An interior point method for semidefinite programming
C. Helmberg and F. Rendl and R. J. Vanderbei and H. Wolkowicz
We propose a new interior point based method to minimize a linear
function of a matrix variable
subject to linear equality and inequality constraints over the
set of positive semidefinite matrices. We present a theoretical convergence
analysis, and show that the approach is very efficient for graph bisection
problems, such as max-cut. The approach can also be applied to max-min
Research Report, University of Waterloo.
To appear in SIOPT.