Perspective: Atomistic simulations of water and aqueous systems with machine learning potentials

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 160(17)

Published: May 1, 2024

As the most important solvent, water has been at center of interest since advent computer simulations. While early molecular dynamics and Monte Carlo simulations had to make use simple model potentials describe atomic interactions, accurate ab initio relying on first-principles calculation energies forces have opened way predictive aqueous systems. Still, these are very demanding, which prevents study complex systems their properties. Modern machine learning (MLPs) now reached a mature state, allowing us overcome limitations by combining high accuracy electronic structure calculations with efficiency empirical force fields. In this Perspective, we give concise overview about progress made in simulation employing MLPs, starting from work free molecules clusters via bulk liquid electrolyte solutions solid–liquid interfaces.

Language: Английский

---

Million-atom heat transport simulations of polycrystalline graphene approaching first-principles accuracy enabled by neuroevolution potential on desktop GPUs

Journal of Applied Physics, Journal Year: 2025, Volume and Issue: 137(1)

Published: Jan. 2, 2025

First-principles molecular dynamics simulations of heat transport in systems with large-scale structural features are challenging due to their high computational cost. Here, using polycrystalline graphene as a case study, we demonstrate the feasibility simulating near first-principles accuracy containing over 1.4×106 atoms, achievable even consumer desktop GPUs. This is enabled by highly efficient neuroevolution potential (NEP) approach, implemented open-source GPUMD package. Leveraging NEP model’s and efficiency, quantify reduction thermal conductivity grain boundaries varying sizes, resolving contributions from in-plane out-of-plane (flexural) phonon modes. Additionally, find that can lead finite under significant tensile strain, contrast divergent behavior observed pristine similar conditions, indicating may play crucial role low-dimensional momentum-conserving systems. These findings could offer insights into interpreting experimental observations, given widespread presence both external strains real materials. The demonstrated ability simulate millions atoms near-first-principles on GPUs approach will help make high-fidelity atomistic more accessible broader research community.

Language: Английский

---

Highly efficient path-integral molecular dynamics simulations with GPUMD using neuroevolution potentials: Case studies on thermal properties of materials

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(6)

Published: Feb. 12, 2025

Path-integral molecular dynamics (PIMD) simulations are crucial for accurately capturing nuclear quantum effects in materials. However, their computational intensity often makes it challenging to address potential finite-size effects. Here, we present a specialized graphics processing units (GPUs) implementation of PIMD methods, including ring-polymer (RPMD) and thermostatted (TRPMD), into the open-source Graphics Processing Units Molecular Dynamics (GPUMD) package, combined with highly accurate efficient machine-learned neuroevolution (NEP) models. This approach achieves almost accuracy first-principles calculations efficiency empirical potentials, enabling large-scale atomistic that incorporate effects, effectively overcoming limitations at relatively affordable cost. We validate demonstrate efficacy NEP-PIMD by examining various thermal properties diverse materials, lithium hydride (LiH), three porous metal–organic frameworks (MOFs), liquid water, elemental aluminum. For LiH, our successfully capture isotope effect, reproducing experimentally observed dependence lattice parameter on reduced mass. MOFs, results reveal achieving good agreement experimental data requires consideration both dispersive interactions. significant impact its microscopic structure. aluminum, TRPMD method captures expansion phonon properties, aligning well mechanical predictions. GPU-accelerated GPUMD package provides an alternative, accessible, accurate, scalable tool exploring complex material influenced applications across broad range

Language: Английский

---

Mechanisms of temperature-dependent thermal transport in amorphous silica from machine-learning molecular dynamics

Physical review. B./Physical review. B, Journal Year: 2023, Volume and Issue: 108(18)

Published: Nov. 20, 2023

Amorphous silica (a-${\mathrm{SiO}}_{2}$) is a foundational disordered material for which the thermal transport properties are important various applications. To accurately model interatomic interactions in classical molecular dynamics (MD) simulations of a-${\mathrm{SiO}}_{2}$, we herein develop an accurate yet highly efficient machine-learned potential that allows us to generate a-${\mathrm{SiO}}_{2}$ samples closely resembling experimentally produced ones. Using homogeneous nonequilibrium MD method and proper quantum-statistical correction results, quantitative agreement with experiments achieved conductivities bulk 190-nm-thick films over wide range temperatures. interrogate vibrations at different temperatures, calculated current correlation functions corresponding transverse acoustic longitudinal collective vibrations. The results reveal that, below Ioffe-Regel crossover frequency, phonons as well-defined excitations remain applicable play predominant role low resulting temperature-dependent increase conductivity. In high-temperature region, more excited, accompanied by intense liquidlike diffusion event. We attribute temperature-independent conductivity collaborative involvement excited phonon scattering heat conduction. These findings provide physical insights into expected be applied vast amorphous materials.

Language: Английский

---

Comparing machine learning potentials for water: Kernel-based regression and Behler–Parrinello neural networks

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 160(11)

Published: March 20, 2024

In this paper, we investigate the performance of different machine learning potentials (MLPs) in predicting key thermodynamic properties water using RPBE + D3. Specifically, scrutinize kernel-based regression and high-dimensional neural networks trained on a highly accurate dataset consisting about 1500 structures, as well smaller dataset, half size, obtained only on-the-fly learning. This study reveals that despite minor differences between MLPs, their agreement observables such diffusion constant pair-correlation functions is excellent, especially for large training dataset. Variations predicted density isobars, albeit somewhat larger, are also acceptable, particularly given errors inherent to approximate functional theory. Overall, emphasizes relevance database over fitting method. Finally, underscores limitations root mean square need comprehensive testing, advocating use multiple MLPs enhanced certainty, when simulating complex may not be fully captured by simpler tests.

Language: Английский

---

Molecular dynamics simulations of heat transport using machine-learned potentials: A mini-review and tutorial on GPUMD with neuroevolution potentials

Journal of Applied Physics, Journal Year: 2024, Volume and Issue: 135(16)

Published: April 24, 2024

Molecular dynamics (MD) simulations play an important role in understanding and engineering heat transport properties of complex materials. An essential requirement for reliably predicting is the use accurate efficient interatomic potentials. Recently, machine-learned potentials (MLPs) have shown great promise providing required accuracy a broad range In this mini-review tutorial, we delve into fundamentals transport, explore pertinent MD simulation methods, survey applications MLPs transport. Furthermore, provide step-by-step tutorial on developing highly predictive simulations, utilizing neuroevolution as implemented GPUMD package. Our aim with to empower researchers valuable insights cutting-edge methodologies that can significantly enhance efficiency studies.

Language: Английский

---

Correcting force error-induced underestimation of lattice thermal conductivity in machine learning molecular dynamics

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 161(1)

Published: July 1, 2024

Machine learned potentials (MLPs) have been widely employed in molecular dynamics simulations to study thermal transport. However, the literature results indicate that MLPs generally underestimate lattice conductivity (LTC) of typical solids. Here, we quantitatively analyze this underestimation context neuroevolution potential (NEP), which is a representative MLP balances efficiency and accuracy. Taking crystalline silicon, gallium arsenide, graphene, lead telluride as examples, reveal fitting errors machine-learned forces against reference ones are responsible for underestimated LTC they constitute external perturbations interatomic forces. Since force NEP model random Langevin thermostat both follow Gaussian distribution, propose an approach correcting by intentionally introducing different levels noises via then extrapolating limit zero error. Excellent agreement with experiments obtained using correction all prototypical materials over wide range temperatures. Based on spectral analyses, find mainly arises from increased phonon scatterings low-frequency region caused errors.

Language: Английский

---

Combining linear-scaling quantum transport and machine-learning molecular dynamics to study thermal and electronic transports in complex materials

Journal of Physics Condensed Matter, Journal Year: 2024, Volume and Issue: 36(24), P. 245901 - 245901

Published: March 8, 2024

We propose an efficient approach for simultaneous prediction of thermal and electronic transport properties in complex materials. Firstly, a highly machine-learned neuroevolution potential (NEP) is trained using reference data from quantum-mechanical density-functional theory calculations. This then applied large-scale molecular dynamics simulations, enabling the generation realistic structures accurate characterization properties. In addition, simulations atoms linear-scaling quantum calculations electrons are coupled to account electron-phonon scattering other disorders that affect charge carriers governing demonstrate usefulness this unified by studying pristine graphene thermoelectric antidot lattice, with general-purpose NEP developed carbon systems based on extensive dataset.

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

---

Phonon coherence and minimum thermal conductivity in disordered superlattices

Physical review. B./Physical review. B, Journal Year: 2025, Volume and Issue: 111(8)

Published: Feb. 12, 2025

Language: Английский

---