How to Parallelize DFPT Calculation?

Running Abinit Response calculations
,
1

Hi,

I am doing a calculation to obtain non-linear optical property calculation with optdriver=5. Now, I am stuck at the last stage of calculating third derivative of energy (based on this tutorial tnlo_6.abi DATASET 7).

My system has 56 atoms in it, with 232bands and 2x2x2 kpoint mesh, ecut 60 and pawe
cutdg 120.

I am doing the calculation in HPC with 6 nodes each of 40 cores and 256000 MB memory per node.

In this step the system will read bunch of energy and derivative files from previous datasets to calculate 105 number of perturbations. For each perturbation calculation it’s taking 5hrs+ time, which is very inconvenient for me. In this case I used -

OMP_NUM_THREADS=30 MKL_NUM_THREADS=30 mpirun -np 8

  1. In this context how should I distribute my job so that it becomes faster calculation? I can not afford 105 cycles with each cycle 5hrs+ time.
  2. Is there any parallelization tag specifically for DATSET 7 (3rd derivative of energy in tnlo_6.abi tutorial) that can be used?