Descent methods for optimization on homogeneous manifolds

In this article we present a framework for line search methods for optimization on smooth homogeneous manifolds, with particular emphasis to the Lie group of real orthogonal matrices. We propose strategies of univariate descent (UVD), methods. The main advantage of this approach is that the optimization problem is broken down into one-dimensional optimization problems, so that each optimization step involves little computation effort. In order to assess its numerical performance, we apply the devised method to eigen-problems as well as to independent component analysis in signal processing.