This generalized inequality has a nice interpretation in the context of privacy. See https://arxiv.org/pdf/1511.02381.pdf or message me for the proof.

]]>Best wishes for you,

Rodrigo Nunes Cal ]]>

http://link.springer.com/chapter/10.1007%2F11889342_58#page-1

In particular, it implies that the solution to Thm 9.2 in Csiszar-Korner (and hence the second term in the new result) can be simplified to

which can be solved via linear programming. ]]>