An extension of a variant of a predictor–corrector primal–dual method from linear programming to semidefinite programming

We extend a variant of a predictor–corrector primal–dual method for Linear Programming to Semidefinite Programming. Two versions are proposed. One of the versions uses the HKM direction and the other the NT direction. We present the algorithms associated with these versions and the computational experience using the SDPLIB 1.2 collection of Semidefinite Programming test problems. We show that, in general, the algorithm using the HKM direction is the best and is also better than the one relative to the classical method.