This repository contains the code for the paper:
In this work, we formulate a new neural operator by parameterizing the integral kernel directly in Fourier space, allowing for an expressive and efficient architecture. We perform experiments on Burgers' equation, Darcy flow, and the Navier-Stokes equation (including the turbulent regime). Our Fourier neural operator shows state-of-the-art performance compared to existing neural network methodologies and it is up to three orders of magnitude faster compared to traditional PDE solvers.
It follows from the previous works:
The code is in the form of simple scripts. Each script shall be stand-alone and directly runnable.
fourier_1d.py
is the Fourier Neural Operator for 1D problem such as the (time-independent) Burgers equation discussed in Section 5.1 in the paper.fourier_2d.py
is the Fourier Neural Operator for 2D problem such as the Darcy Flow discussed in Section 5.2 in the paper.fourier_2d_time.py
is the Fourier Neural Operator for 2D problem such as the Navier-Stokes equation discussed in Section 5.3 in the paper,
which uses a recurrent structure to propagates in time.fourier_3d.py
is the Fourier Neural Operator for 3D problem such as the Navier-Stokes equation discussed in Section 5.3 in the paper,
which takes the 2D spatial + 1D temporal equation directly as a 3D problemdata_generation
are the conventional solvers we used to generate the datasets for the Burgers equation, Darcy flow, and Navier-Stokes equation.We provide the Burgers equation, Darcy flow, and Navier-Stokes equation datasets we used in the paper. The data generation configuration can be found in the paper.
The datasets are given in the form of matlab file. They can be loaded with the scripts provided in utilities.py. Each data file is loaded as a tensor. The first index is the samples; the rest of indices are the discretization. For example,
Burgers_R10.mat
contains the dataset for the Burgers equation. It is of the shape [1000, 8192],
meaning it has 1000 training samples on a grid of 8192.NavierStokes_V1e-3_N5000_T50.mat
contains the dataset for the 2D Navier-Stokes equation. It is of the shape [5000, 64, 64, 50],
meaning it has 5000 training samples on a grid of (64, 64) with 50 time steps.We also provide the data generation scripts at data_generation
.
Here are the pre-trained models. It can be evaluated using eval.py or super_resolution.py.
@misc{li2020fourier,
title={Fourier Neural Operator for Parametric Partial Differential Equations},
author={Zongyi Li and Nikola Kovachki and Kamyar Azizzadenesheli and Burigede Liu and Kaushik Bhattacharya and Andrew Stuart and Anima Anandkumar},
year={2020},
eprint={2010.08895},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
@misc{li2020neural,
title={Neural Operator: Graph Kernel Network for Partial Differential Equations},
author={Zongyi Li and Nikola Kovachki and Kamyar Azizzadenesheli and Burigede Liu and Kaushik Bhattacharya and Andrew Stuart and Anima Anandkumar},
year={2020},
eprint={2003.03485},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。