arXiv subjects

Teruyasu Mizoguchi

Publications and source records attributed to Teruyasu Mizoguchi.

35 records · Page 2Linked to original sources

arXiv2026

Towards Agentic Intelligence for Materials Science

The convergence of artificial intelligence and materials science presents a transformative opportunity, but achieving true acceleration in discovery requires moving beyond task-isolated, fine-tuned models toward agentic systems that plan, act, and learn across the full discovery loop. This survey advances a unique pipeline-centric view that spans from corpus curation and pretraining, through domain adaptation and instruction tuning, to goal-conditioned agents interfacing with simulation and experimental platforms. Unlike prior reviews, we treat the entire process as an end-to-end system to be optimized for tangible discovery outcomes rather than proxy benchmarks. This perspective allows us to trace how upstream design choices-such as data curation and training objectives-can be aligned with downstream experimental success through effective credit assignment. To bridge communities and establish a shared frame of reference, we first present an integrated lens that aligns terminology, evaluation, and workflow stages across AI and materials science. We then analyze the field through two focused lenses: From the AI perspective, the survey details LLM strengths in pattern recognition, predictive analytics, and natural language processing for literature mining, materials characterization, and property prediction; from the materials science perspective, it highlights applications in materials design, process optimization, and the acceleration of computational workflows via integration with external tools (e.g., DFT, robotic labs). Finally, we contrast passive, reactive approaches with agentic design, cataloging current contributions while motivating systems that pursue long-horizon goals with autonomy, memory, and tool use. This survey charts a practical roadmap towards autonomous, safety-aware LLM agents aimed at discovering novel and useful materials.

cond-mat.mtrl-sci↗

arXiv2026

Effect of uniaxial compressive stress on polarization switching and domain wall formation in tetragonal phase BaTiO3 via machine learning potential

Ferroelectric materials such as BaTiO3 exhibit spontaneous polarization that can be reoriented by an external electric field, forming the basis of various memory, actuator, and sensor applications. The polarization switching behavior, however, is strongly influenced by mechanical boundary conditions due to the intrinsic electromechanical coupling in ferroelectrics. In this study, we employ a machine learning interatomic potential to investigate the effect of uniaxial compressive stress on polarization switching and domain wall evolution in the tetragonal phase of BaTiO3. This study revealed a critical stress about 120 MPa which 90 degree polarization switching occurs. Beyond the critical stress, larger supercells exhibit lower activation energies for polarization switching with 180-degree domain wall formation and weaker constraints from periodic boundary conditions, thereby facilitating domain-wall formation. Besides, Increasing compressive stress reduces both the remnant polarization and the coercive field, while a double hysteresis loop emerges at a stress level of 80 MPa. These findings provide atomistic insights into stress-controlled ferroelectric switching and highlight the crucial role of mechanical loading in designing reliable ferroelectric devices.

cond-mat.mtrl-sci↗

arXiv2025

Why Physics Still Matters: Improving Machine Learning Prediction of Material Properties with Phonon-Informed Datasets

Machine learning (ML) methods have become powerful tools for predicting material properties with near first-principles accuracy and vastly reduced computational cost. However, the performance of ML models critically depends on the quality, size, and diversity of the training dataset. In materials science, this dependence is particularly important for learning from low-symmetry atomistic configurations that capture thermal excitations, structural defects, and chemical disorder, features that are ubiquitous in real materials but underrepresented in most datasets. The absence of systematic strategies for generating representative training data may therefore limit the predictive power of ML models in technologically critical fields such as energy conversion and photonics. In this work, we assess the effectiveness of graph neural network (GNN) models trained on two fundamentally different types of datasets: one composed of randomly generated atomic configurations and another constructed using physically informed sampling based on lattice vibrations. As a case study, we address the challenging task of predicting electronic and mechanical properties of a prototypical family of optoelectronic materials under realistic finite-temperature conditions. We find that the phonons-informed model consistently outperforms the randomly trained counterpart, despite relying on fewer data points. Explainability analyses further reveal that high-performing models assign greater weight to chemically meaningful bonds that control property variations, underscoring the importance of physically guided data generation. Overall, this work demonstrates that larger datasets do not necessarily yield better GNN predictive models and introduces a simple and general strategy for efficiently constructing high-quality training data in materials informatics.

cond-mat.mtrl-sci↗

arXiv2025

Electric Field-Induced Phase Transitions and Hysteresis in Ferroelectric HfO2 Captured with Machine Learning Potential

Electric field-induced studies, including phase transition and polarization hysteresis, for ferroelectric HfO2 at the atomic scale are critical since they can largely affect its application in ferroelectric and dielectric devices. However, conventional first-principles approaches are computationally limited in capturing large-scale atomic dynamics under realistic field conditions. Here, to enable electric-field-driven molecular dynamics simulations, we develop a machine learning potential (MLP) tailored for HfO2, coupled with an in-situ Born effective charge (BEC) model. This framework enables us to capture key phenomena, including field-induced phase transitions, polarization switching, and strain-dependent dielectric responses, with high fidelity and computational efficiency. Notably, we reproduce hysteresis loops and phase transition barriers consistent with AIMD results and reveal possible electric-field-induced polarization activation in the monoclinic phase. Our approach offers a scalable and transferable tool for atomistic exploration of functional oxides and paves the way for data-driven design of ferroelectric devices.

cond-mat.mtrl-sci↗

arXiv2025

Accelerated Inorganic Materials Design with Generative AI Agents

Designing inorganic crystalline materials with tailored properties is critical to technological innovation, yet current generative computational methods often struggle to efficiently explore desired targets with sufficient interpretability. Here, we present MatAgent, a generative approach for inorganic materials discovery that harnesses the powerful reasoning capabilities of large language models (LLMs). By combining a diffusion-based generative model for crystal structure estimation with a predictive model for property evaluation, MatAgent uses iterative, feedback-driven guidance to steer material exploration precisely toward user-defined targets. Integrated with external cognitive tools-including short-term memory, long-term memory, the periodic table, and a comprehensive materials knowledge base-MatAgent emulates human expert reasoning to vastly expand the accessible compositional space. Our results demonstrate that MatAgent robustly directs exploration toward desired properties while consistently achieving high compositional validity, uniqueness, and material novelty. This framework thus provides a highly interpretable, practical, and versatile AI-driven solution to accelerate the discovery and design of next-generation inorganic materials.

cond-mat.mtrl-sci↗

arXiv2025

Twisted oxide membrane interface by local atomic registry design

Interplay of lattice, orbital, and charge degrees of freedom in complex oxide materials has hosted a plethora of exotic quantum phases and physical properties. Recent advances in synthesis of freestanding complex oxide membranes and twisted heterostructures assembled from membranes provide new opportunities for discovery using moiré design with local lattice control. To this end, we designed moiré crystals at the coincidence site lattice condition, providing commensurate structure within the moiré supercell arising from the multi-atom complex oxide unit cell. We fabricated such twisted bilayers from freestanding SrTiO3 membranes and used depth sectioning-based TEM methods to discover ordered charge states at the moiré interface. By selectively imaging SrTiO3 atomic planes at different depths through the bilayer, we clearly resolved the moiré periodic structure at the twisted interface and found that it exhibits lattice-dependent charge disproportionation in the local atomic registry within the moiré supercell. Our density-functional modelling of the twisted oxide interface predicts that these moiré phenomena are accompanied by the emergence of a two-dimensional flat band that can drive new electronic phases. Our work provides a novel guideline for controlling moiré periodicity in twisted oxides and opens pathways to exploit the new functionalities via moiré lattice-driven charge-orbital correlation.

cond-mat.mtrl-sci↗

arXiv2024

Generative Inverse Design of Crystal Structures via Diffusion Models with Transformers

Recent advances in deep learning have enabled the generation of realistic data by training generative models on large datasets of text, images, and audio. While these models have demonstrated exceptional performance in generating novel and plausible data, it remains an open question whether they can effectively accelerate scientific discovery through the data generation and drive significant advancements across various scientific fields. In particular, the discovery of new inorganic materials with promising properties poses a critical challenge, both scientifically and for industrial applications. However, unlike textual or image data, materials, or more specifically crystal structures, consist of multiple types of variables - including lattice vectors, atom positions, and atomic species. This complexity in data give rise to a variety of approaches for representing and generating such data. Consequently, the design choices of generative models for crystal structures remain an open question. In this study, we explore a new type of diffusion model for the generative inverse design of crystal structures, with a backbone based on a Transformer architecture. We demonstrate our models are superior to previous methods in their versatility for generating crystal structures with desired properties. Furthermore, our empirical results suggest that the optimal conditioning methods vary depending on the dataset.

cond-mat.mtrl-sci↗

arXiv2024

Crystal orbital overlap population based on all-electron ab initio simulation with numeric atom-centered orbitals and its application to chemical-bonding analysis in Li-intercalated layered materials

Crystal Orbital Overlap Population (COOP) is one of the effective tools for chemical-bonding analysis, and thus it has been utilized in the materials development and characterization. In this study, we developed a code to perform the COOP-based chemical-bonding analysis based on the wavefunction obtained from a first principles all-electron calculation with numeric atom-centered orbitals. The chemical-bonding analysis using the developed code was demonstrated for F2 and Si. Furthermore, we applied the method to analyze the chemical-bonding changes associated with a Li intercalation in three representative layered materials: graphite, MoS2, and ZrNCl, because of their great industrial importance, particularly for the applications in battery and superconducting materials. The COOP analysis provided some insights for understanding the intercalation mechanism and the stability of the intercalated materials from a chemical-bonding viewpoint.

cond-mat.mtrl-sci↗

arXiv2024

LCAONet: Message-passing with physically optimized atomic basis functions

A Model capable of handling various elemental species and substances is essential for discovering new materials in the vast phase and compound space. Message-passing neural networks (MPNNs) are promising as such models, in which various vector operations model the atomic interaction with its neighbors. However, conventional MPNNs tend to overlook the importance of physicochemical information for each node atom, relying solely on the geometric features of the material graph. We propose the new three-body MPNN architecture with a message-passing layer that utilizes optimized basis functions based on the electronic structure of the node elemental species. This enables conveying the message that includes physical information and better represents the interaction for each elemental species. Inspired by the LCAO (linear combination of atomic orbitals) method, a classical method for calculating the orbital interactions, the linear combination of atomic basis constructed based on the wave function of hydrogen-like atoms is used for the present message-passing. Our model achieved higher prediction accuracy with smaller parameters than the state-of-the-art models on the challenging crystalline dataset containing many elemental species, including heavy elements. Ablation studies have also shown that the proposed method is effective in improving accuracy. Our implementation is available online.

physics.comp-ph↗

arXiv2022

interface_master: Python package building CSL and approximate CSL interfaces of any two lattices -- an effective tool for interface engineers

Crystalline interfaces are of highly importance in many practical applications. To conduct effective simulation and analysis for coincident site lattice (CSL) interfaces, effective programmes are of high demand in building their CSL bicrystal models to apply periodic boundary condition. The existing reported programmes capable suffer from problems including limitation of available lattice, poor capability for making heterogeneous interfaces and requirement of none-intuitive input parameters which cannot be directly obtained from an experimental observation result for an interface. This work presents a python programme to construct bicrystal model of interfaces consisting of any two lattices, including tilt, twist, mixed, and hetero interfaces and surfaces, which will support in effective simulation and analysis of crystalline interfaces.

cond-mat.mtrl-sci↗

arXiv2021

Robotic fabrication of high-quality lamellae for aberration-corrected transmission electron microscopy

Aberration-corrected scanning transmission electron microscopy (STEM) is widely used for atomic-level imaging of materials. To accelerate the discovery of new materials based on atomic-level investigations, the throughput of aberration-corrected STEM experiments becomes more and more important. However, the throughput of the full workflow of aberration-corrected STEM is still quite low. A fundamental problem is that the preparation of high-quality thin STEM samples (lamellae) depends on manual operation. Here, inspired by the recent successes of "robot scientists", we demonstrate robotic fabrication of high-quality lamellae by focused-ion-beam (FIB) with full automation software. First, we show that robotic FIB can prepare lamellae with a high success rate, where the robotic FIB controls rough-milling, lift-out, and final-thinning processes. Then, we optimize the FIB parameters of the final-thinning process for single crystal Si. Aberration-corrected STEM imaging of these Si lamellae shows atomic-level images with 55 pm resolution. We also demonstrate robotic fabrication of high-quality lamellae of SrTiO3 and sapphire. The robotic FIB system will resolve the current bottleneck of the full workflow of aberration-corrected STEM analysis and accelerate materials discovery based on atomic-level imaging.

cond-mat.mtrl-sci↗

arXiv2021

Accurate Prediction of Bonding Properties by A Machine Learning-based Model using Isolated States Before Bonding

Given the strong dependence of material structure and properties on the length and strength of constituent bonds and the fact that surface adsorption and chemical reactions are initiated by the formation of bonds between two systems, bonding parameters are of key importance for material design and industrial processes. In this study, a machine learning (ML)-based model is used to accurately predict bonding properties from information pertaining to isolated systems before bonding. This model employs the density of states (DOS) before bond formation as the ML descriptor and accurately predicts binding energy, bond distance, covalent electron amount, and Fermi energy even when only 20% of the whole dataset is used for training. The results show that the DOS of isolated systems before bonding is a powerful descriptor for the accurate prediction of bonding and adsorption properties.

cond-mat.mtrl-sci↗

arXiv2020

On the equivalence of molecular graph convolution and molecular wave function with poor basis set

In this study, we demonstrate that the linear combination of atomic orbitals (LCAO), an approximation of quantum physics introduced by Pauling and Lennard-Jones in the 1920s, corresponds to graph convolutional networks (GCNs) for molecules. However, GCNs involve unnecessary nonlinearity and deep architecture. We also verify that molecular GCNs are based on a poor basis function set compared with the standard one used in theoretical calculations or quantum chemical simulations. From these observations, we describe the quantum deep field (QDF), a machine learning (ML) model based on an underlying quantum physics, in particular the density functional theory (DFT). We believe that the QDF model can be easily understood because it can be regarded as a single linear layer GCN. Moreover, it uses two vanilla feedforward neural networks to learn an energy functional and a Hohenberg--Kohn map that have nonlinearities inherent in quantum physics and the DFT. For molecular energy prediction tasks, we demonstrated the viability of an ``extrapolation,'' in which we trained a QDF model with small molecules, tested it with large molecules, and achieved high extrapolation performance. This will lead to reliable and practical applications for discovering effective materials. The implementation is available at https://github.com/masashitsubaki/QuantumDeepField_molecule.

cs.LG↗

arXiv2020

Quantum deep field: data-driven wave function, electron density generation, and atomization energy prediction and extrapolation with machine learning

Deep neural networks (DNNs) have been used to successfully predict molecular properties calculated based on the Kohn--Sham density functional theory (KS-DFT). Although this prediction is fast and accurate, we believe that a DNN model for KS-DFT must not only predict the properties but also provide the electron density of a molecule. This letter presents the quantum deep field (QDF), which provides the electron density with an unsupervised but end-to-end physics-informed modeling by learning the atomization energy on a large-scale dataset. QDF performed well at atomization energy prediction, generated valid electron density, and demonstrated extrapolation. Our QDF implementation is available at https://github.com/masashitsubaki/QuantumDeepField_molecule.

physics.chem-ph↗

arXiv2019

Vibrational Effects in X-ray Absorption Spectra of 2D Layered Materials

With the examples of the C $K$-edge in graphite and the B $K$-edge in hexagonal BN, we demonstrate the impact of vibrational coupling and lattice distortions on the X-ray absorption near-edge structure (XANES) in 2D layered materials. Theoretical XANES spectra are obtained by solving the Bethe-Salpeter equation of many-body perturbation theory, including excitonic effects through the correlated motion of core-hole and excited electron. We show that accounting for zero-point motion is important for the interpretation and understanding of the measured X-ray absorption fine structure in both materials, in particular for describing the $σ^*$-peak structure.

cond-mat.mtrl-sci↗

arXiv2015

Effect of van der Waals interactions on the stability of SiC polytypes

Density functional theory calculations with a correction of the long-range dispersion force, namely the van der Waals (vdW) force, are performed for SiC polytypes. The lattice parameters are in good agreement with those obtained from experiments. Furthermore, the stability of the polytypes in the experiments, which show 3C-SiC as the most stable, are reproduced by the present calculations. The effect of the vdW force on the electronic structure and the stability of polytypes are discussed. We observe that the vdW interaction is more sensitive to the cubic site than the hexagonal site. Thus, the influence of the vdW force increases with decreasing the hexagonality of the polytype, which results in the confirmation that the most stable polytype is 3C-SiC.

cond-mat.mtrl-sci↗

arXiv2015

Prediction of grain boundary structure and energy by machine learning

Grain boundaries dramatically affect the properties of polycrystalline materials because of differences in atomic configuration. To fully understand the relationship between grain boundaries and materials properties, systematic studies of the grain boundary atomic structure are crucial. However, such studies are limited by the extensive computation necessary to determine the structure of a single grain boundary. If the structure could be predicted with more efficient computation, the understanding of the grain boundary would be accelerated significantly. Here, we predict grain boundary structures and energies using a machine-learning technique. Training data for non-linear regression of four symmetric-tilt grain boundaries of copper were used. The results of the regression analysis were used to predict 12 other grain boundary structures. The method accurately predicts both the structures and energies of grain boundaries. The method presented in this study is very general and can be utilized in understanding many complex interfaces.

cond-mat.mtrl-sci↗