O'Regan-Existence Theory For Nonlinear Ordinary Differential Equations

Existence Theory for Nonlinear Ordinary Differential Equations
Mathematics and Its Applications
Managing Editor:
M. HAZEWINKEL
Centre/or Mathematics and Computer Science, Amsterdam, The Netherlands
Volume 398
Existence Theory for

Nonlinear Ordinary
Differential Equations
by
Donal 0 'Regan
Department of Mathematics,
University College Galway,
Galway, Ireland
SPRINGER-SCIENCE+BUSINESS MEDIA, B.Y.
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-4835-6 ISBN 978-94-017-1517-1 (eBook)

DOI 10.1007/978-94-017-1517-1
Printed on acid-free paper
All Rights Reserved
@1997 Springer Science+Business Media Dordrecht

Originally published by Kluwer Academic Publishers in 1997
No part ot- the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any infonnation storage and
retrieval system, without written permission from the copyright owner
To
Mary Rose, Francis, Gerard and Marguerita.
CONTENTS
Chapter 1.
Introduction, notation and preliminaries .......................... 1
Chapter 2.
Fixed point theory ............................................... 4
Chapter 3.
Initial value problems ........................................... 11
Chapter 4.
First order periodic problems .................................... 22
Chapter 5.
Existence principles for second order boundary value problems ... 34
Chapter 6.
Boundary value problems without growth restrictions ............ 46
Chapter 7.
Positone boundary value problems ............................... 58
Chapter 8.
Semi~positone
Chapter 9.
Differential equations singular in the solution variable ............ 82
Chapter 10.
Existence principle for singular boundary problems ............ 105
Chapter 11.
Nonresonance problems in the limit circle case ................. 116
Chapter 12.
Resonance problems in the limit circle case .... , .... , .......... 133
Chapter 13.
Boundary value problems on the half line ...................... 156
boundary value problems ......................... 67
Chapter 14. Existence theory for ordinary differential equations on compact and
noncompact intervals ......................................................... 164
Chapter 15.
Impulsive differential equations ............................... 174
Chapter 16.
Differential equations in abstract spaces ....................... 186
Chapter 1. Introduction, notation and preliminaries.

This book discusses the topological approach to establishing existence
to nonlinear differential equations subject to initial and boundary data. We
try to present some of the more advanced results to date in this field. Also
some well known results are presented in a new way. However, as might
be expected, the selection of topics reflects the particular interests of the
author. We refer the reader to [1-7] for other topics of current interest.
Chapter 2 presents the fixed point theory which will be used throughout
this book. In particular we present two very general fixed point results for
maps between Hausdorff locally convex linear topological spaces. The first
is a result of Leray-Schauder type whereas the second is a recent result due
to Furi and Pera. Chapters 3 and 4 discuss the Cauchy and periodic first
order problems respectively. Here monotonicity and growth conditions are
used to establish existence. These concepts will be made more precise in
the appropriate chapters. In chapter 5 we present very general existence
principles for second order systems of boundary value problems. We will
discuss two cases which arise naturally: the first concerns the case when the
appropriate linear differential operator is invertible and the second discusses
the case when the linear operator is not invertible. Indeed the principles
in chapter .5 could be formulated for n-th order systems. In chapters 6 to
12 the existence principles in chapter .5 are used to establish a variety of
existence results for second order problems. For example in chapter 6 the
notion of upper and lower surfaces is introduced. In particular we show
that if a second order problem has a upper and lower solution and a upper
and lower surface then the problem has a solution. Chapters 7 and R discuss
positone and semi-positone superlinear problems. Wp present an existence
arguement based on showing that no solutions of an appropriate family
of problems lie on the boundary of a suitable open set. As a result our
technique will handle existence for problems wherein the set of norms of all
solutions is unbounded. Chapter 9 discusses problems which are singular
both in the dependent variable and the independent variable. In chapters
10, 11 and 12 resonance and nonresonance problems are discussed in detail.
For example in chapter 12 we present a Landesman-Lazer type result for
singular resonant second order problems. Boundary value problems on
the noncompact interval are discussed in chapters 13 and 14. Existence is
established for a variety of problems via the Arzela-Ascoli theorem and the
fixed point theorem of Furi-Pera type established in chapter 2. Chapter 15
is devoted to the study of first order impulsive differential equations. Much
attention has focused on these problems recently. We present some results
1
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
from the recent literature on impulsive differential equations with variable

(and constant) times. Finally chapter 16 discusses differential equations in
abstract spaces; our existence results rely on the Krasnoselskii fixed point
theorem.
The remainder of this chapter contains notation and some preliminary
results used throughout this book. Euclidean n-space is denoted by R n
and its norm by Ixl. C(I) is the Banach space of continuous functions
(with values in a given Euclidean space) on the interval I = [a, b) equipped
with the maximum norm lulo = max{ lu(x)1 : x E I}. Likewise, cm(I)
is the Banach space of functions whose rn-derivatives are continuous on I
equipped with the norm
lul m = max{ lulo, IU'lo, ..... , lu(m)lo}.

For 1 ~ p < 00, U(I) denotes the Banach space of p-th power integrable
functions with
As usual, L=(I) is the Banach space of essentially bounded measurable

functions together with the essential supremum norm, which is the limit
as p ---+ 00 of the LP-norm. It is convenient to denote the Sobolev class of
functions u such that u(m-l) is absolutely continuous and u(m) E U(I)
by Wm,P(I).
Next we record two well known fixed point results which we will use in
this book. If T : X -+ Y with X a subset of Y then we call x in X
a fixed point of T if x = Tx. Let (X, d) be a metric space. A function
T : X ---+ X is a-contractive if there is a constant a such that 0 ~ a < 1
and d(Tx,Tx') ~ ad(x,x') for all x, x' in X. We call T : X ---+ X a
contraction or contractive if it is a-contractive for some a.
Theorem 1.1. (Banach contraction mapping them'em)
If T : X ---+ X is contractivf: on a complete metric space X then T has a
unique fixed point in X.
A function T : X -, Y between two metric spaces is compact if its
range is a relatively compact subset of Y. Similarly, T is completely
continuous if it maps each bounded subset of X into a relatively compact
subset in Y. In other words, T is completely continuous if it is compact
on each bounded subset in its domain.
Theorem 1.2. (Schauder fixed point theorem)

Let J( be a convex subset of a normed linear space E. Each compact map
T : J( -+ J( has a fixed point.
Also throughout this book the Azela-Ascoli theorem will be used For
convenience we state it here.
Theorem 1.3. (Arzela-Ascoli theorem)
A subset M of C([a, b], Rn) is rdatively compact if and only if it is bounded
and equicontinuous.
References.
[1]. R. P. Agarwal, Boundary value problems for higher order differential

equations, World Scientific, Singapore, 1986.
[2]. A. Ambrosetti and G. Prodi, A primer of nonlinear analysis, Cambridge
studies in advanced math., 34, Cambridge, 199,).
[3). R. F. Brown, A topological introduction to nonlinear analysis,

Birkhauser, Boston, 1993.
[4]. K. Deimling, Multivalued differential equations, Walter de Gruyer,

Berlin, 1992.
[,5]. A. Granas and M. Frigon (Editors), Topological methods in differential equations and inclusions, NATO ASI Series C, Kluwer Acad. Publ.,
Dordrecht, 1995.
[6]. V. Lakshmikantham and S. Leela, Nonlinear differential equations in
abstract spaces, Pergamon Press, Oxford, 1981.
[7). J. Mawhin, Topological degree methods in nonlinear boundary value
problems, AMS Regional Series in Math., 40, Providence, 1979.
Chapter 2. Fixed point theory.

2.1. Introduction and preliminaries.
This chapter presents some general fixed point results for compact mappings between locally convex Hausdorff linear topological spaces. We first
prove a nonlinear alternative of Leray-Schauder type [4,7]. In its simplist
form it can be stated as follows.
Theorem 2.1. Let E be a locally convex Hausdorff linear topological space

with C a complete, convex subset of E. Assume U is a relatively open
subset of C with E U and F : U --+ C a compact map. Then either
(Ai). F has a fixed point; or
(A2). there is a point u E {)U and A E (0,1) with u = AFu.
When E is a Banach space theorem 2.1 will be used extensively in this

book to establish existence results for differential equations. However if E
is not a Banach space, say if E = C[O, 00) and we are discussing differential
equations on the semi infinite interval, then theorem 2.1 cannot be easily
applied. To see this notice in applications that we want to take U to be a
bounded set. If U is a subset of C and U is open in E then U cannot
be bounded since locally convex Hausdorff linear topological spaces do not
admit bounded open sets unless the space is normable. In 1987, Furi and
Pera [6] obtained a more applicable fixed point result for compact maps, in
the spirit of the Schauder-Tychonoff theorem. This result will be presented
in section 2.2.
Section 2.2 has two main parts. We first establish a topological transversality theorem for compact maps between locally convex Hausdorff linear
topological spaces. This result was established by Granas in 1976. The
topological transversality theorem will then be used to prove various fixed
point results which will be used in this book.
Finally in the introduction we present three standard results from functional analysis which will be used in section 2. The first result is known as
the Schauder-Tychonoff fixed point result in the literature [1].
Theorem A. Let E be a locally convex Hausdorff linear topological space

and C a nonempty complete convex subset of E. Suppose f : C --+ C is
a continuous, compact (i.e f( C) is relatively compact) mapping. Then f
has a fixed point.
Theorem B. [9]. Every topological Hausdorff linear space is Tychonoff
(T3l ).
2
Theorem C. l5}. If A is a compact subset of a Tychonoff space X then

4
for every closed set B disjoint from A there exists a continuous function
fL : X ----'- [0,1] such that fLex) = 1 for x E A and fLeX) = for x E B.
2.2. Fixed point theory.

Let C be a complete convex subset of a locally convex Hausdorff linear
topological space E. Now let X ~ C and A ~ X with A closed in X
and X closed in C.
Definition 2.1. We let PAC X, C) denote the set of all continuous, compact
mappings F: X ----'- C such that F is fixed point free on A.
Remark. A map F: X
-+
We call N : X x [0,1]
relatively compact.
C is compact if F(X) is relatively compact.

---->
C a compact mapping if N(X x [0,1]) is
Definition 2.2. A map F E PA (X, C) is essential if every map in

PA(X, C) which agrees with F on A has a fixed point. Otherwise F
is inessential i.e. there exists a fixed point free map G E PA(X,C) with
G = F on A.
Definition 2.3. Two maps F,G E PA(X,C) are homotopic in PA(X,C)
written F ~ G in PA (X, C) if there is a continuous, compact mapping
N : X X [0,1] -+ C with N t ( u) = N( u, t) : X -+ C belonging to PA(X, C)
for each t E [0,1] and No = F, Nt = G.
Remark. Notice
is an equivalence relation in PA(X, C).
Theorem 2.2. Let C, X, A, E be as above with FE PA(X, C). Then the

following conditions are equivalent:
(i). F is inessential
(ii). there is a fixed point free map G EPA (X, C) such that F ~ G in
PA(X,C).
PROOF: We first show that (i) implies (ii). Let G E PA(X,C) be a fixed
point free map with G = F on A. Define N : X X [0,1] -+ C by
N(;}:, t) = tG(x)
+ (1 -
t)F(x).
Clearly N is a continuous, compact map; to see this it is enough to notice

that
N(X X [0, 1]) ~ co (G(X) U F(X)) ,
and to recall [13] that if J( is a compact subset of C then co( J() is
compact. Also since F = (; on A and G is fixed point free on A we have
for x E A that
Nt(x)
= tG(x) + (1 -
t)F(x)
= tG(x) + (1
- t)G(x)
= G(x) # x,
so Nt is fixed point free on A for each t E [0,1). It remains to notice

that Nt E PA(X, C) for each t E [0,1). Finally No = F and Nl = G so
F ~ G in PA(X,C).
We next show that (ii) implies (i). Let N : X X [0,1) ---+ C be the
continuous, compact mapping from G E PA(X, C) to F with No = G
and Nl = F. In particular Nt is fixed point free on A for each t E [0,1).
Let
B = {x EX: x = N(x,t) for some t E [0, I)}.
If B is empty then in particular F = Nl has no fixed points and so F

is inessential. So it remains to consider the case when B is nonempty.
First note B n A = 0. Also B is closed. To see this let (x",) E B be
a net of points of B (i.e x", = N(x"" t",) converging to x. Without loss
of generality assume t", converges to t E [0,1). By the continuity of N
it follows that x = N(x, t) so x E Band B is closed. Also B is a
precompact subset of X since N : X x [0,1) ---+ C is a compact map and
B ~ N(B x [0,1]). Now since B is a closed, precompact subset of X ~ C
and C is complete then B is compact [2,8).
Theorem B implies E (and hence X, with subspace topology) is Tychon off. In addition since An B = 0 then theorem C implies that there is
a continuous function J.L : X ---+ [0,1) with J.L(A) = 1 and J.L(B) = O. Define
J :X
---+
C by J(x)
= N(x,J.L(x.
Clearly J is continuous and compact. We claim that J : X ---+ C is a fixed

point free map with J = F on A. If this is true then J E PA(X,C) and
J is a fixed point free map which agrees with F on A. Consequently F
is inessential and we are finished.
It remains to prove the claim. J is fixed point free since J( x) = x
means N(x,j1(x = x which implies x E B and so J.L(x) =
(i.e
N(x,O) = x), a contradiction since N(x,O)
G(x) is fixed point free.
To see that J = F on A notice if x E A then J.L( x) = 1 and so
J(x) = N(x,j1(x) = N(x, 1) = F(x). 0
Theorem 2.3. Let C, X, A, E be as above. Suppose F and G are two

maps in PA(X,C) such that F ~ G in PA(X,C), Then F is essential
iff G is essential.
PROOF: If F is inessential then theorem 2.2 guarantees a fixed point free
map T E PA(X, C) with F ~ T in PA(X, C). Thus G ~ T in PA(X, C)
and so G is inessential by theorem 2.2. Symmetry will now imply that F
is inessential iff G is inessential. 0
6
Let C be a complete conva subsd of a locally convex

Hausdorff linear topological space E and U an open subset of C with
110 E U. The constant map F: U ---. Uo is eS8(,Tltia/ in PaU(U,C),
Theorem 2.4.
Remark. Notf'
C respectively.
and aU denote the c10su rE' an d the bou ndary of U in
r-
PROO F: Lf't G:
C be any map in !\'w(r. C) with Glau
uo. We must show G has a fixed point in U. DE'fillf'
L( x)
={
= FI8U =
G( x), x E
uo, x E cle.
It is easy to see that L : C - C is a continuous. compact map. Now

theorem A guarantees that L has a fixed point, say .11. in C (i.e. y = L(y)).
In addition since L(x) = Uo for x E CIU and Uo E U we have y E U.
Thus y = G(y) and since Glau = 1to we have .11 E C. Hence G has a fixed
point y E C so F is essential. D
We now use theorem's 2.:3 and 2.4 to establish to fixed point theorems
of Leray-Schauder type.
Theorem 2.5.
Let C be a complete convex 8llbsd of a locally convex

HausdOlff linfa1" topological space E and U an open subset of C with
p E U. II! addition let F : U - C be a contilllLOlis. compact map. Then
either
(A J). F has a jiu:d point in [T; or
(A:!). there is a !L E aU and A E (0, I) with II = AF(u) + (\- A)".
P ROO F: We assume FIDU is fixed point freE' for otherwise (A \) is satisfi .. d.

Let G : U - C be the constant map U f-- P and ronsidE'r the homotop.';
N : [7 X [0,1] - C joining G and F given by
N(u,t)
= tF(u) + (1 -
t)(;(ll).
It is easy to see that N : U X [0, 1] ~ C is a continuous. compact map.

Now either 1\-t is fixed point free on aU for each t E [0,1] or it is not.
If Nt is fixed point free on aU for each t E [0. I] then theorem's 2.:3 and
2.4 imply that F must have a fixed point so (A 1) occurs. If Nt is not
fixed point free on aU for each t E [0,1] then there exists 1t E aU with
1t = AF(u) + (1 - A)p for some A, 0 :S A:S 1. l"ow Ai- 0 since p E U and
Ai-l since Flau was assumed to be fixed point free. D
Essentially the same reasoning as in theorem 2 ..1 establishes thE' following more general result of Leray-SchaudE'r typE'. We will formulate it for
the case when E is a Banach space since we will only use it in this context.
Theorem 2.6. Let C be a closed convex subset of a Banach space E and

U an open subset of C with p E U. Suppose F: U ---. C is a continuous,
compact map. In addition assume there is a continuous, compact mapping
N:Ux[O,l]---'C with Nt(u)=N(u,t):U---.C belonging to Pau(U,C)
for each t E [0,1], and Nl = F, No = p. Then either
(Ai). F has a fixed point in U; or
(A 2). there is a u E aU and A E (0,1) with u = N>.. (u).
We now use theorem 2.5 to establish the Furi-Pera fixed point result.
Theorem 2.7. Let C be a complete convex subset of a metrizable (metric
d) locally convex linear topological space E with Q a closed convex proper
subset of C, 0 E Q, and
Ui
= {x
E E : d( x, Q)
for' i sufficiently large. Assume F : Q

In addition suppose
(2.1)
--->
< ;.}
z
~C
C is a continuous, compact map.
if {(Xj, Aj)}.f=l is a sequence in aQ x [0,1] converging

{ to (x, A) with x = A F(x) and 0:::; A < 1, then
there exists ]0 E {I, 2, .... } with Aj F(xj) E Q
for each j 2 jo
holds. Then F has a fixed point in Q.
PROOF: Let r : E ---. Q be a continuous retraction (the existence of r

follows from Dugundji's extension theorem [4,14]).
Remarks. (i). If 0 E int Q we may take
r( x)
= max {1,J-l ()}

x , xEE
where J-l is the Minkowski functional on Q i.e.

J-l( x)
(ii). If int Q
=0
then aQ
= inf {a > 0 :
= Q.
x E a Q}.
By the above remarks, we may choose (and we do so) the retraction r

above so that r(z) E aQ if z E E/Q. Consider
B={xEC: x=Fr(x)}.
Firstly B =I 0. To see this notice since r is continuous that Fr: C ---;. C

is a continuous, compact map. Theorem A implies that Fr has a fixed
8
point so B =I 0. We next claim that B is closed. To see this let (xc be

a net in B with Xc> ~ Xo E C. By the continuity of F r it follows that
Xo = Fr( xo) so Xo E B. Consequently B is closed. In fact B is compact
since F: Q ~ C is compact and B ~ Fr(B) ~ F(Q).
It remains to show B n Q =I 0. To do this w' argue by contradiction.
Suppos' B n Q = 0. Then sinc' B is compact and Q is closed ther' exists
b > 0 with dist(B,Q) > b. Choos' mE {1,2, ... } such that 1 < bm and
Ui ~ C for i E {m,m+ I, ... }; here
Ui = {
E E : d( x, Q)
<
~}
for i E {m, m
+ 1, .... }.
Fix i E {m,m+I, ... }. Since dist(B,Q) > b then BnUi = 0. Also

Ui is open, 0 E Ui, and Fr : Uj ~ C is a continuous, compact map.
Now theorem 2.5 implies (since B n Uj = 0) that there exists (Yi, Ad E
8Ui X (0, 1) with Yi = Ai F r (y;). Notice in particular since Yi E aUi that
Ai Fr(Yi) ~ Q for each i E {m, m
(2.2 )
+ 1, ... }.
Now consider
D={:rEE: x= AFr(x) forsoIll' AE(O,I]}.

It is easy to check that D is closed and compact (notic' D ~ co (F( Q) U
{O})). This together with d(Yj,Q) = j-, IAjl ::; 1 (for] E {m,m + l, ... })
implies that w' may assum' without loss of generality that Aj ---; A* and
Yj --4 Y* E 8Q. Also since Yj = Aj F r (Yj) w' have that y* = X" F r (y*).
If A* = 1 then y* = F r (y*) which contradicts B n Q = 0. Hence we may
assume 0 ::; A* < 1. But in this case, (2.1) with Xj = r(Yj) E 8Q, x =
y* = r( y*) implies that there exists ]0 E {I, 2, ... } with Aj Fr( Yj) E Q for
each ] ~ jo. This contradicts (2.2). Thus BnQ =10 so there exists x E Q
with :1:= Fr(x) i.e. x=F(x). 0
References.
[1]. C.Corduneanu, Integral equations and applications, Cambridge Univ.

Press, New York, 1990.
[2]. R. Cristescu, Topological vector spaces, Noordhoff Int. Publ., Leyden,

1977.
[3]. J .Danes, Generalized concentrative mappings and their fixed points,

Comment. Math. Univ. Carolinae, 11(1970), 115-1:36.
[4]. J.Dugundji and A.Granas, Fixed point theory, Monografie Matematyczne, PWN, Warsaw, 1982.
[5]. R.Engelking, General topology, Heldermann Verlag, Berlin, 1989.

[6]. M.Furi and P.Pera, A continuation method on locally convex spaces
and applications to ordinary differential equations on noncom pact intervals,
Ann. Polon. Math., 47(1987), 3:31-:346.
[7]. A. Granas, Sur la methode de continuite de Poincare, C.R. Acad. Sci.

Paris, 282( 1976), 983-985.
[8]. J .Kelley, General topology, D. Van Nostrand Company, Toronto, 195.5.
[9]. G.Kothe, Topological vector spaces I, Springer Verlag, New York, 1983.
[10]. W.Krawcewic, Contribution ala tMorie des equations non lineaires
dans les espaces dE' Banach, Diss. Math., 273, Warsaw, 1988.
[11]. D.O'Regan, A fixed point theorem for condensing operators and applications to Hammerstein integral equations in Banach spaces, Computers
and Mathematics with Applications, 30( 199.5), :39-49.
[12]. A. Potter, An elementary version of thE' Leray-Schauder theorem,
Jour. London Math. Soc., 5(1972), 414-416.
[13]. F.Treves, Topological vector spaces, distributions and kernels. Academic Press, New York, 1967.
[14]. E.Zeidler, Nonlinear functional analysis and its applications, Vol I,
Springer Verlag, New York, 1986
10
Chapter 3. Initial value problems.

3.1. Introduction.
We begin our applications of fixed point methods with existence of
solutions to certain first order initial initial value problems. This problem
is relatively easy to treat, illustrates important methods, and in the end
will carry us a good deal further than may first meet the eye. Thus, we
seek solutions to
{ Y'.= I(t,y)
yeO) = r
(1.1 )
where I: I X R n ---+ R n and I = [0, b]. We shall seek solutions that are defined either locally or globally on I, according to the assumptions imposed
on I. Notice that (1.1) is a system of first order equations because I takes
its values in Rn. In section 3.2 we will first establish some basic existence
theorems which guarantee that a solution to (1.1) exists for t > 0 and near
zero. Familiar examples show that the interval of existence can be arbitrarily short, depending on the initial value r and the nonlinear behaviour
of I. As a result we will also examine in section 3.2 the dependence of the
interval of existence on I and r. We mention in passing that, in the results
which follow, the interval I can be replaced by any bounded interval and
the initial value can be specified at any point in I. The reasoning needed
to cover this slightly more general situation requires minor modifications
on the arguments given here. In this chapter we also present the notion of
upper and lower solution for initial value problems.
3.2. Existence theory.
We assume that I is at least continuous in the first three theorems that
follow. Then, evidently, y E C 1 (1) solves (1.1) if and only if y E C(1)
solves
yet)
= r + lot I(s,y(s))ds.
Define an integral operator T : C(1)

Ty(t)
->
C(l) by
= r + lot I(s,y(s))ds.
Then the equivalence above is expressed briefly by y solves (1.1) if and

only if
y = Ty, T: C(1) ...... C(1).
11
In other words classical solutions to (1.1) are fixed points of the integral
operator T. We first present the classical Picard-Lindelof theorem.
Theorem 3.1. Let f : I X R" ----+ R" be continuous and Lip8chitz in y;

that is, there exists () ;:: 0 such that
zl
If(t,y) - f(t.zli ~ alY -
Then then exists a unique y E (' (1) that
for all y, z E R".

80lv(8
(1.1).
PROOF: We apply the Banach contraction mapping theorem to show that

T has a unique fixed point. At first glance. it seems natural to use the
max norm on C(I), but that choice would lead us to only a local solution
defined on a subinterval of I. The trick is to use the weighted maximum
norm
on ('(I) (recall lulo = sUPllu(t)1 for appropriate functions u). Observe
that ('(J) is a Banach space with this norm because it is equivalent to
the max norm, (-ab IYlo ~ IIYII ~ Iylo. Also, T is a contraction on
(C(1), II II) To see this notice
(Ty - Tz)(t) = It[J(.s,y(s)) - f(.s,z(s))]ds
and so
e-Cdl(Ty-Tz)(t)1
< e- at 1t ()(~s('-asIY(8)-z(s)ld8
< (-at (1 t
()enSd.~)
IIy-zll
< e- at (eat_1)"y_zlI~ (l_e-nb)IIy_zll.

Since 1 - e -a b < 1, the Banach contraction mapping theorem implies that
there is a unique y in C(J) such that y = Ty; equivalently, (1.1) has a
unique solution y in ('1 (1). 0
It is worth mentioning here that the proof just given works for the
Cauchy problem in a Banach space; that is, for f : I X E ----+ E, with E a
Banach space. Next we prove the classical Peano theorem.
Theorem 3.2. Let f : I
X R" -+ R" bt, continuous and bounded. Then

there (xists at least one y E ('1 (I) that solves (1.1).
12
PROOF: We still have (1.1) is equivalent to
= Ty,
T: C(I) -. C(I).
Now T need not be a contraction (as in theorem 3.1) but it is a compact

map of C(I), as we can easily check using the Arzela-Ascoli theorem.
Consequently, the Schauder fixed point theorem provides us with a fixed
point for T and a classical solution of (1.1). It remains to check that T is
a compact map of C(I) equipped with the max norm. The continuity of
T comes from the continuity of f. Indeed
Iy" as n
--> 00
Ylo
-+
0 implies Pn = maxlf(,~,yn(s)) - f(s,y(sl-. 0
by the uniform continuity of
f on compact sets. Then
and so
ITYn - Tylo S Pn b ...... 0 as n
--> 00.
So T is continuous. Next, If(t, y)1 S M for some constant M because f

is bounded. Therefore, for any y in C(I),
ITy(t)1
= Ir + lot
f(s,y(s))dsl S
Irl + M b
so ITYlo S
Irl + M b,
and
ITy(t) - Ty(t')1 = Ilt f(s, y(s))dsl S Mit t'
t'l
Thus T(C(I)) is a bounded, equicontinuous subset of C(I); hence, T is

compact by the Arzela-Ascoli theorem. As noted above, the Schauder fixed
point theorem implies that (1.1) has a solution in C 1 (I). 0
Next, we consider the existence question for (1.1) when f is continuous
but may not be bounded or satisfy a Lipschitz condition. If continuity
alone is assumed, we can only expect to find a local solution, as the following
example shows. The initial value problem y' = y2, yeO) = 1 has the unique
solution y = -(1 - t)-1 defined only for 0 S t S 1 despite the fact that
f(t, y) = y2 is defined for all t and all y.
Theorem 3.3. Let f : I X Rn -+ Rn be continuous. Then (1.1) has at
least one local solution. That is thel'e exists a subinterval I' = [0, b'] with
o < b' S b and at least one y E (,'1 (I') that solves (1.1) for t in I'.
PROOF: Let S = {(t,y) : tEl and Iy - rl ~ I}. For P in I X Rn,

let Q be the closest point in S to P. The map P ---> Q is continuous.
Define g : I X Rn ~ Rn by g(P) = f(Q). Clearly g is continuous,
bounded and g = f on S. By theorem 3.2 there is y in C 1( I) that solves
y' = g(t,y), yeO) = r for t in I. Let
b'
= sup{tE/:
(,~,y(.~ E S for 0 ~
Evidently, 0 < b' ~ b. Since g = f on S, we have y'(t)

t in I'. Consequently, y is a local solution to (1.1). 0
t}.
= f(t, y(t
for all
Now we relax the continuity assumption on f and expand the notion of

a solu tion to (1.1) accordingly. We want to do this in a way that preserves
the natural equivalence between (1.1) and the integral equation y = Ty
which was obtained by integrating the Cauchy problem. To this end, we
follow ideas of Caratheodory and make the following definitions.
Definition 3.1. A function y E W1,P(/) is an LP-Caratheodory solution

to (1.1) if y solves (1.1) in the a.e. sense on I.
Definition 3.2. A function f : I X R" ---> Rn is LP-Caratheodory function
if
C1. The map y f-;. f(t, y) is continuous for a.e. t
C2. The map t ........ f(t, y) is measurable for all y
C3. For every c > 0 there exists he E LP(/) such that Iyl ~ c implies
If(t,y)1 ~ hc(t) for a.e. tEl.
In these definitions, I can be any interval. If I is bounded, then
any LP-Caratheodory function is L1-Caratheodory. Furthermore, if as
we shall usually assume, I is compact, then any continuous function is
an LP-Caratheodory function for any p. Finally, we say f is locally LPCaratheodory on I X Rn if f restricted to K X Rn is LP-Caratheodory for
every compact interval K ~ I. If f is LP-Caratheodory then y E W1,P(/)
solves (1.1) if and only if
Y E C(/) and y(t)
= r + lot f(s,y(sds.
In fact, Cl, C2 imply that the integrand on the right is measurable for
any measurable function y and C3 guarantees that it is integrable for any
bounded measurable y. The stated equivalence now is clear. So just as in
the continuous case, (1.1) has a solu tion y if and only if
y
= Ty,
T: C(I)
14
--->
C(I).
We now establish the analogues of theorems 3.1, 3.2 and 3.3 when f is a
LP-Caratheodory function and when' we seek LP-CaratModory solutions.
Theorem 3.4. Let f : I X Rn - Rn be LP-Camtheodory and LPLipschitz in y; that is, there exists 0: E LP(I) such that
If(t, y) - f(t,
z)1
S; o:(t)
Iy - zl
for all y, z ERn.
Then there exists a unique y E W 1 ,p (l) that solves (1.1).
PROOF: The proof is very similar to theorem 3.1 and will only be sketched.
Let
A(t) =
lot 0:(8) ds.
Then A'{t) = o:(t) for a.e. t. Define
lIyll ==
le-A(t)
y(t)lo.
This norm is equivalent to the max norm because e- 11a% IYlo S; lIylI S; IYlo.
Thus (C(I),II.II) is a Banach space and use of the Banach contraction
mapping theorem, essentially as in the proof of theorem 3.1, implies that
there is a unique Y in C(I) such that Y = Ty. It follows that (1.1) has a
unique LP-CaratModory solution on I. 0
Remark. The same proof works for differential equations in a Banach space;
that is, when f: I X E - E. Furthermore, if 0: E Ll[O, 00) and f is LlCaratModory, the line of reasoning above can be carried out in the space
of bounded continuous functions to get an L 1-Caratheodory solution on
the half line.
We next prove the analogue of Peano's theorem in the CaratModory

setting.
Theorem 3.5. Let f : I X R" - Rn be LP -Camtheodory and integmbly

bounded. Then there exists at least one y E W 1'P(I) that solves (1.1).
PROOF: As in theorem 3.2, we show that T is a compact map on C(I).
Then the Schauder fixed point theorem gives the desired fixed point of
T and CaratModory solution of (1.1). We first observe that T is continuous because f is LP-CaratModory. Let IYn - Ylo - 0 as n - 00.
Then IYnlo, IYlo S; c, for some constant c. By properties C1 and C3 of a
CaratModory function, there is a function he in LP(I) such that
Pn(s)
= If(s,Yn(s
- f(s,y(s))I- 0 a.e. on I,
15
and
IPn(s)1 :S 2he(s) a.e. on I.

It follows from the Lebesgue dominated convergence theorem that T
continuous:
IS
and so
ITy" - TYlo :S fob p,,( s) ds
---+
0 as n
---+ 00.
Next, since T is integrably bounded, there is a function h in L1(/) such

that If(t, y)1 :S h(t) a.e. on I. Consequently, for any y in C(/), we have
ITy(t)1 =
Jr+ fot
and
ITy(t) -Ty(t')J
f(s,y(s))dsl:S
IrJ + fob h(s)ds
= Ilt f(s,y(,~))dsJ:S
t'
Ilt h(s)dsl
t'
The last two inequalities and the absolute continuity of the Lebesgue integral imply that T(C(I)) is a bounded, equicontinuous subset in C(I);
hence, T is compact. 0
The reasoning used in theorem 3.5 to show that T is compact can
be used to show that T maps any bounded set of continuous functions
into a bounded, equicontinuous set of functions when f is only an LPCaratheodory function. In the proof above, simply replace h by he obtained from property C3 of a Caratheodory function, where c is a bound
for the functions in the bounded set. Consequently, the reasoning used
to prove theorem 3.5 actually proves the following general result that will
playa key role later in the establishment of general existence principles for
nonlinear boundary value problems in chapter 5.
Theorem 3.6. Let f : I X R m ---+ R n be P -Caratheodory function.

Define its associated Caratheodory operator F : C[a, b] ---> Co[a, b] by
Fy(t)
It
f(s,y(s))ds.
Then F is continuous and completely continuous.

The basic local existence result for CaratModory solutions can also be
proved from theorem 3.2 using the same reasoning and auxiliary problem
16
that enabled us to deduce theorem 3.:3 from theorem :3.2. The result, due
to Caratheodory, is as follows.
Theorem 3.7. Let f : [ X Rn ---+ Rn be P -Caratheodory. Then (1.1)

has at least one local solution. That is then: exists a subinterval [' = [0, b/]
with < b' -:; b and at least one y E W1,P(//) that solves (1.1) for a.e. t
in 1'.
The interval of existence of the local solution obtained when f IS

Caratheodory depends (as we will see in theorem :3.8) on the rate of growth
of the nonlinearity f. The technique we will give was introduced in 1988
by Lee and 0 'Regan [6J.
Theorem 3.8. Let f : [ X R n ---+ R n be an LP -Camtheodory function.

Assume If(t, y)1 -:; "p(lyl) where 1/) : [0,00) ---+ [0,00) is Borel measurable
E LlocUrl, 00). Then (1.1) has a solution on
and a.e. positive with
[ = [0, b] for ellery
[00 du
b < Jlrl "p(u) == Too
t,
MOT-eover this result is best possible for the class of problems considered.
PROOF: By theorem 3.6, T : U -+ C(I) is a compact map for any bounded,

open set U in C(I). By the nonlinear alternative (theorem 2 ..5) if we can
find U with E U such that the family of maps {A T}O<A<l has no fixed
points on au, then T will have a fixed point in U and (1.1) will have
a solution on [0, b]. We determine such a U by establishing an a priori
bound on the possible fixed points of the operators A T for
< A < 1.
Fix b < Too. Then y = ATy implies y' = A f(t, y), yeO) = Ar. Suppose
that Iy( t)1 > Irl for some t E [0, b]. Then, in view of the initial data, there
exists 0 E [0, t) such that ly(,~)1 > 111 on (0, t) and ly(o)1 = Irl. Use
of the Cauchy Schwartz inequality and IY(,~)I = Jy(,~).Y(s) imply that
ly(s)I' -:; ly'(s)1 whenever yes) :I 0. It follows that
ly(s)I' -:; ly'(s)1
= IAf(s,y(sI-:; 1/)(ly(s)1)
and so
a.e. on (o,t)
ly(s)I'ds lIY(t)' du
= Irl --.
1/)(ly(s)l)
"p(u)
Since the integral on the right is strictly bounded away from Too independent of A there is a constant M independent of A in [0,1] such that
IYlo < M.
Now, apply the nonlinear alternative (theorem 2 ..5) to
Too > b> t - a >
-
= {y E C(/)
17
: Iylo
< M}
to obtain a fixed point y of T in U, which is also a Caratheodory solution

to (1.1).
Finally, to see that thp existence result just established is best possiblp
over the class of problems considered, takp n = 1, r = 0 and consider the
Cauchy problem y' = 'I/'(Iyl), y(O) = O. Assume that the solution to this
problem exists on [0, b] for some b. Notice that y ~ 0 on [0, b] and, hence,
[b
b = io
y'(8)' d8
[y(b)
~)(Y(8) = io
du
'l/J(u)
<
[<XJ du
io
'l,h(u)
= Too
Remark. A generalization of theorem a.l:l will be presented in chapter 1.5

(sep theorem 15.2).
For the remainder of this chapter we examine the scalar initial value
problem
{ y'(t) = J(t, yet~

yeO) = r
(2.1 )
a.e. t E [0, T]
where J : [0, T] X R -. R is a Caratheodory function. Here, by a solution

to (2.1), we mean an absolutely continuous function on [O,T] which satisfies the differential equation and the initial condition. We now prove the
existence of a solution to (2.1) between two functions 81 (lower solution)
and 80 (upper solution).
Theorem 3.9. Let J be a L 1 -Caratheod01'y Junction. Assume then; exist

two Junctions So, Sl such that
(i). Si = bi + Ci with Ci E C[O, TJ, and bi a Junction oj bounded variation

on [O,T], i = 0,1;
(ii). Sl(O+) ~ r ~ so(O+), Sl(t) ~ 80(t) J01' all t E [O.T];
J:
J:
(iii).
J(t,80(tdt ~ 80(b-)-80(a+).
J(t,sl(tdt ~ 81(b-)-sda+)
Jor all a < bE [O,T].
Then problem (2.1) ha,~ a solution such that Sl(t) ~ yet) ~ so(t) Jor a.e.
t E [O,T].
PROOF: Define J1 : [0, T] X R -. R by
J(t,80(t, if y> so(t)
{
!t(t, y) =
J(t, y), if ~l(t) ~ Y ~ 80(t)
J(t,81(t, If y < St(t);
1~
and consider the associated problem

(2.2)
y'(t)=ft(t,y(t)) a.e. tE[O,TJ

y(o)
= r.
A solution to (2.2) is a fixed point to the operator N1 : er[O, TJ ---. er[O, TJ

defined by
Nl(y)(t)
= r+ lot
ft(T,Y(T))dT;
here e r [0,1J = {y E e[0,1J: yeO) = r}. Observe that the operator N1

is well defined since f is Caratheodory and 80,81 are measurable and
bounded; and there exists a function h E L~ [0, TJ such that
(2.3)
Ift(t,y(t))1 :::; h(t) a.e. t E [O,TJ and all y E Cr[O,TJ.
This inequality and the Arzela-Ascoli Theorem imply that N1 is compact. Moreover, the continuity of N1 follows from the inequality (2.:1),
the Lebesgue dominated convergence theorem, and the following relation:
ft(t,Yn(t)) ~ ft(t,y(t)) a.e. t E [O,TJ

when Yn ---. Y in C,.[O, T].
Therefore, the Schauder fixed point theorem gives the existence of a
fixed point to N 1 ; that is, a solution y to (2.2).
Now, we want to show that y satisfies 81(t) :::; yet) :::; so(t) for a.e.
t E [0, TJ, and consequently, is a solution to (2.1). Assume y(t) 1:. 80(t)
a.e. t E [0, TJ. Sino' yeO) = r :::; 80(0+), there exist t1 < t2 E [0, T] such
that y(tIl:::; so(tt), y(t2) > '~0(t2)' and yet) 2: 80(t) a.e. t E (tl,h). By
the definition of ft and assumption (iii), we have
This leads to a contradiction. Similarly, we show the other inequality.
We now get immediately a result concerning upper and lower solutions

for a first order problem.
Theorem 3.10. Let f be a LI-Caratheodory function. A8sunu' there exist
two absolutely continuou8 functions 0' :::; (3 such that 0:(0) :::; r :::; /3(0) with
f(t,l3(t)) :::; /3'(t)
and
O"(t):::; f(t,o:(t))
19
a.e. t E [O,T].
Then the problem (2.1) has a solution such that aCt) :S yet) :S {jet) Jor all
tE[O,T).
Remark. A generalization of theorem 3.10 will be presented in chapter 15
(see theorem 15.4).
Theorem 3.11. Let J be a LI-Carathiodory Junction. Assume there

exist two nondecreasing Junctions -S}. So such that 81 :S So, 81 (0+) :S
r :S 80(0+), and
J(t,,~o(t)):S
0:S J(t,S1(t))
a.e. t E [O,T).
Then the problem (2.1) has a 80lution .'mch that S1 (t) :S y( t) :S 80(t) Jor
all t E [0, T).
In Theorem 3.9, we can reverse the inequalities in assumption (iii) if

assumption (ii) is replaced by a stronger one.
Theorem 3.12. Let J be a LI-Carathiodory Junction.
exist two Junctions ,~o, 81 such that
(i). 8i = bi + Ci with
on [0, T], i = 0, 1;
Ci E
Assume there
C[O, TJ, and b; a Junction oj bounded variation
(ii). S1(0+) = r = 80(0+), S1(t):S 80(t) Jor all t E [O,T);
f:
(iii).
J(t,80(t))dt 2: ,~o(b-)-so(a+),
f01' all a < bE [0, T).
f:
J(t,s1(t))dt:S sl(b-)-s1(a+)
Then the problem (2.1) has a solution 8uch that 81 (t) :S y( t) :S 80( t) Jor
a.e. t E [0, T].
PROO F: Let r* E ['~1 (1-),80(1- )]. Consider the problem
(2.4 )
Y'(t)
y( 1)
= J(t, yet))
= r*.
a.e. t E [0, T]
By making the change of variable T = 1 - t, problem (2.4) can be written

as a problem of form (2.1) on which we can apply Theorem :3.9. Therefore,
problem (2.4) has a solution with .~dt) :S yet} :S .~o(t) for a.e. t E [0, T].
By assumption (ii), we deduce that y is a solution to (2.1). 0
20
References.
[1]. A. Bielecki, Une remarque sur la methode de Banach-CaccioppoliTikhonov, Bull. Acad. Polon. Sci., 4(1956), 261-268.
[2]. L.E. Bobisud and D. O'Regan, Existence of solutions to some singular
initial value problems, Jour. Math. Anal. Appl., 133( 1988), 214-230.
[3]. M. Frigon, A. Granas and Z. Guennoun, Sur l'intervalle maximal
d'existence de solutions pour des inclusions differentielles, c. R. Acad. Sci.,
Paris, 306(1988), 747-750.
[4]. M. Frigon and D. O'Regan, Existence results for some initial and
boundary value problems without growth restrictions, Proc. Amer. Math.
Soc., 123(1995),207-216.
[5]. P. Hartman, Ordinary differential equations, Wiley, New York, 1964.
[6] . .J.W. Lee and D. O'Regan, Topological transversality:applications to
initial value problems, Ann. Polon. Math., 48(1988), 31--:36.
[7] . .J.W. Lee and D. O'Regan, Existence principles for differential equations
and systems of equations, Topological methods in Diff. Eqns. and Inclusions (edited by A. Granas and M. Frigon), NATO ASI Series C, Kluwer
Acad. Publ., Dordrecht, 1995, 239-289.
[8]. M.N. Nkashama, A generalized upper and lower solutions method and
multiplicity results for nonlinear first order ordinary differential equations,
J. Math. Anal. Appl., 140(1989),381-;395.
[9]. L.C. Piccinini, G. Stampacchia and G. Vidossich, Ordinary differential
equations in Rn, Applied Mathematical Sciences 39, Springer- Verlag, New
York,1984.
21
Chapter 4. First order periodic problems.

4.1. Introduction.
This chapter discusses the nonlinear first order differential equation
(1.1 )
y' = f( t, y) a.e. on [0, T]

y(O) ::: y(T)
where f : [0. T] X R - t R is a L1-Caratheodory function (see definition

:3.2). By placing a Landesman Lazer type inequality on our nonlinearity f
we will establish the existence of nonnegative solutions to (1.1). Of course
analogue results could be obtained for nonpositive solutions. The proof is
based on a technique initiated by Mawhin [4. chapter 6] and extended by
Nkashama and Santanilla [7] in the early 1990's.
It is worth remarking here that it is also possible to discuss the notion of
upper and lower solutions to first order periodic problems. However since
only minor modifications of the arguement given in chapter :3 are needed
we as result will not include results of this type here; instead we refer the
reader to [6] for a detailed discussion.
To conclude this introduction we state and prove a well known existence principle for first order periodic problems. Notice that the natural
linear differential operator associated with (1.1) is not invertible. However
a simple translation will lead to a satisfactory fixed point problem.
Theorem 4.1. Let f : [0, T] x R -
R lx, a L 1 -Caratheodory function
and let q E L1 [0, T] be such that (JoT q(s) ds

a constant Mo! independent of A, with jyjo
solution y to
{
for'
<
f.
1.
A.~sume
that there exists

M o, for any
= sup[O,T] jy(t)j f.
y' - q(t)y::: A[j(t,y) - q(t)y] a.c. on [O,T]

y(O) = y(T)
A < 1. Then (1.1) has a solution y E
qo, T].
Remark. By a solution to (1.1 h we mean a function y E e[O, T] n AC[O, T]

which satisfies the differential equation in (1.1 h a.e. on [0, T] and y( 0) =
y(T).
PROOF: Let C~[O,T] ::: {u E C1[0,T] : u(O)

linear operator
L: r;[O,T]-- L1[0,T]
by
u(T)} and define the
L(y)::: y' - qy.
22
It is easy to show L is invertible since

[0,1] -> 1[0, T] be defined by
cJ~r q(s) ds i-
1. Let N : e[O, T]
N(y,>')(t) = >'[J(t,y(t)) - q(t)y].

Since f is Caratheodory, N is continuous and for every bounded set f!
C[O, T] there exists hE Ll[O, T] with
N(f!
[0, 1]) ~ {g : Ig(t)1 ~ h(t) a.e. t E [0, T]}.
Therefore j 0 L -IoN: C[O, T] X [0,1] - ('[0, T] is continuow; and completely continuous, where j : C~ [0, T] -> C[O, T] is the continuous inclusion. Moreover, y is a solution of (1.1. if and only if y = j 0 L- 1 oN(y, >.).
The result follows by applying theorem 2.6. 0
Various existence results are established for the periodic problem (1.1) in
this section. We restrict our discussion to the existence of nonnegative
solutions.
Theorem 4.2. Assumc f: [0, T] X R
----+ R is a LI-CamthCodory function.

In addition supposc the following conditions an satisfied:
(2.1 )
f(t,O) ~
for a.e.
t E [O,T]
(2.2) f(t,y)=g(t,y)y+h(t,y)+r(t) fora.f.
(2.:3)
(2.6)
Ih(t, y)1 ~ <Pl(t)yC< + <P2(t) for a.f:.

t E [0, T] and y 2: 0; here
~ Q < 1
<PI, cfJ:2, r E L :2 [0, T]
(2.4 )
(2.5 )
tE[O,T] and all y2:0
thel'e exist 13, TEL:2[O,T] with 13{t) ~ g{t,y)y ~ T(t)y

for
a.e. t E [0, T] and all y ~ 0; herc T 2:
{
a.e. on [0, T] and T >
on a 8ubSf:t of [0, T]
of positive measure
thf:rf: f:xists p E Ll[O, T] with h(t, y) 2: p(t)

for' a.f:. t E [0, T] and y ~
and
(2.7)
foT[-r(t)] dt < foT
Then (1.1) has a solution in
li;~~~f[g(t,x)J;] dt + foT lil~~~f[h(t,x)] dt.

qo, T] n AC(O. T].
PROO F: Consider the family of problems
(2.8h
where 0 < .>. < 1,
y' - Ty = .>.[J*(t.y) - TY] a.e. on [O,T]

y(O) = y(T)
T
is as in (2.5) and
f*(t,y) = { /(t,O) + y, y < 0

fit, y), y 2 o.
We will show that any solution y of (2.8)" satisfies
yet) 20 for t E [0, T].
(2.9)
Let y be a solution of (2.8h. Suppose y has a negative global minimum

at to E [0, T]. Because of the periodicity we may assume to E [0. T).
Now there exists t} > to with yet) < 0 on [to, tl] and yet) 2 y(to) for
t E [to, tl]. Then
l
= ltl
0::; y(tt)- y(to) =
tl
to
to
['>'f*(t,y)
['>'/(t, 0)
+ (1- '>')Ty]dt
+ (1
- '>')TY
+ .>.y] dt < 0,
a contradiction. Thus (2.9) is true.
Remark. The above argument also shows that any solution to (2.8Jt
nonnegative.
IS
Next we claim that there exists a constant Mo with

(2.10)
IYlo
= sup yet) ::;

[O,T]
Mo
for any solution y to (2.8 h. If this is not true then there exists a sequence
(An) in (0, 1) and a sequence (Yn) (here Yn E C[O, T] n AC[O, T]) with
(2.11)
Y:, - T Yn
= '>'n(f(t, Yn) 24
T Yn] a.e. on [0, T]
and
( 2.12)
jYnlo -
00.
From (2.11) we have

Y~! - (1- An)rYn
= An[g(t,Yn)Yn + h(t,Yn) + ret)]
Integrate from 0 to T to obtain
iT
o
[get, Yn)Yn
+ h(t, Yn) + r(t)] dt =
and so
foT[-r(t)] dt
(1- An)
An
foT[g(t,Yn)Yn
This together with the fact that lim inf( Sn

for sequences Sn and tn yields
a.e. on [O,T].
iT
0
ret) Yn dt S 0
+ h(t,Yn)] dt.
+ t n ) ~ lim inf( sn) + lim inf(tn)
rTh(t,Yn)dt.
JorT[-r(t)]dt~liminf JorTg(t,Yn)Yndt + liminf
n->oo Jo
(2.13)
1HOO
Notice g( t, Yn )Yn ~ /3( t) a.e. and h( t, Yn) ~ p( t) a.e. so we may apply

Fatou's lemma to obtain
JorT[_r(t)]dt~ JorTliminf[g(t,Yn)Yn]dt+
n->oo
JorTliminf[h(t,Yn)]dt.
(2.14)
n-HX)
Let
Yn
'Un
= IYnio .
Notice l'Unlo = 1. From (2.8h n we have

I
11"
(1 - An)r 1171
+ Ang( t, Yn )11n + An[h(t,1Yn)1 + ret)]

]
Yn
and so
(2.15)
25
a.e. on
[]
0, T
and this together with (2.12) and (2.15) implies that there exists an integer
no with
Next notice since
..
An/3(t)
[j
(l- An)r(t)vn(t)+Ang(t,Yn(t))vn(t) 2 -1-1- a.e. on O,T
Yn 0
and
1
= -I-I (rYn + An[g(t,Yn)Yn <
Yn 0
r(t)vn(t) a.e. on [O,Tj
rYn])
that
a.e. on [0, Tj. Thus there exists an integer nl with for n 2 nl,
1(1- An)r(t) vn(t)
+ Ang(t,Yn(t))1,,,(t)1
::; max {r(t)vn(t), 1{:J(t)!}
a.e. on [O,Tj. This together with (2.16) implies for n 2 max{no,nl} that
Since Iv"lo
=1
there exists a constant Ml with
(2.18)
Thus for n 2 max {no, nd == n2 we have
(2.19 )
The Arzela-Ascoli theorem with a standard result in functional analysis (if
E is a reflexive Banach space then any norm bounded sequence in E has
a weakly convergent subsequence) implies that there is a subsequence S of
{n2,n2 + I, .... } with
(2.20)
Vn - v in e[O, Tj and v:, ~ v' in L 2 [0, Tj and An - A

26
as n
---+ 00
in S; here
Remark. Notice v ~
denotes weak convergence.
on [0, T] since
1Jn
on [0, T] for all n.
Let us return to the differential equation

a.e. on [0, T]
for n E S. For n E S and '1/) E L2[0,T] we have
I[ v:t 1/ dt = I[[(1 J
(2.21)
+A
An)T1)n
+ An g(t,Y,dvn]1/)dt
rT [h(t,Yn)+r(t))
IYnio
n Jo
.1)
dt.
Notice since
+ r(t)]1jJ(t) 1 <
IYnio
-
I[h(t, Yn(t))
and
IYnio ---+ 00
(2.22)
as n
lim An
n-+oo
---+ 00
1(t)I1/)(t)1 y;:(t)
+ [2(t) + Ir(t)IlI1jJ(t)1
IYnio
we have
rT [h(t, Yn(t)) + ret)] 1jJ(t) dt = 0;
10
a.e
IYnio
here n
---+ 00
in S.
Also (2.20) yields

(2.23)
lim
rT V;t 1/) dt = 10rTv' 1jJ dt;
n-+oo 10
here n
---+ 00
in S.
Now assumption (2.5) implies (as before)

(2.24)
f3(t)
-I-I
Yn
~ [Ang(t, Yn)
+ (1 -
An)r] Vn = JLn(t) ~ r(t)vn{t)
a.e. on [0, T]. Thus, since Vn ---+ v in C[O, T] as n

IYnio ---+ 00, there exists an integer n3 with
(2.25)
IJLn(t)1 ~ max{r(t)[v{t)
+ 1], IfJ(t)!}
---+ 00
in Sand
for n ~ n3 and n E S.
Consequently there exists a constant M2 with

(2.26)
Let S1 denote those n E S with n ~ n3. Notice (2.26) implies that JLn has
a weakly convergent subsequence in L2[0, T] i.e. there exists a subsequence
S2 of S1 with
(2.27)
JLn ~ JL in L2[0, T] as n ~
27
00
in .'h;
here J-L is the weak limit (as n -+ 00 in 8 2 ) in L2[0,T] of J-Ln. Now let
n -+ 00 III 52 in (2.21), using (2.22), (2.23) and (2.27), to obtain
loT v''l/Jdt = loT J-Lt/Jdt.
(2.28)
Also
v(O) = v(T).
(2.29)
Next we claim that
(2.30)
J-L{t)
2: 0 for a.e.
Let m be an integer. Fix m and let
= l.
-f ::; J-Ln(t) ::; r(t)[v(t)
Then from (2.24) there exists
Tn
n4 E 8 2 with
(2.31)
t E [O,T].
+ f]
for n 2: n4 and n E 8 2
Let
= {u E L2[0, T]
: -f ::; u(t) ::; r(t)[v(t)
+ f]
for a.e. t E [0, T]} .
Notice K is convex and strongly closed. Hence K is weakly closed [9].

Now since J-L is the weak limit (as n ~ 00 in .lh) in L2[0, T] of J-Ln and
J-Ln E K for n 2: n4, n E 52 then J-L E K. Hence
(2.32)
- f ::;
J-L{t) ::; r{t)[v(t)
We can do this for each
+ f]
for a.e. t E [O,T].
= 1-,
mE {1,2 .... }.
m
Thus
0::; J-L(t) ::; r(t) t){t) for a.e. t E [0, T]
(2.33)
and so (2.30) is true.

Now (2.30) together with (2.2R) implies that v is nondecreasing on
[0, T]. Consequently, since v(O) = veT),
(2.34)
== c 2: 0, c a constant.
Now if c = 0 we have a contradiction sinn'
(2.:35 )
'11==
Thus there exists

(t)
-Iy"-1=
Yn
n5
vn(t)
Ivlo =
1. Thus
> O.
E oS' with
c
2
2::- for each t E [0, T] and n 2: n5, n E 8.

2H
Hence
(2.36)
Yn(t) ....... 00 for each t
E [0, Tj as n
-+
00 through 8.
Now (2.36) together with (2.14) implies
~ fT
JfT[_r(t)jdt
Jo
o
liminf[g(t,x)xjdt+ fT liminf[h(t,x)jdt.
Jo
x~oo
x-oo
This contradicts (2.7) and so (2.10) is true. Existence of a solution to (1.1)

is now guaranteed from theorem 4.1. 0
Remarks. (i). If /3,1>1, <P2, r, r E U[O, Tj, 1 < p < 2 then the result of
theorem 4.2 is again true; in this case use Ilv~!liLp instead of Ilv~!IIL2 in the
proof.
(ii). We now state a more general version of theorem 4.2. Suppose (2.1)-(2.5) hold and also that there exists () E (-00,1) with the following
conditions satisfied:
(2.6)*
there exists p E L1[0,Tj with JU,y) ~ p(t)y{)

for a.e. t E [0, Tj and y ~ 0
and
(2.7)*
0< fT liminf(f(t,x) x- 8 j dt.
10
x-.,.CX)
Then (1.1) has a solution.

The proof follows the reasoning in theorem 4.2. The only change occurs
from equation (2.12) to (2.14). In this case multiply (2.11) by y;;O and
integrate from 0 to T to obtain
JofT[J(t ,Yn ) Yn-OJ dt' = and so
(1 - An)
An
r(t) 1-8 dt < 0

JfT
Yn
o
o ~ lim
inf fT (f(t, :tin) y,-;,8] dt.
n--(X, io
We may apply Fatou's lemma (l)f>cause of (2.6)*) to obtain
o ~ JfT lim
inf[j(t, Yn) y~gj dt.
o n-(X)
Essentially the same reasoning as in theorem 4.2 establishes the result.
29
It is of interest to establish another type of result when (2.6) may not

be true. Our next theorem gives such a result.
Theorem 4.3. Assunu: f : [0, T] X R -> R is a L1 -Caratheodory function

and suppose (2.1) - -(2 ..5) are satisfied. In addition suppose
there ex.ists a constant M > 0 .9uch that
[get. y(t))y(t) + h(t, y(t)) + r(t)] dt 2 0
{
for all y E C[O, T] n AC[O, T] with
yeO) = yeT) and min[O,T] yet) 2 M
It
(2.37)
is satisfied. Then (1.1) has a solution.

PROOF: Let y be a solution to (2.8).x. Assme (2.10) does not hold. Then
there is a sequence (An) in (0,1) and a sequence (Yn) such that (2.11) and
(2.12) hold. As in theorem 4.2 we have
(2.38)
(l~An)
rT[g(t,Yn)Yn+h(t,Yn)+r(t)]dt=-
Jo
rTr(t)Yn dt .
Jo
Also we know (theorem 4.2) that there exists a subsequence .') of integers
with
Un ~ 'II in C[O, T] as n 00 in .') and 'II == C > 0 ;
here
'lin
(2.39)
= ~.
IYnlO
'lI n
Thus there exists n6 E S with
(t) 2 ~ i.e Yn(t) 2 ~IYnlo for each t E [0, T]
and n 2 n6, n E 5. Let .(h denote those n E 5 with n 2 n6. Since

IYnio -. 00 as n ....... 00 there exists a subsequence 8 4 of 53 with
(2.40)
Yn(t) 2 M for each t E [0, T] and n E 54 ;
here M is as in (2.37). Now (2.38) and (2.40) imply that
faT [g( t, Yn )Yn

notice also min[o,T] Yn(t)
+ h( t, Yn) + r( t)] dt < 0
for n E 8 4
2 M. This contradicts (2.37).
The above results havp "dual versions".

version of theorem 4.2.
:JO
We will just givp the dual
Theorem 4.4. Assume f : [0, T] X R --> R is a LI-Caratheodory function

and suppose (2.1 )- - (2.4) hold. In addition assume the following conditions
are satisfied:
(2.41)
there exist I), T E L2[0, T] with - T( t) Y :s; g( t, y)y

for
a.e. t E [0, T] and all y 2: 0; here T 2:
{
a.e. on [0, T] and T >
on a subset of [0, T]
of positive measur'c
(2.42 )
:s;
there exists p E LI [0, T] with h( t, y)

for a.c. t E [0, T] and y 2:
:s;
jJ( t)
p(t)
and
(2.43)
r [-r( t)] dt > 10r

10
T
lim sup[g( t, x)x] dt

x-+(X)
r
10
lim sup[h( t, x)] dt.

x-+oo

PROOF: Consider the family of problems
(2.44h
y' + Ty = A[J*(t,y)
y(O) = y(T)
where
< A < 1 and
(2.44h. We show
a.e. on [O,T]
is as in theorem 4.2. Let y be a solution to
y(t) 2:
(2.45 )
+ TY]
for t E [0, T].
Suppose y has a negative global minimum at to E [0, T). Then there exists
tl > to with y(t) < on [to, tI] and y(t) 2: y(t o ) for t E [to, iI]. Now the
differential equation yields
Consequently
f, '1 T(X) dx (t )
a
Y 1
eJo'o T(x)dxy(to)
tl
1to
and so
eJo"T(x)dx [A1*(8,0)
- [.'1
y(tI) < e . '0
T(X)
:31
dx
y(tO)
+ AT(c~)y(S) + Ay(s)]ds <

:s;
y(tO),

Next we claim that there exists a constant Mo with IYlo ~ Mo for any
solution Y to (2.44h. If not there exists a sequence (An) in (0,1) and a
sequence (Yn) with
Y:t
+ T Yn = An[J( t, Yn) + T YnJ
a.e. on [0, TJ
and
IYnio -
00.
Of course
i T[
O
9 (t ,Yn )Yn
+ 1~ (t, Yn ) + r ()J
t dt =
(1 - An)
An
iT
T(t) Yn dt >_ 0
and essentially the same reasoning as in theorem 4.2 (except we use lim sup
instead of lim inf) establishes the result. 0
32
References.
[1]. A.Fonda and J.Mawhin, Quadratic forms, weighted eigenfunctions and

boundary value problems for nonlinear second order ordinary differential
equations, Proc. Royal Soc. Edinburgh, 112A(1989), 145-153.
[2]. M.Frigon and D.O'Regan, Existence results for first order impulsive
differential equations, J. Math. Anal. Appl., 193(1995), 96-11:3.
[3]. A.Granas, R.B.Guenther and J.W.Lee, SOlllP general existence principles in the Caratheodory theory of nonlinear differential systems, J. Math.
Pures et Appl., 70(1991), 15:3-196.
[4]. .J .Mawhin, Topological degree methods in nonlinear boundary value
problems, AMS Regional Conf. Series in Math., 40, Providence, 1979.
[5]. J .Mawhin and .J .R.Ward, Periodic solutions of some forced Lienard
differential equations at resonance, Arch. Math., 41( 1983), 337-;35l.
[6]. M.N.Nkashama, A generalized upper and lower solutions method and
J. Math. Anal. Appl., 140( 1989), :381-395.
[7]. M.N.Nkashama and .J.Santanilla, Existence of multiple solutions for
some nonlinear boundary value problems, J. Diff. Eqns., 84{ 1990), 148164.
[8]. D.O'Regan, Landesman Lazer type results for first order periodic problems, Comment. Math. Univ. Carolinae, to appear.
[9]. K.Yosida, Functional analysis, Springer Verlag, Berlin, 1980.
Chapter 5. Existence principles for second order boundary value

problems.
5.1. Introduction.
This chapter presents general principles that can be used to establish

existence to particular boundary value problems. The existence principles
that follow are formulated for 2nd order systems. Corresponding principles
for systems of other orders will be clear from the discussion that follows.
Moreover, except for some minor notational changes, the proofs of the more
general results are the same as for the 2nd order case. The results presented
here are based on ideas presented by Granas, Guenther and Lee [7] in 1991.
In particular we consider the boundary value problem
(1.1 )
y" = J(t, y, y'), y E B
where a ~ t ~ b, J : [a, b] X Rn X Rn ---+ Rn is continuous or a Caratheodory

function, B represents a set of admissible boundary conditions (to be specified shortly), and Bo is the set of corresponding homogeneous boundary
conditions. By a solution of (1.1) we mean a C 2 -solution (classical solution) if J is continuous and a Caratheodory solution if J is a Caratheodory
function. We now reformulate (1.1) as a fixed point problem and then apply
a nonlinear alternative to obtain existence theorems. Two cases emerge naturally; either the natural linear differential operator associated with (1.1) is
invertible and (1.1) reduces directly to a fixed point problem, or this linear
operator is not invertible. In the noninvertible case, a simple translation
often leads to a satisfactory fixed point problem. We will be able to treat
both classical and Caratheodory solutions simultaneously and in a classical
setting if we integrate (1.1), just as Caratheodory did for the initial value
problem, to obtain the equivalent problem:
y solves (1.1) {:}
YE
Define operators L: c~
Ly(t)
= y'(t) -
C~
---.. Co
y'(a)
and y'(t) - y'(a) =
---.. Co
and F: c~
and
Fy(t)
it
J(s,y(s),y'(sds.
by
it J(s,y(s),y'(.<;d<~.
Then the equivalence above can be expressed briefly by
(1.2)
y solves (1.1) {:}
Ly=Fy,
L,F:C~---+Co.

5.2. Existence principles for invertible problems.

Assume L -1 exists. From (1.2).
y solves (1.1)
>
= L- 1 Fy,
L- 1 F: C~
-+
C~.
To apply the nonlinear alternative for compact maps to L -1 F we need to

know that L -1 F is a compact map on the closure of a suitable open set
in CA. The continuity and compactness we need are consequences of the
continuity of L- 1 and the continuity and complete continuity of F.
Theorem 5.1. Let J : [a, b] X R" X R" ----> R" be Caratheodory and
F : C 1 [a, b] ----> Cola. b] be its associated Caratheodory operator. Then F is
continuous and completely continuous.
= F* 0
PROOF: Observe that F
j: (.d([a.b]'R")
j where
C([a, b], R 2 ,,),
---->
J z = z where z = (y,y'),
and
F*: C([a,b),R2n)
-+
Co([a,b],R"), F* z(t) =
it
J(s,z(s))ds.
Evidently j is continuous and maps bounded sets into bounded sets while
F* is continuous and completely continuous by theorem :3.6. Now the
desired conclusion is clear. 0
Next, we develop convenient criteria that guarantee that L -1 exists
and is continuous. For i = 1,2 let Ui : C 1 ----> R" be a continuous linear
operator and '"'Ii E R". Then a (,I-function satisfies the boundary condition B and we write y E B if and only if Ui(y) = '"'Ii for i = 1,2. Likewise
y E Bo if and only if Ui(y) = 0 for i = 1, 2.
The invertibility of the operator
L .. (.,1
-'6
-+
(.'
-'0,
L yet)
= y'(t) -
y'(a),
is conveniently described in terms of the invertibility of two related linear

operators:
A : CAD ----> C, Ay = y",
and
U : R 2"
---->
R 2n ,
where C = (C1,C2) E R n
tions of z" = o.
U ( c)
X
= (U1 (Zl Cl + z2 (2),
U2 (Z1
C1
+ z2 C2) ),
R" and Zl(t), Z2(t) is any scalar basis of solu-
= (/1, ,2) E R 2n.

implies y = o.
Theorem 5.2. Let ,

1. Ay = 0
Then the following are equivalent.
2. A is invertible.
3. U is invertible.
4- L is invertible for each , E R 2n .

Furthennore, each inverse
i,~
continuous whenever it
exi,~t.~.
PROOF: To confirm these elementary facts, first observe that

y"=gEC
and
y'(t) - y'(a)
y=ZICI+Z2C2+G(t),
:}
= 9 E Co
where
G(t)
:}
1t 1
s
= Zl Cl + Z2 C2 + G'(t),
g(u)dud8.
The first equivalence express the fact that the general solution of the inhomogeneous equation is the sum of the general solution of the corresponding homogeneous equation and a particular solution of the inhomogeneous
equation. The second equivalence follows from the fact that G'(t) is a
particular solution to the first order differential equation on the left when
9 E Co and that y'(t) - y'(a) = 0 implies y" = O. Now, in view of the
preceding observations,
there exists A-I
:}
:}
:}
= (Cl' C2)
such that Ui(ZI C1 + Z2 C2 + G) = 0
there exists a unique C = (C1, C2)
such that U(c) = -(U1 (G, U2 (G)
there exists a uniquf'
there exists
[1-1
because the linear operator U- 1 must be injective and hence surjective to

satisfy the middle condition. So (2) and (3) are equivalent. Virtually the
same reasoning establishes the equivalence of (:3) and (4), and of (1) and
(2). Finally, U-I is continuous whenever it exists because U is a linear
map of finite dimensional spaces, and the continuity of the other inverse
operators follows from that of U- l and the equivalences displayed above.
Theorem .5.2 is now established. D
Theorem's .5.1, .5.2 and the nonlinear alternative for compact maps lead
to the following existence principles.
:36
Theorem 5.3. Let f : [a, b] x R" x R" --+ R" be a continuous or a LPCaratheodory function and A be invertible. Assume there is a bounded set
U ~
C1
such that for any A E [0,1) and any solution y to
(2.1h
y"
= A f ( t, y, y'),
y EB
we have y E U. Then
(2.1 )
y"
= f(t, y, y'),
y EB
has a classical 01' Caratheodory solution y E U according as f is continuous or Caratheodory.
Corollary 5.4. Assume there is a bounded open set
n~
R"xR" such that
for' any A E [0,1) and any solution y to (2.1h we have (y(t),y'(t)) E

for every t E [a,b]. Then (2.1) has a solution with (y(t),y'(t)) E IT for
t E [a,b].
Corollary 5.5. Assume there is a constant M < 00 such that Iyh < M
for any A E (0, 1) and any solution y to (2.1 h. Then (2.1) has a solution.
Remark. Recall lulo = SUP[a,b]lu(t)1 and luh = max{lulo, lu'lo} for appropriate functions u.
PROOF (of the corollaries): Corollary 5.4 follows immediately from theorem 5.~3 applied to
l!={YEC1: (y(t),y'(t))En
for tE[a,b]}.
Corollary 5.5 follows from corollary 5.4 applied with
n = {(u,v) E R"
R": lui, Ivl < max(M, 1 + IL- 1 (0)ld}.
Indeed, it is apparent from the basic assumption in corollary .1.5 that

(y(t),y'(t)) E n for all t E [a,b] for any solution y to (2.1h for 0 < A < 1.
So corollary 5.5 will follow from corollary 5.4 if this conclusion holds also
for A = O. If A = 0, (2.1h is y" = 0, Y E B. Integration yields
y'(t) - y'(a) = 0, y E B; that is, Ly = 0 so y = L- 1 (0) and the condition in corollary 5.4 also is satisfied when A = O. Corollary 5.5 has now
been established. 0
PROOF (of theorem 5.3): Recall that
y soIyes (2 . 1)
= L - IF y,
37
L- 1 F'. C'l
'8
C'l
--+'8'
Let U ~ C~ be as in the theorem and observe that p = L -1(0) E U

because p is the solution to (2.1)'\ for A = 0, as we observed above. By
theorem's 5.1 and .5.2, L -1 F is continuous and completely continuous so
the nonlinear alternative for compact maps applies to
It will follow that T = L- 1 F has a fixed point in U, equivalently (2.1)

has a solution of the required type, if alternative (A2) in the nonlinear
alternative (theorem 2.5) does not occur. Assume to the contrary that
(A2) holds. Then there exists A E (0,1) and y E aU such that
y = AT y
Thus
Ly
so L y
= AF y
+ (1
- A)P.
= ALL- 1Fy+(1-A)Lp,
i.e.
y"
= A f (t, y, y'),
YEB
Therefore, y E U by the principal hypothesis of the theorem, which contradicts y E aU. 0

5.3. Existence principles for noninvertible problems.
If the natural linear differential operator associated with a nonlinear
boundary value problem is not invertible, as for example with periodic
boundary conditions, a simple translation enables us to recover invertibility
and to formulate some existence principles very much like those given above.
We start with the observation that for any fER,
y" =
f (t, y, y'),
YEB
:}
y" -
Y=
f (t, y, y') -
{ y, Y E B.
The linear operator determined by y" - f Y and Bo will be invertible for

most E. For such t, we proceed as before to get y solves (1.1) if and only
if
y'(t) - y'(a) -
fit Y($)d$ = I t(J($,y($),y'($)) - EY($)]ds.
Hence
y solves (1.1)
:}
L,y=F,y.
L"F,:c1 ........ C o
and since we are assuming L-;1 exists we get

y solves (1.1)
:}
38
= L-;1
F, y;
here, of course
L, y(t)
and
F,y(t)
= y'(t) -
y'(a) -
fIt
[J(S,Y(S),y'(S)) -
y(s)ds
fy(s)]d<~.
Since f,(t,y,y') = f(t,y,y') - fy is continuous or CaratModory according

as J has the same property, theorem .5.1 has the immediate consequence:
Theorem 5.6.
continuous.
F,: C 1 [a, b]
Cola, b] is continuous and completely
----+
We also have the following counterpart of theorem .5.2.

Theorem 5.7. f is not an eigenvalue oj A if and only iJ L;l exists and
is continuous for' all , E R 2n .
PROOF: Let U1(t), U2(t) be a scalar basis for the solutions of the differential equation y" - f Y = O. Define II : R 2n ----+ R2n by
where c
= (C1,C2)
(:l.l)
E R"
Rn. Observe that any solution y E C 1 of
y' ( t) - y' ( a) -
is in C 2 and satisfies y" -
=0
It
y( 8 ) ds
=0
and hence, has the form
(3.2)
Conversely, any y as in (3.2) solves (3.1).
Now, let 9 E Co and H E C 2 be a particular solution of y" It is easy to check that h = H' E C 1 solves
(:3.3)
y'(t) - y'(a) -
Indeed
h'(t) - h'(a) -
fIt
h(s)ds
fIt
Y = g.
y(S)d8 = g(t).
H"(t) - H"(a) - E(H(t) - H(a))

fH(t)
fH(t)
+ g(t) -
fH(a) - g(a)
+ fH(a) = g(t).
Therefore, (3.3) has the general solution
Such a y satisfies the boundary conditions B precisely when
that is when
(3.4)
Since
U ( c)
= (lh (u 1 C1 + U2 C2 ), U2 ( U1 C1 + U2 C2) )
is linear, (3.4) has a unique solution for c = (C},C2) if and only if U is

injective (hence, surjective). Consequently, Lf is invertible if and only if
U is invertible, which in turn is equivalent to
U(c) = 0
o.
c=
Now, U(c) = 0 if and only if u = Ut Cl + U2C2 solves u" - fU = 0 and

Ui( u) = 0 for i = 1,2. Since Ut, U2 is a basis of solutions, we conclude
that
(U(c)=O ~ c=O) :} (AU=fU~ u=O),
equivalently,
is not an eigenvalue of A.
Theorem's 5.6 and 5.7 and the reasoning used in the proof of theorem
5.3 lead immediately to
Theorem 5.B. Let f : [a, b] X R n X R n - Rn be a continuous or a LPCaratheodory Then the analogues of theorem 5.3 and its corollaries (corollary 5.4 and 5.5) hold with (2.1 h replaced by
(3.5h
y" -
= >.. [jet, y, y') -
y], Y E B.
Of course theorem 5.8 is theorem 5.3 when f = 0 is not an eigenvalue of

A. It turns out that even iftheorem 5.3 is applicable to a particular problem
it may still be advantageous to apply theorem 5.8 with some f :j:. O.
40
5.4. Application.
In this section we present a Bernstein-Nagumo theory for scalar boundary value problems on compact intervals. The existence principles established in section's .5.2 and .5.a will playa major role. The problems we
consider in this section arise naturally in diffusion and osmotic flow theory;
we refer the reader to [6] and its references. The result here is based on
ideas presented by Granas and Guennoun [.5] in 198R. Of course the classical case (Le. when f: [0,1] X R2 ---- R is continuous) has been known for
quite some time [4,6,9].
In particular we examine the general Sturm Liouville problem
y" = f(t,y,y') a . e. on [0,1]
{ yeO) - oy'(O) = r, a ~ 0
y(l) + (3y'(1) = s, (3 ~ 0
(4.1 )
and the periodic problem

y" = f(t, y, y') a.e. on [0,1]
{ y(O)=y(l)
y' (0) = y' ( 1 ).
( 4.2)
Theorem 5.9. Let f : [0,1] X R2 -. R be Caratheod07'y and assume the

following conditions al'e satisfied:
( 4.3)
(4.4 )
and
(4.5)
there is a constant M > 0 such that

yf(t, y, 0) ~ 0 for almo,~t all t E [0,1]
is
IYI > M
implies that
a BOT'el function q: [0,00) ---> (0,00) such that

qfu E L~oc[O,oo) and If(t,y,p)l::; q(lpl) for almost all
t E [0,1] and all y E [-Mo, Mo]
there
1 - () >
00
xdx
q x
2Mo.
In (4.4) and (4.5), for the periodie problem

Mo
=M
and c
=0
whereas f01' the Sturm Liouville problem

Mo
= lllax{M, Irl, lsI}

41
and
C
Is - rl if a = f3 = 0
min{ Mo:lr l , Motlsl} if a ::fi 0 or (3::fi O.
Then each of(4.1) and (4.2) has a solution y E W 2 ,P[0, 1].

Remarks. (i). In our notation c = Motlsl in the case a = 0 and f3 ::fi O. A
similar remark applies if [3 = 0 and a::fi O.
(ii). If q( x) = AX2 + B where A ~ 0, B ~ 0 are constants then (4.,)) is
clearly satisfied.
PROOF: From (4.5) choose and fix MI > 0 and 0 < f < 1 so that
Ml
(4.6)
c.
()
M 0 dx > 2Mo
qX+f
We introduce an "associated" family of problems
(4.7)>.
y" - fy = A[h(t,y,y') - fY] a.e. on [0,1]

y satisfies Sturm Liouville or periodic data
where A E (0,1] and
h(t,y,p)
h : [0,1]
={
R2
->
R is defined by
max{ f(Y - Mo), f(t,y,p)} if y> Mo

f(t,y,p) if - Mo::; y::; Mo
min{ f(Y + M o ), f(t, y,p)} if y < -Mo.
Observe that f is not an eigenvalue of the natural linear operator associated

with y" together with either homogeneous Sturm-Liouville or periodic
boundary conditions.
We claim that each solution y to (4.7. satisfies IYlo ::; Mo and ly'lo::;
MI' It follows from the claim that any solution to (4. 7h is a solution to
(4.1) or (4.2).
To establish the claim suppose y assumes a positive maximum at to E
[0,1]. Consider first the case to E (0,1), in which case y'(to) = O. Suppose
y(to) > Mo. Then by continuity there is an r > 0 such that y(t) > Mo for
t E V = [to,to+r]. By definition of h we have y" ~ fA(y-M o )+(l-A)fY
a.e. on V. Hence y" > 0 a.e. on V. Now since y'(to) = we have
y'(t) =
t y"(s)ds > 0
J~
and y(t) - y(to) =
t y'(s)ds > 0
J~
for t E V.
This contradicts the maximality of y(to). Thus y(to) ::; Mo. Now if y satisfies the periodic boundary data and has a positive maximum at to E [0,1]
42
then in view of the periodicity to E [0, 1) and y'( to) = 0. If to =

then
exactly the same reasoning as above yields y(O) ~ Mo. Next suppose y satisfies the Sturm Liouville boundary condition and has a positive maximum
at O. If a =
then clearly y(o) :so Mo whereas if a i
then
o~
y'(O)
1
= -[y(O)
a
r]
which implies y(O) ~ r ~ Mo. Similarly if y has a positive maximum at 1

then y(l) ~ Mo In conclusion if y satisfies either the periodic or Sturm
Liouville boundary conditions and if y assumes a positive maximum at
to E [0,1] then y(t o) ~ Mo. Similarly, if y assumes a negative minimum
at tl E [0,1] then y(tl) ~ -Mo. Consequently Iylo ~ Mo for any solution
y to (4.7'.
Our next arguement shows Iy'lo ~ MI' First it is easy to see using
the boundary data that there exists aTE [0,1] with ly'(T)1 ~ c; here
c is the constant given in theorem S.9. Now suppose ly'(t)1 > c for some
t E [0,1]. Since ly'(T)1 ~ c and y' is continuous on [0,1] there is an interval
a ~ s ~ t (or t ~ s ~ a) such that ly'(.~)1 > c and ly'(a)1 = c. To be
definite assume y' > c on a < s ~ t and y'(a) = c. From (4.4) and (4.n~
we have
y"(s) ~ ly"(s)1 ~ q(y'(8)) + {M o a.e. on [a,t].
Multiply by y' (s) and integrate from a to t to obtain
yl (t)
q(u)
+ (Mo du
1t
a
y'(8)Y"(8)
q(y'(s + (Mo d8 ~
1t ,
a
y (8) d8 ~ 2Mo.
This together with (4.6) implies ly'(t)1 = y'(t) ~ MI' The other possibilities for the sign of y' and the location of a are treated similarly and the
same bound is obtained. Consequently Iy'lo :so Ml for any solution y to
(4.7},x.
It remains to show (4. 7h has a solution to complete the proof. Existence
of a solution to (4. 7h would follow directly from theorem S.3 or theorem
S.8 except for the fact that the function II may not be CaratModory since
property Cl in the definition may fail. However we will show that the
corresponding operator N h : C 1 --> Co defined by
(Nhu)(X)
= foX II(S,U(8),U'(8d8
is continuous and completely continuous. This is all that is needed to apply

the reasoning used to prove theorem S.3 and theorem .5.8. Consequently,
43
existence of a solution to (4.7h will follow once we prove that NfJ is

continuous and completely continuous. The proof is essentially the same
as in theorem 3.6 except for the verification that if Un -+ U in C t then
h(t, un(t), u~(t)) --+ h(t, u(t), u'(t)) a.e. on [0,1] which is needed to apply
the Lebesgue dominated convergence theorem. For any t with u( t) =I Mo
the pointwise convergence at such a t is dear from the definition of hand
the fact that f is Caratheodory. Let B = {t E [0,1] : 'u(t) = Mo}. Since
u(B) has measure zero, a theorem of Banach's implies that u'(t) = 0 a.e.
on B. For t not in this exceptional set of measure zero, we have u'{t) = o.
Also there exists either a subsequence Nt of integers with u n ( t) :::; Mo
for n E Nt or there is a subsequence N2 of integers with un(t) > Mo for
n E N 2 If Nt exists then
as n
-+ 00
through NI, whereas if N2 exists then (4.3) implies
h(t,un(t),U~(t))
max{ f(Un(t) - Mo),f(t,un(t),u:.(t))}
-+
max{ 0, f(t, M o, O)}

f(t, Mo, 0) a.e. as n
--+ 00
through N 2
So we have the desired pointwise convergence for those t in B with 'u'(t)

The case when u(t) = -Mo is treated similarly. 0
o.
Remarks. (i). In fact it is possible to obtain a more general form of theorem

5.9 by using the notion of upper and lower solution; we refer the reader to
[2] (see also theorem 14.1 in this book).
if
(ii). The theorem of Banach used in theorem 5.9 is the following:
x : [0,1] -+ R is an absolutely continuous function and U ~ [0,1] is a
measurable set such that x( U) is null, then x' (t) = 0 for a.e. t E U.
44
References.
[1]. D.R. Dunninger and J .C. Kurtz, Existence of solutions for some nonlinear singular boundary value problems, J. Math. Anal. Appl., 115(1986),
396-405.
[2]. M. Frigon, Application de la theorie de la transversalite topologies
problemes nonlineaires pour des equations differentielles ordinaires,
Diss. Math., 296(1990), 1-79.
a des
[3]. M. Frigon and D. O'Regan, On a generalisation of a theorem of S.

Bernstein, Ann. Polon. Math., 48( 1988). 297-306.
[4]. R.E. Gaines and J. Mawhin, Coincidence degree and nonlinear differential equations, Springer Lecture notes, 567, New York, 1977.
[5]. A. Granas and Z. Guennoun, Quelques n~sultats dans la theorie de
Bernstein Caratheodory de l'equation y" = J(t, y, y'), C. R. Acad. Sci.
Paris, 306( 1988), 703-706.
[6]. A. Granas, R.B. Guenther and .J.W. Lee, Nonlillf~ar boundary value
problems for ordinary differential equations, Diss. Math., 244. Warsaw.
1985.
[7]. A. Granas, R.B. Guenther and J.W. Lee, Some general existence principles in the Caratheodory theory of nonlinear differential systems, J. Math.
Pures et Appl., 70(1991), 153-196.
[8]. J .W. Lee and D. O'Regan, Existence principles for differential equations
and systems of equations, Topological methods in Diff. Eqns. and Inclusions (edited by A. Granas and M. Frigon), NATO AS} Series C, Kluwer
Acad. Publ., Dordrecht, 1995, 2:39--289.
[9]. J. Mawhin, Topolical degree methods in nonlinear boundary value
problems, AMS Regional Conf. Sf!ries in Math .. 40, Providence, 1979.
[10]. K. Schmidt and R. Thompson, Boundary value problems for infinite
systems of sewnd order differential equations, J. Diff. Eqns., 18(197.5),
277-295.
45
Chapter 6.
tions.
Boundary value problems without growth restric-
6.1. Introduction.
This chapter presents existence results for boundary value problems
without growth restriction on the right member. In particular we examine
the second order problem
(1.1)
(py')'(t) = f(t,y(t),P(t)Y'(t.
{ aoY(O) - bo limt .... o p(t)y'(t)
at y(1) + bt limt .... t p(t)y'(t)
a.e. t E [0,1]
= TO
= Tl
where p E C[O,I] n C l (O,I) with pet) > 0 for t E (0,1), and f :

[0, 1J X R2 -+ R is a Caratheodory function (to be described below). Bya
solution to (1.1), we mean a function y E C[O, 1] n Ct(O, 1) such that py' is
absolutely continuous on [0,1]' and which satisfies the differential equation
and boundary conditions. Boundary value problems of the form (1.1) have
been extensively treated in the literature of ordinary differential equations.
In most of these books and papers, the nonlinearity f satisfies a growth
condition in y'. The results presented in this chapter are obtained without
any growth assumption on f. The ideas rely on the notion of upper and
lower surfaces for (1.1), that is, surfaces having a particular form and on
which the function f ha.c; a given sign. This notion of upper and lower
surfaces was introduced in 1993 by Frigon [3J. The definition was motivated [3,5] by existence results deduced from the location of the zeros of
the nonlinearity [6,7,8,9]. For example in [6], Gran a.c; , Guenther and Lee
drew attention to the fact that
y" = (y')2 + 11"2,

y( 0) = y( 1) =
has no solution, whereas
y" = (y')2 _ 11"2,

yeO) = y(l) =
< t <
< t <
has a solution. In the above examples what seems to be important is not

the presence of the quadratic growth in y' but rather the location of the
zeros of p2 + 11"2 and p2 - 11"2. We will make this statement more preCise in
section 6.2.
46
All our results in this chapter are obtained via the Schauder fixed point
theorem. Throughout this chapter let p E C[O, 1] n C 1 (0, 1) be such that
p(t) > 0 for t E (0,1). We define the Banach space /(1[0,1] by
{y E C[O, 1] n C 1 (0, 1): py' E C[O, I]}

with the norm Ilyilt = max{lIyllo, IIpy'lIo} where
= max{ly(t)i:
IIYllo
and we define
/(2 [0,
E [0, In;
1] by
{y E /(1[0,1] : py' E WI,I[O, I]}.
We denote
/(l [0,1]
by
{y E /(1[0,1] : aoY(O) - bo limp(t)y'(t) = ro,aly(1)

~o
+ btlimp(t)y'(t) =
~t
[(e[O, 1] = /(2[0,1] n /(l[0, 1]; and finally we let C r [O,IJ

yeO) = r}.
= {y
rI},
E e[o, 1] :
(Maximum Principle) Let pE C[c,d]nCt(c,d) be such

that pet) > 0 for all t E (c,d); and let y E J(2[c,dJ be such that
Theorem 6.1.
(py')'(t) 2': 0
a.t~.
t E (c,d),
aoy(c)-bo limt_cp(t)y'(t) ~ 0 and at y(d)+b t limt ..... dP(t)y'(t) ~ 0; where

max{ao, ad > 0, ai,bi 2': 0, and max{ai,b;} > 0, i = 0,1. Then yet) ~ 0
for all t E [c, d).
PROOF: Define
G(t)
={
y(t)2, if ~(t) > 0

0, otherwIse.
Then (py')'(t)G(t) 2': 0 a.e. t E (c,d). Integrating from c to d and using

the integration by parts formula gives:
G(d) lim p(t)y'(t) - G(c) lim p(t)y'(t) _Jd p(t)y'(t)G'(t) dt 2':

t-d
t ..... c
c
We deduce that y( t)
0 for all t E [c, d].
o.
Finally we state a well known result of Banach. This result will be used
in section 6.2 (recall that we used this result previously in chapter 5).
47
Theorem 6.2. Let x : [0,1] -+ R be an absolutely continuous function,

and U ~ [0,1] be a mea!mrable set such that x(U) is null. Then x'(t) =
for a.e. t E U.
6.2. Boundary value problems.
We consider the boundary value problem (1.1) where
ai,b i 2:: 0, max{ai,b;} > 0, i = 0,1, max{ao,ad > 0;

and p and
satisfy the following assumption:
(H). p E C[O, 1] n Cl(O, 1), pet) > for all t E (0,1), f: [0,1] X R2 --> R
is a Caratheodory function, that is:
(i). t ~ f(t,y,q) is measurable for all (y,q) E R2;
(ii). (y, q) ~ f(t, y, q) is continuous for a.e. t E [0,1];
(iii). for any l' > 0, there exists hr E Ll[O,l] such that If(t,y, q)1 :s
hr(t) a.e. t E [0,1] and for all Iyl :s 1', Iql :s 1'. Moreover we have,
I~ p-l(s)I;hr(x)dxds< 00 ifal f= 0, and I~ p-l(8) Islhr(x)dxds< 00
if al = 0; and p-1 E Ll[O, 1] if ao = 1'0 = or al = 1'1 = do not occur.
We start with a particular case in order to illustrate arguments which

will be used in a more general result (theorem 6.6). Consider the problem
(py')'(t) = f(t, y(t),p(t)y'(t))

{ limt ___ op(t)y'(t) = 0
y(l) + blimt--+l p(t)y'(t) =
a.e. t E [0,1]
(2.1)
where b 2 O.
Theorem 6.3. Let p, f be functions satisfying (H). Assume there exist
Mo 2 0 2 Ml such that
f(t,Mo,O) 2
2 f(t,Mt,O)
and the1'e exist two functions So : [0,1]

[Mt.MoJ---- (-00,0] such that:
(2.2)
(2.3)
(2.4 )
a.e. t E [0,1];
X [Ml'
Mo] -. [0,00),
81 :
X f--+ Si(t,X) is continuous for a.e. t E [0,1], i = 0,1;

so(tt, Xl) :s 80(t2, X2) for all tl :s t2,
Sl(tl,XI):S 8l(t2,X2) for all t1 2:: t2,
Xl
Xl
:s X2;
:s x2;
f(t,x,80(t,x)):S o:s f(t,x,s1(t,x)) f01' a.c.

t E [0,1] and all x E [M}, Mo].
4H
and
[0,1] X
Then the problem (2.1) has a solution such that
Ml ~ y(t) ~ Mo. ,ql(t,y(t)) ~ p(t)y'(t) ~ so(t,y(t)) f01' all t E [0,1].

To prove this theorem. we modify the problem (2.1), and we show that
this modified problem has a solution which is also a solution to (2.1). Define
the function fI : [0,1] X Rl -; R by
h(t,y,q) =
f(t.Mo,O), if y> Mo
f(t,Ml,O), if y < MI
f(t,y,so(t,y)), if MI ~ y ~ Mo , q > so(t,y)
j(t,y,SI{t,y)), if MI ~ y ~ Mo, q < SI(t,y)
f(t, y, q), if MI ~ y ~ Mo, SI (t, y) ~ q ~ soft, y).
It is clear that this function may not be Caratheodory. However we have

the following result:
Theorem 6.4. Under the assumptions of theorem 6.3 the opemtor N1 :
[(1 [0,1] -+
Co[O, 1] defined by
N1(y)(t)
= lot h(r,y(r),p(r)y'(r))dr.
is well defined, continuous and compact.

PROOF: Let R > max{IMol, IMII} such that ISi(t,y)1 ~ R for all t E
[0,1] and y E [M}, Mo]; and let hR E Ll[0,1] be the function given in
(H)(iii). Then
(2.5)
Ift(t,y(t),p{t)y'(t))1 ~ hR(t)
a.e. t E [0,1]
and for all y E [(1[0,1]. So, N1 is well defined. Now, assume that Yn -; Y
in [(1[0,1]. By the inequality (2.5) and the Lebesgue dominated convergence theorem, it suffices to show that
(2.6)
h(t,Yn(t),p(t)Y:t(t)) ---. h(t,y(t),p(t)y'(t)) a.e. t E [0,1]
to get that NI Yn ----> NI Y in Co[0,1].

It is clear that (2.6) holds a.e. on
{t E [0,1]: yet)
i' Mo,
and y(t)
On the other hand, by theorem 6.2, y'(t)
{t E [0,1]: yet)
= Mo,
49
i'
M1 }.
a.e. on
or yet)
= M 1 }.
So, S1(t,y(t) ::; p(t)y'(t) = 0::; 80(t,y(t)) a.e. on that set. The assumptions (2.2) and (H)(ii) imply that (2.6) holds a.e. on that set.
The compactness of Nl follows from (2.5) and the Arzela-Ascoli theorem. 0
We consider the modified problem
(2.7)
(py.')'(t) = h(t,y(t),p(t)y'(t))
{ 1imho p(t)y'(t) = 0
y( 1) + blimt_,) p(t)y'(t) = O.
a.e. t E [0,1)
Theorem 6.5. Under' the assumptions of theorem 6.3 the boundary valu(
problem (2.7) has a solution.
PROOF: Set C!} = {g E Co [0,1): Ig(t)1 ::; I~ hR(s)ds} where hR is
the function satisfying inequality (2.5). Define the continuous operator
-1 :
---+
by
L Cr;
J(l
L-l(g)(t) = -bg(1) -
Cr;
J1
g(s)p-l(s)ds.
J(l.
From (2.5), N1 (y) E

for all y E
Remark that y is a solution to
(2.7) if and only if y is a fixed point of L -1 0 N1 :
Kl- By theorem
6.4, the operator L -10 Nl is compact. The Schauder fixed point theorem
gives the existence of a fixed point to this operator, and hence a solution
to (2.7). 0
J(l . . .
Proof of theorem 6.:1: Let y be a solution to (2.7) given by the previous

theorem. We want to show that y is a solution to (2.1).
We claim that Ml ::; y(t) ::; Mo. Indeed, letu(t) = y(t) - Mo. By
assumption and the definition of II, we have
(pu')'(t)
2: 0 a.e. on {t: u(t) > O}.
The boundary conditions and theorem 6.1 imply that y(t) ::; Mo for all
t E [0, 1]. Similarly we get the other inequali ty.
On the other hand, assume there exists t1 E [0,1] such that
p(tJ)y'(tJ) > SO(t1,y(tJ).
=0
::; so(O,y(O)), there exists t2 < t1 such that

so(t 2,y(t 2 )) 2: p(t 2 )y'(t2), and so(t,y(t::; p(t)y'(t) for all t E (t2,td. By
assumption (2.3) and the definition of /1, we get
Since limhOP(t)y'(t)
50
a contradiction. The other inequality is obtained in the same way.

Therefore, the solution y is such that
M1 :'S y(t):'S M o, and S1(t,y(t)):'S p(t)y'(t):'S 80(t,y(t))

for all t E [0, 1J; and consequently, y is a solution to (2.1).
Now, we give a more general result. First of all, we need to introduce

some definitions.
Definition 6.1. A function a E ](2[0,lJ (respectively (3 E ](2[0,1]) is

called a lower solution (resp. upper solution) to (1.1) if
(pa')'(t) 2: f(t,a(t),p(t)a'(t))
aoa(O) - bo limt ..... op(t)a'(t) :'S
( respectively
1'0
a.f' t E [0, 1J;
and a1 0'(1)
+ b1 limt ..... 1P(t)a'(t) <
1'1
(p;3')'(t) :'S f(t,(3(t),p(t)(3'(t)) a.e t E [0, 1J;

ao/J(O) - bo limt ..... op(t)/3'(t) 2:
1'0,
a1 /3(1)
+ b1 limt ..... 1 p(t)O'(t) 2: rd
Let a:'S;3 in C[O, 1J and D = {(t, y) E [0, 1J
R: aCt) :'S y :'S O(t)}.
Definition 6.2. A surface SeD x R is an upper surface to D for (1.1)

if there exist two functions s : D -... [0,00), c E ](1 [0,1 J such that
S
= {(t,x,s(t,x)):
(t,x) ED},
a:'S c:'S /3,
and satisfying the following conditions:
0) (x - c(t))f(t,x,s(t,x)) 2:
a.e. t E [0,1]' (t,x) E D;
(ii). there exist N C [O,lJ and E C R two negligeable sets such that the
function x f-+ s(t,x) is continuous for all (t,x) E D\(N X E);
(iii). p(t)c'(t):'S set, c(t)) for all t E [0, 1J;
(iv). S(t1,X1) 2: S(t2,X2) (resp. S(t1,XI):'S S(t2,X2)) for all t1 :'S t2,
X1 :'S X2 such that (t1' Xl) and (t2' X2) are in the same connected part of
D+ = {(t, x): x > c(t)} (resp. D- = {(t, x): x < c(t)});
(v). aoc(O):'S
1'0
+ bos(O,a(O)),
a1 c(l) 2:
1'1 -
b1 s(1,/3(1)).
Definition 6.3. A surface SeD X R is a lower' surface to D for (1.1)

if there exist two functions s: D -+ (-oo,OJ, c E ](l[O,lJ such that
S
= {(t,x,s(t,x))
: (t,x) ED},
.51
a:'S r :'S /3,
and satisfying (i),(ii) of definition 6.2, and thf' following conditions:

(iii). p(t)c'(t) 2: 8(t, c(t)) for all t E [0,1];
8(t}, xI) :S 8(t2, X2) (resp. 8(t1, xIl 2: 8(t:.21 X2)) for all tl 2: t2,
:S X2 such that (iI, Xl) and (t2, X2) arf' in the same connected part of
D+ = {(t, x): x > c(t)} (resp. D- = {(t, x): x < c{t)});
(v). aoc(O) 2: ro + b0 8(0,/3(O)), al c(I):S rl - bI 8(1,a(1)).
(iv).
Xl
Remark 6.1. If f is a Caratheodory function, and S is an upper (resp.

lower) surface to D for (1.1), without loss of generality, we can assume
that
(1). 5 is bounded;
(2). for y = aft), (3(t) and crt), 8(t,y) = limsup(t,x)ED,x-+y s(t,x) (respectively s(t,y) = liminf(t,x)ED,x-+y8(t,x)).
Theorem 6.6. Let p, f be functions satisfying (H). Assume there exist

a :S (3 r'espectively lower and upper solutions to (1.1). Set
D
= ((t,y)
[0,1]
R: aft) :S y:S ;3(t)},
and suppose there exist 8 1 = 5 1 (81,CIJ and So = 50(80,CO) respectively

lower and upper surfaces to D for (1.1). Then the problem (1.1) has a
solution such that
aft) :S y(t) :S /3(t), sdt,y(t)) :S p(t)y'(t):S 80(t, y(t))
for all t E [0,1].
To (t, y, q) E D x R we associate
if =
if defined
by
so(t,y), if q> so(t.y)

{ q, if 81 (~, y) :S q:S 80(t, y)
81(t, y), If q < 81 (t, y).
Theorem 6.7. Under the assumptions of theorem 6.6 there exi8t four
measurable functions Ci,(>, ci,(3: [0,1]----> [0,00) such that
+ co,(>(t).<
co(t),
if aft) < co(t), so(t,a(t)) < p(t)a'(t),
(3(t) > (3(t) - co,(3(t) > co(t),

co,(3( t) = 0,
other'wise;
if (3(t) > co(t), "'0(t,/3(t)) < p(t)(3'(t),
aft) < aft)

= 0,
CO,(> (t)
other'wzse;
52
moreover, q> So(t,y) JOl'all (y,q) E B O,a(t)UBo,{3(t). SirnilarlyJor i

where
Bi,a(t)
= ((y,q):
(t,y) ED, lI(y,q) - (a(t),p(t)a'(t))11 < ci,a(t)},
iJ ci,a(t) =f. 0, and Bi,a(t)

Bi,{3(t)
= ((y,q):
= 1,
= 0 otherwise;
and
(t,y) E D, lI(y,q) - (f:J(t),p(t)/3'(t1I < Cj,{3(t)},
iJ ci,{3(t) =f. 0, and Bi,{3(t) = 0 other'wise.

PROOF: Let
Ai(t) = CO{(y,Si(t,y: (t,y) ED}
and
di,a(t)
= dist( (a(t),p(t)a'(t,
Ai(t.
Choose
Ci,a(t) :::; min{c;{t) - a(t),di,a(t)}/2.
Remark that if p(t)a'(t)
=f. p(t)a'(t) and Ci(t) - a(t) > 0 then

di,a(t) > 0
by definitions 6.2, 6.3 and remark 6.1(2). We define ci,[J similarly.

Define the function 12 : [0,1]
12(t,y,q)
R2 ---. R by
max{j(t,/3(t),p(t){3'(t,J(t,j3(t),p(t)/3'(t},if y > )3(t)

min{J(t,a(t),p(t)a'(t,J(t,a(t),p(t)a'(t)},if y < a(t)
(1 - Ai,{3(t,y,qmax{J(t,y,q),J(t,Y'(j)}
+Ai,{3(t,y,q)J(t,y,ij), if (y,q) E Bi,{3(t), i = 0,1
(1- Ai,c.(t,y,qmin{j(t,y,q),J(t,y,ij)}
+Ai,a(t,y,q)J(t,y,ij), if (y,q) E Bi,a(t), i = 0,1
J(t, y, ij),
otherwise
where
Ai,{3(t,y,q) = (ci,[J(t-lll(y,q) - (f3(t),p(t)/3'(t11 if (y,q) E Bi,P(t)

and
Ai,a(t,y,q) = (ci,a(t-lll(y,q) - (a(t),p(t)a'(t11 if (y,q) E B;,a(t),

i
= 0,1.
Remark 6.2. On {(t,y,q) E [0,1]

So (t, y) }, 12 (t, y, q) = f (t, y, q ).
Rl: a(t):S; y
:s; /3(t), st{t,y) :s; q :s;
For sake of briefness, we assume that a] of:. 0; the proof of theorem 6.6
is similar for a] = O. Without loss of ~enerality, fix a] = l.
Theorem 6.8. Under the assumptions of theorem 6.6, the operator N2 :
KI [0,1]
-->
ColO, 1] defined by
N 2 (y)(t)
= lot h(T,y(T),p(T)y'(T))dT
is well defined, continuous and compact.

PROOF: Let R> max{llallo,lif3110,llpa'llo,llp/3'llo} such that ISi(t,y)l:S;
R for all (t,y) ED; and let hR E Ll[O, 1] be the function given in (H)(iii).
We have
(2.8) Ih(t,y(t),p(t)y'(t))I:S; hR(t)
a.e. t E [0,1] and all y E KI[O, 1].
Hence, N'2 is well defined and the compactness of N'2 follows from (2.8)
and the Arzela-Ascoli theorem.
To show that N'2 is continuous, according to the inequality (2.8) and
the Lebesgue dominated convergence theorem, it suffices to show that
(2.9)
h(t, y,,(t),p(t)y~(t ...... h(t, y(t),p(t)y'(t))
a.e. t E [0,1]
when y" --> y in KI[O,l].

It is clear that the relation (2.9) holds a.e. on
{t: aCt) of:. yet) and Ott) of:. y(t)}.

On the other hand, on {t: aCt)
a.e. and
h(t, y( t), p( t)y'( t))

If y,,(t)
= yet),
Bi,a(t) of:. 0}, p(t)a'(t)
= p(t)y'(t)
= min {I( t, aCt), p(t)a'(t)), f( t, a(t), pet )a'( t)) }.
aCt), then for n sufficiently large, we have
Hence (2.9) holds a.e. on that set. On {t: aCt) = yet), Bo.a(t) = BI.a(t) =
= p(t)a'(t) = p(t)y'(t) a.e. and (2.9) holds a.e. Similarly, the
relation (2.6) holds a.e. on {t: /3(t) = y(t)}. This completes the proof of
the theorem. 0
0}, p(t)a'(t)
54
Consider the modified problem
(2.10)
(py')'(t) = h.(t,y(t),'P(t),Y,'(t)) a.e. t E [0, 1]

{ ao y(O) - bo limt~op(t)y (t) = ro,
y(l) + b1 limt~lp(t)y'(t) = Tl
Theorem 6.9. Under the assumptwus of theorem 6.6 the problem (2.10)
has a solution.
PROOF: To prove this result, argue as in theorem 6.5 with
L-l( )(t)
g
={
where
A=
-l b
) (1 (TO b;;-l_g(s)) df
- 0
Tl + ro b1 b0 - Ig(l + Jt
pis)
s, 1 ao(TO + bo A)a;)1 + f~(A + g(s))p-l(s) ds, if ao 1- 0
Tl - TO a;) 1
bl g(1) - fol g(s)p-l(s) ds

1
.
b1 + boa;)l + fo p-l(s)ds
-
Proof of theorem 6.6: Let y be a solution to (2.10) given by theorem

6.9. According to remark 6.2, we must show that a(t) ::; y(t) ::; (3(t),
silt, y(t)) ::; p(t)y'(t) ::; so(t, y(t)) for all t E [0,1].
By assumption and the definition of h, we have
(py')'(t) ~ (pr3')'(t)
a.e. on {t: y(t) > (3(t)}
(py')'(t) ::; (pa')'(t)
a.e. on
and
{t: y(t) > a(t)}.
By the boundary conditions and the maximum principle (theorem 6.1), we

deduce that a(t) ::; y(t) ::; ;3(t).
On the other hand, assume that p(t)y'(t) 1:: so(t,y(t)). By definition
6.2 and the boundary conditions, one of the following cases holds:
(a). there exist tl
< t2
E [0,1] such that
y(t) ~ colt), p(t)y'(t) ~ <~o(t, y(t)) a.e.
t E (iI, t2)'
p(tl)y'(td > so(t},y(tl)), p(tZ)y'(t2)::; so(tz,y(tz));

(b). there exist tl
< t z E [0,1] such that
y(t)::; colt), p(t)y'(t) ~ so(t,y(t)) a.e.

p(tl)y'(t 1 )
::;
SO(tl, y(td), p(tz)y'(tz) > so(tz, y(tz)).

55
t E (t},tz),
Without loss of generality, assume (a) holds. Definition 6.2(i) and the
definition of 12 imply that (py')'(t) ~ 0 a.e. t E (tt,t 2 ). Therefore, using
the monotonicity condition on So (definition 6.2 (iv)), we get
a contradiction. Similarly, we can show that Stet, yet)) ~ p(t)y'(t).
56
References.
[1]. A. Cabada, The monotone method for first order problems with linear
and nonlinear boundary conditions, preprint.
[2]. J. Dugundji and A. Granas, Fixed point theory, Vol 1, PWN,
Warszawa, 1982.
[3]. M. Frigon, Theoremes d'existence pour des problemes aux limites sans
condition de croissance, Bull. Sc. Math., 117(1993), ;H7~400.
[4]. M. Frigon and D. O'Regan, Some general existence principles for ordinary differential equations, Topological Meth. Nonlinear Anal., 2(1993),
35~54.
[5]. M. Frigon and D. O'Regan, Existence results for some initial and
boundary value problems without growth restrictions, Proc. Amer. Math.
Soc., 123(1995), 207~216.
[6]. A.Granas, R.B. Guenther and .J .W. Lee, Nonlinear boundary value
problems for ordinary differential equations, Dissertationes Math.,
244(1985), 1~128.
[7]. M.N. Nkashama, A generalized upper and lower solutions method and
Jour. Math. Anal. Appl., 140(1989), 381~395.
[8]. A.G. O'Farrell and D. O'Regan, Existence results for some initial and
boundary-value problems, Proc. Amer. Math. Soc., 110(1990), 661~673.
[9]. L.C. Piccinini, G. Stampacchia and G. Vidossich, Ordinary differential
equations in R n , Applied Mathematical Sciences 39, Springer- Verlag, New
York, 1984.
[10]. A. Rodriguez and A. Tineo, Existence theorems for the Dirichlet
problem without growth restrictions, Jour. Math. Anal. Appl., 135(1988),
1~7.
.57
Chapter 7. Posit one boundary value problems.

7.1. Introduction.
Consider a thin rod with internal heat generation or lateral surface
heating occupying the interval 0 ~ x ~ 1; a mathematical model for the
steady-state absolute temperature y in the rod is
y"(X)
+ J.Lq(x)g(x,y(x)) = 0
for suitable q and g. Here J.Lq( x) may be interpreted as the reciprocal

of the local heat conductivity and 9 as the law governing heat generation
as a function of position and temperature. We assume that the boundary
at x = 0 is insulated, so y'(O) = O. As prototype boundary condition at
x = 1 we consider Stefan's and Newton's laws: -ky'(l) = y(l)C> - TC>,
where k is a positive constant, T is the ambient temperature, and a = 4
or a = 1, respectively. We prefer to rewrite the boundary condition in the
form
h(y'(l)) + y(1) = 0
where
h( z)
=-
(TC> - k z);:;
for
z ~
TC>
y'
We suppose for convenience that this h is artificially extended to z > ~a

as a positive increasing function. Thus it is natural to assume that
(1.1)
h: (-00,00) ----> (-00,00) is a continuous, nondecreasing

{ function, and there exists a unique (3 ~ 0
such that h((3) = O.
In the theorems in this chapter it will be shown that in fact a solution y

satisfies y' (1) ~ (3. In section 7.2 we wish to determine a class of functions
h satisfying (1.1)and for which a solution to our problem exists for some
range of J.L > O. We are interested mostly in the case when 9 is superlinear.
We generalize the above problem to include symmetric solutions of symmetric heat conduction problems in higher dimensions by examining the
problem
(1.2)
(1.3 )
_(1) (p(x)y'(x))'
px
lim p(x)y'(x)
x-+o+
+ J.Lq(x)g(x,y(x)) = 0,
= 0,
h(y'(l))
+ y(l) = OJ
.58
a different boundary condition at x = 0 (and at x = 1) could be considered, but that of (1.3) stems naturally from elliptic partial differential
equations. For such equations in Rn we have that p(x) = x n - l for solutions depending only on the radial variable x; we therefore do not assume
that (lip) E L 1 (0,1), since to do so would eliminate from consideration
elliptic problems in two and more spatial dimensions.
We present an existence arguement [2] based on showing that no solutions of an appropriate family of problems lie on the boundary of a suitable
open set. This suffices for the nonlinear alternative (theorem 2.5), which
requires only the existence of a suitable open set with no fixed points (=
solutions) on its boundary. In the usual application (for example the applications in chapter's 3,4 and 5), all solutions are shown to lie in this open
set, whereas in this chapter we show only the weaker result that no solution
lies on the boundary. As a result this technique is able to handle existence
for problems wherein the set of norms of all solutions is unbounded.
The ideas in this chapter may be used to examine (1.2) with other types
of boundary data. For example we could discuss the generalized Gelfand
problem [3], namely
(1.4 )
y"
+ J.lqg(t,y) = 0,0 < t < 1
yeO)
= a ~ 0,
y(l)
=b ~ 0
where J.l ~ 0 and 9 is superlinear. One can show (under suitable assumptions), using the technique in section 7.2, that (1.4) has a nonnegative
solution for 0 ::; J.l < J.lo where J.lo E (0,00]. As a result, for convenience,
we will restrict ourselves in this chapter to the boundary value problem
(1.2), (1.3).

In view of the connection with symmetric problems for elliptic partial
differential equations, we allow the coefficient p to vanish at zero. We also
allow q to be singular at x = 0 and x = 1. In our first existence result
the following conditions are assumed:
(2.1)
p is continuous and strictly positive on (0,1]
(2.2)
qEC(O,l) satisfies q>O on (0,1),

{ with I~ p(s)q(s)ds < 00, and
I~ p(t) I~ p(s)q(s)dsdt < 00
59
and
(2.3)
g: [0,1] X [0,00) ----'> (-00,00) is continuous,

g(t,O) ~
for t E (0,1), and there exists a continuous
nondecreasing function f: [0.00) ----'> [0,00) such that
f(y) >
for y>
and g(t,y)::::; f(y)
holds on (0,1) X (0,00).
Theorem 7.1. Assume (1.1), (2.1), (2.2) and (2.3) hold. Set
(2.4)
1
r o =max{p(\)1 p(x)q(x)dx, 11
p(~)1x P(t)q(t)dtdx}
and let /10 satisfy
(2 ..5)
sup
zE(O,oo)
J-lorof(z) - h(-J-lorof(z
Then (1.2) - -(1.3) has a nonnegative solution if
) > 1.
: :;
J-l ::::; J-lo
Remark. The supremum in (2.5) is allowed to be infinite.

PROOF: We consider the one-parameter family of problems
(2.6h
~(pyl)'+>'J-lqg*(x,y)=O
limx-->o+ p( x )y'( x) = 0, >. h(y'( 1 + y(1)
for>. E (0,1]' where we define g* on [0, 1J

9 *( x, Y )
(-00,00) by
= {g(X'.O).) yy,~ y0.<

g(x. y,
We first show that any solution of (2.6 h is nonnegative and hence a solution
of (2.6 h with g* replaced by g. If for some >. > the solution y had a
negative minimum at Xo E (0.1). then from (2.6h we would have
g*(xo,Y(xo = g(xo,O) - y(xo) > 0,
a contradiction. If a negative minimum occurs at Xo = 0, then the differential equation shows that (py')' < as far to the right of zero as y remains
negative, whence y' <
there. a contradiction. If a negative minimum
occurs at Xo = 1. then y'( 1) ::::; 0, so from the boundary condition at 1 and
( 1.1) we have that
0> y(l) = ->.h(y'(1 ~ 0.
60
again a contradiction. Consequently y 2:

h(y'(l)) ~ 0, and so y'(l) ~ {3.
For notational convenience set
Yl
= max{sup y,
[0,1]
on [0, 1J. From this we get that
sup ply'l}
[0,1]
Now for x E [O,lJ we have that
y(x)
=
-
11
fat p(s)q{s)g*(s,y(sdsdt
>. h ( - ~~) fal p(t)q(t)g*(t, yet)) dt)
>'/-l
p(t)
and therefore
y( x)
fat p(,~ )q( s) ds dt

>'h (->'/-If(yd P/1) fal P(s)q(s)ds)
< >'/-l f(Yl)

-
fal
p(t)
< /-l TO f(Yl) - h(-/-l TO f(yd),

or
y(x)
::; l.
/-l TO f(yt) - h( -/-l TO f(yt})
Also,
p(x)y'(x) = ->'/-l
faX
p(t)q(t)g*(t,y(tdt,
so
max Ipy'l ::; /-l TO f(yt}
[0,1]
From these two inequalities we get that
(2.7)
Now fix /-l
(2.8)
/-lo and let Mo satisfy

Mo
/-lTof(Mo) - h(-/-lTOf(Mo
> 1.
Let B be the Banach space {y E C[O, 1JnC 1 (0, 1J : py' E C[O, I]} equipped
with the norm
Iyh = max {max Iyl, maxPIY'I} .
[0,1]
61
[0,1]
Set
C~[0,1]={uEC[0,1]nC1(0,1]:pU'EC[0,1], lim p(x)u'(x)=O} C B
x--+o+
and define the mapping N on CJ [0, 1J by
(Ny)(x)
= -h(y'(l + J.l
1
x
pes) 10 p(t)q(t)g*(t,y(tdtds.
By the second inequality of (2.2), Ny E C[O, 1] and
(Ny)'(x)
= p(:)
fox p(t)q(t)g*(t,y(tdt E C(O,IJ.
Thus
N : C~[O, 1J - cJ[O, 1].

Continuity of g* and (2.2) show that N is continuous as a map in CMO,l].
Moreover, the Ascoli-Arzela theorem shows that N maps bounded sets
into compact sets. In addition, the fixed points of A N are precisely the
solutions of (2.6}.x.
Let
= {u E Cd[O,IJ: luh < Mol,
an open subset of the convex subset CJ [0,1] of B. Should there exist

A E (0,1) and y E aU with y = ANy, then y is a solution of (2.6}.x
satisfying Iyll = Mo and hence from (2.7) we have that
-< 1,
Mo
J.lrof(Mo) - h(-J.lrof(Mo
contradicting the choice of Mo in (2.8). Since 0 E U, the nonlinear alternative (theorem 2.5) guarantees that N has a fixed point. 0
Remark. Note that we do not claim that all solutions satisfy Iyh ::; Mo,
but only that the weaker condition Iyh =J Mo is satisfied by any solution
y of(2.6}.x for A E (0,1).
In theorem 7.1 we produced a value Mo such that Iyh =J Mo for any
solution y of (2.6h for A E (0,1). In the proof of the following result we
are able to find only a forbidden value for IYlo; but this suffices, as will be
seen. In theorem 7.2 we set F( z) = f~ f( u) ds.
62
Theorem 7.2. Let (1.1), (2.1), (2.2) and the following hold:
(2.9)
(2.10)
g E CO, 1) X [0,00, .and O:S g(x,y):S fey)

{ on (0,1) X (0,00), where f is continuous
and nondecreasing on (0,00)
there exists a continuous nondecreasing function
{ Q(t) satisfying p(t)2q(t):S Q(t) on (0,1)
and
fl
Jo
y'Q(x) d
p(x)
X
Let
P == p(1)
< 00.
fl
Jo
p(t)q(t)dt,
and suppose J.Lo satisfies

(2.11)
sup
O<s<oo
-h(-/.LoP f(s
dz
> J2J.LO
JF(s) - F(z)
JQ(X)
p(x) dx.
Then (1.2) - -(1.3) has at least one nonnegative solution for J.L E (O,J.Lo],
PROOF: Note that the left hand side of (2.11) increases as J.Lo decreases,
so if this inequality holds for some J.Lo, it also holds for any smaller value.
Also P exists by (2.2). We consider the family of problems
(2.12)'\
~(pY')'+>'J.Lqg(x,y)=O
= 0, >.h(y'(l)
limx-to+ p(x)y'(x)
+ y(l) = 0
for>. E (0,1]. For some>. and J.L let y be any nonnegative solution of
this problem; then (py')' :S 0, so y is nonincreasing. In particular, we
have that y'(l) :S 0 :S {3. Multiplying the differential equation in (2.12),x
by p2 y' and integrating over (0, x) ~ (0,1) leads us to
p( X?y'( X)2 :S - 2>.J.L f; p( t)2q( t )f( y( t) )y'( t) dt

< 2J.L Q(x) Jy(x)
fY(O) f( u) du
-
(2.13)
since y' :S 0 and >. :S 1. Separation of variables and a further integration

over [0,1] yields
(2.14)
Y(O)
dz
<
Y(I) JF(y(O - F(z)
63
J2ii,
fol JQ(X) dx.

p(x)
Jo
Now y must satisfy the integral equation

y(x)
= AJ.lil~)
pet
ft p(8)q(8)g(8,y(8d8dt
Jo
>.. h (- AJ.l
p( 1) Jo
p(t)q(t)g(t, y(t dt)
from which it follows that

y(l) = -Ah (-
P~~) 10
P(t)q(t)g(t,y(t)dt) ::; -h(-J.lP f(y(O)
since y(1) 2: O. Thus from (2.14) we see that any solution y must satisfy
(2.15 )
dz
y(O)
-h(-/.LP f(y(O)))
<
JF(y(O - F(z) -
rcl::l
2J.l
JQ(x)
dx.
pix)
Let J.lo satisfy (2.11); then J.l ::; J.lo also satisfies (2.11) and there must exist
a value Mo E (0,00) such that
dz
MO
-h(-/.LP f(Mo))
v'F(Mo) - F(z)
>
J2i1
JQ(x)
dx.
p(x)
In view of (2.15), Mo is a forbidden value for yeO) = IYlo. From (2.13) we

see that any solution y of (2.12 h that satisfies
y( x) ::; Mo on [0, 1]
also satisfies
: ;
p(x)2IY'(x)1 2 ::; 2J.lQ(l)
(O)
y(l)
f(u)du::; 2J.lQ(1)
fMo
Jo
f(u)du == Ml.
Let B be the Banach space introduced in the proof of theorem 7.1 but
with the equivalent norm
II u II 1 --
max
( max lu(x)l.
-,
xE[O,I]
Mo
rnax P(X)IU'(X)I) .
xE[O,I]
Ml
It follows that no nonnegative solution of (2.12h for A E (0,1) lies on the

boundary of
U = {u E B: Ilulh < I}.
There is no loss of generality in supposing g(x, y) extended to negative
values of y as a nonnegative, continuous function. Let y be any solution
of (2.12h with g so extended. Then (py')'::; 0, so y' ::; and y achieves
its minimum at x = 1. Now y(l) < would force h(y'(Il) > 0, implying
64
that y'(1) > f3 2: 0 and so contradicting the minimalit

solutions of the extended problem are nonnegative and so solutions of the
original problem.
In view of (2.2), the operator N defined by
(N y)(x)
= -h(y'(l)) + Jl
1 1
1
- 1
p(s)
p(t)q(t)g(t, y(t)) dt ds
maps CJ [0,1] continuously into itself. As before, N is also a compact

map, and the fixed points of A N are solutions of (2.12),\, and conversely.
The nonlinear alternative (theorem 2.5) again shows that N has a fixed
point in U, which is the desired nonnegative solution of (1.2) - -(1.3). 0
Remark. Note that in this proof, unlike that of Theorem 7.1, we do not
produce a bound on ply'l for all solutions, but only for solutions satisfying
0':; y(x).:; Mo.
65
References.
[1]. 1. E. Bobisud, J. E. Calvert, and W. D. Royalty, Existence of biological
populations stabilized by diffusion. Diff. Eqs. Dynamical Systems. 3( 1995),
325-344.
[2]. 1. E. Bobisud and D.O'Regan, Existence of positive solutions for singular ordinary differential equations with nonlinear boundary conditions,
Proc. Amer. Math. Soc., 124( 1996), 2081-2087.
[3]. A.M Fink, J .A. Gatica and G.E. Hernandez, Eigenvalues of generalized
Gelfand problems, Nonlinear Anal., 20(1993), 1453-1468.
[4]. S. Wang, On S-shaped bifrucation curves. Nonlinear Anal., 22(1994),
1475-1485.
[5]. Z. Zhao, Positive solutions of nonlinear second order differential equations, Proc. Amer. Math. Soc., 121( 1994), 465-469.
66
Chapter 8. Semi-positone boundary value problems.

8.1. Introduction.
This chapter establishes the existence of nonnegative solutions for the
Dirichlet boundary value problem
(1.1)
y" + fL fey) == 0, 0 :S t :S 1
yeO) == 0, y(l) == b ~ 0 or y(l) == 0, yeO)
=b~ 0
and the mixed problem
( 1.2)
y" + fL f( y) == 0, 0 :S t :S 1
y'(O) = 0, y(1) == b ~ 0 or y'(1) == 0, yeO) == b ~ O.
Here fL > 0 and f(O) < O. Problems of the above type are referred to
in the literature as non-positone (or semi-positone) problems. They occur
in models for steady-state diffusion with reaction (see [2]). Over the last
ten years or so problems (1.1) and (1.2) haw been studied widely by many
authors; see [1,4,5,8,10] and their references. The technique used usually
involves either the shooting method or phase plane methods. One of the
difficulties encountered is that the norm of solutions of the "appropriate"
family of problems considered is usually unbounded. Another difficulty
that arises is that zero is not a lower solution (in fact it is sometimes an
upper solu tion). Section 8.2 shows how to apply the fixed point approach
of chapter 2 (see also chapter 7) successfully in the non-positone case for
the boundary value problem (1.1) and (1.2).
For completeness we state the following existence principle from chapter
5 (see theorem 5.3).
Theorem 8.1. Let fL, a, b E R with a < b. SUpp08f:

f :R
( 1.3)
---7
i8 continuou8.
In addition assume there exists a constant Mo, independent of A, with
sup ly(t)1
[a,b]
"!
Mo
for any solution y (her'e y E C 2 [a, b]) to
(1.4h
y"(t)
+ AfL f(y(t))
== 0,
:S t :S b
yea) == ao, y(b) == bo
67
for each A E (0,1). Then (1.4h has a solution y E C 2 [a,b] with
sup iy(t)i ~ Mo
[a,bj

Our goal in this section is to eventually examine the homogeneous
Dirichlet problem
(2.1 )
y" + I-l f(y) = 0, 0 ~ t ~ 1

y( 0) = y( 1) = 0
and the mixed problem

(2.2)
y" + I-l f(y) = 0, 0 ~ t ~ 1

y'(O) = y(l) = 0;
here I-l > 0 is a constant. Throughout this section we assume (1.3) holds
and also that
(2.3)
f(O) < 0, f'(x) > 0 for x > 0 and there exists

{ positive real numbers f3 and 8, f3 < (} with f(f3)
and F(8)
= 0;
here F(s)
= J; f(x)dx
=0
is satisfied. Before we discuss (2.1) and (2.2) we first establish a general

existence result for the boundary value problem
(2.4)
y" + I-l f(y) = 0, 0 ~ t ~ b, b

y(O) =0, y(b)=a28;
>0
here I-l > 0 is a constant.

Remarks. (i). There is nothing special about [0, b]. In fact we could consider
the differential equation in (2.4) on any compact interval. However since

our goal is to examine (2.1) and (2.2) eventually we will take b =
or
b = 1.
(ii). Analogous results to those obtained in this section could be established
if 0 ~ a < 8 in (2.4). However since our goal is to eventually examine (2.1)
and (2.2) the a's we will choose will automatically satisfy a > 8.
(iii). The result obtained for (2.4) is of independent interest.
Theorem 8.2. Assume (1.3) and (2.3) hold. In addition suppose

(2.5)
J( 1
==
sup
cE(n,oo)
(r
ds
Jo jF(c) - F(s)
6R
>
J2b
I-l
Let
G(x)=
10fX
Kl}
. {G2 (9)
2b2 '2b 2
ds
.j=F[S)' XE(0,9) and !-Ll=mm
If 0 < !-L <!-Ll then (2.4) has a nonnegative solution y (here y E C 2 [0, b]).
PROOF: Fix !-L < !-Ll. Notice 2b 2 !-Ll S; Kf implies that there exists Mo >
Q with
fOMO
ds
> J 2b2 !-L.
10
JF(Mo) - F(s)
Hence there exists
(2.6)
> 0 (here
MO
< Mo small) with
ds
~
JF(Mo) - F(s) >
!-L.
Let no E {I, 2, ... } be chosen so that ...L

<
no
!-L <!-Ll we have
and ...L
<
no
G(9) - J2b2!-L >
Q.
In addition since
o.
Then there exists b > 0 (here b < 9 small) with

(2.7)
G(9) - J2b2!-L _ flJ
11J -1i
Also (see (2.3 there exists nl 2: no with
(~)
>
nl
du
V-Feu)
;1
>
o.
< {3 and
F(9 - b).
Remark. Notice F(O) = F(9) = 0 with F(x) S; 0 for x E [0,9], F(x) 2: 0

for x E (9,00), F is decreasing on [0,{3] and F is inc.reasing on [{3,00).
In addition for n 2: nl we have
(2.8)
Let N+ = {nI' nl
modified problem
+ 1, ... }.
{
To discuss (2.4) we begin by examining the
y" + !-Lf(y) = 0, 0 S; t S; b
y(O) = ~, y(b) = Q, n E N+.
69
We first show (2.9t has a solution for each n E N+. To see this we consider
y" + Af.1J*(y) = 0,0 s:; t s:; b

yeO) = ~, y(b) = a, n E N+
(2.1O)~
where 0 < A < 1,
fey), y ~ a
{
J*(y) =
~ f(a), 0 s:; y
-y, y s:; 0
s:; a
and
(2.11)
a = min
Here C ; [0, OJ
{~,
c-}
n
J -F(u)
du .)}.
(C(O) - J2b Z f.1- (o

}O-b
[0,00) is the increasing function
G(z)
(2.12)
du
, z E [0,0].
o J-F(u)
z
We begin by showing that any solution y to
yet)
(2.13)
~ 0
(2.1O)~
satisfies
for t E [0, b].
To see this suppose y( s) < 0 for some s E (0, b). Then y has a negative
minimum at to E (0, b), so y/(tO) = 0 and y"(tO) ~ O. However
y"(tO)
= -A f.1J*(y(to)) = Af.1y(to)
<
0,
a contradiction. Consequently (2.13) is true. We next show that
yet)
(2.14)
a for t E [0, b].
Suppose y has a global minimum at t} E [0, bJ. If t} = 0 or b then

trivially yet) ~ yetI) ~ ~ ~ a. So it remains to consider the case when
t} E (0,1) (so y'(tt) = 0). If y(tt) ~ a we are finished. As a result we
assume y(t}) < a. Since y(b) = a > ~ there exists an interval [t}, t2] with
0<
- y < ln on [t},t2) and y(t2) = l.
n Notice
y" > 0 on (t}, tz)
( 2.15)
since f(x)
(2.16)
< 0 for x
E [0,(3). Now (2.15) together with y'(td
y'
> 0 on
70
(t},t z ).
=0
implies
In particular
y'(t2) > O.
(2.17)
Also since y(b) = Q > {3 there exists an interval [t2, t3] with 0 ~ y < (3
on [t2, t 3) and y(t3) = {3. Once again y" > 0 on (t 2, t3) and this together
with (2.17) implies
(2.18)
In particular
(2.19)
and
(2.20)
Next since y(b) = Q 2:: () there exists an interval [t3,t4] with 0 ~ y < () on
[t3, t 4) and y(t 4 ) = (). We claim that there exists ts E (t3, t 4) with
(2.21)
1
() 2:: y(ts) > () - 6, y' > 0 on (tl, t s ) and - ~ y ~ y(ts) on [t2, ts];
n
here Ii is as described in (2.7). To see this notice since y'(t3) > 0 that
there are two cases to consider, namely y' > 0 on (t3, t4) or there exists
t6 E (t3, t 4 ) with y' > 0 on (t3, t6) and y'(t6) = O.
Case (i). y' > 0 on (t3, t4).
Since y(t4) = () then trivially (2.21) is satisfied.

Case (ii). There exists t6 E (t3, t4) with y' > 0 on (t 3 , t6) and y'(t6)
In this case since y' > 0 on (t2' t6) we have
and so
f*(y(t)) = f(y(t)) on [t2,t6]'

Also note y(t6) > {3. Now
-y'y"
= >I/..l f(y)y'
so integrate from t2 to t6 to obtain
(2.22)
71
on (t2,t6)
= O.
If y(t6)
:s; fJ - a then (2.8) implies
(:J
(~) ~ F
> F(fJ - a)
so
F(y(t6 - F
(~)
F(y(t6
< 0,
which contradicts (2.22). Thus y(t6) > fJ - b so (2.21) is true.

We have shown that there exists an interval (tl, t s ) with y' > 0 on
(tt,ts), y'(tl) = 0 and fJ ~ y(ts) > fJ - o. Next we show
(2.23)
y' y"
:s; A JL[ - fey )]y' on (tt, ts)
Suppose t E (tt,t5) with yet) ~ (1. Then J*(y(t = f(y(t so (2.23) is

trivially true since y' > 0 on (tl' t5). On the other hand suppose t E (tl' t s )
and 0 :s; yet) :s; (1 . Then J*(y(t = ~ f(1). However since f(x) is
increasing for x > 0 we have f(y(t :s; f(1) and this together with
~ :s; 1, f(1) < 0 yields
= y(t)
J*(y(t
(1
f(1)
f(1)
f(y(t
so (2.23) is again true. Thus (2.23) is true in all cases. Integrate the
inequality in (2.23) from tl to t (t E (tt. ts to obtain
[y'(tW
:s; 2AJL[F(y(tl)) - F(y(t] :s; 2AJL[-F(y(t] :s; 2JL[-F(y(t].
Thus for t E (th t5) we have
y'(t)
<
J-F(y(t -
~
'-JL
and integration from tl to t5 yields
y(tJ)
Now since y(t5) >
(J -
(J-6
y(tIJ
du
y (t s )
--,:.=:;::;:::;:=;::
< V 2b2 JL .
J-F(u)-
6 we have
du
--,:.==,;::;:~
<
J-F(u) -
y (t s )
y(td
72
du
< J2b 2 JL
J-F(u)-
and so (using (2.12)),

(2.24)
G(B - 8) - G(y(tl))
Now we have
G(B - 8) =
e- 6
J2b 2 Jl.
du
= G(B) J-F(u)
1
e
du
e-b J-F(u)
and this together with (2.24) yields

-G(y(td)
J2b 2 Jl
e
du
- G(B).
ie-8 J-F(u)
Consequently
a
> y(t1) :2 G- 1 (G(B) - J2b 2 Jl-
re
ie-Ii
J -F(u)
du
)
> a,

We next claim that
(2.25)
sup iy(t)i =J Mo.

[0.1)
Let the absolute maximum of iyi = y occur at So E [0,1]. If So = 0 or

b then clearly (2.2Ei) is satisfied since Mo > a. So it remains to consider
So E (O,b) (note y'C~o) = 0). If y(80) < a then (2.25) is satisfied. As
a result we need only consider the case y(so) > a. Now since yeO) = ~,
there exists an interval [SI, so] with Y > /3 on (Sl' so] and y( 81) = ;3.
Since f( x) > 0 for x > ;3 we have
(2.26)
This together with y'(so)
=0
implies
y'
> 0 on
(2.27)
In particular
(2.28)
We claim
(2.29)
(0,050)'
Suppose not. Then there exist 82 E (O,st) with y' > 0 on

y'(82) = O. Since (2.14) is true we haw
-y'y"
Integrate from
S2
to
So
= AJ.Lf(y)y'
on
(82,80)
and
(82,80).
to obtain
0= AJ.L[F(y(S2 - F(y(so).
Thus
(2.:30 )
Now y(so) > Q ~ () implies F(y(80 > 0 whereas y(S2) < () implies
F(Y(S2 < O. This contradicts (2.:30). Thus (2.29) is true. Hence on
(0, so) since (2.14) is true we have
-y'y"
= AJ.Lf(y)y'
and integration from t (t E (0,80 to So yields
[y'(t)2 ~ 2J.L[F(y(so - F(y(t).

Thus for t E (0,80) we have
<
y'(t)
JF(y(so - F(y(t))
J2ii
and integration from 0 to So yields
~ y(so)
~
If y( so)
= Mo
{Mo
<
JF(y(so - F(u)
y2b 2 J.L.
then
h
and since 1n
<
du
du
< V2b2 f.l
JF(Mo) - F(u)
we have
Mo
du
JF(Mo)-F(u)
~ V2b 2 f.l.
This contradicts (2.6). Consequently y(so) =1= Mo. As a result (2.25) is

true. Theorem 8.1 implies that (2.10)'1' (and consequently (2.9)") has a
solution y" with 0 ~ (T ~ y,,(t) ~ Mo for t E [O,b).
74
In addition the differential equation guarantees the existence of a constant M1 (independent of n ) with
(2.31)
ly~(t)1 ::; M1
for t E [O,b].
Also there exists T7n E (O,b) with IY:~(11n)1 = tla - *1::; %. This together
with (2.31) implies that there is a constant M2 (independent of n) with
(2.32)
ly~(t)1 ::; M1
for t E [O,b].
The Arzela-Ascoli theorem guarantees the existence of a subsequence S

of {n}, n1 + 1, ... } and a function y E C 1[0, b] with y!j) --> yU) uniformly
on [0, b] as n --> 00 through S, for j = 0,1. Clearly y ~ on [0, b] with
y(b)= a and yeO) = 0. Also notice for n E S that Yn satisfies the integral
equation
Yn (t) -- bat
(2.33)
Let n
(2.34)
+ (b-t)
J1.(b-t) rt f(
nb +
b
Jo S
Yn (
S
+ fbI (b - ,~) f (Yn (S ) ) ds.
--> 00
through S to obtain
yet) =
at
T +
J1(b-t)
b
Jo
Thus y is a solution of (2.4).
sf(y(,~ ds
J1t (t
+T
Jb
(b - s)f(y(s ds.
Next we consider the boundary value problem (here bE (0,1,
(2.35)
here J1 >
y" + J1f(y) = 0, b::; t ::; 1

y( 1) = 0, y(b) = a ~ (J ;
is a constant.
: ;
Remark. In this case also we could consider the differential equation in

(2.35) on any compact interval. In addition the case
a < (J could also
be discussed.
A slight modification of the arguement in theorem 8.2 immediately
yields the following result.
Theorem 8.3. Assume (1.3) and (2.:3) hold. In addition suppose
(2.36)
[(1
==
sup
cE(c<.oo)
(rJo
ds
) >
JF(c) - F(s)
75
V2(I-b)2J1.
Let
G(X)
d~
= Jo ~,;r;
E (O,B)
and f-l2
= min
{(,2(B)
2(; _ b)2 ' 2(1
/(2}
~ b)2 .
If 0 < f-l < f-l2 then (2.:J5) has a nonnegativf solution.

Remark. Notice f-ll = f-l2 if b =
1.
Next we discuss the homogeneous Dirichlet problem (2.1).

Theorem 8.4. Assume (LJ) and (2.:3) hold. In addition suppose
> 0 (.~mall) and a constant

2: (), R( h 0) =
t~ere exist.~ a co.nstant
(2.37)
{ Ii 0 2: B wzth h 0
and R(uj!i for

here
R(u) -
'UE(/(O-T,/(O);
d.s
.jF(u)-F(s)
Let
Jl3
J'i
'or u > B.
J'
= 2 R 2 (B).
If 0 < f-l < f-l3 then (2.1) has a nonnegative solution.
Remark. Notice since F(B) = 0 we have R(B)

defined in the statement of theorem R.2).
G(B) (hen' G is as
PROOF: Let m E {1,2, ... } be chosen so that ~ <

+ 1, ... }. To discuss (2.1) we begin by examining
and let N*
{m, m
(2.38)m
y" + f-lf(y) = 0, 0::; t::;

yeO) = 0, y(
= Ko - ~
1)
and
(2.39)'"
y"
y(
+ f-l fey) = 0, &::; t ::;
&) = Ko -
~, y( 1)
=0
seperately.
Fix m E N*. Now (2.38yn has a solution y", by theorem 8.2. To see
this take b = 12 , a = Ko - lm in theorem R.2. Now (2.5) is satisfied since
(2.37) implies
K1
[R(e)] >
sup
cE(Ko-,k,oo)
76
{g.
V2
Also J.L < J.Ll since J.L < 2R2(O) = 2G 2 (O) and J.L < 2Kf by above.
Similarly (2.39)m has a solution by theorem S.3. In summary (2.38}'" has a
solution Ym for each mE N* with 0 ~ Ym(t) ~ Ko for t E [0,
In
addition the differential equation and the boundary conditions guarantee
the existence of constants K2 and K 3 , independent of m, with
!"
The Arzela-Ascoli theorem guarantees the existence of a subsequence S*

of N* and a function w E C 1 [0,!] with Y~) ----> w(j) uniformly on [O,!]
as n ----> 00 through S*, j = 0,1. Clearly
~ w ~ Ko on [0,
with
w(O) = 0 and w(!) = Ko. Essentially the same reasoning as in theorem 8.2
(see (2.33) and (2.34)implies w"+J.Lf(w) = 0 on [O,i] and WEC 2 [0,i]
Similarly we can show that there exists v E C 2 [!, 1] with v" + J.L f( v) = 0
on [i, 1], v(l) = 0, v(i) = Ko and 0 ~ 1) ~ Ko on [i,l].
Let
y(t) = { w(t), t E [~,!]
v(t), t E b, 1].
iJ
Clearly y E C[O, 1]. We claim

lim w'(t)
(2.40)
1-
=0=
t-"2
lim v'(t).
1+
t-"2
If this is true then y E C 1 [0, 1]. In addition since
and
y"
(~ +) = v" (~ +) = - J.L f (1) (~ +) ) = - J.L f( /(0)
we have y E C 2 [0,1] and of course y" + J.Lf(y) = 0 on [0,1]. It remains

to prove (2.40). Now since
~ w ~ /(0 on [0,
and w(i) = /(0 we
have w'(
2: O. We now show w'(
= 0. Suppose not i.e suppose
w'(
> 0. Then either w' > 0 on (0,
or there exists So E (0,
with
w' > on (.so,!] and w'(so) = O. If there exists an So as described above
then since
t -)
t -)
t -)
i)
77
i]
t)
we have on integration from So to
that
_ (W' (})) 2 = 2fL[F(Ko) - F(w(so],
a contradiction since F(Ko) - F(w(so > o. Consequently w' > 0 on

(0,
Now integrate -w'w" = fLf(w)w' from t to
to obtain
!).
[W'(t)]2
= (w' (~)) 2 + 2fL[F(Ko) -
Thus for t E (0,
F(w(t] > 2fL[F(Ko) - F(w(t].
t),
w'(t)
-r.~~~~rrr
JF(Ko) - F(w(t
and integrate from 0 to

R(Ko)
!<1::
> v 2fL
to obtain
= foKo
>
/F(K:}U_ F(u)
~.
t-)
viet
This contradicts (2.37) and so W'(

= o. To see
+) = 0 suppose
Vl(! +) < o. Then we can show (similar to the above arguement) that
v' < 0 on (!,1). Integrate -v'v" = fLf(v)v' from
to t to obtain
[v'(t)f
= (VI (})) 2 + 2fL[F(Ko) -
Thus for t E
F(v(t]
>
2fL[F(Kol - F(v(t))].
(t, 1),
-v'(t)
JF(Ko } - F(v(t
-~~~::;=:;='<7 >
and integration from
t to
q::
v.c.fL
1 yields
R(Ko) >
which contradicts (2.37). Thus

and we are finished. 0
v'(! +)
~,
=
o.
Consequently (2.40) is true
(2.41 )
there exists a ~onstant (Jo 2: (J with R' ( x) < 0 for

x > (Jo and hm x _ oo R( xl 0;
78
here R is as described in the statement of theorem
If 0
8.4.
Let
< fL < fL4 then (2.1) has a nonnegative solution.
J'i
PROOF: Fix fL < fL4 Then

< R(Oo). Now since R(x) is continuous
for x > 0, R'(x) < 0 for x > 00 and limx~oo R(x) = 0 there exists
/(0 > 00 with R(/(o) =
In addition since R'(x) < 0 for x > 00 there
j'i.
jIi
for U E (/(o - T, /(0).

exists T > 0 with /(0 - T ~ 00 and R( u) >
Thus condition (2.37) is satisfied so the result follows from theorem 8.4. 0
(2.42)
!,,(x) < 0 for x
> 0 and lim f(x)

x-+oo
= 00.
Let
fLs = 2 R2(O);
then (2.1) has a nonnegative solution.
84
If
0<
fL
< fLs
PROOF: It is easy to check that R'(x) < 0 for x> 0 and lim x-+ oo R(x)
O. The result follows from theorem 8.5 with 00 = O. 0
Finally we consider the mixed boundary value problem (2.2).
Theorem 8.7.
(2.43)
As<~ume
(1.3) and (2.3) hold. In addition
there exists a constant T > 0 (small) and a constant

{ /(0 ~ 0 with /(0 - T ~ 0, R(/(o) = ...fIJi
and R( u) >...fIJi for U E (/(o - T, /(0);

fL6
If 0
,~uppose
8.4.
Let
R2( 0)
= -2-'
< fL < fL6 then (2.2) has a nonnegatitle solution.
PROOF: Let m E {I, 2, ... } be chosen so that ~ <

{m, m + 1, ... }. We begin by examining
(2.44)'"
y" + fLf(y) = 0,0::; t::; 1

y( 0) = /(0 - ~, y( 1) = O.
79
and let N*
Fix mE N*. Now (2.44t' has a solution Ym by theorem 8.:3. The ArzelaAscoli theorem (see theorem 8.4) implies that there exists v E C[O, 1] with
v" + f-lf(v) = 0 on [0,1], v(l) = 0, v(O) = Ko and 0 ~ v ~ Ko on [0,1].
We now claim that v'(O) = o. Suppose v'(O) < o. Then v' < 0 on (0,1)
If this is not true there exists So E (0,1) with v' < 0 on (0, so) and
v'(so) = o. Consequently
[V'(0)]2
= 2f-l[F(v(so)
- F(Ko)]
< 0,
a contradiction. Thus v'<O on (0,1). Integrate -v'v"=f-lf(v)v' from

o to t to obtain
[V'(t)]2
= [v'(OW + 2f-l[F(Ko) -
F(v(t)] > 2f-l[F(Ko) - F(v(t].
Thus for t E (0,1) we have
-v'(t)
-r=~=;=,=~~ >
JF(Ko) - F(v(t
and integrate from 0 to 1 to obtain R(Ko) >
17'(0) = 0 so v is a solution of (2.2). 0
J2iif-l
v'2Ji, a contradiction.
Thus
Remark. A similar result could be established for the boundary value problem
Y" + f-l f( y) = 0, 0 ~ t ~ 1
{
Y'( 1) = y(O) = O.
80
References.
[1]. V.Anuradha and R.Shivaji, A quadrature method for classes of multiparameter two point boundary value problems, Appl. Anal., 54(1994),
263-28l.
[2]. R.Aris, Introduction to the analysis of chemical reactors, Prentice-Hall,
New Jersey, 1965.
[3]. L.E.Bobisud and D.O'Regan, Existence of positive solutions for singular ordinary differential equations, Proc. Amer. Math. 80c., 124(1996),
2081-2087.
[4]. A.Castro and A.Kurepa, Infinitely many solutions to a superlinear
Dirichlet problem in a ball, Proc. Amer. Math. 80c., 101(1987),57-64.
[5]. A.Castro and R.Shivaji, Nonnegative solutions for a class of nonpositone problems, Proc. Royal80c. Edinburgh, 10BA(1988), 291- 302.
[6]. L.Erbe and H.Wang, On the existence of positive solutions of ordinary
differential equations, Proc. Amer. Math. 80c., 120(1994), 743-748.
[7]. A.M.Fink, J .A.Gatica and G.E.Hernandez, Eigenvalues of generalized
Gelfand type, Nonlinear Anal., 20(1993), 145:3-1468.
[8]. S. Lin and F .Pai, Existence and multiplicity of positive radial solutions
for semilinear elliptic equations in annular domains, 8IAM J.Math.Anal.,
22( 1991), 1500-1515.
[9]. D.O'Regan, A fixed point approach for a dass of non-positone problems, to appear.
[10]. J.Smoller and A.Wasserman, An existence theorem for positive solutions of semilinear elliptic problems, Arch. Rat. Mecb. Anal., 95(1986),
211-216.
81
Chapter 9. Differential equations singular in the solution variable.

9.1. Introduction.
This chapter discusses the second order differential equation y" =
J(t, y). Here J is not a Caratheodory function due to the singular behavior of its y variable and also the singular behavior of its t variable.
Many physical situations are modelled by problems of this kind. for example problems in gas and fluid dynamics [3,7]. Several problems in nonlinear
mechanics [7] lead to the second order boundary value problem
(1.1 )
y" + q(t)y-C; = 0, 0 < t < 1

yeO) = 0 = y( 1).
In [12] Taliaferro showed that (1.1) has a e[O, 1] n (,'2(0,1) solution; here
a > 0, q E C(O, 1), q > 0 on (0,1) and f~ t(1 - t)q(t)dt < 00. Since
Taliaferro's paper many authors [1,4,8,9,13,14] have examined the more
general problem, namely y" + J(t, y) = 0, 0 < t < 1, where J(t, y) > 0
for t E (0,1) and y > O. Also recently [6] the case where J is allowed
to change sign has been examined. In this chapter we present some of
the more advanced results to date in the theory of differential equations
singular in the solution variable. The technique we use was initiated in
1988 by Bobisud, O'Regan and Royalty [2]. We remark that the case when
J(t, y) < 0 for t E (0,1) and y > has received a lot of attention recently.
However, we will not discuss this problem in our book; instead we refer the
interested reader to [10, chapter 10] and its references.
9.2. Dirichlet boundary condition.

Several existence results are established for the singular Dirichlet boundary value problem
(2.1 )
{ y" + J(t, y)
yeO) = y(l)
= 0, < t <
= O.
These results rely on the following general existence principle for the classical second order boundary value problem
(2.2)
y". + g( t, y) = 0, 0 < t < 1

yeO) = a, yell = b.
Remark. By a solution to (2.2) we mean a function y E C[O, 1] n C 2 (0, 1)

which satisfies the differential equation on (0, 1) and the stated boundary
conditions.
82
Theorem 9.1. Suppose

g : (0,1) X R
(2.3)
R is continuous.
----+
(i). Assume
for each l' >

ther'e exists a hr E C(O, 1) with
J~ t(l - t)hr(t) dt < 00 such that Iyl::; l'
implies Ig(t, y)1 ::; hr(t) for' t E (0,1). Also assume
(2.4)
limt--+o+ e(1- t)hr(t) =

if J~(1 - x)hr(x)dx = 00
and lim t--+ 1 - t(l - t)'2h r (t) =
if J~ xhr(x) dx = 00
holds. In addition suppose there is a constant M, independent of A, with
IYlo =
sup ly(t)1 =P M
[0,1)
for any solution y to

{
<
+ Ag(t, y) =
0,
yeO) = a, y(l) = b
y"
<1
for each A E (0,1). Then (2.1) has a solution.

(ii). Assume
(2.6 )
there exists a hEC(O,l) With. J~t(l-t)h(t)dt<oo such

{ that Ig(t,y)l::; h(t) for' t E (0,1) and y E R. Also assume
limt--+o+ {2(1_ t)h(t) = 0 if J~(1- x)h(x)dx = 00
and limt--+1- t(1- t)2h(t) = 0 if J~ xh(x)dx = 00
holds. Then (2.1) has a solution.
PROOF: (i). Suppose y E C[O,I]. Then :cg(x,y(x)) E L1[0,t] for t

SInCE'
1
xlg(x,y(x))1 dx::; 1- t
Also
lim (1 - t)
1-+1-
it
0
x(l- x)lg(x,y(x1 dx.
10r sg(s,y(sds = 0
from (2.4). Similarly (1 - x)g(x,y(x E L1[t, 1] for t > 0 and

lim t1\1 - s)g(s,y(sds
t--+O+
8:3
= O.
< 1
Consequently
ret)
+ bt + ,x{1- t) 10 t sg(s,y(sds
a{1- t)
,xtl\I-8)g(8,Y(SdsEC[0,I]
with reO) = a and rei) = b.

It is now easy to check that solving (2.5 h is equivalent to finding a
solution y E e[O, 1] to
yet)
(2.7h
+ bt + ,x( 1 - t) I~ sg(s, y~
+,xt 1/(1 - s)g(s, yes~ ds.
= a(I -
t)
Define the operator N : e[O, 1]
Ny(t)
(2.8)
= a(l -
+t J/(l
t)
-'I-
ds
C[O, 1] by
+ bt + (1- t)I~ sg(s,y(sd.~

8)g(8, y~ d8.
Then (2.7h is equivalent to the fixed point problem
y=(I-,x)p+,xNy where p=a(I-t)+b.

We claim that N : e[O, 1] -'I- e[O, 1] is continuous and completely continuous. Continuity follows immediately from the Lebesgue dominated convergence theorem and the fact that
INy,,(t) - Ny(t)1
< (1 - t) 10 t slg~,yn(s)) - g(s,y(s1 ds

+ t
<
10
(1-s)lg(s,Yn{S-g(s,y(s))lds
s(l -
<~)lg~,yn(S
- g(s,y(s1 ds
with Yn, Y E e[O, 1]. To show complete continuity we will use the ArzelaAscoli theorem. To see this let n ~ e[O, 1] be bounded i.e. there exists
Mo > with lulo::; Mo for each u E n. Now if u E n we have
where hMo is determined from the bounded set nand (2.4). It is easy
to check that TMo E Ll[O, 1]. Hence (2.R) and (2.9) imply that Nn is a
bounded equicontinuous family on [0,1]. Consequently the Arzela-Ascoli

theorem implies N : C[O, 1] -+ C[O, 1] is wmpletely continuous. Set
= {u
E C[O, 1]:
lulo < max{ M,p} + 1},
= E = e[O, 1].
Theorem 2.5 guarantees that N has a fixed point i.e. (2.7h has a solution
y E C[O, 1].
(ii). Solving (2.2) is equivalent to the fixed point problem
y == Ny
where N is as described in (2.8). Essentially the same reasoning as in part
(i) implies N : C[O, 1] -+ C[0,1] is continuous and compact (since (2.6)
holds). The Schauder fixed point theorem guarantees that N has a fixed
point. 0
Our first existence result uses theorem 9.1 (i).
Theorem 9.2. Suppose the following conditions aT'e satisfied:
(2.10)
(2.11 )
f : (0,1) X (0,00)
R is continuous
If (t, y) I ::; q} ( t )g( y) + q2 (t ) h( y) on (0, 1) X (0, 00 )

with 9 >
continuous and nonincl'easing
on (0,00), h 2:
continuous on [0,00)
and ~ lwndecl'easing on (0,00); here qi E C(O, 1),
i = 1,2 with qi
(2.12)
-+
>
on (0,1) and I~ qi( x) dx < 00
let n E {3, 4, ... } and associated with each n we

have
a mnstant p" such that {Pn} is a
{
nonincT'easing sequence with lim" ..... oo Pn == 0 and
such thatfoT' ~::; t::; 1- ~ we have f(t,Pn) 2: 0
theT'e exists a constant M
{Z
)0
(2.13)
du
<
g{u} -
T'}
!!:El
+ T'2 g{z) +
j'P'
0
>
such that foT'
> 0,
du
g{u}
implies z::; M; hen.

ri
= max { '2 Io~ t( 1 -
i == 1,2
t)qi(t) dt, 2 I! t( 1 - t)qi(t) dt } ,
there exists a Junction 0' E qo, 1] n ('2 (0, 1) with

{ 0'(0) = 0'(1) = 0, 0' > 0 on (0,1) such that
J(t, y) + a"(t) > 0 Jor>
(t,y) E (0,1) X {y E (0,00): y < aCt)}
(2.14)
and
Jor any R > 0, ~ is diIferentiabZf; on (0, R]
{ with g' < 0 a.f' and
(2.15)
00
addztwn fo
&- E L1 [0, R]; in
1LWli
g(t) dt = 00.
g
Then (2,1) has a solution in C[O, 1]
n C 2 (0, 1).
PROOF: Fix n E {3,4, ... }. We begin by showing that

(2. 16
y"+J*(t,y)=O,O<t< 1
yeO) = y(l) = p"
has a solution; here
J*(t,y)
J(t, y), y 2: p"

{ J(t1,p,,) + p" - y, y < p" and ~:s; t :s; \ - ~
J(-n'P") + p" - y, y < p" and O:S; t :s; n
J(l-~,Pn)+Pn-Y' Y<Pn and 1-~:S;t:S;l.
Remark. Notice (2.10) and (2.11) imply that g = J* satisfies (2.3) and
(2.4).
To show (2.16)" has a solution we consider the family of problems
(2.17)~
y" + AJ*(t,y) = 0, 0 < t < 1,0 < A < 1

y( 0) = y( 1) = p",
We first show that
(2.18)
yet) 2: P", t E [0,1]
for any solution y E e[O, 1] n C 2 (0, 1) to (2.17)~. To see this suppose

y - p" has a negative minimum at to E (0,1), in which case y'(tO) = 0 and
y"(to) 2: o. However
y"(to)
= ->'!'(to,y(to)) = {
-A[J(tO,p,,) + Pn - y(to)], ~ :s; to :s; 1 - ~

-A[J(~,p,,) + Pn - y(to)] if O:S; to :s; ~
-A[J("~l,Pn)+Pn-y(tO)]' n~l :S;to:S; 1
X6
i.e. y"(tO) < 0, a contradiction. Thus (2.18) holds. Now since yeO) =
y( 1) = Pn we may assume the absolute maximum of y occurs at say
tn E (0,1), so y'(tn) = o. For x E (0,1) we have
he y( x ) )
- y" (x)
g(y(x)) :S ql(X)
(2.19)
+ q2(X) g(y(x))
Integrate from t (t < tn) to tn to obtain

y'(t)
g(y(t))
ft n {-g'(y(X))},
+ Jt
g2(y(X))
ft n
< Jt
[y (x)] dx
h( y( t n ))
g(y(t n ))
+
and so
lt
y'(t)
g(y(t)):S
ql(x)dx
h(y(tn ))
+ g(y(tn))
Integrate from 0 to tn to obtain
y (t n )
du
ltn
xql(X) dx
guo
))
+ 9h(y(tn
).
Y tn )
- () :S
qdx)dx
lt
tn
lt
()
q2 X
q2(x)dx.
xq2(X) dx
du
+ la0 -()
9 u
Pn
and so
(2.20)
fy(t n ) du
<
_1_ ftn
g(u) -
JO
I-tn JO
x(l- X)q (x)dx

1
~
_1_ ft n , (1
( ) dx
g(y(t n )) I-tn Jo x
- ')
X q2 X
+ JofP3 ~
g(u)
Similarly if we integrate (2.19) from tn to t (t 2: tn) and then from tn to

1 we obtain
(2.21)
fy(t n
JO
..lfu....
<
g(u) -
fl
tn Jt n
+~f~f!:B
x(1 - x)q 1 (x) dx
L ft~ X(1 -
x)q2(x)dx
Now (2.20) and (2.21) imply
lo
y(t n )
du
h(y(t n ))
- < r l + rZ
g(u) g(y(t n ))
lP3
+ft
~.
du
+
0
g(u)
where rl, rz are as in (2.13). Consequently (2.13) implies y(t n ) :S M and

so
(2.22)
Pn :S yet) :S M
87
for t E [0,1].
Now theorem 9.1 (i) implies (2.16)" has a solution y" E C[O, 1] n C 2 (0, 1)
(in fact in C 1[0, 1] n ('2(0, 1) sinn' J~ qi(X) dx < 00 ) with
(2.23)
p"
s y,,(t) s M
for t E [0,1].
Also Yn is a solution of
{
y" + J(t, y) = 0, 0 < t < 1

y( 0) = y( 1) = p".
Next we will obtain a sharper lower bound on y", namely we will show
(2.24)
aCt) S y,,(t) S M
for t E [0,1]
where a is as in (2.14).
If (2.24) is not true then y" - a would have a negative minimum at say
to E (0,1). In this case y~:(to) - a"(to) ~ o. However since 0 < y,,(to) <
a( to) and y,,( to) ~ p" we have
y~(to) - a"(to)
= -[J(to,y,,(to)) + a"(to)] < 0,
a contradiction. Hence (2.24) is true.

We shall now obtain a solution to (2.1) by means of the Arzela-Ascoli
theorem, as a limit of solutions of (2.16)". To this end, we will show
(2.2.5)
{yn}~=3
is a bounded, equicontinuous family on [0,1].
Of course {Yn} is uniformly bounded by (2.24). To show equicontinuity,

some more estimates are needed.
We have
Divide (2.26) by g(Yn( x) and integrate from 0 to 1 to obtain
Then since y~(O) ~ 0, Y:t(l)

on [0,1]) we have
(2.27)
s0
(note y,,(O)
{1 {_gl(y,,(X))} [ 1 ] 2
g2(y,,(x))
y,,(x) dx
io
= y,,(I) = p"
and Yn ~ p"
_
s io{l[ql(X) + q2(X )h(M)]d
geM) x = Ill
TI
Now consider
I(z) =
r [-g'/U)]t duo
10
g u)
Notice I is an increasing map from [0,00) onto [0,00) with I continuous

on [0, n] for any n > O. For t, s E [0,1] we have from Holder's inequality
that
Yn (t)
[-g'( u)]t
g(u)
yn(s)
dul
11t [-g'(Yn(x]t
y~(x)dxl
g(Yn(x
s
< It - sit (
1
r {-~'(Yn(x}
[y~(xW dX)
g2(Yn(X))
1
10
2"
< Kil t - s l2"

It follows from this inequality, the uniform continuity of I-Ion [O,/(M)]
and
IYn(t) - Yn( s)1 = Irl(l(Yn(t) - rl (I(Yn(s)1
that {Yn} is equicontinuous on [0,1]. Thus (2.25) is established.

The Arzela Ascoli theorem guarantees the existence of a subsequence N
ofintegers and a function Y E C[O, 1] with Yn converging uniformly on [0, 1]
to Y as n -+ 00 through N. Also yeO) = y(l) = 0 and aft) ::; yet) ::; M
for t E [0,1]. Now Yn, n EN, satisfies the integral equation
for t E (0,1). Notice (2.28) (take t = ~ say) implies {y:~(~)}, n E N is a

bounded sequence since aCt) ::; Yn(t) ::; M for t E [0,1]. Thus {Y~(~)}
has a convergent subsequence; for convenience let {y:~(~)}, n E N denote
this subsequence also and let r E R be its limit.
Fix t E (0,1). Since I is uniformly continuous on compact subsets of
[min(!,t),maxG,t)] X (O,M] let n --. 00 through N in (2.28) to obtain
yet) = Y (~)
+r
(t -~) + It(s - t)f(s,y(,qds.

2
Thus Y E C2(0, 1) and _y"(t) = I(t, yet~ for t E (0,1).
89
Our next existence result uses theorem 9.1 (ii).

Theorem 9.3. SUPPOM' (2.10), (2.11), (2.12), (2.14) and (2.1.1) hold. In
addition assume
(2.29)
there exists a funrtion /i E (,[0,1] n e:2(O, 1) with /32: P3

on [0,1] such that fit, li(f)) + 131/(t)
for t E (0,1)
:s:
is satisfied. Then (2.1) has a solution in
qo, 1] n (,2(0,1).
PROOF: Fix n E p,4, ... }. We first show

(2.30)"
y" + f (t, y) = 0, 0
y( 0) = y( 1) = Pn
<t<1
has a solution. The idea is to look at
(2.:H)n
y"
+ f** ( t, y) = 0, 0 < t < 1

= yell = Pn
yeO)
where
j**(t,y)
and 1': R
-+
f(t, (3(t)) + r(f3(t) - y), y > d(t)

f(t,y), Pn :s: y:S: 13(t)
f(t,Pn) + r(Pn - y), y < Pn and *:s: t:S: 1 - *
f(*,Pn)+r(Pn-Y), Y<Pn and O:s:t:s:*
f (1 - *' Pn) + r(Pn - Y), Y < pn an d 1 :s: t :s: 1
[-1, 1] is the radial retraction defined by

1'( u) = {
Remark. Notice 9
= f**
:s:
u if lui
1
fur otherwise.
satisfies (2.3) and (2.6).
Now theorem 9.1 (ii) implies (2.31)n has a solution Yn E C[O, 1] n

C 2 (0, 1). Essentially the same reasoning as in theorem 9.2 yields
( 2.:32)
Yn(t) 2: p" for t E [0,1].
Next we claim
(2.33)
Yn(t)
:s: Il(t)
for t E [0,1].
90
If (2.33) is not true then Yn - (3 would have a positive maximum at say

to E (0,1), in which case Y:~ (to) - (3" (to) ::; o. However since Yn( to) > (3( to)
we have
Y::(to) - (3"(to) = -[J(to,{3(to )) + r(f3(t o ) - Yn(tO))
+ (3"(to)] > 0,
a contradiction. Thus Pn ::; y,,(t) ::; f3(t) for t E [0,1] and so y" is a
solution of (2.30)". Essentially the same reasoning as in theorem 9.2 (from
equation (2.24) onwards) establishes that (2.1) has a solution. 0
We now obtain a "general upper and lower solution theorem" for singular problems.
Theorem 9.4. Suppose (2.1O), (2.12) and (2.14) hold. Assume
(2.34)
IJ(t,y)l::; qt(t)g(y) + q2(t)h(y) on (0,1) X (0,00)

with 9 > 0 continuous and nonincreasing
on (0,00) and h ~ 0 continuous on [0,00);
here qi E C(O,I), i = 1,2 with qi > 0 on
(0,1) and J~ x(1- x)qi(x)dx < 00. Also assume
limt--.o+ t 2(1- t)qi(t) = 0 iJ J~(l- x)qi(x)dx = 00 and
limt-+l- t(l- t)2qi(t) = 0 if J~ xqi(x)dx = 00, i = 1,2
and
for each n E {3, 4, ... } there exists a function
(3" E C[O, 1] n C2(0, 1) with!3n ~ Pn on [0,1]
such that J(t,!3,,(t)) + f3:~(t) ::; 0 for t E (0,1);
in addition Jor each t E [0,1] we have that {!3n(t)}
is a nonincreasing sequence and
lim,,-+CX) f3 n (0) = lim,,-+CX) /3,,(1) = 0
(2.35)
are satisfied. Then (2.1) has a solution in C[O, 1] n C 2 (0, 1).

PROOF: Fix n E {3,4, ... }. Consider (2.30)". The idea is to look at
(2.36)"
y" + J::." (t, y) = 0, 0 < t < 1

yeO) = y(l) = p"
where
J~*(t,y)
J(t,!3,,(t)) + r(f3,,(t) - y), y > !3n {t)

J(t, y), p" ::; y ::; f3,,(t)
J(t,p,,) + r(p" - y), y < p" and ~::; t ::; 1 - ~
J(~,p,,) + r(p", - :1/), y < Pn and 0::; t ::; ~
J(1 - ~,p,,) + r(p" - y), y < p" and 1 - ~ ::; t ::; 1
91
and r: R -+ [-1,1] is the radial retraction. Now theorem 9.1 (ii) implies
that (2.:J6)n has a solution y" E C[O, 1] n C 2 (0, 1). Essentially the same
reasoning as in theorem 9.3 yields
pn :S Yn(t) :S f3n(t) for t E [0,1];

hence Yn is a solution of (2.30)".
Remark. Notice y,,(t) :S /3 3 (t) for t E [0,1] since {!3 n (t)} is nonincreasing
for each t E [0,1].
Also as in theoren 9.2 we have y,,(t) ~ a(t) for t E [0,1].

Look at the interval [1,
Let
The mean value theorem implies that there exists T E (~, ~) with IY~( T)I ==
2 - Yn("3)1
1 :S 6suP[O,l) fh(t)
.
- L3 Hence for t E [1"3'"32] we have
3 1Yn("3)
=
and so
{Yn}~=3
is a bounded, equicontinuous family on
[~,
1] .
The Arzela-Ascoli theorem guarantees the existence of a subsequence N,

of integers and a function Z3 E C[~,~] with y" converging uniformly to
Z3 on [~,~] as n -+ 00 through N 3 . Similarly
{Yn}~=3 is a bounded, equicontinuous family on
[ ~4'4~]
so there is a subsequence N4 of N3 and a function Z4 E C[t,~] with y"

converging uniformly to Z4 on [t, as n -+ 00 through N 4 . Note Z4 == Z3
on [~,~] since N4 ~ N 3 . Proceed inductively to obtain subsequences of
integers
t]
and functions
92
with
Yn converging uniformly to
Zk
[ ~k' 1 -~]
k
on
and
Zk+l
Zk
as
11, ----+
00 through Nk
[~, 1 - ~] .
on
Define afundion y: [0,1]- [0,00) by y(;r) = Zk(X) on [t,l- t] and

y(O) = y(l) = O. Notice y is well defined and a(t) ::; y(t) ::; i33(t) for
t E (0,1). Next fix t E (0,1). Then there exists In E {3,4,5 ... } with
t E (~,1- ~). Now Yn, 11, E N m , satisfies the integral equation
for x E (0,1). Let 11, ----+ 00 through N m to obtain as in theorem 9.2 (here
r E R is as in theorem 9.2),
for x E
[~,1-~] i.e.
y(x)
= y (}) + r
(x -})
+ hX{.~ - x)f(s,y{s))ds
2
for x E [~,1- ~]. Hence y E C 2 {0,1) with yl/(t) = -f(t,y(t)) for each
t E (0,1). It remains to show y is continuous at 0 and 1.
Let E > 0 be given. Now since limn--+oo i3 n (0) = 0 there exists no E
{3,4, ... } with ,l3no{O) < ~. Since /3 no E e[O, 1] there exists ono > 0 with
/3 no (t) < ~ for t E [O,ono]'

Now for n::::: no we have, since {/3,,(t)} is nonincreasing for each t E [0,1]'
f
:2
!3,,{ t) ::; /3 no (t) <
for t E [0, bno )'
This together with the fact that a(t) ::; Yn(t) ::; /3 n (t) for t E (0,1), implies
that for n ::::: no we have
n(t)::; y,,(t)
<
:2
93
for t E [0,0"0)'
Consequently
0::; Q(t) ::; y(t) ::;
"2 < f
for t E [O,b no ]
and so y is continuous at O. Similarly y is continuous at 1 and so y E
C[O,l].
We now discuss condition (2.14). Usually one can construct Q explicitly

from the differential equation; see [9,10]. For convenience we now give a
rather general result.
Theorem 9.5. Suppose (2.10), (2.11) and (2.1,1)) hold. In addition assume
the following condition i.9 satisfied:
let n E {3, 4, ... } and associated with each n we have
a constant Pn such that {Pn} is a
decreasing sequence with limn-+oo pn = 0,
and there exists a constant ko > 0 such that for
~ ::; t::; 1 - ~ and 0 < y ::; Pn we have f(t,y) 2: ko.
(2.37)
Also assume (2.13), here Pn i.9 as in (2.37), holds. Then (2.1) has a solution
in C[0,1]nC 2 (0,1).
PROOF: Clearly (2.37) implies that (2.12) holds. We now show that (2.14)
is satisfied by explicitly constructing Q(t); this is a standard construction
off the sequence of constants {Pn}; see [6] for example.
The details are as follows. Let 0::; x ::;
and
ro(x)
= { Pk (x - t) + L~=k+lPm Cn~1 -~),

0, x
Remarks. (i).
= o.
",,00
L..Jm= +1
Pm
x E (t,
k~I]' k = 4, ...
(_1
__ 1..)
< P3""00
(_1_
- 1..)
= !P3'
m-1
m
.
m-1
m
3'
L..Jm=4
(ii). Notice ro(x) = f;</J(s)d.~ for 0::; x::;! where </J: [O,!] -- [0,00)
the step function defined by
</J(t)
={
0, t = 0
Pk, k E (t, k~1], k
= 4,5, ....
Here ro E C[O,~] and notice ro(t) ::; </J(t) ::; Pk for t E

4,5, ....
Next define
94
(k, k~1]'
IS
(t,
Notice 1/)(0) = 0, 1/, >

on (o,~], '!), E (,'2[0,~] and 1/l'
Also ''!}l(t)::; Pk for t E
k~l]' k = 4,5, .. and so
fit, y)
(2.38)
ko for (t, y) E
Let
(o,~]
>
for t E (0, ~).
X {y E (0,00) : y ::;
'lb(t)}.
0::; t < ~
q(t)':31 ::; t ::; :32
2
~l( 1 - t). :3 < t ::; 1.
'!})( t),
a * (t) =
Here q : [~, ~l--+ (0,P3] is such that q E C<2[~,~] with
and q"(l)
3 = 1//'(!)
3 = q"U)
3
Notice since
< q(t) ::; P3
f(t, y) ~ ko for
(2.39)
for t E [~,~] we have
(t, y) E
[3'132]
X {y E (0,00) : y::;
q(t)}.
Consequently (2.38) and (2.:39) imply
(2.40)
f(t, y)
Also 0:*(0)
= 0*(1) = 0,
ko for (t, y) E (0,1) X {y E (0,00) : y::; o*(t)}.

0:*
>
(2.41)
on (0,1) and 0:* E (,'2[0,1]. Finally define
o(t) = 110*(t)
where
11
= mi n
{1.
(0*
Now Q E (,'2[0,1] with 0:(0) = 0:( 1) =

ott) ::; o*(t) we have from (2.40) that
)~I0+1
}.
and 0: > on (0,1). Also since
fit, y) ~ ko for (t, y) E (0,1) X {y E (0,00) : y < o(t)}.
III addition for (t';Ij) E (0,1)

f(t.y) + o"(t)
~ /,;0 +
{y E (0,00): y < art)} we have
O'''(t)
k I(O'*)//(t)1
~ /,;0 - 1(:*)//10 + 1 > ko - ko
Hence (2.14) is satisfied. Existence is
HOW
95
l?;uarantef'C1. 0
= O.
Similarly we have the following results; we use theorem's 9.3 and 9.4 respectively.
Theorem 9.6. Suppose (2.10), (2.11), (2.15), (2.37) and (2.29), here Pn
is as in (2.37), hold. Then (2.1) has a solution in C[O, 1] n C2(0, 1).
Theorem 9.7. Suppose (2.10), (2.34), (2.:37) and (2.35), here pn is as in
(2.37), hold. Then (2.1) has a solution in C[O, 1] n C2(0, 1).
Example 9.1. The boundary value problem
(2.42)
(y;
y" +
+ "'( t )yB - J.l2) = 0, 0 < t < 1
y( 0) = y( 1) = 0, J.l -:J 0, 'Y > 0, 0 ::; () < 1
has a solution; here", E C(O, 1), 71 > 0 on (0,1) and f~ 71(t) dt <
We will apply theorem 9 ..5. Notice first that (2.:37), with
1
= ( n (2
) * and
J.l + 1
Pn
ko
00.
= 1,
is true since for ~ ::; t ::; 1 - ~ and 0 < y ::; Pn we have
f(t,y)~--J.l ~--J.l ~(J.l +1)-J.l =1.

y"Y
ny"Y
.
Also with ql(t) = t,g(y) = y---r,q2(t) = 11(t) + J.l2 and hey) = yB + 1 it is

easy to check that (2.11), (2.13) since 0::; () < 1, and (2.15) hold. Existence
of a solution to (2.42) is now guaranteed from theorem 9.5.
(2.43)
{ Y"+(~-J.l2)=0,0<t<1
yeO) = y(l) = 0, J.l -:J 0
has a solution.
We will apply theorem 9.6. Notice (2.37), with
Pn
= ( 4n 2( J.l12 + 1) )t
and ko
= 1,
is true. Also let ql(t) = ~,g(y) = y-'2,Q2(t) = J.l2 and hey)

(2.11) and (2.15) hold. Finally (2.29) is satisfied with
f3(t)
= 2J.l + P3
96
= 1.
Clearly
Existencl;' of a solution to (2.4:3) is now guaranteed from theorem 9.6.

{
(2.44)
+ (t 2(1-t)2
+ 1L4 _ J.L2) = 0 '
4y2
y(O) = y( 1) = 0, J.L :f 0
"
Y
0<t < 1
has a solution.
We will apply theorem 9.7. Noticl;' (2.37), with
p"
= ( 4n 4( J.L 2 + 1) )
and ko
= 1,
is true since for ~ ::; t ::; 1 - ~ and 0 < y ::; p" we have
1
J(t,y) ~ - - 0 - J.L
4n p"
= 1.
Also wIth ql(t) =

4
,g(y) = y-'Y,q2(t) = 1 and h(y) = ~ + J.L we
have that (2.34) is satisfied. Finally (2.35), with (3,,(t) = t(l - t) + p",
holds since
t 2 (I-t)2
J(t,p,,(t))
+ p::(t) ::; 4 +
(1
+ )2
:" -
J.L2 - 2 ::;
4+ 1 -
J.L2 - 2 <
o.
Existence of a solution to (2.44) is now guaranteed from theorem 9.7.

9.3. Nonlinear boundary condition.
In this section we discuss the singular problem

y"+J.(t,y)=o,o<t< 1
{ y{O) = 0
O(y'(l)) + y(l) = 0
(3.1 )
where 0 may be nonlinear. Of course our results include the Sturm Liouville condition k y'(1) + y( 1) = 0 where k ~ O.
Theorem 9.B. Suppose (2.10),(2.11),(2.12) and (2.15) hold. In addition
suppose the following conditions are satisfied:
(3.2)
0:R
(3.3)
there exists a constant M > 0 such that Jor z > 0,

{ J~ g(~) ::; J~ xql(x)dx +
J~ xq2(x)dx + ft3 g(~)
implies z::; M
---->
R is continuous and nondecl'easing with 0(0)
*f
97
=0
there
{ with
(0,1)
(t,y)
(3.4 )
exists a Junction 0' E qo, 1] n Cl(O, 1] n ('2(0, 1)

0'(0) = 0(0"(1) + 0'(1) = 0, 0' > on
such that J(t, y) + a"(t) > for
E (0,1) x {y E (0,00): y < aft)}
and
hI
(3.5)
qi(x)g(a(x))dx < 00, i == 1,2.
Then (3.1) has a solution in (,[0,1] n ('1(0, 1] n ('2(0, 1).

PROOF: Fix n E {3,4, ... }. We first show
{ y" + J( t, y) = 0, < t < 1

yeO) = pn, O(y'(1)) + y(l) = p"
(3.6)n
has a (,1[0,1) n C 2 (0, 1) solution. The idea is to look at

(3.7)"
y" + 1*(t, y) == 0, 0< t < 1

yeO) = Pn, O(y'(I + y(1) = Pn-
We establish that (3.7)" has a ('1[0,I]n('2(0,1) solution; here

theorem 9.2.
Look at the family of problems
(3. 0 )7{
()
y" + ). 1*(t, y) = 0, < t < 1,

{ Y(0) = p"
M( y' (1)) + y(1 ) = Pn
1*
isasin
< ). < 1
We claim
yet) 2: p", t E [0,1]
( 3.9)
for any solution y E (,1[0,1] n ('2(0,1) to (;3.8)~. To see this suppose

y - Pn has a negative minimum at to E (0,1). If to E (0,1) then we obtain
a contradiction as in theorem 9.2. It remains to consider the case to = l.
Then y'(l):S and so O(y'(l:S from (;3.2). However
)'O(y'(I)) = p" - y(l) > 0,

a contradiction. Thus (;3.9) holds.
98
Suppose the absolute maximum of y occurs at t" E [0,1]. In fact we

may take tn E (0,1), and so y'(t n ) = O. To see this notice if y(t n ) = p"
then y == p". Next if y(t n ) > Pn then if tn = 1 we have y'(l) ~ 0 and so
0:::; AO(y'(1 = Pn - y(l) < 0,

a contradiction.
For x E (0,1) we have
_y"( x)
h(y(x
g(y(x :::; ql(X) + Q2(X) g(y(x
(3.10)
and integrate from t (t < t n ) to tn and then from 0 to t" to obtain (as
in theorem 9.2),
r(t n l
(3.11) io
du
fl
h(y(tn
fl
[p3 du
g(u):::; io xQl(x)dx+ g(y(t n io xQ2(x)dx+ io g(u)'
Now (3.3) implies yet,,) :::; M and so
p" :::; yet) :::; M for t E [0,1].
(3.12)
Also the mean value theorem implies that there exists r E (0,1) with
ly'(r)1 = ly(l) - y(O)1 :::; 2M. For t E [0,1] we have
ly'(t)1
< ly'(r)1 + 11t If*(x,y(xdxl
heM)
< 2M + g(Pn) io [Ql(X) + Q2(X) geM)] dx == MI.

Define the mappings
L, F: C~JO, 1]
by
Ly(t)
= (y'(t) -
and
Fy(t)
= (-
1t
->
ColO, 1] x R
y'(O), p" - y(l
j*(x,y(xdx, O(y'(1).
Here Co[O,l] = {u E C[O,l]: u(O) = O} and C~JO,l] = {u E 1 [0,1]:

u(O) = Pn}. Now F is completely continuous. Also if Ly = (u(t),1') then
yet) = Pn -t
(1'+ 11 u(x)dx) +1t u(x)dx;

99
hence L -1 exists and is continuous.

Solving (3.8)~ is equivalent to finding a fixed point of Y = )"L -1 Fy =
)"N y where N = L -1 F : (,'~JO, 1] ---+ (,'~JO, 1] is completely continuous.
Let
u = {u E (,'~JO, IJ : luh < max{M, Md + I},
(,'
= (,':JO, 1]
and E = (,'1[0,1]; here luh = max{lulo, lu'lo}. Now theorem 2.5 implies
that (3.7)n has a solution Yn E 1 [0, 1] n (,'2(0.1). Also Pn ~ Yn(t) ~ M
for t E [0,1], so Yn is a solution of (:J.6)".
Next we show
(3.13)
o(t) ~ Yn(t) ~ M for t E [0,1].
If this is not true then Yn - 0 would have a negativp minimum at say

to E (0,1 J. If to E (0,1) then we obtain a contradiction as in theorem 9.2.
If to = 1 then (Yn - 0)'(1) ~ 0 i.e. y~(1) ~ 0'(1) and 0 < Yn(I) < 0(1).
However
0< pn
+ [0(1) -
= O(y~(1)) -
Yn(1)]
0(0'(1)) ~ 0,
a contradiction. Hence (3.13) is true.

We have
-y"(x)
heM)
g(y:(x)) ~ ql(X) + Q2(X) g('M) for x E (0,1).
Integrate from 0 to 1 to obtain
Then since y~(O) 2: 0 and Y:t(l) ~ 0 (notice O(Y:t{I))

implies y~ (1) ~ 0) we have
(3.14)
10{
I { -
Now consider
10t
g' (Yn (x) ) } '

2
g2(Yn(X)) [Yn(x)] dx ~
-l
fez) -
[Ql(X)
[-g'(u)]2'
()
9 u
du
and notice f (as in theorem 9.2) for t,s E [0,1] that
100
- Yn(1) ~ 0
M )]
_
+ Q2(X) h(
geM) dx =
= pn
M 2
Consequently
(3.15)
{Y"}~=:3 is a bounded, equicontinuolls family on [0,1].
The Arzela Ascoli theorem guarantees the existence of a subsequence N of

integers and a function y E e[O, 1] with y" converging uniformly on [0,1]
to y as n
rxJ through N. Also y( 0) =
and 0:( t)
y( t)
M for
t E [0,1]. Now Yn, 11 EN, satisfies the integral equation
---7
y" ( t)
= y" (t)
+ (t - t) n- 1 (Pn
(:3.16)
- Yn
:s
0) - f;(x -
:s
t)J(x,y,,(x))dx)
+ f/(t -1)!(x,y,,(x))dx
given by n(x) = ~ + O(x); notice n
+ Jl(x - t)J(x,y,,(x))dx
2
for t E [0,1]. Here n : R -+ R is

is strictly increasing. We would like to let n First notice
Jl
IJ(x,Yn(x)1 dx <
Jl
g(o:(x)) qJfx)
1
2
00
through N in (3.16).
heM)}
+ q2(X)--)
geM
dx <
00.
Fix t E (0,1]. Let 11 ----t 00 through N in (3.16), and so the Lebesgue

dominated convergence theorem implies
yet)
(~) + (t -~) n-
it
(x
(-y
-~) J(x,y(x))dx +
(~) _ ~1 (x -~) f(x,Y(X)dX)
/1 (t -~)
J(x,y(x))dx
for t E (0,1]. Also for t E (0,1] we have
y'(t) =
n-
(-y
(~) -
ftl
(x
_~) J(X,Y(X)dX) + /1 J(x,y(x))dx
and so y' E C(0,1]. In addition y"(t)

hence y E 2 (0, 1). Finally
0(y'(1))+y(1)
(n-
1n-1
= - J(t,y(t))
for t E (0,1) and
(_y(~) _ ~1 (x- ~)f(x'Y(X)dX))
(-Y(1) -
~1 (x-1)f(X,Y(X))dX)
(1) + it (a: -1) J(x,y(x))dx = 0.

2
101
Now suppose (2.37) is satisfied. Let a* and a be as in theorem 9.5.

Notice a*(l) = 0 and (a*)'(t) = - J~-t ro(x) dx for ~ ::; t ::; 1. Consequently (a*)'(l) = 0 and so O(a'(l + a(1) = 0(0) + 0 = O. However since
(3.5) must be satisfied it is desirable to construct the "best" a. Usually
it is possible to obtain an explicit a from the differential equation. We
conclude by giving a general result for the boundary value problem
y"+J(t,y)=O,O<t< 1
{ yeO) = 0
ky'(l) + y(l) = 0, k ~ 0 a constant.
(3.17)
Remark. Note (3.17) is a special case of (3.1); here O(u) = ku.

Theorem 9.9. Suppose (2.10), (2.11), (2.37) and (3.3) are satisfied.
addition assume
(3.18)
ht
q;(x)g(ao(l - xdx <
00,
= 1,2
Jor any ao
> 0, if k
In
=0
and
(3.19)
there exists T E (0,1) with J(t, y)

Jor t E [T, 1) and 0 < y ::; P3
hold. Then (3.17) has a .mlution in C[O, 1]
>0
n (.'1(0, 1] n C 2 (0, 1).
PROOF: Let, denote the a constructed in theorem 9.5. Without loss of

generality assume ~ < T ::; ~. Define
,(t), 0::; t ::; ~

<t <T
!L(t) = { wet), ~
kl'(r)
(k+t)-r
and a(t)
= 11JL( t)
t>
I'(r)(t-t)
+ (k+l)-T'
'_ T
where
11 =
. { 1. I Ik o }
. JL" 0 + 1
lllIll
Hen' W:[~,TJ---.(0'P3] issurhthat wEC2[~,T] with
1
w(:1)
='(:1)'1
I 1
w (:1)
=,'( 3)'1
,,1
w (:1)
=,"( 3'1 )
and
WeT) = ,(T), W'(T)
= (k -,(;)
+1 102
,W"(T) = O.
T
We now claim that 0' satisfies (:3.4). First notice 0'(0)

We also have as in theorem 9.5 that
(3.20)
f(t, y)
Now for t>
+ a"(t) > 0
for (t, y) E (0, T) X {y E (0,00) : y < aCt)}.
we have al/(t)
0'
( t)
<
- J1
(t)
= ka'(I)+a(l) = O.
<
=0
- (k
and also
kp
(1 - T)p
+ 1).1 _ T + (k + 1) _ ."3T
_
-
P.3
Consequently
(3.21 )
f(t, y)
+ al/(t) =
f(t, y) > 0
for (t,y) E [T,I) X {y E (0,00): y < a(t)}. Thus (3.20) and (3.21) imply
that a satisfies (3.4). Notice also that (3.5) is satisfied with the above a
since (3.18) holds. Existence of a ('[0,1] n ('1(0, 1] n ('2(0,1) solution to
(3.17) is now guaranteed. 0
References.
[1]. L.E.Bobisud, J.E.Calvert and W.D.Royalty, Some existence results for

singular boundary value problems, Ditf. Int. Eq., 6(1993), 553-571.
[2]. L.E.Bobisud, D.O'Regan and W.D.Royalty, Solvability of some nonlinear boundary value problems, Nonlinear Anal., 12( 1988),855-869.
[3]. A.Callegari and A.Nachman, A nonlinear singular boundary value
problem in the theory of pseudoplastic fluids, SIAM J. Appl.Math.,
38(1980),275-281.
[4]. J.A.Gatica, V.Oliker and P.Waltman, Singular nonlinear boundary
value problems for second order ordinary differential equations, J. Ditf.
Eq., 79( 1989), 62-78.
[5]. A.Granas, R.B.Guenther and J.W.Lee, Some general existence principles in the Caratheodory theory of nonlinear differential systems, J. Math.
Pures et Appl., 70(1991), 153-196.
[6]. P.Habets and F.Zanolin, Upper and lower solutions for a generalized
Emden-Fowler equation, J. Math. Anal. Appl., 181(1994),684-700.
[7]. C.D.Luning and W.L.Perry, Positive solutions of negative exponent
generalized Emden-Fowler boundary value problems, SIAM J. Math. Anal,
12( 1981), 874-879.
[8]. C.D.Luning and W.L.Perry, Iterative solutions of negative exponent
Emden Fowler problems, Int. J. Math. and Math. Sci., 13(1990), 159164.
[9]. D.O'Regan, Positive solutions to singular boundary value problems
with at most linear growth, AppJ. Anal., 49( 1993), 171-196.
[10]. D.O'Regan, Theory of singular boundary value problems, World Scientific, Singapore, 1994.
[11]. D.O'Regan, Singular differential equations with linear and nonlinear
boundary conditions, to appear.
[12]. S.Taliaferro, A nonlinear singular boundary value problem, J. Nonlinear Anal., 3(1979), 897-904.
[13]. A.Tineo, Existence theorems for a singular two point dirichlet problem, J. Nonlinear Anal., 19(1992), :l23-:33:l.
[14]. J.Wang and J .Jiang, The existence of positive solutions to a singular
nonlinear boundary value problem, J. Math. Anal. Appl., 176( 1993), 322329.
104
Chapter 10.
problems.
Existence principle for singular boundary value
10.1. Introduction.
In chapters 11 and 12 we discuss "nonresonant" and "resonant" second
order problems. To tackle thesf' problems in their most general setting
we need to establish an existence principle for "singular" boundary value
problems. This chapter presents such a result. In particular we establish
an existence principle for the "nonresonant" singular two point boundary
value problem
p(tlq(t) (p(t)y'(t),
(1.1 )
+ My(t) = J(t, y(t
a.e. on [0,1]
limt-+o+ p(t)y'(t) = 0
y(l) = 0
where M is such that
( 1.2)
;q(PY')' + MY = 0 a.e. on [0,1]

{ limt-+u+ p(t)y'(t) = 0
y( 1) = 0
has only the trivial solution. This result is due to O'Regan [16]. The proof
relies on the Banach contraction principle, on some results in linear and
nonlinear integral operator theory, and on the fixed point results of chapter
2. Throughout the chapter p E C[O, 1] n C 1 (0, 1) together with p > 0 on
(0,1); also q is measurable with q> 0 a.e. on [0,1] and
fal p(x)q(x)d:J; <
00.
It is worth remarking here that if our problem were "regular" then we

could make a change of variables (i.e we could use thf' Liouvillf' transformation [1 J). so that thf' existence principles in chapter 5 (or extensions
thereof) could be used in chapters 11 and 12.
Remal'b. (i). Throughout the condition y( 1) = 0 could bf' replaced by
the more general condition ay(l) + b limt-+l- p(t)y'(t) = 0, a > O,b 2: O.
(ii). We do not assume f~ p(!) < 00.
In addition J : [0,1] X R ~ R will lw a Caratheodory function. By this

we Il1f'an:
(i). t f----7 J(t,y) is measurable for all y E R
10!)
(ii). y
r-+
f(t,y) is continuous for a.e. t E [0,1].
For notational purposes let 'UJ be a weight function. By L:;'[O, 1],

we mean the space of functions 'U such that
11
w(t)lu(tW dt <
:2: 1
00.
In particular L~[O, 1] denotes the space of functions u such that
11
w(t)lu(t)12 dt < 00;
also for u, v E L~[O,I] define {u, v} = f~ w(t)u(t)v(t) dt. Let AC[O,I] be

the space of functions which are absolutely continuous on [0,1].
It is worth remarking here that problems of type (1.1) occur in many
applications in the physical sciences, for example in radially symmetric
nonlinear diffusion [17] in the n-dimensional sphere we have p(t) = tn-I;
these problems involve a homogeneous Neumann condition at zero i.e.
limt-->o+ t n - 1 Y'(t) = O. Another important example is the Poisson Boltzmann equation
(1.3 )
y" + TY' = f(t, y), 0 < t < 1

y'(O+) = y(1) = 0, a:2: 1
which occurs in the theory of thermal explosions [2J and in the theory of
electrohydrodynamics [9].
10.2. Linear problem.
We begin our study of (1.1) by first discussing the singular linear problem. Our goal is to establish a Green's function for the appropriate problems considered in this chapter.
Theorem 10.1. Suppose
P E e[O, IJ n Cl(O, 1) with p
(2.1 )
(2.2)
q E L![O, IJ with q
> 0 on (0,1)
> 0 a.e. on [0, IJ
and
(2.3)
11 (1 P(X)q(:l:)dX)~
p!s)
d8 <
106
00
f07'some constant a> 1
are satisfied.
(i). Then
~(py')' + Ilqy = 0, a.f. on [0,1]
{ limt-+o+ p(t)y'(t) =
y( 0) = ao =1=
(2.4)
has a solution Yl E qo, 1] n (.'1(0,1) with py~ E AC[O, 1]. (By a solution to (2.4) we mean a junction y E qo, 1] n Cl(O, 1), py' E AC[O,I]
which satisfies the differential equation a.e. on [0,1] and the stated initial
conditions.)
(ii). Then
{
(2.5)
~(py')'
+ MY =
0,
a.e. on [0,1]
y( 1) =
limt-+l- p(t)y'(t) = 1
has a solution Y2 E L~q[O, 1) with Y2 E C(O,l]
n Cl(O, 1)
and py~ E
AC[O,l].
PROOF: (i). Let C[O,l) denote the Banach space of continuous functions
on [0,1] endowed with the norm
lulK=
sup le-I{R(tju(t)1 where R(t)= (tp(x)q(x)dx

tE[O,l)
Jo
and
Remark. Here {3
= "'~l
i.e {3 and a are conjugate exponents.
Solving (2.4) is equivalent to finding y E

y(t)=ao-Il
i is
t
Define the operator N : C[O, 1]

Ny(t) = ao -Il
1
-(p s)
-+
qo, 1]
which satisfies
p(x)q(x)y(x)dxds.
(.'[0,1] by
t
- 1
o p( s)
is
107
p(x)q(x)y(x)dxds.
Now N is a contraction since
IILllu -
INu - NvlK :::;
< IILllu - vlK

sup e- K R(t)
tE[O, I)
t:~:I) e
IILI
<
-K R(t)
(13K)7J
tE[O,I)
rs
1 (rs
Jo p(8)
Jo p(x)q(x)dx
tE[O,I)
(1- e-(3K R(I))
<
pqe!3K R(x) dX) 73 ds
)!.- (e!3 K R(s)

13K
1) ~
- 13K
ds
lu - vlK
sup e- K R(t) (e!3 K R(t) _
r ~ rs pqeKR(x) dx ds
Jo p Jo
_1_ (
pq dX) a ( t
Jo p( 8) Jo
Jo
< IILllu - vlK

x
vlK sup e- K R(t)
fj
1) ~
r _1_ ( rs p( x )q( x) dX) a d8

Jo p( 8) Jo
1
lu - vlK
using Holder's integral inequality. The Banach contraction principle now

establishes the result.
(ii). Let L~q[O, 1] denote the Banach space of functions u, with
101 pqlul a dt < 00,

endowed with the norm
lIuliK
(lol p(t)q(t)e-KQ(t)lu(tW dt)!'-
and
K =
7i
( Ifa
IILla
p(t)q(t)
d~
(11P(~)
h.
2
p(8)
1t
(
I02 p(x)q(x) dx
11
p(x)q(x)dx
a
where 13 = a-I
oc since
<~
rs
{I _1_ (
P(x)q(X)dX)!.- d8.
p(8) Jo
Jt
108
d
1
p(8)(f~ p(x)q(x)dx)a
<
dt
(f~ p(x )q(x) dx )a
It p1:) <
Remarks. (i). Notice for example that
(t d8
where Q(t)
(ii). Notice (2.3) implies
r p(t)q(t) (1t
(2.6)
io
To see this let
get) =
(1
t
dS) (} dt <
pes
d
00.
(}-1
pC;)
and fix E, 0 < f < 1. Interchange the order of integration and use Holder's
inequality to obtain
J
l
( p(t)q(t)g(t)
11
t
d~
pC:) dt
J, -pes)1 JS, p(t)q(t)g(t) dt

l
(/,1 p(t)q(t)gf\t) dt)

1,p(s) (Js, p(t)q(t) dt ).!." ds.
1
<
X
73
1 .
-
Consequently
/,1 p(t)q(t) (11p~:J (} dt
<
X
(/,1 p(t)q(t) (11p~:J (} dt)

1 pes (IS, p(t)q(t) dt )1."ds.
1
73
-1)
We will show that
(2.7)
yet)
-11 ~ pes)
fll1 _1_
p(x)q(x)y(x) dx ds
t pes) is
has a solution Y2 E L~q[O, 1]. Also we will show Y2 E C(O, 1] n C 1(0, 1) and
py~ E AC(O, 1] and consequently Y2 will be a solution of (V5).
Define the operator M: L~q[O, 1] - L~q[O, 1] by
My(t)=-
1
1
d~
p( s
~)-fl
11 11
t
-1
p( s)
p(x)q(x)y(x)dxds.
Remark. M is well defined because of (2.6) and
109
(1
:; (11
~
pqlyl dX) a
pqlyla dX)
11 (11p~:))
(11 *11 (1 p~:))
pq
pq dX)
dt
pq
a dt
for any y E L~q[O, 1).

Now M is a contraction since
The Banach contraction principle now establishes that (2.7) has a solution
Y2 E L~q[O, 1). Also
so py~ E AC[0,1) since Y2 E L~q[O, 1) implies pqY2 E Ll[O,1). Thus Y2 is a

solution of (2.5). 0
Consider now
(2.8)
1
')'
-(PY
pq
+ IlY
= h(t)
a.e. on [0,1 )
where (2.1), (2.2), (2.3) and

(2.9)
hE Lp!3q[O, 1]; here (3 = _a_
a-I
hold.
110
Theorem 10.2. Suppo<~e (2.1), (2.2), (2.3) and (2.9) are satisfied. In addition fL is such that (1.2) has only the trivial solution. Then
p1q(py')' + flY = h(t) a.e. on [0,1]

{ limt-+o+ p(t)y'(t) = 0
y(l) = 0
(2.10)
has exactly one solution y (note y E L~q[O, 1] with y E C(O, 1] n C 1(0, 1)

and py' E AC[O, 1]) given by
yet)
(2.11)
= fa1 G(t,s)q(s)h(s)ds
where G( t, s) is the Green's function i. e.

G(t s) _
,
Yl(S)Y2(t)
W(s)
yI!t)Y2(S)
W(s)
= c 0 pes) y1 (s) y.2(t), 0 < s <_ t

= coP(S) Y2 (s) Y1 ()
t, t _< s < 1.
Here Y1 and Y2 are as described in theorem 10.1 and W~) is the Wronskian of Y1 and Y2 at s and notice p(s)W(s) = ~ i- 0 for s E [0,1].
PROOF: This follows the standard construction ofthe Green's function; see
[17,19] for example. We will just justify that p(s)W(s) i= 0 for s E [0,1].
To see this all one needs to show is that Y1 (1) i- o. If Y1 (1) = 0 then
Yl satisfies (1.2) and consequently Y1
o. This contradicts the fact that
y(O)=aoi-0.
Remark. Notice Y in (2.11) is in

fa1 p(t)q(t)
(1
L~q[O, 1] since
p(S)qIY2(S)Y1(t)h(s)1 dS) C> dt
::; fa1 pq1Y11C>
(1 pq1Y21C> d<~) (1 pqlhl i1 dS) ~ dt

1
and also
10.3. Existence principle.

We use theorem 2.5 to establish our existence principle. Before we
do so we gather together some well known results [10] from the theory of
nonlinear integral equations.
111
Q > 1 be a constant and J : [0,1] X R

Caratheodory function. Define the operator
Theorem 10.3. Let
Fy(t)
and suppose F: L~q[O,
and bounded.
1J
---->
---->
R be a
= J(t,y(t))
L~q[O,1J; her'e /3
= "~1'
Then F is continuous
Theorem lOA. Consider' the linear integal operator

Ay(t)
with
(3.1)
11
p(t)q{t)
Then A : L~q[O,
1J
---->
11
p(s)q(s)k(t,s)y(s)ds
101 p(s)q{s)lk(t,s)I"dsdt <

L;q[O,
1],
(i
= c>~1
00
Jor some
1.
is completely continuous.
We next prove an existence principle for (1.1).

Theorem 10.5. Let J : [0,1] X R ----+ R be a Caratheodory function and
suppose (2.1), (2.2) and (2.3) are satisfied. Also suppose
(3.2)
J(t,y(t)) E
L~q[0,1]
whenever y E L;q[O,1J;
here (3
Q:
l'
In addition I-L is such that (1.2) has only the trivial solution. Now suppose
there is a constant M o , independent of A, with
IIyll
(1
p(t)q(t)ly(tW dt) a
:I Mo
for any solution y (here y E L~q[O, 1] with y E C(O, 1]

py' E AC[O, 1]) to
n C 1(O, 1) and
(3.3h
p1q(py')' I-LY = AJ(t,y) a.e. on [0,1]

{ limt~o+ p(t)y'(t) = 0
y(1) = 0
for each A E (0,1). Then (1.1) has at least
aUf
solution.
PROOF: Solving (3.:3h is equivalent to finding y E L;q[O, 1] which satisfies
(3.4 )
112
when'
k (t,s ) -_ {
.
and Y1, Y2,

L~q[O,
Co
Co
Co
yd 8 ) Y2 (t ), 0
Y2 (S ) Y1 (t), t
< .~ s: t
s: .~ < 1,
are as described in theorem 10.2. Define the operator N :
1]-- L~q[O, 1] by
Ny(t)
11 p(s)q(s)k(t,s)f(.~,y(s))ds.
Remark. N is well defined since
and also
11 p(t)q(t) (1 p(8)q('~)\Y1(S)Y2(t)f(s,y(s))\ dS)

t
dt <
00.
Next define F: L~q[O, 1] -- L~q[O, 1] by
Fy(t) = f(t, y(t))

and A: L~q[O, 1]-- L~q[O, 1] by
Ay(t)
11
p(s)q(s)k(t,.s)y(s)ds.
Notice (3.2) and theorem 10.3 implies F is bounded and continuous. A is

completely continuous by theorem 10.4.
Remark. Notice J~ p(t)q(t)J~ p(s)q(s)\k(t,sW dsdt <
00
since
J~ p(t)q(t) J~ p(,~)q(S)\Y1(8)Y2(tW ds dt
J~ p(t)q(t)\Y2(tW J~ p(s )q(sl\yds W ds dt <
s:
00
and also
Consequently N = AF : L~q[O, 1) -+ L~q[O, 1) is completely continuous.

Set
u = {u E L~q[O, 1] : \\u\\ < Mo + I}, C = E = L~q[O, 1).
Then theorem 2.5 implies that N has a fixed point i.e. (1.1) has a solution
y E L~q[0,1]. The fact that y E C(O,1)nC1(0,1) with py' E Ae[O,l)
follows from (3.4) with A = 1. 0
References.
[1]. F.V.Atkinson, Discrete and continuous boundary problems, Academic

[2]. P.L.Chambre, On the solution of the Poisson-Boltzmann equation with
application to the theory of thermal explosions, J. Chem. Phys., 20( 1952),
1795-1797.
[3]. M.M.Chawla and P.N.Shivakumar, On the existence of solutions of a
class of singular nonlinear two point boundary value problems, J. Compo
Appl. Math., 19(1987), 379-:388.
[4]. D.R.Dunninger and J.e.Kurtz, A priori bounds and existence of positive solutions for singular nonlinear boundary value problems, 8IAM J.
Math Anal., 17( 1986), 595-609.
[5]. M.A. El Gebeily, A. Boumenir and A.B.M. Elgindi, Existence and
uniqueness of solutions of a class of two-point singular nonlinear boundary
value problems, J. Compo Appl. Math., 46(1993), 345-355.
[6]. A.Fonda and J .Mawhin, Quadratic forms, weighted eigenfunctions and
equations, Proc. Royal 80c. Edinburgh, 112A( 1989), 14,5-lfi3.
[7]. A.Granas, R.B.Guenther and J.W.Lee, Some general existence principles in the Caratheodory theory of nonlinear differential systems, J. Math.
Pures Appl., 70(1991), 153-196.
[8]. R.Iannacci and M.N .Nkashama, Nonlinear two point boundary value
problems at resonance without Landesman-Lazer conditions, Proc. Amer.
Math. 80c., 106(1989), 94:3-952.
[9]. J .B.Keller, Electrohydrodynamics I. The equilibrium of a charged gas
in a container, J. Rat. Mech. Anal., 5(19.57), 71.5-724.
[10]. M.A.Krasnoselskii, Topological methods in the theory of nonlinear
integral equations, MacMillan Co., New York, 1964.
[11]. J.Mawhin, J.R.Ward and M.Willem, Necessary and sufficient conditions for the solvability of a nonlinear two point boundary value problem,
Proc. Amer. Math. Soc., 93(1985), 667-674.
[12]. J .Mawhin and W.Omano, Two point boundary value problems for
nonlinear perturbations of some singular linear differential equations at
resonance, Gomm. Math. Univ. Carolinae, 30(1989), 5:17-550.
[13]. J.W.Moolley, Numerical schemes for degenerate boundary value problems, J. Phys. A: Math. GeIL, 26( 199:3), 41:~-421.
114
[14]. M.A.Naimark, Linear differential operators Part II, Ungar Publ. Co.,
London, 1968.
[15J. D.O'Regan, Existence principles for second order nonresonant boundary value problems, J. Appl. Math. Stoch. Anal., 7( 1994), 487-507.
[16J. D.O'Regan, Existence theory for nonresonant singular boundary value
problems, Proc. Edinburgh Math. Soc., 38(1995), 431-447.
[17J. D.Powers, Boundary value problems, Harcourt Brace Jovanovich, San
Diego, 1987.
[18J. L.Sanchez, Positive solutions for a class of semilinear two point boundary value problems, Bull. Aust. Math. Soc., 45( 1992), 439-45l.
[19J. I.Stakgold, Greens functions and boundary value problems, John Wiley and Sons, New York, 1979.
11.5
Chapter 11. Nonresonance problems in the limit circle case.

11.1. Introduction.
This chapter presents existence results for thp "nonresonant" singular
boundary value problem
(1.1 )
PhJ(p(t)y'(t)'
+ J-Lq(t)y(t) = I(t,y(t))
limt-+o+ p(t)y'(t)
= y(l) = 0
a.e. on [0,1]
where Am-l ::; J-L ::; Am on [0, 1] (or a more general condition described in
section 11.2) with Am-l < J-L < Am on a subset of [0,1] of positive measure;
here Am, m = 0,1, ... is the (m + 1)st eigenvalue (described in more detail
later) of
( 1.2)
Lu = Au a.e. on [O,IJ
limt ..... o+ p(t)u'(t) = u(l)
where Lu = - .1...(pu')'.
pq
Remark. In the above A-I
= -00.
Throughout this chapter p E C[O, 1] n CI(O, 1) together with p >

on
(0,1); also q is measurable with q> a.e. on [0,1] and f~ p(x)q(x)dx <
00. We do not assume f~ ~ < 00. Thus one of the endpoints t = 1 in
may be singular of limit circle
(1.2) will be regular and the other t =
type. Most of the books and papers on nonresonant nonlinear problems
of the form (1.1) discuss only the case when f~ pC:) < 00 i.e. when both
endpoint are regular. However in this chapter we present results even when
f~ pC:) = 00. The technique we use was initiated in the early 1980's by
Mawhin and Ward [11-12J.
It is worth remarking that other types of boundary data could also be
considered. However, problem (1.2) is the most difficult to discuss (in fact it
is also the "most singular" problem that arises. naturally in applications).
As a result we will only discuss briefly other boundary conditions (see
theorem 11.5).
As in chapter 10 let w be a weight function. L![O, l] denotes the space
of functions u such that f~ w(t)ju(t)j2 dt < 00; also for U,11 E L![O,lJ
I
-define (u, v) = fo w(t)u(t)v(t) dt. Let AC[O,lJ be the space of functions
which are absolutely continuous on [0,1]. We now gather together some
facts [1,3,16,17] on the singular
eigenvalue
problem (1.2). Assume
.
.
( 1.3)
P E C[O, 1]
n CI(O, 1) with p > on (0,1)

116

q E L![O, 1J with q
(1.4 )
> a.e. on [O,lJ
and
e rs
io p(8) io
_1_ (
(1.5 )
p( x )q( x ) dX) 2" ds
<
00
are satisfied.
Remarks. (i). Notice (1.5) implies
{I
io
p(t)q(t)
1
1
PC:)
dt <
00;
see chapter 10.

is a singular point in the limit circle case [1,17]. To see
(ii). Now t =
this we apply Weyl's theorem so all we need show is that all solutions of
Ly = 0 a.e. on [0, 1] lie in L~q[O, 1]. Of course any solution of Ly = 0 is
of the form
y(t) = A
+ B (I
it
d(8)
p s
when' A and B are constants so (1.5) together with remark (i) implies
that we have the limit circle case.
Let
1
')' E Lpq
2 [ 0,1 ]
wEe [
0, ]
1 : w,pw I E AC[O, 1] with -(pw
pq
and lim p(tJW'(t) = w( 1) = o}.
D(L )
1---'>0+
Then L has a countable number [16,17] of real eigenvalues Ai with corresponding eigenfunctions 1./Ji E D( L). The eigenfunctions 'tjJi may be chosen
so that they form all orthonormal set and we may also arrange the eigenvalues so that
Remark. Note Ao
> 0.
III addition the set of eigenfunctions 'I/J; form a basis for L~q[O, 1] and
if h E L~q[O, 1] then II, has a fourier series representation and h satisfies
Parseval's equality i.e.
00
= 2)h,7/\)1./Ji
;=0
and
e pqlhl
in
117
~
dt = L....1(h,
'11\)1 2 .
i==O
As in chapter 10 we say f: [0, 1] X R --+ R is a Caratheodory function if

(i). t 1-+ f(t, y) is measurable for all y E Rand
(ii). y 1-+ f(t,y) is continuous for a.e. t E [0,1].
We also say (see definition 3.2) f : [0,1] X R --+ R is a L 1-Caratheodory
function if
(i). f: [0,1] X R --+ R is a Caratheodory function and
(ii). for any r > there exists hr E Ll[O, 1] such that If(t, y)1 :S hr(t) for
a.e. t E [0,1] and for all Iyl :S r.
In chapter 10 we established the following existence principle.

Theorem 11.1. Let f : [0,1]
X R --+ R be a Caratheodory function and

suppose (1.3), (1.4) and (1.5) hold. Also assume JL is not an eigenvalue of
(1.2) and
f(t,y(t))
2
2 [
]
q(t)
E Lpq[O, 1] whenever y E Lpq 0,1 .
(1.6
)
Now suppose there is a constant M o, independent of A, with

IIYllpq
(1 p(t)q(t)ly(t)12 dt)
1
2"
f:
Mo
for any solution y (here y E L~q[O,I] with y E C(O, 1] n C 1 (0,1) and

py' E AC[O, 1]) to
(1. 7),\
~(py')' + JLqy = Af(t,y) a.e. on [0,1]

limt-+o+ p(t)y'(t) = y(l) =
for each A E (0, 1). Then (1.1) has at least one solution.
Remark. If ~ E L 1 [0, 1] then in fact our solution to (1.1) will lie in QO,l];
see [14, chapter 3].
Finally in section 11.1 we state the following well known compactness criteria [3,8].
Theorem 11.2. (Riesz compactness criteria)
Let M ~ L~q[O, 1]. If
(i).
(ii).
to u
then
M is bounded in L;q[O,l] and

J~ p(t)q(t)lu(t + r) - u(t)12 dt -+
EM
as
M is relatively compact in L;q[O,I].

118
--+
0, uniformly with respect
Remark. Here it is agreed thatu E M is extended to an interval (c, d) :J
[0,1) by letting u(t)
outside [0,1).

In this section we use theorem 11.1 to establish some existence results for
boundary value problems of the form (1.1).
Theorem 11.3. Let f : [0,1) x R -'> R be a Carathe.odoly function
with (1.:3),(1.4),(1.5) and (1.6) holding. In addition assume f has the
decomposition
f(t, u)
where g1,g2,h: [0,1) x R
(2.1 )
ug2(t,U):2:
= g1{t, u)u + g2(t, u) + h(t, u)

-'>
Rare Camtheodory functions and

for a.e. t E [0,1] and u E R
there exist T1 E C[O, 1) with Tl(t)q(t) ::; g1(t,U)::;

for
a.e. t E [0,1] and u E R; here Jk - Tdt) ::; >'0
{
on [0,1) with Jk - Tdt) < >'0 on a subset
of [0, 1) of positive measure
(2.2)
(2.:3)
and
I
(2.4)
f~ p(t)</JI(t) (J/ ~):2

f~
p(t)z(t)
(J/ p7:))
dt
<
.:r.!.
2
dt
and
00
<
00
are satisfied. Then (1.1) has at least one solution.
PROOF: Let y be any solution to (1.7h. Multiply (1.7h by -py and

integrate from to 1 to obtain
la
[P(y'? - flpqy2] dt
= ->.la 1 pyZgdt, y) dt - >.la 1 pyg2(t, y) dt
- >.1a 1pyh(t,y)dt.
This together with (2.1) and (2.2) imply
119
Let
R(y) :::::
101 (p(y')2 -
We claim that there exists
pq[fL - Tl(t)]y2) dt.
> 0 with
Notice first from (2.2) that

(
fl
00
R(y) ~ Jo [P(y')2 - >'Opqy2] dt ::::: t;(>'i - >'o)cf Jo pq'IjJ; dt ~ 0
(2.7)
where
00
y:::::
Also if R(y) ::::: 0 then
Ci
CiVJi and Ci::::: (y, VJi}.

i=O
::::: 0 for i = 1,2, .... Thus y
= coVJo
and so
0::::: R{y)
Now (2.2) implies Co = 0 and so y ::::: o.

If (2.6) is not true then there exists a sequence {Yn} (here Yn E L~q[O, 1]
with Yn E C(O, 1] n CleO, 1) and PY:, E AC[O, 1]) with
2
IIYnllpq
(2.8)
+ llYn, 112p =
0 as n ---.
00.
and
(2.9)
R(Yn)
--->
The Riesz compactness criteria together with a standard result in functional analysis (if E is a reflexive Banach space then any norm bounded
sequence in E has a weakly convergent subsequence) implies that there is
a subsequence S of integers with
(2.10) Yn
here
--->
Y in L;q[O,l] and y~ ~ Y' in L;[0,1] as n
denotes weak convergence.
120
-+ 00
in Sj
Remarks. (i). Note {Yn} is bounded in L~q[O, 1] because of (2.8). Also for
r > 0, Holder's inequality and (2.8) yields
ft p(t)q(t)IYn(t + 7') - Yn(t)1 2 dt
./0
<
_
r1pq11
./0
f1 pq
./0
o as
11
t
d8 dt
p( s)
t+r
ds dt
p(s)
7' -- 0+
by the Lebesgue dominated convergence theorem and (1.5). Then theorem

11.2 implies that bn} is relatively compact in L~q[O, 1].
(ii). It is worth noticing that if ~ E Ll[O, 1] then we may apply the Arzela
Ascoli theorem and (2.10) becomes
(2.10)"" Yn
---4
Y in C(0,1] and Y;, -- Y' in L~[O\ 1] as n
--> 00
in S.
Next a standard result in functional analysis [18] yields

(2.11 )
11 p[y']2 dt ::; lim inf 11 p[y~]2 dt.
Now (2.9), (2.10), (2.11) and the fact that lim inf[sn
lim inf tn for sequences 8 n and tn implies
(2.12)
+ tn ]
2: lim inf Sn
R(y) ::; 0,
since liminff~pq[J.L - Tl(t)]Y;'dt = f~pq[J.L - Tl(t)]y 2 dt. Also (2.7) and

(2.12) imply R(y) = 0 which implies y = o. However
as n --> 00 in S, which is impossible.

Consequently (2.6) is true. This together with (2.f implies
121
Next notice since y( 1)
=0
we have from Holder's inequality that
Thus we have
(2.15)
and
(2.16)
where
Kt =
and
K2
iot
p(t)<Pt(t)
(it pC:)
de;) ~ dt
t
t P(t)<P2(t) (it pesdS)):t.:}!- dt.
io
Put (2.15) and (2.16) into (2.13) to obtain

(2.17)
(lIyll;q + lIy'II;) ~ Ktlly'llp + J(2I1y'lI;+t.
Thus there exists a constant M, independent of A, with

(2.18)
Now (2.18) together with (2.14) yields
(2.19)
J~ pqy2 dt ~ J~ p[y'J2 dt J~ p(t)q(t) J/ ~ dt
~ M J~ pq J/ ~ dt == Mo
Consequently (2.19) together with theorem 11.1 guaranatees that (1.1) has
a solution. 0
Remark. Notice Tt(t)q(t)
9t(t,U)
0 could be replaced by
where 0 ~ b < E is a constant (here E > 0 is as in (2.6 and the result

of theorem 11.3 is again true. A similar type of remark applies also to
theorems 11.4, 11.5 and 11.6.
122
Theorem 11.4. Let f : [0,1] x R -+ R be a Caratheodory function

with (1.3),(1.4),(L5) and (1.6) holding. In addition assume f has the
decomposition f(t,u)=g}(t,u)u+h(t,u) where g},h:[O,l]xR-+R are
Caratheodory functions and
there exist T}, T2 E e[O, 1] with
T}(t)q(t) -::; g}(t,'u) -::; T2(t)q(t)
for a.e. t E [0,1] and u E R; here fL - TI(t) -::; Am
on [0,1] and fL - T2(t) 2: Am_Ion [0,1] together with
fL - T}(t) < Am on a subset of [0,1] of positive measure and
fL - T2(t) > Am_IOn a subset of [0,1] of positive measure.
Also T} -::; 0 on [0,1] and T2 2: 0 on [0,1]
(2.20)
holds. Also assume (2.3) and (2.4) are satisfied. Then (1.1) has at least
one solution.
PROOF: Let y be any solution to (1.7h. First notice that L~q[O, 1] =
n ffi n.L where n = span {'I/)O, .... , 'I/)m-l }, here 1/)i are as described in
section 11.1. Let
y
= u +v
m-l
with u
=L
00
ci'l/)i,
i=O
Remarks. (i). Notice u E

(ii). u = 0 if m = O.
=L
ci'I/Ji and Ci
= (Y,1/Ji)'
i=rn
and v E
n.L.
Multiply (1.7h by p( u - v) and integrate from 0 to 1 to obtain
1
-la (v-u)[(PY')'+JLPqy]dt
Ala 1p(U-V)Y9l(t,Y)dt
10 1 p(u -
v)h(t,y)dt
and so integration by parts yield
fo} [PC V')2

(2.21)
JLpqv 2
+ AP1)2 g1 (t, y)] dt
- f~[p(u')2 - fLpqu 2 + Apu 2g1(t,y)]dt

S f~ plv - u)llh(t, y)1 dt.
Now
Also
123
Put these into (2.21) to obtain

J~[P(V')2 - pq(11 - T1 )v 2] dt - J~[p(u')2 - pq(l1- T2)U 2 ] dt
~ J~ plv - u)llh(t, y)1 dt.
(2.22)
Let
R*(y)
11
- 11
=
(p(v'? - pq[l1- T1(t)]V 2 ) dt

(p( U')2 - pq[11 - T2(t)]U 2) dt.
We claim that there exists { > 0 with

(2.23)
Notice first from (2.20) that
R*(y)
>
11
L.,,(A; '"
,=m
00
> O.
11
",-1
211
(p(v')2 - AmpqV 2) dt -
(p(u'? - Am _lPqu 2 ) dt
pq'lj'i2 dt - 'L.,,(Ai
- Am-dCi2
"
A"')ci
11
0
,=0
pq'i dt
Also if R*(y) = 0 then Ci = 0 for i = O,l, ... m- 2,m+ 1,m+2, ... and
so y = C",-lVJ",-l + Cm 'Ij'm. Thus since u = Cm-lV'm-l and v = cm'm we
have
Hence (2.20) implies Cm = Cm-l = 0 and so y = o.

If (2.23) is not true then there exists a sequence {Yn} with
(2.24)
124
and
R*(Yn) -.... 0 as n -....
(2.25)
Remark. Note Yn =
Un
+ Vn
where
Un
00.
E 0 and
E 01..
Vn
As in theorem 11.3 there is a su bsequence S of integers with
(2.26) Yn -- Y in L;q[O, 1] and y~ -.... y' in L;[0, 1] as n --
00
in S.
Also since strong and weak convergence are the same in finite dimensional
spaces we have
(2.27)
U~
---;.
u' in L;[0,1] as n --
00
in S.
Notice as well
101 p[v']2 dt s: lim inf 101 p[v:.l 2 dt.

Now (2.25), (2.26), (2.27) and (2.28) imply R*(y) s: O. This implies y = O.
(2.28)
However
IIYnll;q + IIY~II; =
R*(Yn)
+ 101pq[v~ + u!] dt + 2101 p[U~]2 dt
+ 101 ([11- T1(t)]pqv; 0 as n -....
--
00
(jL - T2(t)]pqu;) dt
in S,
which is impossible.
Remark. For i
i= j
then J~ pq'ljJi'ljJj dt
101 p1/J:'ljJj dt = -
11
=0
(p'ljJD''ljJj dt
implies
= Ai 101 pq'ljJi'ljJj dt = O.
Consequently (2.23) is true. This together with (2.22) implies
Next notice since yO)

inequality yields
lu(t) - v(t)1
=0
= I Jtfi [v' -
we have u( 1)
u'] dsl
= v( 1)
s: ( Joe p[v' 125
= 0 and so Holder's
u'fl dt
)t (eJt pes)
dS) t
and so
(2.30)
lu(t) _ v(t)i ::;
!iY'lip
(1
d~)!
p(S)
_L
This together with (2.29) yields

(2.31 )
where
'11
Ii 1 ~
and
](2
p( t )1>1 (t)
(1
ds )
-(p S)
~ dt
r p(t)</>2(t) ( itr ~)
4! dt.
p(s)
1
io
Now (2.31) implies that there exists a constant M, independent of A, with

(2.32)
Notice also that (2.30) implies
(2.33)
J~ pqy2 dt = J~ pq[v - uj2 dt ::; Ily'll~ J~ p(t)q(t) J/ p~~) dt

::; M J01pq J/ p~~) dt == Mo.
The result now follows from theorem 11.1.
We now use the ideas of this chapter to establish results for a more
general boundary value problem if ~ E L1 [0.1] i.e if both endpoints of
(1.2) are regular. Consider
(2.:H)
~(PY')'+JlqY~f(t,y,pyl) a.e. on [0,1]

{ y(O) = y(I) = 0 or y(O) = limt __ 1-p(t)y'(t)
or limt-+o+ p(t)y'(t) ~ y(l) = o.
=0
Remark. It is worth noticing that results could also be obtained for Sturm
Liouville, Neumann or periodic boundary data.
For the remainder of this chapter assume (1.3), (1.4) and
(2.:3.5)
1 -<00
1
ds
o p(s)
126
hold.
Theorem 11.5. Let pf: [0,1] X R2 - R Ix an L 1-Caratheodory function with (1.3), (1.4) and (2.35) holding. In addition assume f has the
decomposition
f(t, u, w) = 91(t, u, w)u
(2.36)
(2.37)
+ 92(t, u. w) + h(t, u, w)
Ih(t,u,w)1 :S: (Pt(t) + <P2(t)lul'Y + 4>3(t)lwl(i for

a.e. tE[O,I] withO:S:"O<1
U92(t,u,W) 2::
for a.e. l E [0,1] and u,w E R
{ with 192(t, 'U, w)1 :S: <P4(t) + <Ps(t)lwl" fOl' a.e.
t E [0, 1] and u in a bounded set; here :S: K, < 2
(2.38)
and
(2.39)
there exist Tl E L~q[O,I] with T,l(t)Q(t):S: 91(t,u,W):S:

for
a.Co t E [0,1] and u,w E R; here JL - Tl(t):S: Ao
{
a.e. on [0.1] with JL - Tl(t) < Ao on a
subset of [0,1] of positive meaSU1"(
are satisfied. Then (2.34) has at least one solution.
PROOF: Let y be any solution (here y E C[O, 1] n C 1(O,I) with py' E

AC[O, 1]) to
(2.40h
~(py')' + ILQy = Af(t, y,py') a.e. on [0,1]

{ yeO) = y(l) = or yeO) = lim t ->l- p(t)y'(t)
or limt-+o+p(t)y'(t) = yell =
for
< A < 1. From [14, chapter 3] we must show that there exists a
constant M, independent of A, with
(2.41)
max {sup ly(t)l, sup Ip(t)y'(t)1 } :S: M.

[0.1]
(0,1)
127
Let R( y) be as in theorem 11.3. Essentially the same reasoning as in

theorem 11.3 (except (2.10)* replaces (2.10)) establishes the existence of a
constant f with
(2.42)
and so
(2.43)
(1IYII~q + IIY'II~) :S J~ pIlyl dt + J~ pQ>2Iyl'Y+l dt
+ J~ pQ>3lyllpy'I B dt.
Also Holder's inequality implies
ly(t)l:S Ily'lip (Jo
(2.44 )
ds ) t
Pc:)
for t E [0, 1]. Consequently

(2.45)
10
pQ>Ilyl dt:S
sup ly(t)1
[0,1]
r PI dt :S NlllY'lip
10
and
JOI pQ>3lyllpy'I B dt
(2.47)
:S sup[o,I]ly(t)llly'll! (JoI[p(t)]~[Q>3(t)]2:'0 dt) 2"2

:S N3 1Iy'II!+I
for some constants N I , N2 and N 3 . Put (2.45), (2.46) and (2.47) into (2.43)
and we may deduce that there is a constant Mo with
(2.48)
This together with (2.44) implies that there is a constant MI with

(2.49)
sup ly(t)1
[0,1]
12R
:S
MI.
Now (2.49) together with the differential equation yields
fa1
l(py')'1 dt < 1/LIM1 101 pq dt + M1 10 1 pqlT11 dt + 101 P</>4 dt
+ 101 pc/>5Ipy'l" dt + 10 1 Pc/>l dt

+
Mi fa1 Pc/>2 dt
+ fa1 p3lpy'1 9 dt
and now apply Holder's inequality so we obtain constants Eo, E1 and E2

with
(2.50)
f~
l(py')'1 dt
::; Eo
+ E111y'll~ + E2 11y'll:
K
::; Eo + E1M02
+ E2MJ
Also there exists t1 E [0,1] with limt_tl p(t)y'(t)

with (2.50) yields
Ip(t)y'(t)1 ::; lit

tl
l(py')'1 dsl
== M2
and this together
::; M2
so
(2.51)
sup Ip(t)y'(t)1 ::; M 2.
(0,1)
The result follows from (2.49) and (2.51).
Remark. It is possible to let

t E [0,1].
/'i,
= :2
in (2.:H) if p2(t)</>S(t) is bounded for
Theorem 11.6. Let pf: [0,1] X R2 - R be an L 1-Caratheodoryfunction with (1.3),(1.4) and (2.35) holding. In addition assume f has the
decomposition
f(t,u,w) = 91(t,u,W)u + h(t,u,w)
where P9bph : [0,1]
R2 - R are L 1 -CarathCodory functions. Suppose
(2.36) is satisfied with

(2.52)
{I
~
2
c/>b2 E L![O, 1] with Jo [P(t)]2-9[.1(t)]2-9 dt <
129
00
and
(2.,53)
there exist T},T2 E L~g[O,l] with

Tt{t)q(t) ~ gt(t, u, w) ~ T2(t)q(t)
for a.e. t E [0,1] and u, tl7 E R; here J1, - TI(t) ~ Am
a.e. on [0,1] and J1, - T2(t) 2: Am-I a.e. on [0,1]
together with J1, - Tt (t) < Am on a subset of [0,1]
of positive nu;aSU7e and J1,-T2(t) > Am-I
on a subset of [0,1] of positive measure. Also
TI :S
on [0,1] and T2 2: on [0,1]
holding. Then (2.34) ha,~ at least one solution.

PROOF: Let y be a solution to (2.40h for 0< A < 1 and let u, v, R*(y)
be as in theorem 11.4. Essentially the same reasoning as in theorem 11.4
establishes the existence of a constant f with
(2.54)
(: (1IYII~g + IIY'II~) :S J~ p<pIlu - vi dt + J~ p<p2lu + J~ p<P3lu -
vllyl"! dt
vllpy'l{/ dt.
Notice also Holder's inequality implies
Iv(t) - u(t)l:S
( JofI p(v' -
u')2 dt
) t (fI
d~) t
Jo pC:)
and so
(2.55)
Iv(t) - u(t}l :S Ily'lip
d~) t
(1o -')
1
p(,~
for t E [0,1]. Essentially the same reasoning as in theorem 11.5 establishes

the result. 0
130
References.
[1]. F.V.Atkinson, Discrete and continuous boundary problems, Academic

[2]. M.M.Chawla and P.N.Shivakumar, On the existence of solutions of a
class of singular nonlinear two point boundary value problems, J. Compo
Appl. Math., 19( 1987), 379-388.
[3]. N .Dunford and J.T .Schwartz. Linear operators, Interscience Publ. Inc.,
Wiley, New York, 1958.
[4]. M.A. EI Gebeily, A. Boumenir and A.B.M. Elgindi, Existence and
uniqueness of solutions of a class of two-point singular nonlinear boundary
value problems, J. Compo Appl. Math., 46(199:3), 345-3.'55.
[5]. W.N.Everitt, M.K.Kwong and A.Zettl. Oscillations of eigenfunctions
of weighted regular Sturm Liouville problems, J.London Math. Soc.,
27(1983), 106-120.
equations, Proc. Royal Soc. Edinburgh, 112A( 1989), 14.5-153.
[7]. P.Habets and G.Metzen, Existence of periodic solutions of Duffing
equations, Jour. Diff. Eq., 78(1989), 1-32.
[8]. L.V.Kantorovich and G.P.Akilov, Functional analysis, Pergamon Press,
Oxford, 1982.
[9]. J .Mawhin and W.Omano, Two point boundary value problems for
nonlinear perturbations of some singular linear differential equations at
resonance, Comm. Math. Univ. Carolinae, 30(1989), 537-550.
[10J. J.Mawhin and W.Omano, Bounded nonlinear perturbations of singular boundary value problems at resonance Bull. Cl. Sci. Acad. Roy.
Belgique, 6(1991), 343-3.56.
[11J. J.Mawhin and J.R.Ward, Nonuniform nonresonance conditions at the

first two eigenvalues for periodic solutions of forced Lienard and Duffing
equations, Rocky M. J. Math., 112( 1982),643-654.
[12J. J .Mawhin and J .R.Ward, Periodic solutions of some forced Lienard
differential equations at resonance, Arch. Math., 41( 1983), 3:H-:3.51.
[13]. P.Omari and F.Zanolin, Nonresonance conditions on the potential for
a second order periodic boundary value problem, Proc. Amer. Math. Soc.,
117(1993), 125-135.
[14]. D.O'Regan, Theory of singular boundary value problems, World Scientific Press, Singapore, 1994.
131
[15]. D.O'Regan, Existence theory for nonresonant singular boundary value

problems, Proc. Edinburgh Maths. Soc., 38(1995),4:31-447.
[16]. D.O'Regan, Nonresonant nonlinear singular problems in the limit
circle case, Jour. Math. Anal. Appl., 197( 1996), 708-725.
[17]. l.Stakgold, Greens functions and boundary value problems, John Wiley and Sons, New York, 1979.
Chapter 12. Resonance problems in the limit circle case.

12.1. Introduction.
In this chapter we present existence results for the "resonant" singular

boundary value problem
(1.1 )
~lt) (py')' + A,;,q(t)y + yg(t, y) = h(t, y) + vet)

{ bmt-+o+
p(t)y (t) = y{l) = 0.
Here Am, m
of
is the (m + 1)st eigenvalue (described in chapter 11)
= 0,1, ...
{
( 1.2)
Lu = Au a.e. on [0,1]
= u(l) =
lilllt-+o+ p(t)u'(t)
where Lu = - l(pu')'.
pq
Throughout chapter 12,
( 1.3)
(1.4)
PE
a.e. on [0,1]
qo, 1] n (,1(0,1)
q E L;[O, 1] with q
with p>
>
on (0,1)
a.e. on [0,1]
and
(1.5 )
fl _1_
Jo
pes)
(fS
P(x)q(X)dx)t ds < oc
Jo
hold. Also f: [0,1] X R ~ R is a Camtheodory function i.e.

(i). t f---> f(t, y) is measurable for all Y E Rand
(ii). y ........ f(t, y) is continuous for a.e. t E [0,1].
In addition we assume the nonlinearity

( 1.6)
f(t,y(t))
E
q(t )
2 [
Lpq
0,1
satisfies
whenever
2 [
Lpq
0,1 .
In this chapter we use a well known technique, initiated by Mawhin and

Ward [8] in the early 1980's and extended and refined by Iannacci and
Nkashallla [5] in the late 1980's, to establish some Landeslllan-Lazer [5]
type existence results for the singular resonant boundary value problem
(1.1 ).


Theorem 11.1 will now be used to prove a variety of existence results
for the boundary value problem (1.1).
Theorem 12.1. Let f(t, y) = h(t, y) + vet) - yg(t, y) and a,~sume f :

[0,1] X R ~ R i,~ a Camtheodory function with (1.3),(1.4),(1.5) and (1.6)
holding. In addition assume

(2.1)
Ih(t,y)1 ::; <i>l(t)
+ <i>2(t)lyl'Y
for a.e t E [0,1] with 0::; 'Y < 1
(2.2)
io[1 p(t)lv(t)1 (1t
(2.3)
1 d fl ) t
p(~)
dt <
00
and
(2.4)
there exists l' E C[O, 1] with 0::; get, u) ::; T(t)q(t) for a.e
{ t E [0,1] and U E R; he1'e T(t) ::; Am+l - Am for t E [0,1]
with 1'( t) < Am+l - Am on a subset of [0,1]
of positive measure
are satisfied.
(i). Suppose there exists a constant k > 'Y of the form 1 > k
f3 is odd and Q is even, with
(2.,5)
< J
1+
= ~,
where
p(t)[A'ljJm(t)]k+lliminfx-+oo (x 1- k g(t,x) dt
+ J1- p(t)[A'Ij!m(t)]k+lIimsupx-+_oo (xl-kg(t,x) dt
for all constants A of 0; here /+ = {t E [0,1] : A'Ij!m(t) > O} and /{t : A 1Pm (t) < O}. Then (1.1) has at least one solution.
(ii). If 'Y = 0 with
(2.6)
A J~ p(t)v(t)1Pm{t) dt < A Jl+

+A Jl+A Jl+
+A Jl-
p(t)'1,bm(t) lim infx->oo[xg(t, x)] dt

p(t)'Ij!m(t) lim suPx-+_oo[xg(t,x)] dt
pet )'Ij!m( t) lim infx-+oo[ -h(t, x)] dt
p( t )V!m (t) lim sUPx->-oo[
t, x)] dt
134
-he
for all constants A =J O. Then (1.1) has at least one solution.

Remark. As in chapter 10 and 11, by a solu tion to (1.1) we mean a function
y E L~q[O, IJ, with y E C(O, IJ n C 1 (0, 1), py' E AC[O, 1], which satisfies the
differential equation in (1.1) almost everywhere and the stated boundary
conditions.
I
PROOF: For notational purposes let
IIUllpq = (Id pqlul 2 dt) '2
and
Ilu'llp =
(I~ plu'I2 dt) '2 for appropriate functions u.

Choose 11 such that Am < 11 < Am+1 and look at the boundary value
problem
{~(py')' + I1qy = b[h(t,y) + vii) - yg(t,y)
(27)
limt_o+ p(t)y'(t)
= y(l) = 0
+ (11-
Am)qyJ a.e .
for 15 E (0,1). Let y be any solution to (2.7)0. We will show that there
exists a constant M o, independent of A, with IIYllpq :::; Mo so of course
existence of a solution to (1.1) will be guaranteed from theorem 11.1.
Notice L~q[O,IJ = nmn.l where n = span {V)o, ..... ''l/Jm}; here 'I/)i is
as described in section 11.1. Let
10
ex>
n~
ci1/)i,
i=m+ 1
and fJ
= 10 + Uo
Remark. Note fJ
where
=y-
=L
ci'l/)i,
Uo
i=O
Ci
'U1
1n-l
ci1/J;,
= (y,1/)i)'
and y
U1
= 10 + u.
Also if m
=0
Multiply the differential equation in (2.7)6 by p( u from 0 to 1 to obtain

(2.8)
= cm1/)m
i=O
we have Uo
10)
and integrate
Jd (p[w']2 - [Amq + (1- 8)(11- Am)q + bg(t,y)Jp1V2) dt

- Jd (p[u']2 - [A",q + (1 - 15)(11 - Am)q + bg(t, y)Jpu 2 ) dt
= 15 Jd p[h(t,y) + v(t)J(u - w)dt.
Notice
(2.9)
Jd (p[1l7']2 - [Amq + (1 - 15)(11 - Am)q + 8g(t, y)JP1l7 2) dt

~ Jd (p[1l7']2 - [Am + (tL - Am) + r(t)Jpqw 2) dt
= Jo1 (p[w']2 - [An' + r(t)Jpqw 2) dt - (11 - Am) Jd pq1l7 2 dt.
Also
(2.10)
Jd (p[u'J2 - [Amq + (1 - fJ)(l1- Am)q + bg(t,y))pu 2) dt

(1(['J2
:::; JO
p u - AmpqU 2) dt
= Jd (p[u~F - Ampq'ui) dt + Jd (p[u~F - Am Pq u 6) dt
(1(['J2
= JO
P Uo - A",pquo2) dt.
135
= O.
Remark. Note that f~ p'lj<'l/Jj dt = -M(p'lj!i)''lj!j dt = Ai f~ pq'lj!i'ljJj dt.

Put (2.9) and (2.10) into
(2.11)
R(f) - (p - Am)
(V~)
10
to obtain
pqw 2 dt S
101 plh(t, y) + v(t)llu - wi dt
where
We claim that there exists
> 0 with
(2.12)
Notice first however that
1
R(y)
m-I
Jof (p [w ']2 ~
.L
Am+lpqw 2) dt - ~
~ (Ai - Am )Ci2 Jof pq'lj!i2 dt
211 2dt - '~I

L (Ai -
(Ai - Am+ 1 )Ci
1=171+1
pq'~!i
> O.
Also if R(y) = 0 then

C",+1 'l/Jm+l. Thus
Ci
Am )Ci
Joofl pq'l/Ji'2 dt
+ 2, ...
and so y
'2
1=0
= 0 for i = O, ... ,m - I,m
0= R(y)
so (2.4) implies
Cm+1
= 0 i.e. f)
= O.
If (2.12) is not true then there exists a sequence {Yn} = {w n + UO,n}

(here Yn E L~q[O,I] with Yn E C(O,1]nC 1 (O,1) and py:, E AC[O,I]) with
(2.13)
and
(2.14 )
R(y,,) ---. 0 as n -
00.
The Riesz compactness criteria together with a standard result in functional analysis (if E is a reflexive Banach space then any norm bounded
136
sequence in E has a weakly convergent subsequence) implies that there is

a subsequence S of integers with
(2.15)
and
(2.16)
as n
--7
UO,n
00
III
Uo
---'0
in S; here -
2[
Lpq
0,]
1 and
~ U ,o
in Lp2[]
0, 1
denotes weak convergence,
Remark. Note {wn } is bounded in

a
'uO,n
L~q[O,
> 0, Holder's inequality and (2.1:3) yields
1] because of (2.13). Also for
<
<
by the Lebesgue dominated convergence theorem and (1.5). Then theorem

11.2 implies {w n } is relatively compact in L~q[0.1].
Also since weak and strong convergence are the same in finite dimensional spaces we have
(2.17)
U~,n
--7
u~ in L;[0, 1] as n ----
00
in S.
Next a standard result in functional analysis [14] yields
11 plw'12 dt ::;
(2.18)
lim inf
11 p1W~112
dt.
Now (2.14), (2.15), (2.16), (2.17), (2.1R) and the fact that
lim inf[sn
for sequences
Sn
and
tn,
+ t n ] 2': lim inf .5" + lim inf tn
yields (with fJ
= W + uo),
since liminf J~ pq[Am +r(t)Jw;' dt

and so fj = O. However
= J~ pq[Am +r(t)Jw 2 dt.
Ilwnll~q + Ilw~lI~ + IluO,nll~q + Ilu~.nll~ =
as n
--+ 00
Hence R(f})
=0
+ Iol[2P(U~,n)2 + pqu~,n
Ampqu~ n + pqw~ + (Am + r)pqw~J dt
--+
R(f})
in S, which is impossible. Hence (2.12) holds for some ( > O.
Remark. Note ( in (2.12) is independent of fl.
Put (2.13) into (2.12) to obtain

(2.19)
(1Iwll;q + Ilw'll; + Iluoll;q + lIu~II;) S
Now choose fL = Am
Am + ~ < Am+!. Thus
(2.20)
+1
Jo
(fL -
plh(t, y)
Am)lIwlI;q
+ 1J(t)II1L - wi dt.
where ( is chosen sufficiently small so that
~ (lIwll;q + IIw'lI~ + II1LolI~q + lIu~II;) s 101 plh(t, y)+v(t)llu-wl dt.
We next perform some analysis on the right hand side of (2.20).

following equalities are needed:
(2.21)
1If}'II; =
(2.22)
(2.23)
IIw' -
1L'II~
IIw'lI~
+ lIu~lI~
= IIw'lI~ + 111L'II~ = lIy'lI~ = 1I:if'lIp + lIuill~

= IIw'lI~ + 111L~II; + 111L~II~
(2.24)
and
(2.25)
IIw -
1L1I~q = IIwll~q
+ 111LII~q
= IIwll~q
+ l11LolI~q + l11LllI~q
Also for tE(0,1) we have
(2.26)
ly(t)1
111
y'(s) dsl
s lIy'lip
138
(1 P~;J t
1
The
and
1
Iw(t) - u(t)1 ~ (I~ pew' - u')2 dS) 2
(2.27)
= Ily'llp (I/
(I/ ~) 2
p'fu) 2 .
Thus we have
where
[1
K1 = Jo P(t)~1(t)
(2.29)
[1
Jo p(t)~2(t)lyl1'lw - ul dt ~
where
K2
[1
= Jo
p(t)(/>2(t)
(tJ pes)
d,~ ) t
dt;
t
K2
(lIw'lI~ + Ilu~lI~ + lIu~II~)
([1 d~) -,;1
Jt
pC:)
dt;
and
(2.30)
10 1 p(t)lv(t)llw - ul dt ~ K3 (lIw'lI~ + lIu~lI~ + Ilu~II~) t
where
K3 =
10
p(t)lv(t)i
(1 p~:J t dt.
1
Put (2.28), (2.29) and (2.30) into (2.20) to obtain
~ (lIwll~q + Ilw'll~ + lIuoll~q + Ilu~II~)

~
(/(1
+K2
Now since 0
+ K 3 ) (1Iw'lI; + Ilu~lI~ + lIu~II~) 2
(lIw'lI~ + lIu~ll~ + lIu~ II~)
~
2
I < 1 there exist constants K4 and K5 with
Consequently
(2.32)
139
We now claim that (2.32) implies that there is a constant Mo >
with
(2.33)
Su ppose the claim if false. Then there is a sequence (Dn) in (0, 1) and
a sequence (Yn) (here Yn E L~q[O, 1] with Yn E (,(0,1] n (,1(0,1) and
PY:, E AC[O, 1]) with for a.e. t E [0, 1],
(PY;,)'
(2.34)
+ Ampqy" + (1
- bn )(Jl - Am )pqYn
+b"py"g(t,y,,)
= b"p[h(t,Yn) + v(t)]
and
(2.35)
From (2.32)' with Yn
= y~! + 'Ul,,,,
we have
(2.36)
Also
(2.~l2)
and (2.:36) yields
( 2.37)
Let
(2.38)
where
(2.39)
r~,
Yn
= ----=----~1
(IIU1,nll~q + Ilu;,,,II~) 2
and
rl,n
Ul,n
------''-------;--1 .
(IIUl,,,II~q + lIu;,nll~) 2
Fix < f < 1. Now (2.:H), the Riesz compactness criteria and the Arzela
Ascoli theorem, implies that there is a subsequence S* of integers with
and
(2.41 )
r~,
-+
in (' [(, 1] as n ~
140
00
1Il
S".
Remarks. (i). Note {r~.} is bounded in L~q[O.l] because of (2.36) and
11-11 2 =
1I!h!II~q
< /(4 + /(5I1u~,nll~+l
rn pq IIUI,nll~q + lIu~,nll~ - IIUI,nll~q + lIuL.II~
Also for a > 0, Holder's inequality and the fact that {r~.} is bounded in
L~q[O, 1] yields
11 p(t)q(t)lr~(t +
::;
a) -
r~.(t)12 dt
11 /.ds (11pq 11 -(ds.) dt - 11 pq 11 -(-)d8 dt )
--4
p[r~. ]2
0 as a
--4
t+a P
<~
0+
by the Lebesgue dominated convergence theorem. Then theorem 11.2 implies {Tn} is relatively compact in L~q[O, 1].
(ii). For fixed 0 < f < 1 notice {r~.} is bounded in C[i,l] since
Ir~.(t)1
1I!h./llp (1/ ~) 2
1!h.(t)1
----'-=--'--'-"----,--1 ::;
(IIUl.nll~q + lIui.nll~) 2
<
Also for
t,s E [i, 1]
(/(4 + /(5I1 ui,nll;+lr (1/ ~) 2

1
(IIUl,nll~q + lIui,nll~) 2
we have
lJ:y~.'(x)dxl
Tr. ( t) - Tr. (s) 1 = ---'--"-"---'---'----'-~I
::;
IIYn'lIp(J:rfxTr
(IIUl,nll~q + lIui,nll~) 2
The Arzela Ascoli theorem implies {r~.} is relatively compact in C[i,l].

(iii). If ~ E Ll[O, 1] we may take f = 0 in (2.41).
Also
IIYnll~q
IIUI,nll~q
+ IIY:.II~
+ lIui,nll~
141
This together with (2.37), (2.40) and (2.41) implies there is a subsequence
S of integers and a constant A :I 0 with
(2.42)
(2.43)
(2.44)
r .. ---- AI/\" in
C[f, 1] as
n ---.
00
in S
and
rl,n ---- A'!j)m in C[f,l] as n
(2.45)
here 0
<f <
----+ 00
in S;
1 is fixed.
Remarks. (i). Note as well that r~, ~ A'!j)~n in L~[O, 1] as n ---- 00 in S

and r~,n ----+ A'Ij)~ in L~[O, 1] as n ....... 00 in S.
(ii). Notice also that (i) and 11\,,(1)::::: 0 implies rl,n(X) ---. A'Ij)m(x) pointwise for x E (0,1].
(iii). If ~ E Ll[0,1] we may take f::::: 0 in (2.44) and (2.45).
Multiply (2.34) by ~ and integrate from 0 to 1 to obtain
s~ Jdh,,,(py~.)' + AmT l,,,pqYn] dt

(2.46)
+ (1-6n~~-A771) Jd pqrt,,,Yn dt = Jd prl,,, v dt

+ Jot P TI, .. h(t, Yn) dt + Jd p rl,n[ -Yn9(t, Yn)] dt.
Also
=0
and so
(2.47)
Thus
(2.48)
142
Case (i). Suppose (2.5) holds
Multiply (2.48) by (1Iul,,,II~q
-k
+ Ilu~,nll~)""2
to obtain
(2.49)
Recall for t E (0,1) that
h,,,(t)1 ~
d~ ) t
(Jto p[r~,,,12 ds ) t (11 pC:)
and so
Ip(t)v(t)rl,n(t)1
~ IIr~.nllpp(t)lv(t)l (11 p~:)) t .
This together with the remark after (2.45) implies that there is a constant
Eo with for t E (0,1]'
(2.50)
Ip(t)v(t)rl,n(t)i ::; Bop(t)iv(t)l
(11 p~:)) t == g*(t).
Now since g* E Ll[O, 1] we may apply the Lebesgue dominated convergence

theorem to deduce
(2.51)
lim inf f~ p v rl,n dt
Remark. In (2.,51) we have n
= lim J~ P'I) rl ,n dt = J~ lim (p v rl,n)

= A J~ pm/)m dt.
---'0
00
in S.
Hence
(2.52)
fl
p'/J rl,n
(11'1l1,nll~q + 11'1l;.nll~) 2
lim iuf In
= O.
lit
Now (2.,52) together with (2.4!)) yields
02: liminf fci

(2.5:3 )
l'Tl,n[-h(t,Yn)]
lit
(II'L! ,,,II~q+llll; ,,,lit,) 2"
+lim iuf J~
l' Tj ,n Yn!l(t,Yn)
1;.
(11"1, nll~><,+II"; ," II?,) 2
dt.
dt
Remark. In (2.5:3) we have n
-> 00
in S.
Next we claim
(2.54)
To see this recall
(2.55)
Of course for t E (0,1) we have
IYn(t)l:S IIYnllp
I
and
and so
11
Ip rl,n h (t,Yn)1
o (1Iul,nll~q + IluLnll~) 2
k
(1
t
Bll1u~,nllp + B2I1u~,nllp (1I?h,/II~ + lIu~,nll~) 2

(lIul,nll~q + Ilu~,nll~)
where
Bl =
and
B2 =
lo
ds )
pis)
p(t)dt)
This together with (2.:32) yields
(2.56)
144
ill.
2
(11 p(sd") ~ dt
r P(t)<P2(t) (1
Jo
Bll1u;,nllp + B21Iu~,nllpl~lI;
(1Iul,nll~q + lIu~,nll~)
dt:s
1-
:s
~)
d(8
P s)
):zt!- dt.
and so (2 ..14) follows from (2.:36) and the fact that k > ,. Hence (2.53)
reduces to
O>liminfj'
-
(2.57)
1+
P1'I,nYng{t,Yn)
dt
(1luI,n II~q+llu; ,nll~) '2
+ lim inf fI-
prl,n Yng(t,Yn)
dt
(IIUI,nll~q+llu; ,nll~) '2
where 1+ = {t E [0,1] : A'ljJm(t) > O} and 1-
= {t:
A'ljJ",(t)
< O}.
Remark. We know from [1,a] that 1/lm has at a finite number of zero's.
We would like to apply Fatou's lemma in (2.57). To justify this we need
to show that there exists PI E Ll[O, 1] with
(2.58)
p r l,nYng(t,y,,)
---"-------'----"---'-'------'-----.,..-k
2:
( .
.c
PIt) lor a. e . t E
[0 , 1] .
(1IUI,nll~q + Ilu~,nll~) 2
To see this notice
P Ul,ny"g(t,y,,)
prI,n Yng(t, Yn)

k
(1Iul,nll~q + Ilu~,nll~) 2
Thus a.e. on [0,1],

(2.59 )
In addition since
for t E (0,1)
we have
I
(K4 + Ksllu;,,,II;+I) p(t)q{t)T{t) (J/ p'fu) 2

(1Iu 1 ,,,II;q + Ilu;,,,II~)
141)
!!l
2
and (2.36) together with the fact that /,: > I implies that there exists a
constant B3 with
(11 d$)
p(i)q(i)T(t)[y~,(iW
< H (i) (i) (t)
!! _
.3P q T
2
(II Ul.n 11 2pg + II U I1 n 112)
p
(2.60)
( )
Combine (2 ..59) and (2.60) and we have that (2 ..58) is true. Apply Patou's
lemma in (2 ..57) to obtain
o~
(2.61)
II+ lim inf (
+ II- lim inf (
Let B = ~ where
= rt
1+
0:
where Ii
U Jt where It
= 1- n [B, 1]
o~
prl,n Yng(t,Yn)
dt
dt.
(lIulnll~q+llu; ,n II~) 2
and
+ Ir lim inf{prl,n
e
+ IJ+ur liminf(
e
= 1+ n [B, IJ
r = Ii U Ji
Ji = 1- n [O,B).
II+ liminf (prl,n (
li)
li)
E {2,3 .... }. Fix B for the moment. Now
Also
(2.62)
P "I,n Yng(t,Yn)
(lIulnll~q+llu; ,nll~) 2
-Yn
!)
PTJ,nYng(t,Yn)
k)
(1IUI,nll~q+llu; ,nll~) 2
~ 00
= 1+ n [0, B).
Then (2.61) becomes
Yn
!)k
(1IuI,nI12+llu;,n))2)2
(1IUl,n))2+llu;,n))2) 2
Remark. In (2.62) we have n
and .It
y,~-kg(t'Yn))
dt
k [_[ _y,JI-kg(t, Yn)]} dt

di.
in S.
Let (t n ) be any sequence. Recall if a sequence ($n) converges to a

positive real number oS then liminf8 n t" = 8liminft n . Similarly if (8,,)
converges to a negative real number oS then lim inf '~ntn = .slim sup in. This
together with (2.4.5) yields
(2.63)
o ~ II: p[klj)m(t)]k+lliminf (y~,-kg(t,yn)) dt

+ II; p[A'lj)m(t)]k+ 1 Iimsup (y,I,-k g(t,Yn)) dt
+ f + - liminf ( prl,nYng(t,Yn) k) dt.
J e UJe
(I IUI,nl 12pq+1 IUII .n 112)2

p
146
Remark. Note (- l)k = ( -1 ) ~ = 1.

Let t E
It.
Notice (2.45) implies there is an integer
nl
with
+00
as n
---+ 00 III
-00
as n
---+
(2.64)
Remark. Notice the right hand side of (2.64) goes to

S.
Let t E
Ii.
There is an integer nz with
(2.65)
Remark. Notice the right hand side of (2.6.1) goes to
S.
00
in
Now (2.64), (2.65) together with (2.63) yields
o 2: JI +
+ JI (2.66)
+ J/+
XI: p( t)[ A 1/1", (t )]k+ lIim infx->oo
dt
XI; p(t)[A1/I ",(t)]k+llim sUPx---+_oo (xl-kg(t, x)) dt

XJ+ liminf (
PTJ,nYng(t,Yn)
k)
dt
k)
dt.
(1Iul,nll~q+llu;,nll~)2
+ JI -
(x l-k g( t, x))
Tl,n
Xr lim inf (
P
Yng(t,Yn)
e
(1Iul,nll~q+llu;,nll~P"
We now want to let ()

(2.58) implies
---+
0+ (i.e.
=~-
(2.67)
Hence Fatou's lemma implies
(2.68)
147
00)
in (2.66). First notice
Pl(t) for a.e. t E [0,1].
Similarly
k)
liminfj Xr lilllinf(
prl.nYng(t,Yn)
(>-+00 /(}
(1IuI.nll~q + lIui.nll~) 2"
(2.69)
dt~O.
Also
p(t)[A'I/\n(t)]k+llimillf (xl-kg(t,x)) ~ 0 for a.e. t E

x-oo
rt
so Fatou's lemma implies

lim inf(>-+oo f/+ 'XIt p( t)[ AWm( t )]kH lim infx-+oo
(2.70)
~ fI+ liminfa-+oo
= f/+
(X/:
(x l - kg( t, x))
dt
p[A'Ij)m]k+lliminfx-+ oo (xl-kg(t,x)) dt
p(t)[Al/\,,(t)]k+lliminfx-+oo (xl-kg(t,x) dt.
Similarly
(2.71)
lim infa-+oo fl-
XI; p[A'ljJmlk+1lim sUPx-+_oo (xl-kg(t, x) dt
~ fI- p(t)[A'ljJm (t)]k+lJimsupx __ oo (xl-kg(t,x)) dt.
Now (2.68),(2.69),(2.70),(2.71), together with thE:' fact that liminf[sn

tnl ~ liminf Sn + liminftn for sequences '~n and t", and (2.66) yields
(2.72)
o ~ fI+ p(t)[A'ljim(t)]k+lliminfx-+oo (xl-kg(t,x) dt

+ fl- p(t)[A'ljim(t)]k+1limsupx __ (xl-kg(t,x) dt,
00
which contradicts (2.5). Thus (2.:3:3) holds and existence of a solution to

(1.1) is guaranteed from theorem ILL
Case (ii). Suppose (2.6) holds

In this case (2.32) becomes
(2.73)
Also (2.48) together with (2.51) yields
(2.74)
A f~ p 1) 'lji m dt ~ lim inf fl+ p rl." y"g(t, y,,) dt

+ lim inf f I - p rl." Yng(t, Yn) dt
+lim inf fl+ p TI.,,[ -h(t, y,,)] dt
+ lim inf fI- P TI.,,[-h(t, Yn)] dt.
148
We next show that then' exists P2 E 1[0,1] with

(:2.7!))
Pl'l,niJng(t,Yn) ~ P'2(t) for a.p.
t E [0,1].
We have as in (:2.59),
(:2.76 )
Also
(:2.7:~)
yields
p( f )q( t )T( t )[y~,( t
IF
1
(1IIll."II;'I~ + II It;." II;') :2

so thPl'e exists a constallt B4 with
p( t )q( t )T( t )[y~, (t
)]2
1
(1I Il l.nlli,'! + Ilu;.IlIIf,):>

Hence (:2. 7!)) is true.
Next siIlce Ih(t,y)1
IIJ1I.n ll ( t, .lJn)1 :;
:; B4P(t)q(t)T(t)
:; 1>1(t) + </J2(t)
1
1
rl.s
-.
p(8)
<t.e. on [0,1] we have

1
Ip(01 + 0'>:,dlll.,,1
j
(1IItI.nll;'" + 111l;,,,II;,):2
<
111/,;.,,117,P(1)1 + (2) (1/ plt;)):2

1
(1IU 1 n
ll;'q + lIu;,nll~) 2
so there exists P:3 E 1[0,1] with

(:2.77)
Ip(t)rl,n(t)h(t,y,,(t))1
S:
P:l(t) for a.e.
t E [0,1].
Apply Fatou's lemma to (:2.74) to obtain
11+ lim inf (prl,n YnY( t, Yn)) dt

+ fj_lim inf(pTl."Yny(t,y,,)) rlt
Aft~ 7) III/'m dt ~
(:2.7X)
Let () =
+fj+
liminf(p r l,n[-h(t,Yn)]) elt

+ II-lim inf(Pl'l.n [-h(t,Yn)]) dt.
where
(t
E {:2,;~, ... }. Fix () for the mOIIlent and let
149
be as in case (i). Then as in case 0) (using (2.64) and (2.65) we have

AJJ p(t)V(t)1/1m(t)dt
~ AJI+ XI+ p(t)1/1m(t)liminfx-+00 (xg(t,x dt
e
+ A JI- Xr p( t)'ljlm (t) lim supx-+_oo (xg( t, x dt

e
(2.79)
+A JI+ XI+ P(t)'I/\n(t) liminfx-+oo (-h(t,x dt

e
+AJ1- Xr p(t)tPm(t)limsupx-+_oo (-h(t,x dt

e
+ J1+ XJt (liminf[pr1,nYng(t,Yn)] + liminf[-p r 1,n h (t,Yn)]) dt
+ J1- XJ; (liminf[pr1,nYng(t,Yn)]
+ lim inf[-p r 1,n h (t,Yn)])
We now want to let () ---+ 0+ (i.e. a =

---+ 00) in (2.79).
(2.75),(2.77) together with Fatou's lemma implies
(2.80)
liminfj XJ+ liminf[pr1nYng(t,Yn)]dt

<>-+00 1+
e
'
dt.
Notice
~0
and
(2.81)
liminfj
<>-+00
1+
XJ+ liminf(pr1 n[-h(t,Yn)]) dt

e
'
~ O.
Also
Ap(t)1/1m(t)liminf(xg(t,x ~ 0 for a.e. t E
x-+oo
It
so Fatou's lemma implies

(2.82)
liminf,,-+oo A J1+ X1t p(t)1/1m(t)lim infx-+oo (xg(t,x)) dt

~ A JI+ p(t)'ljlm(t)1iminfx-+oo (xg(t,x
dt.
Finally since
then Fat au 's lemma implies

(2.83)
Similar inequalities hold with X/+ replaced by Xr and lim infx-+oo ree
9
placed by lim sUPx-+_oo' These together with (2.79) implies
A10[01
p(t)V(t)1/1m(t)dt
Aj
/+
p(t)'ljJm(t)liminf(xg(t,X dt
:]'-+r)0
150
i- p(t)'ljJ",(t)l~~~~(xg(t,x))
Ai,
i- P(t)'l/Jm(t)l~~~~(-h(t,x))
which contradicts (2.6). Thus
:;
:;
j+
(2.;3;~)
dt
p(t)'ljJ",(t)liminf(-h(t,x)) dt
x--->oo
holds and we are finished.
:;
dt,
Remarks. (i). Notice

g(t, u) :; T(t)q(l) ill (2.4) could be replaced by
g(t,u) :; [T(t) + ",]q(t), where
K, < f is a constant (here f > is
as in (2.12)), and the result of theorem 12.1 is again true. The only change
is that now we have
E
(1Iwll;q + Ilw'll; + Iluoll;q + Ilu~II;) -
Am
(J-L -
+ K,)llwll;q
< 11plh(t,y)+v(t)llu-wldt.
Choose J-L appropriately and we have an analogue of (2.20).
(ii). One could obtain in addition a result if k = I in case (i). Of course
(2.54) does not necessarily hold in this case so we need to adjust (2.5) using
the ideas in case (ii).
fd
If
PV'l/J711 dt = then we have the following additional result. In the
next theorem we will assume that ~ E U [0, 1] and that f : [0, 1] X R - R
is a L 1 -Caratheoclory function i.e
(i). f: [0,1] X R --* R is a Caratheodory function and
(ii). for any r > there exists 11". E 1[0,1] tOuch that If(t,y)1 :; hr(t) for
a.e. t E [0,1] and for all Iyl :; r.
Theorem 12.2. Let
I(t, y) = h(t, y)
and assume f : [0,1]
+ v(t) -
yg(t, y)
R ~ R i8 a 1 -Camtheodory function with

and (2.4) holding. In addition assume
(1.3),(1.4),(L5),(2.1),(2.2),(2.;~)
(2.84)
(2.8.5)
loolp(t)v(t)1/Jm(t)dt = and lol -dss() <

0
h(t, u) :;
f01' a.e. t E [0,1],

151
and u>
00
and
(2.86)
h(t,u) 2: 0 f01' a.Co t E [0,1)'
and 'u < 0
hold. Then (1.1) has a solution.

PROOF: Let y be a solution to (2.7)5 (from [10, chapter :3] we know that
y E C[O, 1] n C 1 (0, 1) with py' E AC[O, 1]). Follow the same arguement in
theorem 12.1 to equation (2.47). In this case since (2.84) holds we have
(2.87)
and so since 1/Jm has a finite number of zeros

(2.88)
o > JI+ (p r'I,,, Yng( t, Yn) + P 1'I,n[ -h(t, y,,)]) dt

+ JI- (prl,,, y"g(t, Yn) + prl,n[-h(t, y,,)])
dt.
For t E [+, there is a integer nl (independent of t) with

(2.89)
rl,n(t) > 0 and
y,,(t) > 0 for n 2: nl and n E S; t E [+
and for t E [- , there is a integer

(2.90)
Uz
with
rl,n(t) < 0 and Yn(t) < 0 for n 2:
n2
and n E S; t E [-;
here S is some subsequence of integers. We will just prove the second

part part of (2.89). It is easy to check (using (2.:34), (2.36), (2.44) (with
remark) and the Arzela-Ascoli theorem) that there exists a subsequence
S* of integers with
(2.91 )
pr:, ~ Ap1/<n
Notice p( 1 h/J~n (1)
-::J 0
In
C[O, 1] as n
-> 00
In
S*.
(i.e. if p( 1 h'J~n (l ) = 0 then
which implies Wrn == 0, a contradiction). Also we know that 'ljJrn has a

finite number of zeros in (0,1); let ro denote the smallest one and So the
largest zero. Without loss of generality assume A'ljJ", > 0 in (so,l). Now
SlUce
152
< 0. Thus then' exists
we have Ap( 1 h/I~~ (1)
tl E
(so, 1) with
for t E [tl' 1]. Also there exists (by (2.91)) an inte!!;er kl with
for 71, ~ kl' n E S* and t E [tbI]. Thus p(t)r~~(t) < ~p(I)1/I~n(1) for
n ~ kl, n E 5* and t E [t1' 1]. Consequently since Yn(1) = 0,
for n ~ kl' n E 5* and t E [tl' 1). Similarly there exists tz with So < t2 <
it and there exists an integer kz with y,,(t) > for n ~ k2' n E S* and
t E (so, t2] (note in this case for a fixed t3 E (so, tz) then (2.64) implies
that Yn (t3) >
for n sufficiently large: now consider t E (t3, t2] and
tE(so,h. Finally since r,,~A1jlm in C[O, 1] as n-oo in 5* there
exists (since min[t2,td A 1j1",(t) > 0) an integer k3 with Yn(t) > for 71, ~
k3, 71, E 5* and t E [tz, tl]' Consequently for n ~ max{k 1 , kz, k:3 }, 71, E 5*
we have Yn(t) > on (80,1). Sinn' '~)1Ti has only a finite number of zeros
in (0, 1) then there exists an integer 71,1 with Y" (t) > for n ~ 71,1, 71, E S*
and t E /+, so (2.H9) is true.
>
Remark. Notice that 1/1",(0) "I

so if A1/!",
on (0,1'0) then since
rn - + AW", in C[O,I] as n --. 00 in S* there exists fJ E (0,1'0) and an
integer k4 with 1',,( t) > 0 for n ~ k4' 11. E S* and t E [0, fJ].
A similar arguement shows that (2.90) is true. Thus (2.89) and (2.90)
together with (2.85) and (2.86) yield
< j
[+
+
for
71,
[-
(prl,nYng(t,Yn)+prl,n[-h(t,Yn)])dt
(p rl,n y"g( t, Yn)
+ P 7"1,,,[ -he t, Yn)])
2: max{nl,nz} and n E S. This contradicts (2.HH).
dt
0
The above results have "dual" versions.

Theorem 12.3. Let f(t,y) = h(t,y)+v(t)-yg(t,y) and assume f: [0, I]x
R - R is a Caratheod07'yfunction with (l.:~),(1.4),(1.5),(1.6),(2.I), (2.2)
15;~
and (2.3) holding. In addition assunu
(2.92)
there exists T E C[0,.1] with - T(t)q(t):S:; g(t,u):S:; for a.e

{ t ~ [0,1] and u E R, hen T(t):S:; Am - Am-I for t.E. [0,1]
wzth T(t) < Am - Am_Ion a subset of [0,1] of posztzve
measure; also A-I = -00
is satisfied.
(i). Suppose there exists a constant k > , of the form 1 > k
/3 is odd and 0: is even, with
= ~,
> J/+ p(t)[A'ljJm(t)]k+ I limsupx-+oo (xl-kg(t,x)) dt
(2.93)
+ J1-
p(t)[A'lj!m(t)]k+1 liminf x -+_ oo (XI-kg(t,x)) dt
for all constants A f; 0; here 1+ = {t E [0,1] : A'lj!m(t)

{t: A'ljJm(t) < O}. Then (1.1) has at least one solution.
(ii). If,
where
> o} and 1- =
with
A J~ p(t)v(t)'l/Jm(t) dt
> A J1+ p(t)'lj!m(t) lim suPx-+oo[xg(t, x)] dt
+A f1-
p( t )'lj!m(t) lim infx-+_oo[xg( t, x)] dt

+A f1+ pet h'J,,,(t) lim sUPx-+oo[ -h(t, x)] dt
+A fI- p(t)'lj!m (t) lim infx -+_ oo [ -he t, x)] dt
(2.94)
for all constants A f; 0. Then (1.1) has at least one solution.

PROOF: In this case choose IL such that Am-I < IL < Am and let y be a
solution to
(2.95)8 {
~(py')' + ILqy,= o[h(t, y) + vet) - yg(t, y) + (IL - Arn)qy] a.e.

hmt->o+ p(t)y (t)
= y(l) =
Cit/!; , Wo
for 0 E (0,1). Let
rn-l
00
w=
~=1n
ci'lj!i,
=L
,=0
<X)
ci'lj!i, and WI
= cm'ljJm
i=m+1
in this case. The same type of analysis as that in theorem 12.1 establishes
the result. 0
154
References.
[1]. F.V.Atkinson, Disc1'etf' and continuous boundary pmblems, Academic

[2]. L.E.Bobisud and D.O'Regan, Positiw solutions for a class of nonlinear
singular boundary value problems at resonance, Jour. Math. Anal. Appl.,
184,263-284 (1994).
[3]. W.N.Everitt, M.K.Kwong and A.Zettl, Oscillations of eigenfunctions
of weighted regular Sturm Liouville problems, J. London Math. Soc., 27,
106-120 (1983).
equations, Proc. Royal Soc. Edinburgh. 112A, 145-1.53 (1989).
[5]. R.lannacci and M.N .Nkashama, Unbounded perturbations of forced
second order ordinary differential equations at resonance, Jour. Diff. Eq.,
69, 289-309 (1987).
[6]. L.V.Kantorovich and G.P.Akilov, Functional analysis, Pergamon Press,
Oxford, 1982.
[7]. J .Mawhin, Topological degree methods in nonlinear boundary value
problems, AMS Regional Coni. Series in Math., 40, Providence, 1978.
[8]. J .Mawhin and J .R.Ward, Nonuniform nonresonance conditions at the
first two eigenvalues for periodic solutions of forced lienard and Duffing
equations, Rocky Mount. J. Math., 112, 643-654 (1982).
[9]. J .Mawhin and J .R.Ward, Periodic solutions of some forced Lienard
differential equations at resonance, A1och. Math., 41, 337-351 (1983).
[10]. D.O'Regan, Theory of singular boundary value pmblems, World Scientific Press, Singapore, 1994.
[11]. D.O'Regan, Resonant nonlinear singular problems in the limit circle
case, Nonlinear Diff. Eqns. and Applications, 3( 1996), 55-77.
[12]. L.Sanchez, Positive solutions for a class of semilinear two point boundary value problems, Bull. Aust. Math. Soc., 45,439-4,51 (1992).
[13]. I.Stakgold, Greens functions and boundary value pmblems, John Wiley
and Sons, New York, 1979.
155
Chapter 13. Boundary value problems on the half line.

13.1. Introduction.
This chapter presents existenc!' principles and theory for two semlinfinite problems, namely
y" + m 2y'. = f(t, y) a.e. on [0,00)
{ y(O) = a
limt-+oo y( t) = 0
(1.1 )
and
y" - m 2 y = f(t, y) a.e. on [0,00)

{ y(O) = a
limt-+oo y( t) = O.
( 1.2)
Problems of the abov!' type occur frequently in the science and engineering
literature, see [3,6,7,9J. The fixed point theory developed in chapter 2 (in
particular theorem 2.7) will playa major role in establishing existence for
the boundary value problems (1.1) and (1.2). The technique we use was
initiated in the lat!' 1980's by Furi and Pera [.5J. It is worthwhile remarking
here that the ideas in section 13.2 could also be used to establish existence
of solutions to various other boundary value problems on the semi-infinite
interval. We also remark here that the general theory of nonlinear boundary
value problems on the semi-infinite is not very well developed. Most of the
results in the literature require rather technical hypothesis and apply only
to narrowly defined classes of problems.

In this section general existence principles are established for second
order boundary value problems. In particular we will consider
y" + m 2 y' = f(t,y) a.e. on [0,00)
{ y(O) = a
limt-+oo y(t) = 0
(2.1 )
and
y" - m 2 y = f(t,y) a.e. on [0,00)

{ y(O) = a
limt-+oo y( t) = 0:
(2.2)
here m
is a constant and y takes values in R".

156

The following notation will be used throughout this section. We let

C([O, 00), Rn) be the space of continuous mappings from [0,00) to R n ,
the topology being that of uniform convergence on any compact interval of
[0,00). If u E C([O,oo),Rn) then for every m, E {1,2, ... } we define the
semi norms Pm ( u) by
Pm('U) = sup lu(t)1
[o,tml
where tm i OOj the metric (since the topology is determined by a countable

number of seminorms) is defined by
_ ~
(x, y) -
Pm (x - y)
~ 2m 1 + P (x _ y) ,
rn=l
'Tn
Note C([O,oo),Rn) is a Frechet space. BC([O,oo),Rn) will denote the

space of bounded continuous mappings from [0,00) to R". If
u E BC([O, 00), Rn)

then we write
lul cx =
sup
[0,(0)
lu(t)l.
Remark. The Arzela-Ascoli theorem says that a set n ~ C([O,oo),R n ) is

compact iff
is uniformly bounded and equicontinuous on each compact
interval of [0,00).
We now use theorem 2.7 to establish an existence principle for (2.1).

Theorem 13.1. Assume the following conditions are satisfied:
f : [0,00) X Rn -+ Rn is a L - Caratheodory function:

- 4 f(s,y)
is measurable f01' any y E R"
(ii). y - 4 f( s, y) is continuous for a,e. .5 E [0,00)
(iii). for each l' > 0 there exists Tr E tI [0,00) such
that IYI :S l' implies If( 8, Y II :S Tr {8) for almost all
.5 E [0,00) and lilll t --->ooe- m2t J~ e m2S Tr(s)ds =
(i).8
(2.3)
and
(2.4 )
then exists a constant Mo > lal with lu{t)1 :S M o, t E [0,00)

for
any functionu E BC([O,oo),R") n WI!';([O,oo),Rn)
{
which satisjie8 u" + m,2 1l' = Af( t, 1L) a.e, on [0,00),
u(O) = a, limt--->()() u(t) = f01' :S A < 1.

157
Then (2.1) has a solution in BC([O,oo),R") n WI:';([O,oo),R").

Remarks. (i). By a solution to (2.1) we mean a function
Y E BC([O, 00), R") n WI~';([O, 00), R")
which satisfies the differential equation almost everywhere on [0,00) and
the stated boundary conditions.
(ii). If in addition f : [0,00) x R" --+ R" is continuous then by a solution to
(2.1) we mean a function y E C 2 ([0, 00), Rn) which satisfies the differential
equation everywhere on [0,00) and the stated boundary conditions. Also
in assumption (2.4), in this case, u E C 2 ( [0,00), R") and u satisfies the
differential equation everywhere.
(iii). Let I = [a, b]. The Sobolev class of functions u such that u(m-l) is
absolutely continuous and u(m) E LP(I) is denoted by wm,p(I,Rn).
PROOF: It is easy to see from (2.3) that solving (2.1) is equivalent to
finding ayE BC([O, 00), R") which satisfies
yet) = ae- m2t
(2.5 )
_e~~2t f~
Let E
n~2
e",2s
ftC>
f(s,y(sds
f(8,y(s))ds
+ e~,~2t fooo f(s,y(sds.
= C([O, 00), R"),
Q = {y E C([O, 00), R"): y E Be([O, 00), R"), IYloo 'S: Mo
+ 1 = No}
and
Fy(t)
Now Q is a closed, convex, bounded subset of C([O,oo),Rn ). We claim

F : Q --+ C([O, 00), R") is compact. First we show continuity. Let Yn --+ Y
in Q. We must show Fy" --+ Fy in Q. There exists TNo EL I [0,00) with
If(s,Yn(s1 'S: TNo(S) and If(s,y(s)1 'S: TNo(S) for almost all .~ E [0,00).
Also for each t E [0,00) we have
f(s,Yn(s))
-->
f(s,y(.s)) for a.e. s E [0,00).
This together with the Lebesgue dominated convergence theorem implies

FYn(s) --> Fy(s) pointwise on [O,tm]. Let x,t E [O,t m] with t < x. Then
/Fy,,(t) - Fy,,(:r)1
'S:
158
A similar bound can be obtained for IFy( t) - Fy( x)l. So given I: >
exists an b > such that t, x E [0, t",] and It - xl < b implies
(2.6)
IFYn(t) - FYn(x)1 <
I:
there
for all n
and
(2.7)
IFy(t) - Fy(x)1 <
Hence (2.6), (2.7) and the fact that FYn(8) ~ FY(8) pointwise on [O,t",]
implies the convergence is uniform on [O,t"J Thus F: Q ----+ C([O,oo),R n )
is continuous.
We now show F( Q) is relatively compact in C([O, 00), Rn). This follows
once we show F( Q) is uniformly bounded and equicontinuous on [0, t",].
We know there exists TNa E L1 [0,00) with If( 8, u)1 :::; TNa (8) for a.e. 8 E
[0,00) and lui :::; Mo. The equicontinuity of F( Q) on [0, t m ] follows
essentially the same reasoning as that used to prove (2.7). Also F(Q) is
uniformly bounded since for t E [0, t m ] we have
for each u E Q. Thus F(Q) is relatively compact in C([O,oo),R") and so

F : Q ----+ C([O, 00), Rn) is compact.
The resul t follows immediately from theorem 2.7 once we show condition
(2.1) in chapter 2 is satisfied. Take a sequence {(Yj,Aj)}~l in 8Q X [0,1]
with Aj ----+ A and Yj ----+ Y with y = AF(y) and
A < 1. We need
to show that Aj F( yj) E Q for j sufficiently large. Take any 11 E E with
11I(tll S No for t E [0,00). Then
: :;
IFv(t)1 <
159
Notice limt->cx) II! No (t) = O. This to)!;ether with the fact that Yj E Q implies
that there exists ao :::: 0 with IFYj(t)1
Mo + 1 = No for t E lao, 00) and
] E {L2, ... }. Consequently
(2.8)
IA.iFYj(t)i:S; No, t E [ao, 00) and] E {l,2, ... }.
Next consider the situation when t E [0, aoJ. Since F is continuous on Q

we have FYl ~ Fy uniformly on [O,aoj. In addition since Aj - A and
F(Q) is a subset of a bounded set ill E we have AjFYj'--" )"Fy uniformly
on [0, aoJ. Thus there exists ]0 E {1. 2, ... } with
(2.9)
Now Y
IAjFYj(t)1 :s; IAFy(t)1
= AF(y)
+ 1.
t E [0, ao] for]:::: ]0.
so (2.4) implies
I)"Fy(t)i :s; Mo
and this together with (2.9) implies for] ::::]0 that
(2.10 )
Now (2.8) and (2.10) imply that AjF(Yj) E Q for] :::: ]0' Consequently
all the conditions of theorem 2.7 are satisfied so (2.1) has a solution. 0
Remark. Note (2.4) amounts to findinl!; a priori bounds on the family of
problems 'U = AFu with O:S; A < 1.
We also have an existence principle for (2.2).
Theorem 13.2. Assume the following condition8 are satisfied:
(2.11)
f : [0,00) X R" -. R" is a L * - Camtlufodory function:

(i).8 - 4 f( 8, y) is measurable for any y E Rn
(ii). y - 4 f( 8, y) is continuou8 for a.e. 8 E [0,00)
(iii). for' each l' > 0 there exists Ir E L1 [0,00) such that
Iyl:s; l' implies If(8,y)l:s; Tr('~) f01' almost all 8 E [0,00)
togethe1' with limt_cxoc-mt J~ f"LSlr(s)d,~ =
,
an d IIIIlt->(x)
f tnt j'()O
t
f -ms Tr ( S ) d.5 =
and
there exists a CO,nstant Mo > lal with IU,(t)1 :s; Mo, t E [0,00)
'2 1
for
any function It E BC([O, 00), R") n Wlo~ ([0, 00), Rn)
( 2.12) {
which satisfies u" - m 2 u = Af(t, 'u) a.e. on [0,00),
u(O) == a, 1imhoo u(t) = 0 for 0 S A < 1.
160
Then (2.2) has a solution in BC([O, 00), R")
n WI!; ([0,00), Rn).
PROOF: It follows from (2.11) that solving (2.2) is equivalent to finding a

y E BC([O,oo),R") which satisfies
y(t)
ar- mt
-('Tnt
2m
(,-mt
+ -.-
1=
2m
1= (,-
ms
f(8,y(sds
f - 1II1
f-mSf(8,y(.~)d.s---
2m
lt
0
(,""f(.s,y(.sd.s.
Let
Q = {y E C([O,oo),R"): y E BC([O,oo),R"),
iyioo
~ Mo
+ 1 = No}
and
af-mt
Fy(t)
emt
2m
- ,-
+ f.-mt
1C'
.
2m
f-
f''''j f-msf(s,y(sds
io
.,-1IIt
llLS f(s,y(.sds- .2m
lt
0
e""f(s,y(sds.
Essentially the same reasoning as in theorem n.1 will now establishe the
result. 0
Remark. Existence principles analogous to theorem's 13.1 and 13.2 may be

established for other types of boundary conditions.
The general existence principles derived above can be easily used to
establish existence of solutions to various boundary value problems.
Examples.
(i). Consider the boundary value problem
y"
m 2 y'
f(t, y).
{ y( 0) = 0
Ii Ult-+(K, y( t) = o.
(2.13 )
o~
< 00,
m -:J 0
Assume (2.3), with n = 1, and
f : [0,00) X R
(2.14 )
(2.15)
-+
R is continuous
there exists a constant Mo > 0 such that

implies yf(t,y) > 0 for all t E [0.00)
hold. Then (2.1;{) has a solution in y E B('( [0,00), R")

161
iyi >
Mo
n ('2([0,00), R").
This follows immediately from theorem 1:3.1 once we show condition

(2.4) holds. To see this let u E BC([O,oo),Rn) n C 2([0,00),Rn ) be a
solution to
u" + m 2 u' = AJ(t, u), 0:::; t < 00,0:::; A < 1

{ u(O) = 0
limt-+oo u(t) = O.
We claim lu(t)1 :::; Mo for t E [0,00). If A = 0 then this is true since u == O.
So suppose 0 < A < 1. If there exists atE (0,00) with lu(t)1 > Mo
then max[o,oo) lu(t)1 = lu(to)1 > Mo with to E (0,00) and u'(to) = O.
Consequently
u ( to ) u " (to)
= u(to). [ u " (to) + m 2 u (to) = AU(to)J(to, u(to)) > 0,

I
which contradicts the maximality of lu(to)l. Hence lu(t)1 :::; Mo for t E

[0,00) and so (2.4) holds. Existence of a solution to (2.1:3) is guaranteed
from theorem 1:3.I.
(ii). Consider the boundary value problem
y" - m 2 y = J(t, y), 0 :::; t < 00, m

{ yeO) = 0
limt--+oo yet) = O.
(2.16)
Assume (2.11), with n

(2.17)
1= 0
= 1, (2.14) and
there exists a constant Mo > 0 such that

implies yJ(t,y) ~ 0 for all t E [0,00)
Iyl > Mo
hold. Then (2.16) has a solution in y E BC([O, 00), Rn) n C 2 ([0, 00), Rn).
The result follows immediately from theorem 1;3.2. We need only check
(2.12) holds. Let u E BC([O, 00), Rn) n (,'2([0,00 ),Rn ) be a solution to
u" - m 2 u = AJ(t, u), 0:::; t < 00,0:::; A < 1

{ u(O) = 0
limt-+oou(t) = O.
Again assume 0 < A < 1 and max[o,oo) lu(t)1
(0,00). Then
u(to)u " (to) =
Tn 2 [u(to)] 2
= lu(to)!
> Mo with to E
+ AU(to)J(to,u(to )) > 0,
a contradiction.
162
References.
[1]. J.V. Baxley, Existence and uniqueness for nonlinear boundary value
problems, J. Math. Anal. Appl., 147( 1990), 12:1-1;};1.
[2]. J.W. Bebernes and L.K. Jackson, Infinite interval problems for y"
f(t,y), Duke Math. J., 34(1967). :}9-47.
[3]. L.E.Bobisud, Existence of positive solutions to SOlIle nonlinear singular

boundary value problems on finite and infinite intervals, J. Math. Anal.
Appl., 173(1993),69-83.
[4]. H.Brezis and F.E.Browder, Existence theorems for nonlinear integral
equations of Hammerstein type, Bull. Amer. Math. Soc., 81(1975), 73-78.
[5]. M.Furi and P.Pera, A continuation method on locally convex spaces
and applications to ordinary differential equations on noncompact intervals,
Ann. Polan. Math., 47(1987), 3:11-:346.
[6]. A.Granas, R.B.Guenther, .J.W.Lee and D.O'Regan, Boundary value
problems on infinite intervals and semiconductor devices, J. Math. Anal.
Appl., 116(1986), 335-:348.
[7]. W.Okrasinski, On a nonlinear ordinary differential equation, Ann.
Polon. Math., 49(1989), 237-245.
[8]. D.O'Regan, Theory of singular boundary value problems, World Scientific Press, Singapore, 1994.
[9]. D.O'Regan, Positive solutions for a class of boundary value problems
on infinite intervals, Nonlinear Diff. Eqns. Appl., 1( 1994), 203-228.
[10]. D.O'Regan, Continuation fixed point theorems for locally convex linear topological spaces, Mathematical and Computer Modelling, 24(1996),
57-70.
[11]. R.Precup, Nonlinear boundary value problems for infinite systems of
second order functional differential equations, "Babes-Bolyai Univ. Seminar on Diff. Eqns., 8( 1988), 17-:10.
[12]. E.Zeidler, Nonlinear functional analysis and its applications, Vol I,
Springer Verlag, New York, 1986
Chapter 14. Existence theory for ordinary differential equations

on compact and noncom pact intervals.
14.1. Introduction.
This chapter presents existence results for second order differential equations on finite, semi-finite, infinite, open and half open intervals. In particular, we examine the problem
(1.1 )
y":= f(t,y,y')
a.e. on (a,b)
where -00 ::; a < b ::; 00 and f is a locally Caratheodory function. We

also discuss seperately the problem (1.1) with boundary conditions:
( 1.2)
y(a)
= c,
y(b)
:=
and/or
( 1.:3)
d.
The main objective of this chapter is to give a systematic treatment of

the smoothness properties of a solution of (1.1). For example, when we
examine (1.1)(1.2), it is of some importance to know what conditions must
one put on the nonlinearity f to have a solution in e[a, b) n WI!'} (a, b) or
1
2 1
in C [a,b)n Wlo'c(a,b).
Throughout this chapter, we will assume the existence of lower and upper solutions of (1.1). In addition, our nonlinearity f will satisfy a growth
condition of Nagumo type. Our technique (which was initiated in [6]) uses
results from the theory of boundary value problems on compact intervals
(see chapter 5) together with a simple fixed point result. More precisely, the
fixed point is obtained as a limit of fixed points of an appropriate sequence
of operators.
The following notation will be used throughout this chapter. Let I be
a real interval. We denote by ('k(I) the space of k-times continuously
differentiable functions endowed with the topology of uniform convergence
on each compact subset of I. This space is a Banach space if I is compact,
otherwise, it is a metrizable locally convex space. By C;:"(I), we mean the
space of infinitely differentiable functions with compact support in I. We
denote by Lfoc(I) the set of functions integrable on each compact subset of
I. Observe that if I is compact, L1 (l) = Lloc(l). Similarly, we denote by
Wl~':(l) the set of functions y E W 2,1(J) for every compact subset J c I.
164
Definition 14.1. We say that J: I X R2 -. R is a Lloc(I)-Caratheodory

function if
(i). the map t ~ J(t, y, z) is measurable for all (y, z) E R2;
(ii). the map (y, z) ~ J(t, y, z) is continuous for almost all tEl;
(iii). for any r > 0, there exists hr E Lloc(I) such that IJ( t, y, z)1 :S h r {t)
for almost all tEl and for all Iyl :S r, Izi :S r.
Observe that if I is a compact interval, this definition coincides with
the usual definition of Caratheodory function.
Now we give the definition of lower and upper solutions of (1.1), (1.1)
(1.2) and (1.1) (1.3). We take I a real interval such that (a,b) C Ie [a,b]
where -00 :S a < b :S 00.
Definition 14.2. A function a E WI~'~ (I) is called a lower solution of

(1.1) if
a"(t) 2 J(t,a(t),a'(t
almost everywhere on I. When a E I, a is called a lower solution of (1.1)
(1.2) if it is a lower solution of (1.1) and satisfies a( a) :S c. Similarly, when
bEl, a is called a lower' solution of (1.1) (1.3) if a is a lower solution of
(1.1) and satisfies a( b) :S d.
We define an upper solution of (1.1), (1.1) (1.2), and (1.1) (1.3) by reversing
the inequalities.
The next result can be established by a slight variation of the arguement
given in chapter 5 (see theorem 5.9 with its remark). We refer the interested
reader to [4,5].
Theorem 14.1. Let -00 < a < b < 00 and J: [a,b] X R2 -. R be a L1Caratheodory function. Assume there exists a, /3 E W 2 ,1[a,b] respectively
lower and upper solutions oJ(1.1) (1.2) (1.:3) with a(t) :S /3(t) on [a,b].
In addition, assume there exist p E [1,00], a function I E U[a,b] with
1/p+ l/q = 1, M > 0, and a Borel measurable function 'Ii' : [0,00) -'> (0,00)
1
such that xP/'tIJ(x) E LLc[O,oo),

IJ(t,y,z)1 :S I(t)w(lzl) a.e. t E [a,b] and all a(t) :S y:S [J(t),
and
1M 'Ii~~)
1
[A]* 1I/IILQ[a.bj <
dx,
where r = Id-cl/lb-al, and A = sup{[J(t)-a(x): t,x E [a,b]}.

Then the problem (1.1) (1.2) (1.3) has a .~olution y E W 2 ,1[a,b] such that
a(t) :S y(t):S {3(t) and ly'(t)1 :S M f01' all t E [a,b].
165
Now, we formulate a simple fixed point theorem. This will be used

section 14.2 to establish existence results for second order differential
equations on semi~infinite, infinite, open and half open intervals.
Let {En}n~l be a sequence of topological spaces such that ... C E2 C
E1 and n~=l En = E. We say that a sequence {X',}n~l such that Xn E En
for all n E N converges to x E E if the sequence {Xn}n~k converges to x
in Ek for every kEN.
III
Theorem 14.2. (Fixed point theorem) Let E be a space and T : E ---.

E be a multivalued operator. Suppose there exist a sequence of metrizable
locally convex spaces . .. C E2 C E 1 , and a sequence of multivalued operators {Tn: En ---. En}n>l such that nn>lEn = E, and for each sequences
{xnkh~l and {vnkh~; with 1Jnk E Tn)x nk ), converging to x and v respectively, we have v E T( x), A.ssu1/u that for evcr'y n EN, Tn has a fixed
point Xn E En' In addition, suppose that {Xn}n~k is relatively compact in
Ek for every kEN, Then T has a fixed point in E,
Remarks. (i) The proof is immediate, The assumptions yield the existence
of a subsequence S of integers and an x E E with Xn converging to x as
n
00 in S.
---7
(ii) In the applications in section 14.2, En will be Banach spaces, the operators Tn : En -+ En will be single-valued and the operator T : E -+ E
will have the following property: for any subsequence S of integers and
any x E E with Xn converging to x as n -+ 00 in S, there exists vEE
such that Tn Xn converges to v E Tx as n ---. 00 in S.
14.2. Existence theory

Let I be a real interval (open, half-open, closed, finite, or infinite) and
let f: I X R2 ---. R be a LL.(I)~Caratheodory function. So, (a,b) C I C
[a, bJ for some -00 :S a < b:S 00.
We denote by E a space to be specified in the applications and such
that E C C1(a, b). Define the llluitivalued operator T: E -+ E by
T(u)
{g E En
Wl~':(a,b)
for all <p E
1
6
g(t)"(t)dt
1
6
f(t,u(t),u')(t)dt
C~(a,b)}.
Of course, for the moment, we don't know if the operator T is well defined.
Let {akh~l' {bdk~l' {Cdk2':l' {ddk2':l be sequences of real numbers
such that
166
and (a, b) C Uk2: 1 [ak, bk ] C I. For kEN, denote by Ek a Banach space,

and by Tk : Ek --> Ek an operator to be specified later and such that
E = nk~lEb Ek+l C Ek, Ek C
(d [ak,
bkj,
and Tk(U)(t) = Fk(1l)(t) for every t E [ak,bd, where Fk(U) E W:.!,l[ak,bk] is

the unique function satisfying
Fdu)"(t) = f(t,u(t),u'(t)) a.e. on (ak,bk),

Fk(u)(ak) = Ck, Fk(U)(bkl = d k
Observe that for every u E Ek,
for every 4> E C;:'( ak, bd. Our first result establishes the existence of a
solution of (1.1) in WI~';(a,b).
Theorem 14.3. Let f : I X R2 --> R be a LL(a,b)-Caratheodory function. Assume there exist 0 :; {j E WI:'; (a, b) respectively lower' and upper solution of (1.1). In addition, assume there exist p E [1,00], a function 'I E Lioc(a,b) with lip + l/q = 1, and a Bor'e/ measurable function
1/': [0,00)
-->
(0,00) such that X P/1j!(X) E Lloc[O,oo),
If(t,y,z)1 :; 'I(th')(lzl)
and
;
00
a.e. tEl and all o(t):; y:; (3(t),
:J;~
1/)(:1:)
dx
= 00.
Then (1.1) has a solution y E W2,1(a,b) such that
0::; y :; f3.
Remark. In Theorem 14.3, we can replace

00
;
where
rk
and Ak
x~
'lj;(X) d:1:
= 00
= min{I(3(bk) -
= sup{ /3(t) -
o(adl, 1/3(ak) - O(bk)1 }/Ibk - akl,
nix) : t,:1: E [abbk]}'
PROOF: As before, choose {adk>l' {bdk>l' {Ckh>l, {ddk>l sequences

of real numbers such that Uk~da:,bd = (;;,b), ala:) ::; Ck ~ /3(ak)' and
167
a(b k ) ~ dk ~ f3(b k ). Fix E = r: 1 (a,b), Ek = C 1 [ak,bkJ, and Tk : Ek -+ Ek

the operator defined by Td u) = Fk('U) for a1l1l E Ek where Fd u) is defined
previously.
It is clear that a fixed point of T in E is a solution of (1.1). Also,
E = nk2:1 Ek, and if ::t:" E En is a sequence converging to x E E and
Tn(x n ) -+ vEE, then (2.1) and the Lebesgue dominated convergence
theorem imply v E T( x).
On the other hand, a fixed point Yk E Ek of Tk is a solution in
W 2 ,1[ak, bk ] to the problem
(2.2)
y"(t) = f(t,y(t),y'(t)) a.e. on (ak,b k ),

y(ak) = Ck, y(b k ) = d k
By theorem 14.1, Tk has a fixed point Yk E Ek such that a(t) ~ Yk(t) ~

f3(t) and ly~(t)1 ~ Mk for every t E [ak, bkj with
where
and
In fact, for every kEN and for every n ;::: k, the following inequalities are
satisfied:
and
ly::(t)1 ::; hd t ) a.e.
on [ak,bkJ,
where hI.: E l[ak, bd is a function given in condition (iii) of definition

14.1.
The Arzela-Ascoli theorem implies that {Yn}n2:1.: is relatively compact
in Ek. The conclusion follows from Theorem 14.2. 0
Now, we assume that - 00 < a E I and we want to point out some
additional assumptions that could lw imposed in order to guarantee that
the solution lies in C[a, b) n W 2 ,1 (a, b).
Theorem 14.4. Let f: IxR2 -+ R be a Lloc(a,b)-Caratheodoryfunction

with condition (iii) of definition 14. J replaced by
168
(i). for any l' > 0, there exi..'1ts hr E Lloc(a,b) such that If(t,y,z)1 ~
hr(t) for almost all t E (a, b) and for' all Iyl ~ 1', z E R; (t - a)hr(t) E
Lloc[a, b), and
In addition, assume there exist 0' :::; j3 E era, b) n Wl:~ (a, b) 1'espectit)ely
lower' and upper solutions of (1.1)(1.2). Then the pmblem (1.1)(1.2) has a
21
solution Y E C[a, b) n Wlo~
(a, b).
PROOF: As before, choose the sequences
such that
and
ick -
cl = min{ Ix - cl : O'(ak) :::; x
:::; f3(ak)}.
Take E = C[a,b)nC 1(a,b), Ek = C[a,bklnC1[ak,bkl, and define Tk : Ek--+

Ek by
Fk( uHt),
Tk(U)(t) = {
Ck- C (t - a) + c
ak- a
'
It is dear that E = nk?lEk, and if Xk

x and Tdxk) --+ v in E, then
VET(x).
Again, by considering the problem (2.2) and applying theorem 14.1,
we get the existence of a fixed point Yk E Ek to Tk for every kEN.
Moreover, for every n :2: k, the following inequalities are satisfied:
---,0
inf {O'(t)} ~ Yn(t) ~
tE[a,ak)
Iy~:(t)l ~
hdt)
sup {{3(t)}
tE[a,ad
for every t E [a,akl, and
almost everywhere on [ak, bkl,
where Mk > 0 is as in the proof of theorem 14.3, and hk E Lloc{ a, b) is

given in (i) (in the statement of theorem 14.4).
By the Arzela-Ascoli theorem,
{Y,.}n?k is relatively compact in C 1 [ak' bkl

169
for every kEN. Now, we want to show that it is relatively compact in

bk ]. We already know that this set is bounded. All that remains to be
shown is the equicontinuity at to for every to E [a, bk ].
First, we examine the case when to -I a. Choose N 2: k such that
a < aN < to. Then for n 2 N and aN ::; s < t E [a, bd, we have
era,
IYn{t)-Yn(8)1
IYn(bkJ-Yn(aN)(t_8)
bk - aN
bk
k JT !(T,Yn(T),y;,(T))dTdr
Jb
aN
aN
-.5
1J"
"N
aN
< J(kl t - 81
with
J(k
= (l/(b k -
!(T,Yn(T),y;,(TdTdrl
+ 211hNIIU[aNhlit - 81,
alSUp{ 1/3(t)l, la(t)1 : t E [a,b k ]}.
This implies the equicontinuity of {Yn}n>k at to E (a, bk].

On the other hand, observe that for ;~ 2: k, we have
if tE[a,a n ],
1 ~(t-a)l,
ftn-a
/( cn - c)
+ Yn(bki-yn(an)(t
bk-an
a)
- bbk-t
k-an
J.t
f/k(b k - r)!(r,Yn,y;,)dr/, t E (an,bk].
btk-~~~'
an
(r - an)!(r,Yn(r),y~Jrdr
This immediately yields
I~~=~ (t IYn(t) - y,,(a)1 $ {
a)l, if t E [a, an],
len - cl + J(kl t -
bt __an
k
an
al + b~~~~ fL(r - an)hk(r)dr

I,tbk(bk - r)hdr)dr, if t E (an,b k ],
and so
I~~=~(t -
Iy,,(t) - y,,(al! $ {
all, if t E [a, an],
len - c/ + J(klt -
+ b~=~l
al
+ b~k.:;,tl
f:(r - a)hk(r)dr
ftbk (bk - r )hk{ r) dr, if t E (an, bk].

170
By choice of Cn , Cn ---. c. Also, by (i) and Hospital's rule, (r - a)hk(r) E

Lloc[a,b) and
lim(t
- a)
t __ a
l
t
bk (bk
- r)hdr)dr
= O.
Therefore, {Yn}n2: k is equicontinuous at a, and the proof is complete.
Remark. Observe that the solution y obtained in the previous theorem is

not necessarily in Cl[a, b) but y' E Ltoc[a, b).
Theorem 14.5. Let f : I X R2 ----> R be a Ltoc[a,b)-Camtheodory function. Assume there exist a ::; (3 E W/!'; [a, b) 'respectively lower and upper
solutions of(1.1) (1.2). In addition, assume there exist p E [1,00]' afunction 'Y E Lioc[a,b) with lip + llq = 1, and a Borel measumble function
~): [0,00)
---->
(0,00) such that XP f'Ij)(x) E LL[o, 00) ,
If(t,y,z)1 ::; 'Y(t)1/)(lzl)

and
JOO
a.e. t E I and all aCt)
~~:) dx =
(3(t),
00.
Then (1.1) (1.2) has a solution y E W/!';[a,b) such that a::; y::; (3.
PROOF: The proof is similar to that in Theorem 14.:3, We choose ak == a,
and Ck == C. 0
Remark. In Theorem 14.5, we can replace
where
and
Ak
= sup{(3(t)- a(x)
: t,x
[a,b k ]}.
Similar results could be obtained if we are interested in solutions in

W/!;(a, b) n C( a, b), or in W/!'c1 (a, b], or in Wl!~ (a, b) n C[a, b).
We give three very simple examples to illustrate the above results. Consider the problem
(2.3)
t r y"(t)
= f(t,y(t))
171
a.e. t E (O,b),
and/or
(2.4 )
y(o)
where f: (O,b) X R
-+
= c,
R is a LOO(O,b)-CaratModory function.
Example 14.1. Assume there exist
0 :::;
/3 E
21
w,o~
(0, b) such that
t To//(t) 2: f(t,o(t)) and t Tf3//(t) :::; f(t,/3(t))

almost everywhere on (O,b). Then, for any r E R, the problem (2.3) has a
solution in W,!';(O,b).
To get more regularity on the solution, we add some restrictions.
Example 14.2. Assume there exist
0:::; /3
E W,!'}(O,b)nC[O,b) such that
tTo//(t) 2: f(t,o(t)) and tT{3//(t):::; f(t,(3(t))

almost everywhere on (0, b), and 0(0) :::; c :::; /3(0). Then, for any r < 2,
the problem (2.;3) (2.4) has a solution in W,!'} (0, b) n qo, b).
Example 14.3. Under the assumptions of the previous example with
0::; (3 E W,!';[O, b), for any r < 1, the problem (2.:3) (2.4) has a solution in
WI!': [0, b).
172
References.
[1]. P. B. Bailey, L. F. Shampine and P. E. Waltman, Nonlinear two points

boundary value problems, Academic Press, New York, 1968.
[2]. J. W. Bebernes and L. K. Jackson, Infinite interval problems for y" =
f(t,y), Duke Math. J., 34 (1967), 37-47.
[3]. C. Y. Chan and Y. C. Hon, Computational methods for generalised
Emden Fowler models of neutral atoms, Quart. Appl. Math., 46 (1988),
711-726.
[4]. M. Frigon, Application de la theorie de la transversalite topologique
Dissertationes Math. 296 (1990), 1-79.
a des
[5]. M. Frigon, Boundary and periodic value problems for systems of nonlinear second order differential equations, Topol. Methods Nonlinear Anal.,
1 (1993),259-274.
[6]. M. Frigon and D.O'Regan, Existence theory for compact and noncompact intervals, ComIll. on Applied Nonlinear Analysis, 2(1995), 75-82.
[7]. R. E. Gaines and .J. Mawhin, Coincidence degree and nonlinear differential equations, Lecture Notes in Math., 567, Springer Verlag, New York,
1977.
[8]. A. Granas, R. B. Guenther and J. W. Lee, Some general existence
principles in the Caratheodory theory of nonlinear differential systems, J.
Math. Pures Appl., 70 (1991), lri3-196.
[9]. A. Granas, R. B. Guenther, .J. W. Lee and D. O'Regan, Boundary
value problems on infinite intervals and semiconductor devices, J. Math.
Ana1. Appl., 116 (1986),335-348.
[10]. V. Lakshmikantham and S. Leela, Nonlinear differential equations in
abstract spaces, Pergamon Press, New York, 1981.
[11]. J. Mawhin, Topological degree methods in nonlinear boundary value
problems, AMS Regional Conf. Series in Math., 40, Providence, 1979.
[12]. K. Schmitt and R. Thompson, Boundary value problems for infinite
systems of second order differential equations, J. Differential Equations, 18
(1975),277-295.
173
Chapter 15. Impulsive differential equations.

15.1. Introduction.
In this chapter we discuss first order impulsive differential equations.
Many physical situations are modelled by problems of this kind, for examplp
problems in optimal control theory and problems in threshold theory in
Biology. The last ten years or so have seen major developments in the
theory of impulsive differential equations. In this chapter we present some
of the more advanced results to date in the existence theory of nonlinear
first order impulsive differential equations with variable times. Let k be
a positive integer and T E (0,00]. In section 1.5.;3 we establish existence
results for the impulsive differential equation (IDE).
(1.1)
y' = f(t,y) for a.e. t E [O,T), t i: Ti(y(t)),

{ y(t+)=Ii(Y(C)) if t=Ti(y(t)),i=l, ..... ,k,
y(O) = Yo.
Here f : [0, T) X R" --7 Rn is a LLc[O, T)-Caratheodory function, Ii maps

R n into R" and Ti: Rn -" R for i = i, ... ,k.
We let n" = { u : [a, T) -" R" : u is continuous except for a countable
number of points ti E (a, T ),u( tt) = lilIlh_O+ u( ti +h) and u( ti ) exist and
u( ti) = 'u( til} and n~ = {'u E
'll is differentiable almost everywhere
on (a,T) and u' E LL[a,T)}; here 0 ::; a < T. By a solution to (1.1)
we mean a function y E n~ with y(t+) = Ii(Y(C)) if t = Ti(y(t)), i =
l, ..... ,k, y(O) = Yo and y satisfying the differential equation y' = f(t,y)
for a.e. t E [O,T), t i: Ti(y(t)),i = i, ..... ,k.
Most of the literature to datI' examine impulsive differential equations
with fixed moments i.e. when Ti is a constant for each i and Ti+1 >
Ti for i = 1, ... , k. Only very recently [:3,4,7] have attempts been made
to establish existence results for the more general problem (1.1) i.e. for
impulsive differential equations with variable times. In section 1.5.;~ two
types of existence results are established for ( i.l). Our first result uses the
notion of upper and lower solutions. Conditions are put on f, Ii and Ti
to guarantee that (1.1) has a solution y E n~ with art) ::; y(t) ::; p(t)
for t E [0, T). Our second result establishes existence of a solu tion to (1.1 )
when our nonlinearity f satisfies a growth condition of Wintner type [2,8].
na :
15.2. Preliminaries.
In this section we establish two existpl\('p results for tlH' initial valup
i74
problem
(2.1)
y' = f(t, y) a.e. on [a, T),

yea) = ao,
which will be needed when we examine the IDE (1.1): here 0 S a < T S 00.
We will establish the existence of a solution to (2.1) in AC/oc([a,T),Rn).
Recall AC/oc([a,T),Rn) is the set of functions u E C([a,T),Rn ) which
are absolutely continuous on every compact subset of [a, T). Also recall
if u E C([a,T),Rn) then for every m E {1,2, ... } = N+, we define the
semi norm Pm ( u) by
Pm( u) = sup lu(t)1
[a,tmJ
where tm i T. Notice C([a, T), Rn) is a locally convex linear topological space. The topology on C ([ a, T), Rn), induced by {Pm} mEN+, is the
topology of uniform convergence on every compact interval of [a, T). From
the Arzela-Ascoli theorem, a set n ~ C([a, T), Rn) is compact iff n is
uniformly bounded and equicontinuous on each compact interval of [a, T).
We use the Schauder-Tychonoff theorem [1] to establish existence results for (2.1). For completeness we state the fixed point result.
Theorem 15.1. Let ]( be a closed convex subset of a locally convex linear
topological space E. A ssume that f : ]( ---. ]( is continuous and that

F (]() is relatively compact in E. Then f has at least one fixed point in
](.
We will assume throughout this section that f: [a, T) X R n ---. R n is

an LJoc[a, T)-Caratheodory function; by this we mean
(a). the map s 1--+ f( S, y) is measurable for all y E R".
(b). the map y 1--+ f( s, y) is continuous for almost all <~ E [a, T),
(c). for each r > 0 there exists tLr E LJoc[a,T) such that Iyl S r implies
If(s,z)1 ~ tLr(s) for almost all s E [a,T).
Theorem 15.2. Let f : [a, T) X Rn ---+ Rn be a Ltoc[a, T)-Camtheodory
function. Assume the following conditions ar'e satisfied:
(2.2)
(2.3)
If(t, y)1 ~ q(t)'IjJ(lyl) for almost all t E [a, T) with

{ 'IjJ: [0,00) ---+ (0,00) a Borel measumble function
with t, E LJocla, 00) and q E LJocla, T);
1/) is
nondec7'ea<~ing;
175
and
(2.4)
t*
q( s) d<~ <
1(X) _(u
d
for' any f" < T.
laol '1/) u)
Then (2.1) has a solution y E ACloc([a,T),Rn).

PROOF: A solution to (2.1) is a fixed point of the operator
N : C([a,T),R n
defined by
= ao +
Ny(t)
Let
f(
it
= {y E C([a,T),R n ):
where
(2.5)
bet)
= J- 1
and
J(z)
C([a,T),R n )
) -
f(s,y(s))ds.
ly(t)l:S bet), t E [a,T)}
(it
q(X)dX)
dx
-.-.
lao I '1/)( x)
Notice f( is a closed, convex, bounded subset of C([a, T), R"). We next

claim that N maps f( into f(. To see this let y E f(. Notice for t < T
that
INy(t)1
< laol
=
sInce
laol
it
+ it
q(s)1/)(ly(s)i)ds:S laol
b'(s)ds
b(t)
it
q(s)1/J(b(s))ds
= bet)
dx
.1.( )
it
q(x)dx.
laol 0/ X
a
Thus Ny E f( and so N : f( --. f(.
It remains to show that N : C([a,T),Rn ) --. C([a,T),R") is continuous and completely continuous. To see continuity let y" ---> y in
C([a,T),R") i.e. Pm(Yn) ---> Pm(Y) for each m. The Lebesgue dominated
convergence theorem together with the fact that f : [a, T) X R" ---> R n
is a LloJ a, T)-Caratheodory function implies N Yn ---> Ny uniformly on
[a,t m ] for each t m . Consequently N : C([a,T),R") ---> C([a,T),Rn) is
176
continuous. To show N is completely continuous, let A ~ C([a, T), R n)

be bounded. It is easy to see that N (A) is uniformly bounded and equicontinuous on [a,t",] for each tm. Hence N : C([a,T),R") -- C([a,T),R")
is completely continuous.
The Schauder Tychonoff theorem implies that N has a fixed point in
K i.e. (2.1) has a solution y E AC/oc([a,T),RU ). 0
Remark. We note that assumption (2.3) can be removed in theorem 15.2;
see [9].
Theorem 15.2 immediately yields the following result which is useful

when we know that there exists upper and lower solutions to (2.1).
Theorem 15.3. Let f : [a, T) X R" -- R U be a LL[a, T)-Caratheodory
function with the following condition satisfied:
(2.6)
then: exists q E LL[a,T) with If(t,y)1 :'S q(t)

for alrrwst all t E [a, T) and all y E R".
Then (2.1) has a solution y E AC/oc([a,T),Rn
).
We now use theorem 15.3 to obtain an extra existence result for (2.1)
when f : [a, T)xR -> R is a Llocla, T)-Caratheodory function. A function
a E n~ is called a lower solution of (2.1) if
a'(t):'S f(t,a(t) a.e. tE[a,T),
{ a(t+):'S a(r) for all t E [a,T),
a(a) :'S ao
Similarly we define an upper' solution of (2.1) by reversing the inequalities.

Theorem 15.4. Let f : [a, T) X R - R be a Lloc[a, T)-Caratheodory
function. Assume the following conditions ar'e satisfied:
(2.7)
there exists a, Ii E n~ r'espectiV.ely lower' and upper

solutions of (2.1) with a(t):'S 13(t) for t E [a,T);
and
(2.8)
there exists Si.E (a,T), i = 1,2, ... , a < 81. <:.~2 < ....
{ and Si T T wzth a(t+)
a(r), /3(t+)
/3(t )
for all t E [a,T)/{81'<~2'''''}'

Then (2.1) has a solution y E AC/oc([a. T), R") with a(t)
on [a,T).
177
:'S yet) :'S (3(t)
PROOF: Consider the modified problem

{ y' = r(t,y) a.e.
:lJ(a) = ao,
(2.9)
where
[a,T),
OIl
J(t,p(t)). y > d(t),

I*(t, y) = { J(t, y), aCt) :::; y :::; (:J(t),
J(t,a(t). y
< art).
Notice that
is a LIoe[ a, T)-Caratheodory function since 0', (3 are continuous a.e. on [0, T). Theorem 15.:3 implies that (2.9) has a solution
y E AC[oe([a, T), Rn). We now claim that aCt) :::; yet) :::; (:J(t) for t E [a, T).
First consider t E [a,81) where 81 is as described in (2.8). We show
yet) :::; (3(t) for t E [a,sl). Assume this is false. Then since yea) = ao :::;
(3(a) there exists Xl < X'2 E (a,81] such that y(xtl :::; /3(xtl, Y(X2) > (3(X2)
and yet) ~ (3(t) for t E (Xl,X2). Consequently
a contradiction. Thus y(t) :::; ;l(t) for t E [a,81] and a similar argument
yields yet) ~ aCt) for t E [a,81)' Thus
(2.10)
aCt) :::; yet) :::; (3(t) for t E [a,81).
This together with the fact that

solutions of (2.1) implies
0',
f3 are respectively lower and upper
O'(st):::; a(sl):::; y(81):::; /3(8 1):::; f3(st).
That is
(2.11)
Next consider t E [81,S'2)' Essentially the sauli" reasoning (since (2.11) is

true) establishes that aCt) :::; yet) :::; !3(t) for t E [Sl,S'2)' Continue this
process and so the theorem is established. 0
15.3. Impulsive Differential equations.
In this section various existence results are established for the impulsive
differential equation
(3.1)
y' = J(t, y) for a.e. t E [0, T), t =1= Ti(y(t,

{ y(t+) = Ji(y(t)) if t = Ti(y(t)), i = 1, ..... ,k,
yeO) = Yo.
178
We first present a result for scalar IDE's based on the notion of upper and
lower solutions. A function a E n6 is called a lower solution of (31 ) if
a/(t) ~ f(t,a(t for a.e. t E [O,T),

{ a(t+) ~ a(r) for all t E [O,T),
0(0) ~ Yo.
Similarly we define an upper solution of (3.1) by reversing the inequalities.
Theorem 15.5. Let f : [0, T)
X R -. R be a L]ocl0, T)-Caratheodory
function. Also let Ii : R ---- Rand Ti : R ~ R with Ti E Cl(R) for each
i = 1, ... , k. Assumf' the following conditions are satisfied:
(3.2)
(3.3)
(3.4)
Ti(YO) :j:.
for i = 1, ... , k;
there exists Q, (3 E nA respectively lower and upper solutions

of
(3.1) with a(t) ~ (i(t) for t E [O,T); alsoforany (t,x)
{
and i E {1,2, .. ,k} such that t = Ti(X), aCt) ~ x ~ (i(t)
we have a(t+) ~ Ii(x) ~ (3(t+);
there exist f > 0, ao ~ 0, bo > 1 such that for all tl < t2
(tt, t2 E (0, T and for all y E C[tl, t2J such that
Tj(y(t2 = t2 for some j and y(tt} E Uf=lIi(Ti-l(tl,
T(tl,t2,Tj,y) = f/12 f(t,y(tTj(y(tdt + Tj(Y(h
satisfies one of the following conditions
(i). T(tt,t2,Tj,Y) < t2,
(ii). T(tl,t2,Tj,y) > t2,
(iii). T(tl,t2,Tj,Y) ~ f + tl - aO(t2 - tl),
(iv). T(tt,t2,Tj,Y) ~ - f + tl + bo(t2 - tt};
and
(3 ..5 )
either Ti(X) :j:. Tj(X) for all x E Rand i:j:. j with

{ i,j E.{I, ..",k} or if Ti(X) = Tj(X) for some x E R
and ~:j:. J then I;(x) = Ij{x).
Then (3.1) has a solution y E nb with ott) ~ yet) ~ (3(t) for t E [0, T).
Remark. In (3.4), Ti- l denotes the inverse image.
PROOF: Let Yl E C[O, T) be a solution (guaranteed by theorem 1.5.4) of
the IVP,
(3.6)
y' = f(t, y) a.e. on [0, T),

yeO) = Yo,
179
with aCt) ~ YI(t) ~ j3(t) for t E
[0, T). Define
= Ti(YI(t) - t for t 2:: o.

=f 0 for i = l, ... ,k. If ri.l(t) =f
ri,l(t)
Now (3.2) implies ri,l(O)

0 on [O,T) for
i = 1, ... ,k (i.e. t =f Ti(Yl(t on [O,T) for i = 1, ... ,k) then the result of
the theorem follows i.e. YI is a solution of (3.1). It remains to consider
the case when ri,l(t) = 0 for some t and SOllle i. Now since ri,l(O)
0
for i = 1, .. , k and ri,l are continuous, there exists tl > 0 such that
ri1,l(td = 0 for SOllle it E {1, .. ,k} and rj,dt) =f 0 for all t E [O,td and
j = 1, .. , k. Assumption (:3.3) implies
t=
(3.7)
Now let Y2 E C[tl, T) be a solution (guaranteed from (3.7) and theorem
15.4) of the IVP,
(3.8)
Y' = J(t, y) a.e. OIl ttl, T),

y(tl) = Iil(YI(ttl),
with aCt) ~ Y2(t) ~ j3(t) for t E [t l , T). Observe that
Define
ri,2(t) = Ti(Y2(t - t for t 2:: t l .

If ri,2(t)
=f
0 on {tt,T) for all i = l, ... ,k then

Y
={
on [0, tl],
on (tI,T),
YI,
Y2,
is a solution of (:3.1). It remains to consider the case when there exists

t* > tl with ri,2(t*) = 0 for some i. Now since y~ = J(t, Y2) a.e. on
[tt, T) we have
t*
[Ti(Y2(t* - Ti(Y2(ttl)]
+ Ti(Y2(ttl)
= t* J(t, Y2(tT[(Y2(tdt + TiCY2(tt) = T(tl,t*,T;,Y2)'
itl
By assumption (3.4), if (i) or (ii) is satisfied, we get a contradiction (and

we are finished). If (iii) is satisfied then
180
If (iv) is satisfied then f" - t1 2: bo~ l ' Consequently

i * - i1
(3.9)
f}"
== { .
.{ --, - > lIun

-
1 + ao bo - 1
Now (3.9) together with the continuity of ri,'}. for i = 1, .. , k implies that
there exists t2 2: t1 + ('" such that l"i,'}.(t) f. 0 for all t E (i1, t2) and
i = 1, .. ,k with ri2,'}.(t'}.) = 0 for SOUle i'}.. Assumption (3.3) implies
Let Y3 E C[i 2, T) be a solution of the TVP,
(3.1O)
with a(t) :::; Y3(t) :::; f3(t) for t E [t'}., T). Define
If ri,3{t)
f.
0 on (t')., T) for all i = 1, ... , k then
is a solution of (3.1). It remains to consider the case when there exists

t> t2 with ri.3(I) = 0 for some i. Then, as above, t = T(t2, t, Ti, Y3)' By
assumption (3.4), if (i) or (ii) is satisfied, we get a contradiction. If (iii) or
(iv) is satisfied then t - t'}. 2: f*. Hence there exists t3 2: t'}. + (* such that
ri,3(t) f. 0 for all t E (t'}.,t3) and i = 1, .. ,k with ri 1,3(t3) = 0 for some
t3
Continue this process and the result of the theorem follows. Observe
that if T < 00 the process will stop after a finite number of steps. 0
Remark. In assumption (3.4) we may replace "for all Y E C[t1,t2]" by "for
all Y E C[tl,t'}.] which satisfy a(t):::; y(t):::; i3(t), t E [t 1 ,t'}.]". Also, we can
replace f by f( i]) with f an appropriate function.
Examples.
(i). A problem with fixed moments satisfies (:3.4), indeed suppose T;(X) = r;
for each x E R with t1 < t'}. < .... < tl,;. Wf' clailll that (;~.4)( iii) is satisfied.
181
To see this, sino' fl < t'2 < .... < tk notice there exists ( >
with
distCt;,tj) ~ ( for all i,j E {1, .. ,k} and i 'f j. Notice also that Tj(X) =
so
T(t I ,t'2,Tj,Y) = Tj(y(td) = tj;
here t I ,t'2 E (O,T), tl < t'2 and y E (,[t I ,t'2l with Tj(y(t 2
j and y(td E Ut=lli(Ti-I(td). Notice
tj
= t'2
and tl E
This follows since

-I(
T
I
)_
tl -
= t'2
for some
{fl, ... , td
{0,
if tl 'f r;,
.
R, If tl = ti,
and so y(tJ) E U7=lli(Ti-l(id) if and only if il E {fl, ... ,td. Thus if

tl < t2 with t2 = tj and tl = r; for SOlIle i then j > i (since tl < t2)
and so
T(t 1 ,t2,Tj,Y) = ij ~ tj_1 + f ~ r; + f = t] + L
Consequently (3.4)(iii) is satisfied.
(ii). Suppose for each J; E R we have Ti(X) = T(X) and l;(x) = lex) for
all i == 1, ... ,k. In addition assume T: R ---+ R is increasing and l(x)::; x
for all x E R. Also suppose J( t, x )T'( x) < 1 for all t E (0, T) and x E R
(need only for all x with art) ::; x ::; OCt), t E (O,T and j E {l, .. ,k}.
We claim that (3.4)(i) is satisfied.
To see this notice
here tl,t2 E (O,T), t] < t'2 and y E (,[t l ,t'2l with T(y(t2 == t2 and
yetI) E I(T-I(ttl). Now since T is increasing y(ttl E I(T-](tI) means
y(tl) == I(T-I(tI)). Also since l(x)::; x we have
Thus
T(t],t'2,T,y) < t'2 - i l
+ T(y(ttJ)::; t2
and so (3.4)(i) is satisfied.
Remark. Notice we could replace "l(x) ::; ;I: and f(t,x)r'(x) < 1" by
"l(x) < x and J(t,X)T'(X)::; 1" in the above example.
182
We now establish a result when our nonlinearity f satisfies a growth

condition of Wintner type. First wp prove the result for scalar equations
and then we state the analogup for systems of equations (since the proof is
essentially the same we will omit it).
Theorem 15.6. Let f : [0, T) X R
R be a LloclO, T)-Caratheodory
function. Also let Ii : R -. Rand Ti : R - R with Ti E (.'1 (R) for each
i = 1, ... ,k. Suppose (3.2),(3.4) and (3.5) an ::;atisfied. In addition aS8U1nf
the following conditions hold:
-7
(:3.11)
( 3.12)
Ii : R
-7
R is bounded on bounded sets, i
= 1, ... , k;
If(t, y)1 ~ q(t )'ljJ( Iyl) for almost all t E [0, T) with
1/) : [0,00) -. (0,00) a Borel measurable function
with
E LLJo, 00) and q E LLcl0, T);
(3.13)
1/' is nondecl'easing;
and
(3.14 )
Then (3.1) has a solution y E
PROOF: Let Y1 E
(3.6) with
(3.15)
qo, T)
IYl (t)1
dx
'V)(x)
OCi
nb.
be a solution (guaranteed by theorem 15.2) of
./-1
(1 q(x) dX)
t
where
J(z)=
Define
= 00.
for t E [0, T)
dx
-.
Iyo I 1/'( J:)
2
= Ti(ydt)) - t for
for all i = 1, ... ,k. wp
t 2 0.
where Qo
= ./-1
1'i,dt)
If ri,dt) ion [O,T)

arp finishpd. It remains to
consider the case when there exists tl > such that 1'i 1 ,1 (tIl = for SOUle
ZI with rj,1(t) ifor all t E [0, t1) and j = 1, ... , k. Observp that
IIi 1(Yl(tJ))1
max
[-Qa,QoJ
IIi 1 (:r)1
183
(fotl q(x)dx).
Jo
Let Y2 E C[tt, T) be a solution (guaranteed by theorem 15.2) of (3.8).

Define
ri,2(t) = Ti(Y2(t)) - t for t 2: tl.
Essentially the same reasoning as in theorem 15.5 (theorem 1.5.2 is used
instead of theorem 15.4) guarantees the result. 0
Remark. Assumption (3.1:3) could be removed in theorem 15.6 if we use a
result in [9].
We now state the analogue of theorem 1.5.6 for systems of first order
IDE's.
Theorem 15.7. Let f : [0, T) x R" -.. R" be a Lloc[O, T)-Caratheodory
function. Also let Ii : R" -> R" and Ti : R" ---. R with Ti E CI(Rn) for
each i = 1, ... , k. Suppose the conditions (3.2), (:3..5), (3.11), (3.12), (3.13)
and (3.14), with R replaced by Rn, are satisfied. In addition assume
(3.16)
there exist f > 0, ao 2: 0, bo > 1 8uch that for all tl < t2

(tI, t2 E (0, T)) and for all y E C[tI' t2] such that
Tj(y(t2)) = t2 for sonu~ j and y(tl) E Uf=IIi(T;I(tl)),
T(tt,t2,Tj,Y) = ftt12 f(t, y(t).'VTj(y(t) dt Tj(y(ttJ)
satisfies one of the following conditions
(i). T(tI,t2,Tj,Y) < t2,
(ii). T(tl,t 2,Tj,y) > t2,
(iii). T(t I ,t2 ,Tj,y) 2: f + t1 - aO(t2 - tIl,
(iv). T(t I ,t2,Tj,y)::; - f + tl + bo(t2 - td.
Then (3.1) has a solution y E
f26.
Remark. The 'V in (3.16) denotes the gradient and . denotes the usual
inner product in R".
184
References.
[1]. J. Dugundji and A. Granas, Fixed Point Theory, VoLl, Monografie

Matematyczne, PWN, Warsaw, 19R2.
[2]. M. Frigon, Application de la theorie de la transversalite topologique
Dissertationes Math. 296(1990), 1-79.
a des
[3]. M. Frigon and D. O'Regan, Impulsive differential equations with variable times, Nonlinear Anal., 26(1996), 1913-1922.
[4]. S. Kaul, V. Lakshmikantham and S. Leela, Extremal solutions, comparison principle and stability criteria for impulsive differential equations
with variable times, Nonlinear .4.nal., 22( 1994), 1263-1270.
[5]. V. Lakshmikantham, D.D. Bainov and P.S. Simeonov, Theory of impulsive differential equations, World Scientific Press, Singapore, 1989.
[6]. V. Lakshmikantham, Trends in the theory of impulsive differential
equations, Proc. Int. Conf Theory Appl. Ditf. Eg., Ohio University
Press, Athens, Ohio, 1988, 76-87.
[7]. V. Lakshmikantham, N. S. Papageorgiou and J. Vasundhara, The
method of upper and lower solutions and monotone technique for impulsive
differential equations with variable moments, Appl. Anal., 15( 1993), 4158.
[8]. J.W. Lee and D. O'Regan, Topological transversality: applications to
initial value problems, Ann. Polon. Math., 48(1988), 31-36.
[9]. J .W. Lee and D. O'Regan, Existence principles for differential equations
and systems of equations, Topological methods in Ditf. Egns. and Inclusions (edited by A. Granas and M. Frigon), NATO ASI Series C, Kluwer
Acad. Publ., Dordrecht, 1995, 239-2R9.
[10]. X. Liu, Nonlinear boundary value problems for first order impulsive
differential equations, Appl. Anal. 36( 1990), 119-130.
185
Chapter 16. Differential equations in abstract spaces.

16.1. Introduction and preliminaries.
In this chapter, we are concerned with the initial value problem:
(1.1 )
y'(t)
y(O)
= f(t,Y,(t))
= a E E;
t E [O,Tj
where E is a real Banach space and f : [0, Tj X E --. E has a decomposition

f = 9 + h with 9 and h CaratModory functions satisfying respectively, a
compactness and Lipschitz assumptions. Our results rely on Krasnoselskii

fixed point theorem for contraction plus compact mappings and don't use
homotopy arguments. It is worth remarking here that the periodic problem
could also be discussed in this setting (we leave this as an exercise). Our
main existence principle is obtained in section 16.2. This result will be
used in section 16.3 to obtain more applicable existence results. More
precisely, in section 16.3, we give existence theorems of Wintner type. Also
in section 16.3, existence theorems are obtained under an assumption which
is equivalent to an assumption of existence of upper and lower solutions to
(1.1) in the scalar case. No growth condition is assumed.
Throughout E is a real Banach space with norm II '11. In case E = H
a Hilbert space, we denote the inner product by (, ) so that lIull2 = (u, u)
for u E H. We denote by C([O, Tj, E) the space of continuous functions
u: [O,Tj ----> E. We set
Ca([O,Tj, E) = {u E C([O,Tj,E): u(O) = a}.
Jt
Let u : [0, Tj -+ E be a measurable function. By

u( t) dt, we mean
the Bochner integral of u, assuming it exists. See [3j for properties of
the Bochner integral. We define the Sobolev classe W 1 1 ([0, T], E) by the
space of continuous functions u such that there exists v E Ll([O, T], E) with
u(t) - u(O) = J~ V(8) d8, for all tin [0, Tj. Notice that if u E W1.1([0, Tj, E)
then u is differentiable almost everywhere on [O,Tj, u' E L 1 ([0,Tj,E), and
n(t) - u(O) = J~ U'(8) d8 for t in [0, Tj. Also, if E is a reflexive Banach
space, u E Wl.1 ([0, Tj, E) if and only if 'U is absolutely continuous. By
a solution to (1.1), we mean a function 'U E W1.1([0,T],E) satisfying the
differential equation (1.1) a.e. on [O,Tj and y(O) = a.
A function 9 : [0, T] X E - E is a Camthcodo1'Y function if: (1) the
map t --> g(t, z) is measurable for each:; in E; (2) the map z ---. g(t, z)
is continuous for almost all t in [0, Tj; (:1) for each l' > 0, there exists
186
hr E L1([0, Tj, R) such that Ilzll ::; r implies IIg(t. z)II ::; hr(t) for almost all
tin [0, Tj.
For sake of completness, we state the Krasnoselskii fixed point Theorem,
see [7, 12j for a proof.
Theorem 16.1. Let C Ix: a nonempty closed convex subset of a Banach
space E. Suppose T 1 , T2 map C into E and
(i). T 1 (C)+T2(C)CC;
(ii). T 1 : C ---l- E is continuou8 and compact;
(iii). T 2 : C -+ E is a contmction mapping.
Then there exists y E C with T 1 (y)
+ T2(y) = y.
16.2. General existence principle.

In this section, we establish an existence result for the initial value
problem (1.1), which will be used to prove forthcoming theorems.
Theorem 16.2. Assume f : [0, Tj X E ~ E has the decomposition f =
g* + h* with g* and h* Camtheodory functions such that
(2.1)
for each t E [0, Tj, the set

g*(s,u(s)) ds : u E Ca([O,Tj,E)}
is relatively compact;
{fci
and
ther'e exists q* E Ll([O,TJ,R) with
{ IIh*(t, u) - h*(t, '0)11 ::; q*(t)lIu - vII
for a.e. t E [0, Tj and all u, vEE.
(2.2)
Then ( 1.1) has a solution.

PROOF: Let us endow C([O, TJ, E) with the norm
IlullQ
max Ile-Q(t)u(t)1I
tE[O.T]
where
Q(t)
= lot q*(s)ds.
A solution to (1.1) is a fixed point of the operator T: Ca([O,Tj,E)---lCa([O, Tj, E) defined by:
(Ty)(t)
ltf~,y(s))d.~=a+ lth*(8,y(s))ds + l
(T2Y)( t) + (Tl y)(t),
a+
-
187
tg *(8,Y(S))dS
where (Tly)(t)
I~g*($,y(8d.s and (T2y)(t)
a + I~ h*(s,y(sds.
Since g* and h* are Caratheodory, Tl and T2 are well defined and continuous. Condition (2.1) together with the Arzela-Ascoli Theorem imply that
Tl is compact. Also, T2 is a contraction mapping since
max Ile-Q(t) ft [h*($, u(s -
10
tE[O,Tj
<
h*(,~, vCs] ds II
max Ile-Q(t) ft q*(s)e Q(s)e- Q(s)lIu(s) - v(s)1I ds
10
tE[o.TI
II
< Ilu - vllQ max Ile-Q(t)[eQ(t) - 1]11

tE[O,Tj
(1 - e-Q(T))
lIu - vllQ == kollu - vllQ
with ko < 1. Then, Krasnoselskii's fixed point Theorem (Theorem 16.1)

gives the existence of a fixed point to T i.e. a solution to (1.1). 0
Remarks. (a). Condition (2.1) is satisfied if g* is a [( -Caratheodory function i.e. g* is Caratheodory and satisfies property- [( (see [4]).
{
for each r > 0, there exist a nonnegative function

'T]r E Ll([O, TJ, R) and a compact set [(1' in E such that
IIzll ::; r implies g*(t,z) E 11r(t)[(r for almost all t in [O,T].
(b). A more general version of this theorem could be given in terms of

measure of non-compactness and condensing mappings.
16.3. Existence results.

W(> first present an existence result if E
= H, a real
Hilbert space.
Theorem 16.3. Assume f : [0, Tj X H ----> H has the decomposition f =

9 + h with 9 and h Camtheodory functions such that
there arc r E L 1 ([0,Tj,[0,00 and 'P: [0,00) -. (0,00)

{ a Borel measumble function such that
(y,/(t,y ::; r(tj'P(lIyll) for a.t:. t E [O,Tj
and all y E H, and It r( 8 j ds < ~;:II <p(x) dx.
(3.1 )
= Nail <p(x} dx and notice I : [ lIall, 00) ....... [0,00 j

Define M = I-I <It r( s ) ds j and assumt:
Let I( z)
ing.
(3.2)
i.~
strictly increas-

{ {f~ g( s, u( s ds : u E Ca([O, Tj, H) with lIu( s)1I ::; M
for all oS E [0. Tj} is I'elatively compact;
188
(3.3)
there exi8t8 q E L 1 ([0,Tj,R) with

{ IIh(t, u) - h(t, 11)11 ~ q(t)lIu - 1111
for a.e. t E [0, T] and all U,l1 E H
with
Ilull, IIvll
M.
PROOF: Consider the initial value problem:

(3.4)
with gl(t,y)
y'(t) = g1(t,y(t))
yeO) = a;
+ hl(t,y(t))
t E [O,T]
= g(t,p(y)), h1(t,y) = h(t,p(y and

={
p(y)
y,
if
t!r,
Ilyll
~ M
Ilyll > M
if
is the radial retraction of H onto B(O, M)

check that p is nonexpansive i.e.
= {y : Ilyll
M}. It is easy to
This together with (3.3) implies that

Ilhl(t,u) - h 1 (t,v)11
IIh(t,p(u)) - h(t,p(v)JII ~ q(tJllp(u) - p(v)11
<
q(t)llu - 1111;
so (2.2) is satisfied with h* = hI and q* = q. Notice as well that g* = gl

satisfies (2.1). Consequently, Theorem 16.2 implies that (3.4) has a solution
y. The following Lemma 16.4 applied with R = Iiall, 'IjJ(x) = <p(x)jx,
and z(t) = lIy(t)1I implies that Ily(t)11 ~ M for all t E [0, T]. Indeed, if
Ily(t)11 > M for some t, then there exists [to, tl] C [0, T] such that lIyll E
W 1,1([to,td,R), lIy(toJIl = lIall, ~y(tdll = M and Iiall ~ Ily(tJII ~ M
for t E [to, tl]' If ft~l res) ds = fo res) ds, ret) = 0 a.e. on [0, T]\[to, tl]'
Thus Ily(t)11 = M for t > tt and lIy(t)1I = lIall for t < to, a contradiction.
Otherwise, by Lemma 16.4, Ily(t)11 < M for t E [to, ttl, a contradiction.
Consequently, y is a solution to the original problem (1.1). 0
Lemma 16.4. Let R 2: 0, r E L1 ([0, T], [0,00)) and 1/): [0,00)
a Borel function 8uch that
T
res) d8
u
<
100 -1- dx.

R
189
~(x)
--+
(0,00) be
Let Mo be such that It r(s)ds = 1%0 1l)x) dx. Then for any [to,t I ] C [O,T]
and any z in wl,1([to, td, [0, 00)) with z'(t) ::::; r(t)'ljJ(z(t for a.e. t E
[to, it] and z(tv :S R, we ham: z(t) :S Mo f01' all t E [to, tI]' Moreover, if
It~l r(s)ds < 10 r(s)ds, zIt) < Mo f01' all t E [to,td
PROOF: Since z'(t) ~ r(t)'ljJ(z(t)) for a.e. t E [to,td and z{to) ~ R, for
any t E [to,tI] such that z(t) > R. then' exists to E [to,t] with z(to) =
R. Dividing by 'ljJ(z(t), integrating from to to t, and using the change of
variable formula give:
(t)
it
1
--dx:S
r(s)ds~
W( x)
to
Therefore z(t) :S Mo, and z(t)

completes the proof. 0
iT 1'(s)ds= MO
R
< Mo if
No 1'(s) ds
1
-(-)dx.
'ljJ x
< It 1'(s) ds. This
We next deriV' an existence result for a real Banach space E.
Theorem 16.5. Assume f : [0, T] X E ~ E has the decomposition f

with 9 and h Camtheod01'Y functions such that
(3.5 )
= g+h
there exist a nondec7"f:asing Bard measumble function

t.p:
[0,00) --.. (0,00), and l' E LI([O,T],[O,oo) such that
{
Ilf(t, y)11 :S 1'(t)t.p(llyll) for a.e. t E [0, T] and all
y E E, and loT 1'(s)ds < ~;:;I 'P(x) dx.
Let J(z)
= liIail 'P(x) dx
Define M =
and notice J :
[llall,oo)
---+
[0,00) is strictly in-
creasing.
J-l(Jt r(s)ds) and assume (3.2) and (3.3) hold
with H = E. Then ( 1.1) has a solution.
PROOF: Consider the initial value problem (3.4). First, notice that in that
case, the radial retraction p is Lipschitz and not necessarily nonexpansive.
More precisely,
This together with (:3.5) implies
Ilht{t,u) - ht{t,v)ll:S q(t)lIp(u) - p(v)li::::; 2q(t)llu - vii;

so (2.2) is satisfied with h* = hI and q* = 2q. Notice as well that g* = gl
satisfies (2.1). Consequently, Theorem 16.2 implies that (3.4) has a solution
y. For such a y,
Ily(t)11
Ily(O) +
r y'(s)dsll ~ Iiall + Inr Ily'(s)11 ds == p(t).

t
Jo
190
Clearly, p(t) is absolutely continuous and p'(t) ::::

Since <p is non decreasing,
P'(t)::::
Ily'(t)1i
Ily'(tlll almost everywhere.
~ r(t)<p(liy(t)II) ~ r(t)<p(p(t
almost everywhere. Lemma 16.4 applied with R :::: Ilall,1/J(x) :::: <p(x), and
z :::: p implies that liy(t)1I ~ p(t) ~ M for all t E [0, Tj and consequently
that y is a solution to the original problem (1.1). 0
Next we establish existence results without a growth restriction. Let
E :::: H be a real Hilbert space.
Theorem 16.6. Assume J : [0, Tj X H -+ H has the decomposition J ::::
g + h with g and h Camtheodory functions such that
(3.6)
there exist v E Wl,l([O,Tj,H) and ME WI, 1 ([O,T], [0,00))

such that (y - v(t),J(t,y) - v'(t) ~ M(t)M'(t) for a.e.
t E [0, Tj and all y E H with Ily - v(t)1I :::: M(t);
J(t,v(t:::: v'(t) a.e. on {t E [O,Tj: M(t):::: o};
lIa - v(O)1I ~ M(O);
{ {f~g(s,u(sds: U E Ca([O,T],H) with
lIu(s) - v(s)1I ~ M(s) f07' all s E [O,T]}
is l-elatively compact;
(3.7)
there exists q ELI ([0, Tj, R) with

IIh(t,utlh(t,U2)1i ~ q(t)IIUI - u211
{
(3.8)
for a.e. t E [0, Tj and all Ui E H with

lIu; - v(t)1I ~ M(t), i :::: 1,2.
Then (1.1) has a solution such that lIy(t) - v(t)1i ~ M(t) for all t E [O,Tj.
Remark. Observe that in the scalar case, if 0' ~ (3 E Wl,1([O,Tj,R) are
respectively lower and upper solutions to (1.1) i.e.
0'(0) ~ a ~ /3(0), O"(t) ~ f(t,O'(t, {3'(t) ~ J(t,;3(t,
then v:::: (0'
+ ;3)/2 and M
:::: (tJ - 0')/2 satisfy (3.6).
PROOF: Consider the initial value problem:

(3.9)
y'(t):::: h(t,y(t
y(O) :::: a:
191
t E [O,Tj
with h(t,y)
= f(t,p(t,y)
_{y,
and
if Ily - v(t)11 ::; M(t)

M(t) (1I~::::;g!lI) + vet), if Ily - v(tlll > M(t)
pet, y) -
is the radial retraction of H onto B(v(t), M(t = {y : lIy - v(t)11 ::; M(t)}.
It is easy to check that p is continuous and nonexpansive i.e.
IIp(t,yd - p(t,Y2)11 ::; IIYI - Y211
for all YbYZ E H;
and condition (2.2) i,; satisfied with h*(t,y) = h(t,p(t,y and q* = q.

Notice as well that g*(t,y) = g(t,p(t,y satisfies (2.1). Consequently,
Theorem 16.2 implies that (3.9) has a solution y. We now claim that
Ily(t) - v(t)11 ::; M(t) for all t E [O,T), and consequently, y is a solution to
the original problem (1.1).
Suppose there exists tl E (0, T) with Ily(tI) - V(tl )11 > M(tl). Since
IIY(O) - v(O)11 ::; M(O), there exists to E [0, tIl such that Ily(to) - v(to)1I =
M(to) and Ily(t) - v(t)11 > M(t) for t E (to, td. Now, (:3.6) implies that a.e.
on (to, tJ),
> (p(t,y(t - v(t),h(t,y(t - V'(t))
M(t)M'(t)
M(t)
.
I
Ily(t) _ v(t)11 (y(t) - vet), h( t , yet)) - v (t).
Thus
(3.10)
M '(t) > (y(t) - v(t),h(t,y(t - V'(t)

Ily(t) - v(t)11
a.e. tE(to,tl).
On the other hand, Ily(t) - v(t)11 is differentiable a.e. on (to,lIl and

(3.11 )
Ily(tJ) - V(t1 )11 = Ily(to) - v(to)11
+ ftl
to
(y(t)-V(t1,yl(t)-V I (t)
lIy(t -v(t)1I
Combining (:3.10) and (:3.1 1) gives
M(td < Ily(tI) - v(tI)II::; M(to)

a contradiction.
tl
to
dt
M'(t)dt = M(tIl,
Corollary 16.7. AS8umc f : [0, T] X H -+ H has the decomposition f

9 + h with 9 and h Camtheodory functions SItch that
(3.12)
thcn cxists a positivf constant.. M 2: Iiall such that

{ (y,f(t,y)::; for a.f. t E [O,T] and all y E H
with Ilyll = M;
192
and satisfying (3.2) and (3.;~). Then (1.1) has a .solution such that Ily(t)11 ::;
M for' all t E [0, T].
We next give an existence result for a real Banach space E.
Theorem 16.8. Assume f : [0, T] X E ~ E has the decomposition

with g and h Caratheodor'y functions such that
= g+h
there exist VEWl,l([O.T],E) and MEW1,1([0,T],[0,oo))

(313) { a nondecreasing function such that Iia - v(O)11 ::; M(O) and
Ilf(t,y) - v'(t)11 ::; M'(t) for a.e. t E [O,T] and all y E H
.
with Ily - v(tlll = M(t);
and satisfying (:3.7) and (3.8) with H = E. Tlwn (1.1) has a solution sl1ch
that Ily(t) - v(t)11 ::; M(t) for all t E [0. T].
PROOF: Consider the initial value problem (:3.9). As before, Theorem 16.2
implies the existence of a solution y to (3.9).
Suppose there exists h E (0, T] with Ily(tJ}-v(h )11 > M(td. The initial
conditions implies the existence of to E [0, tl) such that Ily(to) - v(to)11 =
M(to) and Ily(t) - v(t)11 > M(t) for t E (to, tl)' By condition (3.13), we get
M(td < Ily(ttJ- v(tIlII
Ily(to) - veto)
< M(to)
It
. to
+ tl [y'(t) Jto
v'et)] dt II
M'(t)dt = M(td,
a contradiction. Therefore Ily(t) - n(tlll ::; M(t) for all t E [O,T] and thus
we have shown that y is a solution to (1.1). 0
Remark. It is possible also to give an analogue version of theorem 16.6 in

the Banach space setting.
References.
[lJ. R.R. Akhmerov, M.1. Kamenskii, A.S. Potapov, A.E. Rodkina and
B.N. Sadovskii, Measures of noncompactness and condensing operators,
Birkhiiuser, Basel, 1992.
[2J. K. Deimling, Ordinary differential equations in Banach spaces, Springer
Verlag, New York/Berlin, 1977.
[3J. N. Dunford and J.T. Schwartz, Linear operators part.!: General theory,
Interscience, New York, 1967.
[4J. M. Frigon and .J.W. Lee, Existence principles for CaratModory differential equations in Banach spaces, Topological Methods in Nonlinear
Analysis, 1,(1993), 91-106.
[5J. M. Frigon and D. O'Regan, Existence results for initial value problems
in Banach spaces, Ditf. Eqns. and Dynamical Systems, 2(1994), 41-48.
[6J. A. Granas, R.B. Guenther and .l.W. Lee, Some general existence principles in the CaratModory theory of nonlinear differential systems J. Math.
pures et appl. ,70(1991), Hi3-196.
[7J. M.A. Krasnoselskii, Two remarks on the method of successive approximations, Uspehi Mat. Nauk., 10(1955), 12:J-127.
[8J. V. Lakshmikanthan and S. Leela, Nonlinear differential equations in
abstract spaces, Pergamon, New York, 19R1.
[9J . .l.W. Lee and D. O'Regan, Existence results for differential equations
in Banach spaces, Commentat. Math. Univ. Carol. ,34(199:3), 239-251.
[10J. J.W. Lee and D. O'Regan, Topological transversality. Applications

to initial-value problems, Ann. Polonici Math., 48(1988),247-252.
[I1J. R.H. Martin, Nonlinear operators and differential equations in Banach

spaces, Wiley, New York, 1976.
[12J. D.R. Smart, Fixed point theorems, Cambridge University Press, 1980.
[13J. A. Wintner, The nonlocal existence problems for ordinary differential
equations, Am. J. Math., 67( 194.5), 227-284.
194
Index.
Arzela-Ascoli
basis
35
Bernstein-Nagumo
C(I)
41
cm(I)
Caratheodory
105
Contraction
compact
completely continuous
conjugate exponents
eigenvalue
107
39, 116
eigenfunction
essential
116
forbidden value
62
Fourier series
11 7
Frechet space
157
Fmi Pera fixed point theorem

half line
156
homotopic
inessential
invertible problems
35
impulsive equations
174
Krasnoselskii fixed point theorem
LP(T)
L=(I)
187
2
2
LP-Caratheodory
LP-Lipschitz
14
I!)
Landesman Laser condition

Lipschitz
limit circle
22, 13:3
12
116
195
lower solution
15, 19, 51, 165, 179
lower surface
51
maximum principle
47
Minkowski functional
moments
174
noninvertible problem
38
nonlinear alternative
4, 7
nonresonance
116
orthonormal set
117
PA(X,C)
5
Parseval's equality 117
Peano theorem 12
periodic 22
Picard Lindelof 12
positone 58
radial retraction 90
resonance 138
retraction 8
Riesz compactness criteria 118
Schauder fixed point theorem 2
Schauder Tychonoff fixed point theorem
semi norms 157
semi-positone 67
singular 82
Sturm Liouville 41
topological transversality 4, 6
Tychonoff 4
upper solution 18, 19, 51, 165, 179
upper surface 51
Wm,P(I)
weak limit
weakly closed
28
28
196
Other Mathematics and Its Applications titles of interest:
A.M. Samoilenko: Elements of the Mathematical Theory of Multi-Frequency

ISBN 0-7923-1438-7
Oscillations. 1991,314 pp.
Yu.L. Dalecky and S.V. Fomin: Measures and Differential Equations in InfiniteDimensional Space. 1991,338 pp.
ISBN 0-7923-1517-0
W. Mlak: Hilbert Space and Operator Theory. 1991,296 pp. ISBN 0-7923-1042-X
N.la. Vilenkin and A.V. Klimyk: Representation of Lie Groups and Special
Functions. Volume 1: Simplest Lie Groups, Special Functions, and Integral
ISBN 0-7923-1466-2
Transforms. 1991,608 pp.
N.la. Vilenkin and A.V. Klimyk: Representation of Lie Groups and Special
Functions. Volume 2: Class I Representations, Special Functions, and Integral
ISBN 0-7923-1492-1
Transforms. 1992,630 pp.
NJa. Vilenkin and A.V. Klimyk: Representation of Lie Groups and Special
Functions. Volume 3: Classical and Quantum Groups and Special Functions. 1992,
650 pp.
ISBN 0-7923-1493-X
(Set ISBN for Vols. 1, 2 and 3: 0-7923-1494-8)
K. Gopalsamy: Stability and Oscillations in Delay Differential Equations of
Population Dynamics. 1992, 502 pp.
ISBN 0-7923-1594-4
N.M. Korobov: Exponential Sums and their Applications. 1992,210 pp.
ISBN 0-7923-1647-9
Chuang-Gan Hu and Chung-Chun Yang: Vector-Valued Functions and their
ISBN 0-7923-1605-3
Applications. 1991,172 pp.
Z. Szmydt and B. Ziemian: The Mellin Transformation and Fuchsian Type Partial
Differential Equations. 1992, 224 pp.
ISBN 0-7923-1683-5
L.I. Ronkin: Functions of Completely Regular Growth. 1992, 394 pp.
ISBN 0-7923-1677-0
R. Delanghe, F. Sommen and V. Soucek: Clifford Algebra and Spinor-valued
Functions. A Function Theory of the Dirac Operator. 1992,486 pp.
ISBN 0-7923-0229-X
A. Tempelman: Ergodic Theorems for Group Actions. 1992, 400 pp.
ISBN 0-7923-1717-3
D. Bainov and P. Simenov: Integral Inequalities and Applications. 1992, 426 pp.
ISBN 0-7923-1714-9
I. Imai: Applied Hyperfunction Theory. 1992, 460 pp.
ISBN 0-7923-1507-3
Yu.I. Neimark and P.S. Landa: Stochastic and Chaotic Oscillations. 1992,502 pp.
ISBN 0-7923-1530-8
H.M. Srivastava and R.G. Buschman: Theory and Applications of Convolution
ISBN 0-7923-1891-9
Integral Equations. 1992,240 pp.
A. van der Burgh and J. Simonis (eds.): Topics in Engineering Mathematics. 1992,
266 pp.
ISBN 0-7923-2005-3
F. Neuman: Global Properties of Linear Ordinary Differential Equations. 1992,
320 pp.
ISBN 0-7923-1269-4
A. Dvurecenskij: Gleason's Theorem and its Applications. 1992, 334 pp.
ISBN 0-7923-1990-7
D.S. Mitrinovic, J.E. Pecaric and A.M. Fink: Classical and New Inequalities in
ISBN 0-7923-2064-6
Analysis. 1992,740 pp.
H.M. Hapaev: Averaging in Stability Theory. 1992,280 pp.
ISBN 0-7923-1581-2
S. Gindinkin and L.R. Volevich: The Method of Newton's Polyhedron in the

ISBN 0-7923-2037-9
Theory ofPDE's. 1992,276 pp.
Yu.A. Mitropolsky, A.M. Samoilenko and D.I. Martinyuk: Systems of Evolution
Equations with Periodic and Quasiperiodic Coefficients. 1992,280 pp.
ISBN 0-7923-2054-9
LT. Kiguradze and T.A. Chanturia: Asymptotic Properties of Solutions of Nonautonomous Ordinary Differential Equations. 1992,332 pp. ISBN 0-7923-2059-X
V.L. Kocic and G. Ladas: Global Behavior of Nonlinear Difference Equations of
Higher Order with Applications. 1993,228 pp.
ISBN 0-7923-2286-X
S. Levendorskii: Degenerate Elliptic Equations. 1993,445 pp.
ISBN 0-7923-2305-X
D. Mitrinovic and J.D. Keckic: The Cauchy Method of Residues, Volume 2. Theory
and Applications. 1993, 202 pp.
ISBN 0-7923-2311-8
R.P. Agarwal and P.J.Y Wong: Error Inequalities in Polynomial Interpolation and
ISBN 0-7923-2337-8
Their Applications. 1993,376 pp.
A.G. Butkovskiy and L.M. Pustyl'nikov (eds.): Characteristics of DistributedISBN 0-7923-2499-4
Parameter Systems. 1993, 386 pp.
B. Stemin and V. Shatalov: Differential Equations on Complex Manifolds. 1994,
504 pp.
ISBN 0-7923-2710-1
S.B. Yakubovich and Y.F. Luchko: The Hypergeometric Approach to Integral
Transforms and Convolutions. 1994, 324 pp.
ISBN 0-7923-2856-6
C. Gu, X. Ding and c.-c. Yang: Partial Differential Equations in China. 1994, 181
pp.
ISBN 0-7923-2857-4
V.G. Kravchenko and G.S. Litvinchuk: Introduction to the Theory of Singular
Integral Operators with Shift. 1994,288 pp.
ISBN 0-7923-2864-7
A. Cuyt (ed.): Nonlinear Numerical Methods and Rational Approximation II. 1994,
446 pp.
ISBN 0-7923-2967-8
G. Gaeta: Nonlinear Symmetries and Nonlinear Equations. 1994,258 pp.

ISBN 0-7923-3048-X
V.A. Vassiliev: Ramified Integrals, Singularities and Lacunas. 1995,289 pp.
ISBN 0-7923-3193-1
NJa. Vilenkin and A.U. Klimyk: Representation of Lie Groups and Special
Functions. Recent Advances. 1995,497 pp.
ISBN 0-7923-3210-5
Yu. A. Mitropolsky and A.K. Lopatin: Nonlinear Mechanics, Groups and Symmetry. 1995,388 pp.
ISBN 0-7923-3339-X
R.P. Agarwal and P.Y.H. Pang: Opial Inequalities with Applications in Differential
and Difference Equations. 1995,393 pp.
ISBN 0-7923-3365-9
A.G. Kusraev and S.S. Kutateladze: Subdifferentials: Theory and Applications.
1995,408 pp.
ISBN 0-7923-3389-6
M. Cheng, D.-G. Deng, S. Gong and c.-C. Yang (eds.): Harmonic Analysis in
China. 1995,318 pp.
ISBN 0-7923-3566-X
M.S. Livsic, N. Kravitsky, A.S. Markus and V. Vinnikov: Theory of Commuting
ISBN 0-7923-3588-0
NonselfadJoint Operators. 1995,314 pp.
A.1. Stepanets: Classification and Approximation of Periodic Functions. 1995,360
pp.
ISBN 0-7923-3603-8
C.-G. Ambrozie and F.-H. Vasilescu: Banach Space Complexes. 1995,205 pp.
ISBN 0-7923-3630-5
E. Pap: Null-Additive Set Functions. 1995,312 pp.
ISBN 0-7923-3658-5
c.J. Colboum and E.S. Mahmoodian (eds.): Combinatorics Advances. 1995, 338
pp.
ISBN 0-7923-3574-0
V.G. Danilov, V.P. Maslov and K.A. Volosov: Mathematical Modelling of Heat
and Mass Transfer Processes. 1995,330 pp.
ISBN 0-7923-3789-1
A. Laurincikas: Limit Theorems for the Riemann Zeta-Function. 1996, 312 pp.
ISBN 0-7923-3824-3
A. Kuzhel: Characteristic Functions and Models of Nonself-AdJoint Operators.
1996,283 pp.
ISBN 0-7923-3879-0
G.A. Leonov, I.M. Burkin and A.1. Shepeljavyi: Frequency Methods in Oscillation
Theory. 1996,415 pp.
ISBN 0-7923-3896-0
B. Li, S. Wang, S. Yan and C.-C. Yang (eds.): Functional Analysis in China. 1996,
390 pp.
ISBN 0-7923-3880-4
P.S. Landa: Nonlinear Oscillations and Waves in Dynamical Systems. 1996, 554
pp.
ISBN 0-7923-3931-2
A.J. Jerri: Linear Difference Equations with Discrete Transform Methods. 1996,
462 pp.
ISBN 0-7923-3940-1
I. Novikov and E. Semenov: Haar Series and Linear Operators. 1997,234 pp.
ISBN 0-7923-4006-X
L. Zhizhiashvili: Trigonometric Fourier Series and Their Conjugates. 1996,

312 pp.
ISBN 0-7923-4088-4
R.G. Buschman: Integral Transformation, Operational Calculus, and Generalized

ISBN 0-7923-4183-X
Functions. 1996,246 pp.
V. Lakshmikantham, S. Sivasundaram and B. Kaymakcalan: Dynamic Systems on
ISBN 0-7923-4116-3
Measure Chains. 1996,296 pp.
D. Guo, V. Lakshmikantham and X. Liu: Nonlinear Integral Equations in Abstract
ISBN 0-7923-4144-9
Spaces. 1996,350 pp.
Y. Roitberg: Elliptic Boundary Value Problems in the Spaces of Distributions.
1996, 427 pp.
ISBN 0-7923-4303-4
Y. Komatu: Distortion Theorems in Relation to Linear Integral Operators. 1996,
313 pp.
ISBN 0-7923-4304-2
A.G. Chentsov: Asymptotic Attainability. 1997,336 pp.
ISBN 0-7923-4302-6
S.T. Zavalishchin and A.N. Sesekin: Dynamic Impulse Systems. Theory and
Applications. 1997,268 pp.
ISBN 0-7923-4394-8
U. Elias: Oscillation Theory of Two-Term Differential Equations. 1997,226 pp.
ISBN 0-7923-4447-2
D. O'Regan: Existence Theory for Nonlinear Ordinary Differential Equations.
1997,204 pp.
ISBN 0-7923-4511-8

O'Regan-Existence Theory For Nonlinear Ordinary Differential Equations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

O'Regan-Existence Theory For Nonlinear Ordinary Differential Equations

Uploaded by

Copyright:

Available Formats

Existence Theory for Nonlinear Ordinary Differential Equations

Mathematics and Its Applications

Existence Theory for

SPRINGER-SCIENCE+BUSINESS MEDIA, B.Y.

ISBN 978-90-481-4835-6 ISBN 978-94-017-1517-1 (eBook)

Printed on acid-free paper

All Rights Reserved

@1997 Springer Science+Business Media Dordrecht

Mary Rose, Francis, Gerard and Marguerita.

Introduction, notation and preliminaries .......................... 1

Fixed point theory ............................................... 4

Initial value problems ........................................... 11

First order periodic problems .................................... 22

Existence principles for second order boundary value problems ... 34

Boundary value problems without growth restrictions ............ 46

Positone boundary value problems ............................... 58

Differential equations singular in the solution variable ............ 82

Existence principle for singular boundary problems ............ 105

Nonresonance problems in the limit circle case ................. 116

Boundary value problems on the half line ...................... 156

boundary value problems ......................... 67

Impulsive differential equations ............................... 174

Differential equations in abstract spaces ....................... 186

Chapter 1. Introduction, notation and preliminaries.

from the recent literature on impulsive differential equations with variable

lul m = max{ lulo, IU'lo, ..... , lu(m)lo}.

As usual, L=(I) is the Banach space of essentially bounded measurable

Theorem 1.2. (Schauder fixed point theorem)

[1]. R. P. Agarwal, Boundary value problems for higher order differential

[3). R. F. Brown, A topological introduction to nonlinear analysis,

[4]. K. Deimling, Multivalued differential equations, Walter de Gruyer,

Chapter 2. Fixed point theory.

Theorem 2.1. Let E be a locally convex Hausdorff linear topological space

When E is a Banach space theorem 2.1 will be used extensively in this

Theorem A. Let E be a locally convex Hausdorff linear topological space

Theorem C. l5}. If A is a compact subset of a Tychonoff space X then

2.2. Fixed point theory.

C is compact if F(X) is relatively compact.

C a compact mapping if N(X x [0,1]) is

Definition 2.2. A map F E PA (X, C) is essential if every map in

is an equivalence relation in PA(X, C).

Theorem 2.2. Let C, X, A, E be as above with FE PA(X, C). Then the

Clearly N is a continuous, compact map; to see this it is enough to notice

so Nt is fixed point free on A for each t E [0,1). It remains to notice

If B is empty then in particular F = Nl has no fixed points and so F

Clearly J is continuous and compact. We claim that J : X ---+ C is a fixed

Theorem 2.3. Let C, X, A, E be as above. Suppose F and G are two

Let C be a complete conva subsd of a locally convex

and aU denote the c10su rE' an d the bou ndary of U in

It is easy to see that L : C - C is a continuous. compact map. Now

Let C be a complete convex 8llbsd of a locally convex

P ROO F: We assume FIDU is fixed point freE' for otherwise (A \) is satisfi .. d.

It is easy to see that N : U X [0, 1] ~ C is a continuous. compact map.

Theorem 2.6. Let C be a closed convex subset of a Banach space E and

for' i sufficiently large. Assume F : Q

C is a continuous, compact map.

if {(Xj, Aj)}.f=l is a sequence in aQ x [0,1] converging

holds. Then F has a fixed point in Q.

PROOF: Let r : E ---. Q be a continuous retraction (the existence of r

= max {1,J-l ()}

where J-l is the Minkowski functional on Q i.e.

By the above remarks, we may choose (and we do so) the retraction r

Firstly B =I 0. To see this notice since r is continuous that Fr: C ---;. C