[CP2K-user] [CP2K:10810] XC_HYB_MGGA_XC_WB97M_V functional in libxc

Stanislav Šimko stanisl... at gmail.com
Fri Jun 21 14:22:49 UTC 2019


and the saga continues. 
While trying to run some "benchmarks" to see whether I am setting up the 
M-(r)V and X-(r)V functionals correctly, I came across SCF convergence 
problems for some systems. E.g., single oxygen, in the ground (triplet) 
state - at first, SCF gets close to converging but eventually fails to 
converge and, most worrying, diverges. The "problem" seems to come from the 
functional itself - replacing it by PBE (while keeping HF and VdW) leads to 
no convergence problems, while switching off the HF and VdW has no effect.

I observed similar convergence problems with different system (benzene-Ne 
dimer) with large (aug-pc-2, aug-pc-3) basis sets, while a smaller pc-3 
converged well. I didn't test PBE here though.

I use 4.2.3 libxc version.

I would be glad for any suggestions.
Attached are the input and output files for the failed wB97M-v oxygen 
calculation.
Kind regards,
Stanislav.

On Sunday, April 28, 2019 at 9:28:00 PM UTC+2, Stanislav Šimko wrote:
>
> No problem. Actually, I'm slowly working on it a bit more because there 
> are some modifications for the wB97M-V settings, but there's not enough 
> days in a week and hours in a day.... That's also why I haven't submitted 
> via repo.
>
> On Sunday, April 28, 2019 at 9:21:29 PM UTC+2, Frederick Stein wrote:
>>
>> Dear Stanislav,
>>
>> Sorry for the delay. Thank you for your contribution. I included your 
>> patch, now.
>> If you have contributions, it is easier to add them yourself. (see 
>> https://www.cp2k.org/dev:starting)
>>
>> Best regards.
>> Frederick
>>
>> Am Dienstag, 9. Oktober 2018 10:10:15 UTC+2 schrieb Stanislav Šimko:
>>>
>>> If anybody would like to try, here are the patch files for xc_libxc.F 
>>> and xc_libxc_wrap.F files that are in the cp2k/src/xc folder. You can apply 
>>> them from the xc folder like:
>>> patch -p0 -i xc_libxc.F.patch
>>>
>>> On Monday, October 8, 2018 at 1:10:18 PM UTC+2, Stanislav Šimko wrote:
>>>>
>>>> Hello,
>>>>
>>>> I'm using cp2k 6.1 r.18604, git version. I compiled it myself (after
>>>> quite some struggle, probably because of problems with Intel
>>>> compilers), with MKL  18.0.3 and gcc 5.2.0 (both installed by
>>>> HPC admins).
>>>>
>>>> Also, to clarify on the (2) - SCF energy usually goes to huge values 
>>>> after this warning (always for the benzene-Ar system). My box size is now 
>>>> 31A for the benzene-Ar system (I tried up to 36A with no obvious success). 
>>>> I'm troubled by the huge/wrong energy in SCF that does not want to get 
>>>> reasonable not by the warning itself (see my partial logfile).
>>>>
>>>> Thank you!
>>>> Regards,
>>>> stanislav.
>>>>
>>>> On Monday, October 8, 2018 at 11:27:45 AM UTC+2, jgh wrote:
>>>>>
>>>>> Hi 
>>>>>
>>>>> 1) Which version of CP2K are you using? This is an error 
>>>>>    for an internal optimization. Should not appear in newer versions 
>>>>> of the code. 
>>>>> 2) This is a warning. Not serious if your molecule is in the center 
>>>>>    of the box. If you enlarge the box the warning should disappear. 
>>>>>
>>>>> 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: "Stanislav Šimko" 
>>>>> Sent by: c... at googlegroups.com 
>>>>> Date: 10/08/2018 10:37AM 
>>>>> Subject: Re: [CP2K:10810] XC_HYB_MGGA_XC_WB97M_V functional in libxc 
>>>>>
>>>>> Hello, 
>>>>>
>>>>> hopefully, someone will be able to help me further. As suggested, I'm 
>>>>> trying to reproduce Benzene-Ar interaction potential, as well as some water 
>>>>> dimers and some other things. When running with Ahlrichs-def2-QZVP basis, 
>>>>> results seem OK-ish but not convincing enough in my opinion. Therefore, I'm 
>>>>> now trying to get working (aug-)pc-3 basis set with the Benzene-Ar. My 
>>>>> setup does not work though. System energy is exploding during SCF. I get 2 
>>>>> warnings/errors: 
>>>>> 1) POWELL| Error in trust region 
>>>>> 2)  *** WARNING in pw/ps_wavelet_methods.F:236 :: Density non-zero on 
>>>>> the *** 
>>>>>  *** edges of the unit cell: wrong results in WAVELET solver           
>>>>> *** 
>>>>>
>>>>> I have no clue about the first one. Is it something related to math 
>>>>> libraries (MKL)? 
>>>>>
>>>>> The second one can be somewhat avoided when I increase cell size. 
>>>>> However, this does not work all the time and I'm using 30A box already. 
>>>>> Attached is my input. 
>>>>> Do you please have got any suggestions? 
>>>>> Thank you! 
>>>>>
>>>>> On Tuesday, October 2, 2018 at 2:56:13 PM UTC+2, Stanislav Šimko 
>>>>> wrote: 
>>>>> Hi, 
>>>>> Thank you for suggestions. I'm already working on some comparisons 
>>>>> with what I hope could be the wB97M-V functional. Results are not perfect, 
>>>>> but they are not bad as well. I will try to report more soon. 
>>>>>
>>>>> On Tuesday, October 2, 2018 at 10:55:49 AM UTC+2, Frederick Stein 
>>>>> wrote: 
>>>>> Hi, 
>>>>>
>>>>> I am familiar with the related wB97X-functionals by Head-Gordon, 
>>>>> unfortunately not with those containing VV10 dispersion corrections like 
>>>>> wB97(M)-V. Your XC_FUNCTIONAL section is correct. 
>>>>> Use the following INTERACTION_POTENTIAL section 
>>>>>         &INTERACTION_POTENTIAL 
>>>>>           # Chose it a bit less than half the shortest 
>>>>> nearest-neighbour distance (i.e. cubic box with a=4 A, cutoff = 1.99 A; or 
>>>>> cp2k starts complaining) 
>>>>>           CUTOFF_RADIUS 4.0 
>>>>>           # c_sr V_sr + V_lr = c_sr V_Coul+(1-c_sr) V_lr 
>>>>>           SCALE_LONGRANGE 0.833 
>>>>>           SCALE_COULOMB 0.167  # Affects the scaling of the truncated 
>>>>> potential for periodic systems 
>>>>>           # For non-periodic systems use MIX_CL instead 
>>>>>           POTENTIAL_TYPE MIX_CL_TRUNC 
>>>>>           T_C_G_DATA t_c_g.dat 
>>>>>         &END INTERACTION_POTENTIAL 
>>>>> Finally, you miss the VDW-section to include the VV10 correction (See 
>>>>> https://groups.google.com/forum/#!searchin/cp2k/wb97x%7Csort:date/cp2k/JEvnnfAEgPY/oryDShPd_w0J). 
>>>>> They used something like this 
>>>>>       &vdW_POTENTIAL 
>>>>>          DISPERSION_FUNCTIONAL NON_LOCAL 
>>>>>          &NON_LOCAL 
>>>>>            TYPE RVV10 
>>>>>            PARAMETERS 6.3 0.0093 
>>>>>            VERBOSE_OUTPUT 
>>>>>            KERNEL_FILE_NAME 
>>>>> /path/to/your/VV10/table/rVV10_kernel_table.dat 
>>>>>            CUTOFF  150 
>>>>>          &END NON_LOCAL 
>>>>>       &END vdW_POTENTIAL 
>>>>> To my knowledge, cp2k provides only parameters for the rVV10 model, 
>>>>> not for the VV10 model. You should test wether the rVV10 parameters are 
>>>>> suitable. You might have a further look at the discussions to the wB97X-V 
>>>>> functional in this forum. (
>>>>> https://groups.google.com/forum/#!searchin/cp2k/wb97x%7Csort:date/cp2k/JEvnnfAEgPY/oryDShPd_w0J, 
>>>>>
>>>>> https://groups.google.com/forum/#!searchin/cp2k/wb97x|sort:date/cp2k/VsG6j38wdwI/rsJYd0ZpBwAJ, 
>>>>>
>>>>> https://groups.google.com/forum/#!searchin/cp2k/wb97x|sort:date/cp2k/le-ojRVXWa0/acojYhQ0AwAJ) 
>>>>>
>>>>> There are also some examples to test wB97X-V. According to the paper 
>>>>> where wB97M-V was published, you might try some of the databases or the 
>>>>> dissociation curve of this functional (benzene-argon-dimer) depending on 
>>>>> what you can afford. When I tested some functionals, dissociation curves 
>>>>> can be a good choice because the shape differs between different 
>>>>> functionals as well as optimal distances and bonding energies. 
>>>>> I think Juerg wrote in one of the mentioned threads that you should 
>>>>> test the different parts (DFT, HF, VV10) of the functional seperately. 
>>>>> Because there are different flavours of that functional (B97, wB97X, 
>>>>> wB97X-V, these functionals with different dispersion corrections), there 
>>>>> should not be a problem to find reference data. 
>>>>>
>>>>> regards 
>>>>>
>>>>> Frederick   
>>>>>   -- 
>>>>>  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 cp... at googlegroups.com. 
>>>>>  To post to this group, send email to c... at googlegroups.com. 
>>>>>  Visit this group at https://groups.google.com/group/cp2k. 
>>>>>  For more options, visit https://groups.google.com/d/optout. 
>>>>>   
>>>>>
>>>>> [attachment "4.xyz" removed by Jürg Hutter/at/UZH] 
>>>>> [attachment "4.in" removed by Jürg Hutter/at/UZH] 
>>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190621/30fb9087/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: o.in
Type: application/octet-stream
Size: 6510 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190621/30fb9087/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: o.out
Type: application/octet-stream
Size: 90355 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190621/30fb9087/attachment-0001.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: TMD02_O.xyz
Type: chemical/x-xyz
Size: 49 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190621/30fb9087/attachment.xyz>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: TMD02_O-RESTART.wfn
Type: application/octet-stream
Size: 4612 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190621/30fb9087/attachment-0002.obj>


More information about the CP2K-user mailing list