Axiomatisability and hardness for universal Horn classes of hypergraphs

We characterise finite axiomatisability and intractability of deciding membership for universal Horn classes generated by finite loop-free hypergraphs.

💡 Research Summary

The paper investigates two fundamental aspects of universal Horn classes (UHCs) generated by finite, loop‑free hypergraphs: (1) when such classes admit a finite axiomatization in first‑order logic (or any finite fragment thereof), and (2) the computational complexity of the membership problem for these classes. The authors extend the well‑known classification for simple graphs due to Caicedo to the much richer setting of hypergraphs, and they obtain a complete dichotomy.

Main results.
Theorem 1 provides a precise characterisation. Let H be a universal Horn class generated by a family of finite, loop‑free hypergraphs, and assume that both the chromatic number χ(H) and the maximal hyperedge cardinality k are bounded. Then H has a finite axiomatization (in first‑order logic) iff every member of H is a disjoint union of bipartite graphs (i.e., the class is essentially a collection of 2‑uniform hypergraphs with no odd cycles). In all other cases—most notably whenever a hyperedge of size ≥ 3 appears—no finite set of universal Horn sentences (of any quantifier depth) can capture H.

The proof of the “only‑if” direction uses two complementary techniques. The first is a probabilistic construction originally due to Erdős and Hajnal: for any prescribed lower bound ℓ on cycle length and any number of colours n, one can build a finite k‑uniform hypergraph with no short cycles that nevertheless is not n‑colourable. Such hypergraphs witness the failure of any finite axiomatization because any finite set of Horn sentences can only forbid finitely many small cycles. The second technique relies on the authors’ recent “All‑or‑Nothing” theorem. By embedding a suitable hypergraph into a template structure, they show that the membership problem for the corresponding UHC is NP‑complete (or harder), which immediately implies non‑axiomatisability (finite axiomatizations always yield polynomial‑time decidable membership).

Richness of the lattice of UHCs.
Section 2 shows that the lattice of universal Horn classes of hypergraphs is extremely dense. Using a method of Bonato, the authors prove that every interval in the homomorphism order of hypergraphs contains a continuum of distinct UHCs. This extends Nešetřil‑Pultr’s density results for graphs to hypergraphs and demonstrates that there are uncountably many inequivalent Horn theories even when the underlying signature is fixed.

Finite‑structure restriction.
In Section 3 the authors turn to the finite‑model setting, a topic of growing interest in finite model theory. They prove that the dichotomy of Theorem 1 remains valid when attention is restricted to finite hypergraphs. The proof employs an Ehrenfeucht–Fraïssé game argument to establish a general lemma: any hereditary class of finite structures closed under disjoint unions and direct powers satisfies the SP‑closure property (closed under substructures, direct powers, and direct sums). Using this lemma they show that the SP‑class generated by a single hyperedge Eₖ contains all finite k‑uniform hyperforests, and consequently the same finite‑axiomatisability criterion holds for finite structures.

Complexity consequences.
Section 4 revisits the hardness results via the All‑or‑Nothing theorem, providing an alternative route that avoids probabilistic constructions. The theorem yields explicit reductions from classic NP‑complete problems (e.g., NAE‑3SAT) to the membership problem for certain hypergraph UHCs, establishing NP‑completeness (and, in some cases, higher hardness) for a wide range of templates. This reinforces the intuition that, except for the degenerate bipartite‑graph case, deciding whether a given finite hypergraph belongs to a generated UHC is computationally intractable.

Methodological contributions.
The paper introduces several technical tools that may be of independent interest:

• The “set(·)” operator that closes a k‑ary relation under set‑equivalence, allowing a uniform treatment of hyperedges of varying size within a single relational signature.
• Lemma 11, showing that the SP‑class of a single hyperedge Eₖ contains all k‑uniform hyperforests, which is a key step in the finite‑structure analysis.
• A careful adaptation of the separation conditions (SEP1–SEP3) for universal Horn classes to the hypergraph setting, linking homomorphisms, colourings, and non‑hyperedge preservation.

Conclusion and outlook.
The authors have achieved a comprehensive understanding of when universal Horn classes of finite loop‑free hypergraphs are finitely axiomatizable and when their membership problems are computationally hard. The results parallel the known picture for simple graphs, confirming that the introduction of higher‑arity hyperedges does not soften the dichotomy but rather sharpens it: any presence of a hyperedge of size ≥ 3 forces non‑axiomatisability and NP‑hardness. The paper also opens several avenues for future work, such as extending the analysis to hypergraphs with loops, to signatures with multiple relations, or to parameterised complexity analyses of the membership problem. Overall, the work makes a significant contribution at the intersection of model theory, combinatorics, and computational complexity.