Professional Documents
Culture Documents
Managing Editor:
M. HAZEWINKEL
Centre/or Mathematics and Computer Science, Amsterdam, The Netherlands
Volume 398
Donal 0 'Regan
Department of Mathematics,
University College Galway,
Galway, Ireland
A C.I.P. Catalogue record for this book is available from the Library of Congress.
To
CONTENTS
Chapter 1.
Chapter 2.
Chapter 3.
Chapter 4.
Chapter 5.
Chapter 6.
Chapter 7.
Chapter 8.
Semi~positone
Chapter 9.
Chapter 10.
Chapter 11.
Chapter 12.
Resonance problems in the limit circle case .... , .... , .......... 133
Chapter 13.
Chapter 14. Existence theory for ordinary differential equations on compact and
noncompact intervals ......................................................... 164
Chapter 15.
Chapter 16.
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.
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.
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.
+ (1 -
t)F(x).
= tG(x) + (1 -
t)F(x)
= tG(x) + (1
- t)G(x)
= G(x) # x,
J :X
---+
C by J(x)
= N(x,J.L(x.
Theorem 2.4.
Remark. Notf'
C respectively.
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.
Theorem 2.5.
N(u,t)
= tF(u) + (1 -
t)(;(ll).
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)
(2.1)
--->
< ;.}
z
~C
r( x)
(ii). If int Q
=0
then aQ
= inf {a > 0 :
= Q.
x E a Q}.
Ui = {
E E : d( x, Q)
<
~}
for i E {m, m
+ 1, .... }.
(2.2 )
+ 1, ... }.
Now consider
References.
10
(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.
->
C(l) by
= r + lot I(s,y(s))ds.
In other words classical solutions to (1.1) are fixed points of the integral
operator T. We first present the classical Picard-Lindelof theorem.
zl
(1.1).
and so
e-Cdl(Ty-Tz)(t)1
< e- at 1t ()(~s('-asIY(8)-z(s)ld8
< (-at (1 t
()enSd.~)
IIy-zll
12
= Ty,
T: C(I) -. C(I).
Iy" as n
--> 00
Ylo
-+
and so
--> 00.
ITy(t)1
= Ir + lot
f(s,y(s))dsl S
Irl + M b
so ITYlo S
Irl + M b,
and
t'l
b'
= sup{tE/:
(,~,y(.~ E S for 0 ~
t}.
= f(t, y(t
for all
= 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
PROOF: The proof is very similar to theorem 3.1 and will only be sketched.
Let
A(t) =
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.
= If(s,Yn(s
- f(s,y(s))I- 0 a.e. on I,
15
and
IS
and so
---+
0 as n
---+ 00.
ITy(t)1 =
Jr+ fot
and
ITy(t) -Ty(t')J
f(s,y(s))dsl:S
= 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.
It
f(s,y(s))ds.
that enabled us to deduce theorem 3.:3 from theorem :3.2. The result, due
to Caratheodory, is as follows.
t,
MOT-eover this result is best possible for the class of problems considered.
= 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}
b = io
y'(8)' d8
[y(b)
~)(Y(8) = io
du
'l/J(u)
<
[<XJ du
io
'l,h(u)
= Too
(2.1 )
a.e. t E [0, 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~
= r.
Nl(y)(t)
= r+ lot
ft(T,Y(T))dT;
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:
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).
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).
Ci E
Assume there
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]
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
(1.1 )
<
f.
1.
A.~sume
= sup[O,T] jy(t)j f.
qo, T].
L: r;[O,T]-- L1[0,T]
by
22
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
cJ~r q(s) ds i-
1. Let N : e[O, T]
N(f!
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
4.2. Existence theory.
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
(2.1 )
f(t,O) ~
for a.e.
t E [O,T]
(2.:3)
(2.6)
(2.4 )
(2.5 )
and
(2.7)
(2.8h
is as in (2.5) and
(2.9)
l
= ltl
tl
to
to
['>'f*(t,y)
['>'/(t, 0)
+ (1- '>')Ty]dt
+ (1
- '>')TY
+ .>.y] dt < 0,
Remark. The above argument also shows that any solution to (2.8Jt
nonnegative.
IS
IYlo
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
and
( 2.12)
jYnlo -
00.
iT
o
[get, Yn)Yn
and so
foT[-r(t)] dt
(1- An)
An
foT[g(t,Yn)Yn
a.e. on [O,T].
iT
0
ret) Yn dt S 0
+ h(t,Yn)] dt.
rTh(t,Yn)dt.
JorT[-r(t)]dt~liminf JorTg(t,Yn)Yndt + liminf
n->oo Jo
(2.13)
1HOO
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 .
11"
(1 - An)r 1171
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
..
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
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,
+ Ang(t,Yn(t))1,,,(t)1
a.e. on [O,Tj. This together with (2.16) implies for n 2 max{no,nl} that
Since Iv"lo
=1
(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)
as n
---+ 00
in S; here
Remark. Notice v ~
on [0, T] since
1Jn
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
10
a.e
IYnio
here n
---+ 00
in S.
lim
n-+oo 10
here n
---+ 00
in S.
f3(t)
-I-I
Yn
~ [Ang(t, Yn)
+ (1 -
IJLn(t)1 ~ max{r(t)[v{t)
+ 1], IfJ(t)!}
---+ 00
in Sand
for n ~ n3 and n E S.
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
(2.28)
Also
v(O) = v(T).
(2.29)
Next we claim that
(2.30)
J-L{t)
2: 0 for a.e.
= l.
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]
- f ::;
+ f]
= 1-,
mE {1,2 .... }.
m
Thus
(2.33)
== c 2: 0, c a constant.
(2.:35 )
'11==
Yn
n5
vn(t)
Ivlo =
1. Thus
> O.
E oS' with
c
2
Hence
(2.36)
E [0, Tj as n
-+
00 through 8.
~ fT
JfT[_r(t)jdt
Jo
o
liminf[g(t,x)xjdt+ fT liminf[h(t,x)jdt.
Jo
x~oo
x-oo
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)*
and
(2.7)*
10
x-.,.CX)
(1 - An)
An
o ~ lim
inf fT (f(t, :tin) y,-;,8] dt.
n--(X, io
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
(2.37)
(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
for n E 8 4
:JO
(2.41)
(2.42 )
:s;
:s;
jJ( t)
p(t)
and
(2.43)
r
10
(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]
y(t) 2:
(2.45 )
+ TY]
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
T(X)
:31
dx
y(tO)
y(tO),
Y:t
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.
YE
Define operators L: c~
Ly(t)
= y'(t) -
C~
---.. Co
y'(a)
---.. Co
and F: c~
and
Fy(t)
it
J(s,y(s),y'(sds.
by
it J(s,y(s),y'(.<;d<~.
(1.2)
Ly=Fy,
L,F:C~---+Co.
>
= L- 1 Fy,
L- 1 F: C~
-+
C~.
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
j: (.d([a.b]'R")
j where
---->
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),
U : R 2"
---->
R 2n ,
where C = (C1,C2) E R n
tions of z" = o.
U ( c)
X
U2 (Z1
C1
+ z2 C2) ),
2. A is invertible.
3. U is invertible.
i,~
continuous whenever it
exi,~t.~.
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
: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
(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
n~
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]}.
n = {(u,v) E R"
= L - IF y,
37
L- 1 F'. C'l
'8
C'l
--+'8'
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
y" =
f (t, y, y'),
YEB
:}
y" -
Y=
f (t, y, y') -
{ y, Y E B.
Hence
y solves (1.1)
:}
L,y=F,y.
L"F,:c1 ........ C o
:}
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<~.
F,: C 1 [a, b]
----+
where c
= (C1,C2)
(:l.l)
E R"
y' ( t) - y' ( a) -
=0
It
y( 8 ) ds
=0
(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).
fH(t)
+ g(t) -
fH(a) - g(a)
+ fH(a) = g(t).
that is when
(3.4)
Since
U ( c)
= (lh (u 1 C1 + U2 C2 ), U2 ( U1 C1 + U2 C2) )
o.
c=
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" -
y], Y E B.
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 )
( 4.2)
(4.4 )
and
(4.5)
is
IYI > M
implies that
1 - () >
00
xdx
q x
2Mo.
=M
and c
=0
and
C
Is - rl if a = f3 = 0
min{ Mo:lr l , Motlsl} if a ::fi 0 or (3::fi O.
Ml
(4.6)
c.
()
M 0 dx > 2Mo
qX+f
(4.7)>.
h(t,y,p)
h : [0,1]
={
R2
->
R is defined by
y'(t) =
t y"(s)ds > 0
J~
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
o~
y'(O)
1
= -[y(O)
a
r]
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
as n
-+ 00
h(t,un(t),U~(t))
-+
--+ 00
through N 2
o.
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
45
Chapter 6.
tions.
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
< t <
< t <
46
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
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
= 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
+ btlimp(t)y'(t) =
~t
= {y
rI},
E e[o, 1] :
Theorem 6.1.
(py')'(t) 2': 0
a.t~.
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
(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.
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)
(2.4 )
a.e. t E [0,1];
X [Ml'
Mo] -. [0,00),
81 :
Xl
Xl
:s X2;
:s x2;
and
[0,1] X
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).
Co[O, 1] defined by
N1(y)(t)
= lot h(r,y(r),p(r)y'(r))dr.
(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)
{t E [0,1]: yet)
i' Mo,
and 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.
J(l . . .
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
=0
50
1'0
and a1 0'(1)
1'1
1'0,
a1 /3(1)
= {(t,x,s(t,x)):
(t,x) ED},
0) (x - c(t))f(t,x,s(t,x)) 2:
(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)).
= {(t,x,s(t,x))
: (t,x) ED},
.51
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
= ((t,y)
[0,1]
if =
if defined
by
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),
CO,(> (t)
other'wzse;
52
= ((y,q):
= ((y,q):
= 1,
= 0 otherwise;
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.
12(t,y,q)
R2 ---. R by
where
= 0,1.
Rl: a(t):S; y
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
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)
a.e. t E [0,1]
= yet),
= p(t)y'(t)
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
(2.10)
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
a.e. on
and
< t2
t E (iI, t2)'
::;
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
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
multiplicity results for nonlinear first order ordinary differential equations,
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
+ J.Lq(x)g(x,y(x)) = 0
=-
(TC> - k z);:;
for
z ~
TC>
y'
(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
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
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"
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).
(2.2)
and
(2.3)
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}
(2 ..5)
sup
zE(O,oo)
J-lorof(z) - h(-J-lorof(z
) > 1.
: :;
~(pyl)'+>'J-lqg*(x,y)=O
limx-->o+ p( x )y'( x) = 0, >. h(y'( 1 + y(1)
(-00,00) by
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
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
Yl
= max{sup y,
[0,1]
sup ply'l}
[0,1]
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)
fal
p(t)
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]
(2.7)
Now fix /-l
(2.8)
> 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+
(Ny)(x)
= -h(y'(l + J.l
1
x
pes) 10 p(t)q(t)g*(t,y(tdtds.
(Ny)'(x)
= p(:)
Thus
-< 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)
fl
Jo
y'Q(x) d
p(x)
X
Let
P == p(1)
< 00.
fl
Jo
p(t)q(t)dt,
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
(2.13)
Y(O)
dz
<
Y(I) JF(y(O - F(z)
63
J2ii,
Jo
= AJ.lil~)
pet
ft p(8)q(8)g(8,y(8d8dt
Jo
>.. h (- AJ.l
p( 1) Jo
P~~) 10
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)
: ;
(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
64
(N y)(x)
= -h(y'(l)) + Jl
1 1
1
- 1
p(s)
p(t)q(t)g(t, y(t)) dt ds
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
(1.1)
y" + fL fey) == 0, 0 :S t :S 1
yeO) == 0, y(l) == b ~ 0 or y(l) == 0, yeO)
=b~ 0
( 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).
( 1.3)
---7
i8 continuou8.
sup ly(t)1
[a,b]
"!
Mo
(1.4h
y"(t)
+ AfL f(y(t))
== 0,
:S t :S b
67
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
sup iy(t)i ~ Mo
[a,bj
here I-l > 0 is a constant. Throughout this section we assume (1.3) holds
and also that
(2.3)
= 0;
here F(s)
= J; f(x)dx
=0
>0
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
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
ds
~
JF(Mo) - F(s) >
!-L.
and ...L
<
no
Q.
In addition since
o.
11J -1i
(~)
>
nl
du
V-Feu)
;1
>
o.
< {3 and
F(9 - b).
+ 1, ... }.
{
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
(2.1O)~
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 .)}.
G(z)
(2.12)
du
, z E [0,0].
o J-F(u)
z
yet)
(2.13)
~ 0
(2.1O)~
satisfies
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,
yet)
(2.14)
( 2.15)
since f(x)
(2.16)
< 0 for x
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).
and so
-y'y"
= >I/..l f(y)y'
(2.22)
71
on (t2,t6)
= O.
If y(t6)
(:J
(~) ~ F
> F(fJ - a)
so
F(y(t6 - F
(~)
F(y(t6
< 0,
y' y"
= 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
y'(t)
<
J-F(y(t -
~
'-JL
y(tJ)
(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)-
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)
J2b 2 Jl
e
du
- G(B).
ie-8 J-F(u)
Consequently
a
re
ie-Ii
J -F(u)
du
)
> a,
=0
implies
y'
> 0 on
(2.27)
In particular
(2.28)
We claim
(2.29)
(0,050)'
-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'
<
y'(t)
JF(y(so - F(y(t))
J2ii
~ 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.
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].
Yn (t) -- bat
(2.33)
Let n
(2.34)
+ (b-t)
J1.(b-t) rt f(
nb +
b
Jo S
Yn (
S
--> 00
through S to obtain
yet) =
at
T +
J1(b-t)
b
Jo
sf(y(,~ ds
J1t (t
+T
Jb
(b - s)f(y(s ds.
(2.35)
here J1 >
is a constant.
: ;
(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 .
1.
(2.37)
{ Ii 0 2: B wzth h 0
'UE(/(O-T,/(O);
d.s
.jF(u)-F(s)
Let
Jl3
J'i
'or u > B.
J'
= 2 R 2 (B).
G(B) (hen' G is as
and let N*
{m, m
(2.38)m
1)
and
(2.39)'"
y"
y(
&) = 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
!"
iJ
(2.40)
1-
=0=
t-"2
lim v'(t).
1+
t-"2
and
y"
t -)
t -)
t -)
i)
77
i]
t)
that
!).
[W'(t)]2
t),
w'(t)
-r.~~~~rrr
JF(Ko) - F(w(t
!<1::
> v 2fL
to obtain
= foKo
>
/F(K:}U_ F(u)
~.
t-)
viet
[v'(t)f
Thus for t E
F(v(t]
>
2fL[F(Kol - F(v(t))].
(t, 1),
-v'(t)
JF(Ko } - F(v(t
-~~~::;=:;='<7 >
t to
q::
v.c.fL
1 yields
R(Ko) >
v'(! +)
~,
=
o.
(2.41 )
78
If 0
8.4.
Let
J'i
j'i.
jIi
(2.42)
= 00.
Let
fLs = 2 R2(O);
here R is as described in the statement of theorem
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
If 0
,~uppose
8.4.
Let
R2( 0)
= -2-'
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,
[V'(t)]2
= [v'(OW + 2f-l[F(Ko) -
-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
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.
{ 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)
82
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
(2.3)
R is continuous.
----+
(i). Assume
(2.4)
IYlo =
sup ly(t)1 =P M
[0,1)
<
+ Ag(t, y) =
0,
yeO) = a, y(l) = b
y"
<1
(2.6 )
1
xlg(x,y(x))1 dx::; 1- t
Also
lim (1 - t)
1-+1-
it
0
10r sg(s,y(sds = 0
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]
yet)
(2.7h
+ bt + ,x( 1 - t) I~ sg(s, y~
+,xt 1/(1 - s)g(s, yes~ ds.
= a(I -
t)
Ny(t)
(2.8)
= a(l -
+t J/(l
t)
-'I-
ds
C[O, 1] by
INy,,(t) - Ny(t)1
<
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
= {u
E C[O, 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
i = 1,2 with qi
(2.12)
-+
>
(2.13)
du
<
g{u} -
T'}
!!:El
+ T'2 g{z) +
j'P'
0
>
> 0,
du
g{u}
i == 1,2
(2.14)
and
Jor any R > 0, ~ is diIferentiabZf; on (0, R]
{ with g' < 0 a.f' and
(2.15)
00
addztwn fo
1LWli
g(t) dt = 00.
g
n C 2 (0, 1).
y"+J*(t,y)=O,O<t< 1
yeO) = y(l) = p"
J*(t,y)
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)~
(2.18)
y"(to)
= ->'!'(to,y(to)) = {
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))
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))
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
~
_1_ ft n , (1
( ) dx
g(y(t n )) I-tn Jo x
- ')
X q2 X
+ JofP3 ~
g(u)
fy(t n
JO
..lfu....
<
g(u) -
fl
tn Jt n
+~f~f!:B
L ft~ X(1 -
x)q2(x)dx
lo
y(t n )
du
h(y(t n ))
- < r l + rZ
g(u) g(y(t n ))
lP3
+ft
~.
du
+
0
g(u)
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
{
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)
{yn}~=3
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)
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"
yet) = Y (~)
+r
89
:s:
qo, 1] n (,2(0,1).
y" + f (t, y) = 0, 0
y( 0) = y( 1) = Pn
<t<1
(2.:H)n
y"
yeO)
where
j**(t,y)
and 1': R
-+
Remark. Notice 9
= f**
:s:
u if lui
1
fur otherwise.
Next we claim
(2.33)
Yn(t)
:s: Il(t)
for t E [0,1].
90
+ (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)
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)
where
J~*(t,y)
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
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].
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
[~,
1] .
[ ~4'4~]
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
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
:2
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 ]
C[O,l].
(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)
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, ....
94
(k, k~1]'
IS
(t,
fit, y)
(2.38)
ko for (t, y) E
Let
(o,~]
>
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) =
and q"(l)
3 = 1//'(!)
3 = q"U)
3
Notice since
f(t, y) ~ ko for
(2.39)
(t, y) E
[3'132]
X {y E (0,00) : y::;
q(t)}.
(2.40)
f(t, y)
Also 0:*(0)
= 0*(1) = 0,
>
(2.41)
o(t) = 110*(t)
where
11
= mi n
{1.
(0*
)~I0+1
}.
~ /,;0 +
O'''(t)
k I(O'*)//(t)1
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,
{ 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,
f3(t)
= 2J.l + P3
96
= 1.
Clearly
(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.
J(t,p,,(t))
+ p::(t) ::; 4 +
(1
+ )2
:" -
J.L2 - 2 ::;
4+ 1 -
J.L2 - 2 <
o.
(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)
---->
*f
97
=0
there
{ with
(0,1)
(t,y)
(3.4 )
and
hI
(3.5)
(3.6)n
1*
isasin
< ). < 1
We claim
( 3.9)
98
_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)'
(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
heM)
by
Ly(t)
= (y'(t) -
and
Fy(t)
= (-
1t
->
ColO, 1] x R
j*(x,y(xdx, O(y'(1).
yet) = Pn -t
(,'
= (,':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)
+ [0(1) -
= O(y~(1)) -
Yn(1)]
0(0'(1)) ~ 0,
-y"(x)
heM)
g(y:(x)) ~ ql(X) + Q2(X) g('M) for x E (0,1).
Integrate from 0 to 1 to obtain
10{
I { -
Now consider
10t
-l
fez) -
[Ql(X)
[-g'(u)]2'
()
9 u
du
100
- Yn(1) ~ 0
M )]
_
+ Q2(X) h(
geM) dx =
= pn
M 2
Consequently
(3.15)
---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
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.
yet)
(~) + (t -~) n-
it
(x
(-y
-~) J(x,y(x))dx +
/1 (t -~)
J(x,y(x))dx
y'(t) =
n-
(-y
(~) -
ftl
(x
0(y'(1))+y(1)
(n-
1n-1
= - J(t,y(t))
(-Y(1) -
~1 (x-1)f(X,Y(X))dX)
101
(3.17)
(3.18)
ht
00,
= 1,2
Jor any ao
> 0, if k
In
=0
and
(3.19)
>0
!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
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
(3.20)
f(t, y)
+ a"(t) > 0
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.
[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.
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 )
a.e. on [0,1]
limt-+o+ p(t)y'(t) = 0
y(l) = 0
( 1.2)
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
00.
(ii). y
r-+
11
w(t)lu(tW dt <
:2: 1
00.
11
(1.3 )
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)
and
(2.3)
11 (1 P(X)q(:l:)dX)~
p!s)
d8 <
106
00
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
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=
and
Remark. Here {3
= "'~l
i is
t
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.
IILllu -
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) ~
- 13K
ds
lu - vlK
r ~ rs pqeKR(x) dx ds
Jo p Jo
_1_ (
pq dX) a ( t
Jo p( 8) Jo
Jo
fj
1) ~
lu - vlK
lIuliK
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
It p1:) <
(t d8
where Q(t)
r p(t)q(t) (1t
(2.6)
io
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
<
X
73
1 .
-
Consequently
<
X
73
-1)
(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.
109
(1
:; (11
~
pqlyl dX) a
pqlyla dX)
11 (11p~:))
(11 *11 (1 p~:))
pq
pq dX)
dt
pq
a dt
The Banach contraction principle now establishes that (2.7) has a solution
Y2 E L~q[O, 1). Also
(2.8)
1
')'
-(PY
pq
+ IlY
= h(t)
a.e. on [0,1 )
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
(2.10)
(2.11)
= fa1 G(t,s)q(s)h(s)ds
Yl(S)Y2(t)
W(s)
yI!t)Y2(S)
W(s)
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.
(1
L~q[O, 1] since
and also
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
(3.1)
11
p(t)q{t)
Then A : L~q[O,
1J
---->
11
p(s)q(s)k(t,s)y(s)ds
1],
(i
= c>~1
00
Jor some
1.
is completely continuous.
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
n C 1(O, 1) and
(3.3h
aUf
solution.
(3.4 )
112
when'
k (t,s ) -_ {
.
Co
Co
Co
yd 8 ) Y2 (t ), 0
Y2 (S ) Y1 (t), t
< .~ s: t
s: .~ < 1,
1]-- L~q[O, 1] by
Ny(t)
11 p(s)q(s)k(t,s)f(.~,y(s))ds.
and also
dt <
00.
Ay(t)
11
p(s)q(s)k(t,.s)y(s)ds.
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
References.
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
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.
P E C[O, 1]
q E L![O, 1J with q
(1.4 )
and
e rs
io p(8) io
_1_ (
(1.5 )
<
00
are satisfied.
Remarks. (i). Notice (1.5) implies
{I
io
p(t)q(t)
1
1
PC:)
dt <
00;
+ 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
f(t,y(t))
2
2 [
]
q(t)
E Lpq[O, 1] whenever y E Lpq 0,1 .
(1.6
)
(1 p(t)q(t)ly(t)12 dt)
1
2"
f:
Mo
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
as
--+
outside [0,1).
(2.1 )
ug2(t,U):2:
(2.2)
(2.:3)
and
I
(2.4)
p(t)z(t)
(J/ p7:))
dt
<
.:r.!.
2
dt
and
00
<
00
la
[P(y'? - flpqy2] dt
- >.1a 1pyh(t,y)dt.
This together with (2.1) and (2.2) imply
119
Let
R(y) :::::
101 (p(y')2 -
> 0 with
fl
00
(2.7)
where
00
y:::::
Ci
= coVJo
and so
0::::: R{y)
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
--->
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
./0
<
_
r1pq11
./0
f1 pq
./0
o as
11
t
d8 dt
p( s)
t+r
ds dt
p(s)
7' -- 0+
---4
--> 00
in S.
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,
121
=0
Thus we have
(2.15)
and
(2.16)
where
Kt =
and
K2
iot
p(t)<Pt(t)
(it pC:)
de;) ~ dt
t
io
~ M J~ pq J/ ~ dt == Mo
Consequently (2.19) together with theorem 11.1 guaranatees that (1.1) has
a solution. 0
9t(t,U)
0 could be replaced by
(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
=L
ci'I/Ji and Ci
= (Y,1/Ji)'
i=rn
and v E
n.L.
1
-la (v-u)[(PY')'+JLPqy]dt
Ala 1p(U-V)Y9l(t,Y)dt
10 1 p(u -
v)h(t,y)dt
JLpqv 2
Now
Also
123
(2.22)
Let
R*(y)
11
- 11
=
>
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
124
and
(2.25)
Remark. Note Yn =
Un
+ Vn
where
Un
00.
E 0 and
E 01..
Vn
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
However
IIYnll;q + IIY~II; =
R*(Yn)
--
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.
=0
= I Jtfi [v' -
we have u( 1)
u'] dsl
= v( 1)
= 0 and so Holder's
u'fl dt
)t (eJt pes)
dS) t
and so
(2.30)
!iY'lip
(1
d~)!
p(S)
_L
'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
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)
=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)
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)
for
< A < 1. From [14, chapter 3] we must show that there exists a
constant M, independent of A, with
(2.41)
(0,1)
127
and so
(2.43)
+ J~ pQ>3lyllpy'I B dt.
(2.44 )
ds ) t
Pc:)
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)
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)
sup ly(t)1
[0,1]
12R
:S
MI.
fa1
Mi fa1 Pc/>2 dt
+ fa1 p3lpy'1 9 dt
f~
l(py')'1 dt
::; Eo
+ E111y'll~ + E2 11y'll:
K
::; Eo + E1M02
+ E2MJ
l(py')'1 dsl
== M2
::; M2
so
(2.51)
(0,1)
/'i,
= :2
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]
{I
~
2
c/>b2 E L![O, 1] with Jo [P(t)]2-9[.1(t)]2-9 dt <
129
00
and
(2.,53)
vllyl"! dt
vllpy'l{/ dt.
Iv(t) - u(t)l:S
( JofI p(v' -
u')2 dt
) t (fI
d~) t
Jo pC:)
and so
(2.55)
d~) t
(1o -')
1
p(,~
130
References.
Here Am, m
of
= 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
f(t,y(t))
E
q(t )
2 [
Lpq
0,1
satisfies
whenever
2 [
Lpq
0,1 .
+ <i>2(t)lyl'Y
(2.2)
(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
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)
134
-he
and
Ilu'llp =
(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
= cm1/)m
i=O
we have Uo
10)
and integrate
Notice
(2.9)
Also
(2.10)
= O.
R(f) - (p - Am)
(V~)
10
to obtain
pqw 2 dt S
where
> 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
1=171+1
pq'~!i
> O.
Ci
Am )Ci
Joofl pq'l/Ji'2 dt
+ 2, ...
and so y
'2
1=0
0= R(y)
so (2.4) implies
Cm+1
= 0 i.e. f)
= O.
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
--7
UO,n
00
III
Uo
---'0
in S; here -
2[
Lpq
0,]
1 and
~ U ,o
in Lp2[]
0, 1
'uO,n
L~q[O,
<
<
U~,n
--7
u~ in L;[0, 1] as n ----
00
in S.
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,
yields (with fJ
= W + uo),
as n
--+ 00
Hence R(f})
=0
+ Iol[2P(U~,n)2 + pqu~,n
Ampqu~ n + pqw~ + (Am + r)pqw~J dt
--+
R(f})
Now choose fL = Am
Am + ~ < Am+!. Thus
(2.20)
+1
Jo
(fL -
plh(t, y)
Am)lIwlI;q
+ 1J(t)II1L - wi dt.
1If}'II; =
(2.22)
(2.23)
IIw' -
1L'II~
IIw'lI~
+ lIu~lI~
(2.24)
and
(2.25)
IIw -
1L1I~q = IIwll~q
+ 111LII~q
= IIwll~q
+ l11LolI~q + l11LllI~q
(2.26)
ly(t)1
111
y'(s) dsl
s lIy'lip
138
(1 P~;J t
1
The
and
1
(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
Jt
pC:)
dt;
and
(2.30)
where
K3 =
10
p(t)lv(t)i
(1 p~:J t dt.
1
(/(1
+K2
Now since 0
~
2
Consequently
(2.32)
139
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)
( 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".
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
--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
(IIUl.nll~q + lIui.nll~) 2
<
Also for
t,s E [i, 1]
(IIUl,nll~q + lIui,nll~) 2
we have
lJ:y~.'(x)dxl
(IIUl.nll~q + lIui.nll~) 2
::;
IIYn'lIp(J:rfxTr
(IIUl,nll~q + lIui,nll~) 2
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.
Also
=0
and so
(2.47)
Thus
(2.48)
142
-k
+ Ilu~,nll~)""2
to obtain
(2.49)
h,,,(t)1 ~
d~ ) t
(Jto p[r~,,,12 ds ) t (11 pC:)
and so
Ip(t)v(t)rl,n(t)1
This together with the remark after (2.45) implies that there is a constant
Eo with for t E (0,1]'
(2.50)
00
in S.
Hence
(2.52)
fl
p'/J rl,n
(11'1l1,nll~q + 11'1l;.nll~) 2
lim iuf In
= O.
lit
l'Tl,n[-h(t,Yn)]
lit
+lim iuf J~
l' Tj ,n Yn!l(t,Yn)
1;.
dt.
dt
-> 00
in S.
Next we claim
(2.54)
(2.55)
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
where
Bl =
and
B2 =
lo
ds )
pis)
p(t)dt)
(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
prl,n Yng(t,Yn)
dt
= {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,,)
(1Iul,nll~q + Ilu~,nll~) 2
In addition since
for t E (0,1)
we have
I
!!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)
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).
li)
li)
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).
Yn
!)k
(1IuI,nI12+llu;,n))2)2
(1IUl,n))2+llu;,n))2) 2
and .It
y,~-kg(t'Yn))
dt
in S.
(2.63)
It.
nl
with
+00
as n
---+ 00 III
-00
as n
---+
(2.64)
Ii.
(2.65)
S.
00
in
o 2: JI +
+ JI (2.66)
+ J/+
dt
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"
---+
0+ (i.e.
=~-
(2.67)
(2.68)
147
00)
Similarly
k)
liminfj Xr lilllinf(
prl.nYng(t,Yn)
(>-+00 /(}
(1IuI.nll~q + lIui.nll~) 2"
(2.69)
dt~O.
Also
rt
(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
Similarly
(2.71)
(2.72)
(2.73)
Also (2.48) together with (2.51) yields
(2.74)
t E [0,1].
We have as in (:2.59),
(:2.76 )
Also
(:2.7:~)
yields
IF
1
)]2
1
IIJ1I.n ll ( t, .lJn)1 :;
:; B4P(t)q(t)T(t)
:; 1>1(t) + </J2(t)
1
1
rl.s
-.
p(8)
Ip(01 + 0'>:,dlll.,,1
j
(1IItI.nll;'" + 111l;,,,II;,):2
<
(1IU 1 n
ll;'q + lIu;,nll~) 2
Ip(t)rl,n(t)h(t,y,,(t))1
S:
t E [0,1].
Aft~ 7) III/'m dt ~
(:2.7X)
Let () =
+fj+
where
(t
149
(2.79)
dt.
Notice
~0
and
(2.81)
liminfj
<>-+00
1+
~ O.
Also
Ap(t)1/1m(t)liminf(xg(t,x ~ 0 for a.e. t E
x-+oo
It
dt.
Finally since
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))
:;
:;
j+
(2.;3;~)
dt
p(t)'ljJ",(t)liminf(-h(t,x)) dt
x--->oo
:;
dt,
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.
I(t, y) = h(t, y)
and assume f : [0,1]
+ v(t) -
yg(t, y)
(1.3),(1.4),(L5),(2.1),(2.2),(2.;~)
(2.84)
(2.8.5)
h(t, u) :;
and u>
00
and
(2.86)
dt.
Uz
with
n2
and n E S; t E [-;
pr:, ~ Ap1/<n
-::J 0
In
C[O, 1] as n
-> 00
In
S*.
152
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.
>
< j
[+
+
for
71,
[-
(prl,nYng(t,Yn)+prl,n[-h(t,Yn)])dt
(p rl,n y"g( t, Yn)
dt
0
(2.92)
is satisfied.
(i). Suppose there exists a constant k > , of the form 1 > k
/3 is odd and 0: is even, with
= ~,
(2.93)
+ J1-
where
> o} and 1- =
with
A J~ p(t)v(t)'l/Jm(t) dt
+A f1-
(2.94)
= y(l) =
Cit/!; , Wo
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.
155
(1.1 )
and
( 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.
(2.1 )
and
(2.2)
here m
_ ~
(x, y) -
Pm (x - y)
~ 2m 1 + P (x _ y) ,
rn=l
'Tn
lul cx =
sup
[0,(0)
lu(t)l.
(2.3)
and
(2.4 )
157
(2.5 )
_e~~2t f~
Let E
n~2
e",2s
ftC>
f(s,y(sds
f(8,y(s))ds
+ 1 = No}
and
Fy(t)
f(s,Yn(s))
-->
/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)
I:
there
for all n
and
(2.7)
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
: :;
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)
(2.9)
Now Y
= AF(y)
+ 1.
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.
(2.11)
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
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
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
f : [0,00) X R
(2.14 )
(2.15)
-+
R is continuous
iyi >
Mo
n ('2([0,00), R").
u ( to ) u " (to)
(2.16)
1= 0
= 1, (2.14) and
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
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.
y":= f(t,y,y')
a.e. on (a,b)
y(a)
= c,
y(b)
:=
and/or
( 1.:3)
d.
164
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
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
1M 'Ii~~)
1
dx,
(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.
{g E En
Wl~':(a,b)
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
(d [ak,
bkj,
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)
-->
If(t,y,z)1 :; 'I(th')(lzl)
and
;
00
:J;~
1/)(:1:)
dx
= 00.
0::; y :; f3.
where
rk
and Ak
x~
'lj;(X) d:1:
= 00
= min{I(3(bk) -
= sup{ /3(t) -
where
and
In fact, for every kEN and for every n ;::: k, the following inequalities are
satisfied:
and
on [ak,bkJ,
(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 -
:::; f3(ak)}.
'
tE[a,ak)
Iy~:(t)l ~
hdt)
sup {{3(t)}
tE[a,ad
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,
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
btk-~~~'
an
(r - an)!(r,Yn(r),y~Jrdr
len - cl + J(kl t -
bt __an
k
an
and so
I~~=~(t -
Iy,,(t) - y,,(al! $ {
len - c/ + J(klt -
+ b~=~l
al
+ b~k.:;,tl
f:(r - a)hk(r)dr
lim(t
- a)
t __ a
l
t
bk (bk
- r)hdr)dr
= O.
---->
JOO
~~:) 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 ]}.
(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.
0 :::;
/3 E
21
w,o~
0:::; /3
172
References.
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
(1.1)
na :
15.2. Preliminaries.
In this section we establish two existpl\('p results for tlH' initial valup
i74
D. ORegan, Existence Theory for Nonlinear Ordinary Differential Equations
Springer Science+Business Media Dordrecht 1997
problem
(2.1)
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
(2.2)
(2.3)
1/) is
nondec7'ea<~ing;
175
and
(2.4)
t*
q( s) d<~ <
1(X) _(u
d
laol '1/) u)
= 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.
(it
q(X)dX)
dx
-.-.
lao I '1/)( x)
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
(2.6)
).
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
(2.7)
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 )
177
(2.9)
where
[a,T),
OIl
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)
0',
That is
(2.11)
In this section various existence results are established for the impulsive
differential equation
(3.1)
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
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;
and
(3 ..5 )
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)
ri,l(t)
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)
Define
=f
={
on [0, tl],
on (tI,T),
YI,
Y2,
t*
[Ti(Y2(t* - Ti(Y2(ttl)]
+ Ti(Y2(ttl)
itl
180
(3.9)
f}"
== { .
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
(3.1O)
with a(t) :::; Y3(t) :::; f3(t) for t E [t'}., T). Define
If ri,3{t)
f.
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
T
I
)_
tl -
= t'2
for some
{fl, ... , td
{0,
if tl 'f r;,
.
R, If tl = ti,
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(y(ttJ)::; t2
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
(:3.11)
( 3.12)
Ii : R
-7
= 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 )
PROOF: Let Y1 E
(3.6) with
(3.15)
qo, T)
IYl (t)1
dx
'V)(x)
OCi
nb.
./-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)
IIi 1(Yl(tJ))1
max
[-Qa,QoJ
IIi 1 (:r)1
183
(fotl q(x)dx).
Jo
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)
f26.
Remark. The 'V in (3.16) denotes the gradient and . denotes the usual
inner product in R".
184
References.
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
y'(t)
y(O)
= f(t,Y,(t))
= a E E;
t E [O,Tj
Jt
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.
(2.1)
{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)
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
10
tE[o.TI
II
(1 - e-Q(T))
Remarks. (a). Condition (2.1) is satisfied if g* is a [( -Caratheodory function i.e. g* is Caratheodory and satisfies property- [( (see [4]).
{
= H, a real
Hilbert space.
(3.1 )
Let I( z)
ing.
(3.2)
i.~
strictly increas-
(3.3)
with
Ilull, IIvll
M.
y'(t) = g1(t,y(t))
yeO) = a;
+ hl(t,y(t))
t E [O,T]
p(y)
y,
if
t!r,
Ilyll
~ M
Ilyll > M
if
= {y : Ilyll
M}. It is easy to
<
q(t)llu - 1111;
T
res) d8
u
<
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
iT 1'(s)ds= MO
R
< Mo if
No 1'(s) ds
1
-(-)dx.
'ljJ x
(3.5 )
= g+h
Let J(z)
= liIail 'P(x) dx
Define M =
and notice J :
[llall,oo)
---+
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,
Ily(t)11
Ily(O) +
Jo
190
P'(t)::::
Ily'(t)1i
~ 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)
(3.7)
(3.8)
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
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
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.
M(t)M'(t)
M(t)
.
I
Ily(t) _ v(t)11 (y(t) - vet), h( t , yet)) - v (t).
Thus
(3.10)
a.e. tE(to,tl).
+ ftl
to
(y(t)-V(t1,yl(t)-V I (t)
lIy(t -v(t)1I
tl
to
dt
M'(t)dt = M(tIl,
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.
= g+h
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
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.
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
156
homotopic
inessential
invertible problems
35
impulsive equations
174
LP(T)
L=(I)
187
2
2
LP-Caratheodory
LP-Lipschitz
14
I!)
22, 13:3
12
116
195
lower solution
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
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
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
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