NLPQLP: A Fortran Implementation of a Sequential Quadratic Programming
Algorithm with Distributed and Non-Monotone Line Search - User's Guide, Version
K. Schittkowski, Report, Department of Computer Science, University of Bayreuth (2014)
The Fortran subroutine NLPQLP solves
smooth nonlinear programming problems by a sequential quadratic programming (SQP)
algorithm. This version is specifically tuned to run under distributed systems
controlled by an input parameter. In case of computational errors as for example
caused by inaccurate function or gradient evaluations, a non-monotone line
search is activated.
Numerical results are included which show that in case of noisy function values,
a significant improvement of the performance is achieved compared to the version
with monotone line search. Further stabilization is obtained by performing
internal restarts in case of errors when computing the search direction due to
inaccurate derivatives. The new version of NLPQLP successfully solves more than
90 % of our 306 test examples subject to a stopping tolerance of 10-7,
although at most two digits in function values are correct and although
numerical differentiation leads to additional truncation errors.
In addition, automated initial and periodic scaling with restarts is
implemented. The usage of the code is documented and illustrated by an example.
To download the report, click here: NLPQLP.pdf