Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

Addressing uncertainty in atomistic machine learning.

Andrew A Peterson¹, Rune Christensen, Alireza Khorshidi

¹School of Engineering, Brown University, Providence, Rhode Island 02912, USA. andrew_peterson@brown.edu.

Physical Chemistry Chemical Physics : PCCP

|April 19, 2017

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Grand Challenges and Opportunities in Stimulated Dynamic and Resonant Catalysis.

ACS catalysis·2026

Same author

Carbon-Doped Graphitic Carbon Nitride Inorganic Filler in Solid Polymer Electrolytes for All-Solid-State Batteries.

ACS applied materials & interfaces·2025

Same author

Anion engineering in lithium cobalt oxide for application in high-performance supercapacitors.

Scientific reports·2025

Same author

Influence of Coverage Dependence on the Thermophysical Properties of Adsorbates and Its Impact on Microkinetic Models.

Journal of chemical information and modeling·2025

Same author

Sulfonated polystyrene foam waste as an efficient catalyst for Friedel-Crafts type reactions.

Scientific reports·2025

Same author

Unifying thermochemistry concepts in computational heterogeneous catalysis.

Chemical Society reviews·2024

Same journal

Localization and delocalization of defect states in 2D polyaramid with carbon and nitrogen vacancies.

Physical chemistry chemical physics : PCCP·2026

Same journal

The impact of macrocyclization: electronic structures and excited state dynamics of pillar[4]arene[1]quinone.

Physical chemistry chemical physics : PCCP·2026

Same journal

Tuning the transport properties of penta-graphene nanoribbons.

Physical chemistry chemical physics : PCCP·2026

Same journal

High-throughput screening of M-based layered compounds as solid-state electrolytes for chloride-ion batteries.

Physical chemistry chemical physics : PCCP·2026

Same journal

Lower bound of the capacitance of constant phase elements based on electrochemical impedance spectra.

Physical chemistry chemical physics : PCCP·2026

Same journal

Stability constants of lanthanide-nitrate complexes in aqueous solutions: a theoretical study.

Physical chemistry chemical physics : PCCP·2026

See all related articles

Machine learning models can predict atomic energies and forces but require uncertainty quantification for new predictions. Uncertainty analysis helps validate machine learning models in atomistic simulations, enabling reliable acceleration of complex calculations.

Area of Science:

Computational chemistry
Materials science
Machine learning

Background:

Machine-learning regression accurately emulates potential energy and forces from electronic-structure calculations.
Predicting new regions of potential energy surfaces requires assessing the credibility of machine learning predictions.

Purpose of the Study:

To address errors in atomistic machine learning.
To highlight challenges in machine learning for atomistic simulations.
To demonstrate how uncertainty analysis validates machine learning predictions.

Main Methods:

Utilized a bootstrap ensemble of neural network-based calculators.
Analyzed the width of the ensemble to estimate prediction uncertainty.
Investigated the localization of uncertainty to specific atoms.

Related Experiment Videos

Main Results:

Ensemble width provides an estimate of uncertainty comparable to training data variations.
Uncertainty can be localized to individual atoms within simulations.
This localization offers insights for targeted training data generation.

Conclusions:

Uncertainty analysis is crucial for validating machine learning predictions in atomistic simulations.
Machine learning can be reliably used to accelerate large, high-accuracy, or extended-time simulations.
Localized uncertainty can guide strategic improvements in machine-learned representations.