Optimization and Systems Theory Seminar

May 21, at 13.00, room 3721, Lindstedtsvägen 25, KTH:
Martin Andersen, Division of Automatic Control, Department of Eletrical Engineering, Linköping University.

Logarithmic barriers for sparse matrix cones

In this talk, we present algorithms for evaluating gradients and Hessians of logarithmic barrier functions for two types of convex cones: the cone of positive semidefinite matrices with a given sparsity pattern, and its dual cone, the cone of sparse matrices with the same pattern that have a positive semidefinite completion. The algorithms are based on the multifrontal method for sparse Cholesky factorization. We also discuss some applications, including interior-point methods for nonsymmetric conic formulations of sparse semidefinite programs, sparse covariance selection, and sparse quasi-Newton updates.

Calendar of seminars Last update: May 11, 2012.