[CP2K:22] Re: DCACP potentials

[Axel Kohlmeyer]

dear teo and juerg,

thanks a lot for your replies. this is _very_ helpful.

so it seems, that with our first choice for a test case (co2 dimers) we got
lucky, since i can actually reproduce, e.g., optimal geometries from CPMD
with CP2K using an otherwise as similar as possible setup.

the sad news is, that even in this favorable case the changes due to
the additional
f-projector are miniscule and deviations from high-quality non-DFT calculations
in the inter-molecule distance are still quite substantial (which is
remarkable,
since the oxygen was supposedly parametrized for CO2, but probably with a
different geometry (can't tell for sure, since i could not get my hands on the
paper yet, only: http://lcbcpc21.epfl.ch/DCACP/download/supp_prb/supp.pdf ).

thanks,
    axel.

On 4/29/07, Teodoro Laino <teodor... at gmail.com> wrote:
>
> Ciao Axel,
> There's a quite big problem we discovered recently about DCACP
> potentials within cp2k..
> They have a very long radius for the non-local part of the pseudo
> potential..
>
> Now first let me express some doubt about the fitting procedures..
> how is it possible that something that should fix a short-range
> behavior has a radius of several tens of Angstrom? You may argue that
> is the non-local part to have large radius.. but again this
> is far from being reasonable to me...
>
> My opinion is that the fit procedure adopted to get the DCACP
> potentials had no constraints and maybe a better and more physical
> solution
> exists for sure in which the radius of the non-local part is not such
> a huge number..
>
> Moreover they're mostly unusable within cp2k.. Since we compute
> everything analytical (the core hamiltonian matrix) you have an
> incredible amount of periodic cells to take into account due to the
> large value of the non-local radii..
>
> Sooooooooo.. unless you don't wanna do a PERIODIC NONE calculation
> don't be surprised to see cp2k taking ages in computing the core-
> hamiltonian matrix.. Obviously there's an elegant alternative but
> it's not implemented..
> Call for volunteers?
>
> cheers,
> Teo
>
On 4/29/07, Juerg Hutter <hut... at pci.uzh.ch> wrote:
>
> Hi Axel
>
> the situation with the DCACP potentials is even more
> difficult. The Quickstep code assumes that the sparsity
> in the Kohn-Sham matrix and the basis set overlap matrix are
> identical. This is a good assumption as long as you
> don't have truly non-local potentials, e.g. Hartree-Fock
> exchange or the DCACP potentials.
> It's clear that this assumption has to be considered
> a bug and we will have to remove this assumption.
> However, this is not easy and some really basic
> routines will have to be changed.
>
> Currently, you can only use these potentials for small
> isolated systems (like in QM/MM) where there is no
> sparsity in the overlap matrix.
>
> To get a truly general and efficient code for the DCACPs
> one has to come up with a new clever idea.
>
> best regards
>
> Juerg
>
> ----------------------------------------------------------
> Juerg Hutter                   Phone : ++41 44 635 4491
> Physical Chemistry Institute   FAX   : ++41 44 635 6838
> University of Zurich           E-mail: hut... at pci.unizh.ch
> Winterthurerstrasse 190
> CH-8057 Zurich, Switzerland
> ----------------------------------------------------------
>
>
> On Sat, 28 Apr 2007, akohlmey wrote:
>
> >
> > hi all,
> >
> > we (the CMM) are currently investigating different ways to improve
> > the representation of dispersion forces in DFT and as one option we're
> > trying the density adapted goedecker potential stuff that anatole has
> > done while in lausanne.
> >
> > i've translated the 'library' into CP2k format and uploaded the
> > resulting file into this google group in the hope that somebody
> > else might find it useful (or at least will save the effort of
> > translating it while trying it out).
> >
> > cheers,
> >   axel.
> >
> >
> > >
> >
>
> >
>


-- 
=======================================================================
Axel Kohlmeyer   akoh... at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
  Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.