A generalization of parallel addition (Q1824183): Difference between revisions

Suppose A and B are two positive operators defined on the Hilbert space H. The parallel sum of A and B is denoted by A:B and is defined by the formula \[ ((A:B)x,x)=\inf \{(Ay,y)+(Bz,z);\quad y+z=x,\quad y,z\in H\}. \] \textit{W. N. Anderson} and \textit{G. E. Trapp} [Linear Algebra Appl. 106, 209-217 (1988; Zbl 0657.15018)] and also \textit{E. L. Pekarev} and \textit{J. L. Smul'jan} [Math. USSR, Izv. 10, 351-370 (1977; Zbl 0351.47031)] have obtained several properties of the parallel addition. The authors consider a cone A of positive operators having a binary operation (A,B)\(\to A\circ B\) with properties identical to parallel addition. In aprticular \(A\circ (\lambda A)=g(\lambda)A\) where \(\lambda \in {\mathbb{R}}^+\) and g is a given strictly positive function on \({\mathbb{R}}^+\). They use a result of \textit{F. Bohnenblust} [Duke Math. J. 6, 627-640 (1940; Zbl 0024.04101)] to show that g must have the form \(g(\lambda)=(1+\lambda^{-p})^{-1/p}\) for some \(0<p<\infty\). They also characterize the (E,p) quasi-units of A in terms of the extreme points of the convex set \(\{\) \(B\in A:\) \(B\leq E\}\).