Evaluation of the implementations of the Mehrotra type predictor-corrector method on MARKAL type energy models

In this paper, we focus our attention on an infeasible interior point method and the effects of implementation techniques on computational speed are examined in detail. Infeasible methods can start from any interior point and are considered as one of the more effective methods than affine scaling methods which give a good computational performance in practice. Here one of the infeasible methods which is referred to as the Mehrotra type predictor-corrector method (1992) is considered and the effects of its implementations on the computational efficiency are evaluated by numerical studies. For numerical studies, MARKAL type energy model is used, which is a well known large-scale linear programming problem in energy technology analysis area. In this paper, the algorithm is divided into three parts: i) ordering which is used in order to keep the sparseness of matrices, ii) LU factorization which is used for obtaining a search direction, and iii) basis recovery. The most effective implementation at each part is suggested. By using the combination of the most effective implementations, a problem with 3,246 constraints has been solved about 70 times faster than by using a product form simplex method