Title

"A new quantum version of f-divergence"

Speaker

Keiji Matsumoto (NII)

Abstract

Quantum f-Divergence proposed by Petz (see [1] for comprehensive review) is widely studied, especially due to its implication to perfect error correction. In [2], the present author proposed a new quantum version of "generalized fidelity". This is essentially a new version of quantum f-divergence (Dfmax, below) when f is operator monotone decreasing. In the definition, we supposed two states are strictly positive. In this paper, the definition is generalized to include operator convex function f and, to states which may have null eigenspace. Not only the definition, also the explicit formula for Dfmax(ρ||σ) when ρ and σ have null eigenspaces is given, and several properties of the quantity is studied (convexity, monotonicity by CPTP maps.) Also, the condition Dfmax(ρ||σ) = Dfmax(Λ(ρ)||Λ(σ)), where Λ is CPTP map, is studied. As is well-known, the analogous condition for Df implied reversibility of Λ, or "sufficiency" of the map Λ. Dfmax-version of the condition gives another type of "sufficiency" property of Λ.

[1] F. Hiai, M. Mosonyi, D. Petz, and C. Beny, Rev. Math. Phys. 23, 691 (2011).
[2] K. Matsumoto, arXiv:1311.4722.

Designed by CSS.Design Sample