Preliminary Computational Experience with Modified Log-Barrier Functions for Large-Scale Nonlinear Programming

The paper considers Polyak’s modified logarithmic barrier function for nonlinear programming. Comparisons are made to the classic logarithmic barrier function, and the advantages of the modified log-barrier method, including starting from nonfeasible starting points, inclusion of equality constraints, and better conditioning are discussed. Extensive computational results are included which demonstrate that the method is clearly superior to the classic method and holds definite promise as a viable method for large-scale nonlinear programming.