Active Δ-learning with universal potentials for global structure optimization
- 1 December 2025
- journal article
- research article
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 28 (1) , 912-926
Abstract
Active learning and Δ-model construction are employed to improve universal machine learning interatomic potentials (uMLIPs) used for global structure optimization.Keywords
All Related Versions
Funding Information
- Danmarks Grundforskningsfond (DNRF150)
- Villum Fonden (16562)
This publication has 80 references indexed in Scilit:
- Universally Accurate or Specifically Inadequate? Stress‐Testing General Purpose Machine Learning Interatomic PotentialsAdvanced Intelligent Discovery, 2025
- Machine‐Learning‐Driven Exploration of Surface Reconstructions of Reduced Rutile TiO 2Angewandte Chemie International Edition, 2025
- Accelerating high-throughput phonon calculations via machine learning universal potentialsMaterials Today Physics, 2025
- Machine-learning interatomic potentials for materials scienceActa Materialia, 2021
- DScribe: Library of descriptors for machine learning in materials scienceComputer Physics Communications, 2020
- Systematic study on the structures and properties of (Ag2S)n (n = 1–8) clustersJournal of Molecular Modeling, 2019
- Amp: A modular approach to machine learning in atomistic simulationsComputer Physics Communications, 2016
- Adsorption of sulfur on Ag(100)Surface Science, 2011
- Practical Methods of Optimization2000
- Parallel tempering algorithm for conformational studies of biological moleculesChemical Physics Letters, 1997