Unstable genius: DeepMind cracks a century-old physics mystery with AI (www.businessinsider.com)

DeepMind announced that its researchers used bespoke, physics-aware machine learning models to discover new families of unstable singularities in three distinct fluid-dynamics equations — long-standing, hard-to-solve mathematical descriptions of how fluids move. Singularities are “blow-ups” where equations predict nonphysical extremes (infinite pressure or velocity), and unstable singularities are especially elusive. The team embedded the structure of the governing equations directly into neural solvers, trained and optimized them in stages, and reached near–machine-precision solutions that mathematicians could then verify formally. DeepMind frames this as a new playbook for applying AI to rigorous mathematical physics problems rather than heuristic modeling alone. The result matters because singularities underpin turbulence and other unpredictable fluid behaviors that plague engineering and natural systems. By revealing previously unknown unstable singularities, the work tightens understanding of where common fluid equations break down, which could improve computational fluid dynamics, turbulence monitoring, and models used in aerospace drag estimation, weather forecasting, blood-flow analysis, and energy systems. More broadly, the project demonstrates how tailored ML architectures that respect physical structure can produce verifiable, high-precision discoveries — a shift from generative hype toward tools that advance fundamental science.