Fourier Decomposition

Every signal is a chord of pure tones.

Suggested next → The Harmonic Oscillator · PHYS

The brief

In 1807 Joseph Fourier, studying how heat spreads through metal, made a claim his examiners — Lagrange and Laplace among them — found so implausible they withheld the prize: any repeating shape, however jagged, can be built from nothing but smooth sine waves added together. It sounded impossible. That a sum of gentle curves could reproduce a sharp corner or a sudden jump offended the best mathematicians of the age, and pinning down exactly when it works took another century of analysis. Fourier was wrong in the details and right in the essential, and his idea turned out to be one of the most useful in all of applied mathematics — today every signal we record, send, or compress passes through some descendant of it.

Fourier's idea is that the world has two faces. There is the time domain, where a signal rises and falls moment by moment, and the frequency domain, where that same signal appears as a fixed recipe of pure tones — how much of each frequency it contains. The two descriptions hold exactly the same information, and the Fourier transform is the dictionary that carries you between them. What makes this more than bookkeeping is that many physical systems — heat, sound, light, the quantum wavefunction — become far simpler in the frequency domain, because a sine wave is the one shape that differentiation leaves essentially unchanged. Problems that are tangled in time fall apart into independent, easily solved pieces once viewed as frequencies. A square wave makes the trick concrete: take a sine at the base pitch, add a third of the sine at three times the frequency, a fifth of the one at five times, and the running sum steadily squares off — though near each jump it always overshoots by about 9%, a stubborn ripple called the Gibbs phenomenon that is exactly the sense in which Fourier's flat claim was 'wrong.' The property that made the whole thing indispensable came later: the convolution theorem, which says that the messy operation of filtering — smearing every point of a signal across its neighbors — becomes plain multiplication once you are in the frequency domain. That single fact is why sharpening a photo, cleaning up audio, or solving a differential equation is so often easiest there, and why technologies from image compression to medical imaging quietly do their real work not in time but in frequency.

Why nowThe idea only conquered the world once it became fast. The Fast Fourier Transform, published by Cooley and Tukey in 1965, cut the cost of the computation so drastically that the entire digital age — wireless standards, audio and video codecs, medical scanners — became practical almost overnight; it is among the most consequential algorithms ever written. The same duality also imposes a limit that rhymes with Heisenberg's: a signal cannot be pinned down sharply in both time and frequency at once, the trade-off that later gave rise to wavelets for studying sudden, bursty events. The governing image — that every signal is a chord of pure tones — is that rare metaphor which is also literally true.

Further readingBracewell's The Fourier Transform and Its Applications (1965, multiple editions) is the standard engineer's text — clear, applied, durable. For the underlying mathematics, Stein and Shakarchi's Fourier Analysis: An Introduction (2003) is the modern graduate entry. The compressed-sensing perspective and the FFT's algorithmic life are best caught in Strang's Introduction to Linear Algebra and his MIT lectures. For the historical arc — heat conduction to MP3 — Higgins's Fourier Analysis (1998) tells the story without losing the math.