曲面扰动理论之C2曲面的扰动
依然假设邻域:
(
48
)
N
=
{
x
∈
R
n
+
1
:
∣
d
(
x
)
∣
<
1
2
K
∞
}
(48) \quad \mathcal{N}=\left\{\mathrm{x} \in \mathbb{R}^{n+1}:|d(\mathrm{x})|<\frac{1}{2 K_{\infty}}\right\}
(48)N={x∈Rn+1:∣d(x)∣<2K∞1}
假设有曲面
Γ
\Gamma
Γ,它到
γ
\gamma
γ的最近点投影是个双射:
P
d
=
I
−
d
∇
d
:
Γ
→
γ
\mathbf{P}_{d}=\mathbf{I}-d \nabla d: \Gamma \rightarrow \gamma
Pd=I−d∇d:Γ→γ
v
v
v是
v
~
\tilde v
v~的沿法向的自然扩充,即:
v
=
v
~
∘
P
d
v=\tilde{v} \circ \mathbf{P}_{d}
v=v~∘Pd
我们有如下引理,描述了两个曲面上函数梯度的关系:
Lemma
20
(relation between tangential gradients). If
v
~
:
γ
→
R
is of class
H
1
,
then the tangential gradients
∇
γ
v
~
and
∇
Γ
v
satisfy for all
x
∈
Γ
(
49
)
∇
Γ
v
(
x
)
=
Π
Γ
(
x
)
(
I
−
d
W
)
(
x
)
Π
(
x
)
∇
γ
v
~
(
P
d
(
x
)
)
(50)
∇
γ
v
~
(
P
d
(
x
)
)
=
(
I
−
d
W
)
−
1
(
x
)
(
I
−
ν
Γ
(
x
)
⊗
ν
(
x
)
ν
Γ
(
x
)
⋅
ν
(
x
)
)
∇
Γ
v
(
x
)
\begin{array}{l}{\text { Lemma } 20 \text { (relation between tangential gradients). If } \tilde{v}: \gamma \rightarrow \mathbb{R} \text { is of class } H^{1} \text { , }} \\ {\text { then the tangential gradients } \nabla_{\gamma} \tilde{v} \text { and } \nabla_{\Gamma} v \text { satisfy for all } \mathrm{x} \in \Gamma} \\ {(49) \quad \nabla_{\Gamma} v(\mathrm{x})=\Pi_{\Gamma}(\mathrm{x})(\mathrm{I}-d \mathbf{W})(\mathrm{x}) \Pi(\mathrm{x}) \nabla_{\gamma} \tilde{v}\left(\mathrm{P}_{d}(\mathrm{x})\right)} \\ {\begin{array}{llll}{\text { (50) }} & {\nabla_{\gamma} \tilde{v}\left(\mathbf{P}_{d}(\mathrm{x})\right)=} & {(\mathbf{I}-d \mathbf{W})^{-1}(\mathbf{x})\left(\mathbf{I}-\frac{\nu_{\Gamma}(\mathbf{x}) \otimes \nu(\mathbf{x})}{\nu_{\Gamma}(\mathbf{x}) \cdot \nu(\mathbf{x})}\right) \nabla_{\Gamma} v(\mathbf{x})}\end{array}}\end{array}
Lemma 20 (relation between tangential gradients). If v~:γ→R is of class H1 , then the tangential gradients ∇γv~ and ∇Γv satisfy for all x∈Γ(49)∇Γv(x)=ΠΓ(x)(I−dW)(x)Π(x)∇γv~(Pd(x)) (50) ∇γv~(Pd(x))=(I−dW)−1(x)(I−νΓ(x)⋅ν(x)νΓ(x)⊗ν(x))∇Γv(x)
证明:
由曲面梯度的第二定义以及式(27)所描述的曲面外一点的梯度和投影点曲面梯度关系,我们立马可以得到:
∇
Γ
v
(
x
)
=
Π
Γ
(
x
)
∇
v
(
x
)
=
Π
Γ
(
x
)
(
I
−
d
W
)
(
x
)
Π
(
x
)
∇
γ
v
~
(
P
d
(
x
)
)
\nabla_{\Gamma} v(\mathrm{x})=\Pi_{\Gamma}(\mathrm{x}) \nabla v(\mathrm{x})=\Pi_{\Gamma}(\mathrm{x})(\mathrm{I}-d \mathrm{W})(\mathrm{x}) \Pi(\mathrm{x}) \nabla_{\gamma} \tilde{v}\left(\mathrm{P}_{d}(\mathrm{x})\right)
∇Γv(x)=ΠΓ(x)∇v(x)=ΠΓ(x)(I−dW)(x)Π(x)∇γv~(Pd(x))
第一个结论得证。
为了证明第二个结论,我们必须找出
∇
v
\nabla v
∇v和
∇
Γ
v
\nabla_{\Gamma} v
∇Γv的一个关系。
首先,有两个式子:
∇
v
=
(
I
−
ν
Γ
⊗
ν
Γ
)
∇
v
+
ν
Γ
⊗
ν
Γ
∇
v
=
∇
Γ
v
+
(
∇
v
⋅
ν
Γ
)
ν
Γ
\nabla v=\left(\mathbf{I}-\nu_{\Gamma} \otimes \nu_{\Gamma}\right) \nabla v+\nu_{\Gamma} \otimes \nu_{\Gamma} \nabla v=\nabla_{\Gamma} v+\left(\nabla v \cdot \nu_{\Gamma}\right) \nu_{\Gamma}
∇v=(I−νΓ⊗νΓ)∇v+νΓ⊗νΓ∇v=∇Γv+(∇v⋅νΓ)νΓ
∇
Γ
v
⋅
ν
+
(
ν
Γ
⋅
ν
)
∇
v
⋅
ν
Γ
=
0
⇒
∇
v
⋅
ν
Γ
=
−
1
ν
Γ
⋅
ν
∇
Γ
v
⋅
ν
\nabla_{\Gamma} v \cdot \nu+\left(\nu_{\Gamma} \cdot \nu\right) \nabla v \cdot \nu_{\Gamma}=0 \quad \Rightarrow \quad \nabla v \cdot \nu_{\Gamma}=-\frac{1}{\nu_{\Gamma} \cdot \nu} \nabla_{\Gamma} v \cdot \nu
∇Γv⋅ν+(νΓ⋅ν)∇v⋅νΓ=0⇒∇v⋅νΓ=−νΓ⋅ν1∇Γv⋅ν
第二个式子是因为
∇
v
(
x
)
⋅
ν
(
x
)
=
0
\nabla v(\mathbf{x}) \cdot \boldsymbol{\nu}(\mathbf{x})=0
∇v(x)⋅ν(x)=0。
整合两个式子,得到:
∇
v
(
x
)
=
(
I
−
ν
Γ
(
x
)
⊗
ν
(
x
)
ν
Γ
(
x
)
⋅
ν
(
x
)
)
∇
Γ
v
(
x
)
∀
x
∈
Γ
\nabla v(\mathrm{x})=\left(\mathbf{I}-\frac{\nu_{\Gamma}(\mathrm{x}) \otimes \nu(\mathrm{x})}{\nu_{\Gamma}(\mathrm{x}) \cdot \nu(\mathrm{x})}\right) \nabla_{\Gamma} v(\mathrm{x}) \quad \forall \mathrm{x} \in \Gamma
∇v(x)=(I−νΓ(x)⋅ν(x)νΓ(x)⊗ν(x))∇Γv(x)∀x∈Γ
代入到
∇
v
\nabla v
∇v和
∇
γ
v
\nabla_{\gamma} v
∇γv的关系,立证。
Lemma
21
(geometric consistency). The error matrices
E
,
E
Γ
∈
R
(
n
+
1
)
×
(
n
+
1
)
in
(34) are given on
Γ
by
(51)
E
∘
P
d
:
=
q
Γ
q
Π
(
I
−
d
W
)
Π
Γ
(
I
−
d
W
)
Π
−
Π
(52)
E
Γ
:
=
q
q
Γ
(
I
−
ν
⊗
ν
Γ
ν
⋅
ν
Γ
)
(
I
−
d
W
)
−
2
(
I
−
ν
Γ
⊗
ν
ν
⋅
ν
Γ
)
−
Π
Γ
\begin{array}{l}{\text { Lemma } 21 \text { (geometric consistency). The error matrices } \mathbf{E}, \mathbf{E}_{\Gamma} \in \mathbb{R}^{(n+1) \times(n+1)} \text { in }} \\ {\begin{array}{ll}{\text { (34) are given on } \Gamma} & {\text { by }} \\ {\text { (51) }} & {\mathbf{E} \circ \mathbf{P}_{d}:=\frac{q_{\Gamma}}{q} \Pi(\mathbf{I}-d \mathbf{W}) \Pi_{\Gamma}(\mathbf{I}-d \mathbf{W}) \Pi-\Pi} \\ {\text { (52) }} & {\mathbf{E}_{\Gamma}:=\frac{q}{q_{\Gamma}}\left(I-\frac{\nu \otimes \nu_{\Gamma}}{\nu \cdot \nu_{\Gamma}}\right)(\mathbf{I}-d \mathbf{W})^{-2}\left(I-\frac{\nu_{\Gamma} \otimes \nu}{\nu \cdot \nu_{\Gamma}}\right)-\Pi_{\Gamma}}\end{array}}\end{array}
Lemma 21 (geometric consistency). The error matrices E,EΓ∈R(n+1)×(n+1) in (34) are given on Γ (51) (52) by E∘Pd:=qqΓΠ(I−dW)ΠΓ(I−dW)Π−ΠEΓ:=qΓq(I−ν⋅νΓν⊗νΓ)(I−dW)−2(I−ν⋅νΓνΓ⊗ν)−ΠΓ
证明:
由上面一个引理,立马得到:
∫
Γ
∇
Γ
w
⋅
∇
Γ
v
=
∫
γ
∇
γ
w
~
⋅
(
q
Γ
q
Π
(
I
−
d
W
)
Π
Γ
(
I
−
d
W
)
Π
)
∇
γ
v
~
\int_{\Gamma} \nabla_{\Gamma} w \cdot \nabla_{\Gamma} v=\int_{\gamma} \nabla_{\gamma} \widetilde{w} \cdot\left(\frac{q_{\Gamma}}{q} \Pi(\mathbf{I}-d \mathbf{W}) \Pi_{\Gamma}(\mathbf{I}-d \mathbf{W}) \Pi\right) \nabla_{\gamma} \widetilde{v}
∫Γ∇Γw⋅∇Γv=∫γ∇γw
⋅(qqΓΠ(I−dW)ΠΓ(I−dW)Π)∇γv
另一个类似可证。
Lemma
22
(relation between
q
and
q
Γ
)
.
Given any parametrization
χ
Γ
of
Γ
,
let
χ
:
=
P
d
∘
χ
Γ
be the parametrization of
γ
.
If
ν
(
x
)
⋅
ν
Γ
(
x
)
≥
0
for all
x
∈
Γ
,
then
the ratio of area elements
q
(
y
)
/
q
Γ
(
y
)
with
y
=
χ
Γ
−
1
(
x
)
satisfies
(53)
q
(
y
)
q
Γ
(
y
)
=
det
(
I
−
d
(
x
)
W
(
x
)
)
(
ν
(
x
)
⋅
ν
Γ
(
x
)
)
∀
x
∈
Γ
\begin{array}{l}{ \text { Lemma }\left.22 \text { (relation between } q \text { and } q_{\Gamma}\right) . \text { Given any parametrization } \chi_{\Gamma} \text { of } \Gamma, \text { let }} \\ {\chi:=\mathbf{P}_{d} \circ \chi_{\Gamma} \text { be the parametrization of } \gamma . \text { If } \nu(\mathrm{x}) \cdot \nu_{\Gamma}(\mathrm{x}) \geq 0 \text { for all } \mathrm{x} \in \Gamma, \text { then }} \\ {\text { the ratio of area elements } q(\mathrm{y}) / q_{\Gamma}(\mathrm{y}) \text { with } \mathrm{y}=\chi_{\Gamma}^{-1}(\mathrm{x}) \text { satisfies }} \\ {\begin{array}{llll}{\text { (53) }} & {\frac{q(\mathrm{y})}{q_{\Gamma}(\mathrm{y})}=\operatorname{det}(\mathrm{I}-d(\mathrm{x}) \mathrm{W}(\mathrm{x}))\left(\nu(\mathrm{x}) \cdot \nu_{\Gamma}(\mathrm{x})\right)} & {\forall \mathrm{x} \in \Gamma}\end{array}}\end{array}
Lemma 22 (relation between q and qΓ). Given any parametrization χΓ of Γ, let χ:=Pd∘χΓ be the parametrization of γ. If ν(x)⋅νΓ(x)≥0 for all x∈Γ, then the ratio of area elements q(y)/qΓ(y) with y=χΓ−1(x) satisfies (53) qΓ(y)q(y)=det(I−d(x)W(x))(ν(x)⋅νΓ(x))∀x∈Γ
证明:
由(5)式,我们有:
q
q
Γ
=
det
(
[
ν
,
D
χ
]
[
ν
Γ
,
D
χ
Γ
]
−
1
)
\frac{q}{q_{\Gamma}}=\operatorname{det}\left([\nu, D \chi]\left[\nu_{\Gamma}, D \chi_{\Gamma}\right]^{-1}\right)
qΓq=det([ν,Dχ][νΓ,DχΓ]−1)
我们假设有:
[
ν
Γ
,
D
χ
Γ
]
−
1
=
[
v
,
A
]
t
\left[\nu_{\Gamma}, D \chi_{\Gamma}\right]^{-1}=[\mathbf{v}, \mathbf{A}]^{t}
[νΓ,DχΓ]−1=[v,A]t
那么由
[
v
,
A
]
t
[
ν
Γ
,
D
χ
Γ
]
=
I
[\mathbf{v}, \mathbf{A}]^{t}\left[\boldsymbol{\nu}_{\Gamma}, D \chi_{\Gamma}\right]=\mathbf{I}
[v,A]t[νΓ,DχΓ]=I和
[
ν
Γ
,
D
χ
Γ
]
[
v
,
A
]
t
=
I
\left[\nu_{\Gamma}, D \chi_{\Gamma}\right][\mathbf{v}, \mathbf{A}]^{t}=\mathbf{I}
[νΓ,DχΓ][v,A]t=I,我们有:
v
=
ν
Γ
\mathbf{v}=\nu_{\Gamma}
v=νΓ
D
χ
Γ
A
t
=
I
−
ν
Γ
⊗
ν
Γ
=
Π
Γ
D \chi_{\Gamma} \mathbf{A}^{t}=\mathbf{I}-\nu_{\Gamma} \otimes \nu_{\Gamma}=\Pi_{\Gamma}
DχΓAt=I−νΓ⊗νΓ=ΠΓ
[
ν
,
D
χ
]
[
ν
Γ
,
D
χ
Γ
]
−
1
=
ν
⊗
ν
Γ
+
D
χ
A
t
[\nu, D \chi]\left[\nu_{\Gamma}, D \chi_{\Gamma}\right]^{-1}=\nu \otimes \nu_{\Gamma}+D \chi \mathbf{A}^{t}
[ν,Dχ][νΓ,DχΓ]−1=ν⊗νΓ+DχAt
令
x
=
χ
Γ
(
y
)
∈
Γ
\mathbf{x}=\chi_{\Gamma}(\mathbf{y}) \in \Gamma
x=χΓ(y)∈Γ以及
χ
(
y
)
=
P
d
(
x
)
=
x
−
d
(
x
)
∇
d
(
x
)
∈
γ
\chi(\mathrm{y})=\mathrm{P}_{d}(\mathrm{x})=\mathrm{x}-d(\mathrm{x}) \nabla d(\mathrm{x}) \in \gamma
χ(y)=Pd(x)=x−d(x)∇d(x)∈γ,使用链式法则,有:
D
χ
(
y
)
=
(
I
−
d
(
x
)
W
(
x
)
)
Π
(
x
)
D
χ
Γ
(
y
)
∀
y
∈
V
D \chi(\mathbf{y})=(\mathbf{I}-d(\mathbf{x}) \mathbf{W}(\mathbf{x})) \Pi(\mathbf{x}) D \chi_{\Gamma}(\mathbf{y}) \quad \forall \mathbf{y} \in \mathcal{V}
Dχ(y)=(I−d(x)W(x))Π(x)DχΓ(y)∀y∈V
进而,我们有:
q
q
Γ
=
det
(
ν
⊗
ν
Γ
+
(
I
−
d
W
)
Π
Π
Γ
)
=
det
(
(
I
−
d
W
)
(
ν
⊗
ν
Γ
+
Π
Π
Γ
)
)
=
det
(
(
I
−
d
W
)
)
det
B
\begin{aligned} \frac{q}{q_{\Gamma}} &=\operatorname{det}\left(\nu \otimes \nu_{\Gamma}+(\mathbf{I}-d \mathbf{W}) \Pi \Pi_{\Gamma}\right) \\ &=\operatorname{det}\left((\mathbf{I}-d \mathbf{W})\left(\nu \otimes \nu_{\Gamma}+\Pi \Pi_{\Gamma}\right)\right)=\operatorname{det}((\mathbf{I}-d \mathbf{W})) \operatorname{det} \mathbf{B} \end{aligned}
qΓq=det(ν⊗νΓ+(I−dW)ΠΠΓ)=det((I−dW)(ν⊗νΓ+ΠΠΓ))=det((I−dW))detB
此处
B
:
=
ν
⊗
ν
Γ
+
Π
Π
Γ
\mathbf{B}:=\boldsymbol{\nu} \otimes \boldsymbol{\nu}_{\Gamma}+\Pi \Pi_{\Gamma}
B:=ν⊗νΓ+ΠΠΓ。这里第二个等号用到了
W
ν
=
0
\mathbf{W} \nu=0
Wν=0。至此,我们只需要证明
det
B
=
ν
⋅
ν
Γ
\operatorname{det} \mathbf{B}=\nu \cdot \nu_{\Gamma}
detB=ν⋅νΓ。
如果
ν
=
ν
Γ
\nu=\nu_{\Gamma}
ν=νΓ,结论显然。故我们假设二者是相互独立的。能展成一个空间
X
=
span
{
ν
,
ν
Γ
}
\mathbb{X}=\operatorname{span}\left\{\boldsymbol{\nu},\boldsymbol{\nu}_{\Gamma}\right\}
X=span{ν,νΓ},有一对正交基
ν
\nu
ν和
e
e
e。
我们考虑旋转算子
R
∈
R
(
n
+
1
)
×
(
n
+
1
)
\mathbf{R} \in \mathbb{R}^{(n+1) \times(n+1)}
R∈R(n+1)×(n+1),使得:
R
ν
=
ν
Γ
=
cos
θ
ν
+
sin
θ
e
,
R
e
=
−
sin
θ
ν
+
cos
θ
e
\mathbf{R} \nu=\nu_{\Gamma}=\cos \theta \nu+\sin \theta \mathbf{e}, \quad \mathbf{R e}=-\sin \theta \nu+\cos \theta \mathbf{e}
Rν=νΓ=cosθν+sinθe,Re=−sinθν+cosθe
那么,我们最终有:
B
=
(
ν
⊗
ν
+
Π
R
Π
)
R
t
⇒
det
B
=
det
(
ν
⊗
ν
+
Π
R
Π
)
\mathbf{B}=(\nu \otimes \nu+\Pi \mathbf{R} \Pi) \mathbf{R}^{t} \Rightarrow \operatorname{det} \mathbf{B}=\operatorname{det}(\nu \otimes \nu+\Pi \mathbf{R} \Pi)
B=(ν⊗ν+ΠRΠ)Rt⇒detB=det(ν⊗ν+ΠRΠ)
进一步,因为
cos
θ
=
ν
⋅
ν
Γ
\cos \theta=\nu \cdot \nu_{\Gamma}
cosθ=ν⋅νΓ,且
B
\mathbf{B}
B对应的特征值为1(特征向量
ν
\nu
ν)、
c
o
s
(
θ
)
cos(\theta)
cos(θ)(特征向量
e
\mathbf{e}
e)和1(特征向量在垂空间中)。证毕。
我们定义一些几何量:
(
54
)
d
∞
:
=
∥
d
∥
L
∞
(
Γ
)
,
ν
∞
:
=
∥
ν
−
ν
Γ
∥
L
∞
(
Γ
)
,
K
∞
:
=
∥
K
∥
L
∞
(
γ
)
(54) \quad d_{\infty}:=\|d\|_{L_{\infty}(\Gamma)}, \quad \nu_{\infty}:=\left\|\nu-\nu_{\Gamma}\right\|_{L_{\infty}(\Gamma)}, \quad K_{\infty}:=\|K\|_{L_{\infty}(\gamma)}
(54)d∞:=∥d∥L∞(Γ),ν∞:=∥ν−νΓ∥L∞(Γ),K∞:=∥K∥L∞(γ)
我们有如下引理:
Lemma
23
(perturbation error estimate for
C
2
surfaces). Let
u
solve (Ig) and
u
Γ
solve
(
33
)
with
Γ
⊂
N
. Let
χ
Γ
and
χ
:
=
P
d
∘
χ
Γ
be the parametrizations of
Γ
and
γ
that give rise to the area elements
q
Γ
and
q
.
If the normal vectors satisfy
ν
⋅
ν
Γ
≥
c
>
0
,
then
(55)
∥
∇
γ
(
u
−
u
Γ
)
∥
L
2
(
γ
)
≲
(
d
∞
K
∞
+
ν
∞
2
)
∥
f
Γ
∥
H
#
−
1
(
Γ
)
+
∥
f
q
q
Γ
−
1
−
f
Γ
∥
H
#
−
1
(
Γ
)
\begin{array}{l}{\text { Lemma } 23 \text { (perturbation error estimate for } C^{2} \text { surfaces). Let } u \text { solve (Ig) and }} \\ {u_{\Gamma} \text { solve }(33) \text { with } \Gamma \subset \mathcal{N} \text { . Let } \chi_{\Gamma} \text { and } \chi:=\mathbf{P}_{d} \circ \chi_{\Gamma} \text { be the parametrizations of }} \\ {\Gamma \text { and } \gamma \text { that give rise to the area elements } q_{\Gamma} \text { and } q . \text { If the normal vectors satisfy }} \\ {\nu \cdot \nu_{\Gamma} \geq c>0, \text { then }} \\ {\begin{array}{llll}{\text { (55) }} & {\left\|\nabla_{\gamma}\left(u-u_{\Gamma}\right)\right\|_{L_{2}(\gamma)}} & {\lesssim\left(d_{\infty} K_{\infty}+\nu_{\infty}^{2}\right)\left\|f_{\Gamma}\right\|_{H_{\#}^{-1}(\Gamma)}+\left\|f q q_{\Gamma}^{-1}-f_{\Gamma}\right\|_{H_{\#}^{-1}(\Gamma)}}\end{array}}\end{array}
Lemma 23 (perturbation error estimate for C2 surfaces). Let u solve (Ig) and uΓ solve (33) with Γ⊂N . Let χΓ and χ:=Pd∘χΓ be the parametrizations of Γ and γ that give rise to the area elements qΓ and q. If the normal vectors satisfy ν⋅νΓ≥c>0, then (55) ∥∇γ(u−uΓ)∥L2(γ)≲(d∞K∞+ν∞2)∥fΓ∥H#−1(Γ)+∥∥fqqΓ−1−fΓ∥∥H#−1(Γ)
证明:
和引理19相似地证明,Step 1和Step 3是完全一样的。只要估计几何误差矩阵。如果我们证明了:
(56)
∥
E
∥
L
∞
(
γ
)
≲
ν
∞
2
+
d
∞
K
∞
\text { (56) } \quad\|\mathbf{E}\|_{L_{\infty}(\gamma)} \lesssim \nu_{\infty}^{2}+d_{\infty} K_{\infty}
(56) ∥E∥L∞(γ)≲ν∞2+d∞K∞
结论立明。我们首先写
E
∘
P
d
=
I
1
+
I
2
+
I
3
\mathbf{E} \circ \mathbf{P}_{d}=\mathbf{I}_{1}+\mathbf{I}_{2}+\mathbf{I}_{3}
E∘Pd=I1+I2+I3,其中:
I
1
:
=
(
q
Γ
q
−
1
)
Π
(
I
−
d
W
)
Π
Γ
(
I
−
d
W
)
Π
I
2
:
=
(
Π
(
I
−
d
W
)
Π
Γ
(
I
−
d
W
)
Π
−
Π
Π
Γ
Π
)
I
3
:
=
(
Π
Π
Γ
Π
−
Π
)
\begin{array}{l}{\mathbf{I}_{1}:=\left(\frac{q_{\Gamma}}{q}-1\right) \Pi(\mathbf{I}-d \mathbf{W}) \Pi_{\Gamma}(\mathbf{I}-d \mathbf{W}) \Pi} \\ {\mathbf{I}_{2}:=\left(\Pi(\mathbf{I}-d \mathbf{W}) \Pi_{\Gamma}(\mathbf{I}-d \mathbf{W}) \Pi-\Pi \Pi_{\Gamma} \Pi\right)} \\ {\mathbf{I}_{3}:=\left(\Pi \Pi_{\Gamma} \Pi-\Pi\right)}\end{array}
I1:=(qqΓ−1)Π(I−dW)ΠΓ(I−dW)ΠI2:=(Π(I−dW)ΠΓ(I−dW)Π−ΠΠΓΠ)I3:=(ΠΠΓΠ−Π)
由面积元之间的关系,我们有:
q
(
y
)
q
Γ
(
y
)
−
1
=
(
(
ν
(
x
)
⋅
ν
Γ
(
x
)
−
1
)
∏
i
=
1
n
(
1
−
d
(
x
)
κ
i
(
x
)
)
)
+
(
∏
i
=
1
n
(
1
−
d
(
x
)
κ
i
(
x
)
)
−
1
)
\frac{q(\mathbf{y})}{q_{\Gamma}(\mathbf{y})}-1=\left(\left(\nu(\mathbf{x}) \cdot \nu_{\Gamma}(\mathbf{x})-1\right) \prod_{i=1}^{n}\left(1-d(\mathbf{x}) \kappa_{i}(\mathbf{x})\right)\right)+\left(\prod_{i=1}^{n}\left(1-d(\mathbf{x}) \kappa_{i}(\mathbf{x})\right)-1\right)
qΓ(y)q(y)−1=((ν(x)⋅νΓ(x)−1)i=1∏n(1−d(x)κi(x)))+(i=1∏n(1−d(x)κi(x))−1)
既然
1
−
ν
⋅
ν
Γ
=
1
2
∣
ν
−
ν
Γ
∣
2
≤
1
2
ν
∞
2
1-\nu \cdot \nu_{\Gamma}=\frac{1}{2}\left|\nu-\nu_{\Gamma}\right|^{2} \leq \frac{1}{2} \nu_{\infty}^{2}
1−ν⋅νΓ=21∣ν−νΓ∣2≤21ν∞2,我们可到:
(
57
)
∣
q
(
y
)
q
Γ
(
y
)
−
1
∣
≲
ν
∞
2
+
d
∞
K
∞
∀
y
∈
V
(57) \quad\left|\frac{q(\mathbf{y})}{q_{\Gamma}(\mathbf{y})}-1\right| \lesssim \nu_{\infty}^{2}+d_{\infty} K_{\infty} \quad \forall \mathbf{y} \in \mathcal{V}
(57)∣∣∣∣qΓ(y)q(y)−1∣∣∣∣≲ν∞2+d∞K∞∀y∈V
q
Γ
q
\frac{q_{\Gamma}}{q}
qqΓ的界根据
ν
⋅
ν
Γ
≥
c
>
0
\boldsymbol{\nu} \cdot \boldsymbol{\nu}_{\Gamma} \geq c>0
ν⋅νΓ≥c>0类似可以得到。
对于第二部分:
I
2
=
−
Π
Π
Γ
d
W
Π
−
Π
d
W
Π
Γ
Π
+
Π
d
W
Π
Γ
d
W
Π
\mathbf{I}_{2}=-\Pi \Pi_{\Gamma} d \mathbf{W} \Pi-\Pi d \mathbf{W} \Pi_{\Gamma} \Pi+\Pi d \mathbf{W} \Pi_{\Gamma} d \mathbf{W} \Pi
I2=−ΠΠΓdWΠ−ΠdWΠΓΠ+ΠdWΠΓdWΠ
容易有,
∥
I
2
∥
L
∞
(
γ
)
≲
d
∞
K
∞
\left\|\mathbf{I}_{2}\right\|_{L_{\infty}(\gamma)} \lesssim d_{\infty} K_{\infty}
∥I2∥L∞(γ)≲d∞K∞。
对于第三部分,可写成:
I
3
=
−
Π
ν
Γ
⊗
Π
ν
Γ
=
−
(
ν
Γ
−
(
ν
⋅
ν
Γ
)
ν
)
⊗
(
ν
Γ
−
(
ν
⋅
ν
Γ
)
ν
)
\mathbf{I}_{3}=-\Pi \nu_{\Gamma} \otimes \Pi \nu_{\Gamma}=-\left(\nu_{\Gamma}-\left(\nu \cdot \nu_{\Gamma}\right) \nu\right) \otimes\left(\nu_{\Gamma}-\left(\nu \cdot \nu_{\Gamma}\right) \nu\right)
I3=−ΠνΓ⊗ΠνΓ=−(νΓ−(ν⋅νΓ)ν)⊗(νΓ−(ν⋅νΓ)ν)
因为
ν
Γ
−
(
ν
⋅
ν
Γ
)
ν
=
(
ν
Γ
−
ν
)
+
(
1
−
ν
⋅
ν
Γ
)
ν
\nu_{\Gamma}-\left(\nu \cdot \nu_{\Gamma}\right) \nu=\left(\nu_{\Gamma}-\nu\right)+\left(1-\nu \cdot \nu_{\Gamma}\right) \nu
νΓ−(ν⋅νΓ)ν=(νΓ−ν)+(1−ν⋅νΓ)ν,我们推断
∥
I
3
∥
L
∞
(
γ
)
≲
ν
∞
2
\left\|\mathbf{I}_{3}\right\|_{L_{\infty}(\gamma)} \lesssim \nu_{\infty}^{2}
∥I3∥L∞(γ)≲ν∞2。立证结论。
下面使用
λ
∞
\lambda_\infty
λ∞给出
v
∞
v_\infty
v∞的估计。有如下引理:
Lemma
24
(error estimate for normals). The errors
ν
∞
and
λ
∞
defined in
(
54
)
and
(
39
)
satisfy
(58)
ν
∞
≲
λ
∞
where the hidden constant depends on
S
χ
defined in
(
38
)
\begin{array}{l}{\text { Lemma } 24 \text { (error estimate for normals). The errors } \nu_{\infty} \text { and } \lambda_{\infty} \text { defined in }(54)} \\ {\text { and }(39) \text { satisfy }} \\ {\begin{array}{ll}{\text { (58) }} & {\nu_{\infty} \lesssim \lambda_{\infty}} \\ {\text { where the hidden constant depends on } S_{\chi} \text { defined in }(38)}\end{array}}\end{array}
Lemma 24 (error estimate for normals). The errors ν∞ and λ∞ defined in (54) and (39) satisfy (58) where the hidden constant depends on Sχ defined in (38)ν∞≲λ∞
证明:
由定义,我们有:
ν
−
ν
Γ
=
N
−
N
Γ
∣
N
∣
+
∣
N
Γ
∣
−
∣
N
∣
∣
N
∣
N
Γ
∣
N
Γ
∣
⇒
∣
ν
−
ν
Γ
∣
≤
2
∣
N
−
N
Γ
∣
∣
N
∣
\nu-\nu_{\Gamma}=\frac{N-N_{\Gamma}}{|N|}+\frac{\left|N_{\Gamma}\right|-|N|}{|N|} \frac{N_{\Gamma}}{\left|N_{\Gamma}\right|} \Rightarrow\left|\nu-\nu_{\Gamma}\right| \leq 2 \frac{\left|N-N_{\Gamma}\right|}{|N|}
ν−νΓ=∣N∣N−NΓ+∣N∣∣NΓ∣−∣N∣∣NΓ∣NΓ⇒∣ν−νΓ∣≤2∣N∣∣N−NΓ∣
因为
N
=
∑
i
=
1
n
+
1
det
(
[
e
i
,
D
χ
]
)
e
i
\mathbf{N}=\sum_{i=1}^{n+1} \operatorname{det}\left(\left[\mathbf{e}_{i}, D \chi\right]\right) \mathbf{e}_{i}
N=∑i=1n+1det([ei,Dχ])ei且
det
(
[
e
i
,
D
X
]
)
−
det
(
[
e
i
,
D
X
Γ
]
)
\operatorname{det}\left(\left[\mathbf{e}_{i}, D_{\mathcal{X}}\right]\right)-\operatorname{det}\left(\left[\mathbf{e}_{i}, D_{\mathcal{X}_{\Gamma}}\right]\right)
det([ei,DX])−det([ei,DXΓ])是众多
∂
j
(
χ
−
χ
Γ
)
⋅
e
k
\partial_{j}\left(\chi-\chi_{\Gamma}\right) \cdot \mathbf{e}_{k}
∂j(χ−χΓ)⋅ek和n-1个因子
∂
ℓ
χ
m
\partial \ell \chi_{m}
∂ℓχm乘积的西格玛和,我们有:
∣
det
(
[
e
i
,
D
χ
]
)
−
det
(
[
e
i
,
D
χ
Γ
]
)
∣
≲
∣
D
(
χ
−
χ
Γ
)
∣
∣
D
χ
∣
n
−
1
\left|\operatorname{det}\left(\left[\mathbf{e}_{i}, D \chi\right]\right)-\operatorname{det}\left(\left[\mathbf{e}_{i}, D \chi_{\Gamma}\right]\right)\right| \lesssim\left|D\left(\chi-\chi_{\Gamma}\right)\right||D \chi|^{n-1}
∣det([ei,Dχ])−det([ei,DχΓ])∣≲∣D(χ−χΓ)∣∣Dχ∣n−1
再诉诸于
∣
N
∣
=
q
|\mathbf{N}|=q
∣N∣=q以及引理16证明中得到的
q
≈
∣
D
χ
∣
n
q \approx|D \chi|^{n}
q≈∣Dχ∣n,立得结论。