2026-10-14T15:41:51Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33691693 2026-09-13T15:19:10Z category_29161 item_type_3 month_year_09_2026

Prophet-OAME-MBD: Matbench Discovery prediction files Kairos Materials (24850590) Deep learning Matbench discovery Machine Learning Interatomic Potentials (MLIPs) Equivariant Neural Networks <p dir="ltr">Prediction artifacts of the Prophet-OAME-MBD interatomic potential for the Matbench Discovery benchmark, in the benchmark's canonical file formats: WBM discovery formation-energy predictions (256,961 structures), relaxed geometries with symmetry analyses at symprec 1e-5 and 1e-2 (moyopy 0.7.9), lattice thermal conductivity predictions for the 103 PhononDB structures, homonuclear diatomic curves, and molecular-dynamics benchmark metrics.</p><p><br></p><p dir="ltr">Prophet is an equivariant message-passing interatomic potential by Kairos Materials (62.3M parameters) built on Multi-Cutoff Spectral Decomposition. Inference code: https://github.com/kairosmaterial/prophet. Model weights: https://huggingface.co/kairosmaterial/prophet. </p> 2026-09-13T15:19:10Z Dataset Dataset 10.6084/m9.figshare.33691693.v2 https://figshare.com/articles/dataset/Prophet-OAME-MBD_Matbench_Discovery_prediction_files/33691693 CC BY 4.0