Reproduce “Energy-Guided Transport for Projection-Free Physics-Informed Flow Matching” 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
Giuliano Galadini, Aurelio Uncini, Danilo Comminiello
EGT guides pretrained flow generation by subtracting a residual-energy gradient from the learned velocity. In a full-PDE 2D Navier–Stokes test with a white-noise prior, it completed every sample at 0.94 seconds per sample; ECI also completed every sample, at 4.40 seconds. EGT achieved 100% completion across four tested systems, without guaranteeing exact endpoint satisfaction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.