Robert Hunter

Math 308, Euclidean Geometry

Term Project, Fall 2003



The Shoemaker's Knife Problem - An Application of Inversion



Pappus of Alexandria (who wrote in the time of Diocletian, 284-305 AD) records in his Collection[i], the following proposition:


Let two semicircles be drawn, a larger one B and a smaller one A fitted inside the first, with centres b and a, respectively, lying on a common baseline (fig. 1).

fig. 1


This figure Pappus calls an άρβηλος, or “shoemaker’s knife”. Now let the area between the two semicircles be filled with circles C0, C1, C2,... tangent to A and B and to each previous circle in the sequence (fig. 2).

fig. 2

Then the following results hold for this figure:


(1)    The centres of circles C0, C1, C2,... all lie on an ellipse (cyan) passing through q.

(2)    The points of contact of same circles all lie on a circle (blue) passing through q.

(3)    Letting the radius of Cn be rn, the height of its centre above the baseline is 2nrn.


That the centres of C0, C1, C2,... all lie on an ellipse passing through q is apparent from the diagram below (fig. 3). The foci of the ellipse are points a and b, and the sum of the distances from the foci to any point on the ellipse is, if we allow a and b to be the radii of circles A and B as well as their centres, a + b. It can also be shown that the lengths of the major and minor axes of the ellipse are a + b and 2√―a―b , respectively.


In order to prove the remaining two statements of the proposition, we shall make use of an insightful trick. Before we can do this, though, we need to make a diversion of some length.






In the complex plane, the algebraic inverse z-1 of a point z = re is the point that results from taking the reciprocal length and then reflecting across the real axis, viz. (1/r)e-iθ. If we just take the reciprocal length but not the reflection, we get the geometric inverse of z, or w = (1/r)e, = 1/z―. w is the inversion of z in the unit circle, |z| = 1 (fig. 4).



We can generalize this to inversion in a circle K of radius R centred at point q by:


(1)    translating by –q (centres K at the origin)

(2)    scaling by 1/R (shrinks K to the unit circle)

(3)    inverting in the unit circle

(4)    scaling by R (K returns to its original size)

(5)    translating by q (K returns to its original position).


The resulting formula is w = R2/( z― - q― ) + q. To check that this formula performs as required, note that if z – q = re, w - q = (R2/re-iθ) = (R2/r)e, which is in the same direction as z – q from the centre of K and has the reciprocal length with respect to the scaling factor of R:


(r/R)* ((R2/r)/R) = 1


Also note that |z – q| * |w – q| = R2.


We shall now examine the effect of inversion on circles and lines. Let K, the circle in which we are inverting, have radius R and be centred at q as before, and, for simplicity, assume q = 0.


(1)    Since any point on a line through the origin, with the exception of 0 itself, is mapped by the inversion to another point on the line, and any point on the line, other than 0, is the image of some other point on the line, the image of the line under inversion is, if we ignore this one exception, the same line. In order to patch up the hole at 0, we make the convention that the point at infinity () is mapped to 0, and vice versa. Thus inversion in a circle is 1-1 and onto the extended complex plane. (We denote this function by IK.)

(2)    Lines not through the origin are mapped to circles passing through the origin (understanding that the line contains the point at infinity, which is mapped to 0), and vice versa[ii] (fig. 5).

In this figure, point a, lying on the intersection of L and the perpendicular through 0, is mapped to a point a’ on the perpendicular. Point b, which is mapped to the point b’, is any other finite point on L. C is defined to be the circle with diameter 0a’. We determine that IK(L) = C as follows.















From the formula |z – q| * |w – q| = R2, above, we obtain |a|*|a’| = R2 and |b|*|b’| = R2, which implies |a|/|b| = |b’|/|a’|. Since angles ‹a0b and ‹b’0a’ are the same angle, this implies that the two triangles Δa0b and Δb’0a’ are similar. ‹0b’a’, corresponding to ‹0ab, must therefore be a right angle, i.e. the line segment 0a’ subtends a right angle to b’. b’ thus lies on C.


Since b was taken to be an arbitrary point on L (other than a and , whose images, a’ and 0, clearly lie on the circle) the image of every point on L must lie on this circle.


Analogously, the image of every point on C must lie on L. Thus, the image of L under the inversion must be all of C.


(3)    Circles not passing through the origin are mapped to circles not through the origin (fig. 6).

In this figure, points a and b, lying on the ray from 0 through the centre of C, are mapped to the points a’ and b’, respectively. Point c, which is mapped to c’, is any other point on C. C’ is defined to be that circle with diameter b’a’. We determine that IK(C) = C’ as follows.

As in case (2), above, triangles Δa0c and Δc’0a’ are similar, as are triangles Δb0c and Δc’0b’.[iii] Thus ‹0cb = ‹0b’c’ and ‹0ca = ‹0a’c’, since these are corresponding angles in similar triangles. ‹b’c’a’ = ‹0b’c’ - ‹0a’c’, = ‹0cb - ‹0ca, = ‹acb. Since C is a circle and ab its diameter, ‹acb is a right angle. Thus ‹b’c’a’ is also a right angle, i.e. the line segment b’a’ subtends a right angle to c’. c’ thus lies on C’.

Since c was taken to be an arbitrary point on C (other than a and b, whose images, a’ and b’, clearly lie on C’) the image of every point on C must lie on C’.


Analogously, the image of every point on C’ must lie on C. Thus, the image of C under the inversion must be all of C’.


(4)    Circles orthogonal to the inverting circle K are mapped onto themselves (fig. 7). That this is the case follows from the fact that the images under IK of figures tangent at a point P are again tangent, at the point IK(P).[iv] Circle C is tangent to the lines 0b and 0c at points b and c, respectively. Thus the image of circle C, call it C’, is also tangent to the lines 0b and 0c (the lines 0b and 0c being mapped onto themselves) at points b and c (these being unmoved by the inversion). The only circle that meets these conditions is C itself, so C’ must be C.

Note that, although IK(C) = C, i.e. the points of C are all mapped onto C, the centre of C is not mapped to itself, since the centre of C lies outside of K and its image under the inversion on the inside.[v]


Back to the Shoemaker's Knife Problem

We are now equipped to prove the remaining two parts of the original proposition:

(4)    The points of contact of circles C0, C1, C2,... all lie on a circle passing through q.

(5)    Letting the radius of Cn be rn, the height of its centre above the baseline is 2nrn.

If we consider how to draw the figure (referring to fig. 2), it appears to be the result of taking a vertical column of circles, sandwiched between vertical lines on each side, and then somehow squeezing all this over to a point on the baseline. The bottom circle in the column would remain fixed, the two vertical lines would be bent into the circles A and B, and the remaining circles would be squeezed in to fit between A and B.

This suggests an inversion about a circle orthogonal to C0 and centred at the point at which circles A and B meet (fig. 8).

Here K is the circle centred at q which meets C0 at right angles. Thus IK(C0) is the same circle. The images of the vertical lines A’, B’, and M’ under inversion in K are all circles, A, B, and M, respectively, passing through q and symmetric about the baseline (the perpendicular through the centre of K). Circles C0’, C1’, C2’,... we know are inverted to another string of circles C0, C1, C2,.... Since each point of intersection is mapped to a single point in the image, these must still be in contact with each other and be tangent to the circles A and B. Thus inversion in K produces the desired figure.[vi]

The points of contact of the stack of circles C0’, C1’, C2’,... all lie on the vertical line M’, thus after inversion they must all lie on M, the image of M’ under IK. This proves the second part of the proposition.

In order to prove the third statement, simply choose any circle Cn and invert the figure about that circle Kn, centred at q, which is orthogonal to Cn (fig. 9).

The circle Cn is then unmoved, with C0’, C1’,..., Cn-1’, Cn+1’, Cn+2’... stacked above and below it. Circles A and B are transformed into vertical lines, as before, although now they sandwich this new column. Since the image circles all have the same radius, rn, as Cn, one can immediately see that the centre of Cn lies a distance 2nrn above the baseline. This proves the final statement of the propositon.



[1] Needham, Visual Complex Analysis, Clarendon Press, Oxford, 1997.

[2] Heath, A History of Greek Mathematics, Volume II, Dover, 1981.

[3] Churchill and Brown, Complex Variables and Applications, McGraw-Hill, 1984.

The historical background is to be found in Heath. The proof using inversion, as well as most of the theory on inversion, I took from Needham’s book. The drawings are all original and were produced by PostScript programming.



I would like to thank Professor Casselman for suggesting this topic, as well as for providing two of the three references.



    i.            The Synagoge, Book IV, Section 2.

ii.            The inverse of the function IK is IK, thus, for a line L and a circle C, if IK(L) = C then IK(C) = L.

iii.            It is not in general true, however, that Δabc and Δa’b’c’ are similar.

iv.            See Needham for the proof of a more general result: that geometric inversion is anti-conformal, i.e. the magnitude of the angle between tangent lines of intersecting curves is preserved but the sense is reversed.

    v.            This fact is even more general: For any circle C, the centre of IK(C) is not the image under inversion of the centre of C.

vi.            Note, however, that although the result is what we want, this transformation isn’t quite what was imagined, since, in the image of the inversion, circles A and B appear on the side of the circles opposite to the side the corresponding line was on. In other words, A and B have switched places.