New parallel algorithm for the Proof Analysis Problem

Researchers developed a way to extract satisfying assignments using low-depth circuit classes.

A parallel algorithm was created for the Proof Analysis Problem of Arteche, Atserias, de Rezende, and Khaniki (FOCS 2025). This problem involves a CNF formula φ\varphi on nn variables and a proof π\pi of a size-ss Resolution lower bound.

The algorithm extracts a satisfying assignment for φ\varphi when one exists. It uses polynomial-size TC0\mathbf{TC}^0 circuits and quasipolynomial-size AC0\mathbf{AC}^0 circuits. TC0\mathbf{TC}^0 and AC0\mathbf{AC}^0 are low-depth circuit classes.

The work uses a new formalization of the Atserias-Müller lower bound in the relativized theory T24(α)\mathsf{T}^4_2(\alpha). This theory corresponds to depth-4 Frege systems. This approach provides quasipolynomial-size AC0\mathbf{AC}^0-Frege proofs for many Resolution lower bounds.

The researchers show the Proof Analysis Problem for AC0\mathbf{AC}^0-Frege is NP\mathbf{NP}-hard under quasipolynomial-time reductions. This rules out certain Atserias-Müller-type approaches to the hardness of automating constant-depth Frege. Automating AC0\mathbf{AC}^0-Frege on Resolution lower bound formulas suffices to automate it on all formulas.

Key facts

  • The algorithm performs assignment extraction in low-depth circuit classes.
  • The method works for Resolution lower bounds ¬Refs(φ)\neg \mathrm{Ref}_s(\varphi) for all ss, including s=nO(1)s = n^{O(1)}.
  • The Proof Analysis Problem for AC0\mathbf{AC}^0-Frege is NP\mathbf{NP}-hard under quasipolynomial-time reductions.

All sections

Edition