Questions about the computing time of mobility

[Qiaolin]

Dear Experts:


I try to calculate the mobility of Silicon, but fail to get converged results. And I have tried to increase the grids of k and q, but it takes a lot of time. For

instance, I cannot get the final results using SERTA approach with k-50x50x50, q-50x50x50? although it takes more than 5 days (Server:NSCC, Nodes:16, ppn:192).

I am sorry that I have gotten few questions and I really appreciate your answers.

1.Need I continue to increase Nodes?

2.May I know the computing resources you use, such as server, computing nodes, the number of processor cores and so on?


Best wishes,
QiaoLin Yang,
Institute of Semiconductors,
University of Chinese Academy of Sciences

[sponce]

Dear QiaoLin Yang,

Indeed converging mobility is quite time-consuming. This is because you need to integrate (and therefore converge) both the fine k and q-point grids.

You can speedup the convergence by using random grids with more points close to the band edge (you need to compute their weight correctly using their Voronoi volume for example) for k-point and random grid for q-points.
See for example what I did for silicon in PRB 97,121201 (2018). In particular, see the supplemental.

Note that I'm now working on improving the homogeneous grid implementation to make it significantly faster (using symmetries and the fact that the grids are commensurate).

Typical computational resources is 500-1000 cores for 1-3 days.

Best wishes,
Samuel