[CPMD-list] BO-MD starting problem after successful WFO

[Vladimir Stegailov]

Axel, thanks a lot for a good advice!

I missed that the w.f. compression is lossy. Now everything works fine 
without this option.

best regards,
Vladimir


----- Original Message ----- 
From: "Axel Kohlmeyer" <akohlmey at cmm.chem.upenn.edu>
To: "Vladimir Stegailov" <stegailov at ihed.ras.ru>
Cc: <cpmd-list at cpmd.org>
Sent: Wednesday, June 06, 2007 6:26 PM
Subject: Re: [CPMD-list] BO-MD starting problem after successful WFO


> On Wed, 6 Jun 2007, Vladimir Stegailov wrote:
>
> VS> Dear colleagues,
>
> dear vladimir,
>
> VS> I am trying to do BO-MD of 32 atom system (non-perfect fcc aluminum
> VS> crystal in 3D pbc). First I make a wave function optimization. Then
> VS> I use the resulting restart file to begin MD. The WFO goes fine.
> VS> However at the very beginning of the BO-MD stage the diagonalization
> VS> fails. Program stops after a lot of "FRIESNER_C| EIGENVECTOR ...
> VS> IS VERY BAD!" messages.
>
> it is not fully clear to me from your description, whether
> the problem happens right during initialization or after
> the first MD step.
>
> however, there are a number of things with your input
> that may have some impact.
>
> using compressed write will cut off some accuracy from your
> wavefunction (it is a lossy compression!), so you are not
> starting from the exactly same wavefunction.
>
> using point group symmetry during MD is usually not a good idea.
> the MD may break the detected symmetry and then all of your
> arrays are too small. i remember we had a problem because of
> that a long time ago in geometry optimization, not sure whether
> this is the same here.
>
> VS> It seems strange to me. I expected that starting from a previously
> VS> optimized w.f. there should be no diagonalization problems at least
>
> the compression modifies the wavefunction...
>
> VS> during several MD steps.  Please could you give any advice.
>
> this can depend a _lot_ on how your system moves, how large
> the velocities are and what happens during the steps. also
> the fiddling with the lanczos parameters is almost always required
> to get a stable MD with diagonalization.
>
> it might also help to start from a random wavefunction instead
> of an atomic guess. sometimes the guess leads the optimization
> to a wrong state and some additional randomization then leads
> to conversion problems.
>
> finally, please try looking at the occupations and eigenvalues.
> with 90 states you should be fine, but the error you see sometimes
> also indicates, that for one of the k-points you don't have enough
> empty states.
>
> cheers,
>  axel.
>
> VS>
> VS> My WFO and BO-MD scripts are below.
> VS>
> VS> Thank you.
> VS>
> VS> Vladimir
> VS>
>
> -- 
> =======================================================================
> Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
>   Center for Molecular Modeling   --   University of Pennsylvania
> Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
> tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
> =======================================================================
> If you make something idiot-proof, the universe creates a better idiot.
>
>
>