Edge-End Spaces in Infinite Graphs

• Edge-end spaces are topological spaces defined by classifying rays in infinite graphs via finite edge separations, differing from traditional vertex-end spaces.
• They exhibit unique properties such as metrization, specific covering invariants, and representations using order trees and tree-cut decompositions.
• Their study includes compactness criteria and duality methods through edge-direction spaces and Boolean-algebraic models, offering novel insights into infinite graph theory.

Edge-end spaces are topological spaces obtained from the edge-end structure of infinite graphs. For a graph $G$, the space $\Omega_E(G)$ consists of equivalence classes of rays under finite edge-separation, equipped with the natural edge-end topology. In the locally finite case, edge-ends coincide with the usual graph ends, but in general the edge-end relation is coarser, and the resulting spaces form a proper subfamily of ordinary end spaces. Recent work has given purely topological descriptions of this class, metrization and covering theorems, and representation results through end spaces, order trees, tree-cut decompositions, and edge-direction spaces (Aurichi et al., 2024, Real, 24 Aug 2025, Pitz, 22 Jul 2025).

1. Definition and basic construction

Let $G$ be a graph. Two rays $R,R'$ in $G$ are edge-equivalent if for every finite set $F \subseteq E(G)$, there exist tails $T,T'$ of $R,R'$, respectively, lying in the same connected component of $G \setminus F$. An edge-end is an equivalence class of a ray under this relation, and the set of all edge-ends is denoted by $\Omega_E(G)$ (Carvalho et al., 12 Oct 2025).

The topology on $\Omega_E(G)$0 is defined by finite edge deletions. One standard basis description uses the sets $\Omega_E(G)$1, where $\Omega_E(G)$2 is a finite set of edges and $\Omega_E(G)$3 is a non-rayless component of $\Omega_E(G)$4. An equivalent notation fixes an edge-end $\Omega_E(G)$5 and writes

$\Omega_E(G)$6

where $\Omega_E(G)$7 is the unique component of $\Omega_E(G)$8 containing tails of the rays representing $\Omega_E(G)$9 (Boska et al., 24 Mar 2025).

The basic contrast with ordinary end spaces is that classical ends use finite vertex sets as separators, while edge-ends use finite edge sets. In locally finite graphs, the two notions agree. In general, the edge-end relation is coarser, so a single edge-end may identify several ordinary ends. A concrete example is the graph obtained from a two-way infinite path by adjoining one extra vertex adjacent to every path vertex: it has two ends but only one edge-end (Aurichi et al., 2024).

2. Relation to ordinary end spaces

A central structural fact is that every edge-end space can be realized as an ordinary end space of another graph. More precisely, for every graph $G$0, there exists a graph $G$1 such that $G$2 is homeomorphic to $G$3. One construction expands each vertex that edge-dominates a ray into a clique so that the edge-end topology of $G$4 becomes the usual end topology of the modified graph $G$5 (Aurichi et al., 2024).

The converse fails. There exists a graph $G$6 such that $G$7 is not homeomorphic to $G$8 for any graph $G$9. Equivalently, if

$R,R'$0

then

$R,R'$1

This strict containment isolates edge-end spaces as a genuine subclass inside the broader universe of end spaces (Aurichi et al., 2024).

This separation is topologically significant. It shows that replacing vertex-separation by edge-separation is not merely a change of language; it changes the class of realizable spaces. A common simplification—accurate only for locally finite graphs—is to treat edge-ends and ordinary ends as interchangeable. Outside that setting, the edge-based theory has its own representation theorems, obstruction phenomena, and topological invariants.

3. Purely topological characterizations

The recent classification of edge-end spaces is formulated in terms of clopen subbases. For ordinary end spaces, Pitz showed that a topological space $R,R'$2 is homeomorphic to the end space of some graph if and only if it admits a clopen subbase $R,R'$3 that is nested, noetherian, and hereditarily complete; moreover, the corresponding order tree can be chosen special iff $R,R'$4 is $R,R'$5-disjoint (Pitz, 2023).

For edge-end spaces, the decisive extra condition is the singleton intersection property. The main theorem of "A subbase property for describing edge-end spaces" states that a topological space $R,R'$6 is homeomorphic to the edge-end space of some graph if and only if it admits a clopen subbase $R,R'$7 that is nested, noetherian, hereditarily complete, and satisfies: $R,R'$8 Here nested means that if $R,R'$9 and $G$0, then $G$1 or $G$2; noetherian means that there is no infinite strictly ascending chain; and hereditarily complete means that the property passes to all closed subspaces (Real, 24 Aug 2025).

This theorem gives a purely topological description of the class of edge-end spaces. It also identifies the precise difference from the vertex-end case: edge-end spaces replace $G$3-disjointness by the singleton intersection property. The latter is strictly stronger than $G$4-disjointness, so all edge-end spaces are vertex-end spaces, but not vice versa. The proof uses representations as ray spaces of order trees together with a trimming method that restricts the order-tree model to the edge-end setting (Real, 24 Aug 2025).

4. Metrization and covering properties

The metrization theory of edge-end spaces is now explicit. The paper "A metrization theorem for edge-end spaces of infinite graphs" proves that the edge-end space of an infinite graph is metrizable if and only if it is first-countable. In fact, for $G$5, the following are equivalent: $G$6 is first-countable, $G$7 is metrizable, $G$8 is completely ultrametrizable, and $G$9 is homeomorphic to the end space $F \subseteq E(G)$0 of a tree $F \subseteq E(G)$1 (Pitz, 22 Jul 2025).

The proof is based on tree-cut decompositions of finite adhesion into $F \subseteq E(G)$2-edge blocks. This yields a representation of every edge-end space as a subspace $F \subseteq E(G)$3 with $F \subseteq E(G)$4 for a tree $F \subseteq E(G)$5, and first-countability becomes a combinatorial restriction on the relevant child cones of nodes in that tree model. The same work also gives a new proof that every infinite graph has a tree-cut decomposition of finite adhesion into its $F \subseteq E(G)$6-edge blocks (Pitz, 22 Jul 2025).

Covering properties admit sharp combinatorial descriptions. If $F \subseteq E(G)$7 denotes the set of non-rayless connected components of $F \subseteq E(G)$8, then

$F \subseteq E(G)$9

Moreover, the extent of $T,T'$0 equals its Lindelöf degree. The Rothberger property, $T,T'$1-compactness, and the Menger property also have graph-theoretic characterizations, and for edge-end spaces the Menger property is equivalent to $T,T'$2-compactness. The same paper shows that all edge-end spaces are $T,T'$3-spaces (Carvalho et al., 12 Oct 2025).

The following summary collects several of these equivalences.

Property of $T,T'$4	Characterization
Lindelöf degree $T,T'$5	For every finite $T,T'$6, $T,T'$7
Rothberger	Lindelöf and no copy of the Cantor space; equivalently Lindelöf and scattered
$T,T'$8-compact	Equivalent to Menger
Menger	Equivalent to $T,T'$9-compact
$R,R'$0-space	Always true

These results place edge-end spaces within classical covering theory while keeping the controlling invariants entirely combinatorial (Carvalho et al., 12 Oct 2025).

5. Compact edge-end spaces and edge-direction spaces

A larger boundary object associated with edge-connectivity is the edge-direction space. For each finite $R,R'$1, let $R,R'$2 be the set of connected components of $R,R'$3. An edge-direction is a coherent choice of component for every such finite $R,R'$4, and the edge-direction space is the inverse limit

$R,R'$5

Each edge-end determines an edge-direction by selecting the component that contains tails of its rays after every finite edge deletion, so $R,R'$6 embeds naturally into $R,R'$7 (Mar et al., 15 Jun 2026).

The space $R,R'$8 is compact, totally disconnected, and Hausdorff, whereas $R,R'$9 is not generally compact and embeds densely in $G \setminus F$0. There is also a line-graph description: $G \setminus F$1 where $G \setminus F$2 is the line graph of $G \setminus F$3. A Boolean-algebraic model is obtained from the finite pre-cut algebra

$G \setminus F$4

the quotient algebra $G \setminus F$5, and Stone duality: $G \setminus F$6 This identifies edge-direction spaces as Stone spaces of edge-cut algebras (Mar et al., 15 Jun 2026).

Compactness inside the class of edge-end spaces is governed by timid vertices, namely vertices that do not edge-dominate any ray. The paper "On edge-direction and compact edge-end spaces" proves the edge-analogue of Diestel’s compactness criterion: $G \setminus F$7 where $G \setminus F$8 is the set of timid vertices and $G \setminus F$9 is the collection of connected components of $\Omega_E(G)$0 that contain a ray. It also shows that every compact edge-end space can be represented as the edge-direction space of a connected graph (Boska et al., 24 Mar 2025).

The compact theory therefore has two complementary descriptions: a combinatorial one in terms of timid vertices and ray-containing components, and a representation-theoretic one through inverse limits, line graphs, and Stone duality.

6. Topological ends, non-dominated rays, and broader frameworks

The relation between edge-ends and topological ends is subtler than the relation between edge-ends and ordinary graph ends. Diestel and Kühn proved that topological ends are precisely the undominated graph ends, yielding a canonical embedding into the space of graph ends. For edge-ends, such an embedding does not exist in general (Aurichi et al., 16 Feb 2026).

The obstruction is purely combinatorial. Let $\Omega_E(G)$1 be the edge-equivalence class of a vertex $\Omega_E(G)$2. Then the following are equivalent: there exists a well-defined injective map $\Omega_E(G)$3 compatible with the canonical maps; whenever $\Omega_E(G)$4 are non-dominated and not vertex-equivalent they are not edge-equivalent; and for every finite vertex set $\Omega_E(G)$5 and every vertex $\Omega_E(G)$6, at most one component $\Omega_E(G)$7 of $\Omega_E(G)$8 is such that $\Omega_E(G)$9 contains a non-dominated ray (Aurichi et al., 16 Feb 2026).

When this embedding exists, its image is exactly the set of edge-ends containing a non-dominated ray. The paper calls these almost non-dominated edge-ends. This gives the edge-analogue of the Diestel–Kühn description only under an explicit graph-theoretic hypothesis, and it shows that distinct topological directions at infinity may collapse under edge-equivalence (Aurichi et al., 16 Feb 2026).

Broader generalizations place edge-end spaces inside a more universal theory of ends. In the connectoid framework, edge-ends correspond to ends of the connectoid $\Omega_E(G)$00, where

$\Omega_E(G)$01

This extends end-space methods beyond undirected graphs to directed graphs, bidirected graphs, hypergraphs, and finitary matroids (Bowler et al., 2024). A parallel development through Boolean algebras and Stone duality introduces edge analogues of tangles and identifies them with edge-directions, providing a compact space in which edge-ends live as a dense subspace (Mar et al., 15 Jun 2026).

Taken together, these results show that edge-end spaces are not merely an edge-based restatement of ordinary end theory. They form a distinct class of spaces with their own topological characterization, their own compactification theory, and their own interaction with domination, tree-cut structure, and Boolean-algebraic duality.