Photocopyingpermitted bylicenseonly theGordonandBreachScience Publishersimprint.
Printed in Singapore.
A QualitativeTheory for Parabolic
Problems under Dynamical Boundary Conditions
JOACHIMVONBELOW*andCOLETTEDE COSTER
LMPAJosephLiouville,EA2597, Universit duLittoralC6ted’Opale, 50, rueF.Buisson,B.R699, F-62228 Calais Cedex,France
(Received16 July 1999; Revised 13September1999)
Fornonlinearparabolicproblemsin aboundeddomainunder dynamical boundarycondi- tions, general comparison techniquesareestablished similar tothe onesunderNeumann orDirichletboundaryconditions.Inparticular, maximum principles andbasicapriori estimates arederived,aswellaslower and upper solution techniques that leadtofunctional band typeestimatesforclassical solutions.Finally, attractivity properties of equilibriaare discussed thatalsoillustratethe damping effectofthe dissipative dynamical boundary condition.
Keywords: Parabolicproblems; Dynamical boundary conditions;Maximumand comparison principles;Upperand lower solutions;Convergencetoequilibria AMS SubjectClassification:35B05; 35B40; 35B45;35B50
1
INTRODUCTION
Theaim of this paperis todevelop a qualitativetheory forparabolic problemsin aboundeddomainunder dynamicalboundaryconditions, i.e.conditionsoftheform
crOtu + cO,
u pu+
hon apart of the timelateral boundary. Throughoutwedealwithupper and lower solutionsor withpairs offunctions withseparatingparabolic
*Corresponding author.
467
defect. First, we derive comparison techniques and monotonicity properties of the flow similartothoseinthenondynamicalcase. Then weestablishthe existence of particular solutionsnotablytheinfimaland supremalsolutionsobtainedforacomparablepair of lowerand upper solutions. Thesetechniques areapplied inorder toobtain attractivity results for equilibria for reaction-diffusion-equations, that, in turn, illustrate the damping effect of the dissipative dynamical boundary condition ontheconvergencebehaviour.
Suppose
fc_
IRn is a bounded domain whose boundary is decom- posedintotwodisjointpartswhere
02f
isofclasst72
and relatively open in 0f.Letu"029t -- n
denotethe outer normal unit vector field on
02f
and0
the outer normalderivative. For
T>
0 we setQr
f[0, T]
and introduce the para- bolicinteriorQ
Trand the parabolicboundaryqr asQr (ft
U02f2)
x(0, T]
and qrOr \ Qr.
This terminology will be justified by the results below. We consider general parabolicequations of the form
Otu F(x,
t,u,Vu, D2u)
=:Flu]
and inequalities associatedtothem,wherethroughoutwesupposethat F"with
Or
respectIR IRton qIRn2D2u. --
IRis increasing(1)
Here the order ,4
<
B between symmetric matrices means that the matrixB-Aispositively semidefinite. Unlessotherwisestated,wedo notrequire strictmonotonicity. Thus,many of the results below include possible degeneracies of theprincipalpart ase.g. the porous medium equation.On
01f (0, T]
weprescribeaninhomogeneousDirichletcondition, while on02ft (0, T]
weconsider the dynamical boundaryconditionB(u)
0 witht()
:=(,, t)o,, + (,, t)Ou ;(, t). (2)
Throughoutwewill assumethedissipativityconditionc>0, r_>0
onoq2fx(0, T]. (3)
Without condition
(3)
blow up and nonuniqueness phenomena can occur.Take e.g. the functionu(x, t) -(T+
Xl0
-1definedontheopenunitball 9t
{x
Et Ilxl12 < }.
ThenusatisfiesOtu 1/2(T+
x, t+2)Au- 2(T +
x,t)llVull
inf x
[0, T)
andxlOtu + O,.,u
0on0f x[0,T),
while ublows up for Tinf andon0f.2
COMPARISON
ANDMAXIMUM PRINCIPLES
Thebasictoolforcomparingclassical solutionsisgivenbythefollowing lemma that generalizes the techniquesdevelopedin
[11
and that can also be used inmoregeneralcases[5].
LEMMA2.1 Let o,
b C(0_.7")
fqC
2’1(O7)
satisfyB(qo) <_ B()
onc92f
x(0, T] (4)
andthetestpointimplication
79
,
7qo2, D2
o<_ D2
===> C9t9 < c9,
at(x, t).
at
(x, t) QT
Then p
< b
onqT-implies qo< b
inQr.
Proof Suppose
thato <
onqT.Sett*
sup{r [0, T]lo < b
on(f
tO02f)
x(0, r)}
and H
(f2
U02f)
x{t* }. Hence
6:=b o >_
0 on H.Suppose
that6(p)=0 for
p=(,t*)H.
If,
then Ot6(p)<_O which isexcludedby
(5).
If E02’2, then Off(p)<_ 0, O,.,6(p)<_0 and(4)
imply tr(p)Ot6(p) 0 andV6(p)
0,whichleadstoDEt(p) _>
0.ThusOt6(p)>
0 by(5),
which contradicts Ot6(p)<_ O. We conclude< b
onH,
andfinally,acompactness argument yieldst* T.
Lemma2.1 yields the following comparison principles and estimates withrespecttothe parabolicboundaryqr.
THEOREM2.2 SupposethatF
satisfies
aone-sidedLipschitzconditionw
>_
u= F(x,
t,w,p,q) F(x,
t,u,p,q) < L(w u) (6)
in
Q
7, Ix]1n X ]n2for
someconstantL>
0, and thereexistsb-
such thatp
<
bcron02f
x(0, T].
Letu,vC(07,)
NC2’1(07,)
satisfyOtu- Flu] < Otv F[v]
<_ t (v)
Thenu
<
v onqT-impliesu<
vinQT..
Proof
We may assume L>
1, b>
1. Fore>
0 set u andb=
v+
eL-
e2Lbt.
Then13() 13(v) +
ee2zbt(2bo"
L-lp) > B()
on0:f
x(0, T]
andata testpointwiththehypothesesfrom
(5)
weconclude,using(1),
0OtV OtU F[V] + Flu]
OtY OtU F[b] + Flu] +
eexp(2Lbt)
<_ Otv Otu +
eexp(2Lbt) < Ot2 Otp.
Lemma2.1 impliesqo
<
b, and,sincee>
0 was arbitrary,u<
vinQT,.
COROLLARY
2.3 Undertheassumptionsof
Theorem 2.2 theinitialbound- ary valueproblem(7)
admitsat mostonesolutioninC(07)
fq C2’1(QT,)"
Otu F(x,
t,u,7u,D2u)
t)O,u + c(x, t)O u p(x, t)u o
U
[qr 3 C(qz).
in
on
02f
x(0, T], (7)
As
usual, thecomparison principle assures the positivity of the flow, if0 has nonnegativeparabolicdefect.COROLLARY
2.4 Under theassumptionsof
Theorem 2.2 and theaddi- tionalhypothesisF(., ., O, O, O) > O,
asolution u EC(T) n
C2’1(aT) of
Otu
F[u] >
0 inQ
T,B(u) >_
0 on02f
x(0, T],
u>O onqr,
satisfies
u>
0inQTy.Next,
wededuceaweak maximumprinciple.THEOREM 2.5
Suppose F(., ., ., O, O) <
0(F(., ., .,
O,O) > O)
and p<
0on02f (0, T].
LetuC(Or)
C2’(QT)
satisfyOtu <_ Flu] <Otu >_ F[u])
inB(u) <
0(B(u) > O)
onO2f (0, T].
Then
max(u,
qrO} m_ax(u, O} (min(u, O} m_in{u, 0}).
Qr qr Qr
Proof
For e>0, apply Lemma 2.1 to qo=u andb=
et+e+maxqr{U, 0}
in the maximum caseandtoo minqr{U 0}
e etandb
uin theminimumcase.Of course, the weakmaximum principle contains a positivitycon- clusion similartotheoneofCorollary2.4under the stronger condition
F(.,.,.,
0,0) >
0,but without(6).
Moreover,
forahomogeneousboundary operatorwededuce the COROLLARY 2.6 Under theconditionsof
Theorem 2.5and, inaddition, p 0on02
X(0, T],
usatisfies
max u maxu
(
min u m_inu).
Another classical a priori estimate
[9]
can be carried over to the dynamical case.THEOREM 2.7
Suppose
uEC(Or)
f3C
2’1(Qr)
is asolutionof
theIBVP(7),
whereFfulfills
anOsgoodtype sign condition3bl,b.
>_ O, V(x,t) QT, Vz
I"zF(x,t,z,O,O) <_ blz + b2 (8)
andp
<_
bcron029t (0, T] for
someconstantb>_
O. ThenOr
A>b,,A_>b qrA- bi (9)
Proof v/ba/(A-bl)}
ApplyLemma
with A>bl,2.1 toA>b and e>0.-
u,b (1 +
We)e
have-’xtmax{maxqr /3()>0 lul,
on029t (0, T]
and with(8)
weconclude thatata testpoint(x0, to)
asin(5),’
Otb Otq A F(xo,
to,,
O, D>_ Ab F(xo,
to,p,O,O)
>_ - -- >_ ( )(- ) > o.
Thusu
< b.
In orderto show -u<
we apply Lemma2.1 to -u andproceedsimilarly.Next,
wederive astrong maximumprinciple forstrongly parabolic quasilinearoperators ofthe formD[u]
:=aik(x,
t,u,Vu) 02u
OxiOXk + f (x,
t,u,Vu)
usingtensornotationandwithpositiveconstants#1and#2such that 0
<_ #1* <_ aJm(
", ", ",")(jn <_ #2(*(
for all
Nn. Moreover,
wehavetoassume p 0 in(2),
thuswedefinet3o(u) (x, t)Otu + (x, t)Ou.
THEOREM2.8
Suppose
thereexists apositiveconstantCsuch thatf(’,
",",P) <_ ClPl (f(’,
",",P) >_ -ClPl) (10)
in
Qr
x".
LetuEC(O.r)
ft72’1 (Qr)
bea solutionof Otu < D[u] (Otu > O[u])
inf2 x(0, T],
and
to(u) _< o (Zo(u) _> o)
ino
x(o, 2]. (12)
Then
/m_in min
maxu=maxu u- uT,
O_.r qr Qr q
and
if
u attains its maximumM(its
minimumm)
atsomepoint(Xo, to)
EQ
T,thenu M u
m)
inQ
toProof
If suffices to show the assertion in the maximum case, in the minimum case weproceed similarly.Ifuattains its maximum Mat
(Xo, to)
withXo 9t thenweconclude u=M for< to
using the classical strong maximum principle for domains, see e.g. [11,IV.26]. By
continuity, this showsmax0Tu
maxqTuandthe strongassertion inthecase Xo
Next,
supposethatuattainsMat(Xo, to)
withXo02f
andu<
Min(0, to].
Then,as029t
is arelativelyopent72-part
of theboundaryof f,we find an open ball B {yEfIly- yol12 < e} c_
9t of radius e>
0 with XoOB. By (10)
and(
11),
usatisfiesalinearinequality with bounded coefficients02U OU
OtU
lJ’m(X, t)
"["J (X, t)
_-’’y__OxOx,.
where we have set
{tJm(x, t) aJm(x,
t,u(x, t), Vu(x, t)), bJ(x, t) Csign((O/Oxj)u(x, t)).
This allows the application of the Friedman- Viborni-Theorem3.4.6of[10]
in ordertoconclude thatOuu(xo, to) >
O.But
this is impossible by(12)
and(3)
and so uhas to attain itsmaxi- muminf x(0, to].
Withthe casealreadyshown aboveweconcludeu M inOr0.
Note that in Corollary 2.6, we had to suppose the differential inequality also in
02F
x(0, T].
Wenote in passing that for simplicity reasons, the same assumption has been made in[6,
Theorem2.7].
But,
following the arguments in the network case[2,4],
the strong maximum principle for quasilinear operators holds also in ramified spaces without the assumption of differential inequality on the interfaces.LOWER AND UPPER
SOLUTIONS
FOR TIME-PERIODICPROBLEMS
Weconsidertheperiodic problem
Otu a(x)Au +f(x,
t,u)
t)O u + t)O ,u t)u o
u=0
u(x, O) u(x, T)
in
F
x(0, T],
on
02F
x(0, T],
on
0
x(0, T],
in ft.
(13)
Weassume
aE
tT(f, (0, 0));
f:
f2 x[0, T]
x/R IRiscontinuousand T-periodic in t;cE
tT(O2f
x[0, T], (0, )),
crEC.(O2f’t
x[0, T], [0, ))
andp E
C(02f
x[0, T])
are T-periodicin t.(14)
Weextend all thecoefficientsbyperiodicitytof x
[0, cx). Moreover,
for the sake of simplicity andas wedonot want to treatexistence results here,weassume tohaveenoughregularity such thatfor every
- >
0, the operator N"C(O_..,.)
xCo(f) C(O__.-)
xCo(f’t),
is continuous,compact, and satisfies
(15)
Range(N) c (C(O-)
fqC2’1(Q))
xC0(f),
where
C0(fZ) (u
EC(fl)
u 0 onc31fl}
and(u, u(., 0)) N(v, uo)
isthesolutionof
Otu a(x)Au + f (x,
t,v) or(x, t)Otu + c(x, t)Ouu p(x, t)v
u=Ou(x, O) Uo(X)
in f x
(0,
on
02fl
x(0, -],
on
01
f x(0,
in f.
Note that the operatorNis well defined dueto Corollary2.3.
More-
over,in viewof knownexistenceresults,e.g.[3,4
(Chap.12),
6 or8],
the hypothesis(15)
isreasonable.DEFINITION 3.1 A
function
OC(OT)
NC2’1(QT)
isalowersolutionof
(13) if
Ota <_ a(x)Aa +f (x,
t,a)
(x, t)o, + (x, t)o <_ p(x,
a<Oa(x, O) < a(x, T)
in f x
(0, T],
on
O9.Ft
x(0, T],
on
01f’t
x(0, T],
intl.
An upper
solution/3 C(Or)
C’ (Qr) of(13)
isdefined
in asimilarwaybyreversingall the above inequalities.
WealsoconsidertheCauchyproblem
Otu a(x)Au + f (x,
t,u) or(x, t)Otu -+- c(x, t)Ouu p(x, t)u
u=Ou(x, O) Uo(X)
in f x
(0, r],
on
Ofl
x(0, r],
on
01 (0, 7-],
in f.
(16)
DEFINITION3.2
(16)/f
A
function
a6(07)
62’1(Qr)
isalower solutionof Ota <_ a(x)Aa + f (x,
t,a)
,,(x, t)o,, + (x, t)o, <_ p(x,
a<O,(x, o) < uo(x)
ina
(o, -],
on02[2x
(0,
on
O1FI
x(0,
intl.
An
uppersolution/3
EC(0)
f3C2’1(Q) of(16)
isdefined
asimilar way byreversingall the above inequalities.PROPOSITION3.3 Assume
that(14)and(15)are satisfiedanduo Co(f).
Letaand be lower and uppersolutions
of (16)
such thata<_
onQ..
Thentheproblem
(16)
hasatleastonesolution uC(O)
Ae
2’1(Q)
with a<_u<_fl onQ.
Proof
Consider the modifiedproblemOtu a(x)Au + f (x,
t,q,(x,
t,u))
t)ot + o(x,
t,u)
u--O
u(x, o) uo(x)
in9t x
(0, -],
on
02’
x(0, T],
on
019t
x(0, -],
in 9t,
(17)
where"y(x,t,
u) a(x, t) + (u a(x, t))
+(u -/3(x, t)) +. By
assumption(15),
we can apply Schauder’s Fixed Point Theorem to prove that(17)
hasatleastonesolutionuC()
NC2’1(Q).
Letusshow thata_<
u onQ.
Similarlyoneshowsthatu_</3onQ.
Setv u aandassumethat
minrv <
0.As
v>_
0onq, there exists(Xo, to) Q
such thatv(x0, to) minv.
Then the strong minimum principle Theorem 2.8 applied locallytoOtv- a(x)Av >_
O,andthe Friedman-Viborni-Theorem 1.c.yieldacontradiction.
Remark 3.4 The assumption a,/3
C(0)
NC2’1(Q)
can be relaxedto: for some 0
to <
tl<...< tn
=’r,Oglfi(ti,ti+l] ][z’X(ti,//+l
EC(e’
X(ti,/i+1])
AC2’1 ((
U02) (ti, ti+l])
and for each x(,
i=l,...,n- 1,
a(x, ti)
lima(x, t) >_
lima(x, t)
t-t;
t--t
and
/(X, ti) lim/3(x, t) < lim/3(x, t).
t-*t;
PROPOSITION 3.5 Assume
(14)
and(15)
aresatisfied
anduo
ECo(f).
Then,thefollowing holds:
(i)
/f
al,a2 are lower solutions and is an upper solutionof (16)
satisfyingal
</3
anda2</3
then there exists asolution uof (16)
satisfying
max{a1, a2} <
u</3;(ii)
/fa
isalowersolutionsand,/32
areuppersolutionsof(16)
satisfying a< 31
anda< 2
then there exists a solution uof (16)
satisfyinga
_<
u< min{/31,/32}.
Proof
Theproofof(i)is exactlythesameasinProposition3.3 with amax{al, O2)
ifweobserve thatV(Xo, to)
is eitherU(Xo, to) Ol(xo, to)
or
u(xo, to) az(Xo, to).
Part (ii)issimilar.PROPOSITION3.6 Under theassumptions
of
Proposition 3.3,theproblem(16)
has aninfimal
and asupremalsolution Uinfand Usup in [,/3 i.e.Uinf,Usup E
C(0-r)
f"lC2’1(Q)
are solutionsof (16)
witha<_
Uinf Usup/
and every solution uC(O)
NC 2’1(Q) of (16)
such that a<
u<
satisfies
Uinf u Usup.
Proof Let (16)}
and(16) }.
Define{#’Q :
a<_
#<_/3,/z
is alowersolution ofH {u" Q :
a<_
u</3,uis anupper solution ofUinf(X, l) inf{u(x, t)"
uUup(X, t) sup{#(x, t)"
#Wewill show thatUinfis an infimal solution. Theproofthat Usupis a supremalsolutionis similar.
Let
{(Xu, t)}=
be a dense subset ofQr
and forN= 1,2,..., let{uV,m}m=
beasequence ofuppersolutionssuch thatlmim
tlN,m(XN, IN) Uinf(XN, tN).
Let
I(X, t) =/l,l(X, t).
Itfollows from Proposition 3.3 that thereexists asolutionUlof(16)
such thata_< u
_</31.Let/2
be definedby/32(t) min(ul (x, t), ul,2(x, t), u,2(x, t)}
then, by Proposition 3.5, there exists a solution u2 of
(16)
such that a<
u2</32.
Letusdefine inductively/i+1 (Y, t) min(ui(x, t),/1,i+1 (y, t),...,/,/i+1,i+1 (y, t)),
then there exists a solution Ui+ of
(16)
such that a<
Ui+ (__i+l"
Hence,
wehaveasequence{ui}i=l
of solutions of(16)
such thatBy
assumption(15)
and monotonicity,wededuce that the sequence{Ui}
converges in
C(Or)
to asolutionuof(16).
Furthermore,it isclearthat, for everyN 1,...,lim
Ui(XN, tN) Uinf(XN, tN).
Hence U(XN, tu)
Uinf(XN,tu)
for allNc
1.As {(XN,
tu)}NiSdenseinQ,
itfollows thatu Uinfon
0r.
In fact, assumebycontradiction that for some(2, -) c Q, u(.2, ) > Uinr(, ). By
definition ofUinf,wecan findu b/so that
Uinf(2, ) _< u(, ) < u(, -)
and for(x, t)
near enough(,-{), u(x, t) < u(x, t).
This is acontradictionif wechoose(x, t)
as an element of theset{ (xv, tu))=l.
Thisconcludes theproofifweobserve that everysolutionuwithc<
u</3
satisfiesu b/andhenceu>
uinf.Notethatseveral authorspreferthe terminologymaximalandmini- malsolutionforsupremaland
infimal
solution.But,
in orderto avoidconfusion withmaximalityinthesenseof existence,wepreferthenotion adoptedhere.
THEOREM 3.7
Assume
that(14)
and(15)
aresatisfied.
Let and3
belower and upper solutions
of (13)
such that c< fl
onOT
andc(., 0),
/3(., O) Co(f).
Then,thefollowing hotds:(i) there exist Uinfand Usup
infimal
andsupremalsolutionsof (13)
in[a, i.e.UinfandUsup are solutions
of (13)
in[,
such that every solutionuof(13)
with<
u</3 satisfies
Uinr u
<
Usup;AQUALITATIVE THEORY FOR PARABOLICPROBLEMS 479
(ii)
there exist 5 andt3
solutionsof (16)
with"
ec and respectivelyUo(.) a(., 0), Uo(’) =/3(., O)
such thatO 5 Uinf Usup
_ 3 /
and
lim
115(., t) Uinf(’, t)[Ic0()
0,t---oc
lim
[[/3(., t) Usup(’, t)[lc0(fi
0;(iii) everysolutionu
of(16)
suchthata<
u</3
onQo satisfies
5<
u<
Oil
Ooco
Remark 3.8
By
(iii),5and/
aretheinfimaland thesupremalsolutions of(16)
in[a,/3
],respectivelywithu0(.) a(., 0)
andu0(.) =/3(., 0).
Proof Let
usprove the result forUinfand 5. The otherpartissimilar.Define asequence
(5n)n
of functionsasfollows. Take as50
theinfimalsolution of
Ot5o a(x)ASo + f (x,
t,50) t)o,C o + t)o C o p(x, t)C o
5o
0C o(x, o) o)
in f x
(0, T],
on
02
x(0, T],
on
01
f(0, T],
in f,
(18)
satisfyinga
< 5o </3.
Such an5o
exists byProposition 3.6 as a and/3
are lower and upper solutions of(18). Moreover 50
satisfies50(., T) _> a(., T) _> a(., 0)--50(., 0).
Then, we define recursively(Sn),
bytaking, forn>_
1,as5n
theinfimalsolutionofOt5n a(x)A5n +f (x,
t,5.)
r(x, t)Ot5n + c(x, t)Ou5n p(x, t)5n
5n
0an(X, o)
in9t x
(0, T],
on
02f
x(0, T],
on
01
fl x(0, T],
inf,
(19)
satisfying 5,-1
<
5,</3.
Again suchan 5, exists by Proposition 3.6 as5n-1
and/3
are lower and upper solution of(19)
withMoreover,
5n satisfies5,(., T)> 5n-1(’, T)= 5n(’,O).
Accordingly,wehave definedasequence
(&n)n
of lowersolutionsof(13)
suchthat,foreachn
>
1,a
_<
,-1_< ,
_</3(20)
and
n(X, O) n-1 (X, T)
in f.(21)
By
monotonicity,(n)n
convergespointwise inQr
tosomefunction usatisfyinga
<
u</3. Moreover,
byassumption(15)
and monotonicity, we deduce that the sequence(cn)n
convergesinC(Qr)
to a solution uof(13).
Now,
define a function’Q I
as follows. If, forsome n E11,(x,t) O
x[nT,(n+ 1)T),we
setC (x, t) ,,r).
It is easy to see that & is continuous,
c(x, t)=
0 on01’-
X(0, C), (., 0) a(., 0)
and, for each n 1t,&[[nr,(,,+l)rl C.(Ft
[nT, (n + 1)T])
f3e2’l((Ft
tO02f) (nT, (n + 1)T]). By
the periodicity of the coefficients,&
also satisfies, for eachn 1, the equationsOt6,
a(x)A& + f (x,
t,t)OtC + c(x, t)O C p(x,
in f
(nT, (n + 1)T],
on
02f
x(nT, (n +-1)T].
We prove that, for each
nN +,
tEC2’I(("I,.J02’)X (0,nT])
andtherefore
c
isasolution ofOt6 a(x)A6 + f (x,
t,60
or(x, t)Ot + c(x, t)Ou& p(x, t)
in 9tx
(0, o),
on
02’2
x(0,
c
0 on01ft
x(0, o),
6(x, O) a(x, 0)
in f.Letusshow thatt
C2’1 ((-
I-J02")
X(0,2T]);
then thegeneralcon- clusion follows by induction.By (15),
let wC(O2r)NcE’I(QEr)
bethe unique solution of the linear initial valueproblem
O,u a(x)Au f (x,
t,(, t)Ou + (. t)O.u p(.
u=0
u(. 0) (.0)
in flx
(0,2T],
on
02f
x(0,2T],
on
01fl (0,2T],
in f.
Sinceboth
(IQ(0,T]
andWlfix(0,T]
aresolutions of thelinearinitialvalue problemOtU a(x)Au f (x,
t,(x, t)Otu + (x, t)Ou p(x, t)c
u--0
u(x. o) (x. o)
in f
(0, T],
on
02f
x(0, T],
on
c91f (0, T],
in f,
byuniqueness,weget
c
winflx[0, T].
Further,both[fx(T,2T)
andWlfi(r,2r)
aresolutionsof thelinearinitial valueproblemOtu a(x)Au f (x,
t,&)
or(x, t)Otu + c(x, t)cg,u p(x, t)
u=0
u(x, T) 6(x, T) w(x, T)
in f
(T, 2T],
on
02fl (T, 2T],
on
019t
x(T,2T],
in ft.
Then,by uniqueness,weget
&
win ftIT, 2T].
Therefore, wecon-clude that
&
win[0, 2T],
sothat&
EC(Q2T)
AC
2’1(Q2T).
Moreover,
by periodicityand construction,wehave limI1(’, t) u(., t)i[c0(fi)
0.To
completetheproofof(i) and (ii), itremains toprove thatevery solutionvof(13)
such thata<
v<
satisfies v>
(n for everyn.Thisis clear as, ifvis such a solution, vis anupper solution of(18)
and by Proposition 3.3, thereis asolution 0 of(18)
witha<
a0<
v</3. As 0
is the infimal solution of(18)
in[a,/3],
we have a< c0 _<
0<
v.Recursively, if
&n- <
v, then v is anupper solutionof(19)
and hence&,_ _< &, <_
vwhichconcludestheproofof(i)and(ii).To prove that every solution v of
(16)
such that a_<
v</3
onQ
satisfiesc _<
v onQo,
weproceed againrecursively, observing first that vis anuppersolution of(18)
and as0
is the infimal solution of(18)
in[a,/3],
a_< 0 <_
v onOT.
Moreoverv(x, + T)
isalsoanuppersolution of
(18).
Hencev(x,t) > &o(x,t- T)
on f x[T,2T].
Recur-sively,
ifv(x,t) > &(x, t)onf
x[O, nT]andv(x,t) > &n-l(X,t-nT)on
f x
[nT, (n + 1)T],
thenv(x, + nT)
is an uppersolution of(19)
andv(x, +
nT) > &n-1 (X, t). As
tnisthe infimal solution of(19)
in[&n-1,/3],
wehave
&n-1 < n (’," + nT)
onQr
i.e.& <
v onQ(n+l)r
andagain,as above,
v(x,t) > &n(x,t- (n + 1)T)
on ) x[(n + 1)T, (n + 2)T].
Now,
aninductionargument shows the assertion.We note in passing that the results of this section extend those obtainedin
[7]
forhomogeneousDirichletboundaryconditions.DAMPING
EFFECT OF
THEDISSIPATIVE
DYNAMICALBOUNDARY CONDITION
The comparison techniques ofSections2 and 3 enable the comparison of solutions under different boundary condition, especially for the Neumann boundaryconditionand
(2).
Letusdiscussthisin amodel case givenbyagloballyattractiveequilibrium.Though theglobalattractor turns out tobeindependentof the conditiononOf x(0, o)
withOf2
Of, the convergence rate decreases with respect to the coefficient The reactionterm issupposedtoadmit twoequilibriaA
<
Bandtobe ofthe form
f(x, t,A) =f(x, t,B)
0 for all(x, t)
f x(0, o), f(x,
t,u) >
0 ifu(A, B), f(x,
t,u) <
0 ifu(B, o). (22)
Ifa
andfdo
notdependon xandt, we can statethe following:THeOReM 4.1 Suppose that a
(0, ), cre-( ., t) L(f) for
all[0, cxz)
andf C([A, o)) fulfills (6)
and(22).
Let uC(Qo)
C’1(Qo)
beasolutionof
Otu aAu + f (u)
t)O,u + t)O u o
u(x, o) >_ A,
on Of x
(0,
cinf,.
(23)
Then
lim
Ilu(., t) nllc0<>
0.t---cx
Proof
Observe first that u>
A for>
0 by Theorem 2.2 and, there- fore, bythe strong minimumprincipleTheorem 2.8 applied locally toOtu aAu >
O,u>
Afor>
0.As
for theNeumann
condition, theenergyE(u)
can serve as aLyapunov
functionalThen
E(u) > E(B) -I1 ]ff f (s)
ds andE(u) E(B)
iffu=B. Green’sformulayields,
d
E(u) f {
a(t Zu Otuf (u) )
dxf(Otu)2
dx-k-o Otua Ou
dsf(Otu)2
dxfocrc-l a(Otu)2
dsThus,
E(u)
is aLyapunov
functional in the considered function class and theLaSalle invariance principle(see
e.g.1])
yields theassertion.Note that we did not use any consistency condition between the diffusioncoefficienta andthe boundaryconductivity
c(x, t)
as it was necessary in the systems considered in[6,
Section5].
For different dynamicalboundarytermsletuscompare thesolutionsinthe following simplecase.THEOREM 4.2 Supposethat
f,
a,cand0<
or1< cr2fulfill
thehypothesesof
Theorem 4.1.Letul,u2EC(Oo; [A, B])
fqC2’1 (Q)
besolutionsof
theIBVP
(23)
withr crl andcr or2,respectively andu2(., 0) < Ul(., 0)
inf.Thenu2
<
Ulinf(0, cxz).
Proof
For13(u; or):= crOtu + cO,
u the differenttime lateralboundary conditionsyield(Ul; 0"2) (0"2 O’l)IOtUl
Thus, if
Otul
>0, the comparison principle Theorem 2.2 permits to conclude.Withthesameargumentasintheproofof Theorem4.1,weseethat
Ul
> A
for>
0. Ifcgtul(, ) <
0atsomepoint(, ’)
EQo,
then uattains alocalminimum m<
Bsomewhere int [’, )
dueto Theorem 4.1.Thehypotheses andthe differential equation imply thatm
A,
which isimpossible.Thus,infact,Otul >
0inQo.
Clearly, thecorresponding resultu2
>
Ul holds forsolutions taking values in[B, o).
Letusillustratethe damping effect of the dissipative dynamicalcon- ditionbymeansof the following simpleexamplethathas beencomputed withtheaidof the convergentfinite differencemethod similartotheone in
[6, (78-79)].
Considerthe equationOtu--Au+u(1-u) in--(O, 1)(O, 1) C2
under the condition
O’OtU + OuU
0 onr {0)
x(0, 1)
and the Neumann condition on the remaining parts of Of. Figure displays thesolutions onthecross-sectiony 0.5 forvalues oftrvarying in[0,
2].
Notethatcris notcontinuouson0f. Though,for allcr>
0,the solutionswithu(., 0) >
0, 0tendtothe equilibrium 1, thedampingof theconvergencerateincreases with increasinga.Under higher regularity assumptions, the attractivity result holds also in thefollowingnonautonomouscase.
THEOREM4.3 Supposethata
C(f, (0, cx))
and thatf,
candcrsatisfy(6), (14), (15)
and(22).
LetutT(0)
NC2’1(Q)
beasolutionof
Otu a(x)Au + f (x,
t,u) t)O,u + c(x, t)O u o
u(x, o) >_ A,
AonOf
(0,
cxzinf,.
(24)
Then
lim
Ilu(,, t) BIIco/ /= o,
0.4 0.35 0.3 0.25
1
0 0.5
1.5
2 0
FIGURE The damping effect of the dynamicalcondition.
Proof
Again, by Theorem 2.2, u_>
A for>
0. Then by the strongmaximumand minimumprincipleTheorem 2.8, there exist d
>
B>
c>
A suchthatd>_ u(x, t) >
conf x[T, o).
Moreovercand darelower and uppersolutionsofOtu a(x)Au +f (x,
t,u)
t)Otu + o(x, t)O u o
infix
(0, ),
on 69f
(0, cx), (25)
u(x, O) u(x, T)
onIfweprove thatBistheonlysolution of
(25)
in[c,d],wecanconclude byTheorem 3.7.IfBisnotthe onlysolution wehave aninfimalsolutionUinfof
(25)
withc<
Uinf BasBis asolution.Hence,
onf(0, T],
OtUinf a(x)Auinf.
By
the minimumprincipleandby periodicity minUinf min Uinf min Uinf.O-
hx{o} x{r}Again, by application of the minimum principle,weconclude thatUinf
is constant in
[0, T],
henceUinf--B.
In the same way, we prove thatUsup B.With the same argument as for Theorem 4.2 we obtain another monotonicity result concerning different dynamical conditions for theproblem
(24).
COROLLARY 4.4 Under thehypotheses
of
Theorem 4.3, two solutionsu,
u
belongingtoC(O; [A, B])
fqt7:z’(Q) of
theIBVP(24)
withor--O"and or=or2, respectively andsatisfying
u(., O)<_ u(., O)
on ffulfill
thelatter inequality
of Q.
References
[1] H. Amann,OrdinaryDifferent&lEquations,Walter deGruyter,Berlin1990.
[2]J.vonBelow, AMaximumPrincipleforSemilinear ParabolicNetworkEquations, in,J.A. Goldstein, F. Kappel andW. Schappacher (Eds.), Differential equations withapplicationsinbiology, physics,andengineering,Lect. Not. PureAppl.Math.
Vol. 133,M.DekkerInc., NewYork, 1991, pp.37-45.
[3]J.vonBelow,AnExistenceResultforSemilinear ParabolicNetworkEquations with DynamicalNode Conditions,in, C.Bandle,J.Bemelmans,M.Chipot,M.Griiterand J.SaintJeanPaulin(Eds.), Progressinpartialdifferentialequations: ellipticand parabolicproblems,PitmanResearchNotesinMath.Ser.Vol. 266.LongmanHarlow Essex,1992, pp. 274-283.
[4]J.vonBelow,ParabolicNetworkEquations,3rd edn.(toappear).
[5]J.vonBelow,A MaximumPrincipleforFullyNonlinearParabolic Equations with TimeDegeneracy,CahiersLMPALiouvilleULCONo.95(1999) (toappear).
[6]J. vonBelow and S. Nicaise, Dynamical InterfaceTransition in Ramified Media withDiffusion, Comm.PartialDifferentialEquations,21(1996),255-279.
[7]C.De CosterandP.Omari, UnstablePeriodic SolutionsforaParabolicProblemin PresenceofNon-well-orderedLowerandUpperSolutions, CahiersLMPALiouville ULCONo.101(1999) (submitted).
[8] T. Hintermann,EvolutionEquationswith Dynamic Boundary Conditions, Proc.
Royal Soc. Edinburgh, l13A(1989),43-60.
[9] O.A.Ladyzhenskaya,V.A.SolonnikovandN.N.Uraltseva,Linearand Quasilinear EquationsofParabolicType, Trans.Math. Monographs 23,A.M.S.,Providence,R.I.
(1968).
[10] M.H. ProtterandH.F.Weinberger,MaximumPrinciplesinDifferentialEquations, Prentice-Hall, Englewood Cliffs, 1967.
[11] W.Walter,Differentialand IntegralInequalities,Springer Verlag, Berlin,1970.