Calculation cost and convergence of OHE

[AKB-OU]

Dear Dr. Q. S. Wu and WT users,

Hi. I am just working on a calculation of angular dependence of magnetoresistance with WT. I would like to ask about the typical calculation cost and convergence of the OHE calculation.

I have already tried the tutorial of Cu and recognized that the typical calculation cost for a single magnetoresistance curve is 500-1700 sec using 80 CPU cores and 81x81x81 kmesh (calculation time seems to depend on the magnetic field direction).

Now, I am working on my target material and noticed that it requires more computational cost than Cu. I think that possible reasons are lower crystal symmetry, more Wannier basis, and more complicated FS etc. than Cu, but I recognized that WT takes extremely long time for the calculation of a specific band compared to the other bands. I show the excerpted part of output:

      Version 2.5.1
      You are using    80 CPU cores

      >> Wannier centers by default, in unit of reciprocal lattice vector
      iwann        R1        R2        R3
      1  0.750000  0.750000  0.760309
      2  0.750000  0.750000  0.760309
      3  0.750000  0.750000  0.760309
      4  0.750000  0.750000  0.760309
      5  0.750000  0.750000  0.760309
      6  0.250000  0.250000  0.239691
      7  0.250000  0.250000  0.239691
      8  0.250000  0.250000  0.239691
      9  0.250000  0.250000  0.239691
      10  0.250000  0.250000  0.239691
      11  0.750000  0.250000  0.000000
      12  0.750000  0.250000  0.000000
      13  0.750000  0.250000  0.000000
      14  0.250000  0.750000  0.000000
      15  0.250000  0.750000  0.000000
      16  0.250000  0.750000  0.000000
      17  0.250000  0.250000  0.669639
      18  0.250000  0.250000  0.669639
      19  0.250000  0.250000  0.669639
      20  0.750000  0.750000  0.330361
      21  0.750000  0.750000  0.330361
      22  0.750000  0.750000  0.330361

      >>>calculation parameters :
      E_arc :          0.00000
      Eta_arc :          0.00100
      Gap_threshold         0.01000
      OmegaMin :          0.00000
      OmegaMax :          0.00000
      OmegaNum :      1
      OmegaNum_unfold :    200
      Nk1 :     51
      Nk2 :     51
      Nk3 :     25
      NP number of principle layers  :      2
      Tmin(Kelvin)  :         50.00000
      Tmax(Kelvin)  :         50.00000
      NumT  :      1
      NBTau  :     50
      Beta  :        100.00000
      Nslice_BTau_Max  :   5000
      BTauMax(Tesla.ps)        50.00000
      Rcut    999999.00000
      Magp               1
      Magp_min               0
      Magp_max               1
      wcc_calc_tol         0.08000
      wcc_neighbour_tol         0.30000
      NumLCZVecs   400
      NumSelectedEigenVals     0
      NumRandomConfs:     1
      Projection weight mode:NORMAL

      >> We finished the symmetry reducing of the kpoints in the kcube
      Number of kpoints mesh is      65025
      Number of symmetry reduced kpoints is      32513

      >> Start to calculate ordinary hall effects

      KBT array in the calculation in unit of Kelvin
      50.00

      NSlice_Btau :        4998
      ik/NK                99   /               407time left     0.008s
      ik/NK               199   /               407time left     0.005s
      ik/NK               299   /               407time left     0.003s
      ik/NK               399   /               407time left     0.000s

      There are      32513 k points generated by the input files
      However there are only        319 k points contribute to the conductance calculations for band          9
      However there are only        692 k points contribute to the conductance calculations for band         10
      However there are only       1434 k points contribute to the conductance calculations for band         11
      However there are only         92 k points contribute to the conductance calculations for band         12

      >> Time cost for calculate MR at ib=     1is            78.90 s
      >> Time cost for calculate MR at ib=     2is           175.67 s
      >> Time cost for calculate MR at ib=     3is          4453.76 s
      >> Time cost for calculate MR at ib=     4is            30.44 s

      Time cost for Boltz_OHE_calc is about         4739.810 s
      End of OHE calculation

      Time cost for whole program is about         4740.275 s

While bands 9, 10, and 12 are very cheap to calculate, band 11 costs 94% of the whole time! Is this reasonable behavior? or what is a possible factor that makes the calculation slow down?

In addition, could you please tell me just for my reference typical computational condition (number of CPU cores etc.), kmesh, and computational time for the cases of Bi and WP2 shown in PRB paper [PRB 99, 035142 (2019)]? I also want to know how to judge the convergence and the typical value of kmesh to obtain a well-converged output.

Best regards, AKB-OU

[AKB-OU]

I also add a comment that the speed of MR loop calculation is as fast as that of other bands, but WT spents most of the time between the last ">> time cost for this loop is xx s" massage and next "Time cost for each k point at iband 3".