Geometry

A shoemaker's knife with the area of a circle

Mark a point on a segment and draw three semicircles: one on the whole segment and one on each piece. The curved region left between them looks like the blade of a shoemaker's knife, and Archimedes called it the arbelos. Its area is exactly the area of the circle whose diameter is the half-chord standing at the point — wherever the point is. The circles fitting on either side of that half-chord are always twins; a chain of circles inside it stands on its own diameters; and the salt cellar of the same Book of Lemmas squares a circle the same way.

Worth reading first: A circle unrolled into a triangle · The five crescents that can be squared.

The circle cannot be squared: no construction with straightedge and compass produces a square of the same area, because π\pi is transcendental. The five crescents that can be squared found that some regions bounded by circular arcs can be squared nonetheless, because their areas involve no π\pi at all — Hippocrates’s lunes have exactly the area of a triangle. There is a second, opposite kind of exactness. Some regions bounded by circular arcs have exactly the area of a circle, with no rational multiple of anything else left over. They cannot be squared any more than a circle can, but they can be circled: a compass draws a disc of exactly their area.

The oldest and best of these is in a collection attributed to Archimedes, the Book of Lemmas, which survives only in an Arabic translation. It describes a figure made of three semicircles, the arbelos — a Greek word for the curved blade of a shoemaker’s knife — and proves that its area equals the area of a circle drawn on one particular segment of the figure. This essay draws the arbelos, proves that and three related facts, and measures what stays the same as the figure changes shape.

Three semicircles and one circle

A shoemaker's knife with the area of a circle. c=0.36: arbelos 0.18095574, circle 0.18095574, sampled 0.18094
Fig. 1 The arbelos, inside the semicircle on AB and outside the semicircles on AC and CB, beside the circle on CD. Both have area π⋅AC⋅CB/4\pi \cdot AC \cdot CB/4, and a grid of samples counted inside the arbelos agrees.

Take a segment ABAB of length 1 and a point CC on it, here with AC=0.36AC = 0.36 and CB=0.64CB = 0.64. Draw the semicircle on ABAB, and on the same side the semicircles on ACAC and on CBCB. The arbelos is what lies inside the big semicircle and outside both small ones: a crescent-like region with three cusps, at AA, CC and BB. At CC, draw the perpendicular to ABAB up to the big semicircle, meeting it at DD. Archimedes’ proposition is that the arbelos has the same area as the circle whose diameter is CDCD.

The figure computes both. The arbelos’s area, by the formula derived below, is π⋅0.36⋅0.64/4=0.180956\pi \cdot 0.36 \cdot 0.64 / 4 = 0.180956; the circle on CDCD has diameter 0.36⋅0.64=0.48\sqrt{0.36 \cdot 0.64} = 0.48 and area π⋅0.482/4=0.180956\pi \cdot 0.48^2 / 4 = 0.180956; and as a check independent of any formula, counting which of 720,000 grid points fall inside the arbelos gives 0.18094. The arbelos is a thin, curved, three-cornered thing, and the circle is a disc; nothing in their shapes suggests they hold the same amount of space.

A circle unrolled proved that a disc’s area is half its circumference times its radius by cutting it into rings and straightening them. The arbelos needs no such cutting. Its proof uses only one fact about circles — that a disc’s area is proportional to the square of its diameter — and one fact about right triangles, and the next section takes them in turn.

Why the areas agree

The half-chord is a geometric mean. c=0.36: CD^2=0.230400
Fig. 2 The triangle ADB stands on the diameter AB, so its angle at D is right and its height CD is the geometric mean of AC and CB. The square on CD equals the rectangle AC by CB.

A semicircle on a segment of length dd has area πd2/8\pi d^2/8. So the arbelos’s area is

π8(AB2−AC2−CB2).\frac{\pi}{8}\left(AB^2 - AC^2 - CB^2\right).

Since AB=AC+CBAB = AC + CB, squaring gives AB2=AC2+2⋅AC⋅CB+CB2AB^2 = AC^2 + 2 \cdot AC \cdot CB + CB^2, and the bracket is just 2⋅AC⋅CB2 \cdot AC \cdot CB. The arbelos’s area is π4⋅AC⋅CB\frac{\pi}{4} \cdot AC \cdot CB. Every π\pi-bearing term from the two small semicircles cancels against part of the big one, and what remains is the cross term of a square — the two rectangles in the familiar picture of (a+b)2(a + b)^2, rewritten as a curved region.

Now the circle on CDCD has area π4CD2\frac{\pi}{4} CD^2, so the claim is that CD2=AC⋅CBCD^2 = AC \cdot CB. That is the figure’s content. The point DD lies on the semicircle with diameter ABAB, so by Thales’s theorem the angle ADBADB is a right angle; an angle that does not move explained why an angle standing on a diameter is always a right angle. In a right triangle the altitude to the hypotenuse divides it into two similar triangles, ACDACD and DCBDCB, and similarity gives AC/CD=CD/CBAC/CD = CD/CB, which is CD2=AC⋅CBCD^2 = AC \cdot CB. The half-chord CDCD is the geometric mean of the two pieces, and the figure draws it literally: the square on CDCD next to the rectangle with sides ACAC and CBCB, both of area 0.2304.

So the proof is two ideas stacked: the arbelos’s area is a rectangle’s worth of π\pi, and the rectangle is a square by the geometric mean. Nothing about the position of CC was used. The equality holds for every split, and the last figure sweeps through all of them.

One right triangle behind the arbelos and the lunes

The right triangle ADBADB that made CDCD a geometric mean is the same triangle that Hippocrates used for his lunes, and seeing both results in one triangle explains why one region’s area is rational and the other’s is a circle’s. Pythagoras’s theorem, which two squares, four triangles, and no algebra proved by rearranging squares, holds for any figures similar to one another drawn on the three sides — semicircles included, since a semicircle’s area is a fixed multiple of the square of its diameter. So the semicircles on ADAD and DBDB together have exactly the area of the semicircle on ABAB.

Hippocrates drew the two small semicircles outward, on the far side of the legs. The big semicircle on ABAB contains the triangle, and the parts of the small semicircles sticking out beyond it are two lunes; subtracting, the two lunes together have exactly the area of the triangle ADBADB — a polygon, with no π\pi left. Archimedes, in the arbelos, drew his small semicircles on the pieces of the hypotenuse instead, ACAC and CBCB, and inward. The same cancellation of squares now leaves the cross term 2⋅AC⋅CB2 \cdot AC \cdot CB, which is 2⋅CD22 \cdot CD^2 by the geometric mean, and the arbelos comes out with a circle’s area.

So the two oldest exact areas of curved regions are two ways of reading one right triangle. Draw semicircles on the legs, and what survives the Pythagorean cancellation is the triangle; draw them on the pieces into which the altitude cuts the hypotenuse, and what survives is the square on the altitude, times π/4\pi/4. The first kind of region can be squared and the second can only be circled, and the difference is whether the surviving quadratic is an area of a polygon or the square of a length.

Archimedes knew both. His proof that a circle’s area lies between two sequences of polygons, which pinned between two sequences followed, is the work that settles what “the area of a circle” means at all; the arbelos is a lemma in the opposite direction, taking the circle’s area as known and showing that a much stranger region shares it. Neither result needs the value of π\pi — the arbelos’s equality holds whatever π\pi is, since both sides carry the same factor — and that is part of what makes it a theorem about shapes rather than a computation.

The twins

Twin circles on either side of the half-chord. c=0.25: rho=0.093750; c=0.4: rho=0.120000; c=0.62: rho=0.117800
Fig. 3 On either side of the half-chord, the circle touching the large arc, one small arc and the line CD. At three different splits, the two circles are always the same size.

The half-chord CDCD divides the arbelos into two curved triangles, one bounded by the arc on ACAC, the big arc and the line CDCD, the other by the arc on CBCB, the big arc and CDCD. Each contains a largest inscribed circle, touching all three of its sides. When CC is the midpoint, the two halves are mirror images and the circles are obviously equal. Archimedes proved that they are equal for every position of CC, however lopsided the arbelos — and these are Archimedes’ twin circles.

The figure draws them at three splits, AC=0.25AC = 0.25, 0.40.4 and 0.620.62, and checks each circle against its three tangencies to twelve decimal places. Their common radius is AC⋅CB/(2⋅AB)AC \cdot CB / (2 \cdot AB): with AB=1AB = 1, 0.25⋅0.75/2=0.093750.25 \cdot 0.75 / 2 = 0.09375 for the first split. The radius is symmetric in ACAC and CBCB, which is the twins’ equality written as a formula, but the circles themselves are not mirror images: the left twin sits higher when CC is towards BB, and the right twin sits higher when CC is towards AA.

The derivation is short. Put the left twin’s centre at horizontal position AC−ρAC - \rho, so that it touches the line CDCD, and require it to touch the arc on ACAC from outside and the arc on ABAB from inside. The two tangency conditions are two equations for its height and radius; subtracting them eliminates the height and leaves an equation linear in ρ\rho, whose solution is AC⋅CB/(2⋅AB)AC \cdot CB / (2 \cdot AB). The same calculation for the right twin, with the roles of ACAC and CBCB exchanged, gives the same expression, because the expression does not care which piece is which. The equality is therefore not a coincidence of the picture but a symmetry of the algebra: the problem of fitting a circle in either half has an answer symmetric in the two halves.

Since Archimedes, dozens more circles with the same radius have been found in the arbelos — circles defined by other tangency conditions that also come out with radius AC⋅CB/(2⋅AB)AC \cdot CB/(2 \cdot AB), and that geometers call Archimedean circles. The twins are the first two, and their common size is the one length every Archimedean circle shares.

A chain standing on its diameters

Pappus's chain stands on its diameters. 1: h/d=1.000000 2: h/d=2.000000 3: h/d=3.000000 4: h/d=4.000000 5: h/d=5.000000 6: h/d=6.000000
Fig. 4 Pappus’s chain, from the semicircle on CB: each circle touches the large arc, the arc on AC and the circle before it. The n-th circle’s centre stands exactly n of its own diameters above AB.

Pappus of Alexandria, around AD 320, added a theorem that is harder to see and easier to prove with modern tools. Starting from the semicircle on CBCB, fit a circle into the arbelos touching the large arc, the arc on ACAC and the semicircle on CBCB; then another touching the large arc, the arc on ACAC and that circle; and so on, a chain of circles shrinking towards AA. Pappus proved that the centre of the nn-th circle stands exactly nn of its own diameters above ABAB. The figure draws six circles with AC=0.66AC = 0.66, marks the heights of the first four, and checks every tangency and every height ratio — 1, 2, 3, 4, 5 and 6 diameters, to twelve places.

The reason is clearest after an inversion, which the map that trades circles for lines introduced: the map that sends each point at distance rr from a centre to the point on the same ray at distance k2/rk^2/r. Inversion centred at AA sends circles through AA to straight lines and preserves tangency. The large arc and the arc on ACAC both pass through AA, so they become two parallel lines perpendicular to ABAB. The chain’s circles do not pass through AA, so they become circles — all touching both parallel lines, and so all of the same size, stacked in a column like coins in a tube. Choosing kk so that the nn-th circle is mapped to itself, its centre is at the same height in both pictures, and in the column picture that height is plainly nn diameters above the axis. Pappus had no inversion; his proof was a long chain of similar triangles. The figure checks the same fact with the explicit formula for the chain, which inversion produces in one line.

The salt cellar

The salinon has the area of the circle on its axis. s=0.28: salinon 0.40715041 circle 0.40715041 sampled 0.40713
Fig. 5 Archimedes’ salinon: the semicircle on AB, minus two equal semicircles at its ends, plus the semicircle below on what remains between them. Its area equals the circle on its whole axis of symmetry.

The Book of Lemmas has a second region of the same kind, the salinon, or salt cellar. On ABAB mark two points at equal distances from the ends; remove the semicircles on the two end pieces from the semicircle on ABAB, and add the semicircle below ABAB on the middle piece. The resulting region is symmetric about the perpendicular bisector of ABAB, and its axis of symmetry runs from the top of the big arc to the bottom of the lower one. Archimedes proved that the salinon’s area equals the area of the circle with that axis as diameter.

The proof is the same algebra. With AB=1AB = 1 and end pieces of length ss, the area is π8(1−2s2+(1−2s)2)=π4(1−s)2\frac{\pi}{8}\left(1 - 2s^2 + (1 - 2s)^2\right) = \frac{\pi}{4}(1 - s)^2, and the axis has length 12+1−2s2=1−s\frac{1}{2} + \frac{1 - 2s}{2} = 1 - s. The figure draws the salinon with s=0.28s = 0.28 and computes both areas, 0.407150, and checks them against a grid of 810,000 samples. Again every square of a diameter has collapsed, by the expansion of a binomial, into the square of one length.

Both of Archimedes’ regions work for the same reason: a combination of semicircles whose diameters add up linearly has area π8\frac{\pi}{8} times a quadratic in the diameters, and when the quadratic happens to be a perfect square, the region has the area of a circle. The arbelos’s quadratic is 2⋅AC⋅CB2 \cdot AC \cdot CB, a square because of the geometric mean; the salinon’s is 2(1−s)22(1 - s)^2, a square outright.

Every position of C at once

Every position of C at once. 0.05:0.03731 0.1:0.07069 0.15:0.10014 0.2:0.12566 0.25:0.14726 0.3:0.16493 0.35:0.17868 0.4:0.18850 0.45:0.19439 0.5:0.19635 0.55:0.19439 0.6:0.18850 0.65:0.17868 0.7:0.16493 0.75:0.14726 0.8:0.12566 0.85:0.10014 0.9:0.07069 0.95:0.03731
Fig. 6 The arbelos’s area for every split, with the circle on CD measured at nineteen positions lying exactly on it. The arbelos is largest at the midpoint, where it fills half the big semicircle.

As CC slides from AA to BB, the arbelos changes from a sliver to a symmetric crescent and back. Its area π4AC⋅CB=π4c(1−c)\frac{\pi}{4} AC \cdot CB = \frac{\pi}{4} c(1 - c) is a parabola in the position cc, zero at both ends and largest at the midpoint, where it is π/16=0.1963\pi/16 = 0.1963, exactly half the area of the big semicircle. The figure plots it and puts a dot at nineteen positions for the circle on CDCD, computed from the measured length of CDCD; every dot lies on the curve. The twin circles’ area, π(c(1−c)/2)2\pi (c(1 - c)/2)^2, is drawn beneath it, a much flatter hump.

One more quantity stays fixed throughout: the arbelos’s boundary. Its upper boundary is the arc on ABAB, of length π2AB\frac{\pi}{2} AB, and its lower boundary is the two small arcs, of total length π2(AC+CB)=π2AB\frac{\pi}{2}(AC + CB) = \frac{\pi}{2} AB. The two inner arcs together are always exactly as long as the outer one, so the arbelos’s perimeter is π⋅AB\pi \cdot AB for every split. That is a much simpler fact than the area — lengths are linear in diameters, areas quadratic — and it says that a tightrope walker going from AA to BB along the top of the arbelos walks exactly as far as one going along the bottom, whatever the position of CC.

What the drawings cannot show

The figures check every equality numerically, to rounding, and the sampled areas to three decimal places. None of that is the proof; the proof is the algebra, which works for every split at once, while each figure checks one split or a few. The sweep comes closest to showing the general fact, but even it samples nineteen positions. What makes the numerical agreement convincing is that the two sides are computed by independent routes — the arbelos from its definition, the circle from a measured length — and agree to every digit at every position tried.

The pictures also hide what makes the twin circles twins. A drawing of two equal circles in an asymmetric figure looks like a coincidence, and the drawing cannot distinguish a coincidence from a theorem; only the algebra, symmetric in the two pieces, shows that equality is forced. Likewise the chain’s heights look like an accident of proportion in the figure; inversion is what explains them, and inversion is a transformation of the whole plane that no single drawing displays.

And the arbelos’s area, though equal to a circle’s, is still transcendental: it is π\pi times a product of lengths. Archimedes’ result does not square anything. It replaces one circular region by another, simpler one, which is all the exactness that a region bounded by full semicircles can have, since the circle that will not square is barred by the transcendence of π\pi from any further reduction.

Still open: how many Archimedean circles there are

The twins are not alone. Leon Bankoff found a third circle of the same size in 1974, and since then catalogues have collected hundreds of circles in the arbelos, each defined by a different construction, each with radius AC⋅CB/(2⋅AB)AC \cdot CB/(2 \cdot AB). There is no known reason why so many constructions produce the same length, and no theory predicting which constructions will. Whether the Archimedean circles form a family with a unifying description — whether they are all, in some sense, the same circle seen through different symmetries of the figure — is an open question of a kind that the Book of Lemmas could not have anticipated. Curvatures that stay whole found that the Apollonian packings, which also fill a region with mutually tangent circles, are governed by a single equation relating four curvatures; nothing of comparable power is known for the circles of the arbelos.