The ABINIT Users Mailing List ( CLOSED )

[hichem bouderba <hichem.bouderba@gmail.com>]

very instructive, thanks.

2009/4/8 Josef Zwanziger <jzwanzig@gmail.com>

yes, you can specify everything from kptrlen rather than ngkpt, if you
want to. thus your input could have

kptopt 1
kptrlen 50

and the system will find a reasonable kptrlatt (combination of ngkpt,
nshiftk and shiftk) for your space group. Be careful, though, because
some features, particularly the phonon analysis in anaddb, are not
compatible with arbitrary kptrlatt's, but insist on using a simpler
ngkpt + shift grid.Therefore the ngkpt + shiftk variant is more robust
though takes more work (sometimes) to generate. If you make a run with
prtkpt 1 it will give you lots of choices of grids to use as a
function of kptrlen, for your system.

On Wed, Apr 8, 2009 at 4:20 AM, hichem bouderba

<hichem.bouderba@gmail.com> wrote:
> thanks matthieu,
> correct me please:
> so, one can check convergence with respect to kptrlen instead of ngkpt
> because, actually, the accuracy is governed by the former. and, I think, the
> following may happen:
> if we define a grid 1 with :
> ngkpt n n n
> and an other grid 2 with:
> ngkpt n+1 n+1 n+1
> it's not trivial that the second one will give a bigger kptrlen ?.
> and suppose that they are shifted in the same way, does it imply that
> kptrlen 2 is bigger ?
> thanks
> 2009/4/7 matthieu verstraete <matthieu.jean.verstraete@gmail.com>
>>
>> "erroneous contribution" is erroneous. The contributions are what they
>> are, but the values we have are on the k-point grid, and that is where
>> the error can be estimated: there are in some sense "perfect" values
>> for the quadrature you choose, whose sum over discrete k would give
>> the exact continuous integral, but obviously we don't know what these
>> are, and the contributions we actually sum are in this sense
>> "erroneous".
>>
>> The finite grid spacing gives an error, and, broadly speaking, the
>> smaller the k grid spacing the more precise your integral. And the
>> smallest grid spacing in k corresponds to the largest real space
>> vector.
>>
>> Matthieu
>>
>> On Tue, Apr 7, 2009 at 9:41 AM,  <hichem.bouderba@gmail.com> wrote:
>> > hello everybody,
>> > in the input file documentation in kptrlen entry we can read the
>> > following:
>> > " ... One can relate the error made by replacing the continuous integral
>> > by a
>> > sum over k point lattice to the Fourier transform of the periodic
>> > quantity.
>> > Erroneous contributions will appear only for the vectors in real space
>> > that
>> > belong to the reciprocal of the k point lattice, except the origin.
>> > Moreover,
>> > the expected size of these contributions usually decreases exponentially
>> > with
>> > the distance. So, the length of the smallest of these real space vectors
>> > is a
>> > measure of the accuracy of the k point grid."
>> >
>> > can somebody be more explicit about the relation between integrals and
>> > "erroneous contributions" ?
>> >
>>
>>
>>
>> --
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> Dr. Matthieu Verstraete
>>
>> European Theoretical Spectroscopy Facility (ETSF)
>> Dpto. Fisica de Materiales,
>> U. del Pais Vasco,
>> Centro Joxe Mari Korta, Av. de Tolosa, 72,   Phone: +34-943018393
>> E-20018 Donostia-San Sebastian, Spain        Fax  : +34-943018390
>>
>> Mail : matthieu.jean.verstraete@gmail.com
>> http://www-users.york.ac.uk/~mjv500
>>
>
>

--
Josef W. Zwanziger
Professor of Chemistry
Canada Research Chair in NMR Studies of Materials
Director, Atlantic Region Magnetic Resonance Centre
Dalhousie University
Halifax, NS B3H 4J3 Canada
tel: +1 (902) 494-1960
fax: +1 (902) 494-1867
web: http://jwz.chem.dal.ca
jzwanzig@gmail.com, jzwanzig@dal.ca