[CP2K-user] [CP2K:12815] SCAN convergence problems for charged systems

[Travis]

Hi,

You're combining a very large cutoff (1200 Ry) with a small rel_cutoff (40 
Ry), which creates an unbalanced grid. Large cutoffs require a larger 
rel_cutoff so the added points aren't just pushed to the coarsest grids.

1200/120
 count for grid        1:          89892          cutoff [a.u.]          
600.00 # lots of points on fine grid, expensive
 count for grid        2:          63191          cutoff [a.u.]          
200.00
 count for grid        3:          42699          cutoff [a.u.]           
66.67
 count for grid        4:          17703          cutoff [a.u.]           
22.22
 total gridlevel count  :         213485

800/50
 count for grid        1:          76679          cutoff [a.u.]          
400.00
 count for grid        2:          57951          cutoff [a.u.]          
133.33
 count for grid        3:          45900          cutoff [a.u.]           
44.44
 count for grid        4:          32955          cutoff [a.u.]           
14.81
 total gridlevel count  :         213485

1200/40
 count for grid        1:          33765          cutoff [a.u.]          
600.00 # too few points on fine grid
 count for grid        2:          56127          cutoff [a.u.]          
200.00
 count for grid        3:          63191          cutoff [a.u.]           
66.67
 count for grid        4:          60402          cutoff [a.u.]           
22.22
 total gridlevel count  :         213485

Simply switching to 800 Ry / 50 Ry (cutoff / rel_cutoff) or keeping your 
initial 1200 Ry but using a rel_cutoff of 120 Ry leads to rapid convergence.

-T

On Monday, January 27, 2020 at 2:27:42 PM UTC-4, Chris Ahart wrote:
>
> Dear Juerg,
>
> Thanks for your response.
>
> The 8x8x8 cell does indeed converge, giving an IP of 288.9 kcal in good 
> agreement with the G12IP dataset SCAN value of 287.9 kcal (J. G. 
> Brandenburg, J. E. Bates, J. Sun and J. P. Perdew, *Phys. Rev. B*, 2016, 
> *94*, 17–19.).  Reducing the cell dimensions is however problematic for 
> other systems (e.g NH3, Cl2), as a cell small enough to give non-diverging 
> SCF then has non-zero density at the cell edges. Increasing the multigrid 
> cutoff appears to make this problem worse.
>
> I have also done some condensed phase tests on bulk iron oxides for 
> polaron studies. Perhaps without any vacuum where the density becomes 
> small, the charged state converges easily. However the resultant potential 
> energy surface is very flat, with localised and delocalised polarons being 
> energy degenerate (HSE and other methods predict small polarons). At least 
> one other group has attempted to use SCAN with CP2K, and they found that 
> SCAN performs even worse than LDA for describing solid gallium (R. C. 
> Remsing, J. Sun, U. V. Waghmare and M. L. Klein, *Mol. Phys.*, 2018, *116*, 
> 3372–3379). Combined with the knowledge that small charged systems can't 
> converge for reasonable cell sizes, it looks like SCAN doesn't work very 
> well with CP2K. 
>
> Would you have any further advice, or does this seem like a methodological 
> problem with SCAN and CP2K?
>
> Regards,
> Chris
>
> On Monday, 27 January 2020 13:17:16 UTC, jgh wrote:
>>
>> Hi 
>>
>> I couldn't find a clear bug for this. My current best guess is 
>> that the problem is related to the numerical instabilities 
>> of the SCAN functional. 
>> Molecules in large boxes have many grid points with small and 
>> 'random' density/gradient values that cause problems. 
>> Your special case works for me in a 8x8x8 box (instead of 10x10x10)! 
>>
>> regards 
>>
>> Juerg Hutter 
>> -------------------------------------------------------------- 
>> Juerg Hutter                         Phone : ++41 44 635 4491 
>> Institut für Chemie C                FAX   : ++41 44 635 6838 
>> Universität Zürich                   E-mail: h... at chem.uzh.ch 
>> Winterthurerstrasse 190 
>> CH-8057 Zürich, Switzerland 
>> --------------------------------------------------------------- 
>>
>> -----c... at googlegroups.com wrote: ----- 
>> To: "cp2k" <c... at googlegroups.com> 
>> From: "Chris Ahart" 
>> Sent by: c... at googlegroups.com 
>> Date: 01/24/2020 04:45PM 
>> Subject: [CP2K:12815] SCAN convergence problems for charged systems 
>>
>> Dear CP2K users, 
>>
>> I am attempting to use the SCAN functional from libXC 4.2.1 with CP2K 6, 
>> performing some basic benchmark calculations from the G12IP dataset. While 
>> SCAN converges well for neutrals systems, I am unable to get convergence 
>> for a charged system. 
>>
>> I attach an example of the CH4 system (input and output for both neutral 
>> and charged), where the neutral state converges within 20 steps while the 
>> charged state immediately diverges. I have observed the same behaviour with 
>> many other systems. I have also tried using LDA pseudopotentials as well as 
>> the SCAN optimised pseudopotentials from github.com/juerghutter/GTH, in 
>> addition to all electron calculations. I have confirmed that the same input 
>> file converges with HF, and with PBE from both the CP2K implementation and 
>> from libXC. 
>>
>> I have also found that the charged system converges when only the 
>> correlation part of the SCAN functional is included, suggesting that the 
>> problem may be related to the exchange part of the SCAN functional. 
>>
>> Any help with this problem would be greatly appreciated. 
>>
>> Regards, 
>> Chris   
>>   -- 
>>  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 c... at googlegroups.com. 
>>  To view this discussion on the web visit 
>> https://groups.google.com/d/msgid/cp2k/4c4384f8-c896-416f-9a8f-98c2b23c0d14%40googlegroups.com. 
>>
>>   
>>
>> [attachment "charged.inp" removed by Jürg Hutter/at/UZH] 
>> [attachment "charged.out" removed by Jürg Hutter/at/UZH] 
>> [attachment "neutral.inp" removed by Jürg Hutter/at/UZH] 
>> [attachment "neutral.out" removed by Jürg Hutter/at/UZH] 
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200127/f01d2b4d/attachment.htm>