Filtern
Dokumenttyp
- Posterpräsentation (2)
- Forschungsdatensatz (2)
- Zeitschriftenartikel (1)
- Preprint (1)
Sprache
- Englisch (6)
Schlagworte
- MACE (3)
- Phonons (3)
- Machine Learning (2)
- Phonon (2)
- Automation (1)
- Batteries (1)
- DFT (1)
- Finetuning (1)
- Foundation Model (1)
- Long-term stability (1)
Organisationseinheit der BAM
Harmonic phonon dataset
(2025)
Mace-mp-03b Phonon Benchmark
(2025)
This repository contains phonon calculations and evaluations for the MACE MP-0b3 model. First release of the scripts used for the phonon benchmark in the paper benchmarking the MACE-MP-03b model. See https://arxiv.org/abs/2401.00096 for a previous version of the paper. Full Changelog: https://github.com/JaGeo/mace-mp-03b-phonon-benchmark/commits/v0.0.1
Machine-learning interatomic potentials are widely used as computationally efficient surrogates for density functional theory in atomistic simulations, enabling large-scale, long-time modeling of materials systems. We investigate how different fine-tuning strategies influence the prediction of harmonic phonon band structures, thermal properties, and the potential energy surface along imaginary phonon modes. We achieve substantial accuracy improvements with minimal additional data, with as few as 10 additional training structures already yielding significant gains. In addition to existing approaches, we introduce Equitrain, a finetuning framework that implements LoRA-based adaptation. Across 53 materials systems, we show that fine-tuned models consistently outperform both the underlying pretrained model and models trained from scratch. Equitrain achieves the best overall performance, and our results demonstrate that fine-tuning enables accurate phonon predictions.
Phonon calculations with ab-initio methods are computationally expensive. The use of universal machine learning models reduces the cost, but raises concerns about prediction quality. Fine-tuning with only a few structures, improves predictions of phonons, thermal properties and especially diffusive thermal conductivity, while reducing computational cost by a factor of 10 in average compared to DFT methods.
Fine-tuning universal interatomic potentials for phonon properties without catastrophic forgetting
(2025)
Accurate phonon predictions are essential for evaluating material stability and thermal behavior. Traditional DFT-based methods are computationally intensive, driving interest in faster, machine-learning-based alternatives. In this work, we fine-tune the machine learning interatomic potential MACE-MP-0b3 (https://arxiv.org/abs/2401.00096) to improve the prediction of harmonic phonons and thermal properties.
A key challenge is the need for highly accurate force calculations. While fine-tuning can enhance phonon accuracy, it may reduce generalizability to other properties.
Despite this, our study shows that fine-tuned models can accurately predict phonon properties (including renormalization of imaginary phonons) at the same time as volume-dependent properties like the bulk modulus. We fine-tune the base model using small sets of rattled structures and introduce a novel fine-tuning method, benchmarking it against existing techniques. These results suggest that refined MACE-based universal potentials are a promising path for efficient and accurate phonon modeling. All implementations are included in the equitrain training code (https://github.com/BAMeScience/equitrain).
Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early machine-learning (ML) force fields have largely been limited by (i) the substantial computational and human effort required to develop and validate potentials for each particular system of interest and (ii) a general lack of transferability from one chemical system to the next. Here, we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model—and its qualitative and at times quantitative accuracy—on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces, and even the dynamics of a small protein. The model can be applied out of the box as a starting or “foundation” model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users obtain reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step toward democratizing the revolution in atomic-scale modeling that has been brought about by ML force fields.
