[gpaw-users] Full optimization in plane wave mode

Thu Nov 8 12:06:19 CET 2012

Hi,

I believe that I am using at least a quite new version of GPAW (0.9rev9473). The installation of the
latest version to the FGI (Finnish Grid Infrastructure) only due to this problem may not be possible
at least as quickly as I would need.

Moreover, during this morning I have learned something new about the stress in GPAW plane wave
calculations (this is the first time I am using PW mode), which just confuses me:

1) Optimization by scanning through different cubic cell sizes gives a unit cell dimension of 7.28Å
(cut-off = 400eV), but there is a positive stress in all three directions (0.016462). The sign of the stress
components actually changes with a cell dimension of 6.96Å (for which energy is much higher), which
corresponds to full optimization with UnitCellFilter. Any ideas why this kind of difference?

2) I am not yet sure if everything is just due to too small energy cut-off, but based on calculations for
the tetragonal structure discussed below, stress values seems to be highly dependent on used cut-off
value. Moreover, with a high cut-off value (800eV) energies behaved much more nicely (previously
reported large fluctuations were related steps where cell size changed drastically), and a better
correspondence to the results of other programs was achieved.

3) Moreover, based on some test calculations with VASP, I was wondering if I should actually define
PW(cell=...) to be able to use same plane waves during the whole optimization instead of regenerating
those for each cell size again and again?

4) There is correlation between energy cut-off and grid spacing - if energy cut-off is increased grid spacing
should decrease. Even with PW mode certain parts of the calculation are performed by taking benefit from
the automatically generated grid. I suppose I shouldn't change this grid spacing?

Best regards,

Janne

From: Marcin Dulak [mailto:Marcin.Dulak at fysik.dtu.dk]
Sent: Thursday, November 08, 2012 12:02 AM
To: Janne Hirvi
Cc: Vladislav Ivanistsev; Jón Steinar Garðarsson Mýrdal; gpaw-users at listserv.fysik.dtu.dk
Subject: Re: [gpaw-users] Full optimization in plane wave mode

On 11/07/12 15:03, Janne Hirvi wrote:
Hi,

I believe that full optimization with UnitCellFilter has been implemented only for plane wave mode (not for finite difference mode)?

However, I will try to increase the cut-off energy (which basically corresponds to decrease in grid spacing for finite difference mode)
and check what happens, even though these energy changes seems to be quite large.
i would first verify the calculations with the latest ASE/GPAW.
I heard rumors that at some point (just before Jens Jørgens leave)
there were problems with stresses. You could maybe also limit the optimization to stress only
with StrainFilter. If the problem persists attach the full script.

Best regards,

Marcin

Regards,

Janne

From: Vladislav Ivanistsev [mailto:olunet at gmail.com]
Sent: Wednesday, November 07, 2012 3:45 PM
To: Jón Steinar Garðarsson Mýrdal
Cc: Janne Hirvi; gpaw-users at listserv.fysik.dtu.dk<mailto:gpaw-users at listserv.fysik.dtu.dk>
Subject: Re: [gpaw-users] Full optimization in plane wave mode

I've also experienced similar problem with grid spacing. What is the lowest h value one can safely use. Is h=12 suitable?
Ph.D. Vladislav Ivanistsev

On Wed, Nov 7, 2012 at 1:33 PM, Jón Steinar Garðarsson Mýrdal <jsmy at dtu.dk<mailto:jsmy at dtu.dk>> wrote:
Hi Janne

When you change the shape or size of your super cell then the grid sampling will change and this can lead to a change in energy.
Therefore it is important to run calculations where you perform cell optimization with extra small grid spacing.
I would try lowering the value of h you are using.

Mvh.
Jón Steinar Garðarsson Mýrdal
Ph.d Student
DTU Energy Conversion
-------------------------------------------
Technical University of Denmark
Department of Energy Conversion and Storage
Frederiksborgvej 399
Building 775
4000 Roskilde
Direct +45 46775818
jsmy at dtu.dk<mailto:jsmy at dtu.dk>
http://www.dtu.dk

________________________________________
From: gpaw-users-bounces at listserv.fysik.dtu.dk<mailto:gpaw-users-bounces at listserv.fysik.dtu.dk> [gpaw-users-bounces at listserv.fysik.dtu.dk<mailto:gpaw-users-bounces at listserv.fysik.dtu.dk>] On Behalf Of Janne Hirvi [janne.hirvi at uef.fi<mailto:janne.hirvi at uef.fi>]
Sent: Wednesday, November 07, 2012 2:01 PM
To: gpaw-users at listserv.fysik.dtu.dk<mailto:gpaw-users at listserv.fysik.dtu.dk>
Subject: [gpaw-users] Full optimization in plane wave mode

Dear GPAW-users,

I tried to perform "full" optimization (in orthogonal cell) for silica structure using plane wave mode.

The script was something like this

-----
mixerown=Mixer(beta=0.1, nmaxold=5, weight=50.0)
calc=GPAW(mode=PW(400), xc='PBE', kpts=(4,4,4), mixer=mixerown, txt=name+'.txt')
Bulk.set_calculator(calc)

ucf=UnitCellFilter(Bulk,mask=[1,1,1,0,0,0])
relax=BFGS(ucf)
traj=PickleTrajectory(name+'.traj','w',Bulk)
relax.attach(traj)
relax.run(fmax=0.03)
----

The corresponding calculation converged (stress and forces are small), but the energies for each step

Zero Kelvin:   -188.134359 (Step 1)
Zero Kelvin:   -187.264322
Zero Kelvin:   -187.394858
Zero Kelvin:   -187.853520
Zero Kelvin:   -187.879200
Zero Kelvin:   -188.081472
Zero Kelvin:   -188.048312
Zero Kelvin:   -188.051650
Zero Kelvin:   -188.081487
Zero Kelvin:   -188.096408 (Step 10)
Zero Kelvin:   -188.091418
Zero Kelvin:   -188.113318
Zero Kelvin:   -188.122375
Zero Kelvin:   -188.072067
Zero Kelvin:   -188.123096
Zero Kelvin:   -188.125491
Zero Kelvin:   -188.097103
Zero Kelvin:   -188.125748
Zero Kelvin:   -188.125478
Zero Kelvin:   -188.130157 (Step 20)
Zero Kelvin:   -188.146106
Zero Kelvin:   -188.122699
Zero Kelvin:   -188.065573
Zero Kelvin:   -188.071948
Zero Kelvin:   -188.033935
Zero Kelvin:   -187.960727
Zero Kelvin:   -187.975459
Zero Kelvin:   -187.954870
Zero Kelvin:   -187.941968
Zero Kelvin:   -187.914349 (Step 30)
Zero Kelvin:   -187.912083
Zero Kelvin:   -187.911463
Zero Kelvin:   -187.909650 (Step 33)

behave quite strangely. The energy minimum has been reached around 20 steps (only slightly smaller than that
for the initial guess) and the final energy is quite notably higher.

Do you know if this is related to large changes in cell dimensions when using plane wave mode or what might
be the reason for this?

Best regards,

Janne Hirvi

-----
Janne Hirvi
PhD (Physical Chemistry), Researcher
University of Eastern Finland, Department of Chemistry
P.O.Box 111, 80101 Joensuu, FINLAND
Tel: +358 50 4422594 & +358 50 3474223

_______________________________________________
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<mailto: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/20121108/5338d744/attachment.html 