The Geometry of Mbius Transformations

John Olsen

University of Rochester Spring 2010

Contents
1 Introduction 2 Ane Transformations 3 The Stereographic Projection 4 The Inversion Map 5 Mbius Transformations 6 The Cross Ratio 2 2 4 9 11 14

7 The Symmetry Principle and Maps of the Unit Disk and the Upper Halfplane 17 8 Conjugacy Classes in M(
)

23 26 28 28 31 34 35

9 Geometric Classication of Conjugacy Classes 10 Mbius Transformations of Finite Period 11 Rotations of

12 Finite Groups of Mbius Transformations 13 Bibliography 14 Problems

Introduction

The purpose of these notes is to explore some basic properties of Mbius transformations (linear fractional transformations) which are one-to-one, onto and conformal (angle preserving) maps of the socalled extended complex plane. We will develop the basic properties of these maps and classify the one-to-one and onto conformal maps of the unit disk and the upper half plane using the symmetry principle. The one-to-one, onto and conformal maps of the extended complex plane form a group denoted P SL2 ( ). We will study the conjugacy classes of this group and nd an explicit invariant that determines the conjugacy class of a given map. We nish with a classication of the nite subgroups of P SL2 ( ). The theory of Mbius Transformations is developed without any use of and only one reference to complex analysis. This point of view certainly requires more work, but I feel the eort is worth it, since it allows somebody with no knowledge of complex analysis to study the subject. The prerequisite is some basic knowledge of group theory, which is certainly met if the students have taken an undergraduate algebra course. If not, a couple of lectures at the beginning of the course where one introduces the basics of group theory should suce. I would like to thank John Harper and the students in the course Introduction to Geometry (MTH 250, Spring 2010) for valuable comments and proofreading.

Ane Transformations

Let us briey recall a few basic properties of the complex numbers. If z , then we can write z = r(cos() + i sin()), where r is the modulus, |z|, of z and the argument, arg(z), of z. We will denote the real and imaginary part of z by (z) and (z), respectively. We have the following ve basic maps, which we will study in the following: 1. z cz, c

, scaling
, translation

2. z z + A, A

3. z Az, A = ei , rotation 4. z z, complex conjugation

Denition 1. A direct ane transformation is a combination of (1), (2) and (3), i.e. a map of the form T (z) = Az + B. 2

Remark 1. Since we have |T (z) T (z )| = |(Az + B) (Az + B)| = |A||z z |, we see that a direct ane transformation is an isometry if and only if |A| = 1. Lemma 2.1. The direct ane transformation T (z) = Az + B is a translation if and B only if A = 1. If |A| = 1, A = 1 then T is a rotation about 1A by an angle arg(A). Proof. If A = 1, then T (z) = z + B is a translation. If |A| = 1 and A = 1, we let F be a xed point of T , i.e. a point where T (F ) = F . Then we have F = AF + B, B which implies F = 1A . Now T (z) F = Az A2 z + B BA B 1A 2 Az A z BA = 1A Az(1 A) B = 1A = A(z F ).

This shows that if |A| = 1 and A = 1 then T is a rotation about F by an angle arg(A). Proposition 2.1. The set of direct ane transformations form a group under composition, which is denoted by Af f ( ). Proof. The identity map I(z) = z is the unit in the group Af f ( ). Since composition of functions is always associative, we only need to check that the set is closed and that inverses exist. Let T (z) = Az + B and S(z) = A z + B , then we have S T = A (Az + B) + B = A Az + A B + B , which is again an ane transformation. To nd an inverse to T (z) = Az + B, we guess that T 1 (z) = Cz + D. Then we have T 1 T (z) = z which means that C(Az+B)+D = z. So we get CAz+CB+D = z, which means that C = A1 , and D = AB 1 . Corollary 2.1. A direct ane transformation preserves circles and lines. Proof. A direct ane transformation, T (z) = Az+B, where |A| = 1 is by Lemma 2.1 B a rotation about 1A which clearly preserves circles and lines. A direct ane transformation, T (z) = Az + B, where |A| = 1 can be written as T (z) = r(A z + B ), B where r is real and |A | = 1. Again by Lemma 2.1 this map is a rotation about 1A scaled by r, which preserves circles and lines. 3

Remark 2. A map that preserves angles is called conformal. By the same argument as above, a direct ane transformation is conformal.

The Stereographic Projection

O z

Consider the unit sphere S 2 in

3, that is S 2 = {(u, v, w) 3 | u2 + v 2 + w 2 = 1}.

We embed S 2 such that the origin and the center of the sphere coincide. We identify the complex plane with the equitorial plane. That is, for coordinates (u, v, w) of 3 , is the plane where w = 0. The North pole has coordinates N = (0, 0, 1). We will denote a complex number z by x + iy. Notice that the unit circle in coincides with the equator of S 2 . For any point P S 2 , there is a unique line from N to P , which we denote by NP . This line intersects the complex plane in exactly one point z . Denition 2. The map from SP : S 2 \ N which assigns to a point P the point z given by the intersection NP is called the stereographic projection. We will now derive the coordinate expressions for both stereographic projection and its inverse.

Proposition 3.1. For P S 2 \ N and z , P = (u, v, w) and z = x + iy the coordinates of the stereographic projection SP : S 2 \ N and its inverse 1 2 SP : S \ N are given by: v u +i (1) SP ((u, v, w)) = 1w 1w and SP 1 (x + iy) = = Proof.
N 1w L w z P

2x 2y x2 + y 2 1 , 2 , 2 x2 + y 2 + 1 x + y 2 + 1 x + y 2 + 1 z + z i(z z) zz 1 . , , zz + 1 zz + 1 zz + 1

(2) (3)

From the picture we see that |NO| = 1, |LO| = w, |LN| = 1 w, |zO| = r = x2 + y 2 . Set |P L| = and note that = u2 + v 2 + w 2 w 2 = u2 + v 2 . Since the triangles NLP and NOZ are similar we have 1w u v = = = . r 1 x y from which is follows that u u 1w = x= 1 x 1w 1w v v = y= . 1 y 1w

v u This proves that SP ((u, v, w)) = 1w + i 1w . For the second statement we observe that since z = x + iy we have

z =

v u + iv u +i = 1w 1w 1w v u iv u i = z = 1w 1w 1w 5

This implies that x2 + y 2 = zz = u + iv u iv = 1w 1w (1 w)(1 + w) u2 + v 2 = = = 2 (1 w) (1 w)2 1 1+w = 1 + . = 1w 1w

where we used u2 + v 2 = 1 w 2 in the fourth equality. This gives 1w = Since we have x + iy = and we get u + iv u + iv 1w = 1w x + iy 2 2x + 2iy (x + iy) = 2 . 2+1 +y x + y2 + 1 2 x2 + y 2 1 zz 1 w= 2 = . 2 + y2 + 1 2+1 x x +y zz + 1

u + iv = (1 w)(x + iy) =

x2

Comparing real and imaginary parts gives u= x2 2x z+z 2y i(z z) = v= 2 = , 2+1 2+1 +y zz + 1 x +y zz + 1 = {}.

which nishes the proof. Denition 3. The extended complex plane is given by

Remark 3. If we identify, via stereographic projection, points in the complex plane with points in S 2 \N and further identify with N then we have a bijection between the extended complex plane and S 2 . Under this identication S 2 is known as the Riemann sphere. It is clear that stereographic projection is continuous as a map S 2 \ N with a continuous inverse, since both maps are given as fractions of polynomials where the denominator is never zero. With the above identication of N with we get a continuous map SP : S 2 with continuous inverse, SP 1 : S 2 . A continuous map with continuous inverse is called a homeomorphism. We note that since S 2 is compact and stereographic projection is a homeomorphism, is compact as well. The space is a1 socalled one-point compactication of .
1

There are other one-point compatications of

The stereographic projection gives a way of mapping a region of the sphere onto a plane. Using2 the methods of Riemannian geometry, Gauss proved that it is impossible to nd such a map that preserves both distances and angles. We will now see that the stereographic projection preserves angles, i.e. it is conformal. Proposition 3.2. The map SP : S 2 given by stereographic projection and its inverse SP 1 : S 2 are conformal maps. Proof. Since the identity is conformal it is enough to prove that one of the maps is conformal, since the composition of two conformal maps is again conformal. We prove that SP 1 : S 2 preserves angles by showing that the angle between two is the same as the angle bewteen the two tangent vector at the point of lines in intersection of the two curves lifted to S 2 . We can assume that one of the lines in is the x-axis and that the intersect at an angle at the point p. That is consider the two lines r(t) = p + t and s(t) = t ei +p. z+z We have SP 1(z) = zz+1 , i(zz) , zz1 , so we get zz+1 zz+1 r(t) = SP 1 (r(t)) = (2(p + t), 0, (p + t)2 1) 1 + (p + t)2

s(t) = SP 1 (s(t)) = (2(p + t cos()), 2t sin(), (p + t cos())2 + t2 sin2 () 1) . = 1 + (p + t cos())2 + t2 sin2 () These two curves intersect at t = 0, so to nd the angle between r(t) and s(t) we dierentiate and evaluate at t = 0. (2(p2 1), 0, 4p) (p2 + 1)2 (2(p2 1) cos(), 2(p2 + 1) sin(), 4p cos()) . s (0) = (p2 + 1)2 r (0) = We let be the angle between r (0) and s (0) and by calculating the dot product we
A more accurate way of saying this is that by inventing Riemannian geometry Gauss proved a theorem usually referred to as Theorema Egregium, or, in English, totally awesome theorem, which states that it is impossible to construct a distance and angle preserving map between two spaces if has curvature 0 and S 2 has the Gaussian curvature of the two space is not equal. In our case, curvature 1.
2

have cos() = (4(p2 1)2 + 16p2)(4(p2 1)2 cos2 () + 4(p2 + 1)2 sin2 () + 16p2 cos2 ()) 4(p2 + 1)2 cos() = = 2(p2 + 1)2(p2 + 1) = cos(). So we see that = , which means the angle is preserved. In the following it will be useful to have a clear denition of what is meant by a circle on S 2 . A great circle on S 2 is given by the intersection of 2-dimensional subspace with S 2 , and in general, a circle on S 2 is the intersection of a 2-dimensional plane with S 2 . One extremely useful observation is the following. Proposition 3.3. Stereographic projection takes circles to circles and lines. Proof. Let us rst consider the case where the circle goes through N. Let P be any point on the circle, then the line NP lies in the plane. Since the intersection of the plane with is a line, stereographic projection takes the circle through N to a line in . The other case is proved by considering the equation for the plane in 3 , Au + Bv+Cw+D = 0, and express u, v, w in terms of x, y via stereographic projection and see that it denes a circle. From Proposition 3.1 we have w = zz1 , u = (1 w)x = zz+1 2 2 x and v = (1 w)y = zz+1 y. Plugging this into the equation for the plane we zz+1 get = 4(p2 1)2 cos() + 16p2 cos()

A(1 w)x + B(1 w)y + Cw + D = 0 2 2 zz 1 zz + 1 A x+B y+C +D = 0. zz + 1 zz + 1 zz + 1 zz + 1 This gives 2Ax + 2By + C(zz 1) + D(zz + 1) = 0 2Ax + 2By + C(x2 + y 2 1) + D(x2 + y 2 + 1) = 0 (C + D)(x2 + y 2) + 2Ax + 2By + D C = 0, which is a circle in 8

Remark 4. What does an open neighborhood around look like? A open neighborhood around N S 2 is the interior of a circle with center N in S 2 . The circle is under stereographic projection mapped to a circle with the origin as its center of some radius, say r. This means that the interior of the circle around N is mapped to the points {z | |z| > r}. So since stereographic projection is a homeomorphism an open neighborhood around in is the set {z | |z| > r}.

The Inversion Map

1 In this section we will study the map T , T (z) = z , called inversion. If z = 0 then 1 there is a unique w = z , so T : \ 0 \ 0 is a bijection. Our goal is to extend T to 1 z a homeomorphism, T : . Notice that z = |z|2 , so T is the composition of z the two maps S(z) = z and R(z) = |z|2 . Furthermore we see that arg(R(z)) = arg(z) 1 and that |R(z)| = |z| , so R is inversion in the unit circle.

Proposition 4.1. Let T denote the inversion. Under the identications that T (0) = and T () = 0, T is a homeomorphism T : . Proof. The map T is clearly a bijection T : C so since T is continuous (it is the composition of the continuous maps R and S) as a map T : C \ {0} \ {0} all we need to check is that it is continuous at 0 and . We have 1 lim T (z) = lim T ( ) = lim z = , |z| |z|0 |z| z 1 lim T (z) = lim T ( ) = lim z = 0. |z| |z|0 |z| z This shows that T : C is a homeomorphism.

is continuous. Since the map T is its own inverse, it

Proposition 4.2. Let T denote the inversion. The map T takes circles and lines to circles and lines. Proof. We will not give all the details, since this is the same argument as in the proof of Proposition 3.3. Let w = 1 be the image of z under T . If w = a + ib and z = x + iy then we have z x a a = x2 +y2 , b = x2y 2 , x = a2 +b2 and y = a2b 2 . The equation +y +b A(x2 + y 2 ) + Bx + Cy + D = 0

is a circle or, if A = 0, a line. Using the expressions for x, y in terms of a, b and substituting it into the equation for the circle or line, we get that a, b satisfy D(a2 + b2 ) + Ba Cb + A = 0 which is the equation for a circle or a line. Remark 5. Note that if the circle goes through the origin it must be mapped to a line, since the image is unbounded, and a circle is bounded. Example 1. Let us nd the image of the vertical line x = c1 under the inversion map. According to Equation 4, the image is the circle c1 (a2 + b2 ) + a = 0 (put 1 1 A = 0, B = 1, C = 0 and D = c1 ). This can be rewritten as (a 2 c1 ) + b2 = ( 2 c1 )2 , 1 which a circle centered at ( 2 c1 , 0) of radius 1 c1 . Note that the circle goes through 2 the origin, which we knew, since the image is a line. Example 2. Let us nd the image of the line x = 0 under the inversion map. Since x a = x2 +y2 , we see that the image has x-coordinate equal to zero, hence the image lies on the imaginary axis. Now, b = x2y 2 = y = 1 . So the inversion takes y 1 +y y2 y y and maps 0 . Next we want to show that the inversion is a conformal map . This can be done in two dierent ways, we will show it geometrically by using the stereographic projection. Remark 6. If we take z = x + iy and z = x iy and map them onto S 2 by SP 1. How do the the points SP 1 (z) and SP 1(z) dier? We use the equation for SP 1 to get SP 1(z) = SP 1 (x + iy) = SP 1(z) = SP 1 (x iy) = 2x 2y x2 + y 2 1 , 2 , 2 x2 + y 2 + 1 x + y 2 + 1 x + y 2 + 1 2y x2 + y 2 1 2x , 2 , 2 x2 + y 2 + 1 x + y 2 + 1 x + y 2 + 1

This means that the two points SP 1 (z) and SP 1 (z) dier by a reection in the u, w-plane. If we let : S 2 S 2 denote the reection in the u, w-plane, we see that the composition SP SP 1 corresponds to complex conjugation, i.e. SP SP 1(z) = z. z Similarly we can take the points z and |z|2 (invert z in the unit circle) and map them onto S 2 by SP 1 , and ask how they dier. First note that since the unit circle is xed under inversion in the unit circle, the corresponding map on S 2 must x the equator. Furthermore, it must interchange the points N and S (the South pole), 10

since the inversion in the unit circle maps 0 to . A calculation shows that SP 1 (z) z and SP 1( |z|2 ) dier by a reection in the equitorial plane. If we let : S 2 S 2 denote the reection in the equitorial plane, we have that SP SP 1 corresponds z to inversion in the unit circle, i.e. SP SP 1(z) = |z|2 . This means that as a map of S 2 , the inversion map T is given by or as a map of it can be written as T = SP SP 1 : Proposition 4.3. The inversion T :

is a conformal map.

Proof. Recall that the inversion is the composition of complex conjugation and inversion in the unit circle. Using the notation and results from the previous remark, the map SP SP 1 : is equal to the inversion map, T . Since the maps SP , , and SP 1 are conformal (that reections are conformal is easy to check), we see that T is conformal.

Mbius Transformations

Having shown the basic properties of ane transformations, stereographic projection and the inversion map, we are now in a position to study Mbius transformations. Denition 4. A Mbius transformation f : f (z) = az + b , cz + d a, b, c, d

is a map

and ad bc = 0.

We have the following basic theorem about Mbius transformations. Theorem 1. Let f be any Mbius transformations, then 1. f can be expressed as a composition of ane transformations and inversions. 2. f maps

one-to-one onto itself, and is continuous.

3. f maps circles and lines to circles and lines. 4. f is conformal. Proof. 1. We write f as
a (cz + d) ad + b az + b a b ad c c = c = + . cz + d cz + d c cz + d

11

1 If we let w1 , w2 and w3 be the maps w1 = cz + d, w2 = w1 and w3 = (b ad )w2 + a , c c then f = w3 w2 w1 . Note that if c = 0, there is no inversion in the decomposition of f . 3. Since we have shown that both ane transformations and inversions take circles and lines to circles and lines, it follows from 1.) that f takes circles and lines to circles and lines. az+b 2. If z = d and w = cz+d , then z = dw+b . So, at every point z , z = d , f is c cwa c well-dened, one-to-one, onto and continuous. We extend f to a map f : by setting f () = , if c = 0. If c = 0, then we set f ( d ) = . One can check c that this makes f continuous as a function f : . The inverse of f is given by

f 1 (w) =

dw + b . cw a

Again, if c = 0, we set f 1 () = . If c = 0, then we set f 1 ( a ) = . With c these choices, one can check that f 1 : is continuous. In summary, f : is a homeomorphism. 4. This also follows from 1. since we have shown that ane transformations and inversions are conformal, see Remark 2 and Proposition 4.3. Remark 7. As a map f : it is continuous and conformal at every point d z = c . As a map f : it is continuous and conformal everywhere. One can determine if the image of a circle is a line or a circle by using the following az+b argument. The map f = cz+d is continuous as a map at every point except z = d (this point is called a pole of f ) and since a continuous function maps a c bounded set to a bounded set, we conclude the if z = d does not lie on the circle, c the image is bounded and hence a circle. If z = d lies on the circle, z = d is c c mapped to and the image is a line. Denition 5. Let M( M(
) )

denote the set of Mbius transformations

= {f :

| f (z) =

az + b , ad bc = 0}. cz + d

This set has some additional structure; it is a group under composition of functions and can naturally be identied with GL2 ( ). We state this as a theorem. Theorem 2. The set M( ) is a group under composition of functions and there is a surjective group homomorphism : GL2 ( ) M( ) with kernel the diagonal matrices. 12

Proof. We must show that the composition of two Mbius transformations is again a2 z+b2 1 z+b a Mbius transformation. Let f1 (z) = a1 z+d1 and f2 (z) = c2 z+d2 be Mbius transforc 1 mations. One easily checks that f2 f1 (z) = (a2 a1 + b2 c1 )z + a2 b1 + b2 d1 , (c2 a1 + d2 c1 )z + c2 b1 + d2 d1

which again a Mbius transformation.3 The identity f (z) = z is a Mbius transformation. In the proof of Theorem 1 we saw that for a Mbius transformation f (z) = az+b the inverse is given by cz+d f 1 (w) = dw + b . cw a

Furthermore, the composition of functions is associative, so we have shown that the set M( ) is a group under composition of functions. We have a map : GL2 ( ) M( ) given by a b c d az + b . cz + d

The map sends the identity matrix to the map f (z) = z, which is the identity in the group M( ). We must show that the product of two matrices is mapped to a1 z+b1 a2 z+b2 the product of two Mbius transformations. Let f1 (z) = c1 z+d1 and f2 (z) = c2 z+d2 be Mbius transformations. a2 a1 + b2 c1 a2 b1 + b2 d1 a1 b1 a2 b2 = c1 d 1 c2 d 2 c2 a1 + d2 c1 c2 b1 + d2 d1 (a2 a1 + b2 c1 )z + a2 b1 + b2 d1 = f2 f1 (z). (c2 a1 + d2 c1 )z + c2 b1 + d2 d1

This means that under the map : GL2 ( ) M( ) matrix multiplication corresponds to composition of functions, which means that it is a homomorphism. It is az+b onto since the Mbius transformation f (z) = cz+d is hit by the matrix with the same entries. The kernel, Ker(), is the subgroup of matrices that are mapped to the az+0 identity map. The Mbius transformation f (z) = z can be written as f (z) = 0z+a , which is exactly the image of the diagonal matrices.
One should add that f2 f1 cannot reduce to a constant. This is clear since the composition of two functions both of which are one-to-one and onto is again one-to-one and onto.
3

13

Remark 8. The kernel of is Ker() = kI, k and I the identity matrix. Hence induces a group isomorphism : GL2 ( )/kI M( ). The group GL2 ( )/kI is denoted P GL2 ( ). If we let SL2 ( ) denote the complex matrices with determinant one, one can show that the map : SL2 ( ) M( ) is onto, and has kernel I, so one gets an isomorphism : SL2 ( )/ I M( ). This just says that, instead of the condition ad bc = 0, we could just as well have used ad bc = 1. We end this section with a few examples. Example 3. Find the image of interior of the circle |z 2| = 2 under the map z f (z) = 2z8 . Note rst that z = 4 is on the circle, so the image must be a line. We i see that f (0) = 0 and f (2 + 2i) = 2 , so the image is the imaginary axis. Since a continuous map preserves connectedness, the interior of the circle is either mapped 1 to the right or left half plane. Since f (2) = 2 , the set {z | |z 2| < 2} is mapped to the set {z | (z) < 0}.

Example 5. Determine the image of the second quadrant {z | (z) < z+i 0 and (z) > 0} under the mapping f (z) = zi . Let us rst see where the two axis are mapped. Since f (2i) = 3 and f (3i) = 2 the imaginary axis is mapped to the real axis. We have f (0) = 1 and f (1) = i, so the imaginary axis is mapped to the line {z = x+iy | y = x1}. We also have f (1+3i) = 92i , so we see that f maps the 5 set {z | (z) < 0 and (z) > 0} to the set {z = x + iy | y > x 1 and y < 0}.

Example 4. Construct a conformal map {z | |z| < 1} {z | (z) > 0}. Since Mbius transformations are conformal, we construct a Mbius transformation that takes one set to the other. First we look for a map that takes the unit circle to the imaginary axis. The map must have a pole on the unit circle, since the z+1 imaginary axis is unbounded. Look at f1 (z) = z1 , which satises f1 (1) = and f1 (1) = 0, so it maps the unit circle onto some straight line through the origin. Since f1 (i) = i it maps {z | |z| = 1} to {z | (z) = 0}. We see that the interior {z | |z| < 1} is mapped to {z | (z) < 0}, since f1 (0) = 1 and z+1 since f maps connected sets to connected sets. Hence the map f (z) = z1 maps {z | |z| < 1} {z | (z) > 0}.

The Cross Ratio

We have already seen that Mbius transformations map circles to circles4 . In this section we want to nd a specic Mbius transformation that takes a specic circle
In this section we will not destinguish between circles and lines, since a line in closed at .
4

is a circle in

14

to another specic circle. Recall from Euclidean geometry that three points uniquely determine a circle. Let us denote one circle by C1 and one by C2 . We choose points z1 , z2 and z3 on C1 and w1 , w2 and w3 on C2 . Then if we nd a Mbius transformation h that takes h(z1 ) = w1 , h(z2 ) = w2 , h(z3 ) = w3 (4)

then h must map C1 to C2 . The trick is rst to map C1 onto the real axis, then map the real axis onto C2 . To map C1 onto the real axis is the same as solving Equation 4 for w1 = 0, w2 = 1 and w3 = . If the points zi = , we dene a Mbius transformation f by f (z) = (z z1 )(z2 z3 ) (z z3 )(z2 z1 ) (5)

which clearly takes f (z1 ) = 0, f (z2 ) = 1, f (z3 ) = . If one of the three points zi = (which means that C1 is a line) we have f (z) = z2 z3 (z1 = ), z z3 f (z) = z z1 (z2 = ), z z3 f (z) =

which satisfy f (z1 ) = 0, f (z2 ) = 1, f (z3 ) = . Now let g be another Mbius transformation which takes g(w1) = 0, g(w2 ) = 1, g(w3) = . Then we see that the map h = g 1 f satises h(z1 ) = g 1 f (z1 ) = g 1 (0) = w1 h(z2 ) = g 1 f (z2 ) = g 1 (1) = w2 h(z3 ) = g 1 f (z3 ) = g 1 () = w3 . Notice that the equation h(z) = w can be written as g 1(f (z)) = w which means that (z z1 )(z2 z3 ) (w w1 )(w2 w3 ) = . (z z3 )(z2 z1 ) (w w3 )(w2 w1 ) These fractions are called cross ratios. Denition 6. Let z, z1 , z2 , and z3 be four points in (z, z1 , z2 , z3 ) =
.

z z1 (z1 = ) (6) z2 z1

g(w) = f (z)

(7)

(8)

Then the expression

(z z1 )(z2 z3 ) (z z3 )(z2 z1 )

is called the cross ratio of the four points z, z1 , z2 , and z3 . 15

To map z1 , z2 , and z3 onto w1 , w2 , and w3 we have to solve Equation 7 for w in terms of z, which by Equation 8 is the same as equating the two cross ratios and solving for w in terms of z. Note that the order of the terms in the cross ratio is essential. We now give some examples. Example 6. Find a Mbius transformation that takes 0 to i, 1 to 2 and 1 to 4. We calculate the approprate cross ratios 2z (z 0)(1 (1) = (z (1))(1 0) z+1 w i(2 4) 2(w i) (w, i, 2, 4) = = . (w 4)(2 i) (w 4)(2 i) (z, 0, 1, 1) = So to nd the Mbius transformation we must equate the two expressions and solve for w. 2z 2(w i) = (w 4)(2 i) z+1 which gives w = h(z) = (16 6i)z + 2i (6 2i)z + 2

which is the desired Mbius transformation. The next example is analogous to an earlier example which we treated by an ad hoc method. The concept of an orientation gives us an algorithm to solve similar problems. Remark 9. A circle is not only determined by the three points z1 , z2 and z3 but also oriented by the three points. A line also needs three points to orient it, since it closes up at . The orientation is given by proceeding through the three points in succession. This orientation determines a left region, namely the region that is to the left of an observer traversing the circle in the direction of the orientation. Using conformality of the Mbius transformation it can be shown that given two circles C1 and C2 with orientations determined by the points z1 , z2 and z3 and w1 , w2 and w3 , the Mbius transformation maps the left region of C1 to the left region of C2 . Example 7. Find a Mbius transformation that takes the region D1 = {z |z| > 1} to the region D2 = {z | (z) < 0}. 16 |

We choose both D1 and D2 to be left regions. That is accomplished by choosing z1 = 1, z2 = i, z3 = 1 and w1 = 0, w2 = i, w3 = . Since Mbius transformations take left regions to left regions, a solution to the problems is a Mbius transformation that takes 1 to 0, i to i and 1 to . As in the previous example we nd such a Mbius transformation by equating the two cross ratios, i.e. (w, 0, i, ) = (z, 1, i, 1) which is the same as (z 1)(i + 1) w0 = , i0 (z + 1)(i 1) where we have used the rst formula in Equation 6 to calculate the cross ratio for w. This gives the desired Mbius transformation w = h(z) = 1z . 1+z

The Symmetry Principle and Maps of the Unit Disk and the Upper Halfplane

We wish to determine all possible Mbius transformations from the unit disk to itself, and from the upper halfplane to itself. For this, we need the symmetry principle. First we need a few denitions. Denition 7. Two points z1 and z2 are symmetric with respect to a straight line L if L is the perpendicular bisector of the line joining z1 and z2 . A circle is orthogonal to a line L if the tangent of the circle is orthogonal to the line at the point of intersection. This is equivalent to the center of the circle lying on L. Denition 8. Two points z1 and z2 are symmetric with respect to a circle C if every straight line or circle passing through z1 and z2 intersects C orthogonally. Note that this in particular means that the center a of C and are symmetric with respect to C. Since a circle is bounded there is no circle through both a and and any line through a (and ) is orthogonal to C. We can now state the symmetry-preserving properties of Mbius transformations.

17

Theorem 3 (The Symmetry Principle). Let C be a line or circle in and let f be a Mbius transformation. Two points z1 and z2 are symmetric with respect to C if and only if their images under f , f (z1 ) and f (z2 ) are symmetric with respect to the image of C under f , f (C). Proof. Since Mbius transformations are conformal they preserve orthogonality. Two points are symmetric with respect to a circle or line if every circle or line containing the points intersects the given circle or line orthogonally. Since they preserve the class of circles and lines and preserve orthogonality, they preserve the symmetry condition. Given a circle C and a point we would like a formula for the point , the point symmetric to with respect to C. Proposition 7.1. Given a point and a circle C with center a and radius R. Then = R2 + a, a (9)

is the point symmetric to with respect to C. Proof. By equating the cross ratios (w, 0, 1, ) and (z, a R, a + Ri, a + R) we observe that points a R, a + Ri and a + R (and hence all of C) is mapped to the real axis by f (z) = i z (a R) . z (a + R) (10)

By the symmetry principle is symmetric to with respect to C if and only if f () is symmetric to f ( ) with respect to the real axis. That f () is symmetric to f ( ) with respect to the real axis is clearly equivalent to f () and f ( ) are complex conjugate points in , f () = f ( ). Using Equation 10 this is equivalent to i (a R) (a R) z (a R) =i = i (a + R) z (a + R) (a + R) R2 + a. a

which when solving for yields =

This also shows that the point is unique. 18

Remark 10. From Equation 9 we see that arg( a) = arg R2 a = arg R2 ( a) | a|2 = arg( a).

This means that the symmetric points lie on the same line from the center a. Furthermore we have | a| = R2 R2 = | a|| a|. | a|

Next we classify the Mbius transformations that take the unit disk to itself5 . We denote the open unit disk by D 2 . Theorem 4. Let f be a Mbius transformation that takes D 2 to itself, then f (z) = ei z , z 1 where and || < 1.

Proof. Let f be such a Mbius transformation. Then f maps the unit circle Cz to the unit circle Cw . Furthermore, since the interior is mapped to the interior there must be a point , || < 1 which is mapped to the origin, i.e. f () = 0. According to Equation 9 with a = 0 the point = 12 1 = 0

is symmetric to with respect to Cz . By the symmetry principle this implies that 1 f ( ) is symmetric to f () = 0 with respect to Cw . Since the origin is the center of Cw its symmetric point is . This implies f 1 = .
1 .

This means that f has a zero at and a pole at form f (z) = k for some constant k.

This implies that f is of the

z z 1 = k z 1 z

More generally it can be shown that the Mbius transformations are the one-to-one analytic (complex dierentiable) maps of the unit disk to itself.

19

We also know that f (1) must be mapped to some point on Cw , so we have 1 = |f (1)| = |k|| |1 | = |k|. | 1|

This implies that k = ei for some [0, 2]. We see that f must look like f (z) = ei z , z 1 where || < 1.

Now we just need to check that all such maps takes the disk to itself. Let = a + ib, then we have |f (i)| = ei |f (1)| = ei |i | | a + (1 b)i| i = = = i 1 |i 1| |b 1 + ai| (a)2 + (1 b)2 (b 1)2 + a2 = 1,

1 = 1, 1 1 |f (1)| = ei = 1. 1

Since f () = 0 and || < 1 the interior goes to the interior, which proves that f maps the disk to itself. We nish this section with a classication of the Mbius transformations that takes 2 , 2 = {z | (z) > 0}, to 2 . We formulate this in the following theorem.

Theorem 5. The Mbius transformations f (z) = az + b , a, b, c, d cz + d

2 2 are the maps and ad bc > 0.

For the proof we need the following proposition. Proposition 7.2. A Mbius transformation f maps the real line to the real line if and only if the coecients a, b, c and d are real. Proof. We rst assume that the coecients a, b, c and d of f are real. Then the image of the three points 0, 1 and 2 is clearly three points on the real axis. Since the image of a line is determined by the image of three points on that line, this proves that the real axis is mapped to the real axis.

20

Denote the real axis by 6 . Assume now that f ( ) = . This implies that f maps three points on the real axis q1 , q2 and q3 to three points on the real axis r1 , r2 and r3 (we assume that all points are nite, if this is not the case, modify the following cross ratios acordingly, see Equation 6). This means that f is determined by equating the two cross ratios (w, r1 , r2 , r3 ) = (z, q1 , q2 , q3 ) which is the same as (w r1 )(r2 r3 ) (z q1 )(q2 q3 ) = . (z q3 )(q2 q1 ) (w r3 )(r2 r1 ) When we solve for w we get a Mbius transformation w = f (z) with real coecients, since q1 , q2 , q3 , r1 , r2 and r3 are real. Since the points q1 , q2 , q3 , r1 , r2 and r3 were arbitrary, this nishes the proof. This proposition allows us to give a quick proof of the theorem. Proof. Let f be a Mbius transformation that maps 2 to 2 . Since the real axis partitions into two connected components, and a Mbius transformation maps a connected component to a connected component, f must map the real axis to the real axis. By Proposition 7.2 we can choose f as f (z) = We have f (z) = az + b cz + d az + b = (cz + d) |cz + d|2 1 (ac|z|2 + bd + bcz + adz). = |cz + d|2 1 (ac|z|2 + bd + bcz + adz) 2 |cz + d| ad bc (z). |cz + d|2 az + b , a, b, c, d cz + d

Hence we have (f (z)) = az + b cz + d = =

This means that f maps

2 to itself if and only if ad bc > 0.

The subscript is a reminder that this is a circle in

which is closed up at

21

Remark 11. The set of maps M (

f (z) =

az + b | a, b, c, d cz + d

is a subgroup of the the group of Mbius transformations which analogously to the case where the coecients are complex can be identied with the set of matrices GL2 ( )/kI = P GL2 ( ). The subgroup of matrices in GL2 ( ) with positive determinant is denoted by GL+ ( ). We have a surjective map 2

: GL+ ( ) 2

az + b | a, b, c, d cz + d

and ad bc > 0
)

= M+ (

with Ker() = kI, where k an isomorphism

, k > 0, and I is the identity matrix. Hence we get

The group GL+ ( )/kI is denoted P GL+ ( ). 2 2

: GL+ ( )/kI M+ ( 2

One can ask, if there are more conformal bijections of D 2 and 2 than the ones we have determined in the preceeding theorems. The answer relies on some dicult results in complex analysis, so we state the theorem without a proof. Theorem 6. The conformal maps of are precisely the Mbius transformations. In particular, the maps D 2 D 2 and 2 2 determined in Theorem 4 and Theorem 5 are the only conformal bijections of D 2 and 2 .

Since 2 and D 2 both serve as models of hyperbolic geometry, it is important to nd all conformal maps from 2 D 2 . The map 7

(z) =

is a conformal bijection 2 D 2 and we use this map and the calculation we have done for 2 to determine all conformal bijections 2 D 2 . We state the result as a theorem.

zi z+i

(11)

Theorem 7. The set of conformal bijections g : {g = h | h(z) = az + b , cz + d

2 D2 is the set of maps a, b, c, d and ad bc > 0}

(12)

where is given by Equation 11. This set is in one-to-one correspondence with the group P GL+ ( ) 2

The are of course many other conformal bijections

22

Proof. Since is a conformal bijection 2 D 2 it is in view of Theorem 6 clear that all possible conformal bijections 2 D 2 are given by Equation 12. A bijection

{g = h | h(z) = {h | h(z) =

az + b , a, b, c, d cz + d

az + b , a, b, c, d cz + d

and ad bc > 0}

and ad bc > 0}

is given by h h, with inverse h h. We have previously noted that the set of conformal bijections 2 2 is P GL+ ( ). 2 Remark 12. Since : 2 D 2 is a conformal bijection, the set of conformal bijections of D 2 is in bijective correspondence with the set of conformal bijections of 2 . If we let f be any conformal bijection of 2 and g be any conformal bijection of D 2 , then the bijection is given by conjugating with and 1 , that is, f f 1 and g 1 g . For this reason the set of conformal bijections of D 2 must be in one-to-one correspondence with P GL+ ( ), which is not obvious at all from the 2 description in Theorem 4.

Conjugacy Classes in M(

In this section we will work in the group P SL2 ( ), which is isomorphic to M( ). The assumption Let f P SL2 ( ) means that f is a Mbius transformation which corresponds to an equivalence class in P SL2 ( ) given by A, A SL2 ( ). Denition 9. Two elements f, g P SL2 ( ) are conjugate if there exists h P SL2 ( ) such that f = h g h1 . Remark 13. Note that conjugacy is an equivalence relation, which partitions P SL2 ( ) into disjoint equivalence classes. The equivalence class containing the identity map, id, contains no other elements, since for all g P SL2 ( ) we have g id g 1 = g g 1 = id. The following easy fact will be used again and again in the sequel: If z0 is a xed point of f then h(z0 ) is a xed point of h f h1 , and the maps f and h f h1 have the same number of xed points. We start with the following theorem. Theorem 8. Let f P SL2 ( ), f (z) = az+b . If (a + d)2 = 4, then f has two xed cz+d points in . If (a + d)2 = 4, f not equal to the identity, then f has one xed point in . 23

Proof. The map f xes if and only if c = 0. If c = 0, z is a xed point if and only if it is a root in cz 2 + (d a)z b = 0. This has two roots unless the discriminant (d a)2 + 4bc = 0. Now we have (d a)2 + 4bc = 0 a2 + d2 2ad + 4bc = 0 a2 + d2 2ad + 4bc + 4 = 4 a2 + d2 2ad + 4bc + 4(ad bc) = 4 a2 + d2 + 2ad = 4.

This implies that if c = 0, then f has a single xed point if and only if (a + d)2 = 4. If c = 0, then ad = 1 and we know that f (z) = az+b = a2 z + ba, since ad = 1. d ab This map has as a xed point and the second xed point is 1a2 . The point ab = if and only if and only if a2 = 1 or equivalently (a + d)2 = 4, since ad = 1. 1a2 If a2 = 1 then f (z) = z b, so either f (z) = z or f has as its only xed point. For a matrix A M22 ( ) we have the trace tr : M22 ( ) given by tr a b c d = a + d.

For two matrices A and B we have tr(AB) = tr(BA) tr(ABA1 ) = tr(B) tr(A) = tr(A). In particular we see that the trace only depends on the conjugacy class of a matrix. Since an equivalence class in P SL2 ( ) is represented by the matrices A, A SL2 ( ), we see that the trace is not well-dened as a function tr : P SL2 ( ) . However, when we square the trace we get a well-dened function tr 2 : P SL2 ( ) We state this as a propositon. Proposition 8.1. Let f P SL2 ( ), f = , given by tr 2 a b c d = (a + d)2 . a b . The function tr 2 : P SL2 ( ) c d .

is well-dened and depends only on the conjugacy class of f . 24

Example 8. Let f (z) = z, the matrices f = and so we have tr 2 (f ) = ( +

\ {0, 1}. Then f is represented in SL2 ( ) by 0 0

1 )2

=+

+ 2.

Denition 10. Dene the functions f P SL2 ( ) by f (z) = z = 0, 1, z + 1 = 1.

These simple maps turn out to represent all the conjugacy classes in P SL2 ( ). We state this as a theorem. Theorem 9. If f is a non-identity element in P SL2 ( ), then there exists some \ {0} such that f is conjugate to f . Proof. Suppose rst that f has only one xed point, z0 . Using cross ratios we nd a g such that g(z0 ) = . The map g f g 1 xes only g(z0 ) = , since f only has one xed point. Since is the only xed point we must have g f g 1 (z) = z + t for some t. If we let h(z) = z we see that t h g f g 1 h1 (z) = z + 1 which means that (h g) f (h g)1 (z) = z + 1, and so f is conjugate to f1 . If f has two xed points z1 and z2 , we use cross ratios to nd a map g such that g(z1 ) = 0 and g(z2 ) = . This implies that g f g 1 xes 0 and and hence is a dilation, that is, g f g 1(z) = z, \ {0, 1}. This means that f is conjugate to f , \ {0, 1}. To determine the conjugacy classes completely we must see when two maps f and f are conjugate.
1 Theorem 10. The maps f and f are conjugate if and only if = or = , i.e. tr 2 (f ) = tr 2 (f ).

Proof. First we consider the case = 1. Since f1 xes only , the map g f1 g 1 xes only g(). This implies that f1 is not conjugate to f , = 1, since f , = 1 xes both 0 and . 25

Now we let = 1 and = 1. Suppose that f and f are conjugate, then 1 1 tr 2 (f ) = tr 2 (f ). This implies + + 2 = + + 2, which implies that = or 1 = . On the other hand, if T (z) = 1 we have z T f T 1 (z) = which means that f is conjugate to f 1 .

1 1 z = f 1 (z) 1 = z

Corollary 8.1. Two non-identity elements f, g P SL2 ( ) are conjugate if and only if tr 2 (f ) = tr 2 (g). Proof. We already know that if f and g are conjugate, then tr 2 (f ) = tr 2 (g). Now assume tr 2 (f ) = tr 2 (g) and pick representatives f and f in the conjugacy classes of f and g. This means that there exists h1 , h2 P SL2 ( ) such that f = h1 f h1 and f = h2 g h1 . Since tr 2 (f ) = tr 2 (g) we have tr 2 (f ) = tr 2 (f ), 1 2 and so by the previous theorem f and f are conjugate. Now f is conjugate to f , f is conjugate to f and f is conjugate to g, which by transitivity of the relation implies that f is conjugate to g. Denition 11. Let f P SL2 ( ). Then f is conjugate to f and f 1 for some 1 \ {0}. The pair (, ) is called the multiplier of f . Remark 14. It follows from the theorem that two maps are conjugate if and only if they have the same multiplier. The multiplier and the tr 2 are related by the fact 1 that (, ) are roots in the equation z 2 + (2 tr 2 (f ))z + 1 = 0.

Geometric Classication of Conjugacy Classes

In this section we will study the limiting behaviour of representatives in the dif1 ferent conjugacy classes. The behaviour is determined by the multiplier, (, ) 1 or equivalently by tr 2 (f ) = + + 2, since f is conjugate to f if and only if 1 2 2 tr (f ) = tr (f ) = + + 2. Let f P SL2 ( ) and assume rst that tr 2 (f ) = 4. This is equivalent to the multiplier of f being (1, 1) or = 1 and hence f is conjugate to f1 . Such an f has only one xed point, z0 . We let g P SL2 ( ) be a map such that g(z0 ) = . Then the conjugate h = g f g 1 xes only . That means h is a translation and since there is only one xed point, we have h(z) = z + k, k = 0. It follows that hn (z) = z + nk. We have
n

lim hn (z) = lim z + nk =

26

for all z

. This implies
n

lim hn (z) = lim g f n g 1(z)

which implies
n

lim g 1 hn (z) = lim g 1 g f n g 1(z).

By composing with g we get

lim g 1 hn g(z) = lim f n g 1 g(z) = lim f n (z).

Since limn hn (z) = limn z + nk = for all z . This implies

we have limn hn g(z) =

lim f n (z) = g 1() = z0

for all z . Hence we see that by applying f repeatedly, all points in are moved towards the xed point, z0 . A map with = 1 is called parabolic. Next we consider the case where tr 2 (f ) = 4. In this case f is conjugate to f , = 0, 1, and f has two xed points, z1 and z2 . n If || = 1 then = ei . This means that the limit of f does not exist and hence neither does the limit of f n . Such a map is called elliptic. There are two case remaining: || < 1 and || > 1. Using cross ratios we nd a map g that takes z1 to 0 and z2 to . Then g f g 1 \ {0, 1}. We clearly have xes 0 and and hence g f g 1 = f for n f (z) = n z. n If || < 1 we see that limn f (z) = 0 for all z = . Similarly to above this implies that
n

lim f n (z) = g 1 (0) = z1 for all z = z2

. so we see that
.

n If || > 1 we have limn f (z) = for all z = 0 n

lim f n (z) = g 1 () = z2 for all z = z1

So this means that f progessively moves a point z = z1 , z2 away from one of these xed points and towards the other one. A map with || = 1 and is real and positive is called hyperbolic. Otherwise it is called loxodromic.

27

1 Since tr 2 (f ) = + + 2, we have the following classication of the limiting behaviour of a Mbius transformation, f in terms of the tr 2 (f ). We formulate this in the following table:

f f f f

is is is is

parabolic if and only if tr 2 (f ) = 4 hyperbolic if and only if tr 2 (f ) > 4 loxodromic if and only if tr 2 (f ) < 0 or tr 2 (f ) / 2 elliptic if and only if 0 tr (f ) < 4.

If f is elliptic f is conjugate to f , || = 1. For a unit complex number = 1 we 1 1 have + and satises 2 + < 2, which implies that f is elliptic if and 2 only if 0 tr (f ) < 4.

10

Mbius Transformations of Finite Period

Certain maps f M( ) satisfy f m (z) = z. These maps generate nite subgroups of M( ) and we wish to classify which maps have this property. We begin with a denition. Denition 12. An element f P SL2 ( ) has period m if m is the smallest integer such that f m (z) = z. If no such integer m exists, f has innite period. Theorem 11. If f P SL2 ( ) is a non-identity element with nite period, then f is elliptic.
n Proof. First note that f is conjugate to f for some , so f n is conjugate to f . Since we assume that f has nite period, say m, we see that f also has nite period, m. n n We know that f1 (z) = z + n, so f1 has innite period and we conclude that f is not n conjugate to f1 . This implies that = 1 and hence f (z) = n z. Since f also has m m period m we see that z = z and hence = 1. Hence we see that f is elliptic.

Remark 15. It is not true that all elliptic maps have nite period. If we put = ei we see that f is elliptic if and only if is not an integer multiple of 2 and that it has nite period if and only if is a rational (but not integer) multiple of 2. If is irrational, f is an elliptic map which does not have nite period.

11

Rotations of

In this section we will study certain Mbius transformations which are rotations. We start by dening rotation in 3 and then explain what we mean by a rotation

28

of . We will show that rotations of are Mbius transformations, and will classify which Mbius transformations are rotations. Denition 13. A rotation of S 2 is a linear mapwith positive determinant that maps S 2 to itself. Remark 16. By a theorem of Euler, such a rotation always has an axis that is xed. By choosing a suitable orthonormal basis, with the vector that is xed as a basis vector, of 3 , such a map is given by the matrix cos() sin() 0 sin() cos() 0 0 0 1

This means that in the orthogonal complement to the vector that is xed, the map is given by an element of SO(2). A rotation of S 2 extends to a rotation of 3 by linearity. The set of rotations of 3 is denoted by SO(3) and forms a group under matrix multiplication. One can show that rotations are conformal (angle preserving) maps of 3 . We start with the following basic denition.

Denition 14. A map f : is called a rotation of f SP : S 2 S 2 is a rotation, i.e. an element of SO(3).

if the map SP 1

Remark 17. The set of rotations of is denoted Rot( ) and is a group under composition. Since both stereographic projection and the rotation SP 1 f SP : S 2 S 2 are conformal, we see that f has to be conformal as well, i.e. f is an element of M( ) or P SL2 ( ), since the Mbius transformations are exactly the conformal maps of . In particular, Rot( ) M( ) is a subgroup, which can be identied with a subgroup of P SL2 ( ) in the usual way. In the following we will describe this subgroup concretely. If P = (u, v, w) S 2 then we call P = (u, v, w) S 2 the antipodal point of P . If z = SP ((u, v, w)) C the antipodal point of z, z is given by z = SP ((u, v, w)). It follows from the formulas for SP that the antipodal point of z is given by 1 . Note that if f Rot( ) then an antipodal pair (z, z ) z Since rotating S 2 and then apply the is mapped to an antipodal pair (f (z), f (z)) antipodal map is the same as applying the antipodal map and then rotating, we have f () = f (z), which means for a rotation f we have z f 1 z = f (z) 1 .

This formula is important for the proof of the next theorem. 29

Theorem 12. The group of rotations of Rot( Proof. Let f Rot( have
)

is given by = P SU2 ( ).

az + b | aa + bb = 1 bz + a f (z) =
az+b cz+d

),

and ad bc = 1. By the above formula we = 1 .

1 +b z 1 c z +d

az+b cz+d

which implies cz d bz a = . dz c az + b Comparing these two matrices in P SL2 ( ) we get b a d c = c d a b

If the factor is 1 we get b = c and a = d, which means that 1 = ad bc = aa bb = (|a|2 + |b|2 ) < 0, which is clearly impossible. If the factor is +1 we az+b have b = c and a = d and we get f (z) = b+a , aa + bb = 1. These maps form a subgroup of P SL2 ( ), which we denote by P SU2 ( ). Hence we have shown that Rot( ) P SU2 ( ). az+b For the other inclusion, we rst noice that if f (z) = b+a , aa + bb = 1 xes 0, then b = 0. This implies that aa = 1 and hence f (z) = az = a2 z, |a2 | = 1. So f a az+b is a rotation by an angle , where a2 = ei . Now let f (z) = b+a , aa + bb = 1 be any transformation and let f (z0 ) = 0. We let z0 correspond to P S 2 and as usual 0 corresponds to S. Then there exists a rotation of S 2 , R, with R(P ) = S. The composit SP R SP 1 = R1 takes z0 to 0, and is by denition a rotation of . 1 The map f R1 xes 0 and since P SU2 ( ) is a group, is an element of P SU2( ). By the previous argument, an element of P SU2 ( ) that xes 0 is a rotation, so 1 f R1 (z) = ei z = R2 (z), for some . Now we have f = R2 R1 Rot( ), which shows that P SU2 ( ) Rot( ). Corollary 11.1. There is a group isomorphism P SU2 ( ) SO(3). = Proof. Every element in Rot( ) corresponds to a rotation of S 2 , which extends to a rotation of 3 . This means that Rot( ) SO(3). The statement now follows = from the previous theorem.

30

12

Finite Groups of Mbius Transformations

In this section we want to classify all nite subgroups of P SL2 ( ). The presentation will be a bit less detailed then the previous sections. Some algebraic arguments needed to complete the proofs of the theorems are beyond the scope of these notes, and will not be presented. Recall that f is elliptic if 0 tr 2 (f ) < 4 and that by Theorem 11 an element of nite order is elliptic. Hence the nite subgroups of P SL2 ( ) must consist of elliptic elements and the identity. Recall that two subgroups 1 , 2 P SL2 ( ) are conjugate in P SL2 ( ) if there exists a g P SL2 ( ) such that 1 = gg 1. This means that the elements in 1 are simultaneously conjugate (conjugate by the same element) to the elements in 2 . The next theorem says that a subgroup of elliptic transformations together with the identity is conjugate to a subgroup of rotations. Theorem 13. Let be a subgroup of P SL2 ( ) consisting of elliptic elements together with the identity. Then is conjugate in P SL2 ( ) to a subgroup of P SU2 ( ). Sketch of Proof. Let be a subgroup of P SL2 ( ) consisting of elliptic elements together with the identity, and let be the image of in SL2 ( ). Step 1: Show that by conjugating we may assume that contains the element 0 , where || = 1. S= 0 a b Step 2: Consider an arbitrary element of , T = . By calculating tr 2 (ST ) c d one shows that a = d. Step 3: By calculating ST S 1T 1 one shows that b = 0 if and only if c = 0. Step 4: One shows that there exists an r {0} depending on , but not T such that c = rb. Put v = 4 |r| and dene a transformation V = v 0 SL2 ( ). Now we have 0 v 1

V T V 1 =

a v2b v2b a

r > 0,

( )

a v2b v 2 b a

r < 0. ( )

This implies that by replacing by V V 1 we may assume that the elements in 31

are of the form ( ) or ( ). Since SU(2) =

a b | aa + bb = 1 it is enough b a to show that the elements of are of the form ( ). We also see that V SV 1 = S, by V V 1 , still contains S. even after replacing a b Step 5: Consider an element = of type ( ) in . Since S, T the b a product ST S 1 T 1 is an element of . One now checks that tr 2 (ST S 1 T 1 ) > 4, so ST S 1T 1 is neither the identity, nor elliptic. This is a contradiction and hence consists of elements of type ( ), which means that is conjugate to a subgroup of P SU2 ( ).

Corollary 12.1. Every nite group of Mbius transformations is conjugate to a group of rotations of . Proof. As previously noted, by Theorem 11 an element of nite order is elliptic. Hence the nite subgroups of P SL2 ( ) must consist of elliptic transformations and the identity. By the previous theorem, such a subgroup is conjugate in P SL2 ( ) to a subgroup of rotations. The proof of the following theorem relies on some results in group theorey, and will be omitted. Theorem 14. Let be a nite group of rotations in holds. 1. is cyclic, 2. is dihedral, 3. is the symmetry group of a regular tetrahedron (A4 ), octahedron (S4 ) or icosahedron (A5 ). One can show that two nite subgroups in P SL2 ( ) are conjugate if and only if they are isomorphic. Combining the previous two theorems yields the following classication result. Corollary 12.2. Every nite subgroup of P SL2 ( ) is cyclic, dihedral, or isomorphic to A4 , S4 or A5 . Proof. By Theorem 13 and the remark before it, every nite subgroup of rotations, i.e. a subgroup of P SU2 ( ). The corollary now follows from Theorem 14 and the fact that two subgroups of P SL2 ( ) are conjugate if and only if they are isomorphic. 32
.

Then one of the following

13

Bibliography

1. Churchill, Ruel V. and Brown, James Ward Complex variables and applications McGraw-Hill Book Co. New York 1984. 2. Conway, John B. Functions of one complex variable Graduate Texts in Mathematics Springer-Verlag, New York 1978. 3. Jones, Gareth A. and Singerman, David Complex functions - An algebraic and geometric viewpoint Cambridge University Press, Cambridge 1987. 4. Sa, Edward B. and Snider, Arthur David Fundamentals of complex analysis with applications to engineering, Science, and Mathematics Prentice Hall. 2003.

Comments and corrections are much appreciated. Please send them to: [email protected].

33

14

Problems

Problem 1. Show that a line ax + by = c/2 can be written as Az + Az = c, c real. Find the equation for the line through i and 1 + 2i. Problem 2. Find a formula for the reection in the line Az + Az = c, |A| = 1. What is the formula for the reection in the line x + y = 1? Problem 3. Find the formula for the circle {z | |z C| = r} in terms of x and y, z = x + iy. Find the equation for the circle of radius 2 and center i. Problem 4. Write down the formula for the direct ane transformation given by: 1. Translation in the direction (2, 3). 2. Rotation about (0, 1) through /4. Problem 5. Let SP be the stereographic projection from the north pole. Let P = (u, v, w) and z = x + iy. Let SP (P ) = z. Derive the formulas for u, v, w in terms of x, y (that is, ll in the details in the proof of Proposition 3.1). Problem 6. Show that the stereographic projection preserves angles. That is, for inclined at an angle at the origin (i.e. p = 0 in the formulas r and s lines in in the proof of Proposition 3.2), nd the equations for the line on S 2 and show that the angle between the tangent vectors at t = 0 is . Problem 7. A xed point for f : is a point z such that f (z) = z. Find the xed points (if any) of a dilation, translation, a general direct ane transformation and the inversion. Remember to check if innity is a xed point! Problem 8. Find the image of the horizontal line y = c, both for c = 0 and c = 0, under the inversion map. Problem 9. If the unit circle is oriented counter-clockwise, determine the orientation of the image of the unit circle under the inversion map. What is the image of the center of the circle? Problem 10. Determine the image of the disk {z map. Interpret this in terms of inversion on S 2 . | |z| < 1} under the inversion

Problem 11. Find a linear transformation that takes the unit circle {z | |z| = 1} onto the circle {w | |w 5| = 3} and maps the point z = i to the point w = 2. Note the word linear! 34

Problem 12. Show that a Mbius transformation f can have at most two xed points in the complex plane unless f is the identity. Problem 13. Discuss the image of the circle |z 2| = 1 and its interior under the maps 1. f (z) = z 2i 2. f (z) = 3iz 3. f (z) = 4. f (z) =
z2 z1 z4 z3

1 5. f (z) = z .

Problem 14. Find a Mbius transformation that maps {z right half plane {z | (z) > 0} and takes i to the origin.

| |z| < 1} onto the

Problem 15. Find the Mbius transformation that maps (0, 1, ) to the points (0, i, ), (0, 1, 2), (i, , 1) and (1, , 1). Problem 16. What is the image of the third quadrant under the map f (z) = Problem 17. What is the image of the sector {z z the map f (z) = z1 ?
z+i ? zi

| < arg(z) < } under 4 4

Problem 18. Show that a Mbius transformation has as its only xed point if and only if it is a translation, but not the identity. Show that a Mbius transformation has 0 and as its only xed point if and only if it is a dilation, but not the identity. Problem 19. Let f be a Mbius transformation with xed points z1 and z2 . If g is another Mbius transformation, show that g 1f g has xed points g 1 (z1 ) and g 1(z2 ). Problem 20. Two Mbius transformations f and g are said to commute if f g = gf . Let f be a Mbius transformation not equal to the identity. Show that a Mbius transformation g commutes with f if g and f have the same xed points. Hint: Use the statements in the previous problems. Problem 21. Show that the set of elliptic transformations in M( ) together with the identity form a subgroup of M( ) under composition of maps.

35

Problem 22. Consider the Mbius transformation f (z) = 1. Determine the xed points of f . 2. Calculate tr 2 (f ) and the multiplier of f .

z2 . z1

3. Find a representative (not f itself) of the conjugacy class of f . 4. Describe the limiting behavior of f , limn f n (z) for all z Problem 23. Consider the Mbius transformation f (z) = 1. Determine the xed points of f . 2. Calculate tr 2 (f ) and the multiplier of f . 3. Find a representative (not f itself) of the conjugacy class of f . 4. Describe the limiting behavior of f , limn f n (z) for all z
. iz . zi .

Problem 24. By using the formulas for stereographic projection, check that the antipodal point to z is given by 1 . z Problem 25. Show that the set of rotations of of maps.

form a group under composition

Problem 26. Show that P SU2 ( ) form a subgroup of P SL2 ( ). Problem 27. Let (z0 , z1 , z2 , z3 ) be four distinct points in . Show that there are precisely two values of k such that (z0 , z1 , z2 , z3 ) can be mapped to (1, 1, k, k) by a Mbius transformation. Problem 28. Show that a rotation of P SL2 ( ).

is represented by an elliptic element

Problem 29. Show that an elliptic element in P SL2 ( ) whose xed points are antipodal points in is a rotation of . Problem 30. Let f and g be Mbius transformations with a common xed point z0 . Show that the Mbius transformation f g f 1 g 1 is either parabolic or the identity. (Hint: Assume z0 = . Why is this method valid?)

36