Reproduce “TB-Transolver: Parallelized and Coherent Long-Horizon Rollouts via Temporal Bundling for Time-Dependent PDEs” 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

Sidharth S. Menon, Ameya D. Jagtap

TB-Transolver bundles future physical-field snapshots, retaining Transolver’s spatial Physics-Attention while reducing temporal evaluations. The authors guide bundle width with C(K), the spatial correlation between the first and last snapshots, recommending the largest tested width with correlation around 0.8 or higher. On reaction–diffusion, K=10 reduced training time per epoch from 216 to 22.60 seconds and mean relative L2 error from 0.00489 to 0.00113; low-viscosity Navier–Stokes favored K=2, limiting generality.

Summary by alphaXiv.

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