English

Local reductions

Computational Complexity 2014-04-09 v3

Abstract

We reduce non-deterministic time T2nT \ge 2^n to a 3SAT instance ϕ\phi of quasilinear size ϕ=TlogO(1)T|\phi| = T \cdot \log^{O(1)} T such that there is an explicit circuit CC that on input an index ii of logϕ\log |\phi| bits outputs the iith clause, and each output bit of CC depends on O(1)O(1) input bits. The previous best result was CC in NC1^1. Even in the simpler setting of polynomial size ϕ=\poly(T)|\phi| = \poly(T) the previous best result was CC in AC0^0. More generally, for any time TnT \ge n and parameter rnr \leq n we obtain log2ϕ=max(logT,n/r)+O(logn)+O(loglogT)\log_2 |\phi| = \max(\log T, n/r) + O(\log n) + O(\log\log T) and each output bit of CC is a decision tree of depth O(logr)O(\log r). As an application, we tighten Williams' connection between satisfiability algorithms and circuit lower bounds (STOC 2010; SIAM J. Comput. 2013).

Keywords

Cite

@article{arxiv.1311.3171,
  title  = {Local reductions},
  author = {Hamid Jahanjou and Eric Miles and Emanuele Viola},
  journal= {arXiv preprint arXiv:1311.3171},
  year   = {2014}
}
R2 v1 2026-06-22T02:06:45.421Z