A MIXED FINITE ELEMENT ~,~THOD

F O R 2~nd O R D E R E L L I P T IC PROBLEMS

P.A. Raviart and J . M ° T h o m a s

I. I N T R O D U C T I O N

Let ~ be a b o u n d e d open subset of R n w i t h a L i p s h i t z continuous

boundary F. We c o n s i d e r the 2nd o r d e r e l li p t i c model problem

-Au = f in ~ ,

(I .I)

where
I u = 0

f is a g i v e n
oh

function
F ,

of the space L2 (~). A v a r i a t i o n a l form of

problem (1.1), k n o w n as the complementary energy principle, c o n s i s t s
in finding p = gradu which minimizes the complementary energy functi 2
nal

(1.2) ~(q)~ =y I~I~dx

n

over the affine m a n i f o l d W of v e c t o r - v a l u e d functions ~@ (L 2 (~))n

%
which satisfy the equilibrium equation

(1.3) div~ + f = 0 in ~.

The use of c o m p l e m e n t a r y energy p r i n c i p l e for c o n s t r u c t i n g finite

element discretizations of e l l i p t i c problems has been first a d v o c a t e d
by F r a e i j s de V e u b e k e [51, [6],|7] . The s o - c a l l e d equilibrium method
consists first in c o n s t r u c t i n g a finite-dimensional submanifold ~ h of
W~ and then in f i n d i n g ~h 6 ~ h which minimizes the c o m p l e m e n t a r y energy
functional I(q) over the affine m a n i f o l d %~h. For 2nd order e l l i p t i c
problems, the n u m e r i c a l analysis of the e q u i l i b r i u m method has been

Centre de M a t h ~ m a t i q u e s Appliqu~es, Ecole Polytechnique and U n i v e ~

sit~ de Paris VI.
~ Universit~ de Paris VI.
 293

made by Thomas [19],[20].

Now, we note that the p r a c t i c a l c o n s t r u c t i o n of the s u b m a n i f o l d
~h is not in general a simple p t o b l e m since it r e q u i r e s a search for
e x p l i c i t solutions of the e q u i l i b r i u m e q u a t i o n (1.3) in the w h o l e
domain ~.
In order to avoid the above difficulty, we can use a more general
v a r i a t i o n a l principle, known in e l a s t i c i t y theory as the Hellinger-
Reissner principle, in w h i c h the c o n s t r a i n t (1.3) has been removed at
the expense h o w e v e r of introducing a L a g r a n g e multiplier. This paper
will be d e v o t e d to the study of a finite element m e t h o d based on this
v a r i a t i o n a l principle. In fact, this s o - c a l l e d m i x e d m e t h o d has been
found very useful in some p r a c t i c a l p r o b l e m s and refer to [17] for an
a p p l i c a t i o n to the n u m e r i c a l solution of a n o n l i n e a r p r o b l e m of radi~
tive transfer.
For some general results c o n c e r n i n g m i x e d methods, we refer to
Oden [12],[13] , Oden & Reddy [14] , Reddy [16]. M i x e d m e t h o d s for
solving 4th order e l l i p t i c equations have been p a r t i c u l a r l y analyzed:
see Brezzi & Raviart [2] , C i a r l e t & R a v i a r t [4] , J o h n s o n [9 ], [I0 ] ,and
M i y o s h i [11]. For related results we refer also to H a s l i n g e r & Hl~vacek
[8] .

An outline of the paper is as follows. In § 2, we derive the m i x e d

v a r i a t i o n a l f o r m u l a t i o n of p r o b l e m (1.1) and we define the related
d i s c r e t e elements, and in § 4, the error analysis of the a s s o c i a t e d
finite element m e t h o d is made. Finally, in § 5, we g e n e r a l i z e the re-
sults of §§ 3,4 to m i x e d m e t h o d s using r e c t a n g u l a r elements.
Let us d e s c r i b e some of the n o t a t i o n s used t h r o u g h o u t this paper.
Given an integer m > 0 ,

Hm(~) = { v e L 2 (~) ; ~ v e L 2 (~) , I~I < m }

denotes the usual S o b o l e v space p r o v i d e d the n o r m and s e m i - n o r m

i J-

llvlim'a ! (I i
l<m a l a ~ v l 2 d x ) ' IV]m'a : (I !l=m af
Given a v e c t o r - v a l u e d function q = (q, , ..... ,qn) E (Hm(~))n, we set:
%

n 2/- n
llqllm,a = ( [ llqill~,a ) , lqlm,~ = ( [ lqilm,a )2
i=I i=I
/-
2
We d e n o t e by H (I') the space of the traces v LF o v e r I" o f the
functions v E H I (~).
 294

2. T H E MIXED MODEL

In o r d e r to d e r i v e the a p p r o p r i a t e variational form of p r o b l e m

(1.1), we introduce the space

(2.1) H(div ; ~) = {qC (L2 (~)) n ; d i v q • L 2 (~) }

provided with the n o r m

(2.2) IlqIlH (div;~) = (llqll~ + Ildiv qll 2 )

Given a vector-valued function qEH(div ; ~), we m a y define its n o r m a l

component ~ • ~ • H-#(F) where H 2 (7) is the d u a l space of H#(F) and i
is the unit outward normal along F. M o r e o v e r , we h a v e Green's formula

(2.3) VvE H I (~), I{qradv- q + vdivq}dx = I v q-ldy

j ~",P,/~, %
r

where the integral i represents the d u a l i t y between the spaces H-#(F)

F
and H 2 (F) .
We n e x t define problem (P). Find a pair of functions (p,u) E
H(div ; a) x L 2 (a) s u c h that

r
(2.5) Yv6L 2 (~) , | v(div p + f ) d x = 0.
J

Theorem I. T h e p r o b l e m (P) has a unique solution (~,u) E H ( d i v ; ~)×~(e).

In a d d i t i o n , u is the solution of the p r o b l e m (1.1) and we have (2.6)
p = grad u.
q/Vb~/b

Proof. L e t u s first check the u n i q u e n e s s of the solution of problem (P).

Hence, assume that f = 0; f r o m (2.5), we g e t div p = 0. T a k i n g q = p
in (2.4), we obtain p = 0. T h e r e f o r e , we h a v e

f
(2.7) VqEH(div ; ~) , | u div q dx = 0

Now, let w E H I (~) be a f u n c t i o n such that

 295

Aw = u in ~.

Then, by c h o o s i n g q = gradw in (2.7), we g e t u = 0.

It r e m a i n s o n l y to s h o w t h a t the p a i r (p = g r a d u,u) is a solu-
t i o n of p r o b l e m (P), w h e r e u is the s o l u t i o n of p r o b l e m (1.1). O n the
one hand, we h a v e

div p + f = Au + f = O.

O n the o t h e r hand, s i n c e u = 0 on F, we g e t by u s i n g the G r e e n ' s

formula

(2.8) I{p-q~
~ + U d i v q } d x = JI uq'v d y = 0

Remark I. O n e can e a s i l y c h e c k t h a t the s o l u t i o n (p,u) of p r o b l e m (P)

m a y be c h a r a c t e r i z e d as the u n i q u e seddle-point of the q u a d r a t i c
functional

L(q,v).~ = I(q)% + I v ( d i v q% + f)dx

over the s p a c e H ( d i v ; ~) x L2 (£), i.e.,

WqeH(div ; £), ¥ v e L = (~), L(p,v)<_L(p,u)<_L(~,u)

Hence, the f u n c t i o n u is the L a g r a n g e multiplier associated w i t h the
constraint d i v p + f = 0.
%
Let us n o w i n t r o d u c e a general method of d i s c r e t i z a t i o n of p r o -
blem (1.1) b a s e d on the m i x e d v a r i a t i o n a l formulation (2.4),(2.5). We
are g i v e n two f i n i t e - d i m e n s i o n a l s p a c e s £h and V h s u c h t h a t

(2.8) £h C H(div ; £) ; V h C L 2 (~).

T h e n w e d e f i n e problem (Ph) : F i n d a p a i r of f u n c t i o n s (~h ' Uh)

6 Qh x V h s u c h that

(2.9) V qh e £ h " i Ph'~qhdX + i Uh d i v qh dx = 0 ,

(2.10) W vh e V h , V h ( d i v ~h + f)dx 0
 296

Using a general result of Brezzi [I , Theorem 2.1] concerning the

approximation of variational problems, we get the following

Theorem 2. Assume that

~h E Qh
(2.11) [ ~ div ~h = 0
V q dx = 0
v h E V h , j v h div ~h

and that there exists a constant e > 0 such that

i v h div ~h dx
¥ Vh E V h , sup ~ > allVhJl
h li hlJH(div ; -

Then the p r o b l e m (Ph) has a unique solution (~h ' Uh) E ~h × Vh and
there exists a constant ~ > 0 which dependes only on ~ such that

I llp_~hiiH(div;~) + Iiu-uhlIo , ~ <

--

(2.13)
<-- qf i inf IIp-q. I[ + inf llU-Vhll }
~qh @ Qh ~ ~h H(div ; ~) vh e V h o,n

Remark 2. Define the operator V h e L(V h ; Qh ) by

(2.14) ¥ Vh E V h ' V qh E Q h ' I "V'v"

q ' n~ n ~n dx = - I Vh div qh dx "

Clearly, v h can be viewed as an approximation of the operator grad.Now,

the function u h may be characterized as the unique solution of the fol
lowing problem : Find u h E V h such that

(2.15) ¥ Vh E Vh " I "V~U"n "V-v"

~%n n * dx =~ I fvh dx

In fact, from the assumption (2.11) and (2.12), it follows that pro-
blem (2.15) has a unique solution u h E V h . Moreover, it is readily
seen that the pair (~hUh , u h) is the solution of p r o b l e m (Ph) .
Since in general V h : H: (~) (2.15) is non-conforming displacement
0
model for solving p r o b l e m (1.1). For other n O n - c o n f o r m i n g methods ba-
sed on hybrid models, we refer to [15].
 297

It remains to c o n s t r u c t the f i n i t e - d i m e n s i o n a l subspaces ~h and

V h of the spaces H(div ; ~) L2 (~) r e s p e c t i v e l y so that they satisfy
"good" a p p r o x i m a t i o n p r o p e r t i e s and the c o m p a t i b i l i t y conditions
(2.11) and (2.12) w i t h a c o n s t a n t a i n d e p e n d e n t of the p a r a m e t e r h.
For convenience, we shall assume in the sequel that ~ is a
bounded poZygon of ~2. We then e s t a b l i s h a t r i a n g u l a t i o n ~ h of
made up with triangles and p a r a l l e l o g r a m s K whose diameters are ! h .
We begin by c o n s r t u c t i o n f i n i t e - d i m e n s i o n a l ~h of the ~pace H(div;~).
G i v e n a finite element K 6 ~ h ' we denote by ~K the unit outward
normal along the b o u n d a r y ~K of K. Using the Green's formula (2.3) in
each K E ~ h ' one can easily prove that a function q E (L2 (~)) 2
belongs to the space H(div ; D) if and only if the two following
c o n d i t i o n s hold :

(i) for all K e ~ h ' the r e s t r i c t i o n ql K of q to the set K

belongs £o the space H(div ; K) ;
%

(ii) for any pair of a d j a c e n t elements Ks , K2 E ~ h ' we have the

r e c i p r o c i t y relation

(2.16) ql " ~ K + q2 " ~ K = 0 on K' = KI C K2 ,

where ~i stands for qlKi , i = 1,2 .

Hence the functions of ~h will be assumed to be smooth in each

element K E ~ h and to satisfy the r e c i p r o c i t y conditions.
 298

3. MIXED T R I A N G U L A R ELEMENTS

In this § , we shall assume that K is a triangle, w i t h K and

for any integer k >_ 0, we shall a s s o c i a t e a space QK of v e c t o r - v a l u e d
functions q e H ( d i v ; K) such that :

(i) div q is a p o l y n o m i a l of degree < k ;

(ii) the r e s t r i c t i o n of q'~K to any side K' of K is a p o l y n o m i a l of

degree < k.

We begin by introducing the space ~ a s s o c i a t e d w i t h the unit right

triangle K in the (~,q)-plane whose v e r t i c e s are ~, = (I,0), ~2 = (0,1),
@~s= (0,0). Let us first give some notations. We denote by Pk the space
of all p o l y n o m i a l s of d e g r e e < k in the two v a r i a b l e s <,q and by Sk
the space of all functions defined over 9K whose r e s t r i c t i o n s to any
side K' of K are p o l y n o m i a l s of degree < k. Given a p o i n t ~ = (~,q) of
R 2 • we d e n o t e by ~i = ~i(~), I < i <_ 3, the b a r y c e n t r i c c o o r d i n a t e s of
w i t h respect to the vertices ~i of K.
Now, the space ~ is r e q u i r e d to satisfy the following properties:

(3.1) (Pk)2C Q ;
%

(3.2) dim(Q)= (k+1) (k+3) ;

%

(3.3)
9[ ~q

(3.4) ¥~Ek , ~.0k @~S (where ~ stands for ~ ) ;

(3.5) ko = { ~ E ~ ; div ~ = 0} C (pk)~

Lemma I. Assume that the conditions (3.2)-(3.5) hold. Then a function

~E~ is uniquely determined by:

(a) the values of q.v at (k+1) distinct points of each side K' of ~;
% %

(b) the moments of order < k-1 of ~ , i.e.,

qiXl X~ 2 )t~ 3 dx , i = 1,2, al + cL2 + c~3 = k-l.

K

Proof. Since by (3.2) the number of degrees of f r e e d o m (a), (b) is equal

to the d i m e n s i o n of the space Q , it is s u f f i c i e n t to prove that a
function ~ E ~ w h i c h satisfies the two conditions:
 299

(3.6) ~-9~ = 0 at (k+1) d i s t i n c t points of e a c h side K' of K ,

(3.7) i~ix? 1 ~;~ x; ~ dx = o, i = 1,2, ~1 + ~ + ~ = k-1

K

must vanish identically. In fact, conditions (3.4) and (3.6) imply

q-~ = 0 o n ~K. Hence, using (3.7) and a p p l y i n g the G r e e n ' s formula

(2.3) in K, w e o b t a i n for all ~ E P k

i. ~ d iv ~ dR = - { g r a d ~.~ d~ + i "'" d~ = 0
K K ~K

Since, by (3.3), d i v ~ E P k , we get d i v ~

q = 0 so t h a t ~ E Q o .
NOW, i t follows from (3.5) that there exists a polynomial w~2Pk+ ]
uniquely determined up to an a d d i t i v e constant such t h a t

~ ~ )
= ~%~curl Q = I ~n ' - ~--~ "

N o t e t h a t q.~~ = ~-~w
~ = 0 on ~K, w h e r e ~ stands for the t a n g e n t i a l
derivative along ~K. T h u s w e m a y a s s u m e t h a t Q = 0 on ~K a n d we m a y
write

Q = 111213z, ZEPk_ 2 (z = 0 for k = 0,1)

Using again (3.7), w e o b t a i n for a n y t ~ (Pk-1)2

K K K

where ~r2 - ~rl

curl { = ~ ~ E Pk-2 " Clearly " we can choose r so t h a t
= curl ~ and then

i X112X3z2dx = 0 .
K
Therefore, we get ~ = 0 so t h a t Q = 0 and ~ = c u r l Q = 0.

Remark 3. A s r e g a r d s the d e g r e e s of f r e e d o m of a f u n c t i o n ~ E ~ , one

could have equivalently specified the moments of order ~ k

g ~ ' EP k
 300

of ~-9 on the side ~' instead of its v a l u e s at (k+l) distinct points

nj

of R'
Let us give some e x a m p l e s of spaces o

Example 1. Let k > 0 be an even integer; we d e f i n e ~ to be the space

of all f u n c t i o n s ~ of the form
k k
ql = pOlk(~,n) + ao~k+l +al + "''+~k
i ~k $~ + I q~
(3.8) k k
-- -- +1

~t2 = pOlk(g,n) + Bo rlk+l + ( 6 ~ n k + " ' ' + g k ~2 n 2

with
k
2
(3.9) [ (-1)i(~i-6 i) = 0
i=0

In (3.8), pOlk(~,n) denotes any p o l y n o m i a l of d e g r e e k in the two

variable ~,q. Clearly, conditions (3.1),(3.2) hold. Next ~.~ is
n~

obviously a polynomial of d e g r e e <_ k on each side $ = 0 and q = 0 of

A
K. On the other hand, it follows from (3.9) that q.~ is also a poly-
nomial of d e g r e e < k on the side < + q = I . Finally, we have
k
2 , .k-i i
div ~ = pOlk_ I ($,~) + ~ (k+1-i) ~ai% n + Bi~iq k-i) E p k
i=0

so that div ~ = 0 implies

k
~i = 6i = 0 , 0 <_ i <_ ~ - 1 ,

~k + 6k = 0
2

and, by the c o n d i t i o n (3.9)

k
ai = 8i = 0 , 0 <_ i < ~

Hence, hypotheses (3.1)-(3.5) hold.

Consider for i n s t a n c e the case k = 0. Then a f u n c t i o n ~q~ is of
the form

I (~i = ao + a1~
(3.10) q2 = b o + bl q , al = bl ,
 301

and by L e m m a I, the degrees of f r e e d o m of ~ m a y be c h o s e n as the

values of q-~ at the m i d p o i n t s of the sides of the triangle K.

Example 2. Now, let k >_ I be an odd integer; we then define ~ to be

the space of all functions ~ of the form
k+1 k+1
ql = pOlk(~'n) + ~o ~ k + 1 + a ~ k n + "
• .+~k+1 ~ 2 n 2 ,
2
(3.1 1 ) k+1 k+1
2 2
q2 = p O l k ( ~ , ~ ) + ~oqk+l+B1 ~ q k + . . . + S k + 1 ~ n ,
2

with
k+1 k+l
2 2
(3.12) ~ (-I)i~ = ~ (-I) i 6i = 0 .
i= 0 1 i=0

Here again, one can e a s i l y check that conditions (3.1)-(3.5) hold.

For k = I, a f u n c t i o n q ~
~ ~ is of the form

= a O + a1(+a2n +a3(((+q) ,
(3.13)
= b + b,~ + b 2 n +b3~([.+~)
O

and, by L e m m a I, the d e g r e e s of freedom of ~ m a y be c h o s e n as the

values of ~.~ at two d i s t i n c t points of e a c h side of K (for the
Gauss-Legendre points) a n d as the ~ e a n value

1 d£ = ~ ~1
rues (~1) :~ K

of ~ o v e r K.
Next, consider any triangle K in the (x, , x 2 ) - p l a n e whose verti-
ces are d e n o t e d by a i, 1<i<3. We set :

(3.14) h K = diameter of K,

(3.15) PK = d i a m e t e r of the inscribed circle in K.

Let F K : ~ ~ FK(~ ) = BK~ + bK , B K E L~2),b K @ ~2 , be the u n i q u e

affine invertible mapping such that

F K (ai) = a i , 1<i<3.
 302

W i t h any s c a l a r function $ defined on K (resp. on 8K), we a s s o c i a t e

the f u n c t i o n ~ defined on K (resp. on 8K) by

(3.16) ~ = $ o FK I (~ = £0 o F K)

On the o t h e r hand, with any vector-valued function ~ = (q, , q2)

defined on ~, we a s s o c i a t e the f u n c t i o n q d e f i n e d on K by

(3.1 1 q =% I
{ o
--I -i

where JK = det(BK)" We shall constantly use in the s e q u e l the o n e - t o -

one c o r r e s p o n d e n c e s ~ *---~ ~ ~ +-+
The c h o i c e of the t r a n s f o r m a t i o n (3.17) is b a s e d on the f o l l o w o n g
standard result.

Lemma 2. For any f u n c t i o n ~ E (H I (~))2, we have:

(3.18) ¥ ~ E L 2 (~) , d i v ~ dR = ~ d i v q dx ,
K K

(3.19) T4 ~ 6 L2 (~K) , i ~ a~ . 0tb d ~ = i ~ rqb ' ~rbK dy •

~K ~K

F o r the proof, see [18] for i n s t a n c e . We s h a l l a l s o n e e d

Lemma 3. We have for any i n t e g e r £ ~ 0 :

(3.20) v ~ e H~(~), I~1~,~ <_ IIBKII~ ia l i l ,,K ,

J_
(3.21) 2,

] --! --!
where ]IBKil (resp. IB K I]) d e n o t e s the s p e c t r a l n o r m of By (resp. B K ).

Proof. The inequality (3.20) has b e e n d e r i v e d in [3 , inequality

(4.15)]. By u s i n g (3.17), the i n e q u a l i t y (3.21) can be o b t a i n e d in a
very s i m i l a r way.
Now, w i t h the t r i a n g l e E, we a s s o c i a t e the s p a c e

Assume that conditions (3.3) and (3.4) hold. Then, by L e m m a 2, the

functions q of the space £K s a t i s f y the d e s i r e d p r o p e r t i e s (i) a n d (ii).
 303

Concerning the approximation of smooth vector-valued functions q

by functions of the space ~K ' we have

Theorem 3. Assume that the conditions (3.1)-(3.5) hold and let the
space ~K be defined as in (3.22). Then there exist an operator
~K E L((H~ (K))2; and a constant C > 0 independent of K such that:

(i) for each side K' of K and for all ~ E Pk

(3.23) i (~ g-g)-~ ~ d~ = o ,
K'

(ii) for all function q E (Hk+1 (K))2 with div q E H k+1 (K) ,

, k+2

(3.24) II~K q - " H(div;K)i nKPK

q" C (lql k+l,K +i div qlk+l ,K]

Proof. Given a function q 6 (H I (~))2, there exists by Lemma I and

Remark/~. a unique function ~ ~ E ~ such that

(3.25) ~ E Pk ' X (~ ~ - ~ ) "9 ~ dq = 0 for each side K' of K ,

K'

It follows from (3.1) that ~W ~ = ~ for all ~ ~ (PK)2. Then, by

applying Lemma 7 of [3] in vector form, we get for all ~ E (Hk+1 (~)) 2

(3.27) II~-~,~o.~_< c,i~Ik+1,~

for some constant ci = ci (K). On the other hand, using (3.25),(3.26)

and the Green's formula, we obtain for all ~ E Pk

i div(~ ~ - ~ ) ~ dR = - I (~ ~ - ~ ) . g r a d ~ d x + /(~-~ ~).~ ~ dq = 0.

K K ~

Hence div(~ ~) is the orthogonal projection in L2 (K) of div ~ upon PK"

Then, assuming that div ~ 6 H k+1 (~) and applying again [3, Lemma 7],
we obtain for some constant c2 = c2 (~)

(3.28) ,div(~ ~-~), < c~ Idiv

 304

Define now the o p e r a t o r ~ by

~K

~ ^~q E (HI (K))2' ~K~q = "~ ,~

Clearly, (3.23) follows from (3.25) and L e m m a 2. Since

~K q-q~ = 7~ BK (~I-~)o FK '

we have
-f
II~ q-qtl < ttB~<tI1%t tl~ ~-~tlo, ~ .

Thus, by u s i n g i n e q u a l i t i e s (3.27) and (3.21) for £ = k+1, we get for

all q 6 iH k+1 (K)) ~

(3.29) II~zK ~-qll o ,K < cl llBKl[k+21IBK

-i l l l q l k + l , K "

Finally, from (3.18) we have

--I
1
FK

so that

"div(~K ~-~I"o,~ = I%1 "aiv<~ ~-~I"o,~

Therefore, noticing that

div ~ = JK(div~q)

and a p p l y i n g the i n e q u a l i t i e s (3.28) and (3.20) (with £ = k+1 and

= div q), we o b t a i n w h e n div q E H k+1 (K)

(3.30) Lldiv(~ q-z)llo,K <_ c21LB~llk+1 Idiv qlk+1, ~

Since, by [ 15 , Lem/na 2] , we have

hK -i hK
(3.31) llBKil < PK , lIBK" 11 <_--pK ,

the d e s i r e d inequality (3.24) follows from (3.29) and (3.30).

 305

4. E R R O R BOUNDS

Assume that ~ h is a t r i a n g u l a t i o n of ~ m a d e up w i t h t r i a n g l e s
K whose diameters are < h. We n o w i n t r o d u c e the s p a c e

(4.1) Qh = {~qh 6 H ( d i v ; ~) ; V K E~h, qhlK E QK }

where, for all K E ~ h ' the s p a c e ~K is d e f i n e d as in (3.22).

The degrees of f r e e d o m of a f u n c t i o n ~h E ~h are e a s i l y determi-
ned; t h e y can be c h o s e n as

(i) the v a l u e s of ~ h . ~ K , at (k+1) distinct points of e a c h side K'

of the t r i a n g u l a t i o n ~ h ;

(ii) the m o m e n t s of o r d e r ~ k-1 of ~h o v e r e a c h t r i a n g l e K E ~h-

O n the o t h e r hand, for a n y ~h @ ~h a n d any K E ~ h ' we h a v e

(div qh ) IK E Pk " Hence, a natural choice for the s p a c e V h is g i v e n by

(4.2) V h = {v h e L 2 (~) ; ¥ K e ~ h , Vhi K e pk} ,

sO t h a t c o n d i t i o n (2.11) is a u t o m a t i c a l l y satisfied.
N o t e t h a t the f u n c t i o n v h E V h do n o t s a t i s f y a n y c o n t i n u i t y
constraint at the i n t e r e l e m e n t boundaries.
Now, in o r d e r to a p p l y T h e o r e m 2, the e s s e n t i a l step c o n s i s t s in
proving that the c o m p a t i b i l i t y condition (2.12) holds with a constant
independent of h. In fact, we w a n t to s h o w that, for any f u n c t i o n
v h 6 V h , there exists a function ~h @ ~h such that

(4.3) d i v ~h = Vh in

and

(4.4) Ilq.
%n IIH ( d i v ; ~) -< C11Vhllo '~ '

where the c o n s t a n t C is i n d e p e n d e n t of h. F o r the proof, we n e e d some

technical preliminary results.
Let K a triangle of ~ h ; we denote by S k , ~ K the space of all
functions defined over ~K w h o s e restrictions to any side K' of K are
polynomials of d e g r e e < k.

Lemma 4. Let there be given functions v 6 Pk and ~ E Sk,~ K such that

(4.5) i vdx = I ~ dX .
K ~K
 306

Assume that conditions (3.2)-(3.5) hold. Then there exists a function

q • Q K such that

div q = v in K ,
(4.6)
q. w = ~ on ~K ,

and

2 h~ 2 2
(4.7) Ilql]H(div ; K) < C(llVllo 'K + PK Ilull o , ~K )

where the constant C is i n d e p e n d e n t of K.

Proof. Let 91 • Pk and ~i • Sk be f u n c t i o n s such t h a t

(4.8) I 91d£ = f ~id~

K 9K

Then the N e u m a n n probl6m

AW = 91 in K ,

(4.9)
= [~ on ~ ,

has a s o l u t i o n ~ • HI (K) w h i c h is u n i q u e up to an a d d i t i v e constant.

Moreover, there exists a constant ci = ci (K) > 0 s u c h t h a t

2 2 2
(4.~0) t~Ll,~ _< c,(tl~,llo, ~ + tl~,ito,~)

Now, by L e m m a I, t h e r e e x i s t s a unique function ~ • ~ such that

V 9 • __(Pk_1] ( -g.rad Q ) . f dR = 0
K

q'O = ~i on ~K .

From (4.9) and the G r e e n ' s formula, it f o l l o w s that

K ~ ~
 307

= - I grad W grad ~ d£ + [ ~ ~ A~ d~
K ~K K

so t h a t d i v ~ is the o r t h o g o n a l projection in L2 (K) of AW u p o n Pk"

Hence, we get

d i v ~ = 91 in R1 ,
(4.11)
~°~
q i = ~, on B!~.

O n the o t h e r hand, it f o l l o w s from (4.10) that

3_
2 2 2
(4.12) I1~11
~, o ,~, -< c~ (1191, o,P, + il,~,, o,~fi)

for some constant c2 = c2 (R) > 0.

Now, let K E ~ h ; with the functions v E Pk a n d ~ E Sk,~K such
that (4.5) holds, we associate the functions 91 E Pk and ~i 6 ~k
defined by

V ~ E Pk , i 91~ dR = i v ~ dx ,
K K
(4.13)
f
V @ Sk , | ~i ~ d? = ~ ~ d7 •
J
~K BK

Clearly we have (4.8) and there exists a function ~ E ~ such that

(4.11) and (4.12) hold. We n e x t define q E ~K by

I --1
/414) = B, o F.

so that, by (4.11) and L e m m a 2, we get (4.6).

It r e m a i n s only to s h o w the estimate (4.7). We get from (4.12) and
(4.14)

2 2 --1 2 2
(4.15) IIqlI~o,K_<C2UBKIlIJK2 (119111o,K +11~111 o,~R1

Since 91 = IJKlV o FK , we o b t a i n

2 2
(4.16) 1191 []o, ~ = ]JK]llV[Io, K .

On the o t h e r hand, let K' be a side of ~ and let K' = F K ( ~ ' ) . Since
 308

the s u p e r f i c i a l measures of ~' and K' are r e m a t e d by

meas(K') <_ IIBK1 tl[JKlmeas(K')

we obtain

(4.17) u~, U~o,K-,<- IIB~' IIIJKI II~ IIo,~<,

By c o m b i n i n g the i n e q u a l i t i e s (4.15) - (4.17), we get

2 2 2 2
(4.18) llqll~o,K ! c 2211BKII (llV[lo,K + llg-111U~IIKo , 3 K )

Therefore, the d e s i r e d inequality follows from (4.18) and (3.31).

L e t us n e x t i n t r o d u c e the space

(4.19) M h = {~h @ K Sk ; +phl3K ~ = 0 on KI C~ K2

KE~h ,~K Vhl 3Ki

for e v e r y p a i r of a d j a c e n t triangles KI, K~ E ~ h } .

We c o n s i d e r a regular family (~h) of t r i a n g u l a t i o n s of ~ in the sense

of [3] , in t h a t t h e r e e x i s t s a constant o > 0 independent of h s u c h
that

hK
(4.20) max -- < o .
K 6)~ h PK --

Lemma 5. Let there be given spaces V h and M h defined as in (4.2) and

(4.19) which are a s s o c i a t e d with a regular family of triangulations.
Then, with any f u n c t i o n v h E V h , we can associate a f u n c t i o n Uh @ M h
such that for all K @ ~ h

(4.21) v h dx = i Ph dy
K 9K

and
2-
2 2
(4.22) hK[lUhllo,~Kl <_ Cllvhllo, ~ ,
K 6~h

where the constant C > 0 is independent of h .

Proof. We shall c o n s t r u c t the f u n c t i o n Uh by u s i n g a h y b r i d finite

e l e m e n t m e t h o d as it has b e e n d e s c r i b e d and studied in [15] .Hence, the
 309

proof of the Lemma will depend heavily upon the results of [15] .
We first define the space

Xh = {~h • L2 (~) ; V K • ~ h ' ~hiK • Pk+2 )

provided with the norm

ilXh =
K •~ h
(t hL,,K + -2 ii~hLio,K)

Next, we set :

l
a(? h, Sh ) = [ j grad ~ . g r a d ShdX, ~h,$h • X h ,

f
b(~ h, Ph ) = _ ~ J ~hUh d¥ , ~hEXh,UhEMh -
.K •}~ h ~K

Then, by using [15, Theorem 2 and Lemmas 2,3,4], there exists a unique
pair of functions I~ h , ~h ) 6 X h × H h such that

(4.23) W Sh c X h , aI~h , Sh I + bISh , ~h I = S VhShdX '

(4.24) V Ph • Mh ' bI~h ' Ph I = 0.

By choosing in (4.23)

%h = characteristic function of the set K • ~ h

we get (4.21) for all K • ~ h "

Now, in order to prove the inequality (4.22), we introduce the
following subspace of the space X h

Yh = {$h • Xh ; ¥Ph E M h ' b[$h' Ph ] = 0}

Clearly, the function ~h • Yh may be characterized as the solution of

V Sh • Yh ' a(~h ' ~h ) = i VhShdX "

Therefore, we get
 310

a(~ h , ~h ) <_ llVhlIo,~ll~hllo,~

By, [15, inequality (6.18)] , we have the discrete analogue of the

Poincar~-Friedrichs inequality :

¥ Sh 6 Yh ' llShlIo,~ ~ cla($ h , Sh) ,

where the constant ci is independent of h. Hence we obtain

1.
(4.25) a(~oh , ~h )2 < cI [Vhllo, ~

Next, it follows from (4.23) and (4.25) that

2
(4.26) b(, h • ~h ) --
< (ll,hll
o,~ +cla(, h ' ,h ) )11VhLlo , ~ <
--

< C2 II$h[IX h JlVhll0 , 2

-- '

where c2 is a constant independent of h. Thus, the inequality (4.22)

follows from (4.26) and the following inequality

2 2 b($h,~h )
( [ h K ll~h[o,~K ) < C3 sup , c3 = C3 (~)
K 6~h Sh @ Xh JJ~h[JXh

(cfr. [15 , inequality (6.29)]).

We are now able to state

Theorem 4. Let there be given spaces ~h and V h defined as in (4.1) and

(4.2), w h i c h are associated with a regular family of triangulations.
Assume in addition that the conditions (3.2)-(3.5) hold.Then, with any
function v h 6 V h , we can associate a function ~h E ~h which satisfies
the conditions (4.3), (4~4) w i t h a constant C > 0 independent of h.

Proof. Let v h be a function in V h. By Lemma 5, we construct a function

Vh 6 M h such that the conditions (4.21) and (4.22) hold.
Next, using Lemma 4, there exists a function ~h E (L ~ (~))2 such that
for all K E ~ h ' we have :

ZhlK ~ ~K '

div (qh IK) = Vhl K '

(~hL~)'~= ~hl~K •
 311

Since ~h E M h , the reciprocity conditions (2.15) hold so that q~hE~h

and div ~h = Vh in ~. Moreover, it follows from (4.7) and (4.20) that

2 2 2

(4.27) Uqhll
~ ~ (div; ~) ~ C 2 ( IIVhll° '~ + a K @~ h hKIl ~hIlo, 3K ) .

Combining the inequalities (4.22) and (4.27), we obtain the inequality

(4.4).
We now have our main result.

Theorem 5. We assume that u 6 Hk+2(~) and Au E Hk+1(~) for some inte-

ger k ~ 0. Let there be given spaces ~h and V h d e f i n e d as in (4.1)-
(4.2), w h i c h are associated with a regular family of triangulations.
We assume in addition that the conditions (3.1)-(3.5) hold. Then pro-
blem (Ph) has a unique solution and there exists a constant ~ inde-
pendent of h such that

(4.28) 11~-PhJlH(div;~) + llU-Uhllo,~ <_ ~ hk+1 (lUlk+1,e+lUlk+2, +[nUlk+1,e)

Proof. Let v h E V h ; by the previous theorem, we have

l
sup J~ vh d i v ~h dx
1
qh E ~h L'q il >-- ~ HVhUo,e
~h H(div ; ~)
%

I
so that the hypothesis (2.12) holds with a = ~ . Thus, by T h e o r e m 2,
it remains only to evaluate the quantities

inf llp-~hll
~ H (div; ~) and inf llU-Vhllo,~
~h E ~h ~ vh E Vh

On the one hand, by using T h e o r e m 3, we define ~h p E (L 2 (S))2

by

, hPIK = ~ (pl~l

It follows from (3.23) that the reciprocity relations (2.15) hold so

that %~h E ~h" Next, we deduce from (3.24) and (4.20) that for some
constant ci independent of h

(4.29) tL~-~ pll < clh k+1 + IAul

~h H (div;~) -- (lUlk+2,~ k+1,~ )"
 312

O n the o t h e r hand, a straightforward application of [3, T h e o r e m 5]

gives for some c o n s t a n t c~ i n d e p e n d e n t of h

(4.30) inf ~U-Vhll < c2h k+1

o,~ - - lUlk+l,~ "
vh @ Vh

Then, inequality (4.28) follows from inequalities (2.11),(4.29) and

(4.30).

5. M I X E D Q U A D R I L A T E R A L ELEMENTS

We s h a l l b r i e f l y discuss the case of q u a d r i l a t e r a l elements. As

for t r i a n g u l a r elements, we b e g i n by i n t r o d u c i n g the space Q a s s o c i a t e d
w i t h the u n i t s q u a r e K = [ 0,I] z in the (~,q)-plane. Given two i n t e g e r s
k, Z ~ 0, let us d e n o t e by Pk,£ the space of all p o l y n o m i a l s in the two
variables ~,q of the f o r m

k
(5.1) p(~,q) = [ [ ci j ~zq] , ci j e ~ .
i=O j=O

N o w we d e f i n e the s p a c e k by

N o t e that, for ~ E ~, we h a v e :

(i) div ~ = ~ ~ Pk,k

(ii) the r e s t r i c t i o n of ~.2 to a n y side K' of K is a p o l y n o m i a l of

degree < k.

One can p r o v e

Lemma 6. A function ~ e ~ is uniquely determined by:

(a) the values of ~.0 at (k+1) distinct points of each side K'
off<
(b) the quantities

ql ~ i n 3 d x , 0 < i < k-1 , 0 < j < k ,

K

I~ q~iqJd~ , 0 < i <k , 0 < j < k-1

 313

The proof goes along the same lines of that of L e m m a I.

Consider for instance the case k = 0. A F u n c t i o n q • ~ is of the
form

ql = a ° + al ~ ,
(5.3)
q2 b ° + b, n ,

a n d by L e m m a 6, the d e g r e e s of f r e e d o m of ~ m a y .be c h o o s e n as the

values of q.9 at the m i d p o i n t s of the sides of the square ~.
nj

Next, let K be a p a r a l l e l o g r a m in the (x,, x 2 ) - p l a n e . There exists

an a f f i n e invertible mapping FK : R ~ F K(~) = BKR + bK , auch that
K = FK(K). With K, we associate the space

Let ~ 6 ~K ; the restriction of q-v to any side K' of the q u a d r ~

bE
lateral K is a p o l y n o m i a l of d e g r e e ~ k .
Assume now that ~ h is a t r i a n g u l a t i o n of ~ m a d e up w i t h paral-
lelograms K whose diameters are ~ h. We set :

(5.5) Qh = {qh E H ( d i v ; ~); V K E ~h' qh]K e Q }

Note that, for any ~h • ~h a n d any K • ~ h ' we h a v e

(div Zh) IK o F K Pk,k "

So we set

By u s i n g the t e c h n i q u e s of §§ 3,4, one can similarly prove that

problem (Ph) has a unique solution (~h ' Uh) E ~h × V h and that the
error bound (4.28) still holds.
 314

REFERENCES

[I] F.Brezzi : On the existence, uniqueness and approximation of

saddle-point problems arising from lagrangian multipliers.
R.A.I.R.O., R 2 Ao~t 1974, 129-151.

[2] F.Brezzi, P.A.Raviart : Mixed finite element methods for 4 th order

problems. TO appear.
[3] P.G.Ciarlet, P.A.Raviart : General Lagrange and Hermite interpola-
tion in R n with applications to finite element methods.
Arch. Rat. Mech. Anal. 46 (1972), 177-199.

[4] P.G.Ciarlet, P.A.Raviart : A mixed finite element method for the

biharmonic equation, Mathematical Aspects o_f Finite Elements
in Partial Differential E~uat/ons.
pp. 125-145. Academic Press, New-York, 1974.

[5] B.Fraeijs De V e u b e k e : Displacement and equilibrium models in the

finite element method, Stres 6 Analysis (O.C. Zienkiewicz
and G.S.Holister, Editors), ch. 9, pp. 145-197. Wiley, 1965.

[6] B.Fraeijs De V e u b e k e : Diffusive equilibrium models Lecture notes.

University of C a l g a r y , 1973.

[7] B.Fraeijs De V e u b e k e , M.A.Hogge : Dual analysis for heat conductions

problems by finite elements. Int. J. for Num. Methods in Eng.,
5 (I 972) , 65-82.
!
[8] J.Haslinger, I.Hlavacek : A mixed finite element method close to
the e~uilibrium model, I. Dirichlet problem for one equation,
II. Plane elasticity. To appear.

[9] C . J o h n s o n : On the convergence of a mixed finite element method

for plate bending problems. Numer. Math., 21 (1973), 43-62.

[10] C . J o h n s o n : Convergence of another mixed finite element method for

plate bending problems. To a p p e a r .

[11] T . M i y o s h i : A finite element method for the solutions of fourth

order partial differential equations. K u m a m o t o J., (Math.),
9 , (1973), 87-116.

[12] J . T . O d e n : Generalized conjugate functions for mixed finite element

approximations of boundary-value problems, The Mathematical
Foundations of the Finite Element Method (A.K. A z i z Editor),
pp. 629-670, Academic Press, New-York, 1973.

[13] J . T . O d e n : Some contributions to the mathematical theory of mixed

finite element approximations. T~kyo S e m i n a r on F i n i t e Elements,
Tokyo, 1973.
 315

[14] J.T.Oden, J.N.Reddy : On mixed element approximations, Texas

Institute for Computational Mechanics.
The U n i v e r s i t y of Texas at Austin, 1974.

[15] P.A.Raviart, J . M . T h o m a s : Primal hybrid finite element methods

for 2 nd order elliptic equations. To appear.

[16] J . N . R e d d y : A Mathematical Theory of Complementary-Dual Variatio-

nal Principles and Mixed Finite Element Approximations of
Linear B qundary-Value Problems in Continuous Mechanics.
Ph.D. Dissertation, The U n i v e r s i t y of A l a b a m a in
Huntsville, 1973.

[17] B.Scheurer : To appear.

[18] J . M . T h o m a s : Methode des ~l~ments finis hybrides duaux pour les

probl~mes elliptiques du 2 nd ordre. To appear inR.A.I.R.O.,
S~rie A n a l y s e Num~rique.

[19] J . M . T h o m a s : M~thode des ~l~ments finis ~quilibre pour les pro-

blames elliptiques du 2 nd ordre. To appear.

[20] J . M . T h o m a s : Thesis. TO appear.