FFT normalization in ARES/BORG
This page is to summarize the convention used for normalizing Fourier transform, and the rational behind it.
The discrete fourier transform is defined, for a cubic box of mesh size \(N\) as
In cosmology we are mostly interested in the continuous infinite Fourier transform
It can be shown that the continuous transform, under reasonable conditions, can be approximated and matched normalized to the following expression in the discrete case:
\(\delta(\vec{x}) = \frac{1}{L^3} \sum_{\vec{k}} \exp\left(i\frac{2\pi}{L} \vec{x} .\vec{k} \right) \hat{\delta}\left(\vec{k}\frac{2\pi}{L}\right)\)This leads to define the following operator for the discrete Fourier transform:
\(F = \frac{1}{L^3} \mathcal{F}\)which admit the following inverse:
\(F^{-1} = L^3 \mathcal{F}^{-1} = \left(\frac{L}{N}\right)^3 \mathcal{F}^\dagger\)