in the gsl-multimin, the bfgs method is referred to as "vector-bfgs
method".
Is this algorithm the classical BFGS method that needs to store the
matrix
in internal memory or L-BFGS method, also known as Limited Memory
BFGS,
since it does not need to store it ?
Definitely L-BFGS. For the record, I wouldn't mind if GSL had a
classic BFGS method since my problem doesn't have very many
parameters and it's only in cases of lots and lots of parameters that
L-BFGS really helps you. Also, I'm pretty sure GSL only saves the
previous derivatives when approximating the Hessian instead and I
think the general L-BFGS algorithm can use more than one previous round.
Doug
-------------------
Doug McKee
[EMAIL PROTECTED]
http://dmckee.bol.ucla.edu
310-266-2438
_______________________________________________
Help-gsl mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/help-gsl
