Multivariate trace inequalities (Q521995)

The paper provides general extensions of the famous Golden-Thomson and Araki-Lieb-Thirring inequalities to arbitrarily many (but a finite number of) matrices via asymptotic spectral pinching and complex interpolation theory. Namely, the authors introduce the family of probability densities \[ \beta_r(t)=\frac{\sin(\pi r)}{2r\bigl(\cosh(\pi t)+\cos(\pi r)\bigr)}\quad (0<r<1) \] and the limiting cases \[ \beta_0(t):=\lim_{r\searrow0}\beta_r(t)=\frac\pi2\left(\cosh(\pi t)+1\right),\quad \beta_1(t):=\lim_{r\nearrow1}\beta_r(t)=\delta(t), \] where \(\delta(t)\) is the Dirac delta distribution. It is shown that if \(r\in(0,1]\) and \(\{A_k\}_{k=1}^n\) are positive semi-definite matrices, then \[ \log\left\|\left|\prod_{k=1}^nA_k\right|^{\frac1r}\right\|_p\leq\int_{-\infty}^{\infty}\beta_r(t)\log\left\|\prod_{k=1}^nA_k^{1+\mathrm{i}t}\right\|_pdt \] where \(\|\cdot\|_p\) denotes the Schatten \(p\)-norm. As \(r\to0\), for \(\left\{H_k\right\}_{k=1}^n\) Hermitian matrices it yields the inequality \[ \log\left\|\exp\left(\sum_{k=1}^nH_k\right)\right\|_p\leq\int_{-\infty}^{\infty}\beta_0(t)\log\left\|\prod_{k=1}^n\exp\bigl((1+\mathrm{i}t)H_k\bigr)\right\|_pdt \] which contains the Golden-Thomson inequality as a special case when \(n=2\) and \(p=2\). As an application of this result, the monotonicity inequality of quantum entropy is improved. In fact, if for the density matrices \(\rho\), \(\sigma\) the Umegaki relative entropy is denoted by \(D(\rho||\sigma)=\mathrm{tr}\,\rho(\log\rho-\log\sigma)\) and \(\mathcal{N}\) is a trace-preserving completely positive map, then \[ D(\rho||\sigma)-D(\mathcal{N}(\rho)||\mathcal{N}(\sigma)) \] can be bounded from below in terms of the measured relative entropy and the so-called Petz recovery map.