HYPERBOLIC GEOMETRY MA 448 Caroline Series With assistance from Sara Maloni Figures by Sara Maloni and Khadija Paroog 2008; Revised 2010 Last edited: January 4, 2013Contents Bibliography iv Introduction vi 0.1 Introduction vi 0.2 Models of hyperbolic geometry 0.3 Other models of hyperbolic geometry. rn 0.4 Comparison between Euclidean, spherical and hyperbolic go- ometry xiii 0.5 A Potted History sill 1 Linear Fractional Transformations 1 1.1 Linear fractional transformations . 1 11.1 Cross-ratio and transitivity of Aut(€) 4 1.2 Automorphisms of the unit disk D 8 1.3 Automorphisms of the upper half plane . . . 9 2 Basics of hyperbolic geometry 12 2.1 Motries, goodesies and isometries. 12 2.1.1 Geodesies and isometries 13, 2.2. Similarities and differences between hyperbolic and Euclidean geometry 23 221 Circles ce 28 222 Angle sum ofa triangle... .. . ovens 2 2.2.3 Perpendicular projection and the common perpendicular. 25 2.24 Lines meeting on OH 7 2.2.5 Points equidistant from a given line 29 2.2.6 Horocycles . . 30 2.2.7 Angle of parallelisin- 31CONTENTS ii 2.2.8 Schweikart’s constant... ... . ee 32 2.2.9 Pythagoras’ theorem 33 2.2.10 Area of a triangle 34 2.2.11 Congruent triangles 37 2.2.12 Polygons 38 2.2.18 Hyperbolic trigonometry ce 39 3 Isometries of D and HL 3.1. Classification of elements of Aut(C). 3.1.1 Classification by fixed points and trace 3.1.2 Coaxial circles . 3.13 Dynamics and Coa . 3.14 Relation of multiplier to trace 3.1.5 Isometric circles... 22... 3.2 Isometries of Hand D 321 cee 3.2.2 Hyperbolic transformations ©... .......... bl 3.2.3 Elliptic transformations... . Sees 55 3.24 Orientation Reversing Isometries 57 3.2.5 Isometries as product. of reflections, 58 4 Groups of isometries 60 4.1 Subgroups of Isom*(H) 2. ee 60 42 Relation of properly discontinuity and discreteness 68 4.2.1 More properties of discrete groups 7 422 General tests for discreteness 74 5 Fundamental domains 76 5.1 Fundamental domains 76 5.2 Existenee of a fundamental domain: Dirichlet and Ford fun- damental domains 5.2.1 Dirichlet domains . . snes 5.2.2. Ford domains 5.3. Examples of fundamental domains 5.3.1 Dirichlet domains for the elementary groups 5.3.2 Dirichlet domains for the action of Z? on R? .... . . 5.3.3 Dirichlet domains for the action of SL(2,Z) on H.. . 83 5.3.4 Ford domains for the action of SL(2,Z) on H 8 S888CONTENTS 6 Side pairing and the cycle condition 6.1 Side pairings 6.1.1 Relations between the hood of a vertex of R 6.1.2. Reading off the relations mechanically. 6.1.8 How do we know when to stop? 6.14 Examples 6.1.5 Vertex eyele at oo 6.1.6 Poinearé’s Theorem . ide pairings: 7 Hyperbolic structures on surfaces 7.1 Hyperbolic structures on surfaces 7.2 Cusps, funnels and cone points 7.3. Incomplete hyperbolic structures 74 Covering spaces... « 741 Fundamental group . : 74.2 The Universal Covering Space . . 74.3 The developing map ....... 7.5 Proof of the Uniformisation Theorem 7.6 Proof of Poine: 's theorem, 8 The limit set 81 Thelimitset 2... ........005 iii 85 85 the neighbour- o1 94 96 103, 107 107 110 112, 115, . 116 bees LT bees 118, 119 123, 127 cece eee 127Bibliography fF] denotes books closest to this course. [1] L. Ablfors Complex analysis, MeGraw Hill (1953) [2] D. Mumford, C. Series and D. Wright Indra’s Pearls: The vision of Felix Klein, Cambridge University Press (2002) {3} G. Jones and D. Singerman Compler functions, Cambridge University Press (1987). [4] J. Milnor Dynamics in one complex variable, Princeton University Press (1931), [5] J. Stillwell Mathematics and its history, Springer-Verlag (1989) [6] *J. Anderson Hyperbolic Geometry, Spr ager-Verlag (2nd edition 2005) [7] *S. Katok Fuchsian groups, University of Chicago Press (1992). [8] *S. Stahl The Poincaré half-plane; a gateway to modern geometry, Jones and Bartlett Publishers (1993) {9] M. Reid and B. Szendroi Geometry and Topology, Cambridge Univ. Press (2005). [10] A. F. Beardon The Geometry of Discrete Groups, Springer-Verlag (1983). [11] M. do Carmo. Riemannian Geometry, Birkhinser (1992). [12] L. R. Ford Automorphic functions, Chelsea Publishing Company (1929)BIBLIOGRAPHY v [13] J. Lehner Discontinuous Groups and Automorphic Functions, American Mathematical Society (1964) [14] W. P. Thurston, edited by Silvio Levy Three-dimensional geometry and topology, Princeton Univ. Press (1997). [15] W. P. ‘Thurston The geometry and topology of three-manifolds, Princeton Uni Mathematies Department. (1979), lecture notes. (16) F.Bonahon Low dimensional Geometry, New book shortly to appear; available at ntvp://almaak.usc.edu/ tbonahon/Research/Preprints/Preprints. html [17] D. B.A. Epstein Geometric structures on Manifolds, The Mathematical Intelligencer 14, Number 1, (1982), 5-16. [18] D. Hilbert and S. Cohn-Vossen Geometry and the Imagination, Chelsea. Publishing Company, (1952). [19] M. A. Armstrong Basic Topology, Springer-Verlag, (1983). [20] A. Hatcher Algebraic Topology, Cambridge Univ. Press, (2002)Introduction 0.1 Introduction What is hyperbolic geometry and why study it? Spaces of constant curvature Hyperbolic (also called non-Euclidean) go- ometiry is the study of geometry on spaces of constant negative curva- ture. In dimension 2, surfaces of constant curvature are distinguished by whether their curvature Kis positive, zero or negative. Given such a surface in R', if K > 0 all points of such a surface X are on one side of the tangent plane any the point; if K = 0 then there is always a line in X contained in the tangent plane; and if K <0 there are points of nected X on both sides of any tangent plane. If the space is simply cor and if K > 0, then it is a sphere (with K = I/radins); if K = 0 it is the Euclidean plane; and if K < 0 it is the hyperbolic plane, also called 2-dimensional hyperbolic space. Such surfaces look the same at every point and in every dire ion and so ought to have lots of symmetries ‘The goometry of the sphere and the plane are familiar; hyperbolic ge- ometry is the geometry of the third case. Hyperbolic space has many interesting features; some ari those of Euclidean geometry but some are quite different. In parti ular it has a very rich group of isometries, allowing a huge variety of crystallographic symmetry patterns. ‘This makes the geometry both rigid and flexible at the same time. Its properties and symmetries are closely related to tree-like growth patterns and fractals, suggestive of many natural biological objects like ferns and trees.INTRODUCTION, vii Historical importance Another way to deseribe hyperbolic geometry is to say that is the geometry of space in which Euclid’s parallel axiom fails. One way to state the parallel axiom is that for every line L, and point P not on L, there is a unique line L’ through P which does not meet L, even if extended infinitely far in both directions. Historically, hyperbolic geometry had enormous importance. Attempts to deduce the parallel axiom from Euelid’s other axioms led to many developments within Euclidean geometry. Realisation slowly dawned that the reason that: all attempts to deduce the parallel axiom failed, was that one could develop a consistent system of geometry which was not Euclidean, in which the parallel axiom failed while the other axioms remained true. This had enormous scientific and philosophical importance because it showed that mathematically, there was nothing absolute about Euclidean space. Hyperbolic geometry and surfaces Any topological surface can be dowed with a geometric structure. This means that one can find a metric on the surface which in small regions looks like one of the three above types of geometry, and such that the ‘overlap maps’ (c.. the definition of a manifold to understand this) are isometries of the ap- propriate geometry. Thus a cylinder or a torus carries a Euclidean structure while, as we shall see later in the course, every surface of neg- ative Euler characteristic (in particular every closed surface of genus at least 2) carries a hyperbolic structure. In fact, in all cases except the sphere, the surfaces in question carry not one but many such metrics, Being able to put a hyperbolic metric on a surface allows one to study many features of the surface very precisely. This applies especially to the study of diffeomorphisms on surfaces, where hyperbolic geometry plays a crucial role. Revolutionary work by William Thurston in the 1980's opened up the possiblity of a similar description of 3-manifolds. He showed that many 34manifolds are endowed with a natural geometrical structures of one of a fow kinds, of which by far the most common is hyperbolic. He conjectured that any topological 3-manifold could be cut into pieces on the basis of topological information alone, such that each piece carries one of 8 special geometries, among them hyperbolic. ‘This geometrisa- tion conjecture has now almost certainly been proved in consequenINTRODUCTION, viii of the work of Perelman. ‘Thurston's insights led to a great revolution in the study of hyperbolic 3-manifolds, making it a large and very active area of mathematics in which spectacular progress has been made in recent years. Besides ‘Thurston, some of the leading names are: Agol, Bonahon, Brock, C: nary, Gabai, Masur, Minsky, and Otal. Connection with other parts of mathematics Hyperbolic geometry is closely connected to many other parts of mathematics. Here are some: differontial geomotry, complex analysis, topology, dynamical systems ing complex dynamics and ergodic theory, relativity, Diophantine approximation and number theory, geometric group theory, Riemann surfaces, Teichmiiller theory. In this course we will study two closely related models of hyperbolic geometry, that: is, spaces with constant curvature —1. These are the upper half plane Hl = {2 € C : Sz > 1} with the metrie ds = |d2|/Sz and the 2\d2| unit disk D = {2 € C : |z| < 1} with the metrie ds = awh (Here = Vir + dy) Exercise 0.1. From differential geometry, the curvature K of a Riemann surface with metric ds? = dx} + gadr3 is given by the formula 0, 0. 5 (2G) (ee) Verify that the curvature of the above metrics in both HL and D is ~1 0.2 Models of hyperbolic geometry Unlike the situation in spherical geometry, it is impossible to embed an finite simply connected surface of constant negative curvature isometrically a theorem of Hilbert, proved in 1901)? es up more and more as you go towards its ‘edge. Thus to visualize hyperbolic geometry, we have to resort to a model. “Peshaps this is the reason forthe historically late development of hyperbolic geometry.INTRODUCTION ix Figure 1: A crinkled kale Figure 2: Stereographie leaf projection of the sphere to the plane You can think of this as similar to the projection of geometry on the surface of a sphere by stereographic projection (or other map projection) conto the plane. The projection introduces distortion so that either distance or angles measnrements, or both, do not look correct. Like stereographic projection, the two models we will study preserve angles (are conformal) but massively distort distance. We are going to study two such projections or models which are in fact almost equivalent: the upper half plane model Hl = {2 € © : Sz > 1} with the metric ds = |dz|/Sz (also known as the Lobacheysky plane) and the unit 22] ds = (=P) 2€C:|2| <1} with the metric (also known as the Poincaré disk). In both models: Lines are ares of circles (or lines) which meet JD or OHI orthogonally. Notice that in H such ares must be semicireles with centres on R; in ID such, an are never has its centre on OD. Angles The usual Bueli Jean angle. Distance ‘The infinitesimal form of the metric on H is ds = |dz|/3e and on Dis ds = 2\dz|/(1 — |2P). ‘The actual formula for distance between two points can be expressed in terms of cross-ratio. If 2, 22, 24, 24 are distinct points in C then their cross-ratio is defined as E Ar” u—2 +225 2a, Za] = antINTRODUCTION x Now let P,Q be two points in D or H. You can prove by Euclidean geometry that there is a unique (hyperbolic) line joining P to Q. Let P,Q be the points where the extension of this line meets 41D or El, so that the order along the line is P', P,Q,Q!. Th do( P,Q) = logiP’,Q, P,Q] and de(P,Q) = loxiP’,Q, P,Q] Isometries Poincaré (1881) realised that the isometries or distance pre- serving bijections of these two models are exactly the linear fractional 7 az+b transformations or Mobius maps 2 -» . ” (where a,b,c,d € C) which preserve D or Hi respectively. ‘This makes it particularly easy to study and compute with such maps. It turns out that they are also the set of all conformal automorphisms of D or H. We shall review linear fractional transformations in the next. chapter. Figure 3: A tiling of the hyperbolic plane, made by hyperbolically reflecting in the sides of the white triangle whose sides are marked with red, blue and green dots. 0.3. Other models of hyperbolic geometry Here we just give very brief descriptions. Good referenees for this are [9, 14] The hemisphere model ‘Think of the uit disk D in R as contained in the ‘horizontal’ plano r9 = 0. Let $? denote the unit sphere in R*. Stereo- graphically project D from the south pole (~1,0,0) onto the northernINTRODUCTION, xi hemisphere of S*. ‘This transfers the geometry of D to geometry on the hemisphere. The map is conformal or angle preserving (because steroegraphic projection can be deseribed as product of inversions) Geodesies are ‘vertical’ se and orthogonal to AD. nicircles with centres on the lane ry = The hyperboloid model In this model, hyperbolic space is geometry on a ‘sphere of radius. Consider IR® with the indefinite Lorentz metric ds? = —daj, + dx} + dr}, associated to the quadratic form Q( (9,21, ¢2)) = length of a vector x = (#9, 21,29) is Q(x). aft rt +23, The ‘The sphere of radius 4 about the origin is the set, H = {(x0, 01,42) : Q(x) = -1}, Restricted to H,, the metric ds? becomes a positive definite Riemanian you can check that any tangent vector to H has positive length. x € H so that Q(x,x) = -1. Differentiating along any path x(t) € H. gives Q(x,x) = 0. In other words, a vector X in the tangent space at x is in the orthogonal complement of x with respect to Q, and this orthogonal complement has signature +2.) H is a hyperboloid with two components depending on the sign of Hyperbolic space is modelled on one sheet of this hyperboloid, usnally the upper sheet H™ where 1 > 0. (Thinking of the direction (1,0,0) as ‘vertical’.) ‘The hyperbolic distance between two points x,y is given by cosh d(x, y) = —@Q(x,y). Geodesies in H* are exactly the intersection of planes through the origin with H'. There is a nice explanation of all this in [9], starting with the easier case of ‘one dimensional hyperbolic space’ and the Lorenz metric on R?. ‘The set of linear maps which preserve Q, (ie maps A : R* —- IRS such that Q(Ax) = Q(x) for all x € R°) is called O(2,1). The group of (not necessarily orientation preserving) isometries of H' is the index two subgroup of O(2,1) of maps which carry H* to called group. To see why the Minkowski model agrees with the previous ones, one way is to compare the isometries, see eg [14]. Lorentz,INTRODUCTION, xii The Klein model ‘The Klein model (first however introduced by Beltrami!) is also a model in the disk, but now geodesics are Euclidean straight lines. It can be obtained from the hyperboloid model as follows. Any line in R* through the origin O intersects H* in at most one pd ‘The line also intersects the horizontal plane x» = 1 in one point. This defines a projection from H+ to the unit disk. Since a geodesic in H* is the intersection of H+ with a plane through O, the goodesie projects to a straight line in the Klein disk K. Thus geodesics in K are straight lines. Isometries of H* (clements of the Lorenz group) also induce isometries of A. This means that: geodesics in KX look straight, while on the other hand angles are distorted If K is placed as the equatorial disk inside the unit sphere in IR, then Euclidean orthogonal projection in the direction of the vertical veetor (1,0,0) sends K to the upper hemisphere, providing a map from K to the hemisphere model above.INTRODUCTION, xiii 0.4 Comparison between Euclidean, spheri- cal and hyperbolic geometry Euclidean | Spherical | Hyperbolic Basic givens | Points, lines, planes Model Lines DorH. ares orthogonal to boundary Euclidean plane | Surface of sphere Euclidean lines | arcs of great cles Axioms. For ex- ample: Any two distinet, points lie on a unique line Parallel lines | through any PL there is a unique line not meeting L through P ¢ L there is no line not meeting L through Pg L there are in- finitely many lines not meet- ing L >a <0 Angle sum of tri- angle triangles with equal angles congruent: Similar triangles | equal angles doesn’t imply congruence nr triangles — with equal angles congruent Circumference 2a sinr Qn sinhr of circle Area of circle An sin? r/2 4x sinh? r/2 0.5 A Potted History Thales, c. 600 BC Started the formalised study of geometry. Buclid, c. 300 Bc The axiomatic formulation of geometry. ‘Givens’ were points, lines, straight Lines, right angles ete. Postulates:INTRODUCTION, xiv Attempts to prove the parallel postulate Starting from Greck tim . Any two points lie on a unique line. Any straight line can be continued indefinitely in cither direction. . You can draw a circle of any centre and any radius. . All right angles are equal. The parallel postulate: If a straight line, crossing another two straight lines L,I’, makes angles a, 6 with L, [/ on o ifa+ 2a -5 in which case az-+b = —“" 4b 40. 4 : c So we can define f(— Remark 1.2. To be rigourous, you have to replace 2 by L/w in a neighbour hood of oo and to study the behaviour as w — 0.CHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 2 Such f is invertible, hence it is a bijection € — ©. One easily calculates that the inverse map is given by dz I= +a" We can represent the coefficients of a Mébius map by the matrix A = D (: ‘) € GL(2,C). It is casy to check that you can compose Mabius maps ¢ by multiplying the matrices. Observe that for any \-£0, the maps az+b Aaz + Ab ae and 2s SET ard while (7 ” = ») vith £0. c d\o x) \ve ag, To fix the ambiguity, we can require that ad — be = 1 which is achieved by dividing all coefficients by Vdet A = ad — be. ‘This still gives an ambiguity of multiplication by +1. Notice that to invert the map f(2) above you don’t need to divide all coefficients by Vdet A = Vad — bc, as you would to invert A Definition 1.3. A map f : C — C is called conformal or angle pre- serving if, whenever smooth (C1) eurves Cy and C, on € meet at a point P with (signed) angle 0 (measured from C, to C2), then f (C,) and f (Cz) meet at f(P) with the same angle 0 from f (Ci) to f (Ce). are the same function on In particular, a conformal map is necessarily orientation preserving. Definition 1.4. We denote the set of all conformal bijections © — € by Ant(C) Proposition 1.5. Mébius maps are conformal or angle preserving. Proof. This is true of any complex analytic map at any point. at which J") £0. (Locally, the map expands by |f"(zo)| and rotates by Arg J"(2o)-) ad — be —— except at 2 = 00. By (rap * 9% y writing in terms of w = 1/z, you ean also verify that f is conformal at ox. a By computing, we see that f'(z) =CHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 3 Exercise 1.6. Complete the missing cases in the above proposition: Suppose that f is a Mobius map for which f(co) # co. Let w = 1fz and F(w) = (1/2). Show that F'(w) = —1/(c + dw)? and hence that F'(0) # 0 unless ©=0. Deal with this last ease by considering G(w) = 1/F(w). Thus any Mébius map belongs to Aut(C). Clearly Aut(C) is a group under composition, The results above ean be formalized a Proposition 1.7. The map F's GL(2,C) —+ Aut(@) defined by ra) (no-t) is a homomorphism with kernel KerP=t(* °) yee}. OX Proof. Saying that F is an homomorphism just means that we can compose Mébins maps by multiplying matrices. (Ex Check this.) To find the kernel: aztb +d conformal bijection of C ean be repre Theorem 1.8. The map F : GL(2,C) — Aut(C) defined above is a surjec- tion. Proof. ‘This requires Liouville’s theorem and other results from complex anal- ysis, sce Example sheet 1, [3] (p. 9) or [4 (p. 5). a Usually, to avoid the ambiguities in of representation by matrices, we normalise Mabius maps so their coefficients are in SL@,C) = {A € GL(2,C) : det(A) = 1} As above, this can be achieved by dividing all coefficients by /det(A). This still leaves the ambiguity of multiplication by £1. Defining PSL(2,C) = SL, C)/ + I we haveCHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 4 Corollary 1.9 (Normalisation). (1) Any T € Aut(©) can be represented by @ matrix (: ‘) €SL(2,C). (2) Aut(C) © PSL(2,C). Proof. (1) By ‘Theorem 1.8, any T’ € Aut(€) can be represented by M = 75) © GLea.c). Let A = det(M) = ad — be. ‘The matrices ob 4 and Jz M = (@ *) have the same image under F in Aut(€), and ve Va det( 3h) = 1. (2) The obvious map SL(2,C) — Aut(@) has kernel $L(2,C) 9 KerF* AL a 1.1.1 Cross-ratio and transitivity of Aut(C) The group Aut(©) inchudes some especially simple transformations: (1) translations z+» 2 +t where t € C: (2) rotations around a € C: z+ €(2—a) +4; (8) similarities z+ dz with A > 0. (‘This is an expansion if ) > Land a contraction if A < 10.) These transformations all fix 00. It also includes: (4) inversion 2 1/z which interchanges 0 and oo. Exercise 1.10. By writing te) 0th a T@)= Syd aet show that any T' € Aut(C) is @ composition of the above four types. Definition 1.11. A group G is said to act transitively on a space X if, given two points 9 and x, in X, there exists g < G such that g(r) =m. It acts freely if for all ry €X, gro) = x9 implies g = id for all xy < X. Exercise 1.12. Use the above transformations to show that Aut(C) acts transitively but not freely on C. Now do the same for ordered pairs of distinct points (21,22)CHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 5 Proposition 1.13. Aut(C) acts transitively and freely on ordered triples of distinct points in C. More precisely, given two ordered sets of distinct points (2,245 24) and (1, wy, 0) in C3, there exists a unique T € Aut(C) such that T(z;) =, for i= 2,3,4. Proof. ¢ Existence: First, we prove existence in the case (wp, ws,w4) = zy t-% 4-H Check that Tisza.) Maps (22,23, 24) + (0,00,1). In the general case the transformation 7’ we need is obtained by composing Tiess.20) 04 Te yay)? (0,00,1). Denote by Ziz,2,.;) the transformation T(2) that is T = Tegan) ° Mensa © Uniqueness: First we prove that, given $ € Aut(C) such that S(0) = 0,S(c0) = oo and S(1) = 1, then $ = id. Let S(z) = ate If S(0) = ard a+b b/d = 0 and S(co) = a/c = co, then b= = 0. If S(I) = e then a = d. Since we may assume that det (: ) = 1, this gives ad = 1, ed so $= id. Now suppose there were two transform: such that (2,23, 24) (wa, ws, wa). Then Tag) TT e3 .2)° Twas) TT 2524) Poth fix (0,00, 1) and the result follows from the above. a ms Definition 1.14. Let 2, 2,23, 24 be an ordered set of distinct points in ©. The cross-ratio of 2, 2, %, 24 is defined by % ec. (1522529, 24] = (Why is [21, 22, 2a, 24] # 90 for any 21, 29, 25, 24?) The map Tyee,c4) USed in the proof of Proposition 1.13 can be nicely expressed using the cross-ratio: Teeayexsa) (2) = = [2.2228 24] Corollary 1.15. The cross-ratio is invariant under Aut(C), that is if T € Aut(C), then [2, 22,25, 24] = (121), (22), T (2s), T(za)]- Proof. Method 1: Let T(=) att tion substituting in the two sides. ‘The result follows by a direct calculaCHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 6 Then unique Method 2: Define $ ¢ Aut(€) by the formula S(z) = (2,22, 25,24 S sends 2 + 0,24 ++ 00,21 + 1 and by Proposition 1.13, Sis th clement of Aut() with this property. Now write w, = T(2;) for i = 2,3,4 and let W(z) = [z, wg, wy, wy] So W sends w+ 0, ug + 00, w4 + 1, hence W oT sends 22+ 0,25 00, 24+ 1, so WoT = S. Now Wo T(x) = S(z1) = (21,22, 28, 24] while on the other hand, W 0 T(21) = [2(21), 2, ws, wa) = [L(a), T (22), P(s), 2 (z)]- o Remark 1.16. This crucial property of invariance of cross-ratio will allow us to define hyperbolic distance. Notice that the Buclidean distance between 2 points is invariant under translations or rotations and that the ratio of the distances between 3 points is invariant under Euclidean similarities Lemma 1.17. Four distinct points 2,, 2, 2,21 € € are coneyelic if and only if (21, 22, 23, 24] € R. Remark 1.18. (i) Here “concyclic” includes “collinear”, a line as a cirele through oo € C. (Lines through oo are exactly the stereo graphic projections of great circles through the north pole.) (ii) Permuting the order of the points %, 22,23, 24 € C may change [21, 22,23, 24], but it will not change the fact that [2x, 22,2, 24] € R, see Brercise 1.20. since we can view Proof of Lemma 1.17. Suppose that 21, 22, 23,24 € € are concyclic, then (up to reflection) they must. be arranged in one or other of the configurations in Figure 1.1, in which the angles are found by using clementary geometry in cireles. For the configuration shown on the left: Arg (=) =-0 21 25 2). Hence Arg(2—2 a =) =-040=0 [21 22,25, 24] > 0. A-% 4% i c, 24 — 22 For the configuration on the right: Arg(: =O and Arg( = am (0). Hence from which [21,22 24,24] <0. ‘The converse is similCHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 7 Fi ure 1.1: Coneyelie points have real cross ratio. Corollary 1.19. Suppose T € Aut(C). The T maps circles and lines to circles and lines. Proof. Method 1 Lemma 1.17. Method 2: By direct calculation. ‘The equation of a circle centre a radius r is (2-a)(Z-a) = r?. Check that this can be rearranged as |2|?+Bz-+Bz+C =0 where |B|? — C > 0, and that conversely any equation in this second form n Mz —a) = Mz—a) or equivalently Bz + BZ + € = 0. Using Exercise 1.10, show you have only to ‘The result follows immediately from Corollary 1.15 and is a circle. Likewise show that a line ean be writ check the calculation for the transform ion 2 ++ =, which is easy. a Exercise 1.20 (Effect of permutation on the cross-ratio). Check that if x = [e,22.25,21], then by permuting the x we obtain exactly the siz val- wes Xs S51 Xe py Ba Exercise 1.21 (Orientation reversing (anticonformal) automorphisms of €). az+b td Prove that any such map takes [21, 22, %s, 24] to [21, 22, 23, 24] and maps circles lo circles One such map is 2+ 2. Check that the most general such map is 2+CHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 8 1.2 Automorphisms of the unit disk D Let D denote the unit disk {z € C : |2| <1} and set ‘Aut(D) = {r € Aut(@) : 7D) = v}. Notice that any 7 € Aut(C) is automatically a bijection from D to itself. We are interested in Aut(ID) because, as we shall see in the next chapter, it ean be viewed as the set of orientation preserving rigid motions of the Poincaré disk model of hyperbolic space D with the hyperbolic metric. Theorem 1.22. The set Aut(D) is the subgroup of Aut(C) of Mobius maps of the form Se) = (4) or equivalently of the form f= (1.2) Exercise 1.23. Check that functions of type (1.1) form a group. Also explain why the forms (1.1) and (1.2) are equivalent. Remark 1.24. The set ‘ab €GLQ2,C) :d=a,e=d, al? — 2 =1 ed, is called SU(1, 1) Proof. Suppose that f is of the form (1.1). We check: [| <1 <> (x—a)(2—a) < (1~a2)(1—a2) <= (1-22)(1-aa) > 0. So if |a| <1, then |f(2)| <1 <=> |2| <1. Hence f restricts to a bijective map D — D, that is f € Aut(D) Conversely, to show that any conformal automorphis (=a) of this form, first let f(2) = S09) where a = 9-"(0).CHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 9 go 40) = 0 and go f-! is a conformal automorphism. By the Schwarz Lemma from Complex Analysis, if h = go f-) : D — D and h(0) = 0, then [h(2)| < |z| and |z| = |h-1 0 A(z)| < |h(2)]. So |A(z)| = [2]. From na again, it follows that h(z) = ez for some @ € [0, 27). Thus = and the result follows. a 1.3. Automorphisms of the upper half plane ‘The upper half plane I is by definition {2 € € : S(z) > 0}. As we shalll see in Chapter 2, Aut(lH) = {r © Aut() : 7(H) = n} is tho set of orientation preserving rigid motions of the upper half plane model of hyperbolic space. ‘To move from D to HI we introduce the Cayley transformation ari Cine zea Lemma 1.25. The Cayley transformation induces a conformal automor- phism H — D. Proof. We check easily that C(oe) = 1,C(1) = —i,C(0) = —1. It follows that Cmaps the line R U oo to the circle through 1, —i, —1, that is, to AD. ‘Thus C must map the two connected components of € \ BU ce to the two connected components of € \ OD. To see which goes to which, note that C(i) = 0 so HI must map to DD as claimed. ao Exercise 1.26. Check directly that if » = u+iv € Hand w = C(2) then w €D, and conversely. Theorem 1.27. The set Aut(H) of analytic bijections Hl — His the sub- group of Aut(C) of Mébius maps of the form: att with a,b,c,d © R,ad— be > 0. (13) f= Normalising, this shows that Aut(H) = SL(2,R)/ + I = PSL(2,B) Proof. First we first check that any map f of the form (1.3) is a bijection of H to HL We compute that Sf(2) = Since by assumption ad —be > 0 this means that: f(H) CH. Now sinee f is a bijection from € toCHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS 10 itself, to see that it is also a bijection of Hi to Hl, we just have to check it is surjective. This follows since by the same reasoning as before, f~"(IH) C HL So f € Aut(H). Now let, us prove the final statement. Since ad — be > 0, we have b Jad —be € R. So if we normalise the matrix (: ‘) € GL, B) by dividing cach coefficient by Vad —be, we sce that any map f of the form (1.27) is the image of under F (see Theorem 1.8) of a map in SL(2,R). So Aut(H) > SL(2,R)/ +1 = PSL(2,R). Finally we have to show that. an arbitrary F € Aut(H) is of the above form. Defining h: 24 a! we find ho F € Aut(H) and ho F(i) =i. (Why is he Aut(ED)2) Now composing with a rotation g : z+ cos(0)z + sin(0) —sin(0)z + cos(0)’ tion T = go ho F such that T(i) =i and 7"(i) = 1. Then check that CoT oC € Aut(D), T . Using form (1.1) of Theorem 1,22, that this implies that T = id and so T = id. Thus F = he! og”! and you can now easily verify that F is of the form (1.3). o you obtain a fu Remark 1.28. The above theorem can also be proved by conjugating to D using the Cayley transform. The main point is to verify that any T’ € Aut(D) conjugates to some element in SL(2,R). Write 0 0 8/ with fal? = bP = 1 ba As a matrix the Cayley transform © : Hl — DD is written ce Vk ~ifVX va ive ng factor which gives det C= 1. Calculating core. (Rat R? 80-30) Sa—Sb Ra— Rb, where A we find eCHAPTER 1. LINEAR FRACTIONAL TRANSFORMATIONS W which one easily sees (using |al? — [b|? Now given any 5 € Aut(H) we have CS be equated with some T as above. We havi C1CSC™'C = $ € SL(2,R) and we are done. 1) is in SL(2,R). € Aut(D) so that CSC" can, just calculated that C-'1TC =Chapter 2 Basics of hyperbolic geometry 2.1 Metrics, geodesics and isometries. We shalll study the upper half plane and unit disk models of hyperbolie space at the same time. The models are: (i) H= {2 € C: 32) > 0} with the metrie ds? = 2 €C:|2| <1} with the metric ds? = Proposition 2.1. With the above metrics, both HL and D are spaces of con- stant negative curvature —1. Proof. We compute using basic formulae from differential geometry. If ds? gid? + gydy? is a Riemannian metric on a surface, then the curvature K is given by the formula: -1 oO 1a a 1a co tL f8@ (Lg). 2 (LI Vinge alae ( moe ®) +a (Fe y )] Applying this to our metrie on El we have g) = g = 9 = 1/y?, so that a a 1 1 a(jia L aeV9=% gyV9=— yao y i (van) y and finally 1a(ia K=--.£ (22g) =-1 g oy (Jaan lar computation in D. a 12CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 13 ‘The expression for ds gives us the length of an are in Hl by the formula U(q) = f, ds. More precisely, if: [fot] — Hl with >(t) = 2(t) +iy(t) € Hl ue my dx\* (dy tor [gay (Ge) * (ae) For more details about this, see books on differential geometry or [6] Ch. 3. We shall be doing some explicit examples below. 2.1.1 Geodesics and isometries ‘The distance dp or dy: between two points P, Q is defined to be inf [(-7) where the infimum is taken over all paths 7 joining P to Q. Definition 2.2. A pathy from % to 2 is a geodesie if it locally minimises distances. Notice that in general a path may locally minimise distance without ac tually giving the minimum. For example, there are in general two geodesics joining any two points on a sphere (corresponding to the two different ways of going round the great circle from one point to the other). Both routes are geodesic but unless the points are diametrically opposite, one is shorter than the other. ‘As we shall see, in hyperbolic space there is only one geodesic are between any two points, so geodesics do automatically prove: nise distance. Proposition 2.3. (i) Vertical lines are geodesics in H. Moreover if b > a, then. dy(ai, bi) = log b/a. 1 (ii) Radial lines are geodesics in D. In particular dp(0,a) = log ( i + *) Proof. (i) Consider any path + joining ai and bi on the imaginary axis aR, (The proof is the same for any other vertical Tine.) Then yoy dx\* — (dy\* oy oT a) * Ca)” 1 cova ly dt fe uli) S| > Ldy ody + b Yas [ % = ogylt = 10g (2 f wae J OY — hoguit toe (2CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY uM Figure 2.1: An are 7 joining two Figure 22: An are ¥ joining two points on the imaginary axis in El. points on the horizontal radins in D. with equality if and only if 7 is the vertical line (Why?). This argument shows that the vertical line from ia to i actually minimises the distance, both locally and globally, so we deduce that dx (ai, bi) = logb/a. (ii) Take polar coordinates (r,6) in D and recall that dx? + dy? = dr? + r?dé®. ‘Then w= f 2 y( (“) (3) yar f ferret It follows from Proposition 2.3 that for any point x +7 € El on the same horizontal level as i, we have da(n+ i, + it) = —logt — 00 as t — 0. In other words, the distance from xr +7 to the point «x € R vertically below it is infinite. For this reason, R = RU oo is called the boundary at infinity of H, In particular we deduce as in (i) that do(0,a) (= The boundary at infinityCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 15 Figure 2.3: Equally spaced horizontal lines in Hl appear to get exponentially closer as we approach the real axis. written OHI. Point on OHI are sometimes called ideal points. Figure 2.3, which shows equally spaced horizontal levels in H, illustrates the infinite distance from a point in Hi to the boundary. In the disk D, set R = do(0,r). From the formula in Proposition 2.3 we got = 4" Then tanh (21) ‘Thus R~0 Srl and we similarly call 9D the boundary at infinity of D. Exercise 2.4. Prove that concentric cireles centre O € D which are at equal hyperbolic distance apart become exponentially clese together in the Buclidean metric. Isometries Before we find the other geodesics in Hand D, we first investigate isometries in the two models. Formally, a map f : H —> His an orientation preservingCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 16 isometry of H if it is differentiable (as a function R? — B®), if det Df(2) > Oz € HI, and if a(§ (21), £22) = d(21, 22) for all 21, 22 € HL. ‘The sot of all orientation preserving isometries of HL is denoted Isom" (H) ‘he set of orientation preserving isometries of ID, Isom’ (ID), is defined simi- larly. Recall from Chapter 1 that Aut(H) © PSL(2,R) and Aut(D) © PSU(1, 1) are the subgroups of linear fractional transformations which map Hi and D to themselves respectively, Proposition 2.5. The groups Aut(H) and Aut(D) act by isometries on Hl, D respectively. Proof. (i) By Theorem 1.27, iT © Aut(H) then T(2) = w = 4 wisn a,b,od € Rad ~ be = 1, Recall that Sw = —S2_.. (his means that cz + dj? if z € H, then T(z) € H.) Now we also compute that dw = —> and apply the change of variable formula. Let T : H —> Hl defined by T(z) = w. So [dw] fleztd dal 1a) = Jo Sw J, Sz fete See [6] or basic differential geometry books for more details. 19), (i) By Theorem 1.22, if 7 € Aut(D) then T(z) = w = e” €S! and a€D. Then [1 — |aP ld dy = and (=a) P _ a |2P) = Jal? = Goepee et) So Now you ean use the change of variable formula as before aCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 17 After we have discussed hyperbolic circles, we shall prove in Theorem 2.11 that in fact Isom’ (Hl) = PSL(2,IR) and Isom* (1D) = PSU(1, 1). ‘To avoid proving many different, facts twice, as in the above theorem, we can use the Cayley transform which we introduced at the end of Chapter 1. Proposition 2.6. The Cayley transformation C :H — D, defined by zti is an isometry from H — D. Proof. We already showed in Lemma 1.25 that C’ is a bijection between IH and D. Let = C-*(w) = 19) hen l-w dz e.g [te tD] _« dw eS ow ° So dz] 2Idw) 1 w? __2\do) G2 fl-wP 1 The result follows using the change of variable formula as in the proof of Proposition 2.5. a Exercise 2.7. Explain why Proposition 2.6 gives an alternative proof of Proposition 2.5 (ii) above. Geodesies Now we are in a position to find all the other geodesics besides the vertical lines and radii we discussed above. Geodesies are described in the following proposition, illustrated in Figure 24. Proposition 2.8. (i) Let P,Q € Hl. If RP = RQ then the geodesic from P to Q is the vertical line from P to Q, otherwise it is the are of circle, with centre on RU co = OH, joining P to Q. (ii) Let P,Q € D. If P,Q are on the same diameter then the geodesic from P toQ ts the Euclidean line segment joining them, otherwise it is the are of circle, orthogonal to $', joining P to Q. In all cases, the geodesic from P to Q is uniqueCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 18 Proof. The first thing to notice is that an isometry carries a geodesic to a geodesic, so throughout the proof we use isometries to move objects into a. convenient. position, Figure 2.4: Geodesies joining points P,Q in Hl and D. (i) Let n < € be the endpoints of the semicircle C, with centre on R = OH, Then pining P to Q (as deseribed in the figure) and let T= 2+ ; as “1 semicircle C to the imaginary axis because T(€) = 0,7(y) = 00 and because clements of Aut(E) preserve circles and angles. Hence, since the imaginary axis T(C) is the unique geodesic from T(P) to T(Q), C must. be the unique geodesic from P to Q. (ii) One way to prove this is to use the Cayley transformation which maps circle in circles and preserves angles. (Do this!) Alternatively one can make an argument similar to the one just seen. ‘To see there is a circle C as required, Pee Aut(D) whiel 1 T & Aut(H) because det ( 1 ) =€-1 > 0. Moreover T carries the consider the transformation T : 2 > carries P to 1-2P O. ‘Thon let C' be the unique diameter joining O and T(Q) and define c=T"'(c). a Proposition 2.9. (i) The group Isom*(H) (resp. Isom*(ID)) acts transi- tively on equidistant pairs of points in Hi (resp. 3D). (ii) The group Isom (H) (resp. Isom* (D)) acts transitively on ordered triples in OHI (resp. OD).CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 19 Proof. We will do the proof in H. ‘To prove the result in D you can either make direct, calculations or apply the Cayley transformation. (i) We have to show that if P, P',Q,Q! € H with dy(P, P’) = dy(Q,Q’), then there exists an isometry T’€ Isom* (H) such that T(P) = P’ and T(Q) = Q. It will be sufficient, to prove this for the case in which Q = é and Q’ i- cP) (Why?) Let S, be the map used in the previous proposition which maps the circle C to the imaginary axis. Now let S: be the map z+ 2/a which maps the imaginary axis to itself sending ai to i, so in particular S,54(P) = i. Since S25, is an isometry we dz (S25,(P), S25i(P’)) = de(P, P), 80 that: de(i, 525\(P))) = dex(P, P’) ietaalPP), = i cl(PP) = Q wo are done taking T = $25;. Other- wise, apply Sy : 2 + —1/z which fixes Q = i and takes it to i/t for any 1> 0. (Note that Sy = (; “) € SL(2,B).) Then $3$2;(P) = Q and S3525:(P’) = Q! so we are done. (ii) Say &, &, £4 € RUoo. The map 2-& &-& 2 G&G carries £1 + 0, & + 00, &3 + 1. Now det T = Gree), sodetT > 0 7" 2 iff and only if the two triples (0,00, 1) and (£1, &, 5) have the same eyelie ord on i. Then wo can normalize to get. 1’ € SL(2,R). a Ti zr |e,81, 82,65) Hyperbolic circles Let C(2o,) be the hyperbolic circle in D with centre zy € D and hyperbolic radius p. Since the formula for distance in D is clearly symmetric under rotation about 0 € D, it follows from the formula (2.1) that C(0, p) coincides with the Euclidean circle with centre 0 and Euclidean radius r = tanh(/2). Now consider C(z,p). The isometry T(z) — carries % to 0. — 202 Since isometries preserves hyperbolic circles, it follows that T(C(z,p)) = (0, p) and so C(z9,p) = T-\(C(0, p)). Since T carries Euclidean circles to Euclidean circles, C(zo, p) is also a Euclidean cirele. Notice however that theCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 20 hyperbolic centre of C(2q, p) does not in general coincide with the Euclidean centre. Exercise 2.10. Prove that hyperbolic disks are hyperbolically conver. (To say that a set X CH is hyperbotically convex means that given x,y € X then the geodesic from x,y is also in X.) Armed with these facts about circles, we can now prove that MGbius maps ive all the orientation preserving isometries of D and H. Theorem 2.11. Isom*(H) = Aut(H) —_Isom(D) = Aut(D). Proof. From Proposition 2.5 we already know that Isom*(H) > Aut(B) —_Isom(D) > Aut(D). Suppose T € Isom* H. By composing with a suitable clement of SL(2,B) as in the proof of Proposition 2.9 (i), we may assume that T' fixes two points P,P’ on the imaginary axis il. Now let Q € H. Since d(P,Q) = d(P,T(Q)), the point T(Q) lies on the circle radius d(P,Q) centre P. Likewise T(Q) lies on the circle radius d(P*,Q) centre P’. These two cireles are also Eu- clidean circles, they intersect in precisely two points, one on each side of the imaginary axis. Clearly, one of these two points is Q. Since T’ preserves orientation, 7(Q) must be on the same side of the imaginary axis as Q and so T(Q) = Q (why?) and hence T= id. ‘The result in D is similar, or use the Cayley transform. Oo Remark 2.12. From the above discussion it is easy to deduce: (i) T € Isom" (D) és differentiable. (ii) Bvery orientation reversing element in Isom (DD) is the composition of a reflection z+» 2 with an element in Isom (D). More formulae for distance ‘There are several useful ways of writing down the distance between two points without having to integrate along arcs.CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY a 1. Cross-ratio form Recall from Chapter 1 that SL(2,R) preserves cross- ratio. Now we shall see that it can be used to measure distance. Proposition 2.13. da(P,Q) = log[P’,Q, P,Q] and do( P,Q) = log", Q, P,Q] where P’ and Q are the endpoints on AHL or AD of the geodesic that joins P to Q, as shown in Figures 2.5 and 2.6. [Note the importance of the order.] Proof. Suppose that T € Isom* (Hl). We know that Tis an isometry of Hl and preserves cross-ratio. This implies that it maps the geodesic through P,Q to the geodesic through T(P),7(Q), and also that it maps the two endpoints of [P,Q] on R = RU co to the endpoints of ['(P),7(Q)] preserving order ‘Thus it is enough to check for points P and Q on the axis dR. (Why?) If P =i and Q=ai with a ¢R, then P’ = 0 and Q! = ov. In that case da( P,Q) = loga and 0-ai a 0-i co-ai PLQPQl= as we wanted to prove. Now use the Cayley transformation to prove the formula in D. a 2. A distance formula in H A very useful formula for calculating distance in I is: Proposition 2.14. Given tino points 1,22 € Hl: l= 22? coshda(aiv%)) = 14 GE Proof. To prove this formula it is enough to check that both sides are in- variant under isometries and then to check that the formula works when 21,2, €R. All of these are easy to do. aCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 2 @g @ P Q Pm g P Pe Figure 2.5: Cross ratio form of Figure 2.6: Cross ratio form of the distance in Hl the distance in D ‘This formula can be arranged in other ways which are sometimes useful: la =| + la =a! la - | +]a =| lar ~ al 2Bz1B22)2 la — 22] 2321920)? de(21, 22) = In sinb(des(21,22)/2) cosh (dr 21,22)/2) = 5 tanh (e(21,2)/2) = It is an easy exercise to prove that the four equations are equivalent to each other and to the equations of Proposition 2.14. There are also analogous formulae in D: Corollary 2.15. Given two points 2,22 € D we have: a [l- ani? dy(21, 22) = In + cost(do(as22)/2) = Gj apy ler — 2a} sinh?(dp(21, 22)/2) = # banbi(do (215 22)/2) = 7 (1 Jar?) = |ea[?) Proof. ‘This is an exercise using the previous results and the fact that dp(2, 2 d(C ¥(21),C-"(22)) where C': Iki —+ Dis the Cayley transformation, which follows from Proposition 2.6. For details, sce [10] p. 131 a 1 — 2123]CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 23 2.2 Similarities and differences between hy- perbolic and Euclidean geometry 2.2.1 Circles As we saw above, hyperbolic circles are off-centre Buctidean circles. In par- ticular, the hyperbolic circle centre O € DD and radius p coincides with the Euclidean circle centre O € D and Euclidean radius r = tanh p/2. Lemma 2.16. The circumference of a hyperbolic circle of radius p is 2 sinh(p) ‘The area of a hyperbolic dise of radius p is arsinh?(S) Proof. The hyperbolic length of the circumference of a circle with radius p is c where r = tanh(S) But 11? = eosh”*(F), 80 Gi ‘eumference = 4 sinh(£) eosh(2) = 2rsinh(p). ‘The hyperbolic area of the dise with radius p is epee ap ato = LoL wer So Remark 2.17. Observe the very important fact that both the circumference and the area grow exponentially with the radius Exercise 2.18 (Circles in H). We know that a circle in Hi must be an off- centre Euclidean cirele. If the circle has radius p and centre i, you can see easily that it meets imaginary aris at points ic? and ie-*, as shown én Figure 2.7. Clearly it must be symmetric about the imaginary aris. What are its Euclidean centre and radius?CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY By Figure 2.7: Hyperbolic circle centre i in 2.2.2 Angle sum of a triangle One of the most: important differences between Euclidean and hyperbolic geometry is the angle sum of a triangle: Lemma 2.19. The angle sum of a hyperbolic triangle is strictly less than x. Proof. Put the hyperbolic triangle A in D with one vertex at the centre O. ‘The sides through O are radii. If P, Pare the other two vertices, it is easy to see that the Euclidean triangle with vertices OPP strictly contains A, and in particular, that the ZOPP", ZOP'P in A are less than the corresponding Euclidean angles, from which the result: follows. a Corollary 2.20. The angle sum of « hyperbolic polygon with at most n sides is strictly less than (n ~ 2)r. Notice that this includes the case of triangles (or polygons) with one or more vertices on the boundary at infinity. At any such vertex, the angle is zero, because if two lines meet at a point P € AHH, then they are both orthogonal to OH and hence the angle b We shall find the precise relation between area and angle sum later. een them is zero.CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 25 2.2.3 Perpendicular projection and the common per- pendicular. In Euclidean geometry, there is a unique perpendicular projection from a point to a line. The same is true in hyperbolic space: Theorem 2.21. ‘There exists a unique perpendicular from a point P to a line £. This perpendicular minimises the distance from P to £. 2 NY Figure 2.8: The hyperbolic perpendicular from the point P to the line £. Proof. Place P at O € D. Referring to Figure 2.8, it is clear from Euclidean geometry that there is a unique radial line PQ perpendicular to the hyper- bolic line £. ‘The relation dp(P,Q) < do(P,Q') follows since from Euclidean geometry, any other point Q’ € L lies outside the hyperbolic circle centre 0 and radius |PQ|. a Suppose that £ and C’ are are parallel lines in Buclidean space. Then they have infinitely many common perpendiculars, each of which minimises the distance between them. By contrast, in hyperbolic space: ‘Theorem 2.22. Let £ and £’ be disjoint lines which do not mect at oo. Then LL! have a unique common perpendicular which minimizes the distance between them, Proof. ¢ Existence: This time we arrange line £ to be the imaginary axis in H. We may suppose that £’ is a Euclidean semicircle as shown in Figure 2.9.CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 26 Figure 2.9: The circle centre O and rs from L£ to L'. 5 is the common perpendicular (Why?) It follows from Euclidean geometry that the circle with centre O and radius s, where s? = |OP|- |OQ|, cuts £’ orthogonally. ¢ Uniqueness: Any ‘other common perpendicular would produce either a quadrilateral whose an- gle sum is 2m or a triangle of angle sum larger then , as shown in the figure. Both of these cases are impossible by Lemma 2.19 and its corollary. © The common perpendicular minimizes distance. Denote the common per- pendicular by M and let X,X" be the points where M meets £,£’ respec- tively. Let Y,¥" be any other two points on £,L’ respectively. ‘The result follows immediately from the next theorem, on noting that the perpendicular projections from Y,Y’ onto M are exactly the points X,X’. ‘Theorem 2.23 says that dx(¥,¥") > de(X,X"). a Theorem 2.23. Perpendicular projection onto a line M strictly decrease distances. More precisely, suppose that P,Q ¢ M and let P,Q! denote the perpendicular projections of points P,Q onto M. Then d(P,Q) > d(P',Q’) Proof. In K, move M to the imaginary axis by sending the points P’,Q! to i and ei respectively, so that dy(P”,Q’) = R. The line PP’ meets M orthogonally and hence P lies on the geodesic through é perpendicular to the imaginary axis, so P = e'® for some @ € (0,7). Similarly we can write Q = c®e'* for some @ € (0,7).CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY Pr Figure 2.10: Perpendicular projection decreases distances: |P'Q'| < |PQ| The distance formula in Proposition 2.14 gives |e? — eet? sin eR sind 1+ 6% — 2c" (cos(8 — @) 2eF sin sind (=e) sinh? + SpRangaing 2 1 * 2sinh?(R/2). cosh dy(P,Q) = 1+ = 1+ Iv Using cosh d = 1 + 2sinh d/2 this gives sinh? dg(P,Q) > sinh?(2/2) = sinh? da(P’,Q’) so that d(P,Q) > da(P’,Q) with equality if and only if P = P’ and Q = Q We remark that: using the angle of parallelism formula, sce Lemma 2.27 below, we have cosh die(P, P* yy + Hence sing 24 cosh da(@, 2’) = = G- Hence our computations actually show sinh? da(P,Q) > sinh?(R/2) cosh du(P, P’) cosh da(Q, Q’). 2.2.4 Lines meeting on JH ‘Theorem 2.22 shows that if two lines £ and £' do not meet then there is a non-zero shortest distance between them which at infinity, realisedCHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 28 by the unique common perpendicular. If the lines have the same endpoint on dil or AD, however, then it: is not hard to show that the infimum of the distance between points on the two lines vanishes, in symbols: inf{d(P,P): PEL, PEL} =0. ‘The calculation is done by arranging the common point of £ and £! to be oo € H, see Figure 2.11. If we take the two points P, P’ to have the same imaginary part h, then Mde 1 dy(P.P)< [| Fa 5 10 as hv 00. MP PS | R70 a h—+00 Exercise 2.24. In fact dy(P,P’) <1/h. Why? Can you caleulate the actual value? We say the lines £ and £’ 1 thogonally, the angle betwee there is no geodesie which actually realises the infimum. at infinity. Because both meet JH or- h lines is zero. Notice that in this case, Figure 2.11: The lines £ and £' meet at 00; da(P, P!) = 1/h. If a ‘triangle’ or ‘polygon’ has two sides which meet at oo we call the meeting point an ideal verter. ‘The vertex is not in EI but we can still identify it as a definite point on dbl which can be transported around by SL(2,R). Ifall vertices are ideal, the triangle (resp. polygon) is called an ideal triangle (resp. polygon). Exercise 2.25. Show that if A, A’ are two ideal triangles in Hl, there is an isometry carrying on onto the other. Is this isometry unique?CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 29 Figure 2.13: Points equidistant Figure 2.12: ‘The set of points from the line £. ‘These lines cquidistant from the imaginary are sometimes called hyper- axis. cycles. Let us summarize these results, contrasting with Euclidean geometry. In Enclidean geometry, either two lines meet and they have no common per- pendicular, or they are parallel and they have infinitely many common per- pendieulars. In hyperbolic geometry, by contrast: (a) if two lines meet, then as in Euclidean geometry they have no common perpendicular, for this would give a triangle with angle sum > 1. (b) if two lines are “parallel”, that is they meet at oo, th mon perpendicular because that would give a triangle with angle sum equal to , contradicting Lemma 2.19. ‘The lines become asymptotically close as they approach co and the angle between them is zero. (c) if the two lines are “ultra-parallel”, that is they don't meet and they aren’t parallel, th ue common perpendicular which min- imizes the distance betwe 2.2.5 Points equidistant from a given line In Euclidean geometzy, the line parallel to £ through the point P ¢ £ consists of all points Q such that d(@,£) = d(P,£). What about the set of points equidistant from a hyperbolic line? Theorem 2.26. Let £ be a hyperbolic line, The set of all points at a fixed distance from £ is a Euclidean circle with the same endpoints as L on OH (or OB), making an angle 0 with OE (or AD).CHAPTER 2. BASICS OF HYPERBOLIC GEOMETRY 30 Figure 2.15: Horocycles based Figure 2.14: Horocycles based at the point € € R. aloo. Note the distance rela- tionship + = te Proof. First consider the case when L is the imaginary axis in H. Let M be a Euclidean line through 0, making an angle @ € (0,/2) with OE. The L-parameter family of maps {2+ Az: A € RY} fixes M and £ and act transitively on each by isometries. In particular, referring to Figure 2.12, da(P,Q) = da( P’,Q’). In the general case, transporting the points 0, co to the points ,1) € JEL, we see that the locus of points equidistant from a given line £ consists of points on a circle with the same endpoints 7) as £ on JE and which meets AE at, angle 0. a 2.2.6 Horocycles ‘A horocycle is the limit of a circle as the centre approaches OH. In our models, they are Euclidean circles tangent to the boundary at infinity. ‘Thus in H thoy are either horizontal lines if the tangency point € = oo, or circles tangent. to the real axis if € € R, soe Figures 2.15 and 2.14. As we shall see later, horocycles will play an important role. Suppose that. points P, P’ lic on the same horocycle. It is useful to com- pare the distance £ between P,P! measured along the horocycle and the geodesic distance. To do this, as in Figure 2.11 we can arrange the horocycle to be a horizontal line at Euclidean height h > 0 above the real axis in H and by translating and scaling assume that P = ih, P' = ih-+ 1. (Why?)