Neural Ideal Large Eddy Simulation: Modeling Turbulence with Neural Stochastic Differential Equations
Authors: Anudhyan Boral, Zhong Yi Wan, Leonardo Zepeda-Núñez, James Lottes, Qing Wang, Yi-Fan Chen, John Anderson, Fei Sha
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show the effectiveness of our approach (ni LES neural ideal LES) on two challenging chaotic dynamical systems: Kolmogorov flow at a Reynolds number of 20,000 and flow past a cylinder at Reynolds number 500. Compared to competing methods, our method can handle non-uniform geometries using unstructured meshes seamlessly. In particular, ni LES leads to trajectories with more accurate statistics and enhances stability, particularly for long-horizon rollouts. |
| Researcher Affiliation | Industry | Anudhyan Boral Google Research Mountain View, CA 94043, USA anudhyan@google.com Zhong Yi Wan Google Research Mountain View, CA 94043, USA wanzy@google.com Leonardo Zepeda-Núñez Google Research Mountain View, CA 94043, USA lzepedanunez@google.com James Lottes Google Research Mountain View, CA 94043, USA jlottes@google.com Qing Wang Google Research Mountain View, CA 94043, USA wqing@google.com Yi-fan Chen Google Research Mountain View, CA 94043, USA yifanchen@google.com John Roberts Anderson Google Research Mountain View, CA 94043, USA janders@google.com Fei Sha Google Research Mountain View, CA 94043, USA fsha@google.com |
| Pseudocode | Yes | Algorithm 1 Compute M(v(t0)) |
| Open Source Code | No | (Source codes and datasets will be made publicly available.) |
| Open Datasets | No | The paper mentions "The reference data for Kolmogorov flow consists of an ensemble of trajectories generated by randomly perturbing an initial condition." and "The flow past cylinder dataset is generated as a single long, chaotic trajectory...". While it describes how the data was generated, it does not provide concrete access information (link, DOI, citation with author/year) for these datasets, besides stating they "will be made publicly available". |
| Dataset Splits | Yes | The DNS data for Kolmogorov flow consists of an ensemble of 100 trajectories, each 2500 LES time steps long. The first 70 trajectories are used for training, the next 10 for validation, and the final 20 for testing. All results reported here are on the 20 unseen trajectories from the test set. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or specific computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper refers to software (e.g., JAX in citation [15]) but does not provide specific version numbers for any key software components or libraries used in their experiments. |
| Experiment Setup | Yes | For all LES methods, including ni LES, we use a 10 larger time step than the DNS simulation. ni LES uses four SDE samples for both training and inference, and each SDE sample is resolved using 16 uniformly-spaced time steps corresponding to a single LES time step. |