Hierarchical Differentiable Fluid Simulation

Authors:

Xiangyu Kong ^* ^{1 , 2 , 3} , Arnaud Schoentgen ³ , Damien Rioux-Lavoie ³ , Paul G. Kry ¹ , Derek Nowrouzezahrai ^{1 , 2 , 4}

¹ McGill University ² Mila ³ Ubisoft ⁴ Canada CIFAR AI Chair

^* Primary author

Published in Computer Graphics Forum, 2025

📄 Download PDF 📑 Download BibTeX

Teaser

Abstract

Differentiable simulation is an emerging field that offers a powerful and flexible route to fluid control. In grid-based settings, high memory consumption is a long-standing bottleneck that constrains optimization resolution. We introduce a two-step algorithm that significantly reduces memory usage: our method first optimizes for bulk forces at reduced resolution, then refines local details over sub-domains while maintaining differentiability. In trading runtime for memory, it enables optimization at previously unattainable resolutions. We validate its effectiveness and memory savings on a series of fluid control problems.