Autonomous thermodynamically informed database generation for machine-learned interatomic potentials and application to magnesium

2025/08/12 by Vincent G. Fletcher, Albert P. Bartók, Fletcher, Vincent G. +3
Chemistry · Earth and Planetary Sciences · Materials Science · #FOS: Physical sciences #High-pressure geophysics and materials #Inorganic Chemistry and Materials #Machine Learning in Materials Science #Materials Science (cond-mat.mtrl-sci)

paper · pdf · doi:10.48550/arxiv.2508.08864

openalex publication_date 2025/08/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/03

Abstract

We propose a novel approach for constructing training databases for Machine-Learned Interatomic Potential (MLIP) models, specifically designed to capture phase properties across a wide range of conditions. The framework is uniquely appealing due to its ease of automation, its suitability for iterative learning, and its independence from prior knowledge of stable phases, avoiding bias towards pre-existing structural data. The approach uses Nested Sampling (NS) to explore the configuration space and generate thermodynamically relevant configurations, forming the database which undergoes ab initio Density Functional Theory (DFT) evaluation. We use the Atomic Cluster Expansion (ACE) architecture to fit a model on the resulting database. To demonstrate the efficiency of the framework, we apply it to magnesium, developing a model capable of accurately describing behaviour across pressure and temperature ranges of 0-600 GPa and 0-8000 K, respectively. We benchmark the model's performance by calculating phonon spectra and elastic constants, as well as the pressure-temperature phase diagram within this region. The results showcase the power of the framework to produce robust MLIPs while maintaining transferability and generality, for reduced computational cost.

Citations

• Automated generation of structure datasets for machine learning potentials and alloys
• Replica exchange nested sampling
• A Foundational Potential Energy Surface Dataset for Materials
• Fine-tuning foundation models of materials interatomic potentials with frozen transfer learning
• Multi-Phase Dataset for Ti and Ti-6Al-4V
• Orb: A Fast, Scalable Neural Network Potential
• Generalized Representative Structures for Atomistic Systems
• Data-efficient fine-tuning of foundational models for first-principles quality sublimation enthalpies
• MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures
• A foundation model for atomistic materials chemistry
• MACE-OFF: Transferable Short Range Machine Learning Force Fields for Organic Molecules
• Scaling deep learning for materials discovery
• Emissions and energy efficiency on large-scale high performance computing facilities: ARCHER2 UK national supercomputing service case study
• ACEpotentials.jl: A Julia Implementation of the Atomic Cluster Expansion
• Synthetic pre-training for neural-network interatomic potentials
• Robust Training of Machine Learning Interatomic Potentials with Dimensionality Reduction and Stratified Sampling
• Atomic Cluster Expansion for a General-Purpose Interatomic Potential of Magnesium
• First-principles Phonon Calculations with Phonopy and Phono3py
• Implementation strategies in phonopy and phono3py
• Hyperactive Learning (HAL) for Data-Driven Interatomic Potentials
• Graph neural networks for materials science and chemistry
• Optimal data generation for machine learned interatomic potentials
• Nested sampling for physical scientists
• Neural Network Potentials: A Concise Overview of Methods
• GemNet: Universal Directional Graph Neural Networks for Molecules
• Performant implementation of the atomic cluster expansion (PACE): Application to copper and silicon
• E(n) Equivariant Graph Neural Networks
• An Entropy-Maximization Approach to Automated Training Set Generation for Interatomic Potentials
• Atomic Cluster Expansion: Completeness, Efficiency and Stability
• Machine-learned Interatomic Potentials for Alloys and Alloy Phase Diagrams
• SchNet – A deep learning architecture for molecules and materials
• The atomic simulation environment—a Python library for working with atoms
• Crystal Structure Representations for Machine Learning Models of Formation Energies
• Crystal structure representations for machine learning models of formation energies
• Efficient sampling of atomic configurational spaces

Related

• A Combined Tight Binding with Machine Learning Potential Model for Magnesium Compounds
• Atomic Cluster Expansion for a General-Purpose Interatomic Potential of Magnesium
• Optimal data generation for machine learned interatomic potentials
• LightPFP: A Lightweight Route to Ab Initio Accuracy at Scale
• Accelerating CALPHAD-based Phase Diagram Predictions in Complex Alloys Using Universal Machine Learning Potentials: Opportunities and Challenges