Enhancements to the von Neumann trace inequality (Q2469503)

Let \(\lambda_i(X)\) and \(\sigma_i(X)\) be the eigenvalues and the singular values, respectively, of a square complex matrix \(X\). Assume that \[ A=PA_1Q \quad \text{and}\quad B=RB_1S\, , \] are the singular value decomposition of the complex matrices of order \(n\), \(A\) and \(B\), and let \(V\) and \(W\) be unitary matrices of the same order \(n\). Setting \(T=VPQWRS\), this note shows that \[ \text{ Re}\, \text{ tr}\, (VAWB)\leq \sum_{i=1}^n| \text{Re}\,\lambda_i(T)| \sigma_i(A)\sigma_i(B), \] and \[ \text{ Im}\, \text{ tr}\, (VAWB)\leq \sum_{i=1}^n| \text{Im}\,\lambda_i(T)| \sigma_i(A)\sigma_i(B), \] with \(| \text{Re}\,\lambda_i(T)| \) and \(| \text{Im}\,\lambda_i(T)| \) arranged in decreasing order. The first inequality is tighter than \textit{J. von Neumann}'s inequality [Mitt. Forsch.-Inst. Math. Mech. Univ. Tomsk 1, 286--299 (1937; Zbl 0017.09802)].