Skip to content

The Horn That Cannot Be Painted

Monday, November 16, 2026

After writing about German heraldic buffalo horns, I should probably leave the subject of horns alone. Unfortunately, mathematics has one of its own, and it is considerably stranger than anything a medieval German nobleman ever put on a helmet.

It is called Gabriel’s horn, or Torricelli’s trumpet, after the seventeenth-century mathematician Evangelista Torricelli. The horn extends infinitely far in one direction, but encloses a finite volume. That is odd enough, but it also has an infinite surface area. We can fill the horn with a finite quantity of paint, even though the usual explanation says we could never use that paint to coat the surface.

The arithmetic behind the first two claims is quite straightforward, at least for someone familiar with introductory calculus. The claim about paint deserves a little more attention.

Constructing the Horn

We start with a function most students encounter well before calculus:

y=1x,x≥1.y=\frac{1}{x},\qquad x\geq 1.

At (x=1), the value of the function is one. At (x=2), it is one-half. At (x=10), it is one-tenth. The curve gets closer and closer to the horizontal axis as (x) increases, but it never reaches the axis at any finite value of (x).

Now imagine revolving the curve around the (x)-axis. At each value of (x), the distance from the curve to the axis becomes the radius of a circle. We have constructed a three-dimensional horn with a circular opening of radius one at (x=1), narrowing as we move to the right. Its radius is always (1/x), and it continues indefinitely.

This is what calculus calls a solid of revolution. There is nothing exotic about that construction. In an introductory calculus course, we might use the same method to find the volume of a vase or the shape made by turning a curve on a lathe. Gabriel’s horn is different because there is no final value of (x) at which we stop turning.

The absence of a final point does not prevent us from asking how much space the horn encloses.

Filling the Horn

To calculate the volume, we can divide the horn into thin circular disks. Each disk has radius (r=1/x), so its cross-sectional area is (\pi r^2=\pi/x^2). Adding the volumes of the disks from (x=1) onward gives us an improper integral:

V=π∫1∞1x2 dx.V=\pi\int_1^\infty\frac{1}{x^2}\,dx.

An integral extending to infinity is calculated as a limit. We first stop the horn at some finite distance (b), find the volume of that truncated horn, and then see what happens as (b) increases:

V(b)=π∫1bx−2 dx=π[−1x]1b=π(1−1b).\begin{aligned} V(b) &=\pi\int_1^b x^{-2}\,dx\\ &=\pi\left[-\frac{1}{x}\right]_1^b\\ &=\pi\left(1-\frac{1}{b}\right). \end{aligned}

As (b) increases without bound, (1/b) approaches zero. The volume therefore approaches

V=π.\boxed{V=\pi}.

That is (\pi) cubic units, not (\pi) square units or (\pi) units of length. If our coordinate lengths are measured in meters, the idealized horn has a volume of approximately 3.14 cubic meters. It does not matter how far the horn extends. The volume never exceeds that limit.

We can see this without waiting for infinity. At (x=2), the horn already contains half its eventual volume. At (x=10), it contains 90 percent. By (x=100), it contains 99 percent. The horn beyond (x=100) still extends infinitely far, but the entire remaining volume is only (\pi/100) cubic units.

The horn never stops growing in length. Its volume continues increasing, but the additional volume becomes small enough that the total approaches a finite limit. This is the distinction between a sequence continuing indefinitely and the sum of its contributions growing without bound. The two are not the same.

Disks and Shells

There is another way to find the volume, and it provides a useful check on the first calculation. What we just used is the disk method, in which we add up circular cross sections perpendicular to the axis of rotation. In Computational Methods for Numerical Analysis with R, I implemented both the disk method and the cylindrical shell method in the cmna package. The functions are named discmethod() and shellmethod(), and they approximate definite integrals numerically. The implementation uses the usual textbook orientation, with disks about the (x)-axis and shells about the (y)-axis. We can use the same shell geometry around the (x)-axis by taking horizontal, rather than vertical, shells.

Instead of cutting Gabriel’s horn into disks, imagine dividing it into thin cylindrical shells, one nested inside another. A shell at radius (y), with (0<y\leq1), begins at (x=1) and ends where the curve (y=1/x) reaches that radius. Solving for (x), the far end is (x=1/y). The length of the shell is therefore (1/y-1), and its circumference is (2\pi y).

Its volume is the circumference times its length times its infinitesimal thickness. Adding the shells gives

V=2π∫01y(1y−1) dy=2π∫01(1−y) dy=2π[y−y22]01=π.\begin{aligned} V &=2\pi\int_0^1 y\left(\frac{1}{y}-1\right)\,dy\\ &=2\pi\int_0^1(1-y)\,dy\\ &=2\pi\left[y-\frac{y^2}{2}\right]_0^1\\ &=\pi. \end{aligned}

Strictly, the first integral is improper at (y=0), where the shell length grows without bound. We evaluate it by approaching zero from above. The product (y(1/y-1)) has the finite limit one, so the integral converges. Both methods give precisely the same volume, even though one divides the horn lengthwise into shells and the other divides it crosswise into disks.

The CMNA source offers a practical way to experiment with either method on finite intervals. That limitation matters here. A numerical routine does not need to accept infinity as an endpoint for us to study an improper integral. We calculate a sequence of finite approximations and examine what happens as the bound increases, which is exactly what the mathematical limit describes.

Painting the Surface

The surface calculation looks similar, but it has a different result. If we divide the outside into thin bands, each band goes around a circle whose circumference is (2\pi r). We also need to account for the slope of the horn, since the bands lie on a curved surface rather than on a cylinder.

The standard surface-area formula for a solid of revolution is

A=2π∫aby1+(dydx)2 dx.A=2\pi\int_a^b y\sqrt{1+\left(\frac{dy}{dx}\right)^2}\,dx.

In our case, (y=1/x) and (dy/dx=-1/x^2). So, for the horn extending to (x=b),

A(b)=2π∫1b1x1+1x4 dx.A(b)=2\pi\int_1^b\frac{1}{x}\sqrt{1+\frac{1}{x^4}}\,dx.

This looks unpleasant to integrate directly. Fortunately, we do not have to. The square-root term is always greater than one, so the actual surface area must be greater than the result we get by removing it:

A(b)>2π∫1b1x dx=2πln⁡b.\begin{aligned} A(b) &>2\pi\int_1^b\frac{1}{x}\,dx\\ &=2\pi\ln b. \end{aligned}

As (b) approaches infinity, so does (\ln b), although it grows very slowly. The surface area is always larger than a quantity that grows without bound. Accordingly, the horn has infinite surface area.

We have two entirely legitimate calculations. The volume is (\pi) cubic units and the lateral surface area is infinite. Including a circular cap at the wide end would add a finite (\pi) square units of area and change nothing about the result.

The reason this can happen is that the cross-sectional area depends on the square of the radius, while the circumference depends on the radius itself. The horn gets narrower in both respects, but its volume contributions decrease much faster than its surface contributions.

Cutting It into Pieces

There is a nice way to see the difference without relying on the final values of either integral. We can divide the horn into sections, with each section covering twice the lengthwise coordinate range of the preceding section. The first runs from (x=1) to (x=2), the second from (2) to (4), the third from (4) to (8), and so on.

The volume in the section from (x=2^k) to (x=2^{k+1}) is

Vk=π∫2k2k+11x2 dx=π2k+1.\begin{aligned} V_k &=\pi\int_{2^k}^{2^{k+1}}\frac{1}{x^2}\,dx\\ &=\frac{\pi}{2^{k+1}}. \end{aligned}

Each new section contains half as much volume as the preceding one. Adding all the sections gives us the familiar geometric series:

π2+π4+π8+⋯=π.\frac{\pi}{2}+\frac{\pi}{4}+\frac{\pi}{8}+\cdots=\pi.

Now consider the surface area of those same sections. Because the square-root term in the surface-area formula is greater than one, each section has more surface area than

2π∫2k2k+11x dx=2πln⁡2.2\pi\int_{2^k}^{2^{k+1}}\frac{1}{x}\,dx =2\pi\ln 2.

That lower bound is the same for every section. We can keep doubling the distance, and each new section contributes less volume than the one before, but more than (2\pi\ln 2) square units of surface area.

There are infinitely many such sections. Their volumes add to a finite number, while their surface areas cannot possibly add to a finite number. The distinction is not some peculiar failure of calculus. It is the different rates at which the two quantities shrink.

The Trouble with the Paint

The usual statement of the paradox is that we can fill Gabriel’s horn with a finite amount of paint but cannot paint its surface with any finite amount. That sounds persuasive. Once we know the surface area is infinite, it seems obvious that painting it must require infinite paint.

But there is an assumption hiding in that sentence. We have quietly assumed the paint must form a coat of some fixed, positive thickness. If every square unit of surface requires at least a certain positive volume of paint, then an infinite surface requires an infinite volume of paint. No argument there.

Mathematically, though, a coat does not have to be the same thickness everywhere.

Suppose we want to coat the inside of the horn and allow the coating to get thinner as the horn narrows. At any position (x), the inner wall is at radius (1/x). Let the paint occupy the region between radius (1/(2x)) and radius (1/x). The paint reaches the wall everywhere, and its radial thickness is (1/(2x)), which is positive at every finite (x) but approaches zero farther down the horn.

The volume of that paint is the volume of the larger disk minus the smaller disk, integrated over the entire length:

Vpaint=π∫1∞[(1x)2−(12x)2] dx=3π4∫1∞1x2 dx=3π4.\begin{aligned} V_{\mathrm{paint}} &=\pi\int_1^\infty \left[\left(\frac{1}{x}\right)^2 -\left(\frac{1}{2x}\right)^2\right]\,dx\\ &=\frac{3\pi}{4}\int_1^\infty\frac{1}{x^2}\,dx\\ &=\boxed{\frac{3\pi}{4}}. \end{aligned}

So, in our idealized mathematical model, we have covered the entire inner wall of a surface with infinite area using only (3\pi/4) cubic units of paint. The coating gets thinner without having a positive minimum thickness. That is enough to avoid the contradiction.

This is not a demonstration that somebody could manufacture or paint an actual infinitely long object. Real paint consists of molecules, and real materials cannot be divided into arbitrarily thin layers. Nor can a real horn extend indefinitely while maintaining the assumptions of classical geometry. John D. Cook makes the same essential distinction: the idealized horn is allowed to become arbitrarily narrow, while the usual painting argument demands a fixed thickness from the paint.

The equations are not in conflict. The casual statement about painting leaves out a condition that matters.

What Infinity Is Doing Here

The recent discussion of zero and the last integer turned on the difference between being able to produce another integer and treating an infinite collection as an established mathematical object. Gabriel’s horn raises a related issue about mathematical limits, but it is not the same question.

There is no last disk in the horn. No final band that brings the surface-area calculation to a close. For any finite cutoff we choose, the enclosed volume and surface area are both finite. We then ask how those quantities behave as the cutoff increases. One approaches a particular finite value, while the other grows without bound.

And both results are exact.

This is why I like Gabriel’s horn as a calculus example. It requires only an ordinary function, familiar formulas for circles, and two integrals. Yet the result forces us to distinguish the length of an object from the space it encloses, the space it encloses from the area of its boundary, and the properties of a mathematical model from the assumptions we bring to a physical analogy.

We can fill the horn. We can even describe a nonuniform coating that covers it. What we cannot do is give every point on its infinitely large surface a coat of paint with a fixed minimum thickness using a finite volume of paint.

And unlike the German heraldic variety, this is a horn that nobody ought to try wearing on a helmet.

Original illustration of Gabriel's horn with paint and calculus notation, generated with OpenAI on November 13, 2026. Stored locally as WebP.