[Pw_forum] IFC's

[Katalin Gaal-Nagy]

Dear all,

the message I send just half an hour ago is held because of the size ... 
However, I resend it without the attachments. If anyone is interested in 
the files with the routines, please tell me!

Best wishes,
Katalin

---------- Forwarded message ----------
Date: Mon, 3 Sep 2007 10:01:59 +0200 (CEST)
From: Katalin Gaal-Nagy <katalin.gaal-nagy at physik.uni-regensburg.de>
To: PWSCF Forum <pw_forum at pwscf.org>
Subject: Re: [Pw_forum] IFC's

Dear Ezad, dear Paolo,

well ... you can do it, with a trick.
If I understood correctly, you have the q points on a 3x3x3 grid and the q 
points on a 4x4x4 grid. A grid containing both sets of q points is the 12x12x12 
grid.
You can do the following: You calculate the force constants of the 4x4x4 grid 
and do a back Fourier transform and calculate the dynamical matrices on a 
12x12x12 grid. In this set of dynamical matrices, the ones corresponding to the 
4x4x4 grid are identical (up to numerical noise, which is little). Then you 
take these  12x12x12 interpolated dynamical matrices, throw away the 
interpolated ones corresponding to the 3x3x3 mesh and put instead the 3x3x3 
ones you have calculated with ph.x. You again have a full set of dynamical 
matrices and you can calculate from these the force constants, which correspond 
to a mixing of a 4x4x4 grid and a 3x3x3 grid. Note, they DON'T correspond to a 
12x12x12 grid, since there are still interpolated dynamical matrices left. With 
the resulting dynamical matrices also no physical meaning like range of 
corresponding forces can be connected any more.

The idea of this scheme can be found under condmat arXiv:0707.0384.

For the use of this scheme I modified the postprocessing routines matdyn.f90 of 
ESPRESSO (I am not sure about the version, if should have been espresso-3.1). I 
don't remember, if I was putting it in a nice, commented shape (for sure tha 
Makefile not ... ). The main modification was to tell the code to write out the 
dynamical matrices for all q in star(q) in a way, that q2r.x can read it again. 
The modified routines and a makefile are attached. I would apprechiate if you 
could test, if the compilation is working.

However, you have to use the routines q2r.x and matdyn.x ...
The matdyn.x has now an additional flag and the input reads as

  &input
     asr='simple',  amass(1)=28.086,
     flfrc='si.q888.force',
     flfrq='si.q16.16.16.freq.interpol',
     dynmatout=.true.
  /
349

.... list of k points created with kpoints.x ...

It creates (in this case) 349 files containing the interpolated dynamical 
matrices dynint.$number where $number is the number of the q point given in the 
list ...

Paolo, in case you want to include it to the ESPRESSO paackage and there are 
problems, please let me know.


Good luck,
Katalin






On Sun, 2 Sep 2007, Paolo Giannozzi wrote:

> 
> On Sep 2, 2007, at 10:25 , Ezad Shojaee wrote:
> 
>> i want to know that are we able to use the 'IFC's calculated with
>> two grids' together? for example we calculate them by a 3*3*3 grid,
>> and then 4*4*4 , then use all of them in calculating C(q) for an
>> arbitrary point in the Brillouin zone?
> 
> no, it is not: you can use only one specific grid. In principle one
> could
> start with a coarser grid and then add the missing points needed to have
> a denser grid, I think, but it would be a mess
> 
> Paolo
> ---
> Paolo Giannozzi, Democritos and University of Udine, Italy
> 
> 
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum