数学物理学报  2015, Vol. 35 Issue (4): 794-802   PDF (309 KB)    
扩展功能
加入收藏夹
复制引文信息
加入引用管理器
Email Alert
RSS
本文作者相关文章
向妮
石菊花
王玉娥
二维Monge-Ampère型方程的Neumann问题
向妮, 石菊花, 王玉娥     
湖北大学 数学与统计学学院 应用数学湖北省重点实验室, 武汉 430062
摘要: 该文通过构造闸函数将整体约化到边界, 证明了二维Monge-Ampère型方程Neumann边值问题解的二阶导数估计, 进而得到该方程Neumann边值问题经典解的存在性以及正则性.
关键词: 二维Monge-Ampère型方程     Neumann边值条件     二阶导数估计    
The Neumann Boundary Value Problems of Two Dimensional Monge-Ampère Equations
Xiang Ni, Shi Juhua, Wang Yu'e    
Faculty of Mathematics and Statistics, Hubei Key Laboratory of Applied Mathmatics, Hubei University, Wuhan 430062
Abstract: In this paper, we prove the second order derivatives estimates of Monge-Ampère type equations with Neumann boundary condition, using the method of auxiliary function which reduce the global estimates to the boundary. Then, we obtain the existence and regularity of the classical solutions for the equations.
Key words: Two dimensional Monge-Ampère type equations     Neumann boundary condition     Second order derivatives estimates    

1 引言

在二维的情形下,我们考虑带有Neumann边值条件的Monge-Ampère型方程

\begin{equation}\label{aa} \det [D^{2}u-A(x,Du)]=f(x),~\mbox{在$\Omega$内,} \end{equation} (1.1)
\begin{equation}\label{ab} D_{\nu}u=\varphi(x,u),     \mbox{在$\partial \Omega$上,} \end{equation} (1.2)
其中$\Omega$是$R^{2}$中的有界光滑区域,$Du=(u_{x_{1}},~u_{x_{2}})$, $(D^{2}u)=(u_{ij})_{2 \times 2}$, $f$是定义在$\Omega$上的函数且$f > 0$,$\nu$为$\partial \Omega$上的单位外法向量, $A(x,Du)$是$\Omega \times R^{2}$上对称的矩阵值函数,对任意$x \in \Omega$, $A$ 具有以下结构
\begin{equation}\label{ac} A(x,p) \geq -\mu_{0}(1+|p|^{2})I, \end{equation} (1.3)
其中$x \in \Omega$,$p \in R^{2}$,$\mu_{0} > 0$为常数,$I$为$n$阶单位阵, 参见Jiang,Trudinger和Yang的文献[1],$\varphi$是定义在$ \partial \Omega \times R^{2}$ 上的标量函数,满足以下条件
\begin{equation}\label{ad} \mbox{对任意 }\ x\in \partial \Omega,~-\varphi_{u}(x,u) \geq \gamma_{0} > 0, \end{equation} (1.4)
\begin{equation}\label{ae} \gamma_{0}+2\kappa_{1}>0, \end{equation} (1.5)
其中$\gamma_{0}$是常数,$\kappa_{1}(x)$为$\partial \Omega$上点$x$处最小主曲率, $\kappa_{1}=\inf\{\kappa_{1}(x):x \in \partial\Omega\}$.

最优运输问题是Monge-Ampère型方程重要的应用之一, 参见Ma,Trudinger和Wang的文献[2]. 在此应用中,矩阵 $A(x,Du)$由费用函数$c$生成

\begin{eqnarray} A(x,Du)=D^{2}_{x}c(x,Tx) ,x \in \Omega, \end{eqnarray} (1.6)
其中$c:R^{n} \times R^{n} \rightarrow R$是光滑的费用函数, $T$是由$Du(x)=D_{x}c(x,Tx)$决定的最优映射,并称方程的解$u$为势函数. Villani在文献[3]中详细介绍了最优运输问题的背景. Philippis在文献[4]中介绍了Monge-Ampère型方程和最优运输的联系及应用. Trudinger在文献[5]中给出了Monge-Ampère型方程(1.1)的新发展.

假设光滑的费用函数$c$在矩阵中满足以下条件

(A1) 对任意$x,p \in R^{n}$,存在唯一的$y \in R^{n}$使得$D_{x}c(x,y)=p$; 对任意$y,q \in R^{n}$,存在唯一的$x \in R^{n}$使得$D_{y}c(x,y)=q$;

(A2) 对任意$(x,y) \in R^{n} \times R^{n}$,~$\det \{ D^{2}_{xy}c(x,y)\} \neq 0$;

(A3w) 对所有$(x,p) \in \Omega \times R^{n},~\xi,\eta \in R^{n},~\xi \perp \eta,$

\begin{equation}\label{af} A_{ij,kl}(x,p)\xi_{i}\xi_{j}\eta_{k}\eta_{l}=\sum_{i,j,k,l,p,q,r,s}(c^{p,q}c_{ij,p}c_{q,rs}-c_{ij,rs})c^{r,k}c^{s,l}\xi_{i}\xi_{j}\eta_{k}\eta_{l} \geq 0, \end{equation} (1.7)
其中$A_{ij,kl}=D^{2}_{p_{k}p_{l}}A_{ij}$, $c_{i,j}(x,y)=\frac{\partial^{2}c(x,y)}{\partial x_{i} \partial y_{j}}$, $(c^{i,j})$为矩阵$(c_{i,j})$的逆矩阵.

若将(1.7)式右边的零换成 $c_{0}|\xi|^{2}|\eta|^{2}$,其中$c_{0} > 0$,则这个条件被称为(A3), 参见文献[2].

条件(A1)和(A2)用于保证$C^{1}$解的存在性,并且在最优运输中势函数的$C^{1}$正则性是必须的. 因此本文仅考虑Monge-Ampère型方程的二阶导数估计. 条件(A3w)表明,方程(1.1) 解的正则性依赖于矩阵$A$关于$p$的性质. 例如,若$A$关于$p$是凸的,则$A$满足条件(A3w). 有许多费用函数满足条件(1.7),参见文献[2, 6, 7].

Monge-Ampère型方程的各类边值问题已经得到广泛地研究. 对于方程(1.1)Dirichlet边值问题, Jiang,Trudinger和Yang在文献[1]中已给出经典解的存在性. Huang,Jiang和Liu在文献[8]中指出,若$f \in C^{\alpha}$,则有解的全局$C^{2,\alpha}$估计. 当$A \equiv 0$,方程(1.1)退化为标准的Monge-Ampère方程,在这种情况下, 文献[]中已给出第二边值问题的全局估计. 在Urbas的文献[12]中,如下Monge-Ampère方程

\begin{eqnarray} \det [D^{2}u-\sigma(x,u)]=g(x) \end{eqnarray} (1.8)
的斜边值问题已有结论,其中$\sigma:\Omega \times R^{n} \rightarrow R^{n \times n}$ 是对称的矩阵值函数且关于$u$单调不减,$g$是定义在$\Omega$上的标量函数. 若$\sigma(x,u)=\sigma(x)u$,对于这种特殊情况, 李松鹰在文献[6]中用不同的方法得到了解的存在性以及正则性.

Monge-Ampère型方程除了应用于最优传输,还应用于共形几何,几何光学等. 本文中我们采用不同于文献[13]的方法给出了二维情况下Monge-Ampère 型方程(1.1)的Neumann 边值问题解的二阶导数估计. 本文中仅给出了二维情形解的估计, 由于高维情形$\sum w_{ii}$与$\sum w^{ii}$ 不能相互控制,故该方法无法推广到高维的情形.

在文献[14]中,Lions,Trudinger和Urbas研究了带有Neumann边值条件的Monge-Ampère方程

\begin{equation} \det D^{2}u=f(x),  \mbox{在$\Omega$内,} \end{equation} (1.9)
\begin{equation} D_{\nu}u=\varphi(x,u),\qquad   \mbox{在$\partial \Omega$上,} \end{equation} (1.10)
其中$f \in C^{1,1}(\bar{\Omega})$,$f>0$,$\varphi \in C^{2,1}(\bar{\Omega} \times R)$且$\varphi$关于$z$单调不减. 在证明解的二阶导数估计时, 两篇文章均是通过构造闸函数将整体约化到边界. 对于切法估计, 本文的证明与文献[14]中的证明类似; 对于法法估计,文献[14]中利用$u$的凸性可得到, 而在本文中$u$没有凸性条件,故我们利用边界定义函数来构造辅助函数得到; 对于切切估计, 两篇文章构造的闸函数不同.

接下来我们引入一些定义. 令$Mu=D^{2}u-A(x,Du)$,若有$Mu > 0$, 则称方程(1.1)的解$u \in C^{2}(\Omega)$为椭圆解. 若存在定义函数$\phi \in C^{2}(\bar{\Omega})$,使得在$\partial \Omega$上, $\phi = 0$且$D\phi \neq 0$; $\Omega$内,$\phi < 0$, 并且在$\Omega$的邻域$\aleph$内有下列不等式成立

\begin{equation}\label{ak} D_{ij}\phi-D_{p_{k}}A_{ij}(x,Du)D_{k}\phi \geq \delta_{0}I, \end{equation} (1.11)
其中$\delta_{0} > 0$为常数,$I$为$n$阶单位阵,则称$\Omega$是$R^{n}$中严格$A$凸($A$有界)区域.

下面,我们给出这篇文章中的主要结果.

定理1.1 假设$\Omega$是$R^{2}$中严格$A$凸区域, $u \in C^{4}(\Omega) \cap C^{2}(\bar{\Omega})$是Neumann边值问题(1.1)--(1.2)的椭圆解, $\varphi$和$\Omega$满足结构条件(1.4)和(1.5), 矩阵值函数$A(x,Du)$满足结构条件(1.7),则

\begin{eqnarray} \sup_{\Omega}|D^{2}u| \leq C, \end{eqnarray} (1.12)
其中$C$依赖于$\gamma_{0},~\Omega,~|u|_{1;\bar{\Omega}},~A,~\varphi,~f.$

注1.1 由定理1.1和文献[13]中最大模估计和梯度估计,可得问题(1.1)--(1.2)解的C2估计, 进而根据连续性方法得到该问题解的存在性.

注1.2 在Gilbarg和Trudinger的书[15,定理6.31],若条件中函数充分光滑, 则定理1.1中解$u$的正则性可提高,即在定理1.1的基础上, 若$A,f,\varphi,\partial\Omega \in C^{\infty}$,则解$u \in C^{\infty}(\bar{\Omega})$.

2 二阶导数估计

文献[13]中已得到$R^{n}$中问题(1.1)--(1.2)解的最大模估计和梯度估计, 接下来我们采用不同于文献[13]中的办法,讨论$R^{2}$中解的二阶导数估计.

本节假设$u$和$Du$都是有界的,在此条件下,文献[1, 7]中有如下定理

定理2.1[1, 7]假设$u \in C^{4}(\Omega)$是方程(1.1)的椭圆解, 对任意$x \in \Omega,~A(x,Du)$是光滑的矩阵值函数, $f > 0$且$f \in C^{2}(\bar{\Omega})$. 若下列条件之一成立

(i) (A3)成立;

(ii) (A3w)和(1.11)均成立;

(iii) (A3w)成立,并且存在光滑下解$\tilde{u}$满足$\det[D^{2}\tilde{u}-A(x,D\tilde{u})] \geq f(x)$. \\ 则 $$\sup_{ \Omega}|D^{2}u| \leq C(1 + \sup_{\partial \Omega}|D^{2}u|),$$ 其中$C$依赖于$A,~f,~\Omega,~|u|_{1;\Omega}.$

接下来我们引入一些记号. 令 $$F[u]=\log[\det(D^{2}u-A(x,Du))],\quad F^{ij}=\frac{\partial F}{\partial w_{ij}}=w^{ij}, $$ $$F^{ij,kl}=\frac{\partial^{2}F}{w_{ij}w_{kl}}=-w^{ik}w^{jl}, \quad \{w_{ij}\}=\{u_{ij}-A_{ij}\}, $$ 其中$\{w^{ij}\}$为矩阵$\{w_{ij}\}$的逆矩阵. 为了方便起见,对任意向量$\xi,\eta$, 定义$D_{\xi\eta}u=D_{ij}u\xi_{i}\eta_{j}$,$w_{\xi\eta}=w_{ij}\xi_{i}\eta_{j} =D_{ij}u\xi_{i}\eta_{j}-A_{ij}\xi_{i}\eta_{j}.$

由定理2.1可知,对于二阶导数估计已将整体约化到边界, 因此仅需证明边界上的二阶导数估计. 不失一般性,可将$\varphi$和$\nu$分别延拓到 $\bar{\Omega} \times R$和$\bar{\Omega}$.

首先证明在边界上$|D_{\tau\nu}u|$有界. 定义切向导数算子 $\delta=(\delta_{1},\delta_{2})$,$\delta_{i}=(\delta_{ij}-\nu_{i}\nu_{j})D_{j},$ 其中$i,j=1,2.$ 将切向导数算子$\delta_{i}$作用于边值条件(1.2),则在$\partial\Omega$上,

\begin{equation}\label{ao} D_{k}u\delta_{i}\nu_{k}+ \nu_{k}\delta_{i}D_{k}u=\delta_{i}\varphi, \end{equation} (2.1)
\begin{eqnarray*} \tau_{i}\nu_{k}\delta_{i}D_{k}u&=&\tau_{i}\nu_{k}(\delta_{ij}-\nu_{i}\nu_{j})D_{jk}u = \tau_{i}\nu_{k}D_{ik}u-\tau_{i}\nu_{k}\nu_{i}\nu_{j}D_{jk}u\\ &=& \tau_{i}\nu_{k}D_{ik}u = D_{\tau\nu}u, \end{eqnarray*} 因此 $$ D_{\tau\nu}u =\tau_{i}\nu_{k}\delta_{i}D_{k}u = \tau_{i}\delta_{i}\varphi - \tau_{i}D_{k}u\delta_{i}\nu_{k}. $$ 计算可知,对于任意的切向量$\tau$,
\begin{equation}\label{as} |D_{\tau\nu}u|\leq C,  \mbox{在$\partial \Omega $上,} \end{equation} (2.2)
其中$C$依赖于$\varphi,~\Omega,~|u|_{1;\bar{\Omega}}.$

然后,在$\partial\Omega$上估计$D_{\nu\nu}u.$

由方程(1.1),令

\begin{equation}\label{al} F[u]=\log[\det(D^{2}u-A(x,Du))]=g(x), \end{equation} (2.3)
其中$\log f(x)=g(x).$

定义线性化算子

\begin{equation}\label{am} L=F^{ij}(D_{ij}-D_{p_{l}}A_{ij}(x,Du)D_{l}), \end{equation} (2.4)
则计算可得 $$ Lu=F^{ij}(D_{ij}u-D_{p_{l}}A_{ij}(x,Du)D_{l}u) =n+F^{ij}(A_{ij}-D_{p_{l}}A_{ij}(x,Du)D_{l}u). $$ 将方程(2.3)关于$x_{k}$求导可得
\begin{equation}\label{an} F^{ij}(D_{ijk}u-D_{k}A_{ij}(x,Du)-D_{p_{l}}A_{ij}(x,Du)D_{lk}u)=D_{k}g,~~~\mbox{其中$k=1,2,$} \end{equation} (2.5)
\begin{equation} Lu_{k}=D_{k}g+F^{ij}D_{k}A_{ij}(x,Du),~~~\mbox{其中$k=1,2.$} \end{equation} (2.6)

考虑函数$h=\nu_{k}D_{k}u-\varphi(x,u),$ $$ D_{i}h = D_{i}\nu_{k}D_{k}u+\nu_{k}D_{ik}u-\varphi_{i}-\varphi_{u}D_{i}u, $$ \begin{eqnarray*} D_{ij}h &=&D_{ij}\nu_{k}D_{k}u+D_{i}\nu_{k}D_{jk}u+D_{j}\nu_{k}D_{ik}u +\nu_{k}D_{ijk}u-\varphi_{ij}-\varphi_{iu}D_{j}u\\ &&-\varphi_{ui}D_{i}u-\varphi_{uu}D_{i}uD_{j}u-\varphi_{u}D_{ij}u. \end{eqnarray*} 由(2.4)和(2.5)式计算可得

\begin{equation} |Lh|\leq C_{1}(1 + T)\leq C_{2}T,  \mbox{在$\Omega$ 内,} \end{equation} (2.7)
其中$T=\sum F^{ii}$. 由在$\partial\Omega$上$h=0$,故 \[\left\{\begin{array}{ll} |Lh| \leq C_{2}T, &\mbox{在$\Omega$内,}\\ h = 0,&\mbox{在$\partial \Omega $上.} \end{array} \right.\] 令$\phi$为$\Omega$的定义函数,满足条件 \[\left\{\begin{array}{ll} \phi < 0, &\mbox{在$\Omega$内,}\\ \phi = 0,&\mbox{在$\partial \Omega $上.} \end{array} \right.\] 因为$\Omega$是严格$A$凸的(参见文献[15]),对于常数$\delta_{0}>0,$ \[\left\{\begin{array}{ll} L\phi \geq \delta_{0}T, &\mbox{在$\Omega$内$\partial \Omega $附近,}\\ \phi = 0,&\mbox{在$\partial \Omega $上.} \end{array} \right.\] 取$k$充分大,使得
\begin{equation} Lk\phi \geq k \delta_{0}T \geq C_{2}T \geq |Lh|, \end{equation} (2.8)
由比较原理可知
\begin{equation} |D_{\nu}h| \leq C_{3},  \mbox{在$\partial \Omega $上,} \end{equation} (2.9)
其中$C_{3}$与$\Omega$有关.

则综上可得

\begin{equation}\label{at} |D_{\nu \nu}u| \leq C_{4},~~~~~~~\mbox{在$\partial \Omega $上,} \end{equation} (2.10)
其中$C_{4}$与$\Omega $,$|\varphi|_{1;\bar{\Omega}}$,$|u|_{1;\bar{\Omega}}$有关.

由于$|D_{\tau \nu}u|,~|D_{\nu \nu}u|$有界,因此$|D_{\tau \nu}w|,~|D_{\nu \nu}w|$也有界.

最后,估计$u$在$\partial \Omega $上的切向二阶导数.

在边界上的任意一点,任意切向量均可写成

\begin{equation}\label{ar} \xi=〈 \xi,\tau〉 \tau +〈 \xi,\nu 〉 \nu,  〈 \tau,\nu 〉 =0, \end{equation} (2.11)
\begin{equation} w_{\xi\xi}=〈 \xi,\tau〉^{2}w_{\tau\tau}+2〈 \xi,\tau〉 〈 \xi,\nu 〉 w_{\tau \nu} +〈 \xi,\nu 〉^{2}w_{\nu\nu},  \mbox{在$\partial \Omega $上.} \end{equation} (2.12)
由(2.1)式可知,对$\partial \Omega $上任意单位切向量$\tau$,
\begin{equation} w_{\tau\nu}=\delta_{i}\varphi\tau_{i}-D_{k}u(\delta_{i}\nu_{k})\tau_{i}-A_{ij}\tau_{i}\nu_{j}. \end{equation} (2.13)
类似于Lions,Trudinger和Urbas在文献[14]中的方法, 在$\bar{\Omega} \times S^{1}$上构造函数,令
\begin{equation}\label{aq} H(x,\xi)=w_{\xi\xi}-\chi(x,\xi)+2〈 \xi,\tau 〉〈 \xi,\nu 〉 A_{ij}\tau_{i}\nu_{j}+\tilde{k}\phi, \end{equation} (2.14)
其中$\phi $为$\Omega$的定义函数,$\tilde{k}$为待定常数,在$\partial \Omega $上定义 $$\chi(x,\xi)=2〈 \xi,\tau 〉〈 \xi, \nu 〉 [\delta_{i}\varphi\tau_{i}-D_{k}u(\delta_{i}\nu_{k})\tau_{i}], $$ 在$\Omega $内定义$\chi(x,\xi)$是关于$Du$的线性组合.

对(2.3)式求导可得

\begin{equation}\label{ap} F^{ij}[D_{ij}u_{\xi}-D_{\xi}A_{ij}-(D_{p_{k}}A_{ij})D_{k}u_{\xi}]=D_{\xi}g, \end{equation} (2.15)
再次求导可得 \begin{eqnarray*} &&F^{ij}[D_{ij}u_{\xi\xi}-D_{\xi\xi}A_{ij}-2(D_{\xi p_{k}}A_{ij})D_{k}u_{\xi} -(D_{p_{k}p_{l}}A_{ij})D_{k}u_{\xi}D_{l}u_{\xi}-(D_{p_{k}}A_{ij})D_{k}u_{\xi\xi}]\\ &=&F^{ik}F^{jl}D_{\xi}w_{ij}D_{\xi}w_{kl}+D_{\xi \xi}g. \end{eqnarray*} \begin{eqnarray*} Lu_{\xi\xi}&=&w^{ij}[D_{ij}u_{\xi\xi}-(D_{p_{k}}A_{ij})D_{k}u_{\xi\xi}]\\ & \geq & w^{ik}w^{jl}D_{\xi}w_{ij}D_{\xi}w_{kl}+w^{ij}(D_{p_{k}p_{l}}A_{ij})u_{k\xi}u_{l\xi}-Cw^{ii}(1+w_{jj})\\ & \geq & w^{ij}(D_{p_{k}p_{l}}A_{ij})u_{k\xi}u_{l\xi}-Cw^{ii}(1+w_{jj})\\ & \geq & -C[w^{ii}(1+w_{jj})+w_{ii}], \end{eqnarray*} 第二个不等式利用了(A3w)条件. \begin{eqnarray*} LA_{\xi\xi} & = & w^{ij}[D_{ij}A_{\xi\xi}+2D_{i p_{k}}A_{\xi\xi}u_{kj}-D_{p_{k}}A_{ij}D_{k}A_{\xi\xi}-D_{p_{k}}A_{ij}D_{p_{l}}A_{\xi\xi}u_{kl}\\ && + D_{p_{k}p_{l}}A_{\xi\xi}u_{ki}u_{jl}+D_{p_{k}}A_{\xi\xi}u_{kij}]\\ & \leq & C+C[w^{ii}(1+w_{jj})]+w^{ij}D_{p_{k}p_{l}}A_{\xi\xi}u_{ki}u_{jl}+w^{ij}D_{p_{k}}A_{\xi\xi}u_{kij}\\ & \leq & C[w^{ii}(1+w_{jj})+w_{ii}]+w^{ij}D_{p_{k}}A_{\xi\xi}u_{kij}, \end{eqnarray*} 其中$C$与$n,~|A|_{2;\Omega},~|u|_{1;\Omega}$有关.

由(2.15)式可得 \begin{eqnarray*} w^{ij}D_{p_{k}}A_{\xi\xi}u_{kij} & = &D_{p_{k}}A_{\xi\xi}\{w^{ij}[D_{k}A_{ij}+D_{p_{l}}A_{ij}u_{lk}]+g_{k}\}\\ & \leq & C[w^{ii}(1+w_{jj})], \end{eqnarray*} 则

\begin{equation} LA_{\xi\xi} \leq C[w^{ii}(1+w_{jj})+w_{ii}]. \end{equation} (2.16)
同理,$LA_{ij} \leq C[w^{ii}(1+w_{jj})+w_{ii}],$ 即有$L(2〈 \xi,\tau 〉〈 \xi,\nu 〉 A_{ij}\tau_{i}\nu_{j}) \leq C[w^{ii}(1+w_{jj})+w_{ii}],$ 从而可得 $$Lw_{\xi\xi}=Lu_{\xi\xi}-LA_{\xi\xi}\geq -C[w^{ii}(1+w_{jj})+w_{ii}].$$

在$\Omega$内, $$\chi(x,\xi)=\chi(x,u,Du),$$ $$D_{i}\chi=\chi_{i}+\chi_{z}D_{i}u+\chi_{p_{k}}D_{ik}u,$$ \begin{eqnarray*} D_{ij}\chi&=&\chi_{ij}+\chi_{iz}D_{j}u+\chi_{jz}D_{i}u+\chi_{ip_{k}}D_{jk}u+\chi_{zz}D_{i}uD_{j}u+\chi_{zp_{k}}D_{i}uD_{jk}u\\ &&+\chi_{z}D_{ij}u+\chi_{jp_{k}}D_{ik}u+\chi_{zp_{k}}D_{j}uD_{ik}u+\chi_{p_{k}p_{l}}D_{ik}uD_{jl}u+\chi_{p_{k}}D_{ijk}u, \end{eqnarray*} $$ L\chi = w^{ij}[D_{ij}\chi-D_{p_{k}}A_{ij}D_{k}\chi], $$ 由$\chi$关于$Du$是线性的,可得$w^{ij}\chi_{p_{k}p_{l}}D_{ik}uD_{jl}u=0$, 由(2.5)式计算可知$$L\chi \leq C[w^{ii}(1+w_{jj})],$$ 由$\phi$的定义可知 $$L\tilde{k}\phi \geq \tilde{k}\delta_{0}\sum w^{ii}.$$

当$n=2$时,设$\lambda_{1},\lambda_{2}$为矩阵$\{w_{ij}(0)\}_{i,j \leq 2} $的特征值, 则 $$\sum w_{ii}=\lambda_{1}+\lambda_{2},$$ $$\sum w^{ii}=\frac{1}{\lambda_{1}}+\frac{1}{\lambda_{2}}=\frac{\lambda_{1}+\lambda_{2}}{\lambda_{1} \lambda_{2}}=\frac{\sum w_{ii}}{f},$$ $$\sum w_{ii}=f\sum w^{ii},$$ $$\max\limits_{\bar{\Omega}}|f|\sum w^{ii} \geq \sum w_{ii} \geq -\max\limits_{\bar{\Omega}}|f|\sum w^{ii}.$$ 计算可知 \begin{eqnarray*} LH&=&Lw_{\xi\xi}-L\chi+L(2〈 \xi,\tau 〉〈 \xi,\nu 〉 A_{ij}\tau_{i}\nu_{j})+L(\tilde{k}\phi)\\ &\geq&\tilde{k}\delta_{0}w^{ii}-C[w^{ii}(1+w_{jj})+w_{ii}]\\ &\geq& (\tilde{k}\delta_{0}-C)w^{ii}\\ &\geq& 0, \end{eqnarray*} 取$\tilde{k}$充分大,使得最后一个不等式成立. 由极值原理,$H$在边界上某点处取得极大值,设该点为$(x_{0},\xi_{0}) \in \partial \Omega \times S^{1}$,不妨设$\xi_{0}=(0,1)$.

下面分三种情况来讨论.

(a) $\xi_{0}$为$ \partial \Omega $上$x_{0}$处的法向量. 由$H$在$(x_{0},\xi_{0})$处取得极大值可知,$H(x_{0},\tau) \leq H(x_{0},\xi_{0}),$ 将式(2.14)代入可得$$w_{\tau\tau} \leq w_{\nu\nu} \leq C.$$

(b) $\xi_{0}$为$ \partial \Omega $上$x_{0}$处的非切非法向量. \begin{eqnarray*} H(x_{0},\xi_{0})&=&w_{\xi_{0}\xi_{0}}-\chi(x_{0},\xi_{0})-2〈 \xi_{0},\tau 〉〈 \xi_{0},\nu 〉 A_{ij}\tau_{i}\nu_{j}+\tilde{k}\phi\\ &=&〈 \xi_{0},\tau 〉^{2}w_{\tau\tau}+〈 \xi_{0},\nu 〉^{2}w_{\nu\nu}+\tilde{k}\phi\\ &\leq& 〈 \xi_{0},\tau 〉^{2}[H(x_{0},\tau)-\tilde{k}\phi]+〈 \xi_{0},\nu 〉^{2}w_{\nu\nu}+\tilde{k}\phi\\ &\leq& 〈 \xi_{0},\tau 〉^{2}[H(x_{0},\xi_{0})-\tilde{k}\phi]+〈 \xi_{0},\nu 〉^{2}w_{\nu\nu}+\tilde{k}\phi, \end{eqnarray*} 其中第二个等式利用了(2.11)式.

由$〈 \xi_{0},\tau〉^{2} + 〈 \xi_{0},\nu 〉^{2}=1$, 则有$$H(x_{0},\xi_{0}) \leq H(x_{0},\nu),$$ 上式结合$H(x_{0},\tau) \leq H(x_{0},\xi_{0})$,可知$H(x_{0},\tau) \leq H(x_{0},\nu)$, 将式(2.14)代入可得 $$w_{\tau\tau} \leq w_{\nu\nu} \leq C.$$

(c) $\xi_{0}$为$ \partial \Omega $上$x_{0}$处的切向量. 由$H$在$(x_{0},\xi_{0})$处取得极大值可知 $$0 \leq D_{\nu}H(x_{0},\xi_{0})=D_{\nu}w_{\xi_{0}\xi_{0}}-D_{\nu}(\chi(x_{0},\xi_{0})-2〈 \xi_{0},\tau 〉〈 \xi_{0},\nu 〉 A_{ij}\tau_{i}\nu_{j}+\tilde{k}\phi).$$ 则有 \begin{eqnarray*} 0 &\leq& D_{\nu}w_{11}(x_{0})+C\\ &=&D_{\nu}u_{11}(x_{0})-D_{\nu}A_{11}(x_{0},Du(x_{0}))+C\\ &\leq& D_{\nu}u_{11}(x_{0})+C\\ &=&D_{1}D_{\nu}D_{1}u(x_{0})-(D_{1}\nu_{k})D_{k}D_{1}u(x_{0})+C\\ &=&D_{11}D_{\nu}u(x_{0})-D_{1}[(D_{1}\nu_{k})D_{k}u(x_{0})]-(D_{1}\nu_{k})D_{k}D_{1}u(x_{0})+C\\ &=&D_{11}\varphi-(D_{11}\nu_{k})D_{k}u-2(D_{1}\nu_{k})D_{k}D_{1}u(x_{0})+C\\ &=&\varphi_{11}+\varphi_{1u}D_{1}u+\varphi_{u1}D_{1}u+\varphi_{uu}(D_{1}u)^{2} +\varphi_{u}D_{11}u\\ &&-(D_{11}\nu_{k})D_{k}u-2(D_{1}\nu_{k})D_{k}D_{1}u(x_{0})+C\\ &\leq &-\gamma_{0}D_{11}u(x_{0})-2(D_{1}\nu_{k})D_{k}D_{1}u(x_{0})+C, \end{eqnarray*} 其中第一个不等式和第二个不等式均利用了式(2.2)和(2.10), 最后一个不等式利用了条件(1.4).

设$r$为$\Omega$的定义函数,且在$\partial\Omega$上满足条件$|\bigtriangledown r|=1,$ 令$\lambda_{2}(x_{0})=\frac{\partial^{2}r}{\partial x_{1}^{2}},$ 则$$\lambda_{2}(x_{0}) \geq \kappa_{1}(x_{0}) \geq \kappa_{1},$$ 其中$\kappa_{1}(x_{0})$为$\partial\Omega$上点$x_{0}$处最小主曲率, $\kappa_{1}=\inf\{\kappa_{1}(x):x \in \partial\Omega\}.$ 计算可得 \begin{eqnarray*} -2(D_{1}\nu_{k})D_{k}D_{1}u(x_{0})&=&-2\frac{\partial^{2}r}{\partial x_{1}^{2}}D_{11}u(x_{0})-2\frac{\partial^{2}r}{\partial x_{2}\partial x_{1}}D_{2}D_{1}u(x_{0})\\ &\leq&-2\frac{\partial^{2}r}{\partial x_{1}^{2}}D_{11}u(x_{0})+C\\ &=&-2\lambda_{2}(x_{0})D_{11}u(x_{0})+C\\ &\leq&-2\kappa_{1}D_{11}u(x_{0})+C, \end{eqnarray*} 其中第一个不等式利用了式(2.2)和(2.10).

则 \begin{eqnarray*} 0&\leq &-\gamma_{0}D_{11}u(x_{0})-2(D_{1}\nu_{k})D_{1}D_{k}u(x_{0})+C\\ &\leq&-\gamma_{0}D_{11}u(x_{0})-2\kappa_{1}D_{11}u(x_{0})+C\\ &= &(-\gamma_{0}-2\kappa_{1})D_{11}u(x_{0})+C\\ &\leq &(-\gamma_{0}-2\kappa_{1})D_{11}w(x_{0})-(-\gamma_{0}-2\kappa_{1})A_{11}+C\\ &\leq &-(\gamma_{0}+2\kappa_{1})D_{11}w(x_{0})+C, \end{eqnarray*} 即有 $$w_{11}(x_{0}) \leq \frac{C}{\gamma_{0}+2\kappa_{1}}.$$

由以上三种情况,则对$\partial \Omega$上任意的方向$\xi$,则有 $w_{\xi\xi} \leq C,$ 故可得 $u_{\xi\xi} \leq C,$ 由定理2.1则有 $$\sup _{\Omega}|D^{2}u| \leq C.$$ 其中$C$依赖于$\gamma_{0},~\Omega,~|u|_{1;\bar{\Omega}},~A,~\varphi,~f.$

定理1.1 证毕.

参考文献
[1] Jiang F D, Trudinger N S, Yang X P. On the Dirichlet problem for Monge-Ampère type equations. Cale Var PDE, 2014, 49: 1223-1236
[2] Ma X N, Trudinger N S, Wang X J. Regularity of potential functions of the optimal transportation problem. Arch Rat Mech Anal, 2005, 177: 151-183
[3] Villani C. Optimal Transport-Old and New. Berlin: Springer, 2008
[4] Philippis G D, Figalli A. The Monge-Ampère equation and its link to optimal transportation. preprint
[5] Trudinger N S. Recent developments in elliptic partial differential equations of Monge-Ampère type. Proc Int Cong Math, Madrid, 2006, 3: 291-302
[6] Li S Y. On the oblique boundary value problems for Monge-Ampère equations. Pacific J Math, 1999, 190(1): 155-172
[7] Trudinger N S, Wang X J. On the second boundary value problem for Monge-Ampère type equations and optimal transportation. Ann Scuola Norm Sup Pisa Cl Sci, 2009, VIII: 143-174
[8] Huang Y, Jiang F D, Liu J. Dirichlet problem for Monge-Ampère type equations in optimal transportation. preprint
[9] Caffarelli L. Boundary regularity of maps with convex potentials II. Ann of Math, 1996, 144: 453-496
[10] Delanoë P. Classical solvability in dimension two of the second boundary value problem associated with the Monge-Ampère operator. Ann Inst Henri Poincaré, Analyse Non Linéaire, 1991, 8: 443-457
[11] Urbas J. On the second boundary value problem for equations of Monge-Ampère type. J Reine angew Math, 1997, 487: 115-124
[12] Urbas J. Oblique boundary value problems for equations of Monge-Ampère type. Calc Var PDE, 1998, 7: 19-39
[13] Jiang F D, Trudinger N S, Xiang N. On the Neumann problem for Monge-Ampère type equations. preprint
[14] Lions P L, Trudinger N S, Urbas J. The Neumann problem for equations of Monge-Ampère type. Comm Pure Appl Math, 1986, 39: 539-563
[15] Gilbarg D, Trudinger N S. Elliptic Partial Differential Equation of Second Order. Berlin: Springer, 2001
[16] Li S Y. On the Neumann problems for complex Monge-Ampère equations. Indiana Univ Math J, 1994, 43: 1099-1122
[17] Loeper G. On the regularity of solutions of optimal transportation problems. Acta Math, 2009, 202: 241-283
[18] Lieberman G M, Trudinger N S. Nonlinear oblique boundary value problems for nonlinear elliptic equations. Trans Amer Math Soc, 1986, 295: 509-546
[19] Trudinger N S. A note on global regularity in optimal transportation. Bull Math Sci, 2013, 3(3): 551-557