Edge-connectivity between edge-ends of infinite graphs

Extracted source text · arXiv 2404.17106

\documentclass[11pt,a4paper]{article}
\usepackage[margin=1in]{geometry}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{tikz}
\usepackage{hyperref}
\usepackage{amsmath,amssymb,amsfonts,amsthm,amscd, bm, mathtools}
\makeatletter
\newtheorem*{rep@theorem}{\rep@title}
\newcommand{\newreptheorem}[2]{%
\newenvironment{rep#1}[1]{%
 \def\rep@title{#2 \ref{##1}}%
 \begin{rep@theorem}}%
 {\end{rep@theorem}}}
\makeatother
\newtheorem{thm}{Theorem}[section]
\newreptheorem{theorem}{Theorem}
\newtheorem{prop}[thm]{Proposition}
\newtheorem{definition}[thm]{Definition}
\newtheorem{lemma}[thm]{Lemma}
\newtheorem{corol}[thm]{Corollary}

%\usepackage{draftwatermark}
%\SetWatermarkText{DRAFT}
%\SetWatermarkScale{3}

\title{Edge-connectivity between (edge-)ends of infinite graphs}
\author{Leandro Aurichi and Lucas Real}
\date{April 2024}

\begin{document}
\maketitle
\begin{abstract}
    In infinite graph theory, the notion of \textit{ends}, first introduced by Freudenthal and Jung for locally finite graphs, plays an important role when generalizing statements from finite graphs to infinite ones. Nash-Willian's Tree-Packing Theorem and MacLane's Planarity Criteria are examples of results that allow a topological approach, in which ends might be considered as \textit{endpoints} of rays. In fact, there are extensive works in the literature showing that classical theorems of (vertex-)connectivity for finite graphs can be discussed regarding ends, in a more general context. However, aiming to generalize results of edge-connectivity, this paper recalls the definition of \textit{edge-ends} in infinite graphs due to Hahn, Laviolette and \v{S}irá\v{n} in \cite{edgeends}. In terms of that object, we state an edge version of Menger's Theorem (following the work of Polat in \cite{polat}) and generalize the Lovász-Cherkassky Theorem for infinite graphs with edge-ends (inspired by the paper \cite{jacobs} of Jacobs, Joó, Knappe, Kurkofka and Melcher).      
\end{abstract}

\section{Introduction}

\paragraph{}
Most graphs in this paper are \textit{simple}, in the sense that loops and multiple edges are not considered. Exceptionally, graphs that might admit parallel edges are referred as \textit{multigraphs}. We recall that a \textbf{ray} in an (infinite) graph $G$ is an one-way infinite path. In other words, it is a subgraph $r = v_0v_1v_2\dots$, in which $v_i \in V(G)$ and $v_iv_{i+1}\in E(G)$ for every $i \in \mathbb{N}$. Thus, we say that $r$ \textbf{starts} at $v_0$ and that any infinite connected subgraph of $r$ is its \textbf{tail}. Similarly, a \textbf{double ray} is a subgraph of the form $R = \dots v_{-2}v_{-1}v_0v_1v_2\dots$, in which $\{v_iv_{i+1}\}_{i\in \mathbb{Z}}\in E(G)$. For any $n \in \mathbb{Z}$, the subgraphs $r_+ = v_{n}v_{n+1}v_{n+2}\dots$ and $r_- = v_{n}v_{n-1}v_{n-2}\dots$ are called the \textbf{half-rays} of $R$. Intuitively, a ray describe a direction in the ambient graph $G$, which is formalized by the following equivalence relation: we write $r\sim s$ whenever $r$ and $s$ cannot be separated by finitely many vertices, namely, the tails of $r$ and $s$ belong to the same connected component of $G\setminus S$ for every finite set $S\subset V(G)$. If this is not the case, we say that some finite set $S\subset V(G)$ \textbf{separates} $r$ and $s$.


Equivalently, $r\sim s$ if and only if there is an infinite family of $r-s$ disjoint paths. In this notation, given $A$ and $B$ two vertex subsets of $G$, we say that an $A-B$ \textbf{path} is a finite path $v_0v_1v_2\dots v_n$ with precisely one endpoint in $A$ and the other in $B$, i.e., $v_0\in A$, $v_n \in B$ and $v_1,v_2,\dots,v_{n-1}\in G\setminus (A\cup B)$. Hence, it is easily seen that $\sim$ is an equivalence relation over $\mathcal{R}(G)$, the collection of rays of $G$. The quotient $\frac{\mathcal{R}(G)}{\sim}$, denoted by $\Omega(G)$, is called the \textbf{end space} of $G$. An element $[r]\in \Omega(G)$, in its turn, is the \textbf{end} of the ray $r$.


This notion was first approached by Freudenthal in \cite{freudenthal} and by Hopf in \cite{hopf}, within algebraic discussions related to group theory. Halin in \cite{halin}, however, properly stated this definition for infinite graphs, inspiring generalizations of results from finite graph theory since then. For example, Stein in \cite{stein} showed how ends play an important role when extending the Nash-Willian's Three Packing Theorem for locally finite graphs. Her joint work with Bruhn in \cite{bruhn} presents a similar discussion, in which MacLane's Planarity Criteria is stated for locally finite graphs with their ends. Although this paper works under weaker conditions, local finiteness is a convenient hypothesis when preserving theorems from finite graph theory. As pointed out by Diestel in his survey \cite{survey}, the ends are helpful, and occasionally unavoidable, structures to that aim.  


In any case, the results mentioned in the previous paragraph are obtained throughout a topological approach. Indeed, from the definition of $\sim$, a topology for $\hat{V}(G) := V(G)\cup \Omega(G)$ naturally arises: we may see two ends as close each other as connected are their representatives. Formally, given $S\subset V(G)$ finite and $r \in \mathcal{R}(G)$, let $C(S,[r])$ be the set of vertices that belong to the connected component of $G\setminus S$ in which $r$ has a tail. By defining $$\Omega(S,[r]) = \{[s] \in \Omega(G) : S\text{ does not separate }r \text{ and }s\},$$ we declare $S$ and $\hat{V}(S,[r]):= C(S,[r])\cup \Omega(S,[r])$ as basic open sets in $\hat{V}(G)$. Since equivalent rays are infinitely connected, it is easily seen that $\Omega(S,[r])$ is well-defined. 


With this topology for $\hat{V}(G)$, the end $[r]$ is precisely the boundary of the ray $r$ (as a set of vertices), formalizing the idea that $r$ converges to $[r]$ or that $[r]$ might be seen as an endpoint of $r$. Then, roughly speaking, in the previously mentioned generalizations of results from finite graph theory, the ends of $G$ play the same role as \textit{vertices}, but ``at the infinity''. 


On the other hand, the literature related to the \textit{edge}-connectivity of infinity graphs and their ends is still not broad, with few references on the subject. An exception, for instance, is the paper \cite{jacobs}, in which Jacobs, Joó, Knappe, Kurkofka and Melcher state the classical Lovász-Cherkassky Theorem (recalled at the end of the next section) for locally finite graphs considering its ends. Their proof relies on the generalization of the same theorem for countable graphs, obtained by Joó in \cite{joo} without mentioning ends. Actually, the Lovász-Cherkassky Theorem is now written for arbitrary infinite graphs in \cite{joogeral}, where Joó extended his previous work. In this paper, we will revisit all these results by considering infinite graphs and their (edge-)ends. As a conclusion, at the end of Section \ref{sec-paths} we prove the following:

\begin{thm}[Lovász-Cherkassky for infinite graphs with edge-ends]\label{t1}
    Let $G = (V,E)$ be any infinite graph and fix $T\subset V(G)\cup \Omega_E(G)$ a discrete subspace of $\hat{E}(G)$' with $|T|\geq 2$. Suppose that $|\delta(X)|$ is even or infinite for every $X\subset V(G)$ in which $T\subset \overline{X}$. Then, there is a collection $\mathcal{P}$ of edge-disjoint $T-$paths such that, for every $t\in T$, there is a cut separating $t$ from $T\setminus \{t\}$ that lies on the family $\mathcal{P}_t = \{P\in \mathcal{P}: t \text{ is an endpoint of }P\}$.
\end{thm}


The definitions and notations employed in the above statement are presented throughout the next sections. However, we remark that the notion of edge-ends introduced by Theorem \ref{t1} is inspired by the definitions of ends, but slightly modified so that edge-connectivity results can be approached more properly. Thus, given rays $r,s\in \mathcal{R}(G)$, we now say that $r$ and $s$ are \textbf{edge-equivalent}, writing $r\sim_Es$, if, and only if, the tails of $r$ and $s$ belong to the same connected component of $G\setminus F$, for every finite set of \textit{edges} $F\subset E(G)$. Equivalently, there is an infinite family of edge-disjoint paths with disjoint endpoints connecting $r$ and $s$. If this is not the case, we say that a finite set $F\subset E(G)$ \textbf{separates} $r$ and $s$. When $F$ can described by $\{uv \in E(G): u \in X, v \in V(G)\setminus X\}$ for some $X\subset V(G)$, we write $F = \delta(X)$ and call $F$ the \textbf{cut} corresponding to the vertex set $X$. 


As it is easily seen, $\sim_E$ is also an equivalence relation over $\mathcal{R}(G)$, so that the quotient $\frac{\mathcal{R}(G)}{\sim_E}$ defines the \textbf{edge-end space} of $G$, denoted by $\Omega_E(G)$. Similarly, the equivalence class of ray $r$ is denoted by $[r]_E$, said to be an \textbf{edge-end} of the graph. We even observe that $\sim$ and $\sim_E$ are the same relation if $G$ is locally finite, because, in this case, rays which are separated by finitely many vertices can be separated by the finitely many edges incident to them. In particular, the main result of \cite{jacobs} will be preserved by Theorem \ref{t1}. When there are vertices of infinite degree, however, $\sim_E$ may identify more rays, as exemplified by Figure \ref{raioduplodominado}.

\begin{figure}[ht]
    \centering
    \begin{tikzpicture}
  % Double ray
  \draw[line width=2pt, black, -latex] (0,0) -- (3.5,0);
  \draw[line width=2pt, black, -latex] (0,0) -- (-3.5,0);
  
  % Vertices
  \foreach \x in {-3,-2,-1,0,1,2,3}
    \fill (\x,0) circle (2pt) node[below]{$v_{\x}$};

    \fill (0,1) circle (2pt) node[above]{$v_{\infty}$};

    \foreach \x in {-3,-2,-1,0,1,2,3}
    \draw (\x,0) -- (0,1);
  
  % Origin point
  \fill (0,0) circle (2pt);
\end{tikzpicture}
    
    \caption{This graph has two ends, corresponding to the rays $v_0v_1v_2\dots$ and $v_0v_{-1}v_{-2}\dots$, for example. However, it has only one edge-end.}
    \label{raioduplodominado}
\end{figure}


In the literature, the definition of $\sim_E$ was first presented by the paper \cite{edgeends} of Hahn, Laviolette and \v{S}irá\v{n}. Besides this article, the discussions carried out by Georgakopoulos in \cite{georgakopoulos} comprise one of the few other works which approach the notion of edge-ends. In fact, since $\sim$ and $\sim_E$ are the same equivalence relation over locally finite graphs, most references in infinite graph theory may not mention the latter one in their studies.    


Despite that, we observe that Theorem \ref{t1} also requires topological conditions on $V(G)\cup \Omega_E(G)$. Indeed, once presented the topology for $\hat{V}(G)$, the relation $\sim_E$ suggests a topology for $\hat{E}(G) = V(G)\cup \Omega_E(G)$ as well. Now, given a finite $F\subset E(G)$ and an edge-end $[r]_E\in \Omega_E(G)$, we denote by $C_E(F,[r]_E)$ the (vertices of the) connected component of $G\setminus F$ in which $r$ has a tail. In its turn, since edge-equivalent rays are infinitely edge-connected, the following set is also well-defined: $$\Omega_E(F,[r]_E) = \{[s]_E \in \Omega_E(G) : F \text{ does not separate }r\text{ and }s\}.$$


Then, by denoting $\hat{E}(F,[r]_E) := C_E(F,[r]_E)\cup \Omega_E(F,[r]_E)$, the aimed topology for $\hat{E}(G)$ has $\{\{v\} : v \in V(G)\}\cup \{\hat{E}(F,[r]_E): F\subset E(G)\text{ finite}, [r]_E\in \Omega_E(G)\}$ as a basis. In particular, the subspace $\Omega_E(G)\subset \hat{E}(G)$ has $\{\Omega_E(F,[r]_E) : [r]_E\in \Omega_E(G), F\subset E(G)\text{ finite}\}$ as a basis. Then, also considering $\Omega(G)\subset \hat{V}(G)$ with its inherited structure, a comparison between the classes of topological spaces $\Omega:=\{\Omega(G) : G \text{ graph}\}$ and $\Omega_E :=\{\Omega_E(G) : G \text{ graph}\}$ can be found in \cite{compaulo}. Among other studies in that paper, written in a joint work with Paulo Magalhães Júnior, we discuss how $\Omega_E$ is a proper subfamily of $\Omega$.    


Finally, we revisit the work of Polat in \cite{polat} in order to illustrate how the definition of $\sim_E$ is appropriate when studying edge-connectivity properties. More precisely, throughout the next sections we prove the following Mengerian result:

\begin{thm}[Menger's Theorem for edge-ends]\label{t2}
    Let $A,B\subset \Omega_E(G)$ be sets of edge-ends such that $\overline{A}\cap B = A\cap \overline{B} = \emptyset$. Then, the maximum size of a family of edge-disjoint $A-B$ paths is equal to the minimum size of an $A-B$ edge-separator.
\end{thm}

\section{Suitable connecting paths}
\label{sec-paths}
\paragraph{}
In this section, we formalize the definitions of $T-$paths and $A-B$ paths employed by Theorems \ref{t1} e \ref{t2} respectively, besides also proving a first common instance of these statements. To this aim, we first recall that Menger's classical theorem for finite graphs is a duality result regarding connectivity. It claims that the minimum amount of vertices of a graph $G$ that separate two subsets of $V(G)$ is attained by the maximum size of a family of disjoint paths connecting them. In this case, each path of this family must contain precisely one vertex from the minimum separator. It was conjectured by Erd\H{o}s that this property is preserved for infinite graphs, which was only verified by Aharoni and Berger in \cite{erdosmenger} after more than thirty years:

\begin{thm}[Erd\H{o}s-Menger Theorem, \cite{erdosmenger}]\label{erdosmenger}
    Let $G$ be a graph and fix $A,B\subset V(G)$. Then, there exists a family $\mathcal{P}$ of disjoint $A-B$ paths and $S\subset V(G)$ an $A-B$ separator that lies on it.
\end{thm}


In the above formulation, an $A-B$ \textbf{separator} is a vertex set $S\subset V(G)$ such that there is no $A-B$ path in $G\setminus S$. When saying that $S$ \textbf{lies on} a family $\mathcal{P}$ of disjoint paths, we mean that $S$ is obtained by the choice of precisely one vertex from each element of $\mathcal{P}$. 


However, we can rephrase Theorem \ref{erdosmenger} to consider $\mathcal{P}$ a family of \textit{edge-disjoint} $A-B$ paths, rather than analyzing just disjoint ones. In this case, the minimum separator described by the statement must be a set of edges. Then, we say that a cut $F$ in a graph $G$ \textbf{lies on} a family of edge-disjoint paths $\mathcal{P}$ if $F$ is obtained by the choice of precisely one edge from each element of $\mathcal{P}$. In these terms, the result below is folklore, but its proof is presented here for the sake of completeness. Roughly speaking, it is obtained after blowing up vertices to complete graphs and applying the Erd\H{o}s-Menger Theorem:

\begin{corol}[Erd\H{o}s-Menger Theorem for edges]\label{edgesmenger}
    Let $G$ be a graph and fix disjoint subsets $A,B\subset V(G)$. Then, there exist a family $\mathcal{P}$ of edge-disjoint $A-B$ paths and a cut $\delta(X)$ lying on it such that $A\subset X$ and $B\subset V(G)\setminus X$.  
\end{corol}
\begin{proof}
    We will define an auxiliary graph $\tilde{G}$. For every $v\in V(G)$, let $K_v$ be a complete graph of $d(v)$ vertices. Then, the vertex set of $\tilde{G}$ will be the disjoint union $\displaystyle \bigcup_{v\in V(G)}V(K_v)$. For every edge $uv \in E(G)$, we define an edge $u'v'$ between the cliques $K_u$ and $K_v$, referred as an \textit{old edge}. Since $|K_v| = d(v)$, we can assume that every vertex of $K_v$ is an endpoint of at most one old edge. 


    Apply Theorem \ref{erdosmenger} in order to separate the (disjoint) vertex sets $\tilde{A} = \displaystyle \bigcup_{v \in A}V(K_v)$ and $\tilde{B} = \displaystyle \bigcup_{u\in B}V(K_u)$, fixing $\tilde{\mathcal{P}}$ the family of disjoint $\tilde{A}-\tilde{B}$ paths and $\tilde{S}$ the separator lying on it. For every $\tilde{P}\in \tilde{\mathcal{P}}$ and every $v\in V(G)$, we can assume that $|\tilde{P}\cap V(K_v)| \leq 2$. In fact, if $\tilde{v}_1,\tilde{v}_2 \in V(K_v)$ are non-adjacent vertices in $\tilde{P}$, consider the path $\tilde{P}'$ obtained from $\tilde{P}$ after replacing the subpath connecting $\tilde{v}_1-\tilde{v}_2$ by the edge $\tilde{v}_1\tilde{v}_2$. Being an $\tilde{A}-\tilde{B}$ path, $\tilde{P}'$ must meet $\tilde{S}$ in the vertex of $\tilde{P}\cap \tilde{S}$, since $\tilde{\mathcal{P}}$ is composed by disjoint paths. Therefore, $(\tilde{\mathcal{P}}\setminus \{\tilde{P}\})\cup \{\tilde{P}'\}$ is also a family of disjoint $\tilde{A}-\tilde{B}$ paths in which $\tilde{S}$ lie.  


    By considering that $|\tilde{P}\cap V(K_v)| \leq 2$ for every $v\in V(G)$, a path $P$ in $G$ arises from $\tilde{P}\in \tilde{\mathcal{P}}$ after contracting the cliques $\{K_v : v \in V(G)\}$ to their original vertices. Then, $\mathcal{P} = \{P: \tilde{P} \in \tilde{\mathcal{P}}\}$ is a family of edge-disjoint $A-B$ paths. Moreover, each vertex $\tilde{v}\in \tilde{S}$ belongs to a clique of the form $K_v$ and it is the endpoint of an unique old edge $\theta(\tilde{v})$, originally incident in $v\in V(G)$. Note also that $\theta(\tilde{v})$ belongs to the path of $\tilde{\mathcal{P}}$ that contains $\tilde{v}$.


    We observe that every $A-B$ path $Q$ in $G$ must passes through an edge from $\{\theta(\tilde{v}): \tilde{v}\in \tilde{S}\}$. Otherwise, a minimal path in $\tilde{G}$ containing all the old edges of $Q$ will not intersect $\tilde{G}$, contradicting the fact that $\tilde{S}$ separates $\tilde{A}$ and $\tilde{B}$. Therefore, $F = \{\theta(\tilde{v}) : \tilde{v}\in \tilde{S}\}$ is an edge set lying on $\mathcal{P}$ for which there is no $A-B$ path in $G\setminus F$. Although it is no difficult to see that $F = \delta(X)$ for some $X\subset V(G)$ such that $A\subset X$ and $B\subset V(G)\setminus X$, this will later follow by Lemma \ref{cuts}.
\end{proof}


In addition, there also exist statements for Mengerian theorems when considering infinite graphs and their ends, which we shall revisit in order to conclude both Theorem \ref{t1} and \ref{t2}. To that aim, the notions of $A-B$ paths and $A-B$ separators need to be better discussed when $A$ and $B$ are not only vertex subsets, but might contain ends. Nevertheless, when dealing with the topological space $\hat{V}(G)$, different references in the literature present distinct definitions for connecting paths between ends and the corresponding separators, although all the formulations are coincident for locally finite graphs. 


Before comparing these concepts, it is useful to distinguish the role played by some vertices of infinite degree. More precisely, fixed a graph $G$, we say that $v\in V(G)$ \textbf{dominates} a ray $r \in \mathcal{R}(G)$ if it is infinitely connected to $r$, in the following sense: for every finite set $S\subset V(G)\setminus \{v\} $, a tail of $r$ and the vertex $v$ belong to the same connected component of $G\setminus S$, i.e., $v\in C(S,[r])$. Equivalently, there is an infinite family of paths connecting $v$ and $r$, pairwise intersecting precisely at $v$. In this case, $v$ dominates any other ray equivalent to $r$, allowing us to say that $v$ dominates the \textit{end} $[r]$. Then, combining notations from \cite{polat} and \cite{jacobs}, we set the following definitions:

\begin{definition}[Connecting paths]\label{paths}
    Let $G$ be a graph. Depending on whether we are considering connectivities of the form ``vertex-vertex'', ``vertex-end'' and ``end-end'', we define the \textbf{paths} and the \textbf{graphical paths} as follows:
    \begin{itemize}
        \item \texttt{Connectivity between vertices:} For vertices $u,v \in V(G)$, a $u-v$ path is any finite path in $G$ which has $u$ and $v$ as endpoints;
        \item \texttt{Connectivity between vertices and ends:} Fix $v\in V(G)$ and $\omega \in \Omega(G)$. A \textbf{graphic} $v-\omega$ \textbf{path} is a ray starting at $v$ whose end is $\omega$. A $\textbf{v}-\omega$ \textbf{path} is either a graphic $v-\omega$ path or a $v-u$ path, in which $u\in V(G)$ is a vertex that dominates $\omega$;
        \item \texttt{Connectivity between ends:} Fix $\omega_1,\omega_2 \in \Omega(G)$. A \textbf{graphic} $\omega_1-\omega_2$ \textbf{path} is a double ray in which one half-ray has end $\omega_1$ and the other has end $\omega_2$. In its turn, a $\omega_1-\omega_2$ \textbf{path} is either a graphic $\omega_1-\omega_2$ path or a $v-\omega_i$ path, for some $i\in \{1,2\}$ and $v\in V(G)$ a vertex that dominates $\omega_{3-i}$.
    \end{itemize}
    More generally, for sets $A,B\subset \hat{V}(G)$, a (graphic) $A-B$ \textbf{path} is any (graphic) $a-b$ path for some $a\in A$ and some $b\in B$. Additionally, only when $A, B\subset V(G)$, we impose that any $A-B$ path intersects $A\cup B$ precisely at their endpoints, which recovers the definition presented in the introduction. 
\end{definition}



Roughly speaking, the paths in Definition \ref{paths} differ from the graphic ones by allowing dominating vertices to represent some reachable end. If we consider only graphic paths, Bruhn, Diestel and Stein in \cite{versaodiestel} generalized Theorem \ref{erdosmenger} somehow verbatim, under a condition of topological separation:

\begin{thm}[\cite{versaodiestel}, Theorem 1.1]\label{versaodiestel}
    In a given connected graph $G$, fix $A,B\subset \hat{V}(G)$ two sets that are separated topologically, i.e., such that $A\cap \overline{B} = \overline{A}\cap B =\emptyset$. Then, there exist a family $\mathcal{P}$ of disjoint graphic $A-B$ paths and $S\subset \hat{V}(G)$ a subset with the following properties:
    \begin{itemize}
        \item[$i)$] The graphic paths intersect $A$ and $B$ precisely at their endpoints. Formally, for every $P\in \mathcal{P}$ there are $a \in A$ and $b\in B$ such that $\overline{P} \cap (A\cup B) = \{a,b\}$;
        \item[$ii)$] Any two distinct graphic paths $P,Q\in\mathcal{P}$ are disjoint even at the infinity. In other words, $\overline{P}\cap \overline{Q} = \emptyset$;
        \item[$iii)$] $|\overline{P}\cap S| = 1$ for every $P\in \mathcal{P}$ and $S = \displaystyle \bigcup_{P\in\mathcal{P}}S\cap \overline{P}$;
        \item[$iv)$] If $Q$ is any graphic $A-B$ path, then $\overline{Q}\cap S \neq \emptyset$.
    \end{itemize}
\end{thm}


Comparing Theorem \ref{versaodiestel} with the original statement of the Erd\H{o}s-Menger Theorem, the subset $S\subset \hat{V}(G)$ of the above thesis plays the role of an ``$A-B$ separator lying on the family $\mathcal{P}$''. Nevertheless, when applied to the graph $G$ of Figure \ref{raioduplodominado}, there are no two disjoint graphic paths connecting the only two ends $\omega_1,\omega_2\in \Omega(G)$. But, since $v_{\infty}$ is adjacent to every other vertex, at least two \textit{vertices} are needed to separate $\omega_1$ and $\omega_2$. The ``$\{\omega_1\}-\{\omega_2\}$ separator'' claimed by Theorem \ref{versaodiestel}, then, is given by either $\{\omega_1\}$ or $\{\omega_2\}$. However, if we aim to forbid ends in separating sets, the broader definition of connecting paths in Definition \ref{paths} - rather than only the graphic ones - is useful. Relying on this notion, Polat in \cite{polat} obtained the Mengerian result below:


\begin{thm}[\cite{polat}, Theorem 3.2]\label{mengerpolat}
    Let $A,B\subset \Omega(G)$ be two sets of ends such that $\overline{A}\cap B  = A\cap \overline{B} = \emptyset$. Consider a family $\mathcal{P}$ of maximum size of disjoint $A-B$ paths and fix an $A-B$ separator $S$ of minimum size. Then, $|\mathcal{P}| = |S|$.
\end{thm}


In fact, the above statement is also the main inspiration for our Theorem \ref{t2}. The similarity between these results suggests that, in order to conclude the latter, the definitions employed in the former need to be rewritten in their edge-related versions. Hence, inheriting the notion of $\mathcal{E}-$dominance from \cite{edgeends}, we say that a vertex $v\in V(G)$ of a graph $G$ \textbf{edge-dominates} an edge-end $\omega \in \Omega_E(G)$ if $v$ is infinitely edge-connected to any representative $r$ of $\omega$. More precisely, $v\in C_E(F,\omega)$ for every finite set $F\subset E(G)$. In this case, there is an infinite family of edge-disjoint paths connecting $v$ to infinitely many vertices of $r$. Then, we fix the following definitions:

\begin{definition}[Connecting paths - edge version]\label{epaths}
    Let $G$ be a graph. For vertex sets $A,B\subset V(G)$, the notion of an $A-B$ path given by Definition \ref{paths} is preserved. For connectivities between edge-ends and vertices and edge-ends, we consider the following criteria:
    \begin{itemize}
        \item \texttt{Connectivity between vertices and edge-ends:} Fix $v\in V(G)$ and $\omega \in \Omega_E(G)$. A \textbf{graphic} $v-\omega$ \textbf{path} is a ray starting at $v$ whose edge-end is $\omega$. Then, a $v-\omega$ \textbf{path} is either a graphic one or a $\{v\}-\{u\}$ path for some vertex $u\in V(G)$ that edge-dominates $\omega$;
        \item \texttt{Connectivity between edge-ends:} For edge-ends $\omega_1,\omega_2\in \Omega_E(G)$, an $\omega_1-\omega_2$ \textbf{path} is one of objects below:
        \begin{itemize}
            \item[$i)$] A $v-\omega_i$ path, for some $i\in \{1,2\}$ and some vertex $v\in V(G)$ that edge-dominates $\omega_{3-i}$; 
            \item[$ii)$] A double ray in which one half-ray has edge-end $\omega_1$ and the other has edge-end $\omega_2$. This case defines a \textbf{graphic} $\omega_1-\omega_2$ \textbf{path}.
        \end{itemize}
    \end{itemize}
     Finally, given $A,B\subset \hat{E}(G)$ such that $(A\cup B)\cap \Omega_E(G) \neq \emptyset$, an $A-B$ \textbf{path} is simply an $a-b$ path for some $a \in A$ and some $b\in B$.  
\end{definition}


In order to conclude a restricted version of Theorem \ref{t1}, which will be later useful for proving Theorem \ref{t2}, we turn our attention to subsets of $\hat{E}(G)$ that can be separated by finitely many edges. While discussing Menger-type results, we are particularly interested when these separators are small ones, often minimal or with minimum size. Then, in future arguments, we shall rely on the following remark without explicit mention:   

\begin{lemma}\label{cuts}
    Consider sets $A$ and $B$ such that either $A\subset V(G)$ or $A\subset \Omega_E(G)$, as well as either $B\subset V(G)$ or $B\subset \Omega_E(G)$. If it exists, let $F\subset E(G)$ be a finite edge set for which there is no $A-B$ path in $G\setminus F$. Assume even that $F$ is minimal with that property. Then, there is $X\subset V(G)$ such that $F = \delta(X)$.   
\end{lemma}
\begin{proof}
    Considering the topology of $\hat{E}(G)$, define the family $$\mathcal{C} = \{C: C \text{ is a connected component of }G\setminus F \text{ such that }A\cap \overline{C} \neq \emptyset\}.$$ 
    We will show that the claimed subset of $V(G)$ can be chosen as $X = \displaystyle \bigcup_{C\in \mathcal{C}}V(C)$. In fact, for every edge $e\in F$, there is an $A-B$ path in $G\setminus (F\setminus \{e\})$ by the minimality of $F$. This path contains $e$ by the main hypothesis over $F$, although it contains no other edge from $F$. Therefore, $e$ has an endpoint in $X$ and the other in $V(G)\setminus X$, so that $e\in \delta(X)$. 


    Conversely, by definition of $\mathcal{C}$, an edge $e\in \delta(X)$ has endpoints in two distinct connected components of $G\setminus F$. Hence, we must have $e\in F$.
\end{proof}


The precise definition of an $A-B$ separator as mentioned by Theorem \ref{t2} will be better introduced in the next section. From now until the end of the current one, we shall rather conclude the instance below and drawn some related consequences. Roughly speaking, its proof follows from an iterative application of the Erd\H{o}s-Menger Theorem for edges. In the literature, its vertex-analogous statement is the Theorem 4.7 found in \cite{PolatFrances}, from where the details below are adapted: 

\begin{lemma}\label{vertex-end}
    Let $G$ be a connected graph and fix $A\subset \Omega_E(G)$. Assume the existence of a finite edge set $F\subset E(G)$ for which there is no $A-B$ path in $G\setminus F$. Suppose in addition that $F$ is minimal with this property. Then, there is $\mathcal{P}$ a family of $|F|$ edge-disjoint $S-A$ paths.
\end{lemma}
\begin{proof}[Revisited proof of Theorem 4.7 in \cite{PolatFrances}]
    We first consider the case when $A$ is closed in $\Omega_E(G)$. Then, we will construct the required family $\mathcal{P}$ recursively, as the limit object from a sequence $(\mathcal{P}_n)_{n \in \mathbb{N}} = \{P_1^n,P_2^n,\dots,P_k^n\}_{n\in\mathbb{N}}$ of families of edge-disjoint finite paths, where $k = |F|$. By its minimality, we can write $F = \delta(V_0)$ for some $V_0\subset V(G)$, denoting $V_0' = V(G)\setminus V_0$.  Without loss of generality, we assume that $S\subset V_0$ and $A\subset \overline{V_0'}$.
     

    Let $S'$ be the endpoints of the edges in $F$ that belong to $V_0'$. Consider $\hat{G}$ the subgraph of $G$ obtained by adjoining to $G[V_0]$ precisely $S'$ and the edges from $F$. Hence, by the minimality of $F$, the Erd\H{o}s-Menger Theorem for edges guarantees the existence in $\hat{G}$ of a family $\mathcal{P}$ of $k$ edge-disjoint $S'-S$ paths. In particular, each edge from $F$ belongs to precisely one path of $\mathcal{P}$.
    
 Setting $G_0 := G$, $\mathcal{P}_0:=P$ and $F_0:= F$, suppose by induction that we have defined the following objects and its properties:
	\begin{itemize}
		\item The family of paths $\mathcal{P}_n = \{P_i^n: 1 \leq i \leq k\}$. For each $1 \leq i \leq k$, denote by $v_i^n$ the endpoint of $P_i^n$ other than the one in  $S$;
		\item $G_n$ is a subgraph of $G$ in which every end of $A$ has a representative;
		\item A cut $F_n$ in $G_n$ of minimum size that separates a vertex set $S_n\subset V(G_n)$ from $A$, in the sense that there is no $S_n-A$ path in $G_n\setminus F_n$. We assume also that the last edge from each path of $\mathcal{P}_n$ is an element of $F_n$. 
	\end{itemize} 


First, since $F_n$ is a cut of $G_n$, let us write $F_n = \delta(V_n)$ and $V_n':=V(G_n)\setminus V_n$ for some $V_n\subset V(G_n)$. We assume that $G_{n+1} := G_n[V_n]$ is the subgraph induced by the part of the bipartition $\{V_n, V_n'\}$ in which every end of $A$ has a representative.  Then, define $$S_{n+1} = \{v_j^n : \text{ finitely many edges separate }v_j^n\text{ from }A\}.$$


For every $1 \leq i \leq k$ such that $v_i^n \notin S_{n+1}$, we set $P_i^{n+1} = P_i^n$. Since $S_{n+1}$ is finite, its definition allows us to find a finite set $F_{n+1}\subset E(G_{n+1})$ that separates $S_{n+1}$ from $A$. By choosing $F_{n+1}$ of minimum size with that property, $F_{n+1}$ is a cut in $G_{n+1}$. Hence, we write  $F_{n+1} = \delta(V_{n+1})$ for some $V_{n+1}\subset V(G_{n+1})$, considering $V_{n+1}' := V(G_{n+1})\setminus V_{n+1}$ the part of the bipartition $\{V_{n+1}, V_{n+1}'\}$ containing $S_{n+1}$.


Denote by $K_n \subset F_n$ the set of edges of $F_n$ whose endpoints belong to $S_{n+1}$. Let $A'$ and $B'$ be the set of endpoints of edges from $F_{n}$ and $F_{n+1}$, respectively, that do not belong to $G_{n+1}[V_{n+1}']$. Define $\hat{G}_{n}$ as the graph obtained after adding to $G_{n+1}[V_{n+1}']$ the vertices of $A'\cup B'$ and the edges of $K_n\cup F_{n+1}$. By the edge version of the Erd\H{o}s-Menger Theorem, there is a family $\mathcal{P}'$ of edge-disjoint $A'-B'$ paths in $\hat{G}_n$ with a cut $C'$ obtained by the choice of precisely one edge from each path of $\mathcal{P}'$. We observe that $|C'|\geq |K_n|$. Otherwise, $F_n' = (F_n\setminus K_n)\cup C'$ has strictly fewer edges than $F_n$ and separates $A$ from $S_n$, contradicting the definition of $F_n$. Therefore, each edge from $K_n$ lies in precisely one path of $\mathcal{P}'$. By concatenating these paths with the previous paths of $\mathcal{P}_n$ that end in $K_n \subset F_n$, we finish the definition of $\mathcal{P}_{n+1}$.


At the end of this recursive process, consider $P_i' = \displaystyle \bigcup_{n\in\mathbb{N}}P_i^n$ for each $1 \leq i \leq k$. We observe that, if $P_i'$ is a ray, its edge-end belongs to $A$, so that $P_i'$ is a $S-A$ path. Otherwise, since $A$ is closed, there would be a finite set $F'\subset E(G)$ such that $A\cap C_E(F',[P_i']_E) = \emptyset$. However, fixing a path on $m$ edges which connects an edge $e \in F'$ to some edge of $F_0$, we could conclude that $e\notin E(G_{m+1})$. Hence, this would prove that $F' \cap E(G_n) = \emptyset$ for some big enough $n\in\mathbb{N}$, contradicting the construction of $P_i'$. 


Similarly, if $P_i'$ is a finite path for some $1 \leq i \leq k$, then $P_i' = P_i^n$ for all but finitely many numbers $n \in \mathbb{N}$. If its endpoint $v_i$ edge-dominates an edge-end from $A$, then $P_i'$ is also a $S-A$ path by definition. In its turn, consider the set of indices $I = \{1 \leq i \leq k : P_i'\text{ is not a }S-A\text{ path}\}$. Let $J\subset I$ be a subset of maximum size for which there is a family $\mathcal{P}_J = \{P_j : j \in J\}$ of edge-disjoint $S-A$ paths satisfying the following properties:

\begin{itemize}
    \item $\{P_i': i\notin J\}\cup \{P_j : j \in J\}$ is a set of edge-disjoint rays or finite paths;
    \item $P_j'$ is an initial subpath of $P_j$ for every $j\in J$.
\end{itemize}


If we prove that $J = I$, then $\mathcal{P} = \{P_i' : i \notin I\}\cup \{P_i : i \in I\}$ is the claimed family of edge-disjoint $S-A$ paths. Indeed, suppose that there exists $l\in I\setminus J$. Then, $P_l'$ is a finite path whose endpoint $v_l$ (other than the one in $S$) does not edge-dominate an end from $A$. In particular, $v_l$ does not edge-dominate an edge-end from $\{[P_i']_E : P_i' \text{ is a ray}, i\notin I\} \cup \{[P_j]_E : P_j\text{ is a ray}, j \in J\}$. Hence, there is $L\subset E(G)$ a finite set of edges such that, in $G\setminus L$, the connected component $C$ containing $v_l$ does not intersect $\displaystyle \bigcup_{i\notin J}V(P_i')\cup \bigcup_{j\in J}V(P_j)$. However, by construction, $v_l$ cannot be separated from $A$ by finitely many edges, since $P_l'$ is a finite path. Hence, in $C$ there is a ray starting at $v_l'$ whose edge-end belongs to $A$. Concatenating this ray with the finite path $P_l'$, we contradict the maximality of $J$. 


Now, suppose that $A\subset \Omega_E(G)$ is any subset that can be separated from $S$ by the finite set $F\subset E(G)$. If $F$ has minimum size with that property, we observe that $F$ also separates $\overline{A}$ from $S$. For instance, suppose that $r$ is a graphic $S-\overline{A}$ path that misses $F$. Since $[r]_E\in \overline{A}$, fix $\omega \in A\cap \Omega_E(F,[r])$. Then, if $r'$ is a ray whose end is $\omega$, its tail in $G\setminus F$ is contained in $C_E(F,[r]')$. Assuming that $r'$ starts at $S$, we contradict the main hypothesis over $A$.  


Hence, by the case just analyzed, there is $\mathcal{P}$ a family of  $k = |F|$ edge-disjoint $S-\overline{A}$ paths. Then, for every $P\in \mathcal{P}$ there is an edge-end $\omega_P\in \overline{A}$ such that $P$ is either a ray whose edge-end is $\omega_P$ or a finite path such that an endpoint edge-dominates $\omega_P$. In this latter case, by shortening the path $\mathcal{P}$ if necessary, we assume that its endpoint is the unique vertex of $P$ that edge-dominates the edge-end $\omega_P$. 


Consider then the set $X = \{P\in \mathcal{P}:\omega_P \in \overline{A}\setminus A \}$. Suppose that $\mathcal{P}$ is chosen so that $X$ has minimum size. Finishing the proof, we claim that $X = \emptyset$. Otherwise, fix $P\in \mathcal{P}$. Since $\Omega_E(G)$ is a Hausdorff space, there is a finite edge set $L\subset E(G)$ such that $\omega_Q\notin \Omega_E(L,\omega_P)$ for every $Q\in \mathcal{P}$ such that $\omega_Q\neq \omega_P$. Moreover, by possibly adding (finitely many) edges to $L$, we can assume that no edge from $\displaystyle \bigcup_{\substack{Q\in \mathcal{P} \\\omega_Q \neq \omega_ P}}E(Q)$ lies in $C_E(L,\omega_P)$. However, there is $r'$ a ray in $C_E(L,\omega_P)$ whose edge-end $[r']_E$ belongs to $A$, since $\omega_P\in \overline{A}\setminus A$. Unless by changing the choice of $P$ within $\{Q\in \mathcal{P}: \omega_Q = \omega_P\}$, we can assume that $r'$ starts in a vertex of $P$ and does not intersect $\displaystyle \bigcup_{\substack{Q\in \mathcal{P} \\\omega_Q = \omega_ P}}V(Q)$ in any other point, because $[r']_E \neq \omega_P$. This defines a $S-A$ path $P'$ by concatenating $r'$ with an initial segment of $P$, so that $(\mathcal{P}\setminus\{P\})\cup \{P'\}$ contradicts the minimality of $X$.      
    
\end{proof}


Applying the above result for both sides of a bipartition of $V(G)$ given by a minimum cut, we obtain the following corollary as a particular case of Theorem \ref{t1}:

\begin{corol} \label{finiteseparation}
    Let $A,B\subset \Omega_E(G)$ be two sets of edge-ends of a graph $G$. Assume the existence of a finite edge set $F\subset E(G)$ for which there is no $A-B$ path in $G\setminus F$. If $F$ has minimum size with that property, there is a family $\mathcal{P}$ of $|F|-$many edge-disjoint $A-B$ paths.
\end{corol}
\begin{proof}
    Since $F$ has minimum size, we can write $F = \delta(V_1)$ for some $V_1\subset V(G)$ by Lemma \ref{cuts}. Consider $V_2 = V(G)\setminus V_1$ and let $S_i$ be the set of endpoints of edges of $F$ in $V_i$, for $i=1,2$. Assume that the representatives of edge-ends in $A$ have their tails in $G[V_1]$, while the representatives of edge-ends in $B$ have their tails in $G[V_2]$. Hence, $F$ is an edge set of minimal size separating $A$ from $S_2$ in $G[V_1\cup S_2]$. By Lemma \ref{vertex-end}, there is $\mathcal{P}_1 = \{P_1^1,P_2^1,\dots,P_{|F|}^1\}$ a family of edge-disjoint $S_2-A$ paths. Analogously, there is $\mathcal{P}_2 = \{P_1^2,P_2^2,\dots,P_{|F|}^2\}$ a set of edge-disjoint $S_1-B$ paths in $G[V_2\cup S_1]$. By changing the enumeration of the elements in $\mathcal{P}_2$ if necessary, we observe that $P_i^1$ and $P_i^2$ intersects in precisely one edge from $F$ for every $1 \leq i \leq |F|$. Therefore, the concatenation $P_i^1P_i^2$ is a well defined $A-B$ path, so that $\mathcal{P} = \{P_i^1P_i^2 : 1 \leq i \leq |F|\}$ holds as the claimed family.
\end{proof}


Finishing this section, we will discuss how Lemma \ref{vertex-end} is useful for extending already known generalizations of the Lovász-Cherkassky Theorem, specially in order to draw a proof for Theorem \ref{t1} as one of our main results. Before that, we first recall its classical statement within finite graph theory:

\begin{thm}[Lovász-Cherkassky]\label{lovaszfinito}
	In a given finite graph $G$, fix $T\subset V(G)$. If $G$ is inner-Eulerian for $T$, the maximum number o pairwise disjoint edge-disjoint $T-$paths in $G$ is equal to $$\frac{1}{2}\sum_{t\in T}\lambda(t,T\setminus \{t\}),$$ where $\lambda(t,T\setminus \{t\})$ denotes the minimum size of a cut separating $t$ from $T\setminus \{t\}$. 
\end{thm}


In the above result, a $T-$\textbf{path} in $G$ is a $T-T$ path $P$ as in Definition \ref{paths}. On the other hand, $G$ is \textbf{inner-Eulerian} for $T$ if every vertex of $V(G)\setminus T$ has even degree. Under this hypothesis, Theorem \ref{lovaszfinito} claims that there is a family $\mathcal{P}$ of edge-disjoint $T-$paths attaining an optimal size. After all, in a family of at least $1 + \displaystyle \frac{1}{2}\sum_{t\in T}\lambda(t,T\setminus \{t\})$ paths, two of them must intersect in an edge of a minimum cut separating some vertex $t$ from $T\setminus \{t\}$. In particular, for every $t\in T$, the maximum family $\mathcal{P}$ contains precisely $\lambda(t,T\setminus \{t\})$ paths that ends at $t$. Considering this property, Jacobs, Joó, Knappe, Kurkofka and Melcher obtained a version of the Lovász-Cherkassky Theorem for locally finite graphs and their ends: 

 \begin{thm}[\cite{jacobs}, Theorem 1]\label{jooetal}
	Let $G$ be a locally finite graph and fix $T\subset \hat{V}(G) $ a discrete subset. Suppose that $|\delta(X)|$ is even or infinite for every $X\subset V(G)$ for which $T\subset \overline{X}$. Then, there is a family $\mathcal{P}$ of edge-disjoint graphic $T-$paths such that, for every $t\in T$, the number of $\{t\}-(T\setminus \{t\})$ paths is equal to $\lambda(t,T\setminus \{t\})$. 
\end{thm} 


Following the notation from Definition \ref{paths}, a \textbf{graphic} $T-$\textbf{path} is a graphic $T-T$ path $P$ such that $|\overline{P}\cap T| = 2$. Intuitively, $P$ intersects $T$ precisely at its endpoints. Since Theorem \ref{jooetal} is restricted to locally finite graphs, we can rely on both Definitions \ref{paths} and \ref{epaths} to state the above result, once there is no distinction between ends and edge-ends in this graph family. Analogously, $\lambda(t,T\setminus \{t\})$ now denotes the minimum size of a $\{t\} - T\setminus\{t\}$ (edge-)separator, as in Definition \ref{eseparator}. By assuming that $T\subset \hat{V}(G)$ is discrete, $\lambda(t,T\setminus \{t\})$ is a well-defined natural number.  


Comparing the Theorems \ref{lovaszfinito} and \ref{jooetal}, the ``inner-Eulerian'' hypothesis over $G$ is, in the locally finite generalization, replaced by a parity condition over some finite cuts. Despite that, the new assumption restricts to the original hypothesis for finite graphs. More precisely, if $G$ is finite and every vertex of $G\setminus T$ has finite degree, let $\tilde{G}$ be the multigraph obtained by contracting a set $X\supset T$ to a new vertex $v$. Since $\tilde{G}$ has an even number of vertices of odd degree, the degree of $v$ must be even, because so is the degree of every vertex from $V(\tilde{G})\setminus \{v\} = V(G)\setminus X$. Noticing that $|\delta(X)|$ is the degree of $v$ in $\tilde{G}$, the above result is, in fact, a generalization of the classical Lovász-Cherkassky Theorem for finite graphs. On the other hand, its proof relies on the countable version of Theorem \ref{lovaszfinito}, previously obtained by Joó in \cite{joo}. Below, we present an updated statement of Joó's result, written by him in \cite{joogeral} for arbitrary graphs:



\begin{thm}[\cite{joogeral}, Theorem 1.2]\label{joo}
	Let $G$ be a (multi)graph and fix $T\subset V(G)$ with $|T|\geq 2$. Suppose that $|\delta(X)|$ is even or infinite for every $X\subset V(G)$ in which $T\subset X$. Then, there is a family $\mathcal{P}$ of edge-disjoint $T-$paths such that, for every $t\in T$, there is a cut separating $t$ from $T\setminus \{t\}$ obtained by the choice of precisely one edge from each path of $\mathcal{P}_t = \{P\in \mathcal{P}: t \text{ is an endpoint of }P\}$.
\end{thm}



The proof of Theorem \ref{jooetal} is drawn by an appropriate reduction to Theorem \ref{joo}. In \cite{jacobs}, previously known results about the connectivity of locally finite graphs and its ends were also recalled for that study, such as Lemma 10 in \cite{finitos-um}. Now, in order to conclude the Lovász-Cherkassky Theorem for arbitrary infinite graphs and its edge ends, we can combine Lemma \ref{vertex-end} to the main idea for proving Theorem \ref{jooetal}. To this aim, a $T-$\textbf{path} in the result below means a $T-T$ \textbf{path} $P$ as in Definition \ref{epaths} and such that $|\overline{P} \cap T| = 2$: 

\begin{reptheorem}{t1}
	Let $G = (V,E)$ be a graph and fix $T\subset \hat{E}(G)$ a discrete subspace of $\hat{E}(G)$ with $|T|\geq 2$. Suppose that $|\delta(X)|$ is even or infinite for every $X\subset V(G)$ in which $T\subset \overline{X}$. Then, there is a collection $\mathcal{P}$ of edge-disjoint $T-$paths such that, for every $t\in T$, there is a cut separating $t$ from $T\setminus \{t\}$ that lies on the family $\mathcal{P}_t = \{P\in \mathcal{P}: t \text{ is an endpoint of }P\}$.
\end{reptheorem}    
\begin{proof}[Revisited proof of Theorem 1 in \cite{jacobs}]
	We shall rely on Lemma \ref{discreto}, which will be proven in the next section with the support of suitable tools that are developed there. Since $T\subset \hat{E}(G)$ is discrete, that result claims the existence of a family $\{C_{\omega}: \omega \in T\cap \Omega_E(G)\}$ of connected subgraphs of $G$ for which the following properties are verified:

  \begin{itemize}
        \item $C_{\omega_1}\cap C_{\omega_2} = \emptyset$ if $\omega_1 \neq \omega_2$;
        \item For every $\omega \in T\cap \Omega_E(G)$, the cut $F_{\omega}:=\delta(C_{\omega})$ is finite and $\hat{E}(F_{\omega},\omega)\cap T = \{\omega\}$.
    \end{itemize}
 
 
 Then, for every $\omega \in \Omega_E(G)$, let $\hat{C_{\omega}}$ denote the subgraph of $G$ induced by $C_{\omega}$ and the endpoints of the edges in the finite cut $F_{\omega} := \delta(C_{\omega})$. In particular, $F_{\omega}$ turns out to be a finite $S_{\omega}-\{\omega\}$ separator in $G$, where $S_{\omega} = \{x\in V(G)\setminus C_{\omega} : x\text{ is an endpoint of and edge from }F_{\omega}\}$. By passing $C_{\omega}$ to a connected subgraph if necessary, Lemma \ref{cuts} allows us to assume that no edge set in $G$ with fewer than $|F_{\omega}|$ elements is also a $S_{\omega}-\{\omega\}$ separator. Hence, now according to Lemma \ref{vertex-end}, there is a family $\mathcal{P}_{\omega}$ of $|F_{\omega}|-$many edge-disjoint $S_{\omega}-\{\omega\}$ paths in $\hat{C_{\omega}}$.
    
    
    In its turn, let $\tilde{G}$ be the (multi)graph obtained from $G$ by contracting each $C_{\omega}$ to a vertex $v_{\omega}$. In this new graph, for every $\omega \in T\cap \Omega_E(G)$, the edges from $F_{\omega}$ are precisely the ones that are incident in $v_{\omega}$. By hypothesis, when setting $T' = (T\cap V(G))\cup \{v_{\omega} : \omega \in T\cap \Omega_E(G)\}$, the cardinal $|\delta(X)|$ is even or infinite for any $X\subset V(\tilde{G})$ containing $T'$. After all, the edges in $\delta(X)$ are precisely those in $G$ from the cut defined by $\displaystyle (X\cap V(G))\cup \bigcup_{\omega \in T\cap \Omega_E(G)}C_{\omega}$. Therefore, Theorem \ref{joo} (that has no mention of ends or edge-ends) guarantees the existence of a collection $\mathcal{P}'$ of edge-disjoint $T'-$paths such that, for every $t\in T'$, there is a cut $E_t$ separating $t$ from $T'\setminus \{t\}$ which lies on the family $\mathcal{P}_t' = \{P\in \mathcal{P}': t \text{ is an endpoint of }P\}$.
    
     
    
    Finishing the proof, we describe the claimed collection $\mathcal{P}$ of edge-disjoint $T-$paths as follows: for every $\omega \in T\cap \Omega_E(G)$, we concatenate different paths from $\mathcal{P}_{v_{\omega}}'$ with different paths from the family $\mathcal{P}_{\omega}$. Note that $ |\mathcal{P}_{v_{\omega}}'| = |E_{v_{\omega}}|\leq |F_{\omega}| = |\mathcal{P}_{\omega}|$, because $|F_{\omega}|$ is precisely the degree of $v_{\omega}$ in $\tilde{G}$. Then, this construction of $\mathcal{P}$ is well defined, although some paths of $\mathcal{P}_{\omega}$ are not extended to paths of $\mathcal{P}$ if $|E_{v_{\omega}}|<|F_{\omega}|$.     
\end{proof}








\section{Topological and combinatorial separations}
\paragraph{}

We observe that our main results pointed out in the introduction mention topological conditions on relevant sets of vertices or edge-ends. On one hand, the Lovász-Cherkassky theorem for infinite graphs requires that the set $T$ as in Theorem \ref{t1} is discrete. On the other, the sets $A$ and $B$ in Theorem \ref{t2} satisfy $\overline{A}\cap B = A\cap \overline{B} = \emptyset$. However, these hypothesis were not explored so far. In fact, Lemma \ref{vertex-end} was the core of the previous section, but it was obtained when supposing that its relevant objects could be separated by only finitely many edges. In its turn, this section shall explain how the topological assumptions on Theorems \ref{t1} and \ref{t2} ensure the well-definition of infinite edge sets as separators. In particular, we draw some combinatorial interpretations for standard notions within the topology of end and edge-end spaces.


Before that, it is useful to first recall some recent results towards the end structure of infinite graphs. For example, revisiting the work of Kurkofka and Pitz in \cite{representacao}, an \textbf{envelope} for a vertex subset $U\subset V(G)$ of a graph $G$ is a superset $U^*\supset U$ satisfying the two axioms below:


\begin{itemize}
    \item The set $N(C) = \{v \in U^* : v\text{ has a neighbor in }C\}$ is finite for every connected component $C$ of $G\setminus U^*$;
    \item $\overline{U^*}\setminus U^* = \overline{U}\setminus U$, i.e., the boundaries of $U$ and $U^*$ in the topological space $\hat{V}(G)$ define the same set of ends.
\end{itemize}


Envelopes for an arbitrary $U$ are constructed by Theorem 3.2 of \cite{representacao}, and we can even assume that they induce connected subgraphs. Moreover, the first condition above claims precisely that envelopes are subsets of \textbf{finite adhesion}, in the sense that the connected components of their complements have finite neighborhood. If $C$ is such a connected component and it contains a ray $r$, then $\hat{V}(N(C),[r])$ is a basic open set in $\hat{V}(G)$ around the end $[r]$. Due to this simple observation, most definitions and results in this section are statements about convenient subgraphs of finite adhesion.  


Incidentally, rayless normal trees are probably the most well-known examples of these subgraphs. In fact, we recall that a tree $T$ in a graph $G$ is \textbf{normal} if, after fixing a tree-order $\leq$, any path $P = v_0v_1\dots v_n$ in $G$ has comparable endpoints regarding $\leq$ if $V(T)\cap V(P) = \{v_0,v_n\}$. In particular, there is no edge in $G$ connecting incomparable elements of $T$ and, for every connected component $C$ of $G\setminus T$, its neighborhood $N(C) = \{v\in T : v\text{ has a neighbor in }C\}$ is totally ordered by $\leq$. In this case, $T$ has finite adhesion if it contains no rays. Considering that, Kurkofka, Melcher and Pitz in \cite{paracompacidade} discussed how rayless normal trees can approximate the end structure of a given graph:

\begin{thm}[\cite{paracompacidade}, Theorem 1]\label{aproximacao}
    Let $G$ be an infinite graph. Fix an open cover for the subspace $\Omega(G)$ of the form $\mathcal{C} = \{\hat{V}(S_{\omega}, \omega): \omega \in \Omega(G)\}$. Then, there is a rayless normal tree in $G$ which refines $\mathcal{C}$. More precisely, for every connected component $C'$ of $G\setminus T$, there is $C \in \mathcal{C}$ such that $C'\subset C$.
\end{thm}


Compiling many consequences of the above result, the paper \cite{paracompacidade} also contains unified proofs of previous works in infinite graph theory. Below, by relying on the notion of envelopes, we draw one more application of Theorem \ref{aproximacao}: 

\begin{lemma}\label{l1}
    For a fixed graph $G$, let $X\subset \hat{V}(G)$ be a discrete subspace. Then, there is a subgraph $H$ of $G$ containing $X\cap V(G)$ and for which the following properties hold:
    \begin{itemize}
        \item[$i)$] $H$ has finite adhesion;
        \item[$ii)$] For every two different ends $\omega_1,\omega_2 \in X$, there are distinct connected components $C_1$ and $C_2$ of $G\setminus H$ such that $\omega_1 \in \overline{C_1}$ and $\omega_2\in \overline{C_2}$. 
    \end{itemize}
\end{lemma}
\begin{proof}
    Applying Zorn's Lemma, let $\mathcal{W}$ be a $\subseteq-$maximal family of disjoint rays whose ends belong to $\overline{X}\setminus X$. Note that $\mathcal{W} = \emptyset$ if $X$ is closed. Then, consider a connected envelope $U^*$ for the vertex set $U := (X\cap V(G))\cup \displaystyle \bigcup_{s\in \mathcal{W}}V(s)$.
    
    
    We claim that $\overline{U^*}\setminus U^* = \overline{U}\setminus U = \overline{X}\setminus X$, where the first equality follows from the definition of envelopes. In fact, an element of $\overline{X}\setminus X$ is an end $\omega\in \overline{X}$. In this case, any of its representatives $r$ must intersect infinitely many vertices of $U$: otherwise, $\mathcal{W}\cup \{r'\}$ would contradict the $\subseteq-$maximality of $\mathcal{W}$, whether $r'$ denotes a tail of $r$ such that $V(r')\cap U = \emptyset$. In particular, $C(S,\omega)\cap U \neq \emptyset$ for every finite set $S\subset V(G)$, proving that $\omega\in \overline{U}\setminus U$. Conversely, given $\omega\in \overline{U}\setminus U$, we must have $C(S,\omega)\cap U \neq \emptyset$ for every finite set $S\subset V(G)$. Since $\mathcal{W}$ is a family of disjoint rays, we may add finitely many vertices to a set $S\subset V(G)$ in order to assume that $s\setminus S$ is a tail of $s$ for every $s\in \mathcal{W}$. In this case, $C(S,\omega)\cap U$ must contain a vertex from $X\cap V(G)$ or the tail of a ray in $\mathcal{W}$, whose end belongs to $\overline{X}$. In other words, $\hat{V}(S,\omega)\cap \overline{X}\neq \emptyset$ for every finite set $S\subset V(G)$, verifying that $\omega\in \overline{X}\setminus X$. Once showed that $\overline{U^*}\setminus U^* = \overline{X}\setminus X$, it follows that every end $\omega \in X$ has a representative in some connected component of $G\setminus U^*$. If not, we would have $C(S,\omega)\cap U^*\neq \emptyset$ for every finite set $S\subset V(G)$.


    Now, let $D$ be a connected component of $G\setminus U^*$, so that it is finite the neighborhood $N(D) = \{v \in U^*: v \text{ has a neighbor in }D\}$. In particular, $\Omega(D) = \overline{D}\cap \Omega(G)$ is an open subspace of $\Omega(G)$. For every end $\omega \in \Omega(D)$, then either $\omega\in X$ or $\omega\notin \overline{X}$, since $\hat{V}(N(D),\omega)\cap U^* = \emptyset$ and $\overline{U^*}\setminus U^* = \overline{X}\setminus X$. In the first case, there is a finite set $S_{\omega}\subset V(G)$ such that $\hat{V}(S_{\omega},\omega)\cap X = \{\omega\}$, because $X$ is discrete. In the latter one, there is a finite set $S_{\omega}\subset V(G)$ for which $\hat{V}(S_{\omega},\omega)\cap X =\emptyset$. Clearly $\Omega(D)$ is thus covered by the family $\{\hat{V}(S_{\omega},\omega): \omega \in \Omega(D)\}$. Therefore, Theorem \ref{aproximacao} guarantees the existence of a rayless normal tree $T_D$ in $D$ with the following property: for every connected component $C$ of $D\setminus T_D$, there is an end $\omega\in \Omega(D)$ such that $C\subset \hat{V}(S_{\omega},\omega)$. In particular, by the choice of $S_{\omega}$, there is at most one element in $\overline{C}\cap X$. On the other hand, for every $\omega \in \omega(D)\cap X$ there is indeed a connected component $C_{\omega}^D$ of $D\setminus T_D$ such that $\omega \in \overline{C_{\omega}^D}$, once $T_D$ is a rayless normal tree. Then, consider the following family of connected subgraphs of $G$:

    $$\mathcal{K} = \{C_{\omega}^D: \omega \in \Omega(D)\cap X, D\text{ is a connected component of }G\setminus U^*\}.$$


    Finishing the proof, we will show that the graph $H$ claimed by the statement can be chosen as the one induced by $G\setminus \displaystyle \bigcup_{K\in \mathcal{K}}V(K)$. We first note that $H$ contains $U^*$, verifying the inclusion $X\cap V(G)\subset V(H)$. Moreover, there is no edges in $H$ connecting distinct pairs $C_{\omega_1}^{D_1},C_{\omega_2}^{D_2}\in \mathcal{K}$. After all, $C_{\omega_1}^{D_1}$ and $C_{\omega_2}^{D_2}$ are contained in distinct connected components of $G\setminus U^*$ if $D_1 \neq D_2$. If $D: = D_1 = D_2$, then $C_{\omega_1}^{D_1}$ and $C_{\omega_2}^{D_2}$ are distinct connected components of $D\setminus T_D$, because $\omega_1,\omega_2 \in X\cap \Omega(D)$ are different ends. In other words, $\mathcal{K}$ is also the set of connected components of $G\setminus H$. Moreover, for every $\omega \in X$ there is a connected component $D$ of $G\setminus U^*$ in which $\omega$ has a representative, meaning that $\omega \in C_{\omega}^D$. To summarize, we just verified the second item of the statement. Finally, $H$ has finite adhesion since so do the envelope $U^*$ and the rayless normal trees of $\{T_D : D\text{ is a connected component of }G\setminus U^*\}$. 
\end{proof} 


In particular, from the second paragraph in the above proof we can extract the following observation: if $X\subset \Omega(G)$ is a closed set of ends in a graph $G$, there is a subset $U\subset V(G)$ of finite adhesion such that $\overline{U}\setminus U = \overline{X}$. If $X = \overline{A}$ for a given end set $A\subset \Omega(G)$ and we fix $B\subset \Omega(G)$ such that $\overline{A}\cap B = \emptyset$, then every representative $r$ of an end in $B$ has a tail in a connected component $C_{[r]}$ of $G\setminus U$. When we further assume that $A\cap \overline{B} = \emptyset$, then an envelope for the set $\{v \in U : v \text{ has a neighbor in }C_{\omega} \text{ for some end }\omega\in B\}$ plays the role of a graph $H$ as in the statement below:


\begin{lemma}[\cite{polat}, Theorem 1.2]\label{l2}
    Let $A,B\subset \Omega(G)$ be two sets of end in a graph $G$ which are topologically separated, i.e., such that $\overline{A}\cap B = A \cap \overline{B} = \emptyset$. Then, there is a subgraph $H$ of $G$ satisfying the two properties below:
    \begin{itemize}
        \item[$i)$] $H$ has finite adhesion and $\overline{H}\cap (A\cup B) = \emptyset$;
        \item[$ii)$] Given two rays $r$ and $s$ in $G$ such that $[r]\in A$ and $[s]\in B$, their tails in $G\setminus H$ are contained in distinct connected components.
    \end{itemize}
\end{lemma}


The subgraph $H$ in the above result is understood as an object that separates $A$ and $B$ combinatorially. Its existence was first established by Polat in \cite{polat}, who relied in his theory of \textit{multiendings} to that aim. According to his work, $H$ can be taken as an $A-B$ separator in the Menger-type result stated by Theorem \ref{mengerpolat}. In fact, Polat's definition of separators reads as follows: 


\begin{definition}[Separators]\label{separator}
    Fix a graph $G$ and two end subsets $A,B\subset \Omega(G)$. Then, we call $S\subset V(G)$ an $A-B$ separator if $\overline{S}\cap (A\cup B) = \emptyset$ and there is no graphic $A-B$ path in $G\setminus S$.  In particular, every end from $A\cup B$ has a representative ray in $G\setminus S$, but this subgraph contains no $A-B$ path as in Definition \ref{paths}.     
\end{definition}


However, in this paper we are mainly interested in edge-connectivity results. Therefore, we shall now restate the conclusions of Lemmas \ref{l1} and \ref{l2} for studying suitable sets of edge-ends rather than subspaces of $\Omega(G)$. As a familiar tool around these discussions, we recall that the \textbf{line graph} $G'$ of a given graph $G$ is defined by setting $V(G') = E(G)$ and 
\begin{center}
    $ef \in E(G')$ if, and only if, $e$ and $f$ are \textbf{adjacent edges},
\end{center} namely, $e$ and $f$ share a common endpoint. 


 The following observation, then, easily translates paths in $G$ to paths in $G'$ and conversely:

\begin{lemma}\label{41}
	Fix $e$ and $f$ two non-adjacent edges in a connected graph $G$. Therefore, the two statements below hold:
	\begin{itemize}
		\item[i)] If $v_0v_1v_2\dots v_n$ is a path in $G$ such that $e = v_0v_1$ and $f = v_{n-1}v_n$, then $(v_{i}v_{i+1})_{i < n}$ defines a path in $G'$ connecting $e$ and $f$;
		\item[ii)] Conversely, if $e_0e_1e_2\dots e_n$ is a vertex-minimal path in $G'$ connecting the edges $e = e_0$ and $f = f_n$ of $G$, then there is a path $v_0v_1v_2\dots v_{n+1}$ in $G$ such that $e_i = v_iv_{i+1}$ for each $0 \leq i \leq n$.  
	\end{itemize}
\end{lemma}
\begin{proof}
	The first item is trivial. In fact, if $v_0v_1v_2\dots v_n$ is a path in $G$, then $v_{i_1}v_i$ and $v_iv_{i+1}$ are distinct and adjacent edges for every $1 < i < n$, defining a path in $G'$.
	
	
	Now, suppose that $e_0e_1e_2\dots e_n$ is a minimal path in $G'$ connecting the edges $e = e_0$ and $f = f_n$. Hence, for each $i < n$, $e_i$ and $e_{i+1}$ share an endpoint $v_{i+1}$ in $G$, since those are adjacent edges. Moreover, $v_{i+1}\neq v_{j+1}$ for every $j < n$ distinct from $i$. Otherwise, if $v_{i+1} = v_{j+1}$ and $i < j$ for instance, the edges $e_i$ and $e_{j+1}$ would be adjacent, so that the path $e_0e_1 \dots e_i e_{j+1}\dots e_n$ would contradict the minimality of $e_0e_1e_2\dots e_n$.   
\end{proof}


Besides that, the first item in the above lemma also provides a natural identification of some ends of $\Omega_E(G)$ with elements of $\Omega(G')$. More precisely, given a ray $r = v_0v_1v_2\dots$, we denote by $\phi(r)$ the ray in $G'$ presented by $\phi(r) = \{v_iv_{i+1}\}_{i < \omega}$. We thus consider the induced map  
\begin{equation}\label{phi}
     \begin{array}{cccc}
       \Phi:  &  \Omega_E(G) & \to & \Omega(G')\\
         & [r]_E & \mapsto & [\phi(r)]
    \end{array} 
\end{equation}


It is not difficult to prove that $\Phi$ is a well-defined injection, as it is also claimed by Lemma 1 of \cite{edgeends}. However, one cannot ensure that $\Phi$ is surjective, since infinitely many edges incident to a vertex $v$ of $G$ may define a ray in $G'$ whose end belongs to $\Omega(G')\setminus \Omega_E(G)$. In other words, besides being a natural inclusion of $\Omega_E(G)$ in $\Omega(G')$, the map $\Phi$ might not identify these two spaces. Thus, unlike in finite graph theory (where many results of edge-connectivity are restrictions of vertex-analogous statements to line graphs), the interplay between $G$ and $G'$ needs to be analyzed with additional caution if they contain vertices of infinite degree. Considering that, the edge-analogous of Lemma \ref{l1} reads as follows:

\begin{lemma}\label{discreto}
    For a fixed graph $G$, let $T\subset \hat{E}(G)$ be a discrete subset. Then, there is a family $\{C_{\omega}:\omega \in T\cap \Omega_E(G)\}$ of connected subgraphs of $G$ for which the following properties are verified:
    \begin{itemize}
        \item[$i)$] $C_{\omega_1}\cap C_{\omega_2} = \emptyset$ if $\omega_1 \neq \omega_2$;
        \item[$ii)$] For every $\omega \in T\cap \Omega_E(G)$, the cut $F_{\omega}:=\delta(C_{\omega})$ is finite and $\hat{E}(F_{\omega},\omega)\cap T = \{\omega\}$.
    \end{itemize}
\end{lemma}
\begin{proof}
    For every $v\in T\cap V(G)$, let $K_v = \delta(\{v\})$ denote the set of edges in $G$ which have $v$ as an endpoint. Therefore, $K_v$ is a vertex set that induces a clique in the line graph $G'$. We will then show that the set $T' =\displaystyle  \Phi(T\cap \Omega_E(G))\cup \bigcup_{v \in T\cap V(G)}V(K_v)$ is also discrete in $\hat{V}(G')$, where $\Phi$ is the map given by (\ref{phi}). Since vertices are isolated points in this topological space, it is sufficient to conclude that there is no edge-end $[r]_E\in T\cap \Omega_E(G)$ such that $[\phi(r)]\in \overline{T'}$. In fact, for every edge-end $[r]_E \in T\cap \Omega_E(G)$, there is a finite set $F\subset E(G)$ such that $\hat{E}(F,[r]_E)\cap T = \{[r]_E\}$, because $T$ is a discrete subspace of $\hat{E}(G)$. Regarding $F$ as a finite vertex subset of $G'$, we must have $\hat{V}(F,[\phi(r)])\cap T' = \emptyset$. Otherwise, one of the following items is verified:
    \begin{itemize}
        \item If there is $e \in \hat{V}(F,[\phi(r)])\cap T'\cap V(G')$, then $e$ is an edge in $G$ which is incident to some vertex $v\in T$. In this case, by Lemma \ref{41}, any vertex-minimal path in $G'\setminus F$ connecting $e$ to a tail of $\phi(r)$ would describe a path in $G$ connecting $v$ to a tail of $r$, but avoiding the edges in $F$. This contradicts the fact that $v\notin \hat{E}(F,[r]_E)\cap T$;
        \item If there is an end $\omega \in \hat{V}(F,[\phi(r)])\cap T'\cap \Omega(G')$, then $\omega = [\phi(s)]$ for some ray $s$ in $G$ such that $[s]_E\in T$. In this case, again by Lemma \ref{41}, any vertex-minimal path in $G'\setminus F$ connecting the tails of $\phi(r)$ and $\phi(s)$ would describe a path in $G$ connecting the tails of $r$ and $s$, but avoiding the edges of $F$. Since $\hat{E}(F,[r]_E)\cap T = \{[r]_E\}$, we then must have $[r]_E = [s]_E$. Therefore, $\omega = [\phi(r)]$. 
    \end{itemize}
    Hence, $T'$ is indeed a discrete subspace of $\hat{V}(G')$. Then, let $H$ be the subgraph of $G'$ claimed by Lemma \ref{l1} when considering $X = T'$. Therefore, for each edge-end $\omega \in \Omega_E(G)$ there is a connected component $K_{\omega}$ of $G'\setminus H$ such that $\Phi(\omega)\in \overline{K_{\omega}}$. Observing that $V(K_{\omega})\subset V(G') = E(G)$, we define $C_{\omega}$ as the subgraph of $G$ induced by the vertices which are endpoints of some edge from $V(K_{\omega})$. In particular, $C_{\omega}$ is connected by Lemma \ref{41} and must contain all but finitely many edges of any ray $r$ of $G$ such that $\omega = [r]_E$, since a tail of $\phi(r)$ in $G'$ is contained in $K_{\omega}$ in this case.   


    We now note that $C_{\omega_1}\cap C_{\omega_2} = \emptyset$ if $\omega_1 \neq \omega_2$. In fact, if there were a vertex $v \in C_{\omega_1}\cap C_{\omega_2}$, then $v$ would be a common endpoint of some given edges $e \in V(K_{\omega_1})$ and $f \in V(K_{\omega_2})$. This means that $e$ and $f$ would be adjacent edges, belonging (as vertices) to the same connected component $K_{\omega_1} = K_{\omega_2}$ of $G'\setminus H$. However, this would contradict the injectivity of $\Phi$, once Lemma \ref{l1} asserts that $K_{\omega_1} = K_{\omega_2}$ if, and only if, $\Phi(\omega_1) = \Phi(\omega_2)$.


    Finally, for a fixed $\omega \in \Omega_E(G)$, we observe that the following equivalences hold for any edge $e\in E(G)$:

    \begin{align*}
        e \in \delta(C_{\omega}) & \iff e=xy \text{ for some }x\in C_{\omega}\text{ and some }y\in V(G)\setminus C_{\omega} \\
        & \iff e=xy\text{ for some endpoint }x\text{ of an edge }f\in V(K_{\omega})\text{ and some }y \notin V(C_{\omega}) \\
        & \iff e \text{ is adjacent to some }f\in V(K_{\omega})\text{ but do not belong to }K_{\omega}\\
        & \iff e \text{ belongs to }N(K_{\omega}) :=\{h \in H : h\text{ has a neighbor in }K_{\omega}\} \text{ as a vertex of }G'.
    \end{align*}


    In other words, we just verified the equality $\delta(C_{\omega}) = N(K_{\omega})$. From the finite adhesion of $H$, it follows that $F_{\omega}:=\delta(C_{\omega})$ is finite. Hence, it remains to prove that $\hat{E}(F_{\omega},\omega)\cap T \cap V(G) = \emptyset$, once we already concluded that $\omega \in \overline{C_{\omega}} \subset \hat{E}(F_{\omega},\omega)$ and that $C_{\omega}\cap C_{\eta} = \emptyset$ if $\eta \in \Omega_E(G) \cap T\setminus\{\omega\}$. For instance, thus, suppose that there is a vertex $v \in \hat{E}(F_{\omega},\omega)\cap T \cap V(G)$. Then, there is a path $P$ in $C_{\omega}$ connecting $v$ to a tail of a ray $r$ such that $\omega = [r]_E$. In this case, Lemma \ref{41} claims that the edges of $P$ define a path in $G'$ which connects an element from $K_v = \delta(\{v\})$ to a tail of $\phi(r)$. However, contradicting the fact that $E(P)\subset E(C_{\omega})$, this path must intersect $N(K_{\omega}) = \delta(C_{\omega})$, because $K_v\subset L$ and $[\phi(r)] = \Phi(\omega) \in \overline{K_{\omega}}$. 
\end{proof}


Revisiting the previous section, we observe that Lemma \ref{discreto} played a central role in the proof of Theorem \ref{t1}. In addition, from the first paragraphs of the above proof we can extract the following conclusion: if $[\phi(r)]\in \overline{\Phi(T)}$ for some edge-end $[r]_E\in \Omega_E(G)$ and some set $T\subset \Omega_E(G)$, then $[r]_E \in \overline{T}$. In other words, the function $\Phi$ as defined in (\ref{phi}) is an open map over its range. Although simple, this remark is useful for translating Lemma \ref{l2} to its edge-related version:

\begin{lemma}\label{aresta-separador}
    Let $A,B\subset \Omega_E(G)$ be two sets of edge-ends in a graph $G$ which are topologically separated, i.e., such that $\overline{A}\cap B = A\cap \overline{B} = \emptyset$. Then, there is an edge set $H\subset E(G)$ with the following properties:
    \begin{itemize}
        \item[$i)$] For every edge-end $\omega \in \Omega_E(G)\in A \cup B$, there is a finite set $F\subset E(G)$ such that the connected component $C_E(F,\omega)$ of $G\setminus F$ does not contain an edge from $H$;
        \item[$ii)$] For given rays $r$ and $s$ in $G$ such that $[r]_E\in A$ and $[s]_E \in B$, their tails in $G\setminus H$ are contained in distinct connected components.  
    \end{itemize}
\end{lemma}
\begin{proof}
    Let $\Phi$ be the function defined as in (\ref{phi}), which is an open map over its range. Therefore, $\overline{\Phi(A)}\cap \Phi(B) = \Phi(A)\cap \overline{\Phi(B)} = \emptyset$ by the main hypothesis over the sets $A$ and $B$. If $G'$ denotes the line graph of $G$, let $H$ be its subgraph of finite adhesion claimed by Lemma \ref{l2}. More precisely, $H$ verifies the two conditions below:
    \begin{itemize}
        \item Considering the topology of $\hat{V}(G')$, we have $\overline{H}\cap (\Phi(A)\cup \Phi(B)) = \emptyset$;
        \item Given two rays $r$ and $s$ in $G$ such that $[r]_E \in A$ and $[s]_E \in B$, the tails of $\phi(r)$ and $\phi(s)$ in $G'\setminus H$ are contained in distinct connected components. 
    \end{itemize}
   In particular, for an edge-end $\omega = [r]_E$, the connected component $K_{\omega}$ of $G'\setminus H$ containing a tail of $\phi(r)$ has finite neighborhood. In other words, the set $N(K_{\omega}) = \{h \in H : h \text{ has a neighbor in }K_{\omega}\}$ is finite. Nevertheless, we can see $F_{\omega}:=N(K_{\omega})$ as a finite edge set in $G$, once $V(H)\subset E(G)$ by the definition of line graph. In addition, there is no edge $h\in V(H)$ in $C_E(F_{\omega},\omega)$: otherwise, there would exist in this subgraph of $G$ a path $P$ connecting a tail of $r$ to an endpoint of $h$, but this would imply by Lemma \ref{41} in the existence of a path $Q$ in $G'\setminus N(K_{\omega})$ connecting $\phi(r)$ to $h$. Hence, the edge set $V(H)\subset E(G)$ verifies the first property required by the statement.

    
    Therefore, we shall finish the proof by arguing how $V(H)$ fulfills the property $ii)$ as well. To this aim, fix two rays $r$ and $s$ in $G$ such that $[r]_E \in A$ and $[s]_E \in B$. Suppose for a contradiction that their tails are contained in a same connected component of $G\setminus V(H)$, meaning that there is a path in $G$ connecting these rays but avoiding the edges from $V(H)$. However, by Lemma \ref{41}, this would define a path in $G'\setminus H$ which connects the rays $\phi(r)$ and $\phi(s)$, contradicting the above second item for the choice of $H$.
\end{proof}


In its turn, the two properties of the edge set $H$ claimed by Lemma \ref{aresta-separador} inspire an edge-related version of Definition \ref{separator}. More precisely, under the assumption that $A,B\subset \Omega_E(G)$ are topologically separated as in the above result, we just verified that the combinatorial structure below indeed exists within the graph $G$:


\begin{definition}[Separators - edge version]\label{eseparator}
     Fix a graph $G$ and two end subsets $A,B\subset \Omega(G)$. Then, we call $S\subset E(G)$ an $A-B$ separator if there is no graphic $A-B$ path in $G\setminus S$ and, for every edge-end $\omega \in \Omega_E(G)$, there exists a finite set $F\subset E(G)$ such that no edge from $S$ belongs to $C_E(\omega, F)$. In particular, every edge-end from $A\cup B$ has a representative ray in $G\setminus S$, but this subgraph contains no $A-B$ path as in Definition \ref{epaths}.  
\end{definition} 


Therefore, the settings of Theorem \ref{t2} were finally fully discussed: for given sets $A,B\subset \Omega_E(G)$, the $A-B$ paths are those introduced by Definition \ref{epaths}, while the $A-B$ separators are the ones as in Definition \ref{eseparator}. Moreover, Corollary \ref{finiteseparation} already established this Menger-type result precisely when $A$ and $B$ can be separated by finitely many edges. The general case, then, is covered by the proof below:


\begin{reptheorem}{t2}[Erd\H{o}s-Menger Theorem for edge-ends]\label{edge-ends}
	Let $A,B \subset \Omega_E(G)$ be two set of ends such that $\overline{A}\cap B = A\cap \overline{B} = \emptyset$. Fix a family $\mathcal{P}$ of edge-disjoint $A-B$ paths with maximum size. Consider an $A-B$ separator $F$ of minimum cardinality. Then, $|\mathcal{P}| = |F|$.
\end{reptheorem} 
\begin{proof}[Revisited proof of Theorem 2.7 from \cite{polat}]
    Let $A,B\subset \Omega_E(G)$ be the two sets of edge-ends such that $\overline{A}\cap B =A\cap \overline{B} = \emptyset$. Then, consider the cardinal $$\kappa  = \sup \{|\mathcal{P}|: \mathcal{P}\text{ is a family of edge-disjoint }A-B\text{ paths}\}.$$

   If $\kappa$ is finite, the above supremum is clearly attained, even by a family $\mathcal{P}$ that we can suppose to be maximal. If $\kappa$ is infinite, we fix $\mathcal{P}$ a maximal family of edge-disjoint $X-Y$ paths and consider the following two cases:
   \begin{itemize}
       \item Suppose that $\kappa$ is countable but $\mathcal{P}$ is finite. Then, for every $n \in \mathbb{N}$, there is $\mathcal{P}_n$ a family of edge-disjoint $A-B$ paths such that $|\mathcal{P}_n|\geq 2n\cdot|\mathcal{P}|$, by definition of $\kappa$. However, since $\mathcal{P}$ is maximal, $Q\cap \displaystyle \bigcup_{P\in \mathcal{P}}E(P)\neq \emptyset$. In particular, there is $r_n \in \mathcal{P}$ an $A-B$ path such that $|\{Q\in \mathcal{P}_n : Q\cap E(r_n)\neq \emptyset\}|\geq 2n$. Considering a subsequence of $\{r_n\}_{n \in \mathbb{N}}$ if necessary, we can even assume that $r_n = r_m =: r$ for every $n,m\in\mathbb{N}$. Thus, by being infinite, $r$ is either a ray or a double ray. In this latter case, unless by passing $\{\mathcal{P}_n\}_{n\in\mathbb{N}}$ to a subsequence, $r$ contains a ray $r'$ such that $|\{Q\in\mathcal{P}_n : Q\cap E(r')\neq \emptyset\}|\geq n$. Therefore, we can assume that $r$ is a ray such that  $|\{Q\in\mathcal{P}_n : Q\cap E(r)\neq \emptyset\}|\geq n$. Since $r\in \mathcal{P}$ is a $A-B$ path, we have $[r]_E\in A\cup B$. On the other hand, for a finite set $F\subset E(G)$, a tail of $r$ intersects an $A-B$ path $Q$ from $\mathcal{P}_{|F|+1}$. Hence, in $C_E(F,[r]_E)$ there are representatives of ends in $A$ and $B$. This proves that $[r]_E\in \overline{A}\cap \overline{B}$, contradicting the topological separation hypothesis. Therefore, $|\mathcal{P}| = \aleph_0$ if $\kappa = \aleph_0$;  
       \item Suppose that $\kappa$ is uncountable but $|\mathcal{P}|<\kappa$. In particular, there is a family $\mathcal{Q}$ of edge-disjoint $A-B$ paths such that $|\mathcal{P}|< |\mathcal{Q}|$. However, $|\mathcal{P}| = \displaystyle \left|\bigcup_{P\in \mathcal{P}}E(P)\right|$, because $A-B$ paths are countable. Hence, there is $Q\in \mathcal{Q}$ such that $Q\cap\displaystyle \bigcup_{P\in \mathcal{P}}E(P) = \emptyset$. In this case, $\mathcal{P}\cup \{Q\}$ contradicts the maximality of $\mathcal{P}$. Thus, we must have $|\mathcal{P}| = \kappa$ if $\kappa$ is uncountable.
   \end{itemize}
   In any case, $\mathcal{P}$ has maximum size. On the other hand, once $\overline{A}\cap B = A\cap \overline{B} = \emptyset$, we previously discussed how Lemma \ref{aresta-separador} claims the existence of an $A-B$ separator $S$. Considering that, for each $P \in \mathcal{P}$ we define a finite set of edges $T_P$ as follows:
	\begin{itemize}
		\item If $P$ is a finite path, we set $T_P = E(P)$;
		\item If $P$ is a ray of class $\omega \in A\cup B$, consider $C_{\Phi(\omega)}'$ the connected component of $G'\setminus S$ in which $\phi(P)$ has a tail. According to Lemma \ref{41}, $C_{\Phi(\omega)}'$ is indeed the line graph of the connected component $C_{\omega}$ of $G\setminus S$ in which $P$ has a tail. Moreover, by property $(\dagger)$, the neighborhood of $C_{\Phi(\omega)}'$ in $S$ is as a finite set $S_{\omega}$. Considering $S_{\omega}$ as an edge subset of $G$, note that $S_{\omega}$ is the cut $\delta(C_{\omega})$. In this case, we define $T_P$ to be $S_{\omega}\cup(E(P) \cap S)$;
		\item Now, suppose that $P$ is a double ray such that one half-ray $P_1$ has an edge-end $\omega_1 \in A$ and the other, named $P_2$, has an edge-end $\omega_2\in B$. Denote by $C_{\omega_1}$ and $C_{\omega_2}$ the connected components of $G\setminus S$ in which $P_1$ and $P_2$ have their tails, respectively. As in the previous case, the cuts $S_{\omega_1} = \delta(C_{\omega_1})$ and $S_{\omega_2} = \delta(C_{\omega_2})$ are finite subsets of $S$. Then, we set $T_P = S_{\omega_-}\cup S_{\omega_+} \cup (E(P) \cap S)$.
	\end{itemize} 

By the maximality of $\mathcal{P}$, any graphic $A-B$ path $R$ must intersect $F = \displaystyle \bigcup_{P\in \mathcal{P}}T_P $: in $S\cap E(P)$ for some $P\in \mathcal{P}$ or in $S_{\omega}$ for some edge-end $\omega\in \overline{R}$. In other words, $F$ is an $A-B$ separator. Moreover, $F$ has the size of the family $\mathcal{P}$ if $\kappa = |\mathcal{P}|$ is infinite, since $T_P$ is finite for every $P\in\mathcal{P}$. By the same reason, $F$ is finite if so is $\kappa$. In this latter case, Theorem \ref{edge-ends} is reduced to Corollary \ref{finiteseparation}.     
\end{proof}


Closely analyzing this proof, we remark that the existence of an $A-B$ separator was established by Lemma \ref{aresta-separador}, which, in its turn, was obtained from Lemma \ref{l2} via the map $\Phi$ defined in (\ref{phi}). Nevertheless, this suggests that we could conclude Theorem \ref{edge-ends} by directly applying Theorem \ref{mengerpolat} to the line graph of $G$. In fact, it is possible to construct a maximum-sized family $\mathcal{P}$ of edge-disjoint $A-B$ paths, as well as a suitable minimum $A-B$ separator $S$, based on the corresponding objects between $\Phi(A)$ and $\Phi(B)$ in $G'$. However, this requires a clever analysis of which rays in some $\Phi(A)-\Phi(B)$ paths are equivalent to rays in $G'$ that arise from infinitely many edges sharing a common vertex in $G$. If written in details, this approach to Theorem \ref{edge-ends} would be as lengthy as the one just presented, but it would not clarify the role played by its main topological hypothesis over $A$ and $B$.  

\section{Acknowledgments}
\paragraph{}

We thank the support of Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP). The first and the second named authors were sponsored through grant numbers 2023/00595-6 and 2021/13373-6 respectively.


\bibliography{referencias}
\bibliographystyle{plain}

\end{document}

\typeout{get arXiv to do 4 passes: Label(s) may have changed. Rerun}

\begin{thebibliography}{10}

\bibitem{erdosmenger}
Ron Aharoni and Eli Berger.
\newblock Menger’s theorem for infinite graphs.
\newblock {\em Inventiones mathematicae}, 176:1--62, 2005.

\bibitem{compaulo}
Leandro Aurichi, Lucas Real, and Paulo~Magalh{\~{a}}es J{\'{u}}nior.
\newblock Topological remarks on end and edge-end spaces.
\newblock \href{https://arxiv.org/abs/2404.17116}{arXiv:2111.17116}, 2023.

\bibitem{versaodiestel}
Henning Bruhn, Reinhard Diestel, and Maya Stein.
\newblock Menger's theorem for infinite graphs with ends.
\newblock {\em Journal of Graph Theory}, 50(3):199--211, 2005.

\bibitem{bruhn}
Henning Bruhn and Maya Stein.
\newblock Maclane's planarity criterion for locally finite graphs.
\newblock {\em Journal of Combinatorial Theory, Series B}, 96(2):225--239,
  2006.

\bibitem{finitos-um}
Henning Bruhn and Maya Stein.
\newblock On end degrees and infinite cycles in locally finite graphs.
\newblock {\em Combinatorica}, 27:269--291, 05 2007.

\bibitem{survey}
Reinhard Diestel.
\newblock Locally finite graphs with ends: A topological approach, {I}{I}.
  applications.
\newblock {\em Discrete Mathematics}, 310(20):2750--2765, 2010.

\bibitem{freudenthal}
Hans Freudenthal.
\newblock Neuaufbau der endentheorie.
\newblock {\em Annals of Mathematics}, 43(2):261--279, 1942.

\bibitem{georgakopoulos}
Agelos Georgakopoulos.
\newblock Graph topologies induced by edge lengths.
\newblock {\em Discrete Mathematics}, 311(15):1523--1542, 2011.
\newblock Infinite Graphs: Introductions, Connections, Surveys.

\bibitem{edgeends}
Geňa Hahn, François Laviolette, and Jozef {\v{S}}ir{\'{a}}{\v{n}}.
\newblock Edge-ends in countable graphs.
\newblock {\em Journal of Combinatorial Theory, Series B}, 70(2):225--244,
  1997.

\bibitem{halin}
Rudolf Halin.
\newblock Über unendliche wege in graphen.
\newblock {\em Mathematische Annalen}, 157:125--137, 1964/65.

\bibitem{hopf}
Heinz Hopf.
\newblock Enden offener räume und unendliche diskontinuierliche gruppen.
\newblock {\em Commentarii Mathematici Helvetici}, 16:81--100, 1943.

\bibitem{jacobs}
Raphael Jacobs, Attila Joó, Paul Knappe, Jan Kurkofka, and Ruben Melcher.
\newblock The lovász-cherkassky theorem for locally finite graphs with ends.
\newblock {\em Discrete Mathematics}, 346(12):113586, 2023.

\bibitem{joo}
Attila Joó.
\newblock The {L}ovász-{C}herkassky theorem in countable graphs.
\newblock {\em Journal of Combinatorial Theory, Series B}, 159:1--19, 2023.

\bibitem{joogeral}
Attila Joó.
\newblock The {L}ovász-{C}herkassky theorem in infinite graphs.
\newblock \href{https://arxiv.org/abs/2311.06611}{arXiv:2311.06611}, 2023.

\bibitem{paracompacidade}
Jan Kurkofka, Ruben Melcher, and Max Pitz.
\newblock Approximating infinite graphs by normal trees.
\newblock {\em Journal of Combinatorial Theory, Series B}, 148:173--183, 2021.

\bibitem{representacao}
Jan Kurkofka and Max Pitz.
\newblock A representation theorem for end spaces of infinite graphs.
\newblock \href{https://arxiv.org/abs/2111.12670}{arXiv:2111.12670}, 2023.

\bibitem{PolatFrances}
Norbert Polat.
\newblock Aspects topologiques de la s{\'e}paration dans les graphes infinis.
  {I}.
\newblock {\em Mathematische Zeitschrift}, 165:73--100, 1979.

\bibitem{polat}
Norbert Polat.
\newblock A {M}engerian theorem for infinite graphs with ideal points.
\newblock {\em Journal of Combinatorial Theory, Series B}, 51(2):248--255,
  1991.

\bibitem{stein}
Maya Stein.
\newblock Arboricity and tree-packing in locally finite graphs.
\newblock {\em Journal of Combinatorial Theory, Series B}, 96:302--312, 03
  2006.

\end{thebibliography}