← Papers
Graphs

X-MeshGraphNet

MeshGraphNet split into partitions with halos

Paper title X-MeshGraphNet: Scalable Multi-Scale Graph Neural Networks for Physics Simulation

Nabian et al. (NVIDIA) · Aero & CFD · Public: code

Paper arXiv 2411.17164PDF

Builds the graph from points sampled on the STL (k nearest neighbours) instead of the simulation mesh, stacks coarse-to-fine point clouds for long-range messages, and cuts the graph into partitions with halo regions, aggregating gradients so training on partitions equals training on the whole graph.

Share on LinkedIn, X or Bluesky, or or .

Know related work that belongs here?

Something wrong on this page?

More papers

All papers
  1. GraphsR-BiLoGo — Transfer learning for rib effects in sheet-forming surrogates
  2. GraphsRead-Write-Relax — Why neural PDE surrogates need both global and local processing
  3. GraphsDGN / LDGN — Sampling flow distributions on meshes with diffusion graph networks