cs.LGNov 13, 2026

Sample-Efficient Control of Magnetoconvective Flows via Differentiable Simulation

Authors: Jannis Becktepe, Dmitry Krasnov, Nils Thuerey, Jörg Schumacher, Sebastian Peitz

Organizations: TU Dortmund University & Lamarr Institute for Machine Learning and Artificial Intelligence · Institute of Thermodynamics and Fluid Mechanics, TU Ilmenau · Physics-based Simulation Group, TU Munich & Munich Center for Machine Learning · Scientific Machine Learning Group, Paderborn University

Abstract

Magnetohydrodynamic (MHD) fluid flows are central to demanding engineering applications, most prominently fusion blanket cooling, where flow control is a promising tool to optimize key properties such as heat transfer. While active flow control (AFC) has recently advanced through machine learning, its application to MHD flows remains largely unexplored. In this work, we present φφ-PICT, a novel differentiable GPU-accelerated MHD solver targeted at AFC. We validate φφ-PICT against MHD benchmarks and demonstrate its high computational efficiency. Integrated into FluidGym, φφ-PICT enables sample-efficient gradient-based algorithms to improve heat transfer by >25>25% and allows optimizers to achieve 103×10^3\times higher accuracy in inverse design than black-box methods. φφ-PICT is publicly available at https://github.com/upb-sciml/phiPICT.

Figures & tables

Appendix figures & tables26 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. Physics Foundation Models for Efficient Control of Fluid Flows

    Oct 10, 2026Mahindra Rautela, Shaifalee Saxena, Alexander ScheinkerPhysical Foundation ModelsPolicy Learning

  2. Magflow: Differentiable Simulation of Industrial-Grade 6D Magnetic Levitation

    Nov 17, 2026Philipp Hartmann, Jannick Stranghöner, Klaus NeumannFeedback ControlReal-Time Robot Control

  3. The HydroGym Reinforcement Learning Platform for Fluid Dynamics

    Dec 19, 2025Christian Lagemann, Sajeda Mokbel, Miro Gondrum +18RL ControlRL Benchmarks