From maxwell to digital twins: a physics-based framework for future AI-native wireless communications

Wireless digital twins are emerging as a critical technology for next-generation communications. Constructing these twins is challenging because it requires modeling not only environmental geometry, but also complex electromagnetic (EM) properties and interactions. Rooted in classical EM theory, our study shows that accurate wireless digital twin construction depends on properly representing and learning essential intrinsic scene parameters–geometry, material properties, and propagation mechanisms–embedded in the PDE structure of Maxwell’s equations. We propose a neural methodology that leverages deep learning to learn the underlying physics directly from data, achieving high-quality reconstruction and robust generalization across changing environments. Most importantly, we demonstrate that neural models can be guided by physical laws while remaining data-driven, revealing a strong and actionable connection between physics and AI for future AI-native wireless communications that more faithfully sense, understand, and interact with the physical world.