Half Graph in Graph Theory

• Half graph is an ordered bipartite pattern defined on two disjoint vertex sets with nested neighborhoods, encoding a total order.
• It is central to structural graph theory and parameterized complexity, influencing concepts like semi-induced containment and graph sparsity.
• Bounded half-graph index impacts algorithmic tractability, guiding fixed-parameter strategies for problems such as Independent Set.

In graph theory and model theory, a half graph is a canonical ordered bipartite pattern that encodes a total order through nested neighborhoods. One standard formulation uses disjoint vertex sets $A=\{a_1,\ldots,a_t\}$ and $B=\{b_1,\ldots,b_t\}$ with adjacency rule $a_i b_j\in E \iff i\leq j$; an equivalent convention used in the literature writes $(b_i,a_j)\in E \iff i<j$ after a shift of indexing conventions. In both forms, the pattern is distinguished from the matching and the co-matching, and it plays a central role in structural graph theory, parameterized complexity, stability theory, and the study of sparse graph powers (Dreier et al., 7 Feb 2026, Sokołowski, 2021).

1. Canonical ordered pattern

The half-graph is introduced together with two companion bipartite patterns. Given two disjoint sets

$A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$

the matching is defined by

$a_i b_j\in E \iff i=j,$

the co-matching by

$a_i b_j\in E \iff i\neq j,$

and the half-graph by

$a_i b_j\in E \iff i\leq j.$

This makes the half-graph the “order” pattern: vertices on one side have nested neighborhoods into the other side (Dreier et al., 7 Feb 2026).

A related formulation appears in the study of sparse graph powers. There, distinct vertices

$a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$

form a half graph, also called a ladder, of order $\ell$ if

$B=\{b_1,\ldots,b_t\}$0

The same source distinguishes half graphs from semi-ladders, where only the implications for $B=\{b_1,\ldots,b_t\}$1 and the non-edges on the diagonal are required, and from co-matchings, which satisfy $B=\{b_1,\ldots,b_t\}$2 (Sokołowski, 2021).

These definitions emphasize the same structural feature: a half graph is a bipartite adjacency pattern that realizes monotone comparability by index. This is why it is repeatedly treated as a canonical obstruction or witness of order-like behavior.

2. Semi-induced containment and half-graph index

In parameterized and structural graph theory, half-graphs are often studied through the weaker notion of semi-induced containment. A bipartite graph $B=\{b_1,\ldots,b_t\}$3 appears semi-induced in a graph $B=\{b_1,\ldots,b_t\}$4 if there exist $B=\{b_1,\ldots,b_t\}$5 such that the bipartite subgraph between $B=\{b_1,\ldots,b_t\}$6 and $B=\{b_1,\ldots,b_t\}$7 is isomorphic to $B=\{b_1,\ldots,b_t\}$8. Only the cross-edges matter; edges inside $B=\{b_1,\ldots,b_t\}$9 and inside $a_i b_j\in E \iff i\leq j$0 are irrelevant (Dreier et al., 7 Feb 2026).

This leads to the half-graph index of a graph $a_i b_j\in E \iff i\leq j$1: the maximum order of a semi-induced half-graph in $a_i b_j\in E \iff i\leq j$2. Analogously one defines the matching index and co-matching index. For a graph class, having bounded half-graph index means that there exists a constant $a_i b_j\in E \iff i\leq j$3 such that every graph in the class has half-graph index at most $a_i b_j\in E \iff i\leq j$4; the same source also says such classes are half-graph-free. Unbounded half-graph index means that for every $a_i b_j\in E \iff i\leq j$5 there exists a graph in the class containing a semi-induced half-graph of order $a_i b_j\in E \iff i\leq j$6 (Dreier et al., 7 Feb 2026).

A theorem of Ding, Oporowski, Oxley, and Vertigan is the structural starting point for this viewpoint. In the form used in recent complexity work, there exists a function $a_i b_j\in E \iff i\leq j$7 such that every bipartite graph without twins and with at least $a_i b_j\in E \iff i\leq j$8 vertices on one side contains a matching, co-matching, or half-graph of order $a_i b_j\in E \iff i\leq j$9 as an induced subgraph. Via neighborhood diversity and twin classes, this yields a corresponding trichotomy for general graphs: sufficiently large neighborhood diversity forces large matching index, co-matching index, or half-graph index (Dreier et al., 7 Feb 2026).

An important symmetry is that half-graph-freeness is preserved under complementation: a class $(b_i,a_j)\in E \iff i<j$0 is half-graph-free if and only if its complement class $(b_i,a_j)\in E \iff i<j$1 is half-graph-free. By contrast, matching-free and co-matching-free are exchanged by complementation (Dreier et al., 7 Feb 2026).

3. Complexity taxonomy and algorithmic consequences

The boundedness or unboundedness of the matching, co-matching, and half-graph indices partitions graph classes into $(b_i,a_j)\in E \iff i<j$2 cases. Recent work uses this taxonomy to classify the parameterized complexity of Independent Set, Clique, and Dominating Set across all eight cases (Dreier et al., 7 Feb 2026).

The main positive result involving half-graphs is that Independent Set is fixed-parameter tractable on classes with both bounded half-graph index and bounded co-matching index. Formally, there exists a computable function $(b_i,a_j)\in E \iff i<j$3 such that, for every $(b_i,a_j)\in E \iff i<j$4, given an $(b_i,a_j)\in E \iff i<j$5-vertex graph $(b_i,a_j)\in E \iff i<j$6 with half-graph index $(b_i,a_j)\in E \iff i<j$7 and co-matching index $(b_i,a_j)\in E \iff i<j$8, one can decide whether $(b_i,a_j)\in E \iff i<j$9 contains an independent set of size at least $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$0 in time $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$1 (Dreier et al., 7 Feb 2026).

The same paper shows that bounded half-graph index alone is not sufficient for such tractability. It constructs a graph class with half-graph index at most $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$2 on which Independent Set is W[1]-hard, and another graph class with half-graph index at most $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$3 on which Dominating Set is W[1]-hard. These results are used to show that excluding half-graphs alone is structurally meaningful but still too weak to force fixed-parameter tractability for the main problems under study (Dreier et al., 7 Feb 2026).

Half-graph boundedness nonetheless supports approximation. For a graph $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$4 with half-graph index $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$5 and maximum independent-set size $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$6, there is an algorithm running in time $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$7 that finds an independent set of size at least $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$8. This is a direct algorithmic use of the half-graph parameter without the additional assumption of bounded co-matching index (Dreier et al., 7 Feb 2026).

The complement symmetry of the half-graph parameter also propagates immediately to Clique results, because Clique on $A=\{a_1,\ldots,a_t\}, \qquad B=\{b_1,\ldots,b_t\},$9 is Independent Set on $a_i b_j\in E \iff i=j,$0 and half-graph-freeness is complement-invariant (Dreier et al., 7 Feb 2026).

4. Half graphs in powers of sparse graphs

Half graphs remain central when adjacency is defined by bounded distance rather than by direct edges. For a graph $a_i b_j\in E \iff i=j,$1 and integer $a_i b_j\in E \iff i=j,$2, the $a_i b_j\in E \iff i=j,$3-th power $a_i b_j\in E \iff i=j,$4 is defined by

$a_i b_j\in E \iff i=j,$5

A distance-$a_i b_j\in E \iff i=j,$6 half graph is then a sequence of vertices satisfying

$a_i b_j\in E \iff i=j,$7

which is exactly a half graph in $a_i b_j\in E \iff i=j,$8 (Sokołowski, 2021).

This setting connects half graphs to sparsity theory. Adler and Adler proved that nowhere dense classes are stable, hence every fixed first-order interpretation has bounded half-graph order. In the distance setting, this implies that for every nowhere dense class $a_i b_j\in E \iff i=j,$9 and every $a_i b_j\in E \iff i\neq j,$0, the graphs in $a_i b_j\in E \iff i\neq j,$1 have bounded half-graph order. Fabiański et al. later showed boundedness of semi-ladders in $a_i b_j\in E \iff i\neq j,$2, which is stronger in the distance setting. The later quantitative work emphasizes that the earlier bounds were either non-constructive or very loose (Sokołowski, 2021).

For this reason, recent results provide nearly tight asymptotic lower and upper bounds on the maximum order of half graphs in powers of several sparse classes: planar graphs, graphs with bounded maximum degree, graphs with bounded pathwidth or treewidth, and graphs excluding a fixed clique as a minor. For planar graphs, the cited work highlights an improvement from a previous bound of order $a_i b_j\in E \iff i\neq j,$3 to $a_i b_j\in E \iff i\neq j,$4, using structural graph-theoretic analysis of semi-ladders through the notion of cages, together with a new structural result yielding a fully polynomial bound on neighborhood complexity in planar graphs (Sokołowski, 2021).

For graphs of maximum degree at most $a_i b_j\in E \iff i\neq j,$5, the behavior is essentially exponential in $a_i b_j\in E \iff i\neq j,$6. For every $a_i b_j\in E \iff i\neq j,$7 and odd $a_i b_j\in E \iff i\neq j,$8, there exists a graph of maximum degree at most $a_i b_j\in E \iff i\neq j,$9 containing a distance-$a_i b_j\in E \iff i\leq j.$0 half graph of order

$a_i b_j\in E \iff i\leq j.$1

In the same class, the upper bound recorded in the summary is

$a_i b_j\in E \iff i\leq j.$2

Thus, for bounded-degree graphs, the maximum half-graph order in graph powers is pinned down up to constant factors in the exponent (Sokołowski, 2021).

5. Stability, indiscernibility, and regularity phenomena

The half-graph is also important because its absence is tied to stability. Recent complexity work explicitly notes that “the half-graph pattern has strong ties to model theory, where its absence characterizes stable theories” (Dreier et al., 7 Feb 2026). The sparse-graph-power literature makes the same connection in a different language by recalling the stability of nowhere dense classes and the boundedness of half graphs in their distance powers (Sokołowski, 2021).

This model-theoretic perspective becomes algorithmic through indiscernible sequences. One result states that if $a_i b_j\in E \iff i\leq j.$3 has half-graph index $a_i b_j\in E \iff i\leq j.$4, and

$a_i b_j\in E \iff i\leq j.$5

is a $a_i b_j\in E \iff i\leq j.$6-indiscernible sequence of length $a_i b_j\in E \iff i\leq j.$7, then every vertex $a_i b_j\in E \iff i\leq j.$8 satisfies

$a_i b_j\in E \iff i\leq j.$9

The paper interprets this as a “few neighbors or few non-neighbors” dichotomy forced by bounded half-graph index (Dreier et al., 7 Feb 2026).

A stronger statement adds bounded co-matching index. If $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$0 has half-graph index $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$1 and co-matching index $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$2, and $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$3 is a $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$4-indiscernible sequence of length $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$5, then for every $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$6,

$a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$7

In the authors’ interpretation, bounded half-graph index gives “few neighbors or few non-neighbors,” and bounded co-matching index upgrades “few non-neighbors” to “no non-neighbors” (Dreier et al., 7 Feb 2026).

These regularity statements drive the kernelization for Independent Set. They also clarify why the half-graph matters most in tandem with the co-matching: the pair defines the regime in which long sequences exhibit sharply constrained neighborhood behavior, while half-graph-freeness by itself remains too weak to eliminate hard instances (Dreier et al., 7 Feb 2026).

6. Terminological extensions and distinct notions

Although “half graph” has a standard meaning in combinatorics, the same phrase or nearby terminology is used differently in other literatures.

In self-supervised graph representation learning, Pairwise Half-graph Discrimination defines a “half-graph” as one of two induced parts obtained by splitting a graph into two disconnected pieces. The split is node-based, approximately balanced, and produced by choosing a border node; positive examples are pairs of half-graphs from the same source graph, while negative examples replace one half by a half-graph from another graph (Li et al., 2021). This usage is task-specific and not the classical ordered bipartite half-graph.

Several papers explicitly warn against conflating the classical half-graph with nearby terms. Work on half-squares of bipartite graphs states that it is not about the classical half-graph, but about the operation $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$8 or $a_1,\dots,a_\ell,\; b_1,\dots,b_\ell$9 on a bipartite graph $\ell$0 (Le et al., 2018). Likewise, work on half-regular bipartite graphs stresses that it studies half-regularity—constant degree on one side of a bipartition—rather than the chain-graph object usually called a half graph (Aksen et al., 2016).

Outside graph theory proper, the overlap is purely terminological. In quantum-walk theory, a graph with joined half lines $\ell$1 is a star of $\ell$2 semi-infinite rays meeting at one origin and is described as a “joined-half-lines” or “star-like half-line graph,” not as the standard combinatorial half graph (Chisaki et al., 2010). In geometric analysis, titles involving “graphs in a half-space” or “CMC $\ell$3 entire graphs” concern hypersurfaces or sections over geometric domains rather than graph-theoretic half graphs (Mazet, 2013, Edelen et al., 1 Jun 2026).

The resulting picture is therefore two-layered. In the classical sense, the half graph is an ordered bipartite obstruction encoding a total order and governing stability, sparse-graph structure, and parameterized complexity. In broader usage, the term may denote entirely different constructions, and those usages must be disambiguated from the classical object.