search

shankar1729 / jdftx

JDFTx: software for joint density functional theory
http://jdftx.org
84 stars 54 forks source link

Question about SOC and parallel capability #357

Open ShiminZhang21 opened 1 month ago

ShiminZhang21 commented 1 month ago

Hi Shanker,

I am a beginner to the JDFTx code. I am trying to run the scf and phonon calculation of a defect supercell with local magnetization and SOC.

I have two question regarding the parallel capability of JDFTx and SOC setup:

  1. I am running a spin-polarized defect system (with local magnetization of 1 on defect site). For this kind of system i should use "spintype vector-spin" and setup the initial magnetization right?

  2. Another question about the parallel in JDFTx. It seems like JDFTx has only k point parallel. If that's the case, a supercell calculation require only gamma point would run in a sequential way regardless the assigned cores (-n ) ?

Could you have a look on my input to see if that looks reasonable to you? JDFTx_WS2_defect.zip

Best Regard, Shimin

ShiminZhang21 commented 1 month ago

Another question: i used sg15 ONCV pseudopotential, and at the wavefunction initialization step, the output prompt: ---------- Allocating electronic variables ---------- Initializing wave functions: (No orbitals for LCAO) bandwidth-limited random numbers

Is it normal that no LCAO can be constructed with sg15 PP (FR) ?

shankar1729 commented 1 month ago

Hi Shimin,

  1. Yes, you should use vector-spin and set an initial magnetization at the defect. You can skip setting the zero moments in your input and just set it for the single atom that is non-zero. I don't see any obvious issues with the input you attached.
  2. Yes, you can only use one MPI process with a single k-point. The code is still parallelized over threads and GPU cores, so you can use one full compute node for CPUs, and one one full GPU. (It is not "serial".)
  3. The SG15 pseudopotentials distributed by their website does not have atomic orbitals, but the ones distributed with JDFTx do. We had contributed atomic orbital capability to the ONCV code to make this possible. You should be able to download the ONCV input file from the SG15 website and regenerate the pseudopotential with atomic orbitals. Alternately, you can also use FR pseudopotentials from PseudoDOJO which do have atomic orbitals by default.

Best, Shankar