Dielectric functions of intermetallic compounds (Part 1)

Part 1 of the dataset associated with the publication "Broken neural scaling laws in learning the optical properties of solids".The total dataset contains the frequency-dependent dielectric functions and Drude frequencies of 201,361 intermetallic compounds calculated within the independent-particle approximation. It was generated using a high-throughput, ab initio workflow employing the PBE exchange-correlation functional and SG15 pseudopotentials, using the Quantum ESPRESSO and SIMPLE codes.Due to its size (over 20 GB), the dataset is split into two separate Figshare records. Within each record, the data are provided as ZIP archives containing JSON files. The file names of the ZIP archives follow the convention "database_nsite_N_part_P.zip" where "N" denotes the number of atoms per unit cell and "P" denotes the corresponding part index. Each ZIP archive contains data for at most 15,000 materials.Each JSON file stores:the crystal structure (key: "structure", taken from the Alexandria database),the complex interband dielectric function (keys: "eps_inter_re", "eps_inter_im"), uniformly sampled on an energy grid from 0 to 20 eV using 2001 points,the Drude frequency (key: "drude_freq"), given in eV,the complex total dielectric function (keys: "eps_tot_re", "eps_tot_im"), uniformly sampled on an energy grid from 0 to 20 eV using 2001 points,and the joint density of states (key: "jdos"), uniformly sampled on an energy grid from 0 to 20 eV using 2001 points.The JSON files can be loaded in Python using the package "pymatgen":import json from pymatgen.entries.computed_entries import ComputedStructureEntryfile_path = "PATH_TO_A_JSON_FILE" with open(file_path, "r") as file: json_dict = json.load(file)cse = ComputedStructureEntry.from_dict(json_dict) print(cse.data["eps_inter_im"]) The high-throughput ab initio workflow workflow used to generate this dataset is available at: https://github.com/MaxGrossmann/ht_metals