Diff-NekRS: A Scalable Differentiable Framework for Multi-Timestep Solver-in-the-Loop Training

Diff-NekRS introduces a scalable differentiable framework that integrates automatic differentiation capabilities into the spectral element CFD solver NekRS. The framework addresses the challenge of error amplification during autoregressive inference by enabling backpropagation through multi-timestep solver-model rollouts. This enables training trainable correction networks to improve under-resolved fluid dynamics simulations. The approach targets the intersection of computational fluid dynamics and machine learning, specifically for hybrid physics-ML solver development.