[gpaw-users] Convergence issue with mBEEF
Jess Wellendorff
jesswellendorff at gmail.com
Wed Nov 26 16:53:57 CET 2014
On 11/26/2014 03:53 PM, Jens Jørgen Mortensen wrote:
> On 11/25/2014 10:44 PM, Jess Wellendorff wrote:
>> Marcin: The mBEEf functional was fitted on PBEsol electron densities
>> obtained using rather tight SCF criteria. I think something else must
>> be the root cause of this issue. Perhaps GPAW does not supply LibXC
>> with sufficiently accurate kinetic energy densities?
>>
>> Monish: Honestly, I believe the only solution to your problem is to
>> lower the SCF convergence criteria, perhaps even completely eliminate
>> them for the eigenstates and rely only on density and energy
>> criteria. I have successfully used this approach. It is not pretty,
>> but it works :)
>>
>> If you take a look at the mBEEF test script (gpaw/test/mbeef.py), the
>> SCF criteria for very quick molecular calculatios are set to
>> {'eigenstates':1.e+2, 'density':1.e-2, 'energy':1.e-4}. These are
>> bullet-proof settings for a simple test. In general, the density and
>> energy criteria can probably be tightened more. Also, I would expect
>> the eigenstate issue to be much less pronounced in bulk calculations.
>
> What is it that makes mBEEF so special? I'm struggling with a test
> calculation for a nitrogen atom (not mBEEF, but mBEEF-vdW) and I can
> also not converge the eigenstates. Strange thing is that TPSS works
> fine.
JJ: I don't get it, you're saying that this is not an issue with TPSS? I
think it certainly am, it's an issue for all the metaGGAs. Here is my
GPAW output for TPSS and the N2 molecule:
******************************
Number of Valence Electrons: 10
log10-error: Total Iterations:
Time WFS Density Energy Fermi Poisson
iter: 1 08:50:39 -0.44 -20.362686 0 17
iter: 2 08:50:39 -1.11 -20.364991 0
iter: 3 08:50:39 -1.11 -20.363382 0
iter: 4 08:50:41 -1.26 -2.29 -20.362235 0 5
iter: 5 08:50:44 -1.12 -2.41 -20.361605 0 6
iter: 6 08:50:46 -0.89 -3.32 -20.359967 0 3
iter: 7 08:50:49 -0.94 -3.49 -20.360150 0 3
iter: 8 08:50:51 -0.92 -3.73 -20.360137 0 3
iter: 9 08:50:53 -0.92 -4.37 -20.360187 0 2
iter: 10 08:50:55 -0.92 -4.55 -20.360091 0 1
iter: 11 08:50:58 -0.92 -4.73 -20.360269 0 1
iter: 12 08:51:00 -0.92 -4.80 -20.360286 0 1
iter: 13 08:51:03 -0.92 -4.94 -20.360311 0 1
iter: 14 08:51:06 -0.92 -5.10 -20.360363 0 1
iter: 15 08:51:08 -0.92 -5.96 -20.360300 0 1
iter: 16 08:51:10 -0.92 -6.44 -20.360284 0 1
iter: 17 08:51:13 -0.92 -6.60 -20.360280 0 1
iter: 18 08:51:15 -0.92 -6.73 -20.360281 0 1
iter: 19 08:51:17 -0.92 -6.83 -20.360281 0 1
iter: 20 08:51:20 -0.92 -6.92 -20.360281 0 1
iter: 21 08:51:22 -0.92 -7.11 -20.360281 0 1
iter: 22 08:51:24 -0.92 -7.27 -20.360280 0 1
iter: 23 08:51:26 -0.92 -7.59 -20.360280 0 1
iter: 24 08:51:29 -0.92 -7.88 -20.360281 0 1
iter: 25 08:51:31 -0.92 -7.99 -20.360281 0 1
iter: 26 08:51:33 -0.92 -8.08 -20.360281 0 1
iter: 27 08:51:35 -0.92 -8.59 -20.360281 0 1
iter: 28 08:51:38 -0.92 -8.83 -20.360281 0 1
iter: 29 08:51:40 -0.92 -8.99 -20.360281 0 1
iter: 30 08:51:42 -0.92 -9.40 -20.360281 0 1
iter: 31 08:51:44 -0.92 -9.60 -20.360281 0 1
iter: 32 08:51:47 -0.92 -9.80 -20.360281 0 1
iter: 33 08:51:49 -0.92 -10.03 -20.360281 0 1
iter: 34 08:51:51 -0.92 -10.18 -20.360281 0 1
iter: 35 08:51:53 -0.92 -10.44 -20.360281 0 1
iter: 36 08:51:56 -0.92 -10.62 -20.360281 0 1
iter: 37 08:51:58 -0.92 -10.79 -20.360281 0 1
iter: 38 08:52:00 -0.92 -10.96 -20.360281 0 1
iter: 39 08:52:03 -0.92 -11.70 -20.360281 0 1
iter: 40 08:52:06 -0.92 -11.86 -20.360281 0 1
iter: 41 08:52:08 -0.92 -12.03 -20.360281 0 1
iter: 42 08:52:11 -0.92 -12.33 -20.360281 0 1
iter: 43 08:52:13 -0.92 -12.55 -20.360281 0 1
iter: 44 08:52:15 -0.92 -13.25 -20.360281 0 1
******************************
Best,
Jess
>
> It helped a bit to use eigensolver='rmm-diis'.
>
> Jens Jørgen
>
>>
>> Hope this helps.
>> Best,
>> Jess
>>
>> On 11/25/2014 06:53 PM, Marcin Dulak wrote:
>>> On 11/25/2014 04:37 PM, Wellendorff, Jess wrote:
>>>> Hi Monish, in my experience, the eigenstates can be hard to
>>>> converge with metagga. Is this your problem?
>>> i remember even with the older BEEF-vdW molecules are hard to converge.
>>> Maybe this is due to loose SCF thresholds used when fitting the
>>> functional (was it the case also for mBEEF?),
>>> so the resulting functional by design is not reaching a tight
>>> convergence.
>>>
>>> Best regards,
>>>
>>> Marcin
>>>>
>>>> Best,
>>>> Jess
>>>>
>>>>
>>>>> On Nov 25, 2014, at 14:48, mohnish pandey <mohnish.iitk at gmail.com>
>>>>> wrote:
>>>>>
>>>>> Dear all,
>>>>>
>>>>> I am trying to do some test calculations with mBEEF functional for
>>>>> some solids like ZnO. But I am facing difficulty in converging the
>>>>> calculations. I tried tweaking mixing, Fermi temperature etc, but
>>>>> it didn't help. Would really appreciate if somebody helps me out
>>>>> with the issue. Thanks a lot in advance.
>>>>>
>>>>> --
>>>>> Regards,
>>>>> Mohnish,
>>>>> -----------------------------------------------------------------
>>>>> Mohnish Pandey,
>>>>> PhD Student,
>>>>> Center for Atomic-scale Materials Design,
>>>>> Department of Physics,
>>>>> Technical University of Denmark
>>>>> -----------------------------------------------------------------
>>>>> _______________________________________________
>>>>> gpaw-users mailing list
>>>>> 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
>>>>
>>>>
>>>
>>>
>>
>> _______________________________________________
>> gpaw-users mailing list
>> gpaw-users at listserv.fysik.dtu.dk
>> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>
More information about the gpaw-users
mailing list