Semi-Transitive Orientation in Graphs

• Semi-transitive orientation is an acyclic edge-orientation of a graph that forbids induced shortcuts by requiring complete connection along any directed path with a connecting edge.
• It generalizes transitive orientations and encapsulates classes such as comparability, 3-colorable, and circle graphs, thus bridging graph coloring and combinatorial word representations.
• Algorithmic studies reveal that recognizing semi-transitive graphs is NP-complete in general, though efficient methods exist for special cases like split graphs and controlled constructions.

A semi-transitive orientation of a graph is an acyclic edge-orientation in which every shortcut—an induced subdigraph containing a directed path plus a “shortcutting” arc but missing some intermediate arcs— is forbidden. Semi-transitive graphs are precisely the class of word-representable graphs: those for which vertex pairs alternate in some representing word if and only if they are adjacent. Semi-transitive orientability generalizes transitivity (as in comparability graphs) and is central in the structural, algorithmic, and extremal theory of graph orientations, with deep connections to graph colorings, forbidden subgraphs, and combinatorial word structures.

1. Formal Definition and Characterizations

Let $G = (V,E)$ be a simple undirected graph. An orientation of $G$ is an assignment of direction to each edge. The digraph $D$ is acyclic if it contains no directed cycle. The orientation is semi-transitive if the following no-shortcut condition holds: for every directed path

$v_0 \to v_1 \to \cdots \to v_k$

either:

• $\{v_0, v_k\} \notin E$; or
• If $\{v_0, v_k\} \in E$, then $v_i \to v_j$ is an edge for all $0 \leq i < j \leq k$.

An undirected graph is called semi-transitive (or semi-transitively orientable) if such an orientation exists (Kitaev et al., 2019, Halldórsson et al., 2015, Akgün et al., 2018, Kitaev et al., 2021).

Equivalently: in any acyclic orientation, no induced subdigraph on $\{v_0, v_1, \ldots, v_k\}$ consists of a directed path $v_0 \to \cdots \to v_k$, a shortcut arc $v_0 \to v_k$, and a missing arc $v_i \to v_j$ for some $0 < i < j < k$.

Key properties:

• Every transitive orientation (comparability graph) is semi-transitive.
• Semi-transitivity and word-representability are equivalent: a graph is word-representable if and only if it admits a semi-transitive orientation.

2. Relationship to Graph Classes and Word-Representability

Semi-transitive graphs strictly contain important classes:

• 3-colorable graphs: Every 3-colorable graph is semi-transitive; any such coloring yields an explicit semi-transitive orientation (Kitaev et al., 2019, Halldórsson et al., 2015, Kenkireth et al., 10 Feb 2025).
• Comparability graphs: All comparability (i.e., transitively orientable) graphs are semi-transitive (Choi et al., 2018).
• Circle graphs: Intersection graphs of chords in a circle are semi-transitive (Kitaev et al., 2019).

Equivalence with word-representable graphs:

A graph $G$ is word-representable if there exists a word $w$ over $V$ such that two letters alternate in $w$ if and only if their corresponding vertices are adjacent. The orientation-theoretic characterization is: $$G \text{ is word-representable } \iff G \text{ admits a semi-transitive orientation}$$ (Halldórsson et al., 2015, Akgün et al., 2018, Srinivasan et al., 2024, Kenkireth et al., 10 Feb 2025).

3. Algorithmic and Structural Aspects

Complexity:

Deciding if an arbitrary graph is semi-transitive is NP-complete. For the triangle-free case, the problem remains NP-hard (Kitaev et al., 2020, Halldórsson et al., 2015, Kitaev et al., 2021).

Split graphs:

For split graphs (vertex partition into a clique and an independent set), polynomial-time algorithms exist, based on the circular-ones property of $(0,1)$-matrices constructed from the neighborhoods of the independent set (Kitaev et al., 2021).

Hereditary property:

Semi-transitivity is closed under taking induced subgraphs (Akgün et al., 2018).

Minimal forbidden induced subgraphs:

Every minimal non-word-representable graph (or minimal non-semi-transitive graph) corresponds to a forbidden induced subgraph for the class. Complete classification is known for comparability graphs, but not for the semi-transitive class. Recent work identifies infinite families and sporadic minimal non-semi-transitive graphs (Kenkireth et al., 10 Feb 2025).

4. Extremal and Classification Results

Kneser graphs:

For Kneser graphs $K(n,k)$ (vertices: size-$k$ subsets of $[n]$; edges: disjointness), the semi-transitive threshold is rigorously bounded:

• If $k \leq n \leq 2k+1$, $K(n,k)$ is semi-transitive.
• If $n \geq 15k-24$, $K(n,k)$ is not semi-transitive.
• The complement $\overline{K(n,k)}$ is semi-transitive if and only if $n \leq 2k$. These results provide explicit families where semi-transitivity fails for the first time (e.g., the triangle-free $K(8,3)$), and settle the dichotomy for complements (Kitaev et al., 2019).

Triangle-free graphs:

The existence of triangle-free, non-semi-transitive graphs was established probabilistically by Erdős; explicit smallest examples are the Grötzsch graph (11 vertices) and the Chvátal graph (12 vertices, 4-regular) (Kitaev et al., 2020).

Circulant graphs:

All 4-regular circulant graphs are semi-transitive. For consecutive jumps, semi-transitivity can fail or hold depending on parameters; for jumps of at least $(n+1)/4$, circulants admit a simple acyclic semi-transitive orientation (Srinivasan et al., 2024).

Mycielski graphs:

For a given $G$, the Mycielski construction yields a triangle-free graph of higher chromatic number. The Mycielski graph of $G$ is semi-transitive if and only if $G$ is bipartite (Kitaev et al., 2024). This result resolves a conjecture and establishes a sharp classification for both standard and extended Mycielski graphs.

Minimal forbidden subgraphs with dominating vertex:

Adding a universal vertex to a minimal non-comparability graph that is semi-transitive yields minimal non-word-representable graphs with a dominating vertex (Kenkireth et al., 10 Feb 2025).

5. Graph Operations and Preservation Results

Edge deletions/additions/liftings:

For any semi-transitively oriented graph, certain edge deletions, additions, and path liftings preserve the semi-transitive property:

• There always exists an edge whose deletion maintains semi-transitivity.
• There always exists a non-edge whose addition maintains semi-transitivity.
• Path liftings (replacing $u-v-w$ with $u-w$) can sometimes preserve semi-transitivity (Choi et al., 2018).

Negative results for graph products:

Tensor, lexicographic, and strong products do not in general preserve semi-transitive orientability—the property is delicate and easily destroyed by these graph operations (Choi et al., 2018).

Subdivision and edge-deletion equivalence:

If $e$ is an edge, subdividing it (at least once) yields a semi-transitive graph if and only if deleting it does (Choi et al., 2018).

6. Enumeration, Quantitative Results, and Open Problems

Empirical enumeration:

Massive computational campaigns enumerate non-semi-transitive graphs (non-word-representable) up to order 11, correcting prior errors. For 11 vertices, about 65% of connected graphs are not semi-transitive (Akgün et al., 2018).

Representation numbers:

The minimum $k$ such that a graph is $k$-representable by a word is bounded above by $2(n-2)$, and below by $\lfloor n/2\rfloor$ for some graphs. Representation numbers for circulants, crowns, and crown-plus-apex constructions have been determined for certain families (Halldórsson et al., 2015, Akgün et al., 2018, Srinivasan et al., 2024).

Refinements — $k$-semi-transitive orientations:

A $k$-semi-transitive orientation requires the shortcut condition only up to paths of length $k$. This property is strictly stronger than full semi-transitivity: there exist (for $n \geq 9$) graphs that admit a $k$-semi-transitive orientation for small $k$, but not a fully semi-transitive one (Akgün et al., 2018).

Algorithmic boundaries:

Efficient recognition is possible for split graphs and small cliques/independent sets, but remains hard in general. The complexity on other graph classes (e.g., chordal graphs) is open (Kitaev et al., 2021).

Open questions:

• Complete forbidden subgraph characterizations for semi-transitive orientability
• Thresholds for Kneser graph semi-transitivity (narrowing $2k+1 \leq n < 15k-24$)
• Existence and explicit construction of non-semi-transitive graphs of large girth
• Behavior of $k$-semi-transitive vs. semi-transitive classes for larger $k$ (Kitaev et al., 2019, Kenkireth et al., 10 Feb 2025, Akgün et al., 2018, Srinivasan et al., 2024, Kitaev et al., 2021).

7. Tables: Key Thresholds and Explicit Examples

Graph Family	Semi-Transitive if…	Fails if…
Kneser $K(n,k)$	$k\leq n\leq 2k+1$	$n\geq 15k-24$
Complements $\overline{K(n,k)}$	$n\leq 2k$	$n>2k$
Mycielski $p(G)$	$G$ bipartite	$G$ contains odd cycle
Circulant $C(n;R)$	$a_1 \geq (n+1)/4$	certain consecutive jump sets

Explicit Non-Semi-Transitive	Size	Structure
Grötzsch graph	11	Triangle-free, 4-chromatic
Chvátal graph	12	Triangle-free, 4-regular
$K(8,3)$ subgraph	16	Triangle-free, induced

These results collectively frame semi-transitive orientation as a central and richly structured graph invariant, linking acyclic and comparability properties, graph colorings, word alternation theory, and combinatorial extremals (Kitaev et al., 2019, Kenkireth et al., 10 Feb 2025, Kitaev et al., 2020, Halldórsson et al., 2015, Choi et al., 2018, Kitaev et al., 2021, Srinivasan et al., 2024, Akgün et al., 2018, Kitaev et al., 2024).