How a drawing
becomes a sound

The claim

Slow Fourier makes one promise: what you draw and what you hear are a single mathematical object rendered twice. Not a theme. Not a vibe-matched soundtrack. The same list of numbers, poured into your eye as spinning circles and into your ear as harmonics. The pipeline has four honest steps — your line becomes numbers, the numbers become a recipe, the recipe becomes circles, and the recipe becomes sound — and each step is two centuries old, battle-tested, and running in your phone right now for a very different reason than the one it was invented for.

The unlikely life of Joseph Fourier

Jean-Baptiste Joseph Fourier was born in Auxerre in 1768, the son of a tailor, and orphaned before he was ten. The local church schooled him; by fourteen he was devouring mathematics by candle-stubs collected from around the school. He wanted to be an artillery officer — the army of the ancien régime had no place for a tailor's son — so he taught instead, and then the Revolution swallowed everyone's plans including his. He joined his local revolutionary committee, spoke up for victims of the Terror, and was arrested for it in 1794; the story is told that only Robespierre's fall kept his appointment with the guillotine from being kept.

Freed, he studied at the short-lived École Normale under Lagrange and Laplace — the two men who would later spend years telling him he was wrong — and taught at the new École Polytechnique. In 1798 Napoleon took him to Egypt as a scientific advisor, where he became secretary of the Institut d'Égypte and helped assemble the monumental Description de l'Égypte. In 1802 Napoleon made him Prefect of Isère, in Grenoble: a provincial governor's job of draining swamps and building the road to Turin. It was there — administrator by day — that he did the work that carries his name.

The problem was heat: how temperature spreads through a solid body. In 1807 he sent the Paris Institute a memoir with a scandalous method inside it. To solve his heat equation, Fourier claimed that essentially any function — any curve you could draw, corners and jumps included — could be written as a sum of sines and cosines. The committee (Lagrange, Laplace, Monge, Lacroix) balked; Lagrange in particular refused to accept it. Trigonometric sums were smooth, periodic, tame things — how could tame things add up to an arbitrary scrawl? The Institute set the topic as its 1811 grand prize; Fourier won it, with a citation pointedly noting his lack of rigor; publication of the full theory waited until his 1822 masterwork, Théorie analytique de la chaleur. Dirichlet supplied the rigorous convergence conditions in 1829, one year before Fourier died — vindicating the tailor's son over the objections of the two greatest mathematicians of the previous age.

Three footnotes complete the picture. First, the sine-sum idea was born in a fight about music: in the 1750s, d'Alembert, Euler, and Daniel Bernoulli argued over the vibrating string, and Bernoulli insisted a string's motion was a superposition of its pure modes — its harmonics. Fourier turned a musician's intuition into a universal method. Second, Fourier also wrote, in 1824, the first scientific description of what we now call the greenhouse effect — the man who decomposed heat also noticed the atmosphere keeping it in. Third, as the story goes, his Egyptian years left him convinced heat itself was health: he kept his rooms stifling and wrapped himself in blankets to the end, dying in Paris in 1830. He is buried in Père Lachaise; his name is on the Eiffel Tower; and his transform, in its fast modern form, is arguably the most-run mathematics on Earth — inside every song, call, photo, and MRI. Slow Fourier runs it slowly, for the pleasure of watching.

From drawing to math

Step one — your line becomes numbers

When you lift your finger, the app walks along your stroke and takes 128 evenly spaced samples. Each sample is a position on the screen — an x and a y — and here mathematics makes its first elegant move: a point on a plane can be treated as a single number, a complex number z = x + iy. Your whole gesture collapses into a short list: z₁, z₂, … z₁₂₈. Nothing is lost that your eye could see — 128 samples around a loop can capture up to 64 full wiggles, far more than a human hand puts into ten seconds of drawing.

Step two — the recipe is extracted

Now the discrete Fourier transform interrogates the list. Picture a set of probes, one for every whole-number speed k: probe k is a point spinning around a circle exactly k times per trip. For each probe, the transform multiplies your curve against it and averages the result. If your curve contains no motion at that speed, the products swirl in every direction and the average cancels to nothing. If your curve does turn at that speed, the products line up — and what survives the averaging is the coefficient c_k: a single complex number saying how much of speed-k rotation your line contains, and at what angle it starts. It is resonance as arithmetic — each probe is a tuning fork, and your drawing rings the ones it contains.

c_k = ¹⁄₁₂₈ Σ_n z_n e^{−2πikn/128} the interrogation, formally · one multiply-and-average per speed

Step three — the recipe is circles

Each surviving coefficient is, geometrically, a circle: its radius is the amount |c_k|, its rotation speed is k, its starting angle is the phase. Chain the circles tip to tail — the fastest riding on the slower — and let them all spin. The tip of the chain retraces your line. Exactly. That is the animation you watch when you pause: not a metaphor for the math but the partial sums of your own Fourier series, drawing you back to yourself. Ptolemy did this by hand for the planets two thousand years ago; Fourier proved the circles could draw anything.

And here is the fact worth a pull-quote: the number of circles you need is a smoothness meter. Mathematically, the smoother the curve, the faster its coefficients decay — a calm oval collapses into two or three fat, slow circles, while a jittery scribble scatters its energy into dozens of small fast ones. Roughness is high-frequency content. Hold that thought; it is about to become audible.

From math to music

What a sound actually is

A sound is air pressure wobbling. The simplest possible wobble — a pure sine wave — is a pure tone, the sound of nothing but one frequency. Real instruments never make just one: a violin playing A makes A and, stacked quietly on top, 2×A, 3×A, 4×A — its harmonics. The recipe of how strong each harmonic is is what your ear recognizes as violin-ness or flute-ness. That recipe has a name: the spectrum. Timbre is a Fourier series — Helmholtz built the theory in 1863, and your own cochlea agrees, because the basilar membrane in your inner ear physically sorts incoming sound by frequency before your brain ever hears "a sound." You have been doing Fourier analysis, mechanically, since before you were born.

The pour

So the app does the only honest thing available. It takes your curve's harmonic amounts — A_k = |c_k| + |c_−k| — and hands that exact list to the synthesizer as the recipe of a waveform, rooted at 110 Hz, the low A the whole room hums in. Your drawing's spectrum becomes, without translation, a sound's spectrum. Draw a centered circle and the recipe holds a single ingredient: one pure sine, the cleanest tone a speaker can make. Draw a spiky star and you have drawn brightness — energy in the high harmonics, a reedy, complex chord. The SPECTRUM panel in the corner of the app is not decoration; it is the actual array the oscillator received.

The melody — and an honest asterisk

The tune layered on top works differently, and we say so. As the playhead walks your line, the app samples your distance from the gesture's center; farther means higher. Raw distance would land between notes and sound sour, so each value snaps to the nearest tone of the A-minor pentatonic scale — five notes whose harmonics overlap so generously that no pair of them clashes. That snapping is why a random scribble comes out consonant: the lattice contains no wrong notes. But be clear about the epistemology, because your audience will be: the timbre is a discovery — the same object, seen and heard. The melody is a design — a mapping we chose because it is the one you can hear yourself in. Spiral outward and the melody climbs. Discovery plus disclosed design: that is the whole trick, and the app's marginalia states the rule on screen.

Everything an audience asks

Is the sound really my drawing, or is that marketing?

The sustained tone really is: your curve's Fourier amplitudes are handed unaltered to an additive synthesizer, so the drawing's spectrum and the sound's spectrum are the same list of numbers. The melody is a chosen translation (distance → pitch), and the app discloses that on screen. One identity, one design — both honest.

Why does a circle sound pure and a scribble sound bright?

A centered circle is one rotation at one speed — one coefficient, one sine, one pure tone. A scribble's corners and jitters are, mathematically, high-frequency content, which becomes energy in high harmonics — which your ear calls brightness. Roughness is treble, in ink exactly as in sound.

Why don't random drawings sound like random noise?

Three constraints, all disclosed: pitches snap to a pentatonic lattice with no dissonant pairs, everything shares one root note with the ambient drone, and the rhythm rides a fixed slow pulse. Your randomness chooses which consonant thing happens, never whether it is consonant.

What exactly is inside a share link?

About 96 bytes: the top Fourier coefficients of your last stroke — for each, a speed, an amount, and a starting angle — plus your color and symmetry setting. No image, no ID, no server record. The recipient's phone re-derives the drawing and re-synthesizes the sound from the math alone.

Does sampling at 128 points lose my drawing?

128 samples around a closed loop faithfully capture up to 64 oscillations of detail — several times more wiggle than a relaxed hand produces in a stroke. What the samples smooth away is precisely the tremor you were happy to lose.

Why compute it the slow O(N²) way when the FFT exists?

Because 128 points cost 16,384 multiply-adds — a rounding error to a phone — and because the slowness is the brand. The fast Fourier transform (Cooley & Tukey, 1965; Gauss had it in an unpublished manuscript around 1805) exists to serve speed. This one exists to be watched.

Did Fourier invent this for music or art?

Neither — for heat flowing through solid bodies, while he governed a French province. But the idea's grandparents were musical: the 1750s vibrating-string debate, where Daniel Bernoulli insisted a string's motion is a stack of its pure harmonics. Fourier universalized a musician's intuition; Helmholtz carried it back into the ear in 1863; we carry it back onto the page.

What is a "dihedral group," in plain words?

The complete set of ways to rotate and mirror a shape so it lands on itself — a snowflake's symmetries, formalized. In 8-fold mode the app applies all sixteen members of that set to your stroke, live. The folds button is a group selector wearing a cardigan.

Is my ear really "doing Fourier"?

To a good first approximation, mechanically, yes: different frequencies peak at different places along the basilar membrane, so sound arrives at your brain already sorted by frequency. The full story of hearing is richer, but "the cochlea is a frequency analyzer" is textbook, not poetry.

Why the key of A? Why 110 Hz?

The drone, the bells, your line's voice, and the stroke-timbre all share A as a root so that every layer is consonant with every other by construction. 110 Hz (A2) is low enough to feel like a room, high enough to bloom on phone speakers.

The coefficients have angles too — do I hear those?

Barely. For steady tones the ear is largely phase-deaf — an observation old enough to be called Ohm's acoustic law. So the phases do their work where they matter, in the drawing's geometry, and the amplitudes do the singing.

Is any of this AI?

None. Every step is deterministic, two-century-old mathematics computed on your device. No models, no cloud, no data collected — the same input always sings the same song.

Is it "Fourier" or "Fournier"?

Fourier — FOO-ree-ay — no n. Fournier is a fine French surname that belongs to bakers (it means "oven-keeper"), which, given the man died swaddled in blankets chasing warmth, is an almost forgivable mistake.