A key inequality for functions of matrices (Q5929753)

Let \({\Phi}\) be a unital positive linear mapping, i.e., the identity is mapped to the identity, and \({\Phi}(B) \geq 0\) for every positive semi-definite operator \(B\) (a typical example is \({\Phi}(B) = \sum_{j=1}^n P_j B P_j\) with \(P_j = P_j^*\) and \(P_jP_k = P_j \delta_{jk}\)). The authors prove: If \(f: [m,M] \rightarrow {\mathbf R}\) is convex, and if \(L(t) = [(t-m) f(M) + (M-t)f(m)]/(M-m),\) then \({\Phi}(f(A)) \leq L({\Phi}(A)).\) Equality holds for every \({\Phi}\) if and only if either \(f\) is linear or the spectrum of \(A\) is contained in \(\{ m,M \}.\) In terms of matrices, this result gives a substantial generalization and improvement of a variety of previously studied matrix inequalities.