What is inversive geometry?

Andrew Krieger

July 18, 2013

Throughout, Greek letters (↵, , . . . ) denote geometric objects like circles or lines; small Roman letters
(a, b, . . . ) denote distances, and large Roman letters (A, B, . . . ) denote points.. The symbol AB denotes the
!
line segment between points A and B and `(AB) denotes its length. The symbol AB denotes the ray with
endpoint at A and passing through B.

1 Steiner’s Porism
Suppose we have two nonintersecting circles ↵ and in the plane. When is it possible to draw n circles
1 , . . . , n such that each i is tangent to both ↵ and , as well as to i+1 and i 1 (indices taken mod n)?
How many di↵erent such drawings are there for fixed circles ↵ and ?

Figure 1: A solution for n = 12 [Credit: WillowW / Wikimedia Commons / CC-BY-SA]

Suppose ↵, are concentric. Let a and b denote the radii of ↵ and respectively, and assume a > b. Clearly,
such an arrangement of circles i exists only when the centers C1 , . . . , Cn of the n circles 1 , . . . n are the
vertices of a regular n-gon, whose center is the common center O of ↵ and . Consider then a point of
tangency T between two circles i and i+1 , and the triangle 4OCi T . The edge OT is a tangent to i , and
Ci T is a radius, so OT ? Ci T . Furthermore, `(Ci T ) = a 2 b and `(OCi ) = a+b
2 , by straightforward geometry.
So, we have
⇣⇡⌘ `(Ci T ) a b (a/b) 1 ⇢ 1
sin = = = =
n `(OCi ) a+b (a/b) + 1 ⇢+1

where ⇢ := ab .
Rearranging the equation sin(⇡/n) = (⇢ 1)/(⇢ + 1) to solve for ⇢ gives us:
⇣⇡⌘ ⇣⇡⌘ ⇣ ⇣⇡⌘ ⌘ ⇣⇡⌘ ✓ ◆2
1 + sin(⇡/n) 1 + sin(⇡/n)
⇢ sin + sin =⇢ 1 , ⇢ sin 1 = sin 1,⇢= =
n n n n 1 sin(⇡/n) cos(⇡/n)

So, it is possible to draw n circles 1 , . . . , n according to the rules of Steiner’s Porism precisely when
2
the ratio ⇢ = ab satisfies ⇢ = (sec(⇡/n) + tan(⇡/n)) (after simplifying the trigonometric functions from the

1
 

/  2

O


1 T

C1

Figure 2: Geometry of Steiner’s Porism for concentric circles A and B

previous equation). Furthermore, we can obtain uncountably many di↵erent drawings just by rotating about
O through any angle.
Direct analysis of this sort is significantly more difficult in the general (non-concentric) case. So, let’s look
for another method.

2 Inversive Geometry
We will work in the Euclidean plane R2 , with an additional point at infinity (denoted 1) adjoined. We will
consider a line in the plane to be a circle of infinite radius, which included the point at infinity. Any two
lines intersect at 1. If the lines are parallel in the Euclidean sense, so that they meet only at 1, we consider
them to be tangent.
The first example of an inversion map: f (z) : C ! C defined by f (z) = z 1 for z 6= 0, f (0) = 1, f (1) = 0.
This turns the unit circle {z 2 C : |z| = 1} inside-out, exchanging external points and internal points. We
will generalize this to any circle in the Eucldean plane, and express inversion in purely geometric terms.
Definition For any circle in the plane with center O and radius r, the inversion map through takes
!
any point P 6= O to the (unique) point P 0 on the ray OP such that `(OP ) · `(OP 0 ) = r2 . I will denote the
inversion map through by T .
Some properties:
1. T exchanges points inside with points outside.
2. The closer a point on the inside is to O, the farther away its image is.
3. The fixed points of T are precisely the points on the circle .
4. T is an involution, i.e. T T = idR2 [{1} .
5. If ↵ is a circle concentric with , then T (↵) is also a circle concentric with ↵ and .
6. If is a line running through the point O, then T ( ) = .
7. If µ is any line not passing through O, then T (µ) is a circle containing O.
8. If is any circle which included the point O, then T ( ) is a line not passing through O.
The first 6 properties are straightforward to prove. The last two are only slightly more involved, and they
follow from elementary geometry:
Suppose µ is a line not running through O as in Figure 3. We want to show that the image of µ under T is
a circle containing O. If we draw the perpendicular OA to µ, we can find the image A0 = T (A). Then, we
consider the circle with diameter OA0 and show that any point P on µ maps to this circle. Let P 0 be the point
!
where OP intersects the circle with diameter OA0 . By similar triangles, `(OP )/`(OA) = `(OA0 )/`(OP 0 ), so
`(OP ) · `(OP 0 ) = `(OA) · `(OA0 ) = r2 , so P 0 = T (P ). Thus, T (µ) is the circle with diameter OA0 .

2
 
'

 
'

Figure 3: Proof that a line not through O inverts to a circle

Since T = T 1
, it follows that the circle with diameter OA0 is mapped to a line not containing O, and
!
perpendicular to the ray OA0 . This holds as we move A0 throughout the plane, i.e. any circle through O
inverts to a line not passing through O.

3 Concyclic points and separation

Definition If four points A, B, C, D lie on the same circle (or line, since we count lines as circles of infinite
radius), we say following [1] that the pair AC separates BD – written AC//BD – if either arc from A to C
intersects either B or D. In other words, AC//BD if and only if the points lie in the cyclic order A, B, C, D
along the circle (up to reversal of the order). We will extend the definition of separation to four points in
general position after a theorem.
Theorem (Proof from [1]) When A, B, C, D are not concyclic (or colinear), there are two non-intersecting
circles, one through A and C, and another though B and D.
Remark This was a Putnam exam problem in 1965.

Proof. Consider the perpendicular bisectors ↵ and of AC and BD respectively. If these bisectors coincide,
then A, B, C, D are concyclic. So, these bisectors either intersect once in R2 (and again at 1, which
intersection we will ignore) or are parallel.
If the bisectors interesect at the (finite) point O, then there are two circles and with center O, such that
A and C lie on , and B and D lie on . If A, B, C, D are not concycic, these are two distinct concentric
circles, which competes this case of the proof.
! !
Otherwise, ↵ and are parallel. This forces AC and BD to be parallel as well, so these four lines form a
rectangle. Consider the points P and Q, the midpoints of the sides of this rectangle which coincide with ↵
and respectively. There is a circle through A, P , and C and another through B, Q, and D, and because
! !
of the separation between lines AC and BD, these two circles cannot intersect. This completes the proof.



  

 


  


Figure 4: Diagram for proof of theorem [Credit: Reproduction of figure in [1]]

3
With this fact in mind, we can redefine the symbol AC//BD to mean that every circle through A and C
intersects or coincides with every circle through B and D. Then by the theorem, AC//BD only if A, B, C, D
are concyclic, in the concyclic case, this new definition agrees with the old definition of separation.
Remark In [1], a third characterization of AC//BD is presented. The authors prove that for any four
distinct points A, B, C, D,

`(AB) · `(CD) + `(BC) · `(AD) `(AC) · `(BD)

and then prove that equality holds if and only if AC//BD. I will use this fact without proof, due to time
constraints.
Now, we can give a new method of specifyng circles. For any three points A, B, and C, the circle though A,
B, and C is the union of {A, B, C} and the set of all points X such that BC//AX, CA//BX, or AB//CX.
Clearly, any point X satisfying any of these three separation conditions lies on the same circle as A, B, and
C (by the Theorem above), and conversely any point X on the circle is either A, B, or C, or satisfies one of
the separation conditions.
An important result is that inversion through a circle preserves separation. The following short Lemmas are
due to [1]:
Lemma If is any circle with center O and radius r, and A, B are any two points with T (A) = A0 and
T (B) = B 0 , then

r2 `(AB)
`(A0 B 0 ) =
`(OA) · `(OB)

Proof. The triangles 4OAB and 4OB 0 A0 are similar, so we have:

`(A0 B 0 ) `(OA0 ) `(OA) · `(OA0 ) r2

= = =
`(AB) `(OB) `(OA) · `(OB) `(OA) · `(OB)

Definition The cross ratio between two pairs of points AC and BD is {AB, CD} = (`(AC)·`(BD))/(`(AD)·
`(BC)). In particular, AC//BD if and only if {AB, CD} = 1.
Lemma Inversion through circle with center O and radius r preserves cross ratios.

Proof.
r 2 `(AC) r 2 `(BD)
0 0 0 `(A0 C 0 ) · `(B 0 D0 )
0 `(OA)·`(OC) `(OB)·`(OD) `(AC) · `(BD)
{A B , C D } = = = = {AB, CD}
`(A0 D0 ) · `(B 0 C 0 ) r 2 `(AD) r 2 `(BC) `(AD) · `(BC)
`(OA)·`(OD) `(OB)·`(OC)

Combining this with the previous results, we see that inversion also preserves separation. This leads us to
the following theorem:
Theorem Any inversion map takes circles to circles (recall that lines are considered circles of infinite radius).

4
Proof. Suppose that we have a circle through points A, B, and C. Recall that this circle is ↵ = {X :
BC//AX, AC//BX, or AB//CX} [ {A, B, C}. If we invert through some circle , we get A0 = T (A),
B 0 = T (B), and C 0 = T (C). Furthermore, for any other point X on the circle, if (say) BC//AX, then
B 0 C 0 //A0 X 0 , where X 0 = T (X).
Thus,

T (↵) = {X 0 : X 0 = T (X) and BC//AX, AC//BX, or AB//CX} [ {A0 , B 0 , C 0 }

= {X 0 : B 0 C 0 //A0 X 0 , A0 C 0 //B 0 X 0 , or A0 B 0 //C 0 X 0 } [ {A0 , B 0 , C 0 }

is also a circle.

Equally important, inversion preserves the number of intersections of any two circles, so tangent circles invert
to other tangent circles, etc.
Finally, we shall consider some results about orthogonal circles.
Definition Two circles are called orthogonal if they intersect twice at right angles.
Lemma Inversion through a circle with center O preserves the angle of intersection (namely, the angle
between the two tangents at the point of intersection) of any two intersectings circles.

Proof. First, Let ↵, be two intersecting circles. Let be tangent to ↵ and µ tangent to . Under the
inversion map, maps to a circle 0 which touches the center O of the inversion, and whose tangent at O
is parallel to ↵. Likewise, the µ0 has a tangent at O parallel to . By well-known facts about parallel lines,
the angle between the tangents to 0 and µ0 is congruent to the angle between and µ.
Now, if ↵ and were tangent to and µ respectively at the point of intersection of the two lines (call it P )
then ↵0 and 0 are tangent to 0 and µ0 at P 0 , which forces their angle of intersection to equal that of 0 and
µ0 , which is equal to the angle between and µ, which is the angle of intersection of ↵ and .

Corollary Inversion maps orthogonal circles to orthogonal circles.

Theorem Given any two non-intersecting circles, it is possible to invert them into two concentric circles.

Proof. Let ↵, be two non-intersecting circles. By geometry (see [1], Section 5.7), there is a line , called
the radical axis of ↵ and , which is perpendicular to the line of the centers of ↵ and . For many points
P on this line (precisely, those points for which the quantity `(O↵ P )2 r↵2 = `(O P )2 r2 > 0, where O↵
is the center of ↵ and r↵ its radius), the length of the tangents from P to either A or B are equal. A circle
drawn with center P and radius equal to this common length will be orthogonal to both ↵ and .
We can draw two such circles and , which will intersect at two points, say O inside of ↵, and P inside of
. We then invert through any circle with center O.
Let P 0 be the image of P under this inversion, ↵0 the image of ↵, and so on. Then, 0 and 0 are two lines
which meet at O, since and were circles touching O. Since ↵ was orthogonal to both and , we have ↵0
orthogonal to 0 and 0 . Since 0 is a line, it must be a diameter of ↵0 ; likewise 0 is another diameter. these
diameters meet at P 0 , so P 0 must be the center of ↵0 . Likewise, P 0 is the center of 0 , and we are done.

4 Finishing the proof of Steiner’s Porism

. . . is now easy. I leave it as an exercise to the reader in applying the prior theorem.

5
 

'

 

 ' 

' 

Not shown: ', '

(centered around ')

Figure 5: Proof that you can invert two non-intersecting circles into two concentric circles [Credit: This author,
based on a proof and figure from [1]]

5 Bibliography
• Coxeter, H.S.M. and Greitzer, S.L. Geometry Revisited. New York: Random House, 1967. Print.
• https://en.wikipedia.org/wiki/Steiner_chain

6 Image Credits
• Figure 1: “WillowW” / Wikimedia Commons / CC-BY-SA
https://en.wikipedia.org/wiki/File:Steiner_chain_12mer.svg
• Figure 2: Created by the author of these notes
• Figure 3: Created by the author of these notes; similar to part of Figure 5.3B (p. 109) in [1]
• Figure 4: Reproduction by the author of Figure 5.1 from [1]
• Figure 5: Created by the author of these notes; based upon Figure 5.7A of [1]