ArcaNN: automated enhanced sampling generation of training sets for chemically reactive machine learning interatomic potentials

Digital Discovery, Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 1, 2024

ArcaNN is a comprehensive framework that employs concurrent learning to generate training datasets for reactive MLIPs in the condensed phase.

Language: Английский

---

Random Sampling Versus Active Learning Algorithms for Machine Learning Potentials of Quantum Liquid Water

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 14, 2025

Training accurate machine learning potentials requires electronic structure data comprehensively covering the configurational space of system interest. As construction this is computationally demanding, many schemes for identifying most important structures have been proposed. Here, we compare performance high-dimensional neural network (HDNNPs) quantum liquid water at ambient conditions trained to sets constructed using random sampling as well various flavors active based on query by committee. Contrary common understanding learning, find that a given set size, leads smaller test errors not included in training process. In our analysis, show can be related small energy offsets caused bias added which overcome instead correlations an error measure invariant such shifts. Still, all HDNNPs yield very similar and structural properties water, demonstrates robustness procedure with respect algorithm even when few 200 structures. However, preliminary potentials, reasonable initial avoid unnecessary extension covered configuration less relevant regions.

Language: Английский

---

Enhanced sampling of robust molecular datasets with uncertainty-based collective variables

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(3)

Published: Jan. 15, 2025

Generating a dataset that is representative of the accessible configuration space molecular system crucial for robustness machine-learned interatomic potentials. However, complexity systems, characterized by intricate potential energy surfaces, with numerous local minima and barriers, presents significant challenge. Traditional methods data generation, such as random sampling or exhaustive exploration, are either intractable may not capture rare, but highly informative configurations. In this study, we propose method leverages uncertainty collective variable (CV) to guide acquisition chemically relevant points, focusing on regions where ML model predictions most uncertain. This approach employs Gaussian Mixture Model-based metric from single CV biased dynamics simulations. The effectiveness our in overcoming barriers exploring unseen minima, thereby enhancing an active learning framework, demonstrated alanine dipeptide bulk silica.

Language: Английский

---

Neural network potential for dislocation plasticity in ceramics

npj Computational Materials, Journal Year: 2024, Volume and Issue: 10(1)

Published: Nov. 22, 2024

Abstract Dislocations in ceramics are increasingly recognized for their promising potential applications such as toughening intrinsically brittle and tailoring functional properties. However, the atomistic simulation of dislocation plasticity remains challenging due to complex interatomic interactions characteristic ceramics, which include a mix ionic covalent bonds, highly distorted extensive core structures within crystal structures. These complexities exceed capabilities empirical potentials. Therefore, constructing neural network potentials (NNPs) emerges optimal solution. Yet, creating training dataset that includes proves difficult complexity configurations computational demands density theory large atomic models containing cores. In this work, we propose from properties easier compute via high-throughput calculation. Using dataset, have successfully developed NNPs specifically three typical ceramics: ZnO, GaN, SrTiO 3 . effectively capture nonstoichiometric charged slip barriers dislocations, well long-range electrostatic between dislocations. The effectiveness was further validated by measuring similarity uncertainty across snapshots derived large-scale simulations, alongside validation various Utilizing constructed NNPs, examined through nanopillar compression nanoindentation, demonstrated excellent agreement with experimental observations. This study provides an effective framework enable detailed modeling plasticity, opening new avenues exploring plastic behavior ceramics.

Language: Английский

---

Structure and transport properties of LiTFSI-based deep eutectic electrolytes from machine learning interatomic potential simulations

Published: Aug. 9, 2024

Deep Eutectic Solvents have recently gained significant attention as versatile and inexpensive materials with many desirable properties a wide range of applications. In particular, their similar characteristics to ionic liquids, make them promising class liquid electrolytes for electrochemical this study, we utilized local equivariant neural network interatomic potential model study series deep eutectic based on lithium bis(trifluoromethanesulfonyl)imide (LiTFSI) by molecular dynamics (MD) simulations. The use features combined the strict locality result in highly accurate, data-efficient scalable potentials enabling large-scale MD simulations these liquids first-principles accuracy. Comparing structure reported results from classical force field (FF) indicates that ion–ion interactions are not accurately characterized FFs. Furthermore, close contacts between ions bridged oxygen atoms two amide molecules observed. computed cationic transport numbers estimated ratios Li–amide lifetime (τ[Li–amide]) amide’s rotational relaxation time (τ[R]), conductivity trend, suggest more structural Li+ mechanism LiTFSI:urea mixture through exchange molecules. However, vehicular could larger contribution ion LiTFSI:N-methylacetamide electrolyte. Moreover, comparable diffusivities cation TFSI – anion τ[Li–amide]/τ[R] unity, indicate solvent-exchange mechanisms rather equal contributions LiTFSI:acetamide system.

Language: Английский

---

Prediction rigidities for data-driven chemistry

Faraday Discussions, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 23, 2024

The widespread application of machine learning (ML) to the chemical sciences is making it very important understand how ML models learn correlate structures with their properties, and what can be done improve training efficiency whilst guaranteeing interpretability transferability. In this work, we demonstrate wide utility prediction rigidities, a family metrics derived from loss function, in understanding robustness model predictions. We show that rigidities allow assessment not only at global level, but also on local or component-wise level which intermediate (e.g. atomic, body-ordered, range-separated) predictions are made. leverage these behavior different models, guide efficient dataset construction for training. finally implement formalism targeting coarse-grained system applicability an even broader class atomistic modeling problems.

Language: Английский

---

Structure and transport properties of LiTFSI-based deep eutectic electrolytes from machine-learned interatomic potential simulations

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 161(13)

Published: Oct. 1, 2024

Deep eutectic solvents have recently gained significant attention as versatile and inexpensive materials with many desirable properties a wide range of applications. In particular, their characteristics, similar to those ionic liquids, make them promising class liquid electrolytes for electrochemical this study, we utilized local equivariant neural network interatomic potential model study series deep based on lithium bis(trifluoromethanesulfonyl)imide (LiTFSI) using molecular dynamics (MD) simulations. The use features combined strict locality results in highly accurate, data-efficient, scalable potentials, enabling large-scale MD simulations these liquids first-principles accuracy. Comparing the structure reported from classical force field (FF) indicates that ion–ion interactions are not accurately characterized by FFs. Furthermore, close contacts between ions, bridged oxygen atoms two amide molecules, observed. computed cationic transport numbers (t+) estimated ratios Li+–amide lifetime (τLi–amide) amide’s rotational relaxation time (τR), conductivity trend, suggest more structural Li+ mechanism LiTFSI:urea mixture through exchange molecules. However, vehicular could larger contribution ion LiTFSI:N-methylacetamide electrolyte. Moreover, comparable diffusivities cation TFSI− anion τLi–amide/τR unity indicate solvent-exchange mechanisms rather equal contributions LiTFSI:acetamide system.

Language: Английский

---

Efficient ensemble uncertainty estimation in Gaussian Processes Regression

Machine Learning Science and Technology, Journal Year: 2024, Volume and Issue: 5(4), P. 045029 - 045029

Published: Oct. 21, 2024

Abstract Reliable uncertainty measures are required when using data-based machine learning interatomic potentials (MLIPs) for atomistic simulations. In this work, we propose sparse Gaussian process regression (GPR) type MLIPs a stochastic measure akin to the query-by-committee approach often used in conjunction with neural network based MLIPs. The is coined ‘label noise’ ensemble as it emerges from adding noise energy labels training data. We find that method of calculating an well calibrated one obtained closed-form expression posterior variance GPR treated projected process. Comparing two methods, our proposed is, however, faster evaluate than expression. Finally, demonstrate acts better support Bayesian search optimal structure Au 20 clusters.

Language: Английский

---

Data efficient machine learning potentials for modeling catalytic reactivity via active learning and enhanced sampling

npj Computational Materials, Journal Year: 2024, Volume and Issue: 10(1)

Published: Dec. 19, 2024

Language: Английский

---

Uncertainty quantification and propagation in atomistic machine learning

Reviews in Chemical Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 30, 2024

Abstract Machine learning (ML) offers promising new approaches to tackle complex problems and has been increasingly adopted in chemical materials sciences. In general, ML models employ generic mathematical functions attempt learn essential physics chemistry from large amounts of data. The reliability predictions, however, is often not guaranteed, particularly for out-of-distribution data, due the limited physical or principles functional form. Therefore, it critical quantify uncertainty predictions understand its propagation downstream applications. This review examines existing quantification (UQ) (UP) methods atomistic under framework probabilistic modeling. We first categorize UQ explain similarities differences among them. Following this, performance metrics evaluating their accuracy, precision, calibration, efficiency are presented, along with techniques recalibration. These then applied survey benchmark studies that use molecular datasets. Furthermore, we discuss UP propagate widely used simulation techniques, such as dynamics microkinetic conclude remarks on challenges opportunities ML.

Language: Английский

---