Applied Mathematics and Computation 168 (2005) 1004–1019 www.elsevier.com/locate/amc Non-conforming finite element discretisations of a class of non-Newtonian flows A. Agouzal a a,* , K. Allali b, K. Najib c U.M.R. 5585, Equipe de Mathe´matiques Applique´es de Lyon, Universite´ de Lyon 1, Baˆt. 101, 69622 Villeurbanne cedex, France b De´partement de Mathe´matiques, Faculte´ des Sciences et Techniques, B.P. 146, Route de Rabat, 20650 Mohammedia, Morocco c Ecole Nationale de lIndustrie Mine´rale, B.P. 753, Agdal, Rabat, Morocco Abstract We study an approximation for a class of non-Newtonian model, based on the general Ladyzhenskaya law for continuous and non-linear kinematic viscosity functions of the fluid. We introduce a non-conforming finite element approximation of non-linear Stokes problem; we give existence, uniqueness results and optimal error estimates under minimal regularity assumptions. Finally, we introduce a post-processing to obtain conforming approximation of stress, which allow to derive various equivalence with two or three field mixed dual approximations of non-linear Stokes problem. 2004 Elsevier Inc. All rights reserved. Keywords: Non-conforming finite elements; Mixed finite elements; Non-Newtonian model * Corresponding author. E-mail address: [email protected] (A. Agouzal). 0096-3003/$ - see front matter 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.amc.2004.09.042 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 1005 1. Introduction In [9], a model for the motion of ideal incompressible viscous flow has been proposed. Finite element approximation of this model has been studied, in the case of conforming finite elements methods discretisation in primal formulation [5], and more recently in dual formulation in [6]. In this paper, we propose and study the non-conforming finite element approximation method for such model. We obtain optimal a priori error estimate in the quasi-norm for the velocity, and obtain by post-processing a conforming approximation of the stress as in the mixed dual formulation. The model we work with is as follows: consider the motion of stationary ideal incompressible viscous fluids in a bounded open connected, polyhedral domain X Rd with Lipschitz boundary C (d = 2 or 3), and let u denote the velocity field, p the pressure, and f the body force per unit mass. The model is given by 8 Find ðu; pÞ such that; > > > > > > < l0 div ðAðruÞruÞ l1 divðkrukr2 ruÞ þ rp ¼ f ; in X; > > div u ¼ 0; > > > > : u ¼ 0; in X; on C ¼ oX; 2 d where l0, l1 P 0 with l0 + l1 > 0, r 2 ]1, 2], f 2 (L (X)) , and A : Rd d ! Rþ is Lipschitz continuous function satisfying, there are positives constants c1 and c2 such that: for all a, b 2 Rd d , c1 ka bk2 6 ðAðaÞa AðbÞbÞ : ða bÞ ð1:1Þ kAðaÞa AðbÞbk 6 c2 ka bk; ð1:2Þ and (Colon denotes the scalar product in Rd d ). This kind of non-linear Stokes problem appears in the modeling of large class of non-Newtonian fluids. In particular, the Carreau law for viscoplastic flows (see, e.g., [10]) reads, with l1 = 0, 8a 2 R; 2 AðaÞ ¼ k 0 þ k 1 ð1 þ kak Þ b2 2 with k0 P 0, k1 > 0 and b P 1. It is easy to check that the Carreau law satisfies (1.1) and (1.2) for all k0 > 0 and b 2 [1, 2]. In particular, with b = 2 we recover the usual linear Stokes model. Let us recall that this problem has unique weak solution ðu; pÞ 2 H 10 ðXÞ 2 r0 0 r L0 ðXÞ if l0 5 0 and ðu; pÞ 2 W 1;r 0 ðXÞ L ðXÞ with r ¼ r1, if l0 = 0. 1006 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 In the sequel, we denote by W s,p(X) and Ws,p(C), 0 6 s and 1 6 p 6 +1, the usual Sobolev spaces (see e.g., [1]), endowed with the norms jj Æ jjs,p,X and jj Æ jjs,p,C respectively. For a non-integer s we use the notations j Æ js,p,X and j Æ js,p,C, where ZZ p kD½s vðxÞ D½s vðyÞk if p < þ1; jvjps;p;X ¼ dx dy kx ykdþpr X X and jvjps;p;C ¼ ZZ kD½s vðxÞ D½s vðyÞk dx dy; kx ykd1þpr C C if p ¼ 1; p jvjs;þ1;X ¼ sup X X kD½s vðxÞ D½s vðyÞkp ; kx ykr [s] being the integer part of s and r = s [s]. Hs(X) is the usual space Ws,2(X) and H s0 ðXÞ the closure of DðXÞ in Hs(X). 2. Finite element approximation In order to state the precise form of our discrete problem, we specify the hypothesis on the class of finite elements spaces under questions. Let Th be a family of regular triangulation by triangles or tetrahedrons of X in the sense of Ciarlet [4]. We denoted by E the set of all edges (faces) of Th and by EI the set of all interior edges (faces) of Th . Let us introduce the following spaces: 1;p W 0 ðTh Þ :¼ v 2 Lp ðXÞ; 8T 2 Th ; vhjT 2 H p ðT Þ; 8e 2 E; Z ½v dc ¼ 0; 8e edgeðfaceÞ C e Z v dc ¼ 0 e with p 2 [1, 1[; and we set H 10 ðTh Þ ¼ W 1;2 ðTh Þ. d d V h ¼ fvh 2 ðH 10 ðTh ÞÞ ; 8T 2 Th ; vhjT 2 ðP 1 ðT ÞÞ g; M h ¼ fqh 2 L20 ðXÞ; 8T 2 Th ; qhjT 2 P 0 ðT Þg; and W h ¼ fr :¼ ðri;j Þi;j¼1;...;d 2 H ðdiv; XÞ; 8T 2 Th ; d d d r 2 ðRT 0 ðT ÞÞ :¼ ððP 0 ðT ÞÞ þ xP 0 ðT ÞÞ g: A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 We consider the following discrete problem (2.1): 8 Find ðuh ; ph Þ 2 V h M h such that; > > P R > r2 > > 8vh 2 V h ; ðl0 Aðruh Þ þ l1 kruh k Þruh rvh > T > < T 2Th R P R p div vh dx ¼ X fh vh dx; > T h > > T 2Th > > P R > > q div uh dx ¼ 0; : 8q 2 M h ; h where fh is defined by : 1 f h :¼ measd ðT Þ 8T 2 Th ; ð2:1Þ h T T 2Th 1007 Z f dx; on T : T It is clear that this problem has unique solution (uh, ph) 2 Vh · Mh (see e.g. [3,8]), moreover div uh ¼ 0 8T 2 Th ; ð2:2Þ in T : In the sequel, for abbreviation, we frequently write ( )12 X 2 j j1;h ¼ l0 j j1;T T 2Th and ( 8q 21; 1½; j j1;q;h ¼ X )1q j q j1;q;T T 2Th and we introduce the following form defined by d 2 8ðu; vÞ 2 ððW 1;q 0 ðTh ÞÞ Þ ; X Z r2 ðl0 AðruÞ þ l1 kruk Þru rv dx aðu; vÞ :¼ T 2Th T with q = 2 if l0 5 0 and q = r otherwise. In the sequel, we denoted by C, C0, C1, . . . various positive generic constants not dependent of fhT gT 2 Th and not necessarily the same in any place. d Let P : ðW 1;1 0 ðTh ÞÞ ! V h the linear operator defined by: Z d 8v 2 ðW 1;1 ðT ÞÞ ; 8E 2 E; ðPv vÞ dr ¼ 0: h 0 E First, we have d Lemma 2.1. The linear operator P satisfies, for all v 2 ðW 1;1 0 ðTh ÞÞ and for all T 2 Th : 1008 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 Z 8sh 2 ðP 0 ðT ÞÞd d ; sh rðv PvÞ dx ¼ 0; T Z 8qh 2 P 0 ðT Þ qh div ðv PvÞ dx ¼ 0: T Moreover, if v 2 W 1;q ðTh Þ with q 2 ]1, 1[, then ( ) 10 q X Z X q0 d q0 q0 8R 2 ðL ðXÞÞ ; Rðv PvÞ dx 6 Cjvj1;q;h hT kRk0;q0 T ; T 2T T T 2T h 0 where q ¼ h q . q1 d d·d Proof. First, let v 2 ðW 1;1 , by elementwise integration 0 ðXÞÞ and sh 2 (P0(T)) by part, we infer Z Z Z sh rðv PvÞ dx ¼ ðv PvÞ div sh dx þ ðv PvÞsh n dr T T oT Z ¼ ðv PvÞsh n dr ¼ 0: oT Again, by elementwise integration by part, we infer Z 8qh 2 P 0 ðT Þ; qh div ðv PvÞ dx T Z Z ¼ rqh ðv PvÞ dx þ ðv PvÞ nT dr ¼ 0: oT T q0 1,q q . Since Let q 2 ]1, 1[, v 2 W (X) and R 2 L ðXÞ with q0 ¼ q1 kv Pvk0;q;T 6 ChT jvj1;q;T ; we have ( ) 10 q X Z X q0 q0 Rðv PvÞ dx 6 Cjvj1;q;h hT kRk0;q0 T : T 2T T T 2T h h Lemma 2.2. We have the generalise Green formulae, for all q 2 [1, 1[: X Z d 8v 2 ðW 1;q ðT ÞÞ ; 8r 2 W ; ðrh re þ e div rh Þ dx ¼ 0: h h h 0 T 2Th d T Proof. Let ðv; rh Þ 2 ðW 1;q 0 ðTh ÞÞ W h , since Z d 8E 2 EI ; ½vdc ¼ 0 and rh nE 2 ðP 0 ðEÞÞ ; E A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 we have X Z T 2Th X Z ðrh re þ e div rh Þ dx ¼ T T 2Th 1009 r nT v dc oT Z 1X ¼ r nE ½v dc ¼ 0: 2 E2EI E We have also (see e.g., [10]). Lemma 2.3. For all s0, s1, r 2 (Lr(X))d ·d, we have Z ðks0 kr2 s0 ks1 kr2 s1 Þ r dx 6 Ckrk0;r;X X Z 12 r2 fks0 k0;r;X þ ks1 k0;r;X g 2 ðks0 k þ ks0 s1 kÞr2 ks0 s1 k2 dx X and Z r r2 2 ks0 s1 k1;r;X þ ðks0 k þ ks0 s1 kÞ ks0 s1 k dx X Z r2 r2 6 C ðks0 k s0 ks1 k s1 Þ ðs0 s1 Þ dx: X One of our main result in this section is the following. Theorem 2.1. Let (u, p) the weak solution of the model problem and (uh, ph) 2 Vh · Mh the solution of the discrete problem. If l0 5 0, then we have X Z r2 2 l1 ðkruk þ krðu uh ÞkÞ krðu uh Þk dx T 2Th T X Z r 2 kru ruh k dx þ ju uh j1;h þ l1 6C T 2Th ( X T ) h2T kf k20;T T 2Th and ( kp 2 ph k0;X þ inf rh 2W h ;div rh ¼fh 6 C l0 þ l1 kruk0;r;X þ ( C X rh k20;X kr X r kruh k0;r;T T 2Th 1r r2 ) h2T kf k20;T T 2Th with r :¼ l0 AðruÞru þ l1 kruk r2 þ inf rh 2W h ;div rh ¼fh ru pId d . kr rh k20;X ) 1010 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 Proof. We set e = u uh. Recall that X Z l1 ðkruk þ krðu uh ÞkÞr2 krðu uh Þk2 dx T 2Th T þ l1 X Z T 2Th kru ruh kr dx þ ju uh j21;h T 6 Cðaðu; eÞ aðuh ; eÞÞ ¼ C aðu; eÞ Z Z fh e dx þ fh e dx aðuh ; eÞ ¼ I þ II: X X On one hand, since 8T 2 Th , div e = 0 on T, and X Z 8rh 2 W h ; ðrh re þ e div rh Þ dx ¼ 0; T T 2Th we have, "rh 2 Wh, div rh = fh: Z Z X Z aðu; eÞ fh e ¼ aðu; eÞ p div e dx þ e div rh dx X T 2Th X Z ¼ T 2Th X T ðr rh Þ re dx; X which implies that jIj 6 jej1;h inf rh 2W h ;div rh ¼fh kr rh k0;X : On the other hand Z Z Z aðuh ; eÞ fh e dx ¼ aðuh ; PeÞ fh e ¼ fh ðe PeÞ dx; X X X where Pe is the non-conforming interpolation operator. This last equality implies !12 X 2 jIIj 6 Cjej1;h h2T kf k0;T : T 2Th Using the last inequalities and Young inequality, we obtain X Z r2 2 l1 ðkruk þ krðu uh ÞkÞ krðu uh Þk dx T 2Th T þ l1 X Z T 2Th 6C X 2 r kru ruh k dx þ ju uh j1;h T 2 h2 kf k0;T T 2Th T þ inf rh 2W h ;div rh ¼fh kr 2 rh k0;X : A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 Finally, since R X kp ph k0;X 6 C X v2ðH 10 ðXÞÞd and Z ðp ph Þ div v ¼ X Z 1011 ðp ph Þ div v dx jvj1;X p div v X X Z T 2Th ph div Pv dx T Z fv dx aðuh ; PvÞ þ fh Pv dx X X Z Z ¼ aðu; vÞ aðuh ; vÞ þ fh ðPv vÞ dx þ ðfh f Þv dx ¼ aðu; vÞ Z X X and using the fact that (see Lemma 2.3) 8 > < XZ jaðu;vÞ aðuh ;vÞj 6 Cjvj1;X ðkruk > : T 2Th T þkrðu uh ÞkÞ r2 !12 2 krðu uh Þk dx 9 !1r 1r2 2 > = r l1 @kruk0;r;X þ kruh k0;r;T A þ ju uh j1;h ; > ; T 2Th 0 we have 8 > < X 9 !1r 1r2 > = X r kruh k0;r;T A > ; T 2Th ) 0 kp ph k0;X 6 C l0 þ l1 @kruk0;r;X þ > : ( X 2 h2T kf k0;T þ inf 2 T 2Th rh 2W h ;div rh ¼fh 2 kr rh k0;X : Using the similar arguments, we have also Theorem 2.2. Let (u, p) the weak solution of the model problem and (uh, ph) 2 Vh · Mh the solution of the discrete problem. If l0 = 0, then we have X Z r2 2 l1 ðkruk þ krðu uh ÞkÞ krðu uh Þk dx T 2Th T X Z r þ l1 kru ruh k dx T ( T 2Th ) X 0 r0 r0 r 6C hT kf k0;r0 ;T þ inf kr rh k0;r0 ;X T 2Th rh 2W h ;div rh ¼fh 1012 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 and X 1r r2 r kru kp ph k20;X 6 C kruk0;r;X þ h k0;r;T T 2Th ( ) X 0 0 0 r r C hrT kf k0;r0 ;T þ inf kr rh k0;r0 ;X rh 2W h ;div rh ¼fh T 2Th r . with r: = l1k$ukr2$u pIdd and r0 ¼ r1 Using Theorems 2.1 and 2.2, we have unified a priori error estimate. Theorem 2.3. Let (u, p) the weak solution of the model problem and (uh, ph) 2 Vh · Mh the solution of the discrete problem. We have X Z r2 2 ðkruk þ krðu uh ÞkÞ krðu uh Þk dx l1 T T 2Th þ l1 X Z T 2Th ( 6C X 2 r kru ruh k dx þ ju uh j1;h T 0 0 hqT kf kq0;q0 ;T T 2Th þ inf rh 2W h ;div rh ¼fh kr 0 rh kq0;q0 ;X ) and ( kp 2 ph k0;X X 1r r2 r 6 C l0 þ l1 kruk0;r;X þ kru k h 0;r;T T 2T h ( C X 0 q0 hqT kf k0;r0 ;T T 2Th with r :¼ l0 AðruÞru þ l1 kruk q and q0 ¼ q1 . r2 þ inf rh 2W h ;div rh ¼fh kr q0 rh k0;q0 ;X ) ru pId d , q = 2 if l0 5 0, q = r otherwise, We are now in position to state the error estimates for our problem, more precisely, using the classical approximation results for equilibrium finite element method [2], we have Theorem 2.4. Let (u, p) the weak solution of the model problem and (uh, ph) 2 Vh · Mh the solution of the discrete problem. If 0 r :¼ l0 AðruÞru þ l1 krukr2 ru pId d 2 ðW s;q ðXÞÞd d ; s 20; 1: A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 1013 Then we have X Z ðkruk þ krðu uh ÞkÞr2 krðu uh Þk2 dx l1 T 2Th T þ l1 X Z T 2Th ( X 6C 2 r kru ruh k dx þ ku uh k1;h T 0 q0 hqT kf k0;q0 ;T þh sq0 T 2Th q0 krkW s;q0 ðXÞ ) and kp 2 ph k0;X 1 q1 ( 6 Ckf k0;X X 0 q0 hqT kf k0;q0 ;T þh sq0 T 2Th q0 krkW s;q0 ðXÞÞ ) q . with q = 2 if l0 5 0 and q = r otherwise, and q0 ¼ q1 3. Post-processing and mixed finite elements methods First, we have the more general results. Theorem 3.1. Let (uh,ph) 2 Vh · Mh the solution of discrete problem. The tensor field rh defined by 8T 2 Th ; rh :¼ l0 Aðruh Þruh þ l1 kruh kr2 ruh fh ðx xg Þ ; d where xg is the barycenter of T and by using the notations: 8f ; g 2 Rd ; ðf gÞij ¼ fi gj ; i; j ¼ 1; . . . ; d; satisfies rh ph Id d 2 H ðdiv; XÞ and div ðrh ph Id d Þ ¼ fh Proof. Remark that 8T 2 Th ; d rh ph Id d 2 ðRT 0 ðT ÞÞ : Let e ¼ oT 1 \ oT 2 2 EI and vh 2 (Vh)d such that: Z 8f 2 E; vh dc ¼ dfe : f on X: on T ; 1014 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 Since 8T 2 Th ; Z 8vh 2 V h ; T fh ðx xg Þ rvh dx ¼ 0 d and using Green formulae, we have: Z ½ðrh ph IdÞ ne ¼ ½ðrh ph Id d Þ:nvh dc e ¼ (Z 2 X ) ðrh ph IdÞ rvh þ vh div ðrh ph IdÞ dx Ti i¼1 2 Z X ¼ l0 Aðruh Þ þ l1 kruh kr2 ruh Ti i¼1 ph Id d ¼ 2 Z X l0 Aðruh Þ þ l1 kruh kr2 ruh rvh Ti i¼1 Z fh ðx xg Þ fh vh dx rvh d Ti Z ph div vh dx Ti Z fh vh dx Ti ¼ 0; then rh phIdd 2 H(div,X). Finally, it is clear that div(rh phId) = fh. In the case where l1 = 0, we recovered the classical mixed dual formulations of stokes problem (see e.g., [6,7]). First let d Y h ¼ fvh 2 ðL2 ðXÞÞ ; 8T 2 Th ; d fvh gjT 2 ðP 0 ðT ÞÞ g and qh 2 L20 ðXÞ; X h ¼ fðsh ; qh Þ; and sh qh Id d 2 W h g: 8T 2 Th ; fqh gjT 2 P 0 ðT Þ; To obtain optimal a priori error estimate we need the following regularity result. Lemma 3.1. Let ðu; pÞ 2 ðH 1 ðXÞÞd L20 ðXÞ such that div u = 0 on X. Then r ¼ l0 ru pId d 2 ðH s ðXÞÞd d d u 2 ðH 1þs ðXÞÞ : is equivalent to p 2 H s ðXÞ and A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 1015 Proof. On one hand, it is clear that if p 2 H(s(X)) and u 2 (H1+s(X))d, then $u 2 (Hs(X))d·d which imply r ¼ l0 ru pId d 2 ðH s ðXÞÞd d : On the other hand, if r = l0$u pIdd 2 (Hs(X))d·d, then, since div u = 0, we have traceðrÞ ¼ l0 div u d p ¼ d p 2 H s ðXÞ; which implies d d l0 ru ¼ r þ pId d 2 ðH s ðXÞÞ and then p 2 H s ðXÞ and d u 2 ðH 1þs ðXÞÞ : We have the following a priori error estimates. Theorem 3.2. Let (uh, ph) the solution of discrete problem with l1 = 0 and A ¼ Id d . We set 8T 2 Th ; rh :¼ l0 ruh 8T 2 Th ; 1 ^uh ¼ measd ðT Þ and fh ðx xg Þ ; d Z on T 1 uh dx þ 2 l0 d measd ðT Þ T Z 2 fh kx xg k dx: T Then ððrh ; ph Þ; ^ uh Þ 2 X h Y h is the unique solution of the following problem: uh Þ 2 X h Y h ; such that; Find ððrh ; ph Þ; ^ Z Z 1 rh sh dx þ ^ uh div ðsh qh Id d Þ dx ¼ 0; 8ðsh ; qh Þ 2 X h ; l0 X X Z Z 8vh 2 Y h ; vh div ðrh ph Id d Þ dx ¼ fvh dx: X X s Moreover, if r = l0$u pIdd 2 (H (X)) d ·d , s 2 ]0,1], then kr ðrh ph IdÞk0;X þ kp ph k0;X þ ku ^ uh k0;X 6 Cfhs krkH s ðXÞ þ hkf k0;X g: Proof. Let us recall that rh phIdd 2 Xh and div(rh phIdd) = fh, which implies that Z Z vh div ðrh ph Id d Þ dx ¼ fvh dx: X X 1016 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 Let (sh, qh) 2 Xh, since div uh = 0 on T, for all T 2 Th and for all ðsh ; qh Þ 2 X h ; 8T 2 Th , Z Z fh ðx xg Þ 1 2 : sh ¼ 2 kx xg k dx d meas ðT Þ d T T d Z fh div ðsh ph Id d Þ dx; T we have 1 l0 Z Z 1 rh sh dx ¼ ruh sh dx fh ðx xg Þ:sh dx dl0 X X T 2T Z X 1 ruh ðsh qh Id d Þ dx fh ðx xg Þ ¼ dl0 X T 2T X ðsh qh Id d Þ dx X Z uh þ ¼ T 2Th div ðsh qh Id d Þ dx ¼ Z T 1 2 l0 d measd ðT Þ Z 2 kx xg k dx fh T ^ uh div ðsh qh Id d Þ dx: X Finally, it is clear that 8T 2 Th ; krh ph Id d rk0;T 6 Cfkp ph k0;T þ kru ruh k0;T þ hT kf k0;T g and 8T 2 Th ; ku ^ uh k0;T 6 CfkP 0 u uk0;T þ ku uh k0;T þ hT kf k0;T g; where 8T 2 Th ; 1 P 0u ¼ measd ðT Þ Z u dx: T Using Theorem 2.3, and the fact that u 2 (Hs(X))d, we obtain uh k0;X kr ðrh ph IdÞk0;X þ kp ph k0;X þ ku ^ 6 Cfhs krkH s ðXÞ þ hkf k0;X g: In the general cases, we have the following: Theorem 3.3. Let (uh, ph) the solution of discrete problem. We set 8T 2 Th ; rh :¼ l0 Aðruh Þruh þ l1 kruh k r2 ruh fh ðx xg Þ ; on T ; d A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 8T 2 Th ; uh ¼ 1 measd ðT Þ Z uh dx; th ¼ ruh ; 1017 on T : T d Then ððrh ; ph Þ; th ; uh Þ 2 X h ðY h Þ Y h is the unique solution of two-fold dual mixed formulation 8 d > Find ððrh ; ph Þ; th ; uh Þ 2 X h ðY h Þ Y h such that; > > R > < 8sh 2 ðY h Þd ; ðrh l0 Aðth Þth l1 kth kr2 th Þsh dx ¼ 0; X R > 8ðsh ; qh Þ 2 X h ; ðt sh þ uh div ðsh qh Id d Þ dx ¼ 0; > X h > > R R : 8vh 2 Y h ; X vh div ðrh ph Id d Þ dx ¼ X vh f dx: Proof. Let (uh, ph) 2 Vh · Mh the solution ððrh ; ph Þ; th ; uh Þ 2 X h ðY h Þd Y h defined by: 8T 2 Th ; rh :¼ l0 Aðruh Þruh þ l1 kruh k 8T 2 Th ; uh ¼ 1 measd ðT Þ Z uh dx; of r2 discrete ruh th ¼ ruh ; problem, fh ðx xg Þ ; on T ; d on T : T On one hand, it is clear that div ðrh ph Id d Þ ¼ fh and 8sh 2 ðY h Þd ; Z ðrh l0 Aðrth Þrth l1 kth kr2 th Þsh dx X ¼ X Z fh ðx xg Þ sh dx ¼ 0: d T 2Th T On the other hand, since div uh = 0 on T, for all T 2 Th , we have Z 8ðsh ; qh Þ 2 X h ; th : qh Id d dx ¼ 0; X which implies, using the generalise Green formula: Z ðth : sh þ uh div ðsh qh Id d ÞÞ dx 8ðsh ; qh Þ 2 X h ; X Z ¼ ðth : ðsh qh Id d Þ þ uh div ðsh qh Id d ÞÞ dx ZX ¼ ðth : ðsh qh Id d Þ þ uh div ðsh qh Id d ÞÞ dx ZX ¼ ðruh : ðsh qh Id d Þ þ uh div ðsh qh Id d ÞÞ dx ¼ 0: X and 1018 A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 We deduce that ððrh ; ph Þ; th ; uh Þ 2 X h ðY h Þd Y h is solution of two-fold dual mixed formulation 8 > Find ððrh ; ph Þ; th ; uh Þ 2 X h ðY h Þd Y h such that; > > R > < d r2 8sh 2 ðY h Þ ; ðrh l0 Aðth Þth l1 kth k th Þsh dx ¼ 0; X R > 8ðsh ; qh Þ 2 X h ; ðth sh þ uh div ðsh qh Id d Þ dx ¼ 0; > > > R X R : 8vh 2 Y h ; X vh div ðrh ph Id d Þ dx ¼ X vh f dx: To prove the uniqueness of the solution of this two fold dual mixed problem, let d ððrih ; pih Þ; tih ; uih Þ 2 X h ðY h Þ Y h ; i ¼ 1; 2; two solution of the two field dual mixed problem, and set rh ¼ r1h r2h ; ph ¼ p1h p2h ; uh ¼ u1h u2h and th ¼ t1h t2h : It is clear that div ðrh ph Id d Þ ¼ 0 then rh 2 ðY h Þ d on X; Z and th : rh dx ¼ 0; X which implies aðt1h ; th Þ aðt2h ; th Þ ¼ 0; then th = 0, and since rh 2 (Yh)d, we have Z rh : rh dx ¼ 0; X which is equivalent to rh = 0. Finally, since (see e.g., [3]), div(Xh) = Yh, we obtain uh ¼ 0, which implies that the two-fold mixed problem has at least one solution. h References [1] R.A. Adams, Sobolev Spaces, Academic Press, New York, 1975. [2] A. Agouzal, J.-M. Thomas, An extension theorem for equilibrium finite elements spaces Japan, J. Indust. Appl. Math. 13 (2) (1996) 257–266. [3] F. Brezzi, M. Fortin, Mixed and Hybrid Finite Element Methods, Springer, New York, 1991. [4] P.G. Ciarlet, The Finite Element Method for Elliptic Problems, North-Holland, Amsterdam, New York, Oxford, 1978. [5] Q. Du, M.D. Gunzburger, Finite-element approximations of Ladyzhenskaya model for stationary incompressible viscous flow, SIAM J. Numer. Anal. 27 (1990) 1–19. A. Agouzal et al. / Appl. Math. Comput. 168 (2005) 1004–1019 1019 [6] M. Farhloul, A.M. Zine, A mixed finite element method for a Ladyzhenskaya model, Comput. Meth. Appl. Mech. Eng. 191 (39-40) (2002) 4497–4510. [7] G.N. Gatica, M. Gonza´les, S. Meddahi, A low-order mixed finite element method for a class of casi-Newtonian Stokes flows Technical Report 02-01, Dept. Ingenierı´a Matema´tica, Universidad de Concepcio´n, Chile. [8] V. Girault, P.-A. Raviart, Finite Element Methods for Navier–Stokes Equations, Springer, Berlin, 1986. [9] O.A. Ladyxhenskaya, New equations for the description of the viscous incompressible fluids and solvabilty in the large of the boundary value problems for them, in: Boundary Value Problems of Mathematical Physics V, American Mathematical Society, Providence, RI, 1970. [10] D. Sandri, Sur lapproximation nume´rique des e´coulements quasi-newtoniens dont la viscosite´ suit la loi puissance ou la loi de Carreau, RAIRO Mode´lisation Mathe´matique et Analyse Nume´rique 27 (2) (1993) 131–155.
© Copyright 2024 ExpyDoc