Conditional Information Inequalities and Combinatorial Applications

Abstract

We show that the inequality $H(A \mid B,X) + H(A \mid B,Y) \le H(A\mid B)$ for jointly distributed random variables $A,B,X,Y$, which does not hold in general case, holds under some natural condition on the support of the probability distribution of $A,B,X,Y$. This result generalizes a version of the conditional Ingleton inequality: if for some distribution $I(X: Y \mid A) = H(A\mid X,Y)=0$, then $I(A : B) \le I(A : B \mid X) + I(A: B \mid Y) + I(X : Y)$. We present two applications of our result. The first one is the following easy-to-formulate combinatorial theorem: assume that the edges of a bipartite graph are partitioned into $K$ matchings such that for each pair (left vertex $x$, right vertex $y$) there is at most one matching in the partition involving both $x$ and $y$; assume further that the degree of each left vertex is at least $L$ and the degree of each right vertex is at least $R$. Then $K\ge LR$. The second application is a new method to prove lower bounds for biclique coverings of bipartite graphs.