H. S. M.

C O X E T E R

INVERSIVE GEOMETRY

1. I N T R O D U C T I O N

Euclidean geometry deals mainly with points and straight lines. Its basic
transformation is the reflection, which leaves fixed all the points on one line
and interchanges certain pairs of points on opposite sides of this "mirror".
All other isometries (or "congruent transformations" or "motions") are
expressible in terms of reflections. (For simplicity, we are describing only plane
geometry. In space we would reflect in a plane instead of a line.)
Analogously, inversive geometry deals with points and circles. Its basic
transformation (invented by L. J. Magnus in 1831) is the inversion, which
leaves fixed all the points on one circle and interchanges the inside and out-
side of this "circle of inversion". All other circle-preserving transformations
(including similarities as a special case) are expressible in terms of inversions.
This kind of geometry is worthy of attention not only for the sake of its
intrinsic beauty but also because it is the geometry of complex numbers
[10, 6, pp. 145-147] and because the point pairs and circles of the real
inversive plane provide an isomorphic model for the lines and planes of
hyperbolic (non-Euclidean) space [4, p. 221].

2. A N O U T L I N E OF T H E A X I O M A T I C APPROACH

From the ordinary Euclidean plane we can derive the inversive plane by
regarding a straight line as a kind of circle, namely a circle that passes through
a special point called the point at infinity. This extra point, which is added to
the Euclidean plane to make the inversive plane, enables us to declare,
without any exception, that
A n y three distinct points lie on just one circle. (2.1)
If the three given points happen to be collinear, the "circle" is a straight line.
The idea of adding an ideal line to the Euclidean plane to make a projec-
tive plane has been ascribed to Desargues (1639). For more than two cen-
turies this procedure was believed to be unique, as when Cayley remarked in
1859: "Metrical geometry is a part of descriptive geometry and descriptive
geometry is all geometry". It was M. B6cher, in 1914, who first saw clearly
that the idea of adding a single ideal point to the Euclidean plane (to make

Educational Studies in Mathematics 3 (1971) 310-321. All Rights Reserved

Copyright 9 1971 by D. Reidel Publishing Company, Dordrecht-Holland
 INVERSIVE GEOMETRY 311

an inversive plane) is entirely analogous to what Desargues did, and equally

useful.
Mario Pieri, who gave the first satisfactory set of axioms for projective
geometry (Cayley's "descriptive geometry") in 1899, did the same for inver-
sive geometry in 1912. Subsequent authors have improved the details, so that
now most of the familiar properties of points and circles can be deduced
from just four axioms: (2.1) and the following three:
There exist four points not on a circle. (2.2)
I f a point P is on a circle ~ while Q is not on ~, there is just one
circle through Q whose only common point with a is P. (2.3)
I f each cyclically adjacent pair of four circles have a pair of com-
mon points, forming altogether eight distinct points, and if four of
these points, one from each pair, lie on a circle, then the remaining
four lie on a circle. (2.4)
Two circles are said to be intersecting, tangent, or non-intersecting according
as they have two common points, one common point, or no common point.
Thus the circles considered in (2.3) are tangent, with P as their point of con-
tact.
Notice that this treatment makes no mention of "distance": the only
relation used is that of incidence, which allows us to use such phrases as " a
point P is on a circle ~" or "two circles have a pair of common points".

Fig. 1.
312 u.s.M. COXETER

(Just as some geometers investigate "non-Desarguesian" projective planes

that do not satisfy the theorem of Desargues, so they also investigate "non-
Miquelian" inversive planes that do not satisfy Axiom (2.4), which is the
theorem of Miquel.)
We usually say that two intersecting circles are orthogonal if they have
perpendicular tangents at either of their points of interaction. It is an amu-
sing problem to ask how the same notion can be described without using the
word "perpendicular" or any of its synonyms. The solution is surprisingly
simple (see Figure 1).
Two intersecting circles are orthogonal if one of them belongs to a triad of
mutually tangent circles, touching one another at three distinct points A, B, C,
while the other is the circle ABC.
For any point P not on a circle co, the inverse point P ' can be defined as the
second intersection of any two circles through P orthogonal to co. (In Figure
2, one of the two has been drawn as a straight line, and the other with its
center on this line.)

...
~. T

pt

Fig. 2.

Inversion provides a very simple criterion for orthogonality: two circles

are orthogonal if one of them passes through two points that are inverses in
the other.
Any two distinct circles, ~ and fl, determine a pencil of circles orthogonal
to them, consisting of all the circles PP1P2 where P is a variable point while
P1 and P2 are its inverses in ~ and ft. If y and 6 are two such circles, the pencil
of circles orthogonal to them includes ct and fl, it is the pencil "spanned" by ct
and fl, and is conveniently denoted by aft. One finds that the circles through
two distinct points, L and L', form an intersecting pencil, orthogonal to the
non-intersectingpencil for which L and L' are the limiting points (inverses in
 II'4VERSIVE GEOMETRY 313

each member of the non-intersecting pencil). If c~ and fl are tangent, eft is a

tangent pencil, and the orthogonal circles through the point of contact form
another tangent pencil. Three or more circles that belong to one pencil are
said to be coaxal.
Any three non-coaxal circles, c~, t , ?, span a bundle of circles, say ~fl?,
consisting of the smallest set of circles which includes c~, t , 7 and, with every
two of its members, the whole of the pencil spanned by them. If e, t , 7 have a
common point, eft? consists of all the circles through this point; it is called a
parabolic bundle. If ~, fl, ? are all orthogonal to one circle, aft? consists of
all the circles orthogonal to this one circle; it is called a hyperboBe bundle
(because its members represent the lines of the hyperbolic plane). Finally, if
c~, fl, ? have neither a common point nor a common orthogonal circle (e.g., if
they are mutually orthogonal), ~fl? is called an elliptic bundle (because its
members represent the lines of the elliptic plane).

3. T H E E U C L I D E A N APPROACH

Unlike projective geometry, which is easily developed from its axioms [see,
for instance, 2], the details of pure inversive geometry require so much effort
that the best practical procedure is to specialize one point, call it the point at
infinity, and then use the powerful tool of Euclidean geometry. In other
words, to solve a particular problem concerning circles, we first simplify the
figure by inverting in a circle whose center lies on one or more of the circles
(so as to replace these circles by straight lines) and then deal with the corresp-
onding Euclidean problem. Afterwards we can perform the same inversion
again so as to restore the circles.
Applying one of Euclid's theorems to Figure 2, we see that, if co has center O
and radius OT, O P x OP' = O T z. This remark brings us back to the classical
definition:
Inversion in a circle co, with center O and radius k, is the transformation
that interchanges pairs of points, P and P', such that
OP x OP" = k 2 ,

P and P' being on the same side of O. Thus each point on co is self-inverse,
every point outside co has an inverse inside co, each line through O is self-
inverse, and we can regard the point at infinity as being the inverse of O.
Inversion is a circle-preserving transformation; more precisely, it trans-
forms the set of all straight lines and circles into itself. Fejes T6th [8, p. 85]
proves this as follows.
Consider first the inverse of a circle ? not passing through O. Let the line
of centers of co and ? cut ? in A and B, and let P be any other point on ?.
314 H.S.M. COXETER

Fig. 3.
Since
O P x O P ' = O A x OA" = O B • O B ' ,

we have pairs of oppositely similar triangles

O A P ,~ O P ' A ' , O B P ,,, O P ' B ' ,

as in Figure 3. Thus
, ~ B ' P ' A " = <~ O P ' A ' - , ~ O P ' B '
= <~ O A P - ~ O B B = , ~ A P B = 90 ~,

showing that the inverse of ? is the circle with diameter A ' B ' . This p r o o f is
easily modified to cover the cases when ? is a circle passing through O, or a
line not passing through O. The inverse of such a line ? is a circle ?' through
O whose tangent at O is parallel to ~.
I f follows that the two supplementary angles formed by two hnes through

\
\ P

Fig. 4.
 INVERSIVE GEOMETRY 315

a point P are equal to the angles formed by their inverse circles, which inter-
sect at O and again at P', as in Figure 4. In other words, inversion is an
angle-preserving (or "conformal") transformation. In particular, orthogona-
lity is preserved.
Let us see what our four axioms look like in Euclidean terms. Having
already considered (2.1), we turn to (2.2). Taking one of the four points to be
at infinity, we obtain the obvious statement that there exist three points not
on a line.
When Q is at infinity, (2.3) says that there is just one tangent line at any
point P on a circle ~. More interestingly, when P (instead) is at infinity, (2.3)
says that, i f Q is any point not on a line ~, there is just one line through Q
parallel to ~.

0 C
N M

Fig. 5.

In (2.4), let OA, BN, LD, C M be the four pairs of common points of the
four circles A O B N , N B L D , D L C M , M C O A , with the first of each pair on a
circle, as in Figure 5. Can we assert the existence of a circle A N D M ? Sending
O to infinity, we obtain a triangle A B C with points L, M, N on its three sides,
while circles N L B and C L M (having L in common) meet again in D. Now
the question is whether D lies also on the circle A M N . The answer is Yes, by
the Pivot Theorem [9, p. 17]. I f D is inside the triangle, as in our figure, the
p r o o f is simply
<f. D M C = ~ D L B = <~ D N A .

In other cases it is just as easy.

4. P E N C I L S A N D B U N D L E S

Consider once more the intersecting pencil of circles through any two points
L and L'. When L' is at infinity, this is simply the pencil of lines throgh L.
316 H.S.M. COXETER

The orthogonal non-intersecting pencil, orthogonal to these lines, consists

of all the circles with center L. In other words, any two intersecting circles
can be inverted into intersecting lines, and any two non-intersecting circles
can be inverted into concentric circles.
Similarly, two orthogonal tangent pencils can be inverted into two ortho-
gonal "pencils" of parallel lines. (Parallel lines are tangent circles whose
point of contact is the point at infinity.)
The sides of any triangle are circles having the point at infinity in common.
The parabolic bundle spanned by them is simply the set of all lines in the
Euclidean plane.
If e and fl are two perpendicular diameters of a circle ~, so that a, t , 7 are
mutually orthogonal, the elliptic bundle efly consists of all the diameters of
and all the circles that pass through pairs of diametrically opposite points
of 7- In other words, c~flyconsists of all the circles that arise from the great
circles on a sphere by stereographic projection.
Another Euclidean approach to bundles uses Steiner's concept of power.
With respect to a circle with center O and radius k, the p o w e r of a point P is
Op 2 -- k 2 "

The locus of points of equal power for two circles a and fl is a straight line
(the only line that belongs to the pencil ,fi), called the radical axis of a and ft.
It follows that the three radical axes of pairs of , , fl, y (if not parallel) all
pass through the radical center: a point of equal power for the three circles.
The set of all circles for which a given point P has a given powerp is a bundle:
hyperbolic ifp > 0, parabolic ifp = 0, elliptic ifp < 0.
I f ~ and fl are intersecting circles, their radical axis obviously joins their
two points of intersection. I f , and fl are non-intersecting, their radical axis
can be constructed as in Figure 6, where y is any genuine circle intersecting
both of them. In such a case (and again when three circles are all non-inter-
secting but not coaxal), the radical center of a, fl, y is outside all of them, and
the bundle . f i t is necessarily hyperbolic.

Fig. 6.
 INVERSIVE GEOMETRY 317

5. M I D - C I R C L E S AND ORTHOCYCLIC POINT PAIRS

We recall that two intersecting, tangent, or non-intersecting circles can be

inverted into two intersecting lines, two parallel lines, or two concentric
circles, respectively. When a circle is inverted into a line, inversion in the
circle is transformed into reflection in the line.
Given any two distinct circles ct and t , consider the locus of a point P such
that two circles, tangent to both ce and t , are tangent to each other at P. In-
version shows that this locus consists either of two orthogonal circles or o f a
single circle. More precisely, if ~ and i intersect, we have two circles bisec-
ting their supplementary angles of intersection; if ~ and t are tangent, we
have one circle tangent to both at the same point; and if ~ and fl are non-
intersecting we again have just one circle, belonging to the pencil ceil. Clearly,
inversion in such a circle interchanges ~ and i . The classical name "circle of
antisimilitude" [1, p. 28] is conveniently abbreviated to mid-circle.

Fig. 7.

If ~ and fl are congruent, their mid-circle (or one of their mid-circles) evi-
dently coincides with their radical axis. In other cases the following construc-
tions are available.
If ~ and fl intersect, the centers of their mid-circles lie on the lines that
bisect their angles of intersection.
If cr and fl are non-intersecting, let their diameters on their line of centers
be AA' and BB', so named that A and B' separate B and A', as in Figure 7.
Then a diameter of their mid-circle is formed by the limiting points of the
two non-intersecting circles whose diameters are AB and ,4'B'. For, this
circle, being orthogonal to the circles on AB and A'B', inversts A into B, and
A' into B'.
By making A' and B' coincide (at C, say) we obtain the limiting case when
and fl are tangent circles. Then a diameter o f their mid-circle is formed by
C and its inverse in the circle on AB.
318 H . S . M . COXETER

Notice that the definition of a mid-circle belongs to inversive geometry,

although we have used Euclidean methods to develop its properties. The
advantage of allowing ourselves such freedom is well illustrated by the
following discussion of "orthocyclic point pairs", in which we first use a
strictly inversive method and then, for comparison, the Euclidean method.
Two point pairs, LL' and MM', are said to be orthocyclic [1, p. 100] if
there is a circle through L and L ' that inverts M into M'. In this case, we can
prove that the relation is symmetric: there is a circle through M and M ' that
inverts L into L'.
If all four points lie on one circle ct, as in Figure 8, the circle 2 through L and
L ' that inverts M i n t o M ' inverts ~ into itself and is therefore orthogonal to ~.
The circle/t, through M and M ' orthogonal to ~, is also orthogonal to 2
(since 2 inverts M into M'). Therefore the inversion in/, interchanges the two
intersections of~ and 2, which are L and L'.

M/

Fig. 8. Fig. 9.

Figure 9 illustrates the remaining case, in which L, L', M, M ' do not lie on
one circle. The circle 2 through L and L' that inverts M into M ' is still ortho-
gonal to every circle through M and M'. In particular, 2 is orthogonal to the
two intersecting circles LMM', L'MM', and to each of their two mid-circles,
say # and #'. Both/z and #' invert the two intersecting circles 2 and L M M '
into the two intersecting circles 2 and L'MM'. Since L is one of the inter-
sections of the former pair, and L' of the latter, if/t does not invert L into
L', It does.
Euclidean geometry provides a far easier proof. Take L' to be the point
at infinity. Then there is a line through L that reflects M into M'. Thus L M =
LM'. The circle through M and M ' with center L inverts L into L', as
desired.
 INVERSIVE GEOMETRY 319

6. INVERSIVE DISTANCE

Any two intersecting circles have two supplementary angles of intersection,

which are preserved when the circles are inverted into lines, and are bisected
by the two mid-circles. Somewhat analogously,-any two non-intersecting
circles have an inversive distance, which is preserved when the circles are
inverted into concentric circles, and is bisected by the mid-circle. For con-
centric circles, of radii a and b, the inversive distance is defined to be
a
log~

so that, if three circles of a non-intersecting peneil are inverted into concen-

tric eircles whose radii satisfy a > b > e, we have the proper additive relation
a b a
log~ + log c = log-.e

In particular, concentric circles of radii 1 and e have inversive distance 1.

In other words, the inversive distance between any two non-intersecting
circles is defined to be the logarithm of the ratio of the radii (larger to smaller)
o f two concentric circles into which the given circles can be inverted.
It follows [7, p. 129] that if two non-intersecting circles ~ and fl (neither
surrounding the other) have centers A and B, radii a and b, and inversive
distance v, the (Euclidean) distance between their centers is given by
AB 2 = a 2 + b 2 + 2ab coshv. (6.1)

Moreover [7, pp. 130, 176 (Ex. 4)] if C is the foot of the perpendicular from
A to a line y outside ~, and if2 is the inversive distance between ct and y,
A C = a cosh2. (6.2)
As a limiting case we naturally regard tangent circles as having inversive
distance O.
Three non-intersecting circles a, t , Y are said to be nested, with y separat-
ing ~ and t , if every circle that intersects a and fl intersects ~ too. In particu-
lar, for any three members of a non-intersecting pencil, one separates the
other two in this sense. In the case of three concentric circles, this is the one
whose radius is neither the greatest nor the least. It is interesting to observe
that the mutual inversive distances of three nested circles satisfy a "non-
triangle inequality":
Among the three inversive distances between pairs o f three nested circles,
one is greater than or equal to the sum of the other two. Equality holds only
when the three circles are eoaxal.
320 H.S.M. COXETER

T o p r o v e this, let ? separate ~ and fl, and let the inversive distances be

=(~,,~), ~,=(#,?), v = ( ~ , , f l).

Since a, fl, ? are non-intersecting, there is at least one circle orthogonal to all
of them. T a k i n g one of the intersections o f this circle with ~ to be the point
at infinity, we have a straight line 7 perpendicular to the line A B joining the
centers o f a and fl, as in Figure 10.

c B b

Fig. 10.

Equations (6.1) and (6.2) yield

a 2 -Jr b 2 -I- 2 a b cosh v = A B 2 = ( A C + C B ) 2

= (a cosh 2 + b c o s h / 0 2
= a 2 (1 + sinh2 2) + 2 a b c o s h 2 c o s h # + b 2 (1 + sinh2 # ) ,
whence
2ab{coshv - cosh (2 + p)} = (a s i n h 2 - b s i n h # ) 2 I> 0.
Thus
v/> 2 + # , (6.3)

Fig. 11.
 INVERSIVE GEOMETRY 321

with equality only when a sinh2 = b sinh#. But

(a sinh~l) 2 = (a c o s h 2 ) 2 - a 2 = A C 2 - a 2

is the power o f C with respect to ~, and similarly (b sinhp) 2 is the power o f C

with respect to ft. Hence equality occurs only when ~ is the radical axis o f
c~ and fl, that is, when ~, fl, v are coaxal. (The extreme case when 2 = r is
illustrated in Figure 11.)
The inequality (6.3) is intimately related to Einstein's Twin P a r a d o x (or
Clock Paradox), which depends on the non-triangle inequality for time-like
intervals I-5, p. 11].

University o f Toronto, Canada

REFERENCES

[1] J. L. Coolidge, A treatise on the circle amd the sphere, Oxford, 1916.
[2] H.S.M. Coxeter, Projective geometry, Waltham, Mass., 1964.
[3] H. S. M. Coxeter, Non-Euclideangeometry, Toronto, 1965.
[4] H. S. M. Coxeter, 'The inversive plane and hyperbolic space', Abh. Math. Sere. Univ.
Hamburg 29 (1966), 217-242.
[5] H. S. M. Coxeter, 'The problem of Apollonius', Am. Math. Monthly75 (1968), 5-15.
[6] H. S. M. Coxeter, Introduction to geometry, New York, 1969.
[7] H. S. M. Coxeter and S. L. Greitzer, Geometry revisited, New York, 1967.
[8] IAszl6 Fejes T6th, Regular figures, New York, 1964.
[9] H. G. Forder, Geometry, London, 1950.
[10] Hans Schwerdtfeger, Geometry of complex numbers, Toronto, 1962.