曲面偏微分方程:参数化有限元方法
前面介绍的 P \mathbf{P} P和 P d \mathbf{Pd} Pd,及其对应的同参等值延伸及投影自然延伸,对于曲面上有限元方法的理论分析起着至关重要的作用。
利普希茨参数曲面上的FEM
利普希茨参数化曲面
因为同参映射
P
\mathbf{P}
P的双利普希茨性,存在一个常数
L
L
L,使得
(64)
L
−
1
∣
x
1
−
x
2
∣
≤
∣
x
~
1
−
x
~
2
∣
≤
L
∣
x
1
−
x
2
∣
,
x
~
i
=
P
(
x
i
)
,
i
=
1
,
2
\begin{array}{ll}{\text { (64) }} & {L^{-1}\left|\mathrm{x}_{1}-\mathrm{x}_{2}\right| \leq\left|\widetilde{\mathrm{x}}_{1}-\widetilde{\mathrm{x}}_{2}\right| \leq L\left|\mathrm{x}_{1}-\mathrm{x}_{2}\right|, \quad \tilde{\mathrm{x}}_{i}=\mathrm{P}\left(\mathrm{x}_{i}\right), i=1,2}\end{array}
(64) L−1∣x1−x2∣≤∣x
1−x
2∣≤L∣x1−x2∣,x~i=P(xi),i=1,2
定义
T
T
T周围区域macro patches如下,这里
T
T
T表示的是
γ
\gamma
γ的一次逼近曲面
Γ
\Gamma
Γ的每个分片。
(
65
)
ω
T
=
∪
{
T
′
:
T
′
∩
T
≠
∅
}
,
ω
~
T
=
P
(
ω
T
)
(65) \quad \omega_{T}=\cup\left\{T^{\prime}: T^{\prime} \cap T \neq \emptyset\right\}, \quad \tilde{\omega}_{T}=\mathbf{P}\left(\omega_{T}\right)
(65)ωT=∪{T′:T′∩T=∅},ω~T=P(ωT)
定义形正则化常数,这里
h
T
:
=
∣
T
∣
1
n
h_{T}:=|T|^{\frac{1}{n}}
hT:=∣T∣n1,从表达式可以看出来,这个值越大,正则性越差。
(
66
)
σ
:
=
sup
T
max
T
∈
T
diam
(
T
)
h
T
(66) \quad \sigma:=\sup _{\mathcal{T}} \max _{T \in \mathcal{T}} \frac{\operatorname{diam}(T)}{h_{T}}
(66)σ:=TsupT∈TmaxhTdiam(T)
我们用
T
^
\widehat{T}
T
来表示参考单元单纯形。用
T
=
X
T
(
T
^
)
T=\mathbf{X}_{T}(\widehat{T})
T=XT(T
)来表示逼近曲面的参数化。从(66)可以看得出来下式。这个很重要,要怎么理解呢?
D
X
T
D \mathbf{X}_{T}
DXT就是一个线性变换,它的算子至于映射到
T
T
T上,模和
diam
(
T
)
\operatorname{diam}(T)
diam(T)同阶,那么
D
X
T
D \mathbf{X}_{T}
DXT就是
h
T
h_T
hT阶的。
(67)
h
T
∣
w
∣
≲
∣
D
X
T
w
∣
≲
h
T
∣
w
∣
,
∀
w
∈
R
n
\begin{array}{ll}{\text { (67) }} & {h_{T}|\mathrm{w}| \lesssim\left|D \mathrm{X}_{T} \mathrm{w}\right| \lesssim h_{T}|\mathrm{w}|, \quad \forall \mathrm{w} \in \mathbb{R}^{n}}\end{array}
(67) hT∣w∣≲∣DXTw∣≲hT∣w∣,∀w∈Rn
因为投影的双利普希茨性,我们也知道$
D \chi_{T}
和
和
和h_{T}$是同阶的。
(
68
)
h
T
∣
w
∣
≲
∣
D
χ
T
(
y
)
w
∣
≲
h
T
∣
w
∣
∀
w
∈
R
n
,
y
∈
T
^
(68) \quad h_{T}|\mathrm{w}| \lesssim\left|D \chi_{T}(\mathrm{y}) \mathrm{w}\right| \lesssim h_{T}|\mathrm{w}| \quad \forall \mathrm{w} \in \mathbb{R}^{n}, \mathrm{y} \in \widehat{T}
(68)hT∣w∣≲∣DχT(y)w∣≲hT∣w∣∀w∈Rn,y∈T
我们依然用tilde符号来表示在曲面
γ
\gamma
γ上,hat符号表示在参数域是(参考单元)上。
Γ
\Gamma
Γ的参数化表示为
X
T
\mathbf{X}_{T}
XT,
γ
\gamma
γ的参数化表示为
χ
T
=
P
∘
X
T
\chi_{T}=\mathbf{P} \circ \mathbf{X}_{T}
χT=P∘XT,那么在macro patches上,有,
(69)
L
−
1
h
T
∣
y
1
−
y
2
∣
≤
∣
x
~
1
−
x
~
2
∣
≤
L
h
T
∣
y
1
−
y
2
∣
\begin{array}{ll}{\text { (69) }} & {L^{-1} h_{T}\left|\mathbf{y}_{1}-\mathbf{y}_{2}\right| \leq\left|\widetilde{\mathbf{x}}_{1}-\widetilde{\mathbf{x}}_{2}\right| \leq L h_{T}\left|\mathbf{y}_{1}-\mathbf{y}_{2}\right|}\end{array}
(69) L−1hT∣y1−y2∣≤∣x
1−x
2∣≤LhT∣y1−y2∣
各个区域上的函数值定义如下:
(70)
v
^
T
(
y
)
:
=
v
~
T
(
χ
T
(
y
)
)
∀
x
^
∈
T
^
and
v
T
(
x
)
:
=
v
~
T
(
P
(
x
)
)
∀
x
∈
T
\begin{array}{ll}{\text { (70) } \quad \widehat{v}_{T}(\mathbf{y}):=\tilde{v}_{T}\left(\chi_{T}(\mathbf{y})\right) \quad \forall \widehat{\mathbf{x}} \in \widehat{T}} & {\text { and } \quad v_{T}(\mathbf{x}):=\tilde{v}_{T}(\mathbf{P}(\mathbf{x})) \quad \forall \mathbf{x} \in T}\end{array}
(70) v
T(y):=v~T(χT(y))∀x
∈T
and vT(x):=v~T(P(x))∀x∈T
多边形曲面上的微分几何
同以前的方式,可以定义第一标准型和面积元,
(71)
g
T
:
=
(
D
X
T
)
t
D
X
T
,
q
T
:
=
det
g
T
,
∀
T
∈
T
\begin{array}{ll}{\text { (71) }} & {\mathrm{g}_{T}:=\left(D \mathrm{X}_{T}\right)^{t} D \mathrm{X}_{T}, \quad q_{T}:=\sqrt{\operatorname{det} \mathrm{g}_{T}}, \quad \forall T \in \mathcal{T}}\end{array}
(71) gT:=(DXT)tDXT,qT:=detgT,∀T∈T
且满足,
(72)
eigen
(
g
T
)
≈
h
T
2
,
q
T
≈
h
T
n
,
∀
T
∈
T
\begin{array}{ll}{\text { (72) }} & {\text { eigen }\left(\mathrm{g}_{T}\right) \approx h_{T}^{2}, \quad q_{T} \approx h_{T}^{n}, \quad \forall T \in \mathcal{T}}\end{array}
(72) eigen (gT)≈hT2,qT≈hTn,∀T∈T
因为
D
X
T
D \mathbf{X}_{T}
DXT和
D
χ
T
D \mathbf{\chi}_{T}
DχT的同阶性,所以稳定化常数其实和网格尺寸是没有关系的,即
(
73
)
S
χ
≈
1
(73) \quad S_{\chi} \approx 1
(73)Sχ≈1
因为(72)我们知道
q
q
q和
q
Γ
q_\Gamma
qΓ是同阶的,即
C
1
≤
q
q
Γ
≤
C
2
C_{1} \leq \frac{q}{q_{\Gamma}} \leq C_{2}
C1≤qΓq≤C2
回忆曲面梯度和拉普拉斯算子的定义,有,
(75)
∇
v
^
=
(
D
X
)
t
∇
T
v
,
∇
T
v
=
(
D
X
)
g
Γ
−
1
∇
v
^
\begin{array}{ll}{\text { (75) }} & {\nabla \widehat{v}=(D \mathrm{X})^{t} \nabla_{\mathrm{T}} v, \quad \nabla_{\mathrm{T}} v=(D \mathrm{X}) \mathrm{g}_{\mathrm{\Gamma}}^{-1} \nabla \widehat{v}}\end{array}
(75) ∇v
=(DX)t∇Tv,∇Tv=(DX)gΓ−1∇v
(76)
Δ
Γ
v
=
1
q
Γ
div
(
q
Γ
g
Γ
−
1
∇
v
^
)
\begin{array}{ll}{\text { (76) }} & {\Delta_{\Gamma} v=\frac{1}{q_{\Gamma}} \operatorname{div}\left(q_{\Gamma} \mathbf{g}_{\Gamma}^{-1} \nabla \widehat{v}\right)}\end{array}
(76) ΔΓv=qΓ1div(qΓgΓ−1∇v
)
使用分布积分公式,很容易看到:
∫
Γ
∇
Γ
v
⋅
∇
Γ
w
=
∑
T
∈
T
−
∫
T
w
Δ
Γ
v
+
∫
∂
T
w
∇
Γ
v
⋅
μ
T
=
∑
T
∈
T
−
∫
T
w
Δ
Γ
v
+
∑
S
∈
S
∫
S
w
[
∇
Γ
v
]
\begin{aligned} \int_{\Gamma} \nabla_{\Gamma} v \cdot \nabla_{\Gamma} w &=\sum_{T \in \mathcal{T}}-\int_{T} w \Delta_{\Gamma} v+\int_{\partial T} w \nabla_{\Gamma} v \cdot \boldsymbol{\mu}_{T} \\ &=\sum_{T \in \mathcal{T}}-\int_{T} w \Delta_{\Gamma} v+\sum_{S \in \mathcal{S}} \int_{S} w\left[\nabla_{\Gamma} v\right] \end{aligned}
∫Γ∇Γv⋅∇Γw=T∈T∑−∫TwΔΓv+∫∂Tw∇Γv⋅μT=T∈T∑−∫TwΔΓv+S∈S∑∫Sw[∇Γv]
这里的跳跃,就定义为:
(
77
)
[
∇
Γ
v
]
:
=
∇
Γ
v
+
⋅
μ
+
+
∇
Γ
v
−
⋅
μ
−
(77) \quad\left[\nabla_{\Gamma} v\right]:=\nabla_{\Gamma} v_{+} \cdot \mu_{+}+\nabla_{\Gamma} v_{-} \cdot \mu_{-}
(77)[∇Γv]:=∇Γv+⋅μ++∇Γv−⋅μ−
参数化有限元方法
所谓的参数化有限元方法就是,寻找
U
:
=
U
T
∈
V
#
(
T
)
U:=U_{\mathcal{T}} \in \mathbb{V}_{\#}(\mathcal{T})
U:=UT∈V#(T),使得,
(78)
∫
Γ
∇
Γ
U
⋅
∇
Γ
V
=
∫
Γ
F
V
∀
V
∈
V
#
(
T
)
\begin{array}{ll}{\text { (78) }} & {\int_{\Gamma} \nabla_{\Gamma} U \cdot \nabla_{\Gamma} V=\int_{\Gamma} F V \quad \forall V \in \mathbb{V}_{\#}(\mathcal{T})}\end{array}
(78) ∫Γ∇ΓU⋅∇ΓV=∫ΓFV∀V∈V#(T)
这里的
F
∈
L
2
,
#
(
Γ
)
F \in L_{2, \#}(\Gamma)
F∈L2,#(Γ)。这里定义了分片线性连续空间,及其对应的零均值空间:
V
(
T
)
:
=
{
V
∈
C
0
(
Γ
)
∣
V
∣
T
=
V
^
∘
X
−
1
for some
V
^
∈
P
,
T
∈
T
}
\mathbb{V}(\mathcal{T}):=\left\{V \in C^{0}(\Gamma)|V|_{T}=\widehat{V} \circ \mathrm{X}^{-1} \text { for some } \widehat{V} \in \mathcal{P}, T \in \mathcal{T}\right\}
V(T):={V∈C0(Γ)∣V∣T=V
∘X−1 for some V
∈P,T∈T}
V
#
(
T
)
:
=
V
(
T
)
∩
L
2
,
#
(
Γ
)
\mathbb{V}_{\#}(\mathcal{T}):=\mathbb{V}(\mathcal{T}) \cap L_{2, \#}(\Gamma)
V#(T):=V(T)∩L2,#(Γ)
P
\mathcal{P}
P是线性多项式空间。
因为
F
∈
L
2
,
#
(
Γ
)
F \in L_{2, \#}(\Gamma)
F∈L2,#(Γ),那么,我们也有:
(79)
∫
Γ
∇
Γ
U
⋅
∇
Γ
V
=
∫
Γ
F
V
∀
V
∈
V
(
T
)
\text { (79) } \quad \int_{\Gamma} \nabla_{\Gamma} U \cdot \nabla_{\Gamma} V=\int_{\Gamma} F V \quad \forall V \in \mathbb{V}(\mathcal{T})
(79) ∫Γ∇ΓU⋅∇ΓV=∫ΓFV∀V∈V(T)
介绍一个引理:
证明:用一下误差矩阵的定义以及(78),再用一下
γ
\gamma
γ上的弱形式,并放到
Γ
\Gamma
Γ上,我们有,
∫
Γ
∇
Γ
(
u
−
U
)
⋅
∇
Γ
V
=
∫
γ
∇
γ
u
~
⋅
∇
γ
V
~
+
∫
Γ
∇
Γ
u
⋅
E
Γ
∇
Γ
V
−
∫
Γ
F
V
\int_{\Gamma} \nabla_{\Gamma}(u-U) \cdot \nabla_{\Gamma} V=\int_{\gamma} \nabla_{\gamma} \widetilde{u} \cdot \nabla_{\gamma} \widetilde{V}+\int_{\Gamma} \nabla_{\Gamma} u \cdot \mathbf{E}_{\Gamma} \nabla_{\Gamma} V-\int_{\Gamma} F V
∫Γ∇Γ(u−U)⋅∇ΓV=∫γ∇γu
⋅∇γV
+∫Γ∇Γu⋅EΓ∇ΓV−∫ΓFV
几何一致性
Γ \Gamma Γ上的一致庞加莱估计
回忆一下庞加莱不等式:
(82)
∥
v
∥
L
2
(
Γ
)
≲
∥
∇
v
∥
L
2
(
Γ
)
∀
v
∈
H
#
1
(
Γ
)
\begin{array}{ll}{\text { (82) }} & {\|v\|_{L_{2}(\Gamma)} \lesssim\|\nabla v\|_{L_{2}(\Gamma)} \quad \forall v \in H_{\#}^{1}(\Gamma)}\end{array}
(82) ∥v∥L2(Γ)≲∥∇v∥L2(Γ)∀v∈H#1(Γ)
几何估计量
定义一个几何元指示子,反应的是曲面及其逼近的导数之间的距离:
(83)
λ
T
:
=
∥
D
(
P
−
I
T
P
)
∥
L
∞
(
T
)
=
∥
D
P
−
I
∥
L
∞
(
T
)
∀
T
∈
T
\text { (83) } \quad \lambda_{T}:=\left\|D\left(\mathbf{P}-\mathcal{I}_{\mathcal{T}} \mathbf{P}\right)\right\|_{L_{\infty}(T)}=\|D \mathbf{P}-\mathbf{I}\|_{L_{\infty}(T)} \quad \forall T \in \mathcal{T}
(83) λT:=∥D(P−ITP)∥L∞(T)=∥DP−I∥L∞(T)∀T∈T
相应的几何估计量就定义为:
(84)
λ
T
(
Γ
)
:
=
max
T
∈
T
λ
T
\begin{array}{ll}{\text { (84) }} & {\lambda_{\mathcal{T}}(\Gamma):=\max _{T \in \mathcal{T}} \lambda_{T}}\end{array}
(84) λT(Γ):=maxT∈TλT
用一下复合函数求导,我们有:
(
85
)
max
y
∈
T
ˉ
∣
D
(
χ
T
−
X
T
)
(
y
)
∣
min
{
∣
D
−
χ
T
(
y
)
∣
,
∣
D
−
X
T
(
y
)
∣
}
≤
S
χ
λ
T
∀
T
∈
T
(85) \quad \max _{\mathbf{y} \in \bar{T}} \frac{\left|D\left(\chi_{T}-\mathbf{X}_{T}\right)(\mathbf{y})\right|}{\min \left\{\left|D^{-} \chi_{T}(\mathbf{y})\right|,\left|D^{-} \mathbf{X}_{T}(\mathbf{y})\right|\right\}} \leq S_{\chi} \lambda_{T} \quad \forall T \in \mathcal{T}
(85)y∈Tˉmaxmin{∣D−χT(y)∣,∣D−XT(y)∣}∣D(χT−XT)(y)∣≤SχλT∀T∈T
另外定义两个量,他们是
h
T
h_T
hT的高阶无穷小量,
(86)
β
T
:
=
∥
P
−
I
T
P
∥
L
∞
(
T
)
,
β
T
(
Γ
)
:
=
max
T
∈
T
β
T
\begin{array}{ll}{\text { (86) }} & {\beta_{T}:=\left\|\mathbf{P}-\mathcal{I}_{\mathcal{T}} \mathbf{P}\right\|_{L_{\infty}(T)}, \quad \beta_{\mathcal{T}}(\Gamma):=\max _{T \in \mathcal{T}} \beta_{T}}\end{array}
(86) βT:=∥P−ITP∥L∞(T),βT(Γ):=maxT∈TβT
(
87
)
μ
T
:
=
β
T
+
λ
T
2
,
μ
T
(
Γ
)
:
=
max
T
∈
T
μ
T
(87) \quad \mu_{T}:=\beta_{T}+\lambda_{T}^{2}, \quad \mu_{\mathcal{T}}(\Gamma):=\max _{T \in \mathcal{T}} \mu_{T}
(87)μT:=βT+λT2,μT(Γ):=T∈TmaxμT
C1曲面的几何一致性误差
给一个推论,
Corollary
32
(geometric consistency errors for
C
1
,
α
surfaces). If
X
and
χ
satisfy
(
67
)
and
(
68
)
,
then for all
T
∈
T
we have
(88)
∥
1
−
q
−
1
q
Γ
∥
L
∞
(
T
^
)
,
∥
I
−
grg
−
1
∥
L
∞
(
T
^
)
,
∥
ν
Γ
−
ν
∥
L
∞
(
T
)
≲
λ
T
where the hidden constants depend on
S
χ
≈
1
defined in
(
38
)
.
Moreover,
(89)
∥
E
∥
L
∞
(
T
^
)
+
∥
E
Γ
∥
L
∞
(
T
^
)
≲
λ
T
∀
T
∈
T
\begin{array}{l}{\text { Corollary } 32 \text { (geometric consistency errors for } C^{1, \alpha} \text { surfaces). If } \mathrm{X} \text { and } \chi \text { satisfy }} \\ {(67) \text { and }(68), \text { then for all } T \in \mathcal{T} \text { we have }} \\ {\begin{array}{ll}{\text { (88) }\left\|1-q^{-1} q_{\Gamma}\right\|_{L_{\infty}(\widehat{T})},\left\|\mathbf{I}-\operatorname{grg}^{-1}\right\|_{L_{\infty}(\widehat{T})},\left\|\nu_{\Gamma}-\nu\right\|_{L_{\infty}(T)} \lesssim \lambda_{T}} \\ {\text { where the hidden constants depend on } S_{\chi} \approx 1 \text { defined in }(38) . \text { Moreover, }} \\ {\text { (89) }} & {\|\mathbf{E}\|_{L_{\infty}(\widehat{T})}+\left\|\mathbf{E}_{\Gamma}\right\|_{L_{\infty}(\widehat{T})} \lesssim \lambda_{T} \quad \forall T \in \mathcal{T}}\end{array}}\end{array}
Corollary 32 (geometric consistency errors for C1,α surfaces). If X and χ satisfy (67) and (68), then for all T∈T we have (88) ∥∥1−q−1qΓ∥∥L∞(T
),∥∥I−grg−1∥∥L∞(T
),∥νΓ−ν∥L∞(T)≲λT where the hidden constants depend on Sχ≈1 defined in (38). Moreover, (89) ∥E∥L∞(T
)+∥EΓ∥L∞(T
)≲λT∀T∈T
这个证明和自然而然。用一下引理16和24。(88)的表达通通小于
λ
∞
\lambda_\infty
λ∞,用(85)放一下即可。
E
E
E的估计用(47)。
C2曲面的几何一致性误差
首先要让曲面和逼近曲面足够近,使得,
(90)
β
T
(
Γ
)
<
1
2
K
∞
⇒
Γ
⊂
N
\begin{array}{ll}{\text { (90) }} & {{ \beta }_\mathcal{T}(\Gamma)<\frac{1}{2 K_{\infty}} \Rightarrow \Gamma \subset \mathcal{N}}\end{array}
(90) βT(Γ)<2K∞1⇒Γ⊂N
其次,要让两个不同的投影走一个来回还是属于一个macro patches。
(
91
)
P
d
∘
P
−
1
(
T
~
)
⊂
ω
~
T
∀
T
∈
T
(91) \quad \mathrm{P}_{d} \circ \mathrm{P}^{-1}(\widetilde{T}) \subset \widetilde{\omega}_{T} \quad \forall T \in \mathcal{T}
(91)Pd∘P−1(T
)⊂ω
T∀T∈T
那么走一个来回之后,曲面上的两点的距离满足,
(
92
)
∣
x
~
−
P
d
∘
P
−
1
(
x
~
)
∣
=
∣
P
(
x
)
−
P
d
(
x
)
∣
≤
2
β
T
∀
x
∈
T
(92) \quad\left|\widetilde{\mathbf{x}}-\mathbf{P}_{d} \circ \mathbf{P}^{-1}(\widetilde{\mathbf{x}})\right|=\left|\mathbf{P}(\mathbf{x})-\mathbf{P}_{d}(\mathbf{x})\right| \leq 2 \beta_{T} \quad \forall \mathbf{x} \in T
(92)∣∣x
−Pd∘P−1(x
)∣∣=∣P(x)−Pd(x)∣≤2βT∀x∈T
那么,有这么一个推论,
Corollary
33
(geometric consistency errors for
C
2
surfaces). If
(
90
)
and
(
74
)
hold, then so do the following estimates for all
T
∈
T
(93)
∥
d
∥
L
∞
(
T
)
≲
β
T
,
∥
ν
−
ν
Γ
∥
L
∞
(
T
)
≲
λ
T
,
∥
1
−
q
−
1
q
Γ
∥
L
∞
(
T
)
≲
μ
T
where all the geometric quantities are defined using the parametrizations
χ
=
P
d
∘
X
and
X
.
Moreover,
(94)
∥
E
∥
L
∞
(
T
)
,
∥
E
Γ
∥
L
∞
(
T
)
≲
μ
T
∀
T
∈
T
\begin{array}{l}{\text { Corollary } 33 \text { (geometric consistency errors for } C^{2} \text { surfaces). If }(90) \text { and }(74)} \\ {\text { hold, then so do the following estimates for all } T \in \mathcal{T}} \\ {\text { (93) }\|d\|_{L_{\infty}(T)} \lesssim \beta_{T}, \quad\left\|\nu-\nu_{\Gamma}\right\|_{L_{\infty}(T)} \lesssim \lambda_{T}, \quad\left\|1-q^{-1} q_{\Gamma}\right\|_{L_{\infty}(T)} \lesssim \mu_{T}} \\ {\text { where all the geometric quantities are defined using the parametrizations } \chi=\mathbf{P}_{d} \circ \mathbf{X}} \\ {\text { and } \mathbf{X} . \text { Moreover, }} \\ {\begin{array}{llll}{\text { (94) }} & {\|\mathbf{E}\|_{L_{\infty}(T)},\left\|\mathbf{E}_{\Gamma}\right\|_{L_{\infty}(T)}} & {\lesssim \mu_{T}} & {\forall T \in \mathcal{T}}\end{array}}\end{array}
Corollary 33 (geometric consistency errors for C2 surfaces). If (90) and (74) hold, then so do the following estimates for all T∈T (93) ∥d∥L∞(T)≲βT,∥ν−νΓ∥L∞(T)≲λT,∥∥1−q−1qΓ∥∥L∞(T)≲μT where all the geometric quantities are defined using the parametrizations χ=Pd∘X and X. Moreover, (94) ∥E∥L∞(T),∥EΓ∥L∞(T)≲μT∀T∈T
证明只需用到C2曲面的相应估计式,以及推论已有结果即可。
由上可以得到一些有用的结果,利用
ω
~
\tilde \omega
ω~的利普希茨性质,可以有,
∣
w
~
(
x
~
)
−
w
~
(
P
d
∘
P
−
1
(
x
~
)
)
∣
≤
∥
∇
γ
w
~
∥
L
∞
(
ω
~
T
)
∣
x
~
−
P
d
∘
P
−
1
(
x
~
)
∣
≤
2
∥
∇
γ
w
~
∥
L
∞
(
ω
~
T
)
β
T
\left|\widetilde{w}(\widetilde{\mathbf{x}})-\widetilde{w}\left(\mathbf{P}_{d} \circ \mathbf{P}^{-1}(\widetilde{\mathbf{x}})\right)\right| \leq\left\|\nabla_{\gamma} \widetilde{w}\right\|_{L_{\infty}\left(\widetilde{\omega}_{T}\right)}\left|\widetilde{\mathbf{x}}-\mathbf{P}_{d} \circ \mathbf{P}^{-1}(\widetilde{\mathbf{x}})\right| \leq 2\left\|\nabla_{\gamma} \widetilde{w}\right\|_{L_{\infty}\left(\widetilde{\omega}_{T}\right)} \beta_{T}
∣∣w
(x
)−w
(Pd∘P−1(x
))∣∣≤∥∇γw
∥L∞(ω
T)∣∣x
−Pd∘P−1(x
)∣∣≤2∥∇γw
∥L∞(ω
T)βT
下面介绍一个命题,
Proposition 34 (mismatch between
P
and
P
d
). Assume that (67) as well as the
assumptions (74), (90) and (91) hold. Then there exists
λ
∗
>
0
such for
w
~
∈
H
1
(
γ
)
and
T
∈
T
we have
∥
w
~
−
w
~
∘
P
d
∘
P
−
1
∥
L
2
(
T
~
)
≲
β
T
∥
w
~
∥
H
1
(
ω
~
T
)
provided
λ
T
≤
λ
∗
and
ω
~
T
is a patch in
γ
around
T
~
.
\begin{array}{l}{\text { Proposition 34 (mismatch between } \mathbf{P} \text { and } \mathbf{P}_{d} \text { ). Assume that (67) as well as the }} \\ {\text { assumptions (74), (90) and (91) hold. Then there exists } \lambda_{*}>0 \text { such for } \widetilde{w} \in H^{1}(\gamma)} \\ {\text { and } T \in \mathcal{T} \text { we have }} \\ {\qquad\left\|\widetilde{w}-\widetilde{w} \circ \mathbf{P}_{d} \circ \mathbf{P}^{-1}\right\|_{L_{2}(\widetilde{T})} \lesssim \beta_{T}\|\widetilde{w}\|_{H^{1}\left(\widetilde{\omega}_{T}\right)}} \\ {\text { provided } \lambda_{T} \leq \lambda_{*} \text { and } \widetilde{\omega}_{T} \text { is a patch in } \gamma \text { around } \widetilde{T} .}\end{array}
Proposition 34 (mismatch between P and Pd ). Assume that (67) as well as the assumptions (74), (90) and (91) hold. Then there exists λ∗>0 such for w
∈H1(γ) and T∈T we have ∥∥w
−w
∘Pd∘P−1∥∥L2(T
)≲βT∥w
∥H1(ω
T) provided λT≤λ∗ and ω
T is a patch in γ around T
.
这个命题的证明分为3步:
- reduce到 R n \mathbb{R}^{n} Rn上
- 光滑化
- 估计各项
- 找 ϵ \epsilon ϵ的界
抓关键步骤,简言之,就是先把要证的东西做个参数域的转换,
∥
w
~
−
w
~
∘
ψ
∥
L
2
(
T
~
)
≲
h
T
n
/
2
∥
w
^
−
w
^
∘
ψ
^
∥
L
2
(
T
^
)
\|\widetilde{w}-\widetilde{w} \circ \psi\|_{L_{2}(\widetilde{T})} \lesssim h_{T}^{n / 2}\|\widehat{w}-\widehat{w} \circ \widehat{\psi}\|_{L_{2}(\widehat{T})}
∥w
−w
∘ψ∥L2(T
)≲hTn/2∥w
−w
∘ψ
∥L2(T
)
再把转换出来的东西拆成几部分:
∥
w
^
−
w
^
∘
ψ
^
∥
L
2
(
T
^
)
≲
∥
w
^
−
w
^
ε
∥
L
2
(
T
^
)
+
∥
w
^
ε
−
w
^
ε
∘
ψ
^
∥
L
2
(
T
^
)
+
∥
w
^
ε
∘
ψ
^
−
w
^
∘
ψ
^
∥
L
2
(
T
^
)
\|\widehat{w}-\widehat{w} \circ \widehat{\psi}\|_{L_{2}(\widehat{T})} \lesssim\left\|\widehat{w}-\widehat{w}_{\varepsilon}\right\|_{L_{2}(\widehat{T})}+\left\|\widehat{w}_{\varepsilon}-\widehat{w}_{\varepsilon} \circ \widehat{\psi}\right\|_{L_{2}(\widehat{T})}+\left\|\widehat{w}_{\varepsilon} \circ \widehat{\psi}-\widehat{w} \circ \widehat{\psi}\right\|_{L_{2}(\widehat{T})}
∥w
−w
∘ψ
∥L2(T
)≲∥w
−w
ε∥L2(T
)+∥∥∥w
ε−w
ε∘ψ
∥∥∥L2(T
)+∥∥∥w
ε∘ψ
−w
∘ψ
∥∥∥L2(T
)
估计第一部分……
∥
w
^
−
w
^
ε
∥
L
2
(
T
^
)
≲
ε
∣
w
^
∣
H
1
(
R
n
)
≲
ε
∣
w
^
∣
H
1
(
ω
^
T
)
\left\|\widehat{w}-\widehat{w}_{\varepsilon}\right\|_{L_{2}(\widehat{T})} \lesssim \varepsilon|\widehat{w}|_{H^{1}\left(\mathbb{R}^{n}\right)} \lesssim \varepsilon|\widehat{w}|_{H^{1}\left(\widehat{\omega}_{T}\right)}
∥w
−w
ε∥L2(T
)≲ε∣w
∣H1(Rn)≲ε∣w
∣H1(ω
T)
估计第三部分……
∥
(
w
^
ε
−
w
^
)
∘
ψ
^
∥
L
2
(
T
^
)
≲
∥
w
^
ε
−
w
^
∥
L
2
(
ω
^
T
)
≲
ε
∣
w
^
∣
H
1
(
ω
^
T
)
\left\|\left(\widehat{w}_{\varepsilon}-\widehat{w}\right) \circ \widehat{\psi}\right\|_{L_{2}(\widehat{T})} \lesssim\left\|\widehat{w}_{\varepsilon}-\widehat{w}\right\|_{L_{2}\left(\widehat{\omega}_{T}\right)} \lesssim \varepsilon|\widehat{w}|_{H^{1}\left(\widehat{\omega}_{T}\right)}
∥∥∥(w
ε−w
)∘ψ
∥∥∥L2(T
)≲∥w
ε−w
∥L2(ω
T)≲ε∣w
∣H1(ω
T)
估计第二部分……
∥
w
^
ε
−
w
^
ε
∘
ψ
^
∥
L
2
(
T
^
)
2
≲
ε
n
∑
i
∥
w
^
ε
−
w
^
ε
∘
ψ
^
∥
L
∞
(
B
(
y
i
,
ε
)
∩
T
^
)
≲
ε
2
∑
i
∣
w
^
∣
H
1
2
(
B
(
y
i
,
3
ε
)
)
≲
ε
2
∣
w
^
∣
H
1
(
R
n
)
2
≲
ε
2
∣
w
^
∣
H
1
(
ω
^
T
)
2
\begin{aligned}\left\|\widehat{w}_{\varepsilon}-\widehat{w}_{\varepsilon} \circ \widehat{\psi}\right\|_{L_{2}(\widehat{T})}^{2} & \lesssim \varepsilon^{n} \sum_{i}\left\|\widehat{w}_{\varepsilon}-\widehat{w}_{\varepsilon} \circ \widehat{\psi}\right\|_{L_{\infty}\left(B\left(\mathbf{y}_{i}, \varepsilon\right) \cap \widehat{T}\right)} \\ & \lesssim \varepsilon^{2} \sum_{i}|\widehat{w}|_{H^{1}}^{2}\left(B\left(\mathbf{y}_{i}, 3 \varepsilon\right)\right) \lesssim \varepsilon^{2}|\widehat{w}|_{H^{1}\left(\mathbb{R}^{n}\right)}^{2} \lesssim \varepsilon^{2}|\widehat{w}|_{H^{1}\left(\widehat{\omega}_{T}\right)}^{2} \end{aligned}
∥∥∥w
ε−w
ε∘ψ
∥∥∥L2(T
)2≲εni∑∥∥∥w
ε−w
ε∘ψ
∥∥∥L∞(B(yi,ε)∩T
)≲ε2i∑∣w
∣H12(B(yi,3ε))≲ε2∣w
∣H1(Rn)2≲ε2∣w
∣H1(ω
T)2
估计
ϵ
\epsilon
ϵ……
ε
≤
2
L
h
T
−
1
β
T
\varepsilon \leq 2 L h_{T}^{-1} \beta_{T}
ε≤2LhT−1βT
把所有东西合起来,完事。
∥
w
~
−
w
~
∘
ψ
∥
L
2
(
T
~
)
2
≲
h
T
n
∥
w
^
−
w
^
∘
ψ
^
∥
L
2
(
T
^
)
2
≲
h
T
n
ε
2
∣
w
^
∣
H
1
(
ω
^
T
)
2
≲
h
T
n
h
T
−
2
β
T
2
h
T
2
−
n
∣
w
~
∣
H
1
(
w
~
T
)
2
=
β
T
2
∣
w
~
∣
H
1
(
ω
~
T
)
2
\begin{aligned}\|\widetilde{w}-\widetilde{w} \circ \psi\|_{L_{2}(\widetilde{T})}^{2} & \lesssim h_{T}^{n}\|\widehat{w}-\widehat{w} \circ \widehat{\psi}\|_{L_{2}(\widehat{T})}^{2} \lesssim h_{T}^{n} \varepsilon^{2}|\widehat{w}|_{H^{1}\left(\widehat{\omega}_{T}\right)}^{2} \\ & \lesssim h_{T}^{n} h_{T}^{-2} \beta_{T}^{2} h_{T}^{2-n}|\widetilde{w}|_{H^{1}\left(\widetilde{w}_{T}\right)}^{2}=\beta_{T}^{2}|\widetilde{w}|_{H^{1}\left(\widetilde{\omega}_{T}\right)}^{2} \end{aligned}
∥w
−w
∘ψ∥L2(T
)2≲hTn∥w
−w
∘ψ
∥L2(T
)2≲hTnε2∣w
∣H1(ω
T)2≲hTnhT−2βT2hT2−n∣w
∣H1(w
T)2=βT2∣w
∣H1(ω
T)2
下面有另外一个命题,
Proposition 35 (Lipschitz perturbation). Let
Ω
1
,
Ω
2
⊂
⊂
Ω
⊂
R
n
+
1
be Lipschitz
bounded domains and
L
:
Ω
1
→
Ω
2
be a bi-Lipschitz isomorphism. If
r
:
=
max
x
∈
Ω
1
∣
L
(
x
)
−
x
∣
is sufficiently small so that
(
Ω
1
∪
Ω
2
)
+
B
(
0
,
r
)
⊂
Ω
then for all
g
∈
H
1
(
Ω
)
we have
∥
g
−
g
∘
L
∥
L
2
(
Ω
1
)
≲
r
∥
g
∥
H
1
(
Ω
)
\begin{array}{l}{\text { Proposition 35 (Lipschitz perturbation). Let } \Omega_{1}, \Omega_{2} \subset \subset \Omega \subset \mathbb{R}^{n+1} \text { be Lipschitz}} \\ {\text { bounded domains and } \mathbf{L}: \Omega_{1} \rightarrow \Omega_{2} \text { be a bi-Lipschitz isomorphism. If }} \\ {\qquad r:=\max _{\mathbf{x} \in \Omega_{1}}|\mathbf{L}(\mathbf{x})-\mathbf{x}|} \\ {\text { is sufficiently small so that }\left(\Omega_{1} \cup \Omega_{2}\right)+B(0, r) \subset \Omega \text { then for all } g \in H^{1}(\Omega) \text { we have }} \\ {\qquad\|g-g \circ \mathbf{L}\|_{L^{2}\left(\Omega_{1}\right)} \lesssim r\|g\|_{H^{1}(\Omega)}}\end{array}
Proposition 35 (Lipschitz perturbation). Let Ω1,Ω2⊂⊂Ω⊂Rn+1 be Lipschitz bounded domains and L:Ω1→Ω2 be a bi-Lipschitz isomorphism. If r:=maxx∈Ω1∣L(x)−x∣ is sufficiently small so that (Ω1∪Ω2)+B(0,r)⊂Ω then for all g∈H1(Ω) we have ∥g−g∘L∥L2(Ω1)≲r∥g∥H1(Ω)
先验误差分析
C2曲面的先验误差估计
Lemma
36
(approximability in
H
1
(
Γ
)
)
.
Let
γ
be a surface of class
C
2
and
u
~
∈
H
2
(
γ
)
.
Let
K
∞
be defined in
(
30
)
and
β
T
(
Γ
)
be given in
(
86
)
.
Then we have
(95)
inf
V
∈
V
(
T
)
∥
∇
Γ
(
u
~
∘
P
d
−
V
)
∥
L
2
(
Γ
)
≲
h
T
∣
u
~
∣
H
2
(
γ
)
+
β
T
(
Γ
)
K
∞
∥
∇
γ
u
~
∥
L
2
(
γ
)
\begin{array}{l}{ \text { Lemma }\left.36 \text { (approximability in } H^{1}(\Gamma)\right) . \text { Let } \gamma \text { be a surface of class } C^{2} \text { and } \widetilde{u} \in} \\ {H^{2}(\gamma) . \text { Let } K_{\infty} \text { be defined in }(30) \text { and } \beta_{\mathcal{T}}(\Gamma) \text { be given in }(86) . \text { Then we have }} \\ {\begin{array}{llll}{\text { (95) }} & {\inf _{V \in \mathrm{V}(\mathcal{T})}\left\|\nabla_{\Gamma}\left(\widetilde{u} \circ \mathbf{P}_{d}-V\right)\right\|_{L_{2}(\Gamma)}} & {\lesssim h_{\mathcal{T}}|\widetilde{u}|_{H^{2}(\gamma)}+\beta_{\mathcal{T}}(\Gamma) K_{\infty}\left\|\nabla_{\gamma} \widetilde{u}\right\|_{L_{2}(\gamma)}}\end{array}}\end{array}
Lemma 36 (approximability in H1(Γ)). Let γ be a surface of class C2 and u
∈H2(γ). Let K∞ be defined in (30) and βT(Γ) be given in (86). Then we have (95) infV∈V(T)∥∇Γ(u
∘Pd−V)∥L2(Γ)≲hT∣u
∣H2(γ)+βT(Γ)K∞∥∇γu
∥L2(γ)
这个引理告诉我们的是,解在
H
1
H^1
H1空间中逼近的下界是可控的。
它的证明用到整体下界小等于分片下界和:
(
96
)
inf
V
∈
V
(
T
)
∥
∇
Γ
(
u
~
∘
P
d
−
V
)
∥
L
2
(
Γ
)
2
≲
∑
T
∈
T
inf
V
T
∈
V
(
T
)
∥
∇
Γ
(
u
~
∘
P
d
−
V
T
)
∥
L
2
(
T
)
2
(96) \quad \inf _{V \in \mathbf{V}(\mathcal{T})}\left\|\nabla_{\Gamma}\left(\widetilde{u} \circ \mathbf{P}_{d}-V\right)\right\|_{L_{2}(\Gamma)}^{2} \lesssim \sum_{T \in \mathcal{T}} \inf _{V_{T} \in \mathbf{V}(T)}\left\|\nabla_{\Gamma}\left(\widetilde{u} \circ \mathbf{P}_{d}-V_{T}\right)\right\|_{L_{2}(T)}^{2}
(96)V∈V(T)inf∥∇Γ(u
∘Pd−V)∥L2(Γ)2≲T∈T∑VT∈V(T)inf∥∇Γ(u
∘Pd−VT)∥L2(T)2
下面是
H
1
H^1
H1先验误差估计:
Theorem
37
(
H
1
a-priori error estimate for
C
2
surfaces). Let
γ
be of class
C
2
f
~
∈
L
2
,
#
(
γ
)
and
u
~
∈
H
2
(
γ
)
be the solution of
(
18
)
.
Let
U
∈
V
#
(
T
)
be the solution
to
(
78
)
with
F
=
f
~
∘
P
q
q
r
defined via the lift
P
.
If the geometric assumptions
(
69
)
,
(90), and (91) are valid, then
∥
∇
Γ
(
u
~
∘
P
−
U
)
∥
L
2
(
Γ
)
≲
(
h
T
+
λ
τ
(
Γ
)
)
∥
f
~
∥
L
2
(
γ
)
≲
h
τ
∥
f
~
∥
L
2
(
γ
)
as well as
∥
∇
Γ
(
u
~
∘
P
d
−
U
)
∥
L
2
(
Γ
)
≲
(
h
T
+
μ
T
(
Γ
)
)
∥
f
~
∥
L
2
(
γ
)
≲
h
T
∥
f
~
∥
L
2
(
γ
)
\begin{array}{l}{\text { Theorem } 37\left(H^{1} \text { a-priori error estimate for } C^{2} \text { surfaces). Let } \gamma \text { be of class } C^{2}\right.} \\ {\tilde{f} \in L_{2, \#}(\gamma) \text { and } \widetilde{u} \in H^{2}(\gamma) \text { be the solution of }(18) . \text { Let } U \in \mathbb{V}_{\#}(\mathcal{T}) \text { be the solution }} \\ {\text { to }(78) \text { with } F=\widetilde{f} \circ \mathbf{P} \frac{q}{q_{\mathrm{r}}} \text { defined via the lift } \mathbf{P} . \text { If the geometric assumptions }(69),} \\ {\text { (90), and (91) are valid, then }} \\ {\qquad\left\|\nabla_{\Gamma}(\widetilde{u} \circ \mathbf{P}-U)\right\|_{L_{2}(\Gamma)} \lesssim(h _\mathcal{T}+\lambda \tau(\Gamma))\|\widetilde{f}\|_{L_{2}(\gamma)} \lesssim h \tau\|\widetilde{f}\|_{L_{2}(\gamma)}} \\ {\text { as well as }} \\ {\qquad\left\|\nabla_{\Gamma}\left(\widetilde{u} \circ \mathbf{P}_{d}-U\right)\right\|_{L_{2}(\Gamma)} \lesssim\left(h_{\mathcal{T}}+\mu_{\mathcal{T}}(\Gamma)\right)\|\tilde{f}\|_{L_{2}(\gamma)} \lesssim h_{\mathcal{T}}\|\tilde{f}\|_{L_{2(\gamma)}}}\end{array}
Theorem 37(H1 a-priori error estimate for C2 surfaces). Let γ be of class C2f~∈L2,#(γ) and u
∈H2(γ) be the solution of (18). Let U∈V#(T) be the solution to (78) with F=f
∘Pqrq defined via the lift P. If the geometric assumptions (69), (90), and (91) are valid, then ∥∇Γ(u
∘P−U)∥L2(Γ)≲(hT+λτ(Γ))∥f
∥L2(γ)≲hτ∥f
∥L2(γ) as well as ∥∇Γ(u
∘Pd−U)∥L2(Γ)≲(hT+μT(Γ))∥f~∥L2(γ)≲hT∥f~∥L2(γ)
这个估计表明了参数化方法在
H
1
H^1
H1空间中的一阶收敛性。
在
L
2
L^2
L2空间中,我们也有相应的估计:
Theorem 38 (
L
2
a-priori error estimate for
C
2
surfaces). Let
γ
be of class
C
2
and
be described by a generic lift
P
of class
C
2
.
Let the geometric conditions
(
69
)
,
,
and
(
91
)
be satisfied. Let
u
~
∈
H
#
1
(
γ
)
solve
(
19
)
and
U
∈
V
#
(
T
)
solve
(
78
)
with
F
=
f
~
∘
P
q
q
T
.
Then
(97)
∥
u
~
∘
P
−
U
∥
L
2
(
Γ
)
≲
h
T
2
∥
f
~
∥
L
2
(
γ
)
\begin{array}{l}{\text { Theorem 38 ( } L_{2} \text { a-priori error estimate for } C^{2} \text { surfaces). Let } \gamma \text { be of class } C^{2} \text { and }} \\ {\text { be described by a generic lift } \mathbf{P} \text { of class } C^{2} . \text { Let the geometric conditions }(69), \text { , }} \\ {\text { and }(91) \text { be satisfied. Let } \widetilde{u} \in H_{\#}^{1}(\gamma) \text { solve }(19) \text { and } U \in \mathbb{V}_{\#}(\mathcal{T}) \text { solve }(78) \text { with }} \\ {F=\widetilde{f} \circ \mathbf{P} \frac{q}{q_{\mathrm{T}}} . \text { Then }} \\ {\begin{array}{ll}{\text { (97) }} & {\|\widetilde{u} \circ \mathbf{P}-U\|_{L_{2}(\Gamma)} \lesssim h_{\mathcal{T}}^{2}\|\widetilde{f}\|_{L_{2}(\gamma)}}\end{array}}\end{array}
Theorem 38 ( L2 a-priori error estimate for C2 surfaces). Let γ be of class C2 and be described by a generic lift P of class C2. Let the geometric conditions (69), , and (91) be satisfied. Let u
∈H#1(γ) solve (19) and U∈V#(T) solve (78) with F=f
∘PqTq. Then (97) ∥u
∘P−U∥L2(Γ)≲hT2∥f
∥L2(γ)
它的证明,其实就是用到所谓的对偶技巧。
构造辅助问题:
∫
γ
∇
γ
z
~
⋅
∇
γ
w
~
=
∫
γ
(
u
~
−
U
~
#
)
w
~
∀
w
~
∈
H
#
1
(
γ
)
\int_{\gamma} \nabla_{\gamma} \widetilde{z} \cdot \nabla_{\gamma} \widetilde{w}=\int_{\gamma}\left(\widetilde{u}-\widetilde{U}_{\#}\right) \widetilde{w} \quad \forall \widetilde{w} \in H_{\#}^{1}(\gamma)
∫γ∇γz
⋅∇γw
=∫γ(u
−U
#)w
∀w
∈H#1(γ)
它的有限元逼近是:
∫
Γ
∇
Γ
Z
⋅
∇
Γ
W
=
∫
Γ
(
u
#
−
U
)
W
,
∀
W
∈
V
(
T
)
\int_{\Gamma} \nabla_{\Gamma} Z \cdot \nabla_{\Gamma} W=\int_{\Gamma}\left(u_{\#}-U\right) W, \quad \forall W \in \mathbb{V}(\mathcal{T})
∫Γ∇ΓZ⋅∇ΓW=∫Γ(u#−U)W,∀W∈V(T)
然后想办法将精确解和数值解的零中值化的差值表达成如下:
∥
u
~
−
U
~
#
∥
L
2
(
γ
)
2
=
∫
γ
∇
γ
(
u
~
−
U
~
)
⋅
∇
γ
(
z
~
−
Z
~
)
+
∫
γ
f
~
(
Z
∘
P
d
−
1
−
Z
∘
P
−
1
)
+
∫
γ
∇
γ
U
~
⋅
E
∇
γ
Z
~
\begin{aligned}\left\|\widetilde{u}-\widetilde{U}_{\#}\right\|_{L_{2}(\gamma)}^{2} &=\int_{\gamma} \nabla_{\gamma}(\widetilde{u}-\widetilde{U}) \cdot \nabla_{\gamma}(\widetilde{z}-\widetilde{Z}) \\ &+\int_{\gamma} \widetilde{f}\left(Z \circ \mathbf{P}_{d}^{-1}-Z \circ \mathbf{P}^{-1}\right) \\ &+\int_{\gamma} \nabla_{\gamma} \widetilde{U} \cdot \mathbf{E} \nabla_{\gamma} \widetilde{Z} \end{aligned}
∥∥∥u
−U
#∥∥∥L2(γ)2=∫γ∇γ(u
−U
)⋅∇γ(z
−Z
)+∫γf
(Z∘Pd−1−Z∘P−1)+∫γ∇γU
⋅E∇γZ
之后,分别估计每一项的界,就OK了。
上面的结论是对于
C
2
C^2
C2曲面的,我们相信,对于
C
3
C^3
C3曲面,取
P
=
P
d
\mathbf{P}=\mathbf{P}_{d}
P=Pd,我们有更好的结论如下:
(99)
∥
u
~
∘
P
−
U
∥
L
2
(
Γ
)
≲
h
T
2
∣
d
∣
W
∞
2
(
N
)
∥
f
~
∥
L
2
(
γ
)
\begin{array}{ll}{\text { (99) }} & {\|\widetilde{u} \circ \mathbf{P}-U\|_{L_{2}(\Gamma)} \lesssim h_{\mathcal{T}}^{2}|d|_{W_{\infty}^{2}(\mathcal{N})}\|\widetilde{f}\|_{L_{2}(\gamma)}}\end{array}
(99) ∥u
∘P−U∥L2(Γ)≲hT2∣d∣W∞2(N)∥f
∥L2(γ)
C1曲面的先验误差估计
对于C1曲面,我们同样有
H
1
H^1
H1空间中的逼近性质:
Lemma
39
(approximability in
H
1
(
Γ
)
)
.
Let
γ
be a surface of class
C
1
,
α
and
u
~
∈
H
1
+
s
(
γ
)
,
where
0
<
s
<
α
<
1
or
0
<
s
≤
α
=
1.
Then we have
(100)
inf
V
∈
V
(
T
)
∥
∇
Γ
(
u
~
∘
P
−
V
)
∥
L
2
(
Γ
)
≲
h
T
s
∣
u
~
∣
H
1
+
s
(
γ
)
\begin{array}{l}{ \text { Lemma }\left.39 \text { (approximability in } H^{1}(\Gamma)\right) . \text { Let } \gamma \text { be a surface of class } C^{1, \alpha} \text { and }} \\ {\widetilde{u} \in H^{1+s}(\gamma), \text { where } 0<s<\alpha<1 \text { or } 0<s \leq \alpha=1 . \text { Then we have }} \\ {\begin{array}{lll}{\text { (100) }} & {\inf _{V \in \mathbb{V}(\mathcal{T})}\left\|\nabla_{\Gamma}(\widetilde{u} \circ \mathbf{P}-V)\right\|_{L_{2}(\Gamma)} \lesssim h_{\mathcal{T}}^{s}|\widetilde{u}|_{H^{1+s}(\gamma)}}\end{array}}\end{array}
Lemma 39 (approximability in H1(Γ)). Let γ be a surface of class C1,α and u
∈H1+s(γ), where 0<s<α<1 or 0<s≤α=1. Then we have (100) infV∈V(T)∥∇Γ(u
∘P−V)∥L2(Γ)≲hTs∣u
∣H1+s(γ)
那么C1曲面的
H
1
H^1
H1先验误差估计就变成了:
Theorem
40
(
H
1
a-priori error estimate for
C
1
,
α
surfaces). Let
γ
be of class
C
1
,
α
,
0
<
α
≤
1
,
and assume that the geometric assumptions
(
69
)
,
(
90
)
,
and
(
91
)
are
valid. Let
f
^
∈
L
2
,
#
(
γ
)
and
u
~
∈
H
1
+
s
(
γ
)
be the solution of
(
18
)
and satisfy
\begin{array}{l}{\text { Theorem } 40\left(H^{1} \text { a-priori error estimate for } C^{1, \alpha} \text { surfaces). Let } \gamma \text { be of class } C^{1, \alpha},\right.} \\ {0<\alpha \leq 1, \text { and assume that the geometric assumptions }(69),(90), \text { and }(91) \text { are }} \\ {\text { valid. Let } \widehat{f} \in L_{2, \#}(\gamma) \text { and } \widetilde{u} \in H^{1+s}(\gamma) \text { be the solution of }(18) \text { and satisfy }}\end{array}
Theorem 40(H1 a-priori error estimate for C1,α surfaces). Let γ be of class C1,α,0<α≤1, and assume that the geometric assumptions (69),(90), and (91) are valid. Let f
∈L2,#(γ) and u
∈H1+s(γ) be the solution of (18) and satisfy
∥
u
~
∥
H
1
+
s
(
γ
)
≲
∥
f
~
∥
L
2
(
γ
)
provided
0
<
s
<
α
<
1
or
0
<
s
≤
α
=
1.
If
U
∈
V
#
(
T
)
is the solution to
(
78
)
with
F
=
f
~
∘
P
q
q
r
defined via the lift
P
,
then
∥
∇
Γ
(
u
~
∘
P
−
U
)
∥
L
2
(
Γ
)
≲
h
T
s
∥
u
~
∥
H
1
+
s
(
γ
)
+
λ
T
(
Γ
)
∥
f
~
∥
L
2
(
γ
)
≲
h
T
s
∥
f
~
∥
L
2
(
γ
)
\begin{array}{l}{\qquad\|\widetilde{u}\|_{H^{1+s}(\gamma)} \lesssim\|\tilde{f}\|_{L_{2}(\gamma)}} \\ {\text { provided } 0<s<\alpha<1 \text { or } 0<s \leq \alpha=1 . \text { If } U \in \mathbb{V}_{\#}(\mathcal{T}) \text { is the solution to }(78)} \\ {\text { with } F=\widetilde{f} \circ \mathbf{P} \frac{q}{q_{\mathrm{r}}} \text { defined via the lift } \mathbf{P}, \text { then }} \\ {\quad\left\|\nabla_{\Gamma}(\widetilde{u} \circ \mathbf{P}-U)\right\|_{L_{2}(\Gamma)} \lesssim h_{\mathcal{T}}^{s}\|\widetilde{u}\|_{H^{1+s}(\gamma)}+\lambda_{\mathcal{T}}(\Gamma)\|\tilde{f}\|_{L_{2}(\gamma)} \lesssim h_{\mathcal{T}}^{s}\|\tilde{f}\|_{L_{2}(\gamma)}}\end{array}
∥u
∥H1+s(γ)≲∥f~∥L2(γ) provided 0<s<α<1 or 0<s≤α=1. If U∈V#(T) is the solution to (78) with F=f
∘Pqrq defined via the lift P, then ∥∇Γ(u
∘P−U)∥L2(Γ)≲hTs∥u
∥H1+s(γ)+λT(Γ)∥f~∥L2(γ)≲hTs∥f~∥L2(γ)
后验误差分析
后验误差估计依赖于数值解 U U U和数据,但是不用到精确解 u ~ \tilde u u~。
要说明后验误差估计,我们可以先了解一下scott-zhang插值:
I
T
s
z
:
H
1
(
Γ
)
→
V
(
T
)
\mathcal{I}_{\mathcal{T}}^{\mathrm{sz}}: H^{1}(\Gamma) \rightarrow \mathbb{V}(\mathcal{T})
ITsz:H1(Γ)→V(T)
及其性质,
(
101
)
∥
v
−
I
T
s
z
v
∥
L
2
(
T
)
≲
h
T
∥
∇
Γ
v
∥
L
2
(
ω
T
)
,
∥
∇
Γ
I
T
s
z
v
∥
L
2
(
T
)
≲
∥
∇
Γ
v
∥
L
2
(
ω
T
)
(101) \quad\left\|v-\mathcal{I}_{\mathcal{T}}^{\mathrm{sz}} v\right\|_{L^{2}(T)} \lesssim h_{T}\left\|\nabla_{\Gamma} v\right\|_{L^{2}\left(\omega_{T}\right)}, \quad\left\|\nabla_{\Gamma} \mathcal{I}_{\mathcal{T}}^{\mathrm{sz}} v\right\|_{L^{2}(T)} \lesssim\left\|\nabla_{\Gamma} v\right\|_{L^{2}\left(\omega_{T}\right)}
(101)∥v−ITszv∥L2(T)≲hT∥∇Γv∥L2(ωT),∥∇ΓITszv∥L2(T)≲∥∇Γv∥L2(ωT)
我们还需要定义两个函数量“内部”和“跳跃残差”:
R
T
(
V
)
:
=
F
∣
T
+
Δ
Γ
V
∣
T
∀
T
∈
T
J
S
(
V
)
:
=
∇
Γ
V
+
∣
S
⋅
μ
S
+
+
∇
Γ
V
−
∣
S
⋅
μ
S
−
∀
S
∈
S
T
\begin{aligned} R_{T}(V) &:=\left.F\right|_{T}+\left.\Delta_{\Gamma} V\right|_{T} \quad \forall T \in \mathcal{T} \\ J_{S}(V) &:=\left.\nabla_{\Gamma} V^{+}\right|_{S} \cdot \boldsymbol{\mu}_{S}^{+}+\left.\nabla_{\Gamma} V^{-}\right|_{S} \cdot \boldsymbol{\mu}_{S}^{-} \quad \forall S \in S_{\mathcal{T}} \end{aligned}
RT(V)JS(V):=F∣T+ΔΓV∣T∀T∈T:=∇ΓV+∣∣S⋅μS++∇ΓV−∣∣S⋅μS−∀S∈ST
以及元指示子和误差估计子:
η
T
(
V
,
T
)
2
:
=
h
T
2
∥
R
T
(
V
)
∥
L
2
(
T
)
2
+
h
T
∥
J
∂
T
(
V
)
∥
L
2
(
∂
T
)
2
∀
T
∈
T
\eta _\mathcal{T}(V, T)^{2}:=h_{T}^{2}\left\|R_{T}(V)\right\|_{L^{2}(T)}^{2}+h_{T}\left\|J_{\partial T}(V)\right\|_{L^{2}(\partial T)}^{2} \quad \forall T \in \mathcal{T}
ηT(V,T)2:=hT2∥RT(V)∥L2(T)2+hT∥J∂T(V)∥L2(∂T)2∀T∈T
η
T
(
V
)
2
:
=
∑
T
∈
T
η
T
(
V
,
T
)
2
\eta_{\mathcal{T}}(V)^{2}:=\sum_{T \in \mathcal{T}} \eta \mathcal{T}(V, T)^{2}
ηT(V)2:=T∈T∑ηT(V,T)2
那么,我们首先有C1曲面的后验误差上界(H1空间):
Theorem
41
(a-posteriori upper bound for
C
1
,
α
surfaces). Let
γ
be of class
C
1
,
α
,
be parametrized by
χ
=
P
∘
X
and satisfy the geometric assumption (69). Let
u
~
∈
H
1
(
γ
)
be the solution to
(
18
)
and
U
∈
V
#
(
T
)
be the solution to (78) with
F
=
f
~
∘
P
q
q
r
∈
L
2
,
#
(
Γ
)
.
Then, for
U
~
:
=
U
∘
P
−
1
:
γ
→
R
we have
∥
∇
γ
(
u
~
−
U
~
)
∥
L
2
(
γ
)
2
≲
η
T
(
U
)
2
+
λ
T
2
(
Γ
)
∥
f
~
∥
L
2
(
γ
)
2
\begin{array}{l}{\text { Theorem } 41 \text { (a-posteriori upper bound for } C^{1, \alpha} \text { surfaces). Let } \gamma \text { be of class } C^{1, \alpha},} \\ {\text { be parametrized by } \chi=\mathbf{P} \circ \mathbf{X} \text { and satisfy the geometric assumption (69). Let }} \\ {\widetilde{u} \in H^{1}(\gamma) \text { be the solution to }(18) \text { and } U \in \mathbb{V}_{\#}(\mathcal{T}) \text { be the solution to (78) with }} \\ {F=\tilde{f} \circ \mathbf{P} \frac{q}{q_{\mathrm{r}}} \in L_{2, \#}(\Gamma) . \text { Then, for } \tilde{U}:=U \circ \mathbf{P}^{-1}: \gamma \rightarrow \mathbb{R} \text { we have }} \\ {\qquad\left\|\nabla_{\gamma}(\widetilde{u}-\widetilde{U})\right\|_{L^{2}(\gamma)}^{2} \lesssim \eta_{\mathcal{T}}(U)^{2}+\lambda_{\mathcal{T}}^{2}(\Gamma)\|\widetilde{f}\|_{L_{2}(\gamma)}^{2}}\end{array}
Theorem 41 (a-posteriori upper bound for C1,α surfaces). Let γ be of class C1,α, be parametrized by χ=P∘X and satisfy the geometric assumption (69). Let u
∈H1(γ) be the solution to (18) and U∈V#(T) be the solution to (78) with F=f~∘Pqrq∈L2,#(Γ). Then, for U~:=U∘P−1:γ→R we have ∥∥∥∇γ(u
−U
)∥∥∥L2(γ)2≲ηT(U)2+λT2(Γ)∥f
∥L2(γ)2
证明使用一个非常常用的套路,将误差所构成的双线性型拆成三部分:
(
102
)
∫
γ
∇
γ
(
u
~
−
U
~
)
⋅
∇
γ
v
~
=
I
1
+
I
2
+
I
3
(102) \quad \int_{\gamma} \nabla_{\gamma}(\widetilde{u}-\widetilde{U}) \cdot \nabla_{\gamma} \widetilde{v}=I_{1}+I_{2}+I_{3}
(102)∫γ∇γ(u
−U
)⋅∇γv
=I1+I2+I3
I
1
=
−
∫
Γ
∇
Γ
U
⋅
∇
Γ
(
v
−
V
)
+
∫
Γ
F
(
v
−
V
)
I
2
=
∫
γ
∇
γ
U
~
⋅
E
∇
γ
v
~
I
3
=
∫
γ
f
~
v
~
−
∫
Γ
F
v
\begin{aligned} I_{1} &=-\int_{\Gamma} \nabla_{\Gamma} U \cdot \nabla_{\Gamma}(v-V)+\int_{\Gamma} F(v-V) \\ I_{2} &=\int_{\gamma} \nabla_{\gamma} \widetilde{U} \cdot \mathbf{E} \nabla_{\gamma} \widetilde{v} \\ I_{3} &=\int_{\gamma} \widetilde{f} \widetilde{v}-\int_{\Gamma} F v \end{aligned}
I1I2I3=−∫Γ∇ΓU⋅∇Γ(v−V)+∫ΓF(v−V)=∫γ∇γU
⋅E∇γv
=∫γf
v
−∫ΓFv
易知,
(
103
)
I
1
=
∑
T
∈
T
∫
T
R
T
(
U
)
(
v
−
V
)
+
∑
S
∈
S
∫
S
J
S
(
U
)
(
v
−
V
)
(103) \quad I_{1}=\sum_{T \in \mathcal{T}} \int_{T} R_{T}(U)(v-V)+\sum_{S \in \mathcal{S}} \int_{S} J_{S}(U)(v-V)
(103)I1=T∈T∑∫TRT(U)(v−V)+S∈S∑∫SJS(U)(v−V)
(
104
)
I
1
≲
η
T
(
U
)
∥
∇
Γ
v
∥
L
2
(
Γ
)
≲
η
T
(
U
)
∥
∇
γ
v
~
∥
L
2
(
γ
)
(104) \quad I_{1} \lesssim \eta_{\mathcal{T}}(U)\left\|\nabla_{\Gamma} v\right\|_{L^{2}(\Gamma)} \lesssim \eta_{\mathcal{T}}(U)\left\|\nabla_{\gamma} \tilde{v}\right\|_{L^{2}(\gamma)}
(104)I1≲ηT(U)∥∇Γv∥L2(Γ)≲ηT(U)∥∇γv~∥L2(γ)
之后分别估计每一部分的界即可。
对于C1下界,有如下估计:
Theorem
42
(a-posteriori lower bound for
C
1
,
α
surfaces). Under the same condi-
tions of Theorem
41
(a-posteriori upper bound for
C
1
,
α
surfaces), we have
η
T
(
U
,
T
)
2
≲
∥
∇
γ
(
u
~
−
U
~
)
∥
L
2
(
ω
~
T
)
2
+
osc
T
(
F
,
ω
T
)
2
+
λ
T
2
(
ω
T
)
\begin{array}{l}{\text { Theorem } 42 \text { (a-posteriori lower bound for } C^{1, \alpha} \text { surfaces). Under the same condi- }} \\ {\text { tions of Theorem } 41 \text { (a-posteriori upper bound for } C^{1, \alpha} \text { surfaces), we have }} \\ {\qquad \eta_{\mathcal{T}}(U, T)^{2} \lesssim\left\|\nabla_{\gamma}(\widetilde{u}-\widetilde{U})\right\|_{L^{2}\left(\widetilde{\omega}_{T}\right)}^{2}+\operatorname{osc}_{\mathcal{T}}\left(F, \omega_{T}\right)^{2}+\lambda_{\mathcal{T}}^{2}\left(\omega_{T}\right)}\end{array}
Theorem 42 (a-posteriori lower bound for C1,α surfaces). Under the same condi- tions of Theorem 41 (a-posteriori upper bound for C1,α surfaces), we have ηT(U,T)2≲∥∥∥∇γ(u
−U
)∥∥∥L2(ω
T)2+oscT(F,ωT)2+λT2(ωT)
下面介绍C2曲面的后验误差上界,为此,介绍常量“数据震荡”如下:
osc
T
(
F
,
T
)
2
:
=
h
T
2
∥
F
−
F
ˉ
∥
L
2
(
T
)
2
,
osc
T
(
F
)
2
:
=
∑
T
∈
T
osc
T
(
F
,
T
)
2
\operatorname{osc} \mathcal{T}(F, T)^{2}:=h_{T}^{2}\|F-\bar{F}\|_{L^{2}(T)}^{2}, \quad \operatorname{osc} \mathcal{T}(F)^{2}:=\sum_{T \in \mathcal{T}} \operatorname{osc} \mathcal{T}(F, T)^{2}
oscT(F,T)2:=hT2∥F−Fˉ∥L2(T)2,oscT(F)2:=T∈T∑oscT(F,T)2
那么,我们有如下定理(C2曲面的后验误差上界):
Theorem
43
(a-posteriori upper bound for
C
2
surfaces). Let
γ
be of class
C
2
and
(
67
)
,
(
74
)
,
(
90
)
,
and (91) hold. Let
u
~
be the solution of
(
18
)
with
f
~
∈
L
2
,
#
(
γ
)
and
U
∈
V
(
T
)
be the solution to
(
78
)
with
F
=
f
~
∘
P
q
q
r
,
where
q
corresponds to the
parametrization
χ
=
P
∘
X
of
γ
. Then
∥
∇
γ
(
u
~
−
U
∘
P
d
−
1
)
∥
L
2
(
γ
)
2
≲
η
T
(
U
)
2
+
μ
T
2
(
Γ
)
∥
f
~
∥
L
2
(
γ
)
2
\begin{array}{l}{\text { Theorem } 43 \text { (a-posteriori upper bound for } C^{2} \text { surfaces). Let } \gamma \text { be of class } C^{2} \text { and }} \\ {(67),(74),(90), \text { and (91) hold. Let } \widetilde{u} \text { be the solution of }(18) \text { with } \widetilde{f} \in L_{2, \#}(\gamma) \text { and }} \\ {U \in \mathbb{V}(\mathcal{T}) \text { be the solution to }(78) \text { with } F=\widetilde{f} \circ \mathrm{P} \frac{q}{q_{\mathrm{r}}}, \text { where } q \text { corresponds to the }} \\ {\text { parametrization } \chi=\mathrm{P} \circ \mathrm{X} \text { of } \gamma \text { . Then }} \\ {\qquad\left\|\nabla_{\gamma}\left(\widetilde{u}-U \circ \mathbf{P}_{d}^{-1}\right)\right\|_{L_{2}(\gamma)}^{2} \lesssim \eta _\mathcal{T}(U)^{2}+\mu_{\mathcal{T}}^{2}(\Gamma)\|\tilde{f}\|_{L_{2}(\gamma)}^{2}}\end{array}
Theorem 43 (a-posteriori upper bound for C2 surfaces). Let γ be of class C2 and (67),(74),(90), and (91) hold. Let u
be the solution of (18) with f
∈L2,#(γ) and U∈V(T) be the solution to (78) with F=f
∘Pqrq, where q corresponds to the parametrization χ=P∘X of γ . Then ∥∥∇γ(u
−U∘Pd−1)∥∥L2(γ)2≲ηT(U)2+μT2(Γ)∥f~∥L2(γ)2
证明思路:
和前面是类似的拆分:
(
105
)
∫
γ
∇
γ
(
u
~
−
U
~
)
⋅
∇
γ
v
~
=
I
1
+
I
2
+
I
3
with
I
1
=
−
∫
Γ
∇
Γ
U
⋅
∇
Γ
(
v
−
V
)
+
∫
Γ
F
(
v
−
V
)
I
2
=
∫
γ
∇
γ
U
~
⋅
E
∇
γ
v
~
I
3
=
∫
γ
f
~
v
~
−
∫
Γ
F
v
\begin{array}{l}{(105) \quad \quad \int_{\gamma} \nabla_{\gamma}(\widetilde{u}-\widetilde{U}) \cdot \nabla_{\gamma} \widetilde{v}=I_{1}+I_{2}+I_{3}} \\ {\text { with }} \\ {\qquad \begin{aligned} I_{1} &=-\int_{\Gamma} \nabla_{\Gamma} U \cdot \nabla_{\Gamma}(v-V)+\int_{\Gamma} F(v-V) \\ I_{2} &=\int_{\gamma} \nabla_{\gamma} \widetilde{U} \cdot \mathbf{E} \nabla_{\gamma} \widetilde{v} \\ I_{3} &=\int_{\gamma} \widetilde{f} \tilde{v}-\int_{\Gamma} F v \end{aligned}}\end{array}
(105)∫γ∇γ(u
−U
)⋅∇γv
=I1+I2+I3 with I1I2I3=−∫Γ∇ΓU⋅∇Γ(v−V)+∫ΓF(v−V)=∫γ∇γU
⋅E∇γv
=∫γf
v~−∫ΓFv
只不过这时候,
I
3
I_3
I3不再等于0,而是:
(
106
)
I
3
=
∫
γ
f
~
(
v
~
−
v
~
∘
P
d
∘
P
−
1
)
(106) \quad I_{3}=\int_{\gamma} \widetilde{f}\left(\widetilde{v}-\widetilde{v} \circ \mathbf{P}_{d} \circ \mathbf{P}^{-1}\right)
(106)I3=∫γf
(v
−v
∘Pd∘P−1)
估计一下,有:
I
3
≲
β
T
(
Γ
)
∥
f
~
∥
L
2
(
γ
)
∥
∇
γ
v
~
∥
L
2
(
γ
)
I_{3} \lesssim \beta_{\mathcal{T}}(\Gamma)\|\tilde{f}\|_{L_{2}(\gamma)}\left\|\nabla_{\gamma} \tilde{v}\right\|_{L_{2}(\gamma)}
I3≲βT(Γ)∥f~∥L2(γ)∥∇γv~∥L2(γ)
C2曲面的后验误差下界如下所述:
Theorem
44
(a-posteriori lower bound for
C
2
surfaces). Under the same condi-
tions as Theorem
43
(a-posteriori upper bound for
C
2
surfaces), we have
η
T
(
U
,
T
)
2
≲
∥
∇
γ
(
u
~
−
U
~
)
∥
L
2
(
ω
~
T
)
2
+
osc
T
(
F
,
ω
T
)
2
+
μ
T
(
ω
T
)
2
where
μ
T
(
ω
T
)
=
max
T
′
⊂
ω
T
μ
T
′
\begin{array}{l}{\text { Theorem } 44 \text { (a-posteriori lower bound for } C^{2} \text { surfaces). Under the same condi- }} \\ {\text { tions as Theorem } 43 \text { (a-posteriori upper bound for } C^{2} \text { surfaces), we have }} \\ {\qquad \eta_{\mathcal{T}}(U, T)^{2} \lesssim\left\|\nabla_{\gamma}(\widetilde{u}-\widetilde{U})\right\|_{L^{2}\left(\widetilde{\omega}_{T}\right)}^{2}+\operatorname{osc}_{\mathcal{T}}\left(F, \omega_{T}\right)^{2}+\mu \mathcal{T}\left(\omega_{T}\right)^{2}} \\ {\text { where } \mu_{\mathcal{T}}\left(\omega_{T}\right)=\max _{T^{\prime} \subset \omega_{T}} \mu_{T^{\prime}}}\end{array}
Theorem 44 (a-posteriori lower bound for C2 surfaces). Under the same condi- tions as Theorem 43 (a-posteriori upper bound for C2 surfaces), we have ηT(U,T)2≲∥∥∥∇γ(u
−U
)∥∥∥L2(ω
T)2+oscT(F,ωT)2+μT(ωT)2 where μT(ωT)=maxT′⊂ωTμT′