From electrons to phase diagrams with machine learning potentials using pyiron based automated workflows

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

Published: Nov. 17, 2024

Abstract We present a comprehensive and user-friendly framework built upon the integrated development environment (IDE), enabling researchers to perform entire Machine Learning Potential (MLP) cycle consisting of (i) creating systematic DFT databases, (ii) fitting Density Functional Theory (DFT) data empirical potentials or MLPs, (iii) validating in largely automatic approach. The power performance this are demonstrated for three conceptually very different classes interatomic potentials: an potential (embedded atom method - EAM), neural networks (high-dimensional network HDNNP) expansions basis sets (atomic cluster expansion ACE). As advanced example validation application, we show computation binary composition-temperature phase diagram Al-Li, technologically important lightweight alloy system with applications aerospace industry.

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

---

Machine Learning Potentials for Heterogeneous Catalysis

ACS Catalysis, Journal Year: 2025, Volume and Issue: 15(3), P. 1616 - 1634

Published: Jan. 15, 2025

The production of many bulk chemicals relies on heterogeneous catalysis. rational design or improvement the required catalysts critically depends insights into underlying mechanisms atomic scale. In recent years, substantial progress has been made in applying advanced experimental techniques to complex catalytic reactions operando, but order achieve a comprehensive understanding, additional information from computer simulations is indispensable cases. particular, ab initio molecular dynamics (AIMD) become an important tool explicitly address atomistic level structure, dynamics, and reactivity interfacial systems, high computational costs limit applications systems consisting at most few hundred atoms for simulation times up tens picoseconds. Rapid advances development modern machine learning potentials (MLP) now offer promising approach bridge this gap, enabling with accuracy small fraction costs. Perspective, we provide overview current state art MLPs relevant catalysis along discussion prospects use science years come.

Language: Английский

---

Outlier-detection for reactive machine learned potential energy surfaces

npj Computational Materials, Journal Year: 2025, Volume and Issue: 11(1)

Published: Feb. 15, 2025

Abstract Uncertainty quantification (UQ) to detect samples with large expected errors (outliers) is applied reactive molecular potential energy surfaces (PESs). Three methods–Ensembles, deep evidential regression (DER), and Gaussian Mixture Models (GMM)—were the H-transfer reaction between syn -Criegee vinyl hydroxyperoxide. The results indicate that ensemble models provide best for detecting outliers, followed by GMM. For example, from a pool of 1000 structures largest uncertainty, detection quality outliers ~90% ~50%, respectively, if 25 or are sought. On contrary, limitations statistical assumptions DER greatly impact its prediction capabilities. Finally, structure-based indicator was found be correlated average error, which may help rapidly classify new into those an advantage refining neural network.

Language: Английский

---

A practical guide to machine learning interatomic potentials – Status and future

Current Opinion in Solid State and Materials Science, Journal Year: 2025, Volume and Issue: 35, P. 101214 - 101214

Published: Feb. 26, 2025

Language: Английский

---

Advances in modeling complex materials: The rise of neuroevolution potentials

Chemical Physics Reviews, Journal Year: 2025, Volume and Issue: 6(1)

Published: March 1, 2025

Interatomic potentials are essential for driving molecular dynamics (MD) simulations, directly impacting the reliability of predictions regarding physical and chemical properties materials. In recent years, machine-learned (MLPs), trained against first-principles calculations, have become a new paradigm in materials modeling as they provide desirable balance between accuracy computational cost. The neuroevolution potential (NEP) approach, implemented open-source GPUMD software, has emerged promising potential, exhibiting impressive exceptional efficiency. This review provides comprehensive discussion on methodological practical aspects NEP along with detailed comparison other representative state-of-the-art MLP approaches terms training accuracy, property prediction, We also demonstrate application approach to perform accurate efficient MD addressing complex challenges that traditional force fields typically cannot tackle. Key examples include structural liquid amorphous materials, order alloy systems, phase transitions, surface reconstruction, material growth, primary radiation damage, fracture two-dimensional nanoscale tribology, mechanical behavior compositionally alloys under various loadings. concludes summary perspectives future extensions further advance this rapidly evolving field.

Language: Английский

---

Predicting Solid-state NMR Observables via Machine Learning

Royal Society of Chemistry eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 224 - 255

Published: March 31, 2025

Machine learning is becoming increasingly important in the prediction of nuclear magnetic resonance (NMR) chemical shifts and other observable properties. This chapter provides an introduction to construction machine (ML) models for predicting NMR properties, including discussion feature engineering, common ML model types, Δ-ML transfer learning, curation training testing data. Then it discusses a number recent examples spin–spin coupling constants organic inorganic species. These highlight how decisions made constructing impact its performance, discuss strategies achieving more accurate models, present some representative case studies showing transforming way crystallography performed.

Language: Английский

---

A Theoretical Study on the Structural Evolution of Ru–Zn Bimetallic Nanoparticles

Nanomaterials, Journal Year: 2025, Volume and Issue: 15(8), P. 568 - 568

Published: April 8, 2025

Ru-Zn catalysts exhibit excellent catalytic performance for the selective hydrogenation of benzene to cyclohexene and has been utilized in industrial production. However, structure-performance relationship between remains lacking. In this work, we focused on evolution nanoparticles with size Ru/Zn ratio. The structures Ru bimetallic different sizes were determined by minima-hopping global optimization method combination density functional theory high-dimensional neural network potential. Furthermore, propose growth mechanism processes nanoparticles. Additionally, analyzed structural stability, electronic properties, adsorption properties Zn atoms. This work provides valuable reference guidance future theoretical research applications.

Language: Английский

---

Thermal decomposition mechanism of TKX-50 explored by neural network based molecular dynamics simulation

Fuel, Journal Year: 2025, Volume and Issue: 397, P. 135420 - 135420

Published: April 16, 2025

Language: Английский

---

Efficient treatment of long-range electrostatics in charge equilibration approaches

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(17)

Published: May 2, 2025

A charge equilibration method based on real-space Gaussians as densities is presented. The implementation part of the Electrode package available in Large-scale Atomic/Molecular Massively Parallel Simulator and benefits from its efficient particle-mesh Ewald approach. simple strategy required to switch previously used Slater-type orbital (STO) shielding provided by fitting Coulomb energy two Gaussian distributions repulsion between STOs. Their widths were optimized for O, Si, Ti species, obtaining results consistent with previous studies using STOs case SiO2 polymorphs. In limit sufficiently narrow Gaussians, it shown that converges electronegativity equalization Ti/TiOx interfaces. presented implemented a way potentially beneficial application modern machine-learning force fields include long-range electrostatic interactions.

Language: Английский

---