Photocopying permitted by licenseonly the Gordon andBreachScience Publishers imprint. Printed in Malaysia.
Qualitative
Properties
of
Solutions
to
Elliptic
Singular
Problems*
S.BERHANU
a,
F. GLADIALIband G. PORRUb,t
aMathematicsDepartment,Temple University, Philadelphia,PA19122,USA," bDipartimentodi Matematica, Via Ospedale 72, 09124, Cagliari, Italy
(Received23 February 1998,"Revised7August1998)
WeinvestigatethesingularboundaryvalueproblemAu/u-’ 0 inD,u 0 onOD,where
3’>0.For7>1,wefindtheestimate
lu(x)
bo62/(n+l)(x)l
<whereb0dependson 7only,6(x)denotes thedistance from x toOD and/3is asuitable constant.For3’>0,weprovethat the functionu(+7)/2isconcavewheneverDis convex. Asimilarresultiswellknownfor the equationAu+up 0,with0_<p_<1.Forp 0, p and3’_> weproveconvexity sharpness results.
Keywords." Singular equations; Boundary behaviour; Convexity 1991 MathematicsSubjectClassification:35B45, 35E10
1.
INTRODUCTION
Let N
>
andletDCR
Nbeabounded smoothdomain.In[1-3,5],
the problemAu
upinD,u(x)
--.+oe
asxOD
isinvestigated.Itisproved that such a problem, for p>
1, has a unique positive solutionu(x).
Moreover,
forp>
3thereexists aconstant/3
>
0 such that.(x)
2/(1-P)(x)
</36(x)
in D,(1.1)
*Partially supportedby MURST 40% andbyaUniversity of Cagliari Research Project. Correspondingauthor. E-mail:porru@vaxca1.unica.it.
wherea0is a constantdependingonponlyand
(x)
denotesthedistance from xtoOD [1,3,5].
In
[6,13]
it isprovedthat theproblemAu+uP--0
inD, u-0 on0D,(1.2)
for p
<
0 has a unique positive solutionu(x)
continuous up to the boundaryOD.
Inthesamepapersit isalsoprovedthat,for p<
1,thereexistpositiveconstants
A, A
suchthat/62/(1-P)
(x)
<_
(X)
<_ A62/(1-P)
(x).
InSection2 of the present paperweshall provethat, for p
<
-1 thereexists/3
>
0such thatu(x)
82/(1-P)(x)
<
t6(P+I)/(P-1)(X)
inD,
(1.3)
where
b0
is a constant depending onp only. We emphasize that theconstantsa0in
(1.1)
andb0
in(1.3)
areindependent of the geometry ofthe domainDandevenofthedimensionN.Wealsofind aboundaryestimateforthecase p 1.
Now,
considerproblem(1.2)
with0<
p<
1.Itiswell known that this problemhasapositivesolutionu(x).
Suchasolutionisnot concave evenin the radialcase. Indeed, ifu
u(r)
thenthe corresponding equation readsas(r
N-1ut)
-
rN-1upO,
which implies that
(r
N-1u’) <
0andu’(r) <
0in(0,
R].
Sinceu(R)
0,wehave
N-1
u"
(R)
+
u’ (R)
0,(1.4)
whence,
u"(R) >
0.Thisshowsthatu(r)
is not concave nearr R. It is known[9,10]
that ifu(x)
is a solution of problem(1.2)
with 0
<
p<
in a convex domainD then, the function v u(l-p)/2 isIfp 1, instead of problem
(1.2)
weconsiderthe followingAu+Alu=0
inD, u=0 on0D,where
A1
isthe firsteigenvalue ofD.Ifuisthe (positive) correspond-ing eigenfunction then, v=logu is concave whenever D is convex(see [9-11]).
In
[9,
p.122]
itiswrittenthatifu(x)
is asolutionto(1.2)
ina convexdomainDthen thefunction
V(X)
S-(p+l)/2dsisagoodcandidatetobeconcave.
As
recalledabove,thisfactisknown-to betruefor 0_<
p<
1.InSection3of the present paperweshall prove that thestatementinaboveisalsotruefor p<
0. Furthermore,weshall find the following sharpnessresults. Letu(x)
be asolution of(1.2)
and let e>
0.Then:(1)
ifp-0, the function V--’U1/2+e is not concave in some convexdomain;
(2)
ifp 1, the functionv u is not concave in some convexdomain;(3)
ifp<-1, the function V"-U(1-p)/2+e is not concave even in theradial case.
2.
BOUNDARY
BEHAVlOURIn this section the domain D is assumed to be bounded, smooth and satisfyingauniforminterior and exteriorspherecondition.For7
>
1,we considerthe followingproblemAu+u-7=0
inD, u=0 onOD.
(2.1)
LEMMA
2.1If
u(x)
isapositive solutiontoproblem(2.1)
with "7>
1, thenlim
u(x)
[("7
+
1):
]
1/(+1)Proof
Let R>
0. Consider first the case of DB(R),
a ball with radius R. The corresponding (radial) solution zz(r)
satisfies, for 0<r<R,N-1
z"
z’
+
z-’
O,
z’(O)
O,
z(R)
0.(2.2)
Multiplying(2.2)
byz’
wegetz"z
+
z-z
<
0. After integrationover(0,
r)
wefindThe latter inequality implies
zt
>
z(1-,)/2
(2.3)
Insertionof
(2.3)
intoEq. (2.2)
yieldsN-l(
211/2z(1-’)/2+z-’<O’r
whence
I
z(+’)/2+
<
0.r
Lete
>
0. Sincez(r)
0as rR,
thereexistsr,<
Rsuch thatz"+
z-(1-
e/2)
<
0 in(r,,R).
Sincef(r) <
0,fromthisinequalitywe findand
(z’())
(z’(r,))
2 2e/2
[zl_.(r)
zl_7(re)]
>
0.1-7
Forsomer0
>
re,wealsohave(z’(r))
2 eZ1
2
>
7-1
-7(r)
VrE(ro,
R)
or
z(7_l)/2(r)zt(r)_(l_6.)l/2(,
2,
1/1/2
Integrationover
(r, R)
yieldsz(r) > (1
e)l/(,+_)L
2(’-[(’
+
1)
211)
1/(7+1)
(R- r)
2/(’+1)Vr
(ro, R).
(2.4)
Nowconsiderthe annulusD
B(R,
).
Letww(r)
beasolution toproblem
(2.1)
inB(R,
R).
We have,forR<
r<
R,
N-1
w"
+
w’
+
w-
O,w(R)
O,
w()
O.(2.5)
Ifr
is apointin(R, R)
wherewt(rl)
O,from(2.5)
wefind, forR<
r<
r,(w,(r))2
fr,
2wl-’(r)
wl-f(rl)
(N- 1)
7(w’(t))
dt+
27-1
(2.6)
By (2.5)
wealsofind/r
tN-lw-’(t)
dt<
w’(r)
r--
w-’
(t)
dt. RN-1Using the latter inequality togetherwith de
l’pital
rulewe find lim(’7-1)f(1/t)(w’(t))
2 dt rRr
-1f’
w-’(t)
dt limw’
!r)
<
limRN
Using del’(3pital
rule once more we get(9’- 1)
ff(1/t)(w’(t))
2dt
r
v-1w(r)
lim lim 0.
-
w-()
-<
--w’()
Recall that
w(r)
0asr Randthat, by(2.6),
w’(r)
-+oeas r-
R.If the integral of(w’)
2over
(0, rl)
is finite then we cannot apply del’pital
rule, but in this case, the limit is trivial.As
aconsequence ofthis estimate,
(2.6)
yields, for agiven e>
0,(w’(r))
<
w-(r)
Vr
(,r).
2
7-1
Integratingover
(R, r)
with r<
&, one finds7+1
w(7+1)/2<(2)
1/27-1
(1
+
e)l/2(r-
R),
orw(r) <
(1-I-)1/(7+1)
F2(
’c--ir(’r
+
1):1
1/(7+1)(r-R)2/(+1).
(2.7)
Recall thatDhas auniform interiorand exteriorsphere condition.Take apoint PE
OD. We
may assumethat P(R,
0,...,0),
thatDiscontained in the annulus
B(R, R)
with center in(2R,
0,...,0)
and R large, and thatDcontainsthe ballB(R)
with center in(0,...,
0).
Notethat
B(R, R)
andB(R)
are tangenttoOD
in P. Ifu(x)
is the solutionof problem
(2.1)
inD,ifw(x)
isthesolution intheannulusB(R,[)
and ifz(x)
isthe solutioninB then, bythe comparison principle,wehaveLet 0
<
r<
R. Ifwe take a point x(r,
0,...,0)
then, by using(2.4)
and
(2.7)
itfollows that(1
)
1/(7+1)[(")/+
1):]
1/(3’+1)
1/(7+1)
<
u(x) < (1
+
)2/(7+1)l(")
Sincee is arbitrary, the lemma follows.
THEOREM 2.2
If
U(X)
isapositive solutiontoproblem(2.1)
with "7>
1, thenthereexists>
0such that+
1):.]
’/(7+)(2/(7+1)(X)
i
<
fl(7-1)/(7+l)(x)
VX
ED,
where
6(x)
denotes thedistancefrom
xtoOD.
Proof
Following[5],
putW(x)
bo52/(7+l)(x)
+
fitS(x),
with
and 0
</3
willbefixedlater. Writing instead of8(x),
onefinds2
62/(9,+1)_
flSi
Wi
bo
18i
q-7+1
Since
6i6i:
and-AS= (N- 1)K=
H, Kbeing themeancurvatureof the levelsurfaces of6(x),
wefindA
Wb
-76-27/(7+1) -Jr-b0
2 H8(1-7)/@+1)+
fill.
(2.8)
Onthe otherhand, byapplyingTaylorexpansiontothefunction
f(t)
(bo5
/(+)+
t)
-onefinds
Weclaimthat,for
5(x) _<
5o
smalland/3
large,wehave_q,b--l/35-
+
3’(7
1)
2+
b--r-z/32ts-z/(.r+
2
< b0
Hi5(1-3’)/(3’+1)+/3H.
7+1
(2.10)
Indeed,letus rewrite
(2.10)
as’T
I_bo
_+_/(,.y
_+_i)5(,-I)/(,+I)]
b-’r-2
-y 2</5b
-’-
q-bo
H(52/(’+1)q--H5.
"7+1
The left handside canbe made negative for
5(x)
<
5o
bychoosing/55.r_)/(.+1)
b____o
(2 11)
7+1"
Now,
decrease5o
andincrease/3untilthe right handside ispositive.This ispossiblebecause, bythesmoothness ofD,
HisboundednearOD.Theclaim isproved.
By (2.8)-(2.10)
itfollows thatA
W+W
-’r<0on{xD:5(x)<50}.
Obviously,
W(x)
u(x)
onOD. Increase/3
and decrease5o
accordingto(2.11)
until we haveW(x)> u(x)
for5(x)=5o.
Observe that this isHence,
bythe comparison principle for elliptic equations[7,Theorem9.2]
wefindu(x) <
bo62/(7+l)(x)
-+-
(2.12)
for all xEDwith
6(x) <_
60.
To completetheproofofthetheorem,let
W(x)
with
bo
and/3
asbefore.Wefind-AW
bff’6
-2’/(7+1)+
b0
2H6(1-’)/(’7+1)
"7+1
(2.13)
By
applyingTaylorexpansiontothefunctionf(t)
(b0t52/(’+1)
t)
-onefinds
Weclaimthat,for
6(x) <_
60
smalland/3
largewehave2
b-’)’-lfl6
-1> b0
Ht
(1-’y)/(’+I)3H.
(2.15)
Indeed,inequality
(2.15)
canbewritten asTheclaimfollows easily.
By (2.13)-(2.15)
itfollows thatIf necessary,
increase/3
and decrease60
untilwehaveW(x)< u(x)
for6(x)
0.
Againbythe comparison principlewe find
>
(2.16)
for all x Dwith
(x)
<
0.
The
constants/3
and0
in(2.12)and (2.16)can
be takenwiththesamevalues.Therefore,
u(x)
2/(7+1)
(X)
<
3
(’-1/(’+1
(x),
for allx EDsuch that
6(x)
<
60.
Increasing/3 again, the theorem follows.By
using Theorem 2.2 one finds easily that the gradient ofu(x)
isunboundedneartheboundaryofD.For3’ 1, Theorem 2.2fails.Now
wegiveadirectestimateof thegradientinthiscase, improvingaresult
of[13].
Considerthe followingproblem:
Au+p(x)u
-1=0
inD, u=0 on0D,(2.17)
whereDis a domainsatisfyinganinteriorspherecondition andp(x)is a
smooth function satisfying
0
<
pl<_ p(x).
(2.18)
PROPOSITION 2.3
If
u(x)
is a solution to problem(2.17)
with p(x) satisfying(2.18)
then the gradientof
u(x)
isunboundedineachpointof
OD.
Proof
Letzz(r)
bethesolutionofthe followingproblem: N-lzt
z"
+
plz-
=0, O<
r<
R,
z’(O)
=O,z(R)
=O.(2.19)
By (2.19)
we find(rN-lzt)
<
0, whencez’(r) <
0 in(0, R).
Multiplyingby
f, Eq. (2.19)
impliesIntegratingon
(0,
r)
we find(z’)
+
Pllogz-
<
0, whence,(zz’)
2<
2plz2(r)log
z(0)
The latterinequalityimplies that
limzz O. r-+R
Let
ro
E(R/2, R)
suchthat_zzl
<
pR
forr0<r<R.
4(N-1)
Using
Eq. (2.19)
once more wefind,on(ro, R)
zz p
<-
p<
z r z
\--47r
2zSince z
<
0,from the latter inequalitywefindZtt
Z Pl Z2 Z
Integrationover
(ro,
r)
yields(z’(r))
2(z’(ro))
:z>
PllogFinally,wefindthat
z(ro)
in
(r0,
R)
-z’(r)
>
logNow consider problem
(2.17)
in D. We claim that the interior derivative ofthe solution u approaches+ec
as x OD. Indeed, forPo
EOD,
consider aball BCD and tangent toOD
atPo.
Let zbe thesolutionofproblem
(2.19)
in suchaball.Since Pl<_
p(x),onefindsAz
+
p(x)z
-1>_
0inB.
By
the comparison principle[7,Theorem9.2]
between the last inequality andEq. (2.17)
one findsthatu(x)
>
z(x)
inB.As
aconsequence,ifPis apointinBcloseto
P0
wehaveu(P)
u( 0)
>
z( 0)
Using the last inequality togetherwith
(2.20),
the proposition follows.3.
CONVEXITY
Inthissection,DC]Nisassumedtobebounded,smooth andconvex.
Let
u(x)
beapositivesolution totheproblemAu+u-’-0
inD, u-0 on0D,(3.1)
with0
<
7.Wewant toprove that thefunctionv-/,/(1+,,/)/2
is concave.Using
Eq. (3.1),
wefind[11-
1+.7]
v(x)=0
onOD.
(3.2)
Av-
71Vv12
+
inD,v
+7
2Wefirstshow that thefunction
Q(x)
7iVvl
2+
1+7
(3.3)
LEMMA3.1 Ifu(x)isapositivesolutiontoproblem(3.1)with’y>1,andif
v u
(1+’)/2,
then thefunction
Q(x)defined
asin(3.3)
ispositiveinD.Proof
Wehave,),2
Q(x)=
2 with 2Itsufficestoprovethat,when
<
7,P(x)
<
O.Fore
>
0,consider the solutionu(x)
totheproblemAu+u
-’=0 inD, u=eonOD.
The correspondingfunction
P(x)
IVul2
u’-2satisfies
P Ukblk l’l-’Y1"l
Note
that the summation conventionoverrepeatedindicesisused.ThisequationtogetherwithSchwarz inequality yield
(Pi-
b/-TUi)
2 (Zgkblki)2IVUl2UkiUki,
fromwhich itfollowsthatPi- 2u-’rui
Pi
-+-
u-2"r.
UkiUkiiX7ul
2Usingthis estimate aswellas
Eq. (3.1)
wefindAP
+
2u-Tui- Pi
17ul2
ei>_
O.By
the classical maximum principle, P attains its maximum valueeitherwhen
Vu
0or ontheboundaryofD.SinceDisconvex,Hopf’s boundary lemma prevents Pfrom having its maximum value onOD
(see [14]). Hence,
ul-7
M-7
4-< <0,
-7-
1-7
where
M,
denotes themaximumvalueofu(x)
u,(x).
The lemma followsaseO.
Fordiscussingtheconcavity ofsolutionsto
Eq. (3.2),
weuseKorevaar function[9,11]
C(x,y)
2v((x
+
y)/2)
v(x)
v(y),
x,ye
D.(3.4)
Thefunction
v(x)
is concave inDifandonlyifC(x,
y)>
0inD D.PROPOSITION 3.2
If
v(x)
isa positive solutiontoproblem(3.2)
thenthefunction
C(x,
y)defined
as in(3.4),
cannothaveanegativeminimuminD.Proof
By
Lemma 3.1,thefunctionatthe right handsideofEq. (3.2)
isnegative.
Moreover,
suchafunction isincreasingwithrespectto v.The prooffollows easily by[9,
Theorem3.13,p.116].
Seealso[8,11].
To
get thepositiveness ofC(x,
y)ontheboundaryofDD,
weprove the followingLEMMA
3.3If
V(X)
&apositive solutiontoproblem(3.2)
andif
yEOD,
then the
function
(x)
2v(z)
v(x),
with z(x
+
y)/2
Proof
If xEOD,
wehaveb(x)
2v(z)
>
0.If xED,
bycomputationwefind
v
V(z)
V(x),
x
Xv(z)
Xv(x).
Using
(3.2),
the latter equation yields2v(z)
+
+
v-(-
(Alvv(x)
+
withA
(1
"y)/(1+
"y)andB(1
+
,)/2.
Using(3.5),
this equationcanberewritten as
(A]7v(z)]2 B)
/k/
+
ai)i
2v(z)v(x)
+
(3.6)
with suitable smoothfunctions a
i.
By
Lemma 3.1,the coefficient ofb
in(3.6)
ispositive.Hence,
bythe classicalmaximumprinciple,we inferthatb(x)
attains its minimum value on the boundaryofD. The lemmaisproved.
COROLLARY 3.4
If
v(x)
isa positive solutiontoproblem(3.2),
then thefunction
C(x,
y)defined
asin(3.4)
isnon-negative onD
;.
Proof
Itfollows from Proposition 3.2 andLemma3.3.THEOREM 3.5
If
V(X)
&a positive solutiontoproblem(3.2)
&a convex domainDthenitisstrictlyconcaveinD.Proof
By (3.2)
andLemma3.1, thefunction w -v satisfiesXw-
%
(AlVwl
+
B) >
0,w
withA
(1
7)/(1
+
7)
andB(1
+
"),)/2. By
Corollary 3.4,w is convex.Moreover,
ww(Al7wl
2+
B)
-1 is convex.By
atheorem ofKorevaar and Lewis[12],
weconcludethat theHessianofvhasaconstantrankinD.
Now,
vattainsits maximum inDon acompact subsetKc
D.Iris well knownin this case[4]
that in anyneighborhood UofK,
thereis apointPEUwheretheHessian ofvisstrictly negative.Itfollows thatvisstrictly
concaveinD.
COROLLARY 3.6
If
U(X)
is a positive solution toproblem(3.1)
then it attains its maximumvalueinDatasinglepoint.Nowweprovesomesharpnessresultsonconvexity.
1.Let D
c ]N
bea convexdomain, and letu(x)
beapositivesolution totheproblem
Au+l=0 inD, u=0 on0D.
(3.7)
Werecall that thefunction
v(x)
uis concave inD
[9,10].
Weprove that the exponent1/2
issharp.Let N= 2,and letDbe the trianglewith vertex
(- / x/-, 0),
(1
/ x/,
0)
and
(0, 1).
Thefunctionu(x,
y)
yrl(1
y)2
3x2|=+
solvesproblem
(3.7)
in this domain.Letb(y)4u(0,
y). WehaveqS(y)
y(1
y).
Of course, thefunction is concavein
(0, 1).
Ifa>
1/2,
(b(y)) is not concave neary 1.IfN
>
2,one canobtainthe result in abovebyusing the functionhi(X)
---(1
XN[
XN)
2N-
il
2.Letu
>
0beasolutionoftheproblemSuchafunction isthefirsteigenfunction ofD.Weknow thatv logu is concavewheneverDis convex.
Isthere anye
>
0 such that v u is concavefor allconvex domains?The answer is negative, as one can seeby the followingexample. Let N--2, and let
D-{0<r<a,
0<0<Hererand 0arepolarcoordinates,m is aninteger andaisthefirst zeroof the mth Besselfunction.Thefirsteigenfunction ofDis
u(x,
y)
Jm(r)
sin(toO),
Jm(r)
being the mth Besselfunction.Itisknown thatJm(r)
behaveslike r asr 0.Hence,
we musttake e_<
1/m
ifwewantthefunction u tobeconcave. Since m canbe choosen arbitrarilylarge,the resultfollows. 3.Let
z(x)
beasolutionof theproblemAz+z-l=0
inB, z=0 on0B,(3.8)
whereBis aball.
By
Theorem 3.5,z(x)
is concave.Letusshowthat,given e>
0, thefunction v z1/(1-)is not concave nearOB.
Indeed,Z--1
l-e,
VZ
(1
/Xz
(1
)v-/Xv
(1
)v
--IIX7vl
2,
Substitutinginto
Eq. (3.8),
wefindvl-ZZmv_
ev-2lVvl
2(3.9)
and since, by
(2.20),
]Vv]
as xOB,
the right handsideof(3.9)
becomes positiveneartheboudaryof B.
As
aconsequence,A v>
0nearOBandtherefore,vcannotbeconcave onB.
Similarly,onecanshowthat,if
z(x)
is asolutionof theproblemAz+z
-’=0 inB, z=0 on0B,with 7
>
then,for anye>
0, the function v- z(l+’)/2+eis not concave nearOB.References
[1] L.Andersson andP.T.Chruciel, Solutionsof theconstraintequation ingeneral relativity satisfying "hyperbolic conditions",DissertationesMathematicae355(1996). [2] C. Bandle andM.Marcus, Largesolutionsofsemilinearelliptic equations: existence,
uniqueness and asymptotic behaviour, J. d’Anal. Math.58(1992),9-24.
[3] C.Bandle andM.Marcus,On second ordereffects intheboundarybehaviouroflarge solutionsofsemilinearellipticproblems,Differentialand IntegralEquations11(1998),
23-34.
[4] R. Basener,NonlinearCauchy-Riemann equations and q-pseudoconvexity, Duke Math.J.43(1976),203-215.
[5] S. Berhanu andG.Porru,Qualitativeandquantitativeestimatesfor largesolutions to semilinearequations,CommunicationsinApplied Analysis(toappear).
[6] M.G. Crandall,P.H.RabinowitzandL.Tartar,OnaDirichletproblemwith asingular nonlinearity, Comm. Part.DiffEq.2(1977),193-222.
[7] D.Gilbarg andN.S.Trudinger, Elliptic PartialDifferentialEquationsofSecondOrder, SpringerVerlag, Berlin,1977.
[8] A.GrecoandG.Porru,Convexity ofsolutions to someelliptic partialdifferential equations,SiamJ.Math.Anal. 24(1993),833-839.
[9] B.Kawohl,Rearrangementsand Convexityof LevelSetsinPDE, Lectures Notesin Math. 1150, Springer-Verlag,BerlinHeidelberg,1985.
[10] B. Kawohl, When are solutions to nonlinear elliptic boundary value problems
convex?,Comm.Part.DiffEq.10(1985),1213-1225.
[11] N.J. Korevaar,Convexsolutions to nonlinearelliptic and parabolicboundary value problems,Indiana Univ. Math.J.32(1983),603-614.
[12] N.J. Korevaar andJ. Lewis, Convex solutions of certain elliptic equations have constantrank Hessian, Arch. Rat.Mech. Anal. 97(1987),19-32.
[13] A.C. LazerandP.J. McKenna, On asingularnonlinearelliptic boundary value problem, Pro.American.Math. Soc. 111(1991),721-730.
[14] L.E. Payneand G.A. Philippin, Somemaximumprinciples fornonlinearelliptic equationsindivergence formwithapplicationstocapillary surfaces andtosurfaces of constant meancurvature, J.NonlinearAnal.3(1979),193-211.