FNO Darcy 1

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 1

FNO_Darcy

February 9, 2024

1 Data-driven Darcy Flow Using FNO and its Variants


In this notebook we will cover some theory behind the Fourier Neural Operators and some of their
variants. We will then apply these architectures on a Darcy flow field prediction problem in a
data-driven fashion.

1.0.1 Learning Outcomes


1. A brief theory behind the Fourier Neural Operators (FNO), Adaptive Fourier Neural Opera-
tors (AFNO) and Physics-informed Neural Operators (PINO)
2. How to set up and train the three models in Modulus
3. Differences between the three variants

1.1 Fourier Neural Operators (FNO)


Fourier neural operator (FNO) is a data-driven architecture which can be used to parameterize
solutions for a distribution of PDE solutions. The key feature of FNO is the spectral convolutions:
operations that place the integral kernel in Fourier space. The spectral convolution (Fourier integral
operator) is defined as follows:

(𝒦(w)𝜙)(𝑥) = ℱ−1 (𝑅W ⋅ (ℱ) 𝜙)(𝑥), ∀𝑥 ∈ 𝐷 (1)

where ℱ and ℱ−1 are the forward and inverse Fourier transforms, respectively. 𝑅w is the trans-
formation which contains the learn-able parameters w. Note this operator is calculated over the
entire structured Euclidean domain 𝐷 discretized with 𝑛 points.
Fast Fourier Transform (FFT) is used to perform the Fourier transforms efficiently and the resulting
transformation 𝑅w is just a finite size matrix of learn-able weights. Inside the spectral convolution,
the Fourier coefficients are truncated to only the lower modes which allows explicit control over the
dimensionality of the spectral space and linear operator.
The FNO model is the composition of a fully-connected “lifting” layer, 𝐿 spectral convolutions with
point-wise linear skip connections and a decoding point-wise fully-connected neural network at the
end.

𝑢𝑛𝑒𝑡 (Φ; 𝜃) = 𝒬 ∘ 𝜎(𝑊𝐿 + 𝒦𝐿 ) ∘ ... ∘ 𝜎(𝑊1 + 𝒦1 ) ∘ 𝒫(Φ), Φ = {𝜙(𝑥); ∀𝑥 ∈ 𝐷} (2)

You might also like