Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Pappus’s theorem and the modular group

Pappus’s theorem and the modular group by RICHARD SCHWARTZ 1. Introduction Pappus's theorem is as old as the hills. It refers to the configuration of points and lines shown in Figure 1.1. Fro. 1.1 Pappus's Theorem. ~ If the points A, B, and C are collinear and the points A', B' and C' are cotlinear, then the points A", B" and C" are cotlinear. A slight twist makes this theorem new again. The twist is to iterate, and thereby treat Pappus's theorem as a dynamical system. Pappus's theorem may be considered as a dynamical system defined on objects called marked boxes. Essentially a marked box is a collection of points and lines in the projective plane P which comprises the initial data for Pappus's theorem. (See w 2.2 for a precise definition.) When Pappus's theorem is applied to a marked box, more lines and points are produced. These new lines and points may be used to form new initial data for the theorem, and so on. In fact, one can construct an entire group G of operations to perform on marked boxes. Algebraically G is the modular group, Z/2 9 Z/3, and is generated by the simple operations alluded to above. (For a precise definition of the operations, see w 2.3.) The 188 RICHARD SCHWARTZ orbit ~ of a marked box is an infinite collection of marked boxes, nested inside and outside of each other in a pattern encoded by the modular group. The modular group makes a second entrance into Pappus's theorem, as the group of projective symmetries of ~. Projective transformations are analytic diffeomorphisms of P which take collinear points to collinear points. Projective dualities are analytic diffeomorphisms between P and its dual space of lines P* which take collinear points to coincident lines. Projective transformations and dualities together generate the Lie Group of projective symmetries of P. It turns out that the projective symmetry group M of ~ (those projective symmetries which permute the marked boxes of ~)) is again algebraically the modular group. The group of projective symmetries and the group of operations commute, and so there are two commuting modular group actions on ~. As the original marked box varies, the orbit varies, along with the two group actions. Everything varies smoothly, thereby producing two commuting modular group actions on the space of all marked boxes. There is a certain geometric condition one can put on marked boxes, called convexity. The set of convex marked boxes forms an open subset of the set of all marked boxes. For these marked boxes, there is a certain amount of geometry and topology associated to the group actions described above. In the convex case, there is a fractal curve associated to the orbit ~. Certain " distinguished " points of the marked boxes in ~) are dense in a topological circle A, which is generally non-smooth. Dually, certain " distinguished " lines of the marked boxes in Y~ are dense in a topological circle L of lines. The circles A and L are " self- projective " curves in the sense that they are preserved (or swapped) by the modular group of projective symmetries of ~2. In a certain sense, they are analogues of quasi- circles in the projective plane. Figure 1.2 shows certain of the convex marked boxes in an orbit. Both A and L are " hinted at " in the picture. Projective symmetries induce analytic diffeomorphisms of the projective tangent bundle ~ of the projective plane. When the original marked box is convex, the symmetry group M gives a discrete group action on ~. In fact this group action has a domain of discontinuity and a corresponding quotient 3-manifold. More precisely, let Jff C denote those pairs (p, l) such that p r A and l r L. The fixed points of M are dense in, and contained in, the complement of Jff, and .h r itself is the domain of discontinuity for M. The quotient $" ---- ~r/M is a three-dimensional analytic manifold with the homotopy type of the trefoil knot complement in the three sphere. The manifold $" is a " (~, s " in the sense of IT]. Here fr is the Lie Group of projective symmetries, and s is the universal to the projective tangent bundle ~. There is a classical geometric structure on the trefoil knot complement coming from its description as SL~(R)/SL,(Z). In the case of a " totally symmetric " marked box, one essentially recovers this structure. However, as the marked box varies, one obtains a 2-parameter family of " exotic " geometric structures on the (homotopy) trefoil knot complement. PAPPUS'S THEOREM AND THE MODULAR GROUP 189 FIc. 1.2 In sum, each initial choice of lines and points involved in Pappus's theorem (subject to the convexity condition) gives rise to the following objects: 1. an infinite collection ~ of marked boxes, which is indexed by the modular group, and which has the modular group as its group of projective symmetries; 2. a pair of (usually fractal) curves A ~ P and L ~ P* which are invariant (or swapped) under a modular group of projective symmetries; 3. a representation M of the modular group into the Lie Group of symmetries of the projective tangent bundle ~; 4. a (usually " exotic ") (if, 5~)-structure $" --JV'/M on the (homotopy) trefoil knot complement. This paper is self-contained, but does assume some basic hyperbolic and projective geometry. The hyperbolic geometry can be found in [B], and the projective geometry, as well as an excellent treatment of Pappus's theorem, can be found in [H]. 190 RICHARD SCHWARTZ Here is a list of contents: w 2: Iterating Pappus's Theorem: w 2.1: The Projective Plane w 2.2: Marked Boxes w 2.3: Operations on Marked Boxes w 2.4: Two Commuting Group Actions w 3: Geometry of Marked Boxes w 3.1: Geometry and Depth w 3.2: The Pappus Curve w 3.3: Curve and Linefield w 4: The Three Manifold w 4.1: The Projective Tangent Bundle w 4.2: Domain of Discontinuity w 4.3: Trefoil Knot Complement I would like to thank Hsin-Ta (Frank) Cheng, Peter Doyle, Bill Goldman, Jun Hu, Linda Keen, and Dennis Sullivan for helpful and interesting conversations concerning this work. I would also like to thank the IHES, and the Einstein Chair at CUNY, for their support during the genesis of this paper. 2. Iterating Pappus's Theorem 2.1. THE PROJECTIVE PLANE As usual, we will take the projective plane to be the space P of one-dimensional subspaces of R 3. The ordinary plane R * sits naturally as a subset of P, and consists of all one dimensional subspaces which do not lie on the xy-plane of R 8. Projective transformations of P are just projectivized linear transformations of R 3. The group of projective transformations is naturally isomorphic to SL3(R ). There is a unique projective transformation which takes one specified set of 4 (ordered) general position points to another such set. P* will denote the space of lines of P. This set is naturally in bijection with P. A duality is just a projectivized linear transformation from R s to its dual space. More concretely, a duality is a homeomorphism P-+ P* which takes collinear points to coincident lines. As a special case, a duality which is induced by a positive definite inner product on R ~ is called a polarity. Each duality A induces a dual map A* : P* -+ P, defined by the formula A*(~) = A(p) c~ A(q). A set Q c P is convex if there is a projective transformation T such that T(Q) is a convex subset of the ordinary plane. PAPPUS'S THEOREM AND THE MODULAR GROUP 191 2.2. MARKED BoxEs Informally, a convex marked box is a convex quadrilateral in P with a distinguished top edge, a distinguished bottom edge, a distinguished top point and a distinguished bottom point. Here are some definitions, which culminate in a formal definition of a convex marked box, the main object of study. An overmarked box is a pair of 6 tuples ((p, q, r, s; t, b), (P, Q, R, S; T, B)) having the incidence relations shown in Figure 2.2. There is an involution on the set of overmarked boxes: ((p, q, r, s; t, b), (P, Q, R, S; T, B)) ~ ((q,p, s, r; t, b), (Q, P, S, R; T, B)) A marked box is an equivalence class of overmarked boxes under this involution. Let | be the marked box labelled as in Figure 2.2. The top of t9 is the pair (t, T). The bottom of | is the pair (b, B). The distinguished edges of 19 are T and B. The distinguished points of | are t and b. Fio. 2.2 So far, the definitions make sense in a projective plane over any field. (The exception is the projective plane over Z/2, which doesn't have enough points to contain a marked box.) The remainder of the definitions depend on the ordering properties of R. The marked box | is convex if the following 4 conditions hold: 1. p and q separate t and T n B on the line T. 2. r and s seperate b and T c~ B on the line B. 3. P and Qseperate T and bt on the pencil of lines through t, in the cyclic ordering of lines through b. 4. R and S seperate B and bt on the pencil of lines through b. The first two conditions imply the last two, and vice versa. The convex interior of @ is the open convex quadrilateral whose vertices, in cyclic order, are p, q, r and s. 192 RICHARD SCHWARTZ Projective symmetries act on marked boxes in a way which preserves convexity: Let ~ = T(x) for a projective transformation T. We define: T(| ---- ((/~, ~, r. ?; t, b), (P, Q, R, S; T, B)). Similarly, given a duality A, let x* = A(x) when p is a point, and X* = A*(X) when X is a line. We define: A(| = ((P*, Q*, S*, R*; T*, U*), (q*,p*, r*, s*; t*, b*)). The reordering of elements of A(O) is deliberate. 2.3. OPERATIONS ON MARKED BOXES There are three natural box operations one can perform on marked boxes. Before defining these operations, we introduce some notation. If p, q e P, then pq is the line containing p and q. Likewise, if P and Q are distinct lines in P, then PQis the intersection of P and Q. This notation may be applied recursively; for example, (ab) (cd) is the intersection point of the lines ab and cd. Given | = ((p, q, r, s; t, b), (P, Q., R, S; T, B)) as above, we define: i(| = ((s, r, p, q; b, t), (R, S, Q, P; B, T)). 9 1(| ---- ((P, q, QR, PS; t, (qs) (pr)), (P, Q, qs, pr; T, (QR) (PS))). ":3(| = ((QR, PS, s, r; (qs) (pr), b), (pr, qs, S, R; (QR) (PS), B)). Finally, for completeness, we define the identity operation 1(| = | The three box operations are shown schematicaly in Figure 2.3, for a convex marked box. Fro. 2.3 PAPPUS'S THEOREM AND THE MODULAR GROUP 193 The box operations may be applied iteratively, to form a semigroup G of box operations. We will use the notation ab to mean: "first apply b, then apply a ". It is clear that i and -r~ commute with projective transformations. Two simple computations, one for i and one for (say) ~1, show that dualities commute with these box operations as well. Hence the semigroup of box operations commutes with projective symmetries. Lemma 2.3. -- The following relations always hold: i S = 1 ; "q i'~, = i; "q i~ t = i; v~ izt = v~; v2 iv2 = "q. Proof. -- The first relation is obvious. By symmetry, we only have to verify the third and fourth relations. We compute: i,1(| = ((PS, QR, p, q; (qs) (pr), t), (qs, pr, Q, P; (QR) (PS), T)). ,q i-q(O) = ((PS, QR, (pr) Q, (qs)P; (QR) (PS), (q(QR)) (p(PS))), (...)). "r, i'q(| = ( (pr) ( Q, (qs) P, q, p; ((PS) p) ((QR) q), t), (...)). Looking at Figure 2.2, we see that (pr) Q = r; (qs) P = s; (q(QR)) (p(PS)) = RS = b. Substituting these identities in, we see that the 6 points of-q i~1(0 ) match the 6 points of ~(| and the 6 points of ** iv1(| ) match those of i(| To complete the proof, we note that the 6 points of a marked box clearly determine the entire marked box. 9 The relations in Lemma 2.3 imply that the semigroup G is in fact a group. Clearly G is generated by ~ = i and ~ = iv1. Note that ~3 = iT 1 i.~l iv 1 = iv 1 i.~2 = 1. Looking at the nesting properties of the convex interiors of the marked boxes in the orbit of | in the case | is convex, we see that the two elements ~ = vl and 0c~ ~ = v2 generate a free semigroup. This is enough to identify G as the modular group: G= (~,~:~*= 1,~a= 1). 2.4. Two COMMUTING GROUP ACTIONS Let f2 = f~(| be the orbit of | under G. It is not hard to see, particularly in the convex case, that G acts freely, faithfully, and transitively on fL The structure off~ can be expressed by an incidence graph F. The edges of P correspond to marked boxes in f~, the vertices correspond to tops and bottoms of boxes, and each edge is directed from the top to the bottom. Vertices on distinct edges are identified if the corresponding distinguished sides coincide. A picture of P is shown in Figure 3.1. I. Technically, P is just an abstract graph, but we will always embed P in the hyperbolic plane, as the tiling associated to the hyperbolic reflection group generated by reflections in an ideal hyperbolic triangle. 25 194 RICHARD SCHWARTZ The group G of box operations has a concrete description as a permutation subgroup of edges of I'. The operation i reverses the orientations on each edge. The element -c 1 " rotates " each edge counterclockwise one " click " about the tail point. The element z2 " rotates " each edge one " click " clockwise about the head point. Note that G is not a group of homeomorphisms of the hyperbolic plane. Each marked box induces a faithful representation G of G into the permutation subgroup of marked boxes in the orbit ~. Besides G, there is a second, commuting, modular group action M on I'. This action M is generated by order 2 isometric rotations about centers of edges of U together with order 3 isometric rotations about centers of triangles in U. The marked box | induces a representation M into the Lie Group ~ of projective symmetries of P. Theorem 2.4. -- Let | be a marked box, with orbit ~. There is a faithful representation ~r : M --~ (~ which takes isometrics of I" to projective symmetries of ~ in a way which is natural with respect to the labelling of U. Proof. -- We will show that, for each marked box 9 e ~: 1. There is an order 3 projective transformation having the cycle i(~) ~ ~1(~) -> ~2(~V). 2. There is a polarity having the cycle 9 ~ i(~). The theorem then follows from the fact that everything in sight commutes. Let 9 = ((bl, b3, a3, al; a~, b2) , (...)). Figure 2.4.1 shows a particular norma- lization of ~. We compute i(tF) = ((al, aa, bt, b3; a~, b,), (...)), vl(~F) = ((61, ba, cl, c8; bz, c,), (...)), = ((cl, ca, al, as; c,, (...)). The obvious Euclidean rotation satisfies (1). a2 b3 c3 Fxo. 2.4.1 PAPPUS'S THEOREM AND THE MODULAR GROUP 195 s! \ rl Fie,. 2.4.2 Now let V ---- ((p, q, r, s; t, b), (P, Q, R, S; T, B)). Figure 2.4.2 shows a norma- lization of q~ in which t and b are at infinity, and [IP II [I q ]1 ----- ]1 r II I] s ]] = 1 in the standard norm on the Euclidean plane. The polarity induced by the standard inner product on 113 satisfies (2). 9 All in all, we have produced two different commuting modular group actions G and M on the orbit ~. As the original marked box varies, the orbit varies, and these group actions vary as well. Thus we produce two different commuting group actions on the space of marked boxes. 3. Geometry of Marked Boxes Henceforth, we make the blanket assumption that all our marked boxes are convex. 3.1. DEPTH OF A Box The labeling of the incidence graph I ~ is determined by the choice of O in the orbit ~. It can be arranged that the edges corresponding to i(| ~1(| and "r2(| ) bound the hyperbolic triangle which contains the origin 0 C 112. Given this labelling, we make the following definitions concerning the placement of other boxes in f~ with respect to | Let v be a vertex of P. Say that the depth of v is the minimum number of edges of U which must be crossed by any path from the origin to v. The depths of various vertices are shown in Figure 3.1.1. Each directed edge e ~ U has a half space H, associated to it. The assignment e -+ H e is made so that H, ~ Hf if and only if the convex interior of the marked box labelled by e is contained in the convex interior of the marked box labelled by f. Let H1, H 2 and H 3 be the three half spaces shown in Figure 3.1.1. Assuming that H, C Hj for some j ~ { 1, 2, 3 }, define the major depth of the edge e to be the maximum depth of the two vertices bounding e. Define the minor depth of e to be the minimum depth of its two 196 RICHARD SCHWARTZ 2 i __ 2 1 1 el .0 2 e2 e3 ~ ~. ' 0 FIG. 3.1.1 vertices. There are only finitely many edges at every major depth, and infinitely many at each minor depth. The directed edges of P are in bijection with the marked boxes of f2, so we can transfer the notion of major and minor depth to marked boxes. We will use the nota- tion md(tF) and Md(tF) to denote the minor, and major depth of tF. The purpose of this section is to make precise the statements that boxes having high depth are small and thin. The projective plane inherits a Riemannian metric p from its description as the sphere modulo antipodal points. The metric is not very natural with respect to the full group of projective transformations, but it will nonetheless be useful to us here and in w 4. Unless otherwise stated, all metric measurements (distances and angles) will be made with respect to p. On the other hand, when we speak of convexity, we will always refer to the projective geometric notion, described in w 2.1. Given a marked box ~F ~ f2, let ~F denote the convex interior of ~F, let ~F denote the (smaller) convex quadrilateral shown in Figure 3.1.2, and let a(~) denote the angle shown in Figure 3.1.2. IX I will denote the 9-diameter of the set X C P. ,..', FIG. 3.1.2 PAPPUS'S THEOREM AND THE MODULAR GROUP 197 Depth Lemma. -- Let ~ > 0 be fixed. Then there is a constant N = N(e, O) such that, for any marked box ~ ~ 1. md( r) > N * leVI< 2. Md(~F) > N => t~'[< ~; Proof. -- Without loss of generality, we assume [~l ~ [E x [, and we normalize so that Ex is the unit square. The spherical metric p restricted to the unit square is only boundedly different from the Euclidean metric there, so for ease of computation we will work with the Euclidean metric. This bounded change, and the original norma- lization, merely alter the constants in the Lemma. We may write ~ = w(Et) , where w is a word in the box operations v s. By throwing out uniformly small words, and using symmetry (between -r 1 and .r~) we can assume that w has one of the following properties: 1. The string ... -q v~ ... occurs n times in w, for large n. " for large n. 2. W ~ V2%'1, 3. w ---- v~, for large n. In the first two cases, both md(~) and Md(~) are large. In the third case, only Md(~) is large. Case 1. -- Recall that Et is the unit square. Suppose F ~ ~ is any marked box with FCE 1. We will show that there is a constant ~-~ ~(| 1 such that r d FIG. 3.1.3 Consider the diagonal d shown in Figure 3.1.3. If d is chosen correctly, [~r~[ =ld]. Also, IF[>/ [dc~Fl" All the boxes in ~ are equivalent under a projective symmetry, so the cross ratio of the 4 points shown in Figure 3.1.3 takes on one of finitely many values. Here is a useful trick: Let a, < b, < c, < d, C R be an infinite sequence of 4-tuples of points. Then the ratio l c, -- b, Ilia, -- d,I cannot converge to 1 if the cross ratio 198 RICHARD SCHWARTZ ~, = [a., b,, c,, d,] takes values in a finite set. This trick bounds the diameter of the smaller quadrilateral away from the diameter of the bigger one. Thus [ (~) [ ~< I VF I ~< 0". The argument for ~(W) is similar: one uses a similar trick to show that x(vx v,(F)) ,< ~' a(F), for some other constant ~' < 1. Case 2. -- From Theorem 2.4, there is a projective transformation T which has the infinite orbit E1 ~ v~(E1) ~ x~(E1) .., This map T fixes the distinguished point p at the top of El, and takes the open unit square into itself. It follows easily that T has no fixed points in the open unit square. Since T preserves the two lines x and y shown in Figure 3.1.4, % z~(Ex) = Tk(%(E1)) shrinks as a set to the point p as k grows. Hence ~" and ~ are shrinking to points as well. Fio. 3.1.4 The lines I, and m,, shown in Figure 3.1.4, converge to the top line of El, squeezing the angle ~ down to zero. Case 3. -- The argument here is the similar to that in Case 2, except that the top edge of v~(E1) does not shrink to a point. The only difference in this case is that I CFI remains large. Stringing the three cases together proves the Lemma. 9 3.2. THE PAPPUS CURVE The marked box 6) determines a natural map X from vertices of the incidence graph F into the projective plane. The rule is that ?,(v) is the common distinguished point on all the marked boxes labelled by edges in F emanating from v. Part of the image of X is " hinted at" in Figure 1.2. Recall from w 2.4 that M is the representation of the hyperbolic isometry (modular) group M into the Lie Group of projective symmetries. The index 2 subgroup of M' C M consisting of those group elements taken to projective PAPPUS'S THEOREM AND THE MODULAK GROUP 199 transformations preserves the image A of X. Furthermore, the vertices of P are dense in S x, and it is natural to ask whether or not X extends continuously. If this is true, then the whole image A -- X(S a) is invariant under M'. Theorem 3.2. ~ The map extends to give a continuous homeomorphism from S t onto its M'-equivariant image A C P. This homeomorphism conjugates the action of M' to that of M'. Proof. ~ Recall that ~ is the small convex quadrilateral shown in Figure 3.1.2. The image vertices of X are arranged according to Figure 3.2. Since the nesting properties of the edges of F exactly reflect the nesting properties of the convex interiors of boxes in ~, the following fact is sufficient to prove the theorem: let ea, e2, ... be any infinite sequence of nested edges in F. (This means that H~j~ H~i+,. ) Let tF~ be the corres- ponding marked box. Then I~1 ~ 0 as j -+ oo. But this is an immediate consequence of the Depth Lemma. m FIo. 3.2 3.3. CURVE AND LINEFIELD The dualities of M- M' take the curve A to a curve of lines L, which contains the tops and bottoms of the marked boxes in the orbit ~. The curve L is " hinted at " in Figure 1.2. The two objects L and A not only are dual to each other, but also have a special geometric relationship. A transverse linefield to A is a continuous curve X of lines such that each line of X intersects A in exactly one point, and that each point of A is contained in some line of X. Dually, a section of L is a curve X such that each point of X is contained in exactly one line of L, and each line of L contains exactly one point of X. From the nesting properties of the marked boxes in the orbit ~, it follows that L is a transverse linefield to A, and dually A is a section to L. 200 RICHARD SCHWARTZ Theorem 3.3. -- If A is not a straight line, then L is the unique transverse linefield to A. Dually, if L is not a curve of coincident lines, then A is the unique section to L. Proof. -- We will prove the first statement. By Theorem 3.2, the continuous extension ~ of ~ conjugates the action of M' on S 1 to the action of M' on A. In particular, M' has a dense set of fixed points in A. By continuity, we just have to produce a single point p e A at which any two transverse linefields must agree. Say that a pair of points (p, q) E A � A is a hyperbolic pair if there is a nontrivial element T e M' which fixes both p and q. Such pairs are dense in A x A because the hyperbolic translations of M' have fixed points which are dense in S i X i. S Such an element T attracts about one of the fixed points, say p, and expands around the other one. Since A is not a straight line (and not convex either) there are two points p, q e A such that any line sufficiently close to p-q, emanating from p, intersects A in some point other than p. By density, we can assume that (p, q) is a hyperbolic pair. The situation is shown in Figure 3.3. m' Fio. 3.3 Let l~ be the line of L through p and let lq be the line of L through q. Then T preserves p-q, l~, lq, and fixes the point l~ n l~. Let m be any fourth line through p. We will prove below (Lemma 3.3) that T"(m) converges to l~. Now, if m is sufficiently close to p-q, then m will intersect A elsewhere. By equi- variance, T"(m) will intersect A elsewhere as well. By continuity, all lines containing p between m and T"(m) will intersect A elsewhere. Letting n ~ ~, and m ~pq, we see that all lines between pq and l~ which contain p will intersect A elsewhere. The same argument may be repeated for lines m' on the other side of~. Hence, every line through p except l~ intersects A in at least two points. 9 PAPPUS'S THEOREM AND THE MODULAR GROUP 201 Lemma 3.3. -- Suppose T ~ M' fixes two distinct points p, q ~ A, and attracts about p. Suppose m is a line through p distinct from l~ andS. Then T"(m) converges to l~. Proof. -- The line m intersects l~ at a point x ~ la. This intersection point is neither q or l~ c~ lq. If T"(m) failed to converge to l~, then T would fix x. Fixing three distinct points on l~, T would be the identity on lq. By looking at the conjugated action of T back on S 1, we observe the following fact: if W is any marked box such that q ~ VF, then the open convex interior of T(VF) contains the closure of the convex interior of W. This is not consistent with T fixing Lq pointwise. 9 4. The Three-Manifold 4.1. THE PROJECTIVE TANGENT BUNDLE The projective tangent bundle ~ to the projective plane is the closed analytic 3-manifold of pairs (p, ~), such that p ~ P, P ~ P*, and p ~ P. We will call points of flags. The Lie Group ff of projective symmetries acts as a group of analytic diffeo- morphisms of ~, as follows: T((p, P)) ---- (T(p), T(P)) for a projective transformation T, and A((p, P)) = (A*(P), A(p)) for a duality A. We will say that a pencil of ~ is a set of flags, such that one of the two coordinates varies while the other one takes on all possible values. The objects with which we will deal, in w 4.3 below, are ruled surfaces in the sense that they can be expressed as the union of continuously varying disjoint pencils. 4.2. DOMAIN OF DISCONTINUITY The representation M constructed in Theorem 2.4 gives an action on ~ by analytic diffeomorphisms. Let JV C ~ denote the set of pairs (p, P) such that p r A and P r L. The representation M preserves both J/" and its complement. Lemma 4.2.1. -- The fixed points (flags) of M are dense in, and contained in, the complement of~V. Proof. -- We will prove density first, then containment. Consider the set J/ of pairs (p, P), such that P e L. This set makes up essentially " half" of the complement of.A r. It is sufficient to show that the fixed flags are dense in dr'. There is a continuous map : A � A -- Diagonal --> defined as follows. Let l~ and lq be the lines of L which contain the two points p, q e A. Define q~(p, q) = (l~ n l~, l~). A proof very similar to the one given in Theorem 3.3 implies that the image of cp contains the pencil determined by la provided q is part of a hyperbolic pair (p, q). These 26 202 RICHARD SCHWARTZ pencils are dense in Jg; hence so is the image of ~. Since the hyperbolic pairs are dense in the domain of % their image is dense in the range as well, which is to say that fixed flags are dense in Jg. Now for containment. Recall that M is the modular group of hyperbolic isometrics acting on I'. There are four possibilities for elements of M: order 2, order 3, parabolic and hyperbolic. Consider the corresponding elements of M. 1. By the proof of Theorem 2.4, these are polarities, and have no fixed flags. 2. By the proof of Theorem 2.4, these are conjugate to a Euclidean rotation. Again, no fixed flags. 3. The fourth power of such an element fixes pointwise the distinguished line of some marked box, and also has infinite order. By duality, this fourth power preserves all the lines through the distinguished point of the same marked box. All of these fixed points and lines give rise to two fixed pencils, both of which belong to the complement of Jff. These are the only fixed flags: any more would make the element trivial. 4. The proof of Lemma 3.3 implies that the square of this type of element has exactly 3 fixed points and exactly 3 fixed lines, making 6 flags fixed in all. All of these flags belong to the complement of J/'. 9 One consequence of Lemma 4.2.1 is that M acts freely on Jff. To show that ~U is a domain of discontinuity for M, we need to show that the action here is also proper. This is to say that the set { T ~ M : T(~g') n ~ :~ 0 } is finite for any compact ~ C ..4/'. Before proving this, we state the following equivalent version, which is easier to prove: There is a finite subset H e M such that, for any compact ~r and infinite S C M there is some element T e g and two elements hi, h2 e H such that hi Th2(JY") does not intersect ~. To recover the original statement, one simply applies the new criterion to compact subsets of the form HoU. The reason for this restatement is the following Lemma about the hyperbolic isometry group M. We will refer to the notion of depth used in w 3.1, and also to the fixed edge el described in w 3.1. Lemma 4.2.2. -- There is a finite subset H ~ M having the following property: let S C M be an infinite sequence of distinct elements. Then for each integer N > 0 there are elements hi, h 2 ~ H and an element T ~ S such that hi Th2(el) C va v2(el) and md(h~ Th2(el)) > N. Proof. -- It is an easy exercise in hyperbolic geometry to show that there is a finite set of elements H 1 C M such that all but finitely many elements of S are hyperbolic after composition with some element in H. In fact, it is easy to guarantee that the fixed points of these new hyperbolic elements remain far away from each other in the round metric on the circle. To be more concrete, we can assume that each element of S has fixed points in regions bounded by distinct and non-adjacent edges shown in Figure 4.2. Finally, after Conjugation with another finite set Ha of elements of M, we can guarantee PAPPUS'S THEOREM AND THE MODULAR GROUP 203 FIo. 4.2 that the expanding fixed point of every element of S is contained in the region labelled E, and that the contracting fixed point is contained in the region labelled C. Composing at most twice on the left and on the right by elements and inverses of elements in Hi, we can assume that every element in S expands about a point in E and contracts about a point in C. Most of these elements will have very large translation lengths, and such elements will clearly satisfy the conclusion of the Lemma. 9 Based on Lemma 4.2.2, and the reformulation of the properness criterion, we can without loss of generality assume for the rest of this section that each transformation T ~ M which we consider has the following properties: 1. T has a repelling fixed point in A -- E 1. 2. T has an attracting fixed point in A n Vl v,(E1). 3. T(E1) C vl v2(E1). Recall that 9 is the " spherical metric " on P. Let N~(X) be the r neighborhood about the set X e P. Ifp ~ A, then l~ will denote the line of L which contains p. Lemma 4.2.3. -- Let { T k } be any infinite sequence of distinct projective transformations in M. For any ~ > 0 there is some T, e { T }, k and points p, q ~ A (not necessarily distinct) such T,(P -- N~(/,)) C N~(p) and T,(P -- N~(q)) C N,(/~). Proof. -- Since the fixed points of T k are contained in two disjoint compact subsets, we can normalize so that i. Pk is the origin and qk is at infinity, on they-axis; 2. l~k is the x-axis, and lqk is the line at infinity; 3. the convex interior of -r 1 v2(E1) contains a fixed small square centered at the origin, and is contained in a fixed larger square centered at the origin. 204 RICHARD SCHWARTZ Under this normalization, a sufficiently large ball about the origin contains the complement of N,(lq). Likewise, the set of lines having absolute slope less than a suffi- ciently large (but fixed) number contains the complement of N.(q). Finally, we have Y k ~ ~ Let E' : vl ~2(E1). For most values of k, T,(E') will have large minor depth, and hence by the Depth Lemma of w 3.1, Tk(E' ) will be "small and thin ". Since E' is close in size and shape to the unit square, X,, ~,, and tz,[X, are all very close to zero for large n. In particular, T, contracts a large ball about the origin into a very small one, and contracts lines of bounded large absolute slope to lines having very small absolute slope. 9 Theorem 4.2. -- The set JV" is the domain of discontinuity for the group action M. Proof. -- Let 7:1 and r~ z be the obvious projections from ~ to P and P* respectively. Let dC a denote the set of flags in of whose projections under ~1 are not contained in N~(I,) u N,(p). Dually let 9r denote the set of flags in off whose projections under do not intersect N,(q) u N~(I~). Since Off is compact and is contained in ~4/', the two sets oft cover of once a is small enough. Let { T~ } be an infinite sequence of transfor- mations which (putatively) does not move ~ completely off itself. By Lemma 4.3.3, nl(T,(ofx) ) C N:(p) and ~z(T,(~T'~)) C N~(/~). But then T,(Of~) c~ ~(" = O. Since these two sets cover of, we have T,(Of) c~ 9U = O, a contradiction. We have shown that the fixed points of M are dense in, and contained in the complement ofM r. We have now just seen that M acts properly and freely on~.V'. HenceJV" is a domain of discontinuity forM. 9 We emphasize that this domain X varies with the initial choice of convex marked box. Usually, it is a set with a " fractal boundary " 4.3. TREFOIL KNOT COMPLEMENT Since M r is a domain of discontinuity for M, the quotient manifold $"---- X/M is a three dimensional analytic manifold. $" has a (re, Y') structure in the sense of IT]. Here :Y is the simply connected cover of the projective tangent bundle. In this section, we will prove Theorem 4.3. -- The manifold $" is homotopy equivalent to the complement of the trefoil knot in the three sphere. Before proving Theorem 4.3, we will analyze the bilaterally symmetric case. The analysis here is related to the classical fact that SL2(R)/SL2(Z) is homeomorphic to the trefoil knot complement. To prove the general case, we will show that nothing on the level of homotopy changes as we deform away from the symmetric case. PAPPUS'S THEOREM AND THE MODULAR GROUP 205 If the marked box has bilateral symmetry, then the curve A will be a line of P, and the curve L will consist of coincident lines. We can conjugate by a projective transformation so that A is the line at infinity and L is the set of lines going through the origin in R *. Then the set ~" consists of flags (p, P), where p ~ R * -- { 0 } and P is a line through p which misses the origin. Thus J/" is homeomorphic to R * � 1. S The modular group M=(a,~:a*-----~3-~ 1) acts onJP. ~ is an order three linear map of R * and ~ can be taken as the polarity induced by the standard inner product on 11~. The universal cover ~4 ~ has a description as the set of " flags " (p, P), where p is a point in the universal cover of the punctured plane, and l is a line through p missing the origin. Let c denote the covering transformation of X. The " rotation " ~ M acts on ~d ? by setting ~8 _ c. The polarity a has the following concrete description: take the point p on the unit circle to the tangent line through the antipodal point p' in the unit circle. From this description, we can clearly lift ~ to an action on ~47 which satisfies ~2 = c. Thus, the c-extension of the modular group acts on .A ~. The quotient of this lifted action is the original manifold ~.. Hence g- is the quotient of R 3 by the c-extension of the modular group, which is the trefoil knot complement fundamental group. To analyze the general case, we introduce some notions from point set topology. Let M be a topological 3-manifold. An embedded surface S C M is tamely embedded if each point on the S has a small neighborhood so that the pair (N n S, N) is homeo- morphic to the standard pair (B *, B~), where B ~ is the open unit j ball. Two tamely embedded surfaces S t intersect tamely if each point of the intersection has a neighborhood N such that the triple ($1 n N, S,, n N, N) is homeomorphic to the triple (B~, B~, B3), and if these homeomorphisms are all compatible with the homeomorphisms to standard pairs which take place away from the intersection. Here B~ and B~ are orthogonal open unit disks. Here is a technical fact: Suppose M is a closed topological 3-manifold and A~ and B t are two continuous families of tamely embedded and tamely intersecting closed surfaces. Then the homeomorphism type of M -- A t -- B e remains unchanged as t varies. We omit the proof because it is technical, and would lengthen the exposition considerably. Proof of Theorem 4.3. -- Let d C ~ consist of those flags whose first coordinate is contained in A. Let ~ C ~ consist of those flags whose second coordinate is contained in L. Then J/" ----- ~ -- ~r -- ~. The surface d is a bundle of projective lines sitting over A; it is closed, and ruled in the sense of section w 4.1. Dually for ~. In fact, in the local coordinates shown in Figure 4.3, the two surfaces are locally both ruled surfaces of R 3 in the ordinary sense. Furthermore, they intersect exactly in a curve. It is a standard exercise to show that locally such surfaces are tamely embedded, and intersect tamely, in R 3. Hence the same is true globally, in ~. Taking a family of marked boxes varying 206 RICHARD SCHWARTZ FIO. 4.3 continuously from a bilaterally symmetric one to the one under study, we see that Jg" is homeomorphic to S 1 � R ~, as in the symmetric case. Finally, the two perturbed elements e and ~ may be lifted to the universal cover, since ~ is conjugate to a rotation and e is still a polarity. The identity ~2= ~3__ c cannot change either. Hence the trefoil knot fundamental group acts on the universal cover of Jt r, which is just R 3. The quotient is ~, as desired. I REFERENCES [B] A. BEARDON, The Geometry of Discrete Groups, Springer-Verlag, 1983. [H] HmBE~T and COHN-VossEN, Geometry and the Imagination, Chelsea Publishing Company, 1950. IT] W. THURSTON, The Geometry and Topology of Three-Manifolds, Princeton University Lecture Notes, 1978. Mathematics Department UCSD La Jolla, California 92093 USA Manuscrit refu le 11 juin 1992. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Publications mathématiques de l'IHÉS Springer Journals

Pappus’s theorem and the modular group

Loading next page...
 
/lp/springer-journals/pappus-s-theorem-and-the-modular-group-box0imksTr

References (2)

Publisher
Springer Journals
Copyright
Copyright © 1993 by Publications Mathématiques de L’I.É.E.S.
Subject
Mathematics; Mathematics, general; Algebra; Analysis; Geometry; Number Theory
ISSN
0073-8301
eISSN
1618-1913
DOI
10.1007/BF02712918
Publisher site
See Article on Publisher Site

Abstract

by RICHARD SCHWARTZ 1. Introduction Pappus's theorem is as old as the hills. It refers to the configuration of points and lines shown in Figure 1.1. Fro. 1.1 Pappus's Theorem. ~ If the points A, B, and C are collinear and the points A', B' and C' are cotlinear, then the points A", B" and C" are cotlinear. A slight twist makes this theorem new again. The twist is to iterate, and thereby treat Pappus's theorem as a dynamical system. Pappus's theorem may be considered as a dynamical system defined on objects called marked boxes. Essentially a marked box is a collection of points and lines in the projective plane P which comprises the initial data for Pappus's theorem. (See w 2.2 for a precise definition.) When Pappus's theorem is applied to a marked box, more lines and points are produced. These new lines and points may be used to form new initial data for the theorem, and so on. In fact, one can construct an entire group G of operations to perform on marked boxes. Algebraically G is the modular group, Z/2 9 Z/3, and is generated by the simple operations alluded to above. (For a precise definition of the operations, see w 2.3.) The 188 RICHARD SCHWARTZ orbit ~ of a marked box is an infinite collection of marked boxes, nested inside and outside of each other in a pattern encoded by the modular group. The modular group makes a second entrance into Pappus's theorem, as the group of projective symmetries of ~. Projective transformations are analytic diffeomorphisms of P which take collinear points to collinear points. Projective dualities are analytic diffeomorphisms between P and its dual space of lines P* which take collinear points to coincident lines. Projective transformations and dualities together generate the Lie Group of projective symmetries of P. It turns out that the projective symmetry group M of ~ (those projective symmetries which permute the marked boxes of ~)) is again algebraically the modular group. The group of projective symmetries and the group of operations commute, and so there are two commuting modular group actions on ~. As the original marked box varies, the orbit varies, along with the two group actions. Everything varies smoothly, thereby producing two commuting modular group actions on the space of all marked boxes. There is a certain geometric condition one can put on marked boxes, called convexity. The set of convex marked boxes forms an open subset of the set of all marked boxes. For these marked boxes, there is a certain amount of geometry and topology associated to the group actions described above. In the convex case, there is a fractal curve associated to the orbit ~. Certain " distinguished " points of the marked boxes in ~) are dense in a topological circle A, which is generally non-smooth. Dually, certain " distinguished " lines of the marked boxes in Y~ are dense in a topological circle L of lines. The circles A and L are " self- projective " curves in the sense that they are preserved (or swapped) by the modular group of projective symmetries of ~2. In a certain sense, they are analogues of quasi- circles in the projective plane. Figure 1.2 shows certain of the convex marked boxes in an orbit. Both A and L are " hinted at " in the picture. Projective symmetries induce analytic diffeomorphisms of the projective tangent bundle ~ of the projective plane. When the original marked box is convex, the symmetry group M gives a discrete group action on ~. In fact this group action has a domain of discontinuity and a corresponding quotient 3-manifold. More precisely, let Jff C denote those pairs (p, l) such that p r A and l r L. The fixed points of M are dense in, and contained in, the complement of Jff, and .h r itself is the domain of discontinuity for M. The quotient $" ---- ~r/M is a three-dimensional analytic manifold with the homotopy type of the trefoil knot complement in the three sphere. The manifold $" is a " (~, s " in the sense of IT]. Here fr is the Lie Group of projective symmetries, and s is the universal to the projective tangent bundle ~. There is a classical geometric structure on the trefoil knot complement coming from its description as SL~(R)/SL,(Z). In the case of a " totally symmetric " marked box, one essentially recovers this structure. However, as the marked box varies, one obtains a 2-parameter family of " exotic " geometric structures on the (homotopy) trefoil knot complement. PAPPUS'S THEOREM AND THE MODULAR GROUP 189 FIc. 1.2 In sum, each initial choice of lines and points involved in Pappus's theorem (subject to the convexity condition) gives rise to the following objects: 1. an infinite collection ~ of marked boxes, which is indexed by the modular group, and which has the modular group as its group of projective symmetries; 2. a pair of (usually fractal) curves A ~ P and L ~ P* which are invariant (or swapped) under a modular group of projective symmetries; 3. a representation M of the modular group into the Lie Group of symmetries of the projective tangent bundle ~; 4. a (usually " exotic ") (if, 5~)-structure $" --JV'/M on the (homotopy) trefoil knot complement. This paper is self-contained, but does assume some basic hyperbolic and projective geometry. The hyperbolic geometry can be found in [B], and the projective geometry, as well as an excellent treatment of Pappus's theorem, can be found in [H]. 190 RICHARD SCHWARTZ Here is a list of contents: w 2: Iterating Pappus's Theorem: w 2.1: The Projective Plane w 2.2: Marked Boxes w 2.3: Operations on Marked Boxes w 2.4: Two Commuting Group Actions w 3: Geometry of Marked Boxes w 3.1: Geometry and Depth w 3.2: The Pappus Curve w 3.3: Curve and Linefield w 4: The Three Manifold w 4.1: The Projective Tangent Bundle w 4.2: Domain of Discontinuity w 4.3: Trefoil Knot Complement I would like to thank Hsin-Ta (Frank) Cheng, Peter Doyle, Bill Goldman, Jun Hu, Linda Keen, and Dennis Sullivan for helpful and interesting conversations concerning this work. I would also like to thank the IHES, and the Einstein Chair at CUNY, for their support during the genesis of this paper. 2. Iterating Pappus's Theorem 2.1. THE PROJECTIVE PLANE As usual, we will take the projective plane to be the space P of one-dimensional subspaces of R 3. The ordinary plane R * sits naturally as a subset of P, and consists of all one dimensional subspaces which do not lie on the xy-plane of R 8. Projective transformations of P are just projectivized linear transformations of R 3. The group of projective transformations is naturally isomorphic to SL3(R ). There is a unique projective transformation which takes one specified set of 4 (ordered) general position points to another such set. P* will denote the space of lines of P. This set is naturally in bijection with P. A duality is just a projectivized linear transformation from R s to its dual space. More concretely, a duality is a homeomorphism P-+ P* which takes collinear points to coincident lines. As a special case, a duality which is induced by a positive definite inner product on R ~ is called a polarity. Each duality A induces a dual map A* : P* -+ P, defined by the formula A*(~) = A(p) c~ A(q). A set Q c P is convex if there is a projective transformation T such that T(Q) is a convex subset of the ordinary plane. PAPPUS'S THEOREM AND THE MODULAR GROUP 191 2.2. MARKED BoxEs Informally, a convex marked box is a convex quadrilateral in P with a distinguished top edge, a distinguished bottom edge, a distinguished top point and a distinguished bottom point. Here are some definitions, which culminate in a formal definition of a convex marked box, the main object of study. An overmarked box is a pair of 6 tuples ((p, q, r, s; t, b), (P, Q, R, S; T, B)) having the incidence relations shown in Figure 2.2. There is an involution on the set of overmarked boxes: ((p, q, r, s; t, b), (P, Q, R, S; T, B)) ~ ((q,p, s, r; t, b), (Q, P, S, R; T, B)) A marked box is an equivalence class of overmarked boxes under this involution. Let | be the marked box labelled as in Figure 2.2. The top of t9 is the pair (t, T). The bottom of | is the pair (b, B). The distinguished edges of 19 are T and B. The distinguished points of | are t and b. Fio. 2.2 So far, the definitions make sense in a projective plane over any field. (The exception is the projective plane over Z/2, which doesn't have enough points to contain a marked box.) The remainder of the definitions depend on the ordering properties of R. The marked box | is convex if the following 4 conditions hold: 1. p and q separate t and T n B on the line T. 2. r and s seperate b and T c~ B on the line B. 3. P and Qseperate T and bt on the pencil of lines through t, in the cyclic ordering of lines through b. 4. R and S seperate B and bt on the pencil of lines through b. The first two conditions imply the last two, and vice versa. The convex interior of @ is the open convex quadrilateral whose vertices, in cyclic order, are p, q, r and s. 192 RICHARD SCHWARTZ Projective symmetries act on marked boxes in a way which preserves convexity: Let ~ = T(x) for a projective transformation T. We define: T(| ---- ((/~, ~, r. ?; t, b), (P, Q, R, S; T, B)). Similarly, given a duality A, let x* = A(x) when p is a point, and X* = A*(X) when X is a line. We define: A(| = ((P*, Q*, S*, R*; T*, U*), (q*,p*, r*, s*; t*, b*)). The reordering of elements of A(O) is deliberate. 2.3. OPERATIONS ON MARKED BOXES There are three natural box operations one can perform on marked boxes. Before defining these operations, we introduce some notation. If p, q e P, then pq is the line containing p and q. Likewise, if P and Q are distinct lines in P, then PQis the intersection of P and Q. This notation may be applied recursively; for example, (ab) (cd) is the intersection point of the lines ab and cd. Given | = ((p, q, r, s; t, b), (P, Q., R, S; T, B)) as above, we define: i(| = ((s, r, p, q; b, t), (R, S, Q, P; B, T)). 9 1(| ---- ((P, q, QR, PS; t, (qs) (pr)), (P, Q, qs, pr; T, (QR) (PS))). ":3(| = ((QR, PS, s, r; (qs) (pr), b), (pr, qs, S, R; (QR) (PS), B)). Finally, for completeness, we define the identity operation 1(| = | The three box operations are shown schematicaly in Figure 2.3, for a convex marked box. Fro. 2.3 PAPPUS'S THEOREM AND THE MODULAR GROUP 193 The box operations may be applied iteratively, to form a semigroup G of box operations. We will use the notation ab to mean: "first apply b, then apply a ". It is clear that i and -r~ commute with projective transformations. Two simple computations, one for i and one for (say) ~1, show that dualities commute with these box operations as well. Hence the semigroup of box operations commutes with projective symmetries. Lemma 2.3. -- The following relations always hold: i S = 1 ; "q i'~, = i; "q i~ t = i; v~ izt = v~; v2 iv2 = "q. Proof. -- The first relation is obvious. By symmetry, we only have to verify the third and fourth relations. We compute: i,1(| = ((PS, QR, p, q; (qs) (pr), t), (qs, pr, Q, P; (QR) (PS), T)). ,q i-q(O) = ((PS, QR, (pr) Q, (qs)P; (QR) (PS), (q(QR)) (p(PS))), (...)). "r, i'q(| = ( (pr) ( Q, (qs) P, q, p; ((PS) p) ((QR) q), t), (...)). Looking at Figure 2.2, we see that (pr) Q = r; (qs) P = s; (q(QR)) (p(PS)) = RS = b. Substituting these identities in, we see that the 6 points of-q i~1(0 ) match the 6 points of ~(| and the 6 points of ** iv1(| ) match those of i(| To complete the proof, we note that the 6 points of a marked box clearly determine the entire marked box. 9 The relations in Lemma 2.3 imply that the semigroup G is in fact a group. Clearly G is generated by ~ = i and ~ = iv1. Note that ~3 = iT 1 i.~l iv 1 = iv 1 i.~2 = 1. Looking at the nesting properties of the convex interiors of the marked boxes in the orbit of | in the case | is convex, we see that the two elements ~ = vl and 0c~ ~ = v2 generate a free semigroup. This is enough to identify G as the modular group: G= (~,~:~*= 1,~a= 1). 2.4. Two COMMUTING GROUP ACTIONS Let f2 = f~(| be the orbit of | under G. It is not hard to see, particularly in the convex case, that G acts freely, faithfully, and transitively on fL The structure off~ can be expressed by an incidence graph F. The edges of P correspond to marked boxes in f~, the vertices correspond to tops and bottoms of boxes, and each edge is directed from the top to the bottom. Vertices on distinct edges are identified if the corresponding distinguished sides coincide. A picture of P is shown in Figure 3.1. I. Technically, P is just an abstract graph, but we will always embed P in the hyperbolic plane, as the tiling associated to the hyperbolic reflection group generated by reflections in an ideal hyperbolic triangle. 25 194 RICHARD SCHWARTZ The group G of box operations has a concrete description as a permutation subgroup of edges of I'. The operation i reverses the orientations on each edge. The element -c 1 " rotates " each edge counterclockwise one " click " about the tail point. The element z2 " rotates " each edge one " click " clockwise about the head point. Note that G is not a group of homeomorphisms of the hyperbolic plane. Each marked box induces a faithful representation G of G into the permutation subgroup of marked boxes in the orbit ~. Besides G, there is a second, commuting, modular group action M on I'. This action M is generated by order 2 isometric rotations about centers of edges of U together with order 3 isometric rotations about centers of triangles in U. The marked box | induces a representation M into the Lie Group ~ of projective symmetries of P. Theorem 2.4. -- Let | be a marked box, with orbit ~. There is a faithful representation ~r : M --~ (~ which takes isometrics of I" to projective symmetries of ~ in a way which is natural with respect to the labelling of U. Proof. -- We will show that, for each marked box 9 e ~: 1. There is an order 3 projective transformation having the cycle i(~) ~ ~1(~) -> ~2(~V). 2. There is a polarity having the cycle 9 ~ i(~). The theorem then follows from the fact that everything in sight commutes. Let 9 = ((bl, b3, a3, al; a~, b2) , (...)). Figure 2.4.1 shows a particular norma- lization of ~. We compute i(tF) = ((al, aa, bt, b3; a~, b,), (...)), vl(~F) = ((61, ba, cl, c8; bz, c,), (...)), = ((cl, ca, al, as; c,, (...)). The obvious Euclidean rotation satisfies (1). a2 b3 c3 Fxo. 2.4.1 PAPPUS'S THEOREM AND THE MODULAR GROUP 195 s! \ rl Fie,. 2.4.2 Now let V ---- ((p, q, r, s; t, b), (P, Q, R, S; T, B)). Figure 2.4.2 shows a norma- lization of q~ in which t and b are at infinity, and [IP II [I q ]1 ----- ]1 r II I] s ]] = 1 in the standard norm on the Euclidean plane. The polarity induced by the standard inner product on 113 satisfies (2). 9 All in all, we have produced two different commuting modular group actions G and M on the orbit ~. As the original marked box varies, the orbit varies, and these group actions vary as well. Thus we produce two different commuting group actions on the space of marked boxes. 3. Geometry of Marked Boxes Henceforth, we make the blanket assumption that all our marked boxes are convex. 3.1. DEPTH OF A Box The labeling of the incidence graph I ~ is determined by the choice of O in the orbit ~. It can be arranged that the edges corresponding to i(| ~1(| and "r2(| ) bound the hyperbolic triangle which contains the origin 0 C 112. Given this labelling, we make the following definitions concerning the placement of other boxes in f~ with respect to | Let v be a vertex of P. Say that the depth of v is the minimum number of edges of U which must be crossed by any path from the origin to v. The depths of various vertices are shown in Figure 3.1.1. Each directed edge e ~ U has a half space H, associated to it. The assignment e -+ H e is made so that H, ~ Hf if and only if the convex interior of the marked box labelled by e is contained in the convex interior of the marked box labelled by f. Let H1, H 2 and H 3 be the three half spaces shown in Figure 3.1.1. Assuming that H, C Hj for some j ~ { 1, 2, 3 }, define the major depth of the edge e to be the maximum depth of the two vertices bounding e. Define the minor depth of e to be the minimum depth of its two 196 RICHARD SCHWARTZ 2 i __ 2 1 1 el .0 2 e2 e3 ~ ~. ' 0 FIG. 3.1.1 vertices. There are only finitely many edges at every major depth, and infinitely many at each minor depth. The directed edges of P are in bijection with the marked boxes of f2, so we can transfer the notion of major and minor depth to marked boxes. We will use the nota- tion md(tF) and Md(tF) to denote the minor, and major depth of tF. The purpose of this section is to make precise the statements that boxes having high depth are small and thin. The projective plane inherits a Riemannian metric p from its description as the sphere modulo antipodal points. The metric is not very natural with respect to the full group of projective transformations, but it will nonetheless be useful to us here and in w 4. Unless otherwise stated, all metric measurements (distances and angles) will be made with respect to p. On the other hand, when we speak of convexity, we will always refer to the projective geometric notion, described in w 2.1. Given a marked box ~F ~ f2, let ~F denote the convex interior of ~F, let ~F denote the (smaller) convex quadrilateral shown in Figure 3.1.2, and let a(~) denote the angle shown in Figure 3.1.2. IX I will denote the 9-diameter of the set X C P. ,..', FIG. 3.1.2 PAPPUS'S THEOREM AND THE MODULAR GROUP 197 Depth Lemma. -- Let ~ > 0 be fixed. Then there is a constant N = N(e, O) such that, for any marked box ~ ~ 1. md( r) > N * leVI< 2. Md(~F) > N => t~'[< ~; Proof. -- Without loss of generality, we assume [~l ~ [E x [, and we normalize so that Ex is the unit square. The spherical metric p restricted to the unit square is only boundedly different from the Euclidean metric there, so for ease of computation we will work with the Euclidean metric. This bounded change, and the original norma- lization, merely alter the constants in the Lemma. We may write ~ = w(Et) , where w is a word in the box operations v s. By throwing out uniformly small words, and using symmetry (between -r 1 and .r~) we can assume that w has one of the following properties: 1. The string ... -q v~ ... occurs n times in w, for large n. " for large n. 2. W ~ V2%'1, 3. w ---- v~, for large n. In the first two cases, both md(~) and Md(~) are large. In the third case, only Md(~) is large. Case 1. -- Recall that Et is the unit square. Suppose F ~ ~ is any marked box with FCE 1. We will show that there is a constant ~-~ ~(| 1 such that r d FIG. 3.1.3 Consider the diagonal d shown in Figure 3.1.3. If d is chosen correctly, [~r~[ =ld]. Also, IF[>/ [dc~Fl" All the boxes in ~ are equivalent under a projective symmetry, so the cross ratio of the 4 points shown in Figure 3.1.3 takes on one of finitely many values. Here is a useful trick: Let a, < b, < c, < d, C R be an infinite sequence of 4-tuples of points. Then the ratio l c, -- b, Ilia, -- d,I cannot converge to 1 if the cross ratio 198 RICHARD SCHWARTZ ~, = [a., b,, c,, d,] takes values in a finite set. This trick bounds the diameter of the smaller quadrilateral away from the diameter of the bigger one. Thus [ (~) [ ~< I VF I ~< 0". The argument for ~(W) is similar: one uses a similar trick to show that x(vx v,(F)) ,< ~' a(F), for some other constant ~' < 1. Case 2. -- From Theorem 2.4, there is a projective transformation T which has the infinite orbit E1 ~ v~(E1) ~ x~(E1) .., This map T fixes the distinguished point p at the top of El, and takes the open unit square into itself. It follows easily that T has no fixed points in the open unit square. Since T preserves the two lines x and y shown in Figure 3.1.4, % z~(Ex) = Tk(%(E1)) shrinks as a set to the point p as k grows. Hence ~" and ~ are shrinking to points as well. Fio. 3.1.4 The lines I, and m,, shown in Figure 3.1.4, converge to the top line of El, squeezing the angle ~ down to zero. Case 3. -- The argument here is the similar to that in Case 2, except that the top edge of v~(E1) does not shrink to a point. The only difference in this case is that I CFI remains large. Stringing the three cases together proves the Lemma. 9 3.2. THE PAPPUS CURVE The marked box 6) determines a natural map X from vertices of the incidence graph F into the projective plane. The rule is that ?,(v) is the common distinguished point on all the marked boxes labelled by edges in F emanating from v. Part of the image of X is " hinted at" in Figure 1.2. Recall from w 2.4 that M is the representation of the hyperbolic isometry (modular) group M into the Lie Group of projective symmetries. The index 2 subgroup of M' C M consisting of those group elements taken to projective PAPPUS'S THEOREM AND THE MODULAK GROUP 199 transformations preserves the image A of X. Furthermore, the vertices of P are dense in S x, and it is natural to ask whether or not X extends continuously. If this is true, then the whole image A -- X(S a) is invariant under M'. Theorem 3.2. ~ The map extends to give a continuous homeomorphism from S t onto its M'-equivariant image A C P. This homeomorphism conjugates the action of M' to that of M'. Proof. ~ Recall that ~ is the small convex quadrilateral shown in Figure 3.1.2. The image vertices of X are arranged according to Figure 3.2. Since the nesting properties of the edges of F exactly reflect the nesting properties of the convex interiors of boxes in ~, the following fact is sufficient to prove the theorem: let ea, e2, ... be any infinite sequence of nested edges in F. (This means that H~j~ H~i+,. ) Let tF~ be the corres- ponding marked box. Then I~1 ~ 0 as j -+ oo. But this is an immediate consequence of the Depth Lemma. m FIo. 3.2 3.3. CURVE AND LINEFIELD The dualities of M- M' take the curve A to a curve of lines L, which contains the tops and bottoms of the marked boxes in the orbit ~. The curve L is " hinted at " in Figure 1.2. The two objects L and A not only are dual to each other, but also have a special geometric relationship. A transverse linefield to A is a continuous curve X of lines such that each line of X intersects A in exactly one point, and that each point of A is contained in some line of X. Dually, a section of L is a curve X such that each point of X is contained in exactly one line of L, and each line of L contains exactly one point of X. From the nesting properties of the marked boxes in the orbit ~, it follows that L is a transverse linefield to A, and dually A is a section to L. 200 RICHARD SCHWARTZ Theorem 3.3. -- If A is not a straight line, then L is the unique transverse linefield to A. Dually, if L is not a curve of coincident lines, then A is the unique section to L. Proof. -- We will prove the first statement. By Theorem 3.2, the continuous extension ~ of ~ conjugates the action of M' on S 1 to the action of M' on A. In particular, M' has a dense set of fixed points in A. By continuity, we just have to produce a single point p e A at which any two transverse linefields must agree. Say that a pair of points (p, q) E A � A is a hyperbolic pair if there is a nontrivial element T e M' which fixes both p and q. Such pairs are dense in A x A because the hyperbolic translations of M' have fixed points which are dense in S i X i. S Such an element T attracts about one of the fixed points, say p, and expands around the other one. Since A is not a straight line (and not convex either) there are two points p, q e A such that any line sufficiently close to p-q, emanating from p, intersects A in some point other than p. By density, we can assume that (p, q) is a hyperbolic pair. The situation is shown in Figure 3.3. m' Fio. 3.3 Let l~ be the line of L through p and let lq be the line of L through q. Then T preserves p-q, l~, lq, and fixes the point l~ n l~. Let m be any fourth line through p. We will prove below (Lemma 3.3) that T"(m) converges to l~. Now, if m is sufficiently close to p-q, then m will intersect A elsewhere. By equi- variance, T"(m) will intersect A elsewhere as well. By continuity, all lines containing p between m and T"(m) will intersect A elsewhere. Letting n ~ ~, and m ~pq, we see that all lines between pq and l~ which contain p will intersect A elsewhere. The same argument may be repeated for lines m' on the other side of~. Hence, every line through p except l~ intersects A in at least two points. 9 PAPPUS'S THEOREM AND THE MODULAR GROUP 201 Lemma 3.3. -- Suppose T ~ M' fixes two distinct points p, q ~ A, and attracts about p. Suppose m is a line through p distinct from l~ andS. Then T"(m) converges to l~. Proof. -- The line m intersects l~ at a point x ~ la. This intersection point is neither q or l~ c~ lq. If T"(m) failed to converge to l~, then T would fix x. Fixing three distinct points on l~, T would be the identity on lq. By looking at the conjugated action of T back on S 1, we observe the following fact: if W is any marked box such that q ~ VF, then the open convex interior of T(VF) contains the closure of the convex interior of W. This is not consistent with T fixing Lq pointwise. 9 4. The Three-Manifold 4.1. THE PROJECTIVE TANGENT BUNDLE The projective tangent bundle ~ to the projective plane is the closed analytic 3-manifold of pairs (p, ~), such that p ~ P, P ~ P*, and p ~ P. We will call points of flags. The Lie Group ff of projective symmetries acts as a group of analytic diffeo- morphisms of ~, as follows: T((p, P)) ---- (T(p), T(P)) for a projective transformation T, and A((p, P)) = (A*(P), A(p)) for a duality A. We will say that a pencil of ~ is a set of flags, such that one of the two coordinates varies while the other one takes on all possible values. The objects with which we will deal, in w 4.3 below, are ruled surfaces in the sense that they can be expressed as the union of continuously varying disjoint pencils. 4.2. DOMAIN OF DISCONTINUITY The representation M constructed in Theorem 2.4 gives an action on ~ by analytic diffeomorphisms. Let JV C ~ denote the set of pairs (p, P) such that p r A and P r L. The representation M preserves both J/" and its complement. Lemma 4.2.1. -- The fixed points (flags) of M are dense in, and contained in, the complement of~V. Proof. -- We will prove density first, then containment. Consider the set J/ of pairs (p, P), such that P e L. This set makes up essentially " half" of the complement of.A r. It is sufficient to show that the fixed flags are dense in dr'. There is a continuous map : A � A -- Diagonal --> defined as follows. Let l~ and lq be the lines of L which contain the two points p, q e A. Define q~(p, q) = (l~ n l~, l~). A proof very similar to the one given in Theorem 3.3 implies that the image of cp contains the pencil determined by la provided q is part of a hyperbolic pair (p, q). These 26 202 RICHARD SCHWARTZ pencils are dense in Jg; hence so is the image of ~. Since the hyperbolic pairs are dense in the domain of % their image is dense in the range as well, which is to say that fixed flags are dense in Jg. Now for containment. Recall that M is the modular group of hyperbolic isometrics acting on I'. There are four possibilities for elements of M: order 2, order 3, parabolic and hyperbolic. Consider the corresponding elements of M. 1. By the proof of Theorem 2.4, these are polarities, and have no fixed flags. 2. By the proof of Theorem 2.4, these are conjugate to a Euclidean rotation. Again, no fixed flags. 3. The fourth power of such an element fixes pointwise the distinguished line of some marked box, and also has infinite order. By duality, this fourth power preserves all the lines through the distinguished point of the same marked box. All of these fixed points and lines give rise to two fixed pencils, both of which belong to the complement of Jff. These are the only fixed flags: any more would make the element trivial. 4. The proof of Lemma 3.3 implies that the square of this type of element has exactly 3 fixed points and exactly 3 fixed lines, making 6 flags fixed in all. All of these flags belong to the complement of J/'. 9 One consequence of Lemma 4.2.1 is that M acts freely on Jff. To show that ~U is a domain of discontinuity for M, we need to show that the action here is also proper. This is to say that the set { T ~ M : T(~g') n ~ :~ 0 } is finite for any compact ~ C ..4/'. Before proving this, we state the following equivalent version, which is easier to prove: There is a finite subset H e M such that, for any compact ~r and infinite S C M there is some element T e g and two elements hi, h2 e H such that hi Th2(JY") does not intersect ~. To recover the original statement, one simply applies the new criterion to compact subsets of the form HoU. The reason for this restatement is the following Lemma about the hyperbolic isometry group M. We will refer to the notion of depth used in w 3.1, and also to the fixed edge el described in w 3.1. Lemma 4.2.2. -- There is a finite subset H ~ M having the following property: let S C M be an infinite sequence of distinct elements. Then for each integer N > 0 there are elements hi, h 2 ~ H and an element T ~ S such that hi Th2(el) C va v2(el) and md(h~ Th2(el)) > N. Proof. -- It is an easy exercise in hyperbolic geometry to show that there is a finite set of elements H 1 C M such that all but finitely many elements of S are hyperbolic after composition with some element in H. In fact, it is easy to guarantee that the fixed points of these new hyperbolic elements remain far away from each other in the round metric on the circle. To be more concrete, we can assume that each element of S has fixed points in regions bounded by distinct and non-adjacent edges shown in Figure 4.2. Finally, after Conjugation with another finite set Ha of elements of M, we can guarantee PAPPUS'S THEOREM AND THE MODULAR GROUP 203 FIo. 4.2 that the expanding fixed point of every element of S is contained in the region labelled E, and that the contracting fixed point is contained in the region labelled C. Composing at most twice on the left and on the right by elements and inverses of elements in Hi, we can assume that every element in S expands about a point in E and contracts about a point in C. Most of these elements will have very large translation lengths, and such elements will clearly satisfy the conclusion of the Lemma. 9 Based on Lemma 4.2.2, and the reformulation of the properness criterion, we can without loss of generality assume for the rest of this section that each transformation T ~ M which we consider has the following properties: 1. T has a repelling fixed point in A -- E 1. 2. T has an attracting fixed point in A n Vl v,(E1). 3. T(E1) C vl v2(E1). Recall that 9 is the " spherical metric " on P. Let N~(X) be the r neighborhood about the set X e P. Ifp ~ A, then l~ will denote the line of L which contains p. Lemma 4.2.3. -- Let { T k } be any infinite sequence of distinct projective transformations in M. For any ~ > 0 there is some T, e { T }, k and points p, q ~ A (not necessarily distinct) such T,(P -- N~(/,)) C N~(p) and T,(P -- N~(q)) C N,(/~). Proof. -- Since the fixed points of T k are contained in two disjoint compact subsets, we can normalize so that i. Pk is the origin and qk is at infinity, on they-axis; 2. l~k is the x-axis, and lqk is the line at infinity; 3. the convex interior of -r 1 v2(E1) contains a fixed small square centered at the origin, and is contained in a fixed larger square centered at the origin. 204 RICHARD SCHWARTZ Under this normalization, a sufficiently large ball about the origin contains the complement of N,(lq). Likewise, the set of lines having absolute slope less than a suffi- ciently large (but fixed) number contains the complement of N.(q). Finally, we have Y k ~ ~ Let E' : vl ~2(E1). For most values of k, T,(E') will have large minor depth, and hence by the Depth Lemma of w 3.1, Tk(E' ) will be "small and thin ". Since E' is close in size and shape to the unit square, X,, ~,, and tz,[X, are all very close to zero for large n. In particular, T, contracts a large ball about the origin into a very small one, and contracts lines of bounded large absolute slope to lines having very small absolute slope. 9 Theorem 4.2. -- The set JV" is the domain of discontinuity for the group action M. Proof. -- Let 7:1 and r~ z be the obvious projections from ~ to P and P* respectively. Let dC a denote the set of flags in of whose projections under ~1 are not contained in N~(I,) u N,(p). Dually let 9r denote the set of flags in off whose projections under do not intersect N,(q) u N~(I~). Since Off is compact and is contained in ~4/', the two sets oft cover of once a is small enough. Let { T~ } be an infinite sequence of transfor- mations which (putatively) does not move ~ completely off itself. By Lemma 4.3.3, nl(T,(ofx) ) C N:(p) and ~z(T,(~T'~)) C N~(/~). But then T,(Of~) c~ ~(" = O. Since these two sets cover of, we have T,(Of) c~ 9U = O, a contradiction. We have shown that the fixed points of M are dense in, and contained in the complement ofM r. We have now just seen that M acts properly and freely on~.V'. HenceJV" is a domain of discontinuity forM. 9 We emphasize that this domain X varies with the initial choice of convex marked box. Usually, it is a set with a " fractal boundary " 4.3. TREFOIL KNOT COMPLEMENT Since M r is a domain of discontinuity for M, the quotient manifold $"---- X/M is a three dimensional analytic manifold. $" has a (re, Y') structure in the sense of IT]. Here :Y is the simply connected cover of the projective tangent bundle. In this section, we will prove Theorem 4.3. -- The manifold $" is homotopy equivalent to the complement of the trefoil knot in the three sphere. Before proving Theorem 4.3, we will analyze the bilaterally symmetric case. The analysis here is related to the classical fact that SL2(R)/SL2(Z) is homeomorphic to the trefoil knot complement. To prove the general case, we will show that nothing on the level of homotopy changes as we deform away from the symmetric case. PAPPUS'S THEOREM AND THE MODULAR GROUP 205 If the marked box has bilateral symmetry, then the curve A will be a line of P, and the curve L will consist of coincident lines. We can conjugate by a projective transformation so that A is the line at infinity and L is the set of lines going through the origin in R *. Then the set ~" consists of flags (p, P), where p ~ R * -- { 0 } and P is a line through p which misses the origin. Thus J/" is homeomorphic to R * � 1. S The modular group M=(a,~:a*-----~3-~ 1) acts onJP. ~ is an order three linear map of R * and ~ can be taken as the polarity induced by the standard inner product on 11~. The universal cover ~4 ~ has a description as the set of " flags " (p, P), where p is a point in the universal cover of the punctured plane, and l is a line through p missing the origin. Let c denote the covering transformation of X. The " rotation " ~ M acts on ~d ? by setting ~8 _ c. The polarity a has the following concrete description: take the point p on the unit circle to the tangent line through the antipodal point p' in the unit circle. From this description, we can clearly lift ~ to an action on ~47 which satisfies ~2 = c. Thus, the c-extension of the modular group acts on .A ~. The quotient of this lifted action is the original manifold ~.. Hence g- is the quotient of R 3 by the c-extension of the modular group, which is the trefoil knot complement fundamental group. To analyze the general case, we introduce some notions from point set topology. Let M be a topological 3-manifold. An embedded surface S C M is tamely embedded if each point on the S has a small neighborhood so that the pair (N n S, N) is homeo- morphic to the standard pair (B *, B~), where B ~ is the open unit j ball. Two tamely embedded surfaces S t intersect tamely if each point of the intersection has a neighborhood N such that the triple ($1 n N, S,, n N, N) is homeomorphic to the triple (B~, B~, B3), and if these homeomorphisms are all compatible with the homeomorphisms to standard pairs which take place away from the intersection. Here B~ and B~ are orthogonal open unit disks. Here is a technical fact: Suppose M is a closed topological 3-manifold and A~ and B t are two continuous families of tamely embedded and tamely intersecting closed surfaces. Then the homeomorphism type of M -- A t -- B e remains unchanged as t varies. We omit the proof because it is technical, and would lengthen the exposition considerably. Proof of Theorem 4.3. -- Let d C ~ consist of those flags whose first coordinate is contained in A. Let ~ C ~ consist of those flags whose second coordinate is contained in L. Then J/" ----- ~ -- ~r -- ~. The surface d is a bundle of projective lines sitting over A; it is closed, and ruled in the sense of section w 4.1. Dually for ~. In fact, in the local coordinates shown in Figure 4.3, the two surfaces are locally both ruled surfaces of R 3 in the ordinary sense. Furthermore, they intersect exactly in a curve. It is a standard exercise to show that locally such surfaces are tamely embedded, and intersect tamely, in R 3. Hence the same is true globally, in ~. Taking a family of marked boxes varying 206 RICHARD SCHWARTZ FIO. 4.3 continuously from a bilaterally symmetric one to the one under study, we see that Jg" is homeomorphic to S 1 � R ~, as in the symmetric case. Finally, the two perturbed elements e and ~ may be lifted to the universal cover, since ~ is conjugate to a rotation and e is still a polarity. The identity ~2= ~3__ c cannot change either. Hence the trefoil knot fundamental group acts on the universal cover of Jt r, which is just R 3. The quotient is ~, as desired. I REFERENCES [B] A. BEARDON, The Geometry of Discrete Groups, Springer-Verlag, 1983. [H] HmBE~T and COHN-VossEN, Geometry and the Imagination, Chelsea Publishing Company, 1950. IT] W. THURSTON, The Geometry and Topology of Three-Manifolds, Princeton University Lecture Notes, 1978. Mathematics Department UCSD La Jolla, California 92093 USA Manuscrit refu le 11 juin 1992.

Journal

Publications mathématiques de l'IHÉSSpringer Journals

Published: Sep 26, 2007

There are no references for this article.