[Pw_forum] BFGS & Wolfe condition

Nicola Marzari marzari at MIT.EDU
Thu Sep 20 18:16:07 CEST 2007


Marcello Rosini wrote:
> I' am performing a bfgs relaxation
> after 3 scf cycles and one bfg relaxation I get
> "wolfe condition not staisfied"
> and the program stops.
> Does anyone have suggestions?
> NB: the forces acting on the atom a reasonably small and all the
> positions are correct.
> Here you can find the input and the output
> http://cdm.unimo.it/home/fisica/rosini.marcello/input.txt
> http://cdm.unimo.it/home/fisica/rosini.marcello/output.txt
> 
> thanks alot
> Marcello


Hi Marcello,

no idea, but Google knows everything. It looks like the message error
you get is from an old bfgs
http://www.democritos.it/pipermail/pw_forum/2004-January/000806.html .

The current bfgs does not print that error message, but checks
for the Wolfe condition:
http://reference.wolfram.com/mathematica/tutorial/UnconstrainedOptimizationLineSearchMethods.html

Now, what that probably means is that there has been an accumulation of
approximations (e.g. in your hessian) that your predicted steps along
the line direction are not going to be very accurate (again, wild
guess).

To solve the problem just (using a newer version of espresso) restart
everything, from scratch, with the current relaxed coordinates, so bfgs
starts from scratch with a new hessian.

There might be better strategies - mine are just guesses.

Moral - if you have a question, google has the answer. It's like
the library of Babel, but searchable.


				nicola

---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu


More information about the Pw_forum mailing list