some optimal error estimates of biharmonic problem using conforming finite element
TRANSCRIPT
Available online at www.sciencedirect.com
Applied Mathematics and Computation 194 (2007) 298–308
www.elsevier.com/locate/amc
Some optimal error estimates of biharmonic problemusing conforming finite element q
Yuan Li, Rong An *, Kaitai Li
Department of Mathematics, Xi’an Jiaotong University, Xi’an 710049, China
Abstract
In this paper, the W 2;p-error and Lp-error estimates of conforming finite element methods for biharmonic problem areestablished via weight function technique, where 2 6 p 6 +1. Finally, the numerical experiments are provided.� 2007 Elsevier Inc. All rights reserved.
Keywords: Optimal error estimates; Biharmonic problem; Finite element approximation
1. Introduction
About 30 yeas ago, many scholars began to investigate the max–norm error estimates of second orderPoisson equation
0096-3
doi:10
q Sup* Co
E-m
�Du ¼ f in X � R2
with Dirichlet boundary
u ¼ 0 on oX
when X was partitioned by regular mesh and irregular mesh, such as [1–11] and references therein. First,Nitsche in [1] gave
ku� uhk0;1 6 ch2j ln hj32juj2;1 ð1Þ
and
ku� uhk1;1 6 chj ln hkuj2;1 ð2Þ
by the method of weight function for linear finite element. But it is obvious that (1) and (2) are not of optimalorder. Subsequently, Nitsche in [2] improved the estimate (1) and showed
003/$ - see front matter � 2007 Elsevier Inc. All rights reserved.
.1016/j.amc.2007.04.040
ported by the Nature Science Foundation of China (Grant No. 50306019, No. 10571142, No. 10471110 and No. 10471109).rresponding author.ail address: [email protected] (R. An).
Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308 299
ku� uhk0;1 6 ch2j ln hkuj2;1: ð3Þ
Fried in [3] gave an example based on radial symmetry which implied that the estimate (3) maybe was of opti-mal order. Hence, it is difficulty to remove the factor jln hj in (3). However, the logarithmic factor in (2) can beremoved. Rannacher and Scott in [4] improved the estimate (2) and obtained some optimal error estimates
ku� uhk1;p 6 chkuk2;p 2 6 p 6 þ1;ku� uhk0;q 6 ch2kuk2;q 2 6 q < þ1:
(ð4Þ
In particular, for high order finite element, the factor jln hj in (1) and (3) disappeared and optimal max–normerror estimates were established
ku� uhk0;1 6 ch2kuk2;1 ð5Þ
and
ku� uhk1;1 6 chkuk2;1: ð6Þ
See [1] for more details. Moreover, if the domain X was partitioned by irregular mesh, Scott in [5] showed that(5) was also true.
Similar results have been obtained for second order nonlinear problems. Frehse and Rannacher in [6]obtained
ku� uhk0;1 6 ch2j ln hkuj2;1
for linear finite element approximation and in [7]ku� uhk0;1 6 chm
for high order m P 3 finite element approximation. If X � RN ;N P 3, they in [6] proved
ku� uhk0;1 6 ch2j ln hjN4þ1juj2;1:
We refer the reader to [8–12] for other similar results.However, to our best knowledge, it is vacant for max–norm error estimate of high order problem. There-
fore, in this paper, we deal with the biharmonic problem
D2u ¼ f ðxÞ in X � R2
with Dirichlet boundary
u ¼ ouon¼ 0 on oX:
We will extend the method in [4] to fourth order problem and show some optimal estimates
ku� uhk2;p 6 ch2kuk4;p; 2 6 p 6 þ1;ku� uhk0;q 6 ch4kuk4;q; 2 6 q < þ1
(ð7Þ
by conforming finite element approximation. If q = +1, we will show
ku� uhk0;1 6 ch3juj4;1; ð8Þ
which is not optimal. But the proof in regular triangular partition is much complicated than in quadrilateralpartition. In particular, if p ¼ q ¼ 2, it is well-known that the estimate (7) is true (cf. [13]).
This paper is organized as follows: in Section 2, we give some preliminary knowledge and define finite ele-ment subspace when X is partitioned by regular quadrilaterals; in Section 3, a stability theorem is established;in Section 4, we will prove the error estimates (7) and (8); in Section 5, the case when X is decomposed intoregular triangles is discussed; in Section 6, the numerical experiments are provided.
Throughout this paper, the symbol c always denotes a positive constant and is independent of h. Moreover,it maybe is different even in the same formulation.
300 Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308
2. Preliminary
Let W m;pðXÞ be the Sobolev space with norm kukm;p. Let C10 ðXÞ be the space of infinite times continuouslydifferentiable functions with compact support in X. W m;p
0 ðXÞ denotes the closure of C10 ðXÞ in W m;pðXÞ. Inparticular, if p = 2, we write H mðXÞ ¼ W m;pðXÞ and H m
0 ðXÞ ¼ W m;p0 ðXÞ.
In this paper, we will study the standard biharmonic problem
D2u ¼ f ðxÞ in X;
u ¼ ouon ¼ 0 on oX;
(ð9Þ
where u : X! R is unknown, X � R2 is a bounded and convex polygonal domain with Lipschitz boundary oX,n denotes the unite outer normal vector to oX.
Introduce the linear space
V ¼ H 20ðXÞ ¼ v 2 H 2ðXÞ; v ¼ ov
on¼ 0 on X
� �
with the normkvkV ¼ aðv; vÞ12 8v 2 V ;
where
aðu; vÞ ¼Z
XDuDv dx:
Given f 2 L2(X), the weak variational formulation associated with (9) is
Find u 2 V such that
aðu; vÞ ¼R
X fvdx 8v 2 V :
�ð10Þ
By the regularity theory of biharmonic problem (cf. [14]), there admits a unique solution u 2 V \ H 4ðXÞ.Let Th be a family of regular quadrilateral partition of X into quadrilaterals of diameter not great than h.
We define finite dimensional space
W h ¼ fu 2 C1ðXÞ; 8s 2 T h; u 2 Q3ðsÞg;
where Q3(s) is the space of all polynomials in the reference space (x,y) of the formPcijxiyj, where the sum
ranges over all integers i and j such that 0 6 i, j 6 3.Let V h ¼ W h \ V be the conforming finite element subspace. Then finite element approximation solution
uh 2 Vh of u is defined by
aðuh; vhÞ ¼Z
Xfvh dx 8vh 2 V h: ð11Þ
Moreover, there exists a unique solution uh 2 Vh.Define a Rietz projection operator P : V! Vh by
aðPu; vhÞ ¼ aðu; vhÞ 8vh 2 V h: ð12Þ
It is well-known thatkPukV 6 kukV :
By the theory of polynomial approximation, there exists two interpolant operators Ih : V! Vh andJh : V! Vh such that (cf. [15])
kv� IhvkV 6 ch2½v�4;2 8v 2 V \ H 4ðXÞ ð13Þ
and
kv� J hvkk;p 6 ch4�kkvk4;p 8v 2 V \ W 4;pðXÞ; k ¼ 0; 1; 2; 2 6 p 6 þ1; ð14Þ
Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308 301
where
½v�4;2 ¼Z
X
o4vox4
1
���� ����2 þ o4vox4
2
���� ����2 dx
!1=2
:
It is obvious that
½vh�4;2;s � 0 8vh 2 V h; s 2 T h: ð15Þ
3. Stability theorem
The proof of error estimate (7) bases on the following stability theorem:
Theorem 1. Given f 2 L2(X). If u 2 V \ W 2;pðXÞ is a unique solution of (10), then we have
kPuk2;p 6 ckuk2;p; ð16Þ
where 2 6 p < +1, c > 0 is independent of h.
Use some notations and parameters in [4]. For any z 2 X such that z is contained in some sz 2 Th, there exists adz 2 C10 ðszÞ such that
Zsz
dz dx ¼ 1; jDkdzj 6 ch�2�k; k ¼ 0; 1; 2; . . . ; ð17Þ
where dz is the Dirac function. By construction,
D/hðzÞ ¼ ðdz;D/hÞ 8/h 2 V h: ð18Þ
Define gz 2 V such thataðgz;/Þ ¼ ðdz;D/Þ 8/ 2 V ; ð19Þ
thenD2gz ¼ Ddz
holds in the sense of V 0 ¼ H�2ðXÞ and following estimate holds:
kgzkV 6 kdzkL2 :
Introduce the weight function
rzðxÞ ¼ ðjx� zj2 þ j2h2Þ12; j P 1;
where the parameter j will be chosen sufficiently large and independent of h. From now on, we drop the sub-script z. The weight function r satisfies (cf. [4])
jDkrj 6 cr1�k6 cðjhÞ1�k
; k ¼ 0; 1; 2; . . . :
Moreover, for sufficiently large j,
maxs2T h
maxx2s
rðxÞ=minx2s
rðxÞ� �
6 c
holds uniformly for z 2 X. Then from (13), we have
ZXrbjDðv� IhvÞj2 dx 6 ch4
Z 0rb j o
4vox4
1
j2 þ o4vox4
2
���� ����2 !
dx ð20Þ
for v 2 V \ ðQ
s2T hH 4ðsÞÞ, where
Z 0� � � dx ¼Xs2T h
Zs� � � dx:
302 Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308
Lemma 1. Let u 2 H10ðXÞ, then for each sufficiently large j, we have
ZXrb�2ju2jdx 6 c1
ZX
rbjDuj2 dx; ð21Þ
where c1 > 0 is some constant and b 2 R.
Proof. According to divergence formulation, one has
2
ZX
rb�2jujDijujxi dx ¼Z
Xrb�2Dijuj2xi dx ¼ �
ZXjuj2Dir
b�2xi dx� 2
ZXju2jrb�2 dx
¼ �ðb� 2ÞZ
Xrb�4jxj2juj2 dx� 2
ZXju2jrb�2 dx:
Thus
ZXrb�2juj2 dx 6Z
Xrb�2jukDukxjdxþ c
ZX
rb�4jxj2juj2 dx:
For sufficiently large j, one has
ZXrb�2juj2 dx 6 c1
ZX
rb�2jukDukxjdx 6 c1
ZX
rb�1jukDujdx 6 c1
ZX
rb�2juj2 dx� �1=2 Z
XrbjDuj2 dx
� �1=2
;
which gives (21). h
Lemma 2. For any given F 2 H 1ðXÞ, let v 2 V satisfy
D2v ¼ F in X;
v ¼ ovon ¼ 0 on oX;
(
then we have
ZXr�4�ajD2vj2 dx 6 cðjhÞ�6
ZX
r4�ajDF j2 dx; ð22Þ
where c > 0 is independent of j and h and 0 < a < 1.
Proof. We argue as in [4]. Using Holder inequality and Sobolev inequality, we have
ZXr�4�ajD2vj2 dx 6Z
Xr�ð4þaÞ=a dx
� �a
kD2vk22=1�a 6 cðjhÞa�4kDD2vk2
2=2�a:
Since
kDD2vk2
2�a2=2�a 6
ZX
r4�ajDD2vj2 dx� � 1
2�aZ
Xr�
4�a1�a dx
� �1�a2�a
:
Then
kDD2vk22=2�a 6 cðjhÞ�2�a
ZX
r4�ajDF j2 dx� �
;
which gives (22). h
Proof of Theorem 1. Let a 2 (0,1) and 2 6 p < +1. Setting / = Pu gives
DPuðzÞ ¼ aðg; PuÞ ¼ aðPg; PuÞ ¼ aðPg; uÞ ¼ ðDu; dÞ � aðu; g � PgÞ: ð23Þ
Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308 303
We estimate the second term of right hand side of (23). Using Holder inequality, one has
jðDu;Dðg � PgÞÞj 6Z
Xr�4�ajDujp dx
� �1pZ
Xr�4�a dx
� �p�22pZ
Xr4þajDðg � PgÞj2 dx
� �12
6 ch�ðp�2Þð2þaÞ
2p Mh
ZX
r�4�ajDujp dx� �1
p
;
where
Mh ¼Z
Xr4þajDðg � PgÞj2 dx
� �12
:
The first term of right hand side of (23) can be estimated by
ðDu; dÞ 6Z
sz
jDujp dx� �1
pZ
sz
jdjp
p�1
� �p�1p
6 ch�2p
Zsz
jDujp dx� �1
p
:
Hence
kDPukpp 6 ch�2
Z Zsz
jDujp dxdzþ cðh�2�aÞp�2
2 Mph
Z Zsz
r�4�ajDujp dxdz� �
6 ckDukpp þ h�
ð2þaÞp2 Mp
hkDukpp:
That is
kDPukp 6 c 1þ h�2�a
2 Mh
kDukp 2 6 p < þ1:
To obtain (16), it is enough that
maxz2X
ZX
r4þajDðg � PgÞj2 dx 6 ch2þa; ð24Þ
where c is independent of z and h. Denote w ¼ r4þaðg � PgÞ, then
ZXr4þajDðg � PgÞj2 dx ¼Z
XDðg � PgÞDðw� IhwÞdx�
ZXðg � PgÞDðg � PgÞDr4þa dx
� 2
ZXrr4þa � rðg � PgÞDðg � PgÞdx:
Using Holder inequality, Young’s inequality, (15) and (20), one has
ZXr4þajDðg � PgÞj2 dx 6Z
Xr�4�aDðw� IhwÞdxþ c
ZX
rajg � Pgj2 dx
6 ch4
Z 0r�4�a o
4wox4
1
���� ����2 þ o4w
ox42
���� ����2 !
dxþ cZ
Xrajg � Pgj2 dx:
Since
Z 0 o4wox41
���� ����2 þ o4wox4
2
���� ����2 dx 6Z 0jD2wj2 dx;
in view of Lemma 1, after a calculation, we obtain
ZXr4þajDðg � PgÞj2 dx 6 ch4
Z 0r4þajD2gj2 dxþ c
ZX
rajg � Pgj2 dx
þ cj�4
ZX
r4þajDðg � PgÞj2 dxþZ
Xrajg � Pgj2 dx
� �:
304 Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308
Then for sufficiently large j, we have
ZXr4þajDðg � PgÞj2 dx 6 ch4
Z 0r4þajD2gj2 dxþ c
ZX
rajg � Pgj2 dx
6 ch4
Zsz
r4þajDdj2 dxþ cZ
Xrajg � Pgj2 dx 6 ch2þa þ c
ZX
rajg � Pgj2 dx: ð25Þ
Consider the auxiliary problem
D2v ¼ raðg � PgÞ in X;
v ¼ ovon ¼ 0 on oX:
(ð26Þ
Then in view of Lemmas 1 and 2, we have
ZXr�4�ajD2vj2 dx 6 cðjhÞ�6
ZX
r4�ajDðraðg � PgÞÞj2 dx 6 cðjhÞ�6
ZX
r2þajg � Pgj2 þ r4þajDðg � PgÞj2 dx
6 cðjhÞ�4
ZX
r4þajDðg � PgÞj2 þ rajg � Pgj2 dx:
Multiplying (26) by g � Pg and integrating over X, we obtain
ZXrajg � Pgj2 dx ¼Z
XDðv� IhvÞDðg � PgÞdx 6
ZX
r4þajDðg � PgÞj2 dx� �1
2Z
Xr�4�ajDðv� IhvÞj2 dx
� �12
6 j�2
ZX
r4þajDðg � PgÞj2 dxþ cj2h4
ZX
r�4�ajD2vj2 dx
6 j�2
ZX
r4þajDðg � PgÞj2 dxþ cj�2
ZX
r4þajDðg � PgÞj2 þ rajg � Pgj2 dx� �
:
Then for sufficiently large j, we have
ZXrajg � Pgj2 dx 6 cj�2
ZX
r4þajDðg � PgÞj2 dx:
Substituting above into (25), we obtain, for sufficiently large j,
ZXr4þajDðg � PgÞj2 dx 6 ch2þa;
which gives (24). Hence we complete the proof of Theorem 1. h
4. Error estimates
By stability Theorem 1, we obtain some optimal error estimates.
Theorem 2. If u 2 V \ W 4;pðXÞ and uh 2 Vh satisfy (10) and (11), respectively, then the following error estimate
holds:
ku� uhk2;p 6 ch2kuk4;p;
ku� uhkp 6 ch4kuk4;p;
(
where c > 0 is independent of h and 2 6 p < +1.Proof. Since the proof of first estimate is similar with Corollary 7.1.12 in [15] and another is similar with Cor-ollary in [4], so we omit it. h
Theorem 3. If u 2 V \ W 4;1ðXÞ and uh 2 Vh satisfy (10) and (11), respectively, then we have following stability
estimate:
Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308 305
kPuk2;1 6 ckuk2;1
and error estimate
ku� uhk2;1 6 ch2kuk4;1;
where c > 0 is independent of h.
Proof. Similar with the argument of Theorem 1, we estimate the right hand side of (23). In fact, if p = +1, wehave
jðDu;Dðg � PgÞÞj 6 kDuk1Z
XjDðg � PgÞjdx
6 kDuk1Z
Xr4þajDðg � PgÞj2 dx
� �12Z
Xr�4�a dx
� �12
6 ch�2þa
2 MhkDuk1
and
jðDu; dÞj 6 kDuk1Z
Xjdjdx 6 ckDuk1:
Hence, by (23),
kDPuk1 6 c 1þ h�2�a
2 Mh
kDuk1:
The rest follows the proofs of Theorems 1 and 2. h
Next, we prove the estimate (8).
Lemma 3. For all vh 2 Vh, we have
kvhk0;1 6 ch�1kvhk0;2:
Proof. By construction, we have
vhðzÞ ¼ ðdz; vhÞ;
where the Dirac function dz is defined in Section 3, then we havekvhk0;1 6
Zsz
jdzkvhjdx 6Z
sz
jdzj2 dx� �1
2Z
Xjvhj2 dx
� �12
6 ch�1kvhk0;2: �
Theorem 4. If u 2 V \ W 4;1ðXÞ and uh 2 Vh satisfy (10) and (11), respectively, then we have
ku� uhk0;1 6 ch3kuk4;1;
where c > 0 is independent of h.
Proof. By Lemma 3, we have
ku� uhk0;1 6 ku� J huk0;1 þ kJ hu� uhk0;1 6 ch4kuk4;1 þ ch�1kJ hu� uhk0;2
6 ch4kuk4;1 þ ch�1ðku� J huk0;2 þ ku� uhk0;2Þ 6 ch4kuk4;1 þ ch3kuk4;2 6 ch3kuk4;1: �
5. Triangular partition
If the domain X is decomposed into regular triangles, Theorems 1–4 also hold. Define finite dimensionalspace
306 Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308
bW h ¼ fu 2 C1ðXÞ; 8s 2 T h; u 2 P 5ðsÞg;
where P5(s) is the space of all polynomials defined on s of degree less than or equal to 5. Let bV h ¼ bW h \ V bethe finite element subspace. In this case, the projection operator P is from V to bV h.
We assume that the following inverse inequality holds:
Z 0rbjD2vhj2 dx 6 ch�4ZX
rbjDvhj2 dx 8vh 2 bV h: ð27Þ
Lemma 4 [16, Theorem A.8]. Let X be a bounded domain of R2 with a Lipschitz continuous boundary. For each
integer k P 0 and real p > 0, there exists a positive constant C such that
kvkk;p 6 Cðkvkp þ ½v�k;pÞ 8v 2 W k;pðXÞ:
By Lemma 4, one hasku� IhukV 6 ch2ðkuk2 þ ½u�4;2Þ 8u 2 V \ H 4ðXÞ ð28Þ
or
ku� IhukV 6 ch2ðkukV þ ½u�4;2Þ 8u 2 V \ H 4ðXÞ; ð29Þ
where c is independent of h. Eqs. (28) (or (29)) plays the same role as (13). Moreover, the inequality (20)
becomes
ZXrbjDðv� IhvÞj2 dx 6 ch4
ZX
rbjvj2 dxþ ch4
Z 0rb o4v
ox41
���� ����2 þ o4vox4
2
���� ����2 !
dx ð30Þ
or
ZXrbjDðv� IhvÞj2 dx 6 ch4
ZX
rbjDvj2 dxþ ch4
Z 0rb j o
4vox4
1
j2 þ j o4v
ox42
j2� �
dx ð31Þ
for v 2 V \ ðQ
s2T hH 4ðsÞÞ.
Theorem 5. Given f 2 L2ðXÞ. If u 2 V \ W 2;pðXÞ and Pu 2 bV h satisfy (10) and (12), respectively, then we have
kPuk2;p 6 ckuk2;p;
where 2 6 p6 +1, c > 0 is independent of h.
Proof. From the argument of Theorems 1 and 3, it is enough to show (24). Moreover, according to (30), onehas
ZXr4þajDðg � PgÞj2 dx 6
ZX
r�4�aDðw� IhwÞdxþ cZ
Xrajg � Pgj2 dx
6 ch4
Z 0r�4�a o4w
ox41
���� ����2 þ o4wox4
2
���� ����2 !
dxþ ch4
ZX
r�4�ajwj2 dxþ cZ
Xrajg � Pgj2 dx:
Since
½wh�4;2;s 6¼ 0 8wh 2 V h;
then in terms of Lemma 1, one has
ZXr4þajDðg � PgÞj2 dx 6 ch4
Z 0r4þajD2gj2 dxþ ch4
Z 0r4þajD2Pgj2 dxþ c
ZX
rajg � Pgj2 dx
þ ch4
Zr4þajg � Pgj2 dxþ cj�4
Zr4þajDðg � PgÞj2 dxþ
Zrajg � Pgj2 dx
� �:
X X X
Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308 307
Then for sufficiently large j, we obtain
TableW 2;p-er
H
8163264
TableLp-erro
H
8163264
ZX
r4þajDðg � PgÞj2 dx 6 ch4
Z 0r4þajD2gj2 dxþ ch4
Z 0r4þajD2Pgj2 dxþ c
ZX
rajg � Pgj2 dx
6 ch4
Zsz
r4þajDdj2 dxþ cZ
Xr4þajDPgj2 dxþ c
ZX
rajg � Pgj2 dx
6 ch2þa þ cZ
sz
r4þajdj2 dxþ cZ
Xrajg � Pgj2 dx 6 ch2þa þ c
ZX
rajg � Pgj2 dx; ð32Þ
which is (25) in the proof of Theorem 1. Let v 2 V satisfy (26). According to (31), we also have
ZXrajg � Pgj2 dx 6Z
Xr4þajDðg � PgÞj2 dx
� �12Z
Xr�4�ajDðv� IhvÞj2 dx
� �12
6 j�2
ZX
r4þajDðg � PgÞj2 dxþ cj2h4
ZX
r�4�ajD2vj2 dxþ cj2h4
ZX
r�4�ajDvj2 dx
6 j�2
ZX
r4þajDðg � PgÞj2 dxþ cj�2
ZX
r4þajDðg � PgÞj2 dxþ cj�2
ZX
rajg � Pgj2 dx:
Hence, for sufficiently large j, we have
ZXrajg � Pgj2 dx 6 cj�2
ZX
r4þajDðg � PgÞj2 dx;
which together with (32) gives (24). Therefore, we complete the proof of Theorem 5. Finally, by Theorem 5,we obtain the error estimates (7) and (8). h
6. Numerical experiments
The numerical experiments in this section are designed to verity the error estimate obtained in Section 4.Here, we use Bogner–Fox–Schmidt element (Q3 Element).
Let X ¼ fðx; yÞ; 0 6 x 6 1; 0 6 y 6 1g and choose a true solution
u ¼ sinðpxÞ2 sinðpyÞ2:
Denote M and N be the division numbers at x-axis and y-axis, respectively. Following table displays somerelative errors in different mesh partitions, from which we can conclude the convergent order.Tables 1 and 2 display that the convergence rates of W 2;p and Lp, p = 2,4,8, errors are almost O(h2) andO(h4), which are consistent with Theorem 2.
1ror estimates under different norms when M = N = H
ku�uhkVkukV
Order ku�uhk2;4
kuk2;4
Orderku�uhk2;8
kuk2;8Order
7.900604 · 10�2 n 7.354391 · 10�2 n 6.911686 · 10�2 n1.987399 · 10�2 1.99 2.012550 · 10�2 1.87 2.148402 · 10�2 1.694.975050 · 10�3 2.00 5.044575 · 10�3 2.00 5.402413 · 10�3 1.991.244164 · 10�3 2.00 1.261903 · 10�3 2.00 1.352011 · 10�3 2.00
2r estimates under different norms when M = N = H
ku�uhk2
kuk2Order ku�uhk4
kuk4Order ku�uhk8
kuk8Order
7.074492 · 10�3 n 6.848307 · 10�3 n 7.128996 · 10�3 n4.568223 · 10�4 3.95 4.837706 · 10�4 3.82 5.557572 · 10�4 3.682.869772 · 10�5 3.99 3.037603 · 10�5 3.99 3.491115 · 10�5 3.991.795645 · 10�6 4.00 1.900341 · 10�6 4.00 2.183457 · 10�6 4.00
308 Y. Li et al. / Applied Mathematics and Computation 194 (2007) 298–308
References
[1] J.A. Nitsche, L1-convergence of finite element approximation, in: Second Conference on Finite Elements, Rennes, France, 1975.[2] J.A. Nitsche, L1-convergence of finite element approximations, in: I. Galligani, E. Magenes (Eds.), Mathematical Aspects of Finite
Element Methods, Lecture Notes in Mathematics, vol. 606, 1977, pp. 261–274.[3] I. Fried, On the optimality of the pointwise accuracy of the finite element solution, Int. J. Numer. Meth. Eng. 15 (1980) 451–456.[4] R. Rannacher, R. Scott, Some optimal error estimates for piecewise linear finite element approximations, Math. Comput. 38 (1982)
437–445.[5] R. Scott, Optimal L1 estimates for the finite element method on irregular meshes, Math. Comput. 30 (1976) 681–697.[6] J. Frehse, R. Rannacher, Asymptotic L1-error estimates for linear finite element approximations of quasilinear boundary value
problems, SIAM J. Numer. Anal. 15 (1978) 418–431.[7] J. Frehse, R. Rannacher, Optimal uniform convergence for the finite element approximation of a quasilinear elliptic boundary value
problem, in: Proc.US-Germany Symp., Formulations and Computational Algorithms in Finite Element Analysis, Mass. Inst. ofTech., Cambridge, MA, 1976.
[8] J. Frehse, Eine gleichmassige asymptotische Fehlerabshatzung zur Methode der finiten Elemente bei quasilinaren Randwertproble-men, Theory of Nonlinear Operator. Constructive Aspects, Tagungsband der Akademie der Wissenschaften, Berlin(DDR), 1976.
[9] C. Johnson, V. Thomee, Error estimates for a finite element approximation of a minimal surface, Math. Comput. 29 (1975) 343–349.[10] H.D. Mittelmann, On pointwise estimates for a finite element solution of nonlinear boundary value problem, Preprint No. 207,
Fachbereich Mathematik, T.H.Darmstadt, Germany, 1975.[11] R. Rannacher, Some asymptotic estimates for finite approximation of minimal surfaces, Rev. Francaise Automat. Informat.
Recherche Operationelle 11 (1997) 181–196.[12] Yingjun Jiang, Jinping Zeng, An L1-error estimate for finite element solution on nonlinear elliptic problem with a source term, Appl.
Math. Comput. 174 (2006) 134–149.[13] Zhangxing Chen, Finite Element Methods and Their Application, Spring-Verlag, 2005.[14] S. Agmon, A. Douglis, L. Nirenberg, Estimates near the boundary for solutions of elliptic partial differential equations satisfying
general boundary conditions. I, Comm. Pure Appl. Math. 12 (1959) 623–727.[15] S.C. Brenner, L.R. Scott, The Mathematical Theory of Finite Element Methods, Springer-Verlag, New York, 1994.[16] V. Girault, P.A. Raviart, Finite Element Methods for Navier–Stokes Equations, Springer-Verlag, New York, 1986.