The ABINIT Users Mailing List ( CLOSED )

[<6671011@163.com>]

Dear Prof. Xavier,

     Thanks for your reply. Yes, I used "gwcomp 1" in both the screening and
self-energy calculations. And the "nband" has been reduced to 160, a rough
estimation according to the examples in your paper PRB 78, 085125 (2008).
 
     But the KSS calculation still takes too much time: I used 80 cores to
calculate a 12-atom (carbon and hydrogen) system, and it took nearly 10 days.
Is this normal? 

    (Usually in an IR/Raman calculation, a whole process for such system can 
be
finished in 3 days with smaller CPU number.)

     I can split the later calculations (SCR and self-energy) into small 
parts,
but the KSS part is a really headache.

Sincerely,
Guangfu Luo


-----------------------------------------------
Dear Guangfu Luo,

Do you use gwcomp=1 for your GW calculations ?
If not, please, do so.

This will allow you to get a much faster convergence with respect to the 
number
of KSS bands, so that the number of bands to be computed for a KSS file might
be reduced by a factor of 5 or so, and the KSS calculation is then reasonably
fast.

Xavier


On 06 Jan 2010, at 04:35, 6671011@163.com wrote:


Dear Friends,

I did not expect that the KSS calculation could be so time-consuming— even 
much longer than that of the IR/Raman calculation.

Since it is only a GS process, can the KSS calculation be divided into parts 
according to the k-points? If so, then they can be calculated at the same 
time,

and, finally, their results be merged into one single file—just like we deal 
with the ddb and scr.

Possibly, there are some parallelism strategies in the KSS calculation, but 
the

above one could be more efficient.

Sincerely,
Gunagfu Luo