The Matrix Multidisk Problem

Abstract

The solutions of a class of matrix optimization problems (including the Nehari problem and its multidisk generalization) can be identified with the solutions of an abstract operator equation of the form T(., ., .) = 0. This equation can be solved numerically by Newton's method if the differentialT' of T is invertible at the points of interest. This is typically too difficult to verify. However, it turns out that under reasonably broad conditions we can identify T' as the sum of a block Toeplitz operator and a compact block Hankel operator. Moreover, we can show that the block Toeplitz operator is a Fredholm operator and and in some cases can calculate its Fredholm index. Thus, T' will also be a Fredholm operator of the same index. In a number of cases that have been checked todate, numerical methods perform well when the Fredholm index is equal to zero and poorly otherwise. The main focus of this paper is on the multidisk problem alluded to above. However, a number of analogies with existing work on matrix optimization have been worked out and incorporated.