[CP2K-user] [CP2K:19524] Switching from GPW to GAPW

[Léon Luntadila Lufungula]

Hi Marcella and Prof. Hutter,

Thanks a lot for your valuable insights and suggestions!

Marcella, I'll definitely explore the Basis Set Exchange and check out the 
EMSL_BASIS_SETS file to find a more suitable basis set for my calculations.

Prof. Hutter, I gave your input a shot with the TZVPP-MOLOPT-GGA-ae basis 
set on an alternate HPC cluster since our usual setup is under maintenance 
until after the weekend. The CP2K version here is 2023.1, so I couldn't try 
the "GAPW_1C_BASIS EXT_SMALL" keyword as this was introduced in version 
2023.2. Unfortunately, the calculation failed with a "Cholesky 
decomposition failed. Matrix ill-conditioned" error. I attached the input 
and output files for your reference.

Any guidance on troubleshooting this problem would be much appreciated.

Also, any advice on optimizing the LEBEDEV_GRID and RADIAL_GRID parameters 
for different atom kinds would be fantastic.

Thanks again for all the help!

Best regards,
Léon

On Thursday, 16 November 2023 at 11:09:54 UTC+1 Jürg Hutter wrote:

> Hi
>
> you can add
> RADIAL_GRID 100
> LEBEDEV_GRID 100
> HARD_EXP_RADIUS 2.10
> to Ti KIND section
> This is my QS section for accurate calculations
> &QS
> METHOD GAPW
> EPS_DEFAULT 1.E-14
> EPSFIT 1.E-4
> EPSISO 1.E-12
> EPSRHO0 1.E-6
> LMAXN0 4
> GAPW_1C_BASIS EXT_SMALL
> &END QS
> The save side for the cutoff would be
> &MGRID
> CUTOFF 400
> REL_CUTOFF 60
> &END MGRID
>
> Yes, you can perform pseudopotential calculations with GAPW. You can also
> mix all electron and pseudopotentials.
>
> Please note that the all electron basis sets in BASIS_MOLOPT_UZH are 
> modeled
> after the def2 Ahlrichs basis sets. In their nomenclature SVP stands for
> split-valence-polarisation. So this is equivalent to a DZVP basis from 
> other schools (double zeta valence + polarisation) and not a single zeta 
> basis.
> You can easily verify this by checking the number of functions.
>
> regards
> JH
>
> ________________________________________
> From: cp... at googlegroups.com <cp... at googlegroups.com> on behalf of Léon 
> Luntadila Lufungula <Leon.luntad... at uantwerpen.be>
> Sent: Tuesday, November 14, 2023 12:04 PM
> To: cp2k
> Subject: Re: [CP2K:19498] Switching from GPW to GAPW
>
> Dear Prof. Hutter,
>
> Thanks for pointing me into the right direction!
>
> Unfortunately, the defaults are not sufficient for my system as I am 
> getting rather large values for the lines you suggested to check:
>
> (Inner SCF loop 1)
> Total charge density (r-space): 0.0000427891
> Total Rho_soft + Rho0_soft (g-space): 0.0001310539
> (Inner SCF loop 2)
> Total charge density (r-space): 0.0000494000
> Total Rho_soft + Rho0_soft (g-space): 0.0001376711
>
> I have attached my input file and the short part that was written to the 
> output if you would like to see it in more detail. My system consists of a 
> TiO2 [anatase (101)] slab with organic molecules adsorbed onto the surface, 
> so I suspect that it might be the Ti atoms for which the defaults are 
> insufficient.
>
> I have been looking into GAPW in the regtests, documentation and forum 
> over the last couple of days and from what I understand, the most important 
> parameters to change should be the following:
> (FORCE_EVAL / SUBSYS / KIND)
>
> * LEBEDEV_GRID: I'm a bit unsure how to get a correct value for this 
> parameter for each atom kind. Does this involve a convergence test for each 
> atom type individually or is there some rule of thumb?
> * RADIAL_GRID: same as for LEBEDEV_GRID.
>
> (FORCE_EVAL /DFT / QS)
>
> * EPS_DEFAULT: This one is already at 1.0E-12, should I reduce it further?
> * EPS_FIT: The default is 1.0E-4, but I've read in another post in a reply 
> from M. Krack<
> https://groups.google.com/g/cp2k/c/FUvjXq9RvWY/m/8VwHvHGUAgAJ> that it is 
> too large for most systems and should be reduced to 1.0E-6. Should I reduce 
> it in my case?
> * EPS_ISO: I've seen it mentioned a couple of times, but never if the 
> value should be reduced or what to do with it exactly.
> * EPS_RHO0: According to the reply of M. Krack<
> https://groups.google.com/g/cp2k/c/FUvjXq9RvWY/m/8VwHvHGUAgAJ> smaller 
> values than the default give better accuracy.
>
> (FORCE_EVAL / DFT / MGRID)
>
> * CUTOFF: This is generally lower than for a GPW calculation as I 
> understand it but I guess I should perform a convergence test to choose an 
> appropriate value. Alternatively, I could also multiply the relative cutoff 
> by the maximum exponent in the soft basis set (PRINT_LEVEL MEDIUM to get 
> the values right?) as you and Marcella suggested in another post<
> https://groups.google.com/g/cp2k/c/JX8bDi_w478/m/okVfMy6EAQAJ>.
> * REL_CUTOFF: I'm a bit unsure about how to choose this as all the 
> convergence schemes I've read about first converge the CUTOFF with a large 
> REL_CUTOFF and then reduce the value of REL_CUTOFF to get the best balance 
> between accuracy and efficiency, but as explained above, the rule of thumb 
> for the cutoff uses the value of REL_CUTOFF to determine the CUTOFF...
>
> Any suggestions on how to optimize these parameters would be greatly 
> appreciated!
>
> I also have two other quick questions about your reply:
>
> 1. I thought that GPW was used for pseudopotential calculations and GAPW 
> was used for all electron calculations, but now you say that you can also 
> do pseudopotential calculations with GAPW. Is there an advantage to using 
> GAPW over GPW? I know that the disadvantage is that some features are not 
> present for GAPW calculations, but is there a reason to use GPW if all your 
> intended features are implemented for GAPW?
> 2. The BASIS_MOLOPT_UZH file contains SVP,- TZVPP- and QZVPP-MOLOPT-GGA-ae 
> basis sets. Is there also a double zeta all electron (MOLOPT) basis set 
> somewhere I can use? I have always been told that single zetas are not 
> intended for production runs, but I suspect that TZVPP is already quite 
> computationally expensive for a pre-optimization.
>
> My apologies for the barrage of questions, but I don't have anyone in my 
> group with experience in CP2K so it's hard sometimes to figure these things 
> out on my own...
>
> Thanks in advance and for your previous reply!
>
> Kind regards,
> Léon
> On Tuesday, 14 November 2023 at 10:51:05 UTC+1 Jürg Hutter wrote:
> Hi
>
> For a pseudopotential calculation you can just change "METHOD GPW" to
> "METHOD GAPW".
> If you want to do all electron calculations, you also need to use 
> corresponding
> basis sets and potentials, e.g.
> BASIS_SET SVP-MOLOPT-GGA-ae
> POTENTIAL ALL
> Use the default settings first, check the lines
> Total charge density (r-space): -0.0000000683
> Total Rho_soft + Rho0_soft (g-space): -0.0000000043
> in the output. If those values are larger than 0.00001 you need probably to
> change defaults.
> See the regtests and manual for more information.
> Come back here for if this basic setup fails.
>
> regards
> JH
>
> ________________________________________
> From: cp... at googlegroups.com <cp... at googlegroups.com> on behalf of Léon 
> Luntadila Lufungula <Leon.luntad... at uantwerpen.be>
> Sent: Wednesday, November 8, 2023 5:21 PM
> To: cp2k
> Subject: [CP2K:19481] Switching from GPW to GAPW
>
> Dear all,
>
> I want to perform an AIM analysis with Critic2 which (for as far as I 
> know) requires an all-electron density (i.e. not the density obtained from 
> a GPW calculation) and also plan to calculate NMR chemical shifts in the 
> future which will also require the all-electron GAPW method. Currently, I 
> have figured out how to do calculations with the GPW method but I'm still a 
> bit unsure about how to go from a GPW to a GAPW calculation. Is there 
> someone who could elaborate on what to look at when modifying my input 
> file? I've looked into it a couple of times, but it's a lot of different 
> parameters and I can't seem to figure out how it works...
>
> All the best,
> Léon
>
>
> --
> 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+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com
> >.
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/cp2k/670d2377-8c38-4444-9130-14ebd3d32d09n%40googlegroups.com
> <
> https://groups.google.com/d/msgid/cp2k/670d2377-8c38-4444-9130-14ebd3d32d09n%40googlegroups.com?utm_medium=email&utm_source=footer
> >.
>
> --
> 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+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com
> >.
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/cp2k/36846b0d-2a8d-4ce1-a35e-991b76854170n%40googlegroups.com
> <
> https://groups.google.com/d/msgid/cp2k/36846b0d-2a8d-4ce1-a35e-991b76854170n%40googlegroups.com?utm_medium=email&utm_source=footer
> >.
>

-- 
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+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/a26cd79f-f8c1-49c3-9d11-47ae715d2a83n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20231116/e4a4d86b/attachment.htm>