[Pw_forum] the efficiency problem of transmission calculation by pwcond.x

alexandr smogunov smogunov at sissa.it
Wed Jan 7 09:38:38 CET 2009


Dear Wang
Could you please provide PW input files for the scattering and lead regions
and also PWCOND output file for one k point and one energy to see what is 
going on? Then, how many CPU do you use? 
Regards,
Alexander

On Wednesday 07 January 2009 07:37, Wang Qinjing wrote:
> Hi all
>
> I am doing transmission calculation by pwcond.x. I have 2 question about
> the efficiency of it.
>
> (1) The most severe problem was that I felt the transmission calculation by
> pwcond.x was too slow. As an example, the system contains a 6-atom lead and
> about 100-atom scattering region. T he kx and ky was 8 and 1 respectively.
> The input file for pwcond.x is as follows:
>
> ----------------
> &inputcond
>     outdir='./outdir',
>     prefixl='lead'
>     prefixs='cen',
>     tran_file='trans.0'
>     ikind = 1,
>     energy0=2.0d0,
>     denergy=-0.2d0
>     ewind=4.d0,
>     epsproj=1.d-7,
>     nz1 = 10,
>  /
>    5
>    0.0000000   0.0000000   0.2500000
>    0.1250000   0.0000000   0.5000000
>    0.2500000   0.0000000   0.5000000
>    0.3750000   0.0000000   0.5000000
>   -0.5000000   0.0000000   0.2500000
>    21
> ------------------
>
> It took about 1 hour to get a transmission value for a certain energy on a
> 8-node parallel cluster with 1 G memory, 2.5 GHz CPU and infiniband. Is it
> a reasonable one?
>
> (2) I notice that the time needed for the transmission is linear
> proportional to the number of energy points. That means if the time needed
> for a single energy point is 1 hour, it will take 20 hours for 20 energy
> points. This is different with some other scientific softwares. Is it
> natural for pwcond.x?
>
> Thank you in advance
>
> Best wishes
>
> Wang
>
>
> ------------------------------------
> 王秦镜,复旦大学物理系研究生
> Wang Qinjing, Physics Department,
> Fudan University



More information about the users mailing list