Random CNFs are Hard for Cutting Planes

The random k-SAT model is the most important and well-studied distribution over k-SAT instances. It is closely connected to statistical physics; it is used as a testbench for satisfiability algorithms, and average-case hardness over this distribution has also been linked to hardness of approximation via Feige’s hypothesis. We prove that any Cutting Planes refutation for random k-SAT requires exponential size, for k that is logarithmic in the number of variables, in the (interesting) regime where the number of clauses guarantees that the formula is unsatisfiable with high probability.

💡 Research Summary

The paper “Random CNFs are Hard for Cutting Planes” establishes the first non‑trivial exponential lower bounds for Cutting Planes (CP) refutations of random k‑SAT formulas when the clause width k grows logarithmically with the number of variables. The authors focus on the regime where the number of clauses m is large enough to make the formula unsatisfiable with high probability (i.e., clause density Δ = m/n exceeds the known unsatisfiability threshold ≈ 2^k ln 2).

The technical contribution proceeds in three stages. First, the authors formalize a hierarchy of proof systems that includes standard CP, its semantic variant (Semantic CP), and communication‑complexity‑restricted versions (CC‑Proofs). They show that low‑weight CP proofs correspond to O(log n)‑bit communication protocols, and that any CP proof can be viewed as a one‑round real‑communication (RCC) proof.

Second, they prove a novel equivalence: the existence of a short CP (or Semantic CP, CC) refutation for any unsatisfiable CNF is equivalent to the existence of a small monotone circuit separating two associated monotone sets. To obtain this equivalence they partition the variables into two sets X and Y, define a search problem Search(F) that, given an assignment split between X and Y, outputs a violated clause, and then translate this search problem into a Karchmer‑Wigderson (KW) game for a monotone constraint‑satisfaction problem (CSP). This generalizes the classic interpolation method, which previously applied only to “split” formulas such as the clique‑coclique family.

Third, they apply the equivalence to random d‑CNF formulas with d = Θ(log n). By exploiting the symmetry of the random formula distribution, they use the symmetric method of approximations to prove that any monotone circuit separating the “yes” and “no” instances of the derived monotone CSP must have size exp(Ω(n)). Consequently, any CP refutation of the original random CNF must also have exponential size. The proof works uniformly for CP, Semantic CP, and the communication‑restricted CC‑proofs.

The paper situates its results within a broader context. It reviews known exponential lower bounds for Resolution, Polynomial Calculus, and Res(k), noting that while these systems are well‑understood for random formulas, CP has remained elusive except for the special clique‑coclique formulas. It also discusses the connection to Feige’s hypothesis, which posits that certifying unsatisfiability of random 3‑SAT (and more generally random d‑SAT) is computationally hard; the authors’ CP lower bound provides evidence that even proof‑system‑based algorithms cannot avoid exponential effort in the average case.

In summary, the authors develop a powerful new proof‑complexity technique that bridges CP refutations and monotone circuit complexity, and they leverage it to show that random CNFs with logarithmic clause width are exponentially hard for Cutting Planes. This advances our understanding of average‑case hardness for strong algebraic proof systems and reinforces the conjectured difficulty of random SAT instances in both theoretical and practical settings.