Mutual information, matroids and extremal dependencies

In this paper, it is shown that the rank function of a matroid can be represented by a "mutual information function" if and only if the matroid is binary. The mutual information function considered is the one measuring the amount of information between the inputs (binary uniform) and the output of a multiple access channel (MAC). Moreover, it is shown that a MAC whose mutual information function is integer valued is "equivalent" to a linear deterministic MAC, in the sense that it essentially contains at the output no more information than some linear forms of the inputs. These notes put emphasis on the connection between mutual information functionals and rank functions in matroid theory, without assuming prior knowledge on these two subjects. The first section introduces mutual information functionals, the second section introduces basic notions of matroid theory, and the third section connects these two subjects. It is also shown that entropic matroids studied in the literature correspond to specific cases of MAC matroids.