A Linearly Convergent Algorithm for Computing the Petz-Augustin Information

Abstract: We propose an iterative algorithm for computing the Petz-Augustin information of order $\alpha\in(1/2,1)\cup(1,\infty)$. The optimization error is guaranteed to converge at a rate of $O\left(\vert 1-1/\alpha \vert^T\right)$, where $T$ is the number of iterations. Let $n$ denote the cardinality of the input alphabet of the classical-quantum channel, and $d$ the dimension of the quantum states. The algorithm has an initialization time complexity of $O\left(n d^{3}\right)$ and a per-iteration time complexity of $O\left(n d^{2}+d^3\right)$. To the best of our knowledge, this is the first algorithm for computing the Petz-Augustin information with a non-asymptotic convergence guarantee.