All-inorganic double perovskites are highly promising optoelectronic functional materials, yet their mechanical properties and microscopic failure characteristics are critical to device operational stability. Using first-principles calculations, we systematically investigate the elastic mechanical response, large-deformation failure mechanisms, and fatigue service behavior of Cs2NaGaX6 (X = Cl, Br). Emphasis is placed on elucidating how halogen coordination regulates the brittle-to-ductile transition and the distinct atomic-scale failure mechanisms of the two halide variants. Both systems are structurally stable; Br− substitution expands the lattice, narrows the band gap, and weakens the intrinsic bonding, effectively enhancing ductility, as indicated by the Pugh and Poisson's ratio criteria and increasing elastic anisotropy. Tensile failure is consistently dominated by the rupture of intrinsic weak Ga-X bonds, independent of halogen species. In contrast, shear failure mechanisms diverge markedly: Cs2NaGaCl6 fails primarily through NaCl ionic bond sliding and dissociation, whereas Cs2NaGaBr6 undergoes preferential GaBr covalent bond rupture. This divergence originates from differences in lattice flexibility and the redistribution of bond-level stress. High-cycle fatigue analysis reveals that the Cl-based brittle system exhibits higher cyclic stress tolerance, while the Br-based high-toughness system is more susceptible to cumulative plastic damage. These findings establish a halogen-coordination-to-mechanical-performance linkage, providing theoretical guidance for durability-oriented design of lead-free perovskite optoelectronics.
Metal-organic frameworks (MOFs) are excellent adsorbents for CO2. The specific surface area (SSA) and CO2 adsorption capacity of MOFs are critical to evaluating their performance of gas adsorption. However, existing experimental investigations and common theoretical calculations are often resource- and time-intensive for high-throughput measurements of MOFs properties. The predictive models based on traditional machine learning algorithms require manual extraction of features to construct databases. These conventional methods hinder more efficient screening of ideal MOFs for CO2 adsorption. In this study, we propose a novel lightweight deep learning model (MOFNet) to accurately predict SSA and CO2 adsorption capacity of MOFs. The proposed MOFNet model integrates deep learning algorithms with adaptive attention and convolutional block attention mechanisms. It innovatively accepts hybrid feature fusion inputs including structural images of MOFs and derived statistical features. The MOFNet simultaneously achieves accurate prediction of SSA and CO2 adsorption capacity, the corresponding coefficients of determination (R2) are 0.968 and 0.931, respectively, outperforming 9 popular baseline models. We also successfully extend the proposed MOFNet to estimate the H2 adsorption capacity and binary adsorption capacity of CO2/N2 mixture, with hybrid feature fusion serving as the model input. The R2 values for the H2 adsorption capacity and CO2/N2 selectivity are 0.988 and 0.867, respectively. Finally, the actual screening ability of the model we proposed is verified by collected experimental SSA data and CO2 adsorption capacity data of bimetallic MOFs synthesized in this study.
The mechanistic role of defect stress fields in modulating solute segregation and heterogeneous nucleation in AlCu alloys remains unclear. Here, we develop a multiphase-field model that couples the Cu concentration field, θ' precipitates, and elastic stress fields to investigate the effects of dislocation lines, low-angle grain boundaries, and coarse θ phases, supplemented by dislocation loop cases to demonstrate the design potential. The simulations quantitatively reproduce the orientation-dependent nucleation and growth of θ' variants, which are validated by STEM observations. A universal cascade mechanism is established: defect stress fields drive directional Cu segregation into enriched zones, which serve as the compositional precursors governing nucleation sites, variant orientation, and growth kinetics. Quantitative analysis further reveals that the spatial distribution and intensity of the stress fields directly influence the degree of solute segregation, while their spatial geometric characteristics lead to significant differences in both the average diameter and volume fraction of θ' precipitates. This study demonstrates that tailoring the geometric features of stress fields offers a viable route for the deliberate design of precipitation microstructures, providing a general framework extendable to other precipitation-hardened alloy systems.
Metal halide perovskites (MHPs) have emerged as promising optoelectronic materials owing to their superior photoelectric properties, among which surface defect tolerance is a fundamental determinant of surface electronic quality and device performance. As a thermally stable all-inorganic perovskite, γ-CsPbI3 suffers from abundant intrinsic surface point defects that severely impair its optoelectronic functionality. However, the termination-dependent defect tolerance and layer-resolved defect behaviors of γ-CsPbI3 remain insufficiently understood. Herein, we systematically investigate the intrinsic point defects in the outermost two atomic layers of γ-CsPbI3 (001) and (110) surfaces with dual CsI and PbI2 terminations using first-principles calculations. The results reveal that γ-CsPbI3 surfaces exhibit distinctly different defect characteristics compared with its bulk counterpart and possess generally poor defect tolerance, with only the CsI-terminated (110) surface showing relatively benign defect behavior. The predominant VI defect maintains a relatively low equilibrium concentration relative to other detrimental deep-level defects under most growth conditions. For the (001) surface, the high-density deep-level recombination centers correspond to first-layer PbCs and Ii on the CsI-terminated facet and first-layer CsI on the PbI2-terminated facet, while only subsurface PbCs and Ii on the PbI2-terminated (001) surface require targeted passivation. In comparison, harmful deep-level defects on the (110) surface, including VPb, IPb, CsI, and PbI, are exclusively localized on the first layer of the PbI2-terminated facet. These defects exhibit low formation energies, deep transition levels and high trap densities, acting as dominant non-radiative recombination centers. Strictly focused on the γ-CsPbI3 system, this work clarifies the termination-dominated defect tolerance mechanism and differentiates the electronic impacts of surface and subsurface defects. The atomistic findings provide reliable, layer-resolved theoretical guidance for precise surface defect modulation and passivation engineering of γ-CsPbI3, offering an effective strategy to improve the intrinsic structural and electronic quality of all-inorganic CsPbI3 perovskite surfaces.
In multi-physics simulations governed by coupled partial differential equations, iteratively solving the mechanical equilibrium problem under elastic inhomogeneity constitutes the dominant computational bottleneck. This work develops a neural surrogate that replaces the iterative elastic solver within a semi-implicit spectral Cahn–Hilliard framework. A gradient enhanced composite loss function is proposed that jointly constrains the predicted field, its first-order gradient, and its Laplacian. Standard mean squared error training yields acceptable pointwise accuracy (R2=0.938) yet produces gradient errors that are amplified by the Laplacian operator in the Cahn–Hilliard update, causing systematic drift in the evolved microstructure. Feature-wise Linear Modulation (FiLM) conditioning is embedded into each convolutional layer of a U-Net backbone, enabling a single model to generalize across an 8-dimensional physical parameter space covering composition, eigenstrain, elastic modulus ratios, and applied stress with only 85 training configurations. Systematic comparison with FNO, standard U-Net, and Transolver baselines on a structured interpolation/extrapolation test set shows that the proposed model achieves a gradient error of 0.074, significantly lower than all compared baselines. The single-step elastic solve is accelerated by 24× on GPU. Surrogate-driven phase-field evolution maintains SSIM >0.95 and autocorrelation error below 30% within the reliable window t≤1.0. A hybrid correction strategy further extends accurate evolution to the full simulation horizon.
Organic – inorganic halide perovskites such as methylammonium lead iodide (MAPbI₃) exhibit exceptional optoelectronic properties but suffer from mechanical fragility that compromises long-term device integrity. In particular, their soft hybrid lattice gives rise to complex, direction-dependent mechanical responses that remain poorly understood. Here, we employ classical molecular dynamics simulations to investigate the anisotropic mechanical behavior of orthorhombic MAPbI₃ under uniaxial tension and compression. The stress–strain response reveals pronounced elastic anisotropy, with substantially lower stiffness and failure strength along the [001] direction compared to the [100] direction. This anisotropy originates from directional differences in PbI octahedral connectivity and weakly bonded organic–inorganic stacking along [001]. Our analysis reveals that uniaxial loading, regardless of direction, suppresses methylammonium (MA) cation rotation dynamics, culminating in molecular locking, with the notable exception of tensile loading along [001]. Molecular locking decouples the organic and inorganic sublattices, driving structural amorphization and ultimately brittle-like mechanical instability. In contrast, under tension along [001], MA molecules retain the freedom to rotate out of the xy-plane; this preserved cation dynamics maintains hydrogen bonding with the inorganic PbI₆ framework and gives rise to ductile-like behavior. Additionally, MAPbI₃ exhibits pronounced tension–compression asymmetry, wherein localized amorphization under tensile loading results in reduced strength and failure strain relative to compression. These mechanistic insights into strain-dependent cation dynamics and structural decoupling provide a foundation for strain-engineering strategies aimed at enhancing the mechanical robustness of hybrid halide perovskites.
Thermodynamic stability prediction remains a fundamental challenge in computational materials science because of the enormous size and structural complexity of crystal design space. In this study, we designed and developed a multimodal and explainable artificial intelligence framework for latent-space characterization of crystalline material stability using graph neural networks, manifold learning, anomaly detection, and explainable artificial intelligence (XAI). A total of 10,000 crystalline materials from the Materials Project database were analyzed using crystallographic and physicochemical descriptors. Hybrid feature representations comprising up to 182 descriptors were constructed from MAGPIE compositional features, lattice parameters, average bond distances, and partial radial distribution function (pRDF) descriptors. Crystal structures were further encoded using Crystal Graph Convolutional Neural Networks (CGCNNs), generating 64 dimensional graph embeddings. Subsequently, five manifold learning techniques, namely UMAP, t SNE, Diffusion Maps, Autoencoder (AE), and Variational Autoencoder (VAE), were comparatively assessed across latent spaces with dimensionalities of 2, 4, 8, 16, and 32. The resulting latent representations were analyzed using six anomaly detection algorithms, yielding 150 experimental configurations. Among all evaluated models, UMAP combined with Deep SVDD in a 4-dimensional latent space achieved the best performance, with an AUROC of 0.684, Precision@K of 0.906, and F1-score of 0.252. Isolation Forest and VAE-ELBO anomaly scoring achieved AUROC values of 0.677 and 0.665, respectively. In contrast, highly compressed 2-dimensional latent spaces produced AUROC values concentrated around 0.494–0.505, indicating limited anomaly separability. XAI analyses using SHAP, LIME, and GNN edge-importance mapping showed that thermodynamic instability was strongly associated with localized manifold-boundary regions and structurally anomalous crystal environments. Overall, the proposed framework demonstrates the potential of manifold-aware explainable AI for interpretable crystal stability analysis and anomaly-sensitive materials discovery.
