Code to run High Throughput IPA calculations on the Alexandria database [1]. The workflow was created as part of the article Deep learning of spectra: Predicting the dielectric function of semiconductors [2]. The final result of the the calculations is available at figshare.
REQUIREMENTS: All version numbers are the versions that were tested - other version might work, but success is not guaranteed.
- Python > v3.9 Installed with miniconda and the following packages (plus their dependencies): numpy = v1.26.4 pandas = v2.2.1 scipy = v1.13.0 matplotlib = v3.8.4 pymatgen = v2024.5.1 ase = v3.22.1
- QuantumEspresso = v7.1 ("/bin" directory loaded into the path, compiled for MPI)
- Yambo = v5.2 ("/bin" directory loaded into the path, compiled for MPI)
- Wannier90 = v3.1 ("/bin" and "/utility" directory loaded into the path, compiled for MPI, used only for the kmesh.pl utility)
USAGE:
- Convert the Alexandria database into .pckl-files using the conversion.ipynb notebook and place them in the input_data folder
- Create an empty folder named "database" and a folder named "CONTROL", which contains the file "NCORES". In this file, specify the total number of cores you want to use (e.g., 640)
- Specify the name of the .pckl-file and the workflows to run in the HT_watcher.py script
- If necessary, adapt the HT_watcher.py script and the start_calc function in basic_utils.py to your cluster
- Run the HT_watcher.py script
- If necessary, restart failed calculations using HT_restart.py or HT_full_restart.py
DATABASE: The main results of the calculations are saved in .json-files the database folder. The database follows the structure of the original database. We add the following data and parameters:
data: ipa_epsI_0: Imaginary part of the xx-component of the dielectric tensor ipa_epsR_0: Real part of the xx-component of the dielectric tensor ipa_indirect_gap: Indirect gap determined on the converged k-point grid for the IPA calculation ipa_direct_gap: Direct gap determined on the converged k-point grid for the IPA calculation ipa_eps_similarity: Similarity coefficient between ipa_epsI_0 using the final and the penultimate k-point density ipa_epsI_1: Imaginary part of the yy-component of the dielectric tensor ipa_epsR_1: Real part of the yy-component of the dielectric tensor ipa_epsI_2: Imaginary part of the zz-component of the dielectric tensor ipa_epsR_2: Real part of the zz-component of the dielectric tensor The latter four properties are only present if they are different from eps_I_0 and eps_R_0 based on symmetry. The gaps differ from the original gaps in the Alexandria database mostly due to the use of different codes, methods, and pseudopotentials. params: pw_conv_k: Converged K-point density in inverse Angstrom for the ground state calculation pw_conv_cutoff: Converged cutoff in Rydberg for the ground state calculation ipa_eps_kppa: Converged K-point density in inverse Angstrom for the IPA calculation ipa_nbands: Number of bands used for the IPA calculation ipa_broad: Value of the broadening in eV used for the IPA calculation
[1] https://alexandria.icams.rub.de/ [2] M. Grunert, M. Großmann and E. Runge, Deep learning of spectra: Predicting the dielectric function of semiconductors, Phys. Rev. Materials (2024)