Motion of level sets by inverse anisotropic mean curvature
Extracted source text ยท arXiv 1804.06639
\documentclass[11pt,a4paper,headinclude,footinclude,fleqn,reqno]{amsart}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage[english]{babel}
\usepackage{graphicx} %
\usepackage[font=small]{quoting} %
\usepackage{caption}
\usepackage[colorlinks=true,linkcolor=blue]{hyperref}
\usepackage[top=.8in,bottom=1.2in,left=1.5in,right=1.5in]{geometry}
%\usepackage{showkeys}
%\usepackage{cite}
%\usepackage{refcheck}
\usepackage{verbatim}
\usepackage{color}
\usepackage{picture}
\usepackage{amsmath,amsthm,amsfonts,amssymb}
\usepackage{comment}
\newcommand{\diesis}{^\#}
\newtheorem{thm}{Theorem}[section]
\newtheorem{lem}[thm]{Lemma}
\newtheorem{cor}[thm]{Corollary}
\newtheorem{prop}[thm]{Proposition}
\newtheorem{exa}{Example}
\newtheorem{defn}{Definition}
\theoremstyle{definition}
\newtheorem{rem}[thm]{Remark}
\theoremstyle{remark}
\newtheorem{nota}{Notation}
\def\di{\displaystyle}
\def\O{\Omega}
\def\e{\epsilon}
\def\S{\Sigma}
\def\n{\nabla}
\def\P{\partial}
\def\p{\partial}
\def\N{n+1}
\def\bn{\overline \nabla}
\def\bd{\overline \Delta}
\def\a{\alpha}
\def\b{\beta}
\def\n{\nabla}
\def\nz{\mathbb{N}}
\def\gz{\mathbb{Z}}
\def\rz{\mathbb{R}}
\def\ef{\mathbb{F}}
\def\CC{\mathbb{C}}
\def\tr{\mathrm{tr}}
\def\o{\omega}
\def\O{\Omega}
\def\p{\partial}
\def\e{\epsilon}
\def\a{\alpha}
\def\b{\beta}
\def\g{\gamma}
\def\d{\delta}
\def\k{\kappa}
\def\l{\lambda}
\def\s{\sigma}
\def\bsl{\backslash}
\def\la{\longrightarrow}
\def\arr{\rightarrow}
\def\ov{\overline}
\def\De{\Delta}
\def\n{\nabla}
\def\<{\langle}
\def\>{\rangle}
\def\div{{\rm div}}
\def\De{\Delta}
\def\n{\nabla}
\def\ode{\overline{\Delta}}
\def\on{\overline{\nabla}}
\def\NN{\mathbb{N}}
\def\gz{\mathbb{Z}}
\def\RR{\mathbb{R}}
\def\FF{\mathbb{F}}
\def\CC{\mathbb{C}}
\def\SS{\mathbb{S}}
\def\HH{\mathbb{H}}
\def\tr{\mathrm{tr}}
\def\o{\omega}
\def\O{\Omega}
\def\p{\partial}
\def\e{\epsilon}
\def\a{\alpha}
\def\b{\beta}
\def\g{\gamma}
\def\d{\delta}
\def\l{\lambda}
\def\s{\sigma}
\def\bsl{\backslash}
\def\la{\longrightarrow}
\def\arr{\rightarrow}
\def\ov{\overline}
\def\ep{\epsilon}
\def\wh{\widehat}
\def\Z{\mathbb{Z}}
\def\Rn{\mathbb R^n}
\def\W{\mathcal W}
\def\de{\partial}
\def\eps{\varepsilon}
\def\To{\longrightarrow}
\def\wS{\overline w_{\eps,\delta}}
\def\ws{\underline w_{\eps,\delta}}
\def\lto{\left(}
\def\rto{\right)}
\def\lqu{\left[}
\def\rqu{\right]}
\def\lgr{\left\{}
\def\R{\mathbb{R}}
\def\N{\mathbb{N}}
\def\C{\mathbb{C}}
\def\ds{\displaystyle}
\DeclareMathOperator{\dive}{div}
\DeclareMathOperator{\epi}{epi}
\DeclareMathOperator{\ess}{ess}
\DeclareMathOperator{\esssup}{ess\,sup}
\DeclareMathOperator{\dist}{dist}
\DeclareMathOperator{\D}{D}
\DeclareMathOperator{\sign}{sign}
\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\Sk}{S}
\newenvironment{sistema}%
{\left\{\begin{array}{@{}l@{}}}{\end{array}\right.}
\patchcmd{\abstract}{\scshape\abstractname}{\textbf{\abstractname}}{}{}
\makeatletter
\def\@makefnmark{}
\makeatother
\newcommand{\myfootnote}[2]{\footnote{\textbf{#1}: #2}}
\begin{document}
\title{Motion of level sets by inverse anisotropic mean curvature}
\author{Francesco Della Pietra}
\address{Universit\`a degli studi di Napoli Federico II, Dipartimento di Matematica e Applicazioni ``R. Caccioppoli'', Via Cintia, Monte S. Angelo - 80126 Napoli, Italia.}
\email{f.dellapietra@unina.it}
\author{Nunzia Gavitone}
\address{Universit\`a degli studi di Napoli Federico II, Dipartimento di Matematica e Applicazioni ``R. Caccioppoli'', Via Cintia, Monte S. Angelo - 80126 Napoli, Italia.}
\email{nunzia.gavitone@unina.it}
\author{Chao Xia}
\address{School of Mathematical Sciences, Xiamen University, 361005, Xiamen, China}
\email{chaoxia@xmu.edu.cn}
\maketitle
\begin{abstract}
In this paper we consider the weak formulation of the inverse anisotropic mean curvature flow, in the spirit of Huisken-Ilmanen \cite{HI1}. By using approximation method involving Finsler-$p$-Laplacian, we prove the existence and uniqueness of weak solutions.
\end{abstract}
{\bf MSC2010:} 53C44, 35D05
{\bf Keywords: }inverse mean curvature flow, anisotropic mean curvature, Finsler-$p$-Laplacian
\section{Introduction}
Let $F\in C^\infty(\mathbb{R}^n\setminus\{0\})$ be a Minkowski norm in $\mathbb{R}^{n}$, i.e.,
\begin{itemize}
\item[(i)] $F$ is a norm in $\mathbb{R}^{n}$, i.e., $F$ is a convex, even, $1$-homogeneous function satisfying $F(\xi)>0$ when $\xi\ne 0$;
\item[(ii)] $F$ satisfies a uniformly elliptic condition: $D^2 (\frac12 F^2)$ is positive definite in $\mathbb{R}^{n}\setminus \{0\}$.
\end{itemize}
Let $X(\cdot,t)\colon M\times[0,T)\to \R^{n}$ be a family of smooth embeddings from a closed manifold $M$ in $\R^{n}$ satisfying the evolution equation
\begin{equation}
\label{iamcf}
\frac{\de}{\de t}X(x,t)=\frac{1}{H_{F}(x,t)}\nu_{F}(x,t),
\end{equation}
where $H_{F}(x,t)>0$ is the anisotropic mean curvature function of the hypersurface $N_{t}=X(M,t)$ and $\nu_{F}(x,t)$ is the unit anisotropic outer normal.
When $F$ is the Euclidean norm, $\nu_F$ and $H_F$ reduce to the unit outer normal and the mean curvature respectively, and in turn \eqref{iamcf} reduces to the classical inverse mean curvature flow (IMCF).
When $F$ is a general Minkowski norm, \eqref{iamcf} is so-called inverse anisotropic mean curvature flow (IAMCF).
Gerhardt \cite{Ge} and Urbas \cite{Ur} proved that the classical IMCF which initiated from a star-shaped and strictly mean convex hypersurface exists for all time and converge to a round sphere after rescaling. For general initial data, the IMCF may develop singularity. Huisken-Ilmanen \cite{HI1} has developed a theory of weak solutions for the IMCF of hypersurfaces in Riemannian manifolds by its level-set formulation and applied it to show the validity of the Riemannian Penrose inequality.
For the anisotropic counterpart, recently the third author \cite{x} has studied the IAMCF which initiated from a star-shaped and strictly $F$-mean convex hypersurface and proved the long time existence and convergence result analogous to Gerhardt and Urbas' result. The aim of this paper is to study Huisken-Ilmanen type weak solutions for the IAMCF by its level-set formulation.
Suppose the evolving hypersurfaces $N_{t}$ are given by level sets of a function $u\colon \R^{n}\to \R$, that is
\[
E_{t}= \{x\in \R^{n}: u(x)<t\}, \quad N_t=\p E_t.
\]
If $u$ is smooth and $\nabla u\ne 0$, then \eqref{iamcf} is equivalent to the degenerate elliptic equation
\begin{equation}
\label{prob1}
\dive\left(F_{\xi}(\nabla u) \right) = F(\nabla u).
\end{equation}
See Section 2.
%This is because the anisotropic mean curvature of $N_t$,
%\[
%H_F=\dive\left(F_{\xi}(\nabla u) \right)
%\] and the anisotropic outer normal of $N_t$
%\[
%\nu_F=F_\xi(\nabla u)\hbox{ and } \<\nabla u, \nu_F\>=F(\nabla u).\]
%See e.g. \cite{WX11}.
When $F$ is Euclidean, it is clear \eqref{prob1} reduces to
\begin{equation}
\label{prob2}
\dive\left(\frac{\nabla u}{|\nabla u|}\right) = |\nabla u|.
\end{equation}
As Huisken-Ilmanen \cite{HI1}, we define a weak solution of \eqref{prob1} by the following minimization principle.
\begin{defn}
\label{defweak}
Let $\O\subset \R^n$ be an open set.
A function $u\in C^{0,1}_{\textrm{loc}}(\Omega)$ is called a weak solution %(subsolution, supersolution respectively)
to \eqref{prob1} if
\begin{equation}
\label{mineq}
J_{F,u}(u)\le J_{F,u}(\varphi)
\end{equation}
for every precompact set $K\subset \Omega$ and for every test function $\varphi\in C^{0,1}_{\textrm{loc}}(\Omega)$ %($\varphi\le u$, $\varphi \ge u$ respectively)
with $\varphi=u$ in $\Omega\setminus K$, and where
\begin{equation}\label{JFu}
J_{F,u}(\varphi):=\int_{K} \big[F(\nabla \varphi)+\varphi F(\nabla u)\big]dx.
\end{equation}
Moreover, $u$ is a proper solution if in addition
\[
\lim_{|x|\to +\infty} u(x)=+\infty.
\]
\end{defn}
Our main result of this paper is the following existence result.
\begin{thm}
\label{thm-existence}
Let $\O\subset \R^n$ be an open set with smooth boundary such that $\O^c=\R^n\setminus \O$ is bounded.
There exists a unique proper weak solution $u\in C^{0,1}_{\textrm{loc}}(\overline \Omega)$ of \eqref{prob1}, in the sense of Definition \ref{defweak}, such that $u=0$ on $\de \O$. Moreover, $u$ satisfies
\begin{align}
\label{boundIMCF}
F(\nabla u(x)) \le \sup_{\de \Omega}H^+_{F}, \quad x \in \overline\Omega,\\
F(\nabla u(x)) \le H^{+}_{F}(x),\quad x \in \de\Omega,
\end{align}
where $H_F^+(x)=\max\{H_F(x),0\}$ and $H_F$ is the anisotropic mean curvature of $\de \Omega$.
\end{thm}
Huisken-Ilmanen's approach in the classical IMCF case to prove the existence is studying an approximate equation of \eqref{prob2}, known as elliptic regularization. One of the key feature of this elliptic regularization is that it corresponds to a family of translating graphs which solves the IMCF in $\R^{n}\times \R$.
It seems such elliptic regularization is not available in the anisotropic case.
Later, Moser \cite{m} found another approximate equation of \eqref{prob2} involving the $p$-Laplacian. It turns out that this approximate equation is also effective to prove the existence of weak solutions for IAMCF.
Inspired by Moser's approach, we consider the approximate equation of \eqref{prob1} involving the Finsler-$p$-Laplacian, that is,
\begin{equation}
\label{p_prob}
\left\{
\begin{array}{ll}
\dive\left( F^{p-1}(\nabla u)F_{\xi}(\nabla u) \right) = F(\nabla u)^{p} & \text{in }\Omega, \\[.2cm]
u=0 &\text{in }\Omega^{c},\\
u\to \infty&\text{as }x\to \infty.
\end{array}
\right.
\end{equation}
We have the following
\begin{thm} \label{theoremapprox}Let $\O\subset \R^n$ be an open set with smooth boundary such that $\O^c=\R^n\setminus \O$ is bounded.
For every $p>1$, there exists a unique solution $u_p\in C^{1,\alpha}_{loc}(\overline{\O})$. Moreover,
for every $\eps >0$, there exists $p_0=p_0(\eps)>1$ such that if $u_p\in C^{1,\alpha}_{loc}(\overline{\O})$ is the solution to \eqref{p_prob} for $1<p\le p_0$, then
\begin{align}
&F(\nabla u_p(x)) \le \sup_{\de \Omega}H^+_{F}+\eps, \quad x \in \overline\Omega\label{estcurv1}\\
&F(\nabla u_{p}(x)) \le H^+_{F}(x)+\eps, \quad x \in \de \Omega.\label{estcurv2}
\end{align}
\end{thm}
Theorem \ref{thm-existence} follows from Theorem \ref{theoremapprox} by approximation.
The rest of this paper is organized as follows. In Section 2, we recall some fundamentals on anisotropic functions and anisotropic mean curvature. In Section 3, we study Huisken-Ilmanen type weak formulation of IAMCF and its properties. In Section 4, we study the approximate equation involving the Finsler-$p$-Laplacian and show the gradient estimate and the existence of weak solution of IAMCF.
\
\section{Notation and preliminaries}
\subsection{Minkowski norm and Wulff shape}\
Let $F$ be a Minkowski norm on $\R^n$.
%\begin{rem}
%We stress that for $p\ge 2$ the condition
%\begin{equation*}
%\nabla^{2}_{\xi}[F^{2}](\xi)\text{ is positive definite in } \Rn\setminus\{0\},
%\end{equation*}
%implies \eqref{strong}.
%\end{rem}
The polar function $F^o\colon \Rn \rightarrow [0,+\infty[$
of $F$, defined as
\begin{equation*}
F^o(x)=\sup_{\xi\ne 0} \frac{\langle \xi, x\rangle}{F(\xi)},
\end{equation*}
is again a Minkowski norm on $\R^n$.
%It is easy to verify that also $F^o$ is a convex function
%which satisfies properties \eqref{eq:omo} and
%\eqref{eq:lin}.
Furthermore,
\begin{equation*}
F(\xi)=\sup_{x\ne 0} \frac{\langle \xi, x\rangle}{F^o(x)}.
\end{equation*}
%From the above property it holds that
%\begin{equation}
%\label{imp}
%|\langle \xi, \eta\rangle| \le F(\xi) F^{o}(\eta), \qquad \forall \xi, \eta \in \Rn.
%\end{equation}
Denote
\[
\mathcal W = \{ x\in \Rn \colon F^o(x)< 1 \}.
\]
This is the so-called Wulff shape centered at the origin. % We put
%$\kappa_n=|\mathcal W|$, where $|\mathcal W|$ denotes the Lebesgue measure
%of $\mathcal W$.
More generally, we denote by $\mathcal W_r(x_0)$
the set $r\mathcal W+x_0$, that is the Wulff shape centered at $x_0$
with radius $r$ and $\mathcal W_r=\mathcal W_r(0)$.
%We observe that $F$ is the support function of $\overline{\mathcal W}$.
The following properties of $F$ and $F^o$ hold true:
for any $x, \xi \in \Rn\setminus \{0\}$,
\begin{gather*}
\label{prima}
\langle F_\xi(\xi) , \xi \rangle= F(\xi), \quad \langle F_x^{o} (x), x \rangle
= F^{o}(x)
\\
\label{seconda} F( F_x^o(x))=F^o( F_\xi(\xi))=1,\\
\label{terza}
F^o(x) F_\xi( F_x^o(x) ) =x, \quad F(\xi)
F_x^o\left( F_\xi(\xi) \right) = \xi.\end{gather*}
See e.g. \cite{X2}, Chapter 2.
\subsection{Anisotropic mean curvature and anisotropic area functional}\
Let $N$ be a smooth closed hypersurface in $\Rn$ and $\nu$ be the unit Euclidean outer normal of $N$.
The anisotropic outer normal of $N$ is defined by \[
\nu_F=F_\xi(\nu).\]
The anisotropic mean curvature of $N$ is defined by
\[H_F=\dive(\nu_F).
\]
Here $\dive$ is the divergence operator with respect to the Euclidean metric.
In this paper we are interested in the case when $N$ is given by a level set of a smooth function $u$, namely, \[
N=N_{t}=\de E_t,\hbox{ where }E_t=\{x\in \Rn: u(x)<t\}.\]
When $\nabla u\neq 0$, it is clear that $\nu=\frac{\nabla u}{|\nabla u|}$ and \begin{eqnarray}\label{levelset}
\nu_F= F_\xi(\nabla u), \quad H_F= \dive(F_\xi(\nabla u)).
\end{eqnarray}
If $N_t$ satisfies the IAMCF, we see that $u(x(t))=t$ and by taking derivative about $t$,
we get \[\left\<\nabla u, \frac{1}{H_F}\nu_F\right\>=1.\] By virtue of \eqref{levelset}, we arrive at \eqref{prob1}.
The anisotropic area functional of $N$ is defined as
\[
\sigma_{F}(N)=\int_{N} F(\nu)\,d \mathcal{H}^{n-1},
\]
where $\mathcal{H}^{n-1}$ is the $(n-1)$-dimensional Hausdorff measure.
It is well-known that a variational characterization of $H_F$ is given by the first variational formula of $\sigma_{F}$, see for instance \cite{Re2, bpr, x}.
More precisely, we have
\begin{prop}[Reilly \cite{Re2}, Bellettini-Novaga-Riey \cite{bpr}]\label{HL}\
Let $N$ be a smooth closed hypersurface given by an embedding $X_0: M\to \Rn$. Let $N_s$ be a variation of $N$ given by $X(\cdot,s): M\to \Rn$, $s\in (- \eps, \eps),$ whose variational vector field $\frac{\p}{\p s}|_{s=0}X(\cdot,s) =V$. Then
\begin{eqnarray}\label{var0}
%\frac{d}{d t}\int_Md\mu_F=
\left.\frac{d}{d s}\right |_{s=0} \sigma_{F}(N_{s})=
\int_{N} \dive_{F,N}(V) F(\nu) d \mathcal{H}^{n-1}=
\int_N H_F(X_0) \left\<V, \nu\right\> d \mathcal{H}^{n-1},
\end{eqnarray}
where
\[
\dive_{F,N}(V) %= {\rm tr}\left(\left(Id-\nu_{F}\otimes \frac{\nu}{F(\nu)}\right)\nabla V\right)
:= {\rm div} V- \left\<\nabla_{\nu_F} V, \frac{\nu}{F(\nu)}\right\>.
\]
%\begin{eqnarray}\label{var}
%\frac{d}{d t}\int_M H_F(X) F(\nu)d\mu_g=\int_M 2\sigma_{2}(\k^F(X)) \psi(X) d\mu_g,
%\end{eqnarray}
%where $$\sigma_2(\k^F)=\sum_{i<j} \k^F_i\k^F_j.$$
\end{prop}
\begin{proof} We refer to \cite{x} for the proof of the second equality. For completeness, we prove the first equality here. We denote $\nu_s$ and $d\sigma_s$ be the unit outer normal and the area element of $N_s$ respectively.
It is well-known that
\[
\frac{d}{ds}\Big|_{s=0}\nu_s=-\<\nu, \nabla_{e_i}V\>e_i,
\]
where $\{e_i\}$ is an orthonormal basis of $TN$.
\[
\frac{d}{ds}\Big|_{s=0}d\sigma_s=\div_N(V) d\sigma.
\]
Thus
\begin{align*}
\frac{d}{ds}\Big|_{s=0} F(\nu_s) d\sigma_s&=\<F_\xi(\nu), -\<\nu, \nabla_{e_i}V\>e_i\>d\sigma+ F(\nu)\dive_N(V) d\sigma
\\&=\left[-\<\nu, \nabla_{e_i}V\>\<F_\xi(\nu), e_i\>+(\dive V- \<\nabla_\nu V, \nu\>)F(\nu)\right] d\sigma
\\&= \left[\div V -\left\<\nabla_{\nu_F} V, \frac{\nu}{F(\nu)}\right\> \right]F(\nu)d\sigma
\end{align*}
The last line follows from \[\nu_F=F_{\xi}(\nu)=F(\nu)\nu+\<F_\xi(\nu), e_i\>e_i.\]
\end{proof}
%and let $u\in C^{2}(\overline\Omega)$ such that $E_t=\{u<t\}$, $N_{t}=\{u=t\}$ and $\nabla u\ne 0$ on $N_t$. The anisotropic outer normal $\nu_{F}$ to $E_t$ on $N_{t}$ is given by
% \[
% \nu_F(x)= F_{\xi}(\nu)= F_{\xi}\left(\nabla u\right),\quad x\in \de \Omega.
% \]
%It holds
% \[
% F^o(\nu_F)=1.
% \]
% The anisotropic mean curvature of $N_t$ is defined as
% \begin{equation*}
% \label{H_F}
% H_{F}(x)= \dive\left( \nu_{F}(x)\right)=
% \dive\left[ \nabla_{\xi}
% F\left({\nabla u(x)}\right) \right], \quad x\in
% N_t.
% \end{equation*}
%\subsection{Anisotropic area functional }
%In this subsection we recall some basic properties and the definition of the anisotropic perimeter (\cite{bu,and}).
%\begin{defn}\label{perdef}
%Let $N$ be a smooth, oriented $(n-1)$-dimensional hypersurface in $\mathbb R^{n}$.
%The anisotropic area functional of $N$ is
%\[
% \sigma_{F}(N):=\int_{N} F(\nu)\,d \sigma,
%\]
%where $\nu$ denotes the outer normal to $N$ and $\sigma$ is the $(n-1)$-dimensional Hausdorff measure.
%\end{defn}
%Let $N$ be a smooth hypersurface
%For smooth hypersurfaces the following variational characterization for $H_{F}$ in terms of the anisotropic area functional is proved (see for instance \cite{Re1, Re2}, \cite{x}, \cite{bpr}).
%Here and in what follows, given two vectors $a,b \in \Rn$, we denote by $a \otimes b$ the matrix $(a \otimes b)_{ij}:= a_{i} b_{j}$.
%\begin{prop}\label{HL} Let $X_0: N\to\R^{n}$ be a smooth closed, oriented hypersurface and $X(\cdot,t)$ be a variation of $X_0$ with variational vector field $\frac{\p}{\p s}X(\cdot,s) =X_{s}(\cdot,s)=\psi(X)\nu(X)$, $- \eps <s<\eps$, where $\psi\in C^{\infty}(N)$. Denoted by $N_{s}=X(N,s)$, then
%\begin{eqnarray}\label{var0}
%\frac{d}{d t}\int_Md\mu_F=
%\left.\frac{d}{d s}\right |_{s=0} \sigma_{F}(N_{s})=
%\int_{N} \left.\dive_{F,N}X_{s}\right|_{s=0} F(\nu)d\sigma=
%\int_N H_F(X) \left.X_{s}\right|_{s=0}\cdot \nu\, d\sigma,
%\end{eqnarray}
%where
%\[
%\dive_{F,N} X_{s} = {\rm Tr}\left(\left(Id-n_{F}\otimes \frac{\nu}{F(\nu)}\right)\nabla X_{s}\right).
%\]
%\begin{eqnarray}\label{var}
%\frac{d}{d t}\int_M H_F(X) F(\nu)d\mu_g=\int_M 2\sigma_{2}(\k^F(X)) \psi(X) d\mu_g,
%\end{eqnarray}
%where $$\sigma_2(\k^F)=\sum_{i<j} \k^F_i\k^F_j.$$
%\end{prop}
%The previous result characterizes the minimizers of the area functional among the smooth sets as the ones with constant anisotropic mean curvature.
\
\section{IAMCF: a variational formulation}
In this section, we review the weak formulation of IAMCF developed by Huisken-Ilmanen \cite{HI1} by using a minimizing principle. We follow closely Huisken-Ilmanen's strategy in \cite{HI1}, Section 1.
%From now on, we will assume that
%\[
%\begin{array}{c}
%\Omega\subsetneq\R^{n}\text{ is an open set with }C^{2}\text{ boundary such that} \\ \Omega^{c}=\R^{n}\setminus \Omega\text{ is bounded, and } H_{F}>0 \text{ on }\de\Omega^{c}.
%\end{array}
%\]
\subsection{Weak formulation of IAMCF}
Recall (Definition \ref{defweak}) that $u$ is called a weak solution (subsolution, supersolution resp.) of \eqref{prob1} in $\O$ if $u\in C^{0,1}_{\textrm{loc}}(\Omega)$ and
$J_{F,u}(u)\le J_{F,u}(\varphi)$
for every precompact set $K\subset \Omega$ and for every test function $\varphi\in C^{0,1}_{\textrm{loc}}(\Omega)$ ($\varphi\le u$, $\varphi \ge u$ resp.) with $\varphi=u$ in $\Omega\setminus K$, where $J_{F,u}$ is defined in \eqref{JFu}.
The fact that
\[
J_{F,u}(\min\{\varphi,u\})+J_{F,u}(\max\{\varphi,u\})=J_{F,u}(\varphi)+J_{F,u}(u)
\]
whenever $\{u\neq \varphi\}$ is precompact implies $u$ is a solution if and only if it is both a weak supersolution and a weak subsolution.
There is an equivalent weak formulation by set functional. For $K\subseteq \Omega$ and $u\in C^{0,1}_{\textrm{loc}}(\Omega)$, define
\begin{equation*}
J_{F, u}(G)=J_{F,u}^{K}(G):=\int_{\de^{*}G\cap K} F(\nu)d\mathcal{H}^{n-1}-\int_{G\cap K}F(\nabla u)dx,
\end{equation*}
for a set $G$ of locally finite perimeter, and $\de^{*}G$ denotes the reduced boundary of $G$.
\begin{defn}
\label{setdef}
We say that $E$ minimizes $J_{F,u}$ in a set $A$ (minimizes on the outside, minimizes on the inside, resp.) if
\begin{equation*}
J_{F,u}(E)\le J_{F,u}(G)
\end{equation*}
for any $G$ such that $E\Delta G\subset\subset A$ ($G\supseteq E$, $G\subseteq E$ resp.) and any compact set $K$ containing $E\Delta G$.
Here $E\Delta G=(E\setminus G)\cup (G\setminus E)$.
\end{defn}
The fact that
\[
J_{F,u}(E\cap G)+J_{F,u}(E\cup G)\le J_{F,u}(E)+J_{F,u}(G)
\]
whenever $E\Delta G$ is precompact guarantees that $E$ minimizes $J_{F,u}$ in $\O$ if and only if $E$ is minimizes $J_{F,u}$ both on the inside and on the outside in $\O$.
The Definitions \ref{defweak} and \ref{setdef} are equivalent in the following sense.
\begin{prop}
\label{equiv}
Let $\Omega$ be an open set and $u\in C^{0,1}_{\textrm{loc}}(\Omega)$, then $u$ is a weak solution of \eqref{prob1} in $\Omega$ if and only if for each $t$,
$E_{t}=\{x\in\Omega \colon u<t\}$
minimizes $J_{F,u}$ in $\Omega$.
\end{prop}
\begin{proof}
%Let $\varphi\in C^{0,1}_{loc}$ such that $\{\varphi\ne u\}\Subset \Omega$ and $K$ be a compact set such that $K\cap \{\varphi\ne u\}$. Then
%\[
%J_{u}^{K}(\varphi)= J_{u}(\varphi<t).
%\]
By the co-area formula, we have for a choice of $a<b$ such that $a<u<b$ and $a<\varphi<b$ in $K$, that
\begin{align}
\notag J_{F,u}(\varphi)&=\int_{K}(F(\nabla \varphi) +\varphi F(\nabla u))
dx=\\ &=\int_{a}^{b} dt\int_{(\de\{\varphi<t\})\cap K}\left(F\left(\frac{\nabla \varphi}{|\nabla \varphi|}\right)+\frac{t}{|\nabla \varphi|} F(\nabla u)\right)d\sigma\\
\notag &= \int_{a}^{b} \int_{(\de\{\varphi<t\})\cap K}F(\nu)\,d\sigma -\int_{K}\int_{a}^{b}\chi_{\{\varphi<t\}}F(\nabla u)d\sigma + \\
&\notag\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad+b\int_{K}F(\nabla u)\,dx\\
\notag &= \int_{a}^{b} J^{K}_{u}(\{\varphi<t\}) + b\int_{K}F(\nabla u)\,dx.
\end{align}
Then, if for any $t$, $E_{t}$ is a minimizer of the set functional $J_{F,u}$, then
\[
J_{F,u}(\varphi) \ge J_{F,u}(u),
\]
that gives the minimality of $u$.
The viceversa can be proved exactly as in the proof of Lemma 1.1 in \cite{HI1}.
%we first assume that $u$ is a supersolution in the sense of Definition \ref{defweak}. Let $t_{0}$ and $G$ such that $G\subseteq E_{t_{0}}$ and $E_{t_{0}}\setminus G\Subset \Omega$, we have to prove that
%\begin{equation}
%\label{aim1}
%J_{u}(E_{t_{0}}) \le J_{u}(G).
%\end{equation}
%By the lower semicontinuity of $J_{u}$ we may assume that
%\[
%J_{u}(G) \le J_{u}(B),
%\]
%for all $B$ with $B \Delta E_{t_{0}} \subset G \Delta E_{t_{0}}$.
\end{proof}
Next we study the weak formulation of IAMCF with initial condition.
\begin{defn}\label{weak2def} Let $E_0$ be an open set with smooth boundary. Let $\{E_t\}_{t>0}$ be a nested family of open sets in $\Rn$.
\begin{itemize}
\item[(i)]
$u$ is called a weak solution of \eqref{prob1} with initial condition $E_0$ if $u\in C^{0,1}_{loc}(\Rn)$, $E_0=\{u<0\}$ and $u$ is a weak solution of \eqref{prob1} in $\Rn\setminus \overline{E_0}$.
% If we denote $\O= \Rn\setminus \overline{E_0}$, then this is equivalent to say that $u\in C^{0,1}_{loc}(\Rn)$ is a weak solution of \eqref{prob1} in $\O$ with $u=0$ on $\de \O$.
\item[(ii)] Define $u$ by the characterization $E_t=\{u<t\}$.
$\{E_t\}_{t>0}$ is called a weak solution of \eqref{iamcf} with initial condition $E_0$ if $u\in C^{0,1}_{loc}(\Rn)$ and $E_t$ minimizes $J_{F,u}$ in $\Rn\setminus E_0$ for each $t>0$.
\end{itemize}
\end{defn}
From Proposition \ref{equiv}, it is easy to see the above two definitions are also equivalent.
\begin{prop} $u$ is a weak solution of \eqref{prob1} with initial condition $E_0$ if and only if $\{u<t\}_{t>0}$ is a weak solution of \eqref{iamcf} with initial condition $E_0$.
\end{prop}
The weak solution is unique.
\begin{prop}[Uniqueness of the weak solutions]
\begin{enumerate}
\item Let $u$ and $v$ be weak solutions to \eqref{prob1} in $\Omega$ in the sense of Definition \ref{defweak}, and $\{v>u\}\subset\subset\Omega$. Then $v\le u$ in $\Omega$;
\item if $\{E_{t}\}_{t>0}$ and $\{F_{t}\}_{t>0}$ solve \eqref{iamcf} in the sense of Definition \ref{weak2def}, with initial data $E_{0},F_{0}$ respectively, and $E_{0}\subseteq F_{0}$, then $E_{t}\subseteq F_{t}$ as long as $E_{t}$ is precompact. In particular, for a given $E_{0}$ there exists at most one solution $\{E_{t}\}_{t>0}$ of \eqref{iamcf} such that $E_{t}$ is precompact.
\end{enumerate}
\end{prop}
\begin{proof}
See Huisken-Ilmanen \cite{HI1}, page 377-378.
\end{proof}
\subsection{Properties of weak IAMCF}
\begin{defn} Let $\O$ be an open set.
\begin{itemize}
\item[(i)] A set $E$ is called an $F$-minimizing hull in $\O$ if
\begin{equation}\label{f-mini}
\sigma_F(\de^*E\cap K)\le \sigma_F(\de^*G\cap K)
\end{equation}
for any $G$ containing $E$ such that $G\setminus E\subset\subset \O$ and any compact set $K$ containing $G\setminus E$.
\item[(ii)] A set $E$ is called a strictly $F$-minimizing hull in $\O$ if it is an $F$-minimizing hull in $\O$ and equality holds in \eqref{f-mini} if and only if
\[
G\cap \O= E\cap \O \hbox{ a.e. }
\]
\item[(iii)] Given a measurable set $E$, the set $E'$ is defined to be the intersection of all the strictly $F$-minimizing hulls in $\O$ that contain $E$.
\end{itemize}
\end{defn}
%In the following, we set
%\[E_t=\{u<t\}, \quad E_t^+={\rm int}\{u\le t\}, \quad N_t=\de E_t, \quad N_t^+=\de E_t^+.\]
One has the following properties for weak solutions of IAMCF.
\begin{prop} Let $u$ be a weak solution of \eqref{prob1} with initial condition $E_0$. Set
\[
E_t=\{u<t\}, \quad E_t^+={\rm int}\{u\le t\}.
\]
Then
\begin{itemize}
\item[(i)] For $t>0$, $E_t$ is an $F$-minimizing hull in $\Rn$;
\item[(ii)] Fot $t\ge 0$, $E_t^+$ is a strictly $F$-minimizing hull in $\Rn$ and $E_t'=E_t^+$ if it is precompact;
\item[(iii)] For $t>0$, $\sigma_F(\de E_t)= \sigma_F(\de E_t^+)$ provided $E_t^+$ is precompact. This holds true for $t=0$ if $E_0$ is an $F$-minimizing hull.
\item[(iv)] $\s_F(E_t)=e^t\s_F(E_0)$ provided $E_0$ is an $F$-minimizing hull.
\end{itemize}
\end{prop}
\begin{proof}
See Huisken-Ilmanen \cite{HI1}, page 372-373.
\end{proof}
%{\color{red} Can you look for some references about regularity of anisotropic minimizing problem like theorem 1.3 in Huisken-Ilmanen's paper?}
%\subsection{Initial value problem and regularity result}
%Let $E_{0}\subseteq \Omega$ an open set with $C^{1}$ boundary. We say that $u$ is a weak solution of \eqref{prob1} with $u=0$ on $\de E_{0}$ or equivalently that $E_{t}=\{u <t\}$, $t>0$ is a weak solution of \eqref{iamcf} with initial condition $E_{0}$ if
%\begin{equation}
%\label{ci}
%u \in C^{0,1}_{loc}(\Omega) \text{ and } E_{t} \text{ minimizes } J_{u} \text{ in } \Omega \setminus E_{0} \text{ for each } t \ge 0.
%\end{equation}
%\subsection{Weak anisotropic mean curvature.}
%When a hypersurface are just in $C^{1}$, the first equality in \eqref{var0} continues to hold true while we look at the second equality as a definition of an anisotropic weak mean curvature as follows.
Analog with the classical case, we define the weak anisotropic mean curvature by the first variational formula, Proposition \ref{HL}.
\begin{defn}
Let $N\subset \Rn$ be a hypersurface of $C^{1} $ or $C^{1}$ with a small singular set and locally finite Hausdorff measure. A locally integrable function $H_{F}$ on $N$ is called weak anisotropic mean curvature provided it satisfies the second equality in \eqref{var0} for every $V \in C_{c}^{\infty}(\Rn)$.
\end{defn}
For smooth IAMCF given by $\{u=t\}$, one sees $H_F=F(\nabla u)$. We show next weak solutions of \eqref{prob1} still have this property.
%Now in the spirit of \cite{HI1} we observe that there is also a connection between the minimizers of $J_{u}$ and the weak anisotropic mean curvature.
%In \cite{x} the author proves that it is equivalent to the following flow
%$$
%\frac{\de}{\de t} X(M,t)=\frac{F(\nu)}{H_{F}}\nu
%$$
\begin{prop}
Let $u$ be a weak solution of \eqref{prob1} with initial condition $E_0$ and let $N_t=\de E_t=\de \{u<t\}$. Then for a.e. $t$, the weak anisotropic mean curvature $H_F$ of $N_t$ satisfies
\[
H_F= F(\nabla u)\hbox{ a.e. }x\in N_t.
\]
\end{prop}
\begin{proof}Let $V \in C_{c}^{\infty}(\Rn)$ and $\Phi^s: \Rn\to \Rn, s\in (-\eps, \eps),$ be the flow of diffeomorphisms generated by $V$ and $\Phi^0=Id$.
Let $W$ be any precompact open set containing ${\rm supp}(V)$.
Because $u$ be a weak solution of \eqref{prob1} in $\Rn\setminus \overline{E_0}$, we see
$J_{F,u}(u\circ \Phi^s)\le J_{F,u}(u).$
Thus $\frac{d}{ds}\Big|_{s=0}J_{F,u}(u\circ \Phi^s)=0$. Next we derive $\frac{d}{ds}\Big|_{s=0}J_{F,u}(u\circ \Phi^s)$.
First, we assume $u$ is smooth. Then
\begin{align*}
&\frac{d}{ds}\Big|_{s=0}\int_K F(\nabla (u\circ\Phi^s)) dx\\&=
\int_W F_{\xi_i}(\nabla u) \nabla_i (\frac{d}{ds}\Big|_{s=0}(u\circ\Phi^s))dx\\
&=\int_W F_{\xi_i}(\nabla u)\nabla_i (\nabla_j u V^j)dx\\
&=\int_W F_{\xi_i}(\nabla u) \nabla^2_{ji} u V^j +F_{\xi_i}(\nabla u) \nabla_j u \nabla_i V^j dx
\\&=\int_W -F_{\xi_i\xi_k}(\nabla u) \nabla^2_{kj} u \nabla_i u V^j -F_{\xi_i}(\nabla u) \nabla_i u \nabla_j V^j +F_{\xi_i}(\nabla u) \nabla_j u \nabla_i V^j dx
\\&=\int_W -F(\nabla u)\dive V+ \<\nabla u, \nabla_{F_{\xi}(\nabla u)} V\> dx.
\end{align*}
By co-area formula,
\begin{align*}
&\int_W -F(\nabla u)\dive V+\<\nabla u, \nabla_{F_{\xi}(\nabla u)} V\> dx
\\&=\int_{-\infty}^\infty \int_{N_t\cap W} -\dive V F(\nu)+ \<\nu, \nabla_{\nu_F} V\> d\sigma_t dt
\\&=\int_{-\infty}^\infty \int_{N_t\cap W} -\dive_{F, N}V F(\nu)d\sigma_t dt.
\end{align*}
Thus \begin{align}\label{firstvariation}
\frac{d}{ds}\Big|_{s=0}\int_K F(\nabla (u\circ\Phi^s)) dx=\int_{-\infty}^\infty \int_{N_t\cap W} -\dive_{F, N}V F(\nu)d\sigma_t dt.
\end{align}
By an approximation argument, we see that the formula \eqref{firstvariation} is still true for $u$ only locally Lipschitz.
On the other hand, it is easy to see
\begin{align}\label{firstvariation2}
\frac{d}{ds}\Big|_{s=0}\int_K (u\circ\Phi^s) F(\nabla u) dx&=\int_W \<\nabla u, V\>F(\nabla u) dx\nonumber
\\&= \int_{-\infty}^\infty \int_{N_t\cap W} \<\nu, V\>F(\nabla u) d\sigma_t dt.
\end{align}
Combining \eqref{firstvariation} and \eqref{firstvariation2}, we get
\begin{align}\label{firstvariation3}
0&=\frac{d}{ds}\Big|_{s=0}J_{F,u}(u\circ \Phi^s)\nonumber\\&= \int_{-\infty}^\infty \int_{N_t\cap W} -\dive_{F, N}V F(\nu)+ \<\nu, V\>F(\nabla u) d\sigma_t dt.
\end{align}
Finally, by the definition of the weak anisotropic mean curvature, we conclude from \eqref{firstvariation3} that
\[
H_F= F(\nabla u)\hbox{ a.e. }x\in N_t \hbox{ a.e. }t.
\]
\end{proof}
%The anisotropic surface measure is
%\[
%\sigma_{F}(M):=\int
%\]
%Then
%\[
%0=\left.\frac{d}{ds}\right|_{s=0} J_{u}(u\circ \Phi^{s}), \quad \Phi^{0}=id
%\]
%Given $v^{s}=u\circ \Phi^{s}$, then by coarea formula
%\begin{align*}
%\left.\frac{d}{ds}\right|_{s=0}\int_{E}F(\nabla v^{s})&=
%\left.\frac{d}{ds}\right|_{s=0}\int_{-\infty}^{+\infty} dt\int_{E\cap\{v^{s}=t\}}F(\nu)d\mu_{g}\\
%&=\left.\frac{d}{ds}\right|_{s=0} \int_{-\infty}^{+\infty}
%\sigma_{F}(\{v^{s}=t\}\circ E)\\
%&=\int_{-\infty}^{+\infty}dt\int_{E\cap \{u=t\}}H_{F}X\cdot \nu \,d\mu_{g}.
%\end{align*}
%On the other hand,
%\begin{align*}
%\left.\frac{d}{ds}\right|_{s=0}\int_{E}v^{s}F(\nabla u)dx&=\int_{E}\nabla u\cdot \left.\frac{d}{ds}\right|_{s=0}\Phi^{s}(x)F(\nabla u)\,dx\\
%&=\int_{E}\nabla u\cdot X \, F(\nabla u) \,dx\\
%&=-\int_{-\infty}^{+\infty}\int_{E\cap\{u=t\}} X\cdot \nu F(\nabla u)\,d\mu_{g}.
%\end{align*}
%Hence, by definition of $J_{u}(v)$ and the minimality of $u$ we have
%As regard the existence of solutions to the problem \eqref{prob1}
\section{Existence of solutions and gradient estimates}
For any $p>1$, we will consider the following auxiliary problem
\begin{equation}
\label{p_change}
\left\{
\begin{array}{ll}
\dive\left( F^{p-1}(\nabla v)F_{\xi}(\nabla v) \right) = 0 & \text{in }\Omega, \\[.2cm]
v=1 &\text{in }\Omega^{c}.\\
v\to 0&\text{as } x\to\infty.
\end{array}
\right.
\end{equation}
\begin{prop}
\label{arm}
If $1<p<n$, then there exists a unique positive solution $v_{p}\in C^{1,\alpha}_{\textrm{loc}}(\overline\Omega)\cap C^\infty(\overline\Omega\setminus \{\nabla v_p=0\})$ of \eqref{p_change}. If $\mathcal W_{r}(x_0)\subset \Omega^{c}\subset \mathcal W_{s}(y_0)$, then
\begin{equation}
\label{th:confr}
\left(\frac{r}{F^{o}(x-x_{0})}\right)^{\frac{n-p}{p-1}}\le v_{p}(x)
\le \left(\frac{s}{F^{o}(x-y_{0})}\right)^{\frac{n-p}{p-1}},\quad\forall x\in \Omega \setminus \{y_{0}\};
\end{equation}
Moreover, $v_{p}$ verifies
\begin{equation}
\label{limgrad}
\lim_{|x| \to \infty} \displaystyle \frac{F(\nabla v_{p})}{v_{p}}=0.
\end{equation}
\end{prop}
\begin{proof}
The proof of existence, uniqueness, regularity, as well as \eqref{th:confr}, follow by nowadays standard arguments; we refer the reader to \cite[Theorem 3.3]{bc} for the general anisotropic case we consider.
Finally we prove \eqref{limgrad}. We argue as in \cite{m}. Let $\eta \in C^{\infty}_{0}(\Omega)$ be a suitable cut-off function. Taking $\psi= v_{p}\eta^{p}$ as test function in the weak formulation for \eqref{p_change} and using the H\"older inequality, it easily follows that:
\[
\int_{\Omega} \eta^{p}F(\nabla v_{p})^{p}dx \le p^{p}\int_{\Omega}v_{p}^{p}F(\nabla \eta)^{p} dx.
\]
By Harnack inequality (see for instance \cite{t}) we get
\[
r^{p-n}\int_{B_{r/4}(x_0)} F(\nabla v_{p})^{p}dx \le C(n,p) \inf_{B_{r/2}(x_0)} v_p^p,
\]
where $C(n,p)$ is a positive constant depending on $n$ and $p$. By applying the result contained in \cite{dib}, we have
\[
\|F(\nabla v_{p})\|_{L^{\infty}(B_{r/8}(x_{0}))} \le \frac{C(n,p)}{r} \inf_{B_{r/2}(x_0)} v_p,
\]
which implies \eqref{limgrad}, and the proof is completed.
\end{proof}
It is direct to see that $$u_p=(1-p)\log v_p\in C^{1,\alpha}(\overline{\Omega})$$ solves \eqref{p_prob} (we refer the reader, e.g., also to \cite{dpgzaa} for problems involving equations as in \eqref{p_prob}). Next we show the gradient estimate in Theorem \ref{theoremapprox}, which is based on the following Lemma.
\begin{lem}\label{sperb}
Let $1<p<n$ and $u_p\in C^{1,\alpha}_{\textrm{loc}}(\overline\Omega)$ be a solution to \eqref{p_prob}, then
\begin{equation}
\label{maxgrad}
\sup_{\bar\Omega}F(\nabla u_p)=\sup_{\de\Omega} F(\nabla u_p).
\end{equation}
\end{lem}
\begin{proof}
We omit the subscript $p$ in $u_p$ in the proof. Let $\tau= \sup_{\de\Omega} F(\nabla u)$. We consider the following set
\[
\Omega_{\beta}=\left\{x\in\Omega\colon F(\nabla u)> \beta\right\},
\]
with $\beta>\tau\ge 0$. From \eqref{limgrad} we see $F(\nabla u)$ vanishes at infinity by \eqref{limgrad}, then $\Omega_{\beta}$ is a bounded, open set such that $\overline\Omega_{\beta}\cap \de\Omega=\emptyset$ and $F(\nabla u)=\beta$ on $\de\Omega_{\beta}$.
In order to prove \eqref{maxgrad}, we will prove that $\Omega_{\beta}=\emptyset$.
Note that in $\overline{\O_\b}$, $\nabla u\neq 0$ and hence $u\in C^\infty(\overline{\O_\b})$.
By writing $$G(\xi)=\frac{1}{2}F^2(\xi),$$ the equation in \eqref{p_prob} becomes
\begin{align}\label{eq-G}
\dive\left( G^{\frac p 2-1}(\nabla u) G_\xi(\nabla u) \right) =G^{\frac p2}(\nabla u).
\end{align}
Hereafter we will adopt the Einstein convention on the repeated indices, and use the notations
\begin{align*}
&G=G(\nabla u),\quad G_{i}=G_{\xi_{i}}(\nabla u),\quad G_{ij}=G_{\xi_{i}\xi_{j}}(\nabla u),\\&\quad u_{i}=u_{x_{i}},\quad u_{ij}=u_{x_{i}x_{j}}, \cdots
\end{align*}
Differentiating \eqref{eq-G} with respect to ${x_{i}}$, we get
\[
\de_{x_{k}}\left(\de_{x_{i}}[G^{\frac p2-1} G_{k}]\right)=\frac p2 G^{\frac p2 -1}G_{j}u_{ij},
\]
and
\[
\de_{x_{k}}\left(G_{i}\de_{x_{i}}[G^{\frac p2-1} G_{k}]\right)=\frac p2 G^{\frac p2 -1}G_{i}G_{j}u_{ij}+G_{il}u_{lk}\de_{x_{i}}[G^{\frac p2-1} G_{k}].
\]
Then
\begin{multline*}
\de_{x_{k}}\left(\left[\frac{p-2}{2}G^{\frac p2-2}G_{j} G_{k}+G^{\frac p2 -1}G_{kj}\right]G_{i}u_{ij}\right)=\\=\frac p2 G^{\frac p2 -1}G_{i}G_{j}u_{ij}+\left[\frac{p-2}{2}G^{\frac p2-2}G_{j} G_{k}+G^{\frac p2 -1}G_{kj}\right]G_{il}u_{lk}u_{ij};
\end{multline*}
hence
\begin{multline}
\label{eq3}
\de_{x_{k}}\left(\left[\frac{p-2}{2}G^{\frac p2-2}G_{i}G_{j} G_{k}+G^{\frac p2 -1}G_{i}G_{kj}\right]u_{ij}\right)=\\=\frac p2 G^{\frac p2 -1}G_{i}G_{j}u_{ij}+\left[\frac{p-2}{2}G^{\frac p2-2}G_{j} G_{k}G_{il}u_{lk}u_{ij}+G^{\frac p2 -1}G_{kj}G_{il}u_{lk}u_{ij}\right]=
\\=
\frac p2 G^{\frac p2 -1}G_{i}G_{j}u_{ij}+\frac{p-1}{2}G^{\frac p2-2}G_{j} G_{k}G_{il}u_{lk}u_{ij} +\\+
G^{\frac p2-2}\left[-\frac{1}{2}G_{j} G_{k}G_{il}u_{lk}u_{ij}+G\,G_{kj}G_{il}u_{lk}u_{ij}\right].
\end{multline}
The Kato type inequality (see \cite[Lemma 2.2]{wx-pac}) implies that
\begin{equation*}
G G_{il}G_{jk}u_{ij}u_{kl}\ge \frac{1}2 G_{il}G_j G_k u_{ij}u_{kl}.
\end{equation*}
Hence, from \eqref{eq3} we get
\begin{multline*}
\de_{x_{k}}\left(\left[\frac{p-2}{2}G^{\frac p2-2}G_{i}G_{j} G_{k}+G^{\frac p2 -1}G_{i}G_{kj}\right]u_{ij}\right)\ge \\ \ge
\frac p2 G^{\frac p2 -1}G_{i}G_{j}u_{ij}+\frac{p-1}{2}G^{\frac p2-2}G_{j} G_{k}G_{il}u_{lk}u_{ij}.
\end{multline*}
The above inequality can be read as
\begin{multline*}
\dive\left[ G^{\frac p2-1}\left(G_{\xi\xi} \nabla_x G +\frac{p-2}{2} \frac{(G_\xi\cdot \nabla_x G)}{G}G_\xi \right) \right]
-\frac{p}{2}G^{\frac p2-1}( G_\xi\cdot\nabla_x G) \ge \\ \ge \frac{p-1}{2}( G_{\xi\xi} \nabla_x G)\cdot\nabla_x G,
\end{multline*}
that is
\begin{equation}
\label{equat}
\dive \left[ G^{\frac p2-1} A \nabla_x G \right]-\frac{p}{2} G^{\frac p2-1} G_\xi \cdot \nabla_x G \ge \frac{p-1}{2} G^{\frac{p}{2}-2} \left(G_{\xi\xi}\nabla_x G\right) \cdot \nabla_x G \ge 0,
\end{equation}
where
\[
A=G_{\xi\xi}+\frac{p-2}{2}\frac{G_\xi\otimes G_\xi}{G}
\]
is a uniformly positive definite matrix. Hence, the functional in the left-hand side of \eqref{equat} can be seen as a linear elliptic operator acting on $G(\nabla u)$, and by the maximum principle we have that
\[
\sup_{\Omega_{\beta}}G(\nabla u) \le \sup_{\de \Omega_{\beta}}G(\nabla u) = \frac {\beta^{2}}{2}.
\]
This implies that $\Omega_{\beta}=\left\{x\in\Omega\colon G(\nabla u)>\frac{\beta^{2}}{2}\right\}$ is empty, and the proof is completed.
\end{proof}
\begin{proof}[Proof of Theorem \ref{theoremapprox}]
We are remained to prove the boundary gradient estimate \eqref{estcurv2}. The global gradient estimate \eqref{estcurv1} follows from Lemma \ref{sperb} and \eqref{estcurv2}.
%By using the homogeneity of $F$, it is immediate to see that the function $u_{p}:=(1-p)\log v_{p}$ is a weak solution of \eqref{p_prob}. Arguing as in \cite{m}, let us
Let $x\in \de \Omega$ such that $\mathcal W_r(x_0) \subset \Omega^c$.
%By Proposition \ref{arm}, we have
%\begin{equation}
%\label{confr1}
%u_{p}(x) \le (n-p) \log \left(\displaystyle \frac{F^o(x-x_0)}{r}\right), \quad x \in \Omega.
%\end{equation}
Since $u_p=0$ on $\de \O$, hence if $\nabla u_p(x)\neq 0$, then $\nu_F(x)=F_\xi(\nabla u_p(x))$ and $F(\nabla u_p(x))=\frac{\de u_p}{\de \nu_F}(x)$.
On the other hand, since $W_r(x_0)$ and $\de \O$ are tangent at $x$, we see $x-x_0= r\nu_F(x)$.
It follows that
\begin{eqnarray*}
F(\nabla u_p(x))&=&\frac{\de u_p}{\de \nu_F}(x)=\lim_{t\to 0}\frac{u_p(x+t\nu_F)}{t}\\&\leq &(n-p)
\lim_{t\to 0}\frac{\log F^0(x+t\nu_F-x_0)-\log r}{t}
\\&=&\frac{n-p}{r}.
\end{eqnarray*}
Thus if we define
\begin{equation}
\label{R}
R:=\sup\{r>0 \colon \forall x \in \de \Omega, \exists \mathcal W_r(x_0) \subset \Omega^c \text{ such that } x \in \de \mathcal W_r(x_0)\}.
\end{equation}
then \begin{equation*}\|F(\nabla u_{p})\|_{L^{\infty}(\de \Omega)}\le \frac{n-p}{R}.\end{equation*}
It follows from Lemma \ref{sperb} that
\begin{equation}
\label{estapp}
\|F(\nabla u_{p})\|_{L^{\infty}(\overline\Omega)}\le \frac{n-p}{R}.
\end{equation}
%in the sense that
%\[
%\int_{\Omega} F(\nabla u_{p})^{p-1}F_{\xi}(\nabla u_{p})\cdot \nabla\psi dx =
%\int_{\Omega} F(\nabla u_{p})^{p} \psi dx,
%\]
%\textcolor{red}{for all $\psi \in W^{1,p}(\Omega)$ with compact support in $\Omega$}.
%and if $ \Omega^c \subset\mathcal W_s(y_0)$ then
%\begin{equation}
%\label{confr2}
%u_{p}(x) \ge (n-p) \log \left(\displaystyle \frac{F^o(x-y_0)}{s}\right), \quad x \in \Omega\setminus\{y_0\}.
%\end{equation}
%Moreover by definition of $u$ and \eqref{limgrad}, we have
%\begin{equation}
%\label{lim}
%\lim_{|x| \to \infty} F(\nabla u_{p}) =0
%\end{equation}
%By definition of $R$ and \eqref{confr1}, we get the estimate
%%\[
%F(\nabla u_{p}) \le \frac{n-p}{R} \text{ on } \de \Omega.
%\]
% Hence by Lemma \ref{sperb} we get the desired estimate \eqref{estapp}.
Next prove the estimate \eqref{estcurv2}. We argue as in \cite{HI1,kn}.
Let $\eps>0$. Choose $\bar w\in C^\infty(\overline{\O})$ such that
\begin{itemize}
\item [i)]$
\bar w=0 \text{ on } \de \Omega$ and $\bar w>0$ in $\Omega$;
\item [ii)]$
H_F^+ < F(\nabla\bar w)\le H_F^++\eps \text{ on } \de \Omega.$
\end{itemize}
Denote
\begin{equation}
\label{Q_p}
\mathcal Q_p[\varphi]:=\dive\left( F^{p-1}(\nabla \varphi)F_{\xi}(\nabla \varphi) \right) - F(\nabla \varphi)^{p}.
\end{equation}
Since $F(\nabla\bar w) >0$ and $\bar w=0$ on $\de \O$, the anisotropic mean curvature of $\de \O$ is given by
\begin{equation*}
H_F(x)= \dive \left(F_\xi(\nabla \bar w)\right). \end{equation*}
Thus
\[
\mathcal Q_1[\bar w](x)= H_F(x)-F(\nabla \bar w(x)) <0 \hbox{ for } x\in \de \O.
\]
Let $\delta >0$ and denote by $U_{\delta}$ the components of the set $\{0 \le \bar w < \delta \} $ containing $\de \Omega$. If we choose $\delta>0$ small enough,
we may have $F(\nabla\bar w) >0$ and $\mathcal Q_1[\bar w]<0$ in $U_\delta$.
Define $w\in C^\infty(U_\delta)$ by \[
w=\displaystyle\frac{\bar w}{1-\frac{\bar w}{\delta}}.
\]
Then \[
\nabla w=\frac{\nabla \bar w}{(1-\frac{\bar w}{\delta})^2}.
\]
A simple computation gives
\[
\mathcal Q_1[ w]= \mathcal Q_1[\bar w]+ \left( 1-\frac{1}{\left( 1-\frac{\bar w}{\delta}\right)^2}\right) F(\nabla \bar w) <0 \quad \text{ on } U_{\delta}.
\]
By \eqref{estapp}, there exists a constant $C=C(\delta)>0$, such that $u_{p}\le C$ in $U_{\delta}$. Denoted by $\tilde{U}_{C+1} $ the component of the set $\{0 \le w \le C+1\}$ in $U_{\delta}$. Since $u_p=w=0$ on $\de \O$, we see
\begin{equation}
\label{cfront}
u_{p} \le w \quad \text{on}\; \de \tilde{U}_{C+1}.
\end{equation}
In order to compare $u_{p}$ and $w$ in $\overline{\tilde{U}}_{C+1}$, we compute
\[
\mathcal Q_p[ w]= F^{p-1}(\nabla w) \displaystyle \left( \mathcal Q_1[ w]+ (p-1) \frac{ w_{ik}F_{\xi_i}(\nabla w) F_{\xi_k}(\nabla w)}{F(\nabla w)} \right) \text{ in }\tilde{U}_{C+1}.
\]
Since $ \mathcal Q_1[ w]<0$ in $\tilde{U}_{C+1}$, one may choose $p-1$ small enough, depending on $\|\bar w\|_{C^2(\tilde{U}_{C+1})}$, $\inf_{\tilde{U}_{C+1}} F(\nabla \bar w)$ and $C$, such that \[
\mathcal Q_p[ w]<0\hbox{ in } \tilde{U}_{C+1}.
\]
Then since $\mathcal Q_p[u_p]=0>\mathcal Q_p[ w]$ in $\tilde{U}_{C+1}$ and \eqref{cfront}, by the comparison principle, we have \[
u_{p} \le w\hbox{ on }\overline{\tilde{U}}_{C+1}.\]
It follows that
\[
F(\nabla u_{p})= \displaystyle \frac{\de u_{p}}{\de \nu_F} \le \displaystyle \frac{\de w}{\de \nu_F}=\displaystyle \frac{\de \bar w}{\de \nu_F} =F(\nabla \bar w) \le H_F^++\eps \quad \text{on }\de \Omega.
\]
The proof of Theorem \ref{theoremapprox} is completed.
\end{proof}
Now we are ready to prove Theorem \ref{thm-existence}.
\begin{proof}[Proof of the Theorem \ref{thm-existence}]
Let $u_{p}$ be the solution of \eqref{p_prob} in Theorem \ref{theoremapprox}.
Then for any precompact set $K\subset \Omega$, $u_{p}$ is also a minimizer of the functional
\begin{equation}
\label{funcjp}
J^{p}_{w}(\varphi)=\int_{K}\left[\frac{1}{p}F(\nabla \varphi)^{p}+\varphi F(\nabla w)^{p}\right]dx,
\end{equation}
in the sense that
\begin{equation}
\label{defsol}
J^{p}_{u_{p}}(u_{p}) \le J^{p}_{u_{p}}(\varphi),\quad \forall \varphi\in W^{1,p}_{\textrm{loc}}(\Omega)\text{ such that }\varphi=u_{p}\text{ in }\Omega\setminus K.
\end{equation}
Indeed, being $u_{p}-\varphi=0$ outside $K$, and using it a as test function for \eqref{p_prob}, we get that
\begin{align*}
\int_{K} F^{p}(\nabla u_{p})(u_{p}-\varphi)dx&=\int_{K} F^{p-1}(\nabla u_{p})F_{\xi}(\nabla u_{p})\cdot\nabla (\varphi-u_{p})dx
\\&\leq \frac{1}{p}\int_K( F(\nabla \varphi)^p- F(\nabla u_p)^p) dx.
\end{align*}
The inequality above follows from the convexity of $F(\xi)^p$.
On the other hand, from \eqref{th:confr}, we know $u_p$ has uniform upper bound on any compact set in $\O$.
Since we also have uniform global gradient estimate \eqref{estcurv1} for $\nabla u_{p}$, by Arzela-Ascoli's theorem, we get that, there exists a subsequence $p_{k}\to 1^{+}$ and $u\in C^{0,1}_{loc}(\overline{\O})$ such that
\begin{equation}
\label{strongconvunif}
u_{p_{k}}\to u\text{ uniformly in any compact sets of }\overline\Omega.
\end{equation}
If we can show $u$ is a weak solution of \eqref{prob1}, then
since the weak solution of \eqref{prob1} is unique, we will get $u_p\to u$ uniformly convergence in any compact sets of $\overline\Omega$ as $p\to 1$.
Next we show that $u$ is a proper weak solution to problem \eqref{prob1}, in the sense of Definition \ref{defweak}.
To this aim, we need to prove that
\begin{equation}
\label{strongconvgrad}
\left|\nabla u_{p_{k}}\right|^{p_{k}} \to |\nabla u| \quad \text{in }L_{\textit{loc}}^{1}(\Omega)\text{ for a subsequence }p_k\to 1^{+}.
\end{equation}
Indeed from \eqref{strongconvunif} and \eqref{strongconvgrad} we can pass to the limit in \eqref{funcjp} obtaining that
\[
J^{K}_{u_{p_{k}}}(\varphi,p_{k}) \to J^{K}_{u}(\varphi) \quad \text{ and } \quad J^{K}_{u_{p_{k}}}(u_{p_{k}},p_{k}) \to J^{K}_{u}(u)
\]
In order to prove \eqref{strongconvgrad}, we argue as in \cite{m,HI1}.
Let $K \subset \Omega$ be a precompact set and consider the following test function $\psi$ in \eqref{defsol}
\[
\psi= \eta \varphi + (1-\eta)u_{p},
\]
where $\eta \in C^{\infty}(\Omega)$ is a cut-off function such that $0 \le \eta \le 1$, $\eta\equiv 1 $ in $K$, and $\varphi\in C^{0,1}_{\textrm{loc}}(\Omega)$. Then we get
\begin{multline*}
\int_{\supp\eta} \left(\frac{F^{p}(\nabla u_{p})}{p} +\eta\left(u_{p}-\varphi\right)F^{p}(\nabla u_{p})\right)dx \le \\ \le
\frac{1}{p}\int_{\supp\eta} F^{p}\left(\nabla (\eta\varphi+(1-\eta)u_{p})\right)dx \le \\ \le \frac{1}{p}\int_{\supp\eta}
\left[(\varphi-u_{p})F(\nabla\eta)+\eta F(\nabla \varphi)+(1-\eta)F(\nabla u_{p})\right]^{p}dx \le\\ \le
\frac{3^{p-1}}{p}\int_{\supp\eta}
\left[|\varphi-u_{p}|^{p}F^{p}(\nabla\eta)+\eta^{p} F^{p}(\nabla \varphi)+(1-\eta)^{p}F^{p}(\nabla u_{p})\right]dx.
\end{multline*}
Choosing $\varphi=u$, and letting $p_k\to 1^{+}$, we obtain
\[
\limsup_{p_{k}\to 1^{+}} \int_{\Omega} \eta F(\nabla u_{p_{k}})^{p_{k}}dx \le \int_{\Omega} \eta F(\nabla u)\,dx,
\]
which together with Fatou's Lemma gives \eqref{strongconvgrad}.
The properness of $u$ follows directly from \eqref{th:confr}.
The estimate follows from that in Theorem \ref{theoremapprox}.
The proof of Theorem \ref{thm-existence} is completed.
\end{proof}
\begin{thebibliography}{10}
%\bibitem{and}
%B. Andrews.
%\newblock {Volume-preserving Anisotropic Mean Curvature Flow}.
%\newblock {\em Indiana Univ. Math. J.}, 50(2):783--827, 2001.
\bibitem{bp}
G. Bellettini and M. Paolini.
\newblock {Anisotropic motion by mean curvature in the context of {F}insler
geometry}.
\newblock {\em Hokkaido Math. J.}, 25:537--566, 1996.
\bibitem{bpr} G. Bellettini, M. Novaga and G. Riey,
First variation of anisotropic energies and crystalline mean curvature for partitions.
Interfaces Free Bound. 5 (2003), no. 3, 331-356.
\bibitem{bc} C. Bianchini and G. Ciraolo, Wulff shape characterizations in overdetermined anisotropic elliptic problems, preprint.
%\bibitem{bu} H. Busemann.
%\newblock {The Isoperimetric Problem for Minkowski Area}.
%\newblock {\em Amer. J. Math.}, 71:743--762, 1949.
\bibitem{dpgzaa}
F. Della Pietra and N. Gavitone,
{Sharp estimates and existence for anisotropic elliptic problems with general growth in the gradient},
{Zeitschrift f\"ur Analysis und ihre Anwendungen (ZAA)}
{35}, {61-80}
{2016}.
\bibitem{dib} E. DiBenedetto, $C^{1+\alpha}$ local regularity of weak solutions of degenerate elliptic equations, Nonlinear Anal. TMA 7 (8), 827-850, 1983.
\bibitem{Ge} C. Gerhardt, Flow of nonconvex hypersurfaces into spheres. J. Differential Geom. 32 (1990), no. 1, 299-314.
\bibitem{GT}
D. Gilbarg and N. Trudinger.
\newblock {Elliptic partial differential equations of second order}.
\newblock {\em Classics in Mathematics. Springer-Verlag, Berlin}, 2001.
\bibitem{HI1} G. Huisken and T. Ilmanen, The inverse mean curvature flow and the Riemannian Penrose inequality. J. Differential Geom. 59 (3), 353-437, 2001.
\bibitem{kn} B. Kotschwar and L. Ni, Local gradient estimates of $p$-harmonic functions, $1/H$-flow, and an entropy formula. Ann. Sci. \'Ec. Norm. Sup\'er. (4) 42 (2009), no. 1, 1-36.
\bibitem{m} R. Moser, The inverse mean curvature flow and $p$-harmonic functions, J. Eur. Math. Soc. 9, 77-83, 2007.
%\bibitem[R1]{Re1} Reilly, R. C., Variational properties of functions of the mean curvatures for hypersurfaces in space forms. J. Differential Geometry 8 (1973), 465-477.
\bibitem{Re2} R. C. Reilly, The relative differential geometry of nonparametric hypersurfaces. Duke Math. J. 43 (1976), no. 4, 705-721.
\bibitem{t} N. Trudinger, On Harnack type inequalities and their applications to quasilinear elliptic equations, Comm. Pure Appl. Math., 20, 721-747, 1967.
\bibitem{x} C. Xia, Inverse anisotropic mean curvature flow and a Minkowski type inequality, Adv. Math., 315 (2017) 102-129.
\bibitem{X2} C. Xia, On a class of anisotropic problem. PhD Thesis, Albert-Ludwigs University Freiburg, 2012.
\bibitem{Ur} J. Urbas, On the expansion of starshaped hypersurfaces by symmetric functions of their principal curvatures. Math. Z. 205 (1990), no. 3, 355-372.
\bibitem{WX11}
G. Wang and C. Xia.
\newblock {A characterization of the Wulff shape by an overdetermined anisotropic PDE}.
\newblock {\em Arch. Ration. Mech. Anal.}, 199(1): 99--115, 2011.
\bibitem{wx-pac}
G. Wang and C. Xia.
\newblock {An optimal anisotropic Poincar\'e inequality for convex domains}.
\newblock {\em Pacific J. Math}, 258(2):305--326, 2012.
\end{thebibliography}
\end{document}
\section{Anisotropic mean curvatures for level set}
The anisotropic principal curvatures $\k_F$ are defined by the eigenvalues of anisotropic Weingarten map $d\nu_F$, where $\nu_F=F_\xi(\nu)$ is the anisotropic Gauss map. Locally, we can compute the anisotropic second fundamental form as
\begin{eqnarray}\label{h_F}
{h_F}_{\a\b}=\<F_{\xi\xi}(\nu)\p_\a \nu, \p_\b\>
\end{eqnarray}
and the anisotropic Weingarten map as
\[
{h_F}^\a_{\b}=g^{\a\g}\<F_{\xi\xi}(\nu)\p_\b \nu, \p_\g\>.
\]
\begin{prop}
Let $u$ be a $C^2$ function on $\R^n$ and $S_t:=\{x\in \R^n: u(x)=t\}$ be the level set of $u$. Let $\k_F$ be the principal curvatures of $S_t$. Then at the point where $\n u\neq 0$, we have
\begin{eqnarray*}
|\k_F|^2=F_{ij}(\n u)F_{kl}(\n u)u_{ik}u_{jl}.
\end{eqnarray*}
\end{prop}
\begin{proof}
We shall locally work around a point $x_0$ with $u(x_0)=t$ and
$|\nabla u(x_0)| \neq 0$. Without loss of generality we assume that
$u_n(x_0)> 0$.
By the implicit function theorem, $S_t$ can be locally represented as a
graph of a function $f$, i.e.
\[
S_t=(x',f(x')),\quad\quad x'=(x_1,x_2,\ldots,x_{n-1}).
\]
We first represent the anisotropic curvatures in terms of $f$. In the following proof, the Greek indices run through $1\leq
\alpha,\beta,\gamma\leq n-1$, while the Roman indices run through $1\leq i,j,k\leq n$,
A basis of the tangent space is given locally by
\begin{eqnarray}\label{xxeq1}
e_{\alpha}=(0,\ldots,1,\ldots,f_\alpha),
\end{eqnarray}
where 1 is on the $\alpha$-th coordinate. The first fundamental form for the graph of $f$ is
given by
\begin{eqnarray}\label{xxeq2}
g_{\alpha\beta}=\langle e_\alpha,e_\beta\rangle=\delta_{\alpha\beta}+f_\alpha f_\beta.
\end{eqnarray}
Here $\delta_{\alpha\beta}$ denotes the Kronecker symbol. %The
%$(n-1)\times (n-1)$ matrix $(g_{\alpha\beta})$ is the first
%fundamental form (or the metric) in the local coordinates. Its
%inverse is denoted by $(g^{\alpha\beta})$.
One can compute that
\begin{eqnarray}\label{xxeq3}
g^{\alpha\beta}=\delta_{\alpha\beta}-\frac{f_\alpha f_\beta}{1+|\nabla f|^2},
\end{eqnarray}
where $\nabla f=(f_1,f_2,\ldots,f_{n-1})$. The unit normal vector
field is given by
\begin{eqnarray}\label{xxeq4}
\nu=(\nu^1,\cdots,\nu^n)=\frac{1}{\sqrt{1+|\nabla f|^2}}(-\nabla f,1).
\end{eqnarray}
We calculate the first derivative of $\nu$ and obtain
$$\p_\a\nu=(\p_\a \nu^1, \p_\a \nu^2,\ldots, \p_\a \nu^n),$$
where
\begin{equation}\label{nu_alpha^beta}
\p_\a \nu^\beta=-\frac{f_{\beta\alpha}}{\sqrt{1+|\nabla
f|^2}}+\frac{f_\beta f_\gamma f_{\gamma\alpha}}{\sqrt{1+|\nabla
f|^2}^3},
\end{equation}
\begin{equation}\label{nu_alpha^n}
\p_\a \nu^n=-\frac{f_\gamma f_{\gamma\alpha}}{\sqrt{1+|\nabla
f|^2}^3}.
\end{equation}
By the local representation of the anisotropic second fundamental form (\ref{h_F}), using \eqref{xxeq1} and \eqref{xxeq3}, we obtain
\begin{eqnarray*}
{h_F}^\a_\b &=& g^{\a\g}F_{ij}(\nu)\p_\b\nu^i e_\g^j\\
&=& g^{\alpha\g}(F_{i\g}(\nu)+F_{in}(\nu)f_\g)\p_\b\nu^i\\
&=&
\bigg(F_{i\alpha}(\nu)+\frac{(F_{in}(\nu)-F_{i\g}(\nu)f_\g)f_\alpha}{1+|\nabla
f|^2}\bigg)\p_\b \nu^i.
\end{eqnarray*}
From Proposition \ref{property} (ii), %(\ref{F_ij u_j})
we have
$$F_{ij}(\nu)\nu^j=0$$
for any $i$. Hence from \eqref{xxeq4} we know $$F_{in}(\nu)-F_{i\g}(\nu)f_\g=0.$$
Therefore the anisotropic Weingarten map is given by
\begin{equation}\label{h_F1}
{h_F}^\a_\b =F_{i\alpha}(\nu)\p_\b \nu^i.
\end{equation}
We return to the level set of $u$. We know \begin{equation}\label{u(x,f(x))}
u(x',f(x'))=t.
\end{equation}
By taking the first and the second derivative of (\ref{u(x,f(x))}), we
obtain that
\begin{equation}\label{f_alpha}
f_\alpha=-\frac{u_\alpha}{u_n},
\end{equation}
\begin{equation}\label{f_alpha beta}
f_{\alpha\beta}=-\frac{u_{\alpha\beta}+u_{\alpha
n}f_\beta}{u_n}+\frac{u_\alpha(u_{n\beta}+u_{nn}f_\beta)}{u_n^2}.
\end{equation}
It follows from (\ref{nu_alpha^beta}), (\ref{nu_alpha^n}), (\ref{f_alpha}) and
(\ref{f_alpha beta}) that
\begin{equation}\label{nu alpha beta2}
\p_\a \nu^i=\frac{|u_n|}{u_n}\left(\frac{1}{|\nabla
u|}(u_{i\alpha}+u_{in}f_\alpha)-\frac{u_i u_j}{|\nabla
u|^3}(u_{j\alpha}+u_{jn}f_\alpha)\right),
\end{equation}
%\begin{equation}\label{nu alpha n2}
%\p_\a\nu^n=\frac{|u_n|}{u_n}\left(\frac{1}{|\nabla
%u|}(u_{n\alpha}+u_{nn}f_\alpha)-\frac{u_n u_i}{|\nabla
%u|^3}(u_{i\alpha}+u_{in}f_\alpha)\right).
%\end{equation}
From (\ref{h_F1}) and (\ref{nu alpha beta2}), we
have
\begin{eqnarray}\label{H_F2}
{h_F}^\a_\b &=&F_{i\alpha}(\nu)\p_\b \nu^i\nonumber\\
&=&\frac{|u_n|}{u_n}\left(F_{i\alpha}(\nu)\frac{1}{|\nabla
u|}(u_{i\b}+u_{in}f_\b)-F_{i\alpha}(\nu)\frac{u_i u_j}{|\nabla
u|^3}(u_{j\b}+u_{jn}f_\b)\right),
\end{eqnarray}
Note that $\nu=\frac{|u_n|\nabla u}{u_n|\nabla u|}$. Again from
Proposition \ref{property} (ii) %(\ref{F_ij u_j})
we have
$$F_{i \a}(\nu)u_i=0,$$
%$$F_{i\alpha}(\nu)u_\alpha+F_{in}(\nu)u_n=0.$$
Applying this and (\ref{f_alpha}) to (\ref{H_F2})
and noting that $F_{ij}(\nu)=F_{ij}(\frac{|u_n|\nabla u}{u_n|\nabla u|})=|\nabla u|F_{ij}(\nabla u)$, we obtain
\begin{eqnarray*}
{h_F}^\a_\b&=&\frac{|u_n|}{u_n} F_{i\alpha}(\nu)\frac{1}{|\nabla
u|}(u_{i\b}+u_{in}f_\b)\\
&=&\frac{|u_n|}{u_n} F_{i\alpha}(\n u)(u_{i\b}-u_{in}\frac{u_\b}{u_n}).\\
\end{eqnarray*}
Therefore,
\begin{eqnarray*}
|\k_F|^2={h_F}^\a_\b{h_F}^\b_\a&=&F_{i\alpha}(\n u)(u_{i\b}-u_{in}\frac{u_\b}{u_n})F_{j\b}(\n u)(u_{j\a}-u_{jn}\frac{u_\a}{u_n})
\\&= & F_{i\alpha}(\n u)F_{j\b}(\n u)u_{i\b}u_{j\a}-2 F_{i\alpha}(\n u) F_{j\b}(\n u) u_{j\a} u_{in}\frac{u_\b}{u_n}\\&&+ F_{i\alpha}(\n u) F_{j \b}(\n u)u_{in}\frac{u_\b}{u_n} u_{jn}\frac{u_\a}{u_n}\\&= &
F_{i\alpha}(\n u)F_{j\b}(\n u)u_{i\b}u_{j\a}+2 F_{i\alpha}(\n u) F_{j n}(\n u) u_{j\a} u_{in}\\&&+ F_{i n}(\n u) F_{j n}(\n u)u_{in} u_{jn}
\\&=& F_{ij}(\n u)F_{kl}(\n u)u_{ik}u_{jl}.
\end{eqnarray*}
Here in the next-to-last equality we again used $F_{j\b}(\n u) \frac{u_\b}{u_n}=-F_{jn}(\n u)$.
\end{proof}
\begin{cor}
Let $u$ be a $C^2$ function on $\R^n$ and $S_t:=\{x\in \R^n: u(x)=t\}$ be the level set of $u$. Let $\k_F$ be the principal curvatures of $S_t$. Then at the point where $\n u\neq 0$, we have
\[
\s_2(\k_F)=\s_2 (F_{ik}(\n u)u_{kj}).
\]
Here $\s_2(\k_F)$ is the second anisotropic mean curvature and $\s_2 (F_{ik}(\n u)u_{kj})$ is the second elementary symmetric function on $n\times n$ matrix $(F_{ik}(\n u)u_{kj}).$
\end{cor}
\begin{proof}
It is sufficient to notice that $H_F=\s_1(\k_F)= F_{ij}u_{ij}$, as shown in \cite{WX11}. Combining this result with the previous Proposition, we get the assertion.
\end{proof}
\begin{rem}
Similarly, I guess that $\s_k(\k_F)=\s_k (F_{ik}(\n u)u_{kj})$ for $k=1,2, \cdots, n-1$.
\end{rem}
\section{A pointwise identity}
Let $\S$ be a hypersurface in $\R^n$. As in Gilbarg-Trudinger \cite{GT}, Chapter 16, for a $C^1$ function $f$ , we define a vector field on $\S$ by
\begin{eqnarray}
\d f=\n f-\<\n f, \nu\>\nu,
\end{eqnarray}
and $\d_if$ be the $i$-th component of $\d f$, i.e.,
\begin{eqnarray}
\d_if=\p_i f-\p_k f \nu^k \nu^i.
\end{eqnarray}
We call $\d$ the tangent derivative operator on $\S$.
\begin{prop}\label{identity}Let $u$ be a $C^2$ function. Define $g_i=\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u)$ at any point where $\n u\neq 0$. Let $S_t$ be the level set of $u$ and $\d$ be the tangent derivative operator on $S_t$. Let $\k_F$ and $H_F$ be the principal curvatures and the mean curvature of $S_t$ respectively. Then we have
\begin{eqnarray}\label{ident1}
\d_i g_i=\frac{1}{|\n u|}|\k_F|^2+\frac{1}{|\n u|}F_k(\n u)\p_{x_k}H_F.
\end{eqnarray}
\end{prop}
\begin{proof} at any point of $S_t$ where $\n u\neq 0$, we have
$\nu^i=\frac{u_i}{|\n u|}.$
Thus
\begin{eqnarray}
\label{proof ident1}
\d_i g_i=\p_i \left[\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u)\right]-\p_k \left[\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u)\right]\frac{u_ku_i}{|\n u|^2}.
\end{eqnarray}
For simplicity, we omit $\n u$ in $F(\n u)$, i.e., we write $F=F(\n u)$, $F_k=F_k(\n u)$, etc.
\begin{eqnarray}\label{proof ident2}
&&\p_i \left[\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u)\right]=\p_i \left[\frac{1}{|\n u|}F_{ij}F_ku_{kj}\right]\nonumber
\\&=&\frac{1}{|\n u|}\left[F_{ijl} u_{li}F_ku_{kj}+ F_{ij}F_{kl}u_{li}u_{kj}+F_{ij}F_ku_{kji}\right]- \frac{1}{|\n u|^3}u_lu_{li}F_{ij}F_ku_{kj}\nonumber
\\&=&\frac{1}{|\n u|}F_{ij}F_{kl}u_{li}u_{kj}+ \frac{1}{|\n u|}F_k \p_k(F_{ij}u_{ij})-\frac{1}{|\n u|^3}u_lu_{li}F_{ij}F_ku_{kj}.
\end{eqnarray}
On the other hand,
\begin{eqnarray}\label{proof ident3}
&&-\p_k \left[\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u)\right]\frac{u_ku_i}{|\n u|^2}\nonumber
\\&=&-\p_k \left[\frac{1}{|\n u|}\p_{x_j} F(\n u)\right]F_{ij}(\n u)\frac{u_ku_i}{|\n u|^2}-\p_k F_{ij}(\n u) \p_{x_j} F(\n u)\frac{u_ku_i}{|\n u|^3}\nonumber
\\&=& -\p_k F_{ij}(\n u) F_l u_{lj}\frac{u_ku_i}{|\n u|^3}.
\end{eqnarray}
Here we used $F_{ij}(\n u)u_j=0$ for each $i$ so that the first term $-\p_k \left[\frac{1}{|\n u|}\p_{x_j} F(\n u)\right]F_{ij}(\n u)\frac{u_ku_i}{|\n u|^2}$ vanishes.
By taking derivative of $F_{ij}(\n u)u_j=0$ we get
\begin{eqnarray}\label{proof ident4}
\p_kF_{ij}(\n u)u_j=-F_{ij}u_{jk}.
\end{eqnarray}
Using \eqref{proof ident4} in \eqref{proof ident3}, we get
\begin{eqnarray}\label{proof ident5}
&&-\p_k \left[\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u)\right]\frac{u_ku_i}{|\n u|^2} = F_{ij}u_{ik} F_l u_{lj}\frac{u_k}{|\n u|^3}.
\end{eqnarray}
Substituting \eqref{proof ident2} and \eqref{proof ident5} into \eqref{proof ident1}, we have
\begin{eqnarray}\label{proof ident6}
&&\d_i g_i= \frac{1}{|\n u|}F_{ij}F_{kl}u_{li}u_{kj}+ \frac{1}{|\n u|}F_k \p_k(F_{ij}u_{ij}).
\end{eqnarray}
Since $|\k_F|^2=F_{ij}F_{kl}u_{li}u_{kj}$ and $H_F=F_{ij}u_{ij}$, we get the assertion from \eqref{proof ident6}.
\end{proof}
We need the following lemma.
\begin{lem}[Gilbarg-Trudinger Lemma 16.1]
For hypersurface $\S\subset \R^n$, the following integral formula holds: for any $C^1$ function $g$,
\begin{eqnarray}
\int_{\S} \d g d\mu_\S=\int_{\S} g H \nu d\mu_\S,
\end{eqnarray}
where $\d$ is the tangent derivative operator on $\S$, $H$ and $\nu$ are the (usual) mean curvature and the unit normal of $\S$ respectively.
\end{lem}
Assume the level set of $u$ is regular, i.e., $\n u\neq 0$ on $S_t$. By
applying this lemma to the level set of $u$ and the function $g_i$ in Proposition \ref{identity} we get
\begin{eqnarray}\label{GTintegral}
\int_{S_t} \d_i g_i d\mu_{S_t}=\int_{S_t} g_i H\nu^i d\mu_{S_t}=0.
\end{eqnarray}
The last equality holds since $g_i\nu^i=\frac{1}{|\n u|}F_{ij}(\n u)\p_{x_j} F(\n u) \frac{u_i}{|\n u|}=0$.
Combining \eqref{GTintegral} with Proposition \ref{identity}, we have the following integral formula for the regular level set of $u$.
\begin{prop}
Assume the level set of $u$ is regular, i.e., $\n u\neq 0$ on $S_t$. Then
\begin{eqnarray*}
\int_{S_t} \frac{1}{|\n u|}F_k(\n u) \p_k H_F d\mu_{S_t}=-\int_{S_t} \frac{1}{|\n u|}|\k_F|^2 d\mu_{S_t}.
\end{eqnarray*}
\end{prop}
%\thanks{{\bf Acknowledgement.} }
The hypothesis \eqref{strong} on $F$ ensures that the operator
\[
\mathcal Q_{p} u:= \dive \left(\frac{1}{p}\nabla_{\xi}[F^{p}](\nabla u)\right)
\]
is elliptic, hence there exists a positive constant $\gamma$ such that
\begin{equation*}
\frac1p\sum_{i,j=1}^{n}{\nabla^{2}_{\xi_{i}\xi_{j}}[F^{p}](\eta)
\xi_i\xi_j}\ge
\gamma |\eta|^{p-2} |\xi|^2,
\end{equation*}
for any $\eta \in
n\setminus\{0\}$ and for any $\xi\in \Rn$.