The ABINIT Users Mailing List ( CLOSED )

[Peter Zhilyaev <peterzhilyaev@gmail.com>]

Dear Vanina,

Thanks for your help!

Sincerely yours,
Peter Zhilyaev

2010/2/15 RECOULES Vanina <vanina.recoules@cea.fr>

Dear Peter,

To compute electrical conductivity you need to run a very  detailed electronic calculation. Electrical conductivity is very sensitive to the number of k-points, the number of bands (your last band must be empty) and to the number of atoms in your simulation cell (for liquid phase). Please have a look on ref PRB80(2009)214105 and PRB78(2005)125306. The authors do not use ABINIT but the implementation of electrical conductivity is similar. For the liquid you might be able to retrieve the results in ref PRB 72(2005)104202. These calculations was done with ABINIT.
In you use the Abinit test t78.in as a starting point, you must change nsym to 0 before starting to increase the number of kpoints or the number of bands. The calculation using PAW potential are really faster  (Test v5/t09.in and t10.in). Here again you must change nsym, nline and nstep to get a good convergency. You have to change the description of fcc lattice too.

Regards,
Vanina Recoules


Peter Zhilyaev a écrit :

Dear All,

I'm trying to calculate conductivity with the help of ABINIT. I launched testing calculations for aluminum (Test v3/t79.in <http://t79.in>), with increased number of bands (20) and k-points (18x18x18 Monkhorst&Pack) in the input file. But, even with such increasing, the results I've obtained are in very bad agreement with experimental one and other ab-initio calculations (Knider et al./JPCM2007, they use Siesta code). Maybe I have to increase the number of bands and k-points more, the test example isn't good for calculations or it is necessary to use another pseudopotential (I use pspnc) .

Can you give me a tip about this problem?

Thank you all in advance!

Sincerely yours,

Peter Zhilyaev