N N N N n+1 N 2 I: N n+1 N N
N N N N n+1 N 2 I: N n+1 N N
GEOMETRY
1. Introduction
The object of this paper is to give a characterizations of Möbius transfor-
mations acting on Sn , under very weak conditions on such a map T : Sn →
Sn , which do not assume invertibility or even continuity of the map.
The standard n-sphere Sn , viewed in Rn+1 is the real algebraic set
n
X
Sn := {(x0 , x1 , · · · , xn ) ∈ Rn+1 : x2i = 1}.
i=0
The set of Möbius transformations are the set of invertible maps F : Sn →
Sn generated by inversions. Such maps send circles to circles and (n − 1)-
spheres to (n−1)-spheres. The study of geometric properties invariant under
such transformations is called inversive geometry.
One can also identify Sn with Rn∞ := Rn ∪ {∞} under sterographic pro-
jection. In the space Rn∞ , an inversion (or a reflection) in an (n − 1)-
sphere S(a, r) := {x ∈ Rn : |x − a| = r} is the function φ defined by
r
φ(x) = a + ( |x−a| )2 (x − a). φ is well defined on Rn∞ − {a, ∞}, and at these
two points, we define φ(a) = ∞ and φ(∞) = a. A reflection in a hyper-
plane is a usual reflection in Rn and fixes the point ∞ in Rn∞ . We define a
Möbius transformation on Rn∞ ∼ = Sn to be a finite composition of reflections
in (n − 1)-spheres or hyperplanes. The group of all Möbius transforma-
tions is called the Generalized Möbius Group GM (Rn∞ ), following Beardon
[2, Chapter 3]. Note that in dimension n = 2, identifying R2∞ with the
Riemann sphere Ĉ, Möbius transformations include all the linear fractional
transformations z 7→ az+bcz+d , which are orientation-preserving maps and also
the conjugate ones z 7→ az̄+b
cz̄+d , which are orientation reversing maps.
1.1. Main Result. We will study mappings satisfying the following very
weak version of the circle-preserving property.
Definition 1.1. A map T of the n-sphere to itself is called weakly circle-
preserving if for every circle C ⊂ Sn , T (C) lies in some circle.
Definition 1.2. A map T of the n-sphere to itself is called weakly sphere-
preserving if for every (n − 1)-sphere Sn−1 ⊂ Sn , T (Sn−1 ) lies in some
(n − 1)-sphere.
In these two definitions we do not assume that T is injective or even con-
tinuous. There are many such maps, including some that are not Möbius
transformations. For example, any map T : Sn → Sn whose image is fi-
nite and consists of (n + 1) points or less is automatically weakly sphere-
preserving and any map on n-sphere with image consisting of 3 points or
less is weakly circle-preserving. Nevertheless such maps are quite restricted
when further assumptions are imposed on them.
The key restrictions we consider are the following “general position” con-
ditions on the image of the map.
Definition 1.3. A subset B of Sn is said to lie in circular general position
if for any circle C, the complement of C contains at least two points of B.
Definition 1.4. A subset B of Sn is said to lie in spherical general position
if for any (n − 1)-sphere Sn−1 , the complement of Sn−1 contains at least two
points of B.
It is obvious from the definition that B must contain at least n + 3 points,
and there do exist many (n + 3) point sets in spherical general positon. If B
lies in spherical general position, then every set containing B lies in spherical
general position. However, if B lies in spherical general position, it is not
clear whether it always contains an n+3 subset of B lies in spherical general
position, and this result does always not hold in dimension n = 2.
Our main result is as follows.
Theorem 1.5. For n ≥ 3, T : Sn −→ Sn be a weakly circle-preserving
map. If T (Sn ) is in spherical general position and there is a 2-sphere S2
with T (S2 ) in circular general position, then T is a Möbius transformation.
This result strengthens many previous characterizations of Möbius trans-
formations, which we discuss below. We remark that the notion of the
weakly-circle preserving, in dimensions 2 and 3, was first studied by the
first author [12], [10], where it was termed “circle-preserving”.
1.2. Previous Results. Rigidity theorems of Möbius transformation have
been investigated extensively. It is clear that Möbius transformations take
generalized (n − 1)-spheres to generalized (n − 1)-spheres and a converse was
known to Möbius, under the assumption that the map T is continuous, see
Blair [4, Theorem 5.6]. A map is called conformal if the map preserves an-
gles. One can define it formally using conformal manifolds, see (Kobayashi
WEAKLY CIRCLE-PRESERVING MAPS IN INVERSIVE GEOMETRY 3
Notice that the image of S2 contains x01 , x02 and a whole circle, so T (S2 )
is in circular general position. By the five-point theorem (Theorem 2.2), T
is a Möbius transformation on S2 . Also notice that S2 must intersect both
components in Sk+1 0 divided by Sk0 , so we let x3 ∈ S2 − Sk0 on the opposite
0
side of x1 , and x3 be its image. Since T is a Möbius transformation on S2 ,
we have x03 ∈ / T (Sk0 ) and x03 6= x01 . Now given any two points y1 , y2 , we
form a 2-sphere through x1 , x3 , y1 , y2 , then this 2-sphere must intersect Sk0
in a circle as it contains x1 and x2 . Therefore, the image of this 2-sphere
consists of x01 , x03 and a whole circle, which means the image is in circular
general position. Therefore, T is a Möbius transformation on this 2-sphere
by five-point theorem (Theorem 2.2), in particular, T (y1 ) 6= T (y2 ). This
shows that T is injective on Sk+1 0 .
6 JOEL C. GIBBONS AND YUSHENG LUO
not, let x1 , x2 be two points in Sk+1 0 such that x02 ∈/ ((Sk0 )0 , x01 ). Let S2 be
0
a 2-sphere through x1 , x2 and intersecting Sk in a circle C, then T (S2 ) is
in circular general position, so T is a Möbius transformation on S2 . But
S20 ∩ ((Sk0 )0 , x01 ) contains x01 and C 0 , so S20 ⊂ ((Sk0 )0 , x01 ), then x02 ∈ S20 ⊂
((Sk0 )0 , x01 ). But this is a contradiction to the assumption x02 ∈ / ((Sk0 )0 , x01 ).
Therefore T (Sk+1 ) lies in some (k + 1)-sphere.
Case m = 1:
(1): This is immediate from the hypothesis of the lemma.
(2): Let T (Sk+1 ) − T (Sk0 ) = {x0 } and fix x ∈ T −1 (x0 ). Suppose for
contradiction that there is y ∈ Sk+1 − Sk0 with y 0 := T (y) ∈ T (Sk0 ).
Claim 1. There is a circle C through x, y and intersect Sk0 at two points
and y 0 ∈
/ T (C ∩ Sk0 ).
Proof. (of Claim 1)
Since Sk0 has codimension 1 in Sk+1 , Sk0 divides Sk+1 into two components.
If x, y are in opposite side of Sk0 , then any circle containing x, y intersects
0
Sk at two points. Let C1 and C2 be two different circles through x, y, then
C1 ∩ C2 ∩ Sk0 = ∅ (as three points determines a circle). Since T is injective
on Sk0 , so T (C1 ∩ Sk0 ) ∩ T (C2 ∩ Sk0 ) = ∅. Hence, y 0 ∈
/ T (Ci ∩ Sk0 ) for at least
one of C1 , C2 .
Otherwise, choose two different point x1 and x2 in the opposite side and
form Ci = (xyxi ). Exact same argument shows that at least one of the two
circles will satisfy the desired property.
Now let y1 , y2 be the intersection point of the circle C with Sk0 , and y10 , y20
be its image respectively. Since T is weakly circle preserving and y 0 , y10 , y20
determines a circle (as no two of them are equal), we have x0 lies on the
circle (y 0 y10 y20 ) ⊂ T (Sk0 ) which is a contradiction to x0 ∈
/ T (Sk0 ).
Case m = 2:
(1): Given Sk+2 containing Sk0 , and suppose for contradiction that |T (Sk+2 )−
T (Sk0 )| ≥ 3. Choose x01 , x02 , x03 ∈ T (Sk+2 ) − T (Sk0 ) and fix x1 , x2 , x3 in its
preimage, i.e. T (xi ) = x0i , and let Sk+1,i := (Sk0 , xi ) for i = 1, 2, 3. Notice
that dim(Sk+1,i ) = k + 1 as xi ∈ / Sk0 , so by the (2) of m = 1, we have
T (Sk+1,i − Sk0 ) = {x0i }. Now given y ∈ Sk0 , and y 0 be its image under T .
Claim 2. There is a circle Cy though y and intersect Sk+1,i − Sk0 for all
i = 1, 2, 3.
Proof. (of Claim 2)
8 JOEL C. GIBBONS AND YUSHENG LUO
Figure 1.
Since T (Sk+1,i − Sk0 ) = {x0i }, we have {x01 , x02 , x03 } ⊂ T (Cy ). Since T is
weakly circle preserving, T (Cy ) ⊂ (x01 x02 x03 ), in particular, y 0 := T (y) ∈
(x01 x02 x03 ). This is true for any y ∈ Sk0 , so T (Sk0 ) ⊂ (x01 x02 x03 ) which is a
contradiction.
(2): Let T (Sk+2 ) − T (Sk0 ) = {x01 , x02 }. Fix x1 , x2 in its preimage and let
Sk+1,i := (Sk0 , xi ). Again, dim(Sk+1,i ) = k + 1, and T (Sk+1,i − Sk0 ) = {x0i }.
Suppose for contradiction that there exists y ∈ Sk+2 − Sk0 such that
y 0 := T (y) ∈ T (Sk0 ), let Sk+1,y = (Sk0 , y). Notice that it is necessary that
T (Sk+1,y ) = T (Sk0 ).
Claim 3. T (Sk+1,y − Sk0 ) = {y 0 }.
Proof. (of Claim 3)
Again, under a Möbius transformation, we may assume that Sk0 = Rk∞ ⊂
Rn∞ and Sk+2 = Rk+2 ∞ ⊂ R∞ .
n
Figure 2.
{x01 , x02 , y10 , y20 } ⊂ T (C), so x01 , x02 , y10 , y20 are on a circle. This is a contra-
diction to dim(Sk+2 0 ) ≥ k + 2 by lemma 3.2.
Now replace Sk+1,3 by Sk+1,y in claim 2, and exact same argument shows
that T (Sk0 ) ⊂ (x01 x02 y 0 ) which is a contradiction.
Case m ≥ 3:
Induction hypothesis: Assume the lemma holds for 1, ..., m − 1.
(1): Given Sk+m containing Sk0 and dim(Sk+m 0 ) ≥ k + m. Suppose for
0
contradiction that |T (Sk+m ) − T (Sk )| ≥ m + 1.
Let {x01 , ..., x0m } ⊂ T (Sk+m ) − T (Sk0 ) such that T (Sk0 ) and {x01 , ..., x0m }
determine a (k + m)-sphere, then T (Sk0 ) and any l-subset of {x01 , ..., x0m }
determine a (k + l)-sphere. Let x0m+1 ∈ T (Sk+m ) − T (Sk0 ) − {x01 , ..., x0m }, and
choose a point in {x01 , ..., x0m } that not in the (k + 1)-sphere determined by
T (Sk0 ) and x0m+1 , say this point is x0m . Fix x1 , ..., xm+1 in its preimage, and
let Sk+m−1 := (Sk0 , x1 , ..., xm−1 ) and Sk+2 := (Sk0 , xm , xm+1 ).
We let D := dim(Sk+m−1 ), then it is clear that D ≤ k + m − 1, and
0
dim(Sk+m−1 ) = k +m−1 ≥ D, so by induction hypothesis, we have m−1 ≤
|T (Sk+m−1 ) − T (Sk0 )| = D − k ≤ m − 1. This implies that D = k + m − 1
and T (Sk+m−1 − Sk0 ) = {x01 , ..., x0m−1 }.
Similarly, we have dim(Sk+2 ) = k + 2 and T (Sk+2 − Sk0 ) = {x0m , x0m+1 }.
Notice that Sk0 ⊂ Sk+m−1 ∩ Sk+2 , so Sk+m−1 ∩ Sk+2 contains at least two
points. By lemma 3.1, we know dim(Sk+m−1 ∩ Sk+2 ) ≥ (k + m − 1) + (k +
2)−(k +m) = k + 1. This means that there exists x ∈ (Sk+2 −Sk0 )∩Sk+m−1 .
Notice that T (x) ∈ T (Sk+2 − Sk0 ) = {x0m , x0m+1 }. Therefore, T (Sk + m −
1) − T (Sk0 ) contains at least m points, which is a contradiction.
(2): Let T (Sk+m )−T (Sk0 ) = {x01 , ..., x0m }, and fix x1 , ..., xm in its preimage,
and let Sk+m−1 := (Sk0 , x1 , ..., xm−1 ). Exact same argument in the proof of
(1) of case m ≥ 3, we have dim(Sk+m−1 ) = k + m − 1 and T (Sk+m−1 − Sk0 ) =
{x01 , ..., x0m−1 }.
Suppose for contradiction that there exists y ∈ Sk+m − Sk0 such that y 0 :=
T (y) ∈ T (Sk0 ). Let Sk+2 := (Sk0 , xm , y). Notice that k + 1 ≤ dim(Sk+2 ) ≤
10 JOEL C. GIBBONS AND YUSHENG LUO
For the induction step, assume that T maps (k + 1)-spheres into (k + 1)-
spheres, for a fixed k + 1 ≥ 2. Given a k-sphere Sk , choose a point x1 ∈ Sn −
0
Sk , let Sk+1,1 := (Sk , x1 ), then by induction hypothesis, dim(Sk+1,1 ) = k+1
(recall here (Sk , x1 ) means the smallest dimension sphere containing Sk and
0
x1 and Sk+1,1 = (T (Sk+1,1 ))). Since T (Sn ) is not contained in some (n − 1)-
sphere, there is an image point x02 ∈ Sn − Sk+1,1 0 . We fix x2 ∈ T −1 (x02 ), and
denote Sk+1,2 := (Sk , x2 ). Notice that Sk ⊂ Sk+1,1 ∩ Sk+1,2 , so that
0 0
T (Sk ) ⊂ T (Sk+1,1 ) ∩ T (Sk+1,2 ) ⊂ Sk+1,1 ∩ Sk+1,2 .
0
But Sk+1,1 0
6= Sk+1,2 as x02 ∈ Sk+1,2
0 0
− Sk+1,1 0
, so Sk+1,1 0
∩ Sk+1,2 is a sphere
of dimension less than or equal to k. This proves that T (Sk ) lies in some
k-sphere, and completes the induction step.
Remark 4.2. Notice that if T (Sn ) is in spherical general position, then
necessarily T (Sn ) is not contained in any (n − 1)-sphere. It follows that
a weakly sphere-preserving map having image in spherical general position
satisfies the hypothesis of Lemma 4.1, hence is a weakly circle preserving
map.
Now we have the following analogue theorem for weakly sphere-preserving
maps.
Theorem 4.3. For n ≥ 3, T : Sn −→ Sn be a weakly sphere-preserving
map. If T (Sn ) is in spherical general position and there is a 2-sphere S2
with T (S2 ) in circular general position, then T is a Möbius transformation.
Proof. This follows immediately from Lemma 4.1 and Theorem 1.5
5. Acknowledgments
The first author thanks N. Nygaard (University of Chicago) for helpful
discussions. The second author worked on this project in an REU Summer
Program at the University of Michigan with supervisor J. C. Lagarias. Both
authors thank J. C. Lagarias for helpful discussions and for editorial assis-
tance. We acknowledge use of the computer package Cinderella 2 [19] to
draw the figures in this paper.
References
[1] J. Aczel, and M. A. McKiernan, On the characterisation of plane projective and com-
plex Mobius transformation. Math. Nachr. 33 (1967), 315-337
[2] A. F. Beardon The Geometry of Discrete Groups, Springer-Verlag: New York 1983.
[3] A. F. Beardon, and D. Minda, Sphere-preserving maps in inversive geometry, Proc.
Amer. Math. Soc. 130 (2001), No. 3, 987-998.
[4] D. E. Blair Inversion Theory and Conformal Mapping, Student Mathematical Library,
v. 9
[5] C. Carathéodory:, The most general transformation of plane regions which transform
circles into circles. Bull. Amer. Math. Sco. 43 (1937), No. 8), 573-579.
[6] A. I. Chubaev and I. Pinelis, Fundamental theorem of geometry without the 1-to-1
assumption, Proc. Amer. Math. Soc. 127 (1999), no.9, 2735-2744.
12 JOEL C. GIBBONS AND YUSHENG LUO
Logistic Research & Trading Co., P.O. Box 63, St. Joseph, MI 49085