曲面上的微积分(一)
参数化方法、迹方法和窄带方法是三种主要方法去求解Laplace-Beltrami问题。首先,我会介绍一些关于曲面上微积分的知识。
参数化曲面
定义曲面的参数化表示,
χ
i
:
V
i
→
U
i
∩
γ
⊂
R
n
+
1
\chi_{i} : \mathcal{V}_{i} \rightarrow \mathcal{U}_{i} \cap \gamma \subset \mathbb{R}^{n+1}
χi:Vi→Ui∩γ⊂Rn+1
这里的
γ
\gamma
γ是紧的,有向的(orientable),属于
C
1
,
α
C^{1,\alpha}
C1,α的。考虑一个单映射是足够的,所以为了方便,我们下面drop掉index i。那么,对于
x
∈
U
∩
γ
\mathbf{x} \in \mathcal U \cap \gamma
x∈U∩γ,
y
:
=
χ
−
1
(
x
)
∈
V
\mathbf{y} :=\chi^{-1}(\mathbf{x}) \in \mathcal{V}
y:=χ−1(x)∈V
那么,
χ
\chi
χ的微分就可以写为,
D
χ
(
y
)
=
(
∂
j
χ
(
y
)
)
j
=
1
n
∈
R
(
n
+
1
)
×
n
D \chi(\mathbf{y})=\left(\partial_{j} \chi(\mathbf{y})\right)_{j=1}^{n} \in \mathbb{R}^{(n+1) \times n}
Dχ(y)=(∂jχ(y))j=1n∈R(n+1)×n
故而,我们定义一个s.p.d的矩阵作为第一基本型,
(1)
g
(
y
)
:
=
D
χ
(
y
)
t
D
χ
(
y
)
∀
y
∈
V
\text { (1) } \quad \mathbf{g}(\mathbf{y}) :=D \chi(\mathbf{y})^{t} D \chi(\mathbf{y}) \quad \forall \mathbf{y} \in \mathcal{V}
(1) g(y):=Dχ(y)tDχ(y)∀y∈V
定义法向量,
N
(
y
)
=
∑
j
=
1
n
+
1
A
j
(
y
)
e
j
\mathbf{N}(\mathbf{y})=\sum_{j=1}^{n+1} A_{j}(\mathbf{y}) \mathbf{e}_{j}
N(y)=j=1∑n+1Aj(y)ej
A
j
:
=
det
(
e
j
,
D
X
)
A_{j} :=\operatorname{det}\left(\mathbf{e}_{j}, D_{\mathcal{X}}\right)
Aj:=det(ej,DX)
容易知道,
N
⋅
∂
i
χ
=
∑
j
=
1
n
+
1
e
j
⋅
∂
i
χ
det
(
e
j
,
D
X
)
=
det
(
∑
j
=
1
n
+
1
(
e
j
⋅
∂
i
χ
)
e
j
,
D
X
)
=
det
(
∂
i
χ
,
D
X
)
=
0
\mathbf{N} \cdot \partial_{i} \chi=\sum_{j=1}^{n+1} \mathbf{e}_{j} \cdot \partial_{i} \chi \operatorname{det}\left(\mathbf{e}_{j}, D_{\mathcal{X}}\right)=\operatorname{det}\left(\sum_{j=1}^{n+1}\left(\mathbf{e}_{j} \cdot \partial_{i} \chi\right) \mathbf{e}_{j}, D_{\mathcal{X}}\right)=\operatorname{det}\left(\partial_{i} \chi, D_{\mathcal{X}}\right)=0
N⋅∂iχ=j=1∑n+1ej⋅∂iχdet(ej,DX)=det(j=1∑n+1(ej⋅∂iχ)ej,DX)=det(∂iχ,DX)=0
所以,我们可以定义单位法向量,
(2)
ν
(
y
)
:
=
N
(
y
)
∣
N
(
y
)
∣
∀
y
∈
V
\text { (2) } \quad \nu(\mathbf{y}) :=\frac{\mathbf{N}(\mathbf{y})}{|\mathbf{N}(\mathbf{y})|} \quad \forall \mathbf{y} \in \mathcal{V}
(2) ν(y):=∣N(y)∣N(y)∀y∈V
定义矩阵
T
(
y
)
:
=
[
D
χ
(
y
)
,
ν
(
y
)
]
∈
R
(
n
+
1
)
×
(
n
+
1
)
∀
y
∈
V
\mathbf{T}(\mathbf{y}) :=[D \chi(\mathbf{y}), \boldsymbol{\nu}(\mathbf{y})] \in \mathbb{R}^{(n+1) \times(n+1)} \quad \forall \mathbf{y} \in \mathcal{V}
T(y):=[Dχ(y),ν(y)]∈R(n+1)×(n+1)∀y∈V
它是可逆的,我们可以把它的逆写成,
T
−
1
=
[
B
v
t
]
,
B
∈
R
n
×
(
n
+
1
)
,
v
∈
R
n
\mathbf{T}^{-1}=\left[\begin{array}{l}{\mathbf{B}} \\ {\mathbf{v}^{t}}\end{array}\right], \quad \mathbf{B} \in \mathbb{R}^{n \times(n+1)}, \mathbf{v} \in \mathbb{R}^{n}
T−1=[Bvt],B∈Rn×(n+1),v∈Rn
注意到,
I
(
n
+
1
)
×
(
n
+
1
)
=
T
−
1
T
=
[
B
D
χ
B
v
v
t
D
χ
v
t
ν
]
\mathbf{I}_{(n+1) \times(n+1)}=\mathbf{T}^{-1} \mathbf{T}=\left[\begin{array}{cc}{\mathbf{B} D \chi} & {\mathbf{B} \mathbf{v}} \\ {\mathbf{v}^{t} D \chi} & {\mathbf{v}^{t} \nu}\end{array}\right]
I(n+1)×(n+1)=T−1T=[BDχvtDχBvvtν]
因此呢,
B
D
χ
=
I
n
×
n
,
v
t
D
χ
=
0
,
v
t
ν
=
1
\mathbf{B} D \chi=\mathbf{I}_{n \times n}, \quad \mathbf{v}^{t} D \chi=0, \quad \mathbf{v}^{t} \nu=1
BDχ=In×n,vtDχ=0,vtν=1
从后面两个式子中,我们知道
v
=
ν
\mathbf{v}=\boldsymbol\nu
v=ν。又注意到,
I
(
n
+
1
)
×
(
n
+
1
)
=
T
T
−
1
=
D
χ
B
+
ν
ν
t
\mathbf{I}_{(n+1) \times(n+1)}=\mathbf{T} \mathbf{T}^{-1}=D \chi \mathbf{B}+\nu \boldsymbol{\nu}^{t}
I(n+1)×(n+1)=TT−1=DχB+ννt
我们可以定义投影矩阵
Π
∈
R
(
n
+
1
)
×
(
n
+
1
)
\Pi \in \mathbb{R}^{(n+1) \times(n+1)}
Π∈R(n+1)×(n+1)为
(3)
Π
:
=
I
−
ν
⊗
ν
=
D
χ
B
\text { (3) } \quad \Pi :=\mathbf{I}-\boldsymbol{\nu} \otimes \boldsymbol{\nu}=D \chi \mathbf{B}
(3) Π:=I−ν⊗ν=DχB
容易得到
B
\mathbf{B}
B的显式表达,通过以下的方式,
D
χ
=
(
I
−
ν
⊗
ν
)
t
D
χ
=
B
t
D
χ
t
D
χ
=
B
t
g
⇒
B
=
g
−
1
D
χ
t
D \chi=(\mathbf{I}-\nu \otimes \nu)^{t} D \chi=\mathbf{B}^{t} D \chi^{t} D \chi=\mathbf{B}^{t} \mathbf{g} \Rightarrow \mathbf{B}=\mathbf{g}^{-1} D \chi^{t}
Dχ=(I−ν⊗ν)tDχ=BtDχtDχ=Btg⇒B=g−1Dχt
那么投影矩阵就可以写为,
(4)
Π
=
D
χ
g
−
1
D
χ
t
\text { (4) } \quad \quad \Pi=D \chi \mathrm{g}^{-1} D \chi^{t}
(4) Π=Dχg−1Dχt
定义面积元
(5)
q
(
y
)
:
=
det
(
[
ν
(
y
)
,
D
χ
(
y
)
]
)
∀
y
∈
V
\text { (5) } \quad \quad q(\mathbf{y}) :=\operatorname{det}([\nu(\mathbf{y}), D \chi(\mathbf{y})]) \quad \forall \mathbf{y} \in \mathcal{V}
(5) q(y):=det([ν(y),Dχ(y)])∀y∈V
容易知道,
(6)
q
=
1
∣
N
∣
det
(
[
N
,
D
χ
]
)
=
1
∣
N
∣
det
(
[
N
,
D
χ
]
t
[
N
,
D
χ
]
)
1
2
=
det
g
\text { (6) } \quad q=\frac{1}{|N|} \operatorname{det}([\mathrm{N}, D \chi])=\frac{1}{|N|} \operatorname{det}\left([\mathrm{N}, D \chi]^{t}[\mathrm{N}, D \chi]\right)^{\frac{1}{2}}=\sqrt{\operatorname{det} \mathrm{g}}
(6) q=∣N∣1det([N,Dχ])=∣N∣1det([N,Dχ]t[N,Dχ])21=detg
进一步,因为
∣
N
∣
2
=
∑
j
=
1
n
+
1
A
j
det
(
[
e
j
,
D
X
]
)
=
det
(
[
N
,
D
X
]
)
|N|^{2}=\sum_{j=1}^{n+1} A_{j} \operatorname{det}\left(\left[\mathbf{e}_{j}, D_{\mathcal{X}}\right]\right)=\operatorname{det}\left(\left[\mathbf{N}, D_{\mathcal{X}}\right]\right)
∣N∣2=j=1∑n+1Ajdet([ej,DX])=det([N,DX])
联合式(6)我们可以得到,
(
7
)
q
=
∣
N
∣
(7) \quad q=|\mathbf{N}|
(7)q=∣N∣
考虑一个可积函数
v
~
:
γ
→
R
\widetilde{v} : \gamma \rightarrow \mathbb{R}
v
:γ→R
及其复合
v
:
V
→
R
v : \mathcal{V} \rightarrow \mathbb{R}
v:V→R
v
=
v
~
∘
χ
v=\widetilde{v} \circ \chi
v=v
∘χ
我们有积分关系如下,
(8)
∫
γ
v
~
=
∫
V
v
q
∀
v
∈
L
1
(
V
)
\text { (8) } \quad \int_{\gamma} \tilde{v}=\int_{\mathcal{V}} v q \quad \forall v \in L_{1}(\mathcal{V})
(8) ∫γv~=∫Vvq∀v∈L1(V)
这个定义不依赖与曲面的参数化(parametrization),事实上,
χ
1
=
χ
2
∘
(
χ
2
−
1
∘
χ
1
)
\chi_{1}=\chi_{2} \circ\left(\chi_{2}^{-1} \circ \chi_{1}\right)
χ1=χ2∘(χ2−1∘χ1)
D
χ
1
=
D
χ
2
D
(
χ
2
−
1
∘
χ
1
)
D \chi_{1}=D \chi_{2} D\left(\chi_{2}^{-1} \circ \chi_{1}\right)
Dχ1=Dχ2D(χ2−1∘χ1)
q
1
=
∣
det
(
D
(
χ
2
−
1
∘
χ
1
)
)
∣
q
2
⇒
∫
V
1
v
q
1
=
∫
V
2
v
q
2
q_{1}=\left|\operatorname{det}\left(D\left(\chi_{2}^{-1} \circ \chi_{1}\right)\right)\right| q_{2} \quad \Rightarrow \quad \int_{\mathcal{V}_{1}} v q_{1}=\int_{\mathcal{V}_{2}} v q_{2}
q1=∣∣det(D(χ2−1∘χ1))∣∣q2⇒∫V1vq1=∫V2vq2
这里的复合函数求导是不是有些问题?
微分算子
我们定义曲面上的切向(或表面)梯度(tangential gradient),它满足链式法则
(9)
∇
v
(
y
)
=
D
χ
(
y
)
t
∇
γ
v
~
(
x
)
∀
y
∈
V
\text { (9) } \quad \nabla v(\mathbf{y})=D \chi(\mathbf{y})^{t} \nabla_{\gamma} \widetilde{v}(\mathbf{x}) \quad \forall \mathbf{y} \in \mathcal{V}
(9) ∇v(y)=Dχ(y)t∇γv
(x)∀y∈V
那么,我们有
(
10
)
∇
γ
v
~
(
x
)
=
D
χ
(
y
)
g
(
y
)
−
1
∇
v
(
y
)
∀
y
∈
V
(10) \quad \nabla_{\gamma} \widetilde{v}(\mathrm{x})=D \chi(\mathbf{y}) \mathbf{g}(\mathbf{y})^{-1} \nabla v(\mathbf{y}) \quad \forall \mathbf{y} \in \mathcal{V}
(10)∇γv
(x)=Dχ(y)g(y)−1∇v(y)∀y∈V
如果
γ
\gamma
γ是
C
2
C^2
C2的,那么单位法向量
ν
\boldsymbol{\nu}
ν是
C
1
C^1
C1的,这时,我们可以定义Weingarten映射(shape map),如下,
(11)
W
(
x
)
=
D
γ
ν
(
x
)
∀
x
∈
γ
\begin{array}{ll}{\text { (11) }} & {\mathbf{W}(\mathbf{x})=D_{\gamma} \boldsymbol{\nu}(\mathbf{x}) \quad \forall \mathbf{x} \in \gamma}\end{array}
(11) W(x)=Dγν(x)∀x∈γ
如果
v
~
=
(
v
~
i
)
i
=
1
n
+
1
:
γ
→
R
n
+
1
\widetilde{\mathbf{v}}=\left(\widetilde{v}_{i}\right)_{i=1}^{n+1} : \gamma \rightarrow \mathbb{R}^{n+1}
v
=(v
i)i=1n+1:γ→Rn+1
那么由(10)我们可以定义其散度,
(
12
)
div
γ
v
~
(
x
)
=
trace
(
D
γ
v
~
(
x
)
)
=
∑
i
,
j
=
1
n
g
i
j
(
y
)
∂
i
χ
(
y
)
⋅
∂
j
v
(
y
)
∀
y
∈
V
(12) \quad \operatorname{div}_{\gamma} \widetilde{\mathbf{v}}(\mathbf{x})=\operatorname{trace}\left(D_{\gamma} \widetilde{\mathbf{v}}(\mathbf{x})\right)=\sum_{i, j=1}^{n} g^{i j}(\mathbf{y}) \partial_{i} \chi(\mathbf{y}) \cdot \partial_{j} \mathbf{v}(\mathbf{y}) \quad \forall \mathbf{y} \in \mathcal{V}
(12)divγv
(x)=trace(Dγv
(x))=i,j=1∑ngij(y)∂iχ(y)⋅∂jv(y)∀y∈V
这里,
g
−
1
=
(
g
i
j
)
i
,
j
=
1
n
\mathbf{g}^{-1}=\left(g^{i j}\right)_{i, j=1}^{n}
g−1=(gij)i,j=1n。那么,这个时候Laplace-Beltrami算子就可以定义为
(
13
)
Δ
γ
v
~
=
1
q
(
y
)
div
(
q
(
y
)
g
(
y
)
−
1
∇
v
(
y
)
)
∀
y
∈
V
(13) \quad \Delta_{\gamma} \widetilde{v}=\frac{1}{q(\mathbf{y})} \operatorname{div}\left(q(\mathbf{y}) \mathbf{g}(\mathbf{y})^{-1} \nabla v(\mathbf{y})\right) \quad \forall \mathbf{y} \in \mathcal{V}
(13)Δγv
=q(y)1div(q(y)g(y)−1∇v(y))∀y∈V
通过这样一个定义,我们可以得到Laplace-Beltrami算子的弱形式,即,弱
ϕ
~
\tilde \phi
ϕ~是
C
1
C^1
C1的,在
γ
\gamma
γ上有紧支集,那么
(
14
)
∫
γ
φ
~
Δ
γ
v
~
=
−
∫
γ
∇
γ
φ
~
⋅
∇
γ
v
~
(14) \quad \int_{\gamma} \widetilde{\varphi} \Delta_{\gamma} \widetilde{v}=-\int_{\gamma} \nabla_{\gamma} \widetilde{\varphi} \cdot \nabla_{\gamma} \widetilde{v}
(14)∫γφ
Δγv
=−∫γ∇γφ
⋅∇γv
证明如下,
∫
γ
φ
~
Δ
γ
v
~
=
∫
V
φ
div
(
q
g
−
1
∇
v
)
=
−
∫
V
∇
φ
⋅
g
−
1
∇
v
q
=
−
∫
V
D
χ
g
−
1
∇
φ
⋅
D
χ
g
−
1
∇
v
q
=
−
∫
γ
∇
γ
φ
~
⋅
∇
γ
v
~
\begin{aligned} \int_{\gamma} \tilde{\varphi} \Delta_{\gamma} \tilde{v} &=\int_{V} \varphi \operatorname{div}\left(q \mathrm{g}^{-1} \nabla v\right) \\ &=-\int_{V} \nabla \varphi \cdot \mathrm{g}^{-1} \nabla v q \\ &=-\int_{V} D \chi \mathrm{g}^{-1} \nabla \varphi \cdot D \chi \mathrm{g}^{-1} \nabla v q \\ &=-\int_{\gamma} \nabla_{\gamma} \tilde{\varphi} \cdot \nabla_{\gamma} \tilde{v} \end{aligned}
∫γφ~Δγv~=∫Vφdiv(qg−1∇v)=−∫V∇φ⋅g−1∇vq=−∫VDχg−1∇φ⋅Dχg−1∇vq=−∫γ∇γφ~⋅∇γv~
定义零平均值的平方可积空间,
L
2
,
#
(
γ
)
:
=
{
v
~
∈
L
2
(
γ
)
∣
∫
γ
v
~
=
0
}
L_{2, \#}(\gamma) :=\left\{\widetilde{v} \in L_{2}(\gamma) | \int_{\gamma} \widetilde{v}=0\right\}
L2,#(γ):={v
∈L2(γ)∣∫γv
=0}
并定义零平均值的索伯洛夫空间,
H
#
1
(
γ
)
:
=
H
1
(
γ
)
∩
L
2
,
#
(
γ
)
,
H
1
(
γ
)
:
=
{
v
~
∈
L
2
(
γ
)
∣
∇
(
v
~
∘
χ
)
∈
[
L
2
(
V
)
]
n
}
H_{\#}^{1}(\gamma) :=H^{1}(\gamma) \cap L_{2, \#}(\gamma), \quad H^{1}(\gamma) :=\left\{\tilde{v} \in L_{2}(\gamma) | \nabla(\tilde{v} \circ \chi) \in\left[L_{2}(\mathcal{V})\right]^{n}\right\}
H#1(γ):=H1(γ)∩L2,#(γ),H1(γ):={v~∈L2(γ)∣∇(v~∘χ)∈[L2(V)]n}
那么,我们有庞加莱不等式(Poincare-Friedrichs)
(15)
∥
v
‾
∥
L
2
(
γ
)
≤
C
∥
∇
γ
v
~
∥
L
2
(
γ
)
∀
v
~
∈
H
#
1
(
γ
)
\begin{array}{llll}{\text { (15) }} & {\|\overline{v}\|_{L_{2}(\gamma)} \leq C\left\|\nabla_{\gamma} \widetilde{v}\right\|_{L_{2}(\gamma)}} & {\forall \widetilde{v} \in H_{\#}^{1}(\gamma)}\end{array}
(15) ∥v∥L2(γ)≤C∥∇γv
∥L2(γ)∀v
∈H#1(γ)
这里的
γ
\gamma
γ是紧的,利普希茨的,常数
C
C
C仅依赖于
γ
\gamma
γ。我们证明一个更一般的情况,
(16) ∥ v ~ ∥ L 2 ( γ ) ≤ C ( ∥ ∇ γ v ~ ∥ L 2 ( γ ) + ∣ ∫ γ v ~ ∣ ) ∀ v ~ ∈ H 1 ( γ ) \begin{array}{lll}{\text { (16) }} & {\|\tilde{v}\|_{L_{2}(\gamma)} \leq C\left(\left\|\nabla_{\gamma} \widetilde{v}\right\|_{L_{2}(\gamma)}+\left|\int_{\gamma} \tilde{v}\right|\right)} & {\forall \widetilde{v} \in H^{1}(\gamma)}\end{array} (16) ∥v~∥L2(γ)≤C(∥∇γv ∥L2(γ)+∣∣∣∫γv~∣∣∣)∀v ∈H1(γ)
用反证法(proof by contradiction)证明如下:
假设存在
v
~
k
∈
H
1
(
γ
)
\widetilde{v}_{k} \in H^{1}(\gamma)
v
k∈H1(γ),使得
∥
v
~
k
∥
L
2
(
γ
)
=
1
,
∥
∇
γ
v
~
k
∥
L
2
(
γ
)
+
∣
∫
γ
v
~
k
∣
→
0
\left\|\widetilde{v}_{k}\right\|_{L_{2}(\gamma)}=1, \quad\left\|\nabla_{\gamma} \widetilde{v}_{k}\right\|_{L_{2}(\gamma)}+\left|\int_{\gamma} \widetilde{v}_{k}\right| \rightarrow 0
∥v
k∥L2(γ)=1,∥∇γv
k∥L2(γ)+∣∣∣∣∫γv
k∣∣∣∣→0
因为
H
1
(
γ
)
⊂
L
2
(
γ
)
H^{1}(\gamma) \subset L_{2}(\gamma)
H1(γ)⊂L2(γ)是紧的,且
{
v
~
k
}
k
\left\{\widetilde{v}_{k}\right\}_{k}
{v
k}k在
H
1
(
γ
)
H^{1}(\gamma)
H1(γ)有界,那么存在一个子列
{
v
~
k
}
k
\left\{\widetilde{v}_{k}\right\}_{k}
{v
k}k在
L
2
L^2
L2中收敛,又因为
∣
∫
γ
v
~
k
∣
→
0
\left|\int_{\gamma} \widetilde{v}_{k}\right| \rightarrow 0
∣∣∣∣∫γv
k∣∣∣∣→0
所以子列收敛于0,这和
∥
v
~
k
∥
L
2
(
γ
)
=
1
\left\|\tilde{v}_{k}\right\|_{L_{2}(\gamma)}=1
∥v~k∥L2(γ)=1
矛盾。
定义对偶空间的范数,
(
17
)
∥
f
~
∥
H
#
−
1
(
γ
)
=
sup
v
~
∈
H
#
1
(
γ
)
∫
γ
f
~
v
‾
∥
∇
γ
v
~
∥
L
2
(
γ
)
(17) \quad\|\widetilde{f}\|_{H_{\#}^{-1}(\gamma)}=\sup _{\tilde{v} \in H_{\#}^{1}(\gamma)} \frac{\int_{\gamma} \widetilde{f} \overline{v}}{\left\|\nabla_{\gamma} \widetilde{v}\right\|_{L_{2}(\gamma)}}
(17)∥f
∥H#−1(γ)=v~∈H#1(γ)sup∥∇γv
∥L2(γ)∫γf
v
方程的弱形式,可以是,对于
f
~
∈
L
2
,
#
(
γ
)
\widetilde{f} \in L_{2, \#}(\gamma)
f
∈L2,#(γ),寻找
u
~
∈
H
#
1
(
γ
)
\widetilde{u} \in H_{\#}^{1}(\gamma)
u
∈H#1(γ),使得
(
18
)
∫
γ
∇
γ
u
~
⋅
∇
γ
v
~
=
∫
γ
f
~
v
~
∀
v
~
∈
H
#
1
(
γ
)
(18) \quad \int_{\gamma} \nabla_{\gamma} \widetilde{u} \cdot \nabla_{\gamma} \widetilde{v}=\int_{\gamma} \widetilde{f} \tilde{v} \quad \forall \widetilde{v} \in H_{\#}^{1}(\gamma)
(18)∫γ∇γu
⋅∇γv
=∫γf
v~∀v
∈H#1(γ)
感谢性质
f
~
∈
L
2
,
#
(
γ
)
\widetilde{f} \in L_{2, \#}(\gamma)
f
∈L2,#(γ),解满足
(
19
)
∫
γ
∇
γ
u
~
⋅
∇
γ
v
~
=
∫
γ
f
~
v
~
∀
v
~
∈
H
1
(
γ
)
(19) \quad \int_{\gamma} \nabla_{\gamma} \widetilde{u} \cdot \nabla_{\gamma} \widetilde{v}=\int_{\gamma} \widetilde{f} \widetilde{v} \quad \forall \widetilde{v} \in H^{1}(\gamma)
(19)∫γ∇γu
⋅∇γv
=∫γf
v
∀v
∈H1(γ)
怎么理解?
下面我们介绍曲面的正则性(regularity),首先,我们先把结果展示在下面:
1、(regularity). If
γ
\gamma
γ is of class
C
2
C^2
C2, then there is a constant
C
C
C only depending on
γ
\gamma
γ such that
(
20
)
∥
u
~
∥
H
2
(
γ
)
≤
C
∥
f
~
∥
L
2
(
γ
)
(20) \quad \quad\|\widetilde{u}\|_{H^{2}(\gamma)} \leq C\|\widetilde{f}\|_{L_{2}(\gamma)}
(20)∥u
∥H2(γ)≤C∥f
∥L2(γ)
2、(regularity for
W
p
2
W^2_p
Wp2 surfaces). If
γ
\gamma
γ is of class
W
p
2
W^2_p
Wp2 with
n
<
p
≤
∞
n<p\le \infty
n<p≤∞, then there is a constant
C
>
0
C>0
C>0 depending on
γ
,
p
\gamma,p
γ,p and
n
n
n such that
(
21
)
∥
u
~
∥
H
2
(
γ
)
≤
C
∥
f
~
∥
L
2
(
γ
)
(21) \quad\|\tilde{u}\|_{H^{2}(\gamma)} \leq C\|\tilde{f}\|_{L_{2}(\gamma)}
(21)∥u~∥H2(γ)≤C∥f~∥L2(γ)
为了证明曲面的正则性,我们先介绍两个来自于别的文献的一些结果。用英文表述如下:
a、Interior regularity therory
b、unique solution
c、
我们先证明第一个。
由
Δ
γ
u
~
i
=
ψ
~
i
f
~
+
2
∇
γ
u
~
⋅
∇
γ
ψ
~
i
+
u
~
Δ
γ
ψ
~
i
=
:
g
~
i
\Delta_{\gamma} \widetilde{u}_{i}=\widetilde{\psi}_{i} \widetilde{f}+2 \nabla_{\gamma} \widetilde{u} \cdot \nabla_{\gamma} \widetilde{\psi}_{i}+\widetilde{u} \Delta_{\gamma} \widetilde{\psi}_{i}=: \widetilde{g}_{i}
Δγu
i=ψ
if
+2∇γu
⋅∇γψ
i+u
Δγψ
i=:g
i
∥
∇
γ
u
~
∥
L
2
(
γ
)
≤
∥
f
~
∥
H
#
−
1
(
γ
)
\left\|\nabla_{\gamma} \widetilde{u}\right\|_{L_{2}(\gamma)} \leq\|\widetilde{f}\|_{H_{\#}^{-1}(\gamma)}
∥∇γu
∥L2(γ)≤∥f
∥H#−1(γ)
以及,庞加莱不等式,我们能得到
∥
g
~
i
∥
L
2
(
γ
)
≤
C
∥
f
~
∥
L
2
(
γ
∩
U
i
)
\left\|\widetilde{g}_{i}\right\|_{L_{2}(\gamma)} \leq C\|\tilde{f}\|_{L^{2}\left(\gamma \cap \mathcal{U}_{i}\right)}
∥g
i∥L2(γ)≤C∥f~∥L2(γ∩Ui)
由
div
(
q
i
(
y
)
g
i
(
y
)
−
1
∇
u
(
y
)
)
=
q
i
(
y
)
g
~
i
(
χ
(
y
)
)
∀
y
∈
V
i
\operatorname{div}\left(q_{i}(\mathbf{y}) \mathbf{g}_{i}(\mathbf{y})^{-1} \nabla u(\mathbf{y})\right)=q_{i}(y) \widetilde{g}_{i}(\chi(\mathbf{y})) \quad \forall \mathbf{y} \in \mathcal{V}_{i}
div(qi(y)gi(y)−1∇u(y))=qi(y)g
i(χ(y))∀y∈Vi
根据引入的定理,得到,
∥
u
i
∥
H
2
(
χ
−
1
(
W
i
)
)
≤
C
∥
g
i
∥
L
2
(
U
i
)
\left\|u_{i}\right\|_{H^{2}\left(\boldsymbol{\chi}^{-1}\left(\mathcal{W}_{i}\right)\right)} \leq C\left\|g_{i}\right\|_{L_{2}\left(\mathcal{U}_{i}\right)}
∥ui∥H2(χ−1(Wi))≤C∥gi∥L2(Ui)
最后,我们可以得到
∥
u
~
∥
H
2
(
γ
)
≤
∑
i
=
1
I
∥
u
~
i
∥
H
2
(
W
i
)
≤
C
∑
i
=
1
I
∥
g
~
i
∥
L
2
(
U
i
)
≤
C
∥
f
~
∥
L
2
(
γ
)
\|\widetilde{u}\|_{H^{2}(\gamma)} \leq \sum_{i=1}^{I}\left\|\widetilde{u}_{i}\right\|_{H^{2}\left(\mathcal{W}_{i}\right)} \leq C \sum_{i=1}^{I}\left\|\widetilde{g}_{i}\right\|_{L_{2}\left(\mathcal{U}_{i}\right)} \leq C\|\widetilde{f}\|_{L_{2}(\gamma)}
∥u
∥H2(γ)≤i=1∑I∥u
i∥H2(Wi)≤Ci=1∑I∥g
i∥L2(Ui)≤C∥f
∥L2(γ)
接下来我们证明第二个。
由
g
,
q
∈
W
p
1
(
V
)
⇒
A
:
=
q
g
−
1
∈
W
p
1
(
V
)
\mathbf{g}, q \in W_{p}^{1}(\mathcal{V}) \quad \Rightarrow \quad \mathbf{A} :=q \mathbf{g}^{-1} \in W_{p}^{1}(\mathcal{V})
g,q∈Wp1(V)⇒A:=qg−1∈Wp1(V)
有
(22)
A
:
D
2
u
=
q
g
−
div
(
A
)
⋅
∇
u
=
ℓ
∈
L
r
0
(
V
)
\text { (22) } \quad \mathbf{A} : D^{2} u=q g-\operatorname{div}(\mathbf{A}) \cdot \nabla u=\ell \in L_{r_{0}}(\mathcal{V})
(22) A:D2u=qg−div(A)⋅∇u=ℓ∈Lr0(V)
则有,
∥
u
∥
W
r
k
2
(
Z
)
≲
∥
ℓ
∥
L
r
k
(
Z
)
\|u\|_{W_{r_{k}}^{2}}(\mathcal{Z}) \lesssim\|\ell\|_{L_{r_{k}}}(\mathcal{Z})
∥u∥Wrk2(Z)≲∥ℓ∥Lrk(Z)
递推可得最小的
k
k
k。
符号距离函数
定义符号距离函数,内正外负
∣
d
(
x
)
∣
=
dist
(
x
,
γ
)
∀
x
∈
N
|d(\mathrm{x})|=\operatorname{dist}(\mathrm{x}, \gamma) \quad \forall \mathrm{x} \in \mathcal{N}
∣d(x)∣=dist(x,γ)∀x∈N
这里,
N
\mathcal{N}
N是
γ
\gamma
γ的tubular领域,对于每一个
x
∈
N
x \in \mathcal{N}
x∈N,定义最近点投影(closest point
projection)
(23)
P
d
(
x
)
=
x
−
d
(
x
)
∇
d
(
x
)
∀
x
∈
N
\text { (23) } \quad \mathbf{P}_{d}(\mathrm{x})=\mathrm{x}-d(\mathrm{x}) \nabla d(\mathrm{x}) \quad \forall \mathrm{x} \in \mathcal{N}
(23) Pd(x)=x−d(x)∇d(x)∀x∈N
一个关于梯度的性质是
(
24
)
∇
d
(
x
)
=
∇
d
(
P
d
(
x
)
)
=
∇
d
(
x
−
d
(
x
)
∇
d
(
x
)
)
∀
x
∈
N
(24) \quad \nabla d(\mathrm{x})=\nabla d\left(\mathrm{P}_{d}(\mathrm{x})\right)=\nabla d(\mathrm{x}-d(\mathrm{x}) \nabla d(\mathrm{x})) \quad \forall \mathrm{x} \in \mathcal{N}
(24)∇d(x)=∇d(Pd(x))=∇d(x−d(x)∇d(x))∀x∈N
二阶Hessian矩阵满足
(25)
D
2
d
(
x
)
∇
d
(
x
)
=
0
∀
x
∈
N
\begin{array}{ll}{\text { (25) }} & {D^{2} d(\mathrm{x}) \nabla d(\mathrm{x})=0 \quad \forall \mathrm{x} \in \mathcal{N}}\end{array}
(25) D2d(x)∇d(x)=0∀x∈N
定义
v
~
\tilde v
v~的一个自然扩充
(
26
)
v
(
x
)
=
v
~
(
P
d
(
x
)
)
=
v
~
(
x
−
d
(
x
)
∇
d
(
x
)
)
∀
x
∈
N
(26) \quad v(\mathrm{x})=\tilde{v}\left(\mathrm{P}_{d}(\mathrm{x})\right)=\tilde{v}(\mathrm{x}-d(\mathrm{x}) \nabla d(\mathrm{x})) \quad \forall \mathrm{x} \in \mathcal{N}
(26)v(x)=v~(Pd(x))=v~(x−d(x)∇d(x))∀x∈N
对上式求导并利用投影算子的定义,我们可以得到(27)
∇
v
(
x
)
=
(
I
−
∇
d
(
x
)
⊗
∇
d
(
x
)
−
d
(
x
)
D
2
d
(
x
)
)
∇
γ
v
~
(
P
d
(
x
)
)
=
(
Π
(
x
)
−
d
(
x
)
D
2
d
(
x
)
)
∇
γ
v
~
(
P
d
(
x
)
)
=
(
I
−
d
(
x
)
D
2
d
(
x
)
)
Π
(
x
)
∇
γ
v
~
(
P
d
(
x
)
)
\begin{aligned} \nabla v(\mathbf{x}) &=\left(\mathbf{I}-\nabla d(\mathbf{x}) \otimes \nabla d(\mathbf{x})-d(\mathbf{x}) D^{2} d(\mathbf{x})\right) \nabla_{\gamma} \widetilde{v}\left(\mathbf{P}_{d}(\mathbf{x})\right) \\ &=\left(\Pi(\mathbf{x})-d(\mathbf{x}) D^{2} d(\mathbf{x})\right) \nabla_{\gamma} \widetilde{v}\left(\mathbf{P}_{d}(\mathbf{x})\right) \\ &=\left(\mathbf{I}-d(\mathbf{x}) D^{2} d(\mathbf{x})\right) \Pi(\mathbf{x}) \nabla_{\gamma} \widetilde{v}\left(\mathbf{P}_{d}(\mathbf{x})\right) \end{aligned}
∇v(x)=(I−∇d(x)⊗∇d(x)−d(x)D2d(x))∇γv
(Pd(x))=(Π(x)−d(x)D2d(x))∇γv
(Pd(x))=(I−d(x)D2d(x))Π(x)∇γv
(Pd(x))
下面我们定义一个新的tangential梯度,如下
(
28
)
∇
γ
v
~
(
x
)
=
(
I
−
ν
(
x
)
⊗
ν
(
x
)
)
∇
v
(
x
)
=
Π
(
x
)
∇
v
(
x
)
∀
x
∈
γ
(28) \quad \nabla_{\gamma} \widetilde{v}(\mathbf{x})=(\mathbf{I}-\boldsymbol{\nu}(\mathbf{x}) \otimes \boldsymbol{\nu}(\mathbf{x})) \nabla v(\mathbf{x})=\Pi(\mathbf{x}) \nabla v(\mathbf{x}) \quad \forall \mathbf{x} \in \gamma
(28)∇γv
(x)=(I−ν(x)⊗ν(x))∇v(x)=Π(x)∇v(x)∀x∈γ
容易想到,这个定义和前面的定义(10)是相容的,事实上,
∇
γ
v
~
(
x
)
⋅
∂
i
χ
(
y
)
=
∇
v
(
x
)
⋅
∂
i
χ
(
y
)
=
∂
i
v
~
(
χ
(
y
)
)
\nabla_{\gamma} \widetilde{v}(\mathbf{x}) \cdot \partial_{i} \chi(\mathbf{y})=\nabla v(\mathbf{x}) \cdot \partial_{i} \chi(\mathbf{y})=\partial_{i} \widetilde{v}(\chi(\mathbf{y}))
∇γv
(x)⋅∂iχ(y)=∇v(x)⋅∂iχ(y)=∂iv
(χ(y))
这个需要细细思考和品味。
两个remark:1、切向梯度和 v ~ \tilde v v~的扩充是无关的(irrelevant)。2、切向梯度和曲面的参数化是无关的(parametrization)。
相同地,我们也可以用新方式的定义曲面上的散度和拉普拉斯算子,他们和原来的定义也是相容的,
div
γ
(
v
~
(
x
)
)
=
trace
(
∇
γ
v
~
(
x
)
)
=
div
(
v
(
x
)
)
−
ν
(
x
)
t
∇
v
(
x
)
ν
(
x
)
∀
x
∈
γ
\operatorname{div}_{\gamma}(\widetilde{\mathbf{v}}(\mathbf{x}))=\operatorname{trace}\left(\nabla_{\gamma} \widetilde{\mathbf{v}}(\mathbf{x})\right)=\operatorname{div}(\mathbf{v}(\mathbf{x}))-\boldsymbol{\nu}(\mathbf{x})^{t} \nabla \mathbf{v}(\mathbf{x}) \boldsymbol{\nu}(\mathbf{x}) \quad \forall \mathbf{x} \in \gamma
divγ(v
(x))=trace(∇γv
(x))=div(v(x))−ν(x)t∇v(x)ν(x)∀x∈γ
Δ
γ
v
~
=
trace
(
(
I
−
ν
⊗
ν
)
D
2
v
)
−
(
∇
v
⋅
ν
)
div
γ
(
ν
)
\Delta_{\gamma} \widetilde{v}=\operatorname{trace}\left((\mathbf{I}-\boldsymbol{\nu} \otimes \boldsymbol{\nu}) D^{2} v\right)-(\nabla v \cdot \boldsymbol{\nu}) \operatorname{div}_{\gamma}(\boldsymbol{\nu})
Δγv
=trace((I−ν⊗ν)D2v)−(∇v⋅ν)divγ(ν)
Δ
γ
v
~
(
x
)
=
Δ
v
(
x
)
−
ν
(
x
)
t
D
2
v
(
x
)
ν
(
x
)
−
(
∇
v
⋅
ν
)
(
x
)
div
γ
(
ν
(
x
)
)
∀
x
∈
γ
\Delta_{\gamma} \widetilde{v}(\mathbf{x})=\Delta v(\mathbf{x})-\boldsymbol{\nu}(\mathbf{x})^{t} D^{2} v(\mathbf{x}) \boldsymbol{\nu}(\mathbf{x})-(\nabla v \cdot \boldsymbol{\nu})(\mathbf{x}) \operatorname{div}_{\gamma}(\boldsymbol{\nu}(\mathbf{x})) \quad \forall \mathbf{x} \in \gamma
Δγv
(x)=Δv(x)−ν(x)tD2v(x)ν(x)−(∇v⋅ν)(x)divγ(ν(x))∀x∈γ
第二个拉普拉斯是因为
∇
γ
(
∇
ν
⋅
ν
)
⋅
ν
=
0
\nabla_{\gamma}(\nabla \boldsymbol{\nu} \cdot \boldsymbol{\nu}) \cdot \boldsymbol{\nu}=0
∇γ(∇ν⋅ν)⋅ν=0