Sharp Analysis of Gaussian Rounding for Boolean Max k-CSP

Abstract

In this note, we show that the approximation algorithm for Boolean Max $k$-CSP presented in [Makarychev and Makarychev 2014] yields a $(1-o_k(1))k/2^k$ approximation, as conjectured in [Makarychev and Makarychev 2017]. This improves the previous guarantee of $(0.626612-o_k(1))k/2^k$ from [Makarychev and Makarychev 2014] and asymptotically matches the known hardness results. The result is a short corollary of the Gaussian stochastic domination theorem of Mulgund.