Searcharxiv⌕ Search

arXiv subjects

Teruyasu Mizoguchi

Publications and source records attributed to Teruyasu Mizoguchi.

At least 19 recordsLinked to original sources
arXiv2026

Atomistic Origin and Strain Control of the Finite-Temperature Dielectric Response in BaTiO3

Soft-mode theory specifies a frequency, not what a local polar unit does. What real-space motion underlies the dielectric response of BaTiO3 therefore remains unclear. Using electric-field-induced molecular dynamics with a machine-learning force field, we show that the permittivity tracks the field-induced angular redistribution of local Ti-O off-centering, not its magnitude. Temperature and biaxial strain modify the local structure through different microscopic routes, yet both dielectric responses collapse onto a common relation with the same orientational descriptor, providing a real-space counterpart to soft-mode behavior.

cond-mat.mtrl-sci↗
arXiv2026

Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM Agents

Self-driving laboratories can explore synthesis conditions autonomously, but their decision-making layer is typically a black-box optimizer, and the output is a set of optimized samples, with the measurements reduced to predefined scalar objectives and the reasons behind success left unarticulated. Here we present SynAgent, a framework in which large language model agents operate an automated experimental system and maintain an explicit, revisable understanding of the synthesis process as the campaign's primary output. Starting with no predefined analysis pipeline, SynAgent adaptively generates analysis skills for newly acquired data and evolves this understanding through multimodal reasoning over experimental data such as X-ray diffraction patterns and electron micrographs. The evolution is guided by a verify-falsify scheme, in which the agent deliberately challenges its own hypotheses by testing conditions predicted to fail as well as those predicted to succeed. In a single campaign of 18 autonomous experiments using LiCoO2 (001) thin-film deposition as a testbed, SynAgent synthesized highly crystalline films and evolved an understanding of how the substrate temperature governs crystallization, discovering an abrupt threshold and a narrow optimal growth window at 650-690 °C. These results extend autonomous experimentation beyond optimized samples to testable, human-readable understanding.

cond-mat.mtrl-sci↗
arXiv2026

All-Electron Single-Atom Reference Correction for Absolute Transition Energies in Fixed-Reference PAW-XCH Calculations

The transition energy of a core-loss spectrum comprises a transferable atomic core-reference contribution and a material-dependent response that determines chemical shifts. Density functional theory calculations commonly underestimate absolute transition energies. In VASP, the same PAW dataset is used for the ground and core-excited occupations, retaining the ground-state atomic reference when the core occupation changes and thereby omitting the associated all-electron atomic reference change. We evaluate this change using self-consistent single-atom calculations and construct a residual atomic reference correction from the valence reference change and isolated-atom total-energy response. Comparisons with an independent ultrasoft-pseudopotential implementation demonstrate numerical consistency of the all-electron reference for K, L1, and L2,3 core holes, while representative corrected spectra approach the experimental absolute energy scale. For a fixed element, edge, and core-hole scheme, the atomic reference terms cancel from energy differences, showing that same-element chemical shifts are governed by the double difference of the supercell total energy. At the Al and Si L2,3 edges, the near-edge spectral shape additionally depends strongly on the PAW representation of low-lying 3d-like unoccupied states.

cond-mat.mtrl-sci↗
arXiv2026

Transient Detour and Cooperative Oxygen Exchange in the Polarization Switching of Ferroelectric Hf0.5Zr0.5O2

Hafnium zirconium oxide (HZO) has attracted significant attention as a core material for next-generation non-volatile memories due to its excellent ferroelectricity in the ultra-thin film regime and its CMOS process compatibility. However, the exploration of its polarization switching mechanism has predominantly relied on static energy barrier analyses, leaving the transient bond formation and cooperative dynamic mechanisms under actual electric field driving unresolved. In this study, we performed Electric-Field-Induced MD simulations on a defect-free ideal HZO lattice using a fine-tuned machine learning force field (MACEField). As a result, we successfully reproduced the P-E hysteresis loop dynamically and demonstrated that the polarization switching in HZO is driven not by conventional simple displacement models (S:N/S:T models), but by the dynamic mutual exchange of 3-coordinated oxygen (O3c) and 4-coordinated oxygen (O4c). Analysis of the oxygen atom displacement trajectories revealed that this pathway is accompanied by a unique "detour" behavior originating from transient cation-oxygen bond formation. Furthermore, we identified an "internal self-compensation mechanism" in which the local volumetric expansion and contraction accompanying the coordination number changes are effectively offset within the cell. These findings provide, from a dynamic perspective, a microscopic physical origin of for HZO's exceptional ability to sustain stable polarization switching without macroscopic strain, a property that has long distinguished HZO from conventional perovskite ferroelectrics yet lacked atomistic explanation. These findings suggest that preserving the integrity of cooperative O3c/O4c exchange pathways, rather than minimizing individual atomic displacements, is the key design principle for endurance and scalability in next-generation ferroelectric memories.

cond-mat.mtrl-sci↗
arXiv2026

Toward Auditable AI Scientists: A Hypothesis Evolution Protocol for LLM Agents

Large language model (LLM) agents are increasingly expected to play a central role in AI-driven scientific discovery. Equipped with broad knowledge, flexible reasoning, and tool use, they have the potential to autonomously explore and solve scientific problems by repeatedly proposing hypotheses, testing them, and revising their beliefs in the light of the evidence. In current agents, however, these hypotheses, tests, and belief updates are buried in unstructured logs, and no mechanism lets the agent or the human researcher audit that process. Here we propose the Hypothesis Evolution Protocol (HEP), an agent harness that provides hypothesis generation, evaluation, and evolution as explicit, auditable operations. On materials-science research tasks, a HEP-equipped agent operates the hypothesis--test--evidence--belief cycle that planning-style agents lack, generalizes across research questions, and exploits the protocol more fully as the base LLM becomes more capable. These results mark a step toward auditable AI scientists, whose scientific reasoning can be inspected, verified, and built upon.

cs.AI↗
arXiv2026

Intrinsic Defect Energetics and Fluorine Doping Effects in Li2CO3 and Li2O2: A First-Principles Study

Lithium carbonate, Li2CO3, is a thermodynamically stable carbonate phase whose defect energetics are closely related to its stability and decomposition behavior in various lithium-based electrochemical systems. These properties of Li2CO3 are particularly important in lithium-oxygen battery environments. In these systems, Li2CO3 can form as a parasitic discharge product alongside Li2O2, the primary discharge product, leading to performance degradation. However, compared with Li2O2, the intrinsic defect thermodynamics of Li2CO3 and how chemical doping modifies its defect energetics remain insufficiently understood. In this study, first-principles calculations were performed to systematically analyze the intrinsic point-defect energetics of Li2CO3 and to evaluate the effects of fluorine doping on vacancy formation energies in Li2CO3 and Li2O2. Intrinsic defect analysis reveals that defect behavior is predominantly governed by lithium-related defects. Upon fluorine doping, lithium and carbon vacancy formation energies decrease selectively in Li2CO3, partially destabilizing the carbonate framework, while a reduction in lithium vacancy formation energy is also observed in Li2O2. These results suggest that fluorine doping modulates the defect energetics of both discharge products, potentially providing a thermodynamic basis for controlling the stability of Li2CO3 and Li2O2 under thermodynamic conditions representative of lithium-oxygen batteries.

cond-mat.mtrl-sci↗
arXiv2026

Transition from Homogeneous to Domain-Wall-Mediated Polarization Switching in BaTiO3: A Machine-Learning Molecular Dynamics Study

Polarization switching in ferroelectric BaTiO3 can proceed through fundamentally different mechanisms - yet the conditions that determine which pathway is realized remain poorly understood. Using machine-learning potential-based molecular dynamics with the MACEField model, we systematically vary supercell size to reveal a clear transition from homogeneous polarization switching to domain-wall-mediated switching, accompanied by a coercive field increase of over 50%. Shannon entropy analysis demonstrates that this transition is driven by size-dependent polarization fluctuations that promote 180 degree domain-wall nucleation - establishing a direct, quantitative link between local configurational disorder and macroscopic switching behavior. Furthermore, the switching pathway and hysteresis response are shown to depend critically on supercell geometry and the relative orientation of applied stress and electric field. These findings reveal that homogeneous and domain-wall-mediated switching are distinct physical regimes in BaTiO3, and that atomistic simulations must account for system size to correctly capture the operative switching mechanism.

cond-mat.mtrl-sci↗
arXiv2026

Scale-Dependent Input Representation and Confidence Estimation for LLMs in Materials Property Prediction

Large language models (LLMs) are increasingly applied to materials science. However, the relationship between prediction accuracy, input representation, and model scale remains unclear, and reliable methods for assessing prediction confidence have not yet been established. In this study, we fine-tune two Llama models of different scales (1B and 8B) using low-rank adaptation (LoRA) on an inorganic crystal structure dataset. We systematically evaluate five input representations, namely chemical composition, crystal summary, local environment description, full text description, and crystallographic information files (CIF), for formation energy and bandgap prediction. Our results show that the optimal input representation depends on model scale. The 1B model performs better with compact representations, whereas the 8B model maintains high accuracy even with longer natural-language descriptions and CIF inputs. Across both model scales, crystal summaries that include space-group information consistently outperform composition-only inputs, indicating that symmetry information serves as a robust and informative feature. We further analyze the relationship between prediction error and the mean negative log-likelihood (mean NLL) of tokens corresponding to predicted numerical values. While no clear correlation is observed in base models, fine-tuned models exhibit a consistent trend in which lower mean NLL corresponds to smaller prediction errors. This result suggests that mean NLL can serve as a practical confidence indicator without requiring additional training. These findings demonstrate that both input representation and model scale play critical roles in LLM-based materials property prediction, and that mean NLL provides an effective and computationally efficient measure of prediction confidence.

cond-mat.mtrl-sci↗
arXiv2026

Inverse Materials Design via Joint Generation of Crystal Structures and Local Electronic Descriptors

Inverse design of inorganic crystals, in which structures are generated to satisfy a target property while preserving diversity and physical plausibility, remains more demanding than ab initio generation, as property conditioning can degrade the structural quality that current generative models otherwise achieve. We propose a diffusion framework that jointly denoises crystal-structure variables and site-resolved local electronic descriptors through a shared score network. As representative descriptors, we adopt Bader charge and atomic density of states (atomic DOS). Under both band-gap and formation energy conditioned generation, the joint models achieved higher success rates than the structure-only baseline in most target conditions, while simultaneously increasing the fraction of generated structures that satisfy uniqueness, novelty, thermodynamic stability, and physical validity (VSUN criteria). A dummy-variable control confirms that these gains originate from the electronic content of the descriptors rather than from auxiliary site-wise variables. The generated Bader charges agree with DFT references with an MAE of 5.5e-2 e on stable structures, and the generated atomic DOS captures the coarse spectral profile of the DFT reference around the modal accuracy range, although finer details and accuracy vary with elemental species. These results establish local electronic descriptors as effective generative variables that serve two complementary roles: broadening the explored materials space through increased structural diversity, and mitigating the trade-off between property targeting and structural quality by guiding the structural trajectory toward electronically plausible configurations during joint denoising.

cond-mat.mtrl-sci↗
arXiv2026

Ultrafast Sliding Ferroelectric Switching in Bilayer Hexagonal Boron Nitride Revealed by Deep Learning Molecular Dynamics

Sliding ferroelectricity in bilayer hexagonal boron nitride (h-BN) offers compelling prospects for next-generation non-volatile memory, yet the atomistic dynamics of electric-field-driven polarization switching remain poorly understood. Here, we present a fully data-driven, coupled atomistic framework that integrates a fine-tuned MACE machine learning potential (MLP) with an equivariant graph convolutional neural network (EGCNN) for real-time Born effective charge (BEC) prediction, enabling large-scale non-equilibrium molecular dynamics simulations of AB-stacked bilayer h-BN under applied electric fields. By implementing a rigorous real-space path-integral polarization formalism combined with a state-constrained Gaussian convolution background extraction procedure, we successfully isolate the intrinsic spontaneous polarization from the dominant dielectric background. Our simulations reveal that coherent single-domain rigid sliding, completing within 5 ps, constitutes a physically viable ultrafast switching mechanism, and reproduces clean ferroelectric hysteresis loops whose shape is qualitatively consistent with experimental observations.

cond-mat.mtrl-sci↗
arXiv2026

Long-range interaction effects on the phase transition, mechanical effect, and electric field response of BaTiO3 by machine learning potentials

Bulk materials are governed by both short-range and long-range interactions, both of which are naturally captured in conventional density functional theory (DFT) calculations through Ewald summation of electrostatic contributions. In contrast, machine learning potentials (MLPs) typically rely on local atomic environment descriptors, and long-range interactions are often neglected. Such approximations may introduce systematic energetic errors and lead to inaccuracies in predicted material properties. To systematically investigate the impact of long-range interactions in ferroelectric BaTiO3 within the framework of MLPs, we developed a long-range MACELES model and compared its performance with the previously reported BaTiO3 MACE model across four key properties (phonon dispersion, phase transition behavior, mechanical response, and ferroelectric properties including dielectric constants). We find that qualitative behaviors, including phase transitions, stress-induced polarization switching, and polarization-electric field hysteresis, are consistently reproduced by both models. In contrast, quantitative properties such as transition temperatures, elastic constants, and dielectric constants exhibit systematic improvements in MACELES model, highlighting the importance of incorporating long-range electrostatics for accurately describing the structural and dielectric responses of BaTiO3. These results suggest that while long-range interactions play a role in improving quantitative accuracy, their omission does not significantly alter the qualitative ferroelectric behavior of BaTiO3.

cond-mat.mtrl-sci↗
arXiv2026

Decoding Dopant-Induced Electronic Modulation in Graphene via Region-Resolved Machine Learning of XANES

Revealing how heteroatom doping alters the local electronic structure of graphene is crucial for understanding and controlling its functional properties. In this study, we combine density functional theory (DFT) and machine learning (ML) to interpret how boron (B) and nitrogen (N) dopants influence the local electronic environments of graphene. A dataset of 415 DFT-simulated XANES spectra from 91 distinct configurations was analyzed using a region-specific approach by decomposing each spectrum into pi*, sigma*, and post-edge regions. Random forest models trained on these spectral segments identified the pi* region as the most informative for predicting key local electronic descriptors, particularly the Bader charge and mean dopant-carbon bond length. The Bader charge quantifies dopant-induced charge redistribution and local bonding polarity, directly reflecting the degree of electronic perturbation introduced by heteroatom substitution. The enhanced predictive power of the pi* region arises from its strong coupling to the perturbed pi-electron network, which captures these charge-transfer and hybridization effects more effectively than sigma* or post-edge regions. These findings establish Bader charge as a robust and physically meaningful descriptor for quantifying dopant-induced electronic modulation and demonstrate that region-resolved ML analysis of XANES spectra provides a powerful pathway to uncover structure-property relationships in doped graphene and related materials.

cond-mat.mtrl-sci↗
arXiv2026

Twist-Angle Engineering of Moiré Potentials for High-Performance Ionics in Bilayer Graphene

Controlling ion transport is a fundamental challenge for advanced energy storage. Bilayer graphene offers a unique platform for modulating ion diffusion via twist-angle-dependent moire superlattices, yet conventional stacking configurations face an inherent trade-off: AA stacking provides stable Li intercalation but high diffusion barriers, while AB stacking enables fast diffusion but poor intercalation stability. Twisted bilayer graphene (tBLG) offers potential to overcome this limitation, yet systematic understanding across different twist angles remains limited. Here, we investigate Li intercalation in tBLG using first-principles density functional theory, evaluating intercalation energies and diffusion barriers across multiple twist angles through potential energy surface (PES) mapping. The Sigma 37 structure (9.43 degrees) simultaneously achieves the most favorable intercalation energy (-2.39 eV) and the lowest diffusion barrier (0.14 eV) among all structures examined, resolving the conventional stacking trade-off. Furthermore, using the Smooth Overlap of Atomic Positions (SOAP) descriptor, we demonstrate that the PES is governed by local atomic environments and that a model trained on limited structures predicts the PES of untested configurations with high accuracy. This transferability enables efficient screening without exhaustive first-principles calculations, establishing a systematic framework for twist-angle engineering of ion transport in two-dimensional layered materials.

cond-mat.mtrl-sci↗
arXiv2026

Origin of Reduced Coercive Field in ScAlN: Synergy of Structural Softening and Dynamic Atomic Correlations

Among wurtzite-type ferroelectrics, scandium-doped aluminum nitride (ScAlN) has emerged as a leading candidate for CMOS-compatible low-voltage memory, combining strong spontaneous polarization with process compatibility. A remarkable feature of this system is the pronounced reduction of the coercive field (Ec) with increasing Sc concentration; however, its microscopic origin remains poorly understood at the atomic scale, particularly under finite temperature and applied electric fields. Here, we integrate a density-functional-theory-accurate machine-learning force field with an equivariant neural-network-based Born effective charge model to perform large-scale electric-field-driven molecular dynamics simulations at near-first-principles accuracy. The framework correctly reproduces the experimentally observed qualitative trends in key experimental trends, including the decrease in the c/a ratio and the monotonic reduction of Ec with increasing Sc content. Beyond static structural softening, we uncover a dynamic mechanism underlying Ec reduction. Sc atoms exhibit larger thermal vibrations and undergo preceding displacements during switching, acting as dynamic triggers for polarization reversal. Moreover, the displacement correlation between Sc and Al atoms evolves systematically with composition, enhancing cooperative atomic rearrangements and lowering the effective switching barrier. These results demonstrate that Ec reduction in ScAlN arises from the synergy of structural softening and dynamic correlation evolution, providing a new perspective for designing hexagonal ferroelectrics.

cond-mat.mtrl-sci↗
arXiv2026

Decoupling structural and bonding effects on ferroelectric switching in ScAlN via molecular dynamics under an applied electric field

ScxAl1-xN has emerged as a promising wurtzite-type ferroelectric material, where increasing the Sc composition reduces both the coercive field (Ec) and remanent polarization (Pr). This composition-dependent behavior is physically attributed to two simultaneous changes: the increase in the internal structural parameter u (structural effect) and the weakening of bond strength (bonding effect). Because these factors are strongly coupled in experiments, their individual contributions to ferroelectric switching remain unclear. In this study, we systematically decoupled these effects using machine-learning force field-based molecular dynamics (MD) simulations under an applied electric field. By artificially tuning u via in-plane strain at a fixed composition, we demonstrated that Pr is determined exclusively by the structural effect, exhibiting a universal linear dependence regardless of the composition. In contrast, Ec deviated from this structural trend, implying an additional compositional contribution. To isolate this, we evaluated configurations with identical u but varying Sc compositions; Pr remained constant, whereas Ec systematically decreased due to bond weakening. Furthermore, static nudged elastic band (NEB) calculations revealed that the static switching barrier depends solely on u, failing to explicitly capture the bonding effect on Ec. These results establish that while Pr is governed strictly by the structural effect, Ec is determined by a superposition of structural and bonding effects. Our findings highlight the necessity of dynamic MD simulations for fully understanding ferroelectric switching in compositionally tunable materials.

cond-mat.mtrl-sci↗
arXiv2026

Achieving Robust Extrapolation in Materials Property Prediction via Decoupled Transfer Learning

Machine learning has revolutionized materials property prediction, yet fails catastrophically when extrapolating beyond training distributions-precisely the capability required for discovering unprecedented materials. Graph neural networks (GNNs) exhibit this collapse because end-to-end training fundamentally couples learned representations to target property distributions, preventing genuine extrapolation. We demonstrate that decoupled transfer learning-separating pretrained GNN feature extractors from simple regressors-overcomes this barrier. Pretrained features provide transferable structural knowledge, while simple regressors enable smooth extrapolation by maintaining learned trends beyond training boundaries. Benchmarked on layered intercalation compounds through four rigorous extrapolation scenarios and a temporal Materials Project split, our framework achieves 68% error reduction (RMSE: 0.881 vs. 2.778 eV/atom) versus end-to-end GNNs for extrapolation. Failure analysis reveals extrapolation succeeds for continuous chemical space but fails for discontinuous space, establishing clear design principles. Validated on Fermi energy prediction, this framework is immediately deployable using existing pretrained models, requiring no architectural innovations-transforming ML-driven materials discovery.

cond-mat.mtrl-sci↗
arXiv2026

Finite-size effects and energy alignment in molecular XANES under periodic boundary conditions: A systematic comparison of core-hole treatments

X-ray absorption near-edge structure (XANES) provides element-specific insight into local electronic and structural environments, but quantitative interpretation of molecular XANES under periodic boundary conditions (PBC) remains challenging due to finite-size effects and core-hole treatments. In this work, we systematically investigate how core-hole approximations and charge compensation schemes affect transition energies, energy alignment, and chemical-shift reproducibility in PBC-DFT-based molecular XANES calculations. Using ethane as a model system, we show that the full core-hole (FCH) approach exhibits pronounced supercell-size dependence originating from interactions between background charge and charged molecules, with transition energies largely changed by leading-order finite-size terms. In contrast, the excited core-hole (XCH) method rapidly converges owing to its neutral final state. We further demonstrate that most finite-size effects in FCH can be removed by Makov-Payne corrections based on multipole expansion of the electrostatic energy of charged supercells under PBC. Furthermore, we propose a simple Fermi-level-based energy correction (EF/2) that provides comparable improvement using only a single supercell. Extending the analysis to an n-alkane series reveals that while intrinsic electronic-structure changes govern peak shifts for small molecules, systematic energy drifts persist in FCH for larger molecules, whereas XCH and FCH+EF/2 remain stable. Finally, for small molecules at the C and N K-edges, XCH and FCH+EF/2 accurately reproduce experimental chemical shifts, whereas uncorrected FCH fails. These results provide practical guidelines for reliable energy alignment and chemical-shift analysis in molecular XANES under PBC, supporting robust applications to molecular, adsorption, and interfacial systems.

cond-mat.mtrl-sci↗
arXiv2026

Generative Inverse Estimation of 3D Atomic Coordination from Near-Edge Spectra via Equivariant Diffusion Models

Extracting 3D atomic coordinates from spectroscopic data is a longstanding inverse problem. We present an equivariant diffusion model that generates site-specific 3D structures directly from near-edge spectra (ELNES/XANES). Trained on Si-O crystals, the model achieves radial accuracy comparable to Extended X-ray Absorption Fine Structure (EXAFS) (RMSD ~0.06 Å) but with superior coordination number precision (errors < 4.3% vs. EXAFS ~20%). Crucially, it reconstructs full 3D geometries including bond angles, overcoming the limitations of 1D radial distribution analysis. The model demonstrates robust out-of-distribution generalization, accurately predicting local structures in amorphous systems despite being trained exclusively on crystalline lattices. Application to experimental O K-edge spectra from α-quartz validates practical applicability. This generative approach outperforms template matching and establishes automated, quantitative 3D structure determination from spectroscopic data.

cond-mat.mtrl-sci↗