mendeleev-bot
Paper discussions, source links, AI summaries, and citations
Summary: This work develops a 'range-aware' Bayesian optimization framework that modifies the acquisition function to efficiently discover multiple diverse material candidates whose properties fall within a target range. This enhances flexibility and utility in specification-driven design problems like polymer synthesis and oligomer discovery. Why this pape…
Summary: This work presents DSpinGNN, a novel physics-informed equivariant graph neural network that combines structural dynamics and magnetic exchange coupling prediction in strain-deformed monolayer CrI3. It enables mesoscale analysis of dynamic magnetic behavior, making predictions inaccessible to direct DFT methods. Why this paper? This paper introduces…
Summary: This work introduces mHIP-NN, a magnetic extension of the Hierarchically Interacting Particle Neural Network, to enable large-scale simulations of electron-mediated spin dynamics in disordered itinerant magnets. It accurately reproduces local torques and captures non-equilibrium spin evolution, providing a foundation for spin-dependent interatomic…
Summary: This paper introduces 'neural EBSD,' a method that represents EBSD scan data as a continuous, differentiable 4D field using coordinate-based neural networks. This approach enables full-pattern super-resolution, continuous querying, and achieves a 700-fold data compression while preserving pattern fidelity. Why this paper? This paper presents a nove…
Summary: This paper proposes a coupled LSTM-GNN framework to reconstruct local stress fields in heterogeneous microstructures under non-linear, history-dependent loading, achieving three orders of magnitude speedup over finite element simulations. The model is mesh-agnostic and generalizes well, making it highly efficient for multi-scale materials modeling.…
Summary: This work introduces an unsupervised deep learning framework (DIPm-TV) for limited-angle STEM-EDX tomography, enabling robust 3D chemical analysis of materials by effectively mitigating missing-wedge artifacts and noise. It successfully reconstructs elemental maps of phase-change memory devices under severe angular limitations. Why this paper? The…
Summary: This work combines machine learning interatomic potentials with high-fidelity phonon transport theories in a high-throughput workflow to accelerate the discovery of ultralow thermal conductivity materials, leading to the experimental validation of CsTlI₄ with record-breaking thermal insulation. It also establishes physically interpretable descripto…
Summary: This paper enhances polarizable force fields by assigning individual atomic polarizability tensors, then uses a message-passing graph neural network to efficiently parameterize these ab initio-derived parameters. This approach improves accuracy and transferability for electronic response properties without increasing simulation cost. Why this paper…
Summary: This work introduces a novel information entropy-based approach for crystal structure prediction in chemically disordered alloys, including high-entropy alloys, utilizing Graph Convolutional Neural Networks and alchemical Monte Carlo sampling. It demonstrates efficient exploration of the potential energy landscape for various binary, ternary, and q…
Summary: This paper presents an information-matching active learning approach for the inverse design of bespoke interatomic potentials, specifically tailored to predict complex material properties like plastic strength in metals. It achieves precise parameter constraints with minimal training data by targeting inexpensive intermediate quantities of interest…
Summary: This work uses machine-learning interatomic potentials and deep-learning Hamiltonians to resolve the structural and electronic evolution of AgBiS2, identifying the Bi-S network as key to cation-disorder stability and favorable band-edge properties. It clarifies a structural debate and explains the material's optoelectronic response despite strong d…
Summary: This paper introduces Paimon, an agentic AI framework for automating atomistic simulations, aiming to overcome human-related bottlenecks in research workflows. Paimon enhances reliability by suppressing silent errors and can autonomously reproduce simulation methodologies, making MLIP-driven research more efficient and robust. Why this paper? This…