On Liouville theorem and the regularity of weak solutions to some nonlinear elliptic systems of higher order
L. Balanda, E. Viszus
Abstract. The aim of this paper is to show that Liouville type property is a sufficient and necessary condition for the regularity of weak solutions of nonlinear elliptic systems of the higher order.
Keywords: regularity of weak solutions, nonlinear elliptic systems Classification: 35J60, 35D10
Introduction.
In this paper, we shall deal with nonlinear elliptic systems. More precisely, we shall consider the following problem.
Let Ω be a bounded domain with Lipschitz boundary inRn,n≥2. Let us denote σ(n, k) = n+k−1k
and̺(n, k) = n+kk
,n, k∈N. We shall study the weak solutions u∈[Hk,∞(Ω)]N to the system
(0.1)
X
|α|≤k
(−1)|α|Dα(aiα(x, γ(u))) = X
|α|≤k
(−1)|α|Dαfαi, i= 1, . . . , N; x∈Ω, k≥1,
γ(u) ={Dαui:i= 1, . . . , N; |α| ≤k}.
By a weak solution of (0.1) we mean a functionu∈[Hk(Ω)]N (Hk(Ω)≡Hk,2(Ω)
— Sobolev space,u= (u1, . . . , uN) — see [4]) such that
(0.2)
N
X
i=1
X
|α|≤k
Z
Ω
aiα(x, γ(u))Dαϕidx=
N
X
i=1
X
|α|≤k
fαiDαϕidx, ϕ∈[D(Ω)]N.
The authors are indebted to Professor J. Neˇcas for the very valuable advice concerning the paper
We shall suppose that:
aiα ∈C1(Ω×RN ̺(n,k)), i= 1, . . . , N, |α| ≤k, (0.3)
N
X
i,j=1
X
|α|=|β|=k
∂aiα
∂ηjβ (x, η)ξαi ξβj >0, (0.4)
(x, η)∈(Ω×RN ̺(n,k)), ξ∈RN σ(n,k), ξ6= 0.
fαi ∈H1,pα(Ω), pα= p
k− |α|+ 1, where p > n, (0.5)
p≥2(k+ 1), i= 1, . . . , N, |α| ≤k.
Let us denote forM >0, G >0:
[M] ={u∈[Hk,∞(Ω)]N :u solves (0.1) and kuk[Hk,∞
(Ω)]N ≤M}, [G] ={fαi ∈H1,pα(Ω) :
N
X
i=1
X
|α|≤k
kfαikH1,pα(Ω)≤G}.
We shall use the notations
γ1(u) ={Dαui:i= 1, . . . , N;|α| ≤k−1}, γ2(u) ={Dαui:i= 1, . . . , N;|α|=k},
PmN ={(P1, . . . , PN) :Pi — polynomial with deg(Pi)≤m, i= 1, . . . , N}, m≥0,
B(x0, R) ={x∈Rn:|x−x0|< R}.
Definition 0.6. The condition (L) of the Liouville type is satisfied for the sys- tem (0.1) if for ∀x0 ∈ Ω, ξ ∈ RN ̺(n,k−1), the only weak solutions in Rn to the system
(0.7) X
|α|=k
(−1)|α|Dα(aiα(x0, ξ, γ2(v))) = 0, i= 1, . . . , N
with bounded derivatives ofk-th order are polynomials of at mostk-th degree (i.e.
v∈PkN).
Definition 0.8. We say that the system (0.1) has a property of regularity (R) if for∀x0 ∈Ω, ξ ∈RN ̺(n,k−1), M >0 there exist η >0, c >0 and µ∈(0,1) such that every weak solutionu(inRn) of the system (0.7) with|Dαui| ≤M, |α|=k, i= 1, . . . , N, belongs to the space [Ck,µ(B(0, η))]N and
kuk[Ck,µ(B(0,η))]N ≤c.
It will be proved in this paper that the property (L) implies the interior regularity, i.e. if u is a weak solution to (0.1) then u ∈ [Ck,µ(Ω′)]N, where Ω′ ⊂⊂ Ω, µ ∈ (0,1−np).
It will be also shown that (R)⇒(L).
These results generalize the results of [2], [3]. In [2], [3], the analogous assertions are proved for nonlinear elliptic systems of the second order. The history of the regularity problem and the Liouville’s property is described in [2], [3], [5].
1. Some lemmas.
By standard arguments (see [6]), we could prove
Lemma 1.1. Letu∈[Hk,∞(Ω)]N be a weak solution to the system(0.1)and let (0.3),(0.4),(0.5)be satisfied. Thenu∈[Hlock+1(Ω)]N.
In our next considerations, we shall use the result from [1]. This result concerns the solutionsu∈[Hk(Ω)]N ∩[Hk−1,∞(Ω)]N of quasilinear elliptic systems of the type
N
X
j=1
X
|α|≤k
|β|=k
(−1)|α|Dα(Aiαjβ(x, γ1(v))Dβvj) = X
|α|≤k
(−1)|α|Dαgαi, (1.2)
x∈Ω, i= 1, . . . , N, with the following assumptions
Aiαjβ∈C(Ω×RN ̺(n,k−1)), (1.3)
giα∈Lpα(Ω), pα= p
k− |α|+ 1, p > n, p≥2(k+ 1), (1.4)
N
X
i,j=1
X
|α|=|β|=k
Aiαjβ(x, ξ)ζαi ζβj >0, (1.5)
for allx∈Ω,ξ∈RN ̺(n,k−1), ζ∈RN σ(n,k),ζ6= 0.
If we denote
[M′] ={u∈[Hk(Ω)]N∩[Hk−1,∞(Ω)]N :u is a solution to (1.2) and u[Hk−1,∞(Ω)]N ≤M′}, M′>0, [G′] ={gαi ∈Lpα(Ω) :
N
X
i=1
X
|α|≤k
kgiαkLpα(Ω)≤G′}, G′>0,
A′ = sup
|ξ|≤M′ x∈Ω
{ X
i,j,α,β
|Aαiβj(x, ξ)|},
U(x0, R) =R−n Z
B(x0,R) N
X
i=1
X
|α|=k−1
|Dαui(x)−(Dαui)x0,R|2 dx,
where u ∈ [Hk−1(Ω)]N, (Dαui)x0,R means the integral mean value Dαui in B(x0, R), we shall state the following
Lemma 1.6. Suppose that u∈ [M′] and the right-hand sides of the system 1.2 belong to [G′]. Let (1.3),(1.4),(1.5) be satisfied. Let Ω′ be a domain such that Ω′⊂⊂Ω. Let
(1.7) lim
R→0+ inf U(x, R) = 0
uniformly with respect tox∈Ω′, u∈[M′]. Thenu∈[Ck−1,µ(Ω′)]N, µ∈(0,1−np) and the a-priori estimate
kuk[Ck−1,µ(Ω′)]N ≤c(M′, G′, A′,Ω′), c >0, holds uniformly with respect to the class[M′]∪[G′].
This lemma (in a slightly generalized form) is proved in [1]. In [1], the problems analogous to those in this paper are solved for quasilinear elliptic systems of higher order. In both papers, the methods of proofs are based on the same idea. The crucial point is to show that the Liouville property implies the assumption (1.7).
But the methods are technically different.
2. Main results.
Theorem 2.1. Let u∈ [M] and the right-hand sides of the system (0.1) belong to[G]. LetΩ′ be a domain such thatΩ′⊂⊂Ω. Suppose that(0.3),(0.4),(0.5)and the condition (L)be satisfied. Then there exists a constant c =c(Ω′, M, G) such that
kuk[Ck,µ(Ω′)]N ≤c, µ∈(0,1−n p).
Proof: For all x0 ∈ Ω′ and R > 0, we define the transformation Tx0R : y = Tx0R(x) =x−xR0. Foru∈[M] (u= (u1, . . . , uN)), we define onOx0R=Tx0R(Ω) :
uix0R(y) = ui(x0+R y)
Rk − X
|γ|<k
Dγui(x0) Rk−|γ|γ! yγ, (2.2)
i= 1, . . . , N.
From (2.2), it follows that fori= 1, . . . , N
(2.3)
Dαuix0R(0) = 0, |α| ≤k−1,
Dαui(x0Ry) =Rk−|α|Dαuix0R(y) + P
|γ|<k α≤γ
R|γ−α|Bγ,αDγui(x0) γ! yγ−α,
|α| ≤k, Dαui(x0+Ry) =Dαuix0R(y) a.e. in Ox0R, |α|=k.
Bγ,α — constants which are related to the derivatives of “yγ”. Let us choose a numbera >0. Then there existsR0>0 such that for∀x0∈Ω′ and 0< R≤R0 B(0,2a) ⊂ Ox0R. From (2.3) it follows that Dαuix0R,|α| ≤ k, i = 1, . . . , N, are bounded uniformly with respect to x0 ∈Ω′ and 0< R≤R0. Clearly there exists a constantt >0 such that for allx0∈Ω′ and 0< R≤R0
(2.4) kux0Rk[Hk(B(0,2a))]N ≤t.
PuttingRkϕi(x−xR0)∈ D(Ω) in (0.2) as a test function and using the transformation x=x0+Ry we have
(2.5)
N
X
i=1
X
|α|≤k
Z
Ox0R
Rk−|α|aiα(x0+Ry, γ1(u(x0+Ry)), γ2(u(x0+Ry)))Dαϕi(y)dy=
=
N
X
i=1
X
|α|≤k
Z
Ox0R
Rk−|α|fαi(x0+Ry)Dαϕi(y)dy.
From Lemma 1.1 it follows thatux0R∈[Hlock+1(Ox0R)]N and forvx0R=∂u∂yx0R
r , the following equation in variations holds:
(2.6)
N
X
i=1
X
|α|≤k
Z
Ox0R
Rk+1−|α|∂aiα
∂xr ·Dαϕi(y)dy+
+
N
X
i,j=1
X
|α|≤k
|β|≤k
Z
Ox0R
Rk−|α|∂aiα
∂ηjβ ·
Rk−|β|Dβvxj0R(y)+
+ X
|γ|<k β+λr≤γ
R|γ−β|Bγ,β+λrDγu(x0)
γ! yγ−β−λr
Dαϕi(y)dy=
=
N
X
i=1
X
|α|≤k
Z
Ox0R
Rk+1−|α|∂fαi
∂xr(x0+Ry)Dαϕi(y)dy.
(We omitted the arguments in ∂a∂xiα
r,∂aiα
∂ηjβ, λr= (0, . . . ,0,1,0, . . . ,0)∈Rn.) From (0.5) it follows that
(2.7) Rk+1−|α|k∂fαi
∂xr (x0+Ry)kL2(B(0,2a)) ≤c1(a)R
p−n pα ·G.
Now putting in (2.6) ϕi = vix0Rχ2k, i = 1, . . . , N, where χ ∈ D(B(0,2a)),0 ≤ χ ≤ 1, χ = 1 on B(0, a) and using the standard argumentation ((0.4), uniform
boundedness of ∂a∂xiα
r,∂aiα
∂ηjβ, H¨older inequality and (2.7)), we obtain the estimates
J :=
N
X
i=1
X
|α|=k
Z
B(0,2a)
(Dαvxi0R·χk)2dy≤
≤c2
N
X
i,j=1
X
|α|=|β|=k
Z
B(0,2a)
∂aiα
∂ηjβ(Dβvjx0R·χk) (Dαvxi0R·χk)dy andJ ≤c3J12 +c4, wherec3, c4>0 are some constants.
The last inequality implies that there exists a constantc5>0 such that
(2.8)
N
X
i=1
X
|α|=k
Z
B(0,a)(Dαvxi0R(y))2dy≤J≤c5
for all x0 ∈ Ω′, u ∈ [M], R ∈ (0, R0]. (2.4) and (2.8) imply that there exists a constantc6 such that
(2.9) kux0Rk[Hk+1(B(0,a))]N ≤c6 for allx0∈Ω′, u∈[M] andR∈(0, R0].
Now we shall prove that
(2.10) lim inf
R→0+ N
X
i=1
X
|α|=k
R−n Z
B(x0,R)
|Dαui(x)−(Dαui)x0,R|2dx
= 0
holds uniformly with respect tox0 ∈Ω′ andu∈[M]. Let us suppose the contrary.
Then there exists{xs}∞s=1 ⊂Ω′, xs →x∈Ω′,{Rs}∞s=1 ⊂R+, Rs →0,{us}∞s=1 ⊂ [M] andε >0 such that
(2.11)
N
X
i=1
X
|α|=k
R−ns Z
B(xs,Rs)|Dαuis(x)−(Dαuis)xs,Rs|2dx≥ε.
Putting a = m, m ∈ N (N = {1,2, . . .}) and using (2.9) and the diagonalization process, we obtain a functionP∈[Hk+1(B(0, m))]N such that for allm∈N:
usxsRs →P in [Hk+1(B(0, m))]N weakly, (2.12)
usxsRs →P in [Hk(B(0, m))]N, (2.13)
DαuisxsRs →DαPi a.e. in B(0, m), (2.14)
|α| ≤k, i= 1, . . . , N.
(2.3), (2.14) imply that there exists a constantτ >0 such that (2.15) |DαPi| ≤τ, |α|=k, i= 1, . . . , N.
Now let ψ ∈ [D(Rn)]N. It is clear that there exist m, R1 such that suppψ ⊂ B(0, m)⊂Ox0Rfor allx0∈Ω′ and 0< R≤R1.
Puttingϕ=ψ in (2.5), we have (2.16)
N
X
i=1
X
|α|≤k
Z
B(0,m)
Rk−|α|s aiα(xs+Rsy, γ1(us(xs+Rsy)), γ2(us(xs+Rsy)))·
·Dαψi(y)dy=
=
N
X
i=1
X
|α|≤k
Z
B(0,m)
Rsk−|α|sfαi(xs+Rsy)Dαψi(y)dy.
For|α|< k,i= 1, . . . , N
|Rk−|α|s aiαDαψi| ≤c7 Rk−|α|s →0, if s→ ∞.
Using the Lebesgue’s dominated convergence theorem, we have
(2.17) lim
s→∞
Z
B(0,m)
Rk−|α|s aiαDαψidy= 0.
From the imbeddingHk,∞(Ω)֒→Hk,p(Ω)֒→֒→Ck−1(Ω) it follows that γ1(us(xs+Rsy))→γ1(P(x)), in B(0, m), s→ ∞,
γ2(us(xs+Rsy)) =γ2(usxsRs(y))→γ2(P(y)) a.e. in B(0, m), s→ ∞.
Then fori= 1, . . . , N,|α|< k, the continuity ofaiα and Lebesgue’s theorem imply
(2.18)
s→∞lim Z
B(0,m)
aiα(xs+Rsy, γ1(us(xs+Rsy)), γ2(us(xs+Rsy)))·
·Dαψi(y)dy=
= Z
B(0,m)
aiα(x, γ1(P(x)), γ2(P(y)))Dαψi(y)dy.
Now fori= 1, . . . , N, if|α|=k, thenpα=pandH1,p(Ω)֒→֒→C(Ω). Using this fact and Lebesgue’s theorem, we have (k≥1)
(2.19) lim
s→∞
Z
B(0,m)
sfαi(xs+Rsy)Dαψi(y)dy=fαi(x) Z
B(0,m)
Dαψi(y)dy= 0.
If|α|< k, pα> nthen using the same argument we have
(2.20) lim
s→∞
Z
B(0,m)
Rk−|α|s sfαi(xs+Rsy)Dαψi(y)dy= 0.
If|α|< k, pα=n, thenH1,pα(Ω)֒→Lq(Ω) for allq≥1.
Choosingqso thatk− |α| −nq >0 and using H¨older inequality, we obtain
(2.21) Z
B(0,m)
Rk−|α|s sfαi(xs+Rsy)Dαψi(y)dy
≤c8Rk−|α|−
n
s qksfαikLq(Ω)≤
≤c9Rk−|α|−
n
s q ·G→0 for s→ ∞.
If|α|< k, pα< n, thenH1,pα(Ω)֒→Lqα(Ω), where q1α =p1α −n1. Using H¨older inequality, we have
(2.22)
Z
B(0,m)
Rk−|α|s sfαi(xs+Rsy)Dαψi(y)dy ≤
≤c10R(k−|α|+1)(1−np)
s · ksfαikLqα(Ω)≤
≤c11R(k−|α|+1)(1−np)
s ·G→0 for s→ ∞.
These facts imply that
N
X
i=1
X
|α|=k
Z
Rn
aiα(x, γ1(P(x)), γ2(P(y)))Dαψi(y)dy= 0 for all ψ∈[D(Rn)]N.
From (2.15) and the condition (L) it follows thatP ∈PkN. (2.3), (2.11) and (2.13) imply
ε≤lim inf
s→∞ R−ns
N
X
i=1
X
|α|=k
Z
B(xs,Rs)
|Dαuis(x)−(Dαuis)xs,Rs|2dx≤
≤lim inf
s→∞
N
X
i=1
X
|α|=k
Z
B(0,1)
|DαusxsRs(y)−DαPi|2dy= 0.
From this it follows that (2.10) holds uniformly with respect tox0∈Ω′andu∈[M].
Now from (0.2), we obtain an equation in variations which has the following form
(2.23)
N
X
i,j=1 n
X
s,t=1
X
|α|≤k
|β|=k
Z
Ω
δst∂aiα
∂ηβj(x, γ(u))Dβ(∂uj
∂xt)Dαϕisdx=
=
N
X
i=1 n
X
s=1
X
|α|≤k
Z
Ω
h∂fαi
∂xs −∂aiα
∂xs(x, γ(u))−
−
N
X
j=1
X
|β|<k
∂aiα
∂ηjβ(x, γ(u))Dβ(∂uj
∂xs)i
Dαϕisdx, whereδst— the symbol of Kronecker delta.
Putting (2.24) gsi
α=∂fαi
∂xs −∂aiα
∂xs(x, γ(u))−
N
X
j=1
X
|β|<k
∂aiα
∂ηβj(x, γ(u))Dβ(∂uj
∂xs) and using the fact thatu∈[M], we obtain the assertion:
There exists a constantG′>0 such that (2.25)
N
X
i=1 n
X
s=1
X
|α|≤k
kgsi
αkLpα(Ω)≤G′. Fori, j= 1, . . . , N;s, t= 1, . . . , n; |α| ≤k,|β|=k, we define
Asij
αβt(x, κ) =δst∂aiα
∂ηjβ(x, γ1(u), κ), x∈Ω, κ∈RN n̺(n,k−1), κ∈RN σ(n,k).
It is clear that (u∈[Ck−1(Ω)]N)
(2.26) Asij
αβt∈C(Ω×RN n̺(n,k−1)).
PuttingUtj = ∂u∂xj
t forj = 1, . . . , N, t= 1, . . . , n in (2.23), we obtain a quasilinear system
(2.27)
N
X
i,j=1 n
X
s,t=1
X
|α|≤k
|β|=k
Z
Ω
Asij
αβt(x, γ1(U))DβUtjDαϕisdx=
=
N
X
i=1 n
X
s=1
X
|α|≤k
Z
Ω
gsi
αDαϕisdx.
It is a matter of routine calculation to verify the assumptions of Lemma 1.6 for the system (2.27). (These assumptions hold uniformly with respect to allu∈[M].)
Now from Lemma 1.6 it follows that there exists a constantc >0 such that for u∈[M], u∈[Ck,µ(Ω′)]N andkukCk,µ(Ω′)]N ≤c.
By the standard method from [2], we shall prove
Theorem 2.28. Suppose that the system(0.1)has the property of regularity(R).
Then Liouville’s property(L)holds.
Proof: Letx0 ∈Ω, ξ∈RN ̺(n,k−1)and ube a solution (inRn) to the system
(2.29)
N
X
i=1
X
|α|=k
Z
Rn
aiα(x0, ξ, γ2(u(x)))Dαϕi(x)dx= 0, ϕ∈[D(Rn)]N,
such that forM >0
(2.30) |Dαui| ≤M, |α|=k, i= 1, . . . , N.
ForR >0, we define
uiR(y) = ui(Ry)
Rk , i= 1, . . . , N.
Putting ϕ(Rx) as a test function in (2.29) and using transformation x = Ry, we obtain
N
X
i=1
X
|α|=k
Z
Rn
aiα(x0, ξ, γ2(uR(y)))Dαϕi(y)dy= 0.
(2.30) and the property (R) imply
(2.31) |DαuiR(y)−DαuiR(0)| ≤c|y|µ,
|α|=k, i= 1, . . . , N, R >0, y∈B(0, η), η >0, µ∈(0,1).
Now let us choosex∈Rn. Then there existsR0 >0 such thatyR=Rx ∈B(0, η) for allR≥R0. Using (2.31), we obtain
|Dαui(x)−Dαui(0)| ≤c|x|µ
Rµ, |α|=k, R≥R0, i= 1, . . . , N.
ForRtending to infinity, we have thatu∈PkN.
References
[1] Balanda L., Viszus E.,On Liouville theorem and H¨older continuity of weak solutions to some quasilinear elliptic systems of higher order, to appear.
[2] Giaquinta M., Neˇcas J.,On the regularity of weak solutions to non-linear elliptic systems of partial differential equations, J. Reine Angew. Math.316(1980), 140–159.
[3] Giaquinta M., Neˇcas J.,On the regularity of weak solutions to nonlinear elliptic systems via Liouville’s type property, Comment. Math. Univ. Carolinae20(1979), 111–122.
[4] Fuˇc´ık S., John O., Kufner A.,Function Spaces, Academia, Prague, 1977.
[5] Neˇcas J.,Introduction to the theory of non-linear elliptic equations, Teubner–Texte zur Ma- thematik, Leipzig, 1983.
[6] Neˇcas J.,Les m´ethodes directes en Th´eorie des ´equations elliptiques, Academia, Prague, 1967.
St´atn´ı v´yzkumn´y ´ustav pro stavbu stroj˚u, 190 11 Praha 9 – Bˇechovice, Czechoslovakia
Faculty of Mathematics and Physics, Comenius University, Mlynsk´a dolina, 842 15 Bratislava, Czechoslovakia
(Received May 31, 1991)