Revisiting Bourgain-Kalai and Fourier Entropies

The total influence of a function is a central notion in analysis of Boolean functions, and characterizing functions that have small total influence is one of the most fundamental questions associated with it. The KKL theorem and the Friedgut junta theorem give a strong characterization of such functions whenever the bound on the total influence is $o(\log n)$, however, both results become useless when the total influence of the function is $\omega(\log n)$. The only case in which this logarithmic barrier has been broken for an interesting class of function was proved by Bourgain and Kalai, who focused on functions that are symmetric under large enough subgroups of $S_n$. In this paper, we revisit the key ideas of the Bourgain-Kalai paper. We prove a new general inequality that upper bounds the correlation between a Boolean function $f$ and a real-valued, low degree function $g$ in terms of their norms, Fourier coefficients and total influences. Some corollaries of this inequality are: 1. The Bourgain-Kalai result. 2. A slightly weaker version of the Fourier-Entropy Conjecture. More precisely, we prove that the Fourier entropy of the low-degree part of $f$ is at most $O(I[f]\log^2 I[f])$, where $I[f]$ is the total influence of $f$. In particular, we conclude that the Fourier spectrum of a Boolean function is concentrated on at most $2^{O(I[f]\log^2 I[f])}$ Fourier coefficients. 3. Using well-known learning algorithms of sparse functions, the previous point implies that the class of functions with sub-logarithmic total influence (i.e. at most $O(\log n/(\log \log n)^2)$) is learnable in polynomial time, using membership queries.