The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

Dear Michael,

On 20 Feb 2007, at 22:43, Haverty, Michael G wrote:

Hi all,

I’m trying to validate Ab-init for conductivity calculations, but the conducti (tests/v3/t78.in and tests/v3/t79.in) results aren’t converging to one conductivity value.  From the CRC handbook the experimental Al conductivity at 300K is 4.13 X 10^5 (Ohm-cm)^5.  

I have the impression that you quote a value for the conductivity that corresponds to static electric fields, hence,

limited by electron-phonon interaction. The conducti utility aims at computing the frequency-dependent

conductivity, for frequencies in the optical range, and for possibly non-zero temperatures, in the non-interacting electron 

approximation (Kubo-Greenwood formalism). It does not include the effects of electron-phonon interaction.

The electron-phonon limited conductivity, for static electric fields, has been recently coded in ABINIT by JP Crocombette,

with the help of M Verstraete and F Jollet, and is available

in v5.3 , not yet in production. You might have a look on the release notes of that version, see point B3 

http://www.abinit.org/Infos_v5.3/release_notes/release_notes.html

Unfortunately, no automatic tests are provided yet.

Good continuation,

Xavier

I’m finding that if I change the sampling of omega when using conducti, my value for Al conductivity (near omega=0) changes by orders of magnitude (from 4.13 X 10^6 to 4.15 X 10^4 for omega sampling from 0.00001 to 0.001).  I’m using a 1-atom supercell, with 20 bands, 40 Hartree energy cutoff,  and 12x12x12 k-point sampling, but have seen similar variability of the conductivity for less dense k-point samplings as well.

From the standard output of the conducti code, I don’t understand what the line “sumrule” means.  For example, my output of this line is:

            sumrule           =   1.24920   2.93610

From the src/main/conducti.f90 file, I see that this output is coming from the following line, but I don’t understand what np_sum and socc signify:

write(6,'(a,3f10.5)')' sumrule           =',np_sum/socc/3.0d0,socc

            To ensure that sum rule evaluation is as close to 1 as possible, which of these 2 results “1.24920” or “2.93610” should we monitor to ensure it is as close to 1 as possible?  Also, what setting of the t78.in (t78.in is the abinit run for self-consistency and DDB, WFK generation) will affect the value of the sum rule?

Can anyone recommend what I should do to get a well-converged conductivity to compare against the experimental data?

                                                Mike