Computing the extremal nonnegative solutions of the M-tensor equation with a nonnegative right side vector

TL;DR

This paper introduces new, efficient numerical methods for computing extremal nonnegative solutions of M-tensor equations with nonnegative right sides, providing theoretical convergence guarantees and extending to Z-tensors and some negative right side elements.

Contribution

The paper provides shorter, more effective proofs of extremal solution existence and develops numerical methods with proven linear convergence for M-tensor equations.

Findings

New proofs for extremal solution existence

Numerical methods with linear convergence

Extensions to Z-tensors and negative right side elements

Abstract

We consider the tensor equation whose coefficient tensor is a nonsingular M-tensor and whose right side vector is nonnegative. Such a tensor equation may have a large number of nonnegative solutions. It is already known that the tensor equation has a maximal nonnegative solution and a minimal nonnegative solution (called extremal solutions collectively). However, the existing proofs do not show how the extremal solutions can be computed. The existing numerical methods can find one of the nonnegative solutions, without knowing whether the computed solution is an extremal solution. In this paper, we present new proofs for the existence of extremal solutions. Our proofs are much shorter than existing ones and more importantly they give numerical methods that can compute the extremal solutions. Linear convergence of these numerical methods is also proved under mild assumptions. Some of our discussions also allow the coefficient tensor to be a Z-tensor or allow the right side vector to have some negative elements.