EMIS/ELibM Electronic Journals

Outdated Archival Version

These pages are not updated anymore. They reflect the state of 20 August 2005. For the current production of this journal, please refer to http://www.math.psu.edu/era/.


%_ **************************************************************************
%_ * The TeX source for AMS journal articles is the publishers TeX code     *
%_ * which may contain special commands defined for the AMS production      *
%_ * environment.  Therefore, it may not be possible to process these files *
%_ * through TeX without errors.  To display a typeset version of a journal *
%_ * article easily, we suggest that you retrieve the article in DVI,       *
%_ * PostScript, or PDF format.                                             *
%_ **************************************************************************
% Author Package file for use with AMS-LaTeX 1.2
\controldates{21-SEP-2004,21-SEP-2004,21-SEP-2004,21-SEP-2004}
 
\RequirePackage[warning,log]{snapshot}
\documentclass{era-l}
\issueinfo{10}{12}{}{2004}
\dateposted{September 27, 2004}
\pagespan{103}{112}
\PII{S 1079-6762(04)00135-0}
\usepackage{amsmath}
\usepackage{amscd}
\usepackage{amssymb}
\usepackage{amsthm}

\copyrightinfo{2004}{American Mathematical Society}
\revertcopyright


\newtheorem{thm}{Theorem}[section]
\newtheorem*{cor}{Corollary}
\newtheorem{lem}{Lemma}[section]
\newtheorem{prop}{Proposition}[section]
\newtheorem{corol}{Corollary}[section]

\theoremstyle{definition}
\newtheorem*{defn}{Definition}
\newtheorem*{exmp}{Example}


\theoremstyle{remark}
\newtheorem*{rmk}{Remark}
\newtheorem{rmrk}{Remark}[section]

\numberwithin{equation}{section}


\newcommand{\myaa}{\alpha}
\newcommand{\bb}{\beta}
\newcommand{\mygg}{\gamma}
\newcommand{\GG}{\Gamma}
\newcommand{\dd}{\delta}
\newcommand{\DD}{\Delta}
\newcommand{\ee}{\varepsilon}
\newcommand{\zz}{\zeta}
\newcommand{\hh}{\eta}
\newcommand{\QQ}{\Theta}
\newcommand{\kk}{\kappa}
\newcommand{\myll}{\lambda}
\newcommand{\LL}{\Lambda}
\newcommand{\myss}{\sigma}
\newcommand{\ff}{\varphi}
\newcommand{\vP}{\varPhi}
\newcommand{\oo}{\omega}
\newcommand{\OO}{\Omega}

\newcommand{\A}{\mathcal{A}}
\newcommand{\B}{\mathcal{B}}
\newcommand{\D}{\mathcal{D}}
\newcommand{\E}{\mathcal{E}}
\newcommand{\F}{\mathcal{F}}
\newcommand{\G}{\mathcal{G}}
\newcommand{\myH}{\mathcal{H}}
\newcommand{\K}{\mathcal{K}}
\newcommand{\myL}{\mathcal{L}}
\newcommand{\M}{\mathcal{M}}
\newcommand{\N}{\mathcal{N}}
\newcommand{\myO}{\mathcal{O}}
\newcommand{\myP}{\mathcal{P}}
\newcommand{\RR}{\mathcal{R}}
\newcommand{\T}{\mathcal{T}}
\newcommand{\U}{\mathcal{U}}
\newcommand{\W}{\mathcal{W}}
\newcommand{\X}{\mathcal{X}}
\newcommand{\Y}{\mathcal{Y}}
\newcommand{\Z}{\mathcal{Z}}

\newcommand{\BB}{\mathbb{B}}
\newcommand{\C}{\mathbb{C}}
\newcommand{\DB}{\mathbb{D}}
\newcommand{\EP}{\mathbb{E}}
\newcommand{\FF}{\mathbb{F}}
\newcommand{\HH}{\mathbb{H}}
\newcommand{\KK}{\mathbb{K}}
\newcommand{\MP}{\mathbb{M}}
\newcommand{\NN}{\mathbb{N}}
\newcommand{\OP}{\mathbb{O}}
\newcommand{\PP}{\mathbb{P}}
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\rd}{\mathbb{R}^d}
\newcommand{\SP}{\mathbb{S}}
\newcommand{\TT}{\mathbb{T}}
\newcommand{\VV}{\mathbb{V}}
\newcommand{\WW}{\mathbb{W}}
\newcommand{\YY}{\mathbb{Y}}
\newcommand{\ZZ}{\mathbb{Z}}
\newcommand{\AF}{\EuFrak{A}}
\newcommand{\ZF}{\EuFrak{Z}}
\newcommand{\GE}{\EuFrak{E}}
\newcommand{\GF}{\EuFrak{F}}
\newcommand{\LF}{\Eufrak{L}}
\newcommand{\MM}{\EuFrak{M}}
\newcommand{\NF}{\EuFrak{N}}
\newcommand{\PF}{\EuFrak{P}}

\newcommand{\pD}{\partial D}
\newcommand{\prD}{\partial_{reg} D}
\newcommand{\pE}{\partial E}
\newcommand{\pQ}{\partial Q}
\newcommand{\rQ}{\partial_r Q}
\newcommand{\prQ}{\partial_{reg} Q}
\newcommand{\paQ}{\partial Q}
\newcommand{\pO}{\partial O}
\newcommand{\grd}{$^\circ$}
\newcommand{\up}{\uparrow}
\newcommand{\down}{\downarrow}
\newcommand{\ar}{\Longrightarrow}
\newcommand{\dar}{\Longleftrightarrow}
\newcommand{\upaa}{\upuparrows}
\newcommand{\raa}{\rightrightarrows}
\newcommand{\foral}{\qquad\text{ for all } \ }

\newcommand{\from}{\qquad\text{ in } \ }
\newcommand{\qt}{\quad\text}

\newcommand{\Levy}{L\'{e}vy}
\newcommand{\Holder}{H\"{o}lder}
\newcommand{\Ito}{It\^{o}}

\newcommand{\prx}{P_{r,x}}
\newcommand{\pirx}{\Pi_{r,x}}
\newcommand{\pix}{\Pi_x}
\newcommand{\px}{P_x}
\newcommand{\rdo}{\mathbb{R}^{d-1}}
\newcommand{\J}{\mathcal {J}}

\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\Supp}{Supp}
\DeclareMathOperator{\diam}{diam} \DeclareMathOperator{\Exp}{Exp}
\DeclareMathOperator*{\limmed}{lim\,med}
\DeclareMathOperator{\const}{const.}
\DeclareMathOperator{\cp}{Cap} \DeclareMathOperator{\CP}{CP}
\DeclareMathOperator{\grad}{grad}
\DeclareMathOperator{\sign}{sign} \DeclareMathOperator{\SBV}{SBV}
\DeclareMathOperator{\LPT}{LPT} \DeclareMathOperator{\PT}{PT}
\DeclareMathOperator{\Sup}{Sup} \DeclareMathOperator{\Inf}{Inf}
\DeclareMathOperator{\SG}{SG} \DeclareMathOperator{\RG}{RG}
\DeclareMathOperator{\tr}{tr} \DeclareMathOperator{\Tr}{Tr}
\DeclareMathOperator{\Div}{div} \DeclareMathOperator{\Ex}{Ex}
\DeclareMathOperator{\NSD}{NSD} \DeclareMathOperator{\PD}{PD}
\DeclareMathOperator{\dist}{dist}

\begin{document}
\title[An upper bound for positive solutions]{An upper bound for positive 
solutions\linebreak[1] of the equation $\Delta u=u^\alpha$}

\author{S. E. Kuznetsov}
\address{Department of Mathematics,  University of Colorado, Boulder, CO
80309-0395}
\email{Sergei.Kuznetsov@Colorado.edu}
\thanks{Partially supported by the National Science Foundation Grant
DMS-9971009}

\subjclass[2000]{Primary 35J15; Secondary 35J25}

\commby{Mark Freidlin}

\date{April 5, 2004}

\begin{abstract}
In 2002 Mselati proved that every positive solution of the
equation $\Delta u=u^2$ in a bounded domain of class $C^4$ is the
limit of an increasing sequence of moderate solutions. (A solution
is called moderate if it is dominated by a harmonic function.)  As
a part of his proof, he established an  upper bound (in terms of the
capacity of $K$) for solutions vanishing off a compact subset $K$
of $\partial E$. We use a different kind of capacity  (we call it the
Poisson capacity) and we establish in terms of this capacity an
upper bound for solutions of $\Delta u=u^\alpha$ with $1<\alpha\le
2$. This is a part of the program: to classify all positive
solutions of this equation.
\end{abstract}

\maketitle

\section{Introduction}
\subsection{Main result}
Let $E\subset \rd$ be a bounded smooth domain of class $C^4$ in
$\rd$. For $x\in E$, we denote by $\rho(x)$ the distance to the
boundary $\pE$ and by $k(x,y)$  the Poisson kernel in $E$ for the
Laplacian $\Delta$.

Let $\M(S)$ stand for the set of all finite measures on a
measurable space $S$.  For every  $\nu\in\M(\pE)$, we denote by
$h_\nu$  the harmonic function $h_\nu(x)= \int_{\partial E} k(x,y)
\nu(dy)$.

For every $\myaa>1$ and every Radon measure $m$ on $E$, there exists
a Choquet capacity given on  compact subsets of $\pE$ by the
formula
\begin{equation}
\label{1.2} \cp(K)=\sup\limits_{\nu\in\myP(K)}  \E(\nu)^{-1}
\end{equation}
where $\myP(K)$ is the  set of all probability measures on $K$ and
\begin{equation}
\label{1.1} \E(\nu)=\int_Eh_\nu(x)^\myaa m(dx).
\end{equation}
We call $\cp$ the Poisson capacity.

Our goal is to establish the following theorem.

\begin{thm}
\label{main thm}
  Suppose $\cp$ is the Poisson capacity corresponding
to $1<\myaa\le 2$ and the measure $m(dx)=\rho(x)dx$. Let $K$ be a
compact subset of $\partial E$. There exists a constant $C(E)$
depending only on $E$, such that, for every compact $K\subset \pE$
and every solution $u$ of the boundary value problem
\begin{equation}
\label{main}
\begin{cases}\Delta u = u^\alpha \qt{ in } E,\\
u = 0 \qt{ on } \partial E\setminus K,
\end{cases}
\end{equation}
we have
\begin{equation}
\label{1.3} u(x)\le C(E) \rho(x) \dist(x,K)^{-d}
\cp(K)^{1/(\myaa-1)}.
\end{equation}
\end{thm}

\subsection{Equivalent definitions of the Poisson capacity}
Put
\begin{equation}
\label{1.4} \hat Kf(y)=\int_Ef(x)m(dx)k(x,y).
\end{equation}

The following definitions of the Poisson capacity are equivalent
to \eqref{1.2}:
\begin{equation}
\label{1.4a} \cp(K)^{1/\myaa}=\sup\{\nu(K): \nu\in\M(K), \E(\nu)\le
1\}
\end{equation}
and
\begin{equation}
\label{1.5} \cp(K)^{1/\myaa} =\inf\{\|f\|_{\myaa'}:\hat Kf\ge 1 \text{ on } K\}
\end{equation}
where $\myaa'=\myaa/(\myaa-1)$ and $\|f\|_{\alpha'}$ stands for the norm
in $L_{\myaa'}(m)$.

The equivalence of  \eqref{1.4a} and \eqref{1.5} is proved, for
instance, in \cite{Dy02} (see Theorem 5.1 in Chapter 13). To prove
the equivalence of \eqref{1.2} and \eqref{1.4}, we note that
$\nu\in\M(K)$ is equal to $t\mu$ where $t=\nu(K)$ and
$\mu=\nu/t\in\myP(K)$ and
\begin{align*}
\sup\limits_{\nu\in\M(K)}\{\nu(K):\E(\nu)\le1\}&=
\sup\limits_{\mu\in\myP(K)}\sup\limits_{t\ge 0}\{t:t^\myaa\E(\mu)\le 1\}\\
&=\sup\limits_{\mu\in\myP(K)}\E(\mu)^{-1/\myaa}=(\cp(K))^{1/\myaa}.
\end{align*}

\subsection{Notation} We denote by $B_r(x)$ a ball of radius $r$
centered at $x$. Let $H$ be a compact subset of $\partial E
\cap B_r(x)$ and let $\phi$ be a $C^\infty$ function on $E$ such
that $0\le \phi \le 1$. We call $\phi$ an \textit{$(R,x)$-truncating
function for $H$} if $\phi = 0$ in a neighborhood of $H$ and
$\phi(y) = 1$ if $\dist(x, y)\ge R$. We call $\phi$ an
\textit{$R$-localizing function} if $\phi= 0$ in a neighborhood of
$H$ and $\phi(y) = 1$ if $\dist(y,H )\ge R$.


\section{Bounds in a halfspace}
\subsection{}
First, we establish some  bounds in the case when $E=(0,\infty)\times
\R^{d-1}$. A generic element of $E$ is denoted  by $z$ or
$(s,x),s\in (0,\infty), x\in \R^{d-1}$. We use the notation $f(s,x)$
and $f^s(x)$ for functions on $E$.

Denote by $\EP$ an infinite strip $\{(s,x): 0\le s<1\}$. The measure
$m(dz)=1_{[0,1]}(s) s\,ds\,dx$ is concentrated on $\EP$. We denote
by $\|f\|_{\alpha}$ the norm in $L_\alpha(m)$. The Poisson kernel
$k$ can be represented by the formula
\[
k((s,x),y)=C q^s(x-y)
\]
where $C$ is a constant depending only on the dimension, and
\begin{equation}
\label{2.1} q^s(x)=\frac {s} {(|x|^2+s^2)^{d/2}}.
\end{equation}
\begin{lem}
\label{Lemma 3.1.1} Suppose that $z_0\in \partial E$ and $H$ is a
compact subset of $\partial E\cap B_1(z_0)$. If $\cp(H)>0$, then
there exists a $(3/2,z_0)$-truncating function $\beta$ for $H$
such that $\beta^s(x)=1$ for $s\ge 1$ and
\begin{equation}
\label{3.1.1-0} \|\nabla^2 \beta\|^{\alpha'} _{\alpha'} + \|
|\nabla \beta|^2\|^{\alpha'}_{\alpha'}+
\|\nabla\beta\|^{\alpha'}_{\alpha'}+\left\|\frac 1s
\frac{\partial\beta}{\partial s}\right\|^{\alpha'}_{\alpha'} \le
C(d) \cp (H)^{1/(\myaa-1)}
\end{equation}
where the constant $C(d)$ depends only on $d$. If $\cp(H)=0$,
then, for every $\epsilon > 0$, there exists a
$(3/2,z_0)$-truncating function $\beta$ for $H$ such that
$\beta^r(x)=1$ for $r\ge 1$ and
\begin{equation}
\label{3.1.1-0a} \|\nabla^2 \beta\|^{\alpha'} _{\alpha'} + \|
|\nabla \beta|^2\|^{\alpha'}_{\alpha'}+
\|\nabla\beta\|^{\alpha'}_{\alpha'}+\left\|\frac 1s
\frac{\partial\beta}{\partial s}\right\|^{\alpha'}_{\alpha'} <
\epsilon.
\end{equation}
\end{lem}
\begin{proof}  Let $\cp(H)>0$. By \eqref{1.5}, there exists a
function $f$ on $E$ such that $\|f\|_{\alpha'}^{\alpha'} \le 2
\cp(H)^{\alpha'/\alpha}$ and $\hat K f\ge 1$ on $H$. We may assume
that $f\ge 0$ (otherwise we just replace $f$ with $f^+$).

Let $A(t), 0\le t<\infty$, be an increasing $C^2$ function such
that $A(t)=0 $ for $t\le 1$ and $A(t) = 1$ for $t\ge \sqrt{2}$.
We set
\begin{equation}
\label{2.2} (\T f)^t(y) = \int_0^1 s\,ds
A(\sqrt{s/t})\int_{\R^{d-1}} f^s(x) q^s(y-x)\,dx \qt{for } t>0
\end{equation}
and
\begin{equation}
\label{2.3} (\T f)^0(y)= \lim\limits_{t\down 0}(\T f)^t(y)=\hat
Kf(y)
\end{equation}
(cf. \cite[formula (6.3), p.~175]{Dy02}).


By \cite[Theorem 13.6.1]{Dy02}, we have
\[
\|\T f\|_{\alpha'}+\|\nabla\T f\|_{\alpha'}+\|\nabla^2 \T
f\|_{\alpha'}+ \left\|\frac 1s \frac{\partial\T f}{\partial
s}\right\|_{\alpha'}\le  C \|f\|_{\alpha'}.
\]
Let $g(s,y)=a(s)b(y)$ be such that $0\le a,b\le 1$, $a,b$ are
$C^2$ functions, $a=1$ in a neighborhood of $0$, $b=1$ in a
neighborhood of $\partial E\cap B_1(z_0)$ and $g=0$ outside
$B_{3/2}(z_0)$. Let $h(t)$ be an increasing $C^2$ function on
$[0,\infty)$ such that $h(t) = 0$ if $t\le 1/4$ and $h(t) = 1$ if
$t\ge 3/4$. As in the proof of Lemma 13.6.5 from \cite{Dy02}, we
put
\[
u=\T f,\quad v = gu, \quad \phi = h(v)
\]
and, finally,
\[
\beta = 1- \phi = 1- h(g \T f).
\]
By \eqref{2.3}, $\T f=\hat K f \ge 1$ on $H$, and $\beta = 0$ in a
neighborhood of $H$ by the choice of $g$ and $h$. By direct
computation,\footnote{See \cite[pp.~181--182]{Dy02}, or
\cite[Section 3]{Ku98}.} we get
\begin{gather}
\label{2.6}
|v|+ |\nabla v|+ |\nabla^2 v| \le C(|u|+ |\nabla u|+ |\nabla^2 u|),\\
\label{2.7} \left|\frac 1s \frac{\partial v}{\partial s}\right| =
\left(\left|\frac as \frac{\partial u}{\partial s} + \frac {a'}s
u\right|\right)b\le C \left(|u|+ \frac 1s \left|\frac{\partial
u}{\partial s}\right|\right).
\end{gather}
More computation yields
\begin{gather}
\label{2.8}
|\nabla \phi|\le C |\nabla v|,\\
\label{2.9}
|\nabla^2 \phi| \le C \left(|\nabla^2 v| + \frac{|\nabla v|^2}v\right),\\
\label{2.10} |\nabla \phi|^2 \le C \left(\frac{|\nabla
v|^2}v\right).
\end{gather}
Therefore
\[
\|\nabla^2 \beta\| _{\alpha'} + \| |\nabla \beta|^2\|_{\alpha'}+
\|\nabla\beta\|_{\alpha'}+\left\|\frac 1s
\frac{\partial\beta}{\partial s}\right\|_{\alpha'} \le C
\|f\|_{\alpha'} \le C \cp(H)^{1/(\myaa-1)}.
\]


If $\cp(H) = 0$, then $\|f\|_{\alpha'}$ can be made arbitrary
small, and the same construction yields \eqref{3.1.1-0a}.
\end{proof}

\subsection{}
For a set $H\subset R^{d-1}$, we put $\lambda H = \{\lambda x:x\in
H\}$.


\begin{lem}
\label{Lemma A} For every compact set $H\subset R^{d-1}$ and every
$0<\lambda<1$,
\begin{equation}
\label{cap-scale} \cp(\lambda H)^{1/(\myaa-1)} \le
\lambda^{d-2\alpha'+1} \cp(H)^{1/(\myaa-1)}.
\end{equation}
\end{lem}
\begin{proof} Let $\myll>0$ and $\nu \in \myP(H)$.
Then $\nu_\lambda(A) = \nu(A/\lambda)$ is concentrated on $\lambda
H$. Note that $q^{\myll s}(\myll x)=\myll^{d-1}q^s(x)$ and  therefore
\begin{equation}
\label{A-1} h_{\nu}(s,x) = \myll^{d-1} h_{\nu_\myll}(\myll s, \myll x).
\end{equation}
Formula \eqref{A-1} and change of variables $t=\myll s, y=\myll x$
yield
\begin{equation*}
\begin{split}
\E(\nu) &= \int_0^1 \int_{\R^{d-1}} h_\nu^\myaa(s,x) sdsdx \\&=
\int_0^1 \int_{\R^{d-1}} \myll^{(d-1)\myaa} h_{\nu_\myll}^\myaa (\myll s,
\myll x) sdsdx \\&= \int_0^\myll \int_{\R^{d-1}} \myll^{(d-1)\myaa}
\myll^{-(d+1)} h_{\nu_\myll}^\myaa (t, y) tdtdy \\&\le \myll^{(d-1)\myaa}
\myll^{-(d+1)} \E(\nu_\myll)
\end{split}
\end{equation*}
and  \eqref{1.2} implies
\begin{equation}
\label{A-2} \cp(H)\ge \myll^{d+1 - (d-1)\myaa} \cp(\myll H).
\end{equation}
Formula \eqref{cap-scale} follows from \eqref{1.2} because
$d-2\myaa'+1=-[d+1-(d-1)\myaa]/(\myaa-1)$.
\end{proof}
\begin{lem}
\label{Lemma B} Let $z_0\in\partial E, 0<\delta<1$ and let
$\Gamma$ be a compact subset of $\partial E\cap B_{\delta}(z_0)$.
Suppose $\cp(\Gamma)>0$. There exists a
$(3\delta/2,z_0)$-truncating function $\mygg=\mygg_{\Gamma,\delta}$
for $\Gamma$ such that
\begin{equation}
\label{B-0} \|\nabla^2 \mygg\|^{\alpha'} _{\alpha'} + \| |\nabla
\mygg|^2\|^{\alpha'}_{\alpha'}+ \left\|\frac
1\delta\nabla\mygg\right\|^{\alpha'}_{\alpha'}+\left\|\frac 1s
\frac{\partial\mygg}{\partial s}\right\|^{\alpha'}_{\alpha'} \le C(d) \cp
(\Gamma)^{1/{\myaa-1}}
\end{equation}
where the constant $C(d)$ depends only on $d$. If $\cp(\Gamma)=0$,
then the left side of \eqref{B-0} can be made smaller than any
$\ee>0$.
\end{lem}
\begin{proof} Let $H = \Gamma/\dd$ and let $\bb(s,x)$ be the function constructed in Lemma \ref{Lemma 3.1.1} applied to $H$ and $z_0/\dd$.
Put $\mygg(s,x) = \beta(r/\delta,x/\delta)$. Since $\beta(s,x) = 1$
if $s\ge 1$, $\mygg(s,x)= 1$ if $s\ge\dd$. Also,
\[
\nabla\mygg(s,x) = (1/\dd) \nabla \beta(s/\dd,x/\dd),\qquad \nabla^2
\mygg(s,x) = (1/\dd^2) \nabla^2 \beta(s/\dd,x/\dd)
\]
and therefore
\[
\begin{split}
\|\nabla \mygg\|_{\myaa'}^{\myaa'} &= \int_0^\dd \int_{\R^{d-1}} |\nabla
\mygg(s,x)|^{\myaa'}\, s\,ds\,dx\\
&= \int_0^\dd \int_{\R^{d-1}} \dd^{-2\myaa'} |\nabla \bb(s/\dd,x/\dd)|^{\myaa'}\, s\,ds\,dx\\
&=\int_0^1 \int_{\R^{d-1}} \dd^{d+1-2\myaa'} |\nabla
\bb(s,x)|^{\myaa'}\, s\,ds\,dx = \dd^{d+1-2\myaa'} \|\bb\|_{\myaa'}^{\myaa'}.
\end{split}
\]
In a similar way,
\[
\| |\nabla \mygg|^2\|^{\alpha'}_{\alpha'} = \dd^{d+1-2\myaa'}\|
|\nabla \bb|^2\|^{\alpha'}_{\alpha'},\qquad \left\|\frac
1\delta\nabla\mygg\right\|^{\alpha'}_{\alpha'} = \dd^{d+1-2\myaa'}
\|\nabla\bb\|^{\alpha'}_{\alpha'}
\]
and
\[
\left\|\frac 1s \frac{\partial\mygg}{\partial
s}\right\|^{\alpha'}_{\alpha'} = \left\|\frac 1s
\frac{\partial\bb}{\partial s}\right\|^{\alpha'}_{\alpha'}.
\]
Therefore \eqref{B-0} follows from \eqref{3.1.1-0} and Lemma
\ref{Lemma A}.
\end{proof}

\section{Bounds in a unit ball}
\subsection{}
Now let $E$ be a ball of radius 1 in
$\rd$ centered at a point $z_0$ with coordinates $s=1,
x=0$. As before, let $\EP =\{(s,x): 0\le s < 1\}$. For a point
$z=(s,x)\in\EP$, we denote by
\[
\phi(z) = x/(1-s)
\]
a projection of $z$ to $\R^{d-1}$ with center at $z_0$. For a
point $z\in E\cap\EP$, we put
\[
\psi(z) = (1-|z-z_0|, \phi(z))
\]
(cf. \cite{Ms02}, Section 3.1.1). The mapping $\psi$ defines a 1-1
correspondence between $E\cap \EP$ and $\EP$.

For a set $H\subset \pE$, denote by $\cp_E(H)$ the Poisson
capacity of $H$ with respect to the domain $E$ and the measure
$m(dz) = \dist(z, \pE) \,dz$. For a set $K\subset \R^{d-1}$, we
denote by $\cp_{\EP}(K)$ the Poisson capacity with respect to the
halfspace and the measure $m(ds,dx) = 1_{[0,1)}(s)\,s\,ds\,dx$.

\begin{lem}
\label{Lemma C} Let $H$ be a compact subset of $\pE$ that is contained in
a ball of radius $1/4$ centered at zero, and let $K =
\psi(H)$. There exists a constant $C$ depending only on the
dimension, such that
\[
C^{-1} \cp_{\EP}(K) \le \cp_E(H) \le C \cp_{\EP}(K).
\]
\end{lem}

\begin{proof} Let $\mu$ be a probability measure on $H$ and let $\nu$ be a measure on $K$ defined by the formula $\nu(\Gamma) = \mu(\psi^{-1}(\Gamma))$. It is enough to show that
\begin{equation}
\label{C-0} C^{-1} \E_E(\mu) < \E_{\EP}(\nu)0$. There exists a $(3\delta/8,0)$-truncating function
$\gamma$ such that
\begin{equation}
\label{s3.1.2-0} \|\nabla^2 \gamma\|^{\alpha'} _{\alpha'} + \|
|\nabla \gamma|^2\|^{\alpha'}_{\alpha'}+ \left\|\frac
1\delta\nabla\gamma\right\|^{\alpha'}_{\alpha'}+\left\|\frac 1\rho
\frac{\partial\gamma}{\partial \rho}\right\|^{\alpha'}_{\alpha'}
\le C(d) \cp_E (H)^{1/(\myaa-1)},
\end{equation}
where the constant $C(d)$ depends only on $d$. If $\cp_E(H)=0$,
then the left side of \eqref{s3.1.2-0} can be made smaller than
any $\ee>0$.
\end{lem}

\begin{proof} We apply Lemma \ref{Lemma B} to the set $K=\psi(H)$. 
Let $\gamma_K$ be the function constructed in Lemma  \ref{Lemma B}. We put
$\gamma(s,x) = \gamma_K(\psi(s,x))$ if $s<1$, and $\gamma(s,x)=0$
otherwise. Similarly to the proof of \cite[Sublemma 3.1.2]{Ms02},
we show that
\begin{equation}
\label{s3.1.2-2}
\begin{split}
\|\nabla^2 \gamma\|^{\alpha'} _{E,\alpha'} &+ \| |\nabla
\gamma|^2\|^{\alpha'}_{E,\alpha'}+ \left\|\frac
1\delta\nabla\gamma\right\|^{\alpha'}_{E,\alpha'}+\left\|\frac
1\rho \frac{\partial\gamma}{\partial
\rho}\right\|^{\alpha'}_{E,\alpha'} \\&\le \|\nabla^2
\gamma_K\|^{\alpha'} _{\EP, \alpha'} + \| |\nabla
\gamma_K|^2\|^{\alpha'}_{\EP, \alpha'}+ \left\|\frac
1\delta\nabla\gamma_K\right\|^{\alpha'}_{\EP,
\alpha'}+\left\|\frac 1s \frac{\partial\gamma_K}{\partial
s}\right\|^{\alpha'}_{\EP, \alpha'}
\end{split}
\end{equation}
where $\|\cdot\|_E$ and $\|\cdot\|_{\EP}$ stand for
$L_{\myaa'}$-norms in $E$ and $\EP$. Finally, we apply Lemma
\ref{Lemma C}.
\end{proof}

\subsection{Localizing functions} Let $H$ be a subset of $\pE$ and let $\gamma$ be a $C^2$-function on $E$.
We call $\gamma$ an $\ee$-localizing function for $H$ if $0\le
\gamma\le 1$, $\gamma = 1$ in a neighborhood of $H$ and $\gamma(z)
= 0$ if $\dist(z,\myH)>\ee$.
\begin{lem}
\label{Lemma 3.1.2} There exists a constant $C(d)$ such that, for
every compact subset $K$ of $\partial E$ with $\cp(K)>0$ and
$\diam (K) \le 4\delta$, there exists a $\delta/2$-localizing
function $\gamma=\gamma_{\delta, K}$ for $K$ such that
\begin{equation}
\label{3.1.2-0} \|\nabla^2 \gamma\|^{\alpha'} _{\alpha'} + \|
|\nabla \gamma|^2\|^{\alpha'}_{\alpha'}+ \left\|\frac
1\delta\nabla\gamma\right\|^{\alpha'}_{\alpha'}+\left\|\frac 1\rho
\frac{\partial\gamma}{\partial \rho}\right\|^{\alpha'}_{\alpha'}
\le C(d) \cp(K)^{1/(\alpha-1)}.
\end{equation}
\end{lem}

\begin{proof} As in \cite[Lemma 3.1.2]{Ms02}, we cover the set $K$ by
finitely many balls $B_{\delta/8}(y_k)$ (the number $n$ of
the balls depends only on the dimension $d$). We apply Lemma
\ref{sublemma 3.1.2} to each of the sets $H_k = K\cap
B_{\delta/8}(y_k)$. Denote by $\gamma_k$ the corresponding
truncating function constructed in Lemma \ref{sublemma 3.1.2} (we
choose $\epsilon = \cp(\Gamma)^{1/(\alpha-1)}$ if $\cp(H_k)=0$ for
some $k$). We set
\[
\gamma = \gamma_1 \cdots \gamma_n.
\]
Note that
\[
|\nabla\gamma| \le \sum_k |\nabla \gamma_k|
\]
and
\[
\begin{split}
|\nabla^2\gamma| &\le \sum_k |\nabla^2 \gamma_k|+ \sum_{k\ne l}
|\nabla \gamma_k|\, |\nabla \gamma_l| \\
&\le \sum_k |\nabla^2 \gamma_k|+ \frac {(n-1)}2 \sum_{k} |\nabla
\gamma_k|^2.
\end{split}
\]
By applying Minkowski inequality, we get
\begin{equation}
\label{3.1.2-1}
\begin{split}
\|\nabla^2 \gamma\|^{\alpha'} _{\alpha'} &+ \| |\nabla
\gamma|^2\|^{\alpha'}_{\alpha'}+ \left\|\frac
1\delta\nabla\gamma\right\|^{\alpha'}_{\alpha'}+\left\|\frac 1\rho
\frac{\partial\gamma}{\partial \rho}\right\|^{\alpha'}_{\alpha'}\\
&\le C(d,n) \sum_k \cp (H_k)^{1/(\alpha-1)} + n C(n)\epsilon
\\&\le n (C(d,n)+ C(n)) \cp (\Gamma)^{1/(\alpha-1)}.
\end{split}
\end{equation}
\end{proof}

\begin{lem}
\label{Sublemma 3.1.4} Let $K$ be a compact subset of $\partial E$
such that $\diam(K)\le 4\delta$ and $\cp(K)>0$. Let $\gamma$ be
the $\delta/2$-localizing function constructed in Lemma \ref{Lemma
3.1.2}. Then
\begin{equation}
\label{3.8}
\begin{split}
\int_E u \gamma^{2\alpha'-1} |\Delta \gamma| \rho &\le C 
\left(\int_E u^\alpha \gamma^{2\alpha'}\rho\right)^{1/\alpha} 
\cp(K)^{1/\alpha},\\
\int_E u \gamma^{2\alpha'-2} |\nabla \gamma|^2 \rho &\le C \left(\int_E 
u^\alpha \gamma^{2\alpha'}\rho\right)^{1/\alpha} \cp(K)^{1/\alpha},\\
\int_E u \gamma^{2\alpha'-1} |\nabla \gamma| \rho &\le C \left(\int_E 
u^\alpha \gamma^{2\alpha'}\rho\right)^{1/\alpha}\delta \cp(K)^{1/\alpha},\\
\int_E u \gamma^{2\alpha'-1} \left|\frac{\partial \gamma}{\partial
\rho}\right|  &\le C \left(\int_E u^\alpha
\gamma^{2\alpha'}\rho\right)^{1/\alpha} \cp(K)^{1/\alpha}
\end{split}
\end{equation}
whenever $u$ satisfies \eqref{main}.
\end{lem}

\begin{proof} The assertion
follows from the H\"older inequality, Lemma \ref{Lemma 3.1.2},
the identity $(\alpha-1)\alpha' = \alpha$  and the inequality
$\gamma^{(2\alpha'-1)\alpha} \le
\gamma^{(2\alpha'-2)\alpha}=\gamma^{2\alpha'}$. For instance, for
the last line in \eqref{3.8}, we have
\[
\begin{split}
\int_E u \gamma^{2\alpha'-1} \left|\frac{\partial \gamma}{\partial
\rho}\right| &= \int_E u \gamma^{2\alpha'-1} \left|\frac 1\rho
\frac{\partial \gamma}{\partial \rho}\right|\rho \\&\le \left(
\int_E u^\alpha
\gamma^{(2\alpha'-1)\alpha}\rho\right)^{1/\alpha}\left( \int_E
\left|\frac 1\rho \frac{\partial \gamma}{\partial
\rho}\right|^{\alpha'}\rho \right)^{1/\alpha'} \\&\le \left(\int_E
u^\alpha \gamma^{2\alpha'}\rho\right)^{1/\alpha} C(d)
\cp(K)^{1/[\alpha'(\alpha-1)]}.
\end{split}
\]
\end{proof}


\begin{lem}
\label{Lemma 3.1.3} Let $K, \gamma, u$ be as in Lemma
\ref{Sublemma 3.1.4}. There exists  a constant $C(d)$ such that
\begin{equation}
\label{3.1.3-0} \int_E u^\alpha \gamma^{2\alpha'} \rho \le C(d)
\cp(K)^{1/(\alpha-1)}
\end{equation}
whenever $u$ satisfies \eqref{main}.
\end{lem}
\begin{proof} This is an adaptation of Lemma 3.1.3 in \cite{Ms02}.
Let $E_s = B(s, z_0)$ and $r=|z-z_0|$. By replacing $u^2$ and
$\gamma^4$ with $u^\alpha$ and $\gamma^{2\alpha'}$ in the arguments
of \cite{Ms02}, we get a bound
\begin{equation}
\label{Ms-3.3.7}
\begin{split}
\int_E u^\alpha \gamma^{2\alpha'} \rho \le &\int_E u \Delta
(\gamma^{2\alpha'}(1-r^2))- 4 \int_E u \frac{\partial
(\gamma^{2\alpha'})}{\partial r} r \\&+ \liminf_{s\to
1-}\int_{\partial E_s} \frac{\partial}{\partial r}
(u\gamma^{2\alpha'}(1-r^2)).
\end{split}
\end{equation}
As in \cite[Sublemma 3.1.3]{Ms02}, one can show that the last term
in \eqref{Ms-3.3.7} is negative and can be dropped.

Finally, we note that $1-r^2\le 2\rho$ and therefore
\begin{equation}
\label{Ms-3.3.9}
\begin{split}
&\left|\int_E u \Delta (\gamma^{2\alpha'}(1-r^2))\right| \\
&\qquad\le 2\alpha' \int_E u \gamma^{2\alpha'-1} |\Delta \gamma|
(1-r^2)+
2\alpha'(2\alpha'-1) \int_E u \gamma^{2\alpha'-2} |\nabla \gamma|^2 \rho \\
&\qquad \le 4\alpha' \int_E u \gamma^{2\alpha'-1} |\Delta \gamma|
(1-r^2)+
4\alpha'(2\alpha'-1) \int_E u \gamma^{2\alpha'-2} |\nabla \gamma|^2 \rho \\
&\qquad \le C \left(\int_E u^\alpha
\gamma^{2\alpha'}\rho\right)^{1/\alpha} \cp(K)^{1/\alpha}.
\end{split}
\end{equation}
In a similar way,
\begin{equation}
\label{Ms-3.3.10}
\begin{split}
\left|\int_E u \frac{\partial (\gamma^{2\alpha'})}{\partial r} r\right| &= 2\alpha'\left|\int_E u \gamma^{2\alpha'-1}\frac{\partial \gamma}{\partial r} r\right| \le 2\alpha'\left|\int_E u \gamma^{2\alpha'-1}\frac{\partial \gamma}{\partial r} \right| \\
&\le C \left(\int_E u^\alpha
\gamma^{2\alpha'}\rho\right)^{1/\alpha} \cp(K)^{1/\alpha}.
\end{split}
\end{equation}
From \eqref{Ms-3.3.7}, \eqref{Ms-3.3.9} and \eqref{Ms-3.3.10}, we
get
\[
\int_E u^\alpha \gamma^{2\alpha'} \rho \le C \left(\int_E u^\alpha
\gamma^{2\alpha'}\rho\right)^{1/\alpha} \cp(K)^{1/\alpha},
\]
which implies \eqref{3.1.3-0}.
\end{proof}

Combining \eqref{3.8} with Lemma \ref{Lemma 3.1.3}, we get:
\begin{lem}
\label{Lemma 3.1.4} Let $K,\gamma,u$ be as in Lemma \ref{Sublemma
3.1.4}. Then
\begin{equation}
\label{3.8bis}
\begin{split}
\int_E u \gamma^{2\alpha'-1} |\Delta \gamma| \rho &\le C  
\cp(K)^{1/(\alpha-1)},\\
\int_E u \gamma^{2\alpha'-2} |\nabla \gamma|^2 \rho &\le C  
\cp(K)^{1/(\alpha-1)},\\
\int_E u \gamma^{2\alpha'-1} |\nabla \gamma| \rho &\le C \delta 
\cp(K)^{1/(\alpha-1)},\\
\int_E u \gamma^{2\alpha'-1} \left|\frac{\partial \gamma}{\partial
\rho}\right|  &\le C  \cp(K)^{1/(\alpha-1)}.
\end{split}
\end{equation}
\end{lem}

\subsection{}
The rest of the proof of Theorem \ref{main thm} is very close to
the corresponding part of the proof of
\cite[Theorem 3.1.1]{Ms02}. We begin with
\begin{lem}
\label{Lemma 3.1.6} Let $K,\gamma,u$ be as in Lemma \ref{Sublemma
3.1.4}, and let $G_E$ be the Green operator of $E$. For every
$y \in E, \beta >0$,
\[
\gamma^\beta(y)u(y) \le \frac12 G_E\left(\gamma^\beta \Delta u -
\Delta (\gamma^\beta u)\right)(y).
\]
\end{lem}

This is a version of \cite[Lemma 3.1.6]{Ms02} (in \cite{Ms02},
$\beta = 4$). Bounds from \cite[Sublemma 3.1.5]{Ms02} must be
replaced with those of \cite[Theorem 7.1]{Dy02}. Other
modifications are obvious.

As a first step, we establish
\begin{lem}
\label{Lemma 3.1.5} There exists a constant $C(d)$ such that the
inequality \eqref{1.3} holds whenever
\[
\dist(x,K) \ge \frac{\diam(K) }4.
\]
\end{lem}

\begin{proof} The proof is an appropriate modification of the proof of
\cite[Lemma 3.1.5]{Ms02}.
Let $\delta = \dist(x,K)$ and let $\gamma$ be the
$\delta/2$-localizing function constructed in Lemma \ref{Lemma
3.1.2}. Clearly, $\gamma=1$ in a neighborhood of $x$ and therefore
\begin{equation}
\label{3.11} u(x) = \gamma^{2\alpha'}(x)u(x) \le \frac12
G_E(\gamma^{2\alpha'} \Delta u - \Delta(\gamma^{2\alpha'}u))(x)
\end{equation}
by Lemma \ref{Lemma 3.1.6}.  As in \cite{Ms02}, the right side can
be evaluated by means of Green's formula applied to the domain
$B(z_0,r)\setminus B(x,\ee)$ and passage to the limit as $\ee \to
0, r\to 1$. Namely, we get
\[
\begin{split}
G_E(\gamma^{2\alpha'} \Delta u &- \Delta(\gamma^{2\alpha'}u))(x)
\\ &= \int_E (2 \nabla_y g_E(x,y)\nabla (\gamma^{2\alpha'}(y)
- g_E(x,y) \Delta (\gamma^{2\alpha'})(y))u(y)\,dy ).
\end{split}
\]
Together with \eqref{3.11}, this implies
\begin{equation}
\label{3.13}
\begin{split}
u(x)&\le  2\alpha'\int_E u(y) \gamma^{2\alpha'-1}(y) \nabla \gamma(y) \nabla_y g_E(x,y)\,dy \\
& \quad +\alpha' \int_E u(y) \gamma^{2\alpha'-1}(y) \Delta \gamma(y) g_E(x,y) \,dy \\
&\quad +\alpha'(2\alpha'-1) \int_E u(y) \gamma^{2\alpha'-2} (y) |
\nabla \gamma(y)|^2 g_E(x,y)\,dy
\end{split}
\end{equation}
(cf. \cite[(3.13)]{Ms02}). Now, $\gamma(y)=1$ for all $y$ such
that $\dist(y,K)>\delta/2$, in particular for all $y$ such that
$|x-y|<\delta/2$, and therefore the integrands are equal to $0$ for
such $y$. Following \cite{Ms02}, from the bounds for the Green's
function and its gradient, we get bounds for the integrals on the
right side of \eqref{3.13} in terms of integrals
\eqref{3.8bis}. For instance,
\[
g_E(x,y) \le C \rho(x)\rho(y) |x-y|^{-d},
\]
and therefore
\[
\begin{split}
{}&\int_E u(y) \gamma^{2\alpha'-2} (y) | \nabla \gamma(y)|^2
g_E(x,y)\,dy \\&\qquad\le C \rho(x) \int_E u(y) \gamma^{2\alpha'-2} (y) |
\nabla \gamma(y)|^2 \rho(y) |x-y|^{-d}\,dy \\ &\qquad=\int_{E\setminus
B(x,\delta/2)} u(y) \gamma^{2\alpha'-2} (y) | \nabla \gamma(y)|^2
\rho(y) |x-y|^{-d}\,dy
\\ &\qquad\le C \rho(x) \delta^{-d}\int_E u \gamma^{2\alpha'-2}(y) |\nabla \gamma|^2(y) \rho
(y)\,dy.
\end{split}
\]
In a similar way, we get
\begin{equation} 
\label{3.17}
\begin{split}
{}&\int_E u(y) \gamma^{2\alpha'-1}(y) \nabla \gamma(y) \nabla_y g_E(x,y)\,dy\\
&\qquad\le C \rho(x) \delta^{-d} \int_E u \gamma^{2\alpha'-1} 
\left|\frac{\partial \gamma}{\partial \rho}\right|  + C\rho(x) 
\delta^{-d-1}\int_E u \gamma^{2\alpha'-1} |\nabla \gamma| \rho, \\
{}&\int_E u(y) \gamma^{2\alpha'-1}(y) \Delta \gamma(y) g_E(x,y) \,dy
\le C \rho(x) \delta^{-d}\int_E u \gamma^{2\alpha'-1} |\Delta
\gamma| \rho.
\end{split}
\end{equation}
It remains to use the bounds of Lemma \ref{Lemma 3.1.4}.
\end{proof}

Theorem \ref{main thm} can be derived from this by using the same
construction as in \cite{Ms02}. Let $x\in E$ and $K,u$ be as in
Theorem \ref{main thm}. Let $\delta = \dist(x, K)$. We set
\[
K_1 = K\cap \overline{B(x,2\delta)},
\]
and
\[
K_n = K\cap\left( \overline{B(x, 2^n\delta)}\setminus B(x,
2^{n-1}\delta)\right),\quad n\ge 2.
\]
Since $K = \bigcap K_n$, we have
\[
u\le u_K\le \sum u_{K_n}
\]
where $u_K$ stands for the maximal solution of problem
\eqref{main}. By construction, we have $\dist (x, K_n )\ge
2^{n-1}\delta$ and $\diam (K_n)\le 2^{n+1}\delta$. Therefore Lemma
\ref{Lemma 3.1.5} is applicable to every $K_n$ and we get
\[
u_{K_n} \le C \rho(x) 2^{-nd} \delta ^{-d}
\cp(K_n)^{1/(\alpha-1)}\le C \rho(x) 2^{-nd} \delta ^{-d}
\cp(K)^{1/(\alpha-1)},
\]
which implies
\[
u(x) \le \sum u_{K_n} \le C \rho(x) \delta ^{-d}
\cp(K)^{1/(\alpha-1)} \sum 2^{-nd}.
\]

The extension of the theorem to arbitrary $C^4$ domains is a
simple modification of the arguments by Mselati \cite{Ms02}.



\bibliographystyle{amsplain}
\begin{thebibliography}{3}

\bibitem{Dy02} E. B. Dynkin, \emph{Diffusions, superdiffusions and partial
differential
  equations},
American Mathematical Society, Providence, RI, 2002. 
\MR{1883198 (2003c:60001)}

\bibitem{Ku98} S. E. Kuznetsov,  \emph{Polar boundary sets for superdiffusions
and removable lateral singularities for nonlinear parabolic PDEs},
Comm. Pure Appl. Math. \textbf{51} (1998), 303--340. 
\MR{1488517 (99c:35111)}

\bibitem{Ms02} B. Mselati, \emph{Classification et repr\'{e}sentation
probabiliste des solutions positives de $\Delta u=u^2$ dans un
domaine}, Th\'{e}se de Doctorat de l'Universit\'{e} Paris 6,
2002. 

\end{thebibliography}

\end{document}

\newcounter{assumption}[subsection]
\newenvironment{asm}%
  { %

\begin{list}{\arabic{section}.\arabic{subsection}.\Alph{assumption}.}%

       { %
          \usecounter{assumption}%
\setlength{\leftmargin}{0pt}\setlength{\rightmargin}{0pt}%

\setlength{\itemindent}{1.55cm}\setlength{\labelsep}{.2cm}%
            \setlength{\labelwidth}{1cm}%
       }%
   }%
  {\end{list}}

\renewcommand{\theassumption}{\thesubsection.\Alph{assumption}}

\newcounter{assumptionn}[subsection]
\newenvironment{asmN}%
  { %

\begin{list}{\arabic{section}.\arabic{subsection}.\arabic{assumptionn}.}%
       { %
          \usecounter{assumptionn}%
    \setlength{\topsep}{4pt}\setlength{\partopsep}{0pt}%
            \setlength{\itemsep}{2pt}%
            \setlength{\leftmargin}{0pt}\setlength{\rightmargin}{0pt}%
\setlength{\itemindent}{1.55cm}\setlength{\labelsep}{.2cm}%
            \setlength{\labelwidth}{1cm}%
       }%
   }%
  {\end{list}}


\renewcommand{\theassumptionn}{\thesubsection.\arabic{assumptionn}}


\newcounter{assumptiona}[section]
\newenvironment{asma}%
  { %
    \begin{list}{\arabic{section}.\Alph{assumptiona}.}%
       { %
          \usecounter{assumptiona}%
\setlength{\leftmargin}{0pt}\setlength{\rightmargin}{0pt}%
\setlength{\itemindent}{1.55cm}\setlength{\labelsep}{.2cm}%
            \setlength{\labelwidth}{1cm}%
       }%
   }%
  {\end{list}}

\renewcommand{\theassumptiona}{\thesection.\Alph{assumptiona}}