Representation-Invariant Analysis of Elastic Tensors via Riemannian Geometry: Crystal Symmetry Classification, Property Prediction, and Anomaly Detection on 11,265 Materials

Read full text Summarise with VIRAT
Authors: Sloan Austermann
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We demonstrate that Euclidean analysis of elastic stiffness tensors is fundamentally flawed: distances between materials change by up to 96,913× depending on whether stiffness or compliance representation is used. The affine-invariant Riemannian metric eliminates this representation dependence exactly. Using proprietary Riemannian descriptors, we achieve 94.4% crystal system classification accuracy from only 5 features (vs. 72.4% for 21D Voigt baseline) and 96.9% with combined features on the de Jong et al. (2015) dataset (1,181 DFT-computed materials). Scaling to 11,265 materials from the Materials Project, we identify 2,018 entries (15.2%) with non-physical elastic tensors that fail positive-definiteness validation. Property prediction achieves R² = 0.989 for bulk modulus and R² = 0.977 for shear modulus. Precomputed Riemannian features for all 11,265 materials are available via commercial API at materials.omnisciences.io.

Fields of Study & Subjects

Machine Learning in Materials Science Material Selection and Properties Microstructure and mechanical properties

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN