A descent cautious BFGS method for computing US-eigenvalues of symmetric complex tensors

Abstract

Unitary symmetric eigenvalues (US-eigenvalues) of symmetric complex tensors and unitary eigenvalues (U-eigenvalues) for non-symmetric complex tensors are very important because of their background of quantum entanglement. US-eigenvalue is a generalization of Z-eigenvalue from the real case to the complex case, which is closely related to the best complex rank-one approximations to higher-order tensors. The problem of finding US-eigenpairs can be converted to an unconstrained nonlinear optimization problem with complex variables, their complex conjugate variables and real variables. However, optimization methods often need a first- or second-order derivative of the objective function, and cannot be applied to real valued functions of complex variables because they are not necessarily analytic in their argument. In this paper, we first establish the first-order complex Taylor series and Wirtinger calculus of complex gradient of real-valued functions with complex variables, their complex conjugate variables and real variables. Based on this theory, we propose a norm descent cautious BFGS method for computing US-eigenpairs of a symmetric complex tensor. Under appropriate conditions, global convergence and superlinear convergence of the proposed method are established. As an application, we give a method to compute U-eigenpairs for a non-symmetric complex tensor by finding the US-eigenpairs of its symmetric embedding. The numerical examples are presented to support the theoretical findings.