Solver lbfgsb minimizes functions subject only to simple-bound
constraints. It is based on L-BFGS-B, by Ciyou Zhu, Richard Byrd,
Peihuang Lu, and Jorge Nocedal; see
http://www.ece.nwu.edu/~nocedal/software.html
for pointers to more information on the limited-memory BFGS algorithm
implemented in L-BFGS-B. Relevant papers include
%A Ciyou Zhu
%A Richard H. Byrd
%A Peihuang Lu
%A Jorge Nocedal
%T Algorithm 778: L-BFGS-B: Fortran Subroutines for
Large-Scale Bound-Constrained
Optimization
%J ACM Trans. Math. Software
%V 23
%N 4
%D 1997
%P 550-560
%A Richard H. Byrd
%A Peihuang Lu
%A Jorge Nocedal
%A Ciyou Zhu
%T A Limited Memory Algorithm for Bound Constrained Optimization
%J SIAM J. Sci. Comput.
%V 16
%N 5
%D 1995
%P 1190-1208
Source file routines.f comes from
ftp://ftp.mcs.anl.gov/pub/neos/Lbfgs/Lbfgsb.2.1.tar.gz
Execute
lbfgsb -?
for a summary of usage and
lbfgsb -=
to see keywords that may appear in assignments in $lbfgsb_options
or the lbfgsb command line.
Possible solve_result_num values for lbfgsb:
000 Solved: maxnorm(projected gradient) <= pgtol.
100 Solved? relative function change <= macheps * factr.
101 No objective: nprob was out of range of stub.nl had no objectives.
400 More than maxfun function and gradient evaluations.
401 More than maxit iterations.
500 Probable bug: inconsistency in linesearch routine.
501 Error -- probably a bug.