[Pw_forum] tpss: too many bands are not converged

Mike Marchywka marchywka at hotmail.com
Fri Apr 25 15:10:34 CEST 2014

> From: paolo.giannozzi at uniud.it
> To: pw_forum at pwscf.org
> Date: Fri, 25 Apr 2014 03:42:06 +0200
> Subject: Re: [Pw_forum] tpss: too many bands are not converged
> On Thu, 2014-04-24 at 09:03 -0400, Mike Marchywka wrote:
>> By free volume, you mean places where n and likely grad everything are numerical noise?
> exactly: the regions of space where the charge has decayed to noise.
> Apparently (the current implementation of) meta-GGA doesn't like the
> presence of such large regions (and it doesn't seem to like H atoms
> as well). Even plain GGA diverges in those regions: it is less of a
> problem, but still, the exact energy values depend upon how the

Does it actually diverge or just reach some kind of limit cycle with
amplitude in energy greater than convergence criteria?  

> potential is cut off. The divergence comes from gradient corrections
> to exchange: there is no divergence in atoms with the correct large-r

so this is related to PP and lack of exponential decay giving more weight
to "free volume" than it should physically have? I had been doing some
tests with analytical hydrogen since it was easy to check :) That is not noise
alone then but a modelling issue. 

> limit of the charge density, but this is not of much help in plane-wave
> calculations of condensed-matter systems.

Molecular crystals better or worse than say metals? 

>From what I can tell of the TPSS "z" anyway, trying to calculate it
as a ratio of intermediate real quantities seems to allow for many errors
such as aliasing that may occur by taking magnitudes.  I did some
algebra it looks like the magnitude of the grad of the wf angle is more direct but I have
not checked the math or tested it much although IIRC some related results
vs analytical hydrogen did not appear obviously wrong but this was hardly a complete test.

> (disclaimer: this is what I remember. I looked at this problem several
> years ago for meta-GGA, many years ago for GGA)

Thanks, I'm just trying to understand a few things and see if I can play with
the problematic terms and evaluate some alternatives.

> Paolo
>> AFAICT, the converge of SCF is largely empirical
>> as there does not seem to be much theory here. In any case, you would like to think that the areas lacking "physics" should not be that
>> big a factor in the overall calculation and if that is not the case there may be something interesting here to explore :)
>>> You can try to decrease the size of the unit-cell in order to minimize the free space it contains. Of course, this will be at the expense of the inter-polyemer interaction you probably want to avoid.
>>> Nevertheless, this will probably not cure your problem, which to my knowledge has no other solution than changing the meta-GGA functional to a GGA one.
>> I took a look at the "z" parameter which being a ratio could have some issue with small denominators. With some initial work,
>> it looks like there are better ways to calculate it than directly from real quantities such as n and grad mags,
>> but I am still trying to test what may be trivially wrong math :)
>> I think my latest result was something to the effect that you could reduce z to A/(A+B)( both non-negative)
>> if you calculate everything from psi and the grad of psi expressed in polar form, not much different from some formula
>> for current densities, and maybe find a limit when A and B are both zero.
>> I had originally hoped to try to test it for trivial issues in JDFTX and c++ or c with libxc using an expanded interface scheme
>> to let me pass z or grad psi in polar form but it will require a bit of effort. Curious if anyone else here has examined
>> this or has literature references they care to share. Seems disconcerting that something with physical meaning would have these
>> kinds of implementation issues - perhaps something else can be learned from investigating.
>> Thanks.
>>> Best regards,
>>> Yves
>>> Le 24 avr. 2014 à 13:36, Davide Tiana a écrit :
>>>> Dear all,
>>>> I've been trying to calculate a polymer using tpss. I already fully
>>>> optimised it with different functional (pbesol, pw91, hse) and
>>>> pseudopotential (nc, paw). For some reason (apologise my ignorance if
>>>> it should be well known) when I try to calculate using tpss nc the
>>>> system simply does not converge crashing with this error:
>>>> Error in routine c_bands (1):
>>>> too many bands are not converged
>>>> After some reaserch in old pw-forum mails, I tried decreasing
>>>> mixing_beta=0.2, increasing mixing_ndim=12, as well as mixing mode TF
>>>> and local-TF.
>>>> everythings failed.
>>>> Does anyone have any tips or idea of why the system can't converge?
>>>> Thanks a lot,
>>>> Davide
>>>> _______________________________________________
>>>> Pw_forum mailing list
>>>> Pw_forum at pwscf.org
>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>> _______________________________________________
>>> Pw_forum mailing list
>>> Pw_forum at pwscf.org
>>> http://pwscf.org/mailman/listinfo/pw_forum
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://pwscf.org/mailman/listinfo/pw_forum
> --
> Paolo Giannozzi, Dept. Chemistry&Physics&Environment,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum

More information about the users mailing list