Gradient estimates for solutions of the Lame´ ffi system with partially infinite coe cients in dimensions greater than two 6 1 0 2 JiGuang Bao HaiGang Li and YanYan Li ∗ ∗† ‡ n a January 29, 2016 J 8 2 ] Abstract P A We establish upper bounds on the blow-up rate of the gradients of solutions . oftheLame´ system withpartially infinitecoefficients indimensions greater than h twoasthedistancebetweenthesurfacesofdiscontinuity ofthecoefficientsofthe t a systemtendstozero. m [ 1 1 Introduction and main results v 9 7 In this paper, we establish upper bounds on the blow-up rate of the gradients of solu- 8 tionsoftheLame´ systemwithpartiallyinfinitecoefficients indimensionsgreater than 7 0 twoasthedistancebetweenthesurfaces ofdiscontinuityofthecoefficientsofthesys- . 1 temtends tozero. Thiswork isstimulatedbythestudyofBabus˘ka, Andersson,Smith 0 and Levin in [10] concerning initiation and growth of damage in composite materi- 6 1 als. The Lame´ system is assumed and they computationally analyzed the damage and : fracture in composite materials. They observed numerically that the size of the strain v i tensor remains bounded when the distance ǫ, between two inclusions, tends to zero. X This was proved by Li and Nirenberg in [32]. Indeed such ǫ-independent gradient r a estimates was established there for solutions of divergence form second order elliptic systems, including linear systems of elasticity, with piecewise Ho¨lder continuous co- efficients inalldimensions. See Bonnetierand Vogelius[16]and Liand Vogelius[33] forcorrespondingresultson divergenceformellipticequations. Theestimatesin[32]and[33]dependontheellipticityofthecoefficients. Ifellip- ticity constants are allowed to deteriorate, the situation is very different. Consider the scalarequation a (x) u = 0 in Ω, k k ∇· ∇ (1.1) u =(cid:16) ϕ (cid:17) on ∂Ω, k School of Mathematical Sciences, Beijing Normal University, Laboratory of Mathematics and ∗ ComplexSystems,MinistryofEducation,Beijing100875,China. Email:[email protected]. Correspondingauthor.Email:[email protected]. † Department of Mathematics, Rutgers University, 110 Frelinghuysen Rd, Piscataway, NJ 08854, ‡ USA.Email:[email protected]. 1 where Ωis a bounded open set of Rd, d 2, containingtwo ǫ-apart convexinclusions ≥ D and D , ϕ C2(∂Ω)isgiven,and 1 2 ∈ k (0, ) in D D , a (x) = ∈ ∞ 1 ∪ 2 k 1 inΩ D D . \ 1 ∪ 2 When k = , the L -norm of u for the solutions u of (1.1) generally becomes ∞ unbounded∞as ǫ tends to 0. The|∇bl∞ow| up rate of u is∞respectively ǫ 1/2 in dimen- − sion d = 2, (ǫ lnǫ ) 1 in dimension d = 3, and|∇ǫ 1∞i|n dimension d 4. See Bao, − − | | ≥ Li and Yin [11], as well as Budiansky and Carrier [18], Markenscoff [36], Ammari, Kang and Lim [7], Ammari, Kang, Lee, Lee and Lim [8] and Yun [41, 42]. Further, moredetailed,characterizationsofthesingularbehaviorof u havebeenobtainedby ∇ ∞ Ammari,Ciraolo,Kang,LeeandYun[3],Ammari,Kang,Lee,LimandZribi[9],Bon- netierandTriki[14,15],GorbandNovikov[24]andKang,LimandYun[25,26]. For relatedworks,see[2,4,5,12,15,17,19,20,21,22,28,29,30,31,34,35,37,39,40] and thereferences therein. In this paper, we mainly investigate the gradient estimates for the Lame´ system withpartiallyinfinitecoefficientsindimensiond = 3,aphysicallyrelevantdimension. This paper is a continuation of [13], where the estimate for dimension d = 2, another physically relevant dimension, is established. We prove that (ǫ lnǫ ) 1 is an upper − | | boundoftheblowuprateofthestraintensorindimensionthree,thesameasthescalar equation case mentioned above. New difficulties need to be overcome, and a number of refined estimates, via appropriate iterations, are used in our proof. We also prove that ǫ 1 is an upper bound of the blow up rate of the strain tensor in dimensiond 4, − ≥ whichisalsothesameasthescalarequationcase. Notethatithasbeenprovedin[11] thattheseupperboundsindimensiond 3 are optimalinthescalarequationcase. ≥ We consider the Lame´ system in linear elasticity with piecewise constant coeffi- cients,which isstimulatedbythestudyofcompositemediawithcloselyspaced inter- facial boundaries. Let Ω R3 be a bounded open set with C2 boundary, and D and 1 ⊂ D are two disjointconvex open sets in Ω withC2,γ boundaries, 0 < γ < 1, which are 2 ǫ apart and faraway from∂Ω, that is, D ,D Ω, theprinciplecurvaturesof∂D ,∂D κ > 0, 1 2 ⊂ 1 2 ≥ 0 (1.2) ǫ := dist(D ,D ) > 0, dist(D D ,∂Ω) > κ > 0, 1 2 1 2 1 ∪ whereκ ,κ areconstantsindependentofǫ. WealsoassumethattheC2,γ normsof∂D 0 1 i are bounded by some constant independent of ǫ. This implies that each D contains a i ballofradius r forsomeconstantr > 0independentofǫ. Denote 0∗ 0∗ Ω := Ω D D . 1 2 \ ∪ Assume that Ω and D D are oeccupied, respectively, by two different isotropicand 1 2 ∪ homogeneous materials with different Lame´ constants (λ,µ) and (λ ,µ ). Then the 1 1 e elasticitytensorsfortheinclusionsandthebackgroundcanbewritten,respectively,as C1 and C0, with C1 = λ δ δ +µ (δ δ +δ δ ), ijkl 1 ij kl 1 ik jl il jk and C0 = λδ δ +µ(δ δ +δ δ ), (1.3) ijkl ij kl ik jl il jk 2 where i, j,k,l = 1,2,3 and δ is the Kronecker symbol: δ = 0 for i , j, δ = 1 for ij ij ij i = j. Letu = (u ,u ,u )T : Ω R3 denotethedisplacementfield. Foragivenvector 1 2 3 → valuedfunctionϕ, weconsiderthefollowingDirichletproblemfortheLame´ system χ C0 +χ C1 e(u) = 0, inΩ, ∇·(cid:18)(cid:16) Ω D1∪D2 (cid:17) (cid:19) (1.4) u = ϕ, e on∂Ω, whereχ is thecharacteristicfunctionof D R3, D ⊂ 1 e(u) = u+( u)T 2 ∇ ∇ (cid:16) (cid:17) isthestraintensor. Assumethat thestandard ellipticityconditionholdsfor(1.4), thatis, µ > 0, 3λ+2µ > 0; µ > 0, 3λ +2µ > 0. (1.5) 1 1 1 For ϕ H1(Ω;R3), it is well known that there exists a unique solution u H1(Ω;R3) ∈ ∈ oftheDirichletproblem(1.4),which isalso theminimizeroftheenergy functional 1 J [u] = χ C0 +χ C1 e(u),e(u) dx 1 2 ZΩ(cid:18)(cid:16) Ω D1∪D2 (cid:17) (cid:19) e on H1(Ω;R3) := u H1(Ω;R3) u ϕ H1(Ω;R3) . ϕ ∈ − ∈ 0 n (cid:12) o Moredetailscan befoundin theAppendixin [1(cid:12)3]. (cid:12) Introducethelinearspaceofrigiddisplacementin R3, Ψ := ψ C1(R3;R3) ψ+( ψ)T = 0 , ∈ ∇ ∇ (cid:26) (cid:27) (cid:12) (cid:12) equivalently, (cid:12) 1 0 0 x x 0 2 3 Ψ = span ψ1 = 0 , ψ2 = 1 , ψ3 = 0 , ψ4 = x , ψ5 = 0 , ψ6 = x . 1 3 If ξ H1(D;R3),e(0ξ) = 0 in D0, and D R13 is a conne−c0ted open set−, txh1en ξ is ali−nxe2ar ∈ ⊂ combination of ψα in D. If an element ξ in Ψ vanishes at three non-collinear points, { } then ξ 0,seeLemma6.1. ≡ For fixed λ and µ satisfying µ > 0 and 3λ+2µ > 0, denoting u the solutionof λ1,µ1 (1.4). Then,as provedintheAppendixin[13], u u in H1(Ω;R3) as min µ ,3λ +2µ , λ1,µ1 → { 1 1 1} → ∞ whereu isa H1(Ω;R3)solutionof u := C0e(u) = 0, in Ω, λ,µ L ∇· u = u , (cid:16) (cid:17) on ∂D ∂D , e((cid:12)(cid:12)(cid:12)+u)∂=u 0(cid:12)(cid:12)(cid:12)−, ψα = 0, iin=De11,∪21,∪Dα2,=21,2, ,6, (1.6) Ru∂D=i ϕ∂ν,0(cid:12)(cid:12)(cid:12)+ · on ∂Ω, ··· 3 where ∂u := C0e(u) ~n = λ( u)~n+µ u+( u)T ~n. ∂ν0(cid:12)(cid:12)+ (cid:16) (cid:17) ∇· (cid:16)∇ ∇ (cid:17) and ~n is the unit out(cid:12)(cid:12)er normal of D, i = 1,2. Here and throughout this paper the (cid:12) i subscript indicatesthelimitfromoutsideandinsidethedomain,respectively. Inthis ± paperwestudysolutionsof (1.6),aLame´ systemwithinfinitecoefficients in D D . 1 2 ∪ The existence, uniqueness and regularity of weak solutions of (1.6), as well as a variational formulation, can be found in the Appendix in [13]. In particular, the H1 weak solution is in C1(Ω;R3) C1(D D ;R3). The solution is also the unique 1 2 ∩ ∪ functionwhichhas theleastenergy inappropriatefunctionalspaces, characterized by e I [u] = minI [v], ∞ v ∞ ∈A where 1 I [v] := C(0)e(v),e(v) dx, ∞ 2 ZΩ (cid:16) (cid:17) and e := u H1(Ω;R3) e(u) = 0 in D D . A ∈ ϕ 1 ∪ 2 n (cid:12) o It is wellknown,see[38], that forany(cid:12)open setO and u,v C2(O), (cid:12) ∈ ∂u C0e(u),e(v) dx = u v+ v. (1.7) λ,µ ZO(cid:16) (cid:17) −ZO(cid:16)L (cid:17)· Z∂O ∂ν0(cid:12)(cid:12)+ · (cid:12) (cid:12) A calculationgives (cid:12) u = µ∆u +(λ+µ)∂ ( u), k = 1,2,3. (1.8) Lλ,µ k k xk ∇· (cid:16) (cid:17) Weassumethatforsomeδ > 0, 0 1 δ µ,3λ+2µ . (1.9) 0 ≤ ≤ δ 0 Since D and D are two strictlyconvex subdomainsof Ω, there exist two points P 1 2 1 ∈ ∂D and P ∂D suchthat 1 2 2 ∈ dist(P ,P ) = dist(∂D ,∂D ) = ǫ. (1.10) 1 2 1 2 Use P P to denote the line segment connecting P and P . Throughout the paper, 1 2 1 2 unlessotherwisestated,C denotesaconstant,whosevaluesmayvaryfromlinetoline, dependingonlyond,κ ,κ ,γ,δ ,andanupperboundoftheC2normof∂ΩandtheC2,γ 0 1 0 normsof∂D and ∂D , but notonǫ. Also,wecall aconstanthaving suchdependence 1 2 auniversalconstant. Themainresult ofthispaper isfordimensionthree. Theorem 1.1. Assume that Ω, D ,D , ǫ are defined in (1.2), λ and µ satisfy (1.9) for 1 2 some δ > 0, and ϕ C2(∂Ω;R3). Let u H1(Ω;R3) C1(Ω;R3) be the solution of 0 ∈ ∈ ∩ (1.6). Then for0 < ǫ < 1/2, wehave e C u ϕ , (1.11) k∇ kL∞(Ω;R3) ≤ ǫ lnǫ k kC2(∂Ω;R3) | | whereC is auniversalconstant. 4 Remark1.1. TheproofofTheorem1.1actuallygivesthefollowingstrongerestimates: C Cdist(x,P P ) u(x) + 1 2 ϕ , x Ω, (1.12) |∇ | ≤ (cid:18) lnǫ ǫ +dist2(x,P P ) ǫ +dist2(x,P P )(cid:19)k kC2(∂Ω;R3) ∈ 1 2 1 2 | | e (cid:0) (cid:1) and u(x) C ϕ , x D D . (1.13) C2(∂Ω;R3) 1 2 |∇ | ≤ k k ∈ ∪ Remark 1.2. The strict convexity assumption on ∂D and ∂D can be replaced by a 1 2 weaker relativestrict convexityassumption,see(3.5)inSection 3. Remark1.3. Hereϕ C2(∂Ω;R3)canbereplacedbyϕ H1/2(∂Ω;R3). Indeed,theH1 ∈ ∈ norm of the solutionu in Ω is bounded by a universalconstant. Then standard elliptic estimatesgiveauniversalboundofuinC2 normin x Ω κ1 < dist(x,∂Ω) < κ1 . We ∈ 4 2 applythetheoremin Ω′ := x ∈ Ω dist(x,∂Ω) > κ31n withϕ(cid:12)(cid:12)(cid:12)′ := u∂Ω′. o n (cid:12) o (cid:12) Remark 1.4. Since the blow up rat(cid:12)e of u for solutions of the(cid:12)scalar equation (1.1) when k = is known to reach the(cid:12)mag|∇nitu∞d|e (ǫ lnǫ ) 1 in dimen(cid:12)sion three, see [11], − ∞ | | estimate(1.11)isexpected tobeoptimal. Following arguments in the proof of Theorem 1.1, we establish the corresponding estimates for higher dimensions d 4. Let Ω Rd, d 4 be a bounded open set ≥ ⊂ ≥ with C2 boundary, and D and D are two disjoint convex open sets in Ω with C2,γ 1 2 boundaries,satisfying(1.2). Let C0 begivenby(1.3)withi, j,k,l = 1,2, ,d,where ··· λ and µsatisfy µ > 0, dλ+2µ > 0, and Ψ := ψ C1(Rd;Rd) ψ+( ψ)T = 0 (1.14) ∈ ∇ ∇ (cid:26) (cid:27) (cid:12) be the linear space of rigid displacement in R(cid:12)(cid:12)d. With e , ,e denoting the standard 1 d ··· basisofRd, e, x e x e 1 i d, 1 j < k d i j k k j − ≤ ≤ ≤ ≤ n (cid:12) o isa basisofΨ. DenotethebasisofΨ(cid:12)(cid:12) as ψα , α = 1,2, , d(d+1). Consider { } ··· 2 u := C0e(u) = 0, in Ω, λ,µ L ∇· u = u , (cid:16) (cid:17) on ∂D ∂D , e((cid:12)(cid:12)(cid:12)+u)∂=u 0(cid:12)(cid:12)(cid:12)−, ψα = 0, iin=De11,∪21,∪Dα2,=21,2, , d(d+1), (1.15) Then wehaveRu∂D=i ϕ∂ν,0(cid:12)(cid:12)(cid:12)+ · on ∂Ω. ··· 2 Theorem 1.2. Assume as above, and ϕ C2(∂Ω;Rd), d 4. Let u H1(Ω;Rd) ∈ ≥ ∈ ∩ C1(Ω;Rd)bethesolutionof (1.15). Then for0 < ǫ < 1/2,we have e C u ϕ , (1.16) k∇ kL∞(Ω;Rd) ≤ ǫ k kC2(∂Ω;Rd) whereC is auniversalconstant. 5 Remark 1.5. The proof of Theorem 1.2 actually gives the following stronger estimate in dimensiond 4: ≥ C ϕ , x Ω, ǫ +dist2(x,P P )k kC2(∂Ω;Rd) ∈ 1 2 |∇u(x)| ≤ e C ϕ , x D D . Wealso haveRemarks 1.2–k1.4kCa2c(∂cΩo;Rrdd)ingly. ∈ 1 ∪ 2 The rest of this paper is organized as follows. In Section 2, we first introduce a setup for the proof of Theorem 1.1. Then we state a proposition, Proposition 2.1, containing key estimates, and deduce Theorem 1.1 from the proposition. In Sections 3 and 4, we prove Proposition 2.1. The proof of Theorem 1.2 is given in Section 5. A linear algebra lemma, Lemma 6.2, used in the proof of Theorem 1.1, is given in Section 6. 2 Outline of the Proof of Theorem 1.1 TheproofofTheorem1.1makesuseofthefollowingdecomposition. Bythethirdline of (1.6), u is a linear combination of ψα in D and D , respectively. Since it is clear 1 2 { } that ξ = 0inΩandξ = 0on∂Ωimplythatξ = 0inΩ,wedecomposethesolution λ,µ L of (1.6), as in[13], as follows: e e e 6 Cαψα, in D , α=1 1 1 u = αPP=661C2αψα, 6 in D2, (2.1) αP=1C1αvα1 +αP=1C2αvα2 +v0, inΩe, wherevα C1(Ω;R3),i = 1,2,α = 1,2, ,6,andv C1(Ω;R3)arerespectivelythe i ∈ ··· 0 ∈ solutionof e vα = 0, inΩ, e Lλ,µ i vα = ψα, on ∂D, (2.2) vαi = 0, one∂Di ∂Ω, j , i, and i j ∪ v = 0, inΩ, λ,µ 0 L v = 0, on∂D ∂D , (2.3) v0 = ϕ, one∂Ω.1 ∪ 2 0 TheconstantsCα :=Cα(ǫ), i= 1,2,α = 1,2, ,6,are uniquelydeterminedbyu. i i ··· By thedecomposition(2.1), wewrite 3 3 2 6 u = Cα Cα vα + Cα (vα +vα)+ Cα vα + v , inΩ, (2.4) ∇ 1 − 2 ∇ 1 2∇ 1 2 i ∇ i ∇ 0 Xα=1 (cid:0) (cid:1) Xα=1 Xi=1 Xα=4 e 6 then 3 3 2 6 u Cα Cα vα + Cα (vα +vα) + Cα vα + v , inΩ. |∇ | ≤ 1 − 2 ∇ 1 2 ∇ 1 2 i ∇ i ∇ 0 Xα=1 (cid:12) (cid:12)(cid:12) (cid:12) Xα=1 (cid:12) (cid:12)(cid:12) (cid:12) Xi=1 Xα=4 (cid:12) (cid:12)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12) (cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12) (cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12) (cid:12)(cid:12) (cid:12)(cid:12) (2e.5) The proof of Theorem 1.1 can be reduced to the following proposition. Without loss ofgenerality,weonlyneedtoproveTheorem1.1for ϕ = 1,andforthegeneral C2(∂Ω) k k caseby consideringu/ ϕ if ϕ > 0. If ϕ = 0,then u = 0. k kC2(∂Ω) k kC2(∂Ω) ∂Ω (cid:12) Proposition2.1. UnderthehypothesesofTheorem1(cid:12)(cid:12).1,andthenormalization ϕ C2(∂Ω) = k k 1, let vα and v be the solution to (2.2) and (2.3), respectively. Then for 0 < ǫ < 1/2, i 0 we have v C; (2.6) ∇ 0 L∞(Ω) ≤ (cid:13)(cid:13)(cid:13)∇(vα1(cid:13)(cid:13)(cid:13)+veα2) L∞(Ω) ≤ C, α = 1,2,3; (2.7) (cid:13)(cid:13) (cid:13)(cid:13) C (cid:13) vα(x) (cid:13) e , i = 1,2, α = 1,2,3, x Ω; (2.8) ∇ i ≤ ǫ +dist2(x,P P ) ∈ (cid:12) (cid:12) 1 2 (cid:12)(cid:12) (cid:12)(cid:12) Cdist(x,P P ) e vα(x) 1 2 +C, i = 1,2, α = 4,5,6, x Ω; (2.9) ∇ i ≤ ǫ +dist2(x,P P ) ∈ (cid:12) (cid:12) 1 2 (cid:12) (cid:12) (cid:12) (cid:12) e and Cα C, i = 1,2, α = 1,2, ,6; (2.10) i ≤ ··· (cid:12) (cid:12) (cid:12) (cid:12) C (cid:12) C(cid:12)α Cα , α = 1,2,3. (2.11) 1 − 2 ≤ lnǫ (cid:12) (cid:12) | | (cid:12) (cid:12) Proofof Theorem 1.1 byus(cid:12)ingProp(cid:12)osition2.1. Clearly,weonlyneedtoprovethethe- orem underthenormalization ϕ = 1. C2(∂Ω) k k Since 0 C4 C5 i i estimate(1.13)follow∇suf=rom−−CC(2ii45.10−).0Ci6 C0i6, in Di, i = 1,2, By (2.5)and Proposition2.1, wehave 3 2 6 u(x) Cα Cα vα(x) + Cα vα +C ∇ ≤ 1 − 2 ∇ 1 i ∇ i (cid:12) (cid:12) Xα=1 (cid:12) (cid:12)(cid:12) (cid:12) Xi=1 Xα=4(cid:12) (cid:12)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12)(cid:12) (cid:12) (cid:12) (cid:12)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12)(cid:12) (cid:12) (cid:12) (cid:12)(cid:12) (cid:12) C Cdist(x,P P ) + 1 2 . (2.12) ≤ lnǫ ǫ +dist2(x,P P ) ǫ +dist2(x,P P ) 1 2 1 2 | | (cid:16) (cid:17) Theorem 1.1followsimmediately. (cid:3) To complete this section, we recall some properties of the tensor C. For the isotropicelasticmaterial,let C := (C ) = λδ δ +µ δ δ +δ δ , µ > 0, dλ+2µ > 0. (2.13) ijkl ij kl ik jl il jk (cid:16) (cid:16) (cid:17)(cid:17) 7 ThecomponentsC satisfythefollowingsymmetriccondition: ijkl C = C = C , i, j,k,l= 1,2, ,d. (2.14) ijkl klij klji ··· Wewillusethefollowingnotations: d d (CA) = C A , and (A,B) := A : B = A B , ij ijkl kl ij ij kX,l=1 iX,j=1 foreverypairofd dmatricesA = (A ),B = (B ). Bythesymmetriccondition(2.14), ij ij × wehave (CA,B) = (A,CB), (2.15) (CA,B) = (CAT,B) = (CA,C) = (CAT,C). Foran arbitrary d d real symmetricmatrixη = (η ), wehave ij × C η η = λη η +2µη η . ijkl kl ij ii kk kj kj It followsfrom (2.13)that Csatisfiestheellipticitycondition min 2µ,dλ+2µ η2 C η η max 2µ,dλ+2µ η2, (2.16) ijkl kl ij | | ≤ ≤ | | n o n o d where η2 = η2. In particular, | | i,j=1 ij P min 2µ,dλ+2µ A+AT 2 C A+AT , A+AT . (2.17) ≤ n o(cid:12) (cid:12) (cid:16) (cid:16) (cid:17) (cid:16) (cid:17)(cid:17) (cid:12) (cid:12) (cid:12) (cid:12) 3 Estimates of v , vα , and (vα + vα) 0 |∇ | |∇ i | |∇ 1 2 | We first fix notations. Use (x ,x ,x ) to denote a point in R3 and x = (x ,x ). By a 1 2 3 ′ 1 2 translationandrotationifnecessary,wemayassumewithoutlossofgeneralitythatthe points P and P in(1.10)satisfy 1 2 ǫ ǫ P = 0 , ∂D , and P = 0 , ∂D . 1 ′ 1 2 ′ 2 2 ∈ −2 ∈ (cid:18) (cid:19) (cid:18) (cid:19) Fix asmalluniversalconstantR, suchthattheportionof∂D and ∂D near P and P , 1 2 1 2 respectively,can berepresented by ǫ ǫ x = +h (x ), and x = +h (x ), for x < 2R. (3.1) 3 1 ′ 3 2 ′ ′ 2 −2 | | Then by the smoothness assumptionson ∂D and ∂D , the functions h (x ) and h (x ) 1 2 1 ′ 2 ′ are ofclassC2,γ(B (0 )), satisfying R ′ ǫ ǫ +h (x ) > +h (x ), for x < 2R, 1 ′ 2 ′ ′ 2 −2 | | h (0 ) = h (0 ) = 0, h (0 ) = h (0 ) = 0, (3.2) 1 ′ 2 ′ 1 ′ 2 ′ ∇ ∇ 8 2h (0 ) κ I, 2h (0 ) κ I, (3.3) 1 ′ 0 2 ′ 0 ∇ ≥ ∇ ≤ − and h + h C. (3.4) 1 C2,γ(B ) 2 C2,γ(B ) k k ′2R k k ′2R ≤ In particular, we only use a weaker relative strict convexity assumption of ∂D and 1 ∂D , thatis 2 h (x ) h (x ) κ x 2, if x < 2R. (3.5) 1 ′ 2 ′ 0 ′ ′ − ≥ | | | | For0 r 2R, denote ≤ ≤ ǫ ǫ Ω := (x ,x ) R3 +h (x ) < x < +h (x ), x < r . r ′ 3 2 ′ 3 1 ′ ′ ∈ − 2 2 | | (cid:26) (cid:27) (cid:12) (cid:12) For0 z < R, let (cid:12) ′ ≤ | | ǫ ǫ Ω (z ) := (x ,x ) R3 +h (x ) < x < +h (x ), x z < s . (3.6) s ′ ′ 3 2 ′ 3 1 ′ ′ ′ ∈ − 2 2 | − | (cid:26) (cid:27) (cid:12) b (cid:12)(cid:12) 3.1 Estimates of v , vα for α = 1,2,3, and (vα + vα) |∇ 0| |∇ i | |∇ 1 2 | Lemma 3.1. v + v C. (3.7) k 0kL∞(Ω) k∇ 0kL∞(Ω) ≤ vα +vα + (vαe+vα) e C, α = 1,2, ,6. (3.8) k 1 2kL∞(Ω) k∇ 1 2 kL∞(Ω) ≤ ··· The proof of Lemma 3e.1 is essentially theesame as in [13] for dimension two. We omitithere. By Lemma3.1, (2.6)and (2.7)is proved. To estimate (2.8), we introduce a scalar function u¯ C2(R3), such that u¯ = 1 on ∈ ∂D , u¯ = 0 on∂D ∂Ω, 1 2 ∪ x h (x )+ ǫ u¯(x) = 3 − 2 ′ 2 , in Ω , (3.9) ǫ +h (x ) h (x ) 2R 1 ′ 2 ′ − and u¯ C. (3.10) k kC2(R3\ΩR) ≤ Define u¯α = u¯ψα, α = 1,2,3, in Ω, (3.11) 1 then u¯α = vα on ∂Ω. e 1 1 Similarly,wedefine e u¯α = uψα, α = 1,2,3, in Ω, (3.12) 2 such that u¯α = vα on ∂Ω, where u is a scalar function ineC2(R3) satisfying u = 1 on 2 2 ∂D , u = 0 on∂D ∂Ω, 2 1 ∪ e x +h (x )+ ǫ u(x) = − 3 1 ′ 2 , in Ω , (3.13) ǫ +h (x ) h (x ) 2R 1 ′ 2 ′ − and u C. (3.14) k kC2(R3\ΩR) ≤ In orderto prove(2.8),it suffices toprovethefollowingproposition. 9 Proposition 3.2. Assume the above, let vα H1(Ω;R3) be the weak solution of (2.2) i ∈ with α = 1,2,3. Then fori = 1,2, α = 1,2,3, e (vα u¯α)2dx C; (3.15) ZΩ ∇ i − i ≤ (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) and e vα C, (3.16) (cid:13)∇ i (cid:13)L∞(Ω\ΩR) ≤ (cid:13)(cid:13) C (cid:13)(cid:13) e , x √ǫ, ′ (vα u¯α)(x) √ǫ | | ≤ x Ω . (3.17) Consequently, (cid:12)(cid:12)(cid:12)∇ i − i (cid:12)(cid:12)(cid:12) ≤ |Cx′|, √ǫ < |x′| ≤ R, ∀ ∈ R C vα(x) , x Ω , (3.18) ∇ i ≤ ǫ + x 2 ∀ ∈ R (cid:12) (cid:12) | ′| (cid:12) (cid:12) and (cid:12) (cid:12) C , x √ǫ, ′ vα(x) √ǫ | | ≤ (3.19) A direct calculationgi(cid:12)(cid:12)(cid:12)∇vexs′,iin v(cid:12)(cid:12)(cid:12)ie≤wo|fCx′(|3,.2)-√(3ǫ.5<),|xth′|a≤t R. C x C ∂ u¯(x) | k| , k = 1,2, ∂ u¯(x) , x Ω . (3.20) | xk | ≤ ǫ + x 2 | x3 | ≤ ǫ + x 2 ∈ R ′ ′ | | | | Thus C u¯α(x) , i = 1,2, α = 1,2,3, x Ω . (3.21) |∇ i | ≤ ǫ + x 2 ∈ R ′ | | Fork,l = 1,2, C C x ∂ u¯(x) , ∂ u¯(x) | ′| , ∂ u¯(x) = 0, x Ω . (3.22) | xkxl | ≤ ǫ + x 2 | xkx3 | ≤ (ǫ + x 2)2 x3x3 ∈ R ′ ′ | | | | For u¯α, defined by (3.11) and (3.12), making use of (1.8) and (3.22), we have, for i i = 1,2, α = 1,2,3, C C x u¯α(x) C ∂ u¯(x) + | ′| , x Ω . (3.23) Lλ,µ i ≤ xkxl ≤ ǫ + x 2 (ǫ + x 2)2 ∈ R (cid:12) (cid:12) kX+l<6(cid:12) (cid:12) | ′| | ′| (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) For z 2R, wealwaysuseδto denote ′ | | ≤ ǫ +h (z ) h (z ) δ := δ(z ) = 1 ′ − 2 ′ . (3.24) ′ 2 By (3.2)-(3.5), 1 ǫ + z 2 δ(z ) C ǫ + z 2 . (3.25) ′ ′ ′ C | | ≤ ≤ | | (cid:16) (cid:17) (cid:16) (cid:17) 10