On the Subadditivity of Mutual f-Information

Abstract

We address an open problem raised by Polyanskiy and Wu in their textbook concerning the subadditivity of mutual \(f\)-information. We study two notions of generalized mutual information: mutual \(f\)-information, defined by \(\widetilde{I}_f(X;Y)=D_f(p_{XY}\|p_Xp_Y)\), and variational \((G,f)\)-information, obtained by minimizing an average of transformed \(f\)-divergences. For the first notion, we derive a sufficient condition on \(g(x)=x^2f''(x)\) that guarantees subadditivity under conditionally independent observations. This condition yields explicit families beyond the previously known examples generated by \(x\log x\) and \((x-1)\log x\). Our proof relies on the concavity of a generalized variance functional. For the variational notion, we extend existing sufficient conditions beyond the transforms \(G(x)\in\{x,\log(1+x),-\log(1-x)\}\). Using a comparison inequality for \(G^{-1}(G(A)+G(B))\), we obtain differential criteria for concave and convex transforms. Explicit constructions illustrate the resulting classes of subadditive information measures.