[Wien] Conjugate Gradient

[Laurence Marks]

Several responses:
1) Conjugant gradient uses derivates; calculating the full Hessian is
the Newton method which would be very expensive. Conjugant gradient is
an obsolete method.
2) I cannot say anything about NEWT, except that I do not expect it to
work. While NEW1 may work, I cannot say anything about it.
3) The main code used in PORT is one of the most respected and stable
quasi-Newton minimization codes freely available -- I did not write
it. Provided that your energies are consistent with your forces it
will not fail. If, however, your energies and forces are not
consistent it will fail, i.e. you have made too many approximations in
how you setup the problem. No code can handle user error.
4) When you mean "fail", I suspect that you mean that you run into
problems where you get ghost bands and/or the mixing diverges. This is
a COMPLETELY different issue, has zero relationship to the PORT code.
This issue is being addressed, but the code to do a better (sometimes
much, much better) is still in the beta-testing stage.

On 5/17/07, John Appleton <banger_deep at yahoo.com> wrote:
> Dear Prof. Blaha and Prof. L. Marks,
> Why can't an accurate code like WIEN perform
> conjugate gradient relaxtion by computing
> the full hessian. Is the computation of the
> second derivative of the energy very
> expensive? Relaxation with WIEN is
> far difficult than I thought. PORT keeps
> failing. It is not easy to find a
> set of deltas and etas for NEWT.
> Is there a thumb rule for selecting
> delta and eta (for NEWT/NEW1)
> for systems with entirely
> different masses? e.g. H and Pt.
>
> Thank you.
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
>


-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
EMM2007 http://ns.crys.ras.ru/EMMM07/
Commission on Electron Diffraction of IUCR
www.numis.northwestern.edu/IUCR_CED