[gpaw-users] GPAW dev r7883: Spin-pol vdW-DF2 diverges

Jess Wellendorff jesswellendorff at gmail.com
Mon Apr 25 21:20:50 CEST 2011


Hi Eskil,

I think you will find that your issue is largely cured by using a 
somewhat less aggressive density mixing for the hydrogen atom, than the 
GPAW default, e.g.
********
from gpaw import GPAW, Mixer
calc = GPAW(....., mixer=Mixer(0.1, 5), ........)
********
with which I found the H atom in the example to converge nicely in less 
than 20 iterations with a=8 and xc='vdW-DF-2'.
The default mixer handles the spinpaired H2 molecule well, but is 
sometimes a bit too aggressive for isolated atoms in vacuum, especially 
if these have several unpaired "magmoms" in their ground state.

Cheers,
Jess Wellendorff

Den 25-04-2011 16:35, Eskil Varenius skrev:
> Thanks for the idea!
> Yes, I've tried changing eigensolver to 'cg' and played a little with 
> the mixer settings, but with no improvement. Trouble is, in other 
> calculations I've sometimes encountered convergence errors which can 
> be solved by following instructions on the convergence page. But those 
> times the energy is always close to some value, it just won't converge 
> the last decimals. Or the log10 error refuses to go below -3. In this 
> case the whole calculation rapidly blows out of proportion. Energy 
> jumps of several hundred eV for a single hydrogen atom seems really 
> weird. And the log errors are often positive between iterations.
>
> Have anyone successfully did a spin-polarised vdW-DF2 calculation? If 
> so, on what system? I would gladly try it here. Also, if one has a 
> working setup for spin-polarised vdW-DF2 it would be interesting to 
> see if the hydrogen atomization example works in that setup.
>
> Best regards,
> Eskil
>
>
> 2011/4/25 Duy Le <ttduyle at gmail.com <mailto:ttduyle at gmail.com>>
>
>     The convergence issues are discussed in:
>     https://wiki.fysik.dtu.dk/gpaw/documentation/convergence/convergence.html#convergence
>     Have you ever tried them? Especially to change the eigensolver and
>     the mixing
>     --------------------------------------------------
>     Duy Le
>     PhD Student
>     Department of Physics
>     University of Central Florida.
>
>     "Men don't need hand to do things"
>
>
>
>     On Mon, Apr 25, 2011 at 9:32 AM, Eskil Varenius
>     <eskil.varenius at gmail.com <mailto:eskil.varenius at gmail.com>> wrote:
>     > Dear GPAW users and developers,
>     > Background
>     > I'm a master student at Chalmers, Gothenburg, Sweden. In my
>     thesis work I
>     > want to use GPAW to calculate total energies with the vdW-DF
>     functional. I
>     > also want to use the recent second version, the vdW-DF2. Since
>     the second
>     > version is not easily accessible in the current release version
>     of GPAW
>     > (r6974) I asked our cluster admins if they could install the lastest
>     > dev-version of GPAW (r7883 at time of asking). In this version
>     the vdW-DF2
>     > is easily accessible. However I encountered a major problem.
>     Since this is a
>     > dev checkout I want to ask if this problem has been fixed in later
>     > dev-versions and if I should ask the cluster admin to upgrade.
>     > Problem
>     > Spin-polarized calculations with vdW-DF2 doesn't work. It more
>     or less
>     > diverges after a few iterations.
>     > However, spin-paired calculations with vdW-DF2 seems to work.
>     > Spin-paired / polarized calculations with vdW-DF (version 1)
>     seems to work
>     > fine.
>     > Steps to reproduce
>     > 1. Install GPAW dev-version 7883.
>     > 2. Run example script for hydrogen atomization energies
>     >
>     (https://wiki.fysik.dtu.dk/gpaw/tutorials/atomization/atomization.html)
>     with
>     > xc='vdW-DF2', and a=8.
>     > 3. Check the output. For the first hydrogen calculation I got:
>     > Number of Valence Electrons:            1
>     >                      log10-error:    Total        Iterations:
>     >            Time      WFS    Density  Energy       Fermi  Poisson
>      MagMom
>     > iter:   1  20:54:30  -0.3            -1.31548     0      14    
>       +1.0000
>     > iter:   2  20:54:30  -1.8            -1.35790     0            
>       +1.0000
>     > iter:   3  20:54:30  -2.3            -1.36544     0            
>       +1.0000
>     > iter:   4  20:54:49  -2.7   -1.2     -29.93035    0      6      
>      -1.0000
>     > iter:   5  20:55:09  -0.4   -0.4     -10.44161    0      6      
>      +1.0000
>     > iter:   6  20:55:28  +0.5   -0.2     -26.34328    0      4      
>      -1.0000
>     > iter:   7  20:55:46  +1.0   -0.1     -40.33089    0      6      
>      +1.0000
>     > iter:   8  20:56:02  +0.8   -0.0     -57.44584    0      6      
>      -1.0000
>     > iter:   9  20:56:18  +1.4   -0.0     -56.50579    0      10    
>       +1.0000
>     > iter:  10  20:56:37  +1.1   +0.0     -170.18543   0      14    
>       -1.0000
>     > iter:  11  20:56:54  +1.9   -0.2     -142.87739   0      8      
>      +1.0000
>     > iter:  12  20:57:12  +0.3   -0.5     -116.98545   0      10    
>       -1.0000
>     > I would be very grateful if someone could check this at least
>     briefly since
>     > my thesis work would benefit a lot from beeing able to do
>     spin-polarized
>     > vdW-DF2 calculations.
>     > Best regards,
>     > Eskil Varenius, Chalmers Master Student
>     > _______________________________________________
>     > gpaw-users mailing list
>     > gpaw-users at listserv.fysik.dtu.dk
>     <mailto:gpaw-users at listserv.fysik.dtu.dk>
>     > https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>     >
>
>
>
> _______________________________________________
> gpaw-users mailing list
> gpaw-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20110425/e9f8cdf8/attachment-0001.html 


More information about the gpaw-users mailing list