This Perspective systematically discusses Physics-Grounded Materials AI (PhysMat AI) as a unifying perspective for integrating physical knowledge into materials intelligence through five complementary roles: physics as prior knowledge, descriptors, constraints, verifiers, and infrastructure.
Abstract
Artificial intelligence (AI) is transforming materials discovery, yet conventional data-driven approaches often suffer from limited interpretability, poor extrapolation, and inconsistency with physical laws. Since materials behavior is fundamentally governed by thermodynamics, kinetics, electronic structure, transport processes, and operating environments, the next generation of materials intelligence must move beyond correlation-based prediction toward physics-grounded reasoning. In this Perspective, we systematically discuss Physics-Grounded Materials AI (PhysMat AI) as a unifying perspective for integrating physical knowledge into materials intelligence through five complementary roles: physics as prior knowledge, descriptors, constraints, verifiers, and infrastructure. Using representative examples from catalysis, solid-state electrolytes in solid-state battery, and hydrogen-storage materials, we illustrate how physical principles guide data representation, model reasoning, validation workflows, and knowledge management. We further present how AI agents can leverage these physics-aware components to perform mechanism-guided discovery within physically feasible search spaces. Finally, we outline a developmental roadmap from physics-aware AI to physics-reasoning AI and ultimately physics-autonomous AI. Looking forward, materials intelligence should evolve from predictive models toward autonomous scientific systems capable of integrating physical reasoning, multiscale simulations, experimental validation, and continuous knowledge updating for reliable materials discovery.
The discovery of next-generation energy-storage materials is increasingly limited by the complexity of the underlying design problem rather than by computational capability alone. Porous transition-metal oxides represent a particularly challenging class of battery materials because their performance emerges from coupled interactions among crystal chemistry, pore architecture, ion transport, electrochemistry, electro-chemo-mechanics, synthesis, manufacturing, and battery-system operation. Recent advances in generative artificial intelligence (AI) have demonstrated remarkable capabilities for generating chemically plausible crystal structures. However, current approaches remain largely focused on crystallographic validity and thermodynamic stability. This perspective presents a roadmap for advancing generative AI beyond crystal generation toward physics-informed, application-aware, and synthesis-aware inverse design. Using porous oxide electrodes as a representative materials platform, we propose a seven-tier physics-informed inverse-design framework integrating chemistry, thermodynamics, transport, electrochemistry, durability, cell compatibility, and manufacturability. We further identify the"Missing Data Problem"as a fundamental bottleneck limiting application-aware AI and introduce an autonomous knowledge-generation framework supported by a Porous Oxide Energy Materials Ontology and a continuously evolving"Knowledge Base". Together, these concepts establish the foundation for Synthesis-Aware, Closed-Loop Autonomous Discovery, providing a general framework for AI-enabled autonomous materials discovery across energy-storage materials and other functional materials.
Machine learning (ML) in materials science refers to computational methods that learn statistical, structural, or physics-informed relationships from experimental, computational, and literature-derived materials data. These methods are used to predict materials properties, identify structure–property and process–structure–property–performance relationships, discover candidate materials, optimize synthesis and processing routes, and guide functional applications. ML is narrower than artificial intelligence (AI), which also includes broader reasoning, planning, search, and automation capabilities. It is also distinct from materials informatics, which is the wider data-centered framework that includes databases, descriptors, metadata, workflows, visualization, and knowledge management. ML can complement high-throughput computation by building surrogate models from density functional theory, finite-element simulation, molecular dynamics, or experimental data, but it is not identical to high-throughput screening itself. Unlike conventional physics-based modeling, which begins with explicit governing equations or mechanistic assumptions, ML usually infers predictive relationships from data; modern approaches increasingly combine both perspectives through physics-informed features, uncertainty quantification, and human expertise.
This work will present the current work on inverse material design, where AI methods—particularly generative pretrained transformers—are used to predict new material candidates based on desired properties, pushing the boundaries of materials innovation.
I. Gonzales, R. Ullberg, Andrew H Salij et al.· ECS Meeting Abstracts· 0 citations
Materials science is underpinned by structure-property relationships that govern the function of a material. These relationships can be encoded into algorithms and integrated into machine-learning models that enable the prediction of materials and their cognate properties. However, machine-learning models are largely being trained on computed data owing to a worldwide shortage of real-world (experimental) datasets. This review describes how to capture and collate experimental data from scientific literature using artificial-intelligence (AI) methods to produce materials-domain-specific datasets or language models. Their application in AI-driven enquiries that facilitate progress in energy-sustainable materials science is then illustrated via six case studies that cover: training machine-learning models, data-driven materials discovery, optimizing manufacturing processes, mapping phases of materials, forecasting materials-centric research trends, and classifying types of materials using automated prompt engineering. The future of materials-domain-specific datasets, language models, and decision-making workflows using AI agents is then envisioned for the energy sector. The intrinsic challenges of accessing historical dark data in materials science are then described and contrasted with timely opportunities for leveraging massive amounts of experimental data from laboratories in going forwards; by exploiting electronic-lab notebooks, high-throughput experiments, and digital-twin technologies. These opportunities are illustrated for energy-sustainable materials science, especially the photovoltaic and battery industries.
Jacqueline M. Cole· Advances in Materials· 0 citations
The discovery of materials with extreme lattice thermal conductivity (κL)—spanning from sub-air thermal insulators to metallic conductors that rival diamond—represents one of the most consequential frontiers in contemporary materials physics and engineering. This Perspective traces the evolution of thermal transport science from its empirical origins through the current AI-driven renaissance. We examine the limitations of conventional density-functional-theory-based phonon workflows, the emergence of universal machine learning interatomic potentials (uMLPs), graph-neural-network screening architectures such as the Crystal Attention Graph Neural Network (CATGNN), and generative inverse-design frameworks including the Physics-Guided Crystal Generative Model (PGCGM) and InvDesFlow-AL. Representative discovery and AI opportunities—including a computationally predicted room-temperature κL of 0.071 Wm−1 K−1 in Cs2HgPtCl6 and approximately 1,100 Wm−1 K−1 in metallic θ-TaN—are presented as concrete demonstrations of AI-driven computational power and human physical intuition, respectively. We conclude with a forward-looking discussion spanning five frontier challenges: machine learning frameworks for phonon–electron and phonon–ion interactions (with emphasis on electron–phonon coupling and its role in superconductivity, thermoelectrics, and carrier mobility); interpretable and symbolic AI approaches that extract closed-form physical principles from learned representations; out-of-distribution generalization strategies essential for discovering genuinely unprecedented materials; autonomous self-driving laboratory ecosystems that close the loop between computational prediction and experimental validation; and AI-accelerated design of interfacial thermal transport for high-power electronics and data center thermal management.
Ming Hu· Journal of Materials Science...· 0 citations
Artificial intelligence (AI) and machine learning (ML) are increasingly used as general-purpose research instruments across the physical sciences, accelerating tasks that were traditionally limited by trial-and-error experimentation, computational cost, or the sheer dimensionality of the underlying physics. Three areas illustrate this shift with particular clarity: computational materials discovery, nanostructured electrode design for energy storage, and space weather / heliophysics forecasting. Despite substantial progress in each area individually, limited work has examined them together, quantitatively, as expressions of a single underlying trend — the convergence of AI methodology with core physical and chemical science. Methods: This study used a narrative and scoping review methodology, incorporating quantitative benchmarks drawn directly from primary and independent critical sources. Peer-reviewed literature, preprints, direct observational monitoring data, and ResearchGate-hosted scholarly works published primarily between 2023 and 2026 were identified through structured searches combining terms from materials informatics, nanostructured energy-storage materials, and AI-based space weather forecasting. Sources were screened for topical relevance and synthesized thematically, with reported quantitative claims cross-checked against independent critical appraisal where available. Results: The synthesis identifies convergent innovation across three domains, with all three now quantitatively documented: (1) AI-driven inverse design (exemplified by a 2.2-million-structure materials search yielding roughly 380,000 candidate stable materials) is accelerating materials discovery, though independent re-analysis found only a small fraction of these structures met joint criteria of novelty, credibility, and utility; (2) nanostructured graphene–metal oxide composite electrodes span a wide reported performance envelope (specific capacitances from roughly 100 F/g to over 1000 F/g; energy densities up to roughly 100+ Wh/kg), with statistically designed synthesis optimization improving reproducibility; and (3) AI-assisted forecasting achieved approximately one-minute precision in reconstructing a major 2024 geomagnetic superstorm, in contrast to a roughly 40% amplitude error and nine-month timing error in the leading pre-cycle statistical/physical forecast of Solar Cycle 25's overall intensity. Discussion: A common methodological pattern recurs across all three domains: AI is used not to replace physical theory but to navigate high-dimensional parameter spaces that are analytically or computationally intractable by classical means alone, and the strongest, most defensible results are those subjected to independent, domain-expert critical appraisal rather than accepted at face value. Conclusion: Continued progress will depend on higher-quality shared datasets, physics-informed model architectures, standardized benchmarking protocols, and routine independent critical appraisal as a formal part of the AI-for-science publication cycle.
Keywords: artificial intelligence; materials discovery; inverse design; nanostructured electrodes; graphene–metal oxide composites; supercapacitors; space weather forecasting; geomagnetic storms; interdisciplinary science
R. Mishra, Divyansh Mishra, R. Agarwal· International journal of phy...· 0 citations