HAL Id: hal-00004903
https://hal.archives-ouvertes.fr/hal-00004903
Submitted on 11 May 2005
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Séverine Bernard, Silvère Paul Nuiro
To cite this version:
Séverine Bernard, Silvère Paul Nuiro. On generalized Sobolev algebras and their applications.. Topo- logical Methods in Nonlinear Analysis, Juliusz Schauder University Centre for Nonlinear Studies, 2005, 25 (2), pp.375-390. �hal-00004903�
ccsd-00004903, version 1 - 11 May 2005
Séverine Bernard and Silvère P. Nuiro
In thelasttwodeades, manyalgebras ofgeneralizedfun-
tionshave been onstruted,partiularly theso-alled gener-
alized Sobolev algebras. Our goal is to study the latter and
some of their main properties. In this framework, we pose
and solvea nonlinear degeneratedDirihlet problem with ir-
regular datasuh as Dirageneralized funtions.
Key words : nonlinear degenerate Dirihlet problem, generalized solu-
tion, Sobolev algebra,non positive solution.
2000 MSC:35J70, 46F30,46E35, 35D05, 35B50.
1. Introdution
A theoretial study of most of thewell-known algebras of generalized fun-
tions haspointed out two fundamental strutures. Therstone is thealge-
brai struture of a solidfator ring C of generalized numbers. The seond one is the topologial struture dened by a family P of seminorms, on
a loally onvex linear spae E, whih is also an algebra. These algebras
have been denoted by A(C, E,P) and one speaks of (C, E,P)-algebras of
generalized objets. The denition overs most of the well-known algebras
of generalized funtions, as for example, the Colombeau simplied algebra
[3℄, Goursat algebras[13℄ and asymptoti algebras [4℄. Ontheother hand,
speial hoies for E, P and C also allow the introdution of some new al- gebras. One of them is the so-alled Egorov extended algebra, beause of
thesimilaritywiththeEgorov [5℄algebra ofgeneralized funtions. We have
beeninterestedinworkingwithinthe frameworkoftheso-alledgeneralized
SobolevalgebrasbasedonthelassialSobolevspaes. AsE isadierential
algebra, the main interest of these algebras is to give a framework whih is
well suitable to solve many non linear dierential problems with irregular
data. Themethodisbasedontheextensionofamappingfrom(E1,P1)into (E2,P2) to a mapping from A(C1, E1,P1) into A(C2, E2,P2). This method
hasbeenintrodued,intheframeworkofasymptotialgebras,byA.Delroix
and D.Sarpalezos [4℄,and used,intheframework of (C, E,P)-algebras, to
solveanonlinearDirihletproblem[12℄andanonlinearNeumannproblem
[11℄,bothwithirregulardatabyJ.-A.Martiand S.P.Nuiro. Inthispaper,
our goal is to lift upthe generalized Sobolev algebras, bygiving more lear
denitions of all the statements and general results in this framework, in
order to work more easily with these algebras. We introdue the rst ex-
ample of ordered generalized Sobolev algebras, whihallows us to pose and
eventually solvean obstaleproblemwithirregular data. We alsopoint out
some suient properties for the existene of an embedding of some spae
into ageneralized Sobolev algebra. Intheframework ofgeneralized Sobolev
algebra,weareableto solveanonlineardegeneratedDirihletproblem[12℄
withweakerassumptions.
Consider Ω an open bounded domain of IRd (d ∈ IN∗) with a lipshitz
ontinuousboundary ∂Ω,we an state thisformalproblem : (P)
−∆Φ (u) +u=f in Ω, u=g on ∂Ω,
wheref andgarenonsmoothfuntionsdenedonΩand∂Ωrespetively,Φ
aninreasing real-valueddierentiablefuntion denedon IRsothatΦ′ isa
ontinuousboundedfuntionthatanvanishonanitesetofdisretepoints
of IR. This is a quasilinear diusion type problem, withnon homogeneous Dirihletonditionontheboundary. Oneanremarkthattheformalseond
order dierential operator L = −div (Φ′(.)∇x) +Id is a degenerated one, beauseΦ′anvanish. Thus,(P)isaDirihletnonlinearelliptidegenerated problem. Inorder to solve this problem, we introdue an auxiliary problem
byusingan artiialvisosityregularization dependingon aparameter ε.
2. Speial types of generalized algebras
2.1. Denitions.Letus,rst,statethatIKisIRorCl,and1I = (1Iε)εwhere 1Iε= 1 for all ε.Thegeneralized algebras onstrutedfromE,a normedIK-
algebra,are partiular aseof (C, E,P)-algebras [10℄, [12℄,[11℄, [13℄.
Consider a subring A of the ring IK]0,1] so that 1I ∈ A, and whih, as a
ring, issolid(with ompatible lattie struture) inthefollowing sense:
Denition 1. A issaid tobe solid iffrom(sε)ε∈Aand|tε| ≤ |sε|foreah ε∈]0,1] it follows that (tε)ε∈A.
We also onsideran idealIAof A whih is solidaswell,and sothat
(1) ∀(rε)ε ∈IA, lim
ε→0rε= 0.
Then, we introdue the fator ring C = A/IA, whih is alled a ring of
generalized numbers.
Denition2. LetE beanormedalgebra. WeshallallN-generalizedalgebra all fator algebra
A(C, E) =HA(E)/IIA(E),
where
HA(E) ={(uε)ε∈E]0,1] / (kuεkE)ε ∈A+}
and
IIA(E) ={(uε)ε∈E]0,1] / (kuεkE)ε∈IA+},
when k · kE is the norm on E, A+ = {(rε)ε ∈ A / ∀ε > 0, rε ∈ IR+} and IA+ = {(rε)ε ∈ IA / ∀ε > 0, rε ∈ IR+}. Its ring of generalized numbers is dened asthe ring
HA(IK)/IIA(IK) =C=A/IA.
Remark 1. We remark that the notation is A(C, E) instead of A(C, E,P)
sine the family P isredued toone single element. The algebra A(C, E) is
also a vetor spae on the eld IK.
Example 1. With
IA=n
r = (rε)ε∈IR]0,1]| ∀k∈IN∗, |rε|=O εko
and
A=n
r= (rε)ε∈IR]0,1]| ∃k∈ZZ, |rε|=O εko
,
we obtain a polynomialgrowthtype N-generalized algebra.
Example 2. We take
IA=n
r= (rε)ε∈IR]0,1]| ∃ε0 ∈]0,1], ∀ε∈]0, ε0], rε= 0o ,
and A = IR]0,1]. With suh A and IA, we obtain another N-generalized algebra.
Example 3. When E is a Sobolev algebra (that is, for example, on the
form Wm+1,p(Ω)∩Wm,∞(Ω), withm ∈]0,+∞[, p∈[1,+∞[andΩ an open
subset of IRd (d∈IN∗)), respetively a Banah algebra, we will speak about generalized Sobolev algebra, respetively generalized Banah algebra, instead
of N-generalized algebra.
2.2. Embeddings and weak equalities.In the following paragraph, we
aregoingto show awayto embed E into A(C, E).
Proposition 1. The mapping i0 dened on E, by :
∀u∈E, i0(u) =cl(u1Iε)ε,
islinear and one-to-one fromE intoA(C, E).
Proof.For every u∈E,we have : (ku1IεkE)ε=kukE1I. Furthermore, as
kukE ∈IK and 1I∈A,there existsλ∈IN sothat
∀ε, kuεkE ≤λ1Iε,
andobviouslyλ1I∈A+. Asaonsequeneofthesolidpropertywhihimplies
that (uε)ε ∈ HA(E),we have i0(u)∈ A(C, E). It an easily be proved that i0 islinear andone-to-one.
Denition 3. The mapping i0 from E into A(C, E), dened in proposition 1, will be the so-alled trivial embedding of E into A(C, E).
WeanalsoembedsometopologialvetorspaeintoA(C, E). Let(G,T)
be a Hausdor topologial vetor spae so that there exists a ontinuous
linearmapping j from (E,k.kE) into (G,T).
Denition 4. T ∈G andU =cl(uε)ε∈ A(C, E) are (G,T)-assoiated if j(uε)→T in (G,T) as ε→0.
It will be denoted by U G,T∼ T.
Remark 2. Thisdenition does not dependon the hosen representative of
U. Indeed, let (eε)ε ∈ IIA(E). Therefore, lim
ε→0keεkE = 0, whih means that eε →0 in (E,k.kE) as ε→0. Consequently, we have j(eε) →0 in (G,T)
as ε→0.
Denition 5. Assume that U =cl(uε)ε, V =cl(vε)ε ∈ A(C, E). We shall
say that U andV are (G,T)-weaklyequals if (U −V)G,T∼ 0.
It will be denoted by U G,T≃ V.
Proposition 2. Assumethat for every T ∈G,there exists(uε)ε∈ HA(E),
so that
j(uε)→T in (G,T), as ε→0.
Then,thereexists,atleast,anembeddingiGfrom(G,T)intotheN-generalized algebraA(C, E). Furthermore, if,for allv∈E,there exists(uε)ε∈ HA(E),
so that (uε−v)ε∈ IIA(E),then
(2) ∀u∈E, (iG◦j) (u)G,T≃ i0(u).
Proof.For every T ∈G, thereexists(uε)ε∈ HA(E),sothat j(uε)→T in (G,T) asε→0.
Let us state iG(T) = cl(uε)ε. The mapping iG from G into A(C, E) is
obviouslylinear. Letus provethatiG isone-to-one. IfiG(T) = 0inA(C, E)
then
iG(T) =cl(eε)ε for (eε)ε∈ IIA(E).
Wehave eε→0 in(E,k.kE)whihimpliesthatj(eε)→0in (G,T),when-
ever ε→0. This leads to T = 0inG, beause (G,T) isa Hausdorspae.
Theseond propertyis obvious.
Remark 3. If there exists anothersuh embedding i′G from(G,T) into the
N-generalized algebra A(C, E) then
∀T ∈G, iG(T)G,T≃ i′G(T).
Example 4. Let j be the anonial embedding of
L∞(Ω),k.kL∞(Ω)
in
H−2(Ω), σ H−2(Ω), H02(Ω)
, where σ H−2(Ω), H02(Ω)
denotes the weak
topology on H−2(Ω). We will say that T ∈ H−2(Ω) and U = cl(uε)ε ∈ A(C, L∞(Ω)) are H−2(Ω)-assoiated if
j(uε)→T in H−2(Ω), σ H−2(Ω), H02(Ω)
, as ε→0,
and we will denote U ∼2 T. Moreover, we will say that U,V ∈ A(C, L∞(Ω))
are H−2(Ω)-weakly equals ifU − V ∼2 0 and we will denote U ≃ V2 .
2.3. Mappingon N-generalizedalgebra.Theideaofextensionofmap-
ping has been introdued by A. Delroix and D. Sarpalezos [4℄, in the
framework of asymptoti algebras. But it is, in fat, a partiular ase of
denitionof mapping onA(C, E)-algebras.
If θ = (θε)ε is a family of mappings from a normed algebra (E,k.kE)
into a normed algebra (F,k.kF), one an view θ asa mapping from theN-
generalized algebra A(C, E) into the N-generalized algebra A(D, F),where
we have set C=A/IA and D=B/IB when A, IA, B and IB areasin 2.1.
One remarksthatthe extension theoremof A.Delroix and D. Sarpalezos
[4℄ deals withthe asewhere θ= (θ)ε.
Theorem 1. Let E and F be two normed algebras and (θε)ε a family of
appliationsof E in F. We assumethat
1) A⊂B andIA⊂IB,
2) there existsafamilyof polynomialfuntions(Ψε)ε ofonevariable with
oeients in A+ so that
∀ε >0 , ∀x∈E , kθε(x)kF ≤Ψε(kxkE),
3) there exists two families of polynomial funtions (Ψ1ε)ε and (Ψ2ε)ε of
one variable with oeients in A+ so that Ψ2ε(0) = 0 for all ε > 0,
and
∀ε >0 , ∀x, ξ∈E , kθε(x+ξ)−θε(x)kF ≤Ψ1ε(kxkE)Ψ2ε(kξkE).
Then there exists an appliation Θ : A(C, E) →A(D, F), whih assoiates cl(θε(xε))ε with cl(xε)ε.
Proof.First, let (xε)ε be in HA(E) and let us show that (θε(xε))ε is in
HB(F). We have (kxεkE)εinA+ so(Ψε(kxεkE))ε isalso inA+,sine(Ψε)ε
has oeients in A+. Thus (kθε(xε)kF)ε) belongs to A+ ⊂ B+, due to
(1) and (2), whih implieswhat we want. Then, let (iε)ε be inIIA(E) and
let us show that (θε(xε+iε)−θε(xε))ε is in IIB(F). Sine (kxεkE)ε and
(kiεkE)ε arerespetivelyinA+and IA+then(Ψ1ε(kxεkE))ε and(Ψ2ε(kiεkE))ε
arerespetivelyinA+ andIA+ ,sine,for i∈ {1,2},(Ψiε)ε hasoeientsin
A+. Then, (Ψ1ε(kxεkE)Ψ2ε(kiεkE))ε isinIA+. Thus(kθε(xε+iε)−θε(xε)kF)ε
belongsto IA+⊂IB+,due to (1)and (3), whih impliestherequired result.
Asa onsequene,we obtain thefollowing result.
Proposition 3. Assume that A ⊂ B and IA ⊂ IB. If (θε)ε is a family of
ontinuous linear mappings from a normed algebra E into a normed algebra F, then (θε)ε also denes a mapping Θfrom A(C, E) into A(D, F).
Example 5. Let Ω be an open subset of IRd and E = H1(Ω)∩L∞(Ω)
with kukE = kukL∞(Ω) +kukH1(Ω). The anonial embedding i : u 7→ u
is ontinuous as well as linear from the Banah algebra E into the Banah
algebra L∞(Ω). Obviously, themapping iveriesallthe assumptions of the previous proposition;this iswhy we an dene itsextension I asa mapping
fromA(C, E) intoA(C, L∞(Ω)).
Inthe same way,one an prove that:
Proposition 4. Assume that (θε)ε is a family of mappings from a normed
algebra E intothe topologial eld (IK,|.|),so that
• there existsafamilyof polynomialfuntions(Ψε)ε ofonevariable with
oeients in A+ so that
∀ε >0 , ∀x∈E , |θε(x)| ≤Ψε(kxkE),
• there exists two families of polynomial funtions (Ψ1ε)ε and (Ψ2ε)ε of
one variable with oeients in A+ so that Ψ2ε(0) = 0 for all ε > 0,
and
∀ε >0 , ∀x, ξ∈E , |θε(x+ξ)−θε(x)| ≤Ψ1ε(kxkE)Ψ2ε(kξkE).
ThenthereexistsanappliationΘ: A(C, E)→ C,whihassoiatescl(θε(xε))ε
withcl(xε)ε.
Remark 4. If θ is a ontinuous linear mapping from a normed algebra (E,k.kE) into the topologial eld (IK,|.|), then θ also denes a mapping,
denoted by Θ, fromA(C, E) into the fatorring C=A/IA.
2.4. An example of ordered generalized Sobolev algebra.Consider
A and IA as in 2.1, the Sobolev algebra L∞(Ω), endowed with its usual
topology,withΩanopenbounded subsetof IRd. Thus, we anonsiderthe
algebra A(C, L∞(Ω)). It is easy to prove, bymeans of theorem 1, thatthe
mapping
p:L∞(Ω) → L∞(Ω)
u 7→ u+= sup{u,0}= 12(u+|u|)
an be extendedasa mappingP fromA(C, L∞(Ω))into itself,dened by:
∀U =cl(uε)ε ∈ A(C, L∞(Ω)), P(U) =cl(p(uε))ε,
due to the following relation :
∀r, s∈IR,
(r+s)+−r+ ≤ |s|.
We arenowableto state thefollowing result:
Proposition 5. The generalized Sobolev algebra A(C, L∞(Ω)) is partially
ordered by the following binaryrelation :
∀U, V ∈ A(C, L∞(Ω)), U ≤V ⇐⇒ P(U −V) = 0.
Proof.Obviously, the relation ≤is reexive, then we have to prove, for U, V, W ∈ A(C, L∞(Ω)),that :
U ≤V andV ≤U ⇒ U =V;
(3)
U ≤V and V ≤W ⇒ U ≤W.
(4)
We state U =cl(uε)ε,V =cl(vε)ε and W =cl(wε)ε.
Proof of property (3) : If U ≤ V and V ≤ U then, there exists (ϕε)ε and (ψε)ε inIIA(L∞(Ω))sothat(uε−vε)+ =ϕε and(vε−uε)+ =ψε. As,
uε−vε = (uε−vε)+−(vε−uε)+=ϕε−ψε,
itfollows that(uε−vε)ε= (ϕε−ψε)ε∈ IIA(L∞(Ω)),whene U =V.
Proof of property(4 ): IfU ≤V and V ≤W thenwe have
(uε−vε)+ L∞(Ω)
ε∈IA ,
(vε−wε)+ L∞(Ω)
ε ∈IA.
Bymeans of thesolidproperty, we dedue,from thefollowing inequality:
(uε−wε)+
L∞(Ω) ≤
(uε−vε)+
L∞(Ω)+
(vε−wε)+ L∞(Ω),
that (uε−wε)+
ε ∈ IIA(L∞(Ω)),whih yields P(U −W) = 0,that is to
sayU ≤W.
Proposition 6. For all u, v ∈ L∞(Ω), we have i0(u) ≤ i0(v) if, and only
if, u≤v in L∞(Ω), that is u≤v almost everywhere in Ω.
Proof.If i0(u) ≤ i0(v) then P(i0(u)−i0(v)) = 0. Consequently, there exists(ϕε)ε,(eε)ε∈ IIA(L∞(Ω))sothat(u−v+eε)+ =ϕε,sine we have
uε−vε=u−v+eε for allε. Taking into aount that ϕε→0 and eε →0 in L∞(Ω) ,as ε→0,
itmaybeseenthat(u−v+eε)+=ϕε →0 a.e. in Ω,whene(u−v)+= 0 a.e. inΩ,sine one aneasily prove that
(u−v+eε)+→(u−v)+ in L∞(Ω) , as ε→0.
It meansthat u≤v a.e. inΩ.
Conversely, ifu ≤v a.e. inΩ then(u−v)+ = 0 a.e. inΩ. By denition
of P andi0,this leads to i0(u)≤i0(v).
Proposition 7. Let u∈L∞(Ω) andU ∈ A(C, L∞(Ω)). If U L1∼(Ω)u (here L1(Ω) isendowed withits usual topology) and U ≤0 then u≤0 a.e. in Ω.
Proof.We set U = cl(uε)ε. Sine U L1∼(Ω) u then, as ε goes to 0, uε →u inL1(Ω),whihgivesu+ε →u+ inL1(Ω),bymeansoftheLebesgue
dominated onvergene theorem. Sine U ≤ 0 then P(U) = 0. Conse-
quently, there exists a sequene of funtions (ϕε)ε ∈ IIA(L∞(Ω)) so that u+ε =ϕε forall ε. Taking into aount that
ϕε→0 in L∞(Ω) , asε→0,
we nd that u+ε = ϕε → 0 a.e. in Ω, whene u+ = 0 a.e. in Ω, whih
impliesu≤0 a.e. inΩ.
3. Solution of the nonlinear degenerate Dirihlet problem
After having solved the auxiliary problem by using an artiial visosity
regularization depending on aparameter ε,we solve our main problem(P)
(see setion 1), ina generalized Sobolev algebra withthe lassial equality
and with the weak one dened in example 4. Then we perform a little
qualitative study ofthe solution.
3.1. The regularized Dirihlet problem.Letus set
V+A=
(rε)ε ∈A+ / ∀ε >0, rε∈]0,1]; lim
ε→0rε= 0 ; 1
rε
ε
∈A+
.
Assumethat V+A6=∅ andthen, for all (rε)ε inV+A,set Φε = Φ +rεId. This
setiononsistsinproving thefollowing proposition:
Proposition 8. If f ∈L∞(Ω) andg ∈L∞(∂Ω) then there exists one, and
onlyone, funtion u∈H1(Ω)∩L∞(Ω)solutionof the regularized problem
(Pε)
−∆Φε(u) +u=f in Ω, u=g on ∂Ω.
Proof.Thisproofgoesinthree steps.
1)Maximum's priniple
We aregoing toprove that ifu∈H1(Ω)isa solutionof thisproblem then m≤u≤M a.e. inΩ,
with m = min{inf
Ω f , inf
∂Ωg} and M = max{sup
Ω
f , sup
∂Ω
g}, whih means
thatu belongsto L∞(Ω).
Indeed,for suh a u,we have, for allv inH01(Ω) Z
Ω
∇Φε(u)∇vdx+ Z
Ω
uvdx= Z
Ω
f vdx,
where dxdenotes the Lebesgue measureon Ω. Letus onsider thefuntion v= (Φε(u)−Φε(M))+ thenv isinH01(Ω),so
Z
Ω
(∇(Φε(u)−Φε(M))+)2dx+ Z
Ω
u(Φε(u)−Φε(M))+dx=
Z
Ω
f(Φε(u)−Φε(M))+dx,
sine Φε(M) isa onstant. Consequently,
k(Φε(u)−Φε(M))+k2H1 0(Ω)=
Z
Ω
(f −M)(Φε(u)−Φε(M))+dx
− Z
Ω
(u−M)(Φε(u)−Φε(M))+dx.
By denitionof M,the rstintegral isnegative and, sine thefuntionsId
and Φε areinreasing,the seond oneis non negative. Then
k(Φε(u)−Φε(M))+k2H1
0(Ω)≤0,
thatisΦε(u)≤Φε(M) a.e. inΩ,whihimpliestherstpartoftherequired
result, sine Φε is an inreasing funtion. For the seond part, we use a
similarmethod bytakingv= (Φε(u)−Φε(m))−.
2)Existene ofa solutioninH1(Ω)
This resultis obtained byusing the Shauder's xed point theoremrelated
to aweaklysequentiallyontinuous mapping froma reexiveand separable
Banah spaeinto itself. Letusonsiderw0 ∈H1(Ω)theunique solutionof
thefollowing linearDirihlet problem:
−∆w0 = 0 inΩ, w0 = g on∂Ω.
Then a solution of the regularized problem is of the form w0 +w, with w∈H01(Ω)and for allv inH01(Ω),onehas
Z
Ω
Φ′ε(w0+w)∇w0∇vdx+
Z
Ω
Φ′ε(w0+w)∇w∇vdx+
Z
Ω
(w0+w)vdx= Z
Ω
f vdx.