You are on page 1of 12

An IntemationalJournal

Available online at

.c,..c. (~o,..cTELSEVIER



with applications

Computers and Mathematics with Applications 51 (2006) 805-816

N e w t o n % M e t h o d for t h e
Navier-Stokes Equations
with Finite-Element
Initial G u e s s of
Stokes E q u a t i o n s
Department of Mathematics
Kyungpook National University
Daegu, Korea
skim©knu, ac. kr

Department of Mathematics
Chonbuk National University
Chonju, Korea
yhlee©math, chonbuk,

Department of Mathematics
Chonnam National University
Kwangju, Korea
bcshin©j nu. ac. k r
is shown that finite element solutions of Stokes equations may be chosen as the
initial guess for the quadratic convergence of Newton's algorithm applied to Navier-Stokes equations
provided there are sufficiently small mesh size h and the moderate Reynold's number. We provide
also a mixed convergence analysis in terms of iterations and finite-error estimates of the initial guess
with a regularity estimate and error analysis for each Newton's step. (~) 2006 Elsevier Ltd. All rights
K e y w o r d s - - N a v i e r - S t o k e s equations, Stokes equations, Convergence, Finite element method,
Newton's method.

1. I N T R O D U C T I O N
In t h e course of solving n o n l i n e a r e q u a t i o n s like N a v i e r - S t o k e s e q u a t i o n s , one m a y e m p l o y N e w t o n
and N e w t o n - l i k e m e t h o d s c o m b i n i n g w i t h o t h e r m e t h o d s (see, for e x a m p l e , [1-3], etc.). It is well
k n o w n t h a t if t h e initial guess is chosen n e a r b y t h e e x a c t s o l u t i o n of t h e g i v e n n o n l i n e a r differential
e q u a t i o n s t h e q u a d r a t i c c o n v e r g e n c e is g u a r a n t e e d (see [4]). It is k n o w n t h a t t h e s o l u t i o n of Stokes
e q u a t i o n s m a y be chosen as t h e initial guess for N e w t o n ' s m e t h o d for N a v i e r - S t o k e s e q u a t i o n s
This work was supported by Korea Research Foundation Grant (KRF-2002-070-C00014).
0898-1221/06/$ - see front matter (~) 2006 Elsevier Ltd. All rights reserved.
doi: 10.1016/j.camwa. 2006.03.007

Typeset by .A~tS-TF_X

1). (1. v E V. In Section 4.z. stand for the Laplacian.) as follows: a(u. the number it is a viscous constant.u = 0. [4] and [5]). This is done by showing that if/A = (u. p) E Hk+l(f~) x Hm(f~) is the weak solution of (1.hll~ + lip -. the Newton's method for Navier-Stokes equations in [4] needs to be rewritten as an equivalent iterative method in a variational form. For this purpose. z.uo.w. As mentioned in earlier studies (see. . the convergence of Newton's sequence can be estimated in terms of mesh size and Reynold's number.v) =/flVu" V v + $ ( ( z . on 0 a . in t2. in f~. KIM et al.1). for all u. an initial guess should be chosen sufficiently close enough to the solution of (1. and divergence operators. 2.~ A u + (u. respectively (Au is the vector of components Au0.hll <~C (hmiiPllm + hkllullk+x + ).V ) u + Vp = f. and V.llull2). a regularity and finite-error estimates are shown for each step of Newton's sequence. then the error can be estimated as liB . In Section 3. we show that the Newton's sequence converges to the unique solution to the NavierStokes equations quadratically and finally provide an error analysis combining the quadratic convergence with the finite-element error of the initial guess and Reynold's number. w. In the implementations of Newton's algorithm. u 2 ) t and the mean-zero condition for the p r e s s u r e p as follows: . V)u + (u. -) and a linear functional (h. we review the quadratic convergence theory of Newton's method and rewrite the Newton's method in an equivalent iterative method. Consequently.1).h) is the finite element solution to Stokes equations.~ for the quadratic convergence of the sequence generated by Newton's method in the finite-element approaches.1) and the initial guess U (°) = (Uo. Hence it is better for us to know how far the finite-element solutions of Stokes equations are away from the solutions of Navier-Stokes equations (1. for example. gradient. The domain ft is a convex polygonal bounded domain with boundary 0fh For the quadratic convergence of Newton's method for Navier-Stokes equations (1. A similar analysis was done in [6] for compressible Navier-Stokes equations. u = 0.Po.2). for a sufficiently large viscosity it (or sufficiently small ~ := 1/it) regularity estimates and finite error analysis for the solution are also provided by each Newton step. V. V .1) have a branch of nonsingular solutions.h.806 s.D. In Section 2.V ) w ) v d f ~ .. for the implementing of Newton's method by the finite-element method we choose the initial guess as the finite-element solution of Stokes equations. The rest of this paper is arranged in the following way. N E W T O N ' S IN A WEAK ALGORITHM FORMULATION Let us define a bilinear form a(. where the symbols A.Po. (1. Therefore the purpose of this paper is to provide that the initial guess as the finite-element approximation solution to incompressible Stokes equations may be chosen if there are sufficiently small mesh size h and a moderate Reynold's number . (2. we assume that the incompressible Navier-Stokes equations (1. With estimation (1. Now interpreting Newton's method as an equivalent iterative method.2) where II" ]Is is the standard norm for the Sobolev space HS(~t) where s is a real number. which we consider now with zero boundary condition for the v e l o c i t y u = ( u l . f is a vector function.1) .1) a p df~ = O. one may use an initial guess as the finite-element approximations to Stokes equations.

8). E X. u. for all v e V. T h e n / 4 -. Conversely. (2. Un. v) for all Y E X. Define the following function spaces: X := V x Q. (A) There is A such that {(A. Un. l e t / 4 := (u. (~ .) and ]].3) and Y := V*.u.. (h. for all v C V and q C Q. we cast the Navier-Stokes equations (2.s) where A is a compact interval in (0. v) for all ~. (2.v)-(p. . This argument shows t h a t / 4 is the solution of (2. .8).. Then B(U. for all v • V. 297]).U) = U + T .(p. (V" u. for example.5) for all q E Q. VV~X. respectively. Now assume t h a t / 4 is the solution of problem (2.5) in the canonical form F(A.V. 0. V) = (g./4) e A x X if and only if (g./4) for (A.1) can be written as: find/4 C X such that. v) . ~)) = (g.( P n + I .O. V . c~). Then the Newton's algorithm in a weak form for (1.p.2) Let Lo2(f~) be the subspace of L2(fl) which consists of the functions of mean zero. v v • x. G(A.8) holds.U) = 0. respectively. e) = { v e x : Ilu . Thus (2.9) L e t / 4 be the unique solution to (1. V " V) = A(--(Un" V)Un + f . Then the weak formulation corresponding to (1.p) and ]2 := (v./4(A)). q) = 0.5) is defined as G : AxX > Y .6) are equivalent.4) where V* denotes the dual space of V. v ) . For the derivation of Newton's algorithm. v). if/4 solves (2.q)=(g.v)+(V. Q = Lo2(ft). v) = V)u + v). V):=a(u.O. let g be defined by (2. for all v E V and q E Q. a(u. p. (2.V) -. (2. (2. the Navier-Stokes equations (1.~+. and g = G(A. + tip .5).ql] < e}. where V = Hol(Ft) 2.6) with T and G given by (2./4) if and only if (2. a(u. (2. Under this Assumption (A). we will use the following assumption.q) = 0. v) = A(f. The L 2 inner product and norm are denoted by (. (2.1) and let an e-neighborhood of/4 be s ( u .7) and a Ca operator G for the incompressible Navier-Stokes equations (2.T g if and only if B(/4.1) have a unique solution/4 E X for a given f E Y (see [4. [4].vii.v).u.q).7) and (2.6) in which a linear solution operator T for the Stokes equations is defined as T : Y ----* X by U = T g for g E Y if and only if B(/4. Throughout this paper. A e A} is a branch of nonsingular solutions of equation (2.~+t) C X such that. v) = J n h . (2.~+l. With a little abuse of notation. I]. v d~2.5) and (2. by g = G(A.Newton ' s M e t h o d and 807 / .6).1o) T h e following convergence result for the sequence {/4n} by Newton's algorithm can be found in.5).1) reads as (see [4]): find b/~+l := (U~+l. (2.

.+++(u++>+ + v) +>u+.+) (2. KI~.v) _-_~ ((u(. ~) u. THEOREM 2.12) as the initial guess..12) ( V . e).12) to rewrite an iterative scheme (2. + b/(n).1) by Newton's method in the rest of this section.. Furthermore the convergence is quadratic (2. we have \~=i := L/(°) +/g(1) + .ull~ + Ilpn+l -PI[) ~ C ( l l u . (P.. v) (2./~) of G is bounded on all bounded subsets of A × X. e') in Newton's algorithm (2. q) = 0.-. . (~.2.+I.v). . First summing (2.0. where ~rt (2. q) -----O.-. V .12). solve for all v E V and q C Q o (o(+.p(J)) E X where j = 1. . First.9) determines a unique sequence {hin+i} C S(b/.])) of F(A. With the solution/4 (°) = (u(°).~.u(. and then adding (2. THEOREM 2.L/) /s Lipschitzcontinuous with respect to V in the ball S(/d.t et al. so that the first weak Fr~chet derivative DuG(A.~ v)= ~(.808 S . The solution Lln by the iterative scheme (2.. PROOF. with a nonnegative integer k... Then there exists e' with 0 < e' <_ e such that for each initial guess hi(o) in S(/.11) (llu=+~ .. . D .z(.-.13) from j = 1 to n + 1. One can see that the weak second Fr~chet-derivative D~tG(A .v)_ (~(0.o.p (°)) of the Stokes equation (2... 2 . hi)(/~.1. e') that converges to the unique solution lA of (2.z(. such that for all v E V and q E Q.13) and show that (2.~. . hi)l~! is Lipschitzcontinuous (see also [4..ullx + Ilpn -Pll) 2- We will choose the initial guess as the weak solution of (2.v) \j=i (V" Un+l.. u (°). which will be shown as the solution to (1.v). o(u(.~ = (u./.).13) =0~ where. p.v)_(~(. z(k) = E u(0" i=0 Now let /g..365]).-~. proceed with the following iterative scheme: find hi(J) = (u(#). j=0 (2i5) j=0 .13) is in fact Newton's algorithm.~ = E u(J) and pn = E : P (j).9). Assume that the first Fr~chet-derivative DuF(A..1).16) .14) n u.13) is the solution generated by Newton's algorithm (2. . (2.

1 <_ m < l. where e(n) n+l = ( u ~ .1 -[-(Uj--1" ~7) U (j)] j=* (2.18) n+l j=l Hence. and the number n of Newton's iterations. q) = 0._. an error analysis will be provided in terms of mesh size h and Reynold's number A for the sequence of Navier-Stokes equations generated by Newton's method if the finite element solution to Stokes equations is chosen as the initial guess. V - v) = A((U~._lb. we will use a generic constant C. q ) = O.1) Vp E Q n Hm(f~). (3. Pn+l. I In this paper. (3. The Newton's algorithm in the finite-element approach (2. it is enough to show that e(n) = 0.h + f. VV) -t. ( V . 3.h.2) and further assume that the following uniform inf-sup condition holds on V h × ~h: for each qh E Qh. Ivhl __ Cllqhl[. A CONVERGENCE ANALYSIS In this section.h. there exists a Vh E Vn such that (qh. for all v E Vh and q E Qh. (3. the mesh size h of triangulations. ( V .h. u~+l. Let Th be a family of triangulations of f~ by a standard finite-element subdivisions of f~ into quasi-uniform triangles with h = max{diam(K) : K E Th}..(Un+ 1 • V)Un] . V) -.V) = • V)un + f.. l<k<l. Assume that the finite-element subspaces Vh C V and Qh C Q satisfy the following approximation properties: inf Ilu . (3. v) + (z17) v).h. V . V v E V n H k + l ( f ~ ) 2. \j=obJ + E a.3) with a positive constant C independent of h. vCVh inf l i p .h. which can be shown by using the following identity: ai = \i=O i=* \ j = O aj i=l a.Newton's Method 809 which can be written as the Newton's iteration form ( r U n + l . Un.(Pn+l.4) . v). otherwise we describe its dependence.~ ([(Un" V)Un+I -[.h. i=l This completes the assertion of theorem.E [( u(j)" V ) U j . u n + l .h) E Vh X Qh such that.h • V)Un. v) -- (Pn+l. vh) = IIq l?.vii 1 _< Chkllullk+l.ql[o < qE~h ChmllPllm.V ) U n + 1 -[-(Un+ 1 • •)U n -. which does not depend on the Reynold's number A. Un. a(un+l.9) can be stated as: find (un+l.

PO. i = 1. for all v • Vh. ( V . Vu • V.D.5) for all q • Qh. . the discrete Newton's algorithm can be written as: find Uh(j) such that ---. a(u0. ( V ' u 0 .h. .h) which is the finite-element approximation solution of Stokes problem for all v • Vu.h. u. the discrete solution to (3. p) E X such that a(u./u(J)h'P(hj)) • xh a [~u h(j). V . \ for all q • Qh. q) = 0. h . for example. there is a v • V such that V.2.(p. for all v • V. v) = A(f.7) P n .6) where j-1 z(J-1) ----E u(i)" i=0 Then. (3. Lemma 7 in [7].8) Ilvlil ~ Cll[ql[x.4 and the proof of Lemma 2. (3. see Corollary 2. and j=o j=o The following technical known results are necessary for a future use. PROOF. v ) . 2 . KIM et al. 0. For (i). v.4) can be written as Un. z h(j-1). V) -. (3. w. . (3. V . for z. (i) Eor every p • Q. LEMMA 3. (ii) is known as Poincare inequality. For (iii) see.v =p. h = ~ P ( h j). (ii) The following Poincar~ inequality holds: Cpllull 2 ~ IlVu[l 2. and w E Hl(fl) / n D i u v w d ~ _< C21]Ull l HVlll llwl[1.A ( ( u (j-l) • V ) u(J-1). u~J)~ = 0. (a. V). z. 1.v) -. where C1 is a positive constant.u) for all q • Q. i. v) = (h. 0. and (3.(p~). with the initial guess b/(°) : = (uo. Equivalently. (iii) For all u. v).9) where Cp is a positive constant depending only on ft.S. q ) = 0.~. v) .(P0. V .h. Zh(J-1).2 in [4]. (q.10) where Di denotes the derivative with respect to the i th coordinate variable in N 2 and C2 N a positive constant. (3. 810 where n = 0.12) . w E V~ where v ~ := z c v I Ilzlll _ ~C-~2~' cp is the Poincare constant . I Consider the following problem: to find (u.2.h=~-~U(hj) (3. . according to Theorem 2. V • v ) = .

PROOF. (3.2 and (3. I LEMMA 3.3.. the solution (u. Since (u. ~ • V.~llulllllvlll and I((u-V)w. w.v)l < c~ll-II~llvlll.14) where C3 is a constant dependent on the Poincare constant Cp.. 2 . z. We have the conclusion.w.~.y l l Cp H1 >.11) satisfies a priori estimate Ilulll <_ c6[Ihll_l and IIpll <.C61lhll-1.Newton's Method LEMMA 3 . w • V~.V)w. PROOF. u) + ((u.. A simple application of (3. Poincare inequality. u) = (h. The proof of (3.4.v)l < c. u) = (h. which implies 2 I]U[ll _< ~--~p]]h]l-1. Vu) + ~((u. where C3 depends on Cp. v . v u ) + a((~. u) = (Vu. w.2. (3.13) comes from Cauchy-Schwarz inequality. For any z . 811 For any u. Let h • V*. w • V~. (3. v)~. I LEMMA 3.17) .16) Then using Lemma 3. PROOF.. V)u. Let z. . v)] _< IlVullllVvll + Cpllulllllvlll <_ c411ulllllvlll. la(u.10) yields the conclusion. ~. and Lemma 3.p) • X of the weak form (3.15) where C6 > 1 is a constant dependent on Cp. u) = (Vu.9). we have a(u. z. we have a(u. yIlulll Vu • v . Vv)l + ~1((~' ' V)u + (u.11). w. V)z.3. u) Cp u2 -> cp]lull~. z. .p) is the weak solution of (3. it follows that Cp 2 yIlu[ll _< a(u. z.13) and C p 2 < a(u. (3.llvll. z. z. (3.y i l u l l l2. u).v)l < Cp~ltull.V)u. That is. From Lemma 3.. u). v • V and z. u) _< Ilhll-lltulll. Then we have I~(-. v)l _< I(Vu. the following holds: 1((7. w • V~.

(3.p) E (V x Q) f3 (Hk+l(f2) x Hm(f~)) be the solution of (2.0. Lemma 3. + AlluI]2) • (3. we have the following.p(O) <_ CAiiui]2.4 that U u(°) 1 + tp .812 S.hl] < C (hmllpl].V ) u in (3.h • (3. C5(1 + 2/Cv). These arguments yield conclusion (3.h 1 + p(O) --Po.19) Now again consider problems (2. o.5).0. By choosing z = w = 0 and h = / ~ ( u .uo. q) ---.h = (uO.hl]l + liP -.25) .5).5) and (Uo. V . z. the following known result in [4] for the initial guess is immediate.1. Ilvllx <-. By triangle inequality.~ ( ( u .12) from (2.15) with a chosen 6'6 = max{2/Cp.D.12). there is a v C V such that V-v--p.hlll + IIP-P0.24). The solution LtO.w.Po.h) be the solution of (3.~ + hkllullk+. PO.22) .11).hll -----C6~llfll-~.23) and (3.h <--C(hkilullk+ +hmllPllm) • (3.3 yields llpll= = ( p . 'tP" <_ Ch(llhll-1 + Hull1)___ c5 (1+ ~--~)Ilhll-1.5) and (3. Iluo. 1}.pO. v ) = . (3.PO.u (°) 1 + p .o. Note that Vh C V and Qh C Q.11). (3. for all v E V.h) E Y h X Qh is the solution of Stokes equations (3. v).hltl -4. for all q E Q.5) and (U0. v ) - (h.11) by choosing z = w = 0 and h = Xf. KIM et aL Since p c Q.hllX + (3. we have o (u_u. where C4 = max{l. Then Ilu .1.V)u.h) E Vh × Qh O[ the finite-element solution to (3. using (3. it can be shown from Lemma 3.v) < C4(11h[I-1 + Ilullx)llvlll. it follows that Ilu.18) By taking such v into (3.h.21) Now subtracting (2. Let (u.24) Combining (3.hll (3.hll <- u .v ) --a(u. C3}. which implies. according to (3. COROLLARY 3. p(O)) of Stokes equations (2.h.lip - po.18) and (3.h.5) in order to estimate Ilu .Clllpll.p) E V × Q is the solution of the Navier-Stokes equations (2.20) in the sense of the mesh size h and the Reynold's number ~ where (u.uo.h ~ + p(0) --Po.5). | Since Vh C H~(f~) 2 and Qh C L02(~) and Stokes equations are a particular type of (3.p ( ° ) + u(O) _ U0.8).17).23) and it can be also shown that the finite-error estimate for Stokes equations (see [4]) u (°) --U0.v) + .U0.5) satisfies IIP0. PROPOSITION 3. This estimate can be done by considering the exact solution (u (°).

11) to show the HI-regularity of the solution generated by Newton's method.p) is the solution of (2. the Newton's sequence converges to the unique solution. Let e be in (2. the solution/4(h°) = (U0.26) In order to show the quadratic convergence of the sequence L/n. the last inequality in (3. | REMARK.e.6). we discuss the Hi-regularity of the solution of the auxiliary problem (3.1) where 5 is a constant..26) that the Reynold's number A is more critical than the mesh size h for using the finite-element solution of Stokes equations as the initial guess. .. This is because one may solve Stokes equations in finite-element sense for sufficiently small mesh size h. Assume that (4.h) of (3. .5)..Newton's Method 813 The above Proposition 3.ll. < 7 and p(O) := II.1 tells us that the finite-element approximate solutions (uo. PROOF.h.h . (hkllullk+l + "XlIulI~ + h"'llPll-. Furthermore the convergence is quadratic with an initial guess/4h(°) qlU. PROOF.4) and (3.5) to solve Navier-Stokes equations if for an e' < 1. 2 .Pn. Then Newton's algorithm (3. e') where/4 = (u. .10).e') ¢3 (Vh × Qh) that converges to the unique solution b/ = (u. It can be noted from (3..p l l ) < C(llu..+l. Assume that (3.ulll + IIp~+l.25).h . i. Furthermore. These arguments may complete the proof.~llfll-1 < 1. THEOREM 3. .1.h) of Stokes equations (3. (3.1) holds.h) can be chosen as an initial guess for the Newton's method (3.3) . for all n = l .27) <_ Cr.1.p) E Hk+l(gt) × Hm(~) of (2.h.) 2"+' . THEOREM 4.Po. 4. (4.h[[ < Cp C~ A' (4. Pn+ l.h generated by Newton's method for the chosen initial guess. Since the hypothesis holds the assumptions of Theorem 2. and ][Un.h[[l+ [[Pn.h[[1EVA. A R E G U L A R I T Y ESTIMATE In this section.. from which u~°) := Iluo. From this observation we infer that for large Reynold's numbers it may not be possible to choose the initial guess as the solution of Stokes equations. where Cn is a positive constant dependent on the iteration number n.hll _< 7.p l l ) 2 (3.27) comes from (3.11) at each iteration step.5). + "~llull~) < ¢.o.10).4) determines a unique sequence {/4n+l.11) satisfies [[Un.26) holds for e' < e. Then the solution (Un.h ) } C S(/4.1.. note that the initial guess satisfies Iluo hill + Ilpo. Let us consider the regularity of the finite-element solution (3.5) belongs to S(/4. Let 7 := c6..Po. we will provide the validity of the assumptions of Theorem 2.1 found in [4]. The proof will be done by mathematical induction for the solution of (3.2) where C v is the Poincare constant and C2 is appeared in (3. Assume that A satisfies o < ~< ~ < 1 c6 (llfll-1 + 2c2/cp)' (4..Xllftl-1. .ll < c6.ull~ + IIp.h . there are a mesh size h and A such that c (hmllPllm + hkllullk+. At first.h} = { (Un+ Lh.h..

4).h = y ~ .814 S. T k k=l -. KIM et al.7) Note that from (4.(.z. u h(o) . since z :----Uj.1) u(:) <_ ~o (u~°) • v ) u ~ °) -. 06 1-3' 08 1 - < O6Xllfll-. Hence. J ~.~ Cp ! 2C2 A" At the jth iteration process in (3.1 - _ Ilfll-.h[] < - p(1) jr_ p(O) < C6 ~ .26'2 ~" (4.V)U(hj)Then (3. < ----~ 1 C6 i=0 (4.hll < 1-~7" (4.1)..~ and OO ][Pl.D.15) implies ulj+.6) and (3. Then. 08 . for the (j + 1) th iteration process. one m a y have from (3. : = uj.11). w i t h z = W = U o . 1< 0 : similarly we have p l j+l) _< C622j+x .6) i=0 and by the same way c6 I]Pj. (4../(llfll-.hll --< ~J u2 ) 1 ___C6 ~ ~2. consider (3.4) Since C6 > 1 and 7 < 1 we have OO Ilu.-.15) and (4. ) u h(o) in (3. i=0 For the last step of mathematical induction. v. h a n d h = that . assume that we have u(/) 1 -< C6"Y and p(h"~) <_ C6"Y.hlll < - U(1) 1 -~ u(O) 1 < C6~ ~ k Cp 1 - 2C2 A' k=l so that Ul.h = Y~i=O u(2) it follows that Ilzlll = tlu~.5) Since ~ < 1. _< ~o u (°~ '<~_~o~' and Note that from (4. cpl + 2o2ob -~) .6). h E V.~ ) e v~.11) with h = --A(U(hJ)..

~+l . from (3.Pn+l) E V × by Newton's method of (2.~.~ 7 2 ~ = C 6 1 7 i=o i=o < -3' - C6 1--3' and j+l c~ "~ [[Pj+l. from (4. COROLLARY 4. ~ u~ ) 1 _ < C 6 ~ . .p ( ° ) ~ hh j satisfies (4.1) are satisfied.h.h[] "~ C i n f {[[u.3) and (4.Pn+l. and constants which depend on the domain ~.Newton's Method 815 Thus. then the finite-element approximation problem (3.7) we conclude that j+l oo ]Iuj+LhNI_<~--~.~+x.hH <_ C (h k Hu. and the coercivity of a(.. where C is a positive constant. Assume that the hypotheses of Theorem 4.8) where the infimum is taken over all Vh 6 V h and qh 6 Qh and C is a positive constant independent of vl~ and qty. T h e above theorem tells that the solution (Un.h} stay always in the set Vx and the upper bound of Reynold's number does depend on f.h[[1 "~-][Pj+l.~).Pn+l. . PROOF.hl[1 + HPn+I -. Pn+l -. Hence.h. (4.h[[1 + liP. the Poincare constant Cp.4) is implied Uj+l.h.2.+1 --Pn+l.Vh[[1 + ][P~+I -. THEOREM[[ ~ Z p(h/) ~ C6 ~ ')'21 : C6 1 3' i=0 i=0 < 66 .~+l -un+l.3).11) satisfies the following: Ilu.h[[ ~ 2c~ 1~ c~ -.1.2) with Theorem 4. p .1) and (3.Un+l. Then the finite-element approximate solution (u. It comes from (3.11) has at least one solution.1 hold and suppose that the solution of the weak form (3.. the uniform inf-sup condition (3.~+lHk+l + h m []Pn+l[[.3).) because of Theorem 4. the weak solution (Un+l.h) E V h X Qh of (3. and ][Uj+l. Now one can easily verify t h a t the following error estimate for the finite-element approximate solution generated by Newton's method (3. 2 .Pn+l...11) can be derived by using the orthogonality property of the error Un+l -.1.h. We note that if (3. for a certain range of Reynold's number A provided the initial guess ~(u(°).+l. + l ) ~ (V × Q) n ( g k + l ( a ) × H m ( a ) ) . we omit its proof.qh[[}.. .1. . Hence an error estimate can be summarized as follows. . We can now show the finite-element discretization error for the solution generated by Newton's method.C 2 )~ I REMARK.Pn.Un+l.h) E V h x Qh of (3.1-3' In particular. . + l .h.h C VA.4) by Newton's method is in ( u . In this case {u.h) generated by Newton's method is always bounded for n = 1.9) and its approximate solution (U. Under the assumption of Theorem 4.4) by finite-element methods satisfy the following error estimate: [[Un+l -.

5. 3. Z. Finite Elemen~ Methods for Navier-Stokes Equations: Theory and Algorithms. 6.D. (2001). Numer. A multigrid preconditioned Newton-Krylov Method. 7. 36.A. A multigrid-preconditioned Newton-Krylov Method for the incompressible Navier-Stokes equations.A. (1980). 23 (2). KIM et al. Knoll and V. Part I: Branches of nonsingular solutions. Analysis of velocity-flux least-squares principles for the Navier-Stokes equations. 262-267. 4. Raviart.B. (2000). Manteuffel and S.A. J. Raviart. 990-1009. 1-25. Rappaz and P. S. 2. Cai. 691-702. (1998). 21 (2).A. F. Comput. 511-524. Comput. . Comput. Numer.F. M. 163. J. Finite-dimensional approximation of nonsingular problems.A. (2003).A. McCormick. Mousseau. 398-418. New York.D. REFERENCES 1.. Sci. (1986). D. D. V. Phys. T. Sci. S I A M J. Numer. Bochev.E. Part I. Springer-Verlag.816 S. Tocci. Pernice and M. Cirault and P. Kim and Y. On Newton-Krylov multigrid methods for the incompressible Navier-Stokes equations. (1999). Lee. Methods P. Rider. Knoll and W. S I A M J. Math.H. P. S I A M J.D. Anal. Error estimate and regularity for the compressible Navier-Stokes equations by Newton's method. 35. Brezzi.D.