[gpaw-users] LCAO basis sets generation

[Ask Hjorth Larsen]

Dear Alexander

2016-03-26 10:34 GMT+01:00 Alexander Romanov <alphard.rm at gmail.com>:
> Dear, Ask
>
> Thanks a lot for your help. Now I can make my calculation with PW91
> functional (and other
> functionals too).
>
> Concerning inaccurate results: Previously I saw strange shifts of different
> parts of optimization
> curve during lattice constant searching.  I made some calculations with
> default dzp basis
> sets and often couldn't converged shifts at all.
>
> After generation proper dzp and tzp basis sets for functionals most of these
> shifts have disappeared. I play two days around this problem and it seems to
> me that I can overcome
> it now. I still have found shifts in some places but now they can be
> converged by slightly
> increasing accuracy of solving Poisson equation or grid space.
>
> I am a little bit surprised why these shifts move only part of the curve far
> from the minimum.
> May be it's a consequence of finite "rcut"?

The functions should approach zero smoothly, so that should not be a problem.

I still don't understand what these "shifts" would be about or what
size they are.  5 meV?  1 eV?

Maybe it has to do with the opimization algorithm.

If anything, try tightening the density convergence criterion.

Best regards
Ask

>
> Best regards,
> Alex
>
>
>
>
> 2016-03-23 18:44 GMT+03:00 Ask Hjorth Larsen <asklarsen at gmail.com>:
>>
>> (Re-added list; please reply to list)
>>
>> 2016-03-23 9:40 GMT+01:00 Alexander Romanov <alphard.rm at gmail.com>:
>> > Thanks for your  reply, Ask.
>> >
>> > I have found the source of discrepancy between command line and script
>> > generation. Command line gpaw-basis has a parameter scalarrel = True as
>> > a
>> > default one. Now I have only little difference in basis sets.
>> >
>> > But I still can not use PW91 basis set in my calculation. When I do
>> > something like this:
>> >
>> > calc = GPAW(mode='lcao', basis='dzp', xc='PW91', kpts=(12,12,12,),
>> > txt=Pt_Lconst.txt)
>> >
>> > I get an error:
>> >
>> > RuntimeError: Could not find required PAW dataset file "Pt.PW91".
>> >
>> > I has no error with other xc (LDA, PBE, RPBE).
>> >
>> > So, the question is: can I do calculation with PW91 functional in lcao
>> > mode?
>>
>> You need to generate a PW91 setup.  gpaw-basis does not save it,
>> although for technical reason it is calculated as part of the
>> calculation.  So either run gpaw-setup <....> and later gpaw-basis
>> with corresponding parameters, or write a script which generates both
>> the setup and its corresponding basis set and saves them.
>>
>> It has been on my stagnant todo-list for years to do allow to save the
>> setup file while running gpaw-basis (and many other improvements).
>>
>> Now *drumroll* there's a merge request which will implement this:
>>
>>   https://gitlab.com/gpaw/gpaw/merge_requests/27
>>
>> Then this will be possible:
>>
>>   gpaw-basis --save-setup <args>
>>
>> You mentioned that some results were inaccurate.  Could you please
>> elaborate?
>>
>> Best regards
>> Ask
>>
>>
>> >
>> > Best regards,
>> > Alex
>> >
>> >
>> > 2016-03-22 17:00 GMT+03:00 Ask Hjorth Larsen <asklarsen at gmail.com>:
>> >>
>> >> Dear Александр
>> >>
>> >> 2016-03-22 13:05 GMT+01:00 Александр Романов <alphard.rm at gmail.com>:
>> >> > Hello,
>> >> >
>> >> > We've use LCAO mode for fast optimization and met some problems with
>> >> > basis
>> >> > sets
>> >> > for transition metals.
>> >> >
>> >> > Default dzp basis sets for Cu, Ag, Pd and Pt give unphysical results
>> >> > for
>> >> > lattice constant.
>> >>
>> >> Which lattice constant do you get then, and which did you expect?
>> >> Would you please attach the input scripts and logs?
>> >>
>> >> > Better convergence of Poisson solver, denser k-points and grid-points
>> >> > meshes
>> >> > give no improvement.
>> >> > We've tried to use gpaw-basis command and manual script for making
>> >> > tzp
>> >> > sets.
>> >> > So, we have
>> >> > two questions:
>> >> >
>> >> > 1)If we use command
>> >> >
>> >> > gpaw-basis -f PW91 --lpol=1 -t tzp Pt
>> >> >
>> >> > we get an error:
>> >> >
>> >> > File "/home/user/bin/gpaw-1.0.0/lib/python/gpaw/atom/all_electron.py"
>> >> > .
>> >> > line
>> >> > 779. in shoot
>> >> > dudrplus = 0.5 * (u[gtp + 1] - u[gtp - 1]) / dr[gtp]
>> >> > IndexError: index out of bounds
>> >> >
>> >> > But all works fine for PBE and RPBE functionals or other elements
>> >> > (for
>> >> > example, Pd).
>> >> >
>> >> > 2) If we use manual script, we get basis set twice as large as
>> >> > gpaw-basis
>> >> > set. As a consequence we've got
>> >> > slow calculations.
>> >>
>> >> Again, please attach the script.
>> >>
>> >> >
>> >> > How can we resolve problem with gpaw-basis command and what's the
>> >> > difference
>> >> > between gpaw-basis
>> >> > and manual script basis set generation?
>> >>
>> >> gpaw-basis is a limited interface.  If you write a Python script to
>> >> generate the basis, you get much more flexibility.  This may in the
>> >> end be what you are looking after, if gpaw-basis does not serve.
>> >>
>> >> Best regards
>> >> Ask
>> >>
>> >> >
>> >> > Best regards,
>> >> >
>> >> > Alex
>> >> >
>> >> >
>> >> >
>> >> >
>> >> > _______________________________________________
>> >> > gpaw-users mailing list
>> >> > gpaw-users at listserv.fysik.dtu.dk
>> >> > https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>> >
>> >
>
>