17
γ
ξ
λ L
1
(Ω)
≤
sup
z∈V
h
(ξ
λ
,
· z)
z
W
1
∞
(Ω)
= sup
z∈V
h
(e
λ
− η
λ
,
· z)
z
W
1
∞
(Ω)
≤
η
λ L
1
(Ω)
+ sup
z∈V
h
(e
λ
,
· z)
z
W
1
∞
(Ω)
=
η
λ L
1
(Ω)
+ sup
z∈V
h
−( e
λ
, z)
z
W
1
∞
(Ω)
≤
η
λ L
1
(Ω)
+ sup
z∈V
h
( e
g
,
z)
z
W
1
∞
(Ω)
≤
η
λ L
1
(Ω)
+
e
g L
1
(Ω)
≤ C +
e
g L
1
(Ω)
.
Then
(4.21)
J
3
|
Ω\Ω
j
≤ C
h
d
j
σ
d
−1/2
j
(C +
e
g L
1
(Ω)
)
It follows from (4.20) and (4.21) that
(4.22)
J
3
≤ Ch
e
g L
2
( ˜
Ω
j
)
+ C
h
d
j
σ
d
−1/2
j
e
g L
1
(Ω)
+ Cd
−1/2
j
(hd
−1
j
+
h
d
j
σ
).
Therefore, estimate for J
1
+ J
4
, J
2
and J
3
, (4.18),(4.19) and (4.22), respectively, give
d
1/2
j
e
g L
2
(Ω
j
)
≤ Chd
−1
j
+ C
h
d
j
σ
+ C
h
d
j
σ
e
g L
1
(Ω)
+ Chd
1/2
j
(
e
g L
2
(Ω
j
)
+
e
g L
2
( ˜
Ω
j
)
).
To summarize,
M
j
≤ C
h
d
j
σ
(1 +
1
ε
) + C
h
d
j
ε
+ C(
hd
1/2
j
ε
+ εd
3/2
j
)
e
g L
2
(Ω
j
)
+
C
ε
h
d
j
σ
e
g L
1
(Ω)
Step 4 (Double kick-back argument). We sum over j in the last expression obtaining
J
j=0
M
j
≤
J
j=0
C
h
d
j
σ
(1 +
1
ε
) + C
h
d
j
ε
+
C
ε
h
d
j
σ
e
g L
1
(Ω)
+C(
h
d
J
ε
+ ε)
J
j=0
d
3/2
j
e
g L
2
(Ω
j
)
Observe that
J
j=0
h
d
j
σ
= h
σ
J
j=0
(2
j
)
σ
= h
σ
(2
σ
)
J +1
− 1
2
σ
− 1
≤
h
d
J
σ
2
σ
2
σ
− 1
≤ CK
−σ
in the last expression C depends on σ which is fixed. Then,
18
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
J
j=0
M
j
≤ C
(1 + ε
−1
)
K
σ
+ C
1
εK
+
C
εK
σ
e
g L
1
(Ω)
+ C(
1
εK
+ ε)
J
j=0
d
3/2
j
e
g L
2
(Ω
j
)
Observing that Ω
j
⊂ Ω
∗
∪
s∈S
Ω
s
, for some finite number S, we can bound the last term
in the right-hand side as follows
J
j=0
d
3/2
j
e
g L
2
(Ω
j
)
≤ C
J
j=0
M
j
+ C(Kh)
3/2
e
g L
2
(Ω
∗
)
≤ C
J
j=0
M
j
+ CK
3/2
.
Choosing K large enough and a sufficiently small ε we have
J
j=0
M
j
≤ C
K,ε
+
C
εK
σ
e
g L
1
(Ω)
This result allows us to conclude in (4.15) that
e
g L
1
(Ω)
≤ C
K,ε
+
C
K
σ
ε
e
g L
1
(Ω)
which, by means of a large enough choice of K, implies the desired result
e
g L
1
(Ω)
≤ C
K,ε
.
This completes the proof.
Proof. ( Proposition 4.1 )
(i) By H
2
-regularity and property of R we have
η
w L
2
(Ω)
+ η
ϕ L
2
(Ω)
≤ Ch( w
H
2
(Ω)
+
ϕ
L
2
(Ω)
) ≤ Ch
the last inequality is due to v
L
2
(Ω
j
)
≤ 1
(ii) We observe that by H¨
older inequality η
w L
∞
(Ω\Ω
j
)
≤ Ch
σ
w
C
1+σ
(Ω\Ω
j
)
Then, since Ω\Ω
j
is separated from Ω
j
by at least d
j
, for x, y ∈ Ω\Ω
j
, using (4.6a) and
(4.8) ,we have
|∂
x
w
k
(x) − ∂
y
w
k
(y)|
|x − y|
σ
≤
3
i=1
Ω
j
∂
x
G
k,i
(x, ξ) − ∂
y
G
k,i
(y, ξ)|
|x − y|
σ
|v(ξ)|dξ
≤ C max
ξ∈Ω
j
(|x − ξ| + |y − ξ|)
−2−σ
Ω
j
|v(ξ)|dξ
≤ Cd
−2−σ
j
d
3/2
j
v
L
2
(Ω
j
)
≤ Cd
−1/2−σ
j
,
for k = 1, 2, 3.
It follows that
η
w L
∞
(Ω\Ω
j
)
≤ C
h
d
j
σ
d
−1/2
j
.
Similarly, for x, y ∈ Ω\Ω
j
, using (4.6b) and (4.8) ,we have
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
19
|ϕ(x) − ϕ(y)|
|x − y|
σ
≤
3
i=1
Ω
j
∂
x
G
4,i
(x, ξ) − ∂
y
G
4,i
(y, ξ)|
|x − y|
σ
|v(ξ)|dξ
≤ C max
ξ∈Ω
j
(|x − ξ| + |y − ξ|)
−2−σ
Ω
j
|v(ξ)|dξ
≤ Cd
−2−σ
j
d
3/2
j
v
L
2
(Ω
j
)
≤ Cd
−1/2−σ
j
,
for k = 1, 2, 3.
Then, by A3 we have
η
ϕ L
∞
(Ω\Ω
j
)
≤ Ch
σ
ϕ
C
σ
(Ω\Ω
j
)
≤
h
d
j
σ
d
−1/2
j
.
(iii) Using (4.6b), (4.7) and dist(Ω
j
, T
z
) = O(d
j
) we have
λ(x)
=
3
k=1
T
z
G
4,k
(x, ξ)(∂
ξ
δ
h
(ξ))δ
i,k
dξ
=
−
T
z
∂
ξ
G
4,i
(x, ξ)δ
h
(ξ)dξ ≤ Cd
−3
j
δ
h L
1
(T
z
)
≤ Cd
−3
j
.
Thus, η
λ L
2
(Ω
j
)
≤ C λ
L
2
(Ω
j
)
≤ Cd
−3/2
j
.
(iv) Using the dyadic decomposition (4.14) and C-S. inequality, we have
η
λ L
1
(Ω)
≤ CK
3/2
h
3/2
η
λ L
2
(Ω
∗
)
+ C
J
j=0
d
3/2
j
η
λ L
2
(Ω
j
)
.
Approximation property of R A2, H
2
-regularity and (4.10b) imply that
h
3/2
η
λ L
2
(Ω
∗
)
≤ Ch
3/2+1
λ
L
2
(Ω)
≤ Ch
5/2
δ
h L
2
(T )
≤ C.
Finally, using (iii) we conclude that
η
λ L
1
(Ω)
≤ CK
3/2
+ C
J
j=0
h
d
j
σ
≤ C
K
.
4.3. Proof of Theorem 1. We start this section with the L
∞
estimate for the velocity. Con-
sider the problem (4.11) with a = 1 and b = 0. We will estimate |∂
x
j
(u)
i
(z)|, where 1 ≤ i, j ≤ 3
are arbitrary and arbitrary z ∈ ¯
Ω. We start the estimate using the definition of the delta
function, then we have
20
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
−∂(u
h
)
i
(z)
=
(u
h
, (∂
x
j
δ
h
)e
i
)
=
(u
h
, −∆g +
λ
=
( u
h
,
g) + (u
h
,
λ)
=
( u
h
,
g) + (u
h
,
λ
h
) + ( u
h
,
(g
h
− g))
=
( u
h
,
g
h
)
=
( u,
g
h
) + ( (p − p
h
), g
h
)
=
( u,
g
h
) + ( p, g
h
)
=
( u,
g
h
) + ( u,
g) + ( p, g
h
− g) + (u,
λ)
=
( u,
g
h
) + (u, −∆g +
λ) + (g − g
h
,
p)
=
( u,
g
h
) − (
∂(u)
i
∂x
j
, δ
h
) − (
· (g − g
h
), p).
We take supremum over all partial derivatives in both sides of the equation, and taking into
account that δ
h L
1
(Ω)
≤ C, then we can conclude that
(4.23)
u
h L
∞
(Ω)
≤ (C +
(g − g
h
)
L
1
(Ω)
)(
u
L
∞
(Ω)
+ p
L
∞
(Ω)
).
The result (4.23) is completed by Lemma 4.2.
Next, we prove the stability of the pressure in the maximum norm.
Let z ∈ T
z
and consider the problem (4.11) with a = 0 and b = 1. Then, using the definition
of the delta function we have
p
h
(z) = (p
h
, δ
h
) = (p
h
, δ
h
− φ) + (p
h
, φ).
We estimate the second term in the right hand side using C-S. inequality and the a priori
error estimate as follows
(p
h
, φ)
=
(p
h
− p, φ) + (p, φ)
≤ C( p − p
h L
2
(Ω)
+ p
L
2
(Ω)
) φ
L
2
(Ω)
≤ C(
u
L
2
(Ω)
+ p
L
2
(Ω)
)
≤ C(
u
L
∞
(Ω)
+ p
L
∞
(Ω)
).
Now, to estimate (p
h
, δ
h
− φ) we use (4.11b)
(p
h
, δ
h
− φ) = (p
h
,
· g) = (p
h
,
· g
h
) = (p,
· g
h
) + (p
h
− p,
· g
h
)
=
(p,
· g) + (p,
· (g
h
− g)) + ( (u − u
h
),
g
h
)
=
(p,
· g) + (p,
· (g
h
− g)) + ( (u − u
h
),
(g
h
− g)) + ( (u − u
h
),
g)
=
(p,
· g) + (p,
· (g
h
− g)) + ( (u − u
h
),
(g
h
− g)) + (
· (u − u
h
), λ)
=
(p, δ
h
− φ) + (p,
· (g
h
− g)) + ( (u − u
h
),
(g
h
− g))
+ (
· (u − u
h
), λ − Rλ)
≤ (
(u − u
h
)
L
∞
(Ω)
+ p
L
∞
(Ω)
)( δ
h L
1
(Ω)
+ φ
L
1
(Ω)
+
(g
h
− g)
L
1
(Ω)
+ λ − Rλ
L
1
(Ω)
)
≤ (
(u − u
h
)
L
∞
(Ω)
+ p
L
∞
(Ω)
)(C +
(g
h
− g)
L
1
(Ω)
+ λ − Rλ
L
1
(Ω)
)
The result (4.23) is completed by Lemma 4.2, Proposition 4.1 and the previous estimate for
the velocity in the L
∞
norm.
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
21
5. Taylor-Hood elements
We consider the Taylor-Hood elements of degree 2 in three dimension (d = 3), i.e.
V
h
=
{v ∈ [C
0
( ¯
Ω)]
3
: v|
T
∈ [P
2
]
3
, ∀T ∈ T
H
, v|
∂Ω
= 0 }
(5.1)
M
h
=
{q ∈ C
0
( ¯
Ω) : q|
T
∈ P
1
, ∀T ∈ T
h
} ∩ L
2
0
(Ω).
(5.2)
Assumptions A1-A3 hold for example by choosing P and R to be the Scott-Zhang [17]
interpolants onto V
h
and M
h
, respectively (see [14] and [3]). It is clear that the A4 assumption
holds in this case. We will prove assumptions A5 and A6 also hold.
We start with the local inf-sup condition A5.
Definition 1. Let b be a vertex of T
h
. We define σ(b) , the patch associated to the vertex b, as
the set of all elements containing b, i.e.
σ(b) := {T ∈ T
h
|
b ∈ T }
Lemma 5.1. Assume that every mesh element has at least 3 edges in int(Ω). Let B ⊂ Ω. Then,
there exists a constant c and a set B
h
⊂ T
h
which contains B and dist(B, ∂B
h
\Ω) ≤ 2h such
that the following inequality holds
sup
v∈V
h
supp(v)⊂B
h
Ω
q
· v
v
H
1
(B
h
)
≥ c
T ∈B
h
h
2
T
|q|
2
H
1
(T )
1/2
≥ ch
2
|q
h
|
H
1
(B)
.
for all q ∈ M
h
.
Proof. (We follow the proof in [4] section 4.2.5., see also [15])
Define the set of vertices
X := {x ∈ int(Ω) : x is a vertex of an element T ∈ T
h
such that T ∩ B = ∅}
Then, we define the set
B
h
:=
x∈X
σ(x),
Note that, the assumption that every mesh element has at least d edges in int(Ω) implies
that B ⊂ B
h
, and dist(B, B
h
) ≤ 2h. We claim that every element of B
h
has at most one face
on ∂B
h
. In fact, let T ∈ B
h
, by definition T belongs to the patch of an interior vertex. Then,
the claim follows from the observation that all the elements of an interior patch has at most one
face on the boundary of the patch.
Let N
i,h
ed
be the number of interior edges in B
h
. For the edge i, with 1 ≤ i ≤ N
i,h
ed
, denote by
d
i
and f
i
its two extremities and by m
i
its midpoint. Set l
i
= f
i
− d
i 3
and τ =
f
i
−d
i
f
i
−d
i 3
, the
length and the unit vector.
Then, for q ∈ M
h
we define v ∈ V
h
for all T ∈ T
h
as follows
v = 0,
if T ∈ T
h
\ int(B
h
)
v = 0,
at the vertices of T , if T ∈ B
h
v(m
i
) = −l
2
i
τ
i
sgn(∂
τ
i
q)|∂
τ
i
q|,
for all the interior edges i of T , if T ∈ B
h
22
MAX-NORM STABILITY OF LOW ORDER TAYLOR-HOOD ELEMENTS
IN THREE DIMENSIONS
Then, it is clear that supp(v) = B
h
and v ∈ V
h
.
Using the following quadrature formula,
T
φ(x)dx =
m
φ(m)
5
−
n
φ(n)
20
|T |
, ∀φ ∈ P
2
(T )
where m spans the set of the edge midpoint of T and n the set of nodes of T , we infer
Ω
q
h
· vdx = −
Ω
v ·
qdx
=
−
T ∈B
h
T
v ·
qdx
=
−
T ∈B
h
m∈T
v(m) ·
q(m)
5
−
n∈T
v(n) ·
q(n)
20
|T |
=
−
T ∈B
h
m
i
∈T
v(m
i
) ·
q(m
i
)
5
|T |
=
T ∈B
h
i:m
i
∈T
l
2
i
| q · τ
i
|
2
|T |
5
≥ c
T ∈B
h
h
2
T
|q|
2
H
1
(T )
.
We observe that the last step (
i:m
i
∈T
| q · τ
i
|
2
≥ | q|
2
) is only possible if every element of B
h
has at least 3 edges on int(B
h
), which is satisfied by our construction of B
h
and hypothesis on
the mesh ( every element has at least 3 edges in Ω). Furthermore, for T ∈ B
h
we have that
v
2
H
1
(T )
≤ ch
2
T
|q|
2
H
1
(T )
then,
v
H
1
(B
h
)
=
T ∈B
h
v
2
H
1
(T )
1/2
≤
T ∈B
h
ch
2
T
|q|
2
H
1
(T )
1/2
Therefore
sup
v∈V
h
supp(v)⊆B
h
Ω
q
· v
v
H
1
(B
h
)
≥ C
T ∈B
h
h
2
T
|q|
2
H
1
(T )
T ∈B
h
h
2
T
|q|
2
H
1
(T )
−1/2
=
C Dostları ilə paylaş: |