[CP2K:3794] Poisson solvers in QMMM

Laino Teodoro teodor... at gmail.com
Fri Apr 20 19:21:31 CEST 2012


Dear Frank,

On 20 Apr 2012, at 18:48, Frank Uhlig <uhlig... at googlemail.com> wrote:

> Dear Teodoro,
> 
> thanks for the answer. It is not true that "nothing prevents [me] from implementing" the recoupling for different Poisson solvers. As you may understand, there are certain constraints given by work and personal obligations.
> 

Absolutely understandable. Similarly, I think is understandable that, for personal and work constraints, not all have the time to read (and possibly show interest for) such long messages.

> If _you_ decide to do it all in your free time that is your choice and also highly appreciated. But please, remember that it takes time to get familiar with the code, cp2k is a rather large project and not all of us are familiar with FORTRAN, good programmers and might have to use cp2k as a black box.
> 
Just lines that nothing add to the real problem communication..

> This brings me to issue on hand, as a "cp2k-black-box-user" I noticed several inconsitencies in what is written in the reference manual and what is happening during the calculation:
> 
> The section %FORCE_EVAL%QMMM%PERIODIC is classified as "cannot be optional" in the manual. If you take the test from the before mentioned post, you can easily check that this section behaves differently and also results will differ depending on its presence in the input file.
> 
Cannot be optional, means: there is no default associated to it: if you want qm/mm periodic you need to specify it.
No periodic: isolated QMMM
Any PS can be used in that case..
> The section %FORCE_EVAL%QMMM%PERIODIC %MULTIPOLE is classified as "can be optional" and the explanation of its purpose is:
> 

See above... It is on by default if you have PERIODIC (with default values)
> "This section is used to set up the decoupling of QM periodic images with the use of density derived atomic point charges."
> 
> Whether this section is included or not does not change the resulting total energy and in both cases the output states "Recoupling energy [...]".
> 
See above.
> Those two things, seem to be either bugs or errors in the input description. I would appreciate if you could comment on what is what.
> 

See above.
> Furthermore, the Wavelet solver can in principle be employed without further implementations.
> 

Please do it. For me is not trivial at all to handle in a larger box the recoupling with WAVELET (given a smaller QM box) (which in 2006 was not even there, btw)
> When one sets the corresponding periodicity in the DFT settings to XYZ the calculation will be equivalent to a plain, periodic plane-wave calculation and the de- and re-coupling can be used for that and is fully equivalent to using the Multipole solver (also really gives fully equivalent results). 
> 

Again, if you know how to do it... Do it.

> CP2k in some places gives error messages for "illegal combination[s]" of keywords and in my opinion it would be helpful to get such an error message for the combination of the Wavelet solver (for isolated system) and what seems to enable recoupling in the input. Btw, as far as I know,
> 
thanks for the suggestion: according my personal and work constraint, I may decide to implement that
> your 2006 JCTC publications about the QMMM boundary conditions does not mention that the recoupling is not possible for other Poisson solvers and the behaviour of CP2k with respect to the input is just a bit deceiving.
> 
Sure... But strange enough in my 2006 JCTC I only discuss about the DDAPC scheme.. Maybe there is a reason (deeply hidden) for that?
> Thanks in advance and best regards,
> 

You're welcome.
Teo
> Frank
> 
> On Apr 20, 2012 5:23 PM, "Teodoro Laino" <teodor... at gmail.com> wrote:
> >
> > Dear Frank,
> >
> > you have already found the reasons. In order to have recoupling the only poisson solver working at the moment is multipole.
> > Nothing prevents you from implementing also for the other solvers something similar.
> > If instead you want to use cp2k as a black box, you have to stay within the implementation borders (see papers and regtests input provided): everything beyond that is not guaranteed to work.
> > Regards,
> > Teo
> >
> > On Apr 20, 2012, at 4:56 PM, Frank Uhlig wrote:
> >
> >> Dear CP2k'lers,
> >>
> >> related to a previous post (https://groups.google.com/d/topic/cp2k/iMsqLpZnRyQ/discussion) I just wanted to highlight the main problems again, because they might have been hidden in the mess.
> >>
> >> The QMMM electrostatic coupling is different for different Poisson solvers if analogous inputs with Wavelet and Multipole solver are used. In the example from the previous post (using Multipole solver) the QM images are recoupled. If one just changes the Poisson solver to the Wavelet solver, no recoupling of the QM images is performed. This behaviour then further depends on the inclusion/exclusion of the non-optional section %FORCE_EVAL%QMMM%PERIODIC.
> >> The input can be changed in a way that one reproduces the total energies with both Poisson solvers (up to ~0.01 eV).
> >>
> >> This behaviour is counterintuitive, imo, and I would appreciate comments of the developers, if it is intended, if I am doing something wrong etc pp
> >>
> >> Cheers,
> >>
> >> Frank
> >>
> >>
> >> -- 
> >> You received this message because you are subscribed to the Google Groups "cp2k" group.
> >> To post to this group, send email to cp... at googlegroups.com.
> >> To unsubscribe from this group, send email to cp2k+uns... at googlegroups.com.
> >> For more options, visit this group at http://groups.google.com/group/cp2k?hl=en.
> >
> >
> > -- 
> > You received this message because you are subscribed to the Google Groups "cp2k" group.
> > To post to this group, send email to cp... at googlegroups.com.
> > To unsubscribe from this group, send email to cp2k+uns... at googlegroups.com.
> > For more options, visit this group at http://groups.google.com/group/cp2k?hl=en.
> -- 
> You received this message because you are subscribed to the Google Groups "cp2k" group.
> To post to this group, send email to cp... at googlegroups.com.
> To unsubscribe from this group, send email to cp2k+uns... at googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/cp2k?hl=en.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20120420/7aee8dcc/attachment.html>


More information about the CP2K-user mailing list