Fourier Transform¶

The Story Behind the Math¶

In the summer of 1807, Jean-Baptiste Joseph Fourier (1768-1830), a French prefect in Egypt under Napoleon, submitted a revolutionary memoir to the Paris Academy of Sciences. He wasn't studying waves or sound, but something seemingly mundane: how heat spreads through solid bodies.

The problem was practical: how to predict temperature in a metal object? Differential equations existed, but no one could solve them for complex geometries.

His brilliant insight: any periodic function (even discontinuous!) can be decomposed into sums of sines and cosines. A "jagged" function like a square wave? Just sines. A ramp? Sines and cosines. Everything.

But when he presented his theory, he ignited a mathematical war. Joseph-Louis Lagrange, the giant of the era, fiercely opposed him. He argued that discontinuous functions couldn't be expressed with smooth functions like sines and cosines. The debate raged for decades.

Dirichlet (1829) finally clarified: under certain conditions (finite discontinuities, absolute integrability), Fourier series converge. Riemann and Lebesgue extended the theory. But Fourier's intuition was correct: almost anything can be decomposed into waves.

The plot twist: What Fourier invented for heat became the fundamental tool for: - Audio (MP3, FLAC) - Images (JPEG, compression) - Quantum physics (wave mechanics) - Machine learning (spectral analysis) - Medicine (MRI, tomography)

Why It Matters¶

The Fourier Transform explains why the world is made of frequencies:

Audio: Breaking music into frequencies (bass, mid, treble)
Images: Analyzing patterns and textures
Communications: Modulating radio waves, filtering noise
Quantum Physics: Position ↔ momentum (wave-particle duality)
Medicine: MRI reconstructs images from frequency signals
Machine Learning: CNNs use convolutions (multiplication in frequency domain)
Finance: Cyclical market analysis

Without Fourier, we wouldn't have: - MP3, JPEG, MPEG - WiFi, 4G, 5G - X-ray microscopy - Spectral analysis

Prerequisites¶

Central Limit Theorem — Uses Fourier in its proof
Integral calculus (improper integrals)
Complex numbers
Trigonometric series
Concept of orthogonality
Simple Harmonic Motion — Building blocks of Fourier analysis

The Core Insight¶

Decomposition into Waves¶

Imagine a vibrating string. It can vibrate in simple modes (harmonics): - Fundamental: half wave on the string - 2nd harmonic: one complete wave - 3rd harmonic: one and a half waves - etc.

Each mode has a characteristic frequency.

Fourier's insight: Any complex vibration is the sum of these simple vibrations!

\[f(x) = \sum_{n=1}^{\infty} \left[a_n \cos(nx) + b_n \sin(nx)\right]\]

The Fourier Series¶

For a periodic function \(f(x)\) with period \(2\pi\):

\[f(x) = \frac{a_0}{2} + \sum_{n=1}^{\infty} \left[a_n \cos(nx) + b_n \sin(nx)\right]\]

Where the coefficients are:

\[a_n = \frac{1}{\pi} \int_{-\pi}^{\pi} f(x) \cos(nx) \, dx\]

\[b_n = \frac{1}{\pi} \int_{-\pi}^{\pi} f(x) \sin(nx) \, dx\]

Why does it work? Sines and cosines form an orthogonal basis (like Cartesian axes). Every function is a linear combination of these "basis vectors".

The Fourier Transform (Continuous)¶

From Series to Transform¶

The series works for periodic functions. But most real signals (audio, images) aren't periodic!

The extension: Consider period \(T \to \infty\). Discrete frequencies become continuous.

The Formula¶

For a function \(f(t)\) (often time):

Forward transform (from time to frequency):

\[\hat{f}(\omega) = \int_{-\infty}^{\infty} f(t) e^{-i\omega t} \, dt\]

Inverse transform (from frequency to time):

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

Where: - \(t\): time variable - \(\omega = 2\pi f\): angular frequency (rad/s) - \(i\): imaginary unit \(\sqrt{-1}\) - \(\hat{f}(\omega)\): frequency spectrum

Interpretation¶

\(\hat{f}(\omega)\) tells us how much of frequency \(\omega\) is present in \(f(t)\).

Optics analogy: A prism decomposes white light into colors (frequencies). The Fourier Transform is the "mathematical prism" for signals.

Derivation: Why This Formula?¶

Step 1: Fourier Series Expansion¶

For periodic functions \(f(t)\) with period \(T\):

\[f(t) = \sum_{n=-\infty}^{\infty} c_n e^{i\omega_n t}\]

Where \(\omega_n = \frac{2\pi n}{T}\) are discrete frequencies.

The coefficients:

\[c_n = \frac{1}{T} \int_{-T/2}^{T/2} f(t) e^{-i\omega_n t} \, dt\]

Step 2: Limit as Period Goes to Infinity¶

Extend the period: \(T \to \infty\)

Discrete frequencies \(\omega_n\) become continuous: - Spacing: \(\Delta\omega = \frac{2\pi}{T} \to d\omega\) - Discrete sum → Continuous integral

Rewrite:

\[c_n = \frac{\Delta\omega}{2\pi} \int_{-T/2}^{T/2} f(t) e^{-i\omega_n t} \, dt\]

Define:

\[\hat{f}(\omega_n) = \int_{-T/2}^{T/2} f(t) e^{-i\omega_n t} \, dt = \frac{2\pi c_n}{\Delta\omega}\]

Step 3: The Integral Representation¶

The series becomes:

\[f(t) = \sum_{n} c_n e^{i\omega_n t} = \sum_{n} \frac{\hat{f}(\omega_n)\Delta\omega}{2\pi} e^{i\omega_n t}\]

In the limit \(T \to \infty\):

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

With:

\[\hat{f}(\omega) = \int_{-\infty}^{\infty} f(t) e^{-i\omega t} \, dt\]

Here are the Fourier Transform formulas!

Key Properties¶

1. Linearity¶

\[\mathcal{F}[af(t) + bg(t)] = a\hat{f}(\omega) + b\hat{g}(\omega)\]

2. Time Shift¶

\[\mathcal{F}[f(t - t_0)] = e^{-i\omega t_0} \hat{f}(\omega)\]

Effect: Phase modulation in frequencies.

3. Frequency Shift (Modulation)¶

\[\mathcal{F}[e^{i\omega_0 t}f(t)] = \hat{f}(\omega - \omega_0)\]

Effect: Spectrum shifting (used in radio).

4. Convolution ↔ Multiplication¶

\[\mathcal{F}[f * g] = \hat{f}(\omega) \cdot \hat{g}(\omega)\]

Where \((f * g)(t) = \int f(\tau)g(t-\tau)\,d\tau\) is convolution.

This is gold!: Convolution (complex) in time becomes simple multiplication in frequency.

5. Differentiation¶

\[\mathcal{F}\left[\frac{df}{dt}\right] = i\omega \hat{f}(\omega)\]

Effect: Differentiating in time = multiplying by \(i\omega\) in frequency.

Fundamental Examples¶

1. Dirac Delta¶

\[f(t) = \delta(t) \quad \Rightarrow \quad \hat{f}(\omega) = 1\]

Meaning: An instantaneous impulse contains all frequencies (white noise).

2. Gaussian¶

\[f(t) = e^{-at^2} \quad \Rightarrow \quad \hat{f}(\omega) = \sqrt{\frac{\pi}{a}} e^{-\omega^2/4a}\]

Meaning: The Fourier transform of a Gaussian is a Gaussian! (Uniqueness of Gaussian).

3. Boxcar (Rectangular Window)¶

\[f(t) = \begin{cases} 1 & |t| \leq T/2 \\ 0 & \text{otherwise} \end{cases}\]

\[\hat{f}(\omega) = T \cdot \text{sinc}\left(\frac{\omega T}{2}\right) = T \cdot \frac{\sin(\omega T/2)}{\omega T/2}\]

Meaning: A narrow time window → wide spectrum (uncertainty principle).

4. Decaying Exponential¶

\[f(t) = e^{-at}u(t) \quad \Rightarrow \quad \hat{f}(\omega) = \frac{1}{a + i\omega}\]

Where \(u(t)\) is the Heaviside step function.

Practical Applications¶

Audio: Equalization and Compression¶

MP3: Removes inaudible frequencies (psychoacoustic model)
Equalizer: Amplifies/attenuates frequency bands
Noise reduction: Filter out specific frequency ranges

Image Processing¶

JPEG: Removes high-frequency components (subtle details) to compress
Blur/sharpen: Low-pass or high-pass filtering
Edge detection: High-frequency enhancement

Communications¶

Modulation: Shift signal spectrum to carrier frequency
Multiplexing: Multiple signals in different frequency bands
Filtering: Remove interference at specific frequencies

Quantum Mechanics¶

Position ↔ Momentum: Fourier transform pair
Wave functions: Decomposed into momentum eigenstates
Heisenberg Uncertainty: Conjugate variables related by Fourier

The Uncertainty Principle¶

In quantum mechanics, position \(x\) and momentum \(p\) are Fourier transform pairs:

\[\psi(p) = \frac{1}{\sqrt{2\pi\hbar}} \int_{-\infty}^{\infty} \psi(x) e^{-ipx/\hbar} \, dx\]

This leads to the Heisenberg Uncertainty Principle:

\[\Delta x \cdot \Delta p \geq \frac{\hbar}{2}\]

General principle: You can't simultaneously know a signal's exact time location and exact frequency. A sharp pulse in time requires all frequencies; a pure tone exists forever in time.

Simple Harmonic Motion — The building blocks
Central Limit Theorem — Uses characteristic functions (Fourier transforms of PDFs)
Gaussian Distribution — The unique self-transform
Laplace Transform — For causal systems (time ≥ 0)
Wavelet Transform — Time-frequency localization
FFT Algorithm — Fast computation (Cooley-Tukey)

References¶

Fourier, J. (1822). Théorie Analytique de la Chaleur. Paris: Firmin Didot.
Dirichlet, P. G. L. (1829). "Sur la convergence des séries trigonométriques." Journal für die reine und angewandte Mathematik.
Bracewell, R. N. (1986). The Fourier Transform and Its Applications (2nd ed.). McGraw-Hill.
Stein, E. M., & Shakarchi, R. (2003). Fourier Analysis: An Introduction. Princeton University Press.