DeepMind PINNs Uncover New Unstable Singularities in Fluid Dynamics
DeepMind researchers used physics-informed neural networks (PINNs) and second-order optimizers to systematically discover new families of unstable singularities in fluid equations (including IPM and Boussinesq). They found striking patterns in blow-up speeds (λ) as instability increased and pushed PINNs to near–machine precision, opening the door to AI-driven, computer-assisted proofs and new advances in mathematics, physics and engineering.