Physics-Guided Transformer (PGT): Physics-Aware Attention Mechanism for PINNs

ArXi:2603.27929v1 Announce Type: cross Reconstructing continuous physical fields from sparse, irregular observations is a central challenge in scientific machine learning, particularly for systems governed by partial differential equations (PDEs). Existing physics-informed methods typically enforce governing equations as soft penalty terms during optimization, often leading to gradient imbalance, instability, and degraded physical consistency under limited data. We