Published 2025-06-30
How to Cite

This work is licensed under a Creative Commons Attribution 4.0 International License.
Abstract
Scientific discovery and physical simulation have traditionally relied on domain-specific numerical solvers and handcrafted feature design. In this paper, we propose a unified AI framework that integrates multimodal scientific representations, physics-aware neural architectures, and task-specific supervision strategies to enable accurate, interpretable, and generalizable simulation across diverse domains. The framework supports structured inputs such as molecular graphs, crystal lattices, and continuous physical fields, and leverages graph neural networks, neural operators, and Transformer-based modules to model system dynamics. Experimental results on molecular property prediction, field simulation, and inverse materials design demonstrate superior accuracy and efficiency compared to classical and deep learning baselines. Ablation studies further validate the importance of geometric encoding and physics-guided regularization. The proposed system enables scalable and transferable AI-driven scientific modeling, offering new opportunities for cross-domain discovery and computational reasoning.