English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Inexpensive machine learning potentials are increasingly being used to speedup structural optimization and molecular dynamics simulations of materials byiteratively predicting and applying interatomic forces. In these settings, itis crucial to detect when predictions are unreliable to avoid wrong ormisleading results. Here, we present a complete framework for training andrecalibrating graph neural network ensemble models to produce accuratepredictions of energy and forces with calibrated uncertainty estimates. Theproposed method considers both epistemic and aleatoric uncertainty and thetotal uncertainties are recalibrated post hoc using a nonlinear scalingfunction to achieve good calibration on previously unseen data, without loss ofpredictive accuracy. The method is demonstrated and evaluated on twochallenging, publicly available datasets, ANI-1x (Smith et al.) andTransition1x (Schreiner et al.), both containing diverse conformations far fromequilibrium. A detailed analysis of the predictive performance and uncertaintycalibration is provided. In all experiments, the proposed method achieved lowprediction error and good uncertainty calibration, with predicted uncertaintycorrelating with expected error, on energy and forces. To the best of ourknowledge, the method presented in this paper is the first to consider acomplete framework for obtaining calibrated epistemic and aleatoric uncertaintypredictions on both energy and forces in ML potentials.

Graph Neural Network Interatomic Potential Ensembles with Calibrated Aleatoric and Epistemic Uncertainty on Energy and Forces