[CP2K-user] [CP2K:11166] general normconserving UPF in CP2K

[Michiel van Setten]

Hi Juerg,

Thank for the comments,

concerning 1, If you like I can send you the inputs and the backtrace I get 
with the nlcc

2) no overhead in the scf is very good news,

3) ok thanks, that confirms my understanding, the factor 4 was include in 
my comparison to abinit. In this I take the smallest rel_cutoff that gives 
me a converged result for a given cutoff value. 

4) sorry for the confusion. Here I meant, the PD-UPF calculation with cp2k 
took more iteration steps than the GTH calculation using cp2k, everything 
else being the same: same basis, same cutoffs, same mixer / 
diagonalization. Here I was hoping that a basis set optimized to fit the 
PseudoDojo potential would improve the convergence. If you have some hints 
an tips for generating them I would appriciate them.

I just generated a q4 Hf potential with ONCVPSP, for this I would need a 
new basis set anyways. Did you ever made an attempt for a q4 Hf?

Until now I only ran convergence test on the total energy, so I have no 
idea yet on the actual relative performance of the pseudo potentials. This 
afternoon I programmed some scripts to run delta-test with cp2k for the 
different pseudo-potentials. This should give a better idea on the 
precision of the fitting procedure. 

Thanks,
Michiel



Op vrijdag 11 januari 2019 14:38:24 UTC+1 schreef jgh:
>
> Hi Michiel 
>
> > 
> >I have run some tests on this topic and would just like to share my 
> experience. 
> > 
> >I ran the pseudo-dojo 0.4 PBE Si pseudo potential combined with the the 
> molopt GTH DZ basis set. 
> > 
> >It took me some time and reading the source code to find out how but 
> eventually cp2k ran. I now use 
> > 
> >    POTENTIAL_FILE_NAME PD04-UPF-POTENTIALS (in the &DFT section) 
> > 
> >and 
> > 
> >    &KIND Si 
> >      POTENTIAL UPF PD04-UPF-POTENTIALS 
> >      BASIS_SET DZVP-MOLOPT-SR-GTH 
> >    &END KIND 
> > 
>
> The POTENTIAL_FILE_NAME should not be necessary. What would you do if 
> there are two different atoms? 
>
> >Comparing calculations with the same cutoff and relcutoff, only the psp 
> is different, I observe the following 
> > 
> >1) I needed to turn of non linear core corrections to prevent cp2k to 
> segfaut in the conversion step 
>
> This is probably a bug? I thought NLCC worked but I cannot find an example 
> to proof it. 
>
> > 
> >2) It seems the gausian fit results in many more gausians than the 
> typical GTH pseudo's 
> > - This however does not seem to deteriorate performace: 
> calculate_rho_elec and integrate_v_rspace take more or less the same time 
> per iteration for both psp's 
>
> The internal format is similar but not exactly the same as the GTHs. Yes, 
> there are much more Gaussians needed for the fit. 
> Performance is not a major concern, as CP2K calculates the PP integrals 
> analytically and adds them to the core Hamiltonian 
> at the start of the calculation. During the SCF there is no overhead. 
>
> >3) I observe a somewhat improved convergence with cutoff and rel_cutoff, 
> comparable to what I am used to when using this pseudo in abinit 
>
> The cutoff in CP2K is related to the density cutoff in abinit, usually 4x 
> the input cutoff. 
> rel_cutoff is a CP2K internal accuracy setting for the integrations of 
> Gaussians, there is no relation to any cutoff in abinit. 
>
> >4) The electronic convergence systematically takes more steps. 
> > - I assume this is caused by the use of basisfunctions not optimized for 
> this pseudo (can someone confirm?) 
>
> This is not clear, more steps than in abinit? The number of steps depends 
> on the optimizer. CP2K is rather good for insulators and using 
> the OT method. For metals and diagonalization methods you need to find the 
> best setting, defaults are not very good. 
>
> >5) in the case of the UPF pseudo I see an additional  read_qs_kind in the 
> timing report, I assume this is the part where the upf is read and fitted. 
> I assume it >should be possible to do this fit once and safe the GTH 
> format. 
>
> see above. The fit is not compatible with the GTH format, you cannot just 
> print and reuse it. 
>
> Thanks for testing this feature. 
>
> best regards 
>
> Juerg 
>
>
> Any comment is welcome. 
>
> Please let me know if someone thinks I'm polluting this channel ;-) 
>
> thanks for reading, 
> Michiel 
>
> Op dinsdag 18 december 2018 09:29:20 UTC+1 schreef jgh:Hi 
>   
> there is minimal support for general normconserving pseudopotentials from 
> UPF 
> in CP2K. The potentials are internally fit to a Gaussian form, similar 
> to the GTH pseudos. The accuracy of the fit has not been thoroughly 
> tested. It should work for most variants (semi-local, non-local) forms, 
> but again only minimal testing has been done. 
> There are plans to implement a direct integration of the UPF pp   
> sometime next year.   
> Another problem will be the available basis sets. You will need 
> to generate also those, or rely on the PBE-GTH optimized basis sets. 
>   
> best regards 
>   
> Juerg 
> -------------------------------------------------------------- 
> Juerg Hutter                         Phone : ++41 44 635 4491 
> Institut für Chemie C                FAX   : ++41 44 635 6838 
> Universität Zürich                   E-mail: hut... at chem.uzh.ch 
> Winterthurerstrasse 190 
> CH-8057 Zürich, Switzerland 
> --------------------------------------------------------------- 
>   
> -----"'Michiel van Setten' via cp2k" <cp... at googlegroups.com> wrote: 
> ----- 
> To: "cp2k" <cp... at googlegroups.com> 
> From: "'Michiel van Setten' via cp2k" <cp... at googlegroups.com> 
> Date: 12/17/2018 11:40AM 
> Subject: [CP2K:11053] general normconserving UPF in CP2K 
>   
> Dear community, 
>   
> I have seen some keywords concerning the use of UPF pseudo potentials in 
> CP2K, I however did not manage to figure out if this means cp2k can use 
> general normconserving UPF pseudopotentials.   
>   
> I would like to make some pseudopotentials with different amounts of 
> valence electrons for some elements. Since I have some experience in 
> generating ONVPSP pseudo potentials, for me the easiest would generate 
> ONCVPSP's and use them in CP2K, either in upf format or, as a second 
> option, via a conversion into HGH format.   
>   
> If any one has some experience, it would be very welcome, reasons why this 
> would be a bad idea would be equally welcome ;-) 
>   
> many thanks, 
> Michiel van Setten 
>   
>     
>   --   
>  You received this message because you are subscribed to the Google Groups 
> "cp2k" group. 
>  To unsubscribe from this group and stop receiving emails from it, send an 
> email to cp2k+... at googlegroups.com. 
>  To post to this group, send email to cp... at googlegroups.com. 
>  Visit this group at https://groups.google.com/group/cp2k. 
>  For more options, visit https://groups.google.com/d/optout. 
>   
>   
>   -- 
>  You received this message because you are subscribed to the Google Groups 
> "cp2k" group. 
>  To unsubscribe from this group and stop receiving emails from it, send an 
> email to cp2k+... at googlegroups.com <javascript:>. 
>  To post to this group, send email to cp... at googlegroups.com <javascript:>. 
>
>  Visit this group at https://groups.google.com/group/cp2k. 
>  For more options, visit https://groups.google.com/d/optout. 
>   
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190111/c1eea2ff/attachment.htm>