arXiv subjects

Won Bo Lee

Publications and source records attributed to Won Bo Lee.

At least 19 recordsLinked to original sources

arXiv2026

Density diversity in training data governs thermodynamic transferability of machine learning interatomic potentials

Machine learning interatomic potentials (MLIPs) offer first-principles accuracy with reduced computational cost, but their transferability across different thermodynamic states remains questionable, particularly for fluid systems where molecules experience local environments far from crystalline equilibrium. Here, we demonstrate that diversifying the density of training configurations, rather than temperature, is the most effective strategy for building thermodynamically transferable MLIPs within a fixed computational budget. We first show that foundation MLIPs trained on solid-state databases accurately describe liquid-like densities but fail at gas-like conditions, while molecular-database-trained models exhibit the opposite behavior. Controlled from-scratch training and distillation experiments confirm that density-diverse datasets resolve both failure modes, whereas temperature-diverse datasets cannot compensate for missing density regimes. Coordination number analysis reveals the physical origin of this behavior: local coordination topology is more susceptible to density than temperature, leading to further structural diversity. These results establish density diversity as a design principle for thermodynamically transferable MLIPs and provide a validation framework for assessing the thermodynamic coverage of both foundation and from-scratch models, enabling reliable atomistic simulation of fluid-phase processes across diverse operating conditions.

physics.chem-ph↗

arXiv2026

Visualizing Millisecond Atomic Dynamics of Nanocrystals in Liquid

Atomic structures of nanomaterials are inherently dynamic, continuously reshaped through interactions with chemical species and external stimuli. Such dynamics are further amplified as the size and dimensionality of nanomaterials are reduced. Despite advances in analytical methods, it remains challenging to capture structural dynamics of nanomaterials in reactive environments with both atomic spatial resolution and commensurate temporal resolution. Here, we directly visualize atomic-scale dynamics of gold (Au) nanocrystals in reactive liquid environments with millisecond-speed liquid cell electron microscopy (EM) and deep-learning denoising. We uncover reversible fluctuations in local crystallinity of Au nanocrystals dependent on the surrounding chemical environment. These transient fluctuations, driven by interactions at nanocrystal-liquid interfaces, critically influence dissolution kinetics and grain boundary relaxation. By overcoming the spatiotemporal limitations in conventional liquid cell EM, our findings provide insights into how transient nanoscale structures dictate the stability and reactivity of nanomaterials.

cond-mat.mtrl-sci↗

arXiv2025

Ionic Liquid Molecular Dynamics Simulation with Machine Learning Force Fields: DPMD and MACE

Machine learning force fields (MLFFs) are gaining attention as an alternative to classical force fields (FFs) by using deep learning models trained on density functional theory (DFT) data to improve interatomic potential accuracy. In this study, we develop and apply MLFFs for ionic liquids (ILs), specifically PYR14BF4 and LiTFSI/PYR14TFSI, using two different MLFF frameworks: DeePMD (DPMD) and MACE. We find that high-quality training datasets are crucial, especially including both equilibrated (EQ) and non-equilibrated (nEQ) structures, to build reliable MLFFs. Both DPMD and MACE MLFFs show good accuracy in force and energy predictions, but MACE performs better in predicting IL density and diffusion. We also analyze molecular configurations from our trained MACE MLFF and notice differences compared to pre-trained MACE models like MPA-0 and OMAT-0. Our results suggest that careful dataset preparation and fine-tuning are necessary to obtain reliable MLFF-based MD simulations for ILs.

physics.chem-ph↗

arXiv2024

Hierarchical Framework for Retrosynthesis Prediction with Enhanced Reaction Center Localization

Retrosynthesis is essential for designing synthetic pathways for complex molecules and can be revolutionized by AI to automate and accelerate chemical synthesis planning for drug discovery and materials science. Here, we propose a hierarchical framework for retrosynthesis prediction that systematically integrates reaction center identification, action prediction, and termination decision into a unified pipeline. Leveraging a molecular encoder pretrained with contrastive learning, the model captures both atom and bond level representations, enabling accurate identification of reaction centers and prediction of chemical actions. The framework addresses the scarcity of multiple reaction center data through augmentation strategies, enhancing the ability of the model to generalize to diverse reaction scenarios. The proposed approach achieves competitive performance across benchmark datasets, with notably high topk accuracy and exceptional reaction center identification capabilities, demonstrating its robustness in handling complex transformations. These advancements position the framework as a promising tool for future applications in material design and drug discovery.

physics.chem-ph↗

arXiv2024

Anomalous Water Penetration in $\text{Al}^{3+}$ Dissolution

The physicochemical characterization of trivalent ions is limited due to a lack of accurate force fields. By leveraging the latest machine learning force field to model aqueous $\text{AlCl}_{3}$, we discover that upon dissolution of $\text{Al}^{3+}$, water molecules beyond the second hydration shell involve in the hydration process. A combination of scissoring of coordinating water is followed by synchronized secondary motion of water in the second solvation shell due to hydrogen bonding. Consequently, the water beyond the second solvation penetrates through the second solvation shell and coordinates to the $\text{Al}^{3+}$. Our study reveals a novel microscopic understanding of solvation dynamics for trivalent ion.

physics.chem-ph↗

arXiv2024

Materials Discovery with Extreme Properties via Reinforcement Learning-Guided Combinatorial Chemistry

The goal of most materials discovery is to discover materials that are superior to those currently known. Fundamentally, this is close to extrapolation, which is a weak point for most machine learning models that learn the probability distribution of data. Herein, we develop reinforcement learning-guided combinatorial chemistry, which is a rule-based molecular designer driven by trained policy for selecting subsequent molecular fragments to get a target molecule. Since our model has the potential to generate all possible molecular structures that can be obtained from combinations of molecular fragments, unknown molecules with superior properties can be discovered. We theoretically and empirically demonstrate that our model is more suitable for discovering better compounds than probability distribution-learning models. In an experiment aimed at discovering molecules that hit seven extreme target properties, our model discovered 1,315 of all target-hitting molecules and 7,629 of five target-hitting molecules out of 100,000 trials, whereas the probability distribution-learning models failed. Moreover, it has been confirmed that every molecule generated under the binding rules of molecular fragments is 100% chemically valid. To illustrate the performance in actual problems, we also demonstrate that our models work well on two practical applications: discovering protein docking molecules and HIV inhibitors.

q-bio.BM↗

arXiv2023

Electric Double Layer from Phase Demixing Reinforced by Strong Coupling Electrostatics

Ionic liquids (ILs) are appealing electrolytes for their favorable physicochemical properties. However, despite their longstanding use, understanding the capacitive behavior of ILs remains challenging. This is largely due to the formation of a non-conventional electric double layer (EDL) at the electrode-electrolyte interface. This study shows that the short-range Yukawa interactions, representing the large anisotropically charged ILs, demix IL to create a spontaneous surface charge separation, which is reinforced by the strongly coupled charge interaction. The properties of the condensed layer, the onset of charge separation, and the rise of overscreening and crowding critically depend on the asymmetry of Yukawa interactions.

cond-mat.soft↗

arXiv2023

Anomalous Diffusion of Lithium-Anion Clusters in Ionic Liquids

Lithium-ion transport is significantly retarded in ionic liquids (ILs). In this work, we performed extensive molecular dynamics (MD) simulations to mimic the kinetics of lithium ions in ILs using [\emph{N}-methyl-\emph{N}-propylpyrrolidium (pyr$_{13}$)][bis(trifluoromethanesulfonyl)imide (Ntf$_{2}$)] with added LiNtf$_{2}$ salt. And we analyzed their transport, developing a two-state model and comparing it to the machine learning-identified states. The transport of lithium ions involves local shell exchanges of the Ntf$_{2}$ in the medium. We calculated train size distributions over various time scales. The train size distribution decays as a power law, representing non-Poissonian bursty shell exchanges. We analyzed the non-Poissonian processes of lithium ions transport as a two-state (soft and hard) model. We analytically calculated the transition probability of the two-state model, which fits well to the lifetime autocorrelation functions of LiNtf$_{2}$ shells. To identify two states, we introduced the graph neutral network incorporating local molecular structure. The results reveal that the shell-soft state mainly contributes to the transport of the lithium ions, and their contribution is more important in low temperatures. Hence, it is the key for enhanced lithium ion transport to increase the fraction of the shell-soft state.

physics.comp-ph↗

arXiv2022

Optimal Planning of Hybrid Energy Storage Systems using Curtailed Renewable Energy through Deep Reinforcement Learning

Energy management systems (EMS) are becoming increasingly important in order to utilize the continuously growing curtailed renewable energy. Promising energy storage systems (ESS), such as batteries and green hydrogen should be employed to maximize the efficiency of energy stakeholders. However, optimal decision-making, i.e., planning the leveraging between different strategies, is confronted with the complexity and uncertainties of large-scale problems. Here, we propose a sophisticated deep reinforcement learning (DRL) methodology with a policy-based algorithm to realize the real-time optimal ESS planning under the curtailed renewable energy uncertainty. A quantitative performance comparison proved that the DRL agent outperforms the scenario-based stochastic optimization (SO) algorithm, even with a wide action and observation space. Owing to the uncertainty rejection capability of the DRL, we could confirm a robust performance, under a large uncertainty of the curtailed renewable energy, with a maximizing net profit and stable system. Action-mapping was performed for visually assessing the action taken by the DRL agent according to the state. The corresponding results confirmed that the DRL agent learns the way like what a human expert would do, suggesting reliable application of the proposed methodology.

cs.LG↗

arXiv2021

Generative Chemical Transformer: Neural Machine Learning of Molecular Geometric Structures from Chemical Language via Attention

Discovering new materials better suited to specific purposes is an important issue in improving the quality of human life. Here, a neural network that creates molecules that meet some desired conditions based on a deep understanding of chemical language is proposed (Generative Chemical Transformer, GCT). The attention mechanism in GCT allows a deeper understanding of molecular structures beyond the limitations of chemical language itself which cause semantic discontinuity by paying attention to characters sparsely. It is investigated that the significance of language models for inverse molecular design problems by quantitatively evaluating the quality of the generated molecules. GCT generates highly realistic chemical strings that satisfy both chemical and linguistic grammar rules. Molecules parsed from generated strings simultaneously satisfy the multiple target properties and vary for a single condition set. These advances will contribute to improving the quality of human life by accelerating the process of desired material discovery.

cs.LG↗

arXiv2021

A2I Transformer: Permutation-equivariant attention network for pairwise and many-body interactions with minimal featurization

The combination of neural network potential (NNP) with molecular simulations plays an important role in an efficient and thorough understanding of a molecular system's potential energy surface (PES). However, grasping the interplay between input features and their local contribution to NNP is growingly evasive due to heavy featurization. In this work, we suggest an end-to-end model which directly predicts per-atom energy from the coordinates of particles, avoiding expert-guided featurization of the network input. Employing self-attention as the main workhorse, our model is intrinsically equivariant under the permutation operation, resulting in the invariance of the total potential energy. We tested our model against several challenges in molecular simulation problems, including periodic boundary condition (PBC), $n$-body interaction, and binary composition. Our model yielded stable predictions in all tested systems with errors significantly smaller than the potential energy fluctuation acquired from molecular dynamics simulations. Thus, our work provides a minimal baseline model that encodes complex interactions in a condensed phase system to facilitate the data-driven analysis of physicochemical systems.

physics.comp-ph↗

arXiv2020

Active microrheology of a bulk metallic glass

The glass transition remains unclarified in condensed matter physics. Investigating the mechanical properties of glass is challenging because any global deformation that may result in shear rejuvenation requires an astronomical relaxation time. Moreover, it is well known that a glass is heterogeneous and a global perturbation cannot explore local mechanical/transport properties. However, an investigation based on a local probe, i.e. microrheology, may overcome these problems. Here, we establish active microrheology of a bulk metallic glass: a probe particle driven into host medium glass. This is a technique amenable for experimental investigations. We show that upon cooling the microscopic friction exhibits a second-order phase transition; this sheds light on the origin of friction in heterogeneous materials. Further, we provide distinct evidence to demonstrate that a strong relationship exists between the microscopic dynamics of the probe particle and the macroscopic properties of the host medium glass. These findings establish active microrheology as a promising technique for investigating the local properties of bulk metallic glass.

cond-mat.soft↗

arXiv2019

Topological extension of the isomorph theory based on the Shannon entropy

Isomorph theory is one of the promising theories to understand the quasi-universal relationship between thermodynamic, dynamic and structural characteristics. Based on the hidden scale invariance of the inverse power law potentials, it rationalizes the excess entropy scaling law of dynamic properties. This work aims to show that this basic idea of isomorph theory can be extended by examining the microstructural features of the system. Using the topological framework in conjunction with the entropy calculation algorithm, we demonstrate that Voronoi entropy, a measure of the topological diversity of single atoms, provides a scaling law for the transport properties of soft-sphere fluids, which is comparable to the frequently used excess entropy scaling. By examining the relationship between the Voronoi entropy and the solid-like fraction of simple fluids, we suggest that the Frenkel line, a rigid-nonrigid crossover line, {be} a topological isomorphic line where the scaling relation qualitatively changes.

cond-mat.stat-mech↗

arXiv2019

A corresponding-state framework for the structural transition of supercritical fluids across the Widom delta

This work proposes a classification algorithm based on the radical Voronoi tessellation to define the Widom delta, supercritical gas-liquid coexistence region, of polyatomic molecules. In specific, we use a weighted mean-field classification method to classify a molecule into either gas-like or liquid-like. Classical percolation theory methods are adopted to understand the generality of the structural transition and to locate the Widom delta. A structural analysis on various supercritical fluids shows that the proposed method detects the influence of the attractive interaction on the structural transition of supercritical fluids. Moreover, we demonstrate that the supercritical gas-liquid coexistence region of water overlap with the ridges of the response function maxima. From the pressure-temperature relation, a three-parameter corresponding state theorem is derived, which states that the fraction of gas-like molecules of a substance is equal to that of another if their reduced pressure, reduced temperature and the critical compressibility factor are the same.

cond-mat.stat-mech↗

arXiv2019

Universality, scaling and collapse in supercritical fluids

The Supercritical Fluid (SCF) is known to exhibit salient dynamic and thermodynamic crossovers and inhomogeneous molecular distribution. But the question as to what basic physics underlies these microscopic and macroscopic anomalies remains open. Here, using an order parameter extracted by machine learning, the fraction of gas-like (or liquid-like) molecules, we find simplicity and universality in SCF: First, all isotherms of a given fluid collapse onto a single master curve described by a scaling relation. The observed power law holds from the high-temperature and pressure regime down to the critical point where it diverges. Second, phase diagrams of different compounds collapse onto their master curves by the same scaling exponent, thereby demonstrating a putative law of corresponding supercritical states in simple fluids. The reported results support a model of the SCF as a mixture of two interchangeable microstates, whose spatiotemporal dynamics gives rise to unique macroscopic properties.

cond-mat.soft↗

arXiv2018

Topological generalization of the rigid-nonrigid transition in soft-sphere and hard-sphere fluids

A fluid particle changes its dynamics from diffusive to oscillatory as the system density increases up to the melting density. Hence, the notion of the Frenkel line was introduced to demarcate the fluid region into rigid and nonrigid liquid subregions based on the collective particle dynamics. In this work, we apply a topological framework to locate the Frenkel lines of the soft-sphere and the hard-sphere models relying on the system configurations. The topological characteristics of the ideal gas and the maximally random jammed state are first analyzed, then the classification scheme designed in our earlier work is applied. The classification result shows that the fraction of solid-like atoms increases from zero to one in the rigid liquid region. The dependence of the solid-like fraction on the bulk density is understood based on the theory of fluid polyamorphism. The percolation behavior of solid-like clusters is described based on the fraction of solid-like molecules in an integrated manner. The crossover densities are obtained by examining the percolation of solid-like clusters. The resultant crossover densities of soft-sphere fluids converge to that of hard-sphere fluid. Hence, the topological method successfully highlights the generality of the Frenkel line.

cond-mat.soft↗

arXiv2018

Topological Characterization of Rigid-Nonrigid Transition across the Frenkel Line

The dynamics of supercritical fluids, a state of matter beyond the gas-liquid critical point, changes from diffusive to oscillatory motions at high pressure. This transition is believed to occur across a locus of thermodynamic states called the Frenkel line. The Frenkel line has been extensively investigated from the viewpoint of the dynamics, but its structural meaning is not still well understood. This letter interprets the mesoscopic picture of the Frenkel line entirely based on a topological and geometrical framework. This discovery makes it possible to understand the mechanism of rigid/non-rigid transition based not on the dynamics of individual atoms, but on their instantaneous configurations. The topological classification method reveals that the percolation of solid-like structures occurs above the rigid-nonrigid crossover densities.

cond-mat.stat-mech↗

arXiv2018

"Two-phase" thermodynamics of the Frenkel line

The Frenkel line, a crossover line between rigid and nonrigid dynamics of fluid particles, has recently been the subject of intense debate regarding its relevance as a partitioning line of supercritical phase, where the main criticism comes from the theoretical treatment of collective particle dynamics. From an independent point of view, this Letter suggests that the two-phase thermodynamics model may alleviate this contentious situation. The model offers new criteria for defining the Frenkel line in the supercritical region and builds a robust connection among the preexisting, seemingly inconsistent definitions. In addition, one of the dynamic criteria locates the rigid-nonrigid transition of the soft-sphere and the hard-sphere models. Hence, we suggest the Frenkel line be considered as a dynamic rigid-nonrigid fluid boundary, without any relation to gas-liquid transition. These findings provide an integrative viewpoint combining fragmentized definitions of the Frenkel line, allowing future studies to be carried out in a more reliable manner.

cond-mat.stat-mech↗