Gaussian Fourier Transform: Theory & Applications

• Gaussian Fourier Transform is an analytical framework that leverages the self-reciprocal and rapid decay properties of Gaussians to yield closed-form solutions in both continuous and discrete settings.
• Generalizations extend the classical transform to families of Gaussian-Fourier operators, providing unitary transforms with complete bases for advanced time-frequency and signal analysis.
• Computational techniques using Gaussian radial basis methods and discrete transforms enable high-precision, efficient evaluations in applications such as optics, quantum mechanics, and adaptive spectral analysis.

A Gaussian Fourier Transform encompasses both the explicit calculation of Fourier transforms of Gaussian-type functions (continuous and discrete) and a spectrum of methodologies, integral transforms, and algorithms exploiting the unique properties of Gaussians in harmonic analysis, numerical integration, and signal processing. The central role of the Gaussian stems from its self-reciprocal property under the Fourier transform, rapid decay, and optimal uncertainty characteristics, and these are leveraged in analytical, computational, and algebraic frameworks throughout both applied and theoretical contexts.

1. Analytical Structure of the Gaussian Fourier Transform

The Fourier transform of multivariate Gaussians exhibits a unique closed-form corresponding to a Gaussian in frequency space. In $n$ dimensions, for a real symmetric positive-definite $A\in GL_n(\mathbb{R})$, the function $g_A(x) = \exp(-\pi x^\top A x)$ has the Fourier transform

$$\mathcal{F}[g_A](\xi) = (\det A)^{-1/2} \exp(-\pi \xi^\top A^{-1} \xi).$$

This is established by completing the square in the integrand and shifting the integration variable, leading to exact Gaussians in the dual variable with explicit prefactors. This self-reciprocal and scaling behavior is fundamental in both theoretical harmonic analysis and applications such as quantum mechanics, optics, and statistics. Polynomial modulations $x^\alpha g_A(x)$ transform to Hermite-type polynomials in the frequency domain, i.e., $$\mathcal{F}[x^\alpha g_A](\xi) = P_\alpha(\xi) e^{-\pi \xi^\top A^{-1} \xi}$$, where $P_\alpha(\xi)$ are Hermite polynomials (Rosenkranz et al., 2020).

2. Generalizations and Gaussian-Family Fourier Operators

The canonical role of the Gaussian as a Fourier eigenfunction carries over to broader classes of integral transforms. Williams, Bodmann, and Kouri introduced a family $\{\Phi_n\}$ of “Gaussian-Fourier” operators, each specified by requiring that an $n$-Gaussian $g_n(t) = \exp(- t^{2n} / (2n))$ is an eigenfunction, together with a dilation intertwining property:

$$\Phi_n D_\alpha = D_{\alpha^{-1}} \Phi_n, \qquad \Phi_n g_n = g_n$$

with $D_\alpha f(t) = |\alpha|^{1/2} f(\alpha t)$. The associated integral transform takes the form

$$(\Phi_n f)(\omega) = \int_{-\infty}^\infty \varphi_n(\omega t) f(t) dt,$$

where the kernel $\varphi_n$ is built from Bessel functions and encodes both even and odd analytic components satisfying a specific Laplacian eigen-equation (Williams et al., 2014).

The classical Fourier transform is obtained when $n=1$, yielding $\varphi_1(\eta) = (2\pi)^{-1/2} \exp(-i\eta)$. For other $n$, the transforms are unitary on $L^2(\mathbb{R})$, satisfy $\Phi_n^4 = I$, and their eigenfunctions span a complete $L^2$ basis. This framework underpins new integral transforms for time-frequency analysis and generalizes the scope of Gaussian-based signal representations.

3. Computational Techniques: Gaussian Radial Basis Methods

Martínez-Finkelshtein, Ramos-López, and Iskander developed fast numerical schemes for general 2D Fourier and diffraction integrals using Gaussian radial basis function (RBF) expansions (Martinez-Finkelshtein et al., 2015). Their method proceeds as follows:

1. Approximation: For a smooth compactly supported $f$, the function is approximated as $$f(x) \approx \sum_{j=1}^N \alpha_j e^{-\lambda_j \|x - c_j\|^2}$$, where $c_j$ are grid centers and $\lambda_j$ are shape parameters (often constant).
2. Fourier Transform: Each term’s 2D Fourier transform is analytic:

$$\int_{\mathbb{R}^2} e^{-\lambda\|x - c\|^2} e^{-i \omega \cdot x} dx = \frac{\pi}{\lambda} \exp\left(-\frac{\|\omega\|^2}{4\lambda}\right) e^{-i \omega\cdot c}$$

1. Series Assembly: By linearity, the transform of the RBF series is a sum of such terms.

This method exhibits “super-spectral” convergence in the fill distance between centers for smooth $f$ and retains high precision and computational efficiency, outperforming the 2D-FFT and extended Nijboer-Zernike approaches for through-focus optical computation. Complexity is $O(N^2 M + N^2 K)$ for an $N\times N$ spatial grid and $M$ defocus values, with negligible incremental cost for additional $f$ once precomputation is complete.

4. Discrete Gaussian Fourier Transform and Jacobi-Theta Theory

The discrete analogue of the Gaussian employs the Jacobi theta function to define minimal-spread packets in a periodic lattice. For $d=2j+1$, the normalized 2D discrete Gaussian is

$$\mathbf{g}_\sigma(n_1, n_2) = \sum_{\alpha_1, \alpha_2 \in \mathbb{Z}} \exp\left[ -\frac{\pi}{d} (n + \alpha d)^\top \sigma (n + \alpha d) \right],$$

where $\sigma$ is a real symmetric positive-definite $2\times2$ matrix. The corresponding discrete Fourier transform satisfies

$$\mathbf{F}[\mathbf{g}_\sigma](k_1, k_2) = \frac{1}{\sqrt{\det\sigma}} \sum_{\beta \in \mathbb{Z}^2} \exp\left[ -\frac{\pi}{d} (k+\beta d)^\top \sigma^{-1} (k+\beta d) \right],$$

or concisely, $$\mathbf{F}[\mathbf{g}_\sigma] = \frac{1}{\sqrt{\det \sigma}}\,\mathbf{g}_{\sigma^{-1}}$$. This discrete self-reciprocity is the finite-dimensional analog of the continuous Gaussian Fourier transform, underpinning minimal-uncertainty states in discrete signal and quantum analysis (Cotfas, 2019).

5. Gaussian-Based Fourier Sampling and Damped Harmonic Series

Gaussian-sampling-based Fourier transform methods enable precise, non-periodic approximations, avoiding the wrap-around effects and aliasing of standard DFT/FFT frameworks (Abrarov et al., 2015). A function $f(t)$ is approximated by shifted Gaussians:

$$f(t) \approx \sum_{n=-N}^N \frac{h}{c\sqrt{\pi}} \exp\left[-\left((t-nh)/c\right)^2\right] f(nh)$$

with step $h$ and width $c$. The Fourier transform is then represented as a sum over complex error (Faddeeva) functions or, equivalently, as a damping harmonic series:

$$F(\nu)\approx h\,e^{-(\pi c\nu)^2}\sum_{n=-N}^N f(nh)\,e^{-2\pi i \nu n h}$$

This representation converges exponentially in $N$, offers analytic convergence properties, and supports high-accuracy non-periodic signal analysis. The method is especially suited for local, non-periodic signals where windowed DFT schemes may introduce artifacts.

6. Special Function Analysis: Gaussians Multiplied by Bessel Functions

Analytically, the Fourier transform of Gaussian-windowed Bessel function modulations, crucial in wave packet and acoustic analysis, reduces to finite sums of confluent hypergeometric (Tricomi $U$-) functions. Explicitly:

$$I_{mn}(\beta, q) = \int_{-\infty}^\infty e^{-\beta^2 x^2 - i q x} x^{m+1/2} J_{n+1/2}(x) dx$$

is given by a finite sum over $k$ involving $$U\left(\frac{m+k+1}{2}, \frac12, \frac{(q \pm 1)^2}{4\beta^2}\right)$$ and simple prefactors. This closed-form analytical approach underpins efficient evaluation of wave-packet radiation integrals and illuminates the deeper structure of composite Gaussian–Fourier transforms (Carley, 2013).

7. Applications, Unitary Structures, and Optimality

Gaussian Fourier transforms underpin time-frequency Gabor analysis, phase-space quantization, through-focus optical computation, finite-dimensional signal processing, and quantum information theory. Their centrality is rooted in properties such as:

• Self-reciprocity and invariance under the Fourier transform;
• Unitarity, normalization, and periodicity properties (e.g., $\Phi_n^4 = I$ for the $\Phi_n$ family);
• Rapid decay and minimal Heisenberg uncertainty realized both in the continuous and discrete settings;
• Tunability in generalizations for adaptive spectral analysis.

In computational contexts, Gaussian-RBF-based Fourier methods yield stable, high-throughput evaluation of multi-parametric or high-resolution transforms with efficient scaling and robust error control. Discrete Gaussian families provide minimal-spread windows for Gabor systems and saturate discrete uncertainty principles, while the algebraic approach via Heisenberg modules unifies many of these phenomena across symbolic, analytic, and computational regimes (Martinez-Finkelshtein et al., 2015, Williams et al., 2014, Rosenkranz et al., 2020, Cotfas, 2019).