On the probability of nonexistence in binomial subsets (Q2184826)

This paper provides a strengthening of what is know as Janson's inequalities. The context is a binomial random subset of a finite set \(\Omega\), where each element \(\omega\) is present independently with probability \(p_\omega\). The ground set \(\Omega\) is equipped with a family of subsets \(\mathcal{X}\). For each such subset \(\gamma\) we have an indicator random variable \(X_\gamma\) that is equal to 1 precisely when \(\gamma\) is entirely included in the random subset. We are interested in \(X=\sum_{\gamma \in \mathcal{X}} X_\gamma\), which counts those subsets/events which have been realised. In particular, we are interested in \(\mathbb{P}(X=0)\). The classic Harris's inequality and Janson's inequality provide lower and upper bounds, respectively. These have had numerous applications in percolation theory, probabilistic combinatorics and analytic number theory. The authors tie down this quantity, under only mild assumptions, expressing it as an exponential of the finite alternating sum of the first \(k\) mixed cumulants of the family \((X_\gamma)_{\gamma \in \mathcal{X}}\) plus an error term which depends on the joint \(k\)th moments of the \(X_i\)s. Moreover, this holds for any \(k\). Furthermore, the authors give applications of this to random hypergraphs (being free from a certain hypergraph) and to analytic combinatorics (being free from arithmetic progressions of given length).