Mobility calculation

[hangfan90]

Dear developers,

I tried to repeat the Silicon mobility calculation example.
There are two points I am confused.
1. How can I determine the fine k-path of electronic states (filkf)?
I am confused why the followeing k-path (kpt.txt) used in the mobility calculation example file(epw2.in). Which one is better for mobility calculation, with the parameter of homogeneous fine grid (nkf1) and specific fine k-path(filkf)?

2. When I tested the degaussw and degaussq parameters, I found the computation took a long time. For example, the computation time is 10 hours with 24 processors, when the fine grid are 40*40*40 k-point and 40*40*40 q-point. So if I increase the fine grid size, the computation time will increase to long time(at least several days) to get convergence I guess. Is there any recommendation for degaussw and degaussq parameters testing?

kpt.txt 105 crystal
0.00 0.00 0.00 0.00952380952
0.09 0.00 0.00 0.00952380952
0.00 0.09 0.00 0.00952380952
0.00 0.00 0.09 0.00952380952
0.09 0.09 0.00 0.00952380952
0.09 0.00 0.09 0.00952380952
0.00 0.09 0.09 0.00952380952
0.09 0.09 0.09 0.00952380952
-0.09 0.00 0.00 0.00952380952
0.00 -0.09 0.00 0.00952380952
0.00 0.00 -0.09 0.00952380952
-0.09 -0.09 0.00 0.00952380952
-0.09 0.00 -0.09 0.00952380952
0.00 -0.09 -0.09 0.00952380952
-0.09 -0.09 -0.09 0.00952380952
0.09 -0.09 0.00 0.00952380952
-0.09 0.09 0.00 0.00952380952
-0.09 0.00 0.09 0.00952380952
0.09 0.00 -0.09 0.00952380952
0.00 -0.09 0.09 0.00952380952
0.00 0.09 -0.09 0.00952380952
-0.09 0.09 0.09 0.00952380952
0.09 -0.09 0.09 0.00952380952
0.09 0.09 -0.09 0.00952380952
-0.09 -0.09 0.09 0.00952380952
-0.09 0.09 -0.09 0.00952380952
0.09 -0.09 -0.09 0.00952380952
0.02 0.00 0.00 0.00952380952
0.00 0.02 0.00 0.00952380952
0.00 0.00 0.02 0.00952380952
0.02 0.02 0.00 0.00952380952
0.02 0.00 0.02 0.00952380952
0.00 0.02 0.02 0.00952380952
0.02 0.02 0.02 0.00952380952
-0.02 0.00 0.00 0.00952380952
0.00 -0.02 0.00 0.00952380952
0.00 0.00 -0.02 0.00952380952
-0.02 -0.02 0.00 0.00952380952
-0.02 0.00 -0.02 0.00952380952
0.00 -0.02 -0.02 0.00952380952
-0.02 -0.02 -0.02 0.00952380952
0.02 -0.02 0.00 0.00952380952
-0.02 0.02 0.00 0.00952380952
-0.02 0.00 0.02 0.00952380952
0.02 0.00 -0.02 0.00952380952
0.00 -0.02 0.02 0.00952380952
0.00 0.02 -0.02 0.00952380952
-0.02 0.02 0.02 0.00952380952
0.02 -0.02 0.02 0.00952380952
0.02 0.02 -0.02 0.00952380952
-0.02 -0.02 0.02 0.00952380952
-0.02 0.02 -0.02 0.00952380952
0.02 -0.02 -0.02 0.00952380952
0.03 0.00 0.00 0.00952380952
0.00 0.03 0.00 0.00952380952
0.00 0.00 0.03 0.00952380952
0.03 0.03 0.00 0.00952380952
0.03 0.00 0.03 0.00952380952
0.00 0.03 0.03 0.00952380952
0.03 0.03 0.03 0.00952380952
-0.03 0.00 0.00 0.00952380952
0.00 -0.03 0.00 0.00952380952
0.00 0.00 -0.03 0.00952380952
-0.03 -0.03 0.00 0.00952380952
-0.03 0.00 -0.03 0.00952380952
0.00 -0.03 -0.03 0.00952380952
-0.03 -0.03 -0.03 0.00952380952
0.03 -0.03 0.00 0.00952380952
-0.03 0.03 0.00 0.00952380952
-0.03 0.00 0.03 0.00952380952
0.03 0.00 -0.03 0.00952380952
0.00 -0.03 0.03 0.00952380952
0.00 0.03 -0.03 0.00952380952
-0.03 0.03 0.03 0.00952380952
0.03 -0.03 0.03 0.00952380952
0.03 0.03 -0.03 0.00952380952
-0.03 -0.03 0.03 0.00952380952
-0.03 0.03 -0.03 0.00952380952
0.03 -0.03 -0.03 0.00952380952
0.47 0.00 0.00 0.00952380952
0.00 0.47 0.00 0.00952380952
0.00 0.00 0.47 0.00952380952
0.47 0.47 0.00 0.00952380952
0.47 0.00 0.47 0.00952380952
0.00 0.47 0.47 0.00952380952
0.47 0.47 0.47 0.00952380952
-0.47 0.00 0.00 0.00952380952
0.00 -0.47 0.00 0.00952380952
0.00 0.00 -0.47 0.00952380952
-0.47 -0.47 0.00 0.00952380952
-0.47 0.00 -0.47 0.00952380952
0.00 -0.47 -0.47 0.00952380952
-0.47 -0.47 -0.47 0.00952380952
0.47 -0.47 0.00 0.00952380952
-0.47 0.47 0.00 0.00952380952
-0.47 0.00 0.47 0.00952380952
0.47 0.00 -0.47 0.00952380952
0.00 -0.47 0.47 0.00952380952
0.00 0.47 -0.47 0.00952380952
-0.47 0.47 0.47 0.00952380952
0.47 -0.47 0.47 0.00952380952
0.47 0.47 -0.47 0.00952380952
-0.47 -0.47 0.47 0.00952380952
-0.47 0.47 -0.47 0.00952380952
0.47 -0.47 -0.47 0.00952380952

Best regards,
Hang

[sponce]

Dear Hang,

1) The mobility example from the school was design to showcase how the code works in a very fast way.
The file kpt.txt contains some points that I know will contribute to mobility. Notice that their weight is incorrect (just 1/nb_points). Therefore one should not expect to get accurate mobility with that calculation.

You can either use homogeneous fine grids or provide random grids (but then be sure to compute the weights correctly, typically using Voronoi volume for the weight). Random grids converge faster. You need to perform two integrations (k and q).

2) degaussw has no impact on speed of the calculation and degaussq is not used for mobility.
Notice that homogeneous grids of 100x100x100 for k-points and q-points are typical.
Computing mobilities is indeed very cpu intensive.

Best wishes,
Samuel