.1 since the positive
cone of LP (0; Rn) is regular. •

Theorem 11.18 Assume all the assumptions of Theorem 11.17 hold. In

addition assume that there are constants a, b > 0 such that

a ~ uo (x) ~ vo (x) ~ b (11.34)

and (11.32) holds for all z, z' E Rn with 0 ~ z ~ z' ~ vo (x), a.e. on O.
Also assume that there is a function ¢ : [alb, 1) -> R with

¢(>..)> >.., ¢(>")f(x,z)~f(x,>"z) (11.35)

for all >.. E [a/b,l) and z E Rn with uo (x) ~ z ~ vo (x), a.e. on O. Then
(11.30) has a unique solution u* in [uo, vo], and

Tk (w) -> u* in £P (0; Rn) for all wE [uo, vol.

Moreover, the sequences (Tk (uo)) and (Tk (vo)) converge monotonically
to u*.

Proof. Let u*, v* be the minimal and maximal solutions given by The-
orem 11.17. Clearly u* ~ v*. We have to prove that u* = v*. Let

>"=sup{j.lE[O,I]: j.lV;(x)~ui(x) a.e.on 0, i=I,2, ... ,n}. (11.36)

From (11.34),
a ui (x)
- < - - < 1 a.e. on 0,
b - v; (x) -
so >.. E [alb, 1]. Clearly, >..v* ~ u*. If >.. < 1, then using (11.35) we obtain
¢ (>..) v* = ¢ (>..) T (v*) ~ T (>..v*) ~ T (u*) = u*.

Since ¢ (>..) > >.. this contradicts the definition of >... Thus >.. = 1 and so
u* = v* . •
A variant of Theorem 11.18 can be stated for the case where T is in-
creasing on [uo, (b/a) uo] instead of [0, vol.
186 Chapter 11

Theorem 11.19 Assume all the assumptions of Theorem 11.17 hold. In

addition assume that there are constants a, b > 0 satisfying (11.34) such
that (11.32) holds for all z, z' E Rn with

b
uo (x) :S z :S z' :S - Uo (x) ,
a

a.e. on O. Also assume that there is a function ¢ : [alb, 1) ----t R satisfying

(11.35) for all >. E [a/b,l) and z E R n with Uo (x) :S z :S (b/a) Uo (x) and
Uo (x) :S >.z, a.e. on O. Then (11.30) has a unique solution u* in [uo, vol ,
and
Tk (w) ----t u* in LP (0; Rn) for all wE [uo, vol.
Moreover, the sequences (Tk (uo)) , (Tk (vo)) converge monotonically to u*.

Proof. Let u*, v* be the minimal and maximal solutions given by The-
orem 11.17. Let>. E [a/b, l] be given by (11.36). Let u = (UI' U2, ... , un)
and v = (VI, V2, ... , v n ) be defined by

Ui (x) = max {UOi (x), >.v; (x)} = >.max {}UOi (x), v; (x)}

and
Vi (x) = max {}U Oi (x), V; (x) } ,

i = 1,2, ... , n. We have

V* :S v :S -uo,
b \ :S u * .
uo:S u = AV
a
Hence if >. < 1 then

¢(>.)v* = ¢(>')T(v*):S ¢(>')T(v):S T(>'v):S T(u*) = u*.

Since ¢ (>.) > >. this contradicts the definition of >.. Thus >. = 1 and so
u* = v* . •
Similar existence and uniqueness results can be established for decreasing
operators. For example, we have the following theorem.

Theorem 11.20 Let p,q E [1,00), A : Lq(O;Rn) ----t LP(O;Rn ) an in-

creasing bounded linear operator and f : OxRn ----t Rn a (p, q)-Caratheodory
function. Let uo, Vo E V (0; Rn) and a, b > 0 be such that (11.34) and

Uo :S ANf (vo) , ANf (uo) :S Vo (11.37)

Monotone Iterative Methods 187

hold. Assume that

f (x, z') :s: f (x, z) (11.38)

for all z, z' E Rn with 0 :s: z :s: z' :s: Vo (x), a.e. on O. Also assume that
there is a function x: [a/b,l) ---t (0,00) with

1
X (.\) < :\"' f (t, .\z) :s: X (.\) f (t, z) (11.39)

for all .\ E [a/b,l) and z ERn with Uo (x):S: z:S: Vo (x), a.e. on O. Then
there exists a unique solution u* E [uo, vol to (11.30) and

Tk (w) ---t u* in £P (0; Rn) for all wE [uo, vol.

Moreover, the sequences

(T2k (u o)) , (T 2k +1 (u o)) , (T2k (v o)) , (T 2k +1 (v o))

converge monotonically to u*.

Proof. The operator T is continuous and decreasing on [uo, vol. In

addition T([uo,voJ) c [uo,vo] , (T2k (uo)) is increasing, (T 2k +1 (uo)) is
decreasing, and T2k (uo) :s: T 2k +1 (uo). The positive cone of £P (0; Rn)
being regular, the two sequences converge to some elements u * and v* ,
respectively. Then

uo:S:u*:S:v*:S:vo, T(u*)=v*, T(v*)=u*.

To prove u* = v* let .\ E [a/b,l] be given by (11.36). Then if .\ < 1, from

o :s: .\v* :s: u* :s: Vo

using the monotonicity of T on [0, vol and (11.39) we deduce

v* = T (u*) :s: T (.\v*) :s: X (.\) T (v*) = X (.\) u*,

where X (.\) < 1/'\, a contradiction to the definition of .\. Thus .\ = 1, and
so u* = v* . •
Our next goal is to obtain lower and upper solutions. We succeed this if
we have more information about f and A, in particular, about the spectrum
of A.
188 Chapter 11

Theorem 11.21 Let p,q E [1,00), A : Lq(O;Rn) ----+ V(O;Rn) an in-

creasing linear operator and f : 0 x Rn ----+ R n a (p, q)-Caratheodory func-
tion. Assume that there are c E R+ and 9 E Lq (0; R+) such that

f (x, z) :S cz + 9 (x) (11.40)

for a.e. x E 0 and all z 2: o. Then any solution Va 2: 0 of the equation

(I - cA) (v) = A (g) (11.41)

(if there is one) is an upper solution of the equation u = AN! (u). If in

addition
- f (x, - z) :S cz + 9 (x) (11.42)
for a.e. x E 0 and all z 2: 0, then Ua := -Va is a lower solution.

Proof. Assume Va 2: 0 solves (11.41). Then, from (11.40) we have

f (x, Va (x)) :S c Va (x) + 9 (x)

and since A is increasing,

AN! (va) :S cA (va) + A (g) = Va·

Hence Va is a upper solution. We leave to the reader to check that -Va IS
a lower solution. •
We conclude this section by two much applicable results involving spec-
tral properties of A.
First we establish an abstract Poincare inequality.

Lemma 11.1 Let X be a Hilbert space and A X ----+ X be a positive

self-adjoint operator. Then

JA(u)J 2 :s JAJ (A(u) ,u), u E X. (11.43)

Proof. Since A is positive, for all u, V E X and t E R, we have

(A (u + tv) , u + tv) 2: 0,
that is
(A (v) , v) t 2 + 2 (A (u) , v) t + (A (u) , u) 2: o.
Consequently
(A(U),V)2:s (A(v),v)(A(u),u).
Monotone Iterative Methods 189

For v = A (u) this inequality becomes

IA (u)14 ~ (A2 (u) , A (u)) (A (u), u) . (11.44)

According to Proposition 6.2,

(A2 (u), A (u)) ~ lAllA (u)12. (11.45)

Now (11.44) and (11.45) yield (11.43) .•

Our next result is an abstract weak maximum principle in L2 (0; Rn).

For a function u : 0 ----t Rn, we let u+, u- be the functions defined by

ut (x) = max {O, Ui (x)}, ui (x) = max {O, -Ui (x)} ,

i = 1,2, ... , n. Clearly, u = u+ - u-, u+ ;:: 0 and u- > O. Also, for a

function u one has u;:: 0, if and only if u- = o.

Lemma 11.2 Let A : L2 (0; Rn) ----t L2 (0; Rn) be a positive self-adjoint
operator. Assume the following conditions are satisfied:

A(u);::O for u;::O; A(u)#O for u#O (11.46)

and

(A(u+),A(u-))2=(A(u+),U-)2=0, UEL2(0;Rn). (11.47)

Then for any constant c < IAI- 1 ,

(I - cA)-l (u) ;:: 0 for all u;:: o. (11.48)

Proof. Let 0' (A) be the spectrum of A, that is

0' (A) = R \ {A E R: A - >..I is bijective} .

It is known that
0' (A) c [-IAI , IAI]
(see Brezis [7], p 94). Since c < IAI- 1 the operator 1 - cA is invertible.
Let u;:: 0 and let v = (1 - cA)-l (u). Clearly

v-cA(v)=u. (11.49)
190 Chapter 11

We have to show that v 2:: 0, equivalently v- = O. Assume the contrary,

i.e., v- -I- O. Then (11.46) guarantees that A (v-) 2:: 0 and A (v-) -I- o. If
we multiply (11.49) by A (v-), and we use (11.47), we obtain

- (A (v-) ,v-)2 + c (A (v-) ,A (v-))2 = (A (v-) ,u)2'

Since both u and A (v-) are positive we have

(A (v-) ,u)2 2:: O.

Therefore
(A (v-) ,v-)2
c> -'---'-----'---'----,;:-'--"'-
- IA (v-)I~
This together with (11.43) implies c 2:: IAI- 1 , a contradiction. Thus v- = 0
and the proof is complete. •
For the last two results assume that 0 c RN is bounded open.

Theorem 11.22 Let A : L2 (0; Rn) - t L2 (0; Rn) be a positive self-adjoint

operator such that (11.46) and (11.47) hold. Let f : 0 x Rn-t R n be a map
satisfying the Caratheodory conditions, such that for each m E (0,00) there
is a constant am E R+ with

f (x, z) + amz increasing in z on [-m, m]

for a.e. x E O. Assume that

f (x, z) :::; cz + c', f (x, -z) 2:: -cz - c' (11.50)

for a.e. x E 0, all z E Rn with z 2:: 0, and some c E R+ with c < IAI- 1
and c' E R+.. In addition assume that the solution of the equation

u - cA (u) = A (c')

belongs to Loo (0; Rn). Then the equation u = ANj (u) has at least one
solution in L2 (0; Rn). Moreover, if the set S+ (S_) of all solutions u 2:: 0
(respectively, u :::; 0) is nonempty, then it has a maximal (respectively,
minimal) element.

Proof. Since c < IAI- 1 ,

the operator I - cA is bijective and so the
equation (11.41) has a unique solution Va for each g. Here g = d. By
Lemma 11.2, Va 2:: o. Now, (11.50) guarantees both (11.40), (11.42). Thus,
by Theorem 11.21, Va is an upper solution and Ua = -Va is a lower solution.
Monotone Iterative Methods 191

Since Vo belongs to Loo (0; Rn), there is mE (0,00) with Vo ::; m. Then
the function fm given by

fm(x,z) = f(x,z) +amz

is increasing in z on [-m, m] . Also the equation u = ANj (u) is equivalent
to
u = (1 + amA)-l ANjm (u) .
Let
Tm = (I + amA)-l ANjm ·
Clearly,
Uo ::; Tm (uo) , Tm (vo) ::; Vo
and Tm is continuous and increasing on [uo, vol. Let u*, v* be the minimal,
respectively maximal solution in [uo, vol. We have

-vo ::; u* ::; v* ::; Vo.

We now show that if w E L2 (0; Rn) , w 2: 0, solves w = ANj (w) then

w ::; Vo. Indeed, from

w = ANj (w) ::; A (cw + c') = cA (w) + A (c')

and
Vo = cA (vo) +A (c') , (11.51)
by subtraction we obtain

Vo - w 2: cA (vo - w) .

Then by the maximum principle, Lemma 11.2, Vo - w 2: o. Hence v* is

maximal in S+. Similarly, if wE L2 (0), W ::; 0 and w = ANj (w), then
-Vo ::; w. Hence u* is minimal in S_ . •
The last theorem is an existence and localization result of a nonnegative
non-zero solution.

Theorem 11.23 Let A : L2 (0; Rn) --. L2 (0; Rn) be a completely con-
tinuous positive self-adjoint operator such that (11.46) and (11.47) hold.
Let f : R+. --) R n be a continuous map such that f (0) = 0 and for each
mE (0,00), there is a constant am E R+ with

fm (z) := f (z) + amz increasing on [0, m].

192 Chapter 11

Assume that
J(z) S;cz+c'
Jor all z E R+., and some c E R+ with c < IAI- 1 , C' E (0, oot , and
J (z) 2: IAI- 1 Z (11.52)

Jor all z E R+. with Izl S; EO, where EO > o. In addition assume that the
solutions oj the equations

u-cA(u)=A(c') and u-IAI-1A(u)=0

belong to £00 (0; Rn). Then the equation u ANf (u) has a maximal
solution u in £2 (0; R+.) and u 1= o.

Proof. As above, the unique solution va of the equation

u - cA (u) = A (c')

belongs to £00 (0; R+.) and is an upper solution of the equation u =

ANf (u). Since J (0) = 0 the null function is a solution, and so a lower
solution. Now we apply Theorem 11.1 to deduce the existence of a maximal
fixed point v* in [0, vol of the operator

Tm = (I + amA)-l ANfrn ,
where m E (0, 00) satisfies Va (x) S; m a. e. on O. As in the proof of
Theorem 11.22 we can show that v* is maximal in the set of all nonnegative
solutions. To show that v* 1= 0, we prove that v* is the maximal fixed point
of Tm in an order subinterval

[uo, vol c [0, vol

with Uo 1= O.
Since A is completely continuous, the supremum in (6.3) is attained at
some Ul with IUll 2 = 1. Furthermore, since A is positive, we may assume
that
IAI = (A(uI),Ulh·
Then, according to (11.47), we have

IAI (A (ut - un ,ut - u 1)2

(A (ut) ,ut)2 + (A (u 1) ,u 1)2
(A (ut + u 1) ,ut + u 1) 2 .
Monotone Iterative Methods 193

Hence we may assume that Ul 2': 0 (otherwise, replace Ul by + u1 ; ut

notice that Iut
+ u l l2 = Iut -
u l l2 = IUll2 = 1). For any fixed E v
L2 (0; Rn) we consider the function

() = (A(Ul+tV),Ul+tvh
g t 2'
IUl + tvl 2
which can be defined on a neighborhood of t = O. This function attains its
maximum IAI at t = 0, so g' (0) = O. Notice

g' (0) = 2 [(A (Ul) ,v)2 - IAI (Ul' v)2l·

Hence
Ul = IAI- 1 A (Ul)
(i.e., IAI is the largest eigenvalue of A and Ul is an eigenfunction). Also,
by hypothesis Ul belongs to Loo (0; Rn). Let

Uo = E IU11~1 Ul,
where 0 <E :=::; EO. Clearly

Uo 2': 0, Uo i- 0, Iuo (x)1 :=::; E a.e. on 0, Uo = IAI- 1 A (uo).

Using (11.52), we deduce

Uo IAI- 1 A (uo) = A (IAI- 1uo)

< ANf (uo).
Thus Uo is a lower solution of U = ANf (u). Also, from

Vo = cA (vo) + A (c') , Uo = IAI- 1 A (Uo) ,

we have

Vo - Uo = cA (vo - uo) + (c -IAI- 1) A (uo) + A (c').

N ow we choose E > 0 small enough so that

(c-IAI- 1 )uo(x)+c'2':0 a.e. on O.

Then
Vo - Uo - cA (vo - uo) 2': 0,
and by the maximum principle, Vo - Uo 2': O. Next we apply Theorem 11.1
to deduce the existence of a maximal fixed point in [uo, vol of Tm. Clearly
it is equal to v*. •
194 Chapter 11

Example 11.5 Let n = 1. The operator

A = (_~)-1

has all the properties required by Theorems 11.22-11.23. Moreover, in this

case u*, v* are, respectively, the minimal and maximal solutions in the set
of all solutions in L2 (0). Indeed, if w E L2 (0) is any solution and we let
f w be defined by
fw (x, z) = { f (x, z) if w (x) > 0,
o ifw(x):S;O,

then
-~w+ = fw (x,w+) :s; cw+ (x) + c'
a.e. on O. Hence
w+ :s; cA (w+) + A (c') .
This together with (11.51) implies

vo - w+ ~ cA (vo - w+) .

Thus w+ :s; Vo. Similarly -Vo :s; -w-. Therefore -vo:S; w :s; Vo·
For other applications of the monotone iterative technique to different
types of equations we refer the reader to Bainov-Hristova [4], Buidl [8],
Carl-Heikkila [13], Constantin [15], De Coster-Habets [19], Fabry-Habets
[22], Liz-Nieto [35]' Pouso [45], Precup [49] and Wang-Cabada-Nieto [61].
 Chapter 12

Quadratically Convergent
Methods

The succesive approximation method as well as the monotone iterative meth-

ods described previously are not very fast. To explain this let us consider
an operator T : X ----t X. All these methods give us convergent sequences
(Uk) of the form
Uk+1 = T (Uk)' kEN
having as limit a fixed point u* of T. For the results in Chapter 10, we
have
d(Uk+l,U*) ~ Md(Uk'U*).
This shows that the convergence is linear.
In Chapter 11 we have not assumed that T was a contraction. However,
if T is Lipschitz in a neighborhood of u*, with the Lipschitz constant c,
then we have

IUk+1 - u*1 IT (Uk)

- T (u*)1
< c IUk - u*1
for all large k, which shows that the convergence is again linear.
Faster iterative methods for solving nonlinear equations can be derived
from the well known Newton's method. In this short chapter we describe just
Newton's method and we explain the way it can be used to solve nonlinear
integral equations. When Newton's method is applied to the differential
equations theirselves, one speaks about the quasilinearization method. This
method offers iterative sequences of approximate solutions that converge

195
R. Precup, Methods in Nonlinear Integral Equations
© Springer Science+Business Media Dordrecht 2002
196 Chapter 12

quadratically to the solution. We say that the convergence of the sequence

(Uk) to u* is of order p if

IUk+1 - u*1 ~ c IUk - u*I P

for all large k and some c> o. When p = 2 we say that the convergence is
quadratic. In case that the nonlinear terms of the operator equations have
behavior properties of convexity type, the iterative sequences of approximate
solutions are also monotone.

12.1 Newton's Method

In this section Newton's method is presented following Deimling [18], Section
15.4 and Granas-Guenther-Lee [26]. For more information, we send to
Janko [31]' Kantorovitch-Akilov [32]' Moore [36] and Potra-Rheinboldt [44].
Let X, Y be Banach spaces, U c X an open set and let F: U -+ Y be
continuously Frechet differentiable. Newton's method is an iterative method
for solving
F (u) = o.
It works as follows: choose an initial approximation Uo E U to the solu-
tion and generate a sequence (Uk) by solving successively, if possible, the
equations
F (Uk) + F' (Uk) (Uk+1 - Uk) = 0, kEN (12.1)
If (Uk) converges to a point u* E U, then by (12.1), F (u*) = o.
Notice that if F' (U) -1 exists for all U E U, then (Uk) is the sequence
of successive approximations for the equation

U=G(U),

where
G (U) = U - F' (u)-l F (u) .
Hence
Uk+! = Uk - F' (Uk)-l F (Uk), kEN. (12.2)
The following theorem contains the convergence criterion for Newton's method.

Theorem 12.1 (Kantorovitch) LetX, Y be Banach spaces and F : Br (uo)

c X --) Y a C1-operator. Assume:
(i) F' (uo)-l E L (Y,X) , IF' (uo)-l F (uo)1 ~a and IF' (uo)-ll ~ /3;
 Quadratically Convergent Methods 197

(ii) 1F'(u)-F'(v)1 :S'Ylu-vl for all u,vEBr(uo);

(iii) 2a/h < 1 and 2a < r.
Then F has a unique zero u* in B2Q (uo) and the sequence (Uk) given
by (12.1) converges quadratically to u* and satisfies

IUk - u*1 :S a2 1- k (2a!J"()2 k -l, kEN. (12.3)

Proof. First we prove that F' (u) is invertible for any U E B2Q (uo).
For this, let U E B2Q (uo) and define the map r : X - t X as

r = 1 - F' (Uo) -1 F' (u) = F' (Uo) -1 (F' (Uo) - F' (u)) .
Then for every v, w E X we have

Ir (v) - r (w)1 :S IF' (uo)-IIIF' (uo) - F' (u)llv - wi

:S JJ"( luo - ullv - wi :S 2a,LJ"( Iv - wi·
Since 2a(3'Y < 1, r is a contraction. Consequently the map

1- r = F' (uO)-1 F' (u)

is bijective. As a result

F' (u) = F' (uo) (1 - r)

is bijective too.
Next we prove that the sequence (Uk) given by (12.2) is well defined and
convergent.
We shall prove by induction that for each k E N\ {O} ,
1
IUk - uol :S 2a, ak:S 2- k+1a, 'Yk <
- -2'
(12.4)

where

'Yk = 'Yak(3k, ak = IF' (Uk-I) -1 F (Uk-I) I, (3k = IF' (uk-I)-II·

Assume (12.4) holds for 1, 2, ... , k. We shall prove that (12.4) is also true
for k + 1. We have

F' (Uk)-1 F (Uk)

F' (Uk)-1 [F (Uk) - F (Uk-I) - F' (Uk-d (Uk - uk-d]

F' (Uk)-1 10 1 [F' (Uk-l + t (Uk - Uk-I)) - F' (Uk-I)] (Uk - Uk-I) dt.
198 Chapter 12

Then from (ii) we obtain

CYk+1 :s i3k+l CYk 11 IF' (Uk-l + t (Uk - Uk-I)) - F' (Uk-I)I dt (12.5)
< ,i3k+l CY 211
k 1
tdt=-,i3k+l CY k2
'
o 2
On the other hand, from

F' (Uk) = F' (Uk-I) [1 + F' (uk_d- l (F' (Uk) - F' (Uk-I))] ,

we deduce that
i3k
i3k+l :s 1 - 'k .
Hence
1,i3kCY 2k
CYk+l :s "2 1 - ,k'
Thus

CYk+1 :s
1 CYkfk
"2 1 - 'k ' < 1 -
'k+l - "2 (1 - ,k)
2 'k
2
(12.6)

Since 'k :s 1/2 we obtain 1

'k+l -< -2'

Then
CYk+1 :s TlCYk :s Tkcy.
In addition

IUk+l - uol < IUk+1 - ukl + IUk - Uk-II + ... + lUI - uol
CYk+1 + CYk + ... + CYl :s (2- k + Tk+1 + .. , + 1) CY

< 2cy.
Hence (12.4) holds for all k E N\ {O}. Thus (Uk) is well defined and is a
Cauchy sequence. Let U* E B200 (uo) be its limit. Clearly F (u*) = O. Now
an inequality like (12.5) guarantees

IUk+l - u*1 :s c IUk - u*12

with c = 2-l,s~Pi3k < 00 since i3k -+ IF' (u*)-ll. Hence (Uk) converges
quadratically to a zero of F. If v* E B 200 (uo) is another zero of F then
 Quadratically Convergent Methods 199

from

v* - u* F' (UO)-l [F (u*) - F (v*) - F' (uo) (u* - v*)]

F' (UO)-l 10 1 [F' (v* + t (U* - v*)) - F' (Uo)] (U* - v*) dt,

we obtain

[u* - v*[ < /3'Y [U* - V*[10 1 [V* + t (U* - V*) - Uo[ dt
< 2Ct/3'Y [U* - V* [.
Thus v* = u*. Finally we prove (12.3). Let 8k = 'Yk (1 - 'Yk) -1. Since
1/2, (12.6) implies 8k ::;
'Yk ::; Hence 8 k ::; 8L1.
1
. Consequently 8t-
1 2k - 2 k+1 2k-l_1
Ctk ::; T 81 Ctk-1::; ... ::; T (2Ct/3'Y) Ct.

Therefore
L
00
k
[Uk - uo[::; Cti::; Ct2 1 - k (2Ct/3'Y)2 -1 .
i=k+1

The proof is complete. _

Example 12.1 Let us present Newton's method for the abstract Hammer-
stein equation in Rn

u=ANf(u), uELP(D;Rn).

We assume that p, q E [1,(0), p > q and the following conditions are

satisfied:

(i) I : D x Rn ---t Rn is differentiable with respect to z and fJ Id fJZ j IS

(p, pq/ (p - q) )-Caratheodory for all i, j E {1, 2, ... , n}.
(ii) A : Lq (D; Rn) ---t V (D; Rn) is bounded linear.
Let F: V (D; Rn) ---t V (D; Rn) be defined as

F (u) = u - ANf (u) .

Then F is a C 1-operator and

F' (u) (h) = h - A [NJ f (u) h] ,

200 Chapter 12

for all u, hE £P (0; Rn), where

J f = [Oli]
OZj l~i,j~n
is the Jacobian matrix of f.
Now let Uo E LP (0; Rn) be an initial approximation of the solution.
The Newton iterates are determined recursively by solving

F (Uk) + F' (Uk) (Uk+l - Uk) = 0,

that is, the linear equation

Uk+1 - A [NJj (Uk) Uk+l] = A [Nf (Uk) - NJj (Uk) Uk] .

Notice F' (U) -1 exists if the only solution to the linear equation

h-A [NJj (u)h] =0

is h = O.
Example 12.2 Here we apply Newton's method to a differential equation
itself. In this situation one speaks about the quasilinearization method.
For the original idea of this method, see Bellman-Kalaba [6]. Consider the
problem
u" = I (x, u, u'), x E [0,1]
{
U (0) = U (1) = O.
Assume I : [0, 1] X R2n - t Rn is continuous and continuously differentiable
with respect to Z E R2n. Let C§ ([0, 1] ; Rn) be the Banach space of all func-
tions from C 2 ([0, 1]; Rn) which satisfy the boundary conditions U (0) =
U (1) = O. Consider the operator F : C§ ([0,1]; Rn) - t C ([0, 1]; Rn) ,
defined by
F(u) = u" - I (.,u,u').
We have that F is a C1- operator with
F' (u) (h) = h" - J} (., u, u') h - J} (., u, u') h'
for h E C~ ([0, 1] ; Rn). Here

J} (x, z, z') [ Oli (X,Z,Z')] ,

OZj l~i,j~n

J} (x, z, z') -+
[ of- (x,z,z ')]
OZj l~i,j~n
 Quadratically Convergent Methods 201

The Newton iterates are here determined by solving in C~ ([0, 1]; Rn) the
linear equation
" - J2f ( ., Uk, Uk' )Uk+l
uk+l ' - Jlf ( ., Uk, Uk, ) Uk+l
j (., Uk, uk) - J] (., Uk, uk) U~ - J] (., Uk, uk) Uk·
Here again p' (u)-l exists if the only solution in C~ ([0, 1] ; Rn) to the
linear equation
h" - J] (.,u,u') h' - J] (.,u,u') h = °
is h = 0. For this, there are known sufficient conditions given in terms of
j, J] and J] (see Granas-Guenther-Lee [26]).
Notice that in applications it is often difficult to find an initial point
Uo and its neighborhood Br (uo) such that all the assumptions of Theorem
12.1 are satisfied. One could overcome this difficulty by asking a convex-
ity property on the nonlinear operator in order to obtain the monotonicity
and the convergence of Newton's iteratives with respect to a certain order
structure. We refer the reader to Vandergraft [60] and Schmidt-Schneider
[57] for Newton's method in ordered spaces, and to Bellman-Kalaba [6] and
Mure§an-Trif [37] for applications to differential equations. The 1990s have
brought up to date the subject by combining Newton's method with the
method of upper and lower solutions together with the monotone iterative
technique, and by the new idea of considering more general nonlinearities
which can be represented as a difference of two convex functions. The re-
sulting mixture is now known as the generalized quasilinearization method.
It provides a tool to construct upper and lower sequences that converge
monotonically and quadratically to a solution of the equation. For several
applications of this method see Lakshmikantham-Leela-Sivasundaram [34],
Carl-Lakshmikantham [14], Cabada-Nieto [10] and the references therein.

12.2 Generalized Quasilinearization for an

Integral Equation with Delay
The results presented in this section was adapted from Precup [48]. We
use the method of quasilinearization generalized in Lakshmikantham-Leela-
Sivasundaram [34], for the quadratic, monotonic, bilateral approximation of
a solution to the delay integral equation

U (t) = l~r j (8, U (8)) d8. (12.7)

202 Chapter 12

We deal with the initial value problem for (12.7) and look for a continuous
solution u (t) for -T::; t ::; tl, when

u(t) = 'P(t) , - T::; t::; o. (12.8)

We assume
'P (0) = [Or f (s, 'P (s)) ds. (12.9)

Under assumption (12.9), problem (12.7)~(12.8) is equivalent with the initial

value problem

u' (t) = f (t, u (t)) - f (t - T, U (t - T)) , O<t::;tl,

{ (12.10)
U (0) = 'P (0) ,

1 - { 'P (t) for - T ::; t ::; 0

where u E C [O,tl] and u(t) = u(t) for 0 < t::; tl .
We shall discuss a more general problem of type (12.10), namely

{
u'(t)=f(t,u(t))+g(t-T,u(t-T)), o< t ::; tl,
(12.11)
u (0) = 'P (0) .

Before we state the main result of this section we present two useful
lemmas. The first one can be proved using Corollary 3.1.2 in Piccinini~
Stampacchia~Vidossich [41].

Lemma 12.1 Let u, v E C l [a, b] be such that u ::; v, and let H (t, z) be
continuous for u (t) ::; z ::; v (t) and every t E [a, b]. Assume that
u' (t) ::; H (t, u (t)) , H (t, v (t)) ::; v' (t)
on [a, b]. Then for each ao E [u (a), v (a)] there exists a solution a E
Cl [a, b] of the problem

a' (t) = H (t, a (t)) , t E [a, b],

{
a (a) = ao

such that u ::; a ::; v.

Proof. For u(t) < v(t) on [a,b] and u(a) < ao < v(a) this result is
Corollary 3.1.2 in Piccinini~Stampacchia~Vidossich [41]. In the general case
use the above particular result on subintervals. We omit the details. •
The next comparison lemma is owed to Lakshmikantham~Leela~Sivasun­
daram [34].
 Quadratically Convergent Methods 203

Lemma 12.2 Let u, v E C 1 [a, b] and F E C ([a, b] x R). Assume that

u' ::; F (t, u) , F (t, v) ::; v',

F (t, Zl) - F (t, Z2) ::; L (Zl - Z2) for Zl ;::: Z2,

where L E R+. Then u (a) ::; v (a) implies u (t) ::; v (t) on [a, b].

Proof. Assume that u (a) ::; v (a). For small c > 0 let

u= u - c e 2L (t-a) , V= v + c e2L (t-a).

We have
u<u, v<v, u(a)<v(a).
In addition

u' u' - 2c Le 2L (t-a)

::; F (t, u) - 2c Le 2L (t-a)
::; F (t, u) + Lc e2L (t-a) - 2c Le 2L (t-a)
< F(t,u).
Similarly
v' > F (t, v).
We claim that u(t) ::; v (t) on [a, b], which proves the result as c ----t O.
Indeed, if not there exists a to E (a, b] with

u(to) = v (to) , u' (to) ;::: v' (to) .

Then we obtain

F (to, U (to) ) > u' (to)

;: : v' (to)
> F (to, v (to)) ,
a contradiction. _
Assume that 'P E C [-T, 0] and let Uo, Vo E C 1 [0, t1] be such that
Uo (0) = Vo (0) = 'P (0) and Uo (t) < Vo (t) on (0, t1]' Let

n= {(t, z) : 0 < t ::; t1, Uo (t) < z < Vo (t)}

and
o=nu{(t,'P(t)): -T::;t<O}.
204 Chapter 12

Theorem 12.2 Assume:

(i) for 0 < t ::; tl, one has

U~(t)::; f(t,uo(t))+g(t-T,1i o (t-T)),

v~ (t) ?:: f (t, Vo (t)) + g (t - T, vo (t - T)) ;

(ii) f E C(n), g E C(O), the derivatives fz, fzz, gz and gzz exist and
are continuous on n, and satisfy

fzz ?:: 0, gz?:: 0, gzz::; 0 on n. (12.12)

Then there exist the sequences (Uk) increasing and (Vk) decreasing which
converge uniformly on [0, tll to the unique solution U E C l [0, tll of (12.11)
satisfying Uo ::; U ::; Vo, and the convergence is quadratic, i.e., there are
constants a, b E R+ such that

IUk - uloo, IVk - uloo ::; a IUk-l - ul~ + b IVk-l - ul~ (12.13)

for all k ?:: 1.

Proof. Notice (12.11) has a unique solution u satisfying Uo ::; u ::; Vo;
this follows by using the step method since f (t, z) is Lipschitz in z on n.
In what follows we shall use the convexity of f and concavity of g by
means of the inequalities

f(t,z) ?:: f(t,z')+fz(t,z') (z-z'), (12.14)

g(t,z) ?:: g(t,z')+gz(t,z)(z-z'),
which are true for all (t, z) , (t, z') E n. Suppose we have already built the
functions Uk and Vk such that

Uo ::; Uk ::; Vk ::; Vo

and
U~ (t) ::; f (t, udt)) + g (t - T, 1idt - T)) , (12.15)
v~(t)?:: f(t,vdt))+g(t-T,vdt-T)).
Then we take Uk+l = a and Vk+1 = {3, a and {3 being the unique solutions
of the following linear initial value problems with delay

{ a'(t)=Fdt,a(t),a(t-T)), O<t::;tl, (12.16)

a (0) = 'P (0) ,
 Quadratically Convergent Methods 205

respectively

(3'(t)=Gk(t,(3(t),{3(t-T)), 0< t ::::; tl,

{ (12.17)
(3 (0) = <p (0) ,

where

Fdt,x,y) = !(t,udt)) +!z(t,udt)) (x-udt)) (12.18)

+g(t-T,udt-T))
+gz (t - T,Vk (t - T)) (y - Uk (t - T))
and

Gk (t, X, y) ! (t,vk (t)) +!z (t,uk (t)) (x - Vk (t)) (12.19)

+g(t-T,vdt-T))
+gz (t - T, Vk (t - T)) (y - Vk (t - T)) ,

where for t E [0, T], by gz (t - T, <p (t - T)) we mean gz (0, <p (0)). Notice
for
Uk (t) ::::; X ::::; vk (t) , uk (t - T) ::::; Y ::::; Vk (t - T) ,
(12.18), (12.14) and gz decreasing yield

Fk (t, x, y) ::::; ! (t, x) + 9 (t - T, y) , (12.20)

whilst (12.19), (12.14) and !z increasing imply

Gdt,x,y) 2: !(t,x)+g(t-T,y). (12.21)

From (12.12), (12.14) and (12.15) we easily see that the following inequalities
hold:
u~ (t)::::; Fk (t,uk (t) ,Uk (t - T)),

V~ (t) 2: H (t,Vk (t) ,Vk (t - T)).

We now successively apply Lemma 12.1 to the intervals [0, T], [T, 2T], ... ,
[mT, tl], where mT < tl ::::; (m + 1) T. Thus we prove the existence of the
solution a = Uk+l of (12.16) satisfying Uk ::::; a ::::; Vk. Similarly we find a
solution (3 of (12.17) such that Uk ::::; (3 ::::; Vk·
Further, by (12.20)-(12.21) one has

a'(t) ::::; !(t,a(t))+g(t-T,a(t-T)), (12.22)

(3'(t) 2: !(t,(3(t))+g(t-T,(3(t-T)).
206 Chapter 12

Hence (12.15) also holds with k+ 1 instead of k. In order now to derive the
inequality a :s; (3, that is Uk+1 :s; Vk+l, we apply Lemma 12.2, successively
on [0, T] , [T,2T] , ... , [mT, tl], by taking

F (t, z) = Fo (t, z) + g (t - T, a (t - T)),

where
!(t,z) for zE [Uo(t),vo(t)],
Fo (t, z) = { ! (t, Uo (t)) for z :s; Uo (t),
!(t,vo(t)) for z~vo(t).
Suppose we have already proved that a (t) :s; (3 (t) for all t :s; jT. Then, for
tE [jT, (j + 1) T] , we know that

a(t-T):s;jj(t-T)
and since g is increasing (recall gz ~ 0),

g(t - T, jj (t - T)) ~ g(t - T, a (t - T)).

Then by (12.22)

a' (t) < ! (t, a (t)) + g(t - T, a (t - T)),

(3'(t) > !(t,(3(t))+g(t-T,a(t-T)).
Now Lemma 2 guarantees a (t) :s; (3 (t) on [jT, (j + 1) T] and so a (t) :s; (3 (t)
for all t:s; (j + 1) T. This argument applied successively yields a (t) :s; (3 (t)
on [0, tl]' Hence the sequences (Uk), (Vk) are well defined. Using standard
arguments it is easy to conclude that they both converge to the unique
solution U of problem (12.11).
Finally, we show that the convergence of the sequences (Uk) and (Vk)
to the unique solution U of (12.11) is quadratic. Let

pdt) = U (t) - udt) , qdt) = vdt) - u (t).

Note that Pk (0) = qk (0) =
Denote
° and Pk (t) ~ 0, qk (t) ~ 0 for t E [0, tl]'

IPkloo,j max{Pk (t) : t E [0, (j + 1) T] n [0, tl]}'

Iqkloo,j max{qk (t) : t E [0, (j + 1) T] n [0, tl]}'
for j = 0, 1, ... , m. We shall prove that for each j E {O, 1, ... , m}, there are
aj, bj E R+ such that

IPk+1loo,j' Iqk+Iloo,j :s; aj IPkl~,j + bj Iqkl~,j' (12.23)

 Quadratically Convergent Methods 207

Clearly, for j = m (12.23) gives (12.13). Let t E [0, T]. Then using the
definition of Uk+1 and the mean value theorem, we obtain

P~+l (t) = f (t, U (t)) + 9 (t - T, 'P (t - T)) - [f (t, Uk (t))

+ fz (t, Uk (t)) (Uk+l (t) - Uk (t)) + 9 (t - T, 'P (t - T))]
f (t, U (t)) - [f (t, Uk (t)) + fz (t, Uk (t)) (Uk+l (t) - Uk (t))]
fz (t, zI) Pk (t) - fz (t, Uk (t)) (Pk (t) - PHl (t))
::; fzz (t, Z2) P~ (t) + fz (t, uk (t)) Pk+1 (t)
< M2Ipkl~,o + M lPk+1 (t),
where Uk (t) ::; Z2 ::; zl ::; U (t), Ifzl ::; Ml and Ifzzl ::; M 2. It follows that
for t E [O,T] ,

Pk+1 (t) ::; M2T IPkl~ ,0 + Ml iot Pk+ds) ds.

Now the Gronwall's inequality implies

Pk+1 (t) ::; M2Te M1T IPkl~,o, t E [0, T].

Hence
IPk+1loo ,0::; M2Te M1T IPkl~ ,o'
Similarly,

q~+1 (t) f (t, Vk (t))+ fz (t, Uk (t)) (Vk+l (t) - Vk (t))

+9 (t - T, 'P (t - T)) - [f (t, U (t)) + 9 (t - T, 'P (t - T))]
f (t, Vk (t)) - f (t, U (t)) + fz (t, Uk (t)) (Vk+l (t) - Vk (t))
fz (t, zI) qk (t) + fz (t, Uk (t)) (qk+1 (t) - qk (t))
::; fzz (t, Z2) (Vk (t) - Uk (t)) qk (t) + fz (t, Uk (t)) qk+1 (t)
< d (qk (t) + Pk (t)) qk (t) + cqk+1 (t),
where U (t) ::; Zl ::; Vk (t) and Uk (t) ::; Z2 ::; Zl. Hence
, 3 2 1 2
qk+l (t) ::; Mlqk+l (t) + 2M2qk (t) + 2M2Pk (t)
3 2 1 2
< Mlqk+l (t) + 2M2lqkloo,o + 2M2IPkloo,o'
Using Gronwall's inequality again, we obtain

qk+1 (t)::; ~M2TeMIT Iqkl~o + ~M2TeMIT IPkl~o, t E [O,T].

2 ' 2 '
208 Chapter 12

Thus
Iqk+Iloo,o ::::: ~M2TeMIT Iqkl~,o + ~M2TeMIT IPkl~,o.
Therefore (12.23) holds for j = 0, with ao = M 2Te M1T , bo = ~M2TeMIT.
Next, assume (12.23) holds for j. We shall prove (12.23) for j + 1. Let
tE [T,(j+1)T]. We have

Pk+1 (t) = f (t, U (t)) + g (t - T, U(t - T)) - [f (t, Uk (t))

+ fz (t, Uk (t)) (Uk+1 (t) - Uk (t)) + g (t - T, Uk (t - T))
+gz (t - T,Vk (t - T)) (Uk+1 (t - T) - Uk (t - T))]
fz (t, zI) Pk (t) + gz (t - T, Z2) Pk (t - T)
+ fz (t, Uk (t)) (Pk+1 (t) - Pk (t))
+gz (t - T, Vk (t - T)) (Pk+1 (t - T) - Pk (t - T))
< [fz(t,u(t))-fz(t,udt))]pdt )
- [gz (t - T,Vk (t - T)) - gz (t - T,Uk (t - T))]Pk (t - T)
+ fz (t, Uk (t)) Pk+1 (t) + gz (t - T, Vk (t - T)) Pk+1 (t - T)
= fzz (t, Z3) p~ (t) - gzz (t - T, Z4) [Vk (t - T) - Uk (t - T)]
XPk (t - T) + fz (t,Uk (t))Pk+1 (t)
+gz (t - T, Vk (t - T)) Pk+1 (t - T),
where Uk (t) < zl, z3 < U (t), Uk (t - T) ::::: Z2 ::::: U (t - T). Also,
-gzz (t - T, Z4) [Vk (t - T) - Uk (t - T)] Pk (t - T)
< N 2 [qk(t-T)+Pk(t-T)]Pk(t-T)
[3 2 1 2
< N22Pdt-T)+2qdt-T) ]

1
< 2N2 [2 2 ].
3lPkl oo,j + Iqkloo,j

Thus,

Pk+1 (t) ::::: M IPk+1 (t) + M2Ipkl~,j+1

1
+NIlpk+lloo,j + 2N2 [2
3lPkl oo ,j 2 ],
+ Iqkloo,j
where Igzl ::::: N I, Igzzl::::: N2. Since

IPk+Iloo,j ::::: aj IPk I~,j + bj Iqk I~,j ,

IPkloo,j ::::: IPkl oo,j+1'
Iqkloo,j ::::: Iqkloo,j+I'
 Quadratically Convergent Methods 209

we obtain

Pk+1 (t) ::; MlPk+l (t) + 0: IPkl~,j+l + /3lqkl~,j+l .

Hence for t E [T, tll we have

Pk+l (t) < Pk+1 (T)

+ [0: IPkl~,j+1 + /3lqkl~,j+1] (tl - T) + Ml it Pk+l (s) ds

< ao IPkl~,o + bo Iqkl~,o + [0: IPkl~,j+l + /3lqkl~,j+1] (tl - T)

+Ml it Pk+l (s) ds

< 'Y IPkl~,j+1 + 8Iqkl~,j+l + Ml it Pk+l (s) ds.

Gronwall's inequality now yields the desired result

IPk+ll oo ,j+1 ::; aj+l IPk 1~,j+1 + bj+l Iqk I~,j+l ,

for some constants aj+l, bj+1 E R+. Similarly,

qk+l (t) f (t, Vk (t))

+ fz (t, Uk (t)) (Vk+1 (t) - Vk (t)) + g (t - T, Vk (t - T))
+gz (t - T,Vk (t - T)) (Vk+l (t - T) - Vk (t - T))
- f (t, U(t)) - g (t - T, U(t - T))
fz (t, qk (t) + gz (t - T, Z2) qk (t - T)
Zl)

+fz (t,Uk (t)) (qk+1 (t) - qk (t))

+gz (t - T,Vk (t - T)) (qk+1 (t - T) - qk (t - T))
< fzz (t, Z3) (Vk (t) - Uk (t)) qk (t) - gzz (t - T, Z4) qk (t - T)
+ fz (t, Uk (t)) qk+l (t) + gz (t - T, Vk (t - T)) qk+1 (t - T)
< fzz (t, Z3) (qk (t) - Pk (t)) qk (t) - gzz (t - T, Z4) qk (t - T)
+ fz (t, Uk (t)) qk+l (t) +gz (t - T, Vk (t - T)) qk+1 (t - T) ,
where U (t) ::; Zl ::; Vk (t), U (t - T) ::; Z2 ::; Vk (t - T) and Uk (t) ::; Z3 ::; Zl.
This yields

qk+1 (t) ::; Mlqk+l (t) + 0:' IPkl~,j+l + /3' Iqkl~,j+l·

As above,
Iqk+lloo,j+l ::; aj+llpkl~,j+l + bj+1lqkl~,j+l'
210 Chapter 12

eventually, with some new constants aJ+l, bj +1 E R+. Thus (12.23) also
holds for j + 1. Therefore (12.23) is true for all j E {O, 1, ... , m} . •

Corollary 12.1 Assume 'P E C [-T, 0] , Uo, Vo E C 1 [0, tl], Uo (0) = Vo (0) =
'P (0), Uo < Vo on (0, tl], f E C(O), and

u~(t) :=; f(t,uo(t))-f(t-T,UO(t-T)),

vb(t) 2: f(t,vo(t))-f(t-T,Vo(t-T)).

If the derivatives fz and fzz exist, are continuous on n, and

fz :=; 0, fzz 2: ° on n,
then there exist the sequences (Uk) increasing and (Vk) decreasing which
converge uniformly and quadratically on [0, tl] to the unique solution u of
(12.10) satisfying Uo :=; u :=; Vo.

An interesting problem is to obtain nonsmooth variants of the quasilin-

earization methods in terms of divided differences instead of derivatives. A
partial result on this point can be found in Balazs~Muntean [5] and Goldner~
Trimbita§ [24]. An extension of the generalized quasilinearization method
in order to obtain convergence of order p> 2 is due to Deo~McGloin Knoll
[21]. Another interesting problem is to express the conditions of such type
of results, in terms of convex or quasiconvex functions of superior order
(see Popoviciu [43], Popoviciu [42] and Cristescu [16]). A general theory in
ordered Banach spaces of both monotone iterative technique and general-
ized quasilinearization method with applications to semilinear problems is
developed in the forthcoming paper by Buidl~Precup [9].
Index

abstract Hammerstein equation, 129 Dirichlet problem, 137

abstract Poincare inequality, 188 discrete continuation principle, 154
abstract weak maximum principle, duality, 85
189
adjoint operator, 86 Egorov theorem, 64
Ambrosetti-Rabinowitz theorem, 112 Ekeland principle, 113
Ascoli-Arzela theorem, 15 energy norm, 131
average function, 19 energy space, 131
equicontinuous, 16
bounded operator, 25
Brouwer theorem, 28 finitely-valued function, 62, 64
first eigenvalue, 137
Caratheodory conditions, 62 flow, 119
Caratheodory-(p, q) function, 62 Frechet derivative, 98
coercive functional, 105 Frechet-Kolmogorov theorem, 21
compact embedding, 27 Fredholm integral equation, 37
compact metric space, 14 Fredholm linear integral operator,
comparison lemma, 202 66
completely continuous embedding, Fredholm operator, 35
27
completely continuous operator, 25 Gateaux derivative, 98
concave functional, 105 generalized metric space, 151
cone, 163 generalized quasiliniarization, 201
continuous embedding, 27 Golomb functional, 103, 130
contractive mapping, 152 gradient, 99
convergent to zero matrix, 152
convex functional, 105 Hammerstein integral equation, 46
critical point, 103 Hammerstein integral operator, 68
Hausdorff theorem, 14
decreasing operator, 167 Holder inequality, 19
delay integral equation, 39
delay integral operator, 40 increasing operator, 167
Dirichlet conditions, 47 initial value problem, 52

217
218 Index

Jacobian matrix, 200 positive operator, 87

positive semidefinite, 110
Kantorovitch theorem, 196 potential, 102
potential type function, 102
Laplacian, 137
pseudo-gradient, 119
Leray-Schauder condition, 43
Leray-Schauder principle, 43 quasiliniarization method, 200
level set, 104
lower semicontinuous, 104 regular cone, 165
lower solution, 169 relatively compact set, 14
LP space, 18
saddle point, 111
matrix kernel, 159 Schauder theorem, 33
maximal solution, 169 Schechter's mountain pass theorem,
Mazur lemma, 33 121
Schechter's Palais-Smale condition,
measurable function, 62
minimal solution, 169 121
self-adjoint operator, 86
minimizing sequence, 104
semi-monotone norm, 165
monotone iterative principle, 168,
Sobolev space, 71
171
splitting, 90
Nemytskii operator, 61 square root operator, 88
net, 14 strictely convex functional, 105
Neumann conditions, 47 strongly measurable function, 64
Newton iterates, 200
translation, 20, 68
Newton method, 196
two-point boundary value problem,
normal cone, 165
50
operator of finite rank, 25 uniform limit of sets, 15
order interval, 164 upper solution, 169
order of convergence, 196
order relation induced by a cone, variational form, 85
164 variational method, 96
ordered Banach space, 164 vector-valued metric, 151
ordered linear space, 164 vector-valued norm, 160
Vitali theorem, 63
Palais-Smale condition, 111 Volterra integral equation, 38
periodic conditions, 47 Volterra operator, 37
Perov theorem, 153 Volterra-Hammerstein equation, 72
point of local minimum, 103
positive cone, 164 weakly lower semicontinuous, 104