A Riemannian symmetric rank-one trust-region method (Q2340330): Difference between revisions

The authors present a Riemannian symmetric rank-one trust region method in solving the problem of minimizing a smooth real-valued function defined on a Riemannian manifold. The method relies on the notions of retraction and vector transport, where the second-order term of the model is generated using a Riemannian generalization of the classical symmetric rank-one update. The global and local convergence of the method are established. A limited-memory version of the method is also proposed. The performance of the proposed method is illustrated by a numerical experiment on a Rayleigh quotient minimization problem on the sphere and on a joint diagonalization problem on the Stiefel manifold.