The ABINIT Users Mailing List ( CLOSED )

["Fernando D. Vila" <fer@tiziano.phys.washington.edu>]

On Wed, 20 Jul 2005 verstraete@pcpm.ucl.ac.be wrote:

> Hello Fer,
>
> Is your DW calculator GPL? Just asking...

Not yet, it's history is a bit convoluted and it's kind of far from useful 
yet... :-)

> I presume you've done the tutorials on Al and so on. There are additional 
> parameters (occopt, tsmear) when you do a metal, and you need a lot more 
> kpoints.

Yes, I did.. I think I have figured out things a bit. My calculations for 
Cu are giving some rasonable numbers for the phonon frequencies (see 
below).

> 1) You definitely do not need to use a supercell. Running anaddb with a mesh 
> of qpoints instead of just gamma (see tutorial 5) will allow you to 
> interpolate the ifc and dynamical matrices for atoms far from the central 
> one. You typically go to 20th neighbor, sometimes more.

OK, this sound like what I want. I have ben trying to find some 
documentation for this without avail. Here is what I would like to get 
from abinit, maybe somebody here would be kind enough to share files, 
experience, etc:

Let's consider the case of Ge. If we think of Ge1 as a central atom, and 
Ge2 as one of its neighbours, I need the following values:

   d2E
---------
dR1a dR2b

where a,b=(x,y,z)

From your comments now I understand that I can get them without having to 
resort to a supercell that includes both Ge1 and Ge2. But after quite a 
bit of experimentation (and trying to find out on the web how to do this) 
I'm at a loss. Here are some specific questions:

1) Let's say I want to get the values mentioned above for the first and 
second neighbours, what kind of a q-grid do I need?

2) Once I have the different DDB files for each q point, I know how to 
merge them. What I don't know is what kind of input files I need for 
anaddb to get the derivatives I need. Any help with this?

> 2) You'll probably need a 10x10x10 kpoint grid (fcc shifted see tutorials), 
> the cutoff should be 20-40 hartree. The rest you'll have to converge. I'd use 
> GGA, but see "Narasimhan S, de Gironcoli S Ab initio calculation of the 
> thermal properties of Cu: Performance of the LDA and GGA PHYSICAL REVIEW B 65 
> (6): Art. No. 064302 FEB 1 2002" for details.

Thanks for the reference. I think I can do Cu reasonably well now (kind of 
blindly, but I think I'll understand a bit more in the future.

> 3) wfoptalg is still quite experimental. Please try it and report any 
> success/failure here.

I tried the wfoptalg=1 option (4 doesn't work even if it's listed in the 
manual). The calculation does run a bit faster (depending on the system), 
but the values for the derivatives are different. So I decided to just 
stick to the standard.

Anyway, I would be very greatful if anybody could share. I'm kind of lost 
regarding the use of anaddb.

Cheers, Fer.

> Matthieu
>
> On Wed, 20 Jul 2005, fdv@u.washington.edu wrote:
>
> > Hello everybody,
> >
> >  Our group has developed a method to calculate Deby-Waller factors for use 
> > in XANES and EXAFS simulations. The only input required are the second 
> > derivatives of the energy. So far we have used force constants from force 
> > fields, but we are interested in obtaining the derivatives ab initio. We 
> > want to have a fast, quasi-black box method to obtain the second 
> > derivatives for a varied number of solids, to within 10-15% accuracy.
> >
> >  Following the tutorials and with a little experience from other plane wave 
> > codes, I was able to set up an abinit calculation for Ge crystal and 
> > obtained a dynamical matrix which results in very good DW factors (compared 
> > to experiment). Now I'm trying to get values for Cu but I'm running into 
> > some problems.
> >
> >  Here are some specific questions that I have:
> >
> > 1) I need to get the second derivatives of an atom and its first and second 
> > neighbours at least. Is there any way to do this besides making a large 
> > supercell? (I don't think there is, but I'm asking just in case.. :-)
> >
> > 2) Does anybody have an input file for copper that is known to give 
> > "reasonable" results (i.e. less than 10% error in the structural constants, 
> > for example). I'm specially interested in the size of the cell, the number 
> > of k-points, the energy cutoff, functional and pseudopotential used and in 
> > the type and amount of smearing used. Most of the plane-wave calculations 
> > that I've done in the past have dealt with insulators. I need an initial 
> > push with metals. (Feel free to suggest any kind of reading that you might 
> > think would be of interest).
> >
> > 3) I have a cluster and have a parallel version of the code compiled. It 
> > scales quite well when the nuber of k-points is large and the size of the 
> > cell is small. When I try a large supercell and a reduced number of points, 
> > some of the processors are not used. I am runnin the 4.5.3 version and when 
> > I tried the wfoptalg=4 option but the code complained that only 0-3 are 
> > allowed. How do I increase the performace for a large cell with fewer 
> > k-points?
> >
> > Well, I think that's all for now and sorry for the long post.
> >
> > Best regards, Fer.
>
> --
> ================================================================
> Matthieu Verstraete             mailto:verstraete@pcpm.ucl.ac.be
> PCPM, Boltzmann, pl. Croix du Sud, 1        tel: +32 10 47 33 59
> B-1348 Louvain-la-Neuve  Belgium            fax: +32 10 47 34 52


Ubi dubium ibi libertas.
*******************************************************************************
Fernando D. Vila                Voice    (206)543-9697
Department of Physics           Fax      (206)685-0635
University of Washington        E-mail   fdv@u.washington.edu
Seattle, WA 98195, USA          WWW      http://faculty.washington.edu/fdv
*******************************************************************************