Abstract:
For linear programs, we consider schemes for the formation of a
generalized central path, which arise under the simultaneous use of interior
and exterior penalty terms in the traditional Lagrange function and the minimax
problems generated by it. The advantage of the new schemes is that they do not
require a priori knowledge of feasible interior points in the primal or dual
problem. Moreover, when applied to problems with inconsistent constraints, the
schemes automatically lead to some of their generalized solutions, which have an
important applied content. Descriptions of the algorithms, their justification,
and results of numerical experiments are presented.
Keywords:linear programming, duality, penalty function methods, regularization methods, improper problems, central path.