A generalized Cauchy-Binet formula and applications to total positivity and majorization (Q1107928)

The classical Cauchy-Binet formula expresses a generalized matrix function founded on a matrix product \(C=AB\) in terms of the corresponding functions founded on A and B separately. The original matrix function was the determinant. Thus, if \(A=(A(x_ i,y_ j))^ n_{i,j=1}\) is an \(n\times n\) matrix, then \[ \det (A)=\sum_{\sigma}sign(\sigma)\prod^{n}_{i=1}A(x_ i,y_{\sigma (j)}), \] where \(\sigma\) varies over the permutations on n symbols. If \(x=(x_ 1,...,x_ n)\) and \(y=(y_ 1,...,y_ n),\) then we can define a kernel function K by \[ K(x,y)=\prod^{n}_{i=1}A(x_ i,y_ i). \] The authors develop an extended Cauchy-Binet formula for more general kernel functions than the particular K above, and eventually go on to consider integral analogues of the form \[ D_ K(x,y)=\int_{G}\chi (g)K(x,gy)dg, \] where x and y are elements of an abstract space \({\mathcal X}\), \(K: {\mathcal X}\times {\mathcal X}\to {\mathbb{R}},\) where G denotes a (compact) group acting on \({\mathcal X}\), dg denotes the Haar measure on G, and where \(\chi\) denotes a character of G. Most of the paper is devoted to the study of a generalized notion of total positivity for certain classes of kernel functions K. Applications to multivariate statistical analysis are briefly discussed.