Loewner matrices and operator convexity (Q1024195): Difference between revisions

Let \(f\) be a \(C^1\)-function from \((0,\infty)\) into itself, and assume further that \(f\) can be extended to a continuous function on \([0,\infty)\). The function \(f\) is called operator monotone if for any two positive semidefinite (p.s.d.) matrices \(A\) and \(B\) such that \(A\leq B\) (that is, \(B-A\) is p.s.d.) we have \(f(A)\leq f(B)\). Given a list \(p_1,\dots,p_n\) of distinct points in \((0,\infty)\) define the Loewner matrix \(L_f(p_1,\dots,p_n)\) to be the \(n\times n\) matrix whose \((i,j)\)th entry is \((f(p_i)-f(p_j)) /(p_i-p_j)\) (\(=f'(p_i)\) when \(i=j\)). In 1934 K. Loewner showed that \(f\) is operator monotone if and only if all the matrices \(L_f (p_1,\dots,p_n)\) are p.s.d. The function \(f\) above is called operator convex if it is a \(C^{2}\)-function and for any p.s.d. matrices \(A\) and \(B\) of the same size we have \[ f((1-t)A+tB)\leq(1-t)f(A)+tf(B) \] for all \(t\in[0,1]\). A Hermitian matrix \(A\) is called conditionally positive definite if \((x,Ax)\geq0\) for every vector \(x\) whose entries sum to \(0\); and \(A\) is conditionally negative definite if the reverse inequality always holds. The main results of the present paper are the following. If \(f\) is an operator convex function, then all Loewner matrices are conditionally negative definite. If \(f(t)=tg(t)\) and \(g\) is an operator convex function, then all Loewner matrices are conditionally positive definite. Appropriate converses hold. More detailed results are obtained for the special case where \(f(t):=t^r\).