Universal Structures and the logic of Forbidden Patterns

Forbidden Patterns Problems (FPPs) are a proper generalisation of Constraint Satisfaction Problems (CSPs). However, we show that when the input is connected and belongs to a class which has low tree-depth decomposition (e.g. structure of bounded degree, proper minor closed class and more generally class of bounded expansion) any FPP becomes a CSP. This result can also be rephrased in terms of expressiveness of the logic MMSNP, introduced by Feder and Vardi in relation with CSPs. Our proof generalises that of a recent paper by Nesetril and Ossona de Mendez. Note that our result holds in the general setting of problems over arbitrary relational structures (not just for graphs).

💡 Research Summary

The paper investigates the relationship between Forbidden Patterns Problems (FPPs) and Constraint Satisfaction Problems (CSPs). While FPPs are known to be a strict generalisation of CSPs in the unrestricted setting, the authors show that this separation collapses when the input structures satisfy two natural conditions: (i) the input is connected, and (ii) the input belongs to a class that admits a low‑tree‑depth decomposition. Typical examples of such classes are graphs of bounded degree, any proper minor‑closed family, and, more generally, any class of bounded expansion.

The authors first formalise FPPs as the decision problem of whether a given relational structure A avoids a finite set Φ of forbidden patterns (small relational substructures). They then reinterpret FPPs in the logical framework MMSNP, a fragment introduced by Feder and Vardi that allows existential quantification together with a global negation of a finite set of patterns. In this setting, an instance of an FPP corresponds to an MMSNP sentence, and the central question becomes whether every MMSNP sentence over a low‑tree‑depth class can be rewritten as an equivalent CSP sentence.

The core technical contribution is a constructive reduction that, for any connected input A with tree‑depth at most t, builds a finite target structure B (depending only on Φ and t) such that A satisfies the forbidden‑pattern condition iff there exists a homomorphism from A to B. The construction proceeds in three stages. First, the forbidden patterns are encoded as a universal structure U_Φ, an infinite relational structure that captures exactly the models of the MMSNP sentence. Second, the low tree‑depth of A is exploited to compress U_Φ: a tree‑depth‑t decomposition of A yields a bounded number of “bag‑types”, each describing the possible local configurations of U_Φ inside a bag. Because the decomposition depth is constant, the set of bag‑types is finite. Third, the authors assemble these bag‑types into a finite relational structure B, preserving the adjacency relations dictated by the decomposition tree. By construction, any homomorphism from A to B respects the original decomposition and therefore cannot embed any forbidden pattern; conversely, any A that avoids the patterns can be mapped bag‑wise into B, yielding a homomorphism.

The reduction is algorithmic and runs in polynomial time: a low‑tree‑depth decomposition can be computed efficiently for bounded‑degree, minor‑closed, or bounded‑expansion classes; the universal structure U_Φ is described syntactically from the MMSNP formula; and the enumeration of bag‑types is feasible because the decomposition depth is a constant. Consequently, the decision problem for the original FPP reduces to a standard CSP instance, for which a wealth of algorithmic tools (Datalog, SAT encodings, algebraic dichotomy theorems) are available.

Beyond the computational consequence, the result yields a logical insight: over any class of connected structures with bounded tree‑depth, MMSNP collapses to CSP. This extends earlier work by Nešetřil and Ossona de Mendez, who proved a similar collapse for graph classes of bounded expansion but only for CSP‑style problems. Here the authors treat arbitrary relational signatures, not just graphs, thereby broadening the applicability of the theory.

The paper also discusses limitations and future directions. The connectivity assumption is essential: disconnected inputs would require handling each component separately, and it is not clear whether a uniform finite target structure exists in that case. Moreover, the technique relies on a constant bound on tree‑depth; classes with unbounded tree‑depth (e.g., dense graphs) remain outside the scope of the reduction. Extending the approach to such settings, or finding tighter bounds on the size of the target structure B for specific pattern families, are natural open problems.

In summary, the authors prove that for any connected input belonging to a low‑tree‑depth class, every forbidden‑patterns problem can be expressed as a CSP. This bridges a gap between two major paradigms in finite model theory and combinatorial optimisation, enriches our understanding of the expressive power of MMSNP, and provides a concrete, polynomial‑time reduction that can be leveraged in both theoretical investigations and practical algorithm design.