A Characterisation of Anti-Lowner Functions

Koenraad M. R. Audenaert

Published 2010-08-17, updated 2011-04-27Version 2

According to a celebrated result by L\"owner, a real-valued function $f$ is operator monotone if and only if its L\"owner matrix, which is the matrix of divided differences $L_f=(\frac{f(x_i)-f(x_j)}{x_i-x_j})_{i,j=1}^N$, is positive semidefinite for every integer $N>0$ and any choice of $x_1,x_2,...,x_N$. In this paper we answer a question of R. Bhatia, who asked for a characterisation of real-valued functions $g$ defined on $(0,+\infty)$ for which the matrix of divided sums $K_g=(\frac{g(x_i)+g(x_j)}{x_i+x_j})_{i,j=1}^N$, which we call its anti-L\"owner matrix, is positive semidefinite for every integer $N>0$ and any choice of $x_1,x_2,...,x_N\in(0,+\infty)$. Such functions, which we call anti-L\"owner functions, have applications in the theory of Lyapunov-type equations.