A time-varying Newton algorithm for adaptive subspace tracking (Q1010056)

Tracking the zeros of a time-varying gradient vector field on Riemannian manifolds is considered. A continuation type approach is proposed for adaptive eigenvalue tracking and asymptotic eigenspace tracking on unbounded intervals. The proposed framework introduces a differential equation called time-varying Newton flow, whose solutions asymptotically converge to a time-varying family of critical points of the corresponding cost function. The differential equation is discretized to obtain a recursive update scheme for the time-varying critical point Simple expressions are obtained for subspace tracking schemes using local parameterization of the Grassmann manifold. The benefits of the algorithms are reduced complexity and guaranteed accuracy illustrated by numerical examples.