Kazuma Seike, Hirotaka Manaka, Yoko Miura . Causal inference in statistics insights into stress-induced ferroelectric states in SrTiO 3 : disentangling piezoelectric and flexoelectric effects2025 · 担当区分： 責任著者 記述言語： 英語 掲載種別： 研究論文（学術雑誌） 出版者・発行元： Taylor & Francis In materials science, experimental conditions are precisely controlled to ensure high reproducibility. This property is well suited for causal inference in statistics, yet its potential remains unrealized. In this study, we integrate causal inference with structural equation modeling (SEM) to analyze birefringence images of the stress-induced ferroelectric SrTiO 3. Random forest analysis identified retardance at 14.1 K, 𝛿(14.1K), as a key predictor of the ferroelectric phase transition temperature, 𝑇F. SEM revealed strong correlations between 𝑇F and 𝛿s, although multicollinearity in 𝛿s necessitated sparse principal component analysis to transform 𝛿(14.1K) and 𝛿(40.0K) into two independent components, 𝑃𝐶1 and 𝑃𝐶2, for subsequent causal analysis. Directed acyclic graphs (DAGs) based on SEM helped infer causal relationships, revealing 𝑃𝐶1‘s predominant influence on 𝑇F in stress/strain-concentrated regions (cluster E12) and 𝑃𝐶2‘s influence in uniform stress/strain regions (cluster E34). Two-model learner (T-Learner) analysis revealed the factors behind the higher 𝑇F in E12 than in E34. Specifically, the difference in magnitude of the piezoelectric effect, which occurs uniformly throughout the substrate, causes 𝑇F to increase by 0.36 K [0.18 K, 0.54 K], while the flexoelectric effect, which occurs only in E12, causes it to increase by an additional 1.49 K [1.23 K, 1.75 K]. These findings demonstrate the utility of causal inference in disentangling piezoelectric and flexoelectric effects in the stress-induced ferroelectric SrTiO 3. DOI： 10.1080/27660400.2025.2503698 Web of Science researchmap
Hirotaka Manaka, Shoutarou Katayama, Soichiro Honda, Yoko Miura . Deep learning framework for analyzing birefringence imaging by incorporating optical polarization overlap in stress-induced fe2025 · 担当区分： 筆頭著者, 責任著者 記述言語： 英語 掲載種別： 研究論文（学術雑誌） 出版者・発行元： Taylor & Francis Optical microscopy is vital in many scientific fields, and various super-resolution techniques have been developed to overcome the resolution limit that restricts the separation of spatially mixed light. However, conventional methods inherently cannot resolve overlapping optical polarization (OP) components, limiting the ‘polarization resolution’ in polarized light microscopy. Instead of quantitatively evaluating ‘polarization resolution’, this study aims to reliably separate intrinsic OP states based on consistent clustering results that are robust to variations in the spatial receptive field (SRF) size. We integrate statistical analysis, machine learning, and deep learning to evaluate overlapping OP states in temperature-dependent birefringence imaging of the stress-induced ferroelectric SrTiO 3 under an external force of 231 MPa. A long short-term memory (LSTM) network is used to extract temperature-dependent features from sequential image data, which effectively captures subtle changes in structural and ferroelectric phase transitions. A 3D convolutional autoencoder (3DCAE) learns spatial relationships between adjacent pixels from these temperature-dependent features, addressing OP overlap at different spatial scales based on different SRF sizes. Although the 3DCAE output considerably depends on the SRF size, clustering results obtained via temperature series forest (Tsf) analysis are highly consistent. This robustness indicates that the extracted OP states reflect physically meaningful spatial distributions rather than convolution artifacts. The proposed sequential analytical framework successfully reconstructs intrinsic OP distributions while balancing local and global structural features, providing a robust foundation for OP-sensitive imaging in materials science. DOI： 10.1080/27660400.2025.2568376 Web of Science researchmap
Hirotaka Manaka, Jun Ishikawa, Yoko Miura . Bayesian-inspired hierarchical modeling of angle-dependent electron paramagnetic resonance spectra in (C 2 H 5 NH 3 ) 2 CuCl 4 . SCIENCE AND TECHNOL2025 · 担当区分： 筆頭著者, 責任著者 記述言語： 英語 掲載種別： 研究論文（学術雑誌） 出版者・発行元： Taylor & Francis Electron paramagnetic resonance (EPR) spectroscopy is a powerful technique for probing magnetic and structural properties in functional materials. However, conventional least-squares analysis methods often fail to propagate uncertainty across related measurements, such as angle-resolved spectra. To address this, we employed a Bayesian-inspired hierarchical modeling framework to consistently estimate spectral parameters across multiple datasets. This framework was implemented for asymmetric EPR spectra of the layered perovskite (C₂H₅NH₃)₂CuCl₄, which exhibits two-dimensional magnetism and potential multiferroicity. Spectral parameters, including g-values and linewidths, were comprehensively estimated across 24 angular datasets. For temperature dependence, a reduced model over 71 temperature points yielded consistent estimates at lower computational cost. These results indicate a structural phase transition at a critical temperature (Tc). Simultaneous scaling-law fits of the temperature dependence of the CuCl6 octahedral tilt angle on both sides of the transition, assuming a common Tc, yielded critical parameters of βh = 0.38 for T > Tc and βl = 0.37 for T < Tc, with Tc = 206.2 K. Although the 95% Bayesian credible intervals were too broad to assign a definitive universality class, the results support a second-order phase transition. To interpret spectral asymmetry, we tested a twin-domain model, which proved insufficient. In contrast, a direct maximum a posteriori (MAP) optimization incorporating absorption and dispersion components successfully reproduced the observed spectra and yielded physically plausible parameters. These results demonstrate that this Bayesian-inspired hierarchical modeling framework provides a practical basis for uncertainty quantification, model evaluation, and structural interpretation in EPR spectroscopy. DOI： 10.1080/27660400.2025.2609362 Web of Science researchmap
Hirotaka Manaka, Kensei Toyoda, Yoko Miura . Multivariate temperature-series analysis of stress-induced ferroelectricity in SrTiO 3 : a machine learning approach with K -shape clustering and h2024 · 担当区分： 筆頭著者, 責任著者 記述言語： 英語 掲載種別： 研究論文（学術雑誌） 出版者・発行元： Taylor & Francis A new machine learning approach that transforms time-series analysis into temperature-series analysis is introduced to analyze stress-induced ferroelectricity in SrTiO3 at 231 MPa using birefringence images observed at successive temperatures. The spatial distribution of the temperature-series data for each of the 42,280 pixels was clustered using the multivariate 𝐾-shape clustering method based on the shape similarity of the temperature dependence. In addition, to obtain the structural and ferroelectric phase transition temperatures, 𝑇c and 𝑇F, hierarchical Bayesian temperature-series estimation was performed at each pixel (as a lower level) constrained over the entire cluster (as a higher level) considering the measurement error. Consequently, the K-shape clustering method revealed four clusters corresponding to elongated ferroelectric domains, explained by slight differences in retardance and fast-axis direction. Statistical analysis of the Bayesian posterior probability distribution showed a uniform distribution of 𝑇c over the sample, but an inhomogeneous distribution of 𝑇F. The higher 𝑇F regions exhibited a concentration of stress and/or strain. The Pearson correlation coefficient calculations suggested a strong to moderate relationship between the distribution of TF and the ferroelectric state, while the correlation between Tc and the ferroelectric state was weak or nonexistent. The combination of machine learning and statistics provides a more reliable and less arbitrary approach to analyzing temperature-series data. These multilevel analyses are particularly useful in studying critical phenomena near phase transition temperatures in condensed matter physics. DOI： 10.1080/27660400.2024.2342234 researchmap
