BVP(1)
LEMMA 12.1. Let f’Po P be a compact map
(i)
Iff (x) v
2xfor
every xS
and every2>=
1, theni(fPp)
1.(ii)
If
there existsan elementp > 0 such that xf (x) v
2pfor
every xS
and every2 >_ 0, theni(f,
Po)
O.Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
Proof
(i) Define acompactmap h’[0,1]
xPo
P by h(z,x)..= zf(x).
Then, by thehomotopyinvarianceand thenormalizationproperties,i(f
Po)
i(O,Po)
1.(ii) Let #..=sup
{llf(x)
[x ePo}
and let21
> (p+ #)/IPll.
Then, by the homotopy invariance,i(f
Po)
i(f+
21p,Po)"
If i(f,
Po)4:
O, then there exists an element xPo
such that xConsequently,
Ilxlt >_-21
PIIf(x)
__>21[p -/ >P, which isimpossible.Hence i(f.Po)
0.THEOREM+ 12.2. Let
f’Po
--* P be a compact map such thatf(x) va
2xfor
xeSo and 2 >_ 1. Then
f
hasafixed
point inProof
Thisfollows from Lemma 12.1 and thesolutionproperty of the fixed point index.Observe that the condition of the above theorem is satisfied if
f(x) =
x forallx e
Sf.
More generally,itsumces
toknow that forsomemajorantf
off,f (x) ;
xfor all xe
Sf. A
similarremark appliestothe following theorem andits corollaries.THEOREM 12.3. Let
f" Po -
Pbeacompactmap, and leta, z e(0, p]witha4: z.Supposethat
(i)
f (x)
=/: 2xfor
every xS
+ and2>__
l,(ii) there exists an element p > 0 such that x
f (x)
:/: )pfor
every xS
+and2 >_O.
Then
f
has atleastonefixed
pointxwithmin
(a, z)
<xll
<max(a, z).
Proof
Let Po min(a,z)
and P max(a,z).
Then by the additivity pro-perty of thefixedpoint index,i(
f Po,
\Poo)
i(f Po
i(f Poo)
Hence,
by Lemma 12.1, i(f,Po,\Poo)
isequal to if Pl a,and equalto if P z.In eachcasetheassertionfollowsfrom thesolution property. V]COROLLARY 12.4
(on
the compression ofacone).
Letf’Po
Pbeacompactmap,and let 0 < a < p.Suppose that
f (x)
xfor
xS
andf (x)
xfor
xS+
Then
f
hasafixed
pointxwitha <]]xl]
< p.COROLLARY 12.5
(on
the expansion ofacone).
Letf’Po
P be a compactmap,Thenand let 0
f
hasafixed
< apoint x with< p.Supposeathat< [x[I <f (x)
$_p.xfor
xS2
andf (x) =
xfor
xS+
Remark 12.6. It should be observed that the above theorems remain true if Eis anarbitrary Banach space andPis aclosed wedgeinE, that is,P hasall the properties ofa conebut P
fl (-P) {0}. (In
thiscase, the relation < is onlya preorderinginE.)
Moreover,
itshould be observed that theproofof Theorem 12.3 shows that the fixedpoint indexoff
overthe "conicalshell"Po, \Poo
is either+
or 1. ThisDownloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
additional information can be very usefulinapplications. For example,suppose itisknown that there existsan isolated fixed point of
f
inPp,\Ppo
whose local index isdifferentfrom+
or 1,respectively. Then,by the additivity of the fixed point index, there mustbeasecond fixed point in this conical shell.13. Fixedpointsof ditferentiablemaps.Inordertoapplythegeneraltheorems of the preceding section,onehasto findmaps
f
which,roughly speaking, pushthe elements ofS
+ either away from zero orcloser to zero.In
order to verify these conditions,it isreasonabletolinearize themapf
at zeroand infinity andto study the behavioroff’
at zero and atinfinity.Throughout this section,we denoteby
(E, P)
an arbitraryOBS,
andbyp a positive number.LEMMA
13.1. Letf’Po
P be a compact map such thatf (O)=
O.Suppose
that
f
hasa rightderivativef’+(O)
atzero suchthat isnotaneigenvalueof f’+(O)
toa positive eigenvector. Then thereexistsa constant
ao (0,
p] such thatfor
everyo
(0, ao],
(i) i(f,
P)= /f f’+(0)
has no positive eigenvector to an eigenvalue greater than one;(ii) i(f,P,)= 0
/f f’+(0)
possesses a positive eigenvecor to an eigenvlue greater than one.Proof.
Sincef’+(O)x
lim_o+,-f(x)
for everyxP,
it follows thatf’+(0)
L+(P P). Moreover,
byTheorem 7.1,f’+(0)lP
is acompletelycontinuous map.Henceid f’+(O)isdosedondosed subsets ofP. Therefore, (idf’+(0))(S +)
is a dosed set and
0
(id-f’+(0))(S +)
by the hypothesis of the 1emma. Hence thereexists apositiveconstant such thatx
f’+ (0)x
x for all xeP.Choose
ao
(0, p] such thatf(x) f’+(0)xl[
x/2
for all xP,o.
Thenfor every o
(0, ao],
every y P satisfying Ilyll< o/2,
and every 2 [0,1],
the map(1 2)(f’+(0) +
y)+ 2f
possesses no fixed point on $+.
Indeed, for everyIIx- (1 2)(f’+(O)x +
y)-2f(x)ll
_>_
xf’+(0)x f(x) f’+(0)x
Ya( /2
Ilyll/) > 0.Hence,
bythehomotopyinvariance property, i(f P,)
i(f ’+ (0) +
y,P,)
(i) Inthiscase,wesety 0and observe that theequationx
2f’+(0)x
0 hasnopositivesolutionfor2[0, ].
Hence bythehomotopyinvarianceand by the solutionproperty,i(
f P)
i(f ’+ (O) P)
i(OP)
1.(ii) In this case, denoteby h S+ an eigenvector of
f’+(0)
to an eigenvalue 2 > 1. Then we claim that, for every/3 >0, the equation xf’+(O)x =h
hasno positivesolution. Indeed,suppose that there existsa solutionx > 0 forsome
Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
fl
> 0.Then there exists anonnegative number Zo such that x>= zoh
and x;
rhfor
>
to.Henceweobtainthe inequalitiesx
f’+(O)x + h >__ f’+(O)oh +/h >__ (to +/)h,
whichcontradictsthe maximality ofo.
Now,
by setting yflh
with0 <fl
<ae/2,
the solution property impliesi(f
P)
i(f’+(0) + flh, P)
O. I--1The following theorem is closely related to Theorem 12.3. It should be observed thatits proofalso gives some informationabout the value of thefixed pointindex inanappropriate conical shell.
TI-mOREM 13.2. Let
f "Pp
--. P be a compact mapsuch thatf (O)
O. Supposethat
f’+(0)
hasaright derivativef’+(0)
atzerosuch that is notaneigenvalueof f’+(O)
to apositive eigenvector.Letone
of
the following conditions besatisfied"
(i)
f(x):/:
2xfor
every xS
and 2>=
1, andf’+(O)
possesses a positiveeigenvectortoaneigenvalue greater than one;
(ii) there exists an element p > 0 such that x
f (x)
:/: )pfor
every xS
and2
>= O,
andf’+(O)
hasno positive eigenvectorto’aneigenvaluegreaterthanone.Then
f
hasa positivefixed
point.Proof By
Lemma 13.1 thereexists a numbera(0,
p)such that i(f,P,)=
0in case(i)andi(f,
Pp)
incase(ii).Hence,
byLemma
12.1 and the additivity of the fixed point index,i(
f Po
\P,,)
i(f Po)
i(f P)
isequalto incase(i)andequalto in case(ii).Therefore thesolutionproperty impliesthe assertion.
It
has been noted in the proofofLemma
13.1 thatf’+(O)eL+(P P)
and thatf’+(0)lP
is a completely continuous map. Let TeL/(P- P)
be arbitrary.Then
TIIP
sup{ll TxII Ix
eP}
<andit canbe shown that
rp(T) lim
Tk P/k
is awell-defined nonnegativenumber,theconespectralradiusofT.Itiseasytosee that
re(T r(T)
ifP gives an open decomposition ofE.Suppose
thatTIP
iscompletely continuous
(that
is, T is P-compact). Then it has been shown by Bonsall[79] (cf.
also10]),
thatthe conespectralradiusofTis aneigenvalue ofT withapositive eigenvector, providedre(T
> 0. Using thisresult,it is obviousthat condition(ii) ofLemma
13.1 is satisfied providedre(f’+(O))
> 1.Moreover,
con-dition (i) is certainly satisfied ifP is total, that is, P P Eandr(f’+(O))
< 1.Finally,ifweassumethatPisgenerating, then
(cf.
Theorem1.6)
theconespectral radiusequalsthe spectralradius.For
simplicityweonlyformulate one corollarytoTheorem 13.2 forarather special,thoughimportantcase.Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
COROLLARY 13.3. Suppose that Phas nonempty interior and let
f "Po -
Pbea compactmap such that
f(O)
O.Moreover,
suppose thatf
hasastronglypositiveright derivativeatzero.Then
f
hasa positivefixed
point,providedoneof
thefollowing conditionsissatisfied"
(i)
f(x) g
xfor
xS
andr(f’+(0)) >
1, (ii)f (x)
xfor
xS
andr(f’+(0))
< 1.Proof It
sufficestoobservethat, byTheorem3.2,thespectralradiusoff’+(0)
isan eigenvaluetoapositiveeigenvectorand,infact,theonlyeigenvalue with this property.
Hence
itis easytoverify thehypothesesofTheorem 13.2.In
thefollowingwelinearizeatinfinityinordertoverifytheconditionsatSo Thefollowinglemmaisthe counterpartatinfinitytoLemma 13.1.Sincetheproof is almost verbal, the same as the proofofLemma
13.1,we leave it to thereader(cf.
also[24,
Lemma2]).
LEMMA
13.4. Letf’P -
P be a completely continuous, asymptotically linear map such that is not aneigenvalueof f’()
toapositive eigenvector. Thenthere
exists a positivenumber tro suchthat,
for
everytr>__
tro,i(f,P)
1if f’()
has nopositive eigenvectorto aneigenvaluegreaterthan one, andi(f,
P)
0otherwise.As
afirstapplication ofthis lemma,we obtainthe following theorem which alreadyhas been used intheproofof Theorem 7.4.THEOREM 13.5. Let
f"
P-. Pbeacompletelycontinuousasymptotically linear map. Thenf
possessesafixed
pointif f’( )
doesnothave a positive eigenvectorto aneigenvaluegreaterthanorequaltoone.Proof By Lemma
13.4,thereexists apositive numberrsuchthati(f,P)
1.Hence
the assertionfollows from thesolutionproperty. V]It is now obvious how Lemma 12.1 and
Lemma
13.4 can be combined to provethe counterparttoTheorem 13.2, namely,toprove the existence ofapositive fixed point off
by imposing conditions forf
atS
and conditions forf’().
We
leaveit tothe readertoformulateandprovethe correspondingstatement.THEOREM 13.6. Let
f"
P--. Pbeacompletelycontinuousmap such thatf (O)
O.Supposethat
f
isasymptotically linear and rightdifferentiable
atzero, such that 1 is not aneigenvalueof f’( )
orof f’+(O)
toa positive eigenvector. Thenf
hasat leastone positive
fixed
point,providedoneof
thefollowing twoconditions issatisfied"
(i)
f’+ (0)
hasno positive eigenvectortoaneigenvaluegreaterthan one, whereasf’( )
possesses suchapositive eigenvector(ii)
f. (0)
possessesa positive eigenvector to aneigenvaluegreater thanonebut thisisnotthecasefor f’().
Proof Lemma
13.1 andLemma
13.4 imply the existence of real numbersp andawith0 < p < trsuch thati(f, Po)
andi(f,P)
0 in case(i),andi(f,Po)
0 andi(f,
P)
incase(ii).Hence,
ineach case,i(f
P\Pp) i(f P)
i(fPp) O,
and the solution property of the fixedpointindeximpliesthe existence ofafixed point x such that p < [xl[ <tr. V]
Suppose,inaddition, thatPhas nonempty interior and that
f’+(0)
andf’()
are strongly positive. Then(cf.
the proof ofCorollary13.3),
conditions (i) and (ii)of the above theoremcanbereplaced by(i’) r(f’+(O))
< andr(f’(c))
> 1, (ii’)r(f’+(O))
> andr(f’())
< 1.Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
Moreover,
inthiscase, 1 is never aneigenvalueoff’+(0)
or off ’(or)
to apositiveeigenvector. Finally, if it is supposed that
r(f’+(O))
< 1(or r(f’())
< 1), thenf’+(0) (or f’())
is not requiredto be stronglypositive.It
should be noted that in the one-dimensional case, the theorems of this paragraph have straightforward easy geometrical interpretations. For example, the lasttheorem expressesthe obvious factthatthegraphofacontinuousfunctionf:+
---,+
withf(0)
0 intersects the45-lineata positive distance from zero, providedoneofthe followingconditions issatisfied"(i)
f’(0) >
andf
hasanasymptote withslopeless than one;(ii)
f’(0) <
andf
hasanasymptote withslopegreater thanone.The one-dimensional case suggests also the useof minorants andmajorants for theproofof fixed point theorems. The following theorem is an example for suchasituation
(cf.
Theorem7.4).
THEOREM 13.7. Let
f
:P P be a map which is compact on order intervals.Suppose
thatf
possessesacompletely continuous, increasing,asymptotically linear majorantf’P
P such thatf’()
does not have a positive eigenvector to aneigenvaluegreaterthanorequaltoone. Then
f
hasafixed
point.Proof. By
Theorem 13.5,there existsafixed point j)off.
Consequently,sincef
is an increasing majorant forf,
the mapf
maps the order interval[0,
p] into itself.Hence theassertionfollows from Schauder’s fixedpointtheorem. [314.
A
multiplicity result.Let f:[9, 93] -
be acontinuous functionon some nontrivial interval[, ]
such that (f(y)-)(.9 -f())
> 0. Then the intermediate value theorem implies the existence ofa fixed pointoff
in(, ).
This elementary fact has been generalized by the theorem on the compression of a cone and the theorem on the expansion of a cone, respectively.
In
fact, in the one-dimensional case, these two theorems reduce precisely to the above assertion.However,in practicalcases,it turns outthat thehypothesesof Theorem 12.3 arerather difficulttoverify.
In
fact,in ordertoestablishthesehypotheses,onehas tofindapriori boundswhich,inthecaseof nonlinearBYP’s,
say,is avery difficult task.For
astraightforward generalization of the one-dimensional case, itseems to be naturaltoreplacethe interval[,
j)]byan order interval andnotbyaconical shell. Indeed, if we suppose that f:[y, p]--,E
is increasing and compact, then Corollary 6.2(or
Schauder’stheorem)
guarantees theexistence ofa fixed point, provided=<
f(y) andf(p)=< p.
The advantage of this theorem is that one has to verify only two conditions, namely one has to show that the "endpoints" of the order interval [y, p] are mapped inside of [y,:9]. In
general, this is a much easierproblem thantoverify the hypotheses ofthetheorem on thecompression ofa cone, where one has to verify infinitely many conditions.Onthe other hand,it iseasytogiveexampleswhichshowthatamapwhich mapsthe "endpoints" oftheorderinterval
[, 39]
outside of[., ],
does notalways have a fixed point. Consequently, it does not seem to be possible to generalize the one-dimensional result on the "expansion of an order interval" to higher dimensions. Nevertheless, it is the purpose of the following considerations to show that,given certain additionalproperties,it ispossibletogeneralize the one-dimensional case inastraightforwardmannertoorderintervals.Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
The multiplicity results of this paragraph are based upon the following generallemma.
LEMMA 14.1. Let Xbearetract
of
someBanach space and letf
X X beacompactmap.SupposethatX and
X2
aredisjointretractsof X,
and letUk,k 1,2, be open subsetsof
X such that Uk c Xk, k 1, 2.Moreover,
suppose thatf (Xk)
c Xkand that
f
hasnofixed
pointsonXk\
Uk,k 1, 2. Thenf
hasatleast three distinctfixed
pointsx,Xl,x2 withXk Xk,k 1,2, andxX\(X1 X2).
Proof. By
the additivity property for the fixed point index,2
i(f
X\(U1 U U2), X)
i(f X,X)
i(f Uk,X).
k=l
The permanence property implies i(f, Uk,
X)-
i(f, Uk,Xk),
and the excision propertygivesi(f Uk,Xk)
i(f Xk,Xk).
Consequently,2
i(f,
X\(UI U2), X)
i(f, X,X)
i(f,Xk,Xk).
k=l
Itiseasytosee
(cf.
theproofofSchauder’sfixedpointtheorem)
thati(f, C,C)
for every compactself-mapf
ofanarbitraryretractC.Hence
i(f,X, X)
i(f,X
1)
i(f, X2,X2)
and, consequently,i(f,X(U1
(.JU2),X)
1.The assertion followsnowfromthe solution property of the fixed pointindex.
THEOREM 14.2. Let
(E, P)
beanOBSwhosepositive conehasnonemptyinterior.Supposethat there exist
four
pointsYk,k E,
k 1,2,withandacompact,
Y
strongly_-< f(;),
increasingf()
map<1, f" .Pl, ]
<f(Y), -
Esuch thatf(J) --< .
Then
f
has at least three distinctfixed
points x, Xx,x
such that1 <-
xl <<f;2 ((
Proof
X22, Let
andXf2 [1, 2] =
X1"
andXk
[Yk,.gk], k 1,2. ThenX,
X1, andX2
areretractsofEwith
Xk
c XandX1 f’l X2
" Hence
Xk,k 1, 2,is a retract ofX. Moreover, sincef
is increasing, the hypotheses imply thatf(X)
X andf(Xk) =
Xk, k 1, 2. Sincef
is strongly increasing andf(l)
<.91,
it follows fromCorollary 6.2 thatf
has amaximal fixed point1
inX1
and1
<<.91.
Con-sequently, X has nonempty interior U in X and
f
has no fixed point on the boundary X \U ofX inX. Similarly, theexistence ofaminimal fixed point off
inX2
and the fact thatf
is strongly increasing imply thatX2
has nonempty interiorU2
in X and thatf
hasno fixed point on Xz\U2.
HenceLemma
14.1is applicableand the assertion follows. [-1
Forthesimplicityof thestatement we have chosen the somewhat restrictive hypothesesgiven above.Itiseasyto seethat thesameproof appliesto moregeneral
Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
situations.Forexample, itsufficestoassumethat
f
isincreasing(insteadof beingstrongly increasing), ifit isknown that the maximal fixed point
1 off
in[1, 1]
satisfies << and that theminimalfixed point
2
in[2, 2]
satisfiesY2
<<2.
Moreover,
the hypothesis thatf
be increasing can be completelydropped if itis known thatf
maps each of the order intervalsX,
X1,andX2
intoitselfsuch thatf
hasnofixedpointsontheboundaries of
Xk
inX, k 1, 2.For example,this can be guaranteed if the existence of strongly increasing majorants and minorants ispresupposed(cf. [19.
Thm.4]).
Ingeneral,it cannot be assertedthat the thirdfixed point xliesbetween
1
and
Ys.
In fact,it cannot evenbe shown that1
< x<
2, wherexl isthemaximal fixed pointoff
in[/1,/91] andff2
isthe minimal fixed pointoff
in[Y2, J2]" However,
since byCorollary 6.2
f
also hasa minimal fixedpoint anda maximal fixedpoint in the large order interval[Yl, P2],
we obtain the followingtrivial corollary to Theorem 14.2.COROLLARY 14.3. Let the hypotheses
of
Theorem 14.2 besatisfied.
Thenf
hasatleast three distinct
fixed
points xl x2,x3such thatxl <<x2 << x3.Suppose
it isonlyknown that the compactstrongly increasingmapf
mapsoneorder intervalintoitself.Then, by Corollary 6.2,
f
hasaminimal fixed pointff and amaximal fixed point. Suppose,
inaddition, that itis known thatff<.
Is ittrue that
f
has athird fixed point in this situation? The following theorem gives an answertothisquestion.THEOREM 14.4. Let
(E, P)
be an OBS whose positive cone has nonempty in-terior.Suppose
that< 19
and let f’[y,19]
E be a strongly increasing compact map such that <f
(y) andf () < .
Suppose, inaddition, that the minimalfixed
point and themaximal
fixed
point are distinct, and thatf
has stronglypositivederivatives at and2, respectively. Then
f
possesses at least three distinctfixed
points provided
r(f’())
4 andr(f’())
4: 1.Proof
It follows from Proposition 7.8 thatff and are weakly stable fixed points.Hence
the above hypotheses imply thatr(f’(2))<
andr(f’(Y))<
1.Since
f
isstrongly increasing,2<<,
andLemma7.5 implies theexistenceof pointsPl
andY2
such that 2 <Pl
<Y2 < , f(Pl) </91,
andf(Y2)
>Y2. Hence,
an application of Theorem 14.2 with1 Y < Pl < Y2 < 2 ;--
gives the asser-tion. 1By
specializing the above theorem to the one-dimensional case, it is easily seenthat thehypothesesconcerning thespectralradiicannotbe omitted.15. Applications to nonlinear elliptic boundary value problems. Throughout this paragraph, we assume that the regularity hypotheses
(R)
of 9 are satisfied.Weconsidermildlynonlinearelliptic
BVP’s
of the form Luf (x, u)
in,(1)
Bu
0 onc,
andwe usethe notationsandconventionsof 9.
Firstweprove the existence ofapositivesolutionfor the BVP
(1)
inthecase thatzero is a solutionand thatf(x,
isasymptotically linear.THEOREM 15.1. Suppose that the
function f C( +) satisfies f
(.,O)
0and
f O,
and letf
havea continuouspartial derivativeDzf
in a neighborhoodof
Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
668 HERBERT AMANN
zero such that
Daf(’, 0) C;’(). Moreover,
suppose that there exists afunction fo C()
such thatlime_, - if(
x,) f(x),
uniformlyinx.
Finally, suppose thatD2f
(x,O)
> 0andfo(x)
> Ofor
almost all x.
Denote by
20
the principal eigenvalueof
the linearEVPLv 2D2
f O)v
in),By 0 on
,
and let
2
bethe principal eigenvalueof
thelinearEVPLv-fv
in,By 0 on).
ThentheBVP
(1)
has atleastone positive solutionprovidedeither (i)o
>and,
<1 or(ii) 20 <
and,
> 1.Proof.
It follows from Theorem 4.3 that the principal eigenvalues2o(D2f(., 0))
and2o 2(f)
exist andare positive.By Lemma
9.7,theBVP (1)
isequivalenttothe fixed point equationuf(u)
in
E,
wheref’P E
is completely continuous. Clearly,f(0)=
0, and(cf.
Lemma 9.1) f
is asymptotically linear and right differentiable at zero with the strongly positive compact derivativesf’. +(0)
andf’(),
respectively. Finally,itr(f())
Hence the assertion is easily seen that21= r(fe.+(0))
and2
follows from Theorem 13.6 and Theorem 3.2.
It is easy to remove the assumptions that
Dzf(x, 0)
> 0 andfo(x) >
0 for almost all x.
Weleavethedetails tothereader.Itis importanttoobserve that the results of 12 and 13 donotpresuppose that
f
be increasing. This has the important consequence that those results are applicabletosemilinearellipticBVP’s,
that is,toproblemsof the formLu
f (x,
u, gradu)
infl,Bu g on
.
For
simplicitywedonotgive further details.Forsomeapplicationsinthis direc-tion wereferto 11, Chap.7].
By
applying Theorem 14.2(more
precisely, Corollary14.3)
to the mapfe
weobtain thefollowingmultiplicity result.
THEOREM 15.2.
Suppose
that thereexist asubsolutiondp astrictsupersolution1,
a strictsubsolutionb2,
andasupersolutiont2for
theBVPLu
f (x, u)
in,
(2)
Bu g onsuch thatdpl <
11
<)2
<112.
Then theBVP(2)
hasatleast threedistinctsolutionsui, 1,2, 3,such that
c1 <=
ul < u2 < u3.<-_ 2.
Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
Thefollowingtheorem appliestothe case where thereisonlyonesubsolution andonesupersolution butit isknown thattheminimalsolutionand the maximal solution
(cf.
Theorem9.4)
are distinct. The proof again follows by a straight-forward applicationof Theorem 14.4tothemapre.
THEOREM 15.3.
Suppose
that there exist a strict subsolutiond
and a strictsupersolution
b for
theBVP(2)
suchtiatd . Moreover,
suppose that the minimal solution u f and the maximal solution u2 fi in the order interval[b, ]
aredistinct. Let
f
be continuouslydifferentiable
with respect to the second variable and suppose that thefunctions
x-oDEf(X ill(X))
i- 1, 2, belong toCt’().
Then thereexists atleast onethirdsolutionu3 such thatu u3 u2,provided eachof
thelinearBlp,s
Lv DE
f (x
ui(x))v 0 inBy 0 on
,
l, 2,has thetrivial solutiononly.
16. Notesantiremarks.Thefixedpointindexis, of course,awell-known tool in fixedpointtheorywhich canbedefined inmuch greater generality
(cf. [80]-[82]).
The permanence property is a special case of the more general commutativity property(e.g.,
[80]-[82]).
The possibility of thederivationof the fixedpointindex from the Leray-Schauder degree is well known to the specialists in this field.Itisimplicitly containedinthe papers byBrowder, Nussbaum,and others.
The theorems"onthe expansion andonthe compression ofa
cone"
aredue to M.A.
Krasnosel’skii (cf. [11], where a lengthy proofis given, which avoids degree theory). In the case wherethe positiveconehas interior points, adegree theoretic proofhas been given in[83].
The short proofsgiven in thispaper have been discovered independently by T. B. Benjamin[84]
and R. Nussbaum[85],
where the latter author had been motivated by Hamilton’s generalizations[86]
of Krasnosel’skii’s theorems. Clearly, these theorems can be proved (by proofs whichareidenticaltotheonesgiven
above)
wheneverafixed point index is known toexist,hence,inparticular,inthecaseofset-contractions(cf. [85], [87]).
Foraresult which iscloselyrelatedtoTheorem 12.3 andwhichcould easily beprovedby the abovemethods,wereferto
[88]. An
extensionof Krasnosel’skii’s theoremstoFr6chetspacesisgiven in[89].
The theorems on cone-compressions and cone-expansions are, as a rule, difficult to apply since their hypotheses are not easy to verify. Applications to two-point
BVP’s
andtointegral equationsare contained inI11
andinthepapers[90], [91
byLaetsch.For applicationstoperiodicsolutions ofordinary differential equations, we refer toI11], [85], [88], [92].
Applications to partial differential equations are given inI11], [93]. In
thisconnection, we again point out the im-portance of the fact that there are no monotonicity assumptions for the map.Consequently, these theorems are applicableto semilinearor even moregeneral quasi-linear elliptic
BVP’s
(cf.[11], 93]).
Part (ii)ofLemma13.1 isduetothe author
[24]. Its
importanceliesinthefact that it gives information about the fixed point indices ofcertain fixed points(cf.
the remarks preceding Theorem13.2).
Theorem 13.2 and Corollary 13.3 arenew,but,of course,theyaretrivialapplicationsof thebasicLemmas 12.1 andDownloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
13.1.
Lemma
13.4 is due to the writer[24].
Theorem 13.6 is contained in [11, Thm.4.11 and Thm.4.16].
The above prooffollows[24] (cf.
also[87]).
Theorem13.7iscloselyrelatedto [11,Thm.
(4.9’)].
The results of 14aredue tothe author
(cf. [19],
where it isassumedthatP is a normalcone).
Originally, Theorem 15.2 and Theorem 15.3 were provedby means ofa direct application of the Leray-Schauder degree theory[94].
These resultsweremotivatedbycertainconjectures andnumericalfindingsin chemical reactortheory (e.g.,[95], [96]). An
application of the above multiplicity resultsto suchaproblemwasgivenbyParter [97].
The multiplicity results of 14 could also be appliedtoaproblem studiedby Rabinowitzin[58].
Further applications are giveninChapterV.CHAPTER
IV.
Nonlinear eigenvalueproblemsand bifurcation.Inthischapter, westudy fixed point equations in OBS’s in the presence ofa nonnegative real parameter.By
using topological methods, we derive some important properties concerning theglobalstructureofthe solutionset. Inaddition, bycombining the topological methods with the results of ChapterII,
we study the question of multiple positivefixedpointsatafixed parameter value.In 17we prove that thesolution set ofan equation of the form x f(2,
x)
containsanonempty closed connected subset which is unbounded in+
P.In18 westudythe situation
wheref(, 0)
0 for all2s/.
Inthis case, bifurcation from thelineof trivial solutions/ {0}
canoccur.We
give sufficient conditions for thistobe the case, andweshowthat, infact, thereexists an unbounded sub-continuum of positive solutions.In
19 we discuss briefly the situation where"bifurcationfrom infinity"canoccur, that is, the case of asymptotically linear maps.
The central section ofthischapter, orevenof thewholepaper,is 20.
Here,
bycombining the topological and the monotonicity results ofChapterII,
wegive amoredetailed studyof thestructureof the solution set. These results enable us thentogive lower estimates for the number offixedpoints forafixedvalue of the parameter 2.Section 21 containssome applicationstoelliptic
BVP’s,
and 22is devoted tohistoricalandbibliographicalremarks.17. Global continua of positive fixed points. Let
(E, P)
be an OBS whose positiveconeis nontrivial, that is,P: {0}.
Inthischapterwestudyequations of the form(1)
xf(2, x),
where
f: +
x P Pis acompletelycontinuous map.In
otherwords, westudyfixedpoints of one-parameter families ofcompletelycontinuous conemaps.
Forsimplicity,wesuppose that the parameter2 belongsto
N+.
There is notmuch loss of generality in this assumption. In fact, every half-open interval
[e,/3)
can be considered as the image ofN+
by a smooth (in fact, analytic)strictly increasing function
. Hence
amap g:[e,fl)
x P-. Pcanbe substituted by the mapf: N+
x P--, P defined byf(2, x):
g(q(2),x),
where, roughly speaking,f
has the same properties as g does.In
the case of an open interval(e, fl)
R or of a closed interval[0, fl] ,
it is easy to see how the following assertionsandproofshavetobe modified in ordertohold in thesecases too.Downloaded 08/01/13 to 128.210.126.199. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php