AI and the World's Hard Problems · Physics · arXiv · 2026-10-20
Self-Supervised In-Context Operator Learning for Optimal Mixing in Incompressible Flows
Classical optimal control for mixing in incompressible flows must be solved separately for each initial condition, which is costly. The authors learn the solution operator that maps initial scalar fields to optimal mixing controls, using a self-supervised method with no labeled data, built on a transformer-based neural operator. Numerical experiments in two and three dimensions show generalization to unseen regimes, and the learned controls transfer to independent Eulerian simulations.