[gpaw-users] Convergence of eigenstate and density
Jun Yan
juya at fysik.dtu.dk
Mon May 2 09:20:29 CEST 2011
Hi, Jens Jørgen,
How many extra bands will be enough ?
Jun
Jens Jørgen Mortensen wrote:
> On Sat, 2011-04-30 at 13:13 +0200, Jun Yan wrote:
>
>> Well, I dont quite understand your question.
>> I use
>> convergence={
>> 'density': 1.0e-6,
>> 'eigenstates': 1.0e-4,
>> 'bands': nband}
>>
>> and nband = 260 while 80 occupied bands are included.
>>
>
> RMM-DIIS has a hard time converging the highest bands, so make sure you
> have more bands than you want converged.
>
> Jens Jørgen
>
>
>> Jun
>>
>> Jess Wellendorff wrote:
>>
>>> ok, but are you sure you are actually converging the unoccupied states?
>>> (perhaps a stupid question)
>>>
>>> Jess
>>>
>>> Den 30-04-2011 12:26, Jun Yan skrev:
>>>
>>>
>>>> Hi, Jess,
>>>>
>>>> I did that, its just I am not sure whether I can lower the
>>>> convergence criteria for eigenstate to as low as 1e-5. It would be much
>>>> better if the occupied states are converged with high criteria and
>>>> unoccupied ones can choose low criteria.
>>>>
>>>> All the best,
>>>> Jun
>>>>
>>>> Jess Wellendorff wrote:
>>>>
>>>>
>>>>> Hi jun,
>>>>>
>>>>> I do not know about the interpretation of the convergence for
>>>>> eigenstates vs. density you observe, but I do know that it is easy to
>>>>> change the convergence criteria:
>>>>> ********
>>>>> conv = {'eigenstates':1.e-5, 'density':1.e-4, 'energy':1.e-4}
>>>>> calc = GPAW(....., convergence=conv, .......)
>>>>> ********
>>>>> where any of the arguments to 'convergence' may be left out (leaving
>>>>> it to default value).
>>>>> It appears to me that your calculation has definately converged in
>>>>> terms of electronic structure, but for some reason this happens for a
>>>>> changes in the eigenstates larger than the default criterion for
>>>>> convergence.
>>>>>
>>>>> cheers,
>>>>> Jess W.
>>>>>
>>>>> Den 30-04-2011 11:31, Jun Yan skrev:
>>>>>
>>>>>
>>>>>> Hi, developers,
>>>>>>
>>>>>> What does it mean if the density is converged while the eigenstate
>>>>>> can't converge, as shown in the following ? The default convergence
>>>>>> criterion is 'density': 1.0e-4, 'eigenstates': 1.0e-9. Can I change
>>>>>> the convergence of eigenstate to 1e-4 ? The system here is Al on Si
>>>>>> surface, with 80 occupied and 180 unoccupied bands using rmm-dis
>>>>>> eigensolver. The mixing is mixer=Mixer(0.05, 5, weight=100.0) and dzp
>>>>>> basis is used for initialization of orbitals. It takes around 40 scf
>>>>>> steps to converge using the default criterion for eigensolver CG
>>>>>> although takes much longer time. This is to test whether rmmdis is able
>>>>>> to get reasonable unoccupied bands compared to CG, but the scf step can
>>>>>> never converge...
>>>>>>
>>>>>> All the best,
>>>>>> Jun
>>>>>>
>>>>>> iter: 450 06:45:14 -5.3 -7.7 -223.31010 1 1
>>>>>> iter: 451 06:48:26 -5.3 -7.9 -223.31010 1 1
>>>>>> iter: 452 06:51:39 -5.3 -8.6 -223.31010 1 1
>>>>>> iter: 453 06:54:52 -5.3 -8.3 -223.31010 1 1
>>>>>> iter: 454 06:58:05 -5.3 -8.4 -223.31010 1 1
>>>>>> iter: 455 07:01:17 -5.3 -8.0 -223.31010 1 1
>>>>>> iter: 456 07:04:30 -5.3 -8.2 -223.31010 1 1
>>>>>> iter: 457 07:07:43 -5.3 -8.2 -223.31010 1 1
>>>>>> iter: 458 07:10:56 -5.3 -8.4 -223.31010 1 1
>>>>>> iter: 459 07:14:09 -5.3 -8.3 -223.31010 1 1
>>>>>> iter: 460 07:17:22 -5.3 -8.4 -223.31010 1 1
>>>>>> iter: 461 07:20:34 -5.3 -8.5 -223.31010 1 1
>>>>>> iter: 462 07:23:47 -5.3 -8.1 -223.31010 1 1
>>>>>> iter: 463 07:27:00 -5.3 -7.6 -223.31010 1 1
>>>>>> iter: 464 07:30:13 -5.3 -7.9 -223.31010 1 1
>>>>>> iter: 465 07:33:26 -5.3 -7.9 -223.31010 1 1
>>>>>> iter: 466 07:36:39 -5.3 -8.4 -223.31010 1 1
>>>>>> iter: 467 07:39:52 -5.3 -8.2 -223.31010 1 1
>>>>>> iter: 468 07:43:04 -5.3 -8.4 -223.31010 1 1
>>>>>> iter: 469 07:46:17 -5.3 -10.0 -223.31010 1 1
>>>>>> iter: 470 07:49:30 -5.3 -8.4 -223.31010 1 1
>>>>>> iter: 471 07:52:43 -5.3 -8.7 -223.31010 1 1
>>>>>> iter: 472 07:55:55 -5.3 -8.5 -223.31010 1 1
>>>>>> iter: 473 07:59:08 -5.3 -9.0 -223.31010 1 1
>>>>>> iter: 474 08:02:21 -5.3 -9.6 -223.31010 1 1
>>>>>> iter: 475 08:05:34 -5.3 -8.5 -223.31010 1 1
>>>>>> iter: 476 08:08:46 -5.3 -8.6 -223.31010 1 1
>>>>>> iter: 477 08:11:59 -5.3 -8.7 -223.31010 1 1
>>>>>> iter: 478 08:15:12 -5.3 -8.2 -223.31010 1 1
>>>>>> iter: 479 08:18:25 -5.3 -7.9 -223.31010 1 1
>>>>>> iter: 480 08:21:38 -5.3 -7.7 -223.31010 1 1
>>>>>> iter: 481 08:24:50 -5.3 -7.8 -223.31010 0 1
>>>>>> iter: 482 08:28:03 -5.3 -7.8 -223.31010 1 1
>>>>>> iter: 483 08:31:16 -5.3 -7.7 -223.31010 1 1
>>>>>> iter: 484 08:34:29 -5.3 -8.7 -223.31010 1 1
>>>>>> iter: 485 08:37:42 -5.3 -8.1 -223.31010 1 1
>>>>>> iter: 486 08:40:54 -5.3 -7.8 -223.31010 1 1
>>>>>> iter: 487 08:44:07 -5.3 -7.9 -223.31010 1 1
>>>>>> iter: 488 08:47:20 -5.3 -8.3 -223.31010 1 1
>>>>>> iter: 489 08:50:32 -5.3 -8.1 -223.31010 1 1
>>>>>> iter: 490 08:53:46 -5.3 -7.8 -223.31010 1 1
>>>>>> iter: 491 08:56:58 -5.3 -7.7 -223.31010 1 1
>>>>>> iter: 492 09:00:11 -5.3 -7.6 -223.31010 1 1
>>>>>> iter: 493 09:03:24 -5.3 -7.6 -223.31010 1 1
>>>>>> iter: 494 09:06:37 -5.3 -7.5 -223.31010 1 1
>>>>>> iter: 495 09:09:50 -5.3 -7.9 -223.31010 1 1
>>>>>> iter: 496 09:13:03 -5.3 -8.0 -223.31010 1 1
>>>>>> iter: 497 09:16:16 -5.3 -8.6 -223.31010 1 1
>>>>>> iter: 498 09:19:28 -5.3 -8.7 -223.31010 1 1
>>>>>> iter: 499 09:22:41 -5.3 -7.7 -223.31010 1 1
>>>>>> iter: 500 09:25:54 -5.3 -7.9 -223.31010 1 1
>>>>>> _______________________________________________
>>>>>> 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
>>>
>>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20110502/a58e1176/attachment-0001.html
More information about the gpaw-users
mailing list