MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
JOHNNY GUZM ´
AN
1
, AND MANUEL S ´
ANCHEZ-URIBE
2
,
Division of Applied Mathematics, Brown University, Providence, RI 02912
Abstract. We prove stability in W
1,∞
(Ω) and L
∞
(Ω) for the velocity and pressure approx-
imations, respectively, using the lowest-order Taylor-Hood finite element spaces to solve the
three dimensional Stokes problem. The domain Ω is assumed to be a convex polyhedra.
Keywords: maximum norm, finite element, optimal error estimates, Stokes.
1. Introduction
Consider the Stokes problem on a convex polyhedral domain Ω ⊂ R
3
−∆u +
p = f
in Ω
(1.1a)
· u = 0
in Ω
(1.1b)
u = 0
on ∂Ω.
(1.1c)
Here u is the velocity and p is the pressure. The aim of this paper is to prove W
1,∞
stability
of the lowest order Taylor-Hood (see for example [1]) approximation in three dimensions. More
specifically, we prove the bound
u
h L
∞
(Ω)
+ p
h L
∞
(Ω)
≤ C(
u
L
∞
(Ω)
+ p
L
∞
(Ω)
).
where u
h
∈ V
h
, p
h
∈ M
h
are the Taylor-Hood approximations.
In previous papers, W
1,∞
[18, 5] stability was proven for many inf-sup stable pair of spaces,
but one major exception was the lowest order Taylor-Hood pair in three dimensions. The reason
for this is that in both papers it was assumed that there exists a Fortin projection Π
h
(i.e. it
commutes with the divergence operator) to the finite element velocity space that is quasi-local,
i.e. Π
h
∈ L(H
1
0
(Ω)
3
, V
h
) satisfies the following properties
(q
h
,
· (Π
h
(w) − w))
Ω
=
0,
∀w ∈ H
1
0
(Ω)
3
,
∀q
h
∈ M
h
.
|Π
h
(v) − v|
W
m,q
(T )
≤ Ch
s−m+3(
1
q
−
1
p
)
T
|v|
W
s,p
(∆T )
,
∀T ∈ T
h
,
∀v ∈ W
s,p
(Ω)
3
for all real numbers 1 ≤ sk + 1, 1 ≤ p, q ≤ ∞, and integer m = 0 or 1 such that W
s,p
(Ω) ⊂
W
m,q
(Ω). The constant C is independent of h and T , and ∆T is a suitable macro-element
containing T . Although such a Fortin projection exists for many inf-sup pair of spaces [16],
existence of a quasi-local Fortin projection for the lowest-order Taylor-Hood element in three
E-mail address:
1
johnny guzman@brown.edu ,
2
manuel sanchez uribe@brown.edu.
Mathematics Subject Classification: 65N30, 65N15.
1
2
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
dimensions is still open. In this paper, we instead use a quasi-local inf-sup condition which holds
for the Taylor-Hood element and avoid the use of a Fortin projection.
The local inf-sup condition has been used before by Arnold and Liu [20] to prove local energy
estimates for Stokes problem. The local energy results in Arnold and Liu were proven only for
interior domains. Chen [19] assuming local energy results (both interior domains and also sub-
domains touching the boundary ∂Ω) proved W
1,∞
stability for finite element approximations to
the Stokes problem for domains Ω that have a smooth boundary.
The techniques used by Chen [19] cannot easily be extended to our setting where we assume
that Ω is a convex polyhedral domain. First, higher elliptic regularity results were used by Chen,
which do not hold in our setting. Second, we cannot use directly the local energy estimates that
Chen assumed because this will require us to estimate the pressure error in a negative order
norm which we do not know how to estimate with the given regularity of the problem. Instead
we prove a local energy estimate that does not contain the error of the pressure which is very
similar to the estimates obtained in [5] ( see also [6]). Of course, the estimates derived in [5]
assumed the existence of a quasi-local Fortin projection.
There will be many similarities between the proofs in this paper and the proofs in article [5].
In order to make our paper self contained we provide many details. However, we will compare
the individual results below to corresponding results in [5]. We prove max-norm estimates for
Stokes elements which satisfy assumptions A1-A6 below. As a corollary we show that the lowest-
order Taylor-Hood element in three dimensions satisfies these assumptions. For simplicity we
only consider Stokes elements that use continuous pressures.
2. W
1,∞
stability result
In this section we state our main result in Theorem 1. The finite element approximation
problems, and the assumptions of our result are presented bellow.
2.1. Preliminaries and Assumptions. For the finite element approximation of the problem,
let T
h
, 0 < h < 1, be a sequence of partitions of Ω, Ω = ∪
T ∈T
h
T , with the elements T mutually
disjoint. Let h
T
denote the diameter of the element T and h := max
T
h
T
. The partitions
are face-to-face so that simplices meet only in full lower-dimensional faces or not at all. The
family of triangulation are shape regular and quasi-uniform. The finite element velocity space
is denoted by V
h
⊂ [H
1
0
(Ω)]
3
and the pressure space is denoted by M
h
⊂ L
2
(Ω). We assume
that V
h
contains the space of piecewise polynomials of degree k (k ≥ 2) and is contained is the
space of piecewise polynomials of degree l. We assume that M
h
contains the space of continuous
piecewise polynomial of degree k − 1.
The finite element approximation (u
h
, p
h
) ∈ V
h
× M
h
solves
( u
h
,
v) − (p
h
,
· v) = (f , v)
∀v ∈ V
h
(2.1a)
(q,
· u
h
) =
0
∀q ∈ M
h
(2.1b)
where (·, ·) denotes the usual L
2
(Ω) inner product. The approximation to the pressure p
h
is unique up to a constant.
We can for example require p, p
h
∈ L
2
0
(Ω), i.e.,
Ω
p(x)dx =
Ω
p
h
(x)dx = 0. Instead, we will require
(2.2)
Ω
p(x)φ(x)dx =
Ω
p
h
(x)φ(x)dx = 0,
where φ(x) is an infinitely differentiable function on Ω that vanishes in a neighborhood of the
edges and satisfies
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
3
(2.3)
Ω
φ(x)dx = 1.
Without loss of generality, we fix φ as above and assume p, p
h
satisfy (2.2). In other words,
we let p and p
h
belong to the space L
2
φ
.
We assume the existence of two projection operators P : [H
1
0
(Ω)]
3
→ V
h
and R : L
2
(Ω) →
M
h
with following properties
A1 (Stability). There exists constants C
1
, C
2
independent of h such that
(2.4a)
Pv
H
1
(Ω)
≤ C
1
v
H
1
(Ω)
,
∀v ∈ [H
1
0
(Ω)]
3
.
(2.4b)
Rq
L
2
(Ω)
≤ C
2
q
L
2
(Ω)
,
∀q ∈ L
2
(Ω).
A2 (Local Approximation) Let Q ⊂ Q
d
⊂ Ω with d ≥ κh, for some fixed κ sufficiently large
and Q
d
= {x ∈ Ω : dist(x, Ω) ≤ d}. For any v ∈ [H
l
(Q
d
)]
3
there exists C independent
of h and v such that
(2.5a)
v − Pv
L
2
(Q)
+ h v − Pv
H
1
(Q)
≤ Ch
l
v
H
l
(Q
d
)
for l = 1, 2.
For any v ∈ [C
1+σ
(Q
d
)]
3
there exists a constant C independent of h such that
(2.5b)
v − Pv
W
t
∞
(Q)
≤ Ch
1+σ−t
v
C
1+σ
(Q
d
)
for t = 0, 1,
where
v
C
1+σ
(Q)
= v
C
1
(Q)
+
sup
x,y∈Q
i∈{1,2,3}
|e
i
· ( v(x) −
v(y))|
|x − y|
σ
For any q ∈ H
1
(Q
d
) there exists a constant C independent of h and Q such that
(2.5c)
q − Rq
L
2
(Q)
≤ Ch q
H
1
(Q
d
)
.
For any q ∈ C
σ
(Q
d
) there exists a constant C independent of h such that
(2.5d)
q − Rq
L
∞
(Q)
≤ Ch
σ
q
C
σ
(Q
d
)
.
A3 (Superapproximation). Let ω ∈ C
∞
0
(Q
d
) be a smooth cut-off function such that ω ≡ 1
on Q and
(2.6a)
|D
s
ω| ≤ Cd
−s
,
s = 0, 1.
We assume that
(2.6b)
ω
2
v − P(ω
2
v)
L
2
(Q)
≤ Chd
−1
v
L
2
(Q
d
)
,
∀v ∈ V
h
(2.6c)
(ω
2
v − P(ω
2
v))
L
2
(Q)
≤ Cd
−1
v
L
2
(Q
d
)
,
∀v ∈ V
h
and
4
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
(2.6d)
(ω
2
q − R(ω
2
q))
L
2
(Q)
≤ Chd
−1
q
L
2
(Q
d
)
,
∀q ∈ M
h
.
A4 (Inverse inequality). There exists a constant C independent of h such that
(2.7a)
v
H
1
(Q)
≤ Ch
−1
v
L
2
(Q
d
)
A5 (Local inf-sup condition). There exists β > 0 and
≥ 1 such that for every set B ⊂ Ω
there exist B
h
⊇ B, with dist(B, ∂B
h
\∂Ω) ≤ h, and β > 0 such that
(2.8)
sup
v∈V
h
\{0}
supp(v)⊂B
h
(q,
· v)
v
H
1
(B
h
)
≥ βh
q
L
2
(B)
,
∀q ∈ M
h
.
A6 (L
1
inf-sup condition). There exists a constant γ > 0 independent of h such that
(2.9)
sup
v∈V
h
\{0}
(q,
· v)
v
W
1
∞
(Ω)
≥ γh
q
L
1
(Ω)
,
∀q ∈ M
h
.
When B = Ω property A5 is the standard inf-sup condition for Stokes finite element spaces.
We now state the main result of the paper.
Theorem 1. Let (u, p) and (u
h
, p
h
) satisfy (1.1) and (2.1), respectively. Under the Assumptions
1-6, there exists a constant C independent of h such that
u
h L
∞
(Ω)
+ p
h L
∞
(Ω)
≤ C(
u
L
∞
(Ω)
+ p
L
∞
(Ω)
).
Of course, as a corollary we have
(u − u
h
)
L
∞
(Ω)
+ p − p
h L
∞
(Ω)
≤ C( sup
v∈V
h
(u − v)
L
∞
(Ω)
+ sup
q∈Q
h
p − q
L
∞
(Ω)
).
The proof of Theorem 1 is presented in section 4. In section 4.1 we state some Green’s function
estimates, established in [9, 7, 8, 11] which are used in section 4.2 to prove a key estimate for
the gradient of the finite element approximation of the Green’s function in the L
1
norm. Finally
in section 4.3 we prove the stability in L
∞
norm of the velocity and the pressure.
3. Local energy estimate
An essential ingredient of our proof is the local energy estimate that we derive in this section.
Consider (v, q) ∈ [H
1
0
(Ω)]
3
× L
2
(Ω) and (v
h
, q
h
) ∈ V × M
h
satisfying the following orthogonality
relation:
( (v − v
h
),
χ) − (q − q
h
,
· χ) = 0
∀χ ∈ V
h
(3.1a)
(w,
· (v − v
h
))
=
0
∀w ∈ M
h
(3.1b)
Theorem 2. Suppose (v, q) ∈ [H
1
0
(Ω)]
3
× L
2
(Ω) and (v
h
, q
h
) ∈ V × M
h
satisfy (3.1). Then,
there exists a constant C > 0 such that for every pair of sets A
1
⊂ A
2
⊂ Ω such that
dist(A
1
, ∂A
2
\∂Ω) ≥ d ≥ κh ( for some fixed large enough constant κ) and for any ε ∈ (0, 1),
the following bound holds:
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
5
(v − v
h
)
L
2
(A
1
)
≤ C ε
−1
(v − Pv)
L
2
(A
2
)
+ (εd)
−1
(v − Pv)
L
2
(A
2
)
+ q − Rq
L
2
(A
2
)
+ε
(v − v
h
)
L
2
(A
2
)
+
C
εd
(v − v
h
)
L
2
(A
2
)
The above result is similar to Theorem 2 in [5].
The main difference is that the term
ε
−1
(v − Pv)
L
2
(A
2
)
appears in our result.
Proof. We first prove the statement with the following assumption for the sets A
1
and A
2
.
A7 Redefine the sets as A
s
= B
sd/2
∩ Ω for s = 1, 2, where B
sd/2
is a ball of radius sd/2
centered at x
0
∈ ¯
Ω and assume that there exists a ball B ⊂ A
1
, such that diam(A
1
) ≤
d < ρ diam(B), where ρ is a fixed constant that only depends on Ω.
We will compete the proof for general sets by a covering argument.
Consider ω ∈ C
∞
0
(A
3/2
) the cut-off function defined in assumption A3, for Q = A
1
and
Q
d
= A
2
. Define e = v − v
h
, η = v − Pv, ξ = Pv − v
h
, e
q
= q − q
h
, η
q
= q − Rq and ξ
q
= Rq − q
h
then
(3.2)
e
L
2
(A
1
)
≤
ω e
L
2
(Ω)
= ( e,
(ω
2
e)) − ( e,
(ω
2
) ⊗ e)
Throughout this proof we will estimate the middle term of (3.2). We first obtain an estimate
for the second term on the right hand side of (3.2), by Cauchy-Schwartz (C-S.) inequality and
the property of ω (2.6a) we obtain
−( e,
(ω
2
) ⊗ e) ≤
C
d
ω e
L
2
(Ω)
e
L
2
(A
3/2
)
.
Applying the arithmetic-geometric mean (a-g.m.) inequality and (3.2), we get
(3.3)
1
2
ω e
L
2
(Ω)
≤ ( e,
(ω
2
e)) +
C
d
2
e
2
L
2
(A
3/2
)
.
Now for the first term on the right hand side of (3.3), we use e = η + ξ, obtaining
( e,
(ω
2
e))
=
( e,
(ω
2
ξ)) + ( e,
(ω
2
η))
≤ ( e,
(ω
2
ξ)) + C ω e
L
2
(Ω)
(
η
L
2
(A
3/2
)
+
1
d
η
L
2
(A
3/2
)
),
(3.4)
in the last line we have estimated the second term using (2.6a). The term ( e,
(ω
2
ξ)) is
more involved, we decompose it as follows
(3.5)
( e,
(ω
2
ξ)) = ( e,
P(ω
2
ξ)) + ( e,
(ω
2
ξ) − P(ω
2
ξ)) =: I
1
+ I
2
.
Summarizing, by (3.4), the a-g.m. inequality , the definition of I
1
and I
2
and (3.3) we have
(3.6)
1
4
ω e
L
2
(Ω)
≤ I
1
+ I
2
+ C
η
2
L
2
(A
3/2
)
+
C
d
2
η
2
L
2
(A
3/2
)
+
C
d
2
e
2
L
2
(A
3/2
)
.
We estimate I
2
applying C-S. inequality, the superapproximation assumption A3 (2.6b) and
the a-g.m. inequality for 0 < ε < 1, obtaining
6
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
I
2
≤
e
L
2
(A
3/2
)
(ω
2
ξ − P(ω
2
ξ))
L
2
(A
3/2
)
≤
e
L
2
(A
3/2
)
C
d
ξ
L
2
(A
2
)
=
ε
e
2
L
2
(A
3/2
)
+
C
εd
2
( η
2
L
2
(A
2
)
+ e
2
L
2
(A
2
)
),
To estimate I
1
we use (3.1a), then adding and subtracting Rq we break I
1
into three parts
I
1
=
−(e
q
,
· P(ω
2
ξ))
=
−(e
q
,
· (ω
2
ξ)) − (η
q
,
· (P(ω
2
ξ) − ω
2
ξ)) − (ξ
q
,
· (P(ω
2
ξ) − ω
2
ξ)) = I
3
+ I
4
+ I
5
Similar to the estimate for I
2
, we estimate I
4
I
4
≤
η
q L
2
(A
3/2
)
· (P(ω
2
ξ) − ω
2
ξ)
L
2
(A
3/2
)
≤
η
q L
2
(A
3/2
)
C
d
ξ
L
2
(A
2
)
=
η
q
2
L
2
(A
3/2
)
+
C
d
2
( η
2
L
2
(A
2
)
+ e
2
L
2
(A
2
)
),
Next we estimate I
5
. We use integration by parts (taking into account that M
h
is continuous),
C-S. inequality, superapproximation assumption A3
I
5
=
( ξ
q
, P(ω
2
ξ) − ω
2
ξ) ≤
ξ
q L
2
(A
3/2
)
P(ω
2
ξ) − ω
2
ξ
L
2
(A
3/2
)
≤
ξ
q L
2
(A
3/2
)
Ch
d
ξ
L
2
(A
2
)
Using the local inf-sup condition assumption A5 we know there exists A
Dostları ilə paylaş: |