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

Advances in machine learning (ML) techniques have enabled the development ofinteratomic potentials that promise both the accuracy of first principlesmethods and the low-cost, linear scaling, and parallel efficiency of empiricalpotentials. Despite rapid progress in the last few years, ML-based potentialsoften struggle to achieve transferability, that is, to provide consistentaccuracy across configurations that significantly differ from those used totrain the model. In order to truly realize the promise of ML-based interatomicpotentials, it is therefore imperative to develop systematic and scalableapproaches for the generation of diverse training sets that ensure broadcoverage of the space of atomic environments. This work explores adiverse-by-construction approach that leverages the optimization of the entropyof atomic descriptors to create a very large ($>2\cdot10^{5}$ configurations,$>7\cdot10^{6}$ atomic environments) training set for tungsten in an automatedmanner, i.e., without any human intervention. This dataset is used to trainpolynomial as well as multiple neural network potentials with differentarchitectures. For comparison, a corresponding family of potentials were alsotrained on an expert-curated dataset for tungsten. The models trained toentropy-optimized data exhibited vastly superior transferability compared tothe expert-curated models. Furthermore, while the models trained with heavyuser input (i.e., domain expertise) yield the lowest errors when tested onsimilar configurations, out-sample predictions are dramatically more robustwhen the models are trained on a deliberately diverse set of training data.Herein we demonstrate the development of both accurate and transferable MLpotentials using automated and data-driven approaches for generating large anddiverse training sets.

Training Data Selection for Accuracy and Transferability of Interatomic Potentials