Volume 73, Number 246, Pages 659–689
S 00255718(03)015503
Article electronically published on October 17, 2003
SUBSTRUCTURING PRECONDITIONERS
FOR THE THREE FIELDS
DOMAIN DECOMPOSITION METHOD
SILVIA BERTOLUZZA
Abstract. We study a class of preconditioners based on substructuring, for
the discrete SteklovPoincar´e operator arising in the three ﬁelds formulation of
domain decomposition in two dimensions. Under extremely general assump
tions on the discretization spaces involved, an upper bound is provided on the
condition number of the preconditioned system, which is shown to grow at most
as log(H/h)
2
(H and h denoting, respectively, the diameter and the discretiza
tion meshsize of the subdomains). Extensive numerical tests—performed on
both a plain and a stabilized version of the method—conﬁrm the optimality
of such bound.
1. Introduction
The use of nonconforming domain decomposition methods is becoming increas
ingly popular. This is due to several factors: the possibility of easily coupling dif
ferent discretizations in diﬀerent subdomains without the need for imposing strong
matching conditions allows us to employ without any adaptation the technologies
developed for treating the problem with a singledomain approach. As an example,
in an adaptive strategy, reﬁnement can be carried out in each subdomain inde
pendently. Moreover, it is possible to use a diﬀerent type of discretization (ﬁnite
elements, spectral methods, wavelets) in the diﬀerent subdomains, and therefore
to employ, in each subdomain, the bestsuited method (spectral methods where
the solution is expected to be very smooth, ﬁnite elements where a complicated
geometry requires it, wavelets where isolated singularities in a regular background
are expected).
Among the diﬀerent nonconforming domain decomposition method, we consider
here the three ﬁelds formulation, as proposed by F. Brezzi and L.D. Marini in [14].
Diﬀerently from other nonconforming formulations [3, 18] of domain decomposition,
weak continuity is not imposed by requiring that the jump across the interface be
orthogonal to some multiplier space, but by introducing a space Φ
h
, approximating
the traces of functions in H
1
(Ω) on the interface Σ. The elements of Φ
h
are to be
used as a Dirichlet boundary condition — imposed by means of Lagrange multipli
ers — for the functions in the subdomains. As with other nonconforming methods,
great freedom is left in the choice of the discretization in the subdomains which,
Received by the editor November 6, 2000 and, in revised form, February 22, 2002.
2000 Mathematics Subject Classiﬁcation. Primary 65N55, 65N22.
Key words and phrases. Three ﬁelds formulation, domain decomposition, stabilized methods,
preconditioning.
c 2003 American Mathematical Society
659
660 SILVIA BERTOLUZZA
after applying a suitable stabilization method if needed [2, 7], can be chosen inde
pendently from each other. The disadvantage of introducing a third unknown ﬁeld
is compensated for by the fact that all subdomains are treated exactly in the same
way, which results in an easier implementation and possibly, when considering the
parallelization of the method, in an easier balancing of the load between processors.
We deal, in this paper, with the problem of the eﬃcient solution of the linear
system arising when applying such a method. By applying a Schur complement
technique, the solution of the resulting discrete problem can be reduced to the
solution of an equation on the “trace” unknown, involving a SteklovPoincar´e type
operator which can be shown to be a pseudodiﬀerential operator of order one.
The resulting linear system will then be solved by an iterative procedure, at each
step of which there will be the need of computing the action of such operator on
a given function. This reduces to solving independent Dirichlet problems on the
subdomains, a task which can be carried out in parallel, subdomain by subdomain
in a completely independent way.
In order to have an eﬃcient scheme for solving the equation for the interface
unknown, we will then need two things:
• eﬃcient solvers for the local Dirichlet problems on the subdomains;
• a preconditioner for the discrete SteklovPoincar´e operator so that the num
ber of iterations is kept as low as possible.
We will concentrate here on the second issue, by analyzing and testing a class
of preconditioners for the discrete SteklovPoincar´e operator. The approach that
we will follow is the substructuring one, proposed by J.H. Bramble, J.E. Pasciak
and A.H. Schatz [9] in the framework of conforming domain decomposition and
already applied in the framework of nonconforming methods to the mortar method
in [1]. This consists in decomposing the trace space Φ
h
in a direct sum of a coarse
subspace (essentially linked to the geometrical decomposition of Ω in subdomains)
and of suitable local subspaces and then considering the related blockJacobi type
preconditioners.
From the theoretical point of view, we study the dependence of the condition
number of the resulting preconditioned matrix on the mesh size of the discretiza
tions, aiming at giving bounds which, under suitable assumptions, are independent
of the number and size of the subdomains. In particular we provide an upper bound
which applies to a wide variety of approximation spaces, including ﬁnite elements
and wavelets. Though only suboptimal in terms of the mesh size on the skele
ton (the condition number of the preconditioned matrix grows like (log(H/h))
2
,
H and h being, respectively, the diameter of the subdomains and the mesh size
of the corresponding discretization), the estimate obtained is uniform with respect
to the number and size of subdomains. The resulting method is then potentially
almost perfectly scalable. Moreover this approach has the advantage over other
asymptotically better preconditioners for the three ﬁelds formulation ([7]) of be
ing itself naturally computable in parallel (each edge in the skeleton being treated
independently of the others).
We conclude the paper by reporting a series of extensive numerical tests, per
formed on both the plain and on a stabilized formulation as proposed in [7]. The
results of the tests are in agreement with the theory and conﬁrm the polylogarithmic
growth (with respect to the mesh size of the discretization) of the preconditioned
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 661
system, as well as the independence (asymptotically) with respect to the number
and size of the subdomains.
2. Domain decomposition method: the three fields formulation
Let Ω ⊂ R
2
be a convex polygonal domain. We will consider the following simple
model problem: given f ∈ L
2
(Ω), ﬁnd u satisfying
(1) −
2
i,j=1
∂
∂x
j
_
a
i,j
(x)
∂u
∂x
i
_
+a
0
(x)u = f in Ω, u = 0 on ∂Ω.
We assume that for almost all x ∈ Ω we have 0 ≤ a
0
(x) ≤ R and that the matrix
(a
ij
(x))
i,j=1,2
is symmetric positive deﬁnite, with smallest eigenvalue ≥ α > 0 and
largest eigenvalue ≤ α
, α, α
independent of x.
In a domain decomposition framework, we will consider the three ﬁelds formula
tion ([14]) of such a problem. Let Ω =
K
k=1
Ω
k
, Ω
k
polygons, which, for simplicity,
we will assume to be quadrangles. We set Γ
k
= ∂Ω
k
, Σ =
k
Γ
k
being the skele
ton of the decomposition. Remark that diﬀerently from [14] here we deﬁne Σ so
that it includes the external boundary of the domain Ω, so that there is no need
of distinguishing between interior subdomains and subdomains which are adjacent
to the boundary ∂Ω. This results in a simpliﬁcation both on the notational and
on the implementation point of view, since all subdomains are treated in the same
way. We will make the following regularity assumptions on the subdomains Ω
k
:
(A1) The subdomains are regular in shape and the geometrical decomposition is
graded, that is,
(a) there exists a positive constant c
0
such that, for all k, Ω
k
contains a
ball of diameter c
0
H
k
, it is contained in a ball of diameter H
k
, and
the length of each side is bounded from below by c
0
H
k
; moreover
any interior angle ω satisﬁes 0 < c
1
< ω < c
2
< π (c
0
, c
1
, and c
2
independent of k);
(b) there exists a positive constant c
3
such that, if , k are such that
[∂Ω
k
∩ ∂Ω
[ > 0, then it holds that
H
k
/H
≤ c
3
.
For technical reasons, we also need to make the following assumption (needed
for the proof of (25) in the following), which in many cases can be shown to be
actually a consequence of the previous one:
(A2) There exists a constant C
0
such that for all y, denoting by
y
the line of
equation x
2
= y and setting K
y
= ¦k : length(
y
∩ Ω
k
) > 0¦, it holds that
k∈Ky
H
k
≤ C
0
.
We are interested here in explicitly studying the dependence of the estimates
that we are going to prove on the number and size of the subdomains. To this end,
in the following we will employ the notation A B (resp. A B) to say that
the quantity A is bounded from above (resp. from below) by cB, with a constant
c independent of k and of the H
k
’s, as well as of any mesh size parameter. The
expression A · B will stand for A B A.
For each k we let the norm and seminorm of H
1
(Ω
k
) be deﬁned as usual by
u
2
H
1
(Ω
k
)
=
_
Ω
k
[u[
2
dx +
_
Ω
k
[∇u[
2
dx and [u[
2
H
1
(Ω
k
)
=
_
Ω
k
[∇u[
2
dx. On the
662 SILVIA BERTOLUZZA
boundary ∂Ω
k
we let the norm of H
1/2
(∂Ω
k
) be deﬁned by
ϕ
H
1/2
(∂Ω
k
)
= inf
u∈H
1
(Ω
k
)
u=ϕ on ∂Ω
k
u
H
1
(Ω
k
)
.
For all s in ]0, 1[, we deﬁne the H
s
(∂Ω
k
) seminorm by
(2) [ϕ[
2
H
s
(∂Ω
k
)
=
_
∂Ω
k
_
∂Ω
k
(ϕ(x) −ϕ(y))
2
[x −y[
2s+1
ds(x) ds(y).
The space H
−1/2
(∂Ω
k
) is deﬁned as the dual of H
1/2
(∂Ω
k
).
For e ⊂ ∂Ω
k
segment of extrema a and b, we deﬁne the H
s
(e) seminorm as usual
by the integral expression (2) where the integrals are taken over e. On e we will
also consider the spaces H
s
0
(e) (s ,= 1/2) and H
1/2
00
(e) of functions whose extension
by zero is in H
s
(∂Ω
k
) (s ,= 1/2) and H
1/2
(∂Ω
k
), respectively, which we will equip
with the norm
ϕ
2
H
1/2
00
(e)
= [ϕ[
2
H
1/2
(e)
+
_
e
[ϕ(x)[
2
[x −a[
2s
ds(x) +
_
e
[ϕ(x)[
2
[x −b[
2s
ds(x).
We recall that for s < 1/2 the two spaces H
s
(e) and H
s
0
(e) coincide, and the
two corresponding norms are equivalent. However, the constant in the equivalence
depends a priori on H
k
and on s. In particular it explodes as s converges to 1/2.
The behaviour of such a constant as s approaches the limit value 1/2 (see (44)) will
play a key role in the forthcoming analysis.
Remark also that, for ϕ ∈ H
1/2
00
(e), letting ˇ ϕ ∈ H
1/2
(∂Ω
k
) denote the function
coinciding with ϕ on e and identically vanishing on ∂Ω
k
¸ e, it holds that
(3) ϕ
H
1/2
00
(e)
· [ϕ[
H
1/2
(∂Ω
k
)
,
the constant in the equivalence being uniformly bounded, provided that the ratio
between the length of ∂Ω
k
and the length of e is uniformly bounded.
The functional setting for the three ﬁelds domain decomposition method is given
by the following spaces:
(4) V =
K
k=1
H
1
(Ω
k
), Λ =
K
k=1
H
−1/2
(∂Ω
k
),
and
(5) Φ = ¦ϕ ∈ L
2
(Σ) : there exists u ∈ H
1
0
(Ω), u = ϕ on Σ¦ = H
1
0
(Ω)[
Σ
,
respectively equipped with the norms:
(6) u
V
=
_
k
u
k

2
H
1
(Ω
k
)
_
1/2
, λ
Λ
=
_
k
λ
k

2
H
−1/2
(∂Ω
k
)
_
1/2
,
and (see [4])
(7) ϕ
ϕ
= inf
u∈H
1
0
(Ω)
u=ϕ on Σ
u
H
1
(Ω)
·
_
k
[ϕ[
2
H
1/2
(∂Ω
k
)
_
1/2
.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 663
Let a
k
: H
1
(Ω
k
) H
1
(Ω
k
) → R denote the bilinear form corresponding to the
linear operator considered:
a
k
(w, v) =
_
Ω
k
_
_
2
i,j=1
a
ij
(x)
∂w
∂x
i
∂v
∂x
j
+a
0
(x)wv
_
_
dx.
Under the assumptions made on the matrix (a
ij
(x))
i,j=1,2
and on a
0
, the bilinear
forms a
k
are uniformly H
1
(Ω
k
)continuous, and semideﬁnite. More precisely, there
exist positive constants α and L independent of k and H
k
such that for all w, v in
H
1
(Ω
k
)
α[w[
2
H
1
(Ω
k
)
≤ a
k
(w, w), [a
k
(w, v)[ ≤ Lw
H
1
(Ω
k
)
v
H
1
(Ω
k
)
.
The three ﬁelds formulation of equation (1) is the following: ﬁnd (u, λ, ϕ) ∈ V
Λ Φ such that
(8)
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
∀k, ∀v
k
∈ H
1
(Ω
k
), ∀µ
k
∈ H
−1/2(∂Ω
k
)
:
a
k
(u
k
, v
k
) −
_
∂Ω
k
v
k
λ
k
ds =
_
Ω
k
fv
k
dx,
_
∂Ω
k
u
k
µ
k
ds −
_
∂Ω
k
µ
k
ϕds = 0,
and ∀ψ ∈ Φ :
k
_
∂
Ω
k
λ
k
ψ ds = 0.
It is known that this problem admits a unique solution (u, λ, ϕ) where u is indeed
the solution of (1) and such that
(9) λ
k
=
∂u
k
∂ν
k
a
on Γ
k
, ϕ = u on Σ,
where ν
k
a
denotes the outer conormal derivative to the subdomain Ω
k
. Remark that
in such a formulation the homogeneous Dirichlet boundary condition is embedded
in the deﬁnition of the space Φ.
Equation (8) can be discretized by a Galerkin scheme: after choosing discretiza
tion spaces V
h
=
K
k=1
V
k
h
⊂
K
k=1
H
1
(Ω
k
), Λ
h
=
K
k=1
Λ
k
h
⊂
K
k=1
H
−1/2
(∂Ω
k
)
and Φ
h
⊂ Φ, we consider the problem: ﬁnd (u
h
, λ
h
, ϕ
h
) ∈ V
h
Λ
h
Φ
h
, such that
664 SILVIA BERTOLUZZA
one has
(10)
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
∀k, ∀v
k
h
∈ V
k
h
, ∀µ
k
h
∈ Λ
k
h
:
a
k
(u
k
h
, v
k
h
) −
_
∂Ω
k
v
k
h
λ
k
h
ds =
_
Ω
k
fv
k
h
dx,
_
∂Ω
k
u
k
h
µ
k
h
ds −
_
∂Ω
k
µ
k
h
ϕ
h
ds = 0,
and ∀ψ
h
∈ Φ
h
:
k
_
∂
Ω
k
λ
k
h
ψ
h
ds = 0.
Existence, uniqueness and stability of the solution of (10) rely on the validity of
suitable infsup conditions. More precisely we can make the following assumption:
(A3) The spaces V
k
h
, Λ
k
h
and Φ
h
are such that
(a) Λ
k
h
contains the constant functions;
(b) the following compatibility condition between V
k
h
and Λ
k
h
holds uni
formly in k:
(11) inf
λ
k
h
∈Λ
k
h
sup
u
k
h
∈V
k
h
_
∂
Ω
k
λ
k
h
u
k
h
ds
u
k
h

H
1
(Ω
k
)
λ
k
h

H
−1/2
(∂Ω
k
)
≥ β > 0;
(c) the following compatibility condition between Λ
h
and Φ
h
holds:
(12) inf
ϕ
h
∈Φ
h
sup
λ
h
∈Λ
h
k
_
∂
Ω
k
λ
k
h
ϕ
h
ds
λ
h

Λ
ϕ
h

Φ
≥ α > 0.
Remark that (A3)(a) implies that for all v
k
h
∈ ker B
k
h
= ¦v
k
h
∈ V
k
h
:
_
∂
Ω
k
v
k
h
µ
k
h
=
0 ∀µ
k
h
∈ Λ
k
h
¦ it holds that
_
∂
Ω
k
v
k
h
ds = 0 and then it is not diﬃcult to see that a
is elliptic on ker B
k
h
. Then, by a wellknown argument ([11]), there exists a unique
solution to problem (10), which converges with an optimal rate to the solution of
(8).
The linear system stemming from (10) takes the form
(13)
_
_
A B
T
0
B 0 C
T
0 C 0
_
_
_
_
u
h
λ
h
ϕ
h
_
_
=
_
_
f
h
0
0
_
_
,
(u
h
, λ
h
, and ϕ
h
being the vectors of the coeﬃcients of u
h
, λ
h
and ϕ
h
in the bases
chosen for V
h
, Λ
h
and Φ
h
, respectively). By a Schur complement argument the
solution of (13) can be reduced to a system in the unknown ϕ
h
of the form
(14) CA
−1
C
T
ϕ
h
= CA
−1
_
f
h
0
_
, C = [ 0 C ], A =
_
A B
T
B 0
_
.
The matrix S = CA
−1
C
T
does not need to be assembled. The system (14) can
rather be solved by an iterative technique (like for instance a conjugate gradient
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 665
method) and therefore only the action of S on a given vector needs to be imple
mented. Multiplying by S implies the need for solving a linear system with matrix
A. This reduces, by a proper reordering of the unknowns, to independently solv
ing K discrete Dirichlet problems with Lagrange multipliers in the K subdomains.
Moreover under the above assumptions it is possible to prove that the bilinear form
s : Φ
h
Φ
h
−→R,
(15) s(ϕ
h
, ψ
h
) = ψ
T
h
Sϕ
h
, S = CA
−1
C
T
,
corresponding to the Schur complement matrix S, satisﬁes the following continuity
and coercivity assumptions:
Lemma 2.1. If (A3) holds, then the bilinear form s is continuous and coercive
with respect to the Φ norm:
s(ϕ
h
, ψ
h
) ϕ
h

Φ
ψ
h

Φ
, (16)
s(ϕ
h
, ϕ
h
) ϕ
h

2
Φ
. (17)
Assumption (A3) is, in practice, quite restrictive. In particular the requirement
that both (A3)(b) and (A3)(c) hold simultaneously poses some serious limitation
on the choice of the discretization spaces for the three unknowns u, λ and ϕ. In
order to be more free in such a choice, and more speciﬁcally in order to be allowed
to choose the discretization space for the interface unknown ϕ independently of
the choice of the other two discretization spaces, it is also possible to consider a
stabilized version. Diﬀerent ways have been proposed for stabilizing the three ﬁelds
formulation ([2, 12, 13]). We consider here the approach of [7, 4]. For all k let
a bilinear form [, ] 1
2
,k
: H
1/2
(∂Ω
k
) H
1/2
(∂Ω
k
) → R be given, satisfying the
following bounds: for all η, ξ ∈ H
1/2
(∂Ω
k
)
(18) [η, η] 1
2
,k
≥ 0, [η, ξ] 1
2
,k
≤ B[η[
1/2,∂Ω
k
[ξ[
1/2,∂Ω
k
,
for B positive constant independent of k and H
k
.
We consider then the following discrete stabilized formulation: ﬁnd u
h
∈ V
h
,
λ
h
∈ Λ
h
, ϕ
h
∈ Φ
h
such that it holds that
(19)
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
∀k, ∀v
k
h
∈ V
k
h
, ∀µ
k
h
∈ Λ
k
h
:
a
k
(u
k
h
, v
k
h
) +γ[u
k
h
, v
k
h
] 1
2
,k
−
_
∂Ω
k
v
k
h
λ
k
h
ds − γ[ϕ
h
, v
k
h
] 1
2
,k
=
_
Ω
k
fv
k
h
dx,
_
∂Ω
k
u
k
h
µ
k
h
ds −
_
∂Ω
k
µ
k
h
ϕ
h
ds = 0,
and ∀ψ
h
∈ Φ
h
:
−
k
γ[u
k
h
, ψ
h
] 1
2
,k
+
k
_
∂
Ω
k
λ
k
h
ψ
h
ds +
k
γ[ϕ
h
, ψ
h
] 1
2
,k
= 0,
where γ > 0 is a parameter independent of the choice of the discretization spaces.
Such formulation is consistent with the original continuous problem, i.e., by sub
stituting the solution (u, λ, ϕ) of (8) at the place of (u
h
, λ
h
, ϕ
h
) in (19), we obtain
an identity.
The assumptions needed to guarantee existence and uniqueness of the solution
of the discrete problem, as well as an error estimate, can now be made on the one
666 SILVIA BERTOLUZZA
hand on the spaces V
h
and Λ
h
and on the other hand in a completely independent
way, on the action of the bilinear forms [, ] 1
2
,k
on Φ
h
[
∂
Ω
k
. For example we can
replace assumption (A3)(c) with
(A3)(c
) There exists a positive constant b such that for all ϕ
h
∈ Φ
h
it holds for
all k that
[ϕ
h
, ϕ
h
] 1
2
,k
≥ b[ϕ
h
[
2
H
1/2
(∂Ω
k
)
.
It is then possible to show that the discrete problem (19) admits a unique solution
satisfying an optimal error estimate ([4]). It is beyond the goals of this paper to
thoroughly discuss the construction of suitable bilinear forms [, ] 1
2
,k
. We recall that
this can be done with the aid of multiscale techniques and we refer to [7, 5, 10, 6]
for a detailed analysis of possible solutions.
The linear system stemming from such a problem takes the following form this
time:
(20)
_
_
ˇ
A B
T
−γD
T
B 0 C
T
−γD C γE
_
_
_
_
u
h
λ
h
ϕ
h
_
_
=
_
_
f
h
0
0
_
_
,
with
ˇ
A = A+γF (the matrices D, E and F deriving from the stabilizing terms).
Again, the solution of (20) can be reduced to a system in the unknown ϕ
h
, this
time taking the form
(21) Sϕ
h
:=
_
−D
ˇ
A
−1
D
T
+γE
_
ϕ
h
= −DA
−1
_
f
h
0
_
with
(22)
ˇ
A =
_
ˇ
A B
T
B 0
_
, D = [ −γD C ].
Once again we let s : Φ
h
Φ
h
−→ R be the bilinear form corresponding to the
Schur complement matrix S
(23) s(ϕ
h
, ψ
h
) = ψ
T
h
Sϕ
h
,
and also for the stabilized formulation the following holds ([4]):
Lemma 2.2. Under assumptions (A3)(a), (b), (c
) the bilinear form s is continuous
and coercive with respect to the Φ norm:
s(ϕ
h
, ψ
h
) ϕ
h

Φ
ψ
h

Φ
, (24)
s(ϕ
h
, ϕ
h
) ϕ
h

2
Φ
. (25)
3. Preconditioning by substructuring
For both original and stabilized three ﬁelds formulation, we end up with an
equation on the interface space Φ
h
of the form
Sϕ
h
= f
h
corresponding, through a relation of the form (23), to a bilinear form s : Φ
h
Φ
h
→
R verifying for all ϕ
h
, ψ
h
∈ Φ
h
that
(26) [s(ϕ
h
, ψ
h
)[ ϕ
h

Φ
ψ
h

Φ
, s(ϕ
h
, ϕ
h
) ϕ
h

2
Φ
.
Our aim is to provide a preconditioner for the matrix S. The approach that we will
follow is the one proposed, in the framework of conforming domain decomposition,
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 667
by J.H. Bramble, J.E. Pasciak and A.H. Schatz in [9] and already studied in [1] for
the mortar method case.
Let us consider a splitting of the interior part of the skeleton Σ¸Ω as the disjoint
union of M segments e
i
of extrema a
i
and b
i
Σ ¸ ∂Ω =
M
_
i=1
e
i
,
(corresponding, for instance, to the splitting of Σ as the union of segments of the
form ∂Ω
k
∩∂Ω
). For each k let E
k
individuate those segments that, together with
∂Ω ∩ ∂Ω
k
, build up the boundary of the subdomain Ω
k
:
∂Ω
k
¸ ∂Ω =
_
i∈E
k
e
i
, with E
k
= ¦i : [e
i
∩ ∂Ω
k
[ > 0¦.
We make the following assumption:
(A4) There exists a constant c
4
such that it holds that
sup
k
sup
i∈E
k
[∂Ω
k
[
[e
i
[
≤ c
4
.
Note that (A4) implies that #(E
k
) ≤ c
4
for all k. This will allow us several times
to write inequalities of the form 
i∈E
k
q
i

2
i∈E
k
q
i

2
, where   stands for
various norms and seminorms (the constant in the inequality depending on c
4
).
Following [9], the main idea for constructing a preconditioner for the matrix S
is to decompose Φ
h
as the direct sum of a coarse space L
H
and some local spaces
Φ
0
h,i
(one for each edge e
i
of the decomposition of Σ) and then considering the
corresponding blockJacobi type preconditioners. More precisely, let L
H
⊂ Φ,
L
H
= ¦ϕ ∈ C
0
(Σ) : ∀i = 1, . . . , M, ϕ[
ei
∈ P
1
(e
i
), ϕ = 0 on ∂Ω¦,
denote the subset of those functions of Φ whose restriction to each segment e
i
is
linear. In order to construct the preconditioner we make the following assumptions
on the approximation space Φ
h
:
(A5) The following two conditions hold:
(a) L
H
⊂ Φ
h
;
(b) for all ϕ
h
∈ Φ
h
such that at the extrema a
i
and b
i
of some edge e
i
it
holds that ϕ
h
(a
i
) = ϕ
h
(b
i
) = 0, the function ϕ
0
h,i
deﬁned as
ϕ
0
h,i
= ϕ
h
on e
i
, ϕ
0
h,i
= 0 on Σ ¸ e
i
veriﬁes
ϕ
0
h,i
∈ Φ
h
.
(A6) Φ
h
⊂ H
1
(Σ) and for all ϕ
h
∈ Φ
h
the following two inverse inequalities hold
for all k, k = 1, . . . , K, and for all r, t, 0 ≤ r < t ≤ 1:
[ϕ
h
[
H
t
(∂Ω
k
)
h
r−t
k
[ϕ
h
[
H
r
(∂Ω
k
)
, (27)
[ϕ
h
[
H
t
(ei)
h
r−t
k
[ϕ
h
[
H
r
(ei)
, ∀i ∈ E
k
. (28)
Remark 3.1. Assumption (A5) is not at all restrictive. In a ﬁnite element framework
it reduces to asking that the extrema a
i
and b
i
of the segments e
i
into which the
skeleton is split be nodes of the ﬁnite element grid for Φ
h
.
Remark 3.2. The parameter h
k
appearing in the inverse inequality (27) has here
the meaning of the smallest mesh size of the restriction to ∂Ω
k
of Φ
h
.
668 SILVIA BERTOLUZZA
Under assumption (A5) it is easy to see that Φ
h
can be split as
Φ
h
= L
H
⊕Φ
0
h
, ϕ
h
= ϕ
H
+ϕ
0
h
,
with Φ
0
h
deﬁned as
Φ
0
h
= ¦ϕ
h
∈ Φ
h
, ϕ
h
= 0 at the extrema a
i
and b
i
of each edge e
i
of Σ¦.
The subspace Φ
0
h
veriﬁes
Φ
0
h
=
M
i=1
¦ϕ
h
∈ Φ
0
h
: ϕ
h
= 0 on Σ ¸ e
i
¦ ∼
M
i=1
Φ
0
h
[
ei
,
where the symbol ∼ means that the two spaces are isomorphic. Note that all ϕ
h
∈
Φ
0
h
verify ϕ
h
[
ei
∈ H
1/2
00
(e
i
) for all i, and in the following we will write ϕ
h

H
1/2
00
(ei)
for ϕ
h
[
ei

H
1/2
00
(ei)
(and analogously for other norms deﬁned on portions of Σ).
For each i = 1, , M, we assume we are given a bilinear form ˇ s
i
: Φ
0
h
[
ei
Φ
0
h
[
ei
→ R, satisfying the following continuity and positivity bounds uniformly in
h: for all η, ξ ∈ Φ
0
h
[
ei
:
(B1) ˇ s
i
(ξ, η) ξ
H
1/2
00
(ei)
η
H
1/2
00
(ei)
, ˇ s
i
(ξ, ξ) ξ
2
H
1/2
00
(ei)
.
Moreover we assume that we are given a bilinear form ˇ s
H
: L
H
L
H
→R such
that for all ξ
H
, ζ
H
∈ L
H
it holds that
(B2) ˇ s
H
(ξ
H
, η
H
) ξ
H

Φ
ζ
H

Φ
, ˇ s
H
(ξ
H
, ξ
H
) ξ
H

2
Φ
.
It is beyond the goals of this paper to fully discuss the diﬀerent ways in which
the bilinear forms ˇ s
i
and ˇ s
H
can be constructed. In the following section we will
brieﬂy recall a possible solution and we refer to [9, 10, 16] for further discussions
on this topic.
Following [9], the preconditioner will be ﬁnally constructed with the aid of a
bilinear form ˇ s : Φ
h
Φ
h
→ R assembled by blocks: for η
h
, ζ
h
∈ Φ
h
, with η
h
=
η
H
+η
0
h
, ζ
h
= ζ
H
+ζ
0
h
, η
H
, ζ
H
∈ L
H
and η
0
h
, ζ
0
h
∈ Φ
0
h
we set
(29) ˇ s(η
h
, ζ
h
) = ˇ s
H
(η
H
, ζ
H
) +
M
i=1
ˇ s
i
(η
0
h
[
ei
, ζ
0
h
[
ei
).
The main result of this paper is the following theorem.
Theorem 3.1. For all ϕ
h
∈ Φ
h
it holds that
(30) s(ϕ
h
, ϕ
h
) ˇ s(ϕ
h
, ϕ
h
) max
k
_
1 + log
H
k
h
k
_
2
s(ϕ
h
, ϕ
h
).
By a wellknown argument, Theorem 3.1 implies that we can derive the following
corollary, where we denote by S and
ˇ
S, respectively, the matrices corresponding to
the Galerkin discretization of the bilinear forms s and ˇ s.
Corollary 3.1. It holds that
(31) cond(
ˇ
S
−1
S) max
k
_
1 + log
H
k
h
k
_
2
.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 669
We recall that by the deﬁnition of the bilinear form s, the matrix S is indeed
the Schur complement matrix deﬁned by either (14) or (21).
In order to prove Theorem 3.1, we will basically follow the guidelines of the proof
of the analogous theorems of [9, 1]. The main diﬀerence is that here we consider a
general discretization space, not necessarily consisting in piecewise linear functions.
Moreover we will be working with functions and operators deﬁned directly on the
interface.
In view of the deﬁnition of ˇ s
i
and ˇ s
H
on the one hand and of property (26) of
s on the other hand, proving (30) is equivalent to proving that ∀ϕ
h
∈ Φ
h
with
ϕ
h
= ϕ
H
+ϕ
0
h
(ϕ
H
∈ L
H
and ϕ
0
h
in Φ
0
h
), it holds that
ϕ
h

2
Φ
ϕ
H

2
Φ
+
M
i=1
ϕ
0
h

2
H
1/2
00
(ei)
max
k
_
1 + log
H
k
h
k
_
2
ϕ
h

2
Φ
.
We start by observing that it holds that
ϕ
h

2
Φ
ϕ
H

2
Φ
+ϕ
0
h

2
Φ
ϕ
H

2
Φ
+
M
i=1
ϕ
0
h

2
H
1/2
00
(ei)
,
where the last bound descends by splitting ϕ
0
h
∈ Φ
0
h
as ϕ
0
h
=
M
i=1
ϕ
0
h,i
which
implies (since #(E
k
) ≤ c
4
and since each i belongs to E
k
for at most two k’s)
ϕ
0
h

2
Φ
k
[ϕ
0
h
[
2
H
1/2
(∂Ω
k
)
k
i∈E
k
[ϕ
0
h,i
[
2
H
1/2
(∂Ω
k
)
M
i=1
ϕ
0
h

2
H
1/2
00
(ei)
,
the last inequality deriving from the H
1/2
00
(e
i
) to H
1/2
(∂Ω
k
) boundedness of the
trivial extensionbyzero operator.
Moreover, by direct computation, using the linearity of ϕ
H
, one can see that it
holds that
(32) ϕ
H

2
Φ
K
k=1
[ϕ
H
[
2
H
1/2
(∂Ω
k
)
M
i=1
(ϕ
H
(a
i
) −ϕ
H
(b
i
))
2
,
a
i
and b
i
being the extrema of e
i
. Then, proving Theorem 3.1 reduces to proving
the following result:
Theorem 3.2. For all ϕ
h
∈ Φ
h
with ϕ
h
= ϕ
H
+ϕ
0
h
(ϕ
H
∈ L
H
and ϕ
0
h
∈ Φ
0
h
), it
holds that
(33)
M
i=1
(ϕ
H
(a
i
) −ϕ
H
(b
i
))
2
+
M
i=1
ϕ
0
h

2
H
1/2
00
(ei)
max
k
_
1 + log
H
k
h
k
_
2
ϕ
h

2
Φ
.
In order to prove Theorem 3.2, we will need several bounds. We start by proving
the following Lemma.
Lemma 3.1. Assume that Φ
h
satisﬁes assumptions (A5) and (A6). Then the
following bounds hold:
(i) for all ϕ
h
∈ Φ
h
such that ϕ
h
(p) = 0 for some p ∈ ¯ e
i
with i ∈ E
k
, we have
(34) ϕ
h

2
L
∞
(ei)
_
1 + log
H
k
h
k
_
[ϕ
h
[
2
H
1/2
(ei)
;
670 SILVIA BERTOLUZZA
(ii) for all ϕ
h
∈ Φ
h
it holds that
(35) (ϕ
h
(a
i
) −ϕ
h
(b
i
))
2
_
1 + log
H
k
h
k
_
[ϕ
h
[
2
H
1/2
(ei)
,
a
i
and b
i
being the extrema of e
i
, i ∈ E
k
;
(iii) for all ϕ
0
h
∈ Φ
0
h
it holds that
(36)
i∈E
k
ϕ
0
h

2
H
1/2
00
(ei)
_
1 + log
H
k
h
k
_
2
[ϕ
0
h
[
2
H
1/2
(∂Ω
k
)
.
Proof. Let i ∈ E
k
. By assumptions (A1)(a) and (A4) we have that [e
i
[ · H
k
. We
start by observing that for any ζ ∈ H
1/2+ε
(e
i
) it holds that
(37) ζ
2
L
∞
(ei)
H
−1
k
ζ
2
L
2
(ei)
+
H
2ε
k
ε
[ζ[
2
H
1/2+ε
(ei)
.
This is easily seen by observing that on the reference segment ˆ e = ]0, 1[ the con
tinuity bound of the injection H
1/2+ε
(ˆ e) ⊂ L
∞
(ˆ e) depends on ε as follows (see
Appendix A, Lemma A.1): for all ζ ∈ H
1/2+ε
(ˆ e)
ζ
2
L
∞
(ˆ e)
ζ
2
L
2
(ˆ e)
+
1
ε
[ζ[
2
H
1/2+ε
(ˆ e)
.
A change of variable then yields (37). Let now ϕ
h
∈ Φ
h
and α ∈ R. By assumption
(A6) it holds that
(38)
H
2ε
k
ε
[ϕ
h
−α[
2
H
1/2+ε
(ei)
=
H
2ε
k
ε
[ϕ
h
[
2
H
1/2+ε
(ei)
H
2ε
k
h
−2ε
k
ε
[ϕ
h
[
2
H
1/2
(ei)
.
After choosing ε = 1/ log(H
k
/h
k
), inequality (37) for ζ = ϕ
h
− α combined with
(38) yields:
(39) ϕ
h
−α
2
L
∞
(ei)
H
−1
k
ϕ
h
−α
2
L
2
(ei)
+ log
H
k
h
k
[ϕ
h
[
2
H
1/2
(ei)
.
In view of (39), (34) can then be proven by applying exactly the arguments of
[9], which we repeat here for completeness: let us choose α to be the average of ϕ
h
on e
i
. Applying Poincar´e inequality gives
(40) H
−1/2
k
ϕ
h
−α
L
2
(ei)
[ϕ
h
[
H
1/2
(ei)
which yields
(41) ϕ
h
−α
2
L
∞
(ei)
_
1 + log
H
k
h
k
_
[ϕ
h
[
2
H
1/2
(ei)
.
Now we remark that ϕ
h
(p) = 0 for some p ∈ ¯ e
i
implies
(42) [α[ ϕ
h
−α
L
∞
(ei)
and we get (34) by triangular inequality.
(ii) then follows by applying (i) to the function ϕ(x) −ϕ(a
i
) (which vanishes in
p = a
i
).
As far as (iii) is concerned, we start by remarking that for ϕ
0
h
∈ Φ
0
h
and for
i ∈ E
k
, denoting by ϕ
0
h,i
the function coinciding with ϕ
0
h
on e
i
and identically
vanishing on ∂Ω
k
¸ e
i
and using (3) and the inverse inequality (27), we obtain the
following bounds
(43) ϕ
0
h

H
1/2
00
(ei)
[ϕ
0
h,i
[
H
1/2
(∂Ω
k
)
h
−ε
k
[ϕ
0
h,i
[
H
1/2−ε
(∂Ω
k
)
[ϕ
0
h
[
H
1/2−ε
0
(ei)
,
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 671
where the second inequality descends from (27) by applying a standard functional
spaces interpolation argument, since the spaces H
r
0
(e
i
), 0 < r < 1/2, can be ob
tained as the interpolants of order 2r between L
2
(e
i
) and H
1/2
00
(e
i
). Again we bound
the H
1/2−ε
0
(e
i
) seminorm of ϕ
0
h
by a change of variable, using the corresponding
bound on the reference segment ˆ e = ]0, 1[ . We recall once more that H
1/2−ε
0
(ˆ e)
and H
1/2−ε
(ˆ e) coincide, with equivalent norms, the constant in the equivalence de
pending on ε. In particular it is possible to prove (see Appendix A, Corollary A.1)
that the following bound holds for all ζ ∈ H
1/2
(ˆ e) and for all α ∈ R:
(44) [ζ[
H
1/2−ε
0
(ˆ e)
1
ε
ζ −α
H
1/2
(ˆ e)
+
1
√
ε
[α[,
which rescales as
[ζ[
H
1/2−ε
0
(ei)
H
ε
k
ε
_
H
−1/2
k
ζ −α
L
2
(ei)
+[ζ −α[
H
1/2
(ei)
_
+
H
ε
k
√
ε
[α[.
Again, taking ζ = ϕ
0
h
and α as its average on e
i
, we can apply on the one hand
Poincar´e inequality (40) and on the other hand the bound (42), which holds since
ϕ
0
h
(a
i
) = 0, and we get
ϕ
0
h

H
1/2
00
(ei)
H
ε
h
−ε
k
ε
[ϕ
0
h
[
H
1/2
(ei)
+
H
ε
h
−ε
k
√
ε
ϕ
0
h
−α
L
∞
(ei)
.
We then take once again ε = 1/ log(H
k
/h
k
), and, thanks to bound (41), we obtain
ϕ
0
h

H
1/2
00
(ei)
_
1 + log
H
k
h
k
_
[ϕ
0
h
[
H
1/2
(ei)
.
Since
i∈E
k
[ [
2
H
1/2
(ei)
[ [
2
H
1/2
(∂Ω
k
)
, by squaring and then taking the sum over
i ∈ E
k
, we obtain (36).
Remark 3.3. Since for s < 1/2 the two spaces H
s
(e
i
) and H
s
0
(e
i
) coincide, the
seminorm [ [
H
s
0
(ei)
in equation (43) could be replaced by the standard H
s
(e
i
)
seminorm. In that case, however, the implicit constant in the bound would depend
on ε.
Analogously to what happens in [9], the main ingredient of the proof of Theorem
3.2 is then the following lemma.
Lemma 3.2. Let ϕ
0
h
∈ Φ
0
h
and ζ
H
∈ L
H
. Then for all k it holds that
(45)
i∈E
k
ϕ
0
h

2
H
1/2
00
(ei)
_
1 + log
H
k
h
k
_
2
[ϕ
0
h
+ζ
H
[
2
H
1/2
(∂Ω
k
)
.
Provided such lemma holds, (33) is obtained, thanks to (ii) of Lemma 3.1, by
applying (45) for ζ
H
= ϕ
H
. In fact, for ϕ
h
= ϕ
H
+ϕ
0
h
, ϕ
H
∈ L
H
, ϕ
0
h
∈ Φ
0
h
, we can
write
M
i=1
ϕ
0
h

2
H
1/2
00
(ei)
=
1
2
K
k=1
i∈E
k
ϕ
0
h

2
H
1/2
00
(ei)
k
_
1 + log
H
k
h
k
_
2
[ϕ
h
[
2
H
1/2
(∂Ω
k
)
max
k
_
1 + log
H
k
h
k
_
2
ϕ
h

2
Φ
,
672 SILVIA BERTOLUZZA
which, combined with (35), yields (33). The only thing that we need to do then is
to prove Lemma 3.2.
Proof of Lemma 3.2. Let k, 1 ≤ k ≤ K. We start by introducing a function ζ
0
∈
Φ
k,0
h
,
(46) Φ
k,0
h
= ¦ϕ
h
∈ Φ
h
[
∂Ω
k
: ϕ
h
= 0 at the cross points of ∂Ω
k
¦ = Φ
0
h
[
∂
Ω
k
,
ζ
0
depending on ζ
H
, deﬁned as the unique element of Φ
k,0
h
satisfying
(ζ
0
, ξ)
H
1/2
(∂Ω
k
)
= (ζ
H
, ξ)
H
1/2
(∂Ω
k
)
, ∀ξ ∈ Φ
k,0
h
,
where (, )
H
1/2
(∂Ω
k
)
is the bilinear form
(ζ, ξ)
H
1/2
(∂Ω
k
)
=
_
∂Ω
k
_
∂Ω
k
(ζ(x) −ζ(y))(ξ(x) −ξ(y))
[x −y[
2
dxdy
inducing the H
1/2
(∂Ω
k
) seminorm. It is easy to check that [ [
H
1/2
(∂Ω
k
)
is a norm
on Φ
k,0
h
and then we can conclude by standard arguments that
(47) [ζ
0
[
H
1/2
(∂Ω
k
)
[ζ
H
[
H
1/2
(∂Ω
k
)
.
Now, adding and subtracting ζ
0
at the lefthand side of (45), we can write:
(48)
i∈E
k
ϕ
0
h

2
H
1/2
00
(ei)
i∈E
k
ϕ
0
h
+ζ
0

2
H
1/2
00
(ei)
+
i∈E
k
ζ
0

2
H
1/2
00
(ei)
.
Since ϕ
0
h
+ζ
0
∈ Φ
k,0
h
, by the deﬁnition of ζ
0
we have that
(49) (ζ
H
−ζ
0
, ϕ
0
h
+ζ
0
)
H
1/2
(∂Ω
k
)
= 0,
which yields
[ϕ
0
h
+ζ
H
[
2
H
1/2
(∂Ω
k
)
≥ [ϕ
0
h
+ζ
0
[
2
H
1/2
(∂Ω
k
)
.
The ﬁrst sum on the righthand side of (48) can then be bound thanks to Lemma
3.1(iii); we have
i∈E
k
ϕ
0
h
+ζ
0

2
H
1/2
00
(ei)
_
1 + log
H
k
h
k
_
2
[ϕ
0
h
+ζ
0
[
2
H
1/2
(∂Ω
k
)
(50)
_
1 + log
H
k
h
k
_
2
[ϕ
0
h
+ζ
H
[
2
H
1/2
(∂Ω
k
).
(51)
Let us bound the second sum on the righthand side of (48). We recall that we
have
(52) ζ
0

2
H
1/2
00
(ei)
= [ζ
0
[
2
H
1/2
(ei)
+I
1
(ζ
0
) +I
2
(ζ
0
),
with
(53) I
1
(ζ
0
) =
_
bi
ai
[ζ
0
(x)[
2
[x −a
i
[
dx, I
2
(ζ
0
) =
_
bi
ai
[ζ
0
(x)[
2
[x −b
i
[
dx,
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 673
(we recall that a
i
and b
i
denote the two vertices of the edge e
i
). Now, using
equations (47), (32) and (35), since ϕ
0
h
vanishes at a
i
and b
i
, we can write
i∈E
k
[ζ
0
[
2
H
1/2
(ei)
[ζ
0
[
2
H
1/2
(∂Ω
k
)
[ζ
H
[
2
H
1/2
(∂Ω
k
)
(54)
i∈E
k
(ζ
H
(a
i
) −ζ
H
(b
i
))
2
(55)
=
i∈E
k
([ϕ
0
h
+ζ
H
](a
i
) −[ϕ
0
h
+ζ
H
](b
i
))
2
(56)
_
1 + log
H
k
h
k
_
[ϕ
0
h
+ζ
H
[
2
H
1/2
(∂Ω
k
)
. (57)
Let us now bound I
1
(ζ
0
). For notational simplicity let us identify e
i
= ]0, T[
with a
i
= 0 and b
i
= T. We have
I
1
(ζ
0
) =
_
T
0
[ζ
0
(x)[
2
[x[
dx (58)
=
_
T
0
[ζ
0
(x) +ζ
H
(x) − ζ
H
(x) +ζ
H
(0) −ζ
H
(0)[
2
[x[
dx (59)
_
T
0
[ζ
0
(x) −ζ
H
(x) + ζ
H
(0)[
2
[x[
dx (60)
+
_
T
0
[ζ
H
(x) −ζ
H
(0)[
2
[x[
dx. (61)
Let us set ζ
⊥
= ζ
0
−ζ
H
. Remarking that ζ
⊥
(0) = −ζ
H
(0), we have
_
T
0
[ζ
0
(x) −ζ
H
(x) +ζ
H
(0)[
2
[x[
dx =
_
T
0
[ζ
⊥
(x) −ζ
⊥
(0)[
2
[x[
dx
=
_
h
k
0
[ζ
⊥
(x) −ζ
⊥
(0)[
2
[x[
dx
+
_
T
h
k
[ζ
⊥
(x) −ζ
⊥
(0)[
2
[x[
dx.
The ﬁrst term is bound, using CauchySchwarz’s inequality and inequality (28), by
_
h
k
0
[ζ
⊥
(x) −ζ
⊥
(0)[
2
[x[
dx =
_
h
k
0
1
[x[
¸
¸
¸
¸
_
x
0
ζ
⊥
x
(τ) dτ
¸
¸
¸
¸
2
dx
_
h
k
0
_
x
0
[ζ
⊥
x
(τ)[
2
dτ dx
h
k
[ζ
⊥
[
2
H
1
(ei)
[ζ
⊥
[
2
H
1/2
(ei)
,
while, using (34), we bound the second by
_
T
h
k
[ζ
⊥
(x) −ζ
⊥
(0)[
2
[x[
dx ζ
⊥
−ζ
⊥
(0)
2
L
∞
(ei)
_
T
h
k
1
x
dx
_
1 + log
H
k
h
k
_
2
[ζ
⊥
[
2
H
1/2
(ei)
,
674 SILVIA BERTOLUZZA
where we used T · H
k
. The second integral in (60) can be bound directly using
the linearity of ζ
H
.
_
T
0
[ζ
H
(x) −ζ
H
(0)[
2
[x[
dx (ζ
H
(b
i
) −ζ
H
(a
i
))
2
_
1 + log
H
k
h
k
_
[ζ
H
+ϕ
0
h
[
2
H
1/2
(ei)
where again we used (35) and the fact that ϕ
0
h
vanishes at a
i
and b
i
. Then, adding
up the diﬀerent contributions, we conclude that
(62) I
1
(ζ
0
)
_
1 + log
H
k
h
k
_
2
[ζ
0
−ζ
H
[
2
H
1/2
(ei)
+
_
1 + log
H
k
h
k
_
[ζ
H
+ϕ
0
h
[
2
H
1/2
(ei)
.
By using the same arguments we get a similar bound for I
2
:
(63) I
2
(ζ
0
)
_
1 + log
H
k
h
k
_
2
[ζ
0
−ζ
H
[
2
H
1/2
(ei)
+
_
1 + log
H
k
h
k
_
[ζ
H
+ϕ
0
h
[
2
H
1/2
(ei)
.
We can now insert bounds (54), (62) and (63) back into (52) and sum over i ∈
E
k
. By observing that for any w ∈ H
1/2
(∂Ω
k
) it holds that
i∈E
k
[w[
2
H
1/2
(ei)
[w[
2
H
1/2
(∂Ω
k
)
, we obtain the bound
i∈E
k
ζ
0

2
H
1/2
00
(ei)
_
1 + log
H
k
h
k
_
[ϕ
0
h
+ζ
H
[
2
H
1/2
(∂Ω
k
)
(64)
+
_
1 + log
H
k
h
k
_
2
[ζ
0
−ζ
H
[
2
H
1/2
(∂Ω
k
)
. (65)
We now observe once again that equation (49) implies that [ζ
H
− ζ
0
[
2
H
1/2
(∂Ω
k
)
[ζ
H
+ϕ
0
h
[
2
H
1/2
(∂Ω
k
)
and we get the thesis.
4. Numerical Tests
We will test the preconditioner we propose on the following problem: ﬁnd u such
that
(66)
_
−∆u = f, in Ω = ]0, 1[ ]0, 1[ ,
u = 0, on Γ = ∂Ω.
We consider a uniform, geometrically conforming, decomposition of Ω = ]0, 1[
]0, 1[ in K = N N equal square subdomains of size H H, H = 1/N.
In each subdomain Ω
k
we take a uniform mesh T
k
composed of n
k
n
k
equal
square elements of size δ
k
δ
k
, δ
k
= H/n
k
= 1/(Nn
k
). We then deﬁne V
k
h
to be
the space of Q
1
ﬁnite elements on the mesh T
k
,
V
k
h
= ¦u
h
∈ C
0
(Ω) : u
h
[
τ
∈ Q
1
(τ), ∀τ ∈ T
k
¦,
where Q
1
(τ) denotes the space of polynomials of degree less than or equal to one
in each variable. The multiplier space Λ
k
h
is then deﬁned as the trace on ∂Ω
k
of
V
k
h
. With such a choice it is possible to prove that (A3) holds (see [4]).
The skeleton Σ¸ ∂Ω is split as Σ¸ ∂Ω =
2(N−1)N
i=1
e
i
with e
i
= Γ
k
∩Γ
for some
k = k(i) and = (i). We remark that [e
i
[ = H for all i. We consider a grid ( on
Σ obtained by uniformly splitting each e
i
into L
i
elements of size h
i
and we deﬁne
Φ
h
to be the space of P
1
ﬁnite elements on such a grid:
Φ
h
= ¦ϕ
h
∈ C
0
(Σ) : ϕ
h
[
κ
∈ P
1
(κ), ∀κ ∈ (, ϕ
h
[
∂Ω
= 0¦.
With this choice, assumption (A6) holds for h
k
= min
i∈E
k
h
i
.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 675
Letting ϕ
0
h,i
(resp. ψ
0
h,i
) denote the vector of the values of ϕ
0
h
(resp ψ
0
h
) ∈ Φ
0
h
at
the interior nodes of the grid ([
ei
, we deﬁne the bilinear form ˇ s
i
: Φ
0
h,i
[
ei
Φ
0
h,i
[
ei
as
ˇ s
i
(ϕ
0
h
[
ei
, ψ
0
h
[
ei
) = (ψ
0
h,i
)
T
(R
i
)
1/2
ϕ
0
h,i
,
where R
i
is the stiﬀness matrix associated to the discretization of the operator
−d
2
/dx
2
by P
1
ﬁnite element on e
i
with homogeneous Dirichlet boundary conditions
at the extrema a
i
and b
i
. With R
i
being positive deﬁnite, its square root is well
deﬁned and it is computed and stored at the start.
As far as the coarse preconditioner ˇ s
H
is concerned, we considered two choices:
• (BPS) Following [9], we set
(67) ˇ s
H
(η
H
, ζ
H
) := 2
M
i=1
(η
H
(a
i
) −η
H
(b
i
))(ζ
H
(a
i
) −ζ
H
(b
i
)).
• (Laplace) Denoting by ˜ η
H
and
˜
ζ
H
the H
1
0
(Ω) functions obtained by lifting
η
H
and ζ
H
harmonically in each subdomain, we set
(68) ˇ s
H
(η
H
, ζ
H
) :=
_
Ω
∇˜ η
H
∇
˜
ζ
H
dx;
note that the set ¦˜ η
H
, η
H
∈ L
H
¦ coincides with the set of Q
1
ﬁnite elements
on the coarse grid corresponding to the decomposition Ω =
Ω
k
, and
then applying the coarse preconditioner reduces to numerically solving the
Laplace equation on such space.
For all tests we set f = 1. We solved the linear system by a preconditioned
conjugate gradient method, using ϕ
h
= 0 as initial guess. We report the number
of iterations needed to reduce the residual of a factor 10
−5
.
4.1. Plain formulation.
4.1.1. Conforming decomposition. We start by considering the case of a conforming
domain decomposition, that is, in the framework described above we set n
k
= n
for all k and L
i
= n for all i. In this case δ
k
= h
i
= h = H/n. With such a choice,
it is not diﬃcult to check that the infsup condition (12) is satisﬁed uniformly in h
and then we are in the range of Lemma 2.1. It is easy to realize that the discrete
solution veriﬁes u
k
h
= ϕ
h
on ∂Ω
k
, and then the function u
h
, deﬁned by u
h
[
Ω
k
= u
k
h
,
veriﬁes u
h
∈ H
1
0
(Ω). In such a case we obtain the same solution that we would
get using a conforming ﬁnite element method on a regular grid of NnNn square
elements of dimension h h. The use of a nonconforming method is therefore, in
this case, only a device to implement a solver for the discrete problem in an easily
parallelizable way.
In order to study both the dependence on H (size of the subdomains) and on h
we tested the preconditioner for diﬀerent values of n in the range [5, 40] and N in
the range [4, 32]. We tested both coarse preconditioners for all combinations of N
and n. The results are summarized in Tables 1 and 2. As one can see, they are in
close agreement with the theory (note that in this case max
k
h
k
/H
k
= h/H = n).
676 SILVIA BERTOLUZZA
Table 1. Number of conjugate gradient iterations needed for re
ducing the residual of a factor 10
−5
, with coarse preconditioner
given by (68) for diﬀerent combinations of the number K = N
2
of
subdomains and n of elements per edge (n
2
elements per subdo
main).
n = 5 n = 10 n = 20 n = 40
K = N
2
# Iter. # Iter. # Iter. # Iter.
16 10 12 14 15
64 14 17 20 23
144 15 18 22 26
256 15 19 23 27
400 16 20 23 27
576 16 20 24 27
784 16 20 24 27
1024 16 20 24 27
Table 2. Number of conjugate gradient iterations needed for re
ducing the residual of a factor 10
−5
, with coarse preconditioner
given by (67) for diﬀerent combinations of the number K = N
2
of
subdomains and n of elements per edge (n
2
elements per subdo
main).
n = 5 n = 10 n = 20 n = 40
K = N
2
# Iter. # Iter. # Iter. # Iter.
16 8 9 11 13
64 10 13 17 22
144 11 15 20 26
256 13 16 22 30
400 13 17 23 32
576 13 17 24 33
784 14 18 25 33
1024 14 18 25 33
4.1.2. Nonconforming decomposition. We consider then the case of a truly noncon
forming decomposition, without stabilization (formulation (10)). In our test for
e
i
= ∂Ω
k
∩ ∂Ω
, we set L
i
= min¦n
k
, n
¦. This discretization does not in general
satisfy the assumptions required for the uniform stability of the discrete problem.
Nevertheless, using a standard technique, it is not diﬃcult to prove that (12) holds
provided that for all , k such that [∂Ω
∩ ∂Ω
k
[ > 0 it either holds that n
= n
k
or
max¦n
, n
k
¦/ min¦n
, n
k
¦ ≥ C for some constant C > 1.
We considered two test conﬁgurations, both on a decomposition of Ω into 5 5
subdomains. In the ﬁrst case (see Figure 1, top), for all subdomains except two,
we set n
k
= 10, while on the remaining two subdomains (subdomains 13 and 14,
if we number all subdomains rowwise) we set n
k
= n, where n varies in the set
[10, . . . , 52]. With the above criterion used to ﬁx L
i
, we have that L
i
= 10 for all
edges e
i
, except the common edge Γ
13
∩ Γ
14
(the two subdomains are adjacent),
where L
i
= max¦n
13
, n
14
¦ = n varies.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 677
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(1)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(2)
Figure 1. The two conﬁgurations on which the preconditioner is
tested for the plain formulation (8) in the nonconforming case.
In the second conﬁguration (see Figure 1, bottom) for all subdomains we again
set n
k
= 10, this time except that on a threebythree block of subdomains where
n
k
= n, with n varying again between 10 and 52. Here, for 12 out of 40 edges of Σ,
we have that L
i
= n. For both conﬁgurations we tested both coarse preconditioners.
The results are provided in Table 3.
678 SILVIA BERTOLUZZA
Table 3. Number of conjugate gradient iterations needed for re
ducing the residual of a factor 10
−5
for Conﬁgurations 1 and 2
(corresponding, respectively, to the top and to the bottom decom
positions illustrated in Figure 1). Both coarse preconditioners (67)
(BPS) and (68) (Laplace) are tested for diﬀerent values of the
number n of elements per edge in the reﬁned part of the skeleton
(one edge for Conﬁguration 1 and twelve edges for Conﬁguration
2). The number of elements per edge in the remaining part of the
skeleton is set to 10.
Laplace BPS
# It. # It.
n = 10 12 8
n = 17 18 11
n = 24 18 13
n = 31 18 13
n = 38 18 14
n = 45 19 14
n = 52 19 14
(a) Conﬁguration 1.
Laplace BPS
# It. # It.
n = 10 12 8
n = 17 20 14
n = 24 23 16
n = 31 24 18
n = 38 25 19
n = 45 26 20
n = 52 27 20
(b) Conﬁguration 2.
As one can easily see, though the theoretical estimate on the condition number
only depends on max
k
H
k
/h
k
, which takes the same value for both conﬁgurations, in
the ﬁrst case in which the overall discretization of the skeleton is sensibly coarser—
the mesh being ﬁne only in a small portion (in our case one) of the edges composing
Σ—the inﬂuence of reﬁning the mesh is weaker than in the second case, in which a
relevant portion of the skeleton is reﬁned.
4.2. Stabilized formulation. We now consider the stabilized formulation (19).
Since the H
1/2
(∂Ω
k
) seminorm is invariant under changes of scale, we can describe
the stabilizing form for subdomains scaled in such a way that [∂Ω
k
[ = 1 (that is,
H = 1/4). We can then identify ∂Ω
k
with the circle T of unitary length. Following
the proposal of [7], the bilinear forms [, ] 1
2
,k
are designed by means of a wavelet
decomposition. More precisely, we restrict the number L
i
of elements of the ith
edge e
i
to be a power of two,
L
i
= 2
ji
for some j
i
≥ 1,
so that for all k, Φ
h
[
∂Ω
k
⊂ V
j
k
+2
where for j > 0, V
j
denotes the space of P
1
ﬁnite
elements on the uniform grid (
j
of T with mesh size 1/2
j
.
The sequence ¦V
j
¦
j≥0
forms a socalled multiresolution analysis of L
2
(T) and
it is well known (see for example [15]) that there exist several wavelet bases asso
ciated with such a multiresolution analysis. More precisely there exist several P
1
compactly supported functions ϑ ∈ C
0
(R) deﬁned on the uniform grid of mesh size
1 and integer nodes, such that, setting
ϑ
m
=
+∞
n=−∞
2
m/2
ϑ(2
m
(x −n) −),
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 679
all functions η
j
∈ V
j
can be written as
η
j
= η
0
+
j−1
m=0
2
m
=1
η
m
ϑ
m
, η
0
constant,
and such that
[η
j
[
2
H
1/2
(∂Ω
k
)
·
j−1
m=0
2
m
=1
2
m
[η
m
[
2
.
Given the values of η
j
at the nodes of the uniform grid, the coeﬃcients (η
m
)
m,l
can be retrieved by a fast (O(2
j
)) wavelet transform (FWT).
If we denote by P
k
: L
2
(T) → V
j
k
+2
the L
2
(T) orthogonal projection, for ζ, ξ ∈
L
2
(T) we can deﬁne corresponding wavelet coeﬃcients
ˇ
ζ
m
and
ˇ
ξ
m
such that
P
k
(ζ) = ζ
0
+
j
k
m=0
2
m
=1
ˇ
ζ
m
ϑ
m
, P
k
(ξ) = ξ
0
+
j
k
m=0
2
m
=1
ˇ
ξ
m
ϑ
m
.
Then we deﬁne the bilinear form [, ] 1
2
,k
as
[ζ, ξ] 1
2
,k
=
j
k
m=0
2
m
=1
2
m
ˇ
ζ
m
ˇ
ξ
m
.
It is possible to prove ([7, 14]) that the bilinear forms thus deﬁned satisfy assump
tions (A4) and (18).
Remark 4.1. Note that the coeﬃcients
ˇ
ζ
m
and
ˇ
ξ
m
are not the classical wavelet
coeﬃcients of the functions ζ and ξ. These would in fact be deﬁned as the scalar
product of, respectively, ζ and ξ with suitable dual functions
˜
ϑ
m
∈ L
2
(T). The
coeﬃcients
ˇ
ζ
m
and
ˇ
ξ
m
are rather the scalar products of the dual functions
˜
ϑ
m
with the projections of ζ and ξ onto V
j
k
+2
, respectively. Naturally, such coeﬃcients
coincide with the classical ones, provided f ∈ V
j
k
+2
.
With this deﬁnition, the computation of [u
k
h
−ϕ
h
, v
k
h
−ψ
h
] 1
2
,k
essentially reduces
to ﬁrst computing the nodal values of P
k
(u
k
h
), P
k
(v
k
h
), P
k
(ϕ
h
) and P
k
(ψ
h
), respec
tively, and then applying a FWT. As far as ϕ
h
and ψ
h
are concerned, the choice
of j
k
implies that ϕ
h
, ψ
h
∈ V
j
k
+2
and then their nodal values are simply computed
by interpolation. As far as u
k
h
and v
k
h
are concerned, we need to compute their
L
2
projection on V
j
k
+2
. Observing that the mass matrix in V
j
k
+2
has a circulant
structure with bandwidth equal to three, this can also be done in O(2
j
k
) operations
by a modiﬁcation of the Croute reduction algorithm for solving linear systems with
tridiagonal matrices. For further details on the implementation of the stabilized
formulation we refer to [8].
We tested the preconditioner (with both form (67) and (68) of the coarse bilinear
form) on four diﬀerent splittings of Ω into, respectively, 4 4, 8 8, 12 12 and
16 16 subdomains. In each case we randomly assigned the values of n
k
in such
a way that for ∼ 1/3 of the subdomains n
k
= 5, for about another third n
k
= 10,
and for the remaining subdomains n
k
= 15. The four conﬁgurations considered are
shown in Figure 2. For approximating the interface function ϕ, we used a uniform
discretization of the interface (L
i
= n for all i). For each conﬁguration, we tested
the preconditioner for diﬀerent values of n and of the stabilization parameter γ.
The function ϑ appearing in the deﬁnition of the stabilizing bilinear form is chosen
680 SILVIA BERTOLUZZA
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(a)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(b)
Figure 2. The four conﬁgurations on which the preconditioner is
tested for the stabilized formulation (10). (Continues)
to be the piecewise linear wavelet corresponding to the 2.2 Bspline biorthogonal
setting ([15]). Note (see Table 4) that for γ = 0 (no stabilization!) if n is big, the
CG algorithm does not converge within the maximum number of iterations (which
was set equals to 100), conﬁrming the need for using some kind of stabilization.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 681
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(c)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(d)
Figure 2. (continued)
As one can see from the results, the value of the parameter γ has an inﬂuence on
the conditioning of the matrix
ˇ
S
−1
S. If γ is too small (see Table 5), the action of the
stabilization term is not suﬃcient to compensate for the lack of validity of the inf
sup condition, and therefore the system behaves almost like the one obtained using
the plain formulation. By increasing γ (see Tables 6 and 7), the situation improves
682 SILVIA BERTOLUZZA
dramatically and one can see that, for instance, for γ = .05 in the case of 16 16
subdomains with n = 32 (which corresponds to 15360 degrees of freedom for ϕ
h
)
the residual is reduced of a factor 10
−5
in about 35 iterations, which is only slightly
worse than the result obtained for the analogous values in the conforming case.
However, when γ is further increased (see Table 8), the performances deteriorate.
This (as well as the diﬀerence with respect to the conforming case) is probably
due to the fact that the constants in the estimates (26)—which have an inﬂuence
on the bound on the condition number—depend on the ratio B/b (B and b being
the constants in (18) and in assumption (A3)(c
), respectively) ampliﬁed by the
factor γ. If such a ratio (which is always greater than or equal to one) in not close
enough to 1, the choice of the right balance between the original bilinear form and
the stabilizing term can become important from the point of view of the spectral
properties of the operator, which are essential for preconditioning.
Table 4. Number of conjugate gradient iterations needed to re
duce the residual of a factor 10
−5
on the four conﬁgurations de
picted in Figure 2 with the plain formulation (10). Both coarse pre
conditioners are tested for diﬀerent values of the number n of ele
ments per edge. As one can see for n ≥ 16, the CG method does not
converge within the maximum number of iteration (= 100), con
ﬁrming the instability of the plain formulation for the discretization
considered.
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 14 17 18 19
8 47 57 60 69
16 — — — —
32 — — — —
Laplace
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 11 13 14 15
8 40 52 54 64
16 — — — —
32 — — — —
BPS
Table 5. Number of conjugate gradient iterations needed to re
duce the residual of a factor 10
−5
on the four conﬁgurations de
picted in Figure 2 with the stabilized formulation (19) for γ =
.00125. Both coarse preconditioners are tested for diﬀerent values
of the number n of elements per edge. One can clearly see that
the method converges very badly or does not converge. This is due
to the fact that, since the stabilization parameter is very small,
the action of the stabilization term is not suﬃcient to compensate
for the lack of validity of the infsup condition, and therefore the
system behaves like the one obtained using the plain formulation.
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 14 17 18 19
8 52 60 64 83
16 86 — — —
32 96 — — —
Laplace
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 11 13 14 15
8 39 51 57 77
16 69 92 — —
32 85 — — —
BPS
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 683
Table 6. Number of conjugate gradient iterations needed to re
duce the residual of a factor 10
−5
on the four conﬁgurations de
picted in Figure 2 with the stabilized formulation (19) for γ = 1.
Both coarse preconditioners are tested for diﬀerent values of the
number n of elements per edge.
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 14 17 18 19
8 22 26 27 29
16 28 34 36 38
32 34 42 44 45
Laplace
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 11 13 14 15
8 18 21 24 27
16 23 31 35 38
32 30 41 46 51
BPS
Table 7. Number of conjugate gradient iterations needed to re
duce the residual of a factor 10
−5
on the four conﬁgurations de
picted in Figure 2 with the stabilized formulation (19) for γ = 5.
Both coarse preconditioners are tested for diﬀerent values of the
number n of elements per edge.
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 14 17 18 19
8 19 22 23 25
16 22 28 29 31
32 24 31 32 34
Laplace
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 11 13 14 15
8 15 17 19 22
16 16 23 26 29
32 19 27 30 35
BPS
Table 8. Number of conjugate gradient iterations needed to re
duce the residual of a factor 10
−5
on the four conﬁgurations de
picted in Figure 2 with the stabilized formulation (19) for γ = 20.
Both coarse preconditioners are tested for diﬀerent values of the
number n of elements per edge.
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 14 17 18 19
8 25 26 27 30
16 31 37 38 41
32 35 41 43 48
Laplace
(a) (b) (c) (d)
n # It. # It. # It. # It.
4 12 13 14 15
8 23 24 24 27
16 28 35 35 37
32 34 40 39 41
BPS
Appendix A: Some Sobolev space injection bounds
The aim of this section is to recall some known bounds for which we provide a
new proof based on the use of equivalent norms expressed through wavelet coeﬃ
cients (see for instance [15]). Such proof will easily reveal the dependence of the
684 SILVIA BERTOLUZZA
constants in the bounds on the smoothness of the spaces considered and the way
these constants explode as the smoothness index approaches the limit of validity of
the bounds.
The key ingredient of the following arguments is the possibility of constructing
two orthonormal bases
B = ¦ρ
k
, k = 1, . . . , 2
j0
¦ ∪ ¦ϑ
j,k
, j ≥ j
0
, k = 1, . . . , 2
j
¦,
B
0
= ¦ρ
0
k
, k = 1, . . . , 2
j0
¦ ∪ ¦ϑ
0
j,k
, j ≥ j
0
, k = 1, . . . , 2
j
¦,
for L
2
(0, 1), tuned up to provide norm equivalences for Sobolev spaces, respectively,
without and with homogeneous boundary conditions. In particular, such bases can
be constructed in such a way that all u ∈ L
2
(0, 1) can be written both ways as
u =
2
j
0
k=1
¸u, ρ
k
)ρ
k
+
∞
j=j0
2
j
k=1
¸u, ϑ
j,k
)ϑ
j,k
,
u =
2
j
0
k=1
¸u, ρ
0
k
)ρ
0
k
+
∞
j=j0
2
j
k=1
¸u, ϑ
0
j,k
)ϑ
0
j,k
and that the following norm equivalences hold for all s with 0 ≤ s ≤ 1:
u
H
s
(0,1)
·
_
_
2
j
0
k=1
[¸u, ρ
k
)[
2
+
∞
j=j0
2
j
k=1
2
2js
[¸u, ϑ
j,k
)[
2
_
_
1/2
, (69)
u
H
s
0
(0,1)
·
_
_
2
j
0
k=1
[¸u, ρ
0
k
)[
2
+
∞
j=j0
2
j
k=1
2
2js
[¸u, ϑ
0
j,k
)[
2
_
_
1/2
, s ,= 1/2, (70)
u
H
1/2
00
(0,1)
·
_
_
2
j
0
k=1
[¸u, ρ
0
k
)[
2
+
∞
j=j0
2
j
k=1
2
j
[¸u, ϑ
0
j,k
)[
2
_
_
1/2
, (71)
with constants independent of s. Such bases are constructed in such a way that the
basis functions satisfy the following properties:
W.1. For all j ≥ j
0
and k = 1, . . . , 2
j
the functions ϑ
j,k
have a certain number
of vanishing moments; in particular ϑ
j,k
has zero mean value.
W.2. The basis functions are well localized and the supports of the basis functions
are properly scaled with respect to j: for some N ﬁxed, satisfying N ≤
2
j0−2
, we have
supp ϑ
j,k
⊂ [(k −N)2
−j
, (k +N)2
−j
] ∩ ]0, 1[ ,
supp ϑ
0
j,k
⊂ [(k −N)2
−j
, (k +N)2
−j
] ∩ ]0, 1[ .
W.3. The interior functions of the two bases coincide: ϑ
j,k
= ϑ
0
j,k
, ∀k =
N, . . . , 2
j
− N; in other words the two bases diﬀer only for those elements
that take into account boundary conditions.
W.4. The basis functions verify
(72) [ρ
k
(x)[ ≤ C, [ρ
0
k
(x)[ ≤ C, [ϑ
j,k
(x)[ 2
j/2
, [ϑ
0
j,k
(x)[ 2
j/2
.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 685
Remark A.1. Note that the constants N and j
0
(and therefore 2
j0
) are ﬁxed once
for all and depend only on the choice of the wavelet basis. We will therefore consider
them as an O(1) as far as the estimates that we are going to prove are concerned.
The normalization estimate (72), together with property W.2 on the size of the
support, trivially yields the following bound on the L
1
norm of the basis functions
(73) ϑ
j,k

L
1
(0,1)
2
−j/2
, ϑ
0
j,k

L
1
(0,1)
2
−j/2
.
In the following we will make use of a certain number of algebraic inequalities:
we start by recalling the following bound on the sum of a geometric series
(74)
j≥1
2
−aj
1
a
,
as well as the wellknown boundedness property of the discrete convolution operator
(75)
j
[
j
a
j
−j
b
j
[
2
_
_
j
[a
j
[
_
_
2
_
_
j
[b
j
[
2
_
_
.
Moreover, since N is a ﬁxed constant, we have
(76)
_
N
k=1
a
k
_
2
N
k=1
[a
k
[
2
.
The ﬁrst bound that we consider regards the injection H
s
(0, 1) ⊂ L
∞
(0, 1),
which is known to hold for all s > 1/2 [17]. As far as the dependence on s of the
constant in the bound is concerned, we have the following lemma.
Lemma A.1. The following bound holds for all u ∈ H
s
(0, 1) with s ∈ ]1/2, 1]:
u
L
∞
(0,1)
u
L
2
(0,1)
+
1
_
s −1/2
[u[
H
s
(0,1)
.
Proof. Let u =
2
j
0
k=1
¸u, ρ
k
)ρ
k
+
∞
j=j0
2
j
k=1
¸u, ϑ
j,k
)ϑ
j,k
. Using (72) and (69), as
well as properties W.1 and W.2 (which guarantees that for x and j ﬁxed, ϑ
j,k
(x) ,= 0
for at most 2N values of k), we have (with ˇ u =
_
1
0
u ds)
sup
x
[u(x)[ sup
x
[
2
j
0
k=1
¸u, ρ
k
)ρ
k
(x) +
∞
j=j0
2
j
k=1
¸u, ϑ
j,k
)ϑ
j,k
(x)[
max
k
[¸u, ρ
k
)[ +
j≥j0
max
k
[¸u − ˇ u, ϑ
j,k
)[2
sj
2
(1/2−s)j
≤ u
L
2
(0,1)
+
_
_
j≥j0
4
(1/2−s)j
_
_
1/2
_
_
j≥j0
2
j
k=1
[¸u − ˇ u, ϑ
j,k
)[
2
2
2sj
_
_
1/2
u
L
2
(0,1)
+
1
_
s −1/2
u − ˇ u
H
s
(0,1)
.
We conclude by applying Poincar´e inequality which yields u − ˇ u
H
s
(0,1)
[u − ˇ u[
H
s
(0,1)
= [u[
H
s
(0,1)
.
686 SILVIA BERTOLUZZA
The second issue that we consider is the equivalence between the spaces H
s
(0, 1)
and H
s
0
(0, 1). This is known to hold for all s ∈ [0, 1/2[ [17]. The dependence on s
of the constant in the bound is the object of the following lemma.
Lemma A.2. The following bound holds for all η ∈ H
s
, 0 ≤ s < 1/2:
(77) η
H
s
0
(0,1)
1
1/2 −s
η
H
s
(0,1)
.
Proof. Let η ∈ H
s
(0, 1) be expanded in terms of the basis B:
η =
2
j
0
k=1
ζ
k
ρ
k
+
j≥j0
2
j
k=1
η
j,k
ϑ
j,k
,
so that
η
2
H
s
(0,1)
·
2
j
0
k=1
[ζ
k
[
2
L
2 +
j≥j0
2
j
2
j
k=1
[η
j,k
[
2
.
We start by decomposing η as the sum of four contributions:
η = η
0
+η
left
+η
center
+η
right
with
η
0
=
2
j
0
k=1
ζ
k
ρ
k
, η
center
=
j≥j0
2
j
−N
k=N
η
j,k
ϑ
j,k
=
j≥j0
2
j
−N
k=N
η
j,k
ϑ
0
j,k
, (78)
η
left
=
j≥j0
N−1
k=1
η
j,k
ϑ
j,k
, η
right
=
j≥j0
2
j
k=2
j
−N+1
η
j,k
ϑ
j,k
(79)
and by triangular inequality we have
η
H
s
0
(0,1)
≤ η
0

H
s
0
(0,1)
+η
left

H
s
0
(0,1)
+η
center

H
s
0
(0,1)
+η
right

H
s
0
(0,1)
.
In order to bound the H
s
0
(0, 1) norm of the four contributions, we will use the
equivalent norm (70). We start by observing that for all (j, n) with n = N, . . . , 2
j
−
N we have ¸ρ
k
, ϑ
0
j,n
) = ¸ρ
k
, ϑ
j,n
) = 0, while for the remaining couples (j, n) we
have
[¸ρ
k
, ϑ
0
j,n
)[ ≤ ρ
k

L
∞
(0,1)
ϑ
0
j,n

L
1
(0,1)
2
j0/2
2
−j/2
;
using (70) and observing that for all j’s the number of indexes n for which ¸ρ
k
, ϑ
0
j,n
)
,= 0 is bounded independently of j, we easily see that
ρ
k

2
H
s
0
(0,1)
2
j0
_
_
ρ
k

2
L
2 +
j≥j0
2
2js
2
−j
_
_
1
1 −2s
.
Using the fact that on R
2
j
0
the
2
and the
1
norms are equivalent (the constant
in the equivalence depending on j
0
, which, we recall, is a ﬁxed constant), we then
easily see that
η
0

2
H
s
0
1
1 −2s
_
_
2
j
0
k=1
[ζ
k
[
_
_
2
1
1 −2s
2
j
0
k=1
[ζ
k
[
2
.
We now consider η
center
. This term contains the contribution of all functions
which are interior; i.e., they do not see the boundary conditions. More precisely,
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 687
we observe that for all j
, k
such that ϑ
j
,k
= ϑ
0
j
,k
it holds that ¸η
center
, ϑ
0
j
,k
) =
η
j
,k
, while for all other j
, k
, ¸η
center
, ϑ
0
j
,k
) = 0 and this yields
η
center

2
H
s
0
(0,1)
j≥j0
2
2sj
2
j
−N
k=N
[η
j,k
[
2
.
Let us now bound the H
s
0
(0, 1) norm of η
left
, which is the contribution of those
basis functions which see the left boundary. Considering that for all N ≤ k
≤
2
j
− N, k ≤ N − 1, it holds that ¸ϑ
j,k
, ϑ
0
j
,k
) = ¸ϑ
j,k
, ϑ
j
,k
) = 0 and that for
k ≤ N −1 and k
> 2
j
−N we have supp ϑ
j,k
∩ supp ϑ
0
j
,k
= ∅, using (70), we can
write
η
left

2
H
s
0
(0,1)
2
j
0
k
=1
[¸η
left
, ρ
0
k
)[
2
+
∞
j
=j0
2
2sj
N−1
k
=1
[¸η
left
, ϑ
0
j
,k
)[
2
.
It is not diﬃcult to realize that the ﬁrst sum on the righthand side can be bound
as follows:
2
j
0
k
=1
[¸η
left
, ρ
0
k
)[
2
η
left

2
L
2
(0,1)
≤ η
H
s
(0,1)
.
In order to bound the second sum, let us now bound [¸η
left
, ϑ
0
j
,k
)[. We observe
that, thanks to (72) and (73), for all j, k, j
, k
we can write
[¸ϑ
j,k
, ϑ
0
j
,k
)[ ≤ ϑ
j,k

L
∞
(0,1)
ϑ
0
j
,k

L
1
(0,1)
2
(j−j
)/2
as well as
[¸ϑ
j,k
, ϑ
0
j
,k
)[ ≤ ϑ
j,k

L
1
(0,1)
ϑ
0
j
,k

L
∞
(0,1)
2
(j
−j)/2
,
which yield
[¸ϑ
j,k
, ϑ
0
j
,k
)[ 2
−j−j
/2
.
Then we have
[¸η
left
, ϑ
0
j
,k
)[ ≤
∞
j=j0
N−1
k=1
[η
j,k
[ [¸ϑ
j,k
, ϑ
0
j
,k
)[
∞
j=j0
2
−j−j
/2
N−1
k=1
[η
j,k
[.
Since the righthand side of the above estimate does not depend on k
, using in
equalities (74), (75) and (76), we can write
∞
j
=j0
2
2j
s
N−1
k
=1
[¸η
left
, ϑ
0
j
,k
)[
2
(N −1)
∞
j
=j0
2
2j
s
_
_
∞
j=j0
2
−j−j
/2
N−1
k=1
[η
j,k
[
_
_
2
∞
j
=j0
_
_
∞
j=j0
2
−(1/2−s)j−j

N−1
k=1
2
js
[η
j,k
[
_
_
2
_
_
∞
j=j0
2
−(1/2−s)j
_
_
2
_
_
∞
j=j0
2
2js
N−1
k=1
[η
j,k
[
2
_
_
,
688 SILVIA BERTOLUZZA
which yields
η
left

2
H
s
0
(0,1)
1
(1/2 −s)
2
_
_
∞
j=0
2
2js
N
k=1
[η
j,k
[
2
_
_
.
Analogously, we can prove
η
right

2
H
s
0
(0,1)
1
(1/2 −s)
2
_
_
∞
j=0
2
2js
2
j
k=2
j
−N+1
[η
j,k
[
2
_
_
.
Adding the bounds for η
0

H
s
0
(0,1)
, η
left

H
s
0
(0,1)
, η
center

H
s
0
(0,1)
and
η
right

H
s
0
(0,1)
, since for s in [0, 1/2[ it holds that 1 ≤ 1/(1/2 − s), in view of
the norm equivalence (69) we obtain the thesis.
Corollary A.1. For all η ∈ H
1/2
(0, 1) and for all α ∈ R it holds, for all s ∈]0, 1/2[,
that
η
H
s
0
(0,1)
≤
1
1/2 −s
η −α
H
1/2
(0,1)
+
[α[
_
1/2 −s
.
Proof. Trivially it holds that
η
H
s
0
(0,1)
≤ η −α
H
s
0
(0,1)
+α
H
s
0
(0,1)
.
In view of bound (77) the only thing that needs to be proven is a bound on
α
H
s
0
(0,1)
, which can be done by direct computation, yielding the thesis.
References
[1] Y. Achdou, Y. Maday, and O. Widlund. Substructuring preconditioners for the mortar
method in dimension two. SIAM J. Numer. Anal., 36:551–580, 1999. MR 99m:65233
[2] C. Baiocchi, F. Brezzi, and D. Marini. Stabilization of Galerkin methods and application
to domain decomposition. In Future Tendencies in Computer Science, Control and Applied
Mathematics. 1992. MR 94g:65119
[3] C. Bernardi, Y. Maday, and A.T. Patera. A new nonconforming approach to domain decom
position: The mortar element method. In H. Brezis & J.L. Lions, editors, Nonlinear Partial
Diﬀerential Equations and their Applications, Coll`ege de France Seminar, volume XI of
Notes Math. Ser. 299, pages 13–51. 1994. MR 95a:65201
[4] S. Bertoluzza. Analysis of a stabilized three ﬁelds domain decomposition method. Technical
Report 1175, I.A.N.C.N.R., 2000.
[5] S. Bertoluzza. Wavelet stabilization of the Lagrange multiplier method. Numer. Math., 86:1–
28, 2000. MR 2001h:65142
[6] S. Bertoluzza, C. Canuto, and A. Tabacco. Stable discretization of convectiondiﬀusion
problems via computable negative order inner products. SINUM, 38:1034–1055, 2000. MR
2001m:65163
[7] S. Bertoluzza and A. Kunoth. Wavelet stabilization and preconditioning for domain decom
position. I.M.A. Jour. Numer. Anal., 20:533–559, 2000. MR 2001h:65156
[8] S. Bertoluzza and G. Manzini. Wavelet stabilization of the three ﬁelds domain decomposition
method: Implementation and numerical tests. In preparation.
[9] J. H. Bramble, J. E. Pasciak, and A. H. Schatz. The construction of preconditioners for elliptic
problems by substructuring. Math. of Comp., 47(175):103–134, 1986. MR 87m:65174
[10] J.H. Bramble, J.E. Pasciak, and P.S. Vassilevski. Computational scales of Sobolev norms with
application to preconditioning. Math. Comp., 69:463–480, 2000. MR 2000k:65088
[11] F. Brezzi and M. Fortin. Mixed and Hybrid Finite Element Methods. Springer, 1991. MR
92d:65187
[12] F. Brezzi, L. Franca, D. Marini, and A. Russo. Stabilization techniques for domain decompo
sition methods with nonmatching grids. In Proc. IX Domain Decomposition Methods Con
ference.
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 689
[13] F. Brezzi and D. Marini. Error estimates for the threeﬁeld formulation with bubble stabi
lization. Math. of Comp., 70:911–934, 2001. MR 2002b:65159
[14] F. Brezzi and D. Marini. A threeﬁeld domain decomposition method. In A. Quarteroni,
J. Periaux, Y.A. Kuznetsov, and O.B. Widlund, editors, Domain Decomposition Methods
in Science and Engineering, volume 157 of American Mathematical Society, Contemporary
Mathematics, pages 27–34, 1994. MR 95a:65202
[15] A. Cohen. Numerical analysis of wavelet methods. In P.G. Ciarlet and J.L. Lions, editors,
Handbook in Numerical Analysis, volume VII. Elsevier Science Publishers, North Holland,
2000. MR 2002c:65252
[16] W. Dahmen and A. Kunoth. Multilevel preconditioning. Numer. Math., 63:315–344, 1992.
MR 93j:65065
[17] J.L. Lions and E. Magenes. Nonhomogeneous Boundary Value Problems and Applications.
Springer, 1972. MR 50:2670, MR 50:2671
[18] P. Le Tallec and T. Sassi. Domain decomposition with nonmatching grids: Augmented La
grangian approach. Math. Comp., 64:1367–1396, 1995. MR 95m:65212
Istituto di Matematica Applicata e Tecnologie Informatiche del Consiglio Nazionale
delle Ricerche, v. Ferrata 1, 27100 Pavia, Italy
Email address: silvia.bertoluzza@imati.cnr.it
URL: http://www.imati.cnr.it/~aivlis
660
SILVIA BERTOLUZZA
after applying a suitable stabilization method if needed [2, 7], can be chosen independently from each other. The disadvantage of introducing a third unknown ﬁeld is compensated for by the fact that all subdomains are treated exactly in the same way, which results in an easier implementation and possibly, when considering the parallelization of the method, in an easier balancing of the load between processors. We deal, in this paper, with the problem of the eﬃcient solution of the linear system arising when applying such a method. By applying a Schur complement technique, the solution of the resulting discrete problem can be reduced to the solution of an equation on the “trace” unknown, involving a SteklovPoincar´ type e operator which can be shown to be a pseudodiﬀerential operator of order one. The resulting linear system will then be solved by an iterative procedure, at each step of which there will be the need of computing the action of such operator on a given function. This reduces to solving independent Dirichlet problems on the subdomains, a task which can be carried out in parallel, subdomain by subdomain in a completely independent way. In order to have an eﬃcient scheme for solving the equation for the interface unknown, we will then need two things: • eﬃcient solvers for the local Dirichlet problems on the subdomains; • a preconditioner for the discrete SteklovPoincar´ operator so that the nume ber of iterations is kept as low as possible. We will concentrate here on the second issue, by analyzing and testing a class of preconditioners for the discrete SteklovPoincar´ operator. The approach that e we will follow is the substructuring one, proposed by J.H. Bramble, J.E. Pasciak and A.H. Schatz [9] in the framework of conforming domain decomposition and already applied in the framework of nonconforming methods to the mortar method in [1]. This consists in decomposing the trace space Φh in a direct sum of a coarse subspace (essentially linked to the geometrical decomposition of Ω in subdomains) and of suitable local subspaces and then considering the related blockJacobi type preconditioners. From the theoretical point of view, we study the dependence of the condition number of the resulting preconditioned matrix on the mesh size of the discretizations, aiming at giving bounds which, under suitable assumptions, are independent of the number and size of the subdomains. In particular we provide an upper bound which applies to a wide variety of approximation spaces, including ﬁnite elements and wavelets. Though only suboptimal in terms of the mesh size on the skeleton (the condition number of the preconditioned matrix grows like (log(H/h))2 , H and h being, respectively, the diameter of the subdomains and the mesh size of the corresponding discretization), the estimate obtained is uniform with respect to the number and size of subdomains. The resulting method is then potentially almost perfectly scalable. Moreover this approach has the advantage over other asymptotically better preconditioners for the three ﬁelds formulation ([7]) of being itself naturally computable in parallel (each edge in the skeleton being treated independently of the others). We conclude the paper by reporting a series of extensive numerical tests, performed on both the plain and on a stabilized formulation as proposed in [7]. The results of the tests are in agreement with the theory and conﬁrm the polylogarithmic growth (with respect to the mesh size of the discretization) of the preconditioned
THE THREE FIELDS DOMAIN DECOMPOSITION METHOD
661
system, as well as the independence (asymptotically) with respect to the number and size of the subdomains. 2. Domain decomposition method: the three fields formulation Let Ω ⊂ R2 be a convex polygonal domain. We will consider the following simple model problem: given f ∈ L2 (Ω), ﬁnd u satisfying
2
(1)
−
i,j=1
∂ ∂xj
ai,j (x)
∂u ∂xi
+ a0 (x)u = f in Ω,
u = 0 on ∂Ω.
We assume that for almost all x ∈ Ω we have 0 ≤ a0 (x) ≤ R and that the matrix (aij (x))i,j=1,2 is symmetric positive deﬁnite, with smallest eigenvalue ≥ α > 0 and largest eigenvalue ≤ α , α, α independent of x. In a domain decomposition framework, we will consider the three ﬁelds formulaK tion ([14]) of such a problem. Let Ω = k=1 Ωk , Ωk polygons, which, for simplicity, we will assume to be quadrangles. We set Γk = ∂Ωk , Σ = k Γk being the skeleton of the decomposition. Remark that diﬀerently from [14] here we deﬁne Σ so that it includes the external boundary of the domain Ω, so that there is no need of distinguishing between interior subdomains and subdomains which are adjacent to the boundary ∂Ω. This results in a simpliﬁcation both on the notational and on the implementation point of view, since all subdomains are treated in the same way. We will make the following regularity assumptions on the subdomains Ωk : (A1) The subdomains are regular in shape and the geometrical decomposition is graded, that is, (a) there exists a positive constant c0 such that, for all k, Ωk contains a ball of diameter c0 Hk , it is contained in a ball of diameter Hk , and the length of each side is bounded from below by c0 Hk ; moreover any interior angle ω satisﬁes 0 < c1 < ω < c2 < π (c0 , c1 , and c2 independent of k); (b) there exists a positive constant c3 such that, if , k are such that ∂Ωk ∩ ∂Ω  > 0, then it holds that Hk /H ≤ c3 . For technical reasons, we also need to make the following assumption (needed for the proof of (25) in the following), which in many cases can be shown to be actually a consequence of the previous one: (A2) There exists a constant C0 such that for all y, denoting by y the line of equation x2 = y and setting Ky = {k : length( y ∩ Ωk ) > 0}, it holds that k∈Ky Hk ≤ C0 . We are interested here in explicitly studying the dependence of the estimates that we are going to prove on the number and size of the subdomains. To this end, in the following we will employ the notation A B (resp. A B) to say that the quantity A is bounded from above (resp. from below) by cB, with a constant c independent of k and of the Hk ’s, as well as of any mesh size parameter. The expression A B will stand for A B A. For each k we let the norm and seminorm of H 1 (Ωk ) be deﬁned as usual by u 2 1 (Ωk ) = Ωk u2 dx + Ωk  u2 dx and u2 1 (Ωk ) = Ωk  u2 dx. On the H H
for ϕ ∈ H00 (e). and the two corresponding norms are equivalent. λ Λ = k λk 2 H −1/2 (∂Ωk ) . u = ϕ on Σ} = H0 (Ω)Σ . For e ⊂ ∂Ωk segment of extrema a and b. provided that the ratio between the length of ∂Ωk and the length of e is uniformly bounded. 1 Λ= k=1 H −1/2 (∂Ωk ). 1[. In particular it explodes as s converges to 1/2. x − y2s+1 ∂Ωk ∂Ωk The space H −1/2 (∂Ωk ) is deﬁned as the dual of H 1/2 (∂Ωk ). it holds that (3) ϕ H00 (e) 1/2 ϕH 1/2 (∂Ωk ) . The behaviour of such a constant as s approaches the limit value 1/2 (see (44)) will play a key role in the forthcoming analysis. we deﬁne the H s (e) seminorm as usual by the integral expression (2) where the integrals are taken over e. letting ϕ ∈ H 1/2 (∂Ωk ) denote the function coinciding with ϕ on e and identically vanishing on ∂Ωk \ e. respectively equipped with the norms: 1/2 1/2 (6) u V = k uk 2 H 1 (Ωk ) . x − b2s s We recall that for s < 1/2 the two spaces H (e) and H0 (e) coincide.662 SILVIA BERTOLUZZA boundary ∂Ωk we let the norm of H 1/2 (∂Ωk ) be deﬁned by ϕ H 1/2 (∂Ωk ) = u∈H 1 (Ωk ) u=ϕ on ∂Ωk inf u H 1 (Ωk ) . respectively. However. 1/2 ˇ Remark also that. the constant in the equivalence being uniformly bounded. and (see [4]) 1/2 (7) ϕ ϕ = 1 u∈H0 (Ω) u=ϕ on Σ inf u H 1 (Ω) k ϕ2 1/2 (∂Ωk ) H . we deﬁne the H s (∂Ωk ) seminorm by (2) ϕ2 s (∂Ωk ) = H (ϕ(x) − ϕ(y))2 ds(x) ds(y). The functional setting for the three ﬁelds domain decomposition method is given by the following spaces: K K (4) and (5) V = k=1 H (Ωk ). which we will equip with the norm ϕ 2 1/2 H00 (e) = ϕ2 1/2 (e) + H e ϕ(x)2 ds(x) + x − a2s s e ϕ(x)2 ds(x). the constant in the equivalence depends a priori on Hk and on s. For all s in ]0. On e we will 1/2 s also consider the spaces H0 (e) (s = 1/2) and H00 (e) of functions whose extension by zero is in H s (∂Ωk ) (s = 1/2) and H 1/2 (∂Ωk ). . 1 1 Φ = {ϕ ∈ L2 (Σ) : there exists u ∈ H0 (Ω).
k where νa denotes the outer conormal derivative to the subdomain Ωk . such that . Λh = k=1 Λk ⊂ k=1 H −1/2 (∂Ωk ) h and Φh ⊂ Φ. Ωk λk ψ ds ∂ = 0. H ak (w. and semideﬁnite. w). ϕ) where u is indeed the solution of (1) and such that ∂uk on Γk .j=1. H 1 (Ωk ) The three ﬁelds formulation of equation (1) is the following: ﬁnd (u. v) ≤ L w v H 1 (Ωk ) . we consider the problem: ﬁnd (uh . ∀v ∈ H (Ωk ).THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 663 Let ak : H 1 (Ωk ) × H 1 (Ωk ) → R denote the bilinear form corresponding to the linear operator considered: ∂w ∂v aij (x) + a0 (x)wv dx. v k ) − v k λk ds ∂Ωk uk µk ds − (8) and ∀ψ ∈ Φ : k ∂Ωk ∂Ωk = Ωk f v k dx. It is known that this problem admits a unique solution (u.2 and on a0 . Remark that in such a formulation the homogeneous Dirichlet boundary condition is embedded in the deﬁnition of the space Φ. ϕ) ∈ V × Λ × Φ such that k 1 k −1/2(∂Ωk ) : ∀k. k ∂νa (9) λk = ϕ = u on Σ. λ. Equation (8) can be discretized by a Galerkin scheme: after choosing discretizaK K K K k tion spaces Vh = k=1 Vh ⊂ k=1 H 1 (Ωk ). v) = ∂xi ∂xj Ωk i. λh . ∀µ ∈ H ak (uk .j=1 2 Under the assumptions made on the matrix (aij (x))i. ak (w. µk ϕ ds = 0. λ. More precisely. ϕh ) ∈ Vh × Λh × Φh . v in H 1 (Ωk ) αw2 1 (Ωk ) ≤ ak (w. there exist positive constants α and L independent of k and Hk such that for all w. the bilinear forms ak are uniformly H 1 (Ωk )continuous.
The matrix S = CA−1 CT does not need to be assembled. there exists a unique solution to problem (10). v k ) − vh λk ds h h h ∂Ωk uk µk ds h h and ∀ψh ∈ Φh : k ∂Ωk = Ωk k f vh dx. ∀vh ∈ Vh . (c) the following compatibility condition between Λh and Φh holds: Ωk λk ϕh ds h (12) ϕh ∈Φh λh ∈Λh ∂ inf sup k λh Λ ϕh Φ ≥ α > 0. uniqueness and stability of the solution of (10) rely on the validity of suitable infsup conditions. ∀µh ∈ Λh : k ak (uk . λh and ϕh in the bases chosen for Vh . (13) ϕh 0 C 0 0 (uh . C = [ 0 C ]. A= A B BT 0 . The linear system stemming from (10) takes the form uh fh 0 A BT B 0 C T · λh = 0 . By a Schur complement argument the solution of (13) can be reduced to a system in the unknown ϕh of the form (14) CA−1 CT ϕh = CA−1 fh 0 . k (b) the following compatibility condition between Vh and Λk holds unih formly in k: Ωk λk uk ds h h (11) λk ∈Λk uk ∈V k h h h h inf sup ∂ uk h H 1 (Ωk ) λk h H −1/2 (∂Ωk ) ≥ β > 0. k k k k k Remark that (A3)(a) implies that for all vh ∈ ker Bh = {vh ∈ Vh : ∂ Ωk vh µk = h k 0 ∀µk ∈ Λk } it holds that ∂ Ωk vh ds = 0 and then it is not diﬃcult to see that a h h k is elliptic on ker Bh . by a wellknown argument ([11]). The system (14) can rather be solved by an iterative technique (like for instance a conjugate gradient . respectively). λh . and ϕh being the vectors of the coeﬃcients of uh . − ∂Ωk µk ϕh ds = 0. h Ωk λk ψh ds h ∂ = 0. which converges with an optimal rate to the solution of (8). Existence. Then. Λh and Φh .664 SILVIA BERTOLUZZA one has (10) k k k k ∀k. Λk and Φh are such that h k (a) Λh contains the constant functions. More precisely we can make the following assumption: k (A3) The spaces Vh .
k = 0.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 665 method) and therefore only the action of S on a given vector needs to be implemented. 2 [η. ψh Φ. ξ] 1 . in practice. This reduces. 2 for B positive constant independent of k and Hk . ξ ∈ H 1/2 (∂Ωk ) (18) [η. h h h h h 2 . v ) + γ[uk .k + Ωk λk ψh ds + γ[ϕh . we obtain an identity. h S = CA−1 CT . We consider then the following discrete stabilized formulation: ﬁnd uh ∈ Vh . ∀vh ∈ Vh . i. ψh ] 1 .k ≥ 0.. We consider here the approach of [7. η] 1 . In particular the requirement that both (A3)(b) and (A3)(c) hold simultaneously poses some serious limitation on the choice of the discretization spaces for the three unknowns u. In order to be more free in such a choice. quite restrictive. v k ] 1 − vh λk ds − γ[ϕh . The assumptions needed to guarantee existence and uniqueness of the solution of the discrete problem. ∀µh ∈ Λh : k k k k k k a (u . h h 2 2 k k ∂ k where γ > 0 is a parameter independent of the choice of the discretization spaces. If (A3) holds. by substituting the solution (u. and more speciﬁcally in order to be allowed to choose the discretization space for the interface unknown ϕ independently of the choice of the other two discretization spaces. h h h ∂Ωk ∂Ωk and ∀ψh ∈ Φh : − γ[uk . (15) s(ϕh . Assumption (A3) is.∂Ωk ξ1/2. satisfying the 2 following bounds: for all η. ψh ) = ψ T Sϕh .e. ϕ) of (8) at the place of (uh . ψh ) s(ϕh . 12. Such formulation is consistent with the original continuous problem. Multiplying by S implies the need for solving a linear system with matrix A. Diﬀerent ways have been proposed for stabilizing the three ﬁelds formulation ([2.k ≤ Bη1/2. λh . λ.k 2 ∂Ωk Ωk uk µk ds − µk ϕh ds = 0.k = f vh dx. λh ∈ Λh .∂Ωk . vh ] 1 . λ and ϕ.1. ϕh ) in (19). satisﬁes the following continuity and coercivity assumptions: Lemma 2. corresponding to the Schur complement matrix S. ϕh ∈ Φh such that it holds that (19) k k k k ∀k.k : H 1/2 (∂Ωk ) × H 1/2 (∂Ωk ) → R be given. it is also possible to consider a stabilized version. ϕh ) ϕh ϕh Φ 2 Φ. as well as an error estimate. 13]). can now be made on the one . ·] 1 . ψh ] 1 . then the bilinear form s is continuous and coercive with respect to the Φ norm: (16) (17) s(ϕh . Moreover under the above assumptions it is possible to prove that the bilinear form s : Φh × Φh −→ R. to independently solving K discrete Dirichlet problems with Lagrange multipliers in the K subdomains. For all k let a bilinear form [·. by a proper reordering of the unknowns. 4].
through a relation of the form (23). Φ ψh Φ. The linear system stemming from such a problem takes the following form this time: ˇ uh fh A B T −γDT B (20) 0 C T · λh = 0 . Under assumptions (A3)(a). H 2 It is then possible to show that the discrete problem (19) admits a unique solution satisfying an optimal error estimate ([4]). 10. ϕh ) ϕh 2 Φ. It is beyond the goals of this paper to thoroughly discuss the construction of suitable bilinear forms [·. ·] 1 . 5. 6] for a detailed analysis of possible solutions. ˇ Sϕh := −DA−1 DT + γE ϕh = −DA−1 fh 0 Once again we let s : Φh × Φh −→ R be the bilinear form corresponding to the Schur complement matrix S (23) s(ϕh . E and F deriving from the stabilizing terms). in the framework of conforming domain decomposition. the solution of (20) can be reduced to a system in the unknown ϕh . ϕh −γD C γE 0 ˇ with A = A + γF (the matrices D. We recall that 2 this can be done with the aid of multiscale techniques and we refer to [7. h and also for the stabilized formulation the following holds ([4]): Lemma 2. this time taking the form (21) with (22) ˇ A= ˇ A B BT 0 .2. . For example we can 2 replace assumption (A3)(c) with (A3)(c ) There exists a positive constant b such that for all ϕh ∈ Φh it holds for all k that [ϕh .k on Φh ∂ Ωk . ψh ) ϕh Φ ψh Φ. ψh ∈ Φh that (26) s(ϕh . 3. The approach that we will follow is the one proposed. on the action of the bilinear forms [·. (c ) the bilinear form s is continuous and coercive with respect to the Φ norm: (24) (25) s(ϕh . ·] 1 . we end up with an equation on the interface space Φh of the form Sϕh = f h corresponding. ψh ) s(ϕh . (b). Our aim is to provide a preconditioner for the matrix S. ψh ) = ψ T Sϕh . Preconditioning by substructuring For both original and stabilized three ﬁelds formulation. D = [ −γD C ]. ϕh ) ϕh Φ ϕh 2 . ϕh ] 1 .666 SILVIA BERTOLUZZA hand on the spaces Vh and Λh and on the other hand in a completely independent way.k ≥ bϕh 2 1/2 (∂Ωk ) . Again.k . to a bilinear form s : Φh × Φh → R verifying for all ϕh . s(ϕh .
H. Following [9]. LH = {ϕ ∈ C 0 (Σ) : ∀i = 1. We make the following assumption: (A4) There exists a constant c4 such that it holds that sup sup k i∈Ek ∂Ωk  ≤ c4 .2. h.i corresponding blockJacobi type preconditioners. ϕ0 = 0 on Σ \ ei h. ϕ = 0 on ∂Ω}. (corresponding. Schatz in [9] and already studied in [1] for the mortar method case. K. with Ek = {i : ei ∩ ∂Ωk  > 0}. In order to construct the preconditioner we make the following assumptions on the approximation space Φh : (A5) The following two conditions hold: (a) LH ⊂ Φh . where · stands for various norms and seminorms (the constant in the inequality depending on c4 ).i veriﬁes ϕ0 ∈ Φh . In a ﬁnite element framework it reduces to asking that the extrema ai and bi of the segments ei into which the skeleton is split be nodes of the ﬁnite element grid for Φh . for instance. k = 1. 0 ≤ r < t ≤ 1: (27) (28) ϕh H t (∂Ωk ) ϕh H t (ei ) hr−t ϕh H r (∂Ωk ) . h. k hr−t ϕh H r (ei ) . Let us consider a splitting of the interior part of the skeleton Σ \ Ω as the disjoint union of M segments ei of extrema ai and bi M Σ \ ∂Ω = i=1 ei .H. More precisely. . the function ϕ0 deﬁned as h. This will allow us several times 2 2 to write inequalities of the form i∈Ek qi i∈Ek qi .1. denote the subset of those functions of Φ whose restriction to each segment ei is linear. . build up the boundary of the subdomain Ωk : ∂Ωk \ ∂Ω = i∈Ek ei .THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 667 by J. .i Remark 3.i (A6) Φh ⊂ H 1 (Σ) and for all ϕh ∈ Φh the following two inverse inequalities hold for all k. (b) for all ϕh ∈ Φh such that at the extrema ai and bi of some edge ei it holds that ϕh (ai ) = ϕh (bi ) = 0. to the splitting of Σ as the union of segments of the form ∂Ωk ∩ ∂Ω ). . k ∀i ∈ Ek . . . The parameter hk appearing in the inverse inequality (27) has here the meaning of the smallest mesh size of the restriction to ∂Ωk of Φh . ei  Note that (A4) implies that #(Ek ) ≤ c4 for all k. Assumption (A5) is not at all restrictive. let LH ⊂ Φ.i ϕ0 = ϕh on ei . . the main idea for constructing a preconditioner for the matrix S is to decompose Φh as the direct sum of a coarse space LH and some local spaces Φ0 (one for each edge ei of the decomposition of Σ) and then considering the h. For each k let Ek individuate those segments that. t. ϕei ∈ P1 (ei ). J.E. . Remark 3. together with ∂Ω ∩ ∂Ωk . Pasciak and A. Bramble. and for all r. . M.
It holds that (31) ˇ cond(S −1 S) max 1 + log k Hk hk 2 . ϕh ) ˇ max 1 + log k Hk hk 2 s(ϕh . h Φ0 h = i=1 {ϕh ∈ Φ0 h : ϕh = 0 on Σ \ ei } ∼ i=1 Φ0 ei . . with ηh = ˇ 0 0 0 0 ηH + ηh . si (ξ. the matrices corresponding to the Galerkin discretization of the bilinear forms s and s. ζh ∈ Φh . h with Φ0 deﬁned as h Φ0 = {ϕh ∈ Φh . the preconditioner will be ﬁnally constructed with the aid of a bilinear form s : Φh × Φh → R assembled by blocks: for ηh . and in the following we will write ϕh H 1/2 (ei ) h 00 for ϕh ei H 1/2 (ei ) (and analogously for other norms deﬁned on portions of Σ). Note that all ϕh ∈ 1/2 Φ0 verify ϕh ei ∈ H00 (ei ) for all i. For all ϕh ∈ Φh it holds that (30) s(ϕh . ζH ) + ˇ i=1 0 si (ηh ei .1. h For each i = 1. where we denote by S and S. In the following section we will ˇ brieﬂy recall a possible solution and we refer to [9.1 implies that we can derive the following ˇ corollary. ξ) ˇ ξ 2 . Theorem 3. By a wellknown argument. Theorem 3. ζh ∈ Φ0 we set h M (29) ˇ s(ηh . ˇ 0 The main result of this paper is the following theorem. ζH ∈ LH it holds that (B2) sH (ξH . 00 η H00 (ei ) 1/2 . h The subspace Φ0 veriﬁes h M M ϕh = ϕH + ϕ0 . ξ ∈ Φ0 ei : h (B1) si (ξ. ϕh ). · · · .668 SILVIA BERTOLUZZA Under assumption (A5) it is easy to see that Φh can be split as Φh = LH ⊕ Φ0 . η) ˇ ξ H00 (ei ) 1/2 where the symbol ∼ means that the two spaces are isomorphic. ϕh ) s(ϕh . M . ˇ Corollary 3.1. 16] for further discussions on this topic. ζh ei ). satisfying the following continuity and positivity bounds uniformly in h: for all η. 10. we assume we are given a bilinear form si : Φ0 ei × ˇ h 0 Φh ei → R. 1/2 H00 (ei ) Moreover we assume that we are given a bilinear form sH : LH × LH → R such ˇ that for all ξH . ξH ) ˇ ξH 2 Φ. ζH ∈ LH and ηh . Following [9]. ζh = ζH + ζh . ϕh = 0 at the extrema ai and bi of each edge ei of Σ}. respectively. ηH ) ˇ ξH Φ ζH Φ. ηH . It is beyond the goals of this paper to fully discuss the diﬀerent ways in which ˇ the bilinear forms si and sH can be constructed. ζh ) = sH (ηH . sH (ξH .
i which implies (since #(Ek ) ≤ c4 and since each i belongs to Ek for at most two k’s) ϕ0 h M ϕ0 h 2 Φ k ϕ0 2 1/2 (∂Ωk ) h H k i∈Ek ϕ0 2 1/2 (∂Ωk ) h. using the linearity of ϕH . Moreover we will be working with functions and operators deﬁned directly on the interface.2. the matrix S is indeed the Schur complement matrix deﬁned by either (14) or (21).i H i=1 1/2 ϕ0 h 2 . Lemma 3. In order to prove Theorem 3. we will need several bounds. we will basically follow the guidelines of the proof of the analogous theorems of [9. it holds that h h h M ϕh 2 Φ ϕH 2 Φ + i=1 ϕ0 h 2 1/2 H00 (ei ) max 1 + log k Hk hk 2 ϕh 2 Φ. Assume that Φh satisﬁes assumptions (A5) and (A6). we have (34) ϕh 2 L∞ (ei ) 1 + log Hk hk ϕh 2 1/2 (ei ) . 1/2 H00 (ei ) M 0 ∈ Φ0 as ϕ0 = where the last bound descends by splitting h h i=1 ϕh. ˇ In view of the deﬁnition of si and sH on the one hand and of property (26) of ˇ s on the other hand. Then.1.1 reduces to proving the following result: Theorem 3. We start by observing that it holds that M ϕh 2 Φ ϕH 2 Φ + ϕ0 h 2 Φ ϕH 2 Φ + i=1 ϕ0 h 2 .2. In order to prove Theorem 3. it h h h holds that M M (33) i=1 (ϕH (ai ) − ϕH (bi ))2 + i=1 ϕ0 h 2 1/2 H00 (ei ) max 1 + log k Hk hk 2 ϕh 2 Φ. 1/2 H00 (ei ) the last inequality deriving from the H00 (ei ) to H 1/2 (∂Ωk ) boundedness of the trivial extensionbyzero operator. The main diﬀerence is that here we consider a general discretization space. proving Theorem 3. ai and bi being the extrema of ei . 1]. Moreover. Then the following bounds hold: ¯ (i) for all ϕh ∈ Φh such that ϕh (p) = 0 for some p ∈ ei with i ∈ Ek . H . by direct computation. For all ϕh ∈ Φh with ϕh = ϕH + ϕ0 (ϕH ∈ LH and ϕ0 ∈ Φ0 ). We start by proving the following Lemma. one can see that it holds that K M (32) ϕH 2 Φ k=1 ϕH 2 1/2 (∂Ωk ) H i=1 (ϕH (ai ) − ϕH (bi ))2 .THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 669 We recall that by the deﬁnition of the bilinear form s.1. not necessarily consisting in piecewise linear functions. proving (30) is equivalent to proving that ∀ϕh ∈ Φh with ϕh = ϕH + ϕ0 (ϕH ∈ LH and ϕ0 in Φ0 ).
Let now ϕh ∈ Φh and α ∈ R. we start by remarking that for ϕ0 ∈ Φ0 and for h h i ∈ Ek . h 0 . H ¯ Now we remark that ϕh (p) = 0 for some p ∈ ei implies (42) α ϕh − α L∞ (ei ) and we get (34) by triangular inequality. Let i ∈ Ek . By assumptions (A1)(a) and (A4) we have that ei  start by observing that for any ζ ∈ H 1/2+ε (ei ) it holds that (37) ζ 2 L∞ (ei ) −1 Hk ζ 2 L2 (ei ) 2ε Hk ζ2 1/2+ε (ei ) . Applying Poincar´ inequality gives e (39) ϕh − α 2 L∞ (ei ) −1 Hk ϕh − α 2 L2 (ei ) + log (40) which yields (41) Hk −1/2 ϕh − α L2 (ei ) ϕh H 1/2 (ei ) Hk hk ϕh − α 2 L∞ (ei ) 1 + log ϕh 2 1/2 (ei ) . inequality (37) for ζ = ϕh − α combined with (38) yields: (38) Hk ϕh 2 1/2 (ei ) . H ai and bi being the extrema of ei . which we repeat here for completeness: let us choose α to be the average of ϕh on ei . H hk In view of (39).i h vanishing on ∂Ωk \ ei and using (3) and the inverse inequality (27).i h−ε ϕ0 H 1/2−ε (∂Ωk ) h. h H Hk . ζ 2 ∞ (ˆ) L e L e e ε A change of variable then yields (37). (ii) then follows by applying (i) to the function ϕ(x) − ϕ(ai ) (which vanishes in p = ai ).i k ϕ0 H 1/2−ε (ei ) . H ε This is easily seen by observing that on the reference segment e = ]0. As far as (iii) is concerned. 1[ the conˆ e e tinuity bound of the injection H 1/2+ε (ˆ) ⊂ L∞ (ˆ) depends on ε as follows (see e Appendix A. we obtain the following bounds (43) ϕ0 h H00 (ei ) 1/2 ϕ0 H 1/2 (∂Ωk ) h. (34) can then be proven by applying exactly the arguments of [9]. By assumption (A6) it holds that + 2ε 2ε Hk h−2ε H 2ε Hk k ϕh − α2 1/2+ε (ei ) = k ϕh 2 1/2+ε (ei ) ϕh 2 1/2 (ei ) .670 SILVIA BERTOLUZZA (ii) for all ϕh ∈ Φh it holds that (35) (ϕh (ai ) − ϕh (bi ))2 1 + log Hk hk ϕh 2 1/2 (ei ) . denoting by ϕ0 the function coinciding with ϕ0 on ei and identically h. H H H ε ε ε After choosing ε = 1/ log(Hk /hk ). Lemma A. i ∈ Ek . We Proof. (iii) for all ϕ0 ∈ Φ0 it holds that h h (36) i∈Ek ϕ0 h 2 1/2 H00 (ei ) 1 + log Hk hk 2 ϕ0 2 1/2 (∂Ωk ) .1): for all ζ ∈ H 1/2+ε (ˆ) 1 2 ζ 2 2 (ˆ) + ζH 1/2+ε (ˆ) .
the main ingredient of the proof of Theorem 3.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 671 where the second inequality descends from (27) by applying a standard functional r spaces interpolation argument. (44) ζH 1/2−ε (ˆ) e e 0 ε ε which rescales as ε Hk Hε −1/2 Hk ζ − α L2 (ei ) + ζ − αH 1/2 (ei ) + √k α. for ϕh = ϕH + ϕ0 . taking ζ = ϕ0 and α as its average on ei . Let ϕ0 ∈ Φ0 and ζH ∈ LH . In that case.1) e that the following bound holds for all ζ ∈ H 1/2 (ˆ) and for all α ∈ R: 1 1 ζ − α H 1/2 (ˆ) + √ α. In particular it is possible to prove (see Appendix A.2 is then the following lemma. with equivalent norms. We then take once again ε = 1/ log(Hk /hk ). the implicit constant in the bound would depend on ε. Since for s < 1/2 the two spaces H s (ei ) and H0 (ei ) coincide. 0 < r < 1/2. H00 (ei ) 1/2 1 + log ϕ0 H 1/2 (ei ) . the s seminorm  · H0 (ei ) in equation (43) could be replaced by the standard H s (ei ) seminorm. ϕH ∈ LH . we obtain ϕ0 h Since i∈Ek  · 2 1/2 (ei ) H i ∈ Ek .1. thanks to bound (41). . In fact. we can apply on the one hand h Poincar´ inequality (40) and on the other hand the bound (42). the constant in the equivalence depending on ε. since the spaces H0 (ei ). and. ϕ0 ∈ Φ0 . can be ob1/2 tained as the interpolants of order 2r between L2 (ei ) and H00 (ei ). using the corresponding the H0 h 1/2−ε (ˆ) e bound on the reference segment e = ]0. Again we bound 1/2−ε (ei ) seminorm of ϕ0 by a change of variable. We recall once more that H0 ˆ e and H 1/2−ε (ˆ) coincide. by squaring and then taking the sum over H s Remark 3. ζH 1/2−ε (ei ) 0 ε ε Again. however. (33) is obtained. by applying (45) for ζH = ϕH . 1[ . which holds since e ϕ0 (ai ) = 0. Lemma 3. Corollary A. Analogously to what happens in [9]. h  · 2 1/2 (∂Ωk ) . we can h h h write M ϕ0 h i=1 2 1/2 H00 (ei ) = 1 2 K ϕ0 h k=1 i∈Ek 2 1/2 H00 (ei ) 2 1 + log k Hk hk Hk hk ϕh 2 1/2 (∂Ωk ) H 2 max 1 + log k ϕh 2 Φ. thanks to (ii) of Lemma 3. h H Provided such lemma holds. Then for all k it holds that h h (45) i∈Ek ϕ0 h 2 1/2 H00 (ei ) 1 + log Hk hk 2 ϕ0 + ζH 2 1/2 (∂Ωk ) .3. we obtain (36).2. and we get h ϕ0 h H00 (ei ) 1/2 H ε h−ε 0 H ε h−ε k ϕh H 1/2 (ei ) + √ k ϕ0 − α h ε ε Hk hk L∞ (ei ) .
h (46) Φk. h H Let us bound the second sum on the righthand side of (48). adding and subtracting ζ 0 at the lefthand side of (45). ξ)H 1/2 (∂Ωk ) = (ζH . H (53) I1 (ζ 0 ) = ai ζ 0 (x)2 dx.672 SILVIA BERTOLUZZA which. 1 ≤ k ≤ K.0 and then we can conclude by standard arguments that h (47) ζ 0 H 1/2 (∂Ωk ) ζH H 1/2 (∂Ωk ) . x − ai  bi I2 (ζ 0 ) = ai ζ 0 (x)2 dx. h ϕ0 + ζ 0 h 2 1/2 H00 (ei ) 1 + log Hk hk Hk hk 2 ϕ0 + ζ 0 2 1/2 (∂Ωk ) h H 2 (51) 1 + log ϕ0 + ζH 2 1/2 (∂Ωk ).0 satisfying h (ζ 0 . ·)H 1/2 (∂Ωk ) is the bilinear form (ζ. we have (50) i∈Ek (ζH − ζ 0 . 1/2 H00 (ei ) Since ϕ0 + ζ 0 ∈ Φk. x − bi  . we can write: (48) i∈Ek ϕ0 h 2 1/2 H00 (ei ) ϕ0 + ζ 0 h i∈Ek 2 1/2 H00 (ei ) + i∈Ek ζ0 2 . It is easy to check that  · H 1/2 (∂Ωk ) is a norm on Φk. yields (33). h ∂Ωk ∂Ωk inducing the H 1/2 (∂Ωk ) seminorm. combined with (35). We start by introducing a function ζ 0 ∈ Φk.1(iii).0 . where (·. by the deﬁnition of ζ 0 we have that h h (49) which yields ϕ0 + ζH 2 1/2 (∂Ωk ) ≥ ϕ0 + ζ 0 2 1/2 (∂Ωk ) . deﬁned as the unique element of Φk. ϕ0 + ζ 0 )H 1/2 (∂Ωk ) = 0. ξ)H 1/2 (∂Ωk ) .0 . The only thing that we need to do then is to prove Lemma 3. We recall that we have (52) with bi ζ0 2 1/2 H00 (ei ) = ζ 0 2 1/2 (ei ) + I1 (ζ 0 ) + I2 (ζ 0 ).2.2. Let k. h h ζ 0 depending on ζH . h h H H The ﬁrst sum on the righthand side of (48) can then be bound thanks to Lemma 3. Proof of Lemma 3. ξ)H 1/2 (∂Ωk ) = (ζ(x) − ζ(y))(ξ(x) − ξ(y)) dx dy x − y2 ∀ξ ∈ Φk.0 = {ϕh ∈ Φh ∂Ωk : ϕh = 0 at the cross points of ∂Ωk } = Φ0 ∂ Ωk .0 . Now.
using (34). H H 0 while. we can write h (54) i∈Ek ζ 0 2 1/2 (ei ) H ζ 0 2 1/2 (∂Ωk ) H ζH 2 1/2 (∂Ωk ) H (55) i∈Ek (ζH (ai ) − ζH (bi ))2 = i∈Ek (56) (57) ([ϕ0 + ζH ](ai ) − [ϕ0 + ζH ](bi ))2 h h 1 + log Hk hk ϕ0 + ζH 2 1/2 (∂Ωk ) . Remarking that ζ ⊥ (0) = −ζH (0). T [ with ai = 0 and bi = T . We have T (58) (59) (60) I1 (ζ 0 ) = 0 T ζ 0 (x)2 dx x ζ 0 (x) + ζH (x) − ζH (x) + ζH (0) − ζH (0)2 dx x ζ 0 (x) − ζH (x) + ζH (0)2 dx x T = 0 T 0 (61) Let us set ζ ⊥ 0 + 0 ζH (x) − ζH (0)2 dx.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 673 (we recall that ai and bi denote the two vertices of the edge ei ). For notational simplicity let us identify ei = ]0. Now. we bound the second by T hk ζ ⊥ (x) − ζ ⊥ (0)2 dx x ζ ⊥ − ζ ⊥ (0) 1 + log Hk hk T 2 L∞ (ei ) 2 hk 1 dx x ζ ⊥ 2 1/2 (ei ) . we have T 0 hk T 0 ζ 0 (x) − ζH (x) + ζH (0)2 dx = x = ζ ⊥ (x) − ζ ⊥ (0)2 dx x ζ ⊥ (x) − ζ ⊥ (0)2 dx x T 0 + hk ζ ⊥ (x) − ζ ⊥ (0)2 dx. (32) and (35). H . by hk 0 ζ ⊥ (x) − ζ ⊥ (0)2 dx x hk = 0 hk 1 x x x 0 ⊥ ζx (τ ) dτ 2 dx ⊥ ζx (τ )2 dτ dx 0 hk ζ ⊥ 2 1 (ei ) ζ ⊥ 2 1/2 (ei ) . since ϕ0 vanishes at ai and bi . h H Let us now bound I1 (ζ 0 ). x The ﬁrst term is bound. x = ζ − ζH . using CauchySchwarz’s inequality and inequality (28). using equations (47).
In each subdomain Ωk we take a uniform mesh T k composed of nk × nk equal k square elements of size δk × δk . 1[ . With such a choice it is possible to prove that (A3) holds (see [4]). (62) and (63) back into (52) and sum over i ∈ Ek . H = 1/N . 1[ in K = N × N equal square subdomains of size H × H. we obtain the bound H (64) i∈Ek ζ0 2 1/2 H00 (ei ) 1 + log Hk hk Hk hk ϕ0 + ζH 2 1/2 (∂Ωk ) h H 2 (65) + 1 + log ζ 0 − ζH 2 1/2 (∂Ωk ) . assumption (A6) holds for hk = mini∈Ek hi . where Q1 (τ ) denotes the space of polynomials of degree less than or equal to one in each variable. With this choice. δk = H/nk = 1/(N nk ). on Γ = ∂Ω. in Ω = ]0. The second integral in (60) can be bound directly using the linearity of ζH . h H ζ 0 − ζH 2 1/2 (ei ) + 1 + log H Hk hk ζH + ϕ0 2 1/2 (ei ) . we conclude that Hk Hk ζ 0 − ζH 2 1/2 (ei ) + 1 + log H hk hk By using the same arguments we get a similar bound for I2 : (62) I1 (ζ 0 ) 1 + log (63) I2 (ζ 0 ) 1 + log Hk hk 2 2 ζH + ϕ0 2 1/2 (ei ) . ∀τ ∈ T k }. By observing that for any w ∈ H 1/2 (∂Ωk ) it holds that i∈Ek w2 1/2 (ei ) H w2 1/2 (∂Ωk ) . Then. The multiplier space Λk is then deﬁned as the trace on ∂Ωk of h k Vh . decomposition of Ω = ]0.674 SILVIA BERTOLUZZA where we used T Hk . adding h up the diﬀerent contributions. T 0 ζH (x) − ζH (0)2 dx x (ζH (bi ) − ζH (ai ))2 1 + log Hk hk ζH + ϕ0 2 1/2 (ei ) h H where again we used (35) and the fact that ϕ0 vanishes at ai and bi . We consider a grid G on Σ obtained by uniformly splitting each ei into Li elements of size hi and we deﬁne Φh to be the space of P1 ﬁnite elements on such a grid: Φh = {ϕh ∈ C 0 (Σ) : ϕh κ ∈ P1 (κ). h H 4. Numerical Tests We will test the preconditioner we propose on the following problem: ﬁnd u such that (66) −∆u = f. H We now observe once again that equation (49) implies that ζH − ζ 0 2 1/2 (∂Ωk ) H ζH + ϕ0 2 1/2 (∂Ωk ) and we get the thesis. h H We can now insert bounds (54). u = 0. . We remark that ei  = H for all i. We then deﬁne Vh to be k the space of Q1 ﬁnite elements on the mesh T . 2(N −1)N ei with ei = Γk ∩ Γ for some The skeleton Σ \ ∂Ω is split as Σ \ ∂Ω = i=1 k = k(i) and = (i). geometrically conforming. ϕh ∂Ω = 0}. k Vh = {uh ∈ C 0 (Ω) : uh τ ∈ Q1 (τ ). 1[ × ]0. 1[ × ]0. ∀κ ∈ G. We consider a uniform.
in the framework described above we set nk = n for all k and Li = n for all i. Plain formulation.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 675 0 Letting ϕ0 (resp. With Ri being positive deﬁnite.i where Ri is the stiﬀness matrix associated to the discretization of the operator −d2 /dx2 by P1 ﬁnite element on ei with homogeneous Dirichlet boundary conditions at the extrema ai and bi . For all tests we set f = 1. The results are summarized in Tables 1 and 2. ηH ∈ LH } coincides with the set of Q1 ﬁnite elements η on the coarse grid corresponding to the decomposition Ω = Ωk . 1 ˜ • (Laplace) Denoting by ηH and ζH the H0 (Ω) functions obtained by lifting ˜ ηH and ζH harmonically in each subdomain.i h. In such a case we obtain the same solution that we would get using a conforming ﬁnite element method on a regular grid of N n × N n square elements of dimension h × h. 4.i as 0 si (ϕ0 ei . we deﬁne the bilinear form si : Φ0 ei × Φ0 ei h. As one can see. deﬁned by uh Ωk = uk . It is easy to realize that the discrete solution veriﬁes uk = ϕh on ∂Ωk . We report the number of iterations needed to reduce the residual of a factor 10−5 . The use of a nonconforming method is therefore. using ϕh = 0 as initial guess. 32]. ζH ) := 2 ˇ i=1 (ηH (ai ) − ηH (bi ))(ζH (ai ) − ζH (bi )).i ˇ the interior nodes of the grid Gei . we set (68) sH (ηH .i h. and then the function uh . ˜ note that the set {˜H . With such a choice. and then applying the coarse preconditioner reduces to numerically solving the Laplace equation on such space. In order to study both the dependence on H (size of the subdomains) and on h we tested the preconditioner for diﬀerent values of n in the range [5. . that is. h h 1 veriﬁes uh ∈ H0 (Ω).1. its square root is well deﬁned and it is computed and stored at the start. in this case. ψh ei ) = (ψ 0 )T (Ri )1/2 ϕ0 . ζH ) := ˇ Ω ˜ ηH ζH dx. We solved the linear system by a preconditioned conjugate gradient method. We start by considering the case of a conforming domain decomposition.1.1. In this case δk = hi = h = H/n.i h. we considered two choices: ˇ • (BPS) Following [9]. 40] and N in the range [4. Conforming decomposition. We tested both coarse preconditioners for all combinations of N and n. As far as the coarse preconditioner sH is concerned. we set M (67) sH (ηH . ˇ h h. only a device to implement a solver for the discrete problem in an easily parallelizable way. it is not diﬃcult to check that the infsup condition (12) is satisﬁed uniformly in h and then we are in the range of Lemma 2. 4. ψ 0 ) denote the vector of the values of ϕ0 (resp ψh ) ∈ Φ0 at h h h. they are in close agreement with the theory (note that in this case maxk hk /Hk = h/H = n).1.
. We considered two test conﬁgurations. Nevertheless. with coarse preconditioner given by (67) for diﬀerent combinations of the number K = N 2 of subdomains and n of elements per edge (n2 elements per subdomain). without stabilization (formulation (10)).1. 10 12 14 15 14 17 20 23 15 18 22 26 15 19 23 27 16 20 23 27 16 20 24 27 16 20 24 27 16 20 24 27 K = N2 16 64 144 256 400 576 784 1024 Table 2. for all subdomains except two.676 SILVIA BERTOLUZZA Table 1. we set Li = min{nk . we set nk = 10. . In the ﬁrst case (see Figure 1. # Iter. except the common edge Γ13 ∩ Γ14 (the two subdomains are adjacent). with coarse preconditioner given by (68) for diﬀerent combinations of the number K = N 2 of subdomains and n of elements per edge (n2 elements per subdomain). if we number all subdomains rowwise) we set nk = n. nk }/ min{n . . 52]. n }. # Iter.2. it is not diﬃcult to prove that (12) holds provided that for all . Nonconforming decomposition. Number of conjugate gradient iterations needed for reducing the residual of a factor 10−5 . # Iter. # Iter. k such that ∂Ω ∩ ∂Ωk  > 0 it either holds that n = nk or max{n . we have that Li = 10 for all edges ei . In our test for ei = ∂Ωk ∩ ∂Ω . while on the remaining two subdomains (subdomains 13 and 14. using a standard technique. n14 } = n varies. . We consider then the case of a truly nonconforming decomposition. Number of conjugate gradient iterations needed for reducing the residual of a factor 10−5 . With the above criterion used to ﬁx Li . n = 5 n = 10 n = 20 n = 40 # Iter. . nk } ≥ C for some constant C > 1. where n varies in the set [10. This discretization does not in general satisfy the assumptions required for the uniform stability of the discrete problem. n = 5 n = 10 n = 20 n = 40 # Iter. both on a decomposition of Ω into 5 × 5 subdomains. # Iter. top). # Iter. 8 9 11 13 10 13 17 22 11 15 20 26 13 16 22 30 13 17 23 32 13 17 24 33 14 18 25 33 14 18 25 33 K = N2 16 64 144 256 400 576 784 1024 4. where Li = max{n13 .
for 12 out of 40 edges of Σ. For both conﬁgurations we tested both coarse preconditioners.6 0.5 0.4 0.2 0.6 0.7 0.3 0. we have that Li = n.1 0 0 0.8 0.3 0. In the second conﬁguration (see Figure 1.1 0 0 0.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 1 677 0. Here.4 0.8 0.5 0.3 0. this time except that on a threebythree block of subdomains where nk = n.2 0. The results are provided in Table 3.9 0.6 0.7 0.5 0.4 0.1 0.9 0. .7 0.9 1 (1) 1 0.8 0. with n varying again between 10 and 52.1 0.9 1 (2) Figure 1.6 0.5 0.2 0.3 0.2 0. The two conﬁgurations on which the preconditioner is tested for the plain formulation (8) in the nonconforming case. bottom) for all subdomains we again set nk = 10.4 0.8 0.7 0.
We now consider the stabilized formulation (19). More precisely. ·] 1 . We can then identify ∂Ωk with the circle T of unitary length.678 SILVIA BERTOLUZZA Table 3. Since the H 1/2 (∂Ωk ) seminorm is invariant under changes of scale. though the theoretical estimate on the condition number only depends on maxk Hk /hk . the bilinear forms [·. n = 10 12 8 20 14 n = 17 23 16 n = 24 24 18 n = 31 25 19 n = 38 26 20 n = 45 27 20 n = 52 (b) Conﬁguration 2. which takes the same value for both conﬁgurations. # It. The number of elements per edge in the remaining part of the skeleton is set to 10. to the top and to the bottom decompositions illustrated in Figure 1). n = 10 12 8 18 11 n = 17 18 13 n = 24 18 13 n = 31 18 14 n = 38 19 14 n = 45 19 14 n = 52 (a) Conﬁguration 1. The sequence {Vj }j≥0 forms a socalled multiresolution analysis of L2 (T ) and it is well known (see for example [15]) that there exist several wavelet bases associated with such a multiresolution analysis. . we can describe the stabilizing form for subdomains scaled in such a way that ∂Ωk  = 1 (that is. Both coarse preconditioners (67) (BPS) and (68) (Laplace) are tested for diﬀerent values of the number n of elements per edge in the reﬁned part of the skeleton (one edge for Conﬁguration 1 and twelve edges for Conﬁguration 2). setting +∞ ϑm = n=−∞ 2m/2 ϑ(2m (x − n) − ). Laplace BPS # It. in which a relevant portion of the skeleton is reﬁned. Φh ∂Ωk ⊂ Vjk +2 where for j > 0. Stabilized formulation. so that for all k. 4. in the ﬁrst case in which the overall discretization of the skeleton is sensibly coarser— the mesh being ﬁne only in a small portion (in our case one) of the edges composing Σ—the inﬂuence of reﬁning the mesh is weaker than in the second case. # It. Following the proposal of [7]. Laplace BPS # It.2.k are designed by means of a wavelet 2 decomposition. respectively. we restrict the number Li of elements of the ith edge ei to be a power of two. Vj denotes the space of P1 ﬁnite elements on the uniform grid Gj of T with mesh size 1/2j . Li = 2ji for some ji ≥ 1. Number of conjugate gradient iterations needed for reducing the residual of a factor 10−5 for Conﬁgurations 1 and 2 (corresponding. More precisely there exist several P1 compactly supported functions ϑ ∈ C 0 (R) deﬁned on the uniform grid of mesh size 1 and integer nodes. As one can easily see. such that. H = 1/4).
vh − ψh ] 1 . It is possible to prove ([7. If we denote by Pk : L2 (T ) → Vjk +2 the L2 (T ) orthogonal projection. In each case we randomly assigned the values of nk in such a way that for ∼ 1/3 of the subdomains nk = 5. Note that the coeﬃcients ζm and ξm are not the classical wavelet coeﬃcients of the functions ζ and ξ. and for the remaining subdomains nk = 15. ·] 1 . k With this deﬁnition. As far as ϕh and ψh are concerned.k essentially reduces h 2 k k to ﬁrst computing the nodal values of Pk (uh ). for ζ. 8 × 8. we tested the preconditioner for diﬀerent values of n and of the stabilization parameter γ. respectively. we need to compute their h 2 L projection on Vjk +2 . For approximating the interface function ϕ. the choice of jk implies that ϕh . Naturally. provided f ∈ Vjk +2 .THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 679 all functions ηj ∈ Vj can be written as j−1 2m ηj = η0 + m=0 =1 ηm ϑm . ξ ∈ 2 ˇ ˇ L (T ) we can deﬁne corresponding wavelet coeﬃcients ζm and ξm such that jk 2m jk 2m Pk (ζ) = ζ0 + m=0 =1 ˇ ζm ϑm . For further details on the implementation of the stabilized formulation we refer to [8]. We tested the preconditioner (with both form (67) and (68) of the coarse bilinear form) on four diﬀerent splittings of Ω into. The ˇ ˜ ˇm and ξm are rather the scalar products of the dual functions ϑm coeﬃcients ζ with the projections of ζ and ξ onto Vjk +2 . Observing that the mass matrix in Vjk +2 has a circulant structure with bandwidth equal to three. For each conﬁguration. These would in fact be deﬁned as the scalar ˜ product of. and such that ηj 2 1/2 (∂Ωk ) H 2m ηm 2 . j−1 2m η0 constant. for about another third nk = 10.l can be retrieved by a fast (O(2j )) wavelet transform (FWT). ζ and ξ with suitable dual functions ϑm ∈ L2 (T ). Pk (vh ). 14]) that the bilinear forms thus deﬁned satisfy assumptions (A4) and (18). respectively. such coeﬃcients coincide with the classical ones. we used a uniform discretization of the interface (Li = n for all i). ψh ∈ Vjk +2 and then their nodal values are simply computed k by interpolation. respectively. 4 × 4. this can also be done in O(2jk ) operations by a modiﬁcation of the Croute reduction algorithm for solving linear systems with tridiagonal matrices.1. ˇ ˇ Remark 4. The four conﬁgurations considered are shown in Figure 2. and then applying a FWT. Then we deﬁne the bilinear form [·. Pk (ϕh ) and Pk (ψh ). the computation of [uk − ϕh .k as 2 jk 2m [ζ. 12 × 12 and 16 × 16 subdomains. the coeﬃcients (ηm )m. As far as uk and vh are concerned.k = 2 m=0 =1 ˇ ˇ 2m ζm ξm . respectively. ξ] 1 . m=0 =1 Given the values of ηj at the nodes of the uniform grid. Pk (ξ) = ξ0 + m=0 =1 ˇ ξm ϑm . The function ϑ appearing in the deﬁnition of the stabilizing bilinear form is chosen .
9 0.6 0.4 0.1 0.8 0.9 1 (a) 1 0.5 0.2 0. The four conﬁgurations on which the preconditioner is tested for the stabilized formulation (10).3 0.2 0.6 0.6 0.6 0.5 0. conﬁrming the need for using some kind of stabilization.7 0.9 1 (b) Figure 2.4 0.2 0.3 0.8 0.3 0.5 0.9 0.7 0.1 0 0 0. Note (see Table 4) that for γ = 0 (no stabilization!) if n is big.1 0.8 0. .2 0.7 0.5 0. the CG algorithm does not converge within the maximum number of iterations (which was set equals to 100).4 0.3 0.8 0.1 0 0 0. (Continues) to be the piecewise linear wavelet corresponding to the 2.2 Bspline biorthogonal setting ([15]).680 1 SILVIA BERTOLUZZA 0.4 0.7 0.
If γ is too small (see Table 5).1 0 0 0. the value of the parameter γ has an inﬂuence on ˇ the conditioning of the matrix S −1 S.5 0.2 0. the action of the stabilization term is not suﬃcient to compensate for the lack of validity of the infsup condition.6 0.5 0.3 0.7 0.1 0.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 1 681 0.1 0 0 0.5 0.8 0.9 1 (d) Figure 2. (continued) As one can see from the results.7 0.4 0.2 0.4 0.6 0.2 0. and therefore the system behaves almost like the one obtained using the plain formulation.9 0.4 0.9 0.3 0.8 0.8 0. the situation improves .7 0.3 0.7 0.6 0.5 0.6 0.1 0.3 0.9 1 (c) 1 0.8 0. By increasing γ (see Tables 6 and 7).4 0.2 0.
the action of the stabilization term is not suﬃcient to compensate for the lack of validity of the infsup condition.05 in the case of 16 × 16 subdomains with n = 32 (which corresponds to 15360 degrees of freedom for ϕh ) the residual is reduced of a factor 10−5 in about 35 iterations. 4 14 17 18 19 47 57 60 69 8 — — — — 16 — — — — 32 Laplace (a) n # It.682 SILVIA BERTOLUZZA dramatically and one can see that. Table 4. and therefore the system behaves like the one obtained using the plain formulation. # It. # It. As one can see for n ≥ 16. for instance. # It. This is due to the fact that. respectively) ampliﬁed by the factor γ.00125. conﬁrming the instability of the plain formulation for the discretization considered. Number of conjugate gradient iterations needed to reduce the residual of a factor 10−5 on the four conﬁgurations depicted in Figure 2 with the stabilized formulation (19) for γ = . the performances deteriorate. 13 14 15 52 54 64 — — — — — — BPS Table 5. # It. # It. for γ = . Both coarse preconditioners are tested for diﬀerent values of the number n of elements per edge. # It. (a) (b) (c) (d) n # It. since the stabilization parameter is very small. # It. # It. 4 11 40 8 — 16 — 32 (b) (c) (d) # It. the choice of the right balance between the original bilinear form and the stabilizing term can become important from the point of view of the spectral properties of the operator. If such a ratio (which is always greater than or equal to one) in not close enough to 1. when γ is further increased (see Table 8). However. 13 14 15 51 57 77 92 — — — — — BPS . which is only slightly worse than the result obtained for the analogous values in the conforming case. (a) (b) (c) (d) n # It. This (as well as the diﬀerence with respect to the conforming case) is probably due to the fact that the constants in the estimates (26)—which have an inﬂuence on the bound on the condition number—depend on the ratio B/b (B and b being the constants in (18) and in assumption (A3)(c ). the CG method does not converge within the maximum number of iteration (= 100). which are essential for preconditioning. # It. Both coarse preconditioners are tested for diﬀerent values of the number n of elements per edge. Number of conjugate gradient iterations needed to reduce the residual of a factor 10−5 on the four conﬁgurations depicted in Figure 2 with the plain formulation (10). # It. 4 11 39 8 69 16 85 32 (b) (c) (d) # It. One can clearly see that the method converges very badly or does not converge. 4 14 17 18 19 52 60 64 83 8 86 — — — 16 96 — — — 32 Laplace (a) n # It.
# It. # It. # It. Number of conjugate gradient iterations needed to reduce the residual of a factor 10−5 on the four conﬁgurations depicted in Figure 2 with the stabilized formulation (19) for γ = 5. (a) (b) (c) (d) n # It. # It. # It. 4 12 23 8 28 16 34 32 (b) (c) (d) # It. # It. Both coarse preconditioners are tested for diﬀerent values of the number n of elements per edge. # It. 13 14 15 17 19 22 23 26 29 27 30 35 BPS Table 8. # It. 4 14 17 18 19 22 26 27 29 8 16 28 34 36 38 34 42 44 45 32 Laplace (a) n # It. # It. Number of conjugate gradient iterations needed to reduce the residual of a factor 10−5 on the four conﬁgurations depicted in Figure 2 with the stabilized formulation (19) for γ = 1. # It. 4 11 15 8 16 16 19 32 (b) (c) (d) # It. Both coarse preconditioners are tested for diﬀerent values of the number n of elements per edge. Such proof will easily reveal the dependence of the . 4 11 18 8 16 23 30 32 (b) (c) (d) # It. Both coarse preconditioners are tested for diﬀerent values of the number n of elements per edge. 4 14 17 18 19 19 22 23 25 8 22 28 29 31 16 24 31 32 34 32 Laplace (a) n # It. 13 14 15 21 24 27 31 35 38 41 46 51 BPS Table 7. 13 14 15 24 24 27 35 35 37 40 39 41 BPS Appendix A: Some Sobolev space injection bounds The aim of this section is to recall some known bounds for which we provide a new proof based on the use of equivalent norms expressed through wavelet coeﬃcients (see for instance [15]).THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 683 Table 6. # It. Number of conjugate gradient iterations needed to reduce the residual of a factor 10−5 on the four conﬁgurations depicted in Figure 2 with the stabilized formulation (19) for γ = 20. # It. # It. # It. 4 14 17 18 19 25 26 27 30 8 31 37 38 41 16 35 41 43 48 32 Laplace (a) n # It. # It. (a) (b) (c) (d) n # It. (a) (b) (c) (d) n # It.
. . ρ0 (x) ≤ C. ρk ρk + j=j0 k=1 ∞ 2j u. s = 1/2. such bases can be constructed in such a way that all u ∈ L2 (0. ρ0 2 + k j=j0 k=1 k=1 with constants independent of s. . 2j − N . ρ0 2 + k j=j0 k=1 ∞ 2j 2j0 .k N. In particular. k = 1. j ≥ j0 . .684 SILVIA BERTOLUZZA constants in the bounds on the smoothness of the spaces considered and the way these constants explode as the smoothness index approaches the limit of validity of the bounds.k ϑj. k k = = 1.k 1/2 2j  u.k 2 1/2 22js  u. 1) can be written both ways as 2j0 ∞ 2j u = k=1 2j0 u. ρ0 k ρ0 k + j=j0 k=1 u. . . . ϑj. Such bases are constructed in such a way that the basis functions satisfy the following properties: W.k has zero mean value. 2j }. ϑ0 2 j. satisfying N ≤ 2j0 −2 . k=1 (71) u H00 (0. k B 0 = {ρ0 . W. 1). k=1 (70) u s H0 (0.k j. . ρk 2 + j=j0 k=1 ∞ 2j 2j0 22js  u.k have a certain number of vanishing moments.1. . without and with homogeneous boundary conditions. (k + N )2−j ] ∩ ]0. ∀k = j. W. 1[ . W. . . ϑj. . .1) 1/2  u.k 2j/2 . . k = 1. 1. in other words the two bases diﬀer only for those elements that take into account boundary conditions.k . ϑ0 ϑ0 j. . [(k − N )2−j . The basis functions are well localized and the supports of the basis functions are properly scaled with respect to j: for some N ﬁxed.k (x) 2j/2 . in particular ϑj.1) 2 0 ∞ 2  u.k supp ϑ0 j. respectively.k .k = ϑ0 . (k + N )2−j ] ∩ ]0. ϑ0 (x) j. 2j0 } ∪ {ϑ0 . tuned up to provide norm equivalences for Sobolev spaces. .k and that the following norm equivalences hold for all s with 0 ≤ s ≤ 1: j 1/2 j (69) u H s (0. The basis functions verify (72) ρk (x) ≤ C. 2j }. .k . k ϑj. For all j ≥ j0 and k = 1. .3. . 1[ . .k ⊂ ⊂ [(k − N )2−j . .2.1)  u. . 2j the functions ϑj. u = k=1 u.k for L2 (0. we have supp ϑj. j.4. . . 2j0 } ∪ {ϑj. The interior functions of the two bases coincide: ϑj. The key ingredient of the following arguments is the possibility of constructing two orthonormal bases B = {ρk . . j ≥ j0 . ϑ0 2 j. .
we have (with u = 0 u ds) ˇ 2j0 ∞ 2j sup u(x) x sup  x k=1 u. 1) ⊂ L∞ (0.k .k (x) max  u. The following bound holds for all u ∈ H s (0.1) 2−j/2 . ϑj.1. we have N 2 N (76) k=1 ak k=1 ak 2 . Using (72) and (69). The ﬁrst bound that we consider regards the injection H s (0.1) u L2 (0.2 on the size of the support. 1) with s ∈ ]1/2. As far as the dependence on s of the constant in the bound is concerned. Lemma A. ϑj.k (x) = 0 1 for at most 2N values of k). ϑj.1.1) + We conclude by applying Poincar´ inequality which yields e u − uH s (0. as well as properties W. s − 1/2 Proof.1 and W.k L1 (0.1) ∞ + 2j 1 uH s (0. 1]: u 2j0 L∞(0. we have the following lemma. Let u = k=1 u. which is known to hold for all s > 1/2 [17].THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 685 Remark A. ρk ρk (x) + j=j0 k=1 u. The normalization estimate (72).k L1 (0. 1).k ϑj. ϑ0 j.k ϑj.1) = uH s (0. since N is a ﬁxed constant.k 2 22sj ˇ j≥j0 k=1 L2(0. ˇ u − u ˇ H s (0. + j≥j0 4(1/2−s)j 1 s − 1/2 u−u ˇ 2j 1/2  u − u. ρk ρk + j=j0 k=1 u. ϑj. trivially yields the following bound on the L1 norm of the basis functions (73) ϑj. We will therefore consider them as an O(1) as far as the estimates that we are going to prove are concerned.1) .1) . In the following we will make use of a certain number of algebraic inequalities: we start by recalling the following bound on the sum of a geometric series (74) j≥1 2−aj 1 .1) 1/2 H s (0. ϑj.1) 2−j/2 .k 2sj 2(1/2−s)j ˇ j≥j0 k ≤ u u L2(0.2 (which guarantees that for x and j ﬁxed. together with property W. Note that the constants N and j0 (and therefore 2j0 ) are ﬁxed once for all and depend only on the choice of the wavelet basis. ρk  + k max  u − u.1) . a as well as the wellknown boundedness property of the discrete convolution operator 2 (75) j  j aj 2 −j bj  j aj  j bj 2 . Moreover.1) .
n) with n = N.k = j≥j0 k=N 2j ηj. they do not see the boundary conditions. (77) η H0 (0. This is known to hold for all s ∈ [0.1) 1/2 − s Proof. 1) be expanded in terms of the basis B: 2j0 2j η= k=1 ζk ρk + j≥j0 k=1 2j0 ηj. is a ﬁxed constant).. η N −1 center = j≥j0 k=N ηj. ϑ0 j. The dependence on s of the constant in the bound is the object of the following lemma. We start by decomposing η as the sum of four contributions: η = η0 + η lef t + η center + η right with 2j0 2j −N 2j −N (78) η0 = k=1 ζk ρk .1) + η center s H0 (0. center .k ϑj.1) .n = 0 is bounded independently of j. i. .1) + η lef t s H0 (0. Let η ∈ H s (0.1) ≤ η0 s H0 (0.n j. We start by observing that for all (j. ϑ0  ≤ ρk L∞ (0. Lemma A.1) . .k ϑj.k . we will use the equivalent norm (70).1) k=1 ζk 2 2 L + j≥j0 2 j k=1 ηj. n) we have  ρk . η right = j≥j0 k=2j −N +1 ηj.k (79) η lef t = j≥j0 k=1 ηj. we then easily see that j 2 2 0 2j0 1 1 2 s ζk  ζk 2 .1) 2j0 /2 2−j/2 .n = 0. j. 0 ≤ s < 1/2: 1 s η H s (0. 1) norm of the four contributions.1) + η right s H0 (0. j. which.1) 2j0 ρk 2 2 + Hs L 1 − 2s j≥j0 the and the 1 norms are equivalent (the constant Using the fact that on R in the equivalence depending on j0 .n = ρk .686 SILVIA BERTOLUZZA The second issue that we consider is the equivalence between the spaces H s (0. . s In order to bound the H0 (0. η0 H0 1 − 2s 1 − 2s 2 k=1 k=1 2j0 . 2j − N we have ρk . we easily see that 1 .2. The following bound holds for all η ∈ H s . This term contains the contribution of all functions We now consider η which are interior. while for the remaining couples (j.k . ϑ0 j. 22js 2−j ρk 2 0 (0.1) ϑ0 L1 (0.k ϑj. so that 2j η 2 H s (0.k and by triangular inequality we have η s H0 (0. 1). 1/2[ [17].e.k ϑj.n using (70) and observing that for all j’s the number of indexes n for which ρk . . we recall. More precisely.k 2 . 1) s and H0 (0. ϑj.k ϑ0 .
1) ≤ η H s (0.k .1) L∞ (0.k  ≤ j j=j0 k=1 ηj. ϑ0 . ϑ0 . k we can write  ϑj. η center . ϑ0 . k .k  j Then we have ∞ N −1 ∞ L1 (0.k  ≤ ϑj.k = ϑj. ρ0 k  + 2 j =j0 2 2sj k =1  η lef t .k . ϑ0 . k such that ϑj .k ∩ supp ϑ0 .1) 2(j−j )/2 ϑ0 .k . ϑ0 .k 2 j ∞ (N − 1) j =j0 N −1 k=1 ∞ N −1 22j s 2 ∞ 2−j−j /2 N −1 k=1 ηj. ϑj . j =j0 j=j0 ∞ 2−(1/2−s)j 22js j=j0 k=1 j=j0 . ϑ0 .k j L∞ (0.k it holds that η center . ϑ0 .k  j=j0 2−(1/2−s)j−j 2  2js ηj.k 2 . while for all other j . s Let us now bound the H0 (0.k 2 . N −1  η lef t . ρ0 2 k k =1 η lef t 2 L2 (0.k . it holds that ϑj. thanks to (72) and (73).k = ϑ0 . We observe j that. ϑ0 .k j as well as  ϑj. Considering that for all N ≤ k ≤ 2j − N .k = ∅. we can write 2 ∞ N −1 ∞ 22j j =j0 s k =1 ∞  η lef t . using inequalities (74).1) . k ≤ N − 1.1) ϑ0 . 2−j−j /2 .k  ·  ϑj.1) Hs k =1 η lef t .k = 0 and this yields j 2j −N = η center 2 s H0 (0.k .k . we can j write 2j0 ∞ N −1 η lef t 2 0 (0.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 687 we observe that for all j . k. In order to bound the second sum. ϑ0 . (75) and (76). j It is not diﬃcult to realize that the ﬁrst sum on the righthand side can be bound as follows: 2j0  η lef t . j . for all j. let us now bound  η lef t .k  ηj.k .1) 2(j −j)/2 .k  ≤ ϑj.k . 1) norm of η lef t .k j which yield  ϑj.k 2 . ϑ0 .k j L1 (0.k  j j=j0 2−j−j /2 ηj. k=1 Since the righthand side of the above estimate does not depend on k . which is the contribution of those basis functions which see the left boundary. using (70).1) j≥j0 22sj k=N ηj. ϑ0 .k j j ηj .k .k = 0 and that for j k ≤ N − 1 and k > 2j − N we have supp ϑj.
MR 2001h:65142 [6] S..1) . Comp. 69:463–480. 1994. In Proc. Wavelet stabilization of the Lagrange multiplier method. Brezzi. Stable discretization of convectiondiﬀusion problems via computable negative order inner products. Achdou. H0 (0. Computational scales of Sobolev norms with application to preconditioning.1) ∞ 1 22js (1/2 − s)2 j=0 ∞ 1 22js (1/2 − s)2 j=0 lef t N ηj. Numer. 1/2[. E. that 1 α s . η − α H 1/2 (0. 20:533–559. Bramble. right s (0. Bertoluzza. Bramble. 1999. SINUM. Brezis & J. 86:1– 28. 1986.1. I. F. Analysis of a stabilized three ﬁelds domain decomposition method. Pasciak. The construction of preconditioners for elliptic problems by substructuring. we can prove η right 2 s H0 (0. Ser. of Comp.1) . Wavelet stabilization and preconditioning for domain decomposition. Marini. Corollary A. Bertoluzza. Kunoth. Russo. A new nonconforming approach to domain decomposition: The mortar element method.1) 2j k=2j −N +1 s s s η η center H0 (0. 38:1034–1055.1) ≤ η−α s H0 (0.T. 2000. Mixed and Hybrid Finite Element Methods. Bertoluzza. MR 2000k:65088 [11] F. [5] S. D. Lions. L. Schatz. SIAM J. Franca. 2000.. Brezzi and M. 1992.1) ≤ 1/2 − s 1/2 − s Proof. Fortin. and O. MR 94g:65119 [3] C. MR 87m:65174 [10] J. and A. Tabacco. 2000. In view of bound (77) the only thing that needs to be proven is a bound on s α H0 (0. MR 92d:65187 [12] F. In Future Tendencies in Computer Science. Springer.N.1) . [9] J.M.A. which can be done by direct computation. Anal. Numer. MR 2001h:65156 [8] S.1) + α s H0 (0. IX Domain Decomposition Methods Conference. Wavelet stabilization of the three ﬁelds domain decomposition method: Implementation and numerical tests. and A.. 47(175):103–134. For all η ∈ H 1/2 (0.S. Marini.k 2 . H. Brezzi. Vassilevski.1) and Adding the bounds for η0 H0 (0.1) . J. 1) and for all α ∈ R it holds. Nonlinear Partial Diﬀerential Equations and their Applications.E. Patera. .. In preparation. Manzini. Bertoluzza and A. 2000. Math. editors. Maday. Stabilization techniques for domain decomposition methods with nonmatching grids.A. MR 2001m:65163 [7] S. pages 13–51. Pasciak. In H. ηj. yielding the thesis. Numer. and D. J.. 1/2[ it holds that 1 ≤ 1/(1/2 − s).R. Trivially it holds that η s H0 (0. MR 95a:65201 [4] S. Math. Canuto.L. MR 99m:65233 [2] C.1) + η H0 (0. Widlund. 299.N. and A. References [1] Y. Baiocchi. I. Anal. Bertoluzza and G.. C. in view of η H0 the norm equivalence (69) we obtain the thesis. 36:551–580.k 2 . volume XI of e Notes Math. Maday. Math. Bernardi.688 SILVIA BERTOLUZZA which yields η lef t 2 s H0 (0. H. Y. Coll`ge de France Seminar. Jour. Substructuring preconditioners for the mortar method in dimension two. and P. Control and Applied Mathematics.1) . and A. 1991. Y.C. 2000.H. since for s in [0. k=1 Analogously. Stabilization of Galerkin methods and application to domain decomposition. for all s ∈]0. Technical Report 1175.
Italy Email address: silvia. 1992. volume 157 of American Mathematical Society. Springer. and O. Math. v. MR 50:2670. MR 2002b:65159 [14] F. Numerical analysis of wavelet methods. MR 93j:65065 [17] J. Lions and E. Math.L.cnr. Periaux. Domain Decomposition Methods in Science and Engineering. MR 50:2671 [18] P. MR 95a:65202 [15] A. Multilevel preconditioning. Contemporary Mathematics. In P.. Elsevier Science Publishers.. Handbook in Numerical Analysis.L. Y. Widlund. Numer. 2001.it URL: http://www. 64:1367–1396. MR 95m:65212 Istituto di Matematica Applicata e Tecnologie Informatiche del Consiglio Nazionale delle Ricerche.A. volume VII. 27100 Pavia.bertoluzza@imati. Marini. Comp. Domain decomposition with nonmatching grids: Augmented Lagrangian approach. MR 2002c:65252 [16] W. Magenes. of Comp. 1995. In A. Marini.G. Cohen. A threeﬁeld domain decomposition method. 1972.it/~aivlis . pages 27–34. J. North Holland. editors. Nonhomogeneous Boundary Value Problems and Applications.THE THREE FIELDS DOMAIN DECOMPOSITION METHOD 689 [13] F. editors. Ciarlet and J. 63:315–344. Le Tallec and T. Kunoth. Brezzi and D. Math. Brezzi and D. Error estimates for the threeﬁeld formulation with bubble stabilization.. Kuznetsov. Dahmen and A. Lions. 70:911–934. Sassi. 2000.cnr. Ferrata 1. Quarteroni. 1994.imati.B.