Fourier Transform

The unifying principle

Any well-behaved function $f(t)$ can be written as a superposition of complex exponentials $e^{i\omega t}$ (pure rotations at frequency $\omega$ ):

$\hat f(\omega) = \int_{-\infty}^{\infty} f(t)\, e^{-i\omega t}\, dt \qquad f(t) = \frac{1}{2\pi}\int_{-\infty}^{\infty} \hat f(\omega)\, e^{i\omega t}\, d\omega$

The reason this works everywhere: the functions $e^{i\omega t}$ form an orthogonal basis, and — crucially — they are the eigenfunctions of the derivative operator:

$\frac{d}{dt}e^{i\omega t} = i\omega\, e^{i\omega t}.$

So any linear, time/shift-invariant operation (differentiation, convolution, wave propagation) becomes multiplication in frequency space. The master identity is the convolution theorem:

$\widehat{(f * g)}(\omega) = \hat f(\omega)\,\hat g(\omega).$

Tangled interactions in the original domain → simple products in the frequency domain. That single fact is why every field below reaches for the same tool.

Here the transform is pure structure: $L^2$ is a Hilbert space, $\{e^{i\omega t}\}$ is an orthonormal-in-the-distributional-sense basis, and Fourier is a unitary change of basis (Plancherel: $\|f\|^2 = \frac{1}{2\pi}\|\hat f\|^2$ , energy is conserved).

Notation: $\mathcal{F}\{f\} = \hat f$ , with $\mathcal{F}\{f'\} = i\omega\,\hat f$ .
Why same idea: diagonalizing the derivative = decomposing into frequencies.
Example: the heat equation $u_t = u_{xx}$ becomes $\hat u_t = -\omega^2 \hat u$ , giving $\hat u(\omega,t)=\hat u(\omega,0)e^{-\omega^2 t}$ — a PDE turned into trivial ODEs, one per frequency.

Physics — quantum states & wave–particle duality

Position and momentum representations of a quantum state are Fourier transforms of each other:

$\psi(x) = \frac{1}{\sqrt{2\pi\hbar}}\int \tilde\psi(p)\, e^{ipx/\hbar}\, dp.$

Why same idea: momentum eigenstates are the pure waves $e^{ipx/\hbar}$ ; "how much of each momentum" is the frequency decomposition.
Consequence: the Heisenberg uncertainty principle $\Delta x\,\Delta p \ge \hbar/2$ is exactly the Fourier fact that a narrow spike in one domain must be broad in the other.
Example: a Gaussian wave packet has a Gaussian momentum spectrum; localize it tighter in $x$ and its spread in $p$ grows.

Coding / CS — the FFT and digital signal processing

Discrete data $x_0,\dots,x_{N-1}$ gets the Discrete Fourier Transform:

$X_k = \sum_{n=0}^{N-1} x_n\, e^{-2\pi i kn/N}.$

The Fast Fourier Transform computes this in $O(N\log N)$ instead of $O(N^2)$ by recursively splitting even/odd samples (divide-and-conquer).

Why same idea: same basis, sampled and finite.
Examples: JPEG uses the cousin DCT to drop high-frequency image detail your eye ignores; MP3 discards inaudible frequency bands; polynomial multiplication is done via FFT (convolution theorem) so big-integer multiplication runs fast.

Hardware — circuits, filters, and spectrum

Real electronics lives in frequency space. With impedances $Z_R=R$ , $Z_C=\frac{1}{i\omega C}$ , $Z_L=i\omega L$ , a differential circuit equation becomes ordinary Ohm's law at each $\omega$ .

Notation: the transfer function $H(\omega)=V_{out}/V_{in}$ is literally the ratio of Fourier components.
Why same idea: capacitors/inductors differentiate & integrate → in frequency space that's just multiplying by $i\omega$ or $1/i\omega$ .
Example: an RC low-pass filter has $H(\omega)=\frac{1}{1+i\omega RC}$ , passing low frequencies and attenuating high ones — exactly frequency selection. Spectrum analyzers, radio tuning, and antenna design all read the world in $\hat f$ .

Why this bridge matters

The convolution ⇄ multiplication swap transfers everywhere. Learn it once (blurring an image, a filter's response, an FFT-based multiply) and you own it in all four fields.
Uncertainty is universal. The physicist's $\Delta x\,\Delta p$ , the audio engineer's time–frequency tradeoff (you can't know exact pitch at an exact instant — the STFT window dilemma), and the mathematician's support-vs-bandwidth bound are the same theorem.
Intuition flows both ways. A CS student's mental picture of the FFT ("sort signal into frequency bins") demystifies the momentum-space wavefunction. A hardware engineer's feel for filters ( $i\omega$ = differentiate) makes the heat-equation solution obvious. And the math notion of a unitary basis change explains why energy (Plancherel), information, and probability are all preserved when you switch views.
The eigenfunction insight is the punchline: whenever your system doesn't care when or where something happens (time/shift invariance), the pure waves are its natural coordinates — so decompose into frequencies and every hard operator turns into simple scalar multiplication.

Connections

01 Hilbert Spaces & Orthogonal Bases
02 Convolution Theorem
03 Heat & Wave PDEs
04 Quantum State Representations
05 Heisenberg Uncertainty Principle
06 FFT & Divide-and-Conquer
07 Digital Signal Processing & Compression
08 Impedance & Transfer Functions
09 Filters & Spectrum Analysis

#bridge