[Pw_forum] Re: Pw_forum digest, Vol 1 #1325 - 4 msgs

Malgorzata Wierzbowska wierzbom at ts.infn.it
Mon Jan 29 18:05:24 CET 2007


On Mon, 29 Jan 2007, Srijan Kumar Saha wrote:

>
>    >>>>>>>And always k_fit/(k_phonon + q_phonon) should be the integer
>   number.
>
>    So if I use 48 48 1 for k_fit
>                     24 24 1 for k_phonon and
>
>                     2/3   0.0  0.0  for a special single q point (it's not
> automatic generated value)
>   Then I think  it  will be  fine , Right???

     It should be right, I mean with respect to the error we spoke about.
     I guess that the z-value of points in (48 48 1) and (24 24 1)
     will be 0.0.

     Another problem is whether this grid is dense enough to get
     accurate results.... This you should test. I would put for
     scf-fit something like 64 or 72 if you have enough time for
     calculations. At the end scf-fit is a single scf, we do it only once.
     Later you can calculate more phonons with the same scf-fit files
     in the tmp-directory. But all your chosen q-points should fit
     to above k-fit mesh.

     Wait a second... what is the point to use la2F if you have only
     one q-point? The whole project was ment to use q2r and matdyn
     in order to get many q-points from a few q-point calculations.
     So, once you make some tests on a single q-point with old
     style calculations for phonons, and you are sure about
     your phonon k-mesh, you should choose all phonons and calculate
     them in one run with the la2F.
     Maybe your phonon grid could be then 3 3 1 or 6 6 1....

     Gosia


>
>
> On 1/29/07, Malgorzata Wierzbowska <wierzbom at ts.infn.it> wrote:
>> 
>> On Mon, 29 Jan 2007, Srijan Kumar Saha wrote:
>> 
>> 
>> >  Respected Users,
>> >
>> >  Would you like to tell me what automatic  (like 48 48 1)
>> >  the dense k-point grid in the first scf calculations
>> >  (i.e. scf-fit). I should  use
>> >  to get  special q point
>> >>
>> >> 0.6666667   0.0   0.0
>> >> as a  the subsets of the scf-fit ????
>> > Is there any definite rule except applying  a trial and
>> > error method like  4 4 1, 6 6 1, 9 9 1, 38 38 1 etc. etc.  ????
>> > Your kind help will be greatly appreciated.
>> > Regards,
>> > Amit
>> >
>> 
>>     If q=(2/3,0.0,0.0) your k-mesh for phonons and k-fit mesh
>>     should be modulo 3. So  (4 4 1) and (38 38 1) are for sure bad.
>>     Meshes (6 6 1) and (9 9 1) are too small, although they would
>>     not give the error if your rare k-mesh (k-mesh for phonons) were
>>     modulo 3.
>>     Actually, for this kind of calculations you should ask for
>>     3 meshes: k_fit, k_phonon, q_phonon.
>>     And always k_fit/(k_phonon + q_phonon) should be the integer
>>     number.
>> 
>>     Gosia
>> 
>> 
>> 
>



More information about the Pw_forum mailing list