A variational expression for the relative entropy (Q1114132): Difference between revisions

Let M be a von Neumann algebra and let \(\phi\) and \(\omega\) be faithful normal states on M. Let \(\Delta\) (\(\phi\),\(\omega)\) be the relative modular operator, and let \(\Omega\) be the normalized vector representative of \(\omega\) chosen in the natural positive cone P. Then the relative entropy S(\(\phi\),\(\omega)\) may be defined by the formula \[ S(\phi,\omega)=-(\Omega,\log \Delta (\phi,\omega)\Omega), \] and it is proved that S(\(\phi\),\(\omega)\) may also be computed from the following variational formula \[ S(\phi,\omega)=\sup \{\omega (h)-\log \phi^ h(1)\}, \] where the supremum is taken over all \(h=h^*\) in M, and \(\phi^ h\) is the perturbed state, introduced by Araki. Specifically, \(\phi^ h\) is obtained from \(\phi\) by application of the time-ordered perturbation expansional to the normalized vector in P which defines \(\phi\). It is proved further that, if the supremum in the formula for S(\(\phi\),\(\omega)\) is attained, at some h, then \(\omega =\phi^ h(1)^{-1}\phi^ h\), i.e., the two states differ by inner perturbation by the element h in M where the maximum is attained. The proofs are based on ideas of \textit{H. Araki} from the paper on the Golden-Thompson inequality, Commun. Math. Phys. 34, 167-178 (1973; Zbl 0274.46048).