Problems parameterized by treewidth tractable in single exponential time: a logical approach

We introduce a variant of modal logic, dubbed EXISTENTIAL COUNTING MODAL LOGIC (ECML), which captures a vast majority of problems known to be tractable in single exponential time when parameterized by treewidth. It appears that all these results can be subsumed by the theorem that model checking of ECML admits an algorithm with such complexity. We extend ECML by adding connectivity requirements and, using the Cut&Count technique introduced by Cygan et al. [4], prove that problems expressible in the extension are also tractable in single exponential time when parameterized by treewidth; however, using randomization. The need for navigationality of the introduced logic is justified by a negative result that two expository problems involving non-acyclic conditions, C_l VERTEX DELETION and GIRTH>l VERTEX DELETION for l>=5, do not admit such a robust algorithm unless Exponential Time Hypothesis fails.

💡 Research Summary

The paper addresses a central question in parameterized algorithmics: which graph problems become tractable in single‑exponential time when the parameter is the treewidth of the input graph? While Courcelle’s theorem guarantees fixed‑parameter tractability for MSO‑definable problems, its running time is typically non‑elementary in the treewidth. The authors therefore propose a new logical framework, Existential Counting Modal Logic (ECML), that captures a broad class of problems known to admit algorithms of the form O*(c^tw).

Core logical ingredients
ECML consists of two layers. The inner layer, called Counting Modal Logic (CML), works on a single vertex and uses modal operators ♦ S and ◻ S, where S is a finitely recognizable set of natural numbers. Such sets are exactly the ultimately periodic subsets of ℕ and can be represented by a finite monoid together with a homomorphism α:ℕ→M. The semantics of ♦ S ψ is: “the number of neighbours w of the current vertex v that satisfy ψ belongs to S”. The dual ◻ S is defined as the negation of ♦ S ¬ψ. In addition to standard propositional connectives, CML contains unary predicates X_i (testing membership of the current vertex in a chosen vertex set) and Y_j (testing whether the edge used to reach the current vertex belongs to a chosen edge set). For directed graphs two extra edge predicates (↓, ↑) indicate the direction of the traversed arc.

The outer layer quantifies over vectors of vertex sets X and edge sets Y, imposes an arithmetic constraint φ (e.g., |X|≤k) and requires that every vertex satisfies a CML formula ψ. Formally the ECML sentence has the shape

 ∃X∃Y ( φ ∧ ∀v ψ )

where φ is a quantifier‑free integer arithmetic expression and ψ is a CML formula evaluated with the chosen X and Y.

Algorithmic meta‑theorem
Given a nice tree decomposition of width tw, the authors design a dynamic programming (DP) scheme that processes the decomposition bottom‑up. For each bag they store a compact state consisting of: (i) a bitmask indicating which vertices/edges of the bag belong to the quantified sets X, Y; (ii) for each vertex in the bag the current value of the monoid element that records how many neighbours satisfying ψ have been seen so far; (iii) the truth value of the global arithmetic constraint φ on the partial solution. Because the monoid is finite and the size of each bag is at most tw+1, the number of possible states per bag is bounded by c^{tw} for a constant c that depends only on the formula (size of the monoids, number of counting modalities, etc.). Consequently the DP runs in O(c^{tw})* time, i.e., single‑exponential in the treewidth, and also counts the number of solutions. This yields a unified algorithmic proof for a large collection of problems, including Vertex Cover, Dominating Set, r‑Dominating Set, Feedback Vertex Set, and many others that were previously handled by ad‑hoc DP constructions.

Extension with connectivity (ECML+C)
Problems that require the chosen vertex/edge set to induce a connected subgraph (e.g., Connected Vertex Cover, Hamiltonian Path) traditionally lead to DP tables of size 2^{O(tw log tw)}. To overcome this barrier the authors incorporate the Cut&Count technique of Cygan et al. (2011). They augment the DP with random hash functions that assign each partial solution a parity; the parity of the total number of “cut” solutions cancels out unless the solution is connected. By combining Cut&Count with the counting modalities of ECML, they obtain Monte‑Carlo algorithms for all ECML‑expressible connectivity problems that also run in O*(c^{tw}) time (with a controllable error probability). The approach is randomized but can be amplified to arbitrarily low error by repetition.

Hardness under ETH
To show that the logical framework cannot be arbitrarily generalized, the paper proves ETH‑based lower bounds for two non‑acyclic problems:

1. C_l‑Vertex Deletion – delete the minimum number of vertices to eliminate all cycles of length at least l.
2. Girth > l Vertex Deletion – delete vertices so that the resulting graph has girth larger than l.

For any fixed l ≥ 5, the authors give reductions from SAT that imply that any algorithm solving these problems in time 2^{o(p²)}·|G|^{O(1)} (where p is the pathwidth) would contradict the Exponential Time Hypothesis. Since ECML inherently works by “navigating” the graph in a tree‑like (acyclic) fashion, these lower bounds justify the need for the acyclicity / navigational restriction in the logic.

Significance and future directions
The work delivers a meta‑theorem that unifies a large family of treewidth‑parameterized algorithms under a single logical umbrella, extending Courcelle’s paradigm to the realm of single‑exponential algorithms. It also demonstrates how modern algebraic techniques (Cut&Count) can be seamlessly integrated into a logical framework, opening the door to systematic treatment of connectivity constraints. The ETH lower bounds delineate the boundary of the approach, indicating that any further logical enrichment must respect acyclicity or accept higher complexity. Future research may explore (i) deterministic replacements for Cut&Count, (ii) richer counting modalities while preserving single‑exponential bounds, and (iii) practical implementations of ECML‑based solvers for real‑world network design problems.