On a special case of the Herbert Stahl theorem (Q515966)

Let \(A\) and \(B\) be \(n\times n\) Hermitian matrices and define \(f_{A,B}(t):=\) trace exp \((tA+B)\) for \(t\in\mathbb{C}\). The BMV conjecture states that \(f_{A,B}\) is representable as a bilateral Laplace transform of a nonnegative measure \(d\sigma_{A,B}(\lambda)\) on a compact real subinterval of \((-\infty,\infty)\) (see [\textit{D. Bessis} et al., J. Math. Phys. 16, 2318--2325 (1975; Zbl 0976.82501)]). This conjecture was recently proved in a long paper by \textit{H. R. Stahl} [Acta Math. 211, No. 2, 255--290 (2013; Zbl 1325.81089)] (simplifications have been given by A. Eremenko). The present paper gives an elementary proof of a different kind in the special case where \(\mathrm{rank~}A=1\) based on the theory of exponentially convex functions as follows. A real-valued, nonnegative function \(f\) of a real variable is called exponentially convex if it is continuous and for each \(N\) and every choice of real numbers \(t_{1},\dots,t_{N}\) the \(N\times N\) matrix \([f(t_{i}+t_{j})]\) is positive semidefinite. Using known results the author shows that the Stahl theorem is equivalent to proving that \(f_{A,B}\) is exponentially convex. He then gives a simple matrix argument to prove that \(f_{A,B}\) is exponentially convex in the special case where \(\mathrm{rank}\)~\(A=1\). He does not discuss whether his argument might extend to other cases.