There is a new version with the stepsize, numeric gradient, and finite
difference code integrated into bfgsmin.cc. This simplifies the code,
eliminates redundant error checking, and leads to about 15% better
performance. There should be no numeric differences in results comparing
to the previous version. The example file bfgsmin_example.m has also
been updated. Please let me know if any problems are seen.
Michael
|
