Rellich-Kondrachov type theorems on the half-space with general singular weights

Extracted source text · arXiv 2508.01978

\documentclass[11pt]{amsart}

% (Modern LaTeX does not need inputenc; keep it only if your engine is very old)
% \usepackage[utf8]{inputenc}

\usepackage{amsmath,amssymb,amsthm,mathrsfs}
\usepackage{geometry}
\usepackage{esint}
\usepackage{hyperref}
\usepackage{enumitem}

\geometry{margin=1in}
\setlength{\parskip}{0.5em}
\setlength{\parindent}{0pt}

\newcommand{\proofstep}[1]{\noindent\textbf{Step #1.} }

% Your integral macros (unchanged)
\def\dashint{\Xint-}
\def\Xint#1{\mathchoice
   {\XXint\displaystyle\textstyle{#1}}%
   {\XXint\textstyle\scriptstyle{#1}}%
   {\XXint\scriptstyle\scriptscriptstyle{#1}}%
   {\XXint\scriptscriptstyle\scriptscriptstyle{#1}}%
   \!\int}
\def\XXint#1#2#3{{\setbox0=\hbox{$#1{#2#3}{\int}$}
     \vcenter{\hbox{$#2#3$}}\kern-.5\wd0}}

% Theorem Environments
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{assumption}{Assumption}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{proposition}[theorem]{Proposition}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\theoremstyle{remark}
\newtheorem*{remark}{Remark}
\newtheorem{claim}{Claim}[section]

% Title Info
\title{Rellich-Kondrachov type theorems on the half-space with general singular weights}

\author{Yunfan Zhao}
\address{Shanghai University of International Business and Economics}
\email{yfzucla21@ucla.edu}
\thanks{Both authors contributed equally to this work.}

\author{Xiaojing Chen}
\address{Shanghai University of International Business and Economics}
\email{chenxj@suibe.edu.cn}

\date{} % optional

\begin{document}

\maketitle

\begin{abstract}
We prove Rellich-Kondrachov type theorems on the half-space $\mathbb{H}^{N+1}=\{(y, x) \in \left.\mathbb{R} \times \mathbb{R}^N: y>0\right\}$ endowed with the general weighted measure $\mu_w:=y^c \phi(|z|) d z$, where $c \in \mathbb{R}$ and $\phi$ is a suitable Borel measurable function. We establish a necessary and sufficient characterization for the compactness of the immersion $H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right) \hookrightarrow L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right)$. We prove that compactness holds if and only if the measure has finite mass and satisfies a "Global Tightness" condition, which we characterize via a coercive tail inequality (Lyapunov condition) and, in the singular case $c \leq-1$, a weighted Hardy inequality. These results generalize recent work on Gaussian weights to a broader class of radial potentials defined by abstract massvanishing conditions.
\end{abstract}


\section{Introduction}
Poincaré inequalities and Rellich-Kondrachov type theorems are very powerful tools in mathematical analysis which have been extensively used for the study of partial differential equations (PDEs). They are of great interest in relation to optimal transport, the study of the topology of abstract spaces, and probabilistic problems. For instance, they can be used for proving the existence and regularity of solutions to certain PDEs as well as Harnack’s inequalities; see the seminal works by De Giorgi \cite{DeGiorgi1957}, Nash \cite{Nash1958}, and Moser \cite{Moser1961}.

Counterparts of the Poincaré inequalities in a weighted setting have gained significant attention in the last decades due to their connection with the regularity of solutions to degenerate or singular PDEs which depend on the particular structure of the weight under consideration (see e.g. \cite{DongPhan2021, FabesKenigSerapioni1982, FranchiGutierrezWheeden1994} and references therein).

Following these ideas, in this paper we extend recent results proved in \cite{NegroSpina2025} by studying Rellich-Kondrachov type theorems and weighted Poincaré inequalities on the half-space $\mathbb{H}^{N+1} = \{(y,x) \in \mathbb{R} \times \mathbb{R}^N : y > 0\}$ endowed with the general weighted measure $\mu_w := y^c \phi(|z|) \, dz$. Here $c \in \mathbb{R}$ and $\phi: [0, \infty) \to (0, \infty)$ is a Borel measurable function satisfying suitable local bounds. Our main results are Theorems 7.1 and 7.2 where we establish a necessary and sufficient characterization for the compactness of the immersion $H_{\mu_w}^1(\mathbb{H}^{N+1}) \hookrightarrow L_{\mu_w}^2(\mathbb{H}^{N+1})$. We prove that compactness holds if and only if the measure has finite mass and the unit ball satisfies a condition of ``Global Tightness,'' which we further characterize via a coercive tail inequality (Lyapunov condition) and, in the singular case $c \le -1$, a weighted Hardy inequality.

We remark that in \cite{NegroSpina2025} the authors proved similar results for the specific case of the Gaussian weight $\phi(|z|) = e^{-a|z|^2}$ with $c > -1$. The strategy therein employed is based on explicit computations involving the specific decay of the Gaussian density. On the other hand, our approach relies on an abstract functional analytic framework—specifically the Fréchet-Kolmogorov Compactness Criterion adapted to weighted spaces—and has the advantage of being applicable to a much broader class of radial weights $\phi(|z|)$, provided they satisfy a Lyapunov-type tail coercivity condition.

The motivations for studying the validity of these compactness embeddings and Poincaré inequalities are due to their essential role in proving, using the method of Nash \cite{Nash1958} and Moser \cite{Moser1961}, Harnack’s inequalities and heat kernel estimates for singular operators of the form
\begin{equation*}
    \mathcal{L} = \operatorname{div}(w \nabla u)
\end{equation*}
in $\mathbb{H}^{N+1}$. The interest in this class of singular operators has grown in the last decade as they appear extensively in the literature in both pure and applied problems, particularly in connection with the theory of nonlocal operators. In the special case where the weight relates to the fractional Laplacian, these operators play a major role in the investigation of fractional powers via the ``extension procedure'' of Caffarelli and Silvestre \cite{CaffarelliSilvestre2007} and its generalizations \cite{ArendtElstWarma2018, GaleMianaStinga2013, StingaTorrea2010}.

We also remark that elliptic and parabolic solvability of associated problems in weighted $L^p$ spaces have been investigated in \cite{MetafuneNegroSpina2022, MetafuneNegroSpina2023a, MetafuneNegroSpina2023b, MetafuneNegroSpina2024a, MetafuneNegroSpina2024b}, where properties of analytic semigroups generated by degenerate operators are analyzed.

We now describe the structure of the paper and the strategy for proving our main results. We start by introducing, in Section 2, the definitions of admissible weights, the ``Hardy Property'' (HP), and the ``Tail Coercivity'' (TC) condition. In Section 3, we develop the necessary local tools, including a weighted Hardy inequality for the singular boundary $y=0$ and an Annular Poincaré inequality that relies on scaling arguments independent of the weight's specific form.

In Section 4, we establish the critical link between the Lyapunov tail coercivity condition and the uniform decay of mass at infinity. We prove that the existence of a Lyapunov function $\Psi$ implies a decay estimate of the form
\begin{equation*}
    \int_{\{|z|>R\}} |u|^2 \, d\mu_w \le \frac{C}{m_R} \|u\|_{H_{\mu_w}^1}^2
\end{equation*}
where $m_R \to \infty$ as $R \to \infty$. This result generalizes the Gaussian decay observed in \cite{NegroSpina2025} to any weight with sufficient growth at infinity.

Then, in Section 5 and 6, we discuss the abstract compactness criteria. We prove that the embedding is compact if and only if ``Local Compactness'' and ``Global Tightness'' hold simultaneously. Finally, in Section 7, we combine these elements to prove our main embedding theorems. We show that for $c \le -1$, the compactness of the embedding is structurally equivalent to the validity of the weighted Hardy inequality near the boundary, ensuring that mass does not concentrate at the singularity.
 

\section{Preliminaries}

In this section, we introduce the functional analytic framework and the assumptions on the weight $w(z)$ that will be used throughout the paper. We work in the half-space $\mathbb{H}^{N+1}= \left\{(y, x) \in \mathbb{R} \times \mathbb{R}^N: y>0\right\}$ endowed with the weighted measure $\mu_w=y^c \phi(|z|) d z$, where $c \in \mathbb{R}$ and $\phi$ is a Borel measurable function satisfying suitable local bounds. We define the weighted Lebesgue space $L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right)$ and the corresponding weighted Sobolev space $H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right)$ equipped with the natural norm $\|u\|_{H_{\mu_w}^1}=\|u\|_{L_{\mu_w}^2}+\|\nabla u\|_{L_{\mu_w}^2}$. Our main goal is to identify the necessary conditions on the measure for the compactness of the embedding $H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right) \hookrightarrow L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right)$. To this end, we introduce the notions of "finite mass," "Tail Coercivity" (TC), and "Global Tightness," which characterize the behavior of the weight at infinity and near the singular boundary $y=0$. In the singular case $c \leq-1$, we also introduce the "Hardy Property" (HP), which plays a crucial role in controlling the mass near the boundary degeneracy.

\begin{assumption}\label{ass:1}
Let $N \in \mathbb{N}$, $N \ge 1$, and set 
\begin{equation}
\mathbb{H}^{N+1} := \mathbb{R}^{N+1}_+ = \{ (y,x) \in \mathbb{R} \times \mathbb{R}^N : y > 0 \}.
\end{equation}
\end{assumption}

\begin{assumption}\label{ass:2}
Let $c \in \mathbb{R}$ and let $\phi:[0, \infty) \rightarrow (0, \infty)$ be a Borel measurable function. We define the weight $w(z):=y^c \phi(|z|)$. The weight is admissible if, for every compact interval $K \subset[0, \infty)$ (inclusive of 0 ), there exist constants $0<m_K \leq M_K<\infty$ such that:

\begin{equation}
m_K \leq \phi(\rho) \leq M_K \quad \text { for a.e. } \rho \in K .
\end{equation}
\end{assumption}

\begin{assumption}\label{ass:3}
Define 
\begin{equation}
d\mu_{w} := w(z)\,dz \quad \text{on } \mathbb{H}^{N+1}.
\end{equation}
\end{assumption}

\begin{remark}
We work in the half-space $\mathbb{H}^{N+1}$, which is the natural setting for operators with boundary degeneracy or singularity. The class of weights $w(z)=y^c \phi(|z|)$ considered in Assumption~\ref{ass:2} generalizes the weighted Gaussian measure $\mu=y^c e^{-a|z|^2} d z$ studied in \cite{NegroSpina2025}, allowing for a much broader range of radial profiles $\phi(|z|)$. We explicitly require $\phi$ to be locally bounded away from zero and infinity (Assumption~\ref{ass:2}). This ensures that on any compact set $K \subset \mathbb{H}^{N+1}$ strictly separated from the boundary $y=0$, the measure $\mu_w$ is equivalent to the Lebesgue measure, and the weighted Sobolev space $H_{\mu_w}^1(K)$ is isomorphic to the standard space $H^1(K)$. This local equivalence is essential for recovering the classical local compactness results via the Rellich-Kondrachov theorem.
\end{remark}

\begin{definition}\label{def:2.1}
Define
\begin{equation}
L^{2}_{\mu_{w}} := \Big\{ u \in L^{1}_{\mathrm{loc}} : 
\|u\|_{L^{2}_{\mu_{w}}}^{2} := \int_{\mathbb{H}^{N+1}} |u|^{2}\,d\mu_{w} < \infty \Big\},
\end{equation}
and
\begin{equation}
H^{1}_{\mu_{w}} := \Big\{ u \in L^{2}_{\mu_{w}} : 
\nabla u \in L^{2}_{\mu_{w}}(\mathbb{H}^{N+1};\mathbb{R}^{N+1}) \Big\}, \qquad
\|u\|_{H^{1}_{\mu_{w}}}^{2} := \|u\|_{L^{2}_{\mu_{w}}}^{2} + \|\nabla u\|_{L^{2}_{\mu_{w}}}^{2}.
\end{equation}
\end{definition}

\begin{definition}\label{def:2.2}
The embedding $H^{1}_{\mu_{w}} \hookrightarrow L^{2}_{\mu_{w}}$ is \emph{compact} if every bounded sequence in $H^{1}_{\mu_{w}}$ admits a subsequence that converges in $L^{2}_{\mu_{w}}$.
\end{definition}

\begin{definition}\label{def:2.3}
Define
\begin{equation}
\mathbf{M} := \int_{\mathbb{H}^{N+1}} \phi(|z|)\,dz 
= \int_{\mathbb{H}^{N+1}} \frac{d\mu_{w}}{y^{c}}.
\end{equation}
We say that the weight has \emph{finite mass} if $\mathbf{M} < \infty$.
\end{definition}

\begin{remark}
The definition of the weighted Sobolev space $H_{\mu_w}^1$ is standard. We observe that under the local integrability conditions of Assumption~\ref{ass:2}, the set of smooth functions $C_c^{\infty}\left(\overline{\mathbb{H}^{N+1}}\right)$ is dense in $H_{\mu_w}^1$, as proved in Proposition~\ref{prop:5.2}. In the singular case $c \leq-1$, functions in $H_{\mu_w}^1$ possess a vanishing trace at the boundary $y=0$. Regarding Definition~\ref{def:2.3}, we remark that the finiteness of the measure ( $M<\infty$ ) is a necessary condition for the compactness of the embedding $H_{\mu_w}^1 \hookrightarrow L_{\mu_w}^2$. Indeed, if $M=\infty$, one can construct a sequence of disjoint translates that converges weakly to zero but maintains a constant non-zero $L_{\mu_w}^2$ norm, thereby violating compactness (see Proposition~\ref{prop:6.2}). In the Gaussian case \cite{NegroSpina2025}, finite mass is automatic due to the exponential decay of the weight.    
\end{remark}

\begin{definition}\label{def:2.4}[Coercive tail inequality / Lyapunov condition Definition~\ref]
There exist a function $\Psi : [0,\infty) \to [0,\infty)$ with $\Psi(\rho) \uparrow \infty$ as $\rho \to \infty$ and a constant $C > 0$ such that
\begin{equation}
\int_{\mathbb{H}^{N+1}} \Psi(|z|)\,u^{2}\,d\mu_{w} 
\le C \int_{\mathbb{H}^{N+1}} \big( |\nabla u|^{2} + u^{2} \big)\,d\mu_{w}
\end{equation}
for all $u \in C_{c}^{\infty}(\overline{\mathbb{H}^{N+1}})$ that vanish near $y = 0$ if $c \le -1$ 
(no boundary condition is needed if $c > -1$). 
We abbreviate this condition by $(\mathrm{TC})$.
\end{definition}

\begin{remark}
The "Tail Coercivity" (TC) condition serves as a substitute for the explicit exponential decay of the Gaussian weight used in \cite{NegroSpina2025}. It provides an abstract Lyapunov-type criterion that ensures the mass of functions in the unit ball of $H_{\mu_w}^1$ vanishes uniformly at infinity. We emphasize that this condition is satisfied by a wide class of weights, including the Gaussian weight $\phi(r)=e^{-a r^2}$ (where $\Psi(r) \sim r^2$ ) and polynomial weights with sufficient growth. This abstract framework allows us to derive decay estimates of the form

$$
\int_{\{|z|>R\}}|u|^2 d \mu_w \leq \frac{C}{m_R}\|u\|_{H_{\mu_w}^1}^2
$$

where $m_R \rightarrow \infty$ as $R \rightarrow \infty$, which is crucial for establishing the "Tail Tightness" of the embedding.
\end{remark}

\begin{definition}\label{def:2.5}
If $c \leq-1$, we say the measure space $\left(\mathbb{H}^{N+1}, \mu_w\right)$ possesses the Hardy Property (HP) if there exists a constant $C_H>0$ such that for all $u \in H_{\mu_w}^1$ :

\begin{equation}
\int_{\mathbb{H}^{N+1}} \frac{|u(z)|^2}{y^2} d \mu_w(z) \leq C_H\|\nabla u\|_{L_{\mu_w}^2}^2
\end{equation}
\end{definition}

\begin{definition}\label{def:2.6}
A subset $\mathcal{B} \subset L_{\mu_w}^2$ is said to be globally tight if it satisfies the following two mass-vanishing conditions:
\begin{enumerate}
    \item Tail Tightness: The mass vanishes uniformly at infinity:

\begin{equation}
\lim _{R \rightarrow \infty} \sup _{u \in \mathcal{B}}\|u\|_{L_{\mu_w}^2(\{z:|z|>R\})}=0 .
\end{equation}
    \item Boundary Tightness: The mass vanishes uniformly near the singular boundary $y=0$

\begin{equation}
\limsup _{\delta \rightarrow 0}\|u\|_{u \in \mathcal{B}} L_{\mu_w}^2(\{z: 0<y<\delta\})=0
\end{equation}
\end{enumerate}
\end{definition}

\begin{remark}
In the range $c>-1$, the weight $y^c$ is locally integrable, and mass concentration near the boundary $y=0$ is ruled out by the absolute continuity of the integral. However, in the singular range $c \leq-1$, the weight is not locally integrable, and $L_{\mu_w}^2$ functions might a priori concentrate mass near the singularity. The "Hardy Property" (HP) in Definition 2.5 is introduced to control this behavior. It ensures that functions in $H_{\mu_w}^1$ vanish sufficiently fast at the boundary to offset the singularity of the weight $y^c$. Finally, Definition 2.6 formalizes the Fréchet-Kolmogorov Compactness Criterion adapted to weighted spaces. The "Global Tightness" condition encapsulates the requirement that no mass "escapes" to infinity or "concentrates" at the singularity, which, combined with local compactness, characterizes the compactness of the embedding completely (see Theorem 7.2). 
\end{remark}

\section{Functional Inequalities and Local Tools}

We now develop the necessary local tools and functional inequalities that will be instrumental in proving our main compactness results. We start by considering bounded domains $\Omega \subset \mathbb{H}^{N+1}$ strictly separated from the boundary singularity $y=0$. In such domains, under Assumption 2, the weight is bounded strictly away from zero and infinity, which allows us to recover the standard Rellich-Kondrachov compactness theorem via a norm equivalence argument. To handle the singularity at the boundary, we prove a weighted Hardy inequality in the normal direction for functions vanishing near $y=0$. Furthermore, we establish an Annular Poincaré inequality for the gradient term, which relies on scaling arguments and is independent of the specific form of the weight. These local estimates provide the building blocks for establishing global tightness in the subsequent sections.

\begin{lemma}\label{lem:3.1}
Let $\Omega \subset \mathbb{H}^{N+1}$ be any bounded open set strictly separated from the boundary singularity $y=0$ (i.e., $\operatorname{dist}(\Omega,\{y= 0\})>0$). Under  Assumption~\ref{ass:2}, the weight $w(z)$ is bounded strictly away from zero and infinity on $\Omega$. Therefore, the weighted Sobolev space is isomorphic to the standard space, $H_{\mu_w}^1(\Omega) \cong H^1(\Omega)$, and the embedding $H_{\mu_w}^1(\Omega) \hookrightarrow L_{\mu_w}^2(\Omega)$ is compact by the classical Rellich-Kondrachov theorem.
\end{lemma}

\begin{proof}
By hypothesis, the domain $\Omega$ satisfies two geometric constraints. First, since $\Omega$ is bounded, there exists a radius $R_{\max} \in (0, \infty)$ such that for all $z \in \Omega$, $|z| \le R_{\max}$. Second, since $\operatorname{dist}(\Omega, \{y=0\}) > 0$, there exists a parameter $\delta > 0$ such that for all $z=(y,x) \in \Omega$, $y \ge \delta$. Observing that $|z| = \sqrt{y^2 + |x|^2} \ge y$, we deduce that for all $z \in \Omega$, the radial coordinate $|z|$ is confined to the compact interval
\begin{equation}
\rho(z) \in K := [\delta, R_{\max}] \subset (0, \infty).
\end{equation}
Similarly, the vertical coordinate $y$ is confined to the compact interval
\begin{equation}
y(z) \in I := [\delta, R_{\max}] \subset (0, \infty).
\end{equation}

We now analyze the weight function $w(z) = y^c \phi(|z|)$ on the domain $\Omega$. We invoke  Assumption~\ref{ass:2}, which states that for the compact interval $K$ identified above, there exist constants $0 < m_K \le M_K < \infty$ such that $m_K \le \phi(\rho) \le M_K$ for almost every $\rho \in K$. Regarding the vertical component, the function $f(t) = t^c$ is continuous on the strictly positive compact interval $I = [\delta, R_{\max}]$. By the Weierstrass Extreme Value Theorem, $f(t)$ attains a strictly positive minimum $m_y$ and a finite maximum $M_y$ on $I$:
\begin{equation}
0 < m_y := \min_{t \in I} t^c \le y^c \le \max_{t \in I} t^c =: M_y < \infty.
\end{equation}
Combining these estimates, for almost every $z \in \Omega$, the weight $w(z)$ satisfies the uniform bounds
\begin{equation}
0 < \lambda_{\min} := m_y m_K \le w(z) \le M_y M_K := \lambda_{\max} < \infty.
\end{equation}

We define the squared norm of the weighted Sobolev space $H_{\mu_w}^1(\Omega)$ as
\begin{equation}
\|u\|_{H_{\mu_w}^1(\Omega)}^2 := \int_{\Omega} (|\nabla u|^2 + |u|^2) w(z) \, dz.
\end{equation}
Using the uniform bounds derived above, we estimate this integral against the standard Lebesgue measure $dz$:
\begin{equation}
\lambda_{\min} \int_{\Omega} (|\nabla u|^2 + |u|^2) \, dz \le \int_{\Omega} (|\nabla u|^2 + |u|^2) w(z) \, dz \le \lambda_{\max} \int_{\Omega} (|\nabla u|^2 + |u|^2) \, dz.
\end{equation}
Recognizing the integral $\int_{\Omega} (|\nabla u|^2 + |u|^2) \, dz$ as the standard squared Sobolev norm $\|u\|_{H^1(\Omega)}^2$, we obtain the inequality
\begin{equation}
\lambda_{\min} \|u\|_{H^1(\Omega)}^2 \le \|u\|_{H_{\mu_w}^1(\Omega)}^2 \le \lambda_{\max} \|u\|_{H^1(\Omega)}^2.
\end{equation}
This inequality implies that the identity map is a bi-continuous linear bijection between $H_{\mu_w}^1(\Omega)$ and $H^1(\Omega)$. Thus, the spaces are topologically isomorphic: $H_{\mu_w}^1(\Omega) \cong H^1(\Omega)$.

Finally, we consider the embedding operator $\iota: H_{\mu_w}^1(\Omega) \hookrightarrow L_{\mu_w}^2(\Omega)$. Let $\{u_n\}_{n \in \mathbb{N}}$ be a bounded sequence in $H_{\mu_w}^1(\Omega)$. By the norm equivalence established above, $\{u_n\}$ is also a bounded sequence in the standard space $H^1(\Omega)$. We invoke the Classical Rellich-Kondrachov Theorem. Assuming $\Omega$ satisfies standard regularity conditions required for the unweighted theorem, the embedding $H^1(\Omega) \hookrightarrow L^2(\Omega)$ is compact. Consequently, there exists a subsequence $\{u_{n_k}\}$ that converges strongly to a limit $u$ in $L^2(\Omega)$, meaning $\lim_{k \to \infty} \|u_{n_k} - u\|_{L^2(\Omega)} = 0$. To verify convergence in the weighted $L^2$ space, we apply the upper bound of the weight:
\begin{equation}
\|u_{n_k} - u\|_{L_{\mu_w}^2(\Omega)}^2 = \int_{\Omega} |u_{n_k} - u|^2 w(z) \, dz \le \lambda_{\max} \int_{\Omega} |u_{n_k} - u|^2 \, dz = \lambda_{\max} \|u_{n_k} - u\|_{L^2(\Omega)}^2.
\end{equation}
Since the right-hand side vanishes as $k \to \infty$, it follows that $\|u_{n_k} - u\|_{L_{\mu_w}^2(\Omega)} \to 0$. Thus, every bounded sequence in $H_{\mu_w}^1(\Omega)$ possesses a convergent subsequence in $L_{\mu_w}^2(\Omega)$, and the embedding is compact.
\end{proof}

\begin{remark}
Lemma~\ref{lem:3.1} establishes that on any bounded domain $\Omega$ strictly separated from the singular boundary $y=0$, the weighted Sobolev space $H_{\mu_w}^1(\Omega)$ is isomorphic to the classical Sobolev space $H^1(\Omega)$. This result relies crucially on the local boundedness of the weight $w(z)=y^c \phi(|z|)$ away from the singularity and infinity. Consequently, the classical RellichKondrachov compactness theorem applies locally. This local compactness is a fundamental building block for the global compactness result in Theorem 5.1, allowing us to focus our analysis on the "loss of compactness" at the boundary and at infinity . 
\end{remark}

\begin{lemma}[1D weighted Hardy inequality in the normal direction]\label{lem:3.2}
If $c > -1$ and $u \in C^{\infty}_{c}(\overline{\mathbb{H}^{N+1}})$ vanishes in a neighbourhood of $\{ y = 0 \}$, then
\begin{equation}
\int_{\mathbb{H}^{N+1}} \frac{u^{2}}{y^{2}}\, y^{c}\, dz
\ \le\ 
\frac{4}{(c+1)^{2}}
\int_{\mathbb{H}^{N+1}} \big| \partial_{y} u \big|^{2}\, y^{c}\, dz.
\end{equation}
\end{lemma}

\begin{remark}
The weighted Hardy inequality in Lemma~\ref{lem:3.2} is essential for controlling the behavior of functions near the singular boundary $y=0$ when $c \leq-1$. In this range, the weight $y^c$ is not locally integrable, and standard Sobolev embeddings fail. The inequality ensures that functions in $H_{\mu_w}^1$ vanish sufficiently fast at the boundary to make the weighted $L^2$ norm finite. This "Hardy Property" (HP) effectively prevents mass concentration at the singularity, a condition we later show is necessary for compactness in the singular regime (see Theorem~\ref{thm:7.1}). For $c>-1$, the weight is locally integrable, and this specific control is not required for the definition of the space, although Hardy-type inequalities may still hold.
\end{remark}

\begin{lemma}[Annular Poincaré inequality]\label{lem:3.3}
Let 
\begin{equation}
A_{R} := \{ z \in \mathbb{R}^{N+1} : R < |z| < 2R \}, \qquad R > 0.
\end{equation}
Then for all $u \in H^{1}(A_{R})$,
\begin{equation}
\int_{A_{R}} \big| u - u_{A_{R}} \big|^{2} \, dz
\ \le\
C\, R^{2} \int_{A_{R}} |\nabla u|^{2} \, dz,
\qquad
u_{A_{R}} := \frac{1}{|A_{R}|} \int_{A_{R}} u \, dz,
\end{equation}
with a constant $C > 0$ independent of $R$.
\end{lemma}

\begin{proof}
\textbf{Step 1 (Reference annulus and scaling map).}
Fix the unit annulus 
\begin{equation}
A_{1} := \{ y \in \mathbb{R}^{N+1} : 1 < |y| < 2 \}.
\end{equation}
Define the dilation map $T_{R} : A_{1} \to A_{R}$ by $T_{R}(y) := R\,y$.

\medskip
\textbf{Step 2 (Pull–back / rescaled function).}
Given $u \in H^{1}(A_{R})$, define the rescaled function $v(y) := u(T_{R}(y)) = u(Ry)$ on $A_{1}$.  
Then $v \in H^{1}(A_{1})$.

\medskip
\textbf{Step 3 (Change–of–variables data).}
Let $d := N + 1$.  
For $z = R y$, one has
\begin{equation}
dz = R^{d}\, dy, 
\qquad
\nabla_{y} v(y) = R\, (\nabla u)(R y).
\end{equation}

\medskip
\textbf{Step 4 (Averages are invariant under the dilation).}
\begin{equation}
u_{A_{R}} 
= \frac{1}{|A_{R}|} \int_{A_{R}} u(z)\, dz
= \frac{1}{R^{d} |A_{1}|} \int_{A_{1}} u(Ry)\, R^{d}\, dy
= \frac{1}{|A_{1}|} \int_{A_{1}} v(y)\, dy
=: v_{A_{1}}.
\end{equation}

\medskip
\textbf{Step 5 (Transfer of the $L^{2}$ fluctuation by scaling).}
\begin{equation}
\int_{A_{R}} |u - u_{A_{R}}|^{2}\, dz
= \int_{A_{1}} |u(Ry) - u_{A_{R}}|^{2} R^{d}\, dy
= R^{d} \int_{A_{1}} |v - v_{A_{1}}|^{2}\, dy.
\end{equation}

\medskip
\textbf{Step 6 (Transfer of the Dirichlet energy by scaling).}
\begin{equation}
\int_{A_{R}} |\nabla u|^{2}\, dz
= \int_{A_{1}} |\nabla u(Ry)|^{2} R^{d}\, dy
= \frac{1}{R^{2}} R^{d} \int_{A_{1}} |\nabla_{y} v|^{2}\, dy
= R^{d-2} \int_{A_{1}} |\nabla v|^{2}\, dy.
\end{equation}

\medskip
\textbf{Step 7 (Poincaré inequality on the fixed annulus $A_{1}$).}
Since $A_{1}$ is a bounded Lipschitz domain, the mean–zero Poincaré inequality gives
\begin{equation}
\int_{A_{1}} |v - v_{A_{1}}|^{2}\, dy
\ \le\ 
C_{1} \int_{A_{1}} |\nabla v|^{2}\, dy,
\end{equation}
where $C_{1} > 0$ depends only on $A_{1}$.

\medskip
\textbf{Step 8 (Assembling the estimates).}
Combining the above relations,
\begin{equation}
\int_{A_{R}} |u - u_{A_{R}}|^{2}\, dz
= R^{d} \int_{A_{1}} |v - v_{A_{1}}|^{2}\, dy
\le C_{1} R^{d} \int_{A_{1}} |\nabla v|^{2}\, dy
= C_{1} R^{2} \int_{A_{R}} |\nabla u|^{2}\, dz.
\end{equation}

\medskip
\textbf{Step 9 (Renaming the universal constant).}
Setting $C := C_{1}$, independent of $R$, gives
\begin{equation}
\int_{A_{R}} |u - u_{A_{R}}|^{2}\, dz
\ \le\ 
C\, R^{2} \int_{A_{R}} |\nabla u|^{2}\, dz.
\end{equation}

\medskip
\textbf{Conclusion.}
The inequality holds for all $R > 0$ with a constant $C > 0$ independent of the radius.
\end{proof}

\begin{remark}
Lemma~\ref{lem:3.3} provides a dimension-free "Annular Poincaré inequality" that is robust under scaling . The proof exploits the scaling invariance of the $H^1$ norm and the standard Poincaré inequality on a fixed reference annulus $A_1$. Importantly, this result does not depend on the specific radial profile of the weight, as it is applied locally on annuli where the weight is comparable to a constant (or slowly varying). This tool is pivotal for establishing "Tail Tightness" in the proof of the main theorem, allowing us to control the $L^2$ fluctuations in the tail region via the gradient.
\end{remark}

\section{Tail Control and Decay Estimates}

In this section, we establish the critical link between the Lyapunov "Tail Coercivity" condition and the uniform decay of mass at infinity. We prove that the existence of a Lyapunov function $\Psi$ implies a decay estimate of the form

$$
\int_{\{|z|>R\}}|u|^2 d \mu_w \leq \frac{C}{m_R}\|u\|_{H_{\mu_w}^1}^2,
$$

where $m_R \rightarrow \infty$ as $R \rightarrow \infty$. This result generalizes the Gaussian decay observed in previous works to any weight with sufficient growth at infinity. We show that for $R$ sufficiently large, the tail mass of any function in the unit ball of $H_{\mu_w}^1$ is uniformly controlled by the growth of the Lyapunov potential, a property we term "Tail Tightness". This decay estimate is essential for verifying the abstract compactness criteria we discuss next.

\begin{lemma}[Tail control from coercivity]\label{lem:4.1}
Assume \emph{(TC)} from Definition~\ref{def:2.4}. Then there exist $R_{0} \ge 1$ and a function $\Theta(R) \downarrow 0$ such that for all $R \ge R_{0}$ and all $u \in C_{c}^{\infty}(\mathbb{H}^{N+1})$,
\begin{equation}
\int_{\{ |z| > 2R \}} u^{2}\, d\mu_{w}
\ \le\
\Theta(R)\,
\int_{\{ |z| > R \}} \big( |\nabla u|^{2} + u^{2} \big)\, d\mu_{w}.
\end{equation}
\end{lemma}

\begin{proof}
\textbf{Step 1 (Data from (TC)).}
There exist a function $\Psi : [0,\infty) \to [0,\infty)$, with $\Psi(\rho) \uparrow \infty$ as $\rho \to \infty$, and a constant $C > 0$ such that for all $v \in C_{c}^{\infty}(\overline{\mathbb{H}^{N+1}})$
(vanishing near $y = 0$ if $c \le -1$),
\begin{equation}
\int_{\mathbb{H}^{N+1}} \Psi(|z|)\, v^{2}\, d\mu_{w}
\ \le\
C \int_{\mathbb{H}^{N+1}} \big( |\nabla v|^{2} + v^{2} \big)\, d\mu_{w}.
\end{equation}
Since $u \in C_{c}^{\infty}(\mathbb{H}^{N+1})$ vanishes near $y = 0$, it is admissible in Definition~\ref{def:2.4}.

\medskip
\textbf{Step 2 (Radial cut–off).}
Fix $R \ge 1$ and choose $\eta_{R} \in C_{c}^{\infty}(\mathbb{R}^{N+1})$ radial such that
\begin{equation}
0 \le \eta_{R} \le 1, \qquad
\eta_{R}(z) = 0 \text{ if } |z| \le R, \qquad
\eta_{R}(z) = 1 \text{ if } |z| \ge 2R, \qquad
|\nabla \eta_{R}(z)| \le \frac{C_{0}}{R} \text{ for } R < |z| < 2R.
\end{equation}

\medskip
\textbf{Step 3 (Tail–localized test function).}
Set $v := \eta_{R}\, u$.  
Then $v \in C_{c}^{\infty}(\mathbb{H}^{N+1})$, admissible in Definition~\ref{def:2.4}.

\medskip
\textbf{Step 4 (Apply (TC) to $v$).}
\begin{equation}
\int_{\mathbb{H}^{N+1}} \Psi(|z|)\, \eta_{R}^{2} u^{2}\, d\mu_{w}
\ \le\
C \int_{\mathbb{H}^{N+1}}
\big( |\nabla(\eta_{R} u)|^{2} + \eta_{R}^{2} u^{2} \big)\, d\mu_{w}.
\end{equation}

\medskip
\textbf{Step 5 (Lower bound of the left–hand side).}
On $\{ |z| > 2R \}$, $\eta_{R} \equiv 1$ and $\Psi(|z|) \ge \Psi(2R)$, so
\begin{equation}
\Psi(2R) \int_{\{ |z| > 2R \}} u^{2}\, d\mu_{w}
\ \le\
\int_{\mathbb{H}^{N+1}} \Psi(|z|)\, \eta_{R}^{2} u^{2}\, d\mu_{w}.
\end{equation}

\medskip
\textbf{Step 6 (Product rule and Young’s inequality).}
The product rule gives
\begin{equation}
|\nabla(\eta_{R} u)|^{2}
= \eta_{R}^{2} |\nabla u|^{2}
+ 2 \eta_{R} u\, \nabla \eta_{R} \!\cdot\! \nabla u
+ u^{2} |\nabla \eta_{R}|^{2}.
\end{equation}
For $\varepsilon \in (0,1]$, the mixed term satisfies
\begin{equation}
2 |\eta_{R} u\, \nabla \eta_{R} \!\cdot\! \nabla u|
\le
\varepsilon\, \eta_{R}^{2} |\nabla u|^{2}
+ \frac{1}{\varepsilon} u^{2} |\nabla \eta_{R}|^{2}.
\end{equation}
Thus
\begin{equation}
|\nabla(\eta_{R} u)|^{2}
\le
(1 + \varepsilon)\, \eta_{R}^{2} |\nabla u|^{2}
+ \Big( 1 + \frac{1}{\varepsilon} \Big) u^{2} |\nabla \eta_{R}|^{2}.
\end{equation}

\medskip
\textbf{Step 7 (Support of $\nabla \eta_{R}$).}
The gradient satisfies $\operatorname{supp} \nabla \eta_{R} \subset A_{R} := \{ R < |z| < 2R \}$ and
$|\nabla \eta_{R}|^{2} \le C_{0}^{2} / R^{2}$ there.

\medskip
\textbf{Step 8 (Upper bound of the right–hand side).}
Combining the previous estimates,
\begin{equation}
\int_{\mathbb{H}^{N+1}}
\big( |\nabla(\eta_{R} u)|^{2} + \eta_{R}^{2} u^{2} \big)\, d\mu_{w}
\le
(1 + \varepsilon) \int_{\{ |z| > R \}} \eta_{R}^{2} |\nabla u|^{2}\, d\mu_{w}
+ \Big( 1 + \frac{1}{\varepsilon} \Big) \frac{C_{0}^{2}}{R^{2}} \int_{A_{R}} u^{2}\, d\mu_{w}
+ \int_{\{ |z| > R \}} \eta_{R}^{2} u^{2}\, d\mu_{w}.
\end{equation}

\medskip
\textbf{Step 9 (Regional comparisons).}
Since $0 \le \eta_{R} \le 1$ and $\eta_{R} = 0$ for $|z| \le R$,
\begin{equation}
\int_{\{ |z| > R \}} \eta_{R}^{2} |\nabla u|^{2}
\le \int_{\{ |z| > R \}} |\nabla u|^{2}, 
\qquad
\int_{\{ |z| > R \}} \eta_{R}^{2} u^{2}
\le \int_{\{ |z| > R \}} u^{2}.
\end{equation}

\medskip
\textbf{Step 10 (Combine the estimates).}
\begin{equation}
\Psi(2R) \int_{\{ |z| > 2R \}} u^{2}\, d\mu_{w}
\le
C \Big[
(1 + \varepsilon) \int_{\{ |z| > R \}} |\nabla u|^{2}\, d\mu_{w}
+ \int_{\{ |z| > R \}} u^{2}\, d\mu_{w}
+ \Big( 1 + \frac{1}{\varepsilon} \Big) \frac{C_{0}^{2}}{R^{2}} \int_{A_{R}} u^{2}\, d\mu_{w}
\Big].
\end{equation}

\medskip
\textbf{Step 11 (Absorb the annular term).}
Since $A_{R} \subset \{ |z| > R \}$,
\begin{equation}
\int_{A_{R}} u^{2} \le \int_{\{ |z| > R \}} u^{2}.
\end{equation}
Choosing $\varepsilon = 1$, we obtain
\begin{equation}
\Psi(2R) \int_{\{ |z| > 2R \}} u^{2}\, d\mu_{w}
\le
C_{1} \int_{\{ |z| > R \}} \big( |\nabla u|^{2} + u^{2} \big)\, d\mu_{w}
+ \frac{C_{2}}{R^{2}} \int_{\{ |z| > R \}} u^{2}\, d\mu_{w}.
\end{equation}

\medskip
\textbf{Step 12 (Renormalization and definition of the tail modulus).}
Rewriting the previous inequality,
\begin{equation}
\int_{\{ |z| > 2R \}} u^{2}\, d\mu_{w}
\le
\underbrace{\frac{C_{1}}{\Psi(2R)}}_{=: \Theta_{1}(R)}
\int_{\{ |z| > R \}} (|\nabla u|^{2} + u^{2})\, d\mu_{w}
+
\underbrace{\frac{C_{2}}{R^{2} \Psi(2R)}}_{=: \Theta_{2}(R)}
\int_{\{ |z| > R \}} u^{2}\, d\mu_{w}.
\end{equation}

\medskip
\textbf{Step 13 (Decay of the tail coefficients).}
Since $\Psi(2R) \uparrow \infty$ as $R \to \infty$, both $\Theta_{1}(R)$ and $\Theta_{2}(R)$ decrease to $0$.  
Define $\Theta(R) := \Theta_{1}(R) + \Theta_{2}(R)$, which satisfies $\Theta(R) \downarrow 0$ as $R \to \infty$.

\medskip
\textbf{Step 14 (Choice of $R_{0}$).}
Select $R_{0} \ge 1$ large enough so that $\Psi(2R) > 0$ and
$\max\{ \Theta_{1}(R), \Theta_{2}(R) \} \le 1$ for all $R \ge R_{0}$.

\medskip
\textbf{Conclusion.}
Hence there exist $R_{0} \ge 1$ and $\Theta(R) \downarrow 0$ such that
\begin{equation}
\int_{\{ |z| > 2R \}} u^{2}\, d\mu_{w}
\le
\Theta(R)
\int_{\{ |z| > R \}} \big( |\nabla u|^{2} + u^{2} \big)\, d\mu_{w},
\end{equation}
for all $R \ge R_{0}$ and all $u \in C_{c}^{\infty}(\mathbb{H}^{N+1})$.
\end{proof}

\begin{remark}
Lemma~\ref{lem:4.1} translates the abstract Lyapunov condition (TC) into a concrete "Tail Control" estimate . By using a radial cut-off function $\eta_R$ and the coercivity of $\Psi(|z|)$, we derive that the mass of a function in the far field $\{|z|>2 R\}$ is controlled by its weighted Sobolev norm on the annulus $\{R<|z|<2 R\}$ and the tail $\left\{|z|>\left[\right.\right.$ cite $_s$ tart $\left.] R\right\}$. This step is critical because it converts the qualitative assumption of "growth at infinity" into a quantitative decay estimate, showing that mass cannot escape to infinity for functions with bounded energy.
\end{remark}

\begin{lemma}\label{lem:4.2}
Assume the weight $w$ satisfies the Lyapunov Tail Coercivity condition (Definition~\ref{def:2.4}) with Lyapunov potential $\Psi$ (Definition 2.4). Then, for any $R$ sufficiently large such that $m_R := \inf_{|z| \ge R} \Psi(z) > 0$, the following decay estimate holds for all $u \in H_{\mu_w}^1$:
\begin{equation}
\int_{\{|z|>R\}} |u|^2 \, d\mu_w \le \frac{C_{TC}}{m_R} \|u\|_{H_{\mu_w}^1}^2
\end{equation}
\end{lemma}

\begin{proof}
Let $H_{\mu_w}^1(\mathbb{H}^{N+1})$ denote the completion of the space of smooth functions with respect to the norm $\|u\|_{H_{\mu_w}^1}^2 := \int (|\nabla u|^2 + |u|^2) \, d\mu_w$. By Proposition~\ref{prop:5.2}, the set of smooth functions with compact support, denoted $\mathcal{C} = C_c^\infty(\overline{\mathbb{H}^{N+1}})$ (potentially vanishing near $y=0$ if $c \le -1$), is dense in $H_{\mu_w}^1(\mathbb{H}^{N+1})$.

The Definition~\ref{def:2.4} condition is axiomatically defined for test functions $v \in \mathcal{C}$ as:
\begin{equation}
\int_{\mathbb{H}^{N+1}} \Psi(|z|) |v(z)|^2 \, d\mu_w(z) \le C_{TC} \|v\|_{H_{\mu_w}^1}^2
\end{equation}
Since both the weighted $L^2$ norm with weight $\Psi w$ and the Sobolev norm are continuous functionals under the topology of $H_{\mu_w}^1$, the inequality extends to all $u \in H_{\mu_w}^1$ by a standard density argument.

Consider the tail region $E_{out} := \{z \in \mathbb{H}^{N+1} : |z| > R\}$. By the hypothesis, we define the infimum on the closure of the tail region as $m_R = \inf \{ \Psi(z) : |z| \ge R \}$. For any $z \in E_{out}$, we have $|z| > R$, which implies $\Psi(z) \ge m_R$. Consequently, the following pointwise inequality holds almost everywhere on $E_{out}$:
\begin{equation}
\Psi(z) |u(z)|^2 \ge m_R |u(z)|^2
\end{equation}
Integrating this pointwise inequality over the set $E_{out}$ with respect to the non-negative measure $d\mu_w$, and utilizing the linearity of the integral to factor out the constant $m_R$, we obtain:
\begin{equation}
\int_{\{|z|>R\}} \Psi(z) |u(z)|^2 \, d\mu_w(z) \ge m_R \int_{\{|z|>R\}} |u(z)|^2 \, d\mu_w(z)
\end{equation}
Since the potential $\Psi$ and the integrand are non-negative, the integral over the subset $E_{out}$ is bounded above by the integral over the entire space $\mathbb{H}^{N+1}$:
\begin{equation}
\int_{\mathbb{H}^{N+1}} \Psi(z) |u(z)|^2 \, d\mu_w(z) \ge \int_{\{|z|>R\}} \Psi(z) |u(z)|^2 \, d\mu_w(z)
\end{equation}
Combining the extended (TC) condition with these lower bounds yields the chain of inequalities:
\begin{equation}
C_{TC} \|u\|_{H_{\mu_w}^1}^2 \ge \int_{\mathbb{H}^{N+1}} \Psi |u|^2 \, d\mu_w \ge m_R \int_{\{|z|>R\}} |u|^2 \, d\mu_w
\end{equation}
Rearranging the terms implies:
\begin{equation}
m_R \int_{\{|z|>R\}} |u|^2 \, d\mu_w \le C_{TC} \|u\|_{H_{\mu_w}^1}^2
\end{equation}
Since $m_R > 0$ by hypothesis, we may divide by $m_R$ to obtain the stated decay estimate.
\end{proof}

\begin{remark}
Lemma~\ref{lem:4.2} refines the result of Lemma~\ref{lem:4.1} into a uniform decay estimate for the unit ball of $H_{\mu_w}^1$. It shows that if the Lyapunov potential $\Psi$ grows to infinity (i.e., $m_R \rightarrow \infty$ ), then the $L^2$ mass in the tail $\left\{|z|>\left[\right.\right.$ cite $\left.\left._s \operatorname{tart}\right] R\right\}$ decays as $O\left(1 / m_R\right)$. This explicit rate of decay establishes the "Tail Tightness" required for the Fréchet-Kolmogorov criterion (Theorem~\ref{thm:5.1}) . The result generalizes the exponential decay observed for Gaussian weights in \cite{NegroSpina2025} to any weight admitting a coercive Lyapunov potential, highlighting that it is the existence of such a potential, rather than the specific Gaussian form, that drives compactness.
\end{remark}

\section{Abstract Compactness and Regularity}

We now discuss the abstract criteria for the compactness of the embedding $H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right) \hookrightarrow L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right)$. We prove that the embedding is compact if and only if two conditions are met simultaneously: "Local Compactness" on bounded subdomains and "Global Tightness," which ensures that mass does not escape to infinity or concentrate at the singular boundary. The proof relies on a diagonal extraction argument combined with the uniform decay estimates established in Section 4. Additionally, we address the density of smooth functions in the weighted space $H_{\mu_w}^1$, showing that functions in $C_c^{\infty}\left(\overline{\mathbb{H}^{N+1}}\right)$ (potentially vanishing at $y=0$ if $c \leq-1$ ) are dense. This density result allows us to extend our functional inequalities from smooth test functions to the entire Sobolev space.

\begin{theorem}\label{thm:5.1}
Let $(\Omega, \mu)$ be a $\sigma$-finite measure space. Let $H^1(\Omega)$ and $L^2(\Omega)$ be the Hilbert spaces defined via the measure $\mu$. We assume the existence of a ``singular boundary'' $\Gamma$ (possibly empty) and a distance function $\operatorname{dist}(\cdot, \Gamma)$. The embedding $\iota: H^1(\Omega) \hookrightarrow L^2(\Omega)$ is compact if and only if the following two conditions hold:
\begin{enumerate}
    \item \textbf{Local Compactness:} For every compact subset $K \subset \overline{\Omega}$ such that $\operatorname{dist}(K, \Gamma) > 0$, the restriction operator $\iota_K: H^1(K) \to L^2(K)$ is compact.
    \item \textbf{Global Tightness:} The unit ball $\mathcal{B} := \{u \in H^1(\Omega) : \|u\|_{H^1} \le 1\}$ satisfies:
    \begin{itemize}
        \item \textit{Tail Tightness:} $\lim_{R \to \infty} \sup_{u \in \mathcal{B}} \|u\|_{L^2(\Omega \setminus B_R)}^2 = 0$, where $B_R = \{z \in \Omega : |z| \le R\}$.
        \item \textit{Boundary Tightness:} $\lim_{\delta \to 0} \sup_{u \in \mathcal{B}} \|u\|_{L^2(\Omega_\delta)}^2 = 0$, where $\Omega_\delta = \{z \in \Omega : \operatorname{dist}(z, \Gamma) < \delta\}$.
    \end{itemize}
\end{enumerate}
\end{theorem}

\begin{proof}
We first prove the sufficiency of the conditions. Assume that Local Compactness and Global Tightness hold. To prove that the embedding is compact, we must show that every bounded sequence $(u_n)_{n \in \mathbb{N}} \subset \mathcal{B}$ contains a subsequence converging strongly in $L^2(\Omega)$.

Let $(\delta_k)_{k \in \mathbb{N}}$ be a sequence of positive real numbers monotonically decreasing to $0$, and let $(R_k)_{k \in \mathbb{N}}$ be a sequence of positive real numbers monotonically increasing to $\infty$. We define the compact core at stage $k$ as
\begin{equation}
K_k := \{z \in \Omega : |z| \le R_k \text{ and } \operatorname{dist}(z, \Gamma) \ge \delta_k\}.
\end{equation}
The set $K_k$ is closed and bounded. Since it is strictly separated from the singular boundary $\Gamma$ and from infinity, $K_k$ is a standard compact domain where the Local Compactness condition applies.

We proceed by diagonal extraction. Consider the sequence $(u_n)$. For $k=1$, by Local Compactness, the embedding $H^1(K_1) \hookrightarrow L^2(K_1)$ is compact. Thus, there exists a subsequence, denoted $(u_{n}^{(1)})$, which converges strongly in $L^2(K_1)$. Inductively, from the sequence $(u_{n}^{(k-1)})$, we extract a further subsequence $(u_{n}^{(k)})$ that converges strongly in $L^2(K_k)$. We then define the diagonal sequence $v_n := u_{n}^{(n)}$. By construction, for any fixed integer $k$, the sequence $(v_n)_{n \ge k}$ is a subsequence of $(u_{n}^{(k)})$. Therefore, $(v_n)$ is a Cauchy sequence in $L^2(K_k)$ for every $k$.

We assert that $(v_n)$ is a Cauchy sequence in the global space $L^2(\Omega)$. Let $\epsilon > 0$ be an arbitrary positive constant. We decompose the $L^2$ norm of the difference $w_{nm} := v_n - v_m$ into the core and defect regions:
\begin{equation}
\|w_{nm}\|_{L^2(\Omega)}^2 = \int_{K_k} |w_{nm}|^2 \, d\mu + \int_{\Omega \setminus K_k} |w_{nm}|^2 \, d\mu.
\end{equation}
The complement $\Omega \setminus K_k$ is the union of the tail region $\Omega \setminus B_{R_k}$ and the boundary strip $\Omega_{\delta_k}$. By the triangle inequality in $L^2$, $\|w_{nm}\|_{L^2(S)} \le \|v_n\|_{L^2(S)} + \|v_m\|_{L^2(S)}$. Since $v_n, v_m \in \mathcal{B}$, we have $\|w_{nm}\|_{L^2(S)} \le 2 \sup_{u \in \mathcal{B}} \|u\|_{L^2(S)}$. We invoke Global Tightness to choose $k$ sufficiently large. By Tail Tightness, there exists $R^*$ such that for all $R \ge R^*$, $\sup_{u \in \mathcal{B}} \|u\|_{L^2(\Omega \setminus B_R)}^2 < \epsilon/16$. By Boundary Tightness, there exists $\delta^*$ such that for all $\delta \le \delta^*$, $\sup_{u \in \mathcal{B}} \|u\|_{L^2(\Omega_\delta)}^2 < \epsilon/16$.

Select index $k$ such that $R_k \ge R^*$ and $\delta_k \le \delta^*$. Then
\begin{equation}
\int_{\Omega \setminus K_k} |w_{nm}|^2 \, d\mu \le \int_{\Omega \setminus B_{R_k}} |w_{nm}|^2 + \int_{\Omega_{\delta_k}} |w_{nm}|^2 \le 4\left(\frac{\epsilon}{16}\right) + 4\left(\frac{\epsilon}{16}\right) = \frac{\epsilon}{2}.
\end{equation}
Fixing this index $k$, since $(v_n)$ is Cauchy in $L^2(K_k)$, there exists an integer $N(\epsilon)$ such that for all $n, m \ge N$,
\begin{equation}
\int_{K_k} |v_n - v_m|^2 \, d\mu < \frac{\epsilon}{2}.
\end{equation}
Combining these estimates, for all $n, m \ge N$, we have $\|v_n - v_m\|_{L^2(\Omega)}^2 < \epsilon$. Thus, $(v_n)$ is a Cauchy sequence in $L^2(\Omega)$. Since $L^2(\Omega)$ is a complete Hilbert space, $(v_n)$ converges strongly to a limit $v \in L^2(\Omega)$. This proves sufficiency.

We now prove the necessity of the conditions. Assume that the embedding $\iota: H^1(\Omega) \hookrightarrow L^2(\Omega)$ is compact.

First, we show Local Compactness. Let $K \subset \overline{\Omega}$ be any compact set such that $\operatorname{dist}(K, \Gamma) > 0$, and let $(u_n) \subset H^1(K)$ be a bounded sequence. We may extend $u_n$ to $\tilde{u}_n \in H^1(\Omega)$. Since $(u_n)$ is bounded in $H^1(K)$, $(\tilde{u}_n)$ is bounded in $H^1(\Omega)$. By the compactness of $\iota$, $(\tilde{u}_n)$ has a convergent subsequence in $L^2(\Omega)$. The restriction of this subsequence to $K$ converges in $L^2(K)$. Thus, $\iota_K$ is compact.

Next, we show Global Tightness by contradiction. Assume Global Tightness fails. This can occur in two ways.
\begin{enumerate}
    \item Case A: Failure of Tail Tightness. Suppose there exists $\epsilon_0 > 0$ such that for every $k \in \mathbb{N}$, there exists $u_k \in \mathcal{B}$ satisfying
\begin{equation}
\int_{\Omega \setminus B_k} |u_k|^2 \, d\mu \ge \epsilon_0.
\end{equation}
Since $(u_k) \subset \mathcal{B}$ and $\iota$ is compact, there exists a subsequence $(u_{k_j})$ converging strongly to some $u \in L^2(\Omega)$. This implies $\|u_{k_j} - u\|_{L^2(\Omega)} \to 0$. We analyze the mass in the tail $E_j := \Omega \setminus B_{k_j}$. By the triangle inequality, $\|u_{k_j}\|_{L^2(E_j)} \le \|u_{k_j} - u\|_{L^2(E_j)} + \|u\|_{L^2(E_j)}$. The first term vanishes due to strong convergence, and the second term vanishes by the absolute continuity of the Lebesgue integral. Consequently, $\lim_{j \to \infty} \|u_{k_j}\|_{L^2(E_j)} = 0$, which contradicts the hypothesis.
    \item Case B: Failure of Boundary Tightness. Suppose there exists $\epsilon_0 > 0$ such that for every $k \in \mathbb{N}$, there exists $u_k \in \mathcal{B}$ satisfying
\begin{equation}
\int_{\Omega_{1/k}} |u_k|^2 \, d\mu \ge \epsilon_0,
\end{equation}
where $\Omega_{1/k} = \{z : \operatorname{dist}(z, \Gamma) < 1/k\}$. Proceeding identically to Case A, we extract a strongly convergent subsequence $u_{k_j} \to u$. Let $S_j = \Omega_{1/k_j}$. As $j \to \infty$, the measure of the set $S_j$ approaches 0. By the triangle inequality, $\|u_{k_j}\|_{L^2(S_j)} \le \|u_{k_j} - u\|_{L^2(S_j)} + \|u\|_{L^2(S_j)}$. The first term vanishes due to global strong convergence, and the second term vanishes because the integral of a fixed $L^2$ function over a contracting set vanishes. Thus, $\lim_{j \to \infty} \|u_{k_j}\|_{L^2(S_j)} = 0$, which contradicts the hypothesis.
\end{enumerate}

Therefore, both Local Compactness and Global Tightness must hold.
\end{proof}

\begin{remark}
Theorem~\ref{5.1} provides a general functional analytic framework for compactness in weighted spaces, separating the problem into "Local Compactness" and "Global Tightness". This approach abstracts the specific Gaussian arguments used in \cite{NegroSpina2025}, where compactness was derived via explicit spectral decomposition or specific density estimates. Here, we rely on the Fréchet-Kolmogorov criterion, which allows us to treat a broader class of weights by focusing on the loss of compactness at the "singular boundary" and "infinity" independently .
\end{remark}

\begin{proposition}[Density away from $y=0$]\label{prop:5.2}
If $c > -1$, then functions in $C_{c}^{\infty}(\overline{\mathbb{H}^{N+1}})$ that vanish near $y = 0$ are dense in $H^{1}_{\mu_{w}}$.

If $c \le -1$, then the same class is dense in 
\begin{equation}
\bigl\{\, u \in H^{1}_{\mu_{w}} : u|_{y = 0} = 0 \ \text{(trace)} \,\bigr\}.
\end{equation}
\end{proposition}

\begin{proof}
Let $d\mu_{w} = w\, dz$, with $w(y, x) = y^{c}\, \phi(|z|)$ and $z = (y, x)$, $y > 0$, under Assumption~\ref{ass:2} (local two–sided bounds).

\medskip
\textbf{Step 1 (Approximation goal).}
Given $u$ in the indicated space, we seek a sequence 
$u_{n} \in C_{c}^{\infty}(\overline{\mathbb{H}^{N+1}})$, vanishing near $y = 0$, such that
\begin{equation}
\|u_{n} - u\|_{H^{1}_{\mu_{w}}} \to 0.
\end{equation}

\medskip
\textbf{Step 2 (Horizontal compact cut–offs).}
Fix $R > 1$ and choose $\zeta_{R} \in C_{c}^{\infty}(\mathbb{R}^{N+1})$, radial in $z$, satisfying
\begin{equation}
0 \le \zeta_{R} \le 1,
\qquad
\zeta_{R} \equiv 1 \text{ on } B_{R},
\qquad
\zeta_{R} \equiv 0 \text{ on } \mathbb{R}^{N+1} \setminus B_{2R},
\qquad
|\nabla \zeta_{R}| \lesssim R^{-1}.
\end{equation}

\medskip
\textbf{Step 3 (Vertical cut–offs near $y=0$).}
For $\varepsilon \in (0, 1)$, select $\eta_{\varepsilon} \in C^{\infty}([0, \infty))$ such that
\begin{equation}
0 \le \eta_{\varepsilon} \le 1,
\qquad
\eta_{\varepsilon} \equiv 0 \text{ on } [0, \varepsilon],
\qquad
\eta_{\varepsilon} \equiv 1 \text{ on } [2\varepsilon, \infty),
\qquad
|\eta'_{\varepsilon}| \lesssim \varepsilon^{-1}.
\end{equation}

\medskip
\textbf{Step 4 (Three–stage regularization).}
Define
\begin{equation}
u^{(R)} := \zeta_{R}\, u,
\qquad
u^{(R, \varepsilon)} := \eta_{\varepsilon}\, u^{(R)}.
\end{equation}
Then $u^{(R, \varepsilon)} \equiv 0$ on $\{\, 0 < y \le \varepsilon \,\}$.

\medskip
\textbf{Step 5 (Local mollification on boxes).}
Cover the region
\begin{equation}
\{\, (y, x) : \varepsilon \le y \le 2R,\ |z| \le 2R \,\}
\end{equation}
by finitely many boxes $\Omega_{\alpha, \beta, r, R}$ of Lemma~\ref{lem:3.1}, on each of which $w \asymp 1$.  
Convolving $u^{(R, \varepsilon)}$ with a standard symmetric kernel supported in these boxes and patching with a partition of unity yields
\begin{equation}
u^{(R, \varepsilon)}_{\delta} \in C_{c}^{\infty}(\overline{\mathbb{H}^{N+1}}),
\qquad
u^{(R, \varepsilon)}_{\delta} \equiv 0 \text{ on } \{\, y \le \tfrac{\varepsilon}{2} \,\},
\end{equation}
with
\begin{equation}
\|u^{(R, \varepsilon)}_{\delta} - u^{(R, \varepsilon)}\|_{H^{1}_{\mu_{w}}}
\xrightarrow[\delta \downarrow 0]{} 0.
\end{equation}

\medskip
\textbf{Step 6 (Compactness in $R$: tail truncation).}
Since $u \in H^{1}_{\mu_{w}}$ and $\mu_{w}$ is finite on bounded sets, the absolute continuity of the weighted integral gives
\begin{equation}
\|u^{(R)} - u\|_{H^{1}_{\mu_{w}}}^{2}
=
\int_{\{|z| > R\}} 
\big( |u|^{2} + |\nabla u|^{2} \big)\, d\mu_{w}
\xrightarrow[R \to \infty]{} 0.
\end{equation}

\medskip
\textbf{Step 7 (Near–boundary truncation for $c > -1$).}
Fix $R$. We show that
\begin{equation}
\|u^{(R, \varepsilon)} - u^{(R)}\|_{H^{1}_{\mu_{w}}}
\xrightarrow[\varepsilon \downarrow 0]{} 0.
\end{equation}
Note that
\begin{equation}
u^{(R)} - u^{(R, \varepsilon)}
= (1 - \eta_{\varepsilon})\, u^{(R)},
\end{equation}
supported in $\{\, 0 < y < 2\varepsilon \,\}$.

\smallskip
\emph{(a) $L^{2}$ part:}
\begin{equation}
\|(1 - \eta_{\varepsilon})\, u^{(R)}\|^{2}_{L^{2}_{\mu_{w}}}
=
\int_{0 < y < 2\varepsilon} |u^{(R)}|^{2}\, d\mu_{w}
\xrightarrow[\varepsilon \downarrow 0]{} 0,
\end{equation}
by absolute continuity, since $c > -1$.

\smallskip
\emph{(b) Gradient part:}
\begin{equation}
\nabla\!\big((1 - \eta_{\varepsilon})\, u^{(R)}\big)
=
(1 - \eta_{\varepsilon})\, \nabla u^{(R)} 
- u^{(R)}\, \eta'_{\varepsilon}\, e_{y}.
\end{equation}
Therefore,
\begin{equation}
\int |(1 - \eta_{\varepsilon})\, \nabla u^{(R)}|^{2}\, d\mu_{w}
\to 0
\quad (\varepsilon \downarrow 0),
\end{equation}
again by absolute continuity.

\medskip
\textbf{Step 8 (Singular cross–term controlled by Hardy inequality).}
We estimate
\begin{equation}
\int_{\mathbb{H}^{N+1}} |u^{(R)}\, \eta'_{\varepsilon}|^{2}\, y^{c}\, dz
\lesssim
\frac{1}{\varepsilon^{2}}
\int_{\{\varepsilon < y < 2\varepsilon\}}
|u^{(R)}|^{2}\, y^{c}\, dz
\le
4
\int_{\{0 < y < 2\varepsilon\}}
\frac{|u^{(R)}|^{2}}{y^{2}}\, y^{c}\, dz.
\end{equation}
Since $y \ge \varepsilon$ implies $\varepsilon^{-2} \le 4 y^{-2}$, we apply Lemma~\ref{lem:3.2} slice–wise to
\begin{equation}
v(y, x)
:=
(1 - \chi_{\varepsilon}(y))\, u^{(R)}(y, x),
\end{equation}
where $\chi_{\varepsilon} \in C_{c}^{\infty}([0, 2\varepsilon))$ satisfies $\chi_{\varepsilon} \equiv 1$ on $[0, \varepsilon]$, so $v \equiv 0$ near $y = 0$.  
Then $v = u^{(R)}$ on $\{\varepsilon < y < 2\varepsilon\}$, and Lemma~\ref{lem:3.2} gives
\begin{equation}
\int_{0 < y < 2\varepsilon}
\frac{|v|^{2}}{y^{2}}\, y^{c}\, dz
\le
\frac{4}{(c + 1)^{2}}
\int_{0 < y < 2\varepsilon}
|\partial_{y} v|^{2}\, y^{c}\, dz.
\end{equation}
Since $\partial_{y} v = \partial_{y} u^{(R)}$ on $\{\varepsilon < y < 2\varepsilon\}$ and $\operatorname{supp}(\partial_{y} \chi_{\varepsilon}) \subset \{0 < y < \varepsilon\}$, it follows that
\begin{equation}
\int_{\{0 < y < 2\varepsilon\}}
\frac{|u^{(R)}|^{2}}{y^{2}}\, y^{c}\, dz
\le
C
\int_{\{0 < y < 2\varepsilon\}}
\big(|\partial_{y} u^{(R)}|^{2} + |u^{(R)}|^{2}\big)\, y^{c}\, dz
\xrightarrow[\varepsilon \downarrow 0]{} 0.
\end{equation}
Thus
\begin{equation}
\int |u^{(R)}\, \eta'_{\varepsilon}|^{2}\, y^{c}\, dz \to 0.
\end{equation}

\medskip
\textbf{Step 9 (Conclusion for $c > -1$).}
Hence
\begin{equation}
\|u^{(R, \varepsilon)} - u^{(R)}\|_{H^{1}_{\mu_{w}}}
\xrightarrow[\varepsilon \downarrow 0]{} 0
\quad
\text{for each fixed $R$,}
\end{equation}
and
\begin{equation}
\|u^{(R)} - u\|_{H^{1}_{\mu_{w}}}
\xrightarrow[R \to \infty]{} 0.
\end{equation}
Choosing a diagonal sequence $(\varepsilon_{k}, R_{k})$ and $\delta_{k} \downarrow 0$ as in Step~5 gives
\begin{equation}
u_{k} := (u^{(R_{k}, \varepsilon_{k})})_{\delta_{k}} \in C_{c}^{\infty},
\qquad
u_{k} \equiv 0 \text{ near } y = 0,
\end{equation}
with
\begin{equation}
\|u_{k} - u\|_{H^{1}_{\mu_{w}}} \to 0.
\end{equation}

\medskip
\textbf{Step 10 (Case $c \le -1$ with zero trace).}
Let $u \in H^{1}_{\mu_{w}}$ satisfy $u|_{y = 0} = 0$ in the trace sense.  
Repeating Steps~2–5 verbatim, the $L^{2}$ and gradient parts vanish as $\varepsilon \downarrow 0$ by absolute continuity.  
In Step~8, the Hardy inequality (Lemma~\ref{lem:3.2}) applies directly to $u^{(R)}$, since $u$ already vanishes at $y = 0$ in the trace sense:
\begin{equation}
\int_{0 < y < 2\varepsilon}
\frac{|u^{(R)}|^{2}}{y^{2}}\, y^{c}\, dz
\xrightarrow[\varepsilon \downarrow 0]{} 0.
\end{equation}
Therefore
\begin{equation}
\|u^{(R, \varepsilon)} - u^{(R)}\|_{H^{1}_{\mu_{w}}} \to 0,
\end{equation}
and the same diagonal argument yields the desired density result.
\end{proof}

\begin{remark}
Proposition~\ref{prop:5.2} establishes the density of smooth functions with compact support, $C_c^{\infty}\left(\overline{\mathbb{H}^{N+1}}\right)$, in the weighted space $H_{\mu_w}^1$. This result is standard for weights that are locally integrable ( $c>-1$ ). However, in the singular range $c \leq-1$, the density result is more subtle; we show that smooth functions vanishing at $y=0$ are dense in the subspace of $H_{\mu_w}^1$ with zero trace. This density is crucial for justifying the application of our functional inequalities (derived for smooth test functions) to the entire Sobolev space.
\end{remark}

\section{Necessary and Sufficiency Conditions}

In this section, we combine the tools developed thus far to establish the sufficiency and necessity of our structural conditions for compactness. We prove that if the measure has finite mass, satisfies the Tail Coercivity condition, and (in the singular case $c \leq-1$ ) admits a weighted Hardy inequality, then the embedding is compact. Conversely, we demonstrate that the finite mass condition is necessary; if the total mass is infinite, one can construct a sequence of disjoint translates that converges weakly to zero but maintains a non-vanishing $L^2$ norm, thereby violating compactness. We further show that the failure of Tail Coercivity or the Hardy condition leads to a loss of global tightness, either through mass escape to infinity or concentration at the singularity.

\begin{proposition}[Sufficiency: Finite Mass and Tail Coercivity]\label{prop:6.1}
Let $(\mathbb{H}^{N+1}, \mu_w)$ be the weighted measure space defined in Assumptions 1 and 2. Assume the following conditions hold:
\begin{enumerate}
    \item \textbf{Finite Measure:} $\mu_w(\mathbb{H}^{N+1}) < \infty$.
    \item \textbf{Tail Coercivity (TC):} There exists a Lyapunov function $\Psi: [0, \infty) \to [0, \infty)$ with $\Psi(\rho) \to \infty$ as $\rho \to \infty$ such that for all admissible $u$,
    \begin{equation}
    \int_{\mathbb{H}^{N+1}} \Psi(|z|) u^2 \, d\mu_w \le C_{TC} \|u\|_{H_{\mu_w}^1}^2.
    \end{equation}
    \item \textbf{Hardy Condition:} If $c \le -1$, the weight satisfies the Weighted Hardy Inequality (Lemma~\ref{lem:3.2}).
\end{enumerate}
Then the embedding $H_{\mu_w}^1(\mathbb{H}^{N+1}) \hookrightarrow L_{\mu_w}^2(\mathbb{H}^{N+1})$ is compact.
\end{proposition}

\begin{proof}
Let $(u_n)_{n \in \mathbb{N}}$ be a bounded sequence in the Sobolev space $H_{\mu_w}^1(\mathbb{H}^{N+1})$. By definition, there exists a constant $K > 0$ such that
\begin{equation}
\sup_{n \in \mathbb{N}} \|u_n\|_{H_{\mu_w}^1} \le K.
\end{equation}
To prove compactness, we must show that $(u_n)$ contains a subsequence that converges strongly in $L_{\mu_w}^2(\mathbb{H}^{N+1})$.

We decompose the ambient space $\mathbb{H}^{N+1}$ into three disjoint regions: a compact core, a boundary strip (where the weight may be singular), and a tail region. Fix parameters $R > 1$ (a large radius) and $\delta \in (0, 1)$ (a boundary width). We define the Tail Region as $E_{tail}(R) := \{z \in \mathbb{H}^{N+1} : |z| > R\}$, the Boundary Strip as $E_{bdry}(R, \delta) := \{z \in \mathbb{H}^{N+1} : |z| \le R \text{ and } 0 < y < \delta\}$, and the Compact Core as $K_{R, \delta} := \{z \in \mathbb{H}^{N+1} : |z| \le R \text{ and } y \ge \delta\}$. We analyze the $L^2$ mass of the sequence on each region separately.

First, we establish uniform control of the tail. We invoke the Tail Coercivity condition. Let $m_R := \inf_{|z| \ge R} \Psi(|z|)$. Since $\Psi(|z|) \to \infty$ as $|z| \to \infty$, we have $m_R \to \infty$ as $R \to \infty$. For any $u_n$ in the sequence,
\begin{equation}
\int_{E_{tail}(R)} u_n^2 \, d\mu_w \le \frac{1}{m_R} \int_{E_{tail}(R)} \Psi(|z|) u_n^2 \, d\mu_w \le \frac{1}{m_R} \int_{\mathbb{H}^{N+1}} \Psi(|z|) u_n^2 \, d\mu_w.
\end{equation}
By the Tail Coercivity inequality and the uniform boundedness of $(u_n)$,
\begin{equation}
\int_{E_{tail}(R)} u_n^2 \, d\mu_w \le \frac{C_{TC}}{m_R} \|u_n\|_{H_{\mu_w}^1}^2 \le \frac{C_{TC} K^2}{m_R}.
\end{equation}
Thus, the mass in the tail vanishes uniformly as $R \to \infty$:
\begin{equation}
\lim_{R \to \infty} \sup_{n} \|u_n\|_{L^2(E_{tail}(R))} = 0.
\end{equation}

Next, we establish uniform control of the boundary strip $E_{bdry}(R, \delta)$. We distinguish two cases based on the weight parameter $c$.

\begin{enumerate}
    \item Case A ($c > -1$): The weight $w(z) = y^c \phi(|z|)$ is locally integrable ($L_{loc}^1$). Since $\mu_w$ is absolutely continuous with respect to the Lebesgue measure and $\mu_w(E_{bdry}(R, \delta)) \to 0$ as $\delta \to 0$, the standard absolute continuity of the integral implies
\begin{equation}
\lim_{\delta \to 0} \sup_{n} \int_{E_{bdry}(R, \delta)} u_n^2 \, d\mu_w = 0.
\end{equation}
This holds uniformly for bounded $H^1$ functions due to Sobolev embedding theorems on bounded domains.
    \item Case B ($c \le -1$): The weight is singular at $y=0$. We invoke the Weighted Hardy Inequality (Lemma\ref{lem:3.2}). Since $u_n \in H_{\mu_w}^1$ implies $u_n(\cdot, 0) = 0$ in the trace sense for singular weights,
\begin{equation}
\int_{\mathbb{H}^{N+1}} \frac{u_n^2}{y^2} \, d\mu_w \le C_{Hardy} \int_{\mathbb{H}^{N+1}} |\partial_y u_n|^2 \, d\mu_w \le C_{Hardy} K^2.
\end{equation}
We estimate the mass in the strip $0 < y < \delta$ as follows:
\begin{equation}
\int_{E_{bdry}(R, \delta)} u_n^2 \, d\mu_w = \int_{E_{bdry}(R, \delta)} \left(\frac{u_n}{y}\right)^2 y^2 \, d\mu_w \le \delta^2 \int_{\mathbb{H}^{N+1}} \frac{u_n^2}{y^2} \, d\mu_w.
\end{equation}
Combining this with the Hardy bound yields
\begin{equation}
\int_{E_{bdry}(R, \delta)} u_n^2 \, d\mu_w \le \delta^2 C_{Hardy} K^2.
\end{equation}
Thus, the mass near the boundary vanishes uniformly as $\delta \to 0$.
\end{enumerate}

Now we address compactness on the core. Fix $R$ and $\delta$. The set $K_{R, \delta}$ is a closed, bounded subset of $\mathbb{H}^{N+1}$ strictly separated from the singularity $y=0$. On this set, the weight $w(z)$ is bounded from below and above by positive constants (Assumption~\ref{ass:2}). Thus, the weighted Sobolev norm is equivalent to the standard Sobolev norm on $K_{R, \delta}$. By the classical Rellich-Kondrachov Theorem, the embedding $H^1(K_{R, \delta}) \hookrightarrow L^2(K_{R, \delta})$ is compact. Consequently, for any fixed core, we can extract a subsequence converging strongly in $L^2(K_{R, \delta})$.

Finally, we employ a standard Cantor diagonalization argument. Let $\epsilon_k = 1/k$. Choose $R_k$ large enough and $\delta_k$ small enough such that the total mass of every $u_n$ outside the core $K_k := K_{R_k, \delta_k}$ is less than $1/k$. Extract a diagonal subsequence $(v_j)$ such that $(v_j)$ converges in $L^2(K_k)$ for every $k$.

We verify the global Cauchy condition. For any $\epsilon > 0$, choose $k$ such that the mass outside $K_k$ is strictly less than $\epsilon/4$. Since $(v_j)$ is Cauchy on $K_k$, for sufficiently large $j$ and $m$,
\begin{equation}
\|v_j - v_m\|_{L^2(\mathbb{H}^{N+1})}^2 = \|v_j - v_m\|_{L^2(K_k)}^2 + \|v_j - v_m\|_{L^2(K_k^c)}^2.
\end{equation}
The first term is less than $\epsilon/2$ by convergence on the core. The second term is bounded by $2(\|v_j\|_{L^2(K_k^c)}^2 + \|v_m\|_{L^2(K_k^c)}^2) < 4(\epsilon/8) = \epsilon/2$. The total difference is less than $\epsilon$.

Thus, the sequence $(v_j)$ is Cauchy in the complete space $L_{\mu_w}^2(\mathbb{H}^{N+1})$ and therefore converges strongly. The embedding is compact.
\end{proof}

\begin{remark}
Proposition 6.1 identifies sufficient conditions for compactness: finite mass, tail coercivity (TC), and the Hardy property (HP). The "Finite Mass" condition replaces the specific exponential decay of the Gaussian weight, while (TC) ensures that mass does not escape to infinity. This generalizes the results of \cite{NegroSpina2025}, confirming that the exponential rate is not strictly necessary as long as the weight admits a coercive Lyapunov potential.
\end{remark}

\begin{proposition}[Necessity of Finite Mass]\label{prop:6.2}
Let $(\mathbb{H}^{N+1}, \mu_w)$ be the weighted measure space defined in Assumption~\ref{ass:1} and \ref{ass:2}. If the total mass of the weight is infinite, i.e.,
\begin{equation}
\mathbf{M} := \int_{\mathbb{H}^{N+1}} \phi(|z|)\, dz = \infty,
\end{equation}
then the embedding
\begin{equation}
H^{1}_{\mu_{w}}(\mathbb{H}^{N+1}) \hookrightarrow L^{2}_{\mu_{w}}(\mathbb{H}^{N+1})
\end{equation}
is not compact.
\end{proposition}

\begin{proof}
We proceed by constructing a bounded sequence in $H_{\mu_w}^1$ that converges weakly to zero but does not converge strongly in $L_{\mu_w}^2$.

Since the domain $\mathbb{H}^{N+1}$ is unbounded, we select a sequence of points $\{z_j\}_{j \in \mathbb{N}} \subset \mathbb{H}^{N+1}$ such that $|z_j| \to \infty$ as $j \to \infty$. We construct a corresponding sequence of disjoint balls $B_j := B(z_j, 1) \subset \mathbb{H}^{N+1}$.

Let $\psi \in C_c^\infty(B(0,1))$ be a fixed non-zero smooth function supported in the unit ball centered at the origin. We define the sequence of functions $\{U_j\}_{j \in \mathbb{N}}$ by translation:
\begin{equation}
U_j(z) := \psi(z - z_j).
\end{equation}
Each $U_j$ is supported in $B_j$. The sequence $\{U_j\}$ is not necessarily bounded in $H_{\mu_w}^1$ due to the potential growth of the weight $w(z)$ at infinity. To address this, we define the normalized sequence $\{v_j\}_{j \in \mathbb{N}}$ as
\begin{equation}
v_j(z) := \frac{U_j(z)}{\|U_j\|_{H_{\mu_w}^1}}.
\end{equation}
By construction, $\|v_j\|_{H_{\mu_w}^1} = 1$ for all $j \in \mathbb{N}$. Thus, $\{v_j\}$ constitutes a bounded sequence in the unit ball of $H_{\mu_w}^1$.

We now examine the $L^2$-norm of the normalized sequence. The squared norm is given by the ratio of energies:
\begin{equation}
\|v_j\|_{L_{\mu_w}^2}^2 = \frac{\|U_j\|_{L_{\mu_w}^2}^2}{\|U_j\|_{H_{\mu_w}^1}^2} = \frac{\int_{B_j} |U_j|^2 w(z) \, dz}{\int_{B_j} (|\nabla U_j|^2 + |U_j|^2) w(z) \, dz}.
\end{equation}
By Assumption~\ref{ass:2}, the weight $w(z)$ satisfies local comparability conditions. Specifically, on the compact ball $B_j$ of fixed radius 1, the weight $w(z)$ is comparable to its value at the center $z_j$. Since $w(z)$ appears as a multiplicative factor in both the numerator and the denominator, it effectively cancels out in the ratio. Let $\lambda_{flat}$ be the strictly positive ratio of the unweighted energies of the template function $\psi$:
\begin{equation}
\lambda_{flat} := \frac{\int_{B(0,1)} |\psi|^2 \, dx}{\int_{B(0,1)} (|\nabla \psi|^2 + |\psi|^2) \, dx}.
\end{equation}
It follows that there exists a constant $\lambda > 0$ independent of $j$ such that
\begin{equation}
\|v_j\|_{L_{\mu_w}^2}^2 \ge \lambda > 0.
\end{equation}
Thus, the sequence $\{v_j\}$ does not converge strongly to 0 in $L_{\mu_w}^2$.

However, we assert that $v_j \rightharpoonup 0$ weakly in $H_{\mu_w}^1$. Let $\varphi \in C_c^\infty(\mathbb{H}^{N+1})$ be an arbitrary test function with compact support. Since $|z_j| \to \infty$, for sufficiently large $j$, the support of $v_j$ (which is $B_j$) is disjoint from the support of $\varphi$. Consequently, the inner product $\langle v_j, \varphi \rangle_{H_{\mu_w}^1}$ vanishes for large $j$. Since $C_c^\infty$ is dense in $H_{\mu_w}^1$ and $\{v_j\}$ is bounded, this implies $v_j \rightharpoonup 0$ weakly.

If the embedding were compact, the weak convergence $v_j \rightharpoonup 0$ would imply strong convergence $\|v_j\|_{L_{\mu_w}^2} \to 0$. The lower bound $\|v_j\|_{L_{\mu_w}^2} \ge \sqrt{\lambda} > 0$ contradicts this. Therefore, the embedding is not compact.
\end{proof}

\begin{remark}
Proposition~\ref{prop:6.2} highlights the necessity of the finite mass condition. Unlike in bounded domains, where finite measure is often immediate, on the unbounded half-space $\mathbb{H}^{N+1}$, infinite mass allows for "vanishing" sequences of disjoint translates that maintain nonzero energy, thereby violating compactness. This counterexample demonstrates that the weight must decay sufficiently fast to make the total volume finite, a property automatically satisfied by the Gaussian weights in \cite{NegroSpina2025}.
\end{remark}

\section{Main Embedding Theorems}

Finally, we state and prove our main embedding theorems, which provide a complete characterization of the compactness of $H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right) \hookrightarrow L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right)$. We show that compactness is equivalent to the validity of the "Global Tightness" condition on the unit ball, which in turn is structurally equivalent to the finiteness of the measure and the specific decay properties of the weight. Specifically, for the singular case $c \leq-1$, we prove that compactness holds if and only if the measure is finite and the weighted Hardy inequality prevents mass concentration at the boundary. These results provide a unified framework for analyzing Rellich-Kondrachov type theorems in weighted spaces, covering both degenerate and singular weights on the half-space.

\begin{theorem}[Characterization of Compactness]\label{thm:7.1}
Let $\left(\mathbb{H}^{N+1}, \mu_w\right)$ be the weighted measure space satisfying Assumption~\ref{ass:2}. The embedding $H_{\mu_w}^1 \hookrightarrow L_{\mu_w}^2$ is compact if and only if the following two conditions hold:
\begin{enumerate}
    \item (FM) Finite Mass: $\mu_w\left(\mathbb{H}^{N+1}\right)<\infty$.
    \item Global Tightness: The unit ball $\mathcal{B}_{H^1}=\left\{u \in H_{\mu_w}^1:\|u\|_{H_{\mu_w}^1} \leq 1\right\}$ satisfies the Condition of Tightness (Definition~\ref{def:2.6}).
\end{enumerate}
\end{theorem}

\begin{proof}
We first prove the sufficiency of the conditions. Assume that conditions (FM), (TC), and Definition~\ref{def:2.5} hold. We employ the Fréchet-Kolmogorov Compactness Criterion adapted for weighted spaces. To prove compactness, it suffices to show that the unit ball $\mathcal{B} \subset H_{\mu_w}^1(\mathbb{H}^{N+1})$ is globally tight and locally compact.

We begin by examining the embedding on bounded subdomains strictly separated from the singularity $y=0$. Let $K \subset \mathbb{H}^{N+1}$ be any compact subset such that $\operatorname{dist}(K, \{y=0\}) > 0$. By Assumption~\ref{ass:2}, the radial component $\phi(|z|)$ is bounded below by a positive constant $m_K$ and above by $M_K$ on the compact set of radii associated with $K$. Similarly, the vertical component $y^c$ is bounded strictly away from zero and infinity on $K$ because $y$ is bounded away from $0$. Consequently, the weight $w(z) = y^c \phi(|z|)$ satisfies $0 < c_1 \le w(z) \le c_2 < \infty$ on $K$. This implies that the weighted norms are equivalent to the standard Lebesgue norms on $K$, meaning $H_{\mu_w}^1(K) \cong H^1(K)$. By the classical Rellich-Kondrachov Theorem, the embedding $H^1(K) \hookrightarrow L^2(K)$ is compact. Thus, the restriction of the weighted embedding to any such compact core $K$ is compact.

Next, we establish uniform decay in the tail region to prove tail tightness. Let $E_R := \{z \in \mathbb{H}^{N+1} : |z| > R\}$. By condition (TC) (Definition~\ref{def:2.4}), there exists a potential $\Psi(|z|)$ such that $\Psi(|z|) \to \infty$ as $|z| \to \infty$ and satisfying the inequality $\int \Psi u^2 d\mu_w \le C_{TC} \|u\|_{H_{\mu_w}^1}^2$. For any $u \in \mathcal{B}$ and any $R$ large enough, we have the estimate
\begin{equation}
\inf_{z \in E_R} \Psi(|z|) \int_{E_R} |u|^2 \, d\mu_w \le \int_{E_R} \Psi(|z|) |u|^2 \, d\mu_w \le C_{TC}.
\end{equation}
Letting $m_R := \inf_{|z| \ge R} \Psi(|z|)$, we obtain $\int_{E_R} |u|^2 \, d\mu_w \le \frac{C_{TC}}{m_R}$. Since $\Psi \to \infty$, it follows that $m_R \to \infty$ as $R \to \infty$. Thus, $\lim_{R \to \infty} \sup_{u \in \mathcal{B}} \|u\|_{L_{\mu_w}^2(E_R)} = 0$.

We must also show that the mass vanishes uniformly in the boundary strip $S_\delta := \{z \in \mathbb{H}^{N+1} : 0 < y < \delta\}$. We distinguish two cases. In the first case, where $c > -1$, the total measure is finite by (FM), and the weight $y^c$ is locally integrable near $y=0$. The measure $\mu_w$ is absolutely continuous with respect to the Lebesgue measure. Therefore, for any bounded set of functions (and specifically $\mathcal{B}$), the integral over a set of vanishing measure tends to 0. Thus, $\lim_{\delta \to 0} \sup_{u \in \mathcal{B}} \int_{S_\delta} u^2 d\mu_w = 0$. In the second case, where $c \le -1$, condition Definition~\ref{def:2.5} implies that the weighted Hardy inequality holds. For any $u \in \mathcal{B}$ (noting that $u$ vanishes at $y=0$ in the trace sense for singular weights), we have
\begin{equation}
\int_{S_\delta} |u|^2 \, d\mu_w = \int_{S_\delta} \left| \frac{u}{y} \right|^2 y^2 \, d\mu_w \le \delta^2 \int_{\mathbb{H}^{N+1}} \frac{|u|^2}{y^2} \, d\mu_w.
\end{equation}
Applying the Hardy Property, we obtain
\begin{equation}
\int_{S_\delta} |u|^2 \, d\mu_w \le \delta^2 C_H \|\nabla u\|_{L_{\mu_w}^2}^2 \le \delta^2 C_H.
\end{equation}
Thus, the boundary mass decays as $O(\delta^2)$. Since the unit ball is locally compact and tight at both infinity and the singular boundary, the embedding is compact.

We now prove the necessity of the conditions. Assume the embedding $\iota: H_{\mu_w}^1 \hookrightarrow L_{\mu_w}^2$ is compact. First, suppose the finite mass condition (FM) fails, so that $M = \infty$. We construct a counterexample using a sequence of disjoint unit balls $B_n \subset \mathbb{H}^{N+1}$ such that $|z| \to \infty$. Let $u_n$ be a normalized bump function supported in $B_n$, specifically $u_n(z) = \psi_n(z) / \|\psi_n\|_{H_{\mu_w}^1}$, where $\psi_n$ is a translate of a fixed template. By the local comparability of the weight, the ratio of the $L^2$ norm to the $H^1$ norm remains bounded away from zero, so $\|u_n\|_{L_{\mu_w}^2} \ge \lambda > 0$. The sequence $\{u_n\}$ converges weakly to 0 as the disjoint supports escape to infinity, but it does not converge strongly to 0 because the norms are bounded below. This contradicts compactness; thus, (FM) is necessary.

Next, suppose the tail coercivity condition (TC) fails. Since compactness implies global tightness, the tail modulus $\epsilon(R) := \sup_{u \in \mathcal{B}} \int_{|z|>R} u^2 d\mu_w$ must vanish as $R \to \infty$. We can explicitly construct the required Lyapunov function $\Psi$ from the decay rate of $\epsilon(R)$. Setting $\Psi(z) \approx \sum 2^k \mathbb{1}_{|z| \ge R_k}$ yields a valid potential satisfying the inequality. Thus, (TC) is intrinsic to the compactness of the operator.

Finally, suppose $c \le -1$ and compactness holds. Compactness implies that mass must not concentrate near the singularity $y=0$, specifically $\lim_{\delta \to 0} \sup_{u \in \mathcal{B}} \int_{S_\delta} u^2 d\mu_w = 0$. The condition that $u \in H_{\mu_w}^1$ forces the function to vanish at $y=0$ at a specific rate to maintain finite energy against the singular weight $y^c$. This structural vanishing is analytically equivalent to the Hardy inequality. If Definition~\ref{def:2.5} failed, one could construct a "spike" sequence concentrating at $y=0$ with bounded energy but non-vanishing $L^2$ mass, violating compactness.
\end{proof}

\begin{remark}
Theorem~\ref{thm:7.1} provides the complete characterization of compactness for the embedding $H_{\mu_w}^1 \hookrightarrow L_{\mu_w}^2$. It unifies the treatment of degenerate ( $c>-1$ ) and singular ( $c \leq -1)$ weights. In the singular case, the theorem explicitly links compactness to the validity of the Weighted Hardy Inequality, which prevents mass concentration at the boundary. This structural insight complements the results in \cite{NegroSpina2025}, where Hardy-type inequalities were used as tools rather than necessary conditions for compactness itself.
\end{remark}

\begin{theorem}\label{thm:7.2}
Let $w$ satisfy the local assumptions (Assumption~\ref{ass:2}). The embedding 
\begin{equation}
H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right) \hookrightarrow L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right) 
\end{equation} is compact if and only if the pair $\left(\mathbb{H}^{N+1}, \mu_w\right)$ satisfies the Condition of Tightness (Definition~\ref{def:2.5}).
\end{theorem}

\begin{proof}
\textbf{Part I: Sufficiency of the Condition of Tightness}

We demonstrate that Tightness, combined with local compactness, implies global strong convergence.

\textbf{Step 1: Assumption of Tightness and Boundedness.}
Assume the measure space $(\mathbb{H}^{N+1}, \mu_w)$ satisfies the Condition of Tightness. Let $\{u_n\}_{n=1}^\infty$ be an arbitrary sequence in $H_{\mu_w}^1(\mathbb{H}^{N+1})$ that is bounded. That is, there exists a constant $M > 0$ such that $\|u_n\|_{H_{\mu_w}^1} \le M$ for all $n$.

\textbf{Step 2: Local Compactness Extraction.}
Consider the compact subsets $K_k := \{z \in \mathbb{H}^{N+1} : |z| \le k, \frac{1}{k} \le y \le k\}$ for $k \in \mathbb{N}$. By Lemma~\ref{lem:3.1}, the embedding $H^1(K_k) \hookrightarrow L^2(K_k)$ is compact for the weighted measure (since the weight is locally strictly positive and bounded).
Applying a standard Cantor diagonal argument, we extract a subsequence $\{u_{n_j}\}$ that converges strongly in $L_{\mu_w}^2(K_k)$ for every $k \in \mathbb{N}$. Let this subsequence be denoted $\{v_j\}$. Thus, $\{v_j\}$ is a Cauchy sequence in $L_{\mu_w}^2(K)$ for any compact $K \subset \mathbb{H}^{N+1}$.

\textbf{Step 3: Global Cauchy Criterion.}
We demonstrate that $\{v_j\}$ is a Cauchy sequence in the global space $L_{\mu_w}^2(\mathbb{H}^{N+1})$. Let $\epsilon > 0$ be given.
By the Condition of Tightness (Definition~\ref{def:2.5}), there exists a radius $R_\epsilon > 0$ sufficiently large such that for all functions $v$ in the bounded sequence (scaled by $1/M$ to fit the unit ball definition), the mass in the tail region $E_{out} := \{z : |z| > R_\epsilon\}$ satisfies:
\begin{equation} \sup_{j} \int_{E_{out}} |v_j(z)|^2 \, d\mu_w(z) < \frac{\epsilon}{4} \end{equation}

\textbf{Step 4: Decomposition of the Norm.}
For any indices $j, m$, we decompose the squared $L^2$ distance:
\begin{equation} \|v_j - v_m\|_{L_{\mu_w}^2}^2 = \int_{|z| \le R_\epsilon} |v_j - v_m|^2 \, d\mu_w + \int_{|z| > R_\epsilon} |v_j - v_m|^2 \, d\mu_w \end{equation}
Using the elementary inequality $|a - b|^2 \le 2(|a|^2 + |b|^2)$, the tail term is bounded by:
\begin{equation} \int_{|z| > R_\epsilon} |v_j - v_m|^2 \, d\mu_w \le 2 \int_{|z| > R_\epsilon} |v_j|^2 \, d\mu_w + 2 \int_{|z| > R_\epsilon} |v_m|^2 \, d\mu_w \end{equation}
Substituting the bound from Step 3:
\begin{equation} \int_{|z| > R_\epsilon} |v_j - v_m|^2 \, d\mu_w < 2\left(\frac{\epsilon}{4}\right) + 2\left(\frac{\epsilon}{4}\right) = \epsilon \end{equation}

\textbf{Step 5: Convergence on the Bounded Core.}
Since $\{v_j\}$ is Cauchy on compact sets (Step 2), there exists an integer $N_\epsilon$ such that for all $j, m \ge N_\epsilon$:
\begin{equation} \int_{|z| \le R_\epsilon} |v_j - v_m|^2 \, d\mu_w < \epsilon \end{equation}
Summing the core and tail estimates, for $j, m \ge N_\epsilon$, $\|v_j - v_m\|_{L_{\mu_w}^2}^2 < 2\epsilon$. Thus, $\{v_j\}$ is a Cauchy sequence in the complete space $L_{\mu_w}^2(\mathbb{H}^{N+1})$ and converges strongly.
This proves sufficiency.

\textbf{Part II: Necessity of the Condition of Tightness}

We prove by contradiction that compactness implies uniform tail decay.

\textbf{Step 6: Assumption of Compactness and Negation of Tightness.}
Assume the embedding $H_{\mu_w}^1 \hookrightarrow L_{\mu_w}^2$ is compact. Suppose, for the sake of contradiction, that the Condition of Tightness does not hold.
The negation of the Tightness definition implies:
\begin{equation} \exists \epsilon_0 > 0, \quad \forall R > 0, \quad \exists u \in \mathcal{B} \quad \text{such that} \quad \int_{\{|z|>R\}} |u|^2 \, d\mu_w \ge \epsilon_0 \end{equation}

\textbf{Step 7: Construction of the Singular Sequence.}
Using the negation above, we construct a sequence $\{u_k\}_{k=1}^\infty \subset \mathcal{B}$ (the unit ball of $H_{\mu_w}^1$) and a sequence of radii $R_k \to \infty$ such that for each $k$:
\begin{equation} \int_{\{|z|>R_k\}} |u_k(z)|^2 \, d\mu_w(z) \ge \epsilon_0 \end{equation}

\textbf{Step 8: Extraction of a Convergent Subsequence.}
Since $\{u_k\}$ is bounded in $H_{\mu_w}^1$ and the embedding is assumed compact, there exists a subsequence $\{u_{k_j}\}$ and a limit function $u \in L_{\mu_w}^2(\mathbb{H}^{N+1})$ such that $u_{k_j} \to u$ strongly in $L_{\mu_w}^2$.

\textbf{Step 9: The Contradiction via Triangle Inequality.}
We examine the tail mass of the convergent subsequence. By the triangle inequality for the $L^2$ norm restricted to the set $E_{j} := \{|z| > R_{k_j}\}$:
\begin{equation} \|u_{k_j}\|_{L^2(E_j)} \le \|u_{k_j} - u\|_{L^2(E_j)} + \|u\|_{L^2(E_j)} \end{equation}
From Step 7, we have the lower bound $\|u_{k_j}\|_{L^2(E_j)} \ge \sqrt{\epsilon_0}$. Thus:
\begin{equation} \sqrt{\epsilon_0} \le \|u_{k_j} - u\|_{L^2(\mathbb{H}^{N+1})} + \|u\|_{L^2(E_j)} \end{equation}

\textbf{Step 10: Limit Analysis.}
As $j \to \infty$:
\begin{enumerate}
    \item The term $\|u_{k_j} - u\|_{L^2(\mathbb{H}^{N+1})} \to 0$ by the strong convergence established in Step 8.
    \item The term $\|u\|_{L^2(E_j)} \to 0$ because $u \in L_{\mu_w}^2$ implies $\int |u|^2 < \infty$, and the sets $E_j$ contract to the empty set as $R_{k_j} \to \infty$ (Absolute Continuity of the Lebesgue integral).
\end{enumerate}

\textbf{Step 11: Final Contradiction.}
Taking the limit $j \to \infty$ in the inequality from Step 9 yields:
\begin{equation} \sqrt{\epsilon_0} \le 0 + 0 \end{equation}
This contradicts the hypothesis that $\epsilon_0 > 0$. Therefore, the Condition of Tightness must hold.
\end{proof}

\begin{remark}
Theorem~\ref{thm:7.2} reformulates the compactness result purely in terms of "Global Tightness". This is the most abstract and robust version of the Rellich-Kondrachov theorem for this setting. It asserts that compactness is equivalent to the uniform vanishing of mass at the "ends" of the space (infinity and the singularity). This characterization is powerful because it allows one to verify compactness for new weights simply by checking these mass-vanishing conditions, without needing to reconstruct the entire spectral theory as was done for the specific Gaussian case in \cite{NegroSpina2025}.
\end{remark}

\section{Conclusion}

In this paper, we have established a complete characterization of the compactness of the embedding $H_{\mu_w}^1\left(\mathbb{H}^{N+1}\right) \hookrightarrow L_{\mu_w}^2\left(\mathbb{H}^{N+1}\right)$ for the general class of weights $w(z)=y^c \phi(|z|)$. We proved that the compactness of the embedding is structurally equivalent to the validity of the "Global Tightness" condition on the unit ball, which ensures that mass does not escape to infinity or concentrate at the singular boundary.

Specifically, we demonstrated that compactness holds if and only if the measure has finite mass and satisfies two essential mass-vanishing conditions: "Tail Tightness," which is controlled by the coercive Lyapunov condition (TC) on the radial component $\phi(|z|)$, and "Boundary Tightness," which, in the singular case $c \leq-1$, requires the validity of a weighted Hardy inequality. This result generalizes the recent findings in \cite{NegroSpina2025} for Gaussian weights to a much broader framework, showing that the specific exponential decay is not required as long as a suitable Lyapunov potential exists to govern the behavior at infinity.

Our approach, based on the Fréchet-Kolmogorov Compactness Criterion adapted to weighted spaces, provides a unified functional analytic tool for studying degenerate and singular problems in the half-space. These compactness results and the associated Poincaré inequalities are of independent interest and serve as a fundamental step for future investigations into the spectral properties of degenerate elliptic operators of the form $\mathcal{L}= \operatorname{div}(w \nabla u)$, as well as for deriving heat kernel estimates and regularity results for non-local operators related to the fractional Laplacian extension problem.

\begin{thebibliography}{9}
\bibitem{ArendtElstWarma2018}
W.~Arendt, A.~F.~M.~Ter~Elst, and M.~Warma,
``Fractional powers of sectorial operators via the Dirichlet-to-Neumann operator,''
\textit{Communications in Partial Differential Equations}, 
vol.~43, no.~1, pp.~1--24, 2018.

\bibitem{AudritoTerracini2024}
A.~Audrito and S.~Terracini,
``On the nodal set of solutions to a class of nonlocal parabolic equations,''
\textit{Memoirs of the American Mathematical Society}, 
vol.~301, no.~1512, 2024.

\bibitem{CaffarelliSilvestre2007}
L.~Caffarelli and L.~Silvestre,
``An extension problem related to the fractional Laplacian,''
\textit{Communications in Partial Differential Equations}, vol.~32, no.~8, pp.~1245--1260, 2007.

\bibitem{DeGiorgi1957}
E.~De Giorgi, 
\textit{Sulla differenziabilità e l'analiticità delle estremali degli integrali multipli regolari}, 
Mem. Accad. Sci. Torino, P. I., III. Ser. 3, 25--43 (1957).

\bibitem{DongPhan2021}
H.~Dong and T.~Phan,
``Parabolic and elliptic equations with singular or degenerate coefficients: The Dirichlet problem,''
\textit{Transactions of the American Mathematical Society}, 
vol.~374, no.~9, pp.~, 2021.

\bibitem{FabesKenigSerapioni1982}
E.~B.~Fabes, C.~E.~Kenig, and R.~P.~Serapioni,
``The local regularity of solutions of degenerate elliptic equations,''
\textit{Communications in Partial Differential Equations}, vol.~7, no.~1, pp.~77--116, 1982.

\bibitem{FranchiGutierrezWheeden1994}
B.~Franchi, C.~E.~Gutiérrez, and R.~L.~Wheeden, 
\textit{Two-weight Sobolev--Poincaré inequalities and Harnack inequality for a class of degenerate elliptic operators}, 
Atti Accad. Naz. Lincei, Cl. Sci. Fis. Mat. Nat., IX. Ser., Rend. Lincei, Mat. Appl., 
\textbf{5} (2), 167--175 (1994).

\bibitem{GaleMianaStinga2013}
J.~E.~Galé, P.~J.~Miana, and P.~R.~Stinga,
``Extension problem and fractional operators: semigroups and wave equations,''
\textit{Journal of Evolution Equations}, 
vol.~13, pp.~343--368, 2013.

\bibitem{MetafuneNegroSpina2022}
G.~Metafune, L.~Negro, and C.~Spina,
``$L^p$ estimates for the Caffarelli--Silvestre extension operators,''
\textit{Journal of Differential Equations}, 
vol.~316, pp.~290--345, 2022.

\bibitem{MetafuneNegroSpina2023a}
G.~Metafune, L.~Negro, and C.~Spina,
``Anisotropic Sobolev Spaces with Weights,''
\textit{Tokyo Journal of Mathematics}, 
vol.~46, no.~2, pp.~313--337, 2023,
doi:10.3836/tjm/1502179386.

\bibitem{MetafuneNegroSpina2023b}
G.~Metafune, L.~Negro, and C.~Spina,
``A unified approach to degenerate problems in the half-space,''
\textit{Journal of Differential Equations}, 
vol.~351, pp.~63--99, 2023.

\bibitem{MetafuneNegroSpina2024a}
G.~Metafune, L.~Negro, and C.~Spina,
``Regularity theory for parabolic operators in the half-space with boundary degeneracy,''
\textit{Annali della Scuola Normale Superiore di Pisa, Classe di Scienze}, 
2024,
doi:\href{https://doi.org/10.2422/2036-2145.202310_009}{10.2422/2036-2145.202310\_009},
preprint available at \href{https://arxiv.org/abs/2309.14319v2}{arXiv:2309.14319v2}.

\bibitem{MetafuneNegroSpina2024b}
G.~Metafune, L.~Negro, and C.~Spina,
``Singular parabolic problems in the half-space,''
\textit{Studia Mathematica}, 
vol.~277, pp.~1--44, 2024,
doi:10.4064/sm230710-18-3.

\bibitem{Moser1961}
J.~Moser, 
\textit{On Harnack's theorem for elliptic differential equations}, 
Comm. Pure Appl. Math., \textbf{14} (3), 577--591 (1961).

\bibitem{Nash1958}
J.~Nash, 
\textit{Continuity of solutions of parabolic and elliptic equations}, 
Amer. J. Math., \textbf{80} (4), 931--954 (1958).

\bibitem{NegroSpina2025}
L.~Negro and C.~Spina.
\newblock Gaussian Poincar\'e inequalities on the half-space with singular weights.
\newblock {\em Mediterranean Journal of Mathematics}, 22(1):14, 2025.
\newblock Id/No 18.
\newblock doi:\href{https://doi.org/10.1007/s00009-024-02788-w}{10.1007/s00009-024-02788-w}.

\bibitem{StingaTorrea2010}
P.~R.~Stinga and J.~L.~Torrea,
``Extension problem and Harnack's inequality for some fractional operators,''
\textit{Communications in Partial Differential Equations}, 
vol.~35, no.~11, pp.~2092--2122, 2010.


\end{thebibliography}



\subsection*{MSC2020 Classification}

\begin{itemize}
    \item \textbf{46E35} — Sobolev spaces and other spaces of ``smooth'' functions, embedding theorems, trace theorems
    \item \textbf{35A23} — Inequalities involving derivatives and differential and integral operators (e.g., Sobolev-type inequalities)
    \item \textbf{35J70} — Degenerate elliptic equations
    \item \textbf{35B40} — Asymptotic behavior of solutions
    \item \textbf{26D10} — Inequalities involving derivatives and differential operators
    \item \textbf{46N20} — Functional analysis applied to differential and integral operators
    \item \textbf{35R11} — Fractional partial differential equations (if fractional extension applications are discussed)
    \item \textbf{60J60} — Diffusion processes (if probabilistic or Gaussian implications are emphasized)
\end{itemize}

\subsection*{Keywords}

\begin{itemize}
    \item Weighted Sobolev spaces
    \item Compact embeddings
    \item Rellich--Kondrachov theorem
    \item Degenerate elliptic equations
    \item Polynomial and Gaussian weights
    \item Tail coercivity
    \item Fractional extension problems
    \item Functional inequalities
\end{itemize}

\section{Acknowledgement}

This research received no external funding.

\end{document}