[Pw_forum] question on electron_dynamics and orthogonalization
Nichols A. Romero
naromero at gmail.com
Wed Mar 29 23:57:37 CEST 2006
Hi Nicola,
Thanks for pointing out the differences in the units for the forces
between the two codes.
It was primarily in the stress which I noticed the wild disagreement.
I *think* that I saw this problem in Q-Epresso 3.0 but I did not make
any comments at the time because I wasn't sure if I was using CP
correctly.
The model system I gave you was nitromethane obtained via PWSCF using
vc-relax with a target stress of 5 GPa at T=0. It's probably near
equilibrium at that pressure.
I will play with it a bit more to see if I can isolate the problem. I
do recall that the stress changes on the order of 0.5 GPa per
component when I changed either
emass_cutoff
or
nr1b,nr2b,nr3b
However, given that the total energy convergence was not set
particularly high in my calculation to begin with, this behavior is
not a surprise.
Bests,
On 3/29/06, Nicola Marzari <marzari at mit.edu> wrote:
>
>
> Hi Nichols,
>
> what you are doing is extremely useful.
>
> The two calculations seem actually in good agreement, but of
> course beauty is in the eye of the beholder.
>
> The total energies are within 1 mRy of each other, and the forces
> seem to have a factor of 2 difference (that is correct - in PWSCF
> the forces are in Ry/bohr, while in CP they are in Ha/bohr; PWSCF
> labels the forces as Ry/au, that is the same as saying Ry/bohr,
> and CP says that they are in atomic units, and atomic units for the
> forces are Ha/bohr).
>
> The eigenvalues are different, but as long as it's a rigid shift between
> the two, we are fine (the two codes have clearly a different measure of
> what is zero in the energy scale - maybe someone could comment on this).
>
> The stresses are wildly different, but this is a bug we also noticed a
> week ago, and Paolo Giannozzi and Carlo Cavazzoni are tracking it down.
> Bug in CP.
>
> Let us know - if you want an ultimate comparison, try to run both in
> serial, make sure that the fft grids are the same (they are, in your
> case, 40 45 64), tighten the tolerance on the energy convergence as much
> as possible, and see what you get - it helps having an unrelaxed, out
> of equilibrium structure, so you get forces that are fairly large.
>
> But the agreement between the codes seems actually very good, at this
> stage - a part from the stress.
>
> Let us know if you find something else that I might have missed.
>
> Thanks,
>
> nicola
>
>
> Nichols A. Romero wrote:
>
> > Nicola (or anyone else on this list),
> >
> > I am going to ask a question that get's asked on this list very
> > frequently (possibly second only to "Why doesn't the code compile?").
> >
> > I am struggling with trying to get PWSCF and the CP code to agree. I
> > am sure the PWSCF is correct, but I don't know about the CP
> > calculation. All I want to immediately do with the CP is to calculate
> > the ground state and have it agree with PWSCF. Though the next step is
> > to do BO MD with the CP code (which I remember correctly, is still in
> > the testing phase).
> >
> > I'm attaching a model problem I've been playing with for the last two
> > days. I include the input and outputfiles. The US_PP are of the PBE
> > RRKJ type that are found on the PWSCF website. It should run fairly
> > quickly on a desktop computer.
> >
> > I would appreciate any help at this point. I'm sure that I am not
> > understand the definition of some of the CP parameters. FYI, I am
> > using the CVS version of Q-ESPRESSO as stated in my previous e-mail.
> >
> >
> > Bests,
> > On 3/28/06, Nicola Marzari <marzari at mit.edu> wrote:
> >
> >>
> >>Nichols,
> >>
> >>I'll let Paolo or Kostya or Ismaila confirm, but my guess is that once
> >>electron-dynamics is "cg" you want a gram-schmidt orthogonalization.
> >>
> >>We might implement an iterative ortho sometimes in the future, but I
> >>believe at this stage gram-schmidt is the only choice.
> >>
> >>Note that the latest CVS (2-3 days ago) has a number of calbec calls
> >>removed, and should be faster.
> >>
> >>We are still optimizing the code, but it should be fairly robust at this
> >>stage - let us know of any feedback you might have.
> >>
> >> nicola
> >>
> >>
> >>Nichols A. Romero wrote:
> >>
> >>
> >>>Hey,
> >>>
> >>>This is a question about CP code (i.e. not PWSCF). I am using the CVS version.
> >>>
> >>>Are the keywords
> >>>
> >>>electron_dynamics & orthogonolization
> >>>
> >>>completely independent of each other when
> >>>electron_dynamics = 'cg'
> >>>
> >>>Forgive, my ignorance. I am mostly familiar with the PWSCF component
> >>>of Q-Espresso.
> >>>
> >>>When doing an initial 'cg' calculation to get the groundstate on the
> >>>BO surface, doesn't the 'cg' routines impose the orthonomality
> >>>constraint automattically? Why does one need a seperate
> >>>*orthogonalization* keyword? (which has two different options?)
> >>>
> >>>Thanks,
> >>>--
> >>>Nichols A. Romero, Ph.D.
> >>>1613 Denise Dr. Apt. D
> >>>Forest Hill, MD 21050
> >>>443-567-8328 (C)
> >>>410-306-0709 (O)
> >>>_______________________________________________
> >>>Pw_forum mailing list
> >>>Pw_forum at pwscf.org
> >>>http://www.democritos.it/mailman/listinfo/pw_forum
> >>
> >>--
> >>---------------------------------------------------------------------
> >>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
> >>_______________________________________________
> >>Pw_forum mailing list
> >>Pw_forum at pwscf.org
> >>http://www.democritos.it/mailman/listinfo/pw_forum
> >>
> >
> >
> >
> > --
> > Nichols A. Romero, Ph.D.
> > 1613 Denise Dr. Apt. D
> > Forest Hill, MD 21050
> > 443-567-8328 (C)
> > 410-306-0709 (O)
>
> --
> ---------------------------------------------------------------------
> 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
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
--
Nichols A. Romero, Ph.D.
1613 Denise Dr. Apt. D
Forest Hill, MD 21050
443-567-8328 (C)
410-306-0709 (O)
More information about the users
mailing list