Negative Power Nonlinear Integral Equations on Bounded Domains

Extracted source text · arXiv 1904.03878

\documentclass[reqno]{amsart}
\usepackage{amscd}
\usepackage[dvips]{graphics}
%\usepackage[notref,notcite]{showkeys}
%\usepackage{citesort}


\def\beq{\begin{equation}}
\def\eeq{\end{equation}}
\def\ba{\begin{array}}
\def\ea{\end{array}}
\def\S{\mathbb S}
\def\R{\mathbb R}
\def\C{\mathbb C}
\def\n{\nabla}
\def\N{\mathbb N}

\def\la{\langle}
\def\ra{\rangle}
\def\cal{\mathcal}
\def \ds{\displaystyle}
\def \vs{\vspace*{0.1cm}}
\def\hs{\hspace*{0.6cm}}


\newcommand{\coker}{\operatorname{coker}}
\newcommand{\im}{\operatorname{im}}
\newcommand{\rk}{\operatorname{rank}}
\newcommand{\aut}{\operatorname{Aut}}
\newcommand{\res}{\operatorname{res}}
\newcommand{\lcm}{\operatorname{lcm}}

\newcommand{\irightarrow}{\stackrel{\sim}{\longrightarrow}}


\renewcommand{\thefootnote}{\fnsymbol{footnote}}
%%%%%%%%%%%%%%
% the following notations are used by someone on constructing the virtual cycle
%\newcommand{\comb}{\operatorname{Comb}}
%\newcommand{\ind}{\operatorname{index}}
%\newcommand{\Syz}{\Sigma_{y,\zeta}}
%\newcommand{\bpat}{\bar{\partial}}
%\newcommand{\Vd}{{V_{deform,\sigma}}}
%\newcommand{\Vm}{{V_{map,\sigma}}}
%\newcommand{\Vmp}{{V'_{map,\sigma}}}
%\newcommand{\Vr}{{V_{resol,\sigma}}}
%\newcommand{\fapp}{{f_{app,y,\zeta}}}
%\newcommand{\tU}{\tilde{U}}
%\newcommand{\ts}{\tilde{s}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%
%\newenvironment{abs}{\textbf{Abstract}\mbox{  }}{ }
%\newenvironment{key words}{\textbf{Keywords}\mbox{  }}{ }
%\newtheorem{thm}{Theorem}[section]
%\newtheorem{lm}[thm]{Lemma}
%\newtheorem{prop}[thm]{Proposition}
%\newtheorem{cor}[thm]{Corollary}
%\newtheorem{conj}[thm]{Conjecture}
%\newtheorem{quest}[thm]{Question}
%\theoremstyle{definition}
%\renewenvironment{proof}{\noindent{\textbf{Proof.}}}{\hfill$\Box$}
%\newtheorem{rem}[thm]{Remark}
%\newtheorem{rems}[thm]{Remarks}
%\newtheorem{df}[thm]{Definition}
%\newtheorem{ex}[thm]{Example}
%\theoremstyle{remark}
%\newtheorem{nota}{Notation} \renewcommand{\thenota}{}
%\numberwithin{equation}{section}

\numberwithin{equation}{section}
\newenvironment{abs}{\textbf{Abstract}\mbox{  }}{ }
\newenvironment{key words}{\textbf{Keywords}\mbox{  }}{ }
\newtheorem{theorem}{Theorem}[section]
\newtheorem{definition}[theorem]{\textbf{Definition}}
\newtheorem{corollary}[theorem]{\textbf{Corollary}}
\newtheorem{proposition}[theorem]{\textbf{Proposition}}
\newtheorem{lemma}[theorem]{Lemma}
\renewenvironment{proof}{\noindent{\textbf{Proof.}}}{\hfill$\Box$}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{\textbf{Remark}}
\theoremstyle{plain}
\pagestyle{plain}
\newenvironment{Ack}{\par\textbf{Acknowledgements:}\mbox{  }}{ }
 \newenvironment{coj}{\par\textbf{Conjecture:}\mbox{  }}{ }


%----------------------------------------------------------------
 \begin{document}
\title{\textbf{Negative power nonlinear integral equations on bounded domains}}
\author[J. Dou, Q. Guo and M. Zhu]{{\it Dedicate to Ha\"im Brezis on occasion of his 75th birthday }
\\
\\Jingbo Dou, Qianqiao Guo and Meijun Zhu}
\address{Jingbo Dou, School of Mathematics and Information Science, Shaanxi Normal University, Xi'an, Shaanxi, 710119, China}

\email{jbdou@snnu.edu.cn}

\address{Qianqiao Guo, Department of Applied Mathematics, Northwestern Polytechnical University, Xi'an, Shaanxi, 710129, China}

\email{gqianqiao@nwpu.edu.cn}

\address{Meijun Zhu, Department of Mathematics,
The University of Oklahoma, Norman, OK 73019, USA}

\email{mzhu@math.ou.edu}


%\date{05/16/2018, hope this is the final version. submitted to TAMS}

%this is the final one. submitted to JDE 3-1-2019
\maketitle

% ----------------------------------------------------------------
\noindent
\begin{abs}
This is the  continuation of our previous work \cite{DZ2017}, where we introduced and studied some  nonlinear integral equations on bounded domains that are related to the sharp Hardy-Littlewood-Sobolev  inequality. In this paper, we introduce some nonlinear integral equations on bounded domains that are related to the sharp reversed Hardy-Littlewood-Sobolev inequality. These are integral equations with nonlinear term involving negative exponents.  Existence results as well as nonexistence results are obtained.
\end{abs}\\
\begin{key words} Reversed sharp Hardy-Littlewood-Sobolev inequality, Integral equations, Existence, nonexistence
\end{key words}\\
\textbf{Mathematics Subject Classification(2000).}
45G10, 35J60 \indent
%---------------------------------------------------------------------------------
\section{\textbf{Introduction}\label{Section 1}}


In \cite{DZ2017}, motivated by the study of certain semi-linear equations and the sharp Sobolev inequality, we introduced and  studied the integral equations (with positive power) related to the sharp Hardy-Littlewood-Sobolev (HLS for short) inequality. Let us briefly recall these as the follows.

%Now we recall it to make the motivation of this paper clear. %Now we recall the results in \cite{DZ2017}.

For $0<\alpha<n$,  on any bounded domain $\Omega \subset \mathbb{R}^n$ with smooth boundary, we considered
\begin{eqnarray*}
\hat{\xi}_\alpha(\Omega)=\sup_{f\in L^\frac{2n}{n+\alpha}(\Omega),f\neq 0}\frac{\int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha)} f(y) dx dy}{||f||^2_{L^\frac{2n}{n+\alpha}(\Omega)}}.
\end{eqnarray*}
It was showed in \cite{DZ2017} that $
\hat{\xi}_\alpha (\Omega)= N_\alpha,$  where $N_\alpha$ is the best constant of the classical sharp HLS inequality (due to Lieb \cite{Lieb1983});
  And $\hat{\xi}_\alpha (\Omega) $ is not attained by any functions if $\Omega \ne \mathbb{R}^n$. This indicates that  there is not any energy maximizing  solution to
 $$
f^\frac{n-\alpha}{n+\alpha}(x)=\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha}}dy,\quad f\ge 0, \quad x\in \overline \Omega.
$$
We then considered a general integral equation
\begin{equation}\label{HB}
f^{q-1}(x)=\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha}}dy+\lambda \int_\Omega\frac {f(y)}{|x-y|^{n-\alpha-1}}dy,\quad f\ge 0, \quad x\in \overline \Omega,
\end{equation}
for $\alpha<n$, and studied  the existence and nonexistence of positive solutions for different power $q$ and parameter $\lambda$.


In this paper we consider integral equation \eqref{HB}  for $\alpha>n$.
%Are there some existence and non-existence results for positive solutions?
This case is related to so called sharp reversed HLS inequality, which was discovered by Dou and Zhu \cite{DZ2015}.

% discovered the reversed sharp HLS inequality, which then naturally leads us to study the integral equation \eqref{HB}  for $\alpha>n$.
%
% On the other hand, even ignoring the reversed sharp HLS inequality, the integral  equation \eqref{HB} is well defined for any $\alpha>0.$ So we concentrate on the existence and nonexistence to the integral equation \eqref{HB} for $\alpha>n$ in this paper.
%

%{\bf put this at the end, 9-24-2017} The case $\alpha=n$ is related to the sharp logarithmic HLS inequality \cite{CarLo1992, B1993}, which will not be considered here.

%From now on,  in this paper we focus on the case of $\alpha>n$.
Recall the sharp reversed  HLS inequality  from \cite{DZ2015} (see also related work by Ng\^{o}, Nguyen \cite{NN2017} and by Beckner \cite{Beck2015}):


\smallskip
\noindent  \textbf{Theorem A.} {\it  For $\alpha>n$,
\begin{equation}\label{R-HLS}
|\int_{\mathbb{R}^n} \int_{\mathbb{R}^n} f(x)|x-y|^{-(n-\alpha)} g(y) dx dy|\ge N_\alpha ||f||_{L^{\frac{2n}{n+\alpha}}(\mathbb{R}^n)}||g||_{L^{\frac {2n}{n+\alpha}}(\mathbb{R}^n)}
\end{equation}
holds for all non-negative functions $f, \ g \in L^{\frac{2n}{n+\alpha}}(\mathbb{R}^n),$
where
\begin{equation}\label{R-extreCon}
N_\alpha=N(\frac{2n}{n+\alpha},\alpha, n)=\pi^{(n-\alpha) /2} \frac{\Gamma(\alpha/2)}{\Gamma(n/2+\alpha/2)} \{\frac{\Gamma(n/2)}{\Gamma(n)} \}^{-\alpha/n};
\end{equation}
And the equality holds if and only if
\begin{equation}\label{R-HLS-ex}
f=c_1g=c_2\big(\frac{1}{c_3+|x-x_0|^2}\big)^{\frac {n+\alpha}2},
\end{equation}
where $c_1, c_2, c_3$ are any positive constants, and $x_0 \in \mathbb{R}^n$.

}\smallskip

\smallskip

Similar to what is done in \cite{DZ2017}: for any smooth domain $\Omega \subset \mathbb{R}^n$,  we consider
\begin{eqnarray*}
\xi_\alpha(\Omega)=\inf_{f\in L^\frac{2n}{n+\alpha}(\Omega),f\ge 0, f \ne 0}\frac{\int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha)} f(y) dx dy}{||f||^2_{L^\frac{2n}{n+\alpha}(\Omega)}}.
\end{eqnarray*}
We will  show that
\begin{eqnarray}\label{R-energy}
 \xi_\alpha (\Omega)= N_\alpha,
  \end{eqnarray}
  and $\xi_\alpha (\Omega) $ is not attained by any functions if $\Omega \ne \mathbb{R}^n$ (see Proposition 2.1 below).
Again, we notice that the corresponding Euler-Lagrange equation for the minimizer (if the minimum is attained) is  integral equation \eqref{HB}  with $\lambda=0$ and a negative power ($\frac{n-\alpha}{n+\alpha}<0$).
We thus know that if $\alpha>n$ there is no energy minimizing solution to  integral equation \eqref{HB} for $q=2n/(n+\alpha)$ and $\lambda=0$.


%{\bf I am here 5-10-2018}

 \medskip

 Let $p_\alpha=2n/(n-\alpha)$, $q_\alpha=2n/(n+\alpha)$ and $d(\Omega)=\sup\limits_{x,y\in\Omega}|x-y|$ be the diameter of the bounded domain $\Omega$.
In this paper, we consider  integral equation \eqref{HB} for $\alpha>n$. We shall prove
\begin{theorem}\label{main}
Assume $\alpha>n$ and $\Omega$ is a bounded domain with smooth boundary.

\noindent (1)  For $0<q<q_\alpha$(subcritical case), and  $-\frac{1}{d(\Omega)}<\lambda$, there is a positive solution $f\in C^1(\overline{\Omega})$ to equation \eqref{HB}.

\noindent(2) For $q=q_\alpha$ (critical case), and $-\frac{1}{d(\Omega)}< \lambda<0$, there is a positive solution $f\in C^1(\overline{\Omega})$ to equation \eqref{HB}.


\noindent(3) For $q_\alpha\le q<1$ (critical and supercritical cases), and   $\lambda\ge 0$, if  $\Omega$ is a star-shaped domain, then  there is not any positive $C^1(\overline{\Omega})$ solution to \eqref{HB}.

\end{theorem}

%{\bf maybe comments on $q\ge 1$ case. We will see that Pohozaev does not work for $q\ge 1.$9-24-2017}

%
%For the existence, we first consider the following general equation
%\begin{equation}\label{R-HB-sub}
%f^{q-1}(x)=\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha}}dy,\quad x\in\overline{\Omega}.
%\end{equation}
%and will prove the following existence results for the subcritical case.
%\begin{theorem}\label{R-sub-existence} Let $\alpha >n$.
%For $q\in (0, \frac{2n}{n+\alpha})$, there is a positive solution $f\in C^1(\overline{\Omega})$ to equation \eqref{R-HB-sub}. {\bf Need $q>0?$ 4/4/2017}
% \end{theorem}
%
% For critical power, we also consider the following Brezis-Nirenberg type equation. \begin{equation}\label{R-HB-critical}
%f^{q_\alpha-1}(x)=\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha}}dy+ \lambda \int_\Omega\frac {f(y)}{|x-y|^{n-\alpha-1}}dy,\quad x\in\overline{\Omega}.
%\end{equation}
%We will show:
%\begin{theorem}\label{R-cri-sub-xistence}
%For $\lambda<0$, there is a positive solution $f\in C^{2}(\overline{\Omega})$ to equation \eqref{R-HB-critical}; for $\lambda\ge 0$ and $\Omega$ is a star-shape domain, there is no positive $C^1$ continuous (up to the boundary) solution to \eqref{R-HB-critical}. {(\bf Check it later--Guo 17/5/29)}
% \end{theorem}


 We emphasis here that $q_\alpha<1$ (since $\alpha>n$). Thus equation \eqref{HB} has a nonlinear term with a negative power. Our results indicate that even though  the integral equation, which is related to the reversed HLS inequality, is of negative nonlinearity, similar phenomena to the integral equation  with positive power can be seen. Contrary to integral equations with positive nonlinearity, it seems that no compact embedding can be used directly for the existence result to the integral equation  with subcritical negative power. Different techniques are needed for deriving the existence as well as the nonexistence results. See more details in Section 3 below.

 We organize the paper as follows: In Section 2, we focus on the nonexistence result (part (3) of Theorem \ref{main}).  In Section 3, we first obtain the existence result  (part (1) of Theorem \ref{main}), and then we show the symmetric and monotonically increasing properties of solutions to the integral equation on a ball, even no pointwise boundary condition is given (Theorem \ref{UB-symmetry} below). In Section 4, we come back to the integral equations with critical exponent and a lower order term and prove the existence result (part (2) of Theorem \ref{main}).

 %{\bf I am here 5-10-2018}

 Notation: for any function $f(x)$ defined on $\Omega$, we always use $\tilde f(x)$ to represent its trivial extension in $\Bbb{R}^n$, namely,
\begin{eqnarray*}
 \tilde{f}(x):= \begin{cases}f(x)&\quad x\in\Omega,\\
 0&\quad x\in\mathbb{R}^n\backslash\Omega.
  \end{cases}
  \end{eqnarray*}
 We also denote
  $$
  I_\alpha f(x):=\int_{\Bbb{R}^n}\frac {f(y)}{|x-y|^{n-\alpha}} dy, \ \ \ \ I_{\alpha, \Omega} f(x):=\int_{\Omega}\frac {f(y)}{|x-y|^{n-\alpha}} dy
  $$
and
 $$  L_+^q(\Omega):=\{ f \in   L^q(\Omega)\setminus\{0\}  \ : \ f \ge 0\}.$$

%{\bf In a follow-up paper, we shall address similar equation for $\alpha=n$, as well as $\alpha>n$. put into last section 3/25/2017}.
%
%
%{\bf all these discussion will be put in a very last section. 3/25/ 2017. Note that one can easily obtains existence result for radial symmetric solution if the domain is an annulus.  It remains as an interesting and challenging question to study \eqref{HB-1} in some domain with non-trivial topology. We shall also study the blowup behavior of solutions as $p\to p_\alpha-$ in the future.}
%
%
%{\bf At the end of this paper, I make some comments on \eqref{HB-1}, (1) can not prescribe boundary value; thus some link to Lihe wang. et. al work; (2) lack of variational structure if lower oder term at different position. 2/10/2017}

\section{nonexistence for critical and supercritical cases}

In this section, we first derive energy estimate \eqref{R-energy} for any domain $\Omega \subset \Bbb{R}^n$ , and then show that the  infimum $\xi_\alpha(\Omega) $ is not achieved by any function once $\Omega \ne \Bbb{R}^n$.



\begin{proposition}\label{prop2-1} For any domain $\Omega \subset \Bbb{R}^n$,
$\xi_\alpha(\Omega)=N_\alpha$. Further, if  $\Omega \ne \Bbb{R}^n$, then the infimum $\xi_\alpha(\Omega)$  is not achieved by any function in $L^{q_\alpha}(\Omega)$.
\end{proposition}



%  devote to prove Theorem \ref{non-existence}.

\begin{proof}
%Let
% \begin{eqnarray*}
% \tilde{f}(x)= \begin{cases}f(x)&\quad x\in\Omega,\\
% 0&\quad x\in\mathbb{R}^n\backslash\Omega.
%  \end{cases}
%  \end{eqnarray*}
If $f\in L_+^{q_\alpha}(\Omega)$, then $\tilde{f}\in L_+^{q_\alpha}(\mathbb{R}^n)$ . It follows that
\begin{eqnarray*}
\xi_\alpha(\Omega)&=&\inf_{{f}\in L_+^{q_\alpha}(\Omega)\setminus\{0\}}\frac{\int_{\mathbb{R}^n} \int_{\mathbb{R}^n} \tilde{f}(x)|x-y|^{-(n-\alpha)} \tilde{f}(y) dx dy}{||\tilde{f}||^2_{L^{q_\alpha}(\mathbb{R}^n)}}\\
&\ge & \inf_{{g}\in L_+^{q_\alpha}(\Bbb{R}^n)\setminus \{0\}}\frac{\int_{\mathbb{R}^n} \int_{\mathbb{R}^n} {g}(x)|x-y|^{-(n-\alpha)} {g}(y) dx dy}{||{g}||^2_{L^{q_\alpha}(\mathbb{R}^n)}} =N_\alpha.
\end{eqnarray*}

On the other hand,  recall  that $f(x)=\big(
\frac{1}{1+|x|^2}\big)^\frac{n+\alpha}2$ is an extremal function to the sharp reversed HLS inequality in {Theorem A}, as well as its conformal equivalent class:
\begin{equation}\label{fe}
f_\epsilon(x)=\epsilon^{-\frac{n+\alpha}2}f(\frac{|x-x_*|}\epsilon)=\big(\frac{\epsilon}{\epsilon^2+|x-x_*|^2}\big)^\frac{n+\alpha}2,
\end{equation}
where $x_*\in\mathbb{R}^n,\epsilon>0$. Thus
\[
\|I_\alpha f\|_{L^{p_\alpha}(\mathbb{R}^n)}=\|I_\alpha f_\epsilon\|_{L^{p_\alpha}(\mathbb{R}^n)},
\quad\|f\|_{L^{q_\alpha}(\mathbb{R}^n)}=\|f_\epsilon\|_{L^{q_\alpha}(\mathbb{R}^n)}.
\]
 Choose $x_0=x_*$ for some point $x_0\in \Omega$ and $R$ small enough so that $B_{R}(x_0)\subset\Omega$. Then we define test function $g (x)$ as
 \begin{eqnarray*}
g(x)= \begin{cases}f_\epsilon(x) &\quad x\in B_{R}(x_0)\subset\Omega,\\
 0&\quad x\in\mathbb{R}^n\backslash B_{R}(x_0).
  \end{cases}
  \end{eqnarray*}
Obviously, $g\in L_+^{q_\alpha}(\Bbb{R}^n).$ Thus,
\begin{eqnarray*}
& &\int_{\Omega} \int_{\Omega} \frac{g(x)g(y)}{|x-y|^{n-\alpha}} dx dy\\
&=&\int_{\mathbb{R}^n} \int_{\mathbb{R}^n} \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy
-2\int_{\mathbb{R}^n} \int_{\mathbb{R}^n\backslash B_{R}(x_0) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy\\
& &+\int_{\mathbb{R}^n\backslash B_{R}(x_0) } \int_{\mathbb{R}^n\backslash B_{R}(x_0) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy\\
&\le&N_\alpha\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}-2I_1 ,
\end{eqnarray*}
where
\begin{eqnarray*}
I_1&:=& \int_{\mathbb{R}^n} \int_{\mathbb{R}^n\backslash B_{R}(x_0) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy.
\end{eqnarray*}
Notice that $f_\epsilon(x)$ is an extremal function for the sharp reversed HLS inequality. Thus it satisfies  integral equation:
\begin{equation*}
f^\frac{n-\alpha}{n+\alpha}(x)=B\int_{\mathbb{R}^n}\frac{f(y)}{|x-y|^{n-\alpha}}dy,
\end{equation*} where $B$ %=\frac{\pi^\frac{n}2\Gamma(\alpha/2) }{\Gamma((n+\alpha)/2)}.$
is a suitable positive constant.
We thus can estimate $I_1$ in the following:
\begin{eqnarray*}
I_1=
%2\int_{\mathbb{R}^n} \int_{\mathbb{R}^n\backslash B_{R}(x_0) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-%\alpha}} dx dy\\
C\int_{\mathbb{R}^n\backslash B_{R}(x_0) } f_\epsilon^{\frac{2n}{n+\alpha}}(x) dx
%&=&C\int_{\frac{R}\epsilon}^\infty \frac{t^{n-1}}{(1+t^2)^{n}}dt\\
=O(\frac{R}\epsilon)^{-n},  \quad\quad\text{as}\quad\epsilon\to0.
\end{eqnarray*}
And we also have
\begin{eqnarray*}
\int_{B_{R}(x_0) } f_\epsilon^{\frac{2n}{n+\alpha}}(x) dx&=&\int_{\mathbb{R}^n} f_\epsilon^{\frac{2n}{n+\alpha}}(x) dx-\int_{\mathbb{R}^n\backslash B_{R}(x_0) } f_\epsilon^{\frac{2n}{n+\alpha}}(x) dx\\
&=&\int_{\mathbb{R}^n} f_\epsilon^{\frac{2n}{n+\alpha}}(x) dx-O(\frac{R}\epsilon)^{-n},  \quad\quad\text{as}\quad\epsilon\to0.
\end{eqnarray*}
Hence, for small enough $\epsilon>0$, we have
 \begin{eqnarray*}
 \xi_\alpha(\Omega)&\le &\frac{\int_{\Omega} \int_{\Omega} \frac{g(x)g(y)}{|x-y|^{n-\alpha}} dx dy}{\|g\|^2_{L^{q_\alpha}(\Omega)}}\\
 %=\frac{\int_{\Omega} \int_{\Omega} \frac{g(x)g(y)}{|x-y|^{n-\alpha}} dx dy}{\|f_\epsilon\|^2_{L^{q_\alpha}(B_{R}(x_0))}}\\
 &\le &\frac{N_\alpha\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}-I_1}{\|f_\epsilon\|^2_{L^{q_\alpha}(B_{R}(x_0))}}\\
 &=&\frac{N_\alpha\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}-O(\frac{R}\epsilon)^{-n}}{\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}-O(\frac{R}\epsilon)^{-n}} ,
  \end{eqnarray*}
which yields $\xi_\alpha(\Omega)\le N_\alpha$ as $\epsilon \to 0$.

Finally,  we show  that $\xi_\alpha(\Omega)$ is not achieved if $\Omega\neq\mathbb{R}^n$.
In fact, if $\xi_\alpha(\Omega)$ were attained by some function $u\in L_+^{q_\alpha}(\Omega)$, then $\tilde u\in L_+^{q_\alpha}(\Bbb{R}^n)$ would be an extremal function to the sharp reversed HLS inequality on $\Bbb{R}^n$, which is impossible due to Theorem A.
%
% \begin{equation*}
%\tilde{ u} =
%\begin{cases}
%u&\quad \text{on}~ \Omega,\\
%0&\quad \text{on}~  \mathbb{R}^n\backslash\Omega,
%\end{cases}
% \end{equation*}
\end{proof}

\medskip

Proposition \ref{prop2-1} indicates that for $\alpha>n$ there is not any minimizing  energy solution to \eqref{HB} for $q=2n/(n+\alpha)$ and $\lambda=0$. In fact, we will show that there is not any positive $C^1$ solution to \eqref{HB}  for $\alpha>n$, $q=2n/(n+\alpha)$ and $\lambda=0$ on any star-shaped domain.

\medskip


\noindent{\bf Proof of part (3) in Theorem \ref{main} (nonexistence part)}. Without loss of generality, here we assume that the origin is in $\Omega$ and the domain is star-shaped with respect to the origin.

Recall the following Pohozaev identity from  \cite{DZ2017}.

\begin{lemma}\label{poh}  Assume that the origin is in $\Omega$ and the domain is star-shaped with respect to the origin. If $u\in C^1(\overline \Omega)$ is a non-negative solution to
\begin{equation}\label{gen_equ}
u(x)=\int_\Omega\frac{u^{p-1}(y)}{|x-y|^{n-\alpha}}dy+ \lambda \int_\Omega\frac{u^{p-1}(y)}{|x-y|^{n-\alpha-1}}dy,\quad x\in \overline\Omega,
\end{equation}
where $p\ne 0$, $\lambda \in \Bbb{R}$,
then
\begin{eqnarray}\label{SB-5}
(\frac{n}{p}+\frac{\alpha-n}2)\int_{\Omega}u^{p}(x)dx= -\frac{\lambda}2\int_{\Omega}\int_{\Omega}\frac{u^{p-1}(x)u^{p-1}(y)}{|x-y|^{n-\alpha}}dydx+\frac1{p}\int_{\partial \Omega}(x\cdot \nu) u^{p}(x)d\sigma,
\end{eqnarray}
where $\nu$ is the outward unit normal vector to $\partial \Omega$.
\end{lemma}

\medskip

Applying Lemma \ref{poh} to \eqref{HB} for $\lambda\ge0$ and $u(x)=f^{q-1}(x)$. Thus $p=\frac{q}{q-1}=q^\prime$. Noticing  that $1>q\ge q_\alpha$ is equivalent to  $p\le p_\alpha<0$, we have
\begin{eqnarray*}
-\frac{\lambda}2\int_{\Omega}\int_{\Omega}\frac{u^{p-1}(x)u^{p-1}(y)}{|x-y|^{n-\alpha-1}}dydx+\frac 1p \int_{\partial \Omega}(x\cdot\nu) u^{p-1}(x)d\sigma\ge 0.
\end{eqnarray*}
Since $\Omega$ is star-shaped domain about the origin, we have $x\cdot \nu>0$  on $\partial\Omega$.   If $\lambda>0$, then $u(x)\equiv\infty $ on $\overline{\Omega}$. If $\lambda=0$, then $u\equiv\infty$ on $\partial\Omega$. Therefore we obtain a contradiction to that $f(x)$ is a positive $C^1(\overline{\Omega})$ solution.

 \medskip

 \begin{remark} Condition $q<1$ is needed in our proof.
 If $q>1$, one  may call it a supercritical exponent  (since  it is bigger than $q_\alpha$). However, in this case $p=q/(q-1)>0$ is also bigger than negative $p_\alpha$. Our nonexistence result may not be true any more in this case.
 \end{remark}

%{(\bf We must make sure star-shaped is not necessary for the following Pohozaev identity) In this paper?? 9-24-2017--%In this paper, star-shaped is OK, Guo 10-07}% Without loss of generality, here we assume that the origin is in $%%\Omega$ and the domain is star-shaped with respect to the origin.


\smallskip



Note that the unit ball is conformally equivalent to the upper half space. We have

\begin{corollary}\label{Pro-Half space}
There is no positive solution $u\in L^{p_\alpha}(\overline{\mathbb{R}^n_+}) \cap C^1(\overline {\mathbb{R}^n_+}) $ to
% \begin{equation}\label{HB-4}
 $$
u(x)=\int_{\mathbb{R}^n_+}\frac{u^{p_\alpha-1}(y)}{|x-y|^{n-\alpha}}dy,\quad x\in \overline{\mathbb{R}^n_+}.
$$
%\end{equation}
\end{corollary}
%\begin{remark}
%In fact, we can derive a stronger statement by removing the $C^1$ condition in Corollary \ref{Pro-Half space} via the method of moving spheres.
%\end{remark}

%{\bf I am here 9/3/2018}

\section{Existence result for subcritical case}

For subcritical exponents we have the following inequality:
\begin{lemma}\label{bound HLS}
Let $q\in (0, {q_\alpha})$.
 There exists a positive constant  $C(n,q,\alpha,\Omega)>0$ such that
 \begin{equation}\label{B-HLS inequality}
 \int_{\Omega} \int_{\Omega} {f}(x)|x-y|^{-(n-\alpha)} {f}(y) dx dy
\ge C(n,q,\alpha,\Omega)\|f\|^2_{L^q(\Omega)}
 \end{equation} holds for any non-negative function  $f\in L^q(\Omega)$.
 %Moreover, operator $I_{\alpha,\Omega}$: $ L^t(\Omega)\hookrightarrow L^{t'}(\Omega)$ is a compact %embedding.
 \end{lemma}
 \begin{proof}
For $f\in L^q(\Omega)$, by using the reversed HLS inequality \eqref{R-HLS} we have
\begin{eqnarray*}
\langle I_{\alpha, \Omega} f,f\rangle&=& \int_{\mathbb{R}^n} \int_{\mathbb{R}^n} \tilde{f}(x)|x-y|^{-(n-\alpha)} \tilde{f}(y) dx dy\\
&\ge& N_\alpha \|\tilde{f}\|^2_{L^{q_\alpha}(\mathbb{R}^n)}=N_\alpha \|\tilde{f}\|^2_{L^{q_\alpha}(\Omega)}\\
&\ge & C(n, q, \alpha, \Omega)\|f\|^2_{L^q(\Omega)}.
\end{eqnarray*}
\end{proof}

We would like to point out that one can also prove the above lemma directly via a Young type inequality as  that in Dou, Guo and Zhu \cite{DGZ17}.

%{\bf Fast way may be to see that $W^{\alpha, t} )\hookrightarrow  L^{t'}(\Omega)$ is compact. $t>p_\alpha$
%if and only if $t' <nt/(n-\alpha t)$. Proof below: for $ \alpha<1$, it may not work. 2/22/2017. Will come back. Will delete later. 3/15/2017}

%Let $\{\Omega_j\}_{j=1}^N$ be a finite covering of $\Omega$ with $\Omega=\cup_{j=1}^\infty\Omega_j$.
 %We divide the proof into two step.

%\textbf{ Step 1.}

%{\bf compactness will be checked out late. 4/7/2017. Different for negative power. 4/24/2017}

%{\bf above compactness will be checked late 4/7/2017. Compactness argument does not work. 4/24/2017}



%{\bf above will be revised. 2/22/2017. Revised on 3/16/2017. No need to assume that $\alpha>1$ here}

Based on the above lemma, we can obtain the existence result for subcritical exponent (part (1) of Theorem \ref{main}). %For simplicity, we only present the proof for $\lambda=0${(\bf Maybe $\lambda>0$ should be given---Guo 6/25/17)}.
 Notice that it is different from the case $0<\alpha<n$ (Lemma 3.2 in \cite{DZ2017}) since no compact embedding can be used directly here. We follow a similar approach used in Dou, Guo and Zhu \cite{DGZ17}.

\begin{lemma}\label{Bound-attained} For $0<q<{q_\alpha}, \lambda>-\frac{1}{d(\Omega)}$, infimum
\begin{equation*}
\xi_{\alpha,q}(\Omega):=\inf_{f\in L_+^q(\Omega)}\frac{\int_{\Omega} \int_{\Omega}(f(x)|x-y|^{-(n-\alpha)} f(y)+\lambda f(x)|x-y|^{-(n-\alpha-1)} f(y)) dy dx}{\|f\|^2_{L^q(\Omega)}}>0,
\end{equation*}
 and it is attained by some nonnegative  function in $L_+^q(\Omega).$
\end{lemma}

\begin{proof}  Notice:  for $x, \ y \in \Omega$ and $0>\lambda>-1/d(\Omega),$ $1+\lambda|x-y|\ge 1+\lambda d(\Omega)>0$.  Thus we know $\xi_{\alpha,q}(\Omega)>0$ by Lemma \ref{bound HLS}.

 Choose a minimizing nonnegative sequence $\{f_j\}_{j=1}^\infty$ in $L^q(\Omega)$. Assume without loss of generality that $f_j\in L^{q_\alpha}(\Omega)$ (see, for example, Proposition 2.5 in \cite{DGZ17}). Then we can normalize it such that $\|f_j\|_{L^{q_\alpha}(\Omega)}=1$. It follows that  there exists a subsequence such that
\[f_j^q\rightharpoonup f_*^q\quad\text{weakly\, in}\quad L^{\frac{q_\alpha}{q}}(\Omega), \ \ \mbox{as} \ j\to\infty.
\]
Then
\begin{equation}\label{equ1-Bound-attained}
\int_{\Omega}f_j^q\to \int_{\Omega}f_*^q, \ \ \mbox{as} \ j\to\infty.
\end{equation}


Claim: $\|f_j\|_{L^{1}(\Omega)}\le C.$

We relegate the proof of this claim to the end.

Once the claim is proved, we have $\int_{\Omega}f_*^q>C>0$ via an interpolation inequality
%({\bf --delete later--write for clear-- since $1=\int_{\Omega}f_j^{q_\alpha}=\int_{\Omega}((f_j)^{\frac{q}{q+1}})^{\frac{q+1}{q}q_\alpha} \le C(\int_{\Omega}(f_j)^{\frac{2q}{q+1}})^{\frac{(q+1)q_\alpha}{2q}} \le C(\epsilon \int_{\Omega}f_j+C(\epsilon)\int_{\Omega}f_j^q)^{\frac{(q+1)q_\alpha}{2q}}\le C\epsilon+C(\epsilon)(\int_{\Omega}f_j^q)^{\frac{(q+1)q_\alpha}{2q}}$ --10/3--Guo})
and $f_j^q\rightharpoonup f_*^q\quad\text{weakly\, in}\quad L^{\frac{1}{q}}(\Omega).$
Then for any fixed $x\in\overline{\Omega}$, $f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)\in L^{\frac{1}{1-q}}(\Omega)$, thus, as $j \to \infty$,
\begin{eqnarray*}
\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy\to\int_{\Omega} f_*(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy.
\end{eqnarray*}

Further, we show  that the above convergence is actually uniformly convergent for  all $x\in\overline\Omega$.

By H\"{o}lder's inequality we have
\begin{eqnarray*}
&&\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy\\
&\le & (\int_{\Omega}f_j^{q\cdot\frac{1}{q}}(y)dy)^q(\int_{\Omega}(f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|))^\frac{1}{1-q}dy)^{1-q}\le C,
\end{eqnarray*}
that is, $\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy$ is uniformly bounded for $x\in\overline{\Omega}$. Notice that for any $x_1,x_2,y\in\overline{\Omega},$
\begin{equation*}
||x_1-y|^{\alpha-n}-|x_2-y|^{\alpha-n}|
\le \begin{cases}
C|x_1-x_2|^{\alpha-n},\ \mbox{if}~0<\alpha-n\le 1,\\
C|x_1-x_2|,\ \ \ \ \ \ \mbox{if}~\alpha-n>1.
\end{cases}
\end{equation*}
Then for any $x_1,x_2\in\overline{\Omega}, $
\begin{eqnarray*}
&&\ \ |\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x_1-y|^{\alpha-n}(1+\lambda|x_1-y|)dy\\
&&-\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x_2-y|^{\alpha-n}(1+\lambda|x_2-y|)dy|\\
&\le &\int_{\Omega}f_j^q(y)f_*^{1-q}(y)||x_1-y|^{\alpha-n}(1+\lambda|x_1-y|)-|x_2-y|^{\alpha-n}(1+\lambda|x_2-y|)|dy
\\
&\le &C\max(|x_1-x_2|^{\alpha-n}, |x_1-x_2|)\int_{\Omega}f_j^q(y)f_*^{1-q}(y)dy
\\
&\le &C\max(|x_1-x_2|^{\alpha-n}, |x_1-x_2|)(\int_{\Omega}f_j(y)dy)^q(\int_{\Omega}f_*(y)dy)^{1-q}\\
&\le& C\max(|x_1-x_2|^{\alpha-n}, |x_1-x_2|).
\end{eqnarray*}
Thus $\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy$ is equicontinuous in $ \overline{\Omega}$. It follows that, as $j\to\infty$,
\[\int_{\Omega}f_j^q(y)f_*^{1-q}(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy\to\int_{\Omega} f_*(y)|x-y|^{\alpha-n}(1+\lambda|x-y|)dy
\]
uniformly for $x\in\overline{\Omega}$.

Therefore for any $\epsilon>0$ small enough, there exists $j_0\in \mathbb{N}$ such that for any $j>j_0$,
\begin{eqnarray*}
&&\int_{\Omega}\int_{\Omega}f_j^q(x)f_*^{1-q}(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j^q(y)f_*^{1-q}(y)dxdy\\
&\ge& \int_{\Omega}f_j^q(x)f_*^{1-q}(x)[\int_{\Omega}|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dy-\epsilon]dx.
\end{eqnarray*}
By H\"{o}lder's inequality we know $\int_{\Omega}f_j^q(x)f_*^{1-q}(x)dx\le C$.  Again,  notice that
\begin{eqnarray*}
&&\int_{\Omega}f_j^q(x)f_*^{1-q}(x)\int_{\Omega}|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dydx \\
&\to& \int_{\Omega}\int_{\Omega}f_*(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dydx
\end{eqnarray*}
since $f_*^{1-q}(x)\int_{\Omega}|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dy\in L^{\frac{1}{1-q}}(\Omega)$. We have for $j>j_0$ large enough,
\begin{eqnarray*}
&&\int_{\Omega}\int_{\Omega}f_j^q(x)f_*^{1-q}(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j^q(y)f_*^{1-q}(y)dxdy\\
&\ge& \int_{\Omega}\int_{\Omega}f_*(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dydx-\epsilon-C\epsilon.
\end{eqnarray*}
Similarly, we also have, for $j>j_0$ large enough,
\begin{eqnarray*}
&&\int_{\Omega}\int_{\Omega}f_j^q(x)f_*^{1-q}(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j^q(y)f_*^{1-q}(y)dxdy\\
&\le& \int_{\Omega}\int_{\Omega}f_*(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dydx+\epsilon+C\epsilon.
\end{eqnarray*}
Therefore, as $j \to \infty$,
\begin{eqnarray*}
&&\int_{\Omega}\int_{\Omega}f_j^q(x)f_*^{1-q}(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j^q(y)f_*^{1-q}(y)dxdy\\
&\to& \int_{\Omega}\int_{\Omega}f_*(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dydx.
\end{eqnarray*}
It follows from the above and H\"{o}lder's inequality that
%we have {\bf not clear!! 9/26/2017}
\begin{eqnarray*}%\nonumber
&&\liminf\limits_{j\to\infty}\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j(y)dxdy\\%\label{equ2-Bound-attained}
&\ge& \int_{\Omega}\int_{\Omega} f_*(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dxdy.
\end{eqnarray*}
Since $||f_j||_{L^q(\Omega)} \to ||f_*||_{L^q(\Omega)} >0$, the above inequality then implies
\begin{eqnarray*}
&&\liminf\limits_{j\to\infty}\frac{\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j(y)dxdy}{\|f_j\|^2_{L^q(\Omega)}}\\
&\ge&\frac{\int_{\Omega}\int_{\Omega} f_*(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_*(y)dxdy}{\|f_*\|^2_{L^q(\Omega)}}.
\end{eqnarray*}
That is: $f_*$ is a minimizer.
% \quad \text{thus} \quad \|f_*\|_{L^t(\Omega)}\le\liminf_{j\to\infty}\|f_j\|_{L^t(\Omega)}

Now we are left to prove the claim: $\|f_j\|_{L^{1}(\Omega)}\le C.$

%{\bf Qianqiao, this is bad wrting, I modify it below 9/25/2017 By $\|f_j\|_{L^{q_\alpha}(\Omega)}=1$, we have $\|f_j\|_{L^{q}(\Omega)}\ge C_1(n,\alpha,q)>0$ since
%\begin{eqnarray*}
%&&\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j(y)dxdy\\
%&\ge&(1+\lambda d(\Omega))\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}f_j(y)dxdy\\
%&\ge&  (1+\lambda d(\Omega)) N_\alpha\|f_j\|_{L^{q_\alpha}(\Omega)}^2=(1+\lambda d(\Omega)) N_\alpha
%\end{eqnarray*} and  $\{f_j\}_{j=1}^\infty$ is a minimizing nonnegative sequence.}

From  $\|f_j\|_{L^{q_\alpha}(\Omega)}=1$, we have
\begin{eqnarray*}
&&\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}(1+\lambda|x-y|)f_j(y)dxdy\\
&\ge&(1-|\lambda| d(\Omega))\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}f_j(y)dxdy\\
&\ge&  (1-|\lambda| d(\Omega)) N_\alpha\|f_j\|_{L^{q_\alpha}(\Omega)}^2\\
&\ge& C>0.
\end{eqnarray*}
 Since  $\{f_j\}_{j=1}^\infty$ is a minimizing nonnegative sequence, we conclude that  $\|f_j\|_{L^{q}(\Omega)}\ge C_1(n,\alpha,q)>0$.
% Choose $q_1 \in (q, q_\alpha)$. Then $ q_1^\prime \in (p_\alpha, q^\prime)$ and
By H\"{o}lder's inequality, we have
\begin{eqnarray*}
0<C_1(n,\alpha,q)\le\|f_j\|_{L^{q}(\Omega)}\le C_2(n,\alpha,q)\|f_j\|_{L^{q_\alpha}(\Omega)}=C_2(n,\alpha,q).
\end{eqnarray*}
The upper bound on $ ||f_j\|_{L^{q}(\Omega)} $ indicates that
$$
\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}f_j(y)dxdy<C_3(n,\alpha,q).
$$
%Therefore on one hand, reversed H\"{o}lder's inequality gives
%\begin{eqnarray*}
%\int_{\Omega}\int_{\Omega}f_j(x)|x-y|^{\alpha-n}f_j(y)dxdy\ge \|f_j\|_{L^{q}(\Omega)}\|I_{\alpha, \Omega} f_j\|_{L^{q^\prime}(\Omega)},
%\end{eqnarray*}
%which implies that
It follows, via reversed H\"{o}lder's inequality, that
 $$\|I_{\alpha, \Omega} f_j\|_{L^{q^\prime}(\Omega)}\le C_4(n,\alpha,q)<\infty.$$

Further,  H\"{o}lder's inequality and reversed HLS inequality yield  that
\begin{eqnarray*}
\infty>C_4(n,\alpha,q) &\ge &\|I_{\alpha, \Omega} f_j\|_{L^{q^\prime}(\Omega)}\\
&\ge & C_5(n,\alpha,q) \|I_{\alpha, \Omega} f_j\|_{L^{p_\alpha}(\Omega)}\\
&\ge & C_6(n,\alpha,q) \|f_j\|_{L^{q_\alpha}(\Omega)}= C_6(n,\alpha,q)>0.
\end{eqnarray*}

Then we can show that for $M>0$ such that $M^{q^\prime}|\Omega|<\frac{1}{2} (C_4(n,\alpha,q))^{q^\prime}$, there exists $0<\delta<|\Omega|$ such that
\begin{equation}
\label{add-1}
m\{x: I_{\alpha, \Omega} f_j(x)\le M\}>\delta, ~{for~all} \ \ j.
\end{equation}
In fact, for  $\Omega_1:=\{x: I_{\alpha, \Omega} f_j(x)\le M\}$,
\begin{eqnarray*}
&&(C_4(n,\alpha,q))^{q^\prime}\\
&\le&
\int_{\Omega} (I_{\alpha, \Omega} f_j)^{q^\prime}dx=\int_{\Omega\setminus\Omega_1} (I_{\alpha, \Omega} f_j)^{q^\prime}dx+\int_{\Omega_1} (I_{\alpha, \Omega} f_j)^{q^\prime}dx\\
&\le & M^{q^\prime}|\Omega|+(\int_{\Omega_1} (I_{\alpha, \Omega} f_i)^{p_\alpha}dx)^{\frac{q^\prime}{p_\alpha}}|\Omega_1|^{1-\frac{q^\prime}{p_\alpha}}\\
&\le & M^{q^\prime}|\Omega|+(\int_{\Omega} (I_{\alpha, \Omega} f_i)^{p_\alpha}dx)^{\frac{q^\prime}{p_\alpha}}|\Omega_1|^{1-\frac{q^\prime}{p_\alpha}}\\
&\le & M^{q^\prime}|\Omega|+(\frac{C_6(n,\alpha,q)}{C_5(n,\alpha,q)})^{q^\prime}|\Omega_1|^{1-\frac{q^\prime}{p_\alpha}}\\
&<&\frac{1}{2}(C_4(n,\alpha,q))^{q^\prime}+|\Omega_1|^{1-\frac{q^\prime}{p_\alpha}}(\frac{C_6(n,\alpha,q)}{C_5(n,\alpha,q)})^{q^\prime},
\end{eqnarray*}
which yields the existence of such $\delta>0$.

Due to \eqref{add-1} we know that  there exists $\epsilon_0>0$,  such that for any $j$, we can find two points $x_j^1, x_j^2\in \Omega$ with the properties that $|x_j^1-x_j^2|\ge\epsilon_0$ and $$I_{\alpha, \Omega} f_j(x_j^1)=\int_{\Omega} f_j(y)|x_j^1-y|^{\alpha-n}\le M,~ I_{\alpha, \Omega} f_j(x_j^2)=\int_{\Omega} f_j(y)|x_j^2-y|^{\alpha-n}\le M.$$
So
\begin{eqnarray*}
\int_{\Omega}f_j(y)dy&\le & \int_{\Omega \setminus B(x_j^1, \frac{\epsilon_0}{4})}f_j(y)dy+\int_{\Omega \setminus B(x_j^2, \frac{\epsilon_0}{4})}f_j(y)dy\\
&\le& (\frac{4}{\epsilon_0})^{\alpha-n}\int_{\Omega \setminus B(x_j^1, \frac{\epsilon_0}{4})}f_j(y)|x_j^1-y|^{\alpha-n}dy\\
&&+(\frac{4}{\epsilon_0})^{\alpha-n}\int_{\Omega \setminus B(x_j^1, \frac{\epsilon_0}{4})}f_j(y)|x_j^2-y|^{\alpha-n}dy\\
&\le&(\frac{4}{\epsilon_0})^{\alpha-n}2M,
\end{eqnarray*}
uniformly for all $j$. We thus verify the claim, and hereby, complete the proof of Lemma \ref{Bound-attained}.
\end{proof}

\smallskip

It is standard to check that  the  minimizer $f(x)$ for energy $\xi_{\alpha,q}(\Omega)$ is positive, and, up to a constant multiplier, satisfies the following equation:
\begin{equation}\label{EL-equ-2}
f^{q-1}(x)=\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha}}dy+\lambda\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha-1}}dy,\quad x\in\overline\Omega.
\end{equation}
From the proof of Lemma \ref{Bound-attained}, we also know that $f(x) \in L^1(\Omega)$, thus $f^{q-1} (x) \in L^\infty(\Omega)$ by equation \eqref{EL-equ-2}.

 Writing $u(x)=f^{q-1}(x), p=q'$,  we thus find a weak positive solution $u(x)\in L^p(\Omega)$ to
\begin{equation}\label{EL-equ-3}
u(x)=\int_\Omega\frac{u^{p-1}(y)}{|x-y|^{n-\alpha}}dy+\lambda\int_\Omega\frac{u^{p-1}(y)}{|x-y|^{n-\alpha-1}}dy,\quad x\in\overline\Omega
\end{equation}
for $0>p>\frac{2n}{n-\alpha}=p_\alpha.$ To complete the proof of part (1) in Theorem \ref{main}, we need to show that $u \in C^1(\overline \Omega)$.
%
%\begin{lemma}\label{regularity-lemma}
%Suppose that $u\in L^p(\Omega)$ is a  positive solution to \eqref{EL-equ-2}. If $ p>p_\alpha,$  then  $u \in C^1(\overline \Omega).$
%\end{lemma}
%\begin{proof} {\bf Again, this is not well written. Where did we use $p>p_\alpha$?, etc. 9-26-2017}
%Firstly, we can see that $u\not\equiv\infty$. Thus we can prove that (by using the same argument of Lemma 2.6 in \cite{DGZ17})

In fact, $f(x)\in L^1(\Omega)$ implies
$$\int_\Omega u^{p-1}(y)dy<\infty.$$
It is easy to see that $u \in C(\overline \Omega)$ from equation \eqref{EL-equ-3}. To show $u\in C^1(\overline{\Omega})$, we can  directly compute, for $i=1,..., n$,
$$\partial_{x_i}u(x)=-(n-\alpha)\int_{\Omega}\frac{u^{p-1}(y)(x_i-y_i)}{|x-y|^{n-\alpha+2}}dy \in C(\overline{\Omega}).
$$
Part (1) of Theorem \ref{main} is hereby proved.

%On the other hand, $u \in C^1(\overline \Omega)$ also implies that $f(x)>0$ for every $x\in\overline\Omega$.
%{\bf(---since $\int_{\Omega}\frac{u^{p-1}(y)(x_i-y_i)}{|x-y|^{n-\alpha+2}}dy\le C\int_{\Omega}\frac{1}{|x-y|^{n-\alpha+1}}dy\le C\int_{B_R(0)}\frac{1}{|y|^{n-\alpha+1}}dy$ for $R>0$ large.---delete later--Guo 7/20/2017)}
%\end{proof}


\medskip

%follows from Lemma \ref{Bound-attained} and \ref{regularity-lemma}.

  It is interesting to study some properties about the  positive solutions to the new integral equation \eqref{EL-equ-3}(such as multiplicity of solutions, blowup behavior as $q \to q_\alpha$ in a star-shaped domain, etc.)
  %{(\bf I cannot see why this sentence is written here--Guo--10-04)}
  In the rest of this section,  as in \cite{DZ2017}, we will  show that
even though the boundary condition is not given pointwise,  the symmetric property for solutions to the integral equation \eqref{EL-equ-3} with $\lambda=0$  on a unit ball still holds. Contrary to the result in \cite{DZ2017}, here we will show that the solution is {\it monotone increasing} due to the monotone increasing property of the kernel.

%{\bf I am here 5-14-2018}

On $B_1:=B_1(0)=\{x\in\mathbb{R}^n\,|\, |x|<1,x\in\mathbb{R}^n\},$ for  $\lambda=0$ we rewrite the equation \eqref{EL-equ-3} as
\begin{equation}\label{UB-2}
u(x)=\int_{B_1}\frac{u^{p-1}(y)}{|x-y|^{n-\alpha}}dy,\quad x\in B_1(0).
\end{equation}
 We have
 \begin{theorem}\label{UB-symmetry}
Let $\alpha>n$, $p\in(p_\alpha, 0)$. Then every positive solution $u\in L^p (\overline{B_1})$ to \eqref{UB-2} is radially symmetric about the origin and strictly increasing in the radial direction.
\end{theorem}



Easy to see from the proof of Part (1) of Theorem \ref{main}  that $u\in C^{1}(\overline{B_1})$. We will use the method of moving planes to  prove Theorem \ref{UB-symmetry}.
%First, we remark that $u \in C(\overline \Omega)$ by Lemma \ref{regularity-lemma}. Theorem \ref{UB-symmetry} will be proved by the method of moving planes in integral form, similar to an early work  by  Chen, Fang and Yang in \cite{CFY2014}.


%{\bf 5-15-2018. nonlinearity has no Lip continuous, will it affect? this is the main different! No worry, since $u>c>0$}

Firstly, we recall the idea of the method of moving planes in $B_1$(see e.g. \cite{CFY2014,DZ2017}).

For any real number $\lambda\in (-1,0)$, define $T_\lambda=\{x\in \mathbb{R}^n \ | \ x_1=\lambda\},$
and  $x^\lambda=(2\lambda-x_1,x_2,\cdots,x_n)$ as the reflection of point $x=(x_1,x_2,\cdots,x_n)$ about  plane $T_\lambda$.
Let
\begin{eqnarray*}
\Sigma_\lambda=\{x=(x_1,x_2,\cdots,x_n)\in B_1|-1<x_1<\lambda\},
\end{eqnarray*}
and $\Sigma^C_\lambda=B_1\backslash \overline \Sigma_\lambda$ be the complement of $\Sigma_\lambda$ in $B_1$. Set $u_\lambda(x)=u(x^\lambda).$
We shall  complete the proof  in two steps. In step 1, we show that for $\lambda$ sufficiently close to $-1$,
\begin{equation}\label{UB-6}
u(x)\ge u_\lambda(x),  \quad \forall x\in \Sigma_\lambda.
\end{equation}
Then we can start to move  plane $T_\lambda$  along the $x_1$ direction. In step 2, we move the plane to the right as long as inequality \eqref{UB-6} holds. We show that the plane can be  moved to $\lambda=0$. So
\begin{equation}\label{UB-7}
u(-x_1,x_2,\cdots,x_n)\geq u(x_1,x_2,\cdots,x_n), \quad\forall x\in B_1,x_1\geq 0.
\end{equation}
Similarly, we can start to move  plane $T_\lambda$ from a place close to $\lambda=1$, and move it to the left limiting position $T_0$. Then
\begin{equation}\label{UB-8}
u(-x_1,x_2,\cdots,x_n)\leq u(x_1,x_2,\cdots,x_n),  \quad\forall x\in B_1,x_1\geq 0.
\end{equation}
By \eqref{UB-7} and \eqref{UB-8}, we have that $u(x)$ is symmetric about the plane $x_1=0$. Similarly, we can show that $u(x)$ is symmetric about any plane passing through the origin,   which then implies that $u(x)$ is radially symmetric about the origin and strictly increasing in the radial direction.

First, we have following comparison inequality.
\begin{lemma}\label{Lm-symmetry-1}
For any $x\in \Sigma_\lambda$ with $\lambda\in (-1,0)$, it holds
\begin{equation}\label{UB-3}
u(x)-u_\lambda(x)\geq\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]({u^{p-1}(y)}-{u^{p-1}_\lambda(y)})dy.
\end{equation}
\end{lemma}


\begin{proof}
Let $\widetilde{\Sigma}_\lambda=\{x^\lambda~|~x\in{\Sigma}_\lambda\}$ be the reflection of $\Sigma_\lambda$ about  plane $T_\lambda$, then
\begin{eqnarray*}
u(x)&=&\int_{\Sigma_\lambda}\frac{u^{p-1}(y)}{|x-y|^{n-\alpha}}dy+\int_{\widetilde{\Sigma}_\lambda}\frac{u^{p-1}(y)}{|x-y|^{n-\alpha}}dy\\
&+&\int_{\Sigma^C_\lambda\backslash\widetilde{\Sigma}_\lambda}\frac{u^{p-1}(y)}{|x-y|^{n-\alpha}}dy.
\end{eqnarray*}
%u_\lambda(x)&=&\int_{\Sigma_\lambda}\frac{u^{p-1}(y)}{|x^\lambda-y|^{n-\alpha}}dy+\int_{\widetilde{\Sigma}_\lambda}\frac{u^{p-1}(y)}{|x^\lambda-y|^{n-\alpha}}dy\\
%&+&\int_{\Sigma^C_\lambda\backslash\widetilde{\Sigma}_\lambda}\frac{u^{p-1}(y)}{|x^\lambda-y|^{n-\alpha}}dy.
%\end{eqnarray*}  {\bf No need 4/4/2017}
Noting that $|x-y|>|x^\lambda-y|$ in $x\in \Sigma_\lambda, y\in\Sigma^C_\lambda\backslash\widetilde{\Sigma}_\lambda $, we have
\begin{eqnarray*}
u(x)-u_\lambda(x)&=&\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy\\
&&+\int_{\widetilde{\Sigma}_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy\\
&&+\int_{\Sigma^C_\lambda\backslash\widetilde{\Sigma}_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy\\
&\geq&\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy\\
&&+\int_{\Sigma_\lambda}\big[\frac1{|x-y^\lambda|^{n-\alpha}}-\frac1{|x^\lambda-y^\lambda|^{n-\alpha}}\big]{u^{p-1}_\lambda(y)}dy\\
&=&\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy\\
&&-\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]{u^{p-1}_\lambda(y)}dy\\
&=&\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]({u^{p-1}(y)}-{u_\lambda^{p-1}(y)})dy.
\end{eqnarray*}
\end{proof}

%We are now ready to prove Theorem 3.4.

\noindent \textbf{Proof of Theorem \ref{UB-symmetry}.}

\noindent  \textbf{Step 1.}  Let $u\in C^1(\overline{B_1})$ be a positive solution to equation  \eqref{UB-2}.  We show that for $\lambda$  sufficiently close to $-1$, inequality \eqref{UB-6} holds.

From \eqref{UB-2} we have
\begin{eqnarray*}
\frac{\partial u(x)}{\partial x_1}|_{x_1=-1}
&=&(\alpha-n)\int_{B_1}|x-y|^{\alpha-n-2}(-1-y_1)u^{p-1}(y)dy\\
&<&0.
\end{eqnarray*}
%For $\lambda$ larger than and sufficiently close to $-1$, $x\in\Sigma_\lambda$, from \eqref{UB-2} we have
%\begin{eqnarray*}
%& &\nabla\big(|x|^\frac{n-\alpha}2u(x)\big)\cdot x=|x|^\frac{n-\alpha}2\big(\nabla u(x)\cdot x+\frac{n-\alpha}2u(x)\big)\\
%&=&(\alpha-n)|x|^\frac{n-\alpha}2\int_{B_1}|x-y|^{\alpha-n-2}\big[\sum_1^n(x_i-y_i)\cdot x_i+\frac{|x-y|^2} 2\big]u^{p-1}(y)dy\\
%&=&\frac{\alpha-n}2|x|^\frac{n-\alpha}2\int_{B_1}|x-y|^{\alpha-n-2}\big[|x|^2-|y|^2\big]u^{p-1}(y)dy\\
%&\ge&0.
%\end{eqnarray*}
Therefore for $\lambda$ sufficiently close to $-1$, we have
\begin{eqnarray*}
 u(x) \ge u_\lambda(x)     \quad\text{for} ~ x\in\Sigma_\lambda.
\end{eqnarray*}
%Since $\alpha>n$  and $|x|>|x_\lambda|$ for $x\in\Sigma_\lambda$, we have
%\[
%u(x) \ge u_\lambda(x).
%\]


\noindent \textbf{Step 2.} Plane $T_\lambda$ can be moved continuously towards right to its limiting position as long as inequality \eqref{UB-6} holds.

Define
\[\lambda_0=\sup\{\lambda\in [-1,0)\ | \ u(y)\geq u_\mu(y),\forall y \in\Sigma_\mu, \ \mu\leq\lambda\}.\]
We claim that $\lambda_0$ must be 0.

We prove it by contradiction. Suppose not, that is,  $\lambda_0<0$.

 We first  show that
\[u(x)>u_{\lambda_0}(x),\quad\text{in} ~\Sigma_{\lambda_0}.\]
Hence, we have
\[u(x)-u_{\lambda_0}(x)>c_1>0,\quad \text{in} ~\Sigma_{\lambda_0-\epsilon_1}
\]for $\epsilon_1>0$ small enough.

In fact, since $|x-y|<|x-y^{\lambda_0}|$ for $x, y\in{\Sigma}_{\lambda_0}$, we have, similar to the calculation in the proof of  Lemma \ref{Lm-symmetry-1}, that
\begin{eqnarray}
u(x)-u_{\lambda_0}(x)
&=&\int_{\Sigma_{\lambda_0}}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^{\lambda_0}-y|^{n-\alpha}}\big]({u^{p-1}(y)}-{u_{\lambda_0}^{p-1}(y)})dy\nonumber\\
&+&\int_{\Sigma^C_{\lambda_0}\backslash\widetilde{\Sigma}_{\lambda_0}}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^{\lambda_0}-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy\nonumber\\
&\ge&\int_{\Sigma^C_{\lambda_0}\backslash\widetilde{\Sigma}_{\lambda_0}}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^{\lambda_0}-y|^{n-\alpha}}\big]{u^{p-1}(y)}dy.\label{UB-10}
\end{eqnarray}
If there exists some point $x_0\in\Sigma_{\lambda_0}$ such that $u(x_0)= u_{\lambda_0}(x_0)$, then  since $|x-y|>|x^{\lambda_0}-y|$ for $x\in{\Sigma}_{\lambda_0},y\in\Sigma^C_{{\lambda_0}}$, we deduce  from \eqref{UB-10} that
\[u(y)\equiv \infty,\quad \forall y\in\Sigma^C_{\lambda_0}\backslash\widetilde{\Sigma}_{\lambda_0}.\]
This contradicts to the  assumption that $u\in C^1(\overline{B_1})$ is a positive solution.
%Hence $u(x)-u_{\lambda_0}(x)>0$ for all $ x \in \Sigma_{\lambda_0}.$

For some small $\delta_1>0$, we choose $\varepsilon \in (0, \epsilon_1)$  small enough  such that for any  $\lambda\in [\lambda_0,\lambda_0+\varepsilon)$, there holds
\[u(x)\ge u_{\lambda}(x), \forall x\in \Sigma_{\lambda_0-\varepsilon_1},\]
%On the other hand, we also can take $\varepsilon_1>0$ and $\varepsilon>0$ small enough such that
and
\[
|\frac1{|x-y|^{n-\alpha}}-\frac1{|x^{\lambda}-y|^{n-\alpha}}|\le\delta_1 ~~~~~~~~~ \quad\text{for}~x\in\Sigma_\lambda\backslash\Sigma_{\lambda_0-\varepsilon_1}.
\] % independent of $\epsilon$,

Write
\begin{equation*}
\Sigma^u_\lambda=\{x\in \Sigma_\lambda |u_\lambda(x)>u(x)\}.
\end{equation*}
It follows from \eqref{UB-3} that for any $x\in\Sigma^u_\lambda$,
\begin{eqnarray*}
0> u(x)-u_\lambda(x)&\ge&\int_{\Sigma_\lambda}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]({u^{p-1}(y)}-{u^{p-1}_\lambda(y)})dy\\
&\ge&\int_{\Sigma_\lambda^u}\big[\frac1{|x-y|^{n-\alpha}}-\frac1{|x^\lambda-y|^{n-\alpha}}\big]({u^{p-1}(y)}-{u^{p-1}_\lambda(y)})dy\\
&\ge&-\delta_1\int_{\Sigma_\lambda^u}({u^{p-1}(y)}-{u^{p-1}_\lambda(y)})dy.
\end{eqnarray*}
 Since $u\in C^1(\overline{B_1})$,  there exists a positive constant $C_0$ such that $\frac 1{C_0}\le u\le C_0$.   It  follows from the above
\begin{eqnarray*}
 \int_{\Sigma_\lambda^u}(u_\lambda(x)-u(x))dx
&\le&\delta_1\int_{\Sigma_\lambda^u}\int_{\Sigma_\lambda^u}({u^{p-1}(y)}-{u^{p-1}_\lambda(y)})dydx\\
&\le& (1-p)\delta_1\int_{\Sigma_\lambda^u}\int_{\Sigma_\lambda^u}u^{p-2}(y)({u_\lambda(y)}-{u(y)})dydx\\
&\le&C\delta_1(\varepsilon+\varepsilon_1)^n\int_{\Sigma_\lambda^u}({u_\lambda(y)}-{u(y)})dy.
\end{eqnarray*}
It implies that
\[\|u_{\lambda}-u\|_{L^1(\Sigma^u_\lambda)}\equiv 0,\]
 for $\delta_1, \varepsilon, \varepsilon_1$ small enough, and hence $\Sigma^u_\lambda$ must have measure zero.

%We arrive at
%\[u(x)-u_{\lambda}(x)\geq 0, \quad a.e. ~x\in \Sigma_\lambda, ~\forall\lambda\in [\lambda_0,\lambda_0+\varepsilon).\]
%Actually we have
We thus have
\[u(x)-u_{\lambda}(x)\geq 0, \quad~ \text{for~ any} ~x\in \Sigma_\lambda, ~\forall\lambda\in [\lambda_0,\lambda_0+\varepsilon)\]
since $u$ is continuous. This contradicts to the definition of $\lambda_0$.
Hence, $\lambda_0=0$. We hereby  complete the proof of Theorem 3.4.
\hfill$\Box$
\medskip


%{(\bf may deleted the following theorem later---Guo 8/24/2017)}
%In a quite similar spirit, we shall use the method of moving sphere
%to establish the following Liouville theorem, what is crucial for blowup analysis in the study of integral equations (see the proof of Proposition 4.2 below).
%
%%{\bf say a few word about why we choose sphere over plane here later. 3/16/2017}
%%{\bf I am here now. 3/17/2017, rest of this section, waiting for Jingbo. You need to make a Kelvin transformation first. 3/26/2017}
%
%
%
%\begin{theorem}\label{nonexistence-Half space}
%There is no positive solution $u\in C(\overline{\mathbb{R}^n_+})$  to \begin{equation}\label{RHB-4}
%u(x)=\int_{\mathbb{R}^n_+}\frac{u^{q_\alpha-1}(y)}{|x-y|^{n-\alpha}}dy,\quad x\in \mathbb{R}^n_+.
%\end{equation}
%% then $u\equiv0$.
%\end{theorem}
%\begin{proof}
%It follows from Lemma \ref{regularity-lemma} that a bounded solution to \eqref{RHB-4} is continuous up to the boundary.
%We prove the above theorem  for $u\in C(\Bbb{R}^n)$ via the method of moving sphere, which is introduced by Li and Zhu in \cite{LZ1995}.
%
%
%
%
%\begin{lemma}\label{Reg-lm-1}
%For  $1\le n<\alpha$ and $q_\alpha<0$, if $ u$  is a  positive Lebesgue measurable solutions to \eqref{RHB-4}, then
%\begin{eqnarray*}
%&(i)&~~\int_{\mathbb{R}^n_+}(1+|y|^{\alpha-n})u^{q_\alpha-1}(y)dy<\infty ;\\
%&(ii)& ~~a:=\lim_{|x|\to\infty}|x|^{n-\alpha}u(x)=\int_{\mathbb{R}^n_+}u^{q_\alpha-1}(y)dy<\infty, \\
%&(iii) &\text{for~some~constants}~C_1>0,\\
%\quad  & &\frac{1+|x|^{\alpha-n}}{C_1}\le u(x)\le C_1(1+|x|^{\alpha-n}),~~~~~~~~~~~~~~\forall\,x\in\mathbb{R}^n_+.
%\end{eqnarray*}
%\end{lemma}
%\begin{proof} The proof is the same as that of Lemma 5.1 in Li \cite{Li2004}. We include details for the completion of the paper.
%
%Since $u \not\equiv \infty$, we know that
%\[meas\{x\in\mathbb{R}^n_+\,:\,u(x)<\infty \}>0.
%\]
%Thus, there exist $R>1$ and some measurable set $E$ such that
%\[E\subset \{x\in \mathbb{R}^n_+\,:\,u(x)<R\}\cap B_R
%\] with $|E|>\frac1R.$
%It follows, for any $y\in\mathbb{R}^n_+$, that
%\begin{eqnarray*}
%u(x)&=&\int_{\mathbb{R}^n_+}|x-y|^{\alpha-n} u^{q_\alpha-1}(y)dy\\
%&\ge&\int_{E}|x-y|^{\alpha-n} u^{q_\alpha-1}(y)dy\\
%&\ge&R^{q_\alpha-1}\int_{E} |x-y|^{\alpha-n}dy.
%\end{eqnarray*}  And then,
%\begin{eqnarray*}
%\lim_{|x|\to\infty}\frac{u(x)}{(1+|x|^{\alpha-n})}
%&\ge&\lim_{|x|\to\infty}\frac{R^{q_\alpha-1}}{(1+|x|^{\alpha-n})}\int_{E} |x-y|^{\alpha-n}dy=CR^{q_\alpha-2}.
%\end{eqnarray*}
%This shows
%\[{u(x)}\ge\frac{(1+|x|^{\alpha-n})}{C_1}.
%\]
%This implies that the left hand side inequalities in $(iii)$ hold.
%
%
%On the other hand, for some $x_0\in\mathbb{R}^n $ with $1\le|x_0|\le2$,
%\begin{eqnarray*}
%\int_{\mathbb{R}^n_+}|x_0-y|^{\alpha-n} u^{q_\alpha-1}(y)dy
%&=&u(x_0)<\infty;
%\end{eqnarray*}
%From the left hand side inequalities in $(iii)$ and the above, we obtain $(i)$.
%
%For $|x|\ge1$,
%\begin{eqnarray*}
%\frac{|x-y|^{\alpha-n}}{|x|^{\alpha-n}} u^{q_\alpha-1}(y)\le(1+|y|^{\alpha-n} u^{q_\alpha-1}(y).
%\end{eqnarray*}
%Combining this  with  $(i)$ and  using the dominated convergence theorem we have $(ii):$
%\begin{eqnarray*}
%a&=&\lim_{|x|\to\infty}|x|^{n-\alpha}u(x)=\lim_{|x|\to\infty}
%\int_{\mathbb{R}^n_+}\frac{|x-y|^{\alpha-n}}{|x|^{\alpha-n}} u^{q_\alpha-1}(y)dy=\int_{\mathbb{R}^n_+} u^{q_\alpha-1}(y)dy<\infty.
%\end{eqnarray*}
%Combining $(i)$ and $(ii)$ with \eqref{RHB-4}, we have the right side inequality in $(iii)$.
%\end{proof}
%
%%\begin{lemma}\label{neg-lm-2}
%%For  $1\le n<\alpha $ and $\theta,\kappa<0$, if $(u,v)$ is a pair of positive Lebesgue measurable solutions to \eqref{RHB-4}, then $u,v\in C^\infty( \mathbb{R}^n)$.
%%\end{lemma}
%%\begin{proof} Again, we adopt the proof given in Li \cite{Li2004}. For $R>0,$ we can split $u$ into  following two parts
%%\begin{eqnarray*}
%%u(y)&=& \int_{|y|\le2R} |x-y|^{\alpha-n} v^{\kappa}(x)dx
%%+\int_{|y|>2R}|x-y|^{\alpha-n} v^{\kappa}(x)dx\\
%%&=&J_1(x)+J_2(x).
%%\end{eqnarray*}
%% From  Lemma \ref{Reg-lm-1} $(i)$ we know that $J_2(x)$  can be differentiated under the integral for $|y|<R,$ so $J_2\in C^\infty({B_R})$. On the other hand, by Lemma \ref{Reg-lm-1} $(iii)$, we have $v^{\kappa}\in L^\infty(B_{2R})$, it is obvious that $J_1$ is at least H\"{o}lder continuous in ${B_R}$. Since $R>0$ is arbitrary, $u$ is H\"{o}lder continuous in ${\mathbb{R}^n}$. Thus, $u^{\theta}$ is H\"{o}lder continuous in $B_{2R}$. Similarly, we have $v,v^{\kappa}$ are H\"{o}lder continuous in $B_{2R}$. By bootstrap, we conclude that $u,v\in C^\infty(\mathbb{R}^n)$.
%%\end{proof}
%
%
%
%For $R>0,$ denote
%$$B_R^+(x)=\{y=(y_1,y_2,\cdots,y_n)\in B_R(x)\,: \, y_n>0,x\in\partial\mathbb{R}^{n}_+\},
%\Sigma_{x,R}^n=\mathbb{R}^n_+\backslash\overline{B_R^+(x)}.$$
%For $x\in\partial \mathbb{R}^n_+$ and $\lambda>0,$ we define
%\[
%\omega_{x,\lambda}(\xi)=\big(\frac\lambda{|\xi-x|}\big)^{n-\alpha}\omega(\xi^{x,\lambda}),\quad\xi\in
%\overline{\mathbb{R}^n_+}\,\backslash\{x\},
%\]
%where
%$$\xi^{x,\lambda}=x+\frac{\lambda^2(\xi-x)}{|\xi-x|^2}$$
%is the Kelvin transformation of $\xi$ with respect to $\partial B_\lambda^+ (x)$. Also we write $\omega^k_{x,\lambda}(\xi):=
%(\omega_{x,\lambda}(\xi))^k$.
%
%First we have following lemma.
%\begin{lemma}\label{Kelvin formula}
%%Let $1<\alpha<n, 1<p<\infty$ and $u$ be a positive solutions to equation \eqref{HB-4}. Then,
%For any $x\in
%\partial\mathbb{R}^{n}_+$,
%\begin{eqnarray*}
%u_{x,\lambda}(\xi)&=&\int_{\mathbb{R}^{n}_+}\frac{u_{x,\lambda}^{q_\alpha-1}(\eta)}
%{|\xi-\eta|^{n-\alpha}}d\eta,\quad \forall \, \xi\in\mathbb{R}^{n}_+.\label{K-1}\\
%\end{eqnarray*}
%Moreover,
%\begin{eqnarray}
%u_{x,\lambda}(\xi)-u(\xi)&=&\int_{\Sigma_{x,\lambda}^n}P(x,\lambda;\xi,\eta)
%\big[ u_{x,\lambda}^{q_\alpha-1}(\eta)-u^{q_\alpha -1}(\eta)\big]d\eta,~~~~~~~~~~~~~~~~~\label{K-2}
%\end{eqnarray} where
%\[
%P(x,\lambda;\xi,\eta)=\frac1{|\xi-\eta|^{n-\alpha}}
%-\big(\frac\lambda{|\xi-x|}\big)^{n-\alpha}\frac1{|\xi^{x,\lambda}-\eta|^{n-\alpha}},
%\]and
%\[
%P(x,\lambda;\xi,\eta)>0,~~~~~~~~~~~~\text{for}~\forall \,\xi,\eta\in\Sigma_{x,\lambda}^n,\lambda>0.
%\]
%\end{lemma}
%
%
%
%
%The proof of Lemma \ref{Kelvin formula} is similar to that in \cite{Li2004}. We shall skip the details here.
%
%
%%It is clear that in Lemma \ref{Kelvin formula} $\tau=0$ holds if and only if $p=q_\alpha-1=\frac{n+\alpha}{n-\alpha}.$ From now on in this section, we assume  $p=q_\alpha-1$.
%
%
%Define $\Sigma_{x,\lambda}^u=\{\xi\in\Sigma_{x,\lambda}^{n}\,|\,u(\xi)<u_{x,\lambda}(\xi)\}$.
%
%\begin{lemma}\label{K-lm-1}
%%Assume the same conditions in Theorem \ref{nonexistence-Half space}. Then
%For any $x\in\partial\mathbb{R}^{n}_+$,
%there exists $\lambda_0(x)>0$ such that: $\forall \ 0<\lambda<\lambda_0(x),$
%\begin{eqnarray*}
%u_{x,\lambda}(\xi) \ge u(\xi),\quad  \, \forall \xi \in   \Sigma_{x,\lambda}^{n}.
%\end{eqnarray*}
%\end{lemma}
%\begin{proof} Without loss of generality, we assume that $x=0$ and $\lambda_0(x)=\lambda_0.$
% Since $n<\alpha$ and $u\in C^1(\mathbb{R}^n_+)$ is a positive function, there exists $r_0\in(0, 1),$ such that
%\[\nabla_\xi\big(|\xi|^{\frac{n-\alpha}2}u(\xi)\big)\cdot \xi<0,~~~~~~~~~~~~~~~~~~\forall\,0<|\xi|<r_0.
%\]
%Thus,
%\begin{equation}\label{cla-1}
%u_{\lambda}(\xi)>u(\xi),~~~~~~~~~~~~~~~~~~~~~~~\forall\,0<\lambda<|\xi|<r_0.
%\end{equation}
%
%Using Lemma \ref{Reg-lm-1} $(iii),$  we have
%\begin{equation*}\label{cla-2}
%u(\xi)\le C(r_0)|\xi|^{\alpha -n},~~~~~~~~~~~~ ~~~~~~~~~\forall ~~~~|\xi|\ge r_0.
%\end{equation*}
%For small $\lambda_0\in(0,r_0)$ and any $0<\lambda<\lambda_0,$ by
%$(iii)$ of Lemma \ref{Reg-lm-1} and \eqref{cla-1}
%\begin{eqnarray*}
%u_{\lambda}(\xi)=\big(\frac\lambda{|\xi|}\big)^{n-\alpha}u(\frac{\lambda^2\xi}{|\xi|^2})
%\ge\big(\frac{|\xi|}{\lambda_0}\big)^{\alpha -n}\inf_{B_{r_0}}u\ge
%u(\xi),~~~~~~~~~~~~~~~~~~~~~\quad|\xi|\ge r_0.
%\end{eqnarray*}
%Combining the above with \eqref{cla-1}, we conclude
% \[
% u_{x,\lambda}(\xi) \ge u(\xi), \ \ \ \, \quad ~~~~~\forall \xi \in~\Sigma_{x,\lambda}
% \]
%with $x=0$ and $\lambda_0(x)=\lambda_0.$
%\end{proof}
%
%%{\bf in the proof of \eqref{K-3}, we use $u$ to control $u_{x, \lambda} $ in $\Sigma^u_{x, \lambda}$ due to Kelvin transformation. So we do not need to worry about $u_{x, \lambda} $ as $|x| \to 0.$  4/3/2017}
%
%
%
%\medskip
%
%Define
%\[\bar{\lambda}(x)=sup\{\mu>0\,|\,u_{x,\lambda}(\xi)\le u (\xi), \  \forall \lambda\in (0, \mu), \forall \xi\in\Sigma_{x,\lambda}^{n}\}.
%\]
%We will show: if the sphere stops, then we have symmetric properties for solutions.
%
%\begin{lemma}\label{K-lm-2}For some $x_0\in
%\partial\mathbb{R}^{n}_+$,
%if $\bar{\lambda}(x_0)<\infty$, then
%\begin{eqnarray*}
%u_{x_0,\bar{\lambda}(x_0)}=u\quad \text{on}~\mathbb{R}^{n}_+.
%\end{eqnarray*}
%\end{lemma}
%\begin{proof}
%
%Again, the proof is similar to that of Lemma 5.5 in \cite{Li2004}.
%
%Without loss of generality, we assume that
%$x_0=0$, and write $\bar{\lambda}=\bar{\lambda}(0), \, u_\lambda=u_{0,\lambda}, \, v_\lambda=v_{0,\lambda}, \,
%\xi^\lambda=\xi^{0,\lambda}.$
%By the definition of $\bar{\lambda}$,
%\begin{eqnarray*}
%u_{\bar{\lambda}}(\xi)\ge
%u(\xi),\quad\forall\,|\xi|\ge\bar{\lambda}.
%\end{eqnarray*}
% Thus, using \eqref{K-2} with $x=0,\lambda=\bar{\lambda},$ and the positivity of the kernel,
% we know that there are following two cases:
%
%  $(\mathbf{a})$ $u_{\bar{\lambda}}(\xi)=u(\xi)$   ;
%  or
%  $(\mathbf{b})$ $u_{\bar{\lambda}}(\xi)>u(\xi)$.
%
%  We show  that case $(\mathbf{b})$ can not happen. More precisely, supposing that  $u_{\bar{\lambda}}(\xi)>u(\xi)$ for all $|\xi|\ge\bar{\lambda}$, we will show that there is a
%$\varepsilon_*>0$, such that, for any $\lambda\in (\bar \lambda, \bar{\lambda}+\varepsilon_*)$, $u_{{\lambda}}(\xi) \ge u(\xi)$ for any
%$|\xi|>{\lambda}.$ This contradicts to the
%definition of $\bar{\lambda}$. We will show this via two steps.
%
% Step $1$. There is a $\varepsilon_1\in(0,1)$, such that for any
%$\varepsilon< \varepsilon_1$, $\bar{\lambda}\le
%\lambda\le\bar{\lambda}+\varepsilon$, if $|\xi|\ge\bar{\lambda}+1$, then
%\begin{eqnarray*}
%u_{\lambda}(\xi)-u(\xi)\ge \frac{\varepsilon_1}2|\xi|^{\alpha-n}.
%\end{eqnarray*}
%
%From Lemma \ref{Kelvin formula}, we know that $P(x,\lambda,\xi,z)>0$
%$\forall \,
%\xi \in \Sigma_{x,\lambda}$.
%By \eqref{K-2} with $x=0,\lambda=\bar{\lambda},$  and  Fatou
%Lemma, we know, for all $|\xi|\ge\bar{\lambda}$, that
%\begin{eqnarray*}
%& &\liminf_{|\xi|\to\infty}|\xi|^{n-\alpha}(u_{\bar{\lambda}}(\xi)-u(\xi))\\
%&\ge&\int_{\Sigma_{\lambda}}\liminf_{|\xi|\to\infty}|\xi|^{n-\alpha} P(0,\bar{\lambda},\xi,\eta)[u^{q_\alpha-1}(\eta)
%-u^{q_\alpha-1}_{\bar{\lambda}}(\eta)]d\eta\\
%&=&\int_{\Sigma_{\lambda}}\big(\big(\frac{\bar{\lambda}}{|\eta|}\big)^{n-\alpha}-1\big)[u^{q_\alpha-1}(\eta)
%-u^{q_\alpha-1}_{\bar{\lambda}}(\eta)]d\eta.
%\end{eqnarray*}
%Thus, using the positivity of $u_{\bar{\lambda}}-u$, we know that there exists $\varepsilon_2\in(0,1)$, such that
%\begin{eqnarray*}
%u_{\bar{\lambda}}(\xi)-u(\xi)\ge
%\varepsilon_2|\xi|^{\alpha-n},~~~~~~~~~~~~~~~~\forall\,|\xi|\ge\bar{\lambda}+1.
%\end{eqnarray*}
%
%Due to the continuity of $u$,  there exists a
%$\varepsilon_3\in(0,\varepsilon_2)$ such that for
%$|\xi|\ge\bar{\lambda}+1$ and $\bar{\lambda}\le
%\lambda\le\bar{\lambda}+\varepsilon_3$,
%\begin{eqnarray*}
%|u_\lambda(\xi)-u_{\bar{\lambda}}(\xi)|&=&|\big(\frac\lambda{|\xi|}\big)^{n-\alpha}u(\frac{\lambda^2\xi}{|\xi|^2})
%-\big(\frac{\bar{\lambda}}{|\xi|}\big)^{n-\alpha}u(\frac{\bar{\lambda}^2\xi}{|\xi|^2})|\\
%&\le& \frac{\varepsilon_3}{2}|\xi|^{\alpha-n}.
%\end{eqnarray*} Thus
%\begin{equation*}
%u_\lambda(\xi)-u(\xi)=u_{\bar{\lambda}}(\xi)-u(\xi)+u_\lambda(\xi)-u_{\bar{\lambda}}(\xi)
%\ge\frac{\varepsilon_2}{2}|\xi|^{\alpha-n},
%\end{equation*}
% for all $|\xi|\ge\bar{\lambda}+1,\bar{\lambda}\le\lambda\le\bar{\lambda}+\varepsilon_2.$
%we complete the proof for Step $1$.
%
%Step $2$. There is a $\varepsilon_* < \varepsilon_1$, such that for any $\varepsilon < \varepsilon_*$,
%$\bar{\lambda}\le\lambda \le\bar{\lambda}+\varepsilon ,$ if $\xi \in
%\mathbb{R}^n_+$ satisfy  $\lambda\le |\xi|\le\bar{\lambda}+1$, then
%$u_{\lambda}(\xi)-u(\xi)\ge0$.
%
%Let $\varepsilon_*\in(0,\varepsilon_1)$.
% We have, for $\bar{\lambda}\le
%\lambda\le\bar{\lambda}+\varepsilon_*$ and
%$\lambda\le|\xi|\le\bar{\lambda}+1$,
%\begin{eqnarray}\label{cla-5}
%u_\lambda(\xi)-u(\xi)&=&\int_{\Sigma_\lambda}P(0,\lambda;\xi,\eta)\big(
%u^{q_\alpha-1}(\eta)-u^{q_\alpha-1}_{\lambda}(\eta)\big)d\eta\nonumber\\
%&\ge&\int_{\Sigma_{\lambda}\backslash\Sigma_{\bar{\lambda}+1}}P(0,\lambda;\xi,\eta)\big(
%u^{q_\alpha-1}(\eta)-u^{q_\alpha-1}_{\lambda}(\eta)\big)d\eta\nonumber\\
%&
%&+\int_{\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{ \bar{\lambda}+3}}P(0,\lambda;\xi,\eta)\big(
%u^{q_\alpha-1}(\eta)-u^{q_\alpha-1}_{\lambda}(\eta)\big)d\eta\nonumber\\
%&\ge&\int_{\Sigma_{\lambda}\backslash\Sigma_{\bar{\lambda}+1}}P(0,\lambda;\xi,\eta)\big(
%u^{q_\alpha-1}_{\bar{\lambda}}(\eta)-u^{q_\alpha-1}_{\lambda}(\eta)\big)d\eta\nonumber\\
%&
%&+\int_{\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{ \bar{\lambda}+3}}P(0,\lambda;\xi,\eta)\big(
%u^{q_\alpha-1}(\eta)-u^{q_\alpha-1}_{\lambda}(\eta)\big)d\eta.
%\end{eqnarray}
%By Step $1$, there exists $\delta_1>0$ such that
%\begin{eqnarray*}
%u^{q_\alpha-1}(\eta)
%-u^{q_\alpha-1}_{\lambda}(\eta)\ge
%\delta_1,~~~~~~~~~~~~~~~~\forall \eta\in\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{ \bar{\lambda}+3}.
%\end{eqnarray*}
% Since
% \[P(0,\lambda;\xi,\eta)=0,~~~~~~~~~~~~~~~~\forall\,|\xi|=\lambda,
% \]
% \[\nabla_\xi P(0,\lambda;\xi,\eta)\cdot\xi|_{|\xi|=\lambda}=(\alpha-n)|\xi-\eta|^{\alpha-n-2}\big(|\eta|^2-|\xi|^2\big)>0,~~~~~~~~~~~~~~~~~~~~~~~~
% ~~~~\forall\,\eta \in\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{ \bar{\lambda}+3},
% \]
%and the function is smooth in the relevant region, it follows, also based on the positivity of kernel, that
%\[P(0,\lambda;\xi,\eta)\ge\delta_2(|\xi|-\lambda),~~~~~~~~~\forall\,\lambda\le|\xi|\le\bar{\lambda}+1,
%\forall \eta\in\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{ \bar{\lambda}+3}.
%\]
% where $\delta_2>0$ is some constant independent of $\varepsilon_*$. It is easy to see that for some constant $C>0$ (independent of $\varepsilon_*$), and $\bar{\lambda}\le\lambda\le\bar{\lambda}+\varepsilon_*$,
% \[
% |u^{q_\alpha-1}_{\bar{\lambda}}(\eta)-u^{q_\alpha-1}_{\lambda}(\eta)|\le C\varepsilon,~~~~~~~~~~
% ~\forall\,\eta\in\Sigma_{\lambda}\backslash\Sigma_{ \bar{\lambda}+1}.
% \]
%Using the mean value theorem, we have, for $\lambda\le|\xi|\le\bar{\lambda}+1$, that
%\begin{eqnarray*}
%& &\int_{\Sigma_{{\lambda}}\backslash\Sigma_{\bar \lambda +1}}P(0,\lambda;\xi,\eta)d\eta
% = \int_{\Sigma_{{\lambda}}\backslash\Sigma_{\bar \lambda +1}}
%\big(\big(\frac{|\xi|}{\lambda}\big)^{\alpha-n}|\xi^\lambda-\eta|^{\alpha-n}-|\xi-\eta|^{\alpha-n}\big)d\eta\\
%&=& \int_{\Sigma_{{\lambda}}\backslash\Sigma_{\bar \lambda +1}}
%\big[(\big(\frac{|\xi|}{\lambda}\big)^{\alpha-n}-1)|\xi^\lambda-\eta|^{\alpha-n}+\big(|\xi^\lambda-\eta|^{\alpha-n}-|\xi-\eta|^{\alpha-n}\big)\big]d\eta\\
%&\le& C(|\xi|-\lambda)+ \int_{\Sigma_{{\lambda}}\backslash\Sigma_{\bar \lambda +1}}
%\big(|\xi^\lambda-\eta|^{\alpha-n}-|\xi-\eta|^{\alpha-n}\big)d\eta\\
%&\le& C(|\xi|-\lambda)+ C|\xi^\lambda-\xi|\\
%&\le&C(|\xi|-\lambda).
%\end{eqnarray*} Thus, for  $\varepsilon\in (0, \varepsilon_*), \, \bar{\lambda}\le\lambda\le\bar{\lambda}+\varepsilon, \, \lambda\le|\xi|\le\bar{\lambda}+1$,  from \eqref{cla-5} it follows
%\begin{eqnarray*}
%u_\lambda(\xi)-u(\xi)&\ge&-C\varepsilon\int_{\Sigma_{\lambda}\backslash \Sigma_{\bar{\lambda}+1} }P(0,\lambda;\xi,\eta)d\eta
%+\delta_1\delta_2(|\xi|-\lambda)\int_{\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{\bar{\lambda}+3}}d\eta\\
%&\ge&\big(\delta_1\delta_2\int_{\Sigma_{\bar{\lambda}+2}\backslash\Sigma_{\bar{\lambda}+3}}d\eta-C\varepsilon\big)(|\xi|-\lambda)\ge0.
%\end{eqnarray*}
%Step $2$ is established. Hence we complete the proof for Lemma \ref{K-lm-2}.
%\end{proof}
%
%
%
%\medskip
%
%The following two key calculus lemmas are needed for carrying out moving sphere procedure. Under a stronger assumption ($f\in
%C^1(\mathbb{R}^n)$), these lemmas were early proved by Li and Zhu \cite{LZ1995} (see, also, Li and Zhang \cite{LZ2003}). The current forms, due to Li and Nirenberg, are adopted from Li \cite{Li2004}. See Frank and Lieb \cite{FL2010} for further extension to nonnegative measures.
%
%\medskip
%
%The following two calculus  lemmas are needed for carrying out moving sphere procedure, which are adopted from Li \cite{Li2004}. Under  stronger assumptions ($f\in C^1(\mathbb{R}^n_+)$), these lemmas were early proved by Li and Zhu \cite{LZ1995}, and by  Li and Zhang \cite{LZ2003}.
%
%%\begin{lemma}\label{Li-1} (Lemma~$5.7$~ in~\cite{Li2004}) For $n\ge 1$ and $\mu \in \mathbb{R}$,
%%if $f$ is a function defined on $\mathbb{R}^n$ and valued in $(-\infty,+\infty)$ satisfying
%%\[
%%\big(\frac\lambda{|y-x|}\big)^\mu f\big(x+\frac{\lambda^2(y-x)}{|y-x|^2}\big)\le f(y),\quad\forall\lambda>0, ~
%%|y-x|\ge\lambda,x,y\in\mathbb{R}^n,
%%\]
%%then $f(x)=$ constant.
%%\end{lemma}
%\begin{lemma}\label{Li-2} (Lemma~$5.8$~ in~\cite{Li2004}) Let $n\ge 1$,  $\mu \in \mathbb{R}$,
% and $f\in C^0(\mathbb{R}^n).$  Suppose that for every $x\in\mathbb{R}^n$, there exists $\lambda>0$ such that
%\[
%\big(\frac\lambda{|y-x|}\big)^\mu f\big(x+\frac{\lambda^2(y-x)}{|y-x|^2}\big)= f(y), \ \ \quad ~ \forall
%y\in\mathbb{R}^n\setminus\{x\}.
%\] Then there are $a\ge0,d>0$ and $\bar{x}\in\mathbb{R}^n$, such that
%\[
%f(x)=\pm a\big(\frac1 {d+|x-\bar{x}|^2}\big)^\frac{\mu}2.
%\]
%\end{lemma}
%
%\begin{lemma}\label{LZ-2}(Lemma~$3.7$~ in~\cite{DZ2015a}) For $n\ge 1$ and  $\mu \in \mathbb{R}$,
%if $f$ is a function defined on $\mathbb{R}^n_+$ and satisfies
%\[
%\big(\frac\lambda{|y-x|}\big)^\mu f\big(x+\frac{\lambda^2(y-x)}{|y-x|^2}\big)\ge f(y),\quad\forall\lambda>0, ~
%|y-x|\ge\lambda,y\in\mathbb{R}^n_+,x\in\partial\mathbb{R}^n_+,
%\]
%then
%\[f(z)=f(z',t)=f(0,t),  ~~~~~~~~~~~~~~~~~~~~~\forall\,z=(z',t)\in\mathbb{R}^n_+.
%\]
%\end{lemma}
% The proof of this lemma is similar to that of Lemma 7.1 in \cite{Li2004}. We skip the details here.
%
%\medskip
%
%We now continue the proof of Theorem \ref{nonexistence-Half space}. There are two cases.
%
%
%
%
%\noindent {\bf Case 1.}  $\bar{\lambda}(x)=\infty $ for all $x \in\partial\mathbb{R}^{n}_+$.
%
% Then for any given $x\in
%\partial\mathbb{R}^{n}_+,$
%\[u_{x,\lambda}(\xi)\ge u(\xi),~~~~~~~~~~~~~~~~~~~~~~\forall\, \xi\in\Sigma_{x,
%{\lambda}}^{n}.
%\]
%
%From Lemma \ref{LZ-2},   we conclude that $u$ only depends on $t$. Thus, for $t>0$,
%\begin{eqnarray*}
%u(\xi',t)&=&u(0,t)=c \int_0^\infty\int_{\mathbb{R}^{n-1}}\frac{u^{q_\alpha-1}(0,s)}{(|y|^2+s^2)^{\frac{n-\alpha}{2}}} dyds\\
%&=&c\int_0^\infty u^{q_\alpha-1}(0,s)s^{\alpha-1}ds \int_{0}^\infty\frac{\rho^{n-1}}{(\rho^2+1)^{\frac{n-\alpha}{2}}}d\rho=+\infty.
%\end{eqnarray*}
%Contradiction!
%%\hfill$\Box$
%
%
%\noindent {\bf Case 2.} There exists some $x_0\in \partial\mathbb{R}^{n}_+$ such that $\bar{\lambda}(x_0)<\infty.$
%
% We first show: $\bar{\lambda}(x)<\infty$ for all $x\in\partial\mathbb{R}^{n}_+$.
%
%For any $x \in\partial\mathbb{R}^{n}_+,$ from the definition of $\bar{\lambda}(x)$, we know: $\forall \, \lambda\in (0, \bar{\lambda}(x)),$
% \[u_{x,\lambda}(\xi)\ge u(\xi),\quad\quad \forall \xi\in\Sigma_{x, {\lambda}}^{n}.
% \]
%It implies
%\begin{equation}\label{K-5}
%0<a:=\liminf_{|\xi|\to\infty}\big(|\xi|^{n-\alpha}u(\xi)\big)
%\le\liminf_{|\xi|\to\infty}\big(|\xi|^{n-\alpha}u_{x,\lambda}(\xi)\big)=\lambda^{n-\alpha}u(x),~~~~~~~~~~~\forall\,
%\lambda\in(0, \bar{\lambda}(x)).
%\end{equation}
%On the other hand, since $\bar{\lambda}(x_0)<\infty$, using Lemma \ref{K-lm-2} we have
%\begin{equation}\label{K-6}
%a=\liminf_{|\xi|\to\infty}\big(|\xi|^{n-\alpha}u(\xi)\big)
%=\liminf_{|\xi|\to\infty}\big(|\xi|^{n-\alpha}u_{x_0,\bar{\lambda}}(\xi)\big)=\bar{\lambda}^{n-\alpha}u(x_0)<\infty.
%\end{equation}
% Combining \eqref{K-5} with \eqref{K-6} we obtain $\bar{\lambda}(x)<\infty$ for all $x\in\partial\mathbb{R}^{n}_+$.
%Applying Lemma \ref{K-lm-2} again, for all $x\in\partial\mathbb{R}^{n}_+$, we know
%\[u_{x,\bar{\lambda}}(\xi)=u(\xi),~~~~~~~~~~~ ~~~~~~~~~~~\forall\,\xi\in \mathbb{R}^{n}_+,
% \]
%which implies that $u \in L^{q_\alpha}(\Bbb{R}^n_+)$. It then follows from Corollary \ref{Pro-Half space} that $u=0$.
%\end{proof}
%{(\bf Until here---the above theorem may be deleted later---Guo 8/24/2017)}
%


\section{Existence result for critical case}



In this section,  we study the existence of positive solutions to the integral equation with critical exponent.

The non-existence of positive solution to \eqref{HB} with critical exponent for $\lambda \ge 0$ on a start-shaped domain follows from  Pohozaev identity \eqref{SB-5}. Next, we shall establish  the existence as well as the regularity results for weak  solutions to \eqref{HB}  with critical exponent for $\lambda<0$. To this end, we consider
\begin{equation*}
Q_\lambda(\Omega):=\inf_{f\in L_+^{q_\alpha}(\Omega)}\frac{\int_{\Omega} \int_{\Omega}f(x)  (|x-y|^{-(n-\alpha)}+\lambda|x-y|^{-(n-\alpha-1)})f(y)dy dx}{\|f\|^2_{L^{q_\alpha}(\Omega)}}.
\end{equation*}
 Notice that the corresponding Euler-Lagrange equation for extremal functions, up to a constant multiplier,  is  integral equation \eqref{HB} with $q=q_\alpha$.


First, we show
\begin{lemma}\label{Lm-inequ}
$Q_\lambda(\Omega)<N_\alpha$ for all $\lambda<0$. Further, $0<Q_\lambda(\Omega)<N_\alpha$ for any $\lambda\in (-\frac{1}{d(\Omega)}, 0)$.
\end{lemma}
\begin{proof} Let $x_*\in \Omega$.
For small positive $\epsilon$ and a fixed $R>0$ so that $B_R(x_*) \subset \Omega$, we define
 \begin{eqnarray*}
\tilde{f}_\epsilon(x)= \begin{cases}f_\epsilon(x) &\quad x\in B_{R}(x_*)\subset\Omega,\\
 0&\quad x\in\mathbb{R}^n\backslash B_{R}(x_*),
  \end{cases}
  \end{eqnarray*}
where $f_\epsilon$ is given by \eqref{fe}.
Obviously, $\tilde{f}_\epsilon\in L^{q_\alpha}(\mathbb{R}^n).$ Thus,  similar to the proof of Proposition \ref{prop2-1},  we have
\begin{eqnarray*}\label{BE-1}
& &\int_{\Omega} \int_{\Omega} \big(\frac{1}{|x-y|^{n-\alpha}}+\frac{\lambda}{|x-y|^{n-\alpha-1}}\big)\tilde{f}_\epsilon(x)\tilde{f}_\epsilon(y) dx dy\\
&&=\int_{\mathbb{R}^n} \int_{\mathbb{R}^n} \frac{1}{|x-y|^{n-\alpha}}{f_\epsilon}(x){f_\epsilon}(y) dx dy\\
& &-2\int_{\mathbb{R}^n} \int_{\mathbb{R}^n\backslash B_{R}(x_*) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy
+\int_{\mathbb{R}^n\backslash B_{R}(x_*) } \int_{\mathbb{R}^n\backslash B_{R}(x_*) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy\\
& &+\lambda \int_{B_{R}(x_*)} \int_{B_{R}(x_*)}\frac{f_\epsilon(x)f_\epsilon(y)}{|x-y|^{n-\alpha-1}} dx dy\\
&\le&N_\alpha\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}-C_1(\frac\epsilon R)^n+\lambda J_1,
%&<&N_\alpha\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}
\end{eqnarray*}
where
\begin{eqnarray*}
%I_1&:=&2\int_{\mathbb{R}^n} \int_{\mathbb{R}^n\backslash B_{R}(x_*) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy\\
%I_2&:=&\int_{\mathbb{R}^n\backslash B_{R}(x_*) } \int_{\mathbb{R}^n\backslash B_{R}(x_*) } \frac{ f_\epsilon(x) f_\epsilon(y)}{|x-y|^{n-\alpha}} dx dy,\\
J_1&:=&\int_{B_{R}(x_*)} \int_{B_{R}(x_*)}\frac{f_\epsilon(x)f_\epsilon(y)}{|x-y|^{n-\alpha-1}} dx dy\\
&=&\int_{B_{R}(x_*)} \int_{B_{R}(x_*)}{|x-y|^{-(n-\alpha-1)}}
\big(\frac{\epsilon}{\epsilon^2+|x-x_*|^2}\big)^\frac{n+\alpha}2\big(\frac{\epsilon}{\epsilon^2+|y-x_*|^2}\big)^\frac{n+\alpha}2 dx dy\\
&=&\epsilon^{-(n-\alpha-1)-(n+\alpha)} \int_{B_{R}(0)} \int_{B_{R}(0)}{\big|\frac{x-y}\epsilon\big|^{-(n-\alpha-1)}}
\big(1+\big|\frac{x}{\epsilon}\big|^2\big)^{-\frac{n+\alpha}2}\big(1+\big|\frac{y}{\epsilon}\big|^2\big)^{-\frac{n+\alpha}2 }dx dy\\
&=&\epsilon\int_{B_{\frac{R}\epsilon}(0)} \int_{B_{\frac{R}\epsilon}(0)}{|\xi-\eta|^{-(n-\alpha-1)}}
\big(1+|\xi|^2\big)^{-\frac{n+\alpha}2}\big(1+|\eta|^2\big)^{-\frac{n+\alpha}2 }d\xi d\eta\\
&\ge&C_0\epsilon.
\end{eqnarray*}
So, for $\lambda<0$ and small enough $\epsilon>0$, we have
\begin{eqnarray*}
& &\int_{\Omega} \int_{\Omega} \big(\frac{1}{|x-y|^{n-\alpha}}+\frac{\lambda}{|x-y|^{n-\alpha-1}}\big)\tilde{f}_\epsilon(x)\tilde{f}_\epsilon(y) dx dy\\
&\le&N_\alpha\|f_\epsilon\|^2_{L^{q_\alpha}(\mathbb{R}^n)}-C_1(\frac{\epsilon}R)^n+\lambda C_0\epsilon.
\end{eqnarray*}
This implies that $Q_\lambda(\Omega)<N_\alpha$ for all $\lambda<0$.
%So, as in the proof of Proposition \ref{prop2-1} we have $Q_\lambda(\Omega)<N_\alpha$ for all $\lambda<0$.

On the other hand, it is easy to see that $Q_\lambda(\Omega)>0$ for any $\lambda\in (-\frac{1}{d(\Omega)}, 0)$.
\end{proof}

The existence of solutions to equation \eqref{HB} will follow from
%{(\bf Maybe not weak solution here. We actually obtain a solution in $C^0$--Guo 9/4/2017)},
 the existence of a minimizer for energy $Q_\lambda(\Omega)$.

\begin{proposition}\label{sub-existence-2}
For any  $\lambda\in (-\frac{1}{d(\Omega)}, 0)$, infimum $Q_\lambda(\Omega) $ is achieved by a positive function $f_* \in L^{q_\alpha}(\Omega).$
%{(\bf I adjust the narration of this proposition--Guo 9/4/2017)}
%If $0<Q_\lambda<N_\alpha$ for a given $\lambda<0$, then $Q_\lambda(\Omega) $ is achieved by a positive function $f_* \in L^{q_\alpha}(\Omega).$
\end{proposition}

For $q<q_\alpha,$  consider
$$Q_{\lambda, q}(\Omega)=\inf_{f\in L_+^{q}(\Omega)}\frac{ \int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha)} f(y) dx dy+\lambda \int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha-1)} f(y) dx dy}{\|f\|^2_{L^{q}(\Omega)}}.
$$

 By Lemma \ref{Bound-attained}, the infimum is attained by a positive function $f_q$, which satisfies
the subcritical equation
\begin{equation}\label{sub-equ2}
Q_{\lambda, q}(\Omega) f^{q-1}(x)=\int_\Omega\frac{f(y)}{|x-y|^{n-\alpha}}dy+ \lambda \int_\Omega\frac {f(y)}{|x-y|^{n-\alpha-1}}dy,\quad x\in\overline{\Omega},
\end{equation}
and $\|f_q\|_{L^{q}(\Omega)}=1.$ Further, we can show easily that $f_q \in C(\overline \Omega)$  and  $Q_{\lambda, q} \to Q_\lambda$ for $q \to (q_\alpha)^-.$

% {(\bf In fact, on one hand, we can take a nondecreasing minimizing sequence $f_i\in L^{q_\alpha}$ for $Q_{\lambda, q}$. We also know that for any $f\in L^{q_\alpha}(\Omega)$, $\|f\|^2_{L^{q}(\Omega)}\to \|f\|^2_{L^{q_\alpha}(\Omega)}$ as $q \to (q_\alpha)^-$, since we can take a nondecreasing sequence $f^k\in L^{q_\alpha}$ such that $f^k\to f$ pointwise and in $L^{q_\alpha}$. For the same reason(need to add something), Thus $\liminf\limits_{q \to (q_\alpha)^-}Q_{\lambda, q}\ge Q_\lambda$.
%
%On the other hand,
%\begin{eqnarray*}
%&&Q_{\lambda, q}(\Omega)\\
%&\le & \inf_{f\in L^{q_\alpha}(\Omega)}\frac{ \int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha)} f(y) dx dy+\lambda \int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha-1)} f(y) dx dy}{\|f\|^2_{L^{q}(\Omega)}}\\
%&\to& \inf_{f\in L^{q_\alpha}(\Omega)}\frac{ \int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha)} f(y) dx dy+\lambda \int_{\Omega} \int_{\Omega} f(x)|x-y|^{-(n-\alpha-1)} f(y) dx dy}{\|f\|^2_{L^{q_\alpha}(\Omega)}}\\
%&=&Q_\lambda.
%\end{eqnarray*}
%-----I write just for clear, which can be deleted later---Guo 7/27/2017
%)}
\begin{lemma}\label{lem3-3}For  $\lambda\in (-\frac{1}{d(\Omega)}, 0)$  and $q\in (0, q_\alpha)$,  let $f_q>0$ be a minimal energy solution to \eqref{sub-equ2} with $\|f_q\|_{L^{q}(\Omega)}=1.$   If  $0<Q_{\lambda, q}\le N_\alpha-\epsilon$ for some $\epsilon>0$, then there exists $C>0$ such that $\frac{1}{C}\le f_q(x)\le C$ uniformly for all  $x\in\overline{\Omega}$ and  $q\in (0, q_\alpha)$.
\end{lemma}

%{\bf 11/13/2017 I am here to take a break}
\begin{proof}
%Notice that $0<q<1$.
 It is  easy to see that  $\max\limits_{\overline \Omega} f_q(x):=f_q(x_q)\le C<\infty$ uniformly for all  $x\in\overline{\Omega}$ and  $q\in (0, q_1)$ provided $0<q_1<q_\alpha$.

We first prove by contradiction that  $\max\limits_{\overline \Omega} f_q(x)=f_q(x_q)\le C<\infty$ uniformly for all  $x\in\overline{\Omega}$ and  $q\in (0, q_\alpha)$. Suppose not. Then $f_q(x_q) \to +\infty$ for $q \to (q_\alpha)^-$. Let
$$\mu_q= f_q^{-\frac {2-q}\alpha}(x_q), \, \, \mbox{ and} \, \, \Omega_\mu=\frac{\Omega-x_q}{\mu_q}:=\{z \ | \ z=\frac {x-x_q}{\mu_q} \ \mbox{for} ~\, x \in \Omega\}.
$$
Define
\begin{equation}\label{blowup-1}
g_q(z)=\mu_q^{\frac{\alpha}{2-q}} f_q(\mu_q z+x_q), \,  \ \, \, \mbox{for} \, \, z \in \overline{\Omega}_\mu.
\end{equation}
Then $g_q$ satisfies
\begin{equation}\label{sub-equ3}
Q_{\lambda, q}(\Omega) g_q^{q-1}(z)=\int_{\Omega_\mu}\frac{g_q(y)}{|z-y|^{n-\alpha}}dy+ \lambda \mu_q \int_{\Omega_\mu}\frac {g_q(y)}{|z-y|^{n-\alpha-1}}dy,\quad z\in\overline{\Omega}_\mu,
\end{equation}
and $g_q(0)=1$, $g_q(z) \in (0, 1]$.

For convenience, denote $h_q(z):=g_q^{q-1}(z)$. Then \eqref{sub-equ3} is equivalent to
\begin{equation}\label{sub-equ4}
Q_{\lambda, q}(\Omega) h_q(z)=\int_{\Omega_\mu}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy+ \lambda \mu_q \int_{\Omega_\mu}\frac {h_q^{p-1}(y)}{|z-y|^{n-\alpha-1}}dy,\quad z\in\overline{\Omega}_\mu,
\end{equation}
where $\frac{1}{p}+\frac{1}{q}=1, \ h_q(0)=1, \ h_q(z) \ge 1$.


Claim:  There exist $C_1, C_2>0$ such that, for all $z$ in a domain $\widehat \Omega$ covered by $\Omega_\mu$, when $q \to (q_\alpha)^-$
\begin{equation}\label{sub-equ7}
0<C_1(1+|z|^{\alpha-n})\le h_q(z) \le C_2(1+|z|^{\alpha-n}), \ \mbox{uniformly.}
\end{equation}

We relegate the proof of this claim to the end.

Once the claim is proved, we can prove that $h_q(z)$ is equicontinuous on any bounded domain $\widehat{\Omega}\subset \Omega_\mu$ when $q \to (q_\alpha)^-$. We write
\begin{eqnarray*}
&&Q_{\lambda, q}(\Omega)h_q(z)\\
&=&\int_{\Omega_\mu\setminus B(0,R)}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy+\int_{\Omega_\mu\cap B(0,R)}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy\\
&+&\lambda \mu_q \int_{\Omega_\mu\setminus B(0,R)}\frac {h_q^{p-1}(y)}{|z-y|^{n-\alpha-1}}dy+\lambda \mu_q \int_{\Omega_\mu\cap B(0,R)}\frac {h_q^{p-1}(y)}{|z-y|^{n-\alpha-1}}dy.
\end{eqnarray*}
Notice that
\begin{equation*}
\int_{\Omega_\mu\setminus B(0,R)}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}(1+\lambda \mu_q|z-y|)dy \ge (1-|\lambda|d(\Omega))\int_{\Omega_\mu\setminus B(0,R)}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy \ge0.
\end{equation*}
 Then for $\epsilon>0$ small enough,  we have, by \eqref{sub-equ7}, that
\begin{eqnarray}\nonumber
0&\le&\int_{\Omega_\mu\setminus B(0,R)}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}(1+\lambda \mu_q|z-y|)dy\\\nonumber
&\le &C\int_{\Omega_\mu\setminus B(0,R)}\frac{h_q^{p-1}(y)}{|y|^{n-\alpha}}dy\\\nonumber%+C\lambda \mu_q \int_{\Omega_\mu\setminus B(0,R)}\frac {h_q^{p-1}(y)}{|y|^{n-\alpha-1}}dy\
&\le&C\int_R^\infty r^{(\alpha-n)(p-1)+\alpha-1}dr\\\label{R-small-term}%+C\lambda \mu_q \int_R^\infty r^{(\alpha-n)(p-1)+\alpha}dr
&=&C R^{(\alpha-n)(p-1)+\alpha}<\epsilon
\end{eqnarray}
for any $z\in\widehat{\Omega}$, by taking $R>0$ large enough and  $q$ close to $q_\alpha$. Similarly, we have
\begin{equation}\label{R-small-term-1}
|\lambda \mu_q \int_{\Omega_\mu\cap B(0,R)}\frac {h_q^{p-1}(y)}{|z-y|^{n-\alpha-1}}dy|<\epsilon
\end{equation}
by taking $R>0$ large enough and $q$ close to $q_\alpha$. On the other hand, it is easy to see that $\int_{\Omega_\mu\cap B(0,R)}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy\in C^1(\widehat{\Omega})$. Hence
for $z_1,z_2\in \widehat{\Omega}$,
\begin{eqnarray}\nonumber
&&|\int_{\Omega_\mu\cap B(0,R)}\frac{h_q^{p-1}(y)}{|z_1-y|^{n-\alpha}}dy-\int_{\Omega_\mu\cap B(0,R)}\frac{h_q^{p-1}(y)}{|z_2-y|^{n-\alpha}}dy|\\\nonumber
&\le& \int_{\Omega_\mu\cap B(0,R)}h_q^{p-1}(y)\frac{1}{|\xi-y|^{n-\alpha+1}}dy|z_1-z_2| \\\label{R-main-term}
&\le&  \int_{B(0,R)}\frac{1}{|\xi-y|^{n-\alpha+1}}dy|z_1-z_2|\le C R^{\alpha-1}|z_1-z_2|,
\end{eqnarray}
where $\xi=tz_1+(1-t)z_2$ for some $t\in(0,1)$. By \eqref{R-small-term}, \eqref{R-small-term-1} and \eqref{R-main-term} we conclude that $h_q(z)$ is equicontinuous on bounded domain $\widehat{\Omega}\in\mathbb{R}^n$ when $q \to (q_\alpha)^-$.

 As $q\to (q_\alpha)^-$, there are two cases:

\noindent{Case 1.} \  $\Omega_\mu \to \Bbb{R}_T^n:= \{(z_1, z_2, \cdots, z_n) \ | \ z_n >T\ge 0\},$ and $h_q(z) \to h(z)\in C({\Bbb{R}_T^n})$ uniformly  in any compact set in $\Bbb{R}_T^n$, where $h(z)$ satisfies
\begin{equation}\label{blowup-3}
Q_\lambda h(z)= \int_{\Bbb{R}_T^n} \frac {h^{p_\alpha-1}(y)}{|z-y|^{n-\alpha}} dy, \ \ \ \ h(0)=1.
\end{equation}
Also, direct computation yields
$$1=\int_\Omega f_q^q(y) dy= \mu_q^{(\frac{n+\alpha}{2-q})\cdot (\frac {2n}{n+\alpha}-q)}\cdot \int_{\Omega_\mu} g_q^qdz \le \int_{\Omega_\mu} g_q^qdz=\int_{\Omega_\mu} h_q^pdz.
$$
On the other hand, by \eqref{sub-equ7} we have $\int_{\Omega_\mu} h_q^pdz\le C$ uniformly.
 Again by \eqref{sub-equ7}, $\int_{\Bbb{R}_T^n} h^{p_\alpha} dz=\lim\limits_{q\to (q_\alpha)^-}\int_{\Omega_\mu} h_q^pdz \ge 1.$  Denote $g(x)=h^{p_\alpha-1}(x)$. Then $\int_{\Bbb{R}_T^n} h^{p_\alpha} dz=\int_{\Bbb{R}_T^n} g^{q_\alpha} dz$. By \eqref{blowup-3}, we have
\begin{eqnarray*}
N_\alpha-\epsilon\ge Q_\lambda&=& \frac{\int_{\Bbb{R}_T^n} \int_{\Bbb{R}^n} \frac {g(x) g(y)}{|x-y|^{n-\alpha}} dx dy}{\|g\|^{q_\alpha}_{L^{q_\alpha}(\Bbb{R}_T^n)}}
\\
&\ge& \frac{\int_{\Bbb{R}_T^n} \int_{\Bbb{R}_T^n} \frac {g(x) g(y)}{|x-y|^{n-\alpha}} dx dy}{\|g\|^{2}_{L^{q_\alpha}(\Bbb{R}_T^n)}}\\
&=&\frac{\int_{\Bbb{R}^n} \int_{\Bbb{R}^n} \frac {\tilde{g}(x) \tilde{g}(y)}{|x-y|^{n-\alpha}} dx dy}{\|\tilde{g}\|^{2}_{L^{q_\alpha}(\Bbb{R}^n)}}\ge N_\alpha.
\end{eqnarray*}
Contradiction!

\noindent {Case 2.} \ $\Omega_\mu \to \Bbb{R}^n,$ and $h_q(z) \to h(z)\in C({\Bbb{R}^n})$ uniformly  in any compact set in $\Bbb{R}^n$, where $h(z)$ satisfies
\begin{equation}\label{blowup-4}
Q_\lambda h(z)= \int_{\Bbb{R}^n} \frac {h^{p_\alpha-1}(y)}{|z-y|^{n-\alpha}} dy, \ \ \ \ h(0)=1.
\end{equation}
Similarly, $C\ge \int_{\Bbb{R}^n} h^{p_\alpha} dz\ge 1.$  Denote $g(x)=h^{p_\alpha-1}(x)$. Then $\int_{\Bbb{R}^n} h^{p_\alpha} dz=\int_{\Bbb{R}^n} g^{q_\alpha} dz$. By \eqref{blowup-4} we have
$$
N_\alpha-\epsilon\ge Q_\lambda= \frac{\int_{\Bbb{R}^n} \int_{\Bbb{R}^n} \frac {g(x) g(y)}{|x-y|^{n-\alpha}} dx dy}{\|g\|^{q_\alpha}_{L^{q_\alpha}(\Bbb{R}^n)}}\ge \frac{\int_{\Bbb{R}^n} \int_{\Bbb{R}^n} \frac {g(x) g(y)}{|x-y|^{n-\alpha}} dx dy}{\|g\|^{2}_{L^{q_\alpha}(\Bbb{R}^n)}}\ge N_\alpha,
$$
which again implies a contradiction.

Thus we conclude that there exists $C>0$ such that $f_q(y)\le C$ uniformly in $y\in\overline{\Omega}$ and  $q\in (0, q_\alpha)$.

%\medskip

On the other hand, if $\min\limits_{\overline \Omega} f_q(x):=f_q(\tilde{x}_q) \to 0$ as $q \to (q_\alpha)^-$, by using $f_q(y)\le C$ uniformly in $y\in\overline{\Omega}$ and  $q\in (0, q_\alpha)$ we have
$$\infty\leftarrow f_q^{q-1}(\tilde{x}_q)=\int_\Omega\frac{f_q(y)}{|\tilde{x}_q-y|^{n-\alpha}}dy+ \lambda \int_\Omega\frac {f_q(y)}{|\tilde{x}_q-y|^{n-\alpha-1}}dy\le C<\infty$$
as $q \to (q_\alpha)^-$, which gives a contradiction.

\smallskip

Now we are left to prove claim \eqref{sub-equ7}.

We first notice that
\begin{equation}\label{sub-equ5}
Q_{\lambda, q}(\Omega)=Q_{\lambda, q}(\Omega)h_q(0)=\int_{\Omega_\mu}\frac{h_q^{p-1}(y)}{|y|^{n-\alpha}}(1+\lambda \mu_q|y|)dy.
\end{equation}
Thus,
%By \eqref{sub-equ5} and noting the definition of $\Omega_\mu$, we have
\begin{eqnarray}\nonumber
\int_{\Omega_\mu}h_q^{p-1}(y)|y|^{\alpha-n}dy&=& \frac{1}{(1-|\lambda| d(\Omega))}\int_{\Omega_\mu}h_q^{p-1}(y)|y|^{\alpha-n}(1-|\lambda| d(\Omega))dy\\\nonumber
&\le & \frac{1}{(1-|\lambda| d(\Omega))}\int_{\Omega_\mu}h_q^{p-1}(y)|y|^{\alpha-n}(1+\lambda \mu_q|y|)dy\\\label{sub-equ5-1}
&\le & C<\infty,
\end{eqnarray}
uniformly as $q \to (q_\alpha)^-$. Since $h_q \ge 1$ and $p<0$, we have
\begin{eqnarray}\label{sub-equ5-2}
\int_{\Omega_\mu}h_q^{p-1}(y)dy\le C<\infty
%&\int_{\Omega_\mu\backslash B(0,1)}h_q^{p-1}(y)dy+ \int_{\Omega_\mu\cap B(0,1)}h_q^{p-1}(y)dy \\\nonumber
%&\le & \int_{\Omega_\mu\backslash B(0,1)}h_q^{p-1}(y)|y|^{\alpha-n}dy+ |B(0,1)|\\\nonumber
%&\le & \int_{\Omega_\mu}h_q^{p-1}(y)|y|^{\alpha-n}dy+ |B(0,1)|\\\label{sub-equ5-2}
%&\le & C<\infty
\end{eqnarray}
uniformly as $q \to (q_\alpha)^-$.

On the other hand, we also have
\begin{equation}\label{sub-equ6}
\int_{\Omega_\mu}h_q^{p-1}(y)dy\ge c_0>0, \ \ \mbox{as} \ \ q \to (q_\alpha)^-.
\end{equation}
Otherwise, if  there exists a sequence $q_n \to (q_\alpha)^-$ such that $\int_{\Omega_\mu}h_{q_n}^{p_n-1}(y)dy\to 0$ with $\frac{1}{p_n}+\frac{1}{q_n}=1$,
then for given $R_0>0$ and $\epsilon>0$ small we can take $R>>R_0$ large enough, such  that for $z\in \Omega_\mu\cap B(0, R_0 )$, as $q_n$ close to $(q_\alpha)^-$,
\begin{eqnarray*}
1&\le & h_{q_n}(z)\\
&=&\frac{1}{Q_{\lambda, q_n}(\Omega)}\big( \int_{\Omega_\mu\setminus B(0,R)}\frac{h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha}}dy+\int_{\Omega_\mu\cap B(0,R)}\frac{h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha}}dy\\
&&+\lambda \mu_{q_n} \int_{\Omega_\mu\setminus B(0,R)}\frac {h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha-1}}dy+\lambda \mu_{q_n} \int_{\Omega_\mu\cap B(0,R)}\frac {h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha-1}}dy\big)\\
&\le&\frac{1}{Q_{\lambda, q_n}(\Omega)}\big( \int_{\Omega_\mu\setminus B(0,R)}\frac{h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha}}dy+\int_{\Omega_\mu\cap B(0,R)}\frac{h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha}}dy\\
&&+\lambda \mu_{q_n} \int_{\Omega_\mu\setminus B(0,R)}\frac {h_{q_n}^{p_n-1}(y)}{|z-y|^{n-\alpha-1}}dy\big) \\
&\le& \frac{1}{Q_{\lambda, {q_n}}(\Omega)}\big( (1+\frac{R_0}{R})^{\alpha-n}\int_{\Omega_\mu\setminus B(0,R)}\frac{h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha}}dy+(R+R_0)^{\alpha-n}\int_{\Omega_\mu\cap B(0,R)}h_{q_n}^{p_n-1}(y)dy\\
&&+\lambda \mu_{q_n} (1-\frac{R_0}{R})^{\alpha+1-n}\int_{\Omega_\mu\setminus B(0,R)}\frac {h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha-1}}dy\big)\\
&=& \frac{1}{Q_{\lambda, {q_n}}(\Omega)}\big( (1+\frac{R_0}{R})^{\alpha-n}\int_{\Omega_\mu\setminus B(0,R)}\frac{h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha}}dy+(R+R_0)^{\alpha-n}\int_{\Omega_\mu\cap B(0,R)}h_{q_n}^{p_n-1}(y)dy\\
&&+\lambda \mu_{q_n} (1-\frac{R_0}{R})^{\alpha+1-n}\int_{\Omega_\mu}\frac {h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha-1}}dy-\lambda \mu_{q_n} (1-\frac{R_0}{R})^{\alpha+1-n}\int_{\Omega_\mu\cap B(0,R)}\frac {h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha-1}}dy\big)\\
&\le & \frac{1}{Q_{\lambda, {q_n}}(\Omega)}\big( (1+\frac{R_0}{R})^{\alpha-n}\int_{\Omega_\mu}\frac{h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha}}dy+\lambda \mu_{q_n} (1-\frac{R_0}{R})^{\alpha+1-n}\int_{\Omega_\mu}\frac {h_{q_n}^{p_n-1}(y)}{|y|^{n-\alpha-1}}dy\big)\\
&&+(R+R_0)^{\alpha-n}\frac{1}{Q_{\lambda, {q_n}}(\Omega)}\int_{\Omega_\mu}h_{q_n}^{p_n-1}(y)dy-\lambda \mu_{q_n} (R-R_0)^{\alpha+1-n}\frac{1}{Q_{\lambda, {q_n}}(\Omega)}\int_{\Omega_\mu}h_{q_n}^{p_n-1}(y)dy\\
&\le & 1+\epsilon.% \ \mbox{for}\ z\in \Omega_\mu\cap B(0, R_0) \ \mbox{uniformly}
\end{eqnarray*}
%by taking $R>0$ large and then, where \eqref{sub-equ5} is used in the last inequality.
That is, $h_{q_n}(z)\to 1, \ z\in \Omega_\mu\cap B(0,R_0)$ uniformly as $q_n \to (q_\alpha)^-$. Then for $R_0>0$ large, since
$\int_{\Omega_\mu\cap B(0, R_0) }h_{q_n}^{p_n-1}(y)dy\le \int_{\Omega_\mu}h_{q_n}^{p_n-1}(y)dy$  and $\Omega_\mu$ goes to either $\Bbb{R}_T^n:= \{(z_1, z_2, \cdots, z_n) \ | \ z_n >T\ge 0\}$ or $\Bbb{R}^n,$  we obtain a contradiction to \eqref{sub-equ5-2}.

By \eqref{sub-equ4}, we have
\begin{eqnarray}\label{sub-equ7-1}
&&\lim_{|z|\to\infty}\frac{1}{|z|^{\alpha-n}}\int_{\Omega_\mu}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy\nonumber\\
&\ge&\lim_{|z|\to\infty} Q_{\lambda, q}(\Omega) \frac{h_q(z)}{|z|^{\alpha-n}}
=\lim_{|z|\to\infty}\frac{\int_{\Omega_\mu}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}(1+ \lambda \mu_q |z-y|)dy}{|z|^{\alpha-n}}\nonumber\\
&\ge&(1-|\lambda|d(\Omega))\lim_{|z|\to\infty}\frac{\int_{\Omega_\mu}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy}{|z|^{\alpha-n}}.
\end{eqnarray}
Since $\frac{1}{|z|^{\alpha-n}}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}\le 2^{\alpha-n}h_q^{p-1}(y)(1+|y|^{\alpha-n})$ as $|z|\to \infty$,  and $\int_{\Omega_\mu}h_q^{p-1}(y)(1+|y|^{\alpha-n})dy\le C$
by \eqref{sub-equ5-1} and \eqref{sub-equ5-2}, we then have
\begin{equation}\label{sub-equ7-2}
\lim_{|z|\to\infty}\frac{1}{|z|^{\alpha-n}}\int_{\Omega_\mu}\frac{h_q^{p-1}(y)}{|z-y|^{n-\alpha}}dy =\int_{\Omega_\mu}h_q^{p-1}(y)dy.
\end{equation}
Hence by \eqref{sub-equ7-1}, \eqref{sub-equ7-2}, \eqref{sub-equ5-2} and \eqref{sub-equ6}  we obtain the claim \eqref{sub-equ7}.  Hereby we complete the proof of Lemma \ref{lem3-3}.
\end{proof}

%{\bf we can prove above lemma via blowup argument based on the new Liouville theorem on the upper half space in this paper, and the classification of extremal functions on whole %space.} or{\bf the regularity arguement as we derive the $C^1(\overline \Omega)$ regularity.}

\noindent{\bf Proof of Proposition \ref{sub-existence-2}}. \   Let $f_q>0$ being  solutions to \eqref{sub-equ2} for $q\in (0, q_\alpha)$, which are also the minimal energy functions to energy $Q_{\lambda, q}$. Then by Lemma \ref{lem3-3}, we know that  $\{f_q\}$ are uniformly bounded above and bounded below by a positive constant. Thus they are  equicontinuous due to equation  \eqref{sub-equ2}.  It follows that $f_q \to f_*$ as $q \to (q_\alpha)^-$ in $C(\overline \Omega),$ and $f_*$ is the energy minimizer for $Q_\lambda$.
 \hfill$\Box$

\medskip

\noindent{\bf Completion of the Proof of Theorem \ref{main}}.  Lemma \ref{Lm-inequ} and Proposition \ref{sub-existence-2} implies the existence of a positive solution  $f \in L^{p_\alpha}(\Omega)\cap C(\overline\Omega)$ to the equation \eqref{HB}  for $q=\frac{2n}{n+\alpha}$,  $\lambda\in (-\frac{1}{d(\Omega)}, 0)$. It is also easy to see that $f \in C^1(\overline\Omega)$. \hfill$\Box$


\medskip





 \vskip 1cm
\noindent {\bf Acknowledgements}\\
\noindent% The author would like to thank the referee for his/her
%careful reading of the manuscript and many good suggestions.
We  dedicate this paper to Professor Ha\"im Brezis to celebrate his seventy five birthday. We thank him for his great influence on us in the study of elliptic equations through his lectures and numerous papers, in particular, the paper with Louis Nirenberg \cite{BN1983}.
The project is partially  supported by  the
National Natural Science Foundation of China (Grant No. 11571268) and the Fundamental Research Funds for the Central Universities (Grant No. GK201802015) and Simmons Collaboration (Grant No. 280487).


%% bibliography--------------------------------------------------------------------
%\begin{center}
\small
\begin{thebibliography}{a}\small

%\bibitem{A1976} T. Aubin, Probl\`{e}mes isop\'{e}rim\'{e}triques et espaces de Sobolev. J. Differential Geom., 11 (1976),  573-598.
%\bibitem{BC1988} A. Bahri and J.-M. Coron, On a nonlinear elliptic equation involving the critical Sobolev exponent: the
%effect of the togology of the domain, Comm. Pure Appl. Math., 41 (1988), 253-294.
%\bibitem{B1993} W. Beckner, Sharp Sobolev inequalities on the sphere and the Moser-Trudinger inequality, Ann. of Math. 138 (1993), 213-242.
%%\bibitem{BLL1974} H. J. Brascamp; E. H. Lieb; J. M. Luttinger, A general rearrangement inequality for multiple integrals. J. Functional Analysis  17  (1974), 227-237.
%\bibitem{BP1989} H. Brezis, L. A. Peletier, Asymptotics for elliptic equations involving critical growth, Partial differential equations and the calculus of variations, Birkh?user Boston, 1989, 1%49-192.

\bibitem{Beck2015} W. Beckner, Functionals for Multilinear Fractional Embedding, Acta Math. Sinica, English Series, 31 (2015), 1-28.



\bibitem{BN1983} H. Brezis, L. Nirenberg, Positive solutions of nonlinear elliptic equations involving critical Sobolev exponents
Commun. Pure Appl. Math., 36 (1983), 437-477.
%\bibitem{BR1992} H. Brezis, Uniform estimates for solutions of $-\Delta u=V(x)u^p$, in Partial Differential Equations and related subjects, Trento, 1990, (M. Miranda, ed.), Longman, 1992, p. 38-52.
%%\bibitem{BL1983} H. Brezis, E. Lieb, A relation between pointwise convergence of functions and convergence of functionals, Proc. Amer. Math. Soc. 88 (1983), 486-490.
%\bibitem{BK1979} H. Brezis, T. Kato, Remarks on the Schr\"{o}dinger operator with singular complex potentials, J. Math. Pures Appl. 58 (1979), 137-151.
%\bibitem{CT2010} X. Cabr\'{e}, J. Tan, Positive solutions of nonlinear problems involving the square root of the Laplacian, Adv. Math., 224 (2010), 2052-2093.
%\bibitem{CS90} L.A. Caffarelli, J. Spruck, Variational problems with critical Sobolev growth and positive Dirichlet data, Indiana Univ. Math. J. 39(1)(1990)1-18.
%\bibitem{CS2007} L. Caffarelli, L. Silvestre, An extension problem related to the fractional Laplacian, Comm. Partial. Diff. Equ. 32 (2007), 1245-1260.
%\bibitem{CarLo1992} E. Carlen, M. Loss, Competing symmetries, the logarithmic HLS inequality and Onofri's inequality on $\S^n$, Geom. Funct. Anal., 2 (1992), 90-104.

%\bibitem{CG2011} S.-Y. Chang, M. Gonzalez, Fractional Laplacian in conformal geometry, Adv. Math. 226 (2011), 1410-1432.
%\bibitem{CL2010} W. Chen, C. Li, Methods on Nonlinear Elliptic Equations, AIMS Ser. Differ. Equ. Dyn. Syst.  vol. 4, 2010.
%\bibitem{CL2011} W. Chen, C. Li, Super polyharmonic property of solutions for PDE systems and its applications, arxiv: 1110.2539.
%\bibitem{CLO2006} W. Chen, C. Li, B. Ou, Classification of solutions for an integral equation, Comm. Pure Appl. Math.  59 (2006), 330-343.
\bibitem{CFY2014} W. Chen, Y. Fang, R.Yang, Semilinear equations involving the fractional Laplacian on domains, arXiv:1309.7499
%\bibitem{CW91} Y. Chen,  L. Wu, Second Order Elliptic Equations and Elliptic Systems, Translations of Mathematical Monographs. AMS, 1998.
%\bibitem{DZ2015a} J. Dou, M. Zhu, Sharp Hardy-Littlewood-Sobolev inequality on the upper half space, Int. Math. Res. Not. 2015 (2015), 651-687.
\bibitem{DZ2015} J. Dou, M. Zhu, Reversed Hardy-Littlewood-Sobolev inequality, Int. Math. Res. Not., 2015 (2015), 9696-9726.
\bibitem{DZ2017} J. Dou, M. Zhu, Integral equations on bounded domains, J. Fun. Anal., (2018), https://doi.org/10.1016/j.jfa.2018.05.020.
%\bibitem{D1989}  W. Ding, Positive solutions of $\Delta u+u^\frac{n+2}{n-2}=0$ on a contactible domain, J. Part. Diff. Eqs. 2(1989), 83-88.
\bibitem{DGZ17} J. Dou, Q. Guo, M. Zhu, Subcritical approach to sharp
Hardy-Littlewood-Sobolev type inequalities on the upper half space, Adv. Math., 312 (2017) 1-45;
Corrigendum to``Subcritical approach to sharp Hardy-Littlewood-Sobolev type inequalities on the upper half space"[Adv. Math., 312 (2017) 1-45], Adv. Math., 317 (2017) 640-644.

%\bibitem{E1988} J. F. Escobar, Sharp constant in a Sobolev trace inequality, Indiana Univ. Math. J.  37,  (1988), 687-698.
%\bibitem{E1992-1} J. F. Escobar, The Yamabe problem on manifolds with boundary, J. Differential Geom. 35 (1992),  21-84.
%\bibitem{E1992} J. F. Escobar, Conformal deformation of a Riemannian metric to a scalar flat metric with constant mean curvature on the boundary, Ann. of Math. 136 (1992), 1-50.
%\bibitem{Evans99} L. C. Evans, book
%\bibitem{FL2010} R. L. Frank, E. H. Lieb, Inversion positivity and the sharp Hardy-Littlewood-Sobolev inequality, Calc. Var. Partial Differential Equations  39 (2010),  85-99.
%    \bibitem{FL2012} R. L. Frank,  E. H. Lieb, Sharp constants in several inequalities on the Heisenberg group, Ann. of Math.  to appear.
%\bibitem{GM2011} M. Gonz\'{a}lez, R. Mazzeo, Y. Sire, Singular solutions of fractional order conformal Laplacians,
%J. Geom. Anal. DOI: 10.1007/s12220-011-9217-9.
%\bibitem{GQ2011} M. Gonz\'{a}lez, J. Qing, Fractional conformal Laplacians and fractional Yamabe problems, arXiv: 1012.0579v1
%    \bibitem{Gr} L. Gross, Logarithmic Sobolev Inequalities,  Amer.  J.  Math.   97  (1976), 1061-1083.
%\bibitem{HL1999} Z. Han,  Y. Y. Li, The Yamabe problem on manifolds with boundary: existence and
%compactness results, Duke Math. J.  99 (1999), 489-542.
%
%\bibitem{HWY2008} F. Hang,  X. Wang,  X. Yan, Sharp integral inequalities for harmonic functions, Comm. Pure Appl. Math. 61 (2008), 0054-0095.

%\bibitem{HL1928} G. H. Hardy,  J. E. Littlewood, Some properties of fractional integrals (1), Math. Zeitschr. 27 (1928), 565-606.

%\bibitem{HL1930} G. H. Hardy, J. E. Littlewood, On certain inequalities connected with the calculus of variations, J. London Math. Soc. 5 (1930), 34-39.
%\bibitem{JLX2011a} T. Jin, Y. Y. Li, J.  Xiong, On a fractional Nirenberg problem, part I: blow up analysis and compactness of solutions, arXiv: 1111.1332v1
%\bibitem{JLX2011b} T. Jin, Y. Y. Li, J.  Xiong, On a fractional Nirenberg problem, part II: existence of solutions, in preparation, 2011.
%\bibitem{JX2011} T. Jin, J. Xiong, A fractional Yamabe flow and some applications, arXiv: 1110. 5664v1
%\bibitem{LP1987} J. M. Lee, T. H. Parker, The Yamabe problem, Bull. Am. Math. Soc., 17 (1987), 37-91.
%\bibitem{LL2016}Y. Lei, C. Li, Sharp criteria of Liouville type for some nonlinear systems, Discrete Contin. Dyn. Syst.-A 36 (2016), 3277-3315.
%\bibitem{Li2004} Y. Y. Li, Remark on some conformally invariant integral equations: the method of moving spheres, J. Eur. Math. Soc.  6 (2004), 153-180.
%\bibitem{LZ2003} Y. Y. Li, L. Zhang, Liouville type theorems and Harnack type inequalities for semilinear elliptic equations, J. D'Anal. Math. 90 (2003), 27-87.
%\bibitem{LZ1995} Y. Y. Li, M. Zhu, Uniqueness theorems through the method of moving spheres, Duke Math. J.  80 (1995),383-417.
%\bibitem{LZ1997} Y. Y. Li, M. Zhu, Sharp Sobolev trace inequalities on
%Riemannian manifolds with boundaries, Comm. Pure Appl. Math.  50 (1997), 449-487.
%\bibitem{Lieb1977} E. H.  Lieb, Existence and Uniqueness of the Minimizing Solution of Choquard's Non-Linear Equation, Studies in Appl. Math. 57, 93-105 (1977).
%\bibitem{LSW2009} D. Li, G. Str\"{o}hmer, L. Wang, Symmetry of integral equations on bounded domains, Proc.Am. Math. Soc., 137 (2009), 3695-3702.
\bibitem{Lieb1983} E. H. Lieb, Sharp constants in the Hardy-Littlewood-Sobolev and related inequalities, Ann. of Math., 118 (1983), 349-374.
%\bibitem{LL2001} E. H. Lieb, M. Loss, Analysis, 2nd ed. Graduate Studies in Mathematics, 14. American Mathematical Society, Providence, R.I. 2001.
%\bibitem{NZ1} Y. Ni, M. Zhu, Steady states for one dimensional curvature flows, Comm. Contemp. Math. 10 (2008), 155-179.
%\bibitem{ONeil} R. O'Neil, Convolution operators and $L(p, q)$ spaces, Duke Math. J. 30 (1963), 129-142.
%\bibitem{SY1988} R. Schoen, S.-T. Yau, Conformally flat manifolds, Kleinian groups and scalar curvature, Inventiones Mathematicae,  92 (1988), 47-71.

\bibitem{NN2017} Q.N. Ng\^{o}, V.H. Nguyen, Sharp reversed  Hardy-Littlewood-Sobolev inequality on $\mathbb{R}^n$, Israel J. Math., 220 (2017), 189-223.

%\bibitem{P1965} S. I. Pohozaev, Eigenfunctions of the equation $-\Delta u +\lambda f(u) = 0$, Soviet Math. Doklady 6 (1965), 1408-1411 (translated from the Russian Dokl. Akad. Nauk SSSR 165 (1965), 33-36.

%\bibitem{T1976} G. Talenti, Best constant in Sobolev inequality, Ann. Mat. Pure Appl.,110 (1976), 353-372.
%\bibitem{T1968} N. S. Trudinger, Remarks concerning the conformal deformation of Riemannian structures on compact manifolds. Annali Sc. Norm. supp. Pisa 22 (1968), 265-274.


%\bibitem{So1963} S. L. Sobolev, On a theorem of functional analysis, Mat. Sb. (N.S.) 4 (1938), 471-479. A. M. S. Transl. Ser. 2, 34 (1963), 39-68.
%\bibitem{Stein} E. M. Stein, Singular integrals and differentiability properties of functions, Princeton Mathematical
%eries, 30. Princeton University Press, Princeton, N.J. 1970.
%\bibitem{Struwe1990} M. Struwe,  Variational Methods,Applications to Nonlinear Partial Differential Equations and Hamiltonian Systems,Fourth Edition,Springer,2008.
%\bibitem{S2006} G. Str\"{o}hmer, About the linear stability of the spherically symmetric solution for the equations
%of a barotropic viscous fluid under the influence of self-gravitation, J. Math. Fluid Mech., 8 (2006), 36-63.
%\bibitem{S2009} G. Str\"{o}hmer About the Stability of Gas Balls, J. Math. Fluid Mech.,  11 (2009), 572-608.

%\bibitem{V2016} J. Villavert, Qualitative properties of solutions for an integral system related to the Hardy-Sobolev inequality, J. DifferentialEquations, 258(2015), 1685-1714.


%\bibitem{Y1960} H. Yamabe, On the deformation of Riemannian structures on compact manifolds, Osaka Math. J., 12 (1960), 21-37.
%\bibitem{Zhu2016} M. Zhu, Prescribing integral curvature equation, Differential and Integral Equations 29(9/10) (2016), 889-904.

\end{thebibliography}

\end{document}