Wavelet Flow Matching Enhances Multi-Scale Physics Emulation
Researchers propose Wavelet Flow Matching (WFM), a generative emulator for multi-scale physical systems governed by PDEs. WFM performs optimal transport directly in wavelet space using a U-Net's hierarchical structure, avoiding separate autoencoder pretraining. It achieves superior long-horizon stability, accuracy, and spectral coherence on three chaotic fluid dynamics systems compared to existing methods.
Key facts
- Wavelet Flow Matching (WFM) is a novel generative emulator for multi-scale physics.
- WFM performs optimal transport directly in wavelet space.
- It leverages the hierarchical structure of a U-Net to predict transport velocities.
- WFM avoids the need for separately pre-trained autoencoders.
- Tested on three challenging chaotic fluid dynamics systems.
- Achieves superior long-horizon stability, accuracy, and spectral coherence.
- Deterministic emulators produce overly-smoothed predictions.
- Generative approaches capture details but are costly.
Entities
—