This repository provides the tools for searching ICSD intermetallic compounds that contain kagome net and the ML models/tools trained on the database to rank unexplored chemical systems.
We adopt a two-step strategy with minimal constraint to filter the intermetallic compounds in order to provide a comprehensive database for fine-tuning use:
Step-1: Topological filter: finding the connectivity of atoms in a structure that matches the kagome connectivity. This step is done using TOPOSPRO software. Prior to TOPOSPRO, a structure is split into several single-element sub-structures based on the compositions of the lattice sites. The tool split_cif.py is provided in the geometrical_filter.
Step-2: Geometrical filter: analyzing the distortions of the nets using the pymatgen structure functionalities. Necessary tools and jupyter notebooks are in geometrical_filter.
The product database is used to train ML models in order to rank chemical systems that are unexplored and can potentially host kagome nets, see the folder ML_example.
The product ML models have identified chemical systems that are not seen in the ICSD but reported in the Pearson's crystal database.