arXiv · 2611.11832
Simulating Nondeterministic Turing Machines in Subexponential Deterministic Time
Abstract
We show that every multitape nondeterministic Turing machine can be simulated deterministically in subexponential time. Specifically, we show that the recent algorithms for $3$-$\mathsf{SAT}$ presented by OpenAI and Anthropic imply $\mathsf{NTIME}[t]\subseteq\mathsf{DTIME}[2^{O(t \log \log t/\sqrt{\log t})}]$, and that a stronger consequence of their proof yields \[\mathsf{NTIME}[t] \subseteq\mathsf{DTIME}[2^{O(t /\sqrt{\log t})}]. \] More generally, we show that any subexponential time algorithm for solving circuit-CSPs transfers to a deterministic simulation of $\mathsf{NTIME}$ with the same savings. Our results follow from new generalizations of Cook--Levin, which may be of independent interest.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Carmen Strassle, Bernardo Subercaseaux. 2026-11-09. Simulating Nondeterministic Turing Machines in Subexponential Deterministic Time. https://arxiv.science/abs/2611.11832
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.