English

On Polynomial Approximations to ${AC}^0$

Computational Complexity 2020-01-01 v2

Abstract

We make progress on some questions related to polynomial approximations of AC0{\rm AC}^0. It is known, by works of Tarui (Theoret. Comput. Sci. 1993) and Beigel, Reingold, and Spielman (Proc. 66th CCC, 1991), that any AC0{\rm AC}^0 circuit of size ss and depth dd has an ε\varepsilon-error probabilistic polynomial over the reals of degree (log(s/ε))O(d)(\log (s/\varepsilon))^{O(d)}. We improve this upper bound to (logs)O(d)log(1/ε)(\log s)^{O(d)}\cdot \log(1/\varepsilon), which is much better for small values of ε\varepsilon. We give an application of this result by using it to resolve a question posed by Tal (ECCC 2014): we show that (logs)O(d)log(1/ε)(\log s)^{O(d)}\cdot \log(1/\varepsilon)-wise independence fools AC0{\rm AC}^0, improving on Tal's strengthening of Braverman's theorem (J. ACM, 2010) that (log(s/ε))O(d)(\log (s/\varepsilon))^{O(d)}-wise independence fools AC0{\rm AC}^0. Up to the constant implicit in the O(d)O(d), our result is tight. As far as we know, this is the first PRG construction for AC0{\rm AC}^0 that achieves optimal dependence on the error ε\varepsilon. We also prove lower bounds on the best polynomial approximations to AC0{\rm AC}^0. We show that any polynomial approximating the OR{\rm OR} function on nn bits to a small constant error must have degree at least Ω~(logn)\widetilde{\Omega}(\sqrt{\log n}). This result improves exponentially on a recent lower bound demonstrated by Meka, Nguyen, and Vu (arXiv 2015).

Cite

@article{arxiv.1604.08121,
  title  = {On Polynomial Approximations to ${AC}^0$},
  author = {Prahladh Harsha and Srikanth Srinivasan},
  journal= {arXiv preprint arXiv:1604.08121},
  year   = {2020}
}
R2 v1 2026-06-22T13:42:37.730Z