Quantum divergences with \(p\)-power means (Q2226462)

For Hermitian matrices \(X\) and \(Y\), let \(\Phi_p=Tr\left[\frac{X+Y}{2}-\left(\frac{X^p+Y^p}{2}\right)^\frac{1}{p}\right] \)and \(\Psi_p=\ln{Tr\left(\frac{X+Y}{2}\right)}-\ln{Tr\left(\left(\frac{X^p+Y^p}{2}\right)^\frac{1}{p}\right)}\), where \(0<p<1\). Then \(\Phi_p\) and \(\Psi_p\) both satisfy the first condition of a metric: \(d\left(X,Y\right)\geq0\) and \(d\left(X,Y\right)=0\ \Leftrightarrow X=Y\). Functions satisfying such condition are called divergences in this paper and other literature. Divergences are often used to describe the difference between two data sets. In this paper, the authors present other properties of the divergences defined above. Particularly, the Taylor expansion of those divergences is discussed as well as data processing inequality, i.e., \(\Phi\left(\mathcal{E}X,\mathcal{E}Y\right)\le\Phi\left(X,Y\right)\), where \(\mathcal{E}\) is a completely positive trace preserving map. Furthermore, the Hellinger distance with log Euclidean mean and the quantum Jensen-Shannon divergence are revisited as consequences of \(\lim_{p\rightarrow0^+}{\Phi_p}\) and \(\lim_{p\rightarrow1^-}{\frac{\Phi_p}{1-p}}\), respectively.