The Logic of Counting Query Answers

We consider the problem of counting the number of answers to a first-order formula on a finite structure. We present and study an extension of first-order logic in which algorithms for this counting problem can be naturally and conveniently expressed, in senses that are made precise and that are motivated by the wish to understand tractable cases of the counting problem.

💡 Research Summary

The paper addresses the fundamental problem of counting the number of answers to a first‑order formula on a finite relational structure, a natural extension of the classic query‑evaluation task in database theory. While traditional first‑order logic (FO) yields a Boolean answer (whether a tuple satisfies the formula), many practical applications require the exact cardinality of satisfying assignments. To capture this computational task within a logical framework, the authors introduce a new logic called 7‑logic.

In 7‑logic, formulas are built from the usual FO atoms, conjunction, disjunction, and quantifiers, but are equipped with two additional operators: C (a “counting” constructor) and P (a “projection” operator). Given a structure B and an assignment h to the free variables of a 7‑formula φ, the evaluation function r_B,φ(h) returns an integer rather than a truth value. The operator C(E, t) binds a tuple of variables t to an atomic predicate E, while P_t ψ fixes an assignment to t and counts how many extensions of this assignment satisfy ψ. By nesting C and P, one can express products, sums, and other arithmetic combinations of sub‑counts, thereby directly encoding the cardinality of the original FO query.

A central technical notion is width, defined as the maximum number of free variables occurring in any sub‑formula. The authors show (Observation 1.4) that if a class Ψ of 7‑sentences has bounded width, then evaluating any ψ∈Ψ on a finite structure can be done in polynomial time. This mirrors the well‑known result that bounded‑width FO sentences are tractable for model checking.

From this observation they formulate the counting condition: for a class Φ of FO formulas, if there exists an algorithm f that maps each φ∈Φ to a 7‑sentence ψ∈Ψ (where Ψ has bounded width) such that ψ represents φ (i.e., ψ’s evaluation equals the number of φ‑answers on any structure), then the counting problem count Φ q is fixed‑parameter tractable (FPT) with respect to the size of φ. This is the counting analogue of the classical condition used for model checking.

The paper’s most significant contribution lies in the setting of existential positive (EP) queries, i.e., FO formulas built from atoms using only conjunction, disjunction, and existential quantification. For EP query classes of bounded arity, the authors prove (Theorem 1.6) that the counting condition is not only sufficient but also necessary for tractability: whenever count Φ q is FPT, a bounded‑width 7‑sentence representation must exist. The proof proceeds via two key theorems.

• Theorem 4.4 shows that any tractable EP counting problem admits a bounded‑width 7‑sentence class Ψ that faithfully represents all formulas in Φ. The construction decomposes EP formulas into elementary components, each translated into a C‑P expression whose width is controlled.

• Theorem 4.5 provides an algorithm that, given an EP formula, computes a 7‑sentence of minimum possible width. The algorithm analyses the dependency graph of free variables, determines an optimal order of projections, and eliminates unnecessary variables, thereby achieving the smallest width achievable by any equivalent 7‑sentence. This result establishes that 7‑logic is not only expressive enough to capture all tractable EP counting queries but also admits an optimal normal form.

Consequently, 7‑logic serves as a computational model for counting query answers, analogous to how ordinary FO serves as a model for model checking. The authors argue that this logical perspective can guide the design of query languages with built‑in counting operators, support systematic query optimisation (width minimisation corresponds to finding optimal execution plans), and provide a clean theoretical foundation for classifying the complexity of counting problems.

The paper also situates its contributions within the broader literature on parameterised complexity, bounded‑width model checking, and the dichotomy theorems for EP queries. By extending the classical width‑based tractability criterion from Boolean model checking to integer‑valued counting, it bridges a gap between theoretical complexity results and practical database features such as the SQL COUNT aggregate.

In summary, the work introduces a novel logic that returns integer values, proves that bounded‑width fragments of this logic are polynomial‑time evaluable, establishes a counting‑condition characterisation of fixed‑parameter tractability for existential positive queries, and supplies an algorithm for optimal width reduction. These results lay the groundwork for future research on richer counting operators, probabilistic extensions, and concrete implementations of 7‑logic‑based query processors.