arXiv subjects

Yuta Yoshimoto

Publications and source records attributed to Yuta Yoshimoto.

9 recordsLinked to original sources

arXiv2025

Knowledge Distillation Framework for Accelerating High-Accuracy Neural Network-Based Molecular Dynamics Simulations

Neural network potentials (NNPs) offer a powerful alternative to traditional force fields for molecular dynamics (MD) simulations. Accurate and stable MD simulations, crucial for evaluating material properties, require training data encompassing both low-energy stable structures and high-energy structures. Conventional knowledge distillation (KD) methods fine-tune a pre-trained NNP as a teacher model to generate training data for a student model. However, in material-specific models, this fine-tuning process increases energy barriers, making it difficult to create training data containing high-energy structures. To address this, we propose a novel KD framework that leverages a non-fine-tuned, off-the-shelf pre-trained NNP as a teacher. Its gentler energy landscape facilitates the exploration of a wider range of structures, including the high-energy structures crucial for stable MD simulations. Our framework employs a two-stage training process: first, the student NNP is trained with a dataset generated by the off-the-shelf teacher; then, it is fine-tuned with a smaller, high-accuracy density functional theory (DFT) dataset. We demonstrate the effectiveness of our framework by applying it to both organic (polyethylene glycol) and inorganic (L$_{10}$GeP$_{2}$S$_{12}$) materials, achieving comparable or superior accuracy in reproducing physical properties compared to existing methods. Importantly, our method reduces the number of expensive DFT calculations by 10x compared to existing NNP generation methods, without sacrificing accuracy. Furthermore, the resulting student NNP achieves up to 106x speedup in inference compared to the teacher NNP, enabling significantly faster and more efficient MD simulations.

cs.LG↗

arXiv2025

Molecular Dynamics Simulations of SrTiO$_3$ with Oxygen Vacancies using Neural Network Potentials

A precise analysis of point defects in solids requires accurate molecular dynamics (MD) simulations of large-scale systems. However, ab initio MD simulations based on density functional theory (DFT) incur high computational cost, while classical MD simulations lack accuracy. We perform MD simulations using a neural network potential (NNP) model (NNP-MD) to predict the physical quantities of both pristine SrTiO$_3$ and SrTiO$_3$ in the presence of oxygen vacancies (V$_{\text{O}}$). To verify the accuracy of the NNP models trained on different data sets, their NNP-MD predictions are compared with the results obtained from DFT calculations. The predictions of the total energy show good agreement with the DFT results for all these NNP models, and the NNP models can also predict the formation energy once SrTiO$_3$:V$_{\text{O}}$ data are included in the training data sets. Even for larger supercell sizes that are difficult to calculate using first-principles calculations, the formation energies evaluated from the NNP-MD simulations well reproduce the extrapolated DFT values. This study offer important knowledge for constructing accurate NNP models to describe point-defect systems including SrTiO$_3$:V$_{\text{O}}$.

cond-mat.mtrl-sci↗

arXiv2025

Active Learning of a Neural Network Potential for Large-Scale Atomistic Simulations of Polymer Electrolyte Membranes

Machine learning interatomic potentials (MLIPs) can achieve near density-functional-theory (DFT) accuracy at force-field computational cost; however, long-time, large-scale molecular dynamics (MD) simulations often fail when trajectories sample local atomic environments that are underrepresented in the training set. Here, we employ an active-learning workflow to develop a robust neural network potential (NNP) for perfluorinated ionomer membranes (Nafion) across a wide range of hydration levels ($\lambda$). A reliable deep potential (DP) model is constructed through iterative dataset expansion within active-learning loops. Specifically, off-equilibrium configurations are generated via non-equilibrium DPMD simulations and selected using an ensemble force-deviation criterion combined with a three-dimensional structural feature space augmented by minimum interatomic distances, which significantly enhances the DP model's robustness. The trained DP model enables stable MD simulations of large Nafion systems containing approximately 10,000-20,000 atoms for an extended duration of 31 ns. Our DPMD simulations reproduce the qualitative hydration dependence of density and yield self-diffusion coefficients of hydrogen atoms and hydronium ions in quantitative agreement with experimental values. Compared with previous ab initio MD and MLIP-MD studies, our simulations show improved agreement with experimental transport properties and remain predictive up to $\lambda = 24$, well beyond the training range ($\lambda \leq 13$). This work provides an efficient and scalable approach for achieving stable, large-scale NNP-MD simulations of heterogeneous polymer electrolyte membranes and related disordered materials.

cond-mat.mtrl-sci↗

arXiv2024

Generator of Neural Network Potential for Molecular Dynamics: Constructing Robust and Accurate Potentials with Active Learning for Nanosecond-scale Simulations

Neural network potentials (NNPs) enable large-scale molecular dynamics (MD) simulations of systems containing >10,000 atoms with the accuracy comparable to ab initio methods and play a crucial role in material studies. Although NNPs are valuable for short-duration MD simulations, maintaining the stability of long-duration MD simulations remains challenging due to the uncharted regions of the potential energy surface (PES). Currently, there is no effective methodology to address this issue. To overcome this challenge, we developed an automatic generator of robust and accurate NNPs based on an active learning (AL) framework. This generator provides a fully integrated solution encompassing initial dataset creation, NNP training, evaluation, sampling of additional structures, screening, and labeling. Crucially, our approach uses a sampling strategy that focuses on generating unstable structures with short interatomic distances, combined with a screening strategy that efficiently samples these configurations based on interatomic distances and structural features. This approach greatly enhances the MD simulation stability, enabling nanosecond-scale simulations. We evaluated the performance of our NNP generator in terms of its MD simulation stability and physical properties by applying it to liquid propylene glycol (PG) and polyethylene glycol (PEG). The generated NNPs enable stable MD simulations of systems with >10,000 atoms for 20 ns. The predicted physical properties, such as the density and self-diffusion coefficient, show excellent agreement with the experimental values. This work represents a remarkable advance in the generation of robust and accurate NNPs for organic materials, paving the way for long-duration MD simulations of complex systems.

cond-mat.mtrl-sci↗

arXiv2024

Transferability of the chemical bond-based machine learning model for dipole moment: the GHz to THz dielectric properties of liquid propylene glycol and polypropylene glycol

We conducted a first-principles study of the dielectric properties of liquid propylene glycol (PG) and polypropylene glycol (PPG) using a recently developed chemical bond-based machine learning (ML) model for dipole moments [T. Amano et al. Phys. Rev. B 110, 165159 (2024)]. The ML dipole models successfully predict the dipole moment of various liquid configurations in close agreement with DFT calculations and generate $20 \mathrm{ns}$ quantum-accuracy dipole moment trajectories to calculate the dielectric function, when combined with ML potentials. The calculated dielectric function of PG closely matches experimental results. We identified a libration peak at $600\, \mathrm{cm}^{-1}$ and an intermolecular mode at $100\, \mathrm{cm}^{-1}$, previously noted experimentally. Furthermore, the models trained on PG2 training data can apply to longer chain PPG not included in the training data. The present research marks the first step toward developing a universal bond-based dipole model.

physics.chem-ph↗

arXiv2022

Gas Adsorption and Diffusion Behaviors in Interfacial Systems Composed of a Polymer of Intrinsic Microporosity and Amorphous Silica: A Molecular Simulation Study

We investigate the adsorption and diffusion behaviors of CO2, CH4, and N2 in interfacial systems composed of a polymer of intrinsic microporosity (PIM-1) and amorphous silica using grand canonical Monte Carlo (GCMC) and molecular dynamics (MD) simulations. We build model systems of mixed matrix membranes (MMMs) with PIM-1 chains sandwiched between silica surfaces. Gas adsorption analysis using GCMC simulations shows that gas molecules are preferentially adsorbed in microcavities distributed near silica surfaces, resulting in an increase in the solubility coefficients of CO2, CH4, and N2 compared to bulk PIM-1. In contrast, diffusion coefficients obtained from MD simulations and then calibrated using the dual-mode sorption model show different tendencies depending on gas species: CO2 diffusivity decreases in MMMs compared to PIM-1, whereas CH4 and N2 diffusivities increase. These differences are attributed to competing effects of silica surfaces: the emergence of larger pores as a result of chain packing disruption, which enhances gas diffusion, and a quadrupole-dipole interaction between gas molecules and silica surface hydroxyl groups, which retards gas diffusion. The former has a greater impact on CH4 and N2 diffusivities, whereas the latter has a greater impact on CO2 diffusivity due to the strong quadrupole-dipole interaction between CO2 and surface hydroxyls. These findings add to our understanding of gas adsorption and diffusion behaviors in the vicinity of PIM-1/silica interfaces, which are unobtainable in experimental studies.

cond-mat.mtrl-sci↗

arXiv2021

Evaluation of gas permeability in porous separators for polymer electrolyte fuel cells: CFD simulation based on micro X-ray computed tomography images

Pore structures and gas transport properties in porous separators for polymer electrolyte fuel cells are evaluated both experimentally and through simulations. In the experiments, the gas permeabilities of two porous samples, a conventional sample and one with low electrical resistivity, are measured by a capillary flow porometer, and the pore size distributions are evaluated with mercury porosimetry. Local pore structures are directly observed with micro X-ray computed tomography (CT). In the simulations, the effective diffusion coefficients of oxygen and the air permeability in porous samples are calculated using random walk Monte Carlo simulations and computational fluid dynamics (CFD) simulations, respectively, based on the X-ray CT images. The calculated porosities and air permeabilities of the porous samples are in good agreement with the experimental values. The simulation results also show that the in-plane permeability is twice the through-plane permeability in the conventional sample, whereas it is slightly higher in the low-resistivity sample. The results of this study show that CFD simulation based on micro X-ray CT images makes it possible to evaluate anisotropic gas permeabilities in anisotropic porous media.

physics.flu-dyn↗

arXiv2021

Molecular Insights into the Mechanical Properties of Polymer-Fullerene Bulk Heterojunctions for Organic Photovoltaic Applications

We investigate the mechanical properties of $π$-conjugated polymeric materials composed of regioregular poly(3-hexylthiophene) (P3HT) and fullerene C$_{60}$ using coarse-grained molecular dynamics simulations. Specifically, we perform tensile simulations of P3HT:C$_{60}$ composites with varied degrees of polymerization and C$_{60}$ mass fractions to obtain their stress-strain responses. Decomposition of stress tensor into kinetic energy and virial contributions indicates that the tensile moduli of the pure P3HT samples are greatly dependent on non-bonded interactions and on bonded interactions associated with bond-stretching, while the addition of C$_{60}$ leads to an increase in the tensile modulus originating from enhanced non-bonded interactions associated with C$_{60}$. Additionally, the tensile strength of the P3HT:C$_{60}$ samples correlates well with molecular chain entanglements, which are characterized by the average number of kinks per chain obtained from primitive path analysis. We also find that the upper and lower yield points characterizing strain softening become more pronounced with increasing C$_{60}$ mass fraction. Persistent homology analysis indicates that the emergence of the yield points correlates well with the coalescence of microvoids in the course of tensile deformation, resulting in the generation of larger voids. These results provide a fundamental understanding of the molecular determinants of the mechanical properties of $π$-conjugated polymer-fullerene composites, which can also help to interpret and predict the mechanical properties of other polymer composites containing fullerene.

cond-mat.mtrl-sci↗

arXiv2020

Relating the thermal properties of a micro pulsating heat pipe to the internal flow characteristics via experiments, image recognition of flow patterns and heat transfer simulations

We investigate the relationship between the thermal properties of a micro pulsating heat pipe (MPHP) and the internal flow characteristics. The MPHP consists of an eleven-turn closed-loop of a meandering square microchannel with a hydraulic diameter of $350\ {}μ{\rm m}$ engraved on a silicon substrate. The MPHP charged with Fluorinert FC-72 tends to exhibit higher effective thermal conductivities for the coolant temperature of $T_{\rm c} = 40\ {}^\circ\mathrm{C}$ compared to $T_{\rm c} = 20\ {}^\circ\mathrm{C}$, and provides the highest effective thermal conductivity of about $700\ {}{\rm W/(m{\cdot}K)}$ for $T_{\rm c} = 40\ {}^\circ\mathrm{C}$ and a filling ratio of 48%. Interestingly, we observe two different self-oscillation modes having different thermal conductivities, even for identical heat input rates. This tendency indicates a hysteresis of the effective thermal conductivity, which originates from the difference in the heat input rates at which the MPHP falls into and recovers from dryout. Subsequently, semantic segmentation-based image recognition is applied to the recorded flow images to identify the flow characteristics, successfully extracting four different flow patterns involving liquid slugs, liquid films, dry walls, and rapid-boiling regions. The image recognition results indicate that high effective thermal conductivities of the MPHP relate to stable self-oscillations with large amplitudes and high frequencies, along with long and thin liquid films beneficial for latent heat transfer. Finally, we perform numerical simulations of latent/sensible heat transfer via vapor plugs and of sensible heat transfer via liquid slugs using the extracted flow patterns as inputs. We find that latent heat transfer via liquid films accounts for a considerable portion of the overall heat transfer, while the sensible heat transfer via liquid slugs is much less significant.

physics.app-ph↗