HoÈlder Continuity for Sub-Elliptic Systems Under the Sub-Quadratic Controllable Growth in Carnot Groups (*)
JIALINWANG(**)(***) - DONGNILIAO(***) - ZEFENGYU(***)
ABSTRACT- This paper is devoted to optimal partial regularity of weak solutions to nonlinear sub-elliptic systems for the case15m52under the controllable growth condition in Carnot groups. We begin with establishing a Sobolev-Poincare type inequality for the functionu2HW1;m(V;RN) withm2(1;2), and then partial reg- ularity with optimal local HoÈlder exponent for horizontal gradients of weak solu- tions to the systems is established by usingA-harmonic approximation technique.
MATHEMATICSSUBJECTCLASSIFICATION(2010). 35H20, 35B65.
KEYWORDS. Carnot group; Nonlinear sub-elliptic system; Sub-quadratic con- trollable growth condition; Optimal partial regularity;A-harmonic approxima- tion technique
1. Introduction and statements of main results
In this paper, we consider nonlinear sub-elliptic systems of second or- der in divergence form under the sub-quadratic (15m52) controllable growth condition in Carnot groups, and settle optimal partial regularity for horizontal gradients of weak solutions.
More precisely, let VGbe a bounded domain in a Carnot groupG with general step, and consider the following system
ÿXk
i1
XiAai(j;u;Xu)Ba(j;u;Xu); j2V; u2RN; Xu2RkN (1:1)
(*) Research supported by National Natural Science Foundation of China (No.
11126294 and No. 11201081), supported by Natural Science Foundation of Jiangxi, China (No. 2010GQS0021), and supported by Science and Technology Planning Project of Jiangxi Province, China (No. GJJ13657).
(**) Indirizzo dell'A.: Key Laboratory of Numerical Simulation Technology of Jiangxi Province, Gannan Normal University, Ganzhou, 341000, Jiangxi, P. R. China.
E-mail: jialinwang1025@hotmail.com
(***) Indirizzo degli Autori: School of Mathematics and Computer Science, Gannan Normal University, Ganzhou 341000, Jiangxi, P. R. China.
with sub-quadratic controllable structure conditions (H1)-(H3) and (C), where
(H1): Aai(j;u;p) is differentiable with respect top, with bounded and continuous derivatives, that is, there exists a constantCsuch that
Aai;pj
b(j;u;p)C(1j jp2)mÿ22 ; (j;u;p)2VRNRkN; 15m52;
(1:2)
where we denote byAai;pj
b()@Aai()
@pjb ;
(H2):Aai(j;u;p) satisfies the following ellipticity condition Aai;pj
b(j;u;p)haihbj l(1j jp2)mÿ22 j jh2; 8h2RkN; (1:3)
wherelis a positive constant;
(H3): Aai(j;u;p) is HoÈlder continuous with exponent g2(0;1) in the first and second variables, i.e.,
Aai(j;u;p)ÿAai(~j;u;~ p)K(j j)(du m(j;~j)uÿu~m)mg(1j jp2)mÿ12 ; (1:4)
where K(): [0;1)![0;1) is monotonously nondecreasing. Without loss of generality, it is convenient to takeK()1;
(C) (Controllable growth condition): Denoter mQ
Qÿmand require Ba(j;u;p)
j j4a pj jm(1ÿ1r)b uj jrÿ1c;
(1:5)
where Q3 is the homogeneous dimension in Carnot groups (see (2.3) below), anda;bandcare positive constants.
Furthermore, (H1) infers that there exists a continuously nonnegative and bounded functionv(s;t): [0;1)[0;1)![0;1), wherev(s;0)0 for alls, andv(s;t) is monotonously nondecreasing insfor fixedt;v(s;t) is concave and monotonously nondecreasing intfor fixeds, such that for all (j;u;p);(~j;u;~ ~p)2VRNRkN,
Aai;pj
b(j;u;p)ÿAai;pj
b(j;u;p)~ C(1j jp2j jp~2)mÿ22 v(j j;p jpÿ~pj):
(1:6)
As is well known, even under reasonable assumptions onAai; Bain the systems, people cannot in general expect that weak solutions of nonlinear elliptic systems of equations will be classical (i.e.C2-solutions) like elliptic scalar equations. This was first shown by De Giorgi [1]. Then the goal is to establish partial regularity of weak solutions for systems. Such regularity
means that for any weak solutionuof a system, there exists an open subset V0Vsuch thatVnV0has a zero Lebesgue measure anduor its gradient Du is locally regular in V0. We refer the reader to monographs of Gia- quinta [2, 3] and Chen-Wu [4].
There are different methods to prove partial regularity: the direct approach was first carried out by Giaquinta-Modica [5]; Guisti-Miranda [6]
was earlier to employ the blow-up method; furthermore, Duzaar and Grotowski in [7] generalized so called A-harmonic approximation techni- que in [8] and gave the remarkable proof of partial regularity for systems with quadratic growth conditions(m2). This method has two major ad- vantages: the first is that we only need to establish a Caccioppoli type inequality; this avoids having to prove an inverse HoÈlder inequality. The other is that one can obtain the optimal HoÈlder exponent in partial reg- ularity. Then, Duzaar et al. [9] studied partial regularity of almost mini- mizers of quasi-convex variational integrals with sub-quadratic growth, and we also note that the degenerate p-Laplacian version of the method has been obtained in [10], and applied to the partial regularity in [11]. In the paper [9], Duzaar et al. actually provide a partially new proof of the original regularity result of Carozza et al. [12]. Later, Chen and Tan in [13, 14] extended Duzaar and Grotowski's results [7] to more general nonlinear elliptic systems under the super-quadratic growth (m>2) and sub- quadratic growth (15m52), respectively.
Several regularity results were focused on systems constructed by basic vector fields in Carnot groups. Capogna and Garofalo in [15] showed the partial HoÈlder regularity for quasi-linear sub-elliptic systems under the quadratic structure conditions in Carnot groups of step two. Shores in [16] considered a homogeneous quasi-linear system under the quadratic growth condition on the Carnot group with general step. She first estab- lished higher differentiability and smoothness for weak solutions of the system with constant coefficients, and then deduced the partial regularity.
Their methods depend mainly on generalization of classical direct method in the Euclidean space. Later, by the method ofA-harmonic approxima- tion, FoÈglein in [17] treated the homogeneous nonlinear system
ÿX2n
i1
XiAai(j;Xu)0; a1; ;N
on the Heisenberg group under super-quadratic structure conditions. She got partial regularity for the horizontal gradient of weak solutions to the initial system. Then the first author and Niu in [18] considered more gen-
eral nonlinear sub-elliptic systems (see (1.1)) in Carnot groups under super- quadratic growth conditions, and established the optimal partial regularity of the horizontal gradient of weak solutions.
In this paper, we will apply the method ofA-harmonic approximation adapted to the setting of Carnot groups to study partial regularity for the system (1.1) under the controllable growth conditions with sub-quadratic case (15m52). The key point is to establish a certain excess-decay esti- mate for the excess functionF. In the casem2, this function is given by
F(j0;r;p0)Br(j0)ÿ1
G
Z
Br(j0)
jXuÿp0j2jXuÿp0jm dj;
whereas in the case 15m52, one uses F(j0;r;p0)Br(j0)ÿ1
G
Z
Br(j0)
V(Xu)ÿV(p0) j j2dj;
(1:7)
where V(A)(1 jAj2)mÿ24 for A2Hom(Rk;RN). It is shown that if F(j0;r;p0) is small enough on a ballBr(j0)V, then for some fixedu2(0;1) one has the excess improvementF(j0;ur;p0)Cu2gF(j0;r;p0). Iteration of this result yields the excess-decay estimate which implies the regularity result. Although the underlying philosophy in this paper is encouraged by that in [14, 9], some different treatment are necessary. Since basic vector fields (see (2.1) below) of Lie algebras corresponding to the Carnot group are more complicated than gradient vector fields in the Euclidean space, we have to find a different auxiliary function in proving Caccioppoli type in- equality. Inspired by [17], we choose horizontal variables to construct such a suitable function (see Remark 1 below); Besides, the non-horizontal deri- vatives of weak solutions will happen in the Taylor type formula on the Carnot group and cannot been effectively controlled in the present hy- potheses. So the method employing Taylor's formula in [14] is not appro- priate in our setting. In order to obtain the desired decay estimate, we need establish and use the Sobolev-Poincare type inequality (3.1) instead.
The main result in this paper is as follows.
THEOREM 1. Assume that coefficients Aai and Ba satisfy conditions (H1)-(H3)and(C), and u2HW1;m(V;RN)be a weak solution to the sys- tem(1.1)withVG, i.e.,
Z
V
Aai(j;u;Xu)XiWadj Z
V
Ba(j;u;Xu)Wadj 8W2C10 (V;RN):
(1:8)
Then there exists an open subsetV0 V;such that u2G1;g(V0;RN), withg in(1.4).Furthermore,VnV0S1[S2and meas(VnV0)0, where
S1n
j02V: lim
r!0sup (Xu)ÿ j0;r
1o
; S2
(
j02V: lim
r!0infjBr(j0)jÿ1 Z
Br(j0)
Xuÿ(Xu)j0;r
mdj>0 )
:
To the best of our konwledge, in the Heisenberg group,G0;gregularity of weak solutions to sub-ellipticp-Laplacian equations is valid for 15p51 and was proved by several people during the 90's in [19, 20, 21], but when turning toG1;gregularity, it is worthy of pointing out that the remarkable contribution of HoÈlder continuity for the gradient of weak solutions to sub- ellipticp-Laplacian equations is due to Capogna [22, 23], Marchi [24, 25], Domokos [26, 27], Manfredi and Mingione et al. [28, 29], but the exponent p should be near 2, and the limitation 2p54 appears in the most recent of the cited works. Recently, Garofalo in [30] obtained theG1;gregularity of weak solution which possess some special symmetries for 2p. When turning to partial continuity of weak solutions, our result shows that the G1;g continuity is also valid for exponent 15p52.
The plan of this paper is organized as follows: In Section 2, we in- troduce two functions and their some useful properties, and collect some basic notions and facts associated to Carnot groups. Since in the sub- quadratic case we are dealing with functions belonging to the horizontal Sobolev spaceHW1;m with 15m52, in the proof of the main results, a Sobolev-Poincare type inequality (see (3.1) below) will be used. So we will prove this inequality and a prior estimate for constant systems in Section 3. In Section 4, we show a Caccioppoli type inequality for weak solutions to the system (1.8). Section 5 is devoted to the proof of The- orem 1, and the process of the proof is split into four steps. An estimate on weak solutions is established in the first step. Second step begins with proving three ``small conditions'' (see (5.13)-(5.15) below), and then obtain the excess-improvement for the functionFin (1.7). Third step is to show that we can iterate the statement in the second step. A decay estimate
F(j0;r;p0)Cr2g
is obtained in the last step, wheregis the same as (1.4). Then we can infer the conclusion of Theorem 1 employing Lemma 2 and Lemma 3 below.
2. Preliminaries
A Carnot groupGof stepris a simply connected, nilpotent Lie group whose Lie algebra eg admits a stratification, i.e., egLr
j1Vj such that [V1;Vj]Vj1; j1; ;rÿ1 and [V1;Vr] f0g:Let Xlidenote a left- invariant basis vector field ofVlwith 1lrand 1iml, wheremlis the dimension ofVl. For the sake of simplicity, we letXiXi1,km1, and denote by X(X1; ;Xk) the sub-elliptic gradient. We will say that Xi(i1; ;k) are the horizontal vector fields with the form
Xi@i Xn
ji1
aij(j)@j; Xi(0)@i; (2:1)
where aij(j) is a polynomial. For a vector valued function u(u1; ;uN):G7!RN, we let Xiua(i1; ;k;a1; ;N) be a horizontal direction derivative, and say thatXuis the horizontal Jacobian with entriesXiua.
Denoting
j(j1;j2; ;jr)x11;x12; ;x1m1;x21; ;x2m2; ;xr1; ;xrml 2G;
the distance from origin defined by
d(j) Xr
l1
Xml
i1
xli2
!r!l 2
4
3 5
2r!1
: (2:2)
For any j;h2G, we set d(j;h)d(hÿ1j), where hÿ1 ÿh (ÿh1; ;ÿhr) is the reverse of h, and is the multiplication rule in G defined by jejjejP(j;ej); j;ej2G; where P:GG7!G has polynomial components.
Following [31], we introduce the gauge ball and sphere Br(j) fh2G d(j;j h)5rg; @Br(j) fh2G d(j;j h)rginG, respectively. In what follows, we denote by vGjB(0;1)jG the volume of unit ball. Then
B(j;r)
j jGvGrQ, where
QXr
l1
l ml
(2:3)
is the homogeneous dimension ofG.
DEFINITION1. Let 1m51, andVGbe an open set. Ifu2Lm(V) satisfies
u
k kHW1;m(V)k ku Lm(V)Xk
i1
Xiu
k kLm(V)51;
(2:4)
we say thatubelongs to the horizontal Sobolev space. The spaceHW01;m(V) is the completion ofC10 (V) under the norm (2.4).
DEFINITION2. LetVG,g2(0;1), and denote
G1;g(V)fu2L1(V)jXiu2Gg(V);i1; ;kg;
whereGg(V)n
u2L1(V)jujg sup
j;~j2V;j6~j
u(j)ÿu(~j) d(j;~j)g 51o
. We say that G1;g(V) is a Folland-Stein space with the norm
u
k kG1;g(V)k ku L1(V)kXukL1(V)jujg jXujg: (2:5)
DEFINITION 3. Let VG; 1m51 and m0. We say that u2Lm(V) belongs to the Campanato spaceLm;m(V) if
u
k kLm;m(V)k ku Lm(V) sup
j2V;r5diamVrÿm Z
Br(j)\V
u(z)ÿuj;r mdz 0
B@
1 CA
m1
51:
(2:6)
Capogna in [23] proved a Campanato type lemma in Carnot groups.
LEMMA2. Letj2VG;m1, r>0, and Q5mQm, then Lm;m(Br(j))Gg(Br(j)); gmÿQ
m : (2:7)
Throughout the paper, we shall use the functions V Vm, WWm:Rn !Rndefined by
V(Q)Q=(1j jQ2)2ÿm4 ; W(Q)Q=(1j jQ2ÿm)12 (2:8)
for eachQ2Rn andm>1. By the elementary inequalityk kx 2
2ÿmk kx 1 2m2k kx 2
2ÿmapplied to the vectorx(1;j jQ2ÿm)2R2 we conclude that 1j jQ2
2ÿm2
1j jQ2ÿm2m21j jQ22ÿm2
; (2:9)
which immediately yields W(Q)
j j jV(Q)j 2m4jW(Q)j:
(2:10)
The purpose of introducing W is the fact that in contrast to j jV m2, the functionj jWm2 is convex (see [9]).
The following lemma includes some useful properties of the functionV, which can be found in Lemma 2.1 of [12].
LEMMA3. Let m2(1;2)and V; W:Rn !Rnbe the functions defined in(2.8 ).Then there holds for anyQ1;Q22Rnand t>0:
(1). 1
p2minj j;Q1 j jQ1m2
jV Q1 j minj j;Q1 j jQ1m2
; (2). jV(tQ1)j maxÿt;tm2
V(Q1) j j;
(3). jV(Q1Q2)j C(m)(jV(Q1)j jV(Q2)j);
(4). m
2 jQ1ÿQ2jjV(Q1)ÿV(Q2)j=(1j jQ12j jQ22)mÿ24 C(m;n)jQ1ÿQ2j;
(5). jV(Q1)ÿV(Q2)j C(m;n)jV(Q1ÿQ2)j;
(6). jV(Q1ÿQ2)j C(m;M)jV(Q1)ÿV(Q2)jfor allQ1 withjQ2j M.
The inequalities(1)-(3)also hold if V is replaced by W.
For later purpose we state the following two simple estimates which can be easily deduced from Lemma 3 (1) and (6). LetQ1;Q22RnwithjQ2j Mit follows that
Q1ÿQ2
j j2C(m;M)jV(Q1)ÿV(Q2)j2; (2:11)
forjQ1ÿQ2j 1;while forjQ1ÿQ2j>1 Q1ÿQ2
j jmC(m;M)jV(Q1)ÿV(Q2)j2: (2:12)
The next lemma due to [9] is a more general version of the Lemma 2.7 in [12].
LEMMA 4. Let0n51;a;b0; v2Lp(Br(j0))and g be a non-nega- tive bounded function satisfying
g(t)ng(s)a Z
Br(j0)
V v sÿt
2djb;
for all r
2t5sr. Then there exists a constant CC(n)such that g r
2 C(n) a Z
Br(j0)
V v r
2djb 0
B@
1 CA:
We conclude this section with an algebraic fact from [32]
LEMMA5. For every t2 ÿ1 2;0
andm0, it holds
1 R1
0m2jps(~pÿp)j2t ds m2j jp2j j~p2
t 8
2t1; (2:13)
for any p;p~2RkN, not both zero ifm0.
3. Sobolev-Poincare type inequality and a prior estimate in the sub- quadratic case
LEMMA 6 (Sobolev-Poincare type inequality). Let m2(1;2) and u2HW1;m(Br(j0);RN)with Br(j0)V, then
(3:1) Br(j0)ÿ1
G
Z
Br(j0)
W uÿuj0;r r
Qÿm2Q
dj 0
B@
1 CA
Qÿm2Q
CP Br(j0)ÿ1
G
Z
Br(j0)
W Xu j j2dj 0
B@
1 CA
12
:
Furthermore, the analogous inequality is valid with W replaced by V defined in (2.8), and in particular, the inequality also holds if we substitute2 for 2Q
Qÿm.
PROOF. Similarly to the proof of Theorem 2 in [9]. Following [33, Lemma 3.2], there existc>1 andC>1 such that forj2B(j0;r)
u(j)ÿuj0;r
C
Z
Bcr(j0)
Xu(h) j j d(j;h)Qÿ1dh:
(3:2)
SinceW2=m(t) is non-decreasing and convex, we applyW2=m(t) to both sides of the last inequality and have by Jensen's inequality and recalling W(0)0:
W2=m u(j)ÿuj0;r r
C r Z
G
d1ÿQ(j;h)We2=m jXu(h)jdh;
(3:3)
where the numberQis the homogeneous dimension inG, and W Xu(h)e j j 0; h2=Bcr(j0);
W Xu(h) j j; h2Bcr(j0):
(
Since W Xu(h) j j 2LmÿBr(j0)
by Lemma 3(1), it implies W Xu(h)e j j 2 Lm . Then Theorem 2.7 in [19] applied withG We2=m jXuj; a1 yields
Z
Br(j0)
W u(j)ÿuj0;r
r
Qÿm2Q
dj 2
64
3 75
QÿmmQ
Z
Br(j0)
W2=m u(j)ÿuj0;r r
QÿmmQ dj 2
64
3 75
QÿmmQ
C r
Z
G
Z
G
d1ÿQ(j;h)We2=m jXu(h)jdh 0
@
1 A
QÿmmQ
dj 2
64
3 75
QÿmmQ
C r
Z
G
We2=m jXu(j)jmdj 2
4
3 5
m1
C r
Z
Br(j0)
W2 jXu(j)jdj 2
64
3 75
m1
;
or
Z
Br(j0)
W u(j)ÿuj0;r 2r
Qÿm2Q
dj 2
64
3 75
Qÿm2Q
Crÿm2 Z
Br(j0)
W2 jXu(j)jdj 2
64
3 75
12
: (3:4)
Note thatW(Q=2)W(Q)=2, we conclude the result (3.1) from (3.4).
LEMMA7. Let u2HW1;1(V;RN)be a weak solution of Z
V
AabijXjubXifadj0 (3:5)
for anyf2C01(V;RN), where Aabij is a constant matrix satisfying the strong Legendre-Hadamard condition:
Aabijhahbmimj >cj jh2j jm2; h2RN; m2Rk:
Then u is smooth and there exists C01such that for any Br(j0)V (3:6) sup
Br=2(j0)uÿuj0;r2r2jXuj2r4X2u2
C0Br(j0)ÿ1
G r2 Z
Br(j0)
Xu j j2dj:
PROOF. Shores [16] has shown that a weak solutionv2HW1;2(V;RN) of (3.5) belongs toC1in the subsetV0V, and Wang and Niu [18] have established the following estimate:
Bsupr=2(j0)vÿvj0;r2r2j jXv2r4 X2v2
C0Br(j0)ÿ1
G r2 Z
Br(j0)
Xv j j2dj:
(3:7)
Then we can argue as in the proof of Proposition 2.10 in [12] to obtain the conclusion.
4. Caccioppoli type inequality
Denote by Bil (RkN) the collection of bi-linear forms defined inRkN, and suppose A 2Bil(RkN). We say that a function h2HW1;m(V;RN) is A-harmonic, if hsatisfies
Z
V
A(Xh;XW)dj0; 8W2C01(V;RN):
(4:1)
Similarly to [9], one can establish the following A-harmonic approx- imation lemma for the case 15m52 in Carnot groups.
LEMMA 8. Let l and L be fixed positive numbers, 15m52, and k;N2N with k2. If for any given e>0, there exists d d(k;N;l;L;e)2(0;1]with the following properties:
(I) for anyA 2Bil(RkN)satisfying
A(n;n)l nj j2 andA(n;n)Lj jn j j;n n;n2RkN; (4:2)
(II) for any g2HW1;m(Br(j0);RN)satisfying Br(j0)
ÿ1
G
Z
Br(j0)
W(Xg)
j j2 dj21and (4:3)
(4:4)
Br(j0)ÿ1
G
Z
Br(j0)
A(Xg;XW)dj
dsup
Br(j0)jXWj; 8W2C01(Br(j0);fRN);
then there exists anA-harmonic function h h2Hn
h2HW1;m(Br(j0);fRN)Br(j0)ÿ1
G
Z
Br(j0)
W(Xh)
j j2 dj1o
such that
Br(j0) ÿ1
G
Z
Br(j0)
W gÿh r
2 dj2e:
(4:5)
To establish the main result, our first aim is to establish a suitable Caccioppoli type inequality.
LEMMA9 (Caccioppoli Type Inequality). Let u2HW1;m(V;fRN)be a weak solution to the system(1.1)under the conditions(H1)-(H3)and(C).
Then for everyj0(j10; ;jr0)2V; u02 RN; p02 fRkN and anyrsat- isfying05r5minf1; dist(j0; @V)g, it holds
Br=2(j0) ÿ1
G
Z
Br=2(j0)
V(Xuÿp0) j j2 dj
Cc Br(j0)ÿ1
G
Z
Br(j0)
V uÿu0ÿp0(j1ÿj10) r
!
2
dj 8>
<
>:
Kÿ
ÿ
1j jp0m=2s
r2gBr(j0)ÿ1
G
Z
Br(j0)
Xu
j jmj jur1
dj
2 64
3 75
mÿ1m 1ÿ1r 9
>>
=
>>
;; (4:6)
where we denote K()K u j j 0 j jp0,j1(j11;j12; ;j1k)is the horizontal component ofj(j1; ;jr)2G,smaxf2m=(mÿ2g);m=(mÿ1ÿg)g>2 and the constant CcCc(a;b;c;n;N;m;L;l;M).
PROOF. Let Br(j0)V, r=2t5sr, and a standard cut off functionh2C01(Bs(j0)) satisfy 0h1; j j5rh C
sÿtandh1 onBt(j0).
Inspired by the way of [17], we let vu(j)ÿu0ÿp0(j1ÿj10), and then define two functions
Whv; f(1ÿh)v;
(4:7) one has
XWXfXuÿp0 (4:8)
and
XW
j jm;jXfjmC(m) j jXvm v sÿt m
: (4:9)
Using hypothesis (H2), from Lemma 5, and as the elementary inequality 1j ja2jbÿaj23 1 j ja2j jb2
; (4:10)
we have Z
Bs(j0)
Aai(j;u;p0XW)ÿAai(j;u;p0)
XiWadj
Z
Bs(j0)
Z1
0
dAai(j;u;p0uXW)
du du
2 4
3 5XiWadj
Z
Bs(j0)
Z1
0
@Aai(j;u;p0uXW)
@pbj duXjWbXiWadj
l Z
Bs(j0)
Z1
0
1jp0u XWp0 ÿp0j2
h i(mÿ2)=2
dujXWj2dj
l Z
Bs(j0)
1j jp02jXWÿp0j2
(mÿ2)=2
XW j j2dj
3(mÿ2)=2l Z
Bs(j0)
1j jp02jXWj2
(mÿ2)=2
jXWj2dj:
(4:11)
From (4.11), it follows that 3(mÿ2)=2l
Z
Bs(j0)
1j jp02jXWj2
(mÿ2)=2
XW j j2dj
Z
Bs(j0)
Aai(j;u;p0XW)ÿAai(j;u;p0)
XiWadj
Z
Bs(j0)
Aai(j;u;Xu)XiWadjÿ Z
Bs(j0)
Aai(j;u;p0)XiWadj
Z
Bs(j0)
Aai(j;u;p0XW)ÿAai(j;u;Xu)
XiWadj
ÿ Z
Bs(j0)
Aai(j;u;p0)ÿAaij;u0p0(j1ÿj10);p0
h i
XiWadj
ÿ Z
Bs(j0)
Aaij;u0p0(j1ÿj10);p0
ÿAai(j0;u0;p0)
h i
XiWadj
ÿ Z
Bs(j0)
Z1
0
@Aai j;u;Xuÿu(XuÿXWÿp0)
@pbj duXjfbXiWadj
Z
Bs(j0)
Bi(j;u;Xu)Wadj :IIIIIIIV;
(4:12)
where we have used (4.8), (1.8) and the fact that Z
BR(j0)
Aai(j0;u0;p0)XWa dj0:
Noting thatWvonBt(j0), and (4.10), the left-hand side in (4.12) can be estimated by
(4:13) 3(mÿ2)=2l Z
Bt(j0)
1j jp02jXWj2
(mÿ2)=2
jXWj2dj
3(mÿ2)=2l Z
Bt(j0)
1j jp02j jXv2
(mÿ2)=2
Xv j j2dj
C(m)l Z
Bt(j0)
1j jp02jXuj2
(mÿ2)=4
Xuÿp0
j j
2
dj
C(k;N;m;l) Z
Bt(j0)
V(Xu)ÿV(p0)
j j2dj
C(k;N;m;l;M) Z
Bt(j0)
V(Xv) j j2dj;
where we have applied Lemma 3(4) in the second inequality and Lemma 3(6) for the final inequality.
The structure condition (H3) yields I
Z
Bs(j0)
K() 1 j jp0m=2j jvgjXWjdj:
(4:14)
To obtaina suitable estimate forI, weneed take the domainBs(j0) into four parts:Bs(j0)\fjv=sj>1g \fjXWj 1g,Bs(j0)\fjv=sj>1g \fjXWj>1g, Bs(j0)\fjv=sj 1g \fjXWj>1g, Bs(j0)\fjv=sj 1g \fjXWj 1g, and will use Young's inequality, (2.11) and (2.12), repeatedly.
CASE1. On the partBs(j0)\fjv=sj>1g \fjXWj 1g, we see K() 1 j jp0m=2 v
s
gsgjXWj C(e)jXWj2C(e)v
s
mC(e)hK() 1 j jp0m=2sgi2m=(mÿ2g) C(e)jV(Xv)j2C(e)V v
s
2C(e)hK() 1 j jp0m=2i2m=(mÿ2g) s2g; (4:15)
where we have used the factsg51 and 2m=(mÿ2g)>2:
CASE2. On the setBs(j0)\fjv=sj>1g \fjXWj>1g, it holds K() 1 j jp0m=2 v
s
gsgjXWj C(e)jXWjmC(e)v
s
mC(e)hK() 1 j jp0m=2sgim=(mÿgÿ1) C(e)jV(Xv)j2C(e)V v
s
2C(e)hK() 1 j jp0m=2im=(mÿgÿ1) s2g; (4:16)
where we have usedm=(mÿgÿ1)>2:
CASE 3. On the set Bs(j0)\fjv=sj 1g \fjXWj>1g, observing m=(mÿ1)>2;one has
K() 1 j jp0m=2j jvgjXWj K() 1 j jp0m=2sgjXWj ejXWjmC(e)
K() 1 j jp0m=2sgm=(mÿ1) C(e)jV(Xv)j2C(e)
K() 1 j jp0m=2m=(mÿ1) s2g: (4:17)
CASE4. ForBs(j0)\fjv=sj 1g \fjXWj 1g, there is K() 1 j jp0m=2j jvgjXWj
K() 1 j jp0m=2sgjXWj ejXWj2C(e)
K() 1 j jp0m=2sg2 C(e)jV(Xv)j2C(e)
K() 1 j jp0m=22 s2g: (4:18)
Combining these estimations with (4.14), we get IC(e)
Z
Bs(j0)
V(Xv)
j j2djC(e) Z
Bs(j0)
V v s 2dj
C(e)hK() 1 j jp0m=2is Bs(j0) j jGs2g; (4:19)
wheresmax 2m=(mf ÿ2g);m=(mÿ1ÿg)g>2.
Similarly toI, it follows that IIC(e)
Z
Bs(j0)
V(Xv)
j j2djC(e) Z
Bs(j0)
V v s
2dj
C(e)hK() 1 j jp0m=2gim=(mÿ1) Bs(j0) j jGs2g: (4:20)
By (H1), Lemma 5 and (4.10), it holds
IIIC Z
Bs(j0)
Z1
0
1jXuu XuÿXfÿXuj2
(mÿ2)=2
du 2
4
3
5jXfjjXWjdj
C Z
Bs(j0)
1jXuj2jXuÿXfj2
(mÿ2)=2
Xf j jjXWjdj
C Z
Bs(j0)
1jXuj2jXfj2
(mÿ2)=2
Xf j jjXWjdj:
(4:21)
Noting thatXfBsnBt andÿ1=25(mÿ2)=250, we split the domain Bs(j0) into four parts:Bs(j0)\fjXfj>1g\fjXWj>1g,Bs(j0)\fjXfj 1g\
XW j j 1
f g, Bs(j0)\fjXfj>1g \fjXWj 1g, and Bs(j0)\fjXfj 1g \ XW
j j>1
f g. Similarly toI, it follows that
IIIC Z
Bs(j0)nBt(j0)
V(Xuÿp0) j j2dj
Z
Bs(j0)
V v sÿt
2dj 0
B@
1 CA:
(4:22)
Using HoÈlder inequality, one has IV C
Z
Bs(j0)
Xu
j jmj jur1
1ÿ1rj jdjW
C Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
Z
Bs(j0)
W j jrdj 0
B@
1 CA
1r
: (4:23)
Analogously asI, we take the domainBs(j0) into two parts.
CASE1. ForBs(j0)\fj jXv >1g, by Sobolev type inequality, Young's inequality and (2.12) it follows that
Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
Z
Bs(j0)
W j jrdj 0
B@
1 CA
1r
C Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
Z
Bs(j0)
Xv j jmdj 0
B@
1 CA
m1
C(e) Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
mÿ1m
Ce Z
Bs(j0)
Xv j jmdj 0
B@
1 CA
C(e) Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
mÿ1m
Ce Z
Bs(j0)
V Xv j j2dj 0
B@
1 CA:
(4:24)
CASE2. On the setBs(j0)\fj j Xv 1g, we have Xv
j jmj jXv21;
and then it follows that Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
Z
Bs(j0)
W j jrdj 0
B@
1 CA
1r
C(e) Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
mÿ1m
Ce Z
Bs(j0)
V Xv j j2dj 0
B@
1
CACejBs(j0)jG; (4:25)
where we have used (2.11).
Combining these estimates inIV, we have
(4:26) IV Ce Z
Bs(j0)
V Xv j j2dj 0
B@
1 CA
C(e) Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
mÿ1m
CejBs(j0)jG:
Substituting (4.13), (4.19), (4.20), (4.22) and (4.26) into (4.12), we finally arrive at
C1
Z
Bt(j0)
V(Xv) j j2dj
C2
Z
Bs(j0)nBt(j0)
V(Xv)
j j2V v sÿt
2
dj
C3 Z
Bs(j0)
V(Xv)
j j2V v sÿt
2
djC4hK() 1 j jp0m=2is Bs(j0) j jGs2g
C5 Z
Bs(j0)
Xu
j jmj jur1
dj
2 64
3 75
1ÿ1r
mÿ1m
:
The proof is now completed by applying Lemma 4.
REMARK1. The functionvtaken in the proof is different from that in [14], essentially.
5. Proof of the main result
In this section, we will complete the proof of Theorem 1 via four steps, and hence consideru2HW1;m(V;RN) (15m52) to be a weak solution of the system (1.1).
FIRST STEP. We claim that if r1 and W2C10 (Br(j0);fRN) with
Bsupr(j0)jXWj 1, then there exist some constants C6C6(m;M;CP)>1 such that
Br(j0) ÿ1
G
Z
Br(j0)
Aai;pj
b(j0;u0;p0)(Xuÿp0)XWa dj C6 sup
Br(j0)jXWj v12j j;p0 F12(j0;r;p0)
F12(j0;r;p0) h
F(j0;r;p0)rgF(j j u0 j j)p0 i
; (5:1)
where we denoteF(t)[K(t)(1t)s0]mÿg2m withs0maxn
m;rÿ1;m 2 go
. PROOF. Noting the fact
Z
Br(j0)
" Z1
0
Aai;pj
b(j0;u0;uXu(1ÿu)p0)(Xuÿp0)du
# XWadj
Z
Br(j0)
" Z1
0
d
duAai(j0;u0;uXu(1ÿu)p0)du
# XWadj
Z
Br(j0)
[Aai(j0;u0;Xu)ÿAai(j0;u0;p0)]XWadj
Z
Br(j0)
hAai(j0;u0;Xu)ÿAai(j;u;Xu)i
XWadj Z
Br(j0)
Ba(j;u;Xu)Wadj;
(5:2)
we have (5:3)
Z
Br(j0)
Aai;pj
b(j0;u0;p0)(Xuÿp0)XWa dj
Z
Br(j0)
Z1
0
Aai;pj
b(j0;u0;p0)du(Xuÿp0) 2
4
3 5XWadj
Z
Br(j0)
Z1
0
Aai;pj
b(j0;u0;p0)ÿAai;pj
b(j0;u0;uXu(1ÿu)p0)jXuÿp0jduo
Bsupr(j0)jXWjdj
Z
Br(j0)
Aai(j0;u0;Xu)ÿAai(j;u0p0(j1ÿj10);Xu)
sup
Br(j0)jXWjdj
Z
Br(j0)
Aai(j;u0p0(j1ÿj10);Xu)ÿAai(j;u;Xu)
sup
Br(j0)jXWjdj
Z
Br(j0)
(a pj jm(1ÿ1r)b uj jrÿ1c)j jW dj :I0II0III0IV0:
Using (H1) and the estimate (1.6) yields (Note thatmÿ250) (5:4) I0
Z
Br(j0)
Z1
0
Aai;pj
b(j0;u0;p0)ÿAai;pj
b(j0;u0;p0u(Xuÿp0))
1212dt 2
4
3
5jXuÿp0jdj
C Z
Br(j0)
Z1
0
1j jp02
mÿ22
1jp0u(Xuÿp0)j2mÿ22
12
8<
:
1j jp02jp0u(Xuÿp0)j2mÿ22
v j j;p0 ju(Xuÿp0)j
12
dt )
Xuÿp0
j jdj C
Z
Br(j0)
1j jp02jXuj2 mÿ24
v12 j j;p0 jXuÿp0jjXuÿp0jdj
C Z
Br(j0)
1jXuÿp0jmÿ22
Xuÿp0
j jv12 j j;p0 jXuÿp0jdj:
Let
B1:Br(j0)\fjXuÿp0j 1g; B2:Br(j0)\fjXuÿp0j>1g:
(5:5)
Then it follows that by first HoÈlder's inequality and then Jensen's in- equality
(5:6) I0 C Z
B1
Xuÿp0
j jv12 j j;p0 jXuÿp0jdj
C Z
B2
Xuÿp0
j jm2v12 j j;p0 jXuÿp0jdj
C Z
B1
Xuÿp0 j j2dj 0
B@
1 CA
12 Z
Br(j0)
v j j;p0 jXuÿp0jdj 0
B@
1 CA
12
C Z
B2
Xuÿp0
j jmdj 0
B@
1 CA
12 Z
Br(j0)
v j j;p0 jXuÿp0jdj 0
B@
1 CA
12
C Z
Br(j0)
V(Xu)ÿV(p0)
j j2dj
0 B@
1 CA
12 Z
Br(j0)
v j j;p0 jXuÿp0jdj 0
B@
1 CA
12
C Z
Br(j0)
V(Xu)ÿV(p0)
j j2dj
0 B@
1 CA
12 Z
Br(j0)
v j j;p0 jXuÿp0jdj 0
B@
1 CA
12
2C B r(j0)
G Br(j0)ÿ1
G
Z
Br(j0)
V(Xu)ÿV(p0)
j j2dj
0 B@
1 CA
12
Br(j0)ÿ1
G
Z
Br(j0)
v j j;p0 jXuÿp0jdj 0
B@
1 CA
12
C B r(j0)
GF12(j0;r;p0)v12 j j;p0 Br(j0)ÿ1
G
Z
B1
Xuÿp0 j j2dj
!12
Br(j0)ÿ1
G
Z
B2
Xuÿp0 j jmdj
!m1!