Decidability and Undecidability Results for Propositional Schemata

We define a logic of propositional formula schemata adding to the syntax of propositional logic indexed propositions and iterated connectives ranging over intervals parameterized by arithmetic variables. The satisfiability problem is shown to be undecidable for this new logic, but we introduce a very general class of schemata, called bound-linear, for which this problem becomes decidable. This result is obtained by reduction to a particular class of schemata called regular, for which we provide a sound and complete terminating proof procedure. This schemata calculus allows one to capture proof patterns corresponding to a large class of problems specified in propositional logic. We also show that the satisfiability problem becomes again undecidable for slight extensions of this class, thus demonstrating that bound-linear schemata represent a good compromise between expressivity and decidability.

💡 Research Summary

The paper introduces a novel logical formalism called propositional schemata, which extends ordinary propositional logic with two orthogonal features: (1) indexed propositional variables (e.g., pₖ where k ranges over natural numbers) and (2) iterated connectives that quantify over a finite interval of an arithmetic variable (e.g., ∧{i=0}^{n} φ(i) or ∨{i=m}^{k} ψ(i)). These extensions allow a single schema to compactly represent an infinite family of concrete propositional formulas, making the formalism attractive for modelling parameterised circuits, infinite‑state systems, or recursive program specifications.

The authors first show that, without any restriction, the satisfiability problem for propositional schemata is undecidable. The proof proceeds by encoding the halting problem of a Turing machine into a schema: the machine’s tape cells, states, and transition rules are represented by indexed propositions, while the evolution of the computation is captured by iterated conjunctions/disjunctions over the time index. Consequently, deciding whether a schema has a model is equivalent to deciding whether the encoded machine halts, which is known to be undecidable. This result establishes that the added expressive power makes the logic computationally as strong as first‑order arithmetic.

To recover decidability, the paper defines a bound‑linear subclass of schemata. A bound‑linear schema obeys three syntactic constraints: (i) every iterated connective ranges over a linear interval of the form 0 ≤ i ≤ n where n is either a constant or a single arithmetic variable; (ii) each schema contains at most one index variable, so nested or simultaneous iterations are forbidden; (iii) indexed propositions may only appear with linear index expressions (e.g., p_{i+2}, p_{2·i+1}) and not with arbitrary arithmetic terms. These restrictions keep the expressive power sufficient for many practical specifications while preventing the kind of self‑reference that leads to undecidability.

The core technical contribution is a reduction from bound‑linear schemata to a more tractable class called regular schemata. The reduction normalises the interval bounds, aligns all iterations to a common “slice” length, and rewrites the internal propositional structure into a canonical normal form (typically conjunctive normal form). Crucially, the transformation runs in polynomial time and does not cause an exponential blow‑up in the size of the schema.

For regular schemata the authors devise a schemata calculus, a proof‑search procedure that systematically simplifies a schema by applying a finite set of rewrite rules: (1) interval shrinking, (2) partial unfolding of iterated connectives, (3) propagation of negation (De Morgan‑style), and (4) merging of identical sub‑patterns. The calculus is shown to be sound (every derived contradiction corresponds to genuine unsatisfiability), complete (if a schema is satisfiable the calculus eventually produces a concrete model), and terminating (no infinite rewrite sequences are possible). The termination argument relies on a well‑founded measure that strictly decreases with each rule application, typically the sum of interval lengths plus the number of unresolved iterated connectives.

Having established a decision procedure for regular schemata, the paper concludes that satisfiability for bound‑linear schemata is decidable: one first translates a bound‑linear schema into a regular one, then runs the calculus. The overall algorithm runs in at most exponential time with respect to the size of the original schema, which is acceptable given the expressive power retained.

The authors also explore the fragility of this decidable frontier. By allowing a modest extension—namely, polynomial bounds on iterated intervals (e.g., 0 ≤ i ≤ n²) or permitting two independent index variables—they obtain a class they call bound‑polynomial. They prove that satisfiability for this extended class is again undecidable. The proof reduces from Diophantine equation solving: the polynomial bounds enable encoding of arbitrary integer equations, and the existence of a satisfying assignment corresponds to the existence of integer solutions, a known undecidable problem (Matiyasevich’s theorem). Thus, the decidability result hinges on the very tight linear and single‑index restrictions.

In the discussion, the paper highlights several practical implications. Bound‑linear schemata can model many verification tasks where a system is parameterised by a size variable (e.g., number of processes, length of a pipeline) yet still admits automated reasoning. The calculus can be implemented as a stand‑alone solver or integrated with existing SAT/SMT engines, using the calculus to eliminate iterated connectives before handing the resulting propositional core to a conventional solver. Conversely, the undecidability of modest extensions warns designers that adding seemingly harmless arithmetic features (quadratic bounds, multiple counters) may render automated analysis infeasible.

Overall, the work makes a clear contribution to the theory of parameterised propositional reasoning: it delineates a maximal expressive fragment (bound‑linear) that remains algorithmically tractable, provides a concrete decision procedure via reduction to regular schemata and a terminating calculus, and demonstrates that any relaxation of the linear, single‑index discipline quickly leads back to undecidability. This balance between expressive power and decidability offers a solid foundation for future tools targeting infinite families of propositional specifications.