The canonical pairs of bounded depth Frege systems

TL;DR

This paper characterizes the canonical pairs of bounded depth Frege proof systems using combinatorial games, generalizing monotone Boolean circuits and feasible interpolation, but leaves open the challenge of proving lower bounds for these games.

Contribution

It introduces a combinatorial game-based characterization of canonical pairs in depth~d Frege systems, extending the monotone feasible interpolation framework.

Findings

Canonical pairs are equivalent to certain game pairs involving winning strategies.

Games can be viewed as generalizations of monotone Boolean circuits.

The framework suggests new directions for proving lower bounds in proof complexity.

Abstract

The canonical pair of a proof system $P$ is the pair of disjoint NP sets where one set is the set of all satisfiable CNF formulas and the other is the set of CNF formulas that have $P$-proofs bounded by some polynomial. We give a combinatorial characterization of the canonical pairs of depth~$d$ Frege systems. Our characterization is based on certain games, introduced in this article, that are parametrized by a number~$k$, also called the depth. We show that the canonical pair of a depth~$d$ Frege system is polynomially equivalent to the pair $(A_{d+2},B_{d+2})$ where $A_{d+2}$ (respectively, $B_{d+1}$) are depth {$d+1$} games in which Player~I (Player II) has a positional winning strategy. Although this characterization is stated in terms of games, we will show that these combinatorial structures can be viewed as generalizations of monotone Boolean circuits. In particular, depth~1 games are essentially monotone Boolean circuits. Thus we get a generalization of the monotone feasible interpolation for Resolution, which is a property that enables one to reduce the task of proving lower bounds on the size of refutations to lower bounds on the size of monotone Boolean circuits. However, we do not have a method yet for proving lower bounds on the size of depth~$d$ games for $d>1$.