TITLE : A non-polyhedral primal active set approach for semidefinite
         programming

SPEAKER : Kartik Krishnan
          Advanced Optimization Laboratory
          Dept. of Computing & Software
          McMaster University
          Hamilton

ABSTRACT :

We present a non-polyhedral primal active set approach for semidefinite
programming (SDP) which mimics the primal simplex method for
linear programming, and exploits the low rank of the extreme
point solutions of the primal feasible region. The goal is to find a proper 
superset of the range space of an optimal extreme point solution. The
algorithm generates a sequence of primal iterates with non-increasing 
objecive values. Under a nondegeneracy assumption, the objective values are 
strictly decreasing. We will discuss the convergence of the algorithm,
and some preliminary computational results. Finally, we 
relate this method to other cutting plane approaches that have been
introduced for the SDP.

This is based on joint works with Gabor Pataki at UNC, Yin
Zhang at Rice University, and John Mitchell at RPI.

------------------------------------------------------------------------------