Staying on the Attractor: Supervising Neural Surrogates of 3D Turbulence Where They Leave It
Abstract
Neural surrogates are trained to predict 3D turbulent flows in place of direct numerical simulation (DNS). For chaotic flows, the goal is short-term pointwise accuracy followed by long-term physical and statistical fidelity. However, small prediction errors can carry a surrogate away from the flow's attractor. Off-attractor states are poorly represented in training data, leaving their evolution weakly constrained. The learned dynamics can then amplify deviations and lead to blow-up, freezing, or statistical drift. We propose off-attractor supervision (OAS) to supervise neural surrogates where they leave the attractor. OAS teaches the model how the true Navier-Stokes dynamics would evolve from these states. Each selected state is paired with its own future computed by DNS. Three generators select a few hundred states for relabeling. The first collects states from the surrogate's own rollouts. The second uses surrogate attacks to target freezing, excessive amplification, and violations of incompressibility and energy balance. The third perturbs training states along an amplified direction and a strongly damped random direction of the dynamics. All attacks run on the surrogate alone, and DNS relabeling is performed offline once per selected state. Experiments on 1283 turbulence show that OAS increases the median time to failure from 21 to 721 steps. The compared baselines achieve medians of at most 110 steps, and the advantage holds across training seeds. OAS also achieves the lowest pointwise error at step 15 and the best long-horizon statistics among the compared methods. OAS integrates physical models into neural simulation by extending supervision from fixed reference trajectories to states where the surrogate is likely to fail. This principle can guide the development of more reliable scientific surrogates when deployment takes models beyond the coverage of their training data.
Related
- Learned Surrogates of Three-Dimensional Magnetohydrodynamic Turbulence
- Neural Fourier Surrogates for Data Reuploading Quantum Neural Networks
- Polynomial neural surrogates for designing photonic quantum experiments
- Benchmarking Quantum Convolutional Neural Networks and their classical surrogates on long-range strongly correlated models
- How Does Distribution Shift Shape Pretraining Gains in Neural PDE Surrogates?
BibTeX
@misc{dai2026staying,
title = {{Staying on the Attractor: Supervising Neural Surrogates of 3D Turbulence Where They Leave It}},
author = {Yilong Dai and Shaswata Mitra and Raj Patel and Yiming Sun and Shengyu Chen and Jiaqi Gong and Sudip Mittal and Shahram Rahimi and Xiaowei Jia and Runlong Yu},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.32864},
eprint = {2609.32864},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.32864},
}
RIS
TY - UNPB
TI - Staying on the Attractor: Supervising Neural Surrogates of 3D Turbulence Where They Leave It
AU - Yilong Dai
AU - Shaswata Mitra
AU - Raj Patel
AU - Yiming Sun
AU - Shengyu Chen
AU - Jiaqi Gong
AU - Sudip Mittal
AU - Shahram Rahimi
AU - Xiaowei Jia
AU - Runlong Yu
PY - 2026
DA - 2026/09/26
PB - arXiv
DO - 10.48550/arxiv.2609.32864
UR - https://arxiv.org/abs/2609.32864
AB - Neural surrogates are trained to predict 3D turbulent flows in place of direct numerical simulation (DNS). For chaotic flows, the goal is short-term pointwise accuracy followed by long-term physical and statistical fidelity. However, small prediction errors can carry a surrogate away from the flow's attractor. Off-attractor states are poorly represented in training data, leaving their evolution weakly constrained. The learned dynamics can then amplify deviations and lead to blow-up, freezing, or statistical drift. We propose off-attractor supervision (OAS) to supervise neural surrogates where they leave the attractor. OAS teaches the model how the true Navier-Stokes dynamics would evolve from these states. Each selected state is paired with its own future computed by DNS. Three generators select a few hundred states for relabeling. The first collects states from the surrogate's own rollouts. The second uses surrogate attacks to target freezing, excessive amplification, and violations of incompressibility and energy balance. The third perturbs training states along an amplified direction and a strongly damped random direction of the dynamics. All attacks run on the surrogate alone, and DNS relabeling is performed offline once per selected state. Experiments on 128^3 turbulence show that OAS increases the median time to failure from 21 to 721 steps. The compared baselines achieve medians of at most 110 steps, and the advantage holds across training seeds. OAS also achieves the lowest pointwise error at step 15 and the best long-horizon statistics among the compared methods. OAS integrates physical models into neural simulation by extending supervision from fixed reference trajectories to states where the surrogate is likely to fail. This principle can guide the development of more reliable scientific surrogates when deployment takes models beyond the coverage of their training data.
ER -