Penalty-Barrier-Methods for Nonlinear Optimization:
Implementation and Numerical Results
J. Franz, M. Liepelt, K. Schittkowski: Report, Dept. of Mathematics, University of Bayreuth (1995)
Abstract:
In this paper we propose some modifications on the penalty-mutiplier
method of Breitfeld and Shanno with the goal to improve numerical efficiency
and reliability.
The mathematical algorithm generates a sequence of bound-constrained
subproblems, where the objective function is a logarithmic barrier function.
The method is described in detail, moreover possible extensions to solve
also large problems. The corresponding computer code is compared with
the original method of Breitfeld and Shanno and a standard sequential
quadratic programming code on a large number of small test problems.
Numerical results indicate a significant improvement w.r.t. the
competitive penalty-multiplier method and even better performance than
that of an SQP method for some very ill-conditioned test problems.