| Scope |
The connections between computational materials tools across different length and time scales remain long-standing challenges. Many physics-based methods still struggle to construct quantitative governing equations from lower-scale simulation data with sufficient accuracy, robustness, and transferability for higher-scale models. These challenges have become increasingly urgent with the rapid expansion of research on chemically complex materials (including high-entropy materials), materials under extreme conditions, and advanced manufacturing and processing. Meanwhile, advances in high-performance computing, automated simulation workflows, and open materials data infrastructures have greatly expanded the availability of high-quality training datasets. In parallel, recent developments in artificial intelligence (AI), including foundation models, equivariant neural networks, active learning, uncertainty-aware learning frameworks, and symbolic regression for equation discovery, are enabling new approaches for integrating physics, data, and computation. These methods create new opportunities to identify governing equations and reduced-order physical models directly from high-fidelity simulation and experimental data. To reflect these trends, this symposium focuses on AI-enabled strategies for building quantitative, physically consistent, and robust multiscale connections to accurately explain and predict complex material behaviors.
Topics include, but are not limited to:
• Physics-informed, hybrid, and data-driven prediction of material properties using first-principles, atomistic, and experimental datasets.
• Next-generation machine learning interatomic potentials, including equivariant and foundation-model-based approaches.
• AI-informed mesoscale modeling (phase field, Monte Carlo, kinetic Monte Carlo, dislocation dynamics, crystal plasticity, etc.) based on data-driven or physics-guided governing equations.
• Data-driven discovery of governing equations and constitutive relations using symbolic regression and equation-learning methods.
• AI-assisted calibration and reduction of phenomenological models using large-scale simulation and multimodal experimental data.
• Data-driven uncertainty quantification and tuning of governing equations for mesoscale and continuum simulations.
• Closed-loop and autonomous multiscale workflows integrating simulation, machine learning, symbolic regression, and experiments. |