We give an example of a boolean function whose information complexity is exponentially smaller than its communication complexity. This was first proven recently by Ganor, Kol and Raz (J. ACM 2016) and our work gives a simpler proof of the same result. In the course of this simplification, we make several new contributions: we introduce a new communication lower bound technique, the notion of a \emph{fooling distribution}, which allows us to separate information and communication complexity, and we also give a more direct proof for the information complexity upper bound. We also prove a generalization of Shearer's Lemma that may be of independent interest.
More details and examples added.
We give an example of a boolean function whose information complexity is exponentially
smaller than its communication complexity. Our result simplifies recent work of Ganor, Kol and
Raz (FOCS'14, STOC'15).
Fixed a typo concerning the indices in the proof.
We give an example of a boolean function whose information complexity is exponentially
smaller than its communication complexity. Our result simplifies recent work of Ganor, Kol and
Raz (FOCS'14, STOC'15).
Fixed an error in the previous definition of the distribution.
We give an example of a boolean function whose information complexity is exponentially
smaller than its communication complexity. Our result simplifies recent work of Ganor, Kol and
Raz (FOCS'14, STOC'15).