English

From Proof Complexity to Circuit Complexity via Interactive Protocols

Computational Complexity 2024-05-06 v1

Abstract

Folklore in complexity theory suspects that circuit lower bounds against NC1\mathbf{NC}^1 or P/poly\mathbf{P}/\operatorname{poly}, currently out of reach, are a necessary step towards proving strong proof complexity lower bounds for systems like Frege or Extended Frege. Establishing such a connection formally, however, is already daunting, as it would imply the breakthrough separation NEXP⊈P/poly\mathbf{NEXP} \not\subseteq \mathbf{P}/\operatorname{poly}, as recently observed by Pich and Santhanam (2023). We show such a connection conditionally for the Implicit Extended Frege proof system (iEF\mathsf{iEF}) introduced by Kraj\'i\v{c}ek (The Journal of Symbolic Logic, 2004), capable of formalizing most of contemporary complexity theory. In particular, we show that if iEF\mathsf{iEF} proves efficiently the standard derandomization assumption that a concrete Boolean function is hard on average for subexponential-size circuits, then any superpolynomial lower bound on the length of iEF\mathsf{iEF} proofs implies #P⊈FP/poly\#\mathbf{P} \not\subseteq \mathbf{FP}/\operatorname{poly} (which would in turn imply, for example, PSPACE⊈P/poly\mathbf{PSPACE} \not\subseteq \mathbf{P}/\operatorname{poly}). Our proof exploits the formalization inside iEF\mathsf{iEF} of the soundness of the sum-check protocol of Lund, Fortnow, Karloff, and Nisan (Journal of the ACM, 1992). This has consequences for the self-provability of circuit upper bounds in iEF\mathsf{iEF}. Interestingly, further improving our result seems to require progress in constructing interactive proof systems with more efficient provers.

Keywords

Cite

@article{arxiv.2405.02232,
  title  = {From Proof Complexity to Circuit Complexity via Interactive Protocols},
  author = {Noel Arteche and Erfan Khaniki and Ján Pich and Rahul Santhanam},
  journal= {arXiv preprint arXiv:2405.02232},
  year   = {2024}
}

Comments

A conference version of this work is accepted to the 51st EATCS International Colloquium on Automata, Languages and Programming (ICALP 2024)