Reproduce “Flow Mesh: A Velocity-Regularized Flow Matching Framework for Stable Distributional Modeling of Physical Fields on Unstructured Meshes” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Siddharth Verma, Jaewan Park, Jay Phil Yoo, Kazuma Kobayashi, Souvik Chakraborty, Syed Bahauddin Alam

FlowMesh generates ensembles of fields on fixed unstructured meshes without remapping them to grids. It encodes mesh coordinates and field values into latent codes, then transports random samples through a learned velocity field and decodes them onto the mesh. On the PWR benchmark, it achieves 91.67% macro-averaged PDF coverage, exceeding the tested latent DDPM, but outputs can violate boundary conditions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.