Branch‑and‑Bound Shortens Unsatisfiability Proofs

A team of computer scientists published a novel approach that dramatically reduces the size of resolution proofs produced when proving logical formulas unsatisfiable. The work, titled “How to Discover Short, Shorter, and the Shortest Proofs of Unsatisfiability,” was released on August 25 2026 in the Journal of Artificial Intelligence Research (JAIR). It lists five co‑authors—Konstantin Sidorov, Koos van der Linden, Gonçalo Homem de Almeida Correia, Mathijs de Weerdt, and Emir Demirović—and appears under the DOI ¹].

Modern SAT solvers are indispensable tools for software verification, planning, and combinatorial design. While they can quickly determine whether a set of Boolean clauses is satisfiable or not, their unsatisfiability proofs often contain thousands of inference steps, making them difficult to analyze manually. This paper tackles precisely that issue by aiming to find the shortest possible resolution proofs for given unsatisfiable instances.

The authors formalize the problem as an optimization task: among all valid resolution derivations leading from the original clause set to contradiction, identify one with minimal length. To explore this space efficiently, they introduce a branch‑and‑bound algorithm that systematically constructs candidate proof trees while pruning branches that cannot beat the current best solution. A key innovation is the layer list representation that removes symmetries caused by arbitrary clause permutations; this reduces redundant exploration and speeds up the search dramatically.

Experimental results demonstrate significant gains over baseline methods. On benchmark formulas drawn from standard SAT competitions, the new approach consistently produced shorter proofs while maintaining comparable runtime performance. The work also shows that many existing solver-generated proofs are far from optimal, underscoring the value of targeted minimization techniques ².

From an expert perspective, the research addresses limitations inherent in modern SAT solvers which often produce lengthy proofs despite being able to determine satisfiability quickly (e.g., DPLL algorithms). By borrowing ideas common in constraint satisfaction problems—such as systematic exploration and aggressive pruning—the proposed method brings a fresh angle to proof compression. Shorter resolution proofs can aid downstream tasks like program verification, where proof size directly impacts readability and computational overhead. Moreover, the technique provides insights into the theoretical gap between solver outputs and true optimum, potentially guiding future solver design toward more efficient proof generation ².

The paper’s contribution lies both in practical applicability for software engineers needing compact correctness certificates and in advancing our understanding of proof complexity within automated reasoning systems. As verification demands grow with increasingly complex software stacks, tools that deliver tighter guarantees without sacrificing speed will become essential.