偏微分方程 - 基础知识 变分问题 极小曲面问题

基础知识

定义1(函数范数)

设标量函数 u:RnΩRu: \mathbb{R}^n\supset \Omega\to \mathbb{R}C(Ω)C(\Omega) 表示在 Ω\Omega 上的连续标量函数构成的线性空间,对于 uC(Ω)u\in C(\Omega),定义

uC(Ω=supxΩu(x),(值域的上确界)||u||_{C(\Omega} = \sup_{x\in\Omega}|u(x)|,\quad\text{(值域的上确界)}

Ck(Ω)C^k(\Omega)表示:由 Ω\Omega 上的 kk 阶连续可微函数构成的线性空间,特别地 C(Ω):=k=1Ck(Ω)\displaystyle C^{\infty}(\Omega):=\bigcup_{k=1}^\infty C^k(\Omega).

对于 uCk(Ω)u\in C^k(\Omega),定义

uCk(Ω)=supxΩu(x)+α=1ksupxΩDαu(x)||u||_{C^k(\Omega)} = \sup_{x\in\Omega}|u(x)|+\sum_{|\alpha|=1}^k\sup_{x\in\Omega}|D^\alpha u(x)|

其中 α=(α1,α2,,α2n)\alpha = (\alpha_1,\alpha_2,\cdots, \alpha_2n) 表示多重指标,且

α=α1+α2++αn, Dαu=αu=αux1α1xnαn.|\alpha|=\alpha_1+\alpha_2+\cdots+\alpha_n,\ D^\alpha u=\nabla^\alpha u=\displaystyle \frac{\partial^{|\alpha|}u}{\partial x_1^{\alpha_1}\cdots\partial x_n^{\alpha_n}}.

定义2(支集)

uC(Ω)\forall u\in C(\Omega),记 uu 的支集为

supp(u):={xΩ:u(x)0}.\text{supp}(u):= \{x\in\Omega:u(x)\neq 0\}.

定义3(紧支集函数类)

记全体 kk 阶连续可微且支集为紧集的函数全体为

C0k(Ω):={uCk(Ω):supp(u)为紧集}.C_0^k(\Omega):=\{u\in C^k(\Omega):\text{supp}(u)\text{为紧集}\}.

定义4(梯度)

u:RnRu:\mathbb{R}^n\to\mathbb{R}uC1u\in C^1,则 uu 的梯度记为

Du=u=(ux1,,uxn)T.D_u = \nabla u = \left(\frac{\partial u}{\partial x_1},\cdots,\frac{\partial u}{\partial x_n}\right)^T.

定义5(散度)

U:RnRnU:\mathbb{R}^n\to\mathbb{R}^n,令 U=(u1,u2,,un)T, ui:RnR, (i=1,2,,n)U=(u_1,u_2,\cdots,u_n)^T,\ u_i:\mathbb{R}^n\to\mathbb{R},\ (i=1,2,\cdots,n),则 UU 的散度记为

div u=u:= (x1,x2,,xn)T(u1,,un)T= u1x1+u2x2++unxn=i=1nuixi.\begin{aligned} \text{div }u = \nabla\cdot u :=&\ \left(\frac{\partial}{\partial x_1},\frac{\partial}{\partial x_2},\cdots,\frac{\partial}{\partial x_n}\right)^T\cdot (u_1,\cdots, u_n)^T\\ =&\ \frac{\partial u_1}{\partial x_1}+\frac{\partial u_2}{\partial x_2}+\cdots+\frac{\partial u_n}{\partial x_n} = \sum_{i=1}^n\frac{\partial u_i}{\partial x_i}. \end{aligned}

定理6(Green公式)

DR2D\subset \mathbb{R}^2 是有界域,D\partial D 分段光滑,P,QC1(D)P, Q\in C^1(D),则

DPdx+Qdy=D(QxPy)dxdy,\int_{\partial D}P\,d x+Q\,d y = \iint_D\left(\frac{\partial Q}{\partial x}-\frac{\partial P}{\partial y}\right)\,d x\,d y,

其中 D\partial D 取正向(逆时针方向).

推论7

DR2D\subset \mathbb{R}^2 是有界域,D\partial D 分段光滑,P,QC1(D)P, Q\in C^1(D),则

D(P,Q)Tnds=D(Px+Qy)dxdy    DUnds=DUdxdy,\int_{\partial D}(P,Q)^T\cdot \boldsymbol{n}\,ds = \iint_D\left(\frac{\partial P}{\partial x}+\frac{\partial Q}{\partial y}\right)\,dx\,dy\iff \int_{\partial D}U\cdot \vec{n}\,ds = \iint_D\nabla\cdot U\,dx\,dy,

其中 n\vec{n}D\partial D 上的单位外法向,U=(P,Q)TU=(P, Q)^T.


定理6推论7的证明请见Green公式在限制条件下的证明 Gauss定理.

定理8(Gauss公式,散度公式)

VR3V\subset \mathbb{R}^3 是有界域,V\partial V 分片光滑,P,Q,RC1(Vˉ)P, Q, R\in C^1(\bar{V}),则

V(Px+Qy+Rz)dxdydz=VPdydz+Qdzdx+Rdxdy,\iiint_V\left(\frac{\partial P}{\partial x}+\frac{\partial Q}{\partial y}+\frac{\partial R}{\partial z}\right)\,dx\,dy\,dz = \iint_{\partial V}P\,dy\,dz + Q\,dz\,dx + R\,dx\,dy,

U=(P,Q,R)TU = (P, Q, R)^T,则

VUdxdydz=VUnds.\iiint_V\nabla\cdot U\,dx\,dy\,dz = \iint_{\partial V} U\cdot \vec{n}\,ds.


分片光滑的定义和定理的证明请见Gauss 定理(散度定理).

定理9(Gauss-Green公式)设 ΩRn\Omega\in\mathbb{R}^n 为有界开集,且 ΩC1\partial \Omega \in C^1,若 U=(u1,,un)T:ΩˉRnU=(u_1,\cdots,u_n)^T:\bar{\Omega}\to\mathbb{R}^nuC1(Ω)C(Ωˉ)u\in C^1(\Omega)\cap C(\bar{\Omega}),则

ΩUdx=ΩUnds,\int_{\Omega}\nabla\cdot U\,d x = \int_{\partial \Omega} U\cdot \vec{n}\,d s,

其中 n\vec{n}Ω\partial \Omega 的单位外法向.

推论10

(1). 若u,vC1(Ω)C(Ωˉ)u, v\in C^1(\Omega)\cap C(\bar{\Omega}),则

Ωuxivdx=Ωuvxidx+Ωuvnids.\int_{\Omega}u_{x_i}v\,d x = -\int_{\Omega}uv_{x_i}\,d x+\int_{\partial \Omega}uvn_i\,d s.

(2). 若uC2(Ω)C1(Ωˉ)u\in C^2(\Omega)\cap C^1(\bar{\Omega}),则

ΩΔudx=Ωunds,\int_{\Omega}\Delta u\,d x=\int_{\partial \Omega}\frac{\partial u}{\partial \vec{n}}\,d s,

其中Δu=(u)=i=1n2uxi2, un=un\displaystyle \Delta u=\nabla\cdot(\nabla u) = \sum_{i=1}^n\frac{\partial^2 u}{\partial x_i^2},\ \frac{\partial u}{\partial \vec{n}}=\nabla u\cdot \vec{n}.

(3). 若u,vC2(Ω)C1(Ωˉ)u, v\in C^2(\Omega)\cap C^1(\bar{\Omega}),则

Ωuvdx=Ωuvdx+Ωuvnds.\int_{\Omega}\nabla u\cdot \nabla v\,d x=-\int_{\Omega} u\nabla v\,d x+\int{\partial \Omega}u\frac{\partial v}{\partial \vec{n}}\,d s.

(4). 若u,vC2(Ω)C1(Ωˉ)u,v\in C^2(\Omega)\cap C^1(\bar{\Omega}),则

Ω(uΔvvΔu)dx=Ω(uvnvun)ds.\int_{\Omega}(u\Delta v-v\Delta u)\,d x=\int_{\partial \Omega}\left(u\frac{\partial v}{\partial\vec{n}}-v\frac{\partial u}{\partial \vec{n}}\right)\,d s.


定理9为Gauss公式和Green公式的推广形式,证明略去(需要复杂的讨论)

证明 推论10
(1). 令 U=(0,,0,uv,0,,0)TU = (0,\cdots, 0, uv, 0,\cdots, 0)^T,即 Uj={uv,j=i,0,ji.U_j = \begin{cases}uv,&\quad j=i,\\ 0,&\quad j\neq i.\end{cases}

 ΩUdx= Ωuvxidx=Ωuxivdx+Ωuvxidx=Gauss-Green ΩUnds=Ωuvnids Ωuxivdx=Ωuvxidx+Ωuvnids.\begin{aligned} &\ \begin{aligned} \int_{\Omega}\nabla\cdot U\,d x =&\ \int_{\Omega}\frac{\partial uv}{\partial x_i}\,d x=\int_{\Omega}u_{x_i}v\,d x+\int_{\Omega}uv_{x_{i}}\,d x \xlongequal{\text{Gauss-Green}}&\ \int_{\partial \Omega}U\cdot \vec{n}\,d s=\int_{\partial \Omega}uvn_i\,d s\\ \end{aligned}\\ \Rightarrow&\ \int_{\Omega}u_{x_i}v\,d x=-\int_{\Omega}uv_{x_i}\,d x+\int_{\partial\Omega}uvn_i\,d s. \end{aligned}

(2).

ΩΔudx=Ω(u)dx=Gauss-GreenΩunds=Ωunds.\int_{\Omega}\Delta u\,d x=\int_{\Omega}\nabla\cdot(\nabla u)\,d x \xlongequal{\text{Gauss-Green}}\int_{\partial \Omega}\nabla u\cdot \vec{n}\,d s = \int_{\partial \Omega}\frac{\partial u}{\partial \vec{n}}\,d s.

(3). 由 (1)(1) 知,令 v=vxi\displaystyle v=\frac{\partial v}{\partial x_i},可得

Ωuxivxidx=Ωu2vxi2dx+Ωuvxinids,(i=1,2,,n),\int_{\Omega}\frac{\partial u}{\partial x_i}\frac{\partial v}{\partial x_i}\,d x =-\int_{\Omega}u\frac{\partial^2 v}{\partial x_i^2}\,d x+\int_{\partial \Omega}u\frac{\partial v}{\partial x_i}n_i\,d s,\quad(i=1,2,\cdots, n),

对上式左右两端同时对 i=1,2,,ni=1,2,\cdots, n 求和可得

Ωuvdx=ΩuΔvdx+Ωuvnds.\int_{\Omega}\nabla u\cdot \nabla v\,d x = -\int_{\Omega}u\Delta v\,d x+\int_{\partial \Omega}u\frac{\partial v}{\partial \vec{n}}\,d s.

(4). 由 (3)(3) 知,交换 u,vu, v 可得

Ωuvdx=ΩuΔvdx+Ωuvnds=ΩvΔudx+Ωvunds,\int_{\Omega}\nabla u\cdot \nabla v\,d x = -\int_{\Omega}u\Delta v\,d x+\int_{\partial \Omega}u\frac{\partial v}{\partial \vec{n}}\,d s=-\int_{\Omega}v\Delta u\,d x+\int_{\partial \Omega}v\frac{\partial u}{\partial \vec{n}}\,d s,

Ω(uΔvvΔu)dx=Ω(uvnvun)ds.\int_{\Omega}(u\Delta v-v\Delta u)\,d x =\int_{\partial \Omega}\left(u\frac{\partial v}{\partial \vec{n}}-v\frac{\partial u}{\partial \vec{n}}\right)\,d s.

三个经典偏微分方程

波动方程

弦振动方程(一维波动方程)

2ut2a22ux2=f,(0<x<l,t>0)\frac{\partial^2 u}{\partial t^2}-a^2\frac{\partial^2u}{\partial x^2} = f,\quad (0 < x < l, t > 0)

高维波动方程u=u(x1,x2,,xn,t)u=u(x_1,x_2,\cdots, x_n, t),高维波动方程为

2ut2a22ux2=f,\frac{\partial^2 u}{\partial t^2}-a^2\frac{\partial^2u}{\partial x^2} = f,

其中 Δu=i=1n2uxi2\Delta u = \sum_{i=1}^n\frac{\partial^2 u}{\partial x_i^2} 为Laplace算子,nn 为物理空间的维数.

热传导方程

u=u(x,y,z,t)u = u(x, y, z, t),则热传导方程为

uta2Δu=f.\frac{\partial u}{\partial t}-a^2\Delta u=f.

连续性方程

ρ=ρ(x,y,z,t)\rho = \rho(x, y, z, t),则连续性方程为

ρt+(ρv)=0,(x,y,z)Ω×(0,).\frac{\partial \rho}{\partial t} + \nabla\cdot(\rho\vec{v}) = 0,\quad (x, y, z)\in\Omega\times (0,\infty).

变分问题

定义1(C0C_0^{\infty}

Ω\OmegaR2\mathbb{R}^2 中的区域,定义在 Ω\Omega 上无穷次可微且在 Ω\Omega 的边界附近为 00 的函数全体记为 C0(Ω)C_0^{\infty}(\Omega).

例一(C0C_0^{\infty} 中的一种核函数)

ρ(x,y)={kexp(11(x2+y2)),x2+y2<1,0,x2+y21.\rho(x, y) = \left\{\begin{aligned} k\cdot \text{exp}\left(-\frac{1}{1-(x^2+y^2)}\right),&\quad x^2+y^2 < 1,\\ 0,&\quad x^2+y^2 \geqslant 1.\end{aligned}\right.ρ(x,y)C0(R2)\rho(x, y)\in C_0^{\infty(\mathbb{R}^2)},可以选取 kk 使得 R2ρ(x,y)dxdy=1\displaystyle \int_{\mathbb{R}^2}\rho(x, y)\,dx\,dy = 1,定义

ρn(x,y)=n2ρ(nx,ny),(n>0)\rho_n(x, y) = n^2\rho(nx, ny),\quad(n > 0)

R2ρn(x,y)dxdy=1\displaystyle \int_{\mathbb{R}^2}\rho_n(x, y)\,dx\,dy = 1 且当 x2+y21n\sqrt{x^2+y^2}\geqslant \frac{1}{n} 时,ρn(x,y)=0\rho_n(x, y)=0,即 supp(ρn)=B1/n\text{supp}(\rho_n) = B_{1/n}. 其中 B1/n={xR2:x2<1n}B_{1/n}=\{x\in\mathbb{R}^2:||x||_2 < \frac{1}{n}\} 即半径为 1n\frac{1}{n} 圆心在原点的开球.

利用 ρn\rho_n 可以将积分域缩小到 B1/nB_{1/n} 中,从而简化计算.

引理2.1

Ω\OmegaR2\mathbb{R}^2 中有界区域, f(x,y)f(x, y)Ω\Omega 上连续,若 φ(x,y)C0(Ω)\forall \varphi(x, y)\in C_0^\infty(\Omega)

Ωf(x,y)φ(x,y)dxdy=0,\iint_{\Omega}f(x, y)\varphi(x, y)\,dx\,dy = 0,

f(x,y)f(x, y)Ω\Omega 上恒为 00.


证明:(反证法,利用 ff 的连续性和例题中的 ρn\rho_n 替换 φ\varphi

反设 (x0,y0)Ω\exists (x_0,y_0)\in\Omega 使得 f(x0,y0)0f(x_0, y_0)\neq 0,不妨令 f(x0,y0)>0f(x_0, y_0) > 0,由于 ffΩ\Omega 上连续,则 δ>0\exists \delta > 0 使得 f(x,y)>0, xBδ(x0,y0)f(x, y) > 0,\ \forall x\in B_{\delta}(x_0, y_0),其中 Bδ(x0,y0B_{\delta}(x_0,y_0 表示以 (x0,y0(x_0, y_0 为圆心半径为 δ\delta 的开球.

对于上述 δ\delta 取充分大的 nn 使得 1nδ\frac{1}{n}\leqslant \delta,令 φ(x,y)=ρn(xx0,yy0)C0(Ω)\varphi(x, y)=\rho_n(x-x_0,y-y_0)\in C_0^\infty(\Omega),于是

0=Ωf(x,y)φ(x,y)dxdy=B1/n(x0,y0)f(x,y)ρn(xx0,yy0)dxdy>00=\iint_{\Omega}f(x, y)\varphi(x, y)\,dx\,dy = \iint_{B_{1/n}}(x_0,y_0)f(x, y)\rho_n(x-x_0,y-y_0)\,dx\,dy > 0

矛盾.

极小曲面问题

极小曲面问题:考虑 R2\mathbb{R}^2 上的有界区域 Ω\OmegaΩ\partial\Omega 充分光滑,令 Ω\partial \Omega 的参数方程为

l:α(s)=(x(s),y(s),φ(s)),(0ss0)l: \alpha(s) = (x(s),y(s),\varphi(s)),\quad(0\leqslant s\leqslant s_0)

其中 x(0)=x(s0),y(0)=y(s0),φ(0)=φ(s0)x(0) = x(s_0), y(0)=y(s_0), \varphi(0)=\varphi(s_0)(封闭曲线).

Ωˉ\bar{\Omega} 上的曲面 SS 满足:

  1. SSll 为周界.(S=l\partial S = l

  2. SS 的表面积最小.


令满足上述条件的曲面参数方程为 S:τ=(x,y,v(x,y))S:\tau = (x, y, v(x, y)),则 SS 的表面积为(更详细的说明请见二维图像的曲面积分

J(v)= Ωτx×τydxdy=Ω(1,0,vx)×(0,1,vy)dxdy= Ω(1+vx2)(1+vy2)vx2vy2dxdy=Ω1+vx2+vy2dxdy\begin{aligned} J(v) =&\ \iint_{\Omega}\left|\frac{\partial \tau}{\partial x}\times\frac{\partial \tau}{\partial y}\right|\,dx\,dy = \iint_\Omega|(1, 0, v_x)\times(0,1,v_y)|\,dx\,dy\\ =&\ \iint_\Omega\sqrt{(1+v_x^2)(1+v_y^2)-v_x^2v_y^2}\,dx\,dy = \iint_{\Omega}\sqrt{1+v_x^2+v_y^2}\,dx\,dy \end{aligned}

设全体满足条件的函数集合为

Mφ={v(x,y):vC1(Ωˉ), vΩ=φ},M_\varphi = \{v(x, y):v\in C^1(\bar{\Omega}),\ v|_{\partial\Omega}=\varphi\},

则该问题可转化为求解如下的极小化问题

minvMφJ(v).\min_{v\in M_{\varphi}} J(v).

假设该问题的解为 uu,则

J(u)=minvMφJ(v)    u=arg minvMφJ(v).(1)J(u) = \min_{v\in M_{\varphi}} J(v)\iff u = \argmin_{v\in M_{\varphi}}J(v).\tag{1}

这里 J(v)J(v) 为定义在函数集合 MφM_\varphi 上的泛函.

变分问题

变分问题就是如上式(1)(1)的一个求解泛函极值的问题.

问题求解

考虑对最优解做一个扰动,说明最优解是扰动后的最值即可.

M0={v(x,y):vC1(Ωˉ), vΩ=0}M_0 = \{v(x, y):v\in C^1(\bar{\Omega}),\ v|_{\partial\Omega} = 0\}εR,vM0\forall \varepsilon \in\mathbb{R}, \forall v\in M_0

则有 u+εvMφu+\varepsilon v\in M_{\varphi}J(u+εv)J(u)J(u+\varepsilon v)\geqslant J(u).

j(ε)=J(u+εv)j(\varepsilon) = J(u + \varepsilon v),则 j(ε)j(0)j(\varepsilon) \geqslant j(0),即 j(ε)ε=0=j(0)=0j'(\varepsilon)|_{\varepsilon=0} = j'(0) = 0(必要性).

j(ε)=J(u+εv)ε=Ω(ux+εvx)vx+(uy+εvy)vy1+(ux+εvx)2+(uy+εvy)2dxdy.\begin{aligned} j'(\varepsilon) = \frac{\partial J(u+\varepsilon v)}{\partial \varepsilon} = \int_{\Omega}\frac{(u_x+\varepsilon v_x)v_x+(u_y+\varepsilon v_y)v_y}{\sqrt{1+(u_x+\varepsilon v_x)^2+(u_y+\varepsilon v_y)^2}}\,dx\,dy. \end{aligned}

j(0)=Ω(ux1+ux2+uy2vx+uy1+ux2+uy2vy)dxdy=Ω11+ux2+uy2uvdxdy=0j'(0) = \int_\Omega\left(\frac{u_x}{\sqrt{1+u_x^2+u_y^2}}v_x+\frac{u_y}{\sqrt{1+u_x^2+u_y^2}}v_y\right)\,dx\,dy = \int_{\Omega}\frac{1}{\sqrt{1+u_x^2+u_y^2}}\nabla u\cdot \nabla v\,dx\,dy=0

uC2(Ωˉ)u\in C^2(\bar{\Omega}),由Green公式(分部积分形式)

Ωuvdxdy=ΩvΔudxdy+Ωvunds\int_{\Omega}\nabla u\cdot \nabla v\, dx\, dy = -\int_{\Omega}v\Delta u\,dx\,dy + \int_{\partial \Omega}v\frac{\partial u}{\partial\vec{n}}\,ds

可知

j(0)=Ω((11+ux2+uy2u))vdxdy+Ωv1+ux2+uy2unds=0j'(0) = -\int_{\Omega}\left(\nabla\cdot \left(\frac{1}{\sqrt{1+u_x^2+u_y^2}}\nabla u\right)\right)v\,dx\,dy + \int_{\partial \Omega}\frac{v}{\sqrt{1+u_x^2+u_y^2}}\frac{\partial u}{\partial \vec{n}}\, ds=0

由于 vΩ=0v|_{\partial \Omega} = 0(上式右侧第二项为 00),且 uC2(Ωˉ)u\in C^2(\bar{\Omega})(上式右侧第一项积分内为 00)可得

(11+ux2+uy2u)=x(ux1+ux2+uy2)+y(uy1+ux2+uy2)=0\nabla\cdot \left(\frac{1}{\sqrt{1+u_x^2+u_y^2}}\nabla u\right)=\frac{\partial}{\partial x}\left(\frac{u_x}{\sqrt{1+u_x^2+u_y^2}}\right)+\frac{\partial}{\partial y}\left(\frac{u_y}{\sqrt{1+u_x^2+u_y^2}}\right) = 0

Euelr方程

将上式和问题的条件称为该变分问题(1)(1)Euler方程

{(11+ux2+uy2u)=x(ux1+ux2+uy2)+y(uy1+ux2+uy2)=0,u(x,y)Ω=φ(x,y).\begin{cases} \displaystyle \nabla\cdot \left(\frac{1}{\sqrt{1+u_x^2+u_y^2}}\nabla u\right)=\frac{\partial}{\partial x}\left(\frac{u_x}{\sqrt{1+u_x^2+u_y^2}}\right)+\frac{\partial}{\partial y}\left(\frac{u_y}{\sqrt{1+u_x^2+u_y^2}}\right) = 0,\\ u(x,y)|_{\partial \Omega} = \varphi(x, y). \end{cases}


上述推导为必要性条件,充分性需考虑 j(ε)j''(\varepsilon),由于

j(ε)=Ωvx2+vy2+((uy+εvy)2vx(ux+εvx)2vy)(1+(ux+εvx)2+(uy+εuy)2)3/2>0j''(\varepsilon) = \int_{\Omega}\frac{v_x^2+v_y^2+((u_y+\varepsilon v_y)^2v_x-(u_x+\varepsilon v_x)^2v_y)}{(1+(u_x+\varepsilon v_x)^2+(u_y+\varepsilon u_y)^2)^{3/2}} > 0

j(0)=0j'(0)=0 即变分问题(1)(1)的充分条件.


偏微分方程 - 基础知识 变分问题 极小曲面问题
https://wty-yy.github.io/posts/51678/
作者
wty
发布于
2022年9月11日
许可协议