Benchmarking Coordination Number Prediction Algorithms on Inorganic Crystal Structures

Abstract

Coordination numbers and geometries form a theoretical framework for understanding and predicting materials properties. Algorithms to determine coordination numbers automatically are increasingly used for machine learning and automatic structural analysis. In this work, we introduce MaterialsCoord, a benchmark suite containing 56 experimentally-derived crystal structures (spanning elements, binaries, and ternary compounds) and their corresponding coordination environments as described in the research literature. We also describe CrystalNN, a novel algorithm for determining near neighbors. We compare CrystalNN against 7 existing near-neighbor algorithms on the MaterialsCoord benchmark, finding CrystalNN to perform similarly to several well-established algorithms. For each algorithm, we also assess computational demand and sensitivity towards small perturbations that mimic thermal motion. Finally, we 1 investigate the similarity between bonding algorithms when applied to the Materials Project database. We expect that this work will aid the development of coordination prediction algorithms as well as improve structural descriptors for machine learning and other applications.

A new material discovery platform of stable layered oxide cathodes for K-ion batteries

“…24 The dopant metals were selected based on a combination of probabilistic model, crystal-near-neighbor (CrystalNN), which is implemented in the Pymatgen library, and available literature reports. [25][26][27][28][29] From many candidates in periodic tables, we selectively applied seven different dopant metals (Zn, Al, Bi, Co, Ni, Fe and Cu) that were verified as an efficient medium in LIB cathode materials. 30 To identify the suitable doping level in K x MnO 2 , thereafter, the combined ML and DFT prediction was further performed in the dopants concentration of 0, 0.1 and 0.2.…”

Towards overcoming data scarcity in materials science: unifying models and datasets with a mixture of experts framework

“…To measure domain shift, CMD was computed in the learned feature space of the last frozen convolutional layer of each extractor. We found that CMD computed in this feature space showed a strong positive trend with CMD computed in the space of features procedurally generated from local atomic structure order parameters [50,51], suggesting the learned feature space faithfully distinguishes atomic structures (see Fig. 6).…”

Automatic Migration Path Exploration for Multivalent Battery Cathodes using Geometrical Descriptors

“…Following the recent findings on benchmarking nearest neighbors algorithms by Pan et al ., we consider an atom a neighbor if the solid angle is larger than 50 % of the maximum solid angle, i. e . ; [38] …”