Tension Bounds for Information Complexity (1408.6285v1)

Abstract: The main contribution of this work is to relate information complexity to "tension" [Prabhakaran and Prabhakaran, 2014] - an information-theoretic quantity defined with no reference to protocols - and to illustrate that it allows deriving strong lower-bounds on information complexity. In particular, we use a very special case of this connection to give a quantitatively tighter connection between information complexity and discrepancy than the one in the work of Braverman and Weinstein (2012) (albeit, restricted to independent inputs). Further, as tension is in fact a multi-dimensional notion, it enables us to bound the 2-dimensional region that represents the trade-off between the amounts of communication in the two directions in a 2-party protocol. This work is also intended to highlight tension as a fundamental measure of correlation between a pair of random variables, with rich connections to a variety of questions in computer science and information theory.