---
title: Half Graph in Graph Theory
url: https://www.emergentmind.com/topics/half-graph
type: topic
---

# Half Graph in Graph Theory

Searching arXiv for recent and foundational papers on “half graph” and related graph-theoretic usages.
Search query: half graph graph theory arXiv ordered bipartite half-graph semi-ladder co-matching
In graph theory, the standard named **half graph** is the ordered bipartite graph \(H_t\) on two disjoint vertex sets \(A=\{a_1,\ldots,a_t\}\) and \(B=\{b_1,\ldots,b_t\}\) with adjacency
\[
a_i b_j\in E \iff i\le j,
\]
or, in an equivalent convention, \(i<j\) [2602.07606]. In this sense a half graph is also called a **ladder**, and its bipartite adjacency matrix is triangular: the vertices \(b_i\) see precisely the later vertices \(a_j\). This makes half graphs canonical order-encoding patterns in graphs [2103.06218]. The same phrase, however, is used in several different ways elsewhere in the literature—notably for degree-balanced subgraphs of Eulerian graphs, for half-squares of bipartite graphs, and in a variety of “half-” constructions that are terminologically related but structurally distinct [1905.06678][1804.05793][1602.04316].

## 1. Classical ordered bipartite half graph

For \(\ell\ge 1\), pairwise different vertices
\[
a_1,\dots,a_\ell,\quad b_1,\dots,b_\ell
\]
form a **half graph** of order \(\ell\) if
\[
(b_i,a_j)\in E \iff i<j \qquad\text{for all }i,j\in[1,\ell].
\]
The order is the integer \(\ell\), equivalently the number of pairs \((a_i,b_i)\) or the length of either witnessing sequence [2103.06218]. In this form the half graph is one of several closely related bipartite patterns. The same sequences form a **semi-ladder** if \((b_i,a_j)\in E\) for all \(i<j\) and \((b_i,a_i)\notin E\) for all \(i\), and a **co-matching** if \((b_i,a_j)\in E \iff i\ne j\) [2103.06218].

The half graph is therefore the most rigid of these patterns: it realizes a strict total order by adjacency. The index order \(1<2<\cdots<\ell\) can be read off from neighborhoods, which is why these objects are said to “encode total orders in graphs” [2103.06218]. In the parameterized-complexity literature the same object is written \(H_t\), with
\[
a_i b_j\in E \Longleftrightarrow i\le j,
\]
and is studied together with the matching \(M_t\) and co-matching \(C_t\) as one of three unavoidable bipartite patterns in large twin-free structures [2602.07606].

This ordered bipartite meaning is the standard graph-theoretic one. It is also the sense most closely connected to structural graph theory, model-theoretic stability, and quantitative bounds in sparse graph classes.

## 2. Half graphs in graph powers and sparse classes

A major modern use of half graphs is in the study of graph powers. For a graph \(G\) and integer \(d\),
\[
V(G^d)=V(G),\qquad E(G^d)=\{(u,v)\mid \operatorname{dist}_G(u,v)\le d\}.
\]
The corresponding distance-\(d\) notion says that \(a_1,\dots,a_\ell,b_1,\dots,b_\ell\) form a **distance-\(d\) half graph** if
\[
\operatorname{dist}(b_i,a_j)\le d \iff i<j.
\]
Thus ordinary half graphs in \(G^d\) are equivalent to distance-\(d\) half graphs in \(G\) [2103.06218].

The central quantitative question is how large a half graph can occur in \(G^d\) when \(G\) belongs to a sparse class. The sharpest available asymptotic bounds currently depend strongly on the class:

| Sparse class | Lower bound on maximum order | Upper bound |
|---|---:|---:|
| Maximum degree at most \(\Delta\) | \(\Delta^{\Omega(d)}\) | \(\Delta^d+1\) |
| Planar graphs | \(2^{\lceil d/2\rceil}\) | \(d^{O(d)}\) |
| Pathwidth at most \(p\) | \(d^{p-O(1)}\) | \((dp)^{O(p)}\) |
| Treewidth at most \(t\) | \(2^{d^{\Omega(t)}}\) | \(d^{O(d^{t+1})}\) |
| Excluding \(K_t\) as a minor | \(2^{d^{\Omega(t)}}\) | \(d^{O(d^{t-1})}\) |

These bounds were established with nearly tight asymptotics for several classes, especially for planar graphs, where the upper bound \(d^{O(d)}\) substantially improves earlier generic nowhere-dense estimates [2103.06218]. The same work gives explicit constructions: for every \(d\ge 1\), planar graphs contain distance-\(d\) half graphs of order \(2^{\lceil d/2\rceil}\); for every \(d\ge 1\), \(p\ge 0\), there is a graph of pathwidth at most \(p+2\) containing a distance-\((4d-1)\) half graph of order \((2d+1)^p\) [2103.06218].

Conceptually, these results show that graph powers of sparse graphs may become dense without admitting arbitrarily large order patterns. The half graph remains a controlled obstruction, and its size becomes a quantitative measure of how much order-like behavior survives in a power.

## 3. Half-graph index and parameterized complexity

When half graphs are sought inside arbitrary graphs, the relevant containment notion is often **semi-induced**. A bipartite graph \(H\) appears semi-induced in a graph \(G\) if there exist disjoint sets \(A,B\subseteq V(G)\) such that the bipartite subgraph between \(A\) and \(B\) is isomorphic to \(H\); edges inside \(A\) and inside \(B\) are ignored [2602.07606]. The **half-graph index** of \(G\) is then the maximum order of a semi-induced half graph in \(G\).

This index is studied jointly with the matching index and co-matching index. A theorem of Ding, Oporowski, Oxley, and Vertigan implies that sufficiently large twin-free graphs must contain a large matching, co-matching, or half graph as a semi-induced subgraph, so graph classes fall into eight regimes according to which of the three indices are bounded [2602.07606]. Half-graph-freeness is also self-dual under complementation: a class is half-graph-free if and only if its complement class is half-graph-free [2602.07606].

The resulting complexity picture is mixed. On the positive side, **Independent Set** is fixed-parameter tractable on every graph class where the half-graph and co-matching indices are simultaneously bounded: if a graph has half-graph index \(<t\) and co-matching index \(<t\), one can decide whether it contains an independent set of size at least \(k\) in time
\[
H(t,k)\cdot n^2
\]
for a computable function \(H\) [2602.07606]. The underlying structural reason is a neighborhood dichotomy on long indiscernible sequences: bounded half-graph index forces every outside vertex to be almost sparse or almost dense on such a sequence, and bounded co-matching index upgrades “almost dense” to “complete” [2602.07606].

On the negative side, bounded half-graph index alone does not imply exact fixed-parameter tractability. There exists a graph class with half-graph index at most \(256\) on which **Independent Set** is W[1]-hard, and a class with half-graph index at most \(16\) on which **Dominating Set** is W[1]-hard [2602.07606]. There is, however, a nontrivial approximation guarantee: given a graph \(G\), there is an algorithm running in
\[
O(n^{2h+2})
\]
that finds an independent set of size at least \(k^{1/h}\), where \(h\) is the half-graph index of \(G\) and \(k\) is the maximum independent-set size [2602.07606].

These results place the half graph at a structural boundary. Its exclusion yields real regularity, but by itself it does not fully tame dense-graph complexity.

## 4. Distinct meanings of “half graph” in the literature

The phrase “half graph” is overloaded. Several arXiv papers use it in senses that are mathematically precise but unrelated to the ordered bipartite graph \(H_t\).

| Usage | Definition in the source | Source |
|---|---|---|
| Standard half graph | \(a_i b_j\in E \iff i\le j\) or \(i<j\) | [2602.07606], [2103.06218] |
| Eulerian half graph | \(d_F(v)=d_G(v)/2\) for all \(v\) | [1905.06678] |
| Half-square | \(B^2[X]\) for one color class \(X\) of a bipartite graph \(B\) | [1804.05793] |
| Half-regularity | One side of a bipartition has constant degree in each color class | [1602.04316] |
| Half-translation-surface graph | Saddle connection graph, not a technical term “half graph” | [1809.06248] |
| Half-Space Proximal graph | A proximity graph used in instance-based learning | [2102.02755] |

In the Eulerian setting, a half graph is a subgraph \(F\subseteq E(G)\) such that
\[
d_F(v)=\frac{d_G(v)}{2}\qquad \forall v\in V(G),
\]
so the word “half” refers to splitting the incident edges evenly at every vertex. This is a degree-constrained spanning-subgraph problem, not the ordered bipartite pattern. The paper proves
\[
\varepsilon(G)\ge h(G),
\]
with equality if and only if \(G\) is bipartite, where \(\varepsilon(G)\) is the number of Eulerian orientations and \(h(G)\) the number of such half graphs [1905.06678].

In the half-square setting, one starts with a bipartite graph \(B=(X,Y,E)\) and forms the graph \(B^2[X]\) on one color class \(X\), joining two vertices when they have a common neighbor in \(Y\). This notion yields exact characterizations: half-squares of biconvex bipartite graphs are exactly unit interval graphs, half-squares of convex bipartite graphs are exactly interval graphs, and half-squares of chordal bipartite graphs are exactly strongly chordal graphs [1804.05793].

The paper on half-regular factorizations studies edge-colored bipartite graphs in which all vertices on one fixed side have the same degree in each color class; it explicitly states that this is different from the standard extremal/comparability-theoretic half graph [1602.04316]. Likewise, the paper on half-translation surfaces explicitly says that it does **not** use “half graph” as a technical term; its relevant object is the saddle connection graph of a half-translation surface [1809.06248].

## 5. Related order patterns, stability, and balance phenomena

Half graphs are closely tied to several neighboring concepts. Every half graph is a semi-ladder, and every co-matching is a semi-ladder; conversely, Ramsey-theoretic arguments show that simultaneous bounds on half graphs and co-matchings imply bounds on semi-ladders [2103.06218]. This makes the half graph one endpoint of a small hierarchy of order-pattern obstructions.

From the model-theoretic side, the half graph has a distinguished status: its absence characterizes stability, and monadically stable graph classes are those in which first-order definable relations with colors are half-graph-free [2602.07606]. In this role the half graph is not merely a forbidden subgraph but a canonical witness that adjacency can simulate linear order.

A different balance phenomenon appears in the Eulerian usage. There the “half graph” is really an \(\underline r\)-factor with
\[
r_v=\frac{d_G(v)}{2},
\]
and the comparison with Eulerian orientations is exact on bipartite graphs:
\[
\varepsilon(G)=h(G)\iff G\text{ is bipartite}.
\]
The same paper derives convolution identities
\[
\varepsilon(G)^2=\sum_{A\subseteq E}\varepsilon(A)\varepsilon(E\setminus A),\qquad
h(G)^2=\sum_{A\subseteq E}h(A)h(E\setminus A),
\]
and an exact formula for the bipartite double cover
\[
\varepsilon(G\times K_2)=h(G\times K_2)=\sum_{A\subseteq E}\varepsilon(A)\,h(E\setminus A)
\]
[1905.06678].

This suggests two recurrent uses of “half” in graph-theoretic language. One is **order-theoretic**, exemplified by the triangular adjacency of the standard half graph. The other is **balance-theoretic**, exemplified by equal splitting of incidences in Eulerian graphs. The two meanings are mathematically unrelated despite the shared terminology.

## 6. Terminological boundaries and common confusions

Several additional papers employ “half” language in ways that are adjacent to, but not instances of, the standard half graph. The graph formulation of Frankl’s union-closed sets conjecture studies vertices belonging to **at most half** of all maximal stable sets; it does not study the named half-graph class, although it works heavily with bipartite graphs and half-threshold counting statements [1212.4175]. In extremal combinatorics, “sparse halves” of triangle-free graphs are vertex subsets of size \(\lfloor n/2\rfloor\) spanning at most \(n^2/50\) edges; again, this is unrelated to \(H_t\) [1311.5818]. Quantum-walk papers on graphs with joined half lines analyze a star-like graph built from several semi-infinite paths sharing one origin, which is a different use of “half” arising from half-lines rather than order patterns [1009.1306].

The Half-Space Proximal graph is another unrelated example. It is a local proximity graph used for parameter-free instance-based learning: starting from a center \(u\), one repeatedly keeps the nearest remaining point \(v\) and removes every candidate \(c\) satisfying
\[
d(c,u)>d(c,v).
\]
Its name refers to a half-space exclusion rule, not to the graph-theoretic half graph \(H_t\) [2102.02755].

Accordingly, the phrase **half graph** is context-sensitive. In graph theory proper, without further qualification, it denotes the ordered bipartite ladder \(H_t\). In adjacent literatures it may refer instead to balanced Eulerian subgraphs, half-squares of bipartite graphs, one-sided regular factorizations, or constructions whose “half” comes from half-lines or half-spaces. For technical work, disambiguation by definition is therefore essential.

Source: https://www.emergentmind.com/topics/half-graph