Benchmarking Coordination Number Prediction Algorithms on Inorganic Crystal Structures

Coordination numbers and geometries form a theoretical framework for understanding predicting materials properties. Algorithms to determine coordination automatically are increasingly used machine learning (ML) automatic structural analysis. In this work, we introduce MaterialsCoord, benchmark suite containing 56 experimentally derived crystal structures (spanning elements, binaries, ternary compounds) their corresponding environments as described in the research literature. We also describe CrystalNN, novel algorithm determining near neighbors. compare CrystalNN against seven existing near-neighbor algorithms on MaterialsCoord benchmark, finding perform similarly several well-established algorithms. For each algorithm, assess computational demand sensitivity toward small perturbations that mimic thermal motion. Finally, investigate similarity between bonding when applied Materials Project database. expect work will aid development of prediction well improve descriptors ML other applications.