Question about parallelization of get_free_energy_hessian in SSCHA

[anyi-mxh]

Dear SSCHA developers,

I am currently using SSCHA to compute the free-energy Hessian for a large supercell (NAT_SC ≈ 440, N_MODE ≈ 1320, with ~30000 configurations). During the calculation I noticed the following behavior:

During the Hessian computation (get_free_energy_hessian), in particular the EMAT / V3 / V4 construction stage (e.g. when the output shows === DEBUG GET_EMAT ===), the calculation also appears to run on a single core only. This step is very slow. It takes 6 days.

I would like to ask:

In the EMAT / V3 / V4 construction stage, how to achieve multi-core parallelism?

Best regards,
mxh

[mesonepigreco]

Hi,
Unfortunately, that is not available in the current version of the code.
The solution is to use the time-dependent sscha, which is parallelized, and compute the static response function from which you can obtain the free energy Hessian for a specific phonon mode.

Here the general documentation for the TDSCHA code:

Here is the template for the input file for computing the dynamical response function on a specific mode: run_local.py

You can then use the utility script plot_hessian_convergence.py to extract the frequency of the free energy hessian