Physics-Informed Spatio‑Temporal Super‑Resolution of Dynamical Systems

We developed a physics‑informed deep learning framework that boosts both spatial and temporal resolution of coarse solutions to time-dependent partial differential equations—without relying on high-resolution training data. The model employs two sequential modules: one enhances spatial detail, followed by another for temporal refinement. A novel composite loss enforces fidelity to governing PDEs, initial/boundary conditions, and inter-time consistency. Applied to an elastodynamics problem, the method achieves high accuracy while enforcing physical constraints and reducing computation and data requirements—accelerating scientific simulation and design workflows

Refer paper (Arora & Shrivastava, 2022) for details.