Open Access

Examining the bonding between their constituent atoms in crystalline materials has played a vital role in understanding material properties.[1–4] For instance, low thermal conductivity in materials is typically attributed to its anharmonicity, which has been reported to arise from strong antibonding interactions and local environment distortions.[5–7] The bonds in the material are often quantified in terms of bond strength and can be extracted from crystalline materials using density-based[8], energy-based[9], and orbital-based[10] methods. LOBSTER[11] is a program that relies on an orbital-based method to extract such bonding information by projecting the plane wave-based wave functions of modern density functional theory computations (DFT) onto a local atomic orbital basis. Since our goal was to use bonding analysis descriptors for material property predictions, we needed to first systematically generate large quantities of bonding analysis data. To streamline this process, we have developed a user-friendly workflow[12], which is now also part of the atomate2[13] package that can generate bonding information data extracted using the LOBSTER program for crystalline materials. This workflow requires only the structure as input from the user. Employing this workflow, we have generated for ~13000 crystalline compounds such bonding analysis data. To create new descriptors from these data, we use our package LobsterPy.[14] The curated descriptors span different types, including statistical representations of bonding characteristics for traditional ML algorithms (e.g., random forests), textual descriptions for large language models (LLMs), and structure graphs for graph neural networks (GNNs). These descriptors are then tested by employing them in several state-of-the-art ML algorithms and architectures to predict the mechanical, vibrational, and thermal properties of crystalline materials. Through this work, we are not only able to demonstrate how one can enhance the model’s predictive accuracy[15] by incorporating quantum chemical bonding-based descriptors alongside typical composition and structure-based descriptors but it also aids in uncovering relationships between bonding and materials properties on a larger scale, which was not possible before.

This repository includes raw data for bonding analysis and lattice thermal conductivity using MLIP-MACE-MP03b, supporting the manuscript “Thermal Transport in Ag₈TS₆ (T= Si, Ge, Sn) Argyrodites: An Integrated Experimental, Quantum-Chemical, and Computational Modelling Study”

Large Language Models (LLMs) are reshaping many aspects of materials science and chemistry research, enabling advances in molecular property prediction, materials design, scientific automation, knowledge extraction, and more. Recent developments demonstrate that the latest class of models are able to integrate structured and unstructured data, assist in hypothesis generation, and streamline research workflows. To explore the frontier of LLM capabilities across the research lifecycle, we review applications of LLMs through 34 total projects developed during the second annual Large Language Model Hackathon for Applications in Materials Science and Chemistry, a global hybrid event. These projects spanned seven key research areas: (1) molecular and material property prediction, (2) molecular and material design, (3) automation and novel interfaces, (4) scientific communication and education, (5) research data management and automation, (6) hypothesis generation and evaluation, and (7) knowledge extraction and reasoning from the scientific literature. Collectively, these applications illustrate how LLMs serve as versatile predictive models, platforms for rapid prototyping of domain-specific tools, and much more. In particular, improvements in both open source and proprietary LLM performance through the addition of reasoning, additional training data, and new techniques have expanded effectiveness, particularly in low-data environments and interdisciplinary research. As LLMs continue to improve, their integration into scientific workflows presents both new opportunities and new challenges, requiring ongoing exploration, continued refinement, and further research to address reliability, interpretability, and reproducibility.