Gabriel's Horn · Torricelli's Trumpet · 1643

The Horn You Can Fill but Never Paint

Spin the curve y = 1/x around and you get a trumpet that never ends. It holds only π cups of paint — a little over three. Yet its inner surface is infinite, so the story goes: you can fill it, but you could never paint it. Pull the horn longer below and watch it happen — the volume creeps to π and stops; the surface never stops. Then watch the paradox come apart in your hands.

In the early 1640s, before calculus had its machinery, Evangelista Torricelli found a solid that was infinitely long and yet enclosed a finite volume. It drew immediate dispute — Roberval, one of the finest geometers in France, first declared it impossible — and fed a running argument over whether an actually-infinite object could be reasoned about at all. Three centuries later someone else noticed its surface was infinite while its volume stayed finite, dressed the pair as a paradox about paint, and hung the archangel's name on it. Here is the object, rebuilt so you can operate it — and here, at the end, is why the paradox was never one.

Companion film — 2:33 The horn drawn from y = 1/x; the race, with the volume flattening against π while the surface sails through and keeps climbing (S ≈ 2π·ln b + 0.70996…); the fill pouring to the tip and the ratio V/S → 0; and the record-correction: Torricelli proved the volume only; the painter's paradox is a much later addition, and an equivocation on the word paint. Every number on screen is computed live by the same functions this page and research/gabriels-horn/verify.mjs (28/28) use; the music is built from the horn's own harmonic series (partials 1/n = the radii 1/x). No claim on screen the verifier does not assert.
The encoded film still says 'under π', calls V/S a film thickness, and gives the incorrect coating argument. Its source text is corrected; the video has not been rerendered. See Corrections.
The horn — pull it longer endpoint b = 10

Volume (paint to fill it)

2.827

V = π(1 − 1/b)  →  π ≈ 3.1416

Inner surface (skin to paint)

15.18

S = 2π∫₁ᵇ (1/x)√(1+1/x⁴) dx  >  2π·ln b · no ceiling

At b = 10 the horn is nine units long. Its volume is already 2.83, most of the way to π. Its surface is 15.2 and climbing with no ceiling in sight.

The volume of the whole endless horn is exactly π — the supremum it approaches but no finite length ever quite reaches. The surface has no such ceiling: because √(1+1/x⁴) > 1 everywhere, the surface is strictly larger than 2π·ln b, and ln b marches to infinity. One quantity saturates; the other diverges — logarithmically, but without bound. Watch them race.

The race — one stops, one never does y-axis: cubic / square units · x-axis: log₁₀ b

Both curves start together at b = 1 (an empty horn). The blue volume flattens against the dashed line at π and can never cross it. The gold surface sails straight through and keeps climbing. That gap — finite below, infinite above — is the whole paradox, drawn.

So: fill it, then try to paint it

Here is the trap the paradox sets. Filling the whole horn takes exactly π cups: finite. Painting the inside, the argument says, must take infinite paint because the surface is infinite. But any paint inside is part of that finite fill. A fixed positive thickness eventually cannot fit the narrowing tail. Operate it and see exactly where the sleight of hand lives.

Fill it · coat it · a finite inside b = 1000

To fill it

3.14

cups · = V(b), capped at π

Inside coat, radial depth h

3.14

cups · C(h,b) ≤ V(b)

Volume-to-area ratio

0.071

V/S → 0 · not local thickness

Pick an inward radial depth h, then pull the horn longer. Each cross-section gets an annulus of paint, capped at the local radius 1/x. Where h ≥ 1/x, the whole disc fills. The coat never exceeds the fill.

There it is. The ideal mathematical fill touches every point of the infinite inner surface with just π cups. Its radial depth varies with x. The ratio V/S tends to zero as the horn lengthens, but it is volume divided by area, not the fill's local thickness. The famous impossibility, "you'd need infinite paint", does hold for a coat of fixed positive radial thickness on the outside. That is a sufficient condition, not a necessary one: some thinning outside coats also have infinite volume. Inside, a coat cannot exceed the fill. Keep those requests distinct and there is no contradiction.

The check — every number recomputed in front of you

The volume is the exact antiderivative π(1−1/b). The surface is the integral 2π∫₁ᵇ(1/x)√(1+1/x⁴)dx, evaluated live (via the substitution u=ln x, which makes it the well-behaved 2π∫₀^{ln b}√(1+e⁻⁴ᵘ)du). These are the figures the offline verifier reproduces by two independent integrators.

endpoint bV(b) · volumeS(b) · surface flat area ln bratio V/S

The volume converges to π ≈ 3.141593; the surface diverges as 2π·ln b + 0.70996… (the constant checked to six places); even the flat region you spun — the area under 1/x — is already infinite, so the finite volume is no accident of a small region but the squeeze of the r² in a volume integral. Run it yourself: node research/gabriels-horn/verify.mjs (28/28).

What's exactly true, what's the sleight of hand, and what Torricelli actually did

Exactly true. The solid of revolution of y=1/x for x≥1 has volume π and infinite lateral surface area — both provable in one line of calculus, both reproduced here live and offline. There is no error in either number, and no error in the sentence "finite volume, infinite surface." That much is not a trick.

The sleight of hand is entirely in the word paint. To "fill" is to supply a volume; to "paint" is usually taken to mean a coat of some fixed positive thickness. Those are different asks, and the paradox smuggles you from one to the other. Under a single honest definition the mystery evaporates: (a) the ideal fill touches the whole inner surface, but V/S is a volume-to-area ratio, not a local film thickness; (b) a fixed positive thickness cannot fit inside the narrowing tail. Our radial model uses C(h,b) = π∫₁ᵇ[1/x² − max(1/x−h,0)²]dx, so 0 ≤ C(h,b) ≤ V(b). The product h·S(b) is only a local thin-coat estimate, not an inside volume for arbitrarily long horns. Outside, a fixed radial thickness has volume π[2h·ln b + h²(b−1)] → ∞. Even the thinning outside coat h(x)=1/ln x for x≥e has infinite volume: its annular area is at least 2π/(x·ln x), whose integral diverges.

The category error underneath it. Comparing "π of volume" with "∞ of area" and calling it a contradiction compares a cubic quantity with a square one. Volume and area are incommensurable; the only bridge is the word paint, and paint is a volume. Once you notice that, the two facts stop arguing with each other — they were never about the same kind of thing.

And a real horn? There isn't one. Past the point where the tube is narrower than a paint molecule — long before x is large — nothing can flow in, so a physical horn can be neither filled nor painted. The object is purely mathematical; "paint" is a metaphor that snaps at the scale of atoms. That the metaphor breaks is the honest end of the story, not a loophole.

What Torricelli actually proved (and what he didn't). In his tract De solido hyperbolico acuto (written 1643, published 1644 in the Opera geometrica), Evangelista Torricelli showed that this "acute hyperbolic solid" — a rectangular hyperbola spun about one of its asymptotes — though infinitely long, has a finite volume, equal to that of a definite finite cylinder. He proved it two ways: with Cavalieri's method of indivisibles (two generations before Newton and Leibniz gave calculus its machinery) and with a classical Archimedean exhaustion argument for the doubters. The result provoked real dispute — Gilles de Roberval at first held it false and impossible, and it fed a seventeenth-century argument about actual infinity and the legitimacy of indivisibles (the surprise, some historians stress, was about method and infinity, not that "finite volume" was intuitively absurd). Crucially, Torricelli's work was about volume only: the infinite surface area, and the "painter's paradox" that pairs it with the finite volume, are a much later, pedagogical addition — not his, and I found no reliable first-namer for it. The names are later still: neither "Torricelli's trumpet" nor "Gabriel's Horn" was used by Torricelli (into the 1700s reference works still called it the hyperbolicum acutum); they are modern coinages. "Gabriel's Horn" evokes the archangel Gabriel, whose horn sounds the Last Judgment — a finite mouthpiece opening onto an infinite bell, the earthly touching the divine.

Ground Truthcalculusinfinity Torricellirecord-correctinginteractive

Corrections

Corrected 2026-10-14: This page said filling the whole horn takes under π cups. Its volume is exactly π; only a finite truncation holds less, found by the assay line, 2026-10-14.

Corrected 2026-10-14: This page treated h·S as an inside-coat volume that grows without bound. Any inside paint is bounded by the fill; a fixed positive thickness eventually cannot fit the narrowing tail. The instrument now shows an inward radial coat capped at the local radius, found by the assay line, 2026-10-14.

Corrected 2026-10-14: This page said infinite paint is needed only for constant thickness. A fixed positive radial thickness outside is sufficient, not necessary: some thinning outside coats also have infinite volume, found by the assay line, 2026-10-14.

Corrected 2026-10-14: This page called V/S the fill's film thickness. It is a volume-to-area ratio, not the fill's local thickness; the ratio still tends to zero as the horn lengthens, found by the assay line, 2026-10-14.

Corrected 2026-10-14: This page said the horn is ten units long at b = 10. From x = 1 to x = 10 its axial length is nine units, found by the assay line, 2026-10-14.

Corrected 2026-10-14: This page gave the film's uploadDate as 2026-07-09. The MP4 first entered this repository at its current path on 2026-07-03, the site's basis for that field, found by the assay line, 2026-10-14.

Corrected 2026-10-14: At the length slider's minimum, this page displayed infinite V/S after rounding b to 1. There V = S = 0 and V/S is undefined; the slider now stays above 1, found by the assay line, 2026-10-14.

The claims record 145 claims re-read against their sources, 14 October 2026

Written 2026-07-03. Claims re-read against their sources on 2026-10-14: 145 checked, 106 confirmed, 34 wrong, 3 unverifiable, 2 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/cited-1014/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex and a Claude agent.

This was the directed 2026-10-14 repair pass over all 37 gabriels-horn triage items, applied to repository baseline 67088387e0. The fixer reread the evidence before editing, corrected the permitted page, Markdown, film source and verifier copies, and retained the listed MINOR findings. The record also carries the checkers' source-backed confirmations with explicit provenance; superseded thickness and length interpretations are narrowed to what the evidence confirms. Three earlier historical claims remain UNVERIFIABLE, including two disputed checker confirmations, with no extra page edits. Encoded media and shared data remain outside the authorized edits, as recorded below. MINOR counts are included in wrong; the by line identifies the prior two-checker source pass, not a new Claude run during fixing.

This record is long (145 claim lines and 36 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.

The assay office: what a record is, and every layer re-read so far.