[Wien] How to select the appropriate nband in hybrid functionals calculations

[Tran, Fabien]

Is the memory usage that you showed for the 3*3*1 or 4*4*1 supercell? Try to figure out if there is no memory problem for the 4*4*1. If you can run a MPI calculation, then do it using more than one node.


When you generate (with "run_kgenhf_lapw -redklist") a reduced k-mesh for a -redklist calculation, the reduced k-mesh is in case.klist_rfbz, but not in case.klist_ibz and case.klist_fbz which are still for the original k-mesh.


If you are interested only in the electronic structure (DOS), then I strongly recommend the use of the option "-diagfh" (see explanations in the user's guide) for a huge speed up of the calculation (supposing that this approximation is accurate enough in your case).


Using -ec 0.6 is nonsense.


________________________________
From: Wien <wien-bounces at zeus.theochem.tuwien.ac.at> on behalf of Yifan Ding <yfding0375 at foxmail.com>
Sent: Saturday, May 15, 2021 10:25 AM
To: wien
Subject: [Wien] How to select the appropriate nband in hybrid functionals calculations

Dear Prof. Tran and Prof. Abo,

Thank you very much for your kindly reply.

The 4*4*1 supercell (including 64 atoms) I want to calculate is really large. When I set the h-BN monolayer 3*3*1 supercell including 36 atoms, the calculation can be successfully completed.

At present, I am doing a calculation of 3*3*1 supercell, and the SCF is running HF in parallel mode. I ssh to the computing node and use the command 'top' to query the following information:

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND

15914 yurc      20   0 2681m 2.6g  916 R 100.0  4.1 953:27.35 hfc
  350 root      39  19     0    0    0 R 99.9  0.0  5128682h kipmi0
15789 yurc      20   0 2682m 2.6g  824 R 99.9  4.1 953:27.78 hfc
15820 yurc      20   0 2682m 2.6g  916 R 99.9  4.1 953:27.85 hfc
15882 yurc      20   0 2681m 2.6g  916 R 99.9  4.1 953:27.38 hfc
15851 yurc      20   0 2682m 2.6g  904 R 99.6  4.1 953:27.47 hfc

The above five "yurc" are my user names, and it seems that there is no problem with memory usage.  I didn't use MPI parallel calculations in 3*3*1 supercell or 4*4*1 supercell calculations, and all calculations are k-point parallel. For the 3*3*1 supercell, the total number of k points is 9, and there are 5 k points in the case.klist_ibz file.

Back to the 4*4*1 supercell calculations with "Error in Parallel HF", although I set nx = 1, ny = 1, nz = 1 for the reduced k-mesh, because the total k points are 10, the number of k points displayed by case.klist_ibz is 5. In 4*4*1 supercell calculations with "Error in Parallel HF", the number of nodes is 1 and the number of cores is 5. Setting -ec 0.6 is indeed very rough, because the value of nband will affect the energy range of DOS, but choosing a large nband will make the hybrid functional calculation very slow. I want to quickly calculate and judge the upper limit of the energy window of DOS by setting different nband, and finally choose the appropriate nband to get DOS of -20 eV ~ 20 eV. In order to complete SCF faster, I used -redklist. Therefore, I set "run_lapw -hf -redklist -ec 0.6 -p -i 999 > output.log" in 4*4*1 supercell calculations.

I am very grateful to the two professors for your enthusiastic help and help me find the questions and answers I didn't found in the mailing list.

I found out from the installation folder that the version information of Wien2k I used is WIEN2k_14.2 (Release 15/10/2014). Because I use the public supercomputer of the Institute of Physics, I found on the network that using the command "grep hf /var/log/syslog" seems to require root, and I don't have permission. At present, many people are using the WIEN2k_14.2 in our institute. If there is a chance later, I will try to upgrade it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20210515/075e966b/attachment-0001.htm>