Portal de Periódicos da CAPES

A projected Lagrangian algorithm and its implementation for sparse nonlinear constraints

1982; Springer Nature; Linguagem: Inglês

10.1007/bfb0120949

2364-8201

B. A. Murtagh, Michael A. Saunders,

Optimization and Variational Analysis

An algorithm is described for solving large-scale nonlinear programs whose objective and constraint functions are smooth and continuously differentiable. The algorithm is of the projected Lagrangian type, involving a sequence of sparse, linearly constrained subproblems whose objective functions include a modified Lagrangian term and a modified quadratic penalty function. The algorithm has been implemented in a general purpose FORTRAN program called MINOS/AUGMENTED. Some aspects of the implementation are described, and computational results are given for some nontrivial test problems. The system is intended for use on problems whose Jacobian matrix is sparse. (Such problems usually include a large set of purely linear constraints.) The bulk of the data for a problem may be assembled using a standard linear-programming matrix generator. Function and gradient values for nonlinear terms are supplied by two user-written subroutines. Future applications could include, some of the problems that are currently being solved in industry by the method of successive linear programming (SLP). We would expect the rate of convergence and the attainable accuracy to be better than that achieved by SLP, but comparisons are not yet available on problems of sinificant size. One of the largest nonlinear programs solved by MINOS/AUGENTED involved about 850 constraints and 4000 variables, with a nonlinear objective function and 32 nonlinear constraints. From a cold start, about 6000 iterations and 1 hour of computer time were required on a DEC VAX 11/780.