The Fourier Transform

The Fourier transform converts a function of a continuous variable into a function of the dual variable (time ↔ frequency, position ↔ momentum). It is one of the most important constructions in analysis and the bridge between continuous functions and their spectral content.

1. Definition

For a suitably nice function $f: \mathbb{R} \to \mathbb{C}$ (for example, $f \in L^1(\mathbb{R})$), the Fourier transform is

$\hat{f}(\xi) = \mathcal{F}[f](\xi) = \int_{-\infty}^{\infty} f(x)\, e^{-2\pi i \xi x}\, dx$

The inverse transform recovers $f$ from $\hat{f}$:

$f(x) = \int_{-\infty}^{\infty} \hat{f}(\xi)\, e^{2\pi i \xi x}\, d\xi$

The symmetric appearance is a consequence of the normalization $e^{-2\pi i \xi x}$; different textbooks use different factors ($e^{-i\omega x}$ with normalization constants $1$, $1/\sqrt{2\pi}$, or $1/(2\pi)$ — always a choice, never a substantive difference).

2. The function class problem

The integral $\int |f|\, dx$ converges iff $f \in L^1(\mathbb{R})$. For such $f$, $\hat{f}$ is continuous and bounded (Riemann–Lebesgue: $\hat{f}(\xi) \to 0$ as $|\xi| \to \infty$). But $L^1$ is too small — the inverse transform may fail — and too large — $\hat{f}$ need not itself be in $L^1$.

The correct home is $L^2(\mathbb{R})$: the square-integrable functions form a Hilbert space on which the Fourier transform is a unitary isomorphism. The extension from $L^1 \cap L^2$ to all of $L^2$ is done by density.

Plancherel’s theorem. For $f, g \in L^2(\mathbb{R})$,

$\int_{-\infty}^{\infty} f(x)\, \overline{g(x)}\, dx = \int_{-\infty}^{\infty} \hat{f}(\xi)\, \overline{\hat{g}(\xi)}\, d\xi$

and in particular $\|f\|_2 = \|\hat{f}\|_2$. Energy is preserved.

3. Key properties

Operation on $f(x)$	Corresponding operation on $\hat{f}(\xi)$
Translation $f(x - a)$	Modulation $e^{-2\pi i a \xi}\hat{f}(\xi)$
Modulation $e^{2\pi i b x} f(x)$	Translation $\hat{f}(\xi - b)$
Dilation $f(\lambda x)$	Inverse dilation $\tfrac{1}{
Differentiation $f'(x)$	Multiplication $2\pi i \xi \, \hat{f}(\xi)$
Multiplication $x f(x)$	Differentiation $\tfrac{-1}{2\pi i}\hat{f}'(\xi)$
Convolution $(f \ast g)(x)$	Product $\hat{f}(\xi)\hat{g}(\xi)$

The convolution theorem (last row) is the reason the Fourier transform dominates signal processing: it turns the costly convolution integral into pointwise multiplication.

4. The Gaussian is self-dual

A beautiful special case: if $f(x) = e^{-\pi x^2}$, then

$\hat{f}(\xi) = e^{-\pi \xi^2}$

The Gaussian is (up to scaling) its own Fourier transform. This is one of the principal reasons the Gaussian distribution is so fundamental in probability and analysis: it is the fixed point of the natural frequency-domain symmetry.

More generally, the Hermite functions $h_n(x) = H_n(x) e^{-\pi x^2}$ (where $H_n$ are Hermite polynomials) are eigenfunctions of $\mathcal{F}$ with eigenvalues $(-i)^n$. Every $L^2(\mathbb{R})$ function has a unique expansion in Hermite functions, giving a spectral decomposition of $\mathcal{F}$ itself.

5. Consequences

Uncertainty principle

A function cannot be simultaneously well-localized in space and in frequency. Quantitatively,

$\sigma_x \cdot \sigma_\xi \geq \frac{1}{4\pi}$

where $\sigma_x$ and $\sigma_\xi$ are the standard deviations of $|f|^2$ and $|\hat{f}|^2$. Equality holds exactly for Gaussians. In quantum mechanics, this is the Heisenberg uncertainty principle $\Delta x \cdot \Delta p \geq \hbar/2$.

Fourier series as a special case

When $f$ is periodic with period $T$, the continuous Fourier integral reduces to the Fourier series:

$f(x) = \sum_{n=-\infty}^{\infty} c_n\, e^{2\pi i n x / T}, \qquad c_n = \frac{1}{T}\int_0^T f(x)\, e^{-2\pi i n x / T}\, dx$

Non-smooth periodic waveforms (square, sawtooth, triangle) decompose into sines and cosines at the fundamental frequency and its integer multiples. The interactive below visualizes these partial sums and Gibbs’ phenomenon — the persistent overshoot at discontinuities.

Fourier transform on other groups

The construction generalizes dramatically:

• On $\mathbb{R}^n$: replace $e^{-2\pi i \xi x}$ by $e^{-2\pi i \langle \xi, x \rangle}$.
• On a torus $\mathbb{T}^n$: recover Fourier series.
• On a finite abelian group $G$: the discrete Fourier transform, used everywhere in digital computing (the DFT and its FFT implementation).
• On a general locally compact abelian group: Pontryagin duality.
• On a non-abelian group: Peter–Weyl theorem, representation theory.

Each of these frameworks inherits its core properties — Plancherel, inversion, convolution — from the prototype on $\mathbb{R}$.

6. Historical note

Joseph Fourier introduced the transform (and the series) in his 1822 Théorie analytique de la chaleur while studying the heat equation $\partial_t u = \partial_{xx} u$. His claim — that arbitrary functions could be decomposed into sines — was initially met with skepticism; Lagrange and Laplace both objected. The mathematical community spent most of the 19th century making the theory rigorous, culminating in Dirichlet’s convergence theorem (1829) and eventually Plancherel (1910) and the modern $L^2$ theory.

The Fast Fourier Transform (Cooley–Tukey, 1965) reduced the computational cost of the discrete transform from $O(N^2)$ to $O(N \log N)$ and is arguably the most widely used algorithm in applied mathematics.

Further reading

• Stein & Shakarchi, Fourier Analysis: An Introduction — rigorous modern textbook.
• Katznelson, An Introduction to Harmonic Analysis — classical.
• Hörmander, The Analysis of Linear Partial Differential Operators I, ch. 7 — distributions and the tempered Fourier transform.
• Folland, A Course in Abstract Harmonic Analysis — Pontryagin duality and the non-abelian theory.

Target waveform

Square Sawtooth Triangle

Number of harmonics

3

Series

square (odd-harmonics)

Coefficients

b_n = 4/(π n), odd n

Move the slider: each additional harmonic sharpens the partial sum toward the target waveform. Near the discontinuities, the overshoot persists — Gibbs' phenomenon.

Frequently asked

Why is the Fourier transform natural from a group-theoretic viewpoint?

The complex exponentials e^(2πiξx) are precisely the characters of the locally compact abelian group (ℝ, +) — the continuous group homomorphisms ℝ → U(1). The Fourier transform is the unique (up to normalization) unitary intertwiner between the regular representation of ℝ on L²(ℝ) and its spectral decomposition. Every abelian Fourier theory (Pontryagin duality) is a generalization of this fact.

What is Plancherel's theorem?

It states that the Fourier transform extends to a unitary isomorphism ℱ: L²(ℝ) → L²(ℝ), preserving inner products: ⟨f, g⟩ = ⟨ℱf, ℱg⟩. In particular ‖f‖₂ = ‖ℱf‖₂. This turns frequency analysis into a genuinely geometric operation on a Hilbert space.

What is the convolution theorem and why does it matter?

The Fourier transform takes convolutions to pointwise products: ℱ(f ∗ g) = ℱ(f) · ℱ(g). Since many operators in physics and signal processing are convolutions (linear time-invariant systems), this turns complicated integral operations into multiplication in the frequency domain.

Why is the Gaussian its own Fourier transform?

Because the Gaussian is an eigenfunction of the differential operator (x − iD), which intertwines multiplication by position with differentiation. The Fourier transform diagonalizes this operator, so its eigenfunctions (Hermite functions, with the Gaussian as ground state) map to themselves up to a phase.

Continue reading

Formula

Euler's Identity

The fundamental relation linking the five essential constants of mathematics through the exponential map on the complex plane — and a gateway into modern analysis.

Read →Formula

The Fundamental Theorem of Calculus

The deep connection between differentiation and integration — the central result that unifies the whole of calculus.

Read →Formula

The Gaussian Distribution

The bell curve — the probability distribution that appears whenever many independent random effects add up. The cornerstone of statistics.

Read →