4101278

A Posteriori Error Estimates for the Mortar Mixed Finite Element Method
Author(s): Mary F. Wheeler and Ivan Yotov

Source: SIAM Journal on Numerical Analysis, Vol. 43, No. 3 (2006), pp. 1021-1042
Published by: Society for Industrial and Applied Mathematics
Stable URL: http://www.jstor.org/stable/4101278
Accessed: 22-12-2015 12:28 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Society for Industrial and Applied Mathematics is collaborating with JSTOR to digitize, preserve and extend access to SIAM
Journal on Numerical Analysis.
http://www.jstor.org
This content downloaded from 202.28.191.34 on Tue, 22 Dec 2015 12:28:42 UTC
All use subject to JSTOR Terms and Conditions
SIAM J. NUMER. ANAL. ( 2005 Society for Industrial and Applied Mathematics
Vol. 43, No. 3, pp. 1021-1042
A POSTERIORI ERROR ESTIMATES FOR THE MORTAR MIXED

FINITE ELEMENT METHOD*
MARY F. WHEELERt AND IVAN YOTOVt
Abstract. Several a posteriori error estimators for mortar mixed finite element discretizations of
elliptic equations are derived. A residual-based estimator provides optimal upper and lower bounds
for the pressure error. An efficient and reliable estimator for the velocity and mortar pressure error
is also derived, which is based on solving local (element) problems in a higher-order space. The
interface flux-jump term that appears in the estimators can be used as an indicator for driving an
adaptive process for the mortar grids only.
Key words. mixed finite element, mortar finite element, nonmatching grids, a posteriori error
estimates, adaptive mesh refinement
AMS subject classifications. 65N12, 65N15, 65N30, 65N50
DOI. 10.1137/S0036142903431687
1. Introduction. We consider the second order elliptic problem written as a

system of two first order equations
(1.1) u =-KVp in Q,
(1.2) Vo u = f in Q,
(1.3) p= g on FD,
(1.4) u- v = 0 on FN,
where C Rd, d - 2 or 3, is a multiblock domain with a boundary 0Q= FD U FN,

FD N = 0, measure (ED) > 0, v is the outward unit normal on 0Q, and K is a
symmetric, uniformly positive definite tensor satisfying, for some 0 < ko < ki < oc,
(1.5) ko0 T TK(x)(s <

< k1iT7 VXe VR E d.
In flow in porous media the above system models single-phase flow where p is the
pressure, u is the Darcy velocity, and K represents the permeability divided by the
viscosity.
A number of papers in recent years have studied the numerical solution of the
above and related problems on multiblock domains with nonmatching grids across the
interfaces. This growing interest is driven by the flexibility provided by the multiblock
paradigm. Complicated geometries can be modeled as unions of relatively simple
subdomains with locally constructed grids. Local features of the solution such as
*Received by the editors July 17, 2003; accepted for publication (in revised form) February 25,
2005; published electronically September 20, 2005.
http://www.siam.org/journals/sinum/43-3/43168.html
tCenter for Subsurface Modeling, Institute for Computational Engineering and Sciences
(ICES), Department of Aerospace Engineering & Engineering Mechanics, and Department of
Petroleum and Geosystems Engineering, The University of Texas at Austin, Austin, TX 78712
(mfw@ices.utexas.edu). The research of this author was partially supported by NSF grants EIA
0121523 and DMS 0411413 and DOE grant DE-FGO2-04ER25617.
fDepartment of Mathematics, University of Pittsburgh, Pittsburgh, PA 15260
(yotov@math.pitt.edu). The research of this author was partially supported by NSF grants
DMS 0107389 and DMS 0411694, DOE grant DE-FG02-04ER25618, and the J. Tinsley Oden
Faculty Fellowship, ICES, The University of Texas at Austin.
1021
1022 MARY F. WHEELER AND IVAN YOTOV
corner singularities or large gradients can be resolved by finer grids in the local region.
Large scale features such as geological faults and layers in subsurface flow can be
modeled with nonmatching grids. Moreover, the resulting algebraic problem can be
efficiently solved via parallel domain decomposition algorithms.
In a multiblock formulation, the equations are imposed locally on each subdomain
and appropriate interface matching conditions are enforced on the interfaces. The use
of mortar finite elements to impose the interface conditions is a popular approach due
to its excellent stability and accuracy. For the use of mortars, the reader is referred
to [10, 8, 20] and references therein for Galerkin finite element and finite volume
methods, and to [40, 3, 9] in the context of mixed finite element methods.
An integral part of any successful computational method is the development of a
posteriori error estimators and adaptive mesh refinement strategies. Although there
is an enormous amount of literature on a posteriori error estimation and adaptivity
on conforming grids (seminal works include [5, 7, 2, 35]), few papers deal with this
issue for mortar finite element methods. In the case of Galerkin finite elements, error
estimators have been developed in [37, 38, 30]. Even fewer results are available for
the mortar mixed finite element method. Goal-oriented estimates and adaptivity are
developed in [6]. Computational results from [36, 29] indicate that a judicious choice
of mortar grids can lead to an accurate solution at low computational cost, but no
rigorous justification is given. The goal of this paper is to develop a posteriori error
estimators and an adaptive mesh refinement strategy for the mortar mixed finite
element method.
Previous works on error estimation for mixed finite element methods on conform-
ing grids include [11, 12, 17, 24, 39, 25]. In [11], mesh-dependent norms are utilized to
obtain optimal residual-based error estimators. Estimators based on superconvergence
error estimates are developed in [12, 24]. In [17, 39], the Helmholtz decomposition
is used to derive optimal residual-based error estimators in the natural pressure and
velocity norms. Hierarchical estimates and implicit estimates based on solving local
problems are also investigated in [39]. Only the three-dimensional results are given in
[25], where a duality argument is employed to obtain residual-based estimates. How-
ever, the velocity bounds derived there depend on a saturation assumption that may
not hold in general.
In this paper we derive a posteriori error estimates that provide lower and upper
bounds for the pressure, velocity, and mortar error in two and three dimensions.
According to the widely accepted terminology, an estimator is referred to as reliableif
it provides an upper bound of the error, whereas it is called efficient if it gives a lower
bound. We employ a duality-type argument to obtain an efficient and reliable residual-
based estimator for the pressure error. In addition to the usual element residual
terms, the estimator involves a flux-jump term and a mortar pressure difference term
on subdomain interfaces. A closely related estimator of the velocity and mortar error
is also derived, which provides an optimal upper bound, but suboptimal (yet sharp)
lower bound. We then proceed to derive an optimal efficient and reliable implicit
estimator for the velocity based on solving local (element) problems in a higher-order
space. Throughout the paper we make several reasonable saturation assumptions
which are motivated by known a priori error estimates.
It was observed in [36, 29] that varying the mortar degrees of freedom while
keeping the subdomain grids fixed has a substantial effect on the convergence of
the interface algorithm employed to solve the algebraic system. At the same time
mortar grids that are too coarse lead to deterioration of the accuracy of the method.
Therefore, finding "optimal" mortar grids for given subdomain grids is an important
A POSTERIORI ESTIMATES FOR MORTAR MIXED METHODS 1023
question. The flux-jump term that appears in all estimators provides a stand-alone
indicator of the nonconformity error in the mortar discretization. It can be used to
drive an adaptive mesh refinement process for the mortar grids.
The rest of the paper is organized as follows. In the next section the mortar
mixed finite element method is defined along with its equivalent interface formula-
tion. In section 3, the residual-based error estimators are derived and analyzed. The
implicit estimator for the velocity is developed in section 4. Computational results
are presented in section 5, followed by some remarks and conclusions in section 6.
2. Formulation of the method and preliminaries. We will make use of the
following standard notation. For a subdomain G c Rd, the L2(G) inner product (or
duality pairing) and norm are denoted by (., )Gand
) I1 1G, respectively, for scalar and
vector valued functions. The Sobolev spaces Wk (G), k E R, 1 < p < oc, are defined
in the usual way [1] with the usual norm I1- Ilk,p,G. Let I1Il-k,G be the norm of the
Hilbert space Hk (G) = W2k(G). We omit G in the subscript if G -= . For a section
of a subdomain boundary S c UnlOQ0i we write (., -)s and I I|s for the L2(S) inner
product (or duality pairing) and norm, respectively.
We assume that problem (1.1)-(1.4) is H2-regular, i.e., there exists a positive
constant C depending only on K and Q such that
(2.1) IIPl12< C(llfll + 1913/2,rD)

We refer the reader to [23, 26, 21] for sufficient conditions for H2-regularity.
To give the weak formulation of (1.1)-(1.4) we recall the usual velocity space [16]
H(div; ) = {v E (L2(Q))d : V E L2(Q)}

with a norm
= (IIVH
IIVIIH(div) 2 + IIV V111/2
and define
V - {v E H(div; Q) :v- v = 0 on FN}.
A weak solution of (1.1)-(1.4) is u C V, p E L2(Q) such that
(2.2) (K-lu, v) = (p, V. v) - (g, veV,

v'-Fv)rD,
(2.3) (V. u,w)= (f,w), w E L2(Q).
It is well known (see, e.g., [16, 32]) that (2.2) and (2.3) have a unique solution.
Let Q = UnL1Q, be a union of nonoverlapping subdomains. Let
n
Fi,j O&Qi
O~aj, F r= ai\aQ.
i = -i n F
i
Let
n
- -
Vi =- {vE H(div; i): v vi E L2(o0i) and v u= 0 on 0i nN}, V =- Vi,
i-= 1
M=L2(F).
Wi=L2(ni), W=OWi-=L2(),
i= 1
It is easy to see that if the solution of (2.2) and (2.3) satisfies u - vr E L2(F) and
p E H1( ), then for 1 < i < n
(2.4) (K-lu, v)Q, = (p, V -v)q, - (A, v - vi)ri - (g, v. v)aninrD, V E Vi,
(2.5) (V - u, w)ni = (f, w) Q, w E Wi,
(2.6) =0, PCEM,

i=1
E(u'vip)r
where A = pir. (2.4)-(2.6) imply that (u,p,A) E V x W x M satisfy
(2.7) A(u, p, A;v, w, p) = L(v, w, ) V (v,w, ) E V x W x M,

where
A(u, p, A;v, w, u)
n
+ (A, v - Vi)r, + a(V. u, w)Q, - a(u - vi,

=i= 1 ((K-lu, v)Qi - (p, V . v)ni p)r,)
and
L(v, w, [) = a(f, w) - (g, v - V)rD .
Here a = 1 or a = -1. If a = -1, A(.; -) is a symmetric bilinear form, which we

denote by A'(.; .). If a = 1, we denote A(.; -) by Ac(.; -) and note that
AC(v, w,I; v, w,I) = (K- v, v);
thus Ac(.; -) is nonsymmetric, but coercive. Note that the solution does not depend
on the choice of a.
Let {Th,i}h be a family of finite element partitions of Qi, 1 < i < n. Let, for any
E E Th,i, hE = diam(E) and let
hi = max hE, h = max hi.

EETh,i 1<i<n
Define PE to be the largest diameter of a ball contained in E. We require that each

subdomain grid satisfies the nondegeneracy condition
hE
max - < co,
EETh,i PE
where the constant co is independent of hi. The partitions Th,i and Th,y may be
nonmatching along Fi,j. Let Th = U= 1Th,i and let Sh be the union of all interior
edges (faces) not including the interfaces and the outer boundary. Let
Vh,i x Wh,i C Vi X Wi
be any of the usual mixed finite element spaces (i.e., the RTN spaces [34, 31, 27],
BDM spaces [15], BDFM spaces [14], BDDF spaces [13], or CD spaces [18]). The
order of the spaces is assumed to be the same on every subdomain. Let
n n
Vh= Vh,i, Wh= Wh,i-

i=1 i=1
Note that this choice leads to a nonconforming approximation since the normal com-
ponents of vectors in Vh do not have to be continuous across F. Throughout the paper
we will abuse notation when using the H(div)-norm. In particular, for v E H(div; Q•),
i=- 1,...,n,
( n 1/2
=
IIVIIH(div) (lVl2 VI I2
iv"
and for v E H(div; E), E E Th,
=
IIVIIH(div) (I2 E 6"Th
v v
For all of the above spaces

- = Wh,i
V Vhi
and there exists a projection operator 1hi of (H1 (Qi))d onto Vh,i satisfying for any
q E (HI(Qi))d
(2.8) (V. (IIh,iq - q), w) = 0, w E Wh,i,

(2.9) ((q - Hh,iq) iV,v vi)Oa
= 0, v E Vh,i.
Let IIh : ((Hl(Qi))d -- Vh be such that IIhqsn = HIh,iq for all q E ((H1(i))d.
Let the mortar interface mesh Th,i,j be a quasi-uniform finite element partition
of Fi,j and let Tr,h = Ul<i<jnTh,i,j. For any 7 E Th,i,j, let
E, = U(E E Th : n 7 0).
"E
We will assume that there exist constants ci and c2 such that
(2.10) h7 ? C2hE VE E E,,

ClhE
_
where the notation hs = diam(S) is used. Denote by Mh,i,j C L2(Fi,j) the mortar
space on Fi,j containing at least either the continuous or discontinuous piecewise
polynomials of degree k + 1 on Th,i,j, where k is associated with the degree of the
polynomials in Vh v. More precisely, if d = 3 and e is a triangle of the mesh, we
take Mh,i,jie = Pk+1(e), the set of polynomials of degree less than or equal to k + 1
on e. If e is a rectangle, we take Mh,i,jie = Qk+l(e), the set of polynomials on e for
which the degree in each variable separately is less than or equal to k + 1. Now let
Mh = • hij
1<i<j<n
be the mortar finite element space on F.

In the mortar mixed finite element method for approximating (2.4)-(2.6) we seek
Uh E Vh, Ph E Wh, and Ah E Mh such that, for 1 < i < n,
(2.11) (K-1uh, V)i = (Ph, -v)Qn - (Ah, V- vi)r, - (g, v - i)gninrD, V Vh,i,
V"
(2.12) (V. uh, w)> = (f,w)Si, w E Wh,i,
n
(2.13) = 0, t E Mh.
(Uha.i, p)r
i= 1
1026 MARY F. WHEELERAND IVAN YOTOV
(2.13) enforces weak (with respect to the mortar space Mh) continuity of the flux
across the block interfaces. Existence and uniqueness of a solution of (2.11)-(2.13)
are shown in [40, 3] along with optimal convergence and superconvergence for both
pressure and velocity under the assumption that for all E Mh,i,j there exists a
constant C independent of h such that
(2.14) IHu|r,S C(IlQh,iIIIpi,

+ IlQh,jIll ),
where Qh,i : L2(dQi) -+ Vh,i - vi lan is the L2-orthogonal projection satisfying for any
q E L2((Qi)
(2.15) ( - Qh,i?, V vi)r, = 0 Vv E Vh,i.

Remark 2.1. Condition (2.14) imposes a limit on the number of mortar degrees
of freedom and is easily satisfied in practice [40, 28].
We recall some a priori error estimates from [3] which will later motivate some
of the saturation assumptions needed in the a posteriori error analysis. Herein 1 is
associated with the degree of the polynomials in Wh and HjI dhis a mortar space norm
defined in the next subsection. Throughout the paper C denotes a generic constant
independent of h.
THEOREM2.1. For the solution of (2.11)-(2.13) if (2.14) holds, then
n
1IV(u - uh)ll C i= 1 IV"u| , hr, 11<r

< r < +1,
n
- UhI C + 1 <r < k + 1,
Iu E(plr+l,i
i= 1
IIUllr,ni)hr,
n
+ HV. ullr,Qi)hr, 1< r < min(k + 1,1 + 1),
Pp- Phl ? CZ
+
(lplr+l,Qi |lullr,i
i= 1
n
IIA-Ahldh ? C E(|pIr+l,ui + 1 <r < k + 1.

i= 1
IuIlr,Qi)hr,
2.1. Interface formulation. Method (2.11)-(2.13) can be reduced to an equiv-

alent interface (mortar) problem. We recall this interface formulation from [22, 40, 3],
as it will be used in estimating the mortar error.
Define dh : L2(r) x L2(F) -+ R for o, / E L2(F) by
n n
(2.16) dh(,) = = vi,/4r,
i=1 dh,i(,Oit) -Z(u()"
i=1
where (u h (p),
UhY/P\Y/~ ht p*(p)) E Vh
V x Wh solve, for 1 < i < n,
-
(2.17) (K-lu({p), v)n> = (p*(Qp),V . v)ni - (, v i)r, v E Vh,i,
(2.18) = 0, w e Wh,i.
(V. u*(c), w)i
Define gh : L2(F) -- R by
n n
gh(A)= h
Zgh,i(A)Z=1i
i=1 .iI
i=1
A POSTERIORI ESTIMATESFOR MORTARMIXED METHODS 1027
where (uih,ph) E Va x Wh solve, for 1 <i < n,
(K-1 ih, v)Q = (Ph, -v)nQ - (g, v - V Vh,i,

wV vWi)o49nFD,
(V.Uhi,w)nQ = (f,w)Qi, w E Wh, .
Then (uh, ph, Ah) satisfies
dh(Ah, p) = gh(p) Vp E Mh, Uh = uh(Ah) + uh, ph = p(hAh) + h.

It is easy to see from (2.16) and (2.17) that
(2.19) dh,i((P, P)=(K-lu uh(p))

(p), ,
which implies that dh(-, ') is positive semidefinite in M x M and, assuming (2.14),
positive definite in Mh x Mh. We define the norm in Mh:
:= dh(P,A)1/2
IIlldh
It is shown in [40, 28] for RTo rectangular elements and very general hanging interface
nodes and mortar grid configurations satisfying (2.14) that
(2.20) I Cdh4tL) Vu C Mh.

1hE)TFh 1•111H/2,,
The proofs in [40, 28] can be generalized in a relatively straightforward way to the
other mixed finite element spaces under consideration and to higher-order elements.
The following construction will also be useful in the analysis of the mortar error.
Define, for WE L2(F),
Uh((P) = u((P) + uih, ph((P) = Pp((P) +Pjh.

We note that (Uh((P),Ph(
()) E Vh x Wh satisfy, for 1 < i < n,
v)ni= (ph(),V . v)ni- (sO,
(K-1uh(p), v. v)r]
(2.21) - (g, v - )aV) nrD,I VVh,i,
(2.22) (V. Uh((P),w)RQ = (f, w E Wh,i.
w)•,
In particular, Uh(Ah) = Uh and ph(Ah) = Ph.
The a priori error bounds from Theorem 2.1 motivate the following assumption
on the mortar error.
Saturation assumption. There exists a constant 7ysuch that
(2.23) IIIA- z
,AhI " hl"-IA-AhH112•
•-ETr , _<-YU/
,h Uhll.
For further justification of (2.23), note that IIIA- Ah is closely related to the discrete
H1/2(F) norm and, by (2.20), to IIA- Ahlidh. Now, assumingthat
-
IlU Uh(A)|| YIU - Uhll,
<-
which is reasonable, since Uh((A) is the numerical solution based on the true interface
data, we have, using (2.19),
CIIA ? Ihuh(~)
- AXhldh -
- u(+h)ll= uUh(Ah)ll
Ituh(A) = Iluh(A)
- uhl
< Ilu- Uh(A)l1 - uhll.
+ Ilu- uhll< (1+ Y1)IIU
Remark 2.2. Condition (2.14) is necessary for the solvability and accuracy of
the method and for the validity of (2.23). See [40, 28] for examples of grids that
satisfy (2.14). Note that (2.14) excludes the case of matching subdomain grids and
a mortar grid that coincides with them. In the case of matching subdomain grids,
the mortar grid has to be at least twice as coarse as their trace on the interface.
Another possibility in the case of matching grids is to use the standard Lagrange
multipliers from the hybrid mixed method [4], in which case the conforming mixed
method solution is recovered. This trivial case is not a special case of the mortar
mixed finite element method, since the mortar spaces consist of polynomials of one
degree higher than the Lagrange multipliers.
2.2. Residual representation and orthogonality of error. Using the nota-
tion from (2.7), the solution of (2.11)-(2.13) (uh, ph, h) E Vh x Wh x Mh satisfies
(2.24) A(uh, ph, Ah;V,w, p) = L(v, w, p) V (v, w, p) E Vh x Wh x Mh.
Our goal is to derive a posteriori estimates of the error functions
= U -Uh, r = P - Ph, and 6 = A - Ah.
Using (2.7), ((, rl,6) E V x W x M satisfies the residual equation
(2.25)
A(, r, 6; v,w,p) = L(v,w, p) -A(uh,ph, Ah;V,W, ) V (v, w,) EV W x M,
which, together with (2.24), implies the orthogonality condition
(2.26) A(?, rl, ; v, w, I) = 0 V (v, w, p) E Vh x Wh X Mh.

2.3. Approximation properties. We present below some of the approximation
properties of the finite element spaces. In addition to the operators defined above, we
will make use of the interpolant 1h in the mortar space Mh, and the L2-projection
onto Wh, defined as
(w-w, h) = 0 Vw Wh.
The following approximation properties hold true. For all E E Th, - E Tr',h, e
Th,ilaQ,, and smooth enough functions v, w, and p,
(2.27) -IhVIIE ? ChElVII1,E,

IIv-
(2.28) II(v- HhV) . VEIE ? ChEIIV- VE
ls,E, S = 0, 1/2,
(2.29) IIw- lEJ? ChEIIWI1,E,
(2.30) I|l - r Ch3/2II113/2,7,
ZhI•l1-?
(2.31) Il| - Qh,iltll Chellllll,e.
Bound (2.27) can be found in [16, 33]; bounds (2.28)-(2.31) are standard interpolation
and L2-projection approximation results [19].
2.4. Some useful inequalities. In the analysis below we will make use of the
trace inequalities
(2.32) VEE Th, e E, l olle C(h-1/211 hlE/2+h( IV E) VEH$(E),

•
(2.33) VE E Th, eE DE, 11l1/2,e -<CII1,E V4E H(E),
(2.34) VE E Th, eE E, - ? Vv E Vh,

Iv ChE1/2IVIIE
v'e
and the well-known inequality
(2.35) 1
ab < ea2 + Ib2 Ve > 0.
4E
3. Residual-based error estimators. In this section we derive upper and
lower bounds on the error in terms of local residuals. The resulting estimators are often
called explicit estimators as they involve only residual terms that depend explicitly on
the input data and the computed solution and do not require the solution of additional
finite element problems.
3.1. Upper bounds. Let, for all E C Th,
E Tr',h,
(3.1) w = K-
+Kuh ph
VphiEh2E + lf - V uh 2 -p2hE
+
(3.2) w2 _v]112h,
-= I[uh
where for any v E V, vIQ, = vi,
[v - ]lr|, = vi - vi + vj j
is the jump operator. We first derive an upper bound on the pressure error i.
THEOREM3.1. There exists a constant C independent of h such that
1W112
Cc ? r,
+ 11-
EETh E Tr,h eEThIrD -Qhgl'he}
Proof. The proof is based on a duality argument. Consider the auxiliary problem
-V - KVw = T in Q,
w= 0 on FD,
KVw v = 0 on FN.
The elliptic regularity assumption (2.1) implies that
(3.3) |Iw12< ClI771.

Let v = -KVw and p = wir. With (2.7), (v, w, p) satisfy
A8(v, w, p; i, w, - (r, w) V (9, w, ) E V x W x M.

,4)=
Then, using (2.26) and (2.26),
1|112= -A'(v, w,
pW; 7i , , 6) = -A8'(?, , 6; v, w, A)
= -AA"(?, 6,6; v - IIhV,w - tb, A - ZThA)
= A'(uh, ph, Ah;V - IIhv, w - - ) + (f, W -- w) + (g, (v - IIhv) - v)rD
-+, p
- v - nhv)E-
1 ((K-1Uh, (ph,V.(V- ~hV))E
- -J,)E)
EETh (V.Uh•,
+? -
, (v HhV) - Vi)r + (Uh . Vi, A -
((A/
i= 1 hhvP)r)
+ (f, w -w'tZ) + (g, (v - HhV) . V)rD.
Applying Green's formula and using (2.9),
1112 EETh ((K-'uh + Vph,v - IhV)E + (f - Uh,W-)E)
+ -
5((A pv
-Ph,(V -InhV) v)rip Vi/ -
(Uh hVi)
i= 1
+ (9g- Qhg, (v - IIhv) . V)rD.
Using the Cauchy-Schwartz inequality and the approximation properties (2.27)-(2.31),
C + VPhIlEhE ?
Ir,12 EETh
IVII1,E if- V. UhEiihE WI1,E
(]jK-luh
+ -
Iv1,E)+ l[Uh -/2 3/2,
IAh Phl
EnFrhl• rETr ,h
+ 5 g-
eETh IrD
QhglllehU/2flVl/2,e}.
An application of the discrete Cauchy-Schwartz inequality, the trace inequality (2.33),
and (3.3) completes the proof. O
Remark 3.1. Because of the approximation property (2.31) of Qh the last term
in the bound of Theorem 3.1 is of higher order than the other terms. Therefore, its
effect becomes negligible for small h.
To derive a bound on = u - uh we need a saturation assumption. Let V',
Whi, and M, be the finite element spaces of one order higher than Vh, Wh, and Mh,
respectively. Let u' E V', p' E W/, and A' E Mh be the mortar mixed finite element
solution in these higher-order spaces (see (2.11)-(2.13)). The a priori error estimates
from Theorem 2.1 motivate the following.
Saturation assumption. There exist constants 3 < 1, Odiv< 1, and /p < 00osuch
that
(3.4) Ilu- U'll< P|U- Uhl,

(3.5) IV (u- u')IIOdivJllV(U- Uh)ll
- p
(3.6) lip p'h1| pllp-
phl,.
Let
77 = - ph, and 6' = A - Ah.
SUh -Uh, Ph
Similar to (2.26) and (2.26), we have that ( 6',i', 6') E V' x W, x Mh~ satisfy the
residual equation
A((',h'6; , Wh,p) = L(v W, -

Wh,p'h) A(Uh,pi , Ah;Vh,Wh,p' I)h
(3.7) hhlv V (V wh, /1') ( E V' X
Wh,
x
Mh
?whh
and the orthogonality condition
(3.8) A((', r', 6'; v, w, p) = 0 V (v, w, ) E Vh X Wh X Mh.
The bounds on ( and 6 will be expressed in terms of weighted local residuals, for
all E E Th, 7 E Tr,h
2
E
= h-2WE IK-1Uh+ Vph II2
E 1f- V. uh lE + IAh-- phil2oEnhE1,
E
E
-2 =
T h-2W2 = [rUh ,V112
11
II[UhII"hr. h,.
THEOREM3.2. Assume that the saturation assumptions (2.23) and (3.4) hold.
Then there exists a constant C independent of 0 such that
g112 c + 2+
HH(div)~1
<- S +V2
E)2EET-h E Th 9l -- Q 112
TETr,h eEThIrD
Proof. The bound on IIV i is trivial. Indeed, for all E E Th,

•-
= Ilf-V -UhIlE
lIV?J(IE <E.
?
To bound JII(|,since (3.4) implies that
1
(3.9) <51-P '11,
i 11
it is enough to bound i'|l'i.Using (3.8) and (3.7),
-1/2 = -
IIK 112 Ac(?', 7',6'; 'I',n', 6') = Ac(', r/, 6';(' Hh ', 77,6')
= LC ' - Il', r', 6') - Ac(uh,p h, Ah; ' - H', '/, 6')
u
=- S
E ETh
((K-u,, ' - Hlh')E
- (Ph,V. (s' - l'))E + (V. Uh,r')E)
- ((Ah, -vi)r -
(7'- (uhV ]i)
i=1 - ( H•') ••
+ (f, T') - (g, (~' - IIh')
?V)rD.
The use of Green's formula and (2.9) gives
5'I2 = - 5
IIK-1/2 ((K-luh + Vph,(' - IIHh')E+ (V -Uh - f, r)E)
EETh
(3.10) - ((- Ph, ( - i) r - (Uh

i= 1 vi, 6')r)
- -
(g- Qhg,(I - Hi') V)rD= T1+... + T5.
For T1, using the Cauchy-Schwartz inequality, (2.27), the inverse inequality, and
(2.35), we have
(3.11) + ' - IInh)EI C uh + VPhlE + E )

j(K-luh Vph, (41 I
Similarly for T2,
1 1
- f,'77)EI fi +
(3.12) I(V -Uh JllV Uh - 2
•I li
To bound T3, the use of (2.28) with s = 0 gives, for e E ri, e E OE,
(3.13) i ('
(A Ph, - I') ~-i) e ? - PhlhE-
Pli2
Similarly for T5,
(3.14) (g - Qhg,
(9 n' ).~)~ c(
- 4 -
ii
Qgllh2-+51
Finally for T4, using (2.35),
n
?
(uh -v1,/') 3([u1 v], )
6'),<
7 h'/211[Uh.v7h-1/211
(3.15) i=1 rEr,h .reTr,h
. v]112h + 641212h--1
(l |[u
-5
IEIr, h
Combining (1.5) with (3.10)-(3.15) for small enough q1, E3, and E5,
(3.16)
I'll,2 c{ (lK-uh + VPh~ + Ilf- V.-Uh2 + -Ph hEn-1
' 2)
EE Th I,•h
1
+ : I-v]lIrh+ 646 h--Qh9 hh1}-
rETr,h [u+ eETh rD
Because of (3.6), the bound on Ilr71l from Theorem 3.1 applies to ITI'llIas well. It
remains to estimate 1116'1112
-= ••Trr,h Using (2.23) (with a constant y' in
]5'112h-h1.
the case of the higher-order spaces) and (3.4), we have
(3.17) < IlA- AhIll

(116'1I + I1A
- ? ~YlU ?
- Uhl + - u'hII
'lllU
- uhl.Alhl-
< (Y+ p)J)lu
Using (3.17), (3.9) and taking E4 in (3.16) small enough completes the proof. O
3.2. Lower bounds. Next, we establish lower bounds on the error, which in-
dicate that the residual error estimators can be used effectively in an adaptive mesh
refinement algorithm.
THEOREM3.3. There exists a constant C independent of h such that
(3.18) 5UE + W2< C II,2 + Eh2i(div;E) + h~l611•7

EETh rETr,h EETh 7E+Tr,h
and, assuming that the saturation assumption (2.23) holds,
(3.19) + E H+
C( b~(div))
EETh
• TrETr,h EETh
• ]
hE2[-
Moreover, the following local bounds hold for any E E Th, e GE, and r E Tr,h:
2
(3.20) IlK-luh+ Vphl2h2+ Ilf- V uh112Eh2C(riH12E
I+ lg(div;E)h2E)
(3.21) . v]l2h3 C|
II[uh (|2HH(div;E,)h-,
(3.22) - phl2hE C(IIT2E

IlAh + lll2H(div;E)h2E
+ 11611hE)
El 1 El
71 73 75
E271 E2T3 E27T5
FIG. 3.1. Construction of

Erk.
Proof. It has been shown in [17], using a bubble function argument, that
IIK-uh + Vph EhE ? C( Iril E + H

?JjEhE),
which, combined with
Ilf - V - Uh IEhE
--IIV" IIEhE,
gives (3.20). To prove (3.21), consider any T E Tr,h. Let T be divided by the
intersection of the two traces of Th on F into elements T1,..., 71. Because of (2.10)
there exists c > 0 such that
(3.23) hk > ch-, k=1,. ..,1.
Next, let us translate any point in Tk in both directions orthogonal to F until an

interior edge (face) of an element of Th is reached. Let Elk be the union of all such
trajectories. Figure 3.1 illustrates this construction in the case of triangular grids in
R2, where the neighboring domains are Q1 and Q2. Note that
= E1Ik U E 7k ,
Ek
where EEk, i - 1, 2, is a subset of an element of Th,i. Let Pk be a continuous piecewise
linear bubble function such that 0 <pk (x) 1 in Ek, Ipk= 1 at the gravity center
of Tk and pk = 0 on OE,,. Such a function can be easily constructed by decomposing
E.k into triangles if d = 2 or tetrahedra if d = 3. We also need an extension of
[Uh V],-k
to Given 7 E H1/2 (Tk), define 1b E Hl(Elk) such that ROZis constant
Ek,.
along lines perpendicular to F. Let
kE -'
kP]R[Uh
-k HI(Erk).
Note that /k = 0 on OE,,. Using a scaling argument similar to the one in [35] it can
be shown that
(3.24) Ikl-k - I[Uh"VIl-k,

(3.25) IIVkE, < Ch-1IkE
(3.26) |kEk Ch21 kI kl-k
(3.27) cll[Uh-] | < (k, [Uh'VI)-k.
1034 MARY F. WHEELERAND IVAN YOTOV
Using (3.27) and that [u - v] = 0,
[Uh.iV]I< (h,I
(3.28) u /1 Uh,2' V2,
.1k)Tk
- - -
= ((Uh,i u) Vl, + ((Uh,2 U) V2,
k)Trk k)irk'
Using Green's formula for the first term on the right-hand side, we have
E + (Ok I
K ElE
S(E1" /21,
•C)kk I(V•)k,•lE + V"
V4kflE +k-](kilE
H1iE1HE k 11V-W E1k
(3.29)
EJlll I•1I•
SCh;]k k + Ii llESEP"11E•
K C(h;1/2 + h/21V '6 Elk)[Uh
1i.4•li i•"]lk
where we have used (3.25) for the second inequality and (3.26), (3.24) for the third
inequality. The second term on the right-hand side of (3.28) can be bounded similarly
in terms of 11|211H(div;E2k). A combination of (3.28), (3.29), and (3.23) gives (3.21).
It remains to show (3.22). By the triangle inequality,
h <
I|Ah - phle - + ip - Phlle.
(3.30) IIhAlleh
For the second term on the right-hand side we employ the trace inequality (2.32)
(3.31)
p - Ph e < C(hE1/2p PhHE ? h'2HV(p - Ph)E)
?C(h1/2Jip + h 2IIK(u - Uh)E)

+ VPhIE
+ h/2IIK-luh
- PhIE
? C(h-1/2llrllE
+ hlj112IIH(div;E)),
using (3.20) for the last inequality. A combination of (3.30), (3.31), and (2.10) com-
pletes the proof of (3.22). The global bound (3.18) follows immediately from (3.20)
to (3.22), using (2.10), and so does (3.19), using (2.23). O
Remark 3.2. The last two terms in (3.18) are of higher order, so JIT11|dominates
for small enough h. Therefore, this bound, combined with Theorem 3.1, implies that
EEThWE + Z ,Th W2 is an efficient and reliable estimator for the pressure error.
Because of the negative power of h in the first term on the right-hand side of (3.19),
the estimator ZEETh + •rCTr,h W~ provides only a suboptimal bound for the
velocity error.
4. Error estimators based on solving local problems. In this section we

derive an implicit error estimator which requires solving local (element) boundary
value problems. These problems approximate the local residual equations satisfied
by the true error. The motivation for considering implicit estimators comes from
the unknown generic constants that appear in the explicit estimators, as well as the
suboptimality in the lower bound for the velocity error. We show that the implicit
estimator provides both optimal upper and lower bounds of the error.
4.1. Global approximation to the error. Similar to the approach in [39],

we first construct a global approximation to the error based on higher-order finite
element spaces. Using (2.4)-(2.6), the true error satisfies the residual equations:
(K-'1, - (i, V - v)Q, + (6, v - = -(g, v v)aninrD

Vi)ri
v)•, - - r(v),
(4.1) - (K-1Uh, v)ni + (Ph, v)n, - (Ah,v Vi)ri v E Vi,
V.
(4.2) Qw)j = (f - V.Uh,, w), w
Wi,
(V.?*,
(4.3) (V i,/j)r = - (u -Vi, P)ri, P E M.

i=1 i=1
Recall from the previous section that V' x Wh x M' are the mortar mixed finite
element spaces of one order higher than Vh x Wh x Mh and (s', r', 8') E V' x Wh M
x?
is the finite element approximation to (6, T, 8) satisfying
(4.4) (K-(', v)n, - (r;',V . v)nQ++ (', ,vv vi)ri

i)1-i=r = r(v),
V
,h,i• v E V
(4.5) (V. ', w)Q = (f - V. Uh, w)n, w E Wi
(4.6) i
-Z(U .V
h .) EM .
•=iip)r
i=-1 i=1
Note that (4.4)-(4.6) implies that (u' = Uh + A',p' = Ph + r7',A' = Ah + 6') is the
finite element approximation to (u,p, A) in V' x W1 x M/ satisfying
(4.7) (K-u', v)n, (p, V. v)Q,- (A',v -) - (, Vvi)a

- nro, VEV'h,
(4.8) (V - u', w)o = (f, W)2, W We,i
(4.9) 1(u' M-.

vi, p)r = 0, e
i=-1
The saturation assumptions (3.4) and (3.5) imply
(4.10) (1 - 1/)?||| |(' | (1 + 3)I1•I1,

(4.11) _-
(1 - ~div)llIIV ?_ I .
(1 + /div)llIV-11,
IV"
'<l
so it is enough to estimate ('.
4.2. Local (element) approximation to the error. For any E E T, the true
error satisfies the local equations:
(4.12) (K-Q?, V)E - (rT,V . V)E T= E(V) - (p, V . VE)aE, V C V(E),

(4.13) (V ??, W)E = (f - V
.Uh,
W)E, W E W(E),
where
rE(v)= -(K-1Uh, V . V)E

V)E+ (Ph,
We construct a higher-order local approximation of the error by solving element sub-
problems: find )' e V' (E) and 0' e W (E) such that
(4.14) (K-V4', v)E - (0', V V)E - rE(v) - (pA, V. VE)OE, V E V(E),

.
(4.15) (V - /',W)E = (f - V . Uh, W)E, W E Wh(E),
where PA - g on FD, PA = Ah on 0E n , and PA P= h on OE n Sh, where Ph is the

Lagrange multiplier for Vh and Wh defined as
(4.16) -
(ph, v VE)aE = -(K-lUh, V)E, v E Vh(E).
V)E + (Ph,
V"
Let p' be the Lagrange multiplier for the higher-order spaces V' and Wh7satisfying
- VE)&E = v)E + (p', V V)E, v
(4.17) (3', v -(K-1h, E V'(E)
We make the following.
Saturation assumption. There exists a constant c such that
- -
(4.18) h eu Uhl
he-l'lp
eEEh
Assumption (4.18) is motivated by the a priori error estimate for the Lagrange mul-
tiplier [16]
?
(h2)
eESh
hlllp- 12 Chk+l
where pi is the L2-projection of p onto Vh -

vblE.
THEOREM4.1. Assume that the saturation assumptions (2.23), (3.4), (3.5), and
(4.18) hold. Then there exist constants C1 and C2 independent of 0 and 3div such
that
+ H(div)
Cl(t|' /lH(div) I vhi)7hl /2)
rh [[Uh
(4.19) "
- H(div) + [u
1 ]rhl/2
- 7ETrI,h
max
where max = max{0, Odiv

" (4.5) and (4.15) imply that on every E E Th,
Proof. We first note that
(4.20) f' = /'.

V. V.
Taking v = 0' - (' in (4.14) and summing over all elements, we have
(4.21)
EETh
- (K-S',
(- 0' - )E + (77/ ' - ))E
EETh
+ rE(4' - ') - (PA,(' - VE)E)

'-
= (- (K-u, )E + (ph, V (' - '))E - hh- ( ') VE)8EnEh
EETh
V'
=- EETh(h' - Ph,('- (Aa
Eh+ - Ah,
(4'- 0 VE)aEnr),
VE)•
using (4.7) and (4.17) for the last equality. For the first term on the right-hand side,
using the saturation assumption (4.18) and (2.34), we have
(4.22)
(' - Ph, '
) ? h 1/21' P hleh./2i11(
E Th ('-0 ) e
') Vell-
VE)OEnSAh-_
Eh -
11 -
?-IIIN
It. II-
For the second term on the right-hand side of (4.21) we write, using (2.34), (2.23),
and (3.17),
5 (A/- ,
Ah)-i -) VE)&Enr= Z(6',(4" -
) .
EETh i= 1
(4.23) < h-;1/2 16' hl1/21[( ' - ') l.

,h "
TETr
1I1'111' - ' :-Cll
111 CRSI 11/'- 0'11.
Combining (4.21)-(4.23) and using (4.20), we obtain
0-1'<? Cfl~jl,
which implies, using the triangle inequality and (4.10),
(4.24) ?
11'11 C||(ll.
Taking w = V - 0' in (4.15) immediately gives
-
IIT. 111 <IIT.011
which, combined with (4.24), implies
1'10IIH(div)
CH' 'HH(div)-
Combining the above bound with (3.21) completes the proof of the left inequality in
(4.19). To show the right inequality in (4.19), taking v = ' in (4.4), and using (4.14),
we have
-
(4.25) (K '- ((' -(', v)r)
') ') i= 1 0', V ')
For the first term on the right-hand side of (4.25) we use (4.20) and the argument
from (4.21) to obtain
( '-'
i=1 .• ( '-',V )
EETh
-
(4.26) = (K-'(' -
'), ')
-
((P' Ph,' VE)&EnEh
EETh
+ (A' - Ah,4 VE)OEnO),
which, combined with (4.25), implies
(4.27) 4 -
_ ((P' -ph, ' +VE)OEqnE
+(Ah Ah, '.VE)&EnF).
E E Th
For the first term on the right-hand side we have, using (4.6),
(4.28) (6'luh E 4c,11[uh Ih

1(6•1,• T h
•Tr,
i-iE F
where the inequality is obtained using the argument in (3.15) and (3.17). The
last two terms on the right-hand side of (4.27) can be bounded similarly to (4.22)
and (4.23):
(4.29)
' ? h- V') i 2+
CC 2g12
E TF,
+
(?'-. . Eio + - ,h, ' " r) ___ II
Combining (4.27)-(4.29),
+
< C(( ci 2 E 111 I +
2
l[' • 2 E
.rET-F,h ?h 2l4E2 11
?-2)1 461rE[uh"4 ,
which implies, using the triangle inequality, (4.10), and taking E1 and E2ssmall enough,
( 'v 5
1 + ]]hi/2)
- [Uh
-<•1 rETrh
An application of (4.11) and (4.20) completes the proof. 0
5. Numerical results. In this section we test the performance of the residual-
based error estimator. The estimator is used as a local error indicator that drives
an adaptive mesh refinement process. The following algorithm describes the adaptive
procedure.
ALGORITHM.
1. Solve the problem on a coarse (both subdomain and mortar) grid.
2. For each subdomain Qi
(a) Compute
( E
h 1
E /h,i
EE rETri h
,
(b) If wi > 0.5 maxl<j<n wj, refine Th,i.
(c) If any neighboring subdomain grid has been refined two times more than
Qi, refine Th,i.
3. For each interface Fi,j, if either Qi or Qj has been refined, refine Th,i,j.
4. Solve the problem on the refined grid. If either the desired error tolerance or
the maximum refinement level has been reached, exit; otherwise go to step 2.
Several comments are in order. First, we employ the pressure error estimator
based on WE and we, defined in (3.1) and (3.2), since it provides an efficient and
reliable estimate of the L2 pressure error, due to Theorems 3.1 and 3.3. Second,
the refinement rule 2(c) is needed to reduce the effect of discretization error due to
large ratios between grid sizes in neighboring subdomains. Third, according to rule 3,
mortar grids are refined if either adjacent subdomain grid is refined.
In the examples below, the subdomains are discretized by the lowest-order
Raviart-Thomas spaces. Discontinuous piecewise linear mortar spaces are used on
the interfaces.
pres pres
1.713 17.1968
1.5988 16.0504
1.4846 14.9039
1.3704 13.7575
1.2562 12.611
1.142 11.4646
1.0278 10.3181
0.9136 9.1717
0.7994 _..I lls . 8.02525
0.6852 I•.•z•;• 6.8788
0.571 I :::5.73235
0.4568 - 4.5859
0.3426 : -•;•,m 3.43945
0.2284 m .__ 2.293
0.1142 1.14655
FIG. 5.1. Computed pressure on the fourth grid level for examples 1 and 2.
0.8
0.7
0.6
Convergence of Pressure Error 0.6
0.5
Convergence of Velocity Error
0.5
0.4
0.4
0.3
0.3
2 0.2 ---- Uniform 0.2 --- Uniform

A Adaptive - A-- Adaptive
2 w
0.1 :0. 1
10
10Number
Numberof 10Elements
of Elements 10Number of Ele10
Numberof 1ments
Elements
FIG. 5.2. Convergence of pressure and velocity error for example 1.
We first illustrate the above algorithm for several two-dimensional problems. In

all examples the domain is the unit square, decomposed into 6 x 6 subdomains. The
coarse grid in each subdomain is 2 x 2 with a single mortar element on each interface.
In the first two examples we test problems with boundary layers. The true pressure
solution is
= 1000 x y
p(x, y) e-k(2+y2),
where k = 100 in example 1 and k = 10 in example 2. In both cases K = I. The

computed pressure after three refinements is shown in Figure 5.1. We note that in
both cases the grids are appropriately refined along the boundary layers. In the
second example the exponential drop is less steep. This causes an extended boundary
layer, which is resolved by a strip of fine subdomain grids along the boundary. In
Figure 5.2, the pressure and velocity errors in example 1 are plotted as functions
of the total number of finite elements. The convergence of the error for all other
examples is similar and is not shown. We observe that the adaptive solution needs
about 20 times fewer elements to provide the same accuracy.
'In
pres pres
0.94 0.93
0.87 0.87
0.81 0.81
0.75 0.75
0.69 0.69
0.62 0.62
0.56 0.56
0.50 0.50
0.44 0.44
..
..0.38 ...0.38
0.31 0.31
0.25 0.25
0.19 0.19
0.13 0.13
S0.06 0.07
won owa
FIG. 5.3. Computed solution on the fourth grid level for example 3. Left: pressure on the full
grid. Right: pressure and velocity near the singularity.
pres
935.92
873.53
811.13
748.74
686.34
623.95
561.55
499.16
436.76 to
374.37
311.97
249.58
187.18
124.79
62.39
FIG. 5.4. Computedsolution on the fourth grid level for example 4. Left: pressure on the full
grid. Right: pressure and velocity zoom.
In the next example, motivated by the modeling flow in heterogeneous porous

media, we test a problem with a discontinuous permeability tensor K. The domain
is divided into four subregions by the lines x = 0.5 and y = 0.5. The permeability
is K = 1001 in the lower-left and upper-right regions and K = I in the other two
regions. Dirichlet boundary conditions p = 1 on the left and p = 0 on the right and
no-flow boundary conditions on the top and bottom force the flow from left to right.
It is known for the true solution that p E Hl+a for some 0 < a < 1 with singularity
occurring at the cross-point. The computed solution after three refinements is shown
in Figure 5.3. As expected the grids are finest near the singularity and are also refined
in the low permeability region to resolve the high pressure gradient. Some of the grids
in the high permeability region are refined as well, due to the refinement rule 2(c).
Finally, a three-dimensional example is presented. The unit cube is divided into
4 x 4 x 3 subdomains. The true pressure
p= 1000e-10(2+y2+z2)
exhibits a steep exponential decay near the origin. The computed pressure and
velocity on the fourth grid level are given in Figure 5.4. The steep pressure gradient
and large velocity are well resolved by the fine computational grids near the origin.
6. Conclusions. In this paper, several two- and three-dimensional a posteriori

error estimators for mortar mixed finite element methods for elliptic equations have
been derived. A residual-based error estimator provides optimal upper and lower
bounds for the pressure error. A closely related error estimator for the velocity gives
an optimal upper bound, but suboptimal lower bound. The negative power of h
that appears is due to the different order of derivatives involved in the L2-norm and
the H(div)-norm. An efficient and reliable implicit estimator for the velocity is also
derived, which is based on solving local (element) problems. All estimators include a
term that measures the jump of flux across subdomain interfaces. This term provides a
measure of nonconformity in the mortar discretization. In cases where the subdomain
grids are fixed and optimal mortar grids need to be obtained, this flux-jump term
can be used to drive an adaptive process for the mortar grids independently of the
subdomain grids.
Acknowledgment. The authors thank Wolfgang Bangerth for his useful

comments.
REFERENCES
[1] R. A. ADAMS,Sobolev spaces, in Pure and Applied Mathematics, Vol. 65, Academic Press, New
York, 1975.
[2] M. AINSWORTH AND J. T. ODEN, A unified approach to a posteriori error estimation using
element residual methods, Numer. Math., 65 (1993), pp. 23-50.
[3] T. ARBOGAST,L. C. COWSAR,M. F. WHEELER,ANDI. YOTOV, Mixed finite element methods
on non-matching multiblock grids, SIAM J. Numer. Anal., 37 (2000), pp. 1295-1315.
[4] D. N. ARNOLDAND F. BREZZI,Mixed and nonconforming finite element methods: Implemen-
tation, postprocessing and error estimates, RAIRO Modbl. Math. Anal. Numbr., 19 (1985),
pp. 7-32.
[5] I. BABUSKAANDW. C. RHEINBOLDT, Error estimates for adaptive finite element computations,
SIAM J. Numer. Anal., 15 (1978), pp. 736-754.
[6] W. BANGERTH ANDM. F. WHEELER,A Posteriori Error Estimates and Adaptivity for a Mixed
Mortar Method for the Laplace Equation, preprint.
[7] R. E. BANK ANDA. WEISER,Some a posteriori error estimators for elliptic partial differential
equations, Math. Comp., 44 (1985), pp. 283-301.
[8] F. BEN BELGACEM,The mortar finite element method with Lagrange multipliers, Numer. Math.,
84 (1999), pp. 173-197.
[9] F. BEN BELGACEM,The mixed mortar finite element method for the incompressible Stokes
problem: Convergence analysis, SIAM J. Numer. Anal., 37 (2000), pp. 1085-1100.
[10] C. BERNARDI,Y. MADAY, AND A. T. PATERA,A new nonconforming approach to domain
decomposition: The mortar element method, in Nonlinear Partial Differential Equations
and Their Applications, H. Brezis and J. L. Lions, eds., Longman Scientific and Technical,
Harlow, UK, 1994, pp. 13-51.
[11] D. BRAESSAND R. VERFilRTH,A posteriori error estimators for the Raviart-Thomas element,
SIAM J. Numer. Anal., 33 (1996), pp. 2431-2444.
[12] J. H. BRANDTS,Superconvergence and a posteriori error estimation for triangular mixed finite
elements, Numer. Math., 68 (1994), pp. 311-324.
[13] F. BREZZI,J. DOUGLAS,JR., R. DURAN, AND M. FORTIN, Mized finite elements for second
order elliptic problems in three variables, Numer. Math., 51 (1987), pp. 237-250.
[14] F. BREZZI, J. DOUGLAS, JR., M. FORTIN, AND L. D. MARINI, Efficient rectangular mixed finite
elements in two and three space variables, RAIRO Modbl. Math. Anal. Numbr., 21 (1987),
pp. 581-604.
[15] F. BREZZI, J. DOUGLAS, JR., AND L. D. MARINI, Two families of mixed elements for second
order elliptic problems, Numer. Math., 88 (1985), pp. 217-235.
[16] F. BREZZIAND M. FORTIN,Mixed and Hybrid Finite Element Methods, Springer-Verlag, New
York, 1991.
[17] C. CARSTENSEN, A posteriori error estimate for the mixed finite element method, Math. Comp.,
66 (1997), pp. 465-476.
[18] Z. CHENAND J. DOUGLAS,JR., Prismatic mixed finite elements for second order elliptic prob-
lems, Calcolo, 26 (1989), pp. 135-148.
[19] P. G. CIARLET,The finite element method for elliptic problems, in Studies in Mathematics and
Its Applications, Vol. 4, North-Holland, Amsterdam, 1978.
[20] R. EWING,R. LAZAROV,T. LIN, ANDY. LIN, Mortar finite volume element approximations of
second order elliptic problems, East-West J. Numer. Math., 8 (2000), pp. 93-110.
[21] D. GILBARGAND N. S. TRUDINGER,Elliptic Partial Differential Equations of Second Order,
Springer-Verlag, Berlin, 1983.
[22] R. GLOWINSKI AND M. F. WHEELER,Domain decomposition and mixed finite element methods
for elliptic problems, in First International Symposium on Domain Decomposition Meth-
ods for Partial Differential Equations, R. Glowinski, G. H. Golub, G. A. Meurant, and
J. Periaux, eds., SIAM, Philadelphia, PA, 1988, pp. 144-172.
[23] P. GRISVARD,Elliptic Problems in Nonsmooth Domains, Pitman, Boston, 1985.
[24] R. H. W. HOPPEANDB. I. WOHLMUTH, Adaptive multilevel techniques for mixed finite element
discretizations of elliptic boundary value problems, SIAM J. Numer. Anal., 34 (1997),
pp. 1658-1681.
[25] R. KIRBY, Residual a posteriori error estimates for the mixed finite element method, Comput.
Geosci., 7 (2003), pp. 197-214.
[26] J. L. LIONSAND E. MAGENES,Non-Homogeneous Boundary Value Problems and Applications,
Vol. 1, Springer-Verlag, New York, 1972.
[27] J. C. NEDELEC,Mixed finite elements in R3, Numer. Math., 35 (1980), pp. 315-341.
[28] G. PENCHEVA ANDI. YOTOV,Balancing domain decomposition for mortar mixed finite element
methods on non-matching grids, Numer. Linear Algebra Appl., 10 (2003), pp. 159-180.
[29] M. PESZYIrSKA, M. F. WHEELER, AND I. YOTOV, Mortar upscaling for multiphase flow in
porous media, Comput. Geosci., 6 (2002), pp. 73-100.
[30] J. POUSINAND T. SASSI, Domain decomposition with non matching grids and adaptive finite
element techniques, East-West J. Numer. Math., 8 (2000), pp. 243-256.
[31] R. A. RAVIART AND J. M. THOMAS,A mixed finite element method for 2nd order elliptic
problems, in Mathematical Aspects of the Finite Element Method, Lecture Notes in Math.
606, Springer-Verlag, New York, 1977, pp. 292-315.
[32] J. E. ROBERTSAND J.-M. THOMAS,Mixed and hybrid methods, in Handbook of Numerical
Analysis, Vol. II, P. G. Ciarlet and J. Lions, eds., Elsevier Science, Amsterdam, 1991,
pp. 523-639.
[33] J. E. ROBERTSAND J.-M. THOMAS,Mixed and hybrid methods, in Handbook of Numerical
Analysis, Vol. II, North-Holland, Amsterdam, 1991, pp. 523-639.
[34] J. M. THOMAS,These de Doctorat d'etat, 'a l'Universite Pierre et Marie Curie, 1977.
[35] R. A posteriori error estimation and adaptive mesh-refinement techniques, J. Com-
VERFiURTH,
put. Appl. Math., 50 (1994), pp. 67-83.
[36] M. F. WHEELERANDI. YOTOV,Physical and computational domain decompositions for model-
ing subsurface flows, in Tenth International Conference on Domain Decomposition Meth-
ods, Contemporary Mathematics, Vol. 218, J. Mandel et al., eds., American Mathematical
Society, Providence, RI, 1998, pp. 217-228.
[37] B. I. WOHLMUTH, Hierarchical a posteriori error estimators for mortar finite element methods
with Lagrange multipliers, SIAM J. Numer. Anal., 36 (1999), pp. 1636-1658.
[38] B. I. WOHLMUTH,A residual based error estimator for mortar finite element discretizations,
Numer. Math., 84 (1999), pp. 143-171.
[39] B. I. WOHLMUTHAND R. H. W. HOPPE, A comparison of a posteriori error estimators for
mixed finite element discretizations by Raviart-Thomas elements, Math. Comp., 68 (1999),
pp. 1347-1378.
[40] I. YoTov, Mized Finite Element Methods for Flow in Porous Media, Ph.D. thesis, Rice Univer-
sity, Houston, TX, 1996. TR96-09, Dept. Comp. Appl. Math., Rice University and TICAM
Report 96-23, University of Texas at Austin.

4101278

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

4101278

Uploaded by

Copyright:

Available Formats

A Posteriori Error Estimates for the Mortar Mixed Finite Element Method

Author(s): Mary F. Wheeler and Ivan Yotov

A POSTERIORI ERROR ESTIMATES FOR THE MORTAR MIXED

AMS subject classifications. 65N12, 65N15, 65N30, 65N50

1. Introduction. We consider the second order elliptic problem written as a

where C Rd, d - 2 or 3, is a multiblock domain with a boundary 0Q= FD U FN,

(1.5) ko0 T TK(x)(s <

(2.1) IIPl12< C(llfll + 1913/2,rD)

H(div; ) = {v E (L2(Q))d : V E L2(Q)}

V - {v E H(div; Q) :v- v = 0 on FN}.

A weak solution of (1.1)-(1.4) is u C V, p E L2(Q) such that

(2.2) (K-lu, v) = (p, V. v) - (g, veV,

(2.6) =0, PCEM,

(2.7) A(u, p, A;v, w, p) = L(v, w, ) V (v,w, ) E V x W x M,

+ (A, v - Vi)r, + a(V. u, w)Q, - a(u - vi,

L(v, w, [) = a(f, w) - (g, v - V)rD .

Here a = 1 or a = -1. If a = -1, A(.; -) is a symmetric bilinear form, which we

AC(v, w,I; v, w,I) = (K- v, v);

hi = max hE, h = max hi.

Define PE to be the largest diameter of a ball contained in E. We require that each

Vh= Vh,i, Wh= Wh,i-

For all of the above spaces

(2.8) (V. (IIh,iq - q), w) = 0, w E Wh,i,

(2.10) h7 ? C2hE VE E E,,

be the mortar finite element space on F.

(2.14) IHu|r,S C(IlQh,iIIIpi,

(2.15) ( - Qh,i?, V vi)r, = 0 Vv E Vh,i.

1IV(u - uh)ll C i= 1 IV"u| , hr, 11<r

IIA-Ahldh ? C E(|pIr+l,ui + 1 <r < k + 1.

2.1. Interface formulation. Method (2.11)-(2.13) can be reduced to an equiv-

where (uih,ph) E Va x Wh solve, for 1 <i < n,

(K-1 ih, v)Q = (Ph, -v)nQ - (g, v - V Vh,i,

Then (uh, ph, Ah) satisfies

dh(Ah, p) = gh(p) Vp E Mh, Uh = uh(Ah) + uh, ph = p(hAh) + h.

(2.19) dh,i((P, P)=(K-lu uh(p))

(2.20) I Cdh4tL) Vu C Mh.

Uh((P) = u((P) + uih, ph((P) = Pp((P) +Pjh.

(2.24) A(uh, ph, Ah;V,w, p) = L(v, w, p) V (v, w, p) E Vh x Wh x Mh.

Our goal is to derive a posteriori estimates of the error functions

= U -Uh, r = P - Ph, and 6 = A - Ah.

Using (2.7), ((, rl,6) E V x W x M satisfies the residual equation

which, together with (2.24), implies the orthogonality condition

(2.26) A(?, rl, ; v, w, I) = 0 V (v, w, p) E Vh x Wh X Mh.

(2.27) -IhVIIE ? ChElVII1,E,

(2.32) VEE Th, e E, l olle C(h-1/211 hlE/2+h( IV E) VEH$(E),

(2.33) VE E Th, eE DE, 11l1/2,e -<CII1,E V4E H(E),

(2.34) VE E Th, eE E, - ? Vv E Vh,

(3.3) |Iw12< ClI771.

A8(v, w, p; i, w, - (r, w) V (9, w, ) E V x W x M.

Applying Green's formula and using (2.9),

1112 EETh ((K-'uh + Vph,v - IhV)E + (f - Uh,W-)E)

(3.4) Ilu- U'll< P|U- Uhl,

A((',h'6; , Wh,p) = L(v W, -

Proof. The bound on IIV i is trivial. Indeed, for all E E Th,

(3.10) - ((- Ph, ( - i) r - (Uh

(3.11) + ' - IInh)EI C uh + VPhlE + E )

Finally for T4, using (2.35),

(3.17) < IlA- AhIll

(3.18) 5UE + W2< C II,2 + Eh2i(div;E) + h~l611•7

and, assuming that the saturation assumption (2.23) holds,

(3.22) - phl2hE C(IIT2E

E271 E2T3 E27T5

FIG. 3.1. Construction of

IIK-uh + Vph EhE ? C( Iril E + H