A reduced-gradient variant of Karmarkar's algorithm and null-space projections

This paper discusses the relationship between Karmarkar's new method for linear programming and the traditional simplex method. It is shown how null-space Karmarkar projections can be done using a basis matrix to compute the projections in the null space. Preliminary computational evidence shows that problems exist in the choice of a basis matrix, but that given a basis, very inexact and computationally efficient projections are computationally sound.