This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
regular realization ofpG requires r > r$ branch cycles.
It simplifies notation to consider this over Q. So, we restrict often to that case. Still, for K finitely generated over Q replacing Q this should hold with ka and kb depending on K. Let r be an integer and C a collection of r conjugacy classes (possibly with repetitions) from PC(G). To this data [MT] canonically attaches a sequence of reduced Hurwitz spaces (App.II): (0.2)
• • • ft(£+1G, C) r d -> U{kpC, C) r d ->•••-> n(°pG, C) r d .
The kth level is the manifold W(*G, C) r d and°VG= G. Special Case: G = Dp is the dihedral group of order 2p, p is an odd prime and C is r = 4 repetitions of the conjugacy class of involutions. Then, (0.2) is the classical sequence of modular curves: • • • -> Yi(p* +2) -> Y1(pk+1) - + . . . - ) . Y1(p). This case agrees with the conclusion of the Main Conjecture. For this case, however, it is easy to show the levels have only one (unobstructed) component. [Fr3, §7] gives the reduction of (f.b) to (f.a) for G = Dp and for any value of r 0 . For Dp and ro = 4 (or 5) the results are very strong; they translate to known results of Frey, Kamienny, Mazur and Merrill. They even give a statement uniform in p ([DFr, §5] or [Fr3, §7]). Conjecture 0.1 asks for less
154
Michael Fried and Yaacov Kopeliovich
in this case: that Q regular Dpk realization requires at least 6 branch points if k > kb. (Also, kb here may depend on p.) The goal is to generalize the following argument. Fix p and let K be any finitely generated extensions over Q. The genus (of the single component) of Yi(pk) goes up (quickly) with k. Thus, Faltings' Theorem implies the K points Yi(pk)(K) on Y\(pk) are finite, excluding finitely many values of pk. The conjecture thus comes to eliminating a possible projective system t)k G Yi(pfc+1) of K rational points. [Fr3, §7] handles this when G = Dp and r 0 = 4 (or 5). For an arbitrary finite group G, each step above encounters problems. [FrK] reveals the main difficulties. This case has G — An and C = Cy, r repeats of the conjugacy class of 3-cycles. The Nielsen class Ni n?r of r 3-cycles in An appears often. Use the notation %n,r for the corresponding (inner) Hurwitz space (App.I). Though most groups produce difficulties that don't appear for the dihedral group, conjecturally Modular Towers do mimic properties of modular curve towers. Even, however, with G = Dp the conjecture is open if VQ > 5. Here are three accomplishments of this paper. (0.3a) For any G it shows that the (f.b) version of Main Conjecture 0.1 reduces to showing (f-a): For k large, K(pG, C) r d has no Q points. (0.3b) For G = A5, p = 2 and C = C^ it describes progress on bounding components of the kih level, k > 0, in the Modular Tower. (0.3c) It shows how obstructed components (§0.C) affect the Main Conjecture. Further, A$ examples support their disappearance for k large. §0.C. Obstructed components of a Modular Tower. The projective limit of sequence (0.1) is the universal p-Frattini cover PG of G. §1.C explains how similar are all the groups kG. Subtheme 0.2 Replacing G with pG produces conclusions on regular realizations of all PG characteristic quotients. For fixed r, computations suggest there is a uniform bound on absolutely irreducible components ofU(kG, C) r d . When r=4, reduced Hurwitz spaces are curves. Bounding the number of components in the fcth level is necessary to assure Faltings' Theorem applies. That is, the genus of absolutely irreducible components at level k must exceed 1 for k large. Still, that doesn't preclude rational points at arbitrary high levels on the most mysterious of components, those we call obstructed. Suppose no points of %(£+1G, C) r d lie above a component W ofU(kG, C) r d . We say %' is obstructed. [MT, §111.D] gave a big invariant detecting ob-
Modular towers
155
structed components. Obstruction Lemma 3.2 reformulates this invariant using modular representations. There is good news and bad news in the appearance of obstructed components. The former occurs if there is exactly one component at level k and it is obstructed. Then, (0.3a) has a positive answer. The reduced Hurwitz space at level A;+l is empty, so it has no rational points. [Frl] gives examples of exactly that (see Thm. 3.1 in §3.A). Many cases, however, have at least one unobstructed component. General results describing components of a level of a Modular Tower require conjugacy classes in C to appear with high multiplicity. The rest of this subsection illustrates this. Definition 0.3 Consider a collection C of conjugacy classes from a finite group G. Let s0 be a positive integer. Then, C has multiplicity at least so if each conjugacy class in C appears at least so times. Suppose C has suitably (explicitly) high multiplicity compared to k. Then, an effective version of a Conway and Parker result precisely bounds the components at level k. The following is a special case of a result from [FrK]. Here M(*G) is the Schur multiplier of £G. Theorem 0.4 Fix a value ofk. Assume G is perfect and centerless, p \ \G\, all classes in C are from pC(G) and C has multiplicity at least SQ. Decompose HipG, C) r d into absolutely irreducible (dimension r—3) components U^ 0 ?i^. Then, there is an explicit so(G, k) = so so the following hold. (0.4a) tk + l = \M(kpG)\. (0.4b) Hf0 is unobstructed. (0.4c) If C is a rational union (see §1.A^, T~Lf0 has field of definition Q. (0.4d) %[ is obstructed, i = 1 , . . . , tk.
§0.D. Disappearance of obstructed components for k large. Suppose G has a nontrivial Schur multiplier (as does An). Then, so does ^G for all k (Schur Multipliers Result 3.3). Thus, Theorem 0.4 implies obstructed components appear in %T^r for r large (notation from §0.B). The crucial investigation must consider minimal values ro = r*o(n, k) of r that produce obstructed components in H^G, C3r) rd , By contrast, a particular Modular Tower considers k large compared to a fixed value of r = r(C). There are serious diophantine troubles for the Main Conjecture should new obstructed components appear at each level. For example, suppose there is an SQ giving the following. (0.5a) Theorem 0.4 holds with SQ independent of k.
156
Michael Fried and Yaacov Kopeliovich
(0.5b) There is a bound on |M(pG)| independent of k. This would produce a bound in (0.3b), at least for r large. That wouldn't, however, contribute to the Main Conjecture. Even the strongest diophantine conjectures can't eliminate rational points on obstructed components at arbitrary high levels of a Modular Tower. Worse yet, (0.5b) doesn't hold. So, it is a surprise that the case G = A$, p = 2 and C = C^r supports the following conjecture. Recall: A projective nonsingular variety V is of general type if it's canonical bundle is ample. If V is a curve this means V has genus at least two. Using the phrase general type on H(pG, C) r d means apply it to some nonsingular compactification of its components. Conjecture 0.5. Assume G is centerless, p\\G\ and C are conjugacy classes from PC(G). Then, there are two constants B — B(G,r) and ko = ko(G, r) so that for k > k0, the following hold. (0.6a) ^(pG, C) r d has at most B absolutely irreducible components. (0.6b) Each component of H(pG, C) r d is of general type. (0.6c) No components ofH(%G,C)rd
are obstructed.
The classical connection has G = Dp, p is an odd prime, and C is r = 2rf repetitions of the conjugacy class of involutions for some integer r'. Braid group action on Nielsen classes here is easy to calculate. As in App.I, orbits of the braid group Br on Nielsen classes Ni(G, C) correspond to the components of %(G,C). [Fr4, §3] shows there is one orbit. So (0.6a) and (0.6c) hold. When r' = 2, §0.B notes (0.6b) holds. This seems to be unknown for higher values of r'. Beyond dihedral groups, evidence for (0.6c) is from the case G = A$. Lemma 3.2 uses the action of the characteristic quotient JG on the characteristic kernel ker^ / kerfc+i for all values of A;. Consider this contrast with A;o fixed and r large. Then, Theorem 0.4 says 7/(2°A 5 ,C 3 r) rd has at least two absolutely irreducible components; one unobstructed and all others obstructed. For r = 4, however, we expect—a result of [PrK] for k small—no obstructed components for any value of k. There should be a minimal value r' = rf(ko) of r for which there are no obstructed components in 7^(2^5, C 3 r /) r d for k > fc0- In support of (0.6c), [PrK] makes an explicit guess for r' = r'(ko).
§1. Precise versions of the main conjecture. Here is the data for a Modular Tower: a finite group G, a prime p \ \G\ and a collection of r conjugacy classes C from PC(G). Recall: Dr is the discriminant locus in projective r-space. We concentrate onrealizations
Modular towers
157
of a centerless group G. [FrV] shows this is equivalent to finding a Q point on an inner Hurwitz space %{G, C) = H(G, C)in for some C. See §0.A and App.I for Nielsen classes and how (G, C) canonically produces the moduli space H(G, C) with a cover $ G ,c = * : H(G, C) -> P r \Dr. Suppose K is a field of definition of U{G, C) and p G U{G, C). Denote the field generated by coordinates of p over K as if (p). In this paper points are geometric points. Also, having coordinates for p means W(G, C) is a quasi-projective (even affine) algebraic set ([Pr2], [FrV] or [V]). §1.A. Notation for Q moduli spaces. Let Afc = N be the least common multiple of the orders of elements in the collection C. The group Z* is the projective limit of the invertible elements of Z/iV over all positive integers N. Consider a field K (a subfield of C). A Galois extension L/K{x) is a K regular (G, C) realization if the following hold. (1.1a) G(L/Q(x)) = G and L n K = if, and (1.1b) points of the x line ramified in L have associated branch cycles in C. Covers associated to such extensions have r branch points. According to the branch cycle argument ([Fr2, before Thm. 5.1], [Pr3], [V, p. 34]), there is a first criterion for such a realization. The minimal field containing all roots of 1 is Q02^, the cyclotomic closure of Q. The formulation uses the if-cyclotomic group: HK = {a G G(QC2/C/Q) ^ Z* | a fixes elements of K n) has field of definition K if and only if C is a if-rational union. If H(G,C)(K) is nonempty, then C is a if-rational union and %(G, C) contains an absolutely irreducible component over if. Also, if points on 7/(G, C) produce if regular (G, C) realizations.
158
Michael Fried and Yaacov Kopeliovich
For simplicity, the remainder of the paper concentrates on K = Q. In particular, assume the conjugacy classes C from G is a rational union. §1.B. Fix conjugacy classes, change the group. Our Main Conjecture divides into cases: (p,Nc) = 1 and p\ NQ- §2 and §3 consider the case (p,Nc) = 1- The Main Conjecture reduces to this case according to Theorem 4.4. Analysis of components of W^G, C) as k varies requires information about PG. To help, we expand here on [FrJ, Chap. 20] and [MT, Part II]. This discussion will continue in [FrK]. The universal j?-Frattini cover (J>G '• PG -» G of G is versal for embedding problems withp-group kernel. The meaning of versal: Given i\) : A —> G —» 1 exact, there is a map ipf : pG —» A giving the obvious commutative diagram. Versal, unlike universal, does not mean ty1 is unique. If G is perfect, the universal p-central extension of G is an example of a natural quotient of PG. It is, however, a very small quotient. We explain this. Let ker0 —> PG -» G be the natural short exact sequence. Then, ker0 is a pro-free p-group of finite rank. Define ker& inductively: It is the closed subgroup of PG that [kerA;_i,kerjt_i]ker^_1 generates. Then, £G = pG/ker^. Follow [FrJ, Chap. 20] in calling the minimal number of elements that generate it its rank. The hypothesis (p, JVc) — 1 allows a special choice of lift of entries of C to corresponding conjugacy classes of pG/kerjt = pG. Choose lifting representatives with the same orders as their images in G. To emphasize this choice of conjugacy classes in pG, keep the notation C for these lifted classes. Further, as conjugacy classes in £G, C is also a rational union, inheriting this property from G. The Frattini cover property appears here. Suppose elements of pG are entries of g' G C. Also, assume g' lifts entries of g G Ni(G, C). Then, (g;) = pG. From the character theory viewpoint—including generalizations of rigidity—(^G, C) realizations are much like (G, C) realizations. Reminder Statement 1.3. [MT, Lemma 3.6] Suppose G is centerless and perfect. Then, so is £G, k > 0. In particular, a Q point on Ti(pG, C) = Hk is equivalent to giving a Q regular (£G, C) realization [PrV]. Such a point automatically gives Q regular (£G, C) realizations, from the images of the corresponding point in Hj, 0 < j < k. One goal of this paper is to show progress on the following problem. Modular Tower Conjecture 1.4. Suppose G is a centerless group and r is a positive integer. With k = k(G,r) suitably large the following hold.
Modular towers
159
(1.2a) For any prime p dividing \G\ and conjugacy classes C supported in k d is empty. pC(G),U( pG,CY (Q) (1.2b) More generally, there are no pG realizations with at most r branch points. Theorem 4.4 shows (1.2a) (for some value of k(G,r)) implies (1.2b) (for a possibly larger value of fc(G, r)). [MT] concentrated on the standard Hurwitz spaces (as in [PrV]) arising from representations of the fundamental group of P r \ Dr acting on Nielsen classes. App.II explains reduced Hurwitz spaces as a PSL2(C) quotient of standard Hurwitz spaces. §1.C. Braid action and PSL2(C) quotients. Let G be a finite group with conjugacy classes C. Consider oneregular (G, C) realization of a group by a cover <\> : X -> P*. This produces infinitely many Q regular realizations; compose <> / with any a E PSL2(Q). Thus, to measure , C) realizations requires counting Q points on reduced Hurwitz spaces , C) r d = nrkd}^LQ. These spaces have natural maps
• • • -» ni^ ->Hrkd^...->Hrod^
Jr.
App.II gives the definition of Jr. Since C is a rational union, these spaces and maps have field of definition Q. Investigating this tower requires information on absolutely irreducible components of %]^ for each level k. We explain more of the group theory behind these diophantine problems. Statement 1.3 says a Q regular (p+1 G, C) realization produces a Q (£G, C) realization. It also suggests these groups are similar, for example, when G is perfect and centerless. Lifting Lemma 4.1 adds to this. It characterizes > G as having the following properties. PG quotients giving a cover ip : H — (1.3a) The kernel of ip has p-group kernel. (1.3b) For g' = {g[,... ,g'8} G H elements of order prime to p, ($') = H if and only if (ip{gf)} = G. Here is a more subtle similarity between H and G satisfying (1.3). Brauer's Theorem [MT, §11.B] says there as many simple ¥p modules as there are p-regular conjugacy classes. Therefore, the two groups have the same simple Fp modules. §3 shows the significance of an appearance of the module 1 (for pG) in the Loewy display of the ¥p module keik /kerk+i- It is the whole story of obstructed components. Fix r 0 and a field K. Theorem 4.4 says the following are equivalent. (1.4a) There are K regular ^G realizations with at most r 0 branch points for each k > 0. (1.4b) For some r < r 0 there is an r-tuple C of p-regular conjugacy classes of G and for each k > 0, there are K regular (^G, C) realizations.
160
Michael Fried and Yaacov Kopeliovich
The Main Conjecture reduces to showing (1.4b) is impossible when K is a number field for each r-tuple C of p-regular conjugacy classes of G. Proving the Main Conjecture comes to considering rational points on reduced Hurwitz spaces (PSL2(C) quotients), curves when r 0 = 4 (App.II). Suppose you show (0.6a) (for large k). Faltings' Theorem [Fa] applies if (0.6b) holds: the genus of the components at level k goes up with k. [FrK] gives evidence for this rise in genus from our special case. §1.D. Projective systems of rational points. Suppose the program of §1.C works. Then, there are only finitely many Q points on some level fco of Modular Towers satisfying the hypotheses. It is, however, a formidable task to eliminate the possibility of a nontrivial set of Q points at every level. This is where hypothesis (0.6c) enters. It assumes there are no obstructed components at level k0 or beyond. Then, having points at each level above k0 produces a projective system of rational points {p^ 6 %(pG, C) rd }^L 0 . That is, the canonical map £ +1 G -> kpG inducing ft(£+1G,C)rd -> ft(£G,C)rd gives pfc+i «->• pk- Precluding projective systems is the last leg for r = r$ = 4. To do so for any value of r is a separate problem independent of other diophantine considerations. Projective System Conjecture 1.5. Assume the setup for the Modular Tower above associated to (G, C), with r arbitrary and K finitely generated over a number field. Then, there exists no projective system of K rational points on the Modular Tower. §2. Construction of universal Prattini covers. [FrK] expands the Normalizer Observation of [MT, Remark 2.10]. It helps decipher the whole universal p-Frattini cover from representation observations on the original group G. Thus, representation facts about G translate to a presentation of the universal p-Frattini cover of G, and so of its infinite string of characteristic quotients. This section presents two preliminary results in this direction. §2.A. Relation of pG to pH with H < G. Let P = Pp = PG,p be a p-Sylow subgroup of G. The notation kero(G) allows simultaneously distinguishing the characteristic kernels of the universal p-Frattini covers for G and for a subgroup H. Use <J>Q : PG —> G for the canonical map. Also, denote keri(G)/keri_|_i(G) by Mi(G). For any subgroup H of G, restricting H to Mi(G) gives an H module, MI(G)H- The rank of any profinite group G is the smallest number of elements that topologically generate it. Finally, is t n e PG,P = ^GH-PG.P) P-Sylow subgroup of PG.
Modular towers
161
Example 2.1. The p-Sylow Frattini hull Let G be any finite group. Suppose Pp has rank u. The free pro-p group on u generators is pFu. Let NG(PP) be the normalizer of Pp in G. Then, the universal p-Frattini cover of NG(Pp) is pFu xsNG(Pp)/Pp [Ri, Th. 3.2]. This a special case of Principle 2.3 (below) by extending the action of N = NG(PP)/PP to PFU. Since PFU is pro-free, there is a subgroup B of Aut(pF w) mapping surjectively to N. The essential point is that the kernel of the map B —> N is a pro-p-group. (Prove this by inductively considering the characteristic Frattini quotients of pFu.) Apply Schur-Zassenhaus to split off a copy of N. Thus produce the action of N on pFu [FrJ, Lemma 20.45]. Important point: Action of N is unique only up to conjugation by an automorphism of PFU. • Let kerj be the kernel of pFu —» Pp as in Ex. 2.1. For Ex. 2.1, the action of NQ{PP) on kerg /ker^ extends to G. Example 2.2. The p-Sylow Frattini hull can be all of PG. Take G = As, p — 2 and NG(PP) = A* < A$. Then, we can identify kerg as the kernel of the universal 2-Frattini of A5 [MT, Prop. 2.9]. When, however, p = 3 or 5 (and G — As) kerg /ker^ is cyclic, and not an A$ module. In particular, kerg for these values of p isn't the kernel of PG -> G. • [FrK] generalizes the observation of Ex. 2.2. Principle 2.3 characterizes the universal p-Frattini cover PG of G to relate it to that of its subgroups. Subgroup Frattini Principle 2.3. Let G be any profinite group and p any prime dividing |G|. Then pG is the smallest group that covers G and has its p-Sylow subgroup pro-free. Suppose H < G are profinite groups. Then, there is a surjective pro-p group homomorphism PG,H '- kero(G) -> kero(i^). Further, PG,H is unique up composition on the left with an automorphism of kero(G). This induces a surjective H module morphism p*G H : M0(G) —> M0(H). Finally, if G is a finite group, the following are equivalent: (2.1a) PQ H is an isomorphism. (2.1b) dimFp(M0(G)) = dim Fp (M 0 (tf)). (2.1c) tcl(H)=pH. Proof. First consider the opening characterization of pG. For G any profinite group, a simple property determines the universal Frattini cover of G. It is the minimal projective cover of G [FrJ, Prop. 20.33]. A profinite group is projective if and only if each p-Sylow subgroup of it is pro-free [FrJ, Prop. 20.47]. The kernel of (f)G is a nilpotent group [FrJ, Prop. 20.44]. So G is the fiber product over G of closed subgroups PG for each prime p | |G|. Suppose p(j)* : pG* —>• G is any profinite group cover of G for which pG*
162
Michael Pried and Yaacov Kopeliovich
has a pro-free p-Sylow subgroup. Consider the fiber product over G of the groups PG* and P>G for all primes //, p' ^ p, dividing |G|. Call this fiber product G*. From the Sylow subgroup characterization of projective groups, G* is projective. Also, it is a minimal projective cover of G if and only if pG* is a minimal cover of G having a pro-free p-Sylow. This concludes the opening characterization of PG. The remainder of the proof concentrates on universal p-Frattini covers. All notation (say, for ker^) refers to the fixed prime p. Now consider H < G profinite groups. Then, H* = pcj)^}{H) is a closed subgroup of pG. The p-Sylow subgroup of H* is a closed subgroup of the p-Sylow of PG. Thus, it is projective and therefore pro-free. Apply the characterization above for the universal p-Frattini cover of H. Conclude there is a surjective map j3 : H* —> PH commuting with the natural surjective maps to H. The image of /? is a closed subgroup of PH mapping surjectively to H. Since PH is a Frattini cover of H, /3 is surjective. So, it induces the surjective map PG,H : kero(G) —>• kero(H) in the statement. The rest follows because the 1st characteristic Frattini quotient determines the rank of a pro-free p-group. • §2.B. Decomposing Frattini cover kernels. The following is a special case of [Be, Exer. 1, p. 11]. We fill in details, isolating it from its generalities. Indecomposability Lemma 2.4. The pG/kerjb(G) module M^{G) is indecomposable. Proof. Let M be any G module. Define fi(M) to be the kernel of a surjective G module homomorphism i/> : P —> M with P a projective module, minimal for direct sum decomposition. Suppose P' is another module with these properties. Then, projectivity of the two modules gives maps a : P —» Pf and p : Pf —» P, each commuting with the surjective maps from P and P' to M. Define / = aof3 and the composition of / , n times, by fn. Similarly, for gn with g the composition f3oa. If n is large, then P = ker(#n) ®Im(^ n ). Note: ker(#n), being a summand of a projective module, is also projective. Suppose P and P' aren't isomorphic. Then ker(#n) goes to 0 under the map from P to M. This contradicts the minimality of P , and proves that fi(M) is well defined. Further, from Schanuel's Lemma [Be, Lemma 1.5.3], up to projective summands Q(M) is well defined. Similarly, there is the operator Q~1(M): the cokernel of the embedding of M in a minimal injective module. In the category of finite dimensional modules over group rings, projectives and injectives are the same [Be, Prop. 1.6.2]. Thus, P above is both the minimal projective covering M and an injective module containing
Modular towers
163
That is, up to addition of a projective module, Q~l(Q(M)) = M. So, M is indecomposable, if and only if fi(M) is. Apply f2 twice to the indecomposable module M — lk^. From [MT, Proj. Indecomp. Lemma 2.3], kerfc/kerfc+i = fi2(M). So, Q2(M) is indecomposable. • Consider any cover <\> : H —> G with abelian kernel M. Then G acts on M: g e G maps m G M to gmg~x where # is any lift of g to if. Remark 2.5. Frattini kernels may decompose. Let Mi and M2 be nonisomorphic simple modules appearing as respective kernels for nonsplit extensions G\ and Gi of G. Consider the fiber product H = G\ XQ G2 and the natural map ^1,2 : H -» G. We showG and G2 —> G are Frattini covers, i?i factors surjectively through them both. Thus, by the Jordan-Holder Theorem, the composition series of the kernel of H\ —>> G includes Mi and M2. It must therefore be Mi 0 M2. Two series of simple groups agree when n = 8: A$ = SL(4, Z/2). Let M4 be the standard 4 dimensional representation of SL(4, Z/2). [Be] shows H2(A$,M4) has dimension 1. This gives a Frattini extension of As not factoring through the universal central extension of A8. It is an example of the above. • §3. Progress on the case A5 and C = C 3 r. The main example of [FrK] is the A5 Modular Tower associated with r = 4, p = 2, and all conjugacy classes those of the 3-cycle in ^ 5 : C = C34. Denote the corresponding reduced Modular Tower (of inner spaces) by (3.1)
• • • -> ^ ( ^ + 1 i 5 , C) r d -> « ( * i 5 , C) r d -4
> «(i4 5 , C) r d .
This example reveals major phenomena that don't occur for modular curves. §3. A. Illustration of obstructed components. Obstructed components produce both encouraging and discouraging diophantine results. First we illustrate the former. Actual examples go precisely as the outline in §0.C. The proof of the next result explains the variety W,(G, C) a b s . Theorem 3.1. ([Prl]) Assume n > 5 is odd. Then, /H{An,C3n-i)rd(Q) is dense in H(An, C 3 n-i) r d . Now assume n > 6 is even. Then, there are no Q regular Q}An,, C 3 n-i) realizations. Further, there is a Q unramified cover U{G, C)in -> U(G, C) a b s , Galois with group Itji with the following properties. (3.2a) Each p <E %(G, C ) a b s corresponds to a degree n cover
164
Michael Fried and Yaacov Kopeliovich
(3.2b) For p G H ( G , C ) a b s , >p : X -> P£ ftas (minimal) field of definition Q(p). /te Galois closure is defined over Q(p') w^/fc p' It/in^ over p.
(3.2c) p G H(G, C) a b s corresponds to a Q(p) regular (An, C 3 n-i) realization if and only if Q(p') = Q(p). For a// n > 5, tfie se£ {p G U(G, C) abs (Q) wz% Q(p') ^ Q(p)} is dense. For n = 5, even {p G U(G, C) abs (Q) with Q(p') 7^ Q(p)} is dense. Proof. This example compares the group theoretic Hurwitz space ideas, with explicit equation calculation. Hurwitz space computations are explicit group theory computations. It's more than a matter of taste, for we are after properties of the spaces, not their equations. Other examples of this are in [Pr3] and [Fr4]. The remaining proof has four parts, starting from statements relating inner Hurwitz spaces to absolute Hurwitz spaces. This lets us focus on the effect of Mestre's calculation for n odd, distinguishing this from the more mysterious case when n is even. Note: If Q(p') ^ Q(p) (notation as in (3.2c)), then the Galois closure of (j)p : X —> P* in (3.2b) has group Sn (see below). Part 1: Absolute Nielsen classes. For k = 0, a special case of [Frl] shows there is one i? n -i orbit on Ni(^4n, C 3 «-i). Another Hurwitz space is handy for this problem. It is closest to Mestre's computations [Me] below. For any G < 5 n , consider the normalizer Nsn (G) of G in 5 n . Let JV(C) = Nsn (G, C) be the subgroup of Nsn (G) whose elements conjugate the entries of C among themselves. We also use G(l), the stabilizer of 1 in this degree n representation. Form the natural quotient Ni(G, C)/N(C) = Ni(G, C) a b s . Then, S n _i, and its Hurwitz monodromy group if n -i = 7Ti(Pn~1 \ -Dn_i) quotient, act on Ni(G, C) a b s . Covering space theory produces a sequence of unramified covers (3.3)
W(G, C) = U(G, C)in -> n(G, C) a b s -> Fn~l \
Dn^.
abs
As in [Fr2], [FrV] and [V], interpret H(G, C) as a moduli space of equivalence classes of degree n covers. For p G H(G, C) a b s , let 0P : Xp ->- P^ be a representing cover. Cover equivalence here is (II.la) (in App.II). This cover has the following properties. (3.4a) The Galois closure X -> ¥]. of >p represents p' G W(G, C)in over p. (3.4b) X is the quotient of X by G(l). Suppose WJ1 is a connected component of Hin and % abs is the image of n Uin i n ^abs T h e n 7 Uin corresponds to an orbit O\ of Br on Ni(G,C). The natural image O abs of O[n in Ni(G,C) abs corresponds to Ulhs. The following set is actually a group: g(C) = {he N(C) I 3 Q G B n _i and 0 G O\n with (g)Q =
hgh'1}.
Modular towers
165
The cover U(G, C)in -> H{G, C) a b s on the left of (3.3) is Galois with group G(C)/G. Note: [MT, Lemma 3.3] says G < G(C). Part 2: Using G(C3n-i)/G = Z/2. With An < Sn the standard representation of the alternating group, N(C) = Sn. Thus, G(C) is a subgroup of Sn/An = Z/2. Also, H(G, C) a b s is the moduli space of degree n covers with 7i—l 3-cycles as branch cycles. From the Riemann-Hurwitz formula, this is a moduli space of genus 0 covers. Further, as in [Fr2], since the subgroup G(l) = An(l) is self normalizing in An, %(G, C) a b s is a fine moduli space. In particular, p G H(G, C) abs (ii0 corresponds to a K cover >p : Xp -> P*. [Frl] says £ n _i is transitive on Ni(;4 n ,C 3 »-i). Thus, (C3»-i) = G = Z/2. An example gives the sense of why this holds. Take n = 5. From the transitivity result, G/G = Z/2 if and only if there exists g G Ni(^4n, C 3 n-i) and Q G B4 with (g)Q = (4 5)g(4 5). Take
Here applying Q 3 (App.I) has the same affect on g as conjugating by (4 5). Apply Hilbert's irreducibility theorem to the sequence of covers in (3.3). This says a dense set of points j: G Pn~1(Q) have above them p' G %{G, C)in lying over p G U(G, C) a b s for which Q(p') ^ Q(p). Excluding the statement about Q points, this completes the proof. Part 3: Interpreting Mestre's calculation. Continue the discussion at the beginning of Part 2. Here is the first place where there is a distinction between n even and odd. If (j)p has odd degree, then Xp has a rational point over Q(p). In particular, 0P is Q(p) equivalent (in the sense of (II. la) in App.II) to a cover P£ -> P*. [DFrl, p. 115] and [DFr2, p. 115] discuss whether this means a space like T ^ n - i IS a fam^y of rational functions. x That would mean, there is a natural map # a b s : 7i^_x xPj -> U^-i ^lAlso, for any p G H*^_ 1? ^ p is a rational function cover representing p. We don't think there is such a map \I>abs, though we haven't excluded it. Still, as in [DFrl, 2], if you add the branch points of the cover to the moduli space, you nearly have coordinates for a family of rational functions. When n is odd that is exactly what [Me] does. In fact, he gives a dense set of rational points in W^'n_1. Here is how it goes [Se, p. 100-101]. Let P(y) = nr=i(2/ ~~ ai) w ^ n ^ ne ai algebraically independent indeterminates over Q. If Q(y) is a polynomial of degree n— 1, then Q/P maps 00 to 0. We look to choose those polynomials Q so the derivative of Q(y)/P(y) is a square (R(y)/P(y))2. If Q and R are suitably generic, that means the cover from Q/P has 3-cycles as branch cycles. It is then in the Nielsen class
166
Michael Fried and Yaacov Kopeliovich
of our interest. Here are the expressions for Q and R:
(3.5a)
Q(y)/p(y) = Yl ~ c '/(y - a«)'
R p
i
=
where the Q S satisfy (3.5b)
] T ^-/(a. - Oj) = 0 for all i.
This is the second place using n odd. The matrix with ixj entry l/(a; — ay) for i ^ j and 0 for i — j is skew-symmetric and n x n. Thus, for n odd its determinant is 0. So, there is a line of solutions for the vector of c» s. [Frl] says 7/(^4 n ,C 3 n-i) m is an absolutely irreducible Q variety. So, Mestre's collection of rational functions shows 7/ abs is a unirational variety ([Me] when n is odd; [Se, p. 100-101] asserts a variant works for even n, too). So it gives a Q dense subset of H a b s . These points are the image of Q points of nin. This is a dense set of p e K(G,C) a b s (Q) for which Q(p') = Q(p). When n = 5, [Fr5, Thm. 5.9] shows there is a dense set of p € « ( G , C) abs (Q) for which Q(p') ^ Q(p). Part 4- For n even, W(A n ,C 3 n-i) is obstructed. There is nothing in the Nielsen class Ni(2^4 n5 C 3 n-i) when n is even. It goes like this. Let An be the universal central exponent 2 extension of An. Then, \An —> An factors through An -> An. Suppose 20 £ Ni(2^Ln,C). The canonical map lAn —> An sends 20 to Q G Arn with II(g) = 1. Let 0 be the image of 0 in Ni(A n ,C 3 n-i). Again use transitivity of Z?n_i on Ni(A n ,C 3 n-i). So, if one element of Ni(A n ,C 3 n-i) lifts to an element of Ni(^4 n ,C 3 n-i), then any element does. On the other hand, [MT, Ex. III. 12] explicitly gives 0 € Ni(i4 n ,C 3 n-i) with no lift to Ni(^4 n ,C 3 n-i). Thus, level one of the Modular Tower for (An, C 3 n-i) is an empty variety. •
§3.B. Invariants for obstructed components. Obstructed components and pure group theory establish Modular Tower Conjecture 1.4 for (An, C 3 r) with even n > 6 and r = n — 1. [Frl], however, shows a different outcome for the cases (n,r), r > n that Theorem 3.1 doesn't cover. There are unobstructed components in all levels of the Modular Tower. Lemma 3.2 is a completely general result applying to all levels of any Modular Tower. As in §0.A, with 9 = ( : H —> G is a central extension. Then, any g G G of order prime to | ker(>)|, has a unique lift g to an element of H of the same order as G. Assume ker(<^>) has order prime
Modular towers
167
to the orders of all entries of g = (#1,... ,gr) G Gr'. For II(g) = 1 let s(g) be II(<7i • • - gr). In Lemma 3.2, the exact sequence has 1 as kernel. Here 1 is the identity module for the action of ^G. Obstruction Lemma 3.2. Suppose O is a Br orbit in g G Ni(pG, C). Let S(O) = {II(g) I g G £ +1 G, g 6 C and g mod kerfc = g}. zs a union 0/ conjugacy classes in p +1 G- TAen, 1 0 5(O) exactly if there exists a sequence of covers 77MS
with the following properties. (3.6a) The kernel of H2 -> H1 w 1. (3.6b) There ezzsfc g* G Ni(#i, C), g* mod kerfc = g and s(g*) ^ 0. Suppose Ho is the level k component of Hk corresponding to O. The previous condition holds exactly when Ho is obstructed. Proof. Use induction on the Loewy layers of kerfc/ker/fe+i. Replace 1 in the kernel in H2 -> Hi by an irreducible module A ^ 1. Consider the set Sf of n(g') as g' runs over all allowable lifts of r-tuples g G H% with II(g) = 1. Then Sf = A. Here is the argument, using that the set Sf is a braid invariant set. That is, let g° be one lift. In the orbit, you can braid g° to something whose lift has the same product, but in the braid gf appears on the right side. Now form a lift replacing gf by agfa~x = gfa9i a" 1 . So, this lift gives Q°agi a" 1 . You can do this for any i and any a G A. If A ^ 1 is irreducible, the possible products of lifts gives g°a with a € A arbitrary. So, the corresponding Nielsen class is empty only if A = 1. • Schur Multipliers Result 3.3. Assume n > 4. For each k > 0, \An has a nontrivial Schur multiplier. Generally, suppose ^G has a nontrivial Schur multiplier. That is, there is a sequence p +1 G —>• H\ —> ^G with 1 —>- Hi —> pG short exact as in Obstruction Lemma 3.2. Then, [ker^ifijiff generates a proper closed subgroup Hi o/kerfc+i. Also, £ +1 G has trivial action on M = kerfc+i /H2, producing a nontrivial Schur multiplier for Proof. As is well-known [MT, §11.C], the exponent 2-part of the Schur multiplier of An (n > 4) is Z/2. Thus, the statement on the alternating groups follows from the general inductive statement. Since ker^ is a profree group, H2 is a proper subgroup of kerfc+i. The action of p +1 G on M is trivial if it is trivial on generators of M. One type of generator is vp with v G ker&. Conjugating v by g G p+1G gives vh for some h G Hi. Modulo
168
Michael Pried and Yaacov Kopeliovich
H\[kerfc, Hi], set hp = 1 for h e Hi and vh = hv for v 6 kerfc and h e H±. Thus, the following hold. (3.7a) gvPg~l = (vh)*> = v? mod H\[ker^/fi]. (3.7b) ^[v, t;']^- 1 = [vfc, i/fi'] = [v, t;'] mod Hp[kerk, #x].
D
§4. Whenp|7V c . Let G be a finite group and let C be a collection of conjugacy classes in G. Theorem 4.4 assumes Modular Tower Conjecture 1.4 in the form (1.2a) holds. From this it concludes Conjecture 1.4 is true. The core of the proof starts by assuming C contains at least one conjugacy class that isn't j9-regular. It then shows, for k large, lifts of C to conjugacy classes in pG can't be a rational union. The gist of this is it suffices to establish our major conjectures when (p, NQ) — 1. §4.A. Lifting elements of order p. Continue notation from §2.A: <j> = (j>o : VG —> G is the canonical map having kernel ker o (G,p) = ker o (G).
Lifting Lemma 4.1. The following are equivalent. (4.1a) C consists of p-regular conjugacy classes. (4.1b) For each k > 1, classes from, C lift uniquely to classes pC of the same order in pG. (4.1c) (4.1b) holds fork = 1. Further, let Q any set of generators of G, with each having order prime to p. Let a : H —> G be a cover of G with p-group as kernel. Then, a is a Frattini cover if and only if lifts of g to elements g of the same order in H implies (g) = H. Proof. Let g be in a conjugacy class from C. Apply Schur-Zassenhaus [MT, Intro, to Part III] to the sequence
This shows (4.1a) implies (4.1b). For the converse, assume p divides the order of g. Suppose g lifts to k g G pG of the same order, for each k. Then, so does ga with a the order of g divided by p. The result follows if we show lifts of g to higher characteristic quotients must increase their order when g has order exactly p. A p-Sylow PQ of PG is a pro-free p-group. The projective limit linioo^-fc kg is an element of order p in the pro-free group PQ- Nontrivial projective profinite groups have no elements of finite order [FrJ, Cor. 20.14]. So, this is impossible.
Modular towers
169
Equivalence of (4.1a) and (4.1b) with (4.1c) follows by showing g, of order p, lifts to no element of order p in *G. Assume P — PQ is a p-Sylow of G containing g. Let Fp be the pro-free p-group of the same rank as P. Choose a surjective map a : Fp —» P. Then, Fp is the universal p-Frattini cover of P [FrJ, Chap. 20]. Denote its kernel by ker o (P). As PQ is also free and covers P, there is a surjective homomorphism i\) : PQ —> Fp commuting with the respective maps <j> and a of PQ and Fp to P. Hypothesis (4.1c) says g lifts to g' of order p in PG/([kev0(G), kero(G)] kero(G)p). Since kero(G) maps surjectively to ker o (P), kero(G)p maps onto kero(P)p and [kero(G),kero(G)] maps onto [ker o (P),ker o (P)]. Thus, ker o (G)/keri(G) maps surjectively to kero(P)/keri(P). The image of g1 in kero(P)/keri(P) has order p and it is also a lift of g. Thus, a lift of g to something of order p in pG implies a lift of it has order p in kero(P)/keri(P). Assume g' G Fp/keri(P) of order p maps to g. Here, apply a similar argument. Consider the pullback a~l((g)) of (g) in Fp. This must map surjectively to Z p , the universal p-Frattini cover of (g). Therefore, an element of order p in Z/p 2 would map to the generator of Z/p. This contradiction concludes the first part of the lemma. Now consider the last condition in the lemma. If H —> G is a Frattini cover the condition holds by definition. Suppose, however, it holds and H —>• G is not a Frattini cover. Then, a proper subgroup Hi of H maps surjectively to G. Apply Schur-Zassenhaus to H\. This lifts the entries of g to elements 91 of Hi of the same respective orders. By hypothesis, Hi > (QI) = H. This contradicts Hi being a proper subgroup of H. • §4.B. Irrational characters. Consider the universal p-Frattini cover of G. For g E G let g G VG be a lift of g. The p'-order of g is the prime to p part of the order of g. This subsection discusses the values of the irreducible characters of ^G on the image kg of g in *(?. The p'-order of g equals that of g. A relevant example might have An = G , n > 5 , p = 3, and the conjugacy classes those of 3-cycles. Then, given a 3-cycle, choose lifts to 3-power orders in kAn. Lift Question 4.2. The values of all irreducible G characters at g generate a field Q^. Similarly, let Q^ be the direct limit of the fields generated by the values of kg at irreducible characters of kG. Suppose G is perfect and p divides the (supernatural) order of kg. Is it possible that Q^ = Q^? If yes, it could be all characteristic PG quotients are groups of regular Galois extensions of Q(x) having a bounded number of branch points, yet infinite inertia groups. The answer, however, is "No!" Call g £ pG rational if its image in kG defines a rational conjugacy class for each k > 0.
170
Michael Fried and Yaacov Kopeliovich
Lemma 4.3. / / the order of g is prime to p, then Qg = Q$. Now suppose g (of any order) defines a rational conjugacy class in G. Then, Lift Question 4-2 has a yes answer if and only if there is a rational lift g £ pG of g. Proof. Suppose g and g have p'-order. Then, characters of kG restricted to (), give sums of characters of the cyclic group (g). Thus, Qkg — Qg for each k > 0. Assume g defines a rational conjugacy class in G. Suppose g is a lift of g with Q = Qg. This means kg has a rational value in each irreducible representation of kG. Assume u is prime to the order of kg. Then, kgu also takes the same values under each irreducible representation of pG, Values of a conjugacy class on irreducible characters determine the conjugacy class. Thus, kgu is conjugate to kg in £G, so kg determines a rational conjugacy class in kG. •
§4.C. Conjecture 1.4 reduces to Modular Tower property (1.2a). Now consider g having order a power of p. Let kP be the p-Sylow of kG. The zeroth characteristic quotient is G; denote ^P by P = Pp. Lemma 4.1 shows any lift of g to *P has larger order than does g. Theorem 4.4. Only p1 elements of PG can be rational. In particular, let 7*0 be any positive integer and let K be a number field. There is an integer ko = ko(ro,K,G) with the following property. Suppose k > ko and there is a K regular realization ofkG. Then, one of the following holds: (4.2a) there is a regular realization corresponding to a point ofH(k°G, for some C with at most 7*0 entries from PC(G); or
C)Td(K)
(4.2b) the regular realization has more than r$ branch points. Proof. Suppose g £ PG is a rational element and g is not a p'-element. For any integer v, gv is also a rational element of PG. Let v be the order of the image of g in G. Then, from Lemma 4.1 as restated above, gv is a nontrivial j9-power element in kero. Replace g by gv. Find the minimal integer n with g G kern \ k e r n + i . In particular, g = gi has nontrivial image in kern /ker n +i. Since ker n is a pro-free pro-p group, any collection of representatives of the nontrivial cosets of kern / k e r n + i give topological generators of ker n . Compliment g\ with elements #2? • • • 19u that freely generate ker n . Now, we show g\ is not a rational element of pG. For each p'-integer ra, our rationality assumption says there is an hm £ m = hmgh^. Since p G/ker m is a finite group, there are pG for which g infinitely many p'-powers g™ of g with corresponding hm in ker n . Let ra' be a nontrivial choice of such an integer.
Modular towers
171
This is our setup. Let B = (g) and D = (#2,»• • ,9u)- Then, the groups B and D freely generate kern. Further, there exists /im> G kern with hm,Bh^ = B. Now apply [HR]. This says, for each h' G kern, either h! G B (and B9 = B) or tiB(ti)-x n B = l. Taking ft7 = ftm/ violates both these conclusions. Therefore, # is not a rational element. To finish the proof, consider the conclusion on regular realizations. Suppose pC is any set of r conjugacy classes (not necessarily p-regular) of pG. A K regular (*G,£C) realization corresponds to a point of H(^G^C)Td(K) (end of App.I). Further, this produces a K regular {PG,JpC) realization with J pC the conjugacy class image of ^C in JG, 0 < j < k. There are only finitely many choices of p-regular conjugacy classes C with at most r$ entries. So, if (4.2a) holds for each &, there exists some C (independent of k) consisting of p-regular classes. Suppose there are pG realizations for every k with at most r 0 branch points, and (4.2a) does not hold with k > kf0. Then, for large k, the following hold. (4.3a) There exists £C = (*Ci,... , fc C r ) and a point of giving & pG realization with r < r$ branch points. (4.3b) At least one entry of ^C mod keiv is not p-regular. Reorder the entries of ^C to assume Ci is not p-regular. The branch cycle argument (§1.A) says ^C must be a if-rational union. Let g G Ci. This puts a bound of r$ on Qk = {gn \ (n, Nk^) = 1}/^G. For suitably large A:, this contradicts the first part of the proof. • Remark 4.5. Effective k0 in Theorem 4-4- The proof of Theorem 4.4 assumes existence of kf0 with no p-regular realizations of pG for k > kf0. Assuming an explicit such k'Q, it is possible to produce an explicit &o. The proof above returns this to the following. Let a G pFu have nontrivial image in the first Frattini quotient of pFu. Then, we must give an explicit lower bound Ck on the number of prime to p powers of a in pFu/kevk conjugate to a, where lirm^oo c& »-> oo. The argument of [HR] can give such a bound. §Appendix I. Nielsen classes and Modular Towers. This is a quick review of fundamental definitions from [MT]. Excluding Theorem 3.1, Hurwitz spaces in this paper are the 7i(G, C ) m inner spaces of [FrV] and [MT]. These parametrize Galois covers X —> Pj. whose branch cycles fall in the Nielsen class Ni(G, C) and have a fixed isomorphism of the automorphism group of X —> P^ with G. Let £C be any conjugacy classes in *&—not necessarily p-regular classes as in §0.B—mapping to C by the canonical quotient with kero-
172
Michael Pried and Yaacov Kopeliovich
Suppose & e C with (g) = G lifts to g G £ C . The Frattini covering property means (g) = kG is automatic. So, after the first level, II(9) = 1 is the significant formula for defining the Nielsen class (§0.A). Specifically: Ni(^G,£C) = {g GkpCI g mod ker0 G Ni(G, C) and II(g) = 1} is the A;th level Nielsen class. Consider the free group on generators Qi, i = l , . . . , r — 1 , with these relations: (Lla)
QiQi+iQi = Qi+iQiQi+u « = 1,..., r - 2 ;
(Lib)
QtQj = QjQi, \i - j \ > 1; and
(Lie)
QiQ2'-Qr-iQr-i~-Qi = l.
Conditions (Lla) and (Lib) define the Artin braid group Br. Add (Lie) t o get the Hurwitz monodromy group Hr of degree r, a quotient of Br. The Qi s in Br act on g € Ni(G, C): (Lid)
(g)Qi = (gv . . . , ^ _ l 7 ftft+i^"1' fti ft+2» • • •»0r)> * = ! , . - . , r - 1 .
Mod out by inner automorphisms of G to induce an action by Hr. Irreducible components of %(G, C) correspond to orbits of this action. Braid group action on Ni(^G,pC) extends that on the level 0 Nielsen class. This produces the corresponding sequence of moduli spaces (1.2)
. . . - » • H(kp+1G,
k +1 p C)
-> H(kpG, kpC) -+ • • • -> H(G, C ) .
Suppose ^G has no center for k > 0. ([FrK] shows this holds if G has no center and no Z / p quotient; in particular, if G is a centerless perfect group [MT, Lemma 3.6].) Let K be a field of characteristic prime to \G\. Then, gives a sequence of K covers as in [MT, Part III], a K point p G H^G^C) (1.3)
k
Xp -> k~lXp ->
> % -> P 1 .
Further, iXp -> P 1 gives a K regular (JpG,JpC) realization, j = 0 , . . . , & . When C = ^ C consists of p-regular conjugacy classes, then (1.2) is what we call a Modular Tower as in §0.B.
Modular towers
173
§Appendix II. Equivalence of covers of the sphere There are two natural equivalences of covers of the sphere. (ILla) fa : Xi -» P 1 , i = 1,2, are equivalent if there exists a : Xi —>• X 2 with 02 ° ot = (j)\.
(Il.lb) As in (ILla), except there is p : P 1 -» P 1 with fa o a = P o fa. §App.ILA. Action of SL2(C) on Fr\Dr. Below, U = U{G, C) refers to a space of covers in a given Nielsen class up to equivalence (ILla) . The group Sr acts on the space (P x ) r by permutation of its coordinates. This gives a natural map \I/r : (P1)'" -» P r . Consider £ = (#1,... , x r ) with none of the coordinates equal 00. The point whose coordinates are the coefficients of the polynomial n i = i ( z ~~ xi) m z represents the image of j: under \£r. (If Xi = 00, replace the factor z — X{ by 1.) Also, \£r takes the fat diagonal A r to Dr. This interprets Ur — P r \ Dr as the space of r distinct unordered points in P 1 . Thus, \I>r : Ur —» Ur is an unramified Galois cover with group Sr. Consider PSL2(C) as linear fractional transformations acting diagonally on the r copies of P x . For a e PSL2(C) and j: G Ur", j: i-> (a(xi),... ,a(xr)). The action of PSL2(C) is on the left, commuting with the coordinate permutation action of Sr. The quotient PSL2(C) \Ur = Ar generalizes the A line minus the points 0,1,00. Further, PSL2(C) \Ur = Jr generalizes the j line minus the point at 00 from the theory of modular curves. It has complex dimension r — 3. The case r = 4 is crucial to us, so we reassure the reader by displaying these identifications. Given j: = {xi,x 2 ,x 3 ,x 4 } up to equivalence (ILla) there is a unique degree 2 cover Xv -> P 1 ramified exactly at j:. Thus, X is a genus 1 curve; its j invariant determines its isomorphism class. Further, XT is equivalent to Xf if and only if there exists a £ PSL2(C) with a(y) = y'. This identifies J4 and the j line minus 00. The natural A line (ramified) cover of the j line is Galois with group S3 [R, 1.59]. Don't confuse this copy of 5 3 with an S3 inside the coordinate permutation action of S4. §App.II.B. Extending PSL2(C) action to U{G, C). Suppose p G H has a representative cover 0 p : Xp -> P 1 . Extend the action of a G PSL2(C) by composing (pp with a to give a o (j)p : Xp -> P 1 , a new G cover in the Nielsen class. Thus, PSL2(C) action extends to W(pG,C); denote its quotient by H(£G,C) rd . Since U{kvG,C) is an affine algebraic set, so is W(£G,C) rd [MFo, Thm. 1.1]. These reduced spaces generalize the spaces of modular curves Y^p^1) [MT, Intro.].
174
Michael Fried and Yaacov Kopeliovich
So, the spaces Wrd are moduli spaces for covers up to equivalence (II.lb): reduced Hurwitz spaces. Many have asked: "Which equivalence is more important?" My answer: Classical geometers often like equivalence (II.lb). (Also, it is usual to use the pulback of Hrd over A r . When, however, there are repetitions in the conjugacy classes C, this pullback is inappropriate, often wiping out the significant arithmetic information.) Their justification is the complex dimension of Hrd is 3 less than that of %. We like that, too. For example, the reduced spaces are curves when r = 4, allowing use of Faltings' Theorem. Further, any rational point on % automatically produces infinitely many others in its PSL2(Q) orbit. Allowing this would defy the finiteness results this paper conjectures. Still, it is equivalence (II.la) that supports the Modular Tower construction. From that construction we compatibly reduce all levels of the tower by the PSL<2(C) action. Conclusion: We require both equivalences for Modular Towers.
References [Be]
D. J. Benson, Representations and cohomology, I: Basic representation theory of finite groups and associative algebras, Cambridge studies in Advanced Mathematics 30, Camb. Univ. Press, 1991. [Be2] D. J. Benson, The Loewy structures for the projective indecomposable modules for A% and A$ in characteristic 2, Comm. in Alg. 11 (1983), 1395-1451. [DFrl] P. Debes and M. Fried, Arithmetic variation of fibers in families: Hurwitz monodromy criteria for rational points, J. Crelle 409 (1990), 106-137. [DFr2] P. Debes and M. Fried, Integral specialization of rational function families, preprint. Jan. 1997, 20 pgs. [Fa] G. Faltings, Diophantine approximation on abelian varieties Annals of Math. 133 (1991), 549-576 [Frl] M. Fried, Alternating groups and lifting invariants, Preprint as of 07/01/96. [Fr2] M. Fried, Fields of Definition of Function Fields and Hurwitz Families and Groups as Galois Groups, Comm. in Algebra 5 (1977), 17-82. [Fr3] M. Fried, Enhanced review of J.-P. Serre's Topics in Galois Theory, with examples illustrating braid ridigity, Proceedings AMS-NSF, Slimmer Conference Cont. Math series 186, Recent Developments in the Inverse Galois Problem (1995), 15-32.
Modular towers
175
[Fr4] M. Fried, Global construction of general exceptional Covers: with motivation for applications to encoding Applications and Algorithms, Cont Math. 168, G.L. Mullen and P.J. Shiue, editors, 1994, 69-100. [Fr5] M. Fried, Arithmetic of 3 and 4 branch point covers: a bridge provided by noncongruence subgroups of SL2(Z), in Progress in Mathematics 81, Birkhauser, 1990, 77-117. [FrJ] M. Fried and M. Jarden, Field Arithmetic, Ergebnisse der Mathematik III, 11, Springer Verlag, Heidelberg, 1986. [FrK] M. Fried and Y. Kopeliovich, A*, Modular Towers, 30 page preprint, 1996. [FrV] M. Fried and H. Volklein, The inverse Galois problem and rational points on moduli spaces, Math. Annalen 290 (1991), 771-800. [MT] M. Fried, Modular Towers: Generalizing the relation between dihedral groups and modular curves, in Proceedings AMS-NSF Summer Conference, Cont. Math series 186, Recent Developments in the Inverse Galois Problem, 1995, 111-171. [HR] W. Herfort and L. Ribes, Torsion elements and centralizers in free products of profinite groups, J. Crelle 358 (1985), 155-161. [Me] J-F. Mestre, Extensions regulieres de Q(T) de groupe de Galois An, J. Alg. 131 (1990), 483-495. [MFo] D. Mumford and J. Fogarty, Geometric Invariant Theory, Ergeb. der Math, und ihrer Grenzgebiete 34, Springer Verlag, 2nd enlarged edition, 1982. [Ri] L. Ribes, Frattini covers of profinite groups, Archiv der Math. 44 (1985), 390-396. [R] A. Robert, Elliptic curves, Lecture Notes in Mathematics 326, Springer-Verlag, Heidelberg-New York, 1973. [Se] J.-P. Serre, Topics in Galois Theory, Bartlett and Jones Publishers, 1992. [V] H. Volklein, Groups as Galois Groups, Cambridge Studies in Advanced Mathematics 53, Cambridge Univ. Press, 1996. University of California at Irvine, Irvine CA 92717 [email protected]
Part III. Galois actions and mapping class groups
Galois group GQ, Singularity 2?7, and Moduli Ms Makoto Matsumoto Monodromy of iterated integrals and non-abelian unipotent periods Zdzislaw Wojtkowiak
Galois group GQ, Singularity £7, and Moduli M3 Makoto Matsumoto
An Artin group is a generalization, defined for any graph, of the usual braid groups. The Artin group of E-? Dynkin diagram has a natural surjection to the mapping class group of genus three curves, defined by sending the canonical generators to a set of Dehn twists, called Humphries generators. We first show that this group homomorphism comes from a geometric morphism between moduli spaces by taking TTI. Next, we describe the notion of tangential morphisms. Let U be an affine variety and D\,..., D/ be normal crossing divisors which are principal. The tangential morphism from D\ n • • • D D\ to U - (Di U • • • U D/) is defined when we specify the generator of the defining ideal of Di for each i. This yields a morphism between the algebraic fundamental groups. A tangential morphism from a point Specfc coincides with the notion of fc-rational tangential base point. Using this, we describe the action of the absolute Galois group on the profmite completion of Artin groups of type An, Bn, C n , Dn, and E7. (The first type gives the Braid group investigated before.) Description uses only the Belyi's coordinates, i.e. the action of the Galois group on the algebraic fundamental group of F 1 minus 0,1, 00. As a corollary, we give an explicit description of the Galois action on the profinite completion of the mapping class group of genus 3 curve.
Contents §0. Introduction §1. E7 and M 3 1.1. Fundamental group of moduli stack 1.2. Humphries generators and Artin group of Ej type 1.3. Variety V(E7) §2. Tangential morphisms 2.1. Tangential morphisms 2.2. Homotopy between tangential morphisms 2.3. Composition of tangential morphisms 2.4. GAGA 2.5. Example
The author was supported by Japan Association for Mathematical Sciences for attending the Luminy conference.
180
Makoto Matsumoto
§3. Galois action on Artin groups 3.1. 3.2. 3.3. 3.4. 3.5. 3.6.
Motivation Tangential base point on V(E?) Pj l o o and P j ± l o o Galois action on each generator The result on E7 and M 3 An, Bn, C n , Dn cases
§0. Introduction This manuscript consists of three parts. The first two parts are logically independent, and the last part uses the results of the first two. The first part of §1 is purely geometric, and is independent of the Galois group. We recall Brieskorn's result [5], [6] on the deformation of ^-singularity, which says that (1) the smooth locus of the deformation space has the E7 Artin group as the fundamental group, and (2) on the smooth locus, each fiber is a genus three curve. By (2), we have a classifying map from the smooth deformation space to the moduli stack of genus three curves. Our first result is that by taking TTI of this morphism, we have a group homomorphism from the E7 Artin group to the mapping class group of genus three curve, which maps each generator of the Artin group to the Humphries' Dehn-twist generators of the mapping class group. This is an interesting coincidence, since the two sets of generators were invented independently. Then the algebraic meaning of the Humphries generators are clear, so we move to the investigation of the Galois action. As a preparation, in the second part of §2, we give the notion of tangential morphisms. This notion is a slight generalization of the tangential base point introduced by Deligne [7], developed and effectively used by Ihara, Nakamura, and the author [15] [17] [22] [18] [27] (Ihara and Nakamura even dealt with the formal scheme version in [18]). Note that there are also log-scheme theoretic interpretations, which fit to motivic fundamental groups [10] [30].) In the third part §3, we give a construction of a tangential base point for the deformation space of the E7 singularity. (The construction mimics that for An in [17].) We give seven tangential morphisms from P 1 minus four points to the deformation space, then the Galois action on the Artin group is written in terms of / a , x(a)i which appears for P 1 minus three points. As a corollary, the Galois action on profinite mapping class group with genus three is described. We also treat some other type of Artin groups, Ani Bn, Cn, Dn. The motivation of this research is as follows. Let V be a geometrically
Galois group GQ, Singularity E7, and Moduli M3
181
connected variety over the rational number field Q, and x be a Q-rational point. Then, we have a group action of the absolute Galois group GQ = Gal(Q/Q) of Q on the profinite completion of the fundamental group of V as a complex variety [11]: pVjX : GQ -> Aut7rT(y(C),x) ^ Aut7r^(F,x) (here V denotes the coefficient extension V ® Q). Grothendieck conjectured1 that V can be recovered from /9y)X, for any "anabelian" variety V. Also he asserted the importance [12] of pv,x with V moduli spaces of (,n)-curves, where g is the genus and n the number of punctures. If V = FQ1OO (i.e.,?1 minus three points 0,1,00), then pv,x is injective (Belyi [1]) (actually is injective for any affine curve V with nonabelian fundamental group [22]). Since PQIOO 1S t n e moduli space of four pointed P 1 , and since it is the simplest, we have a conjecture that pv,x can be "described naturally" in terms of ppi b under good choices of b on Pm™ a n d x on V, Oloo'
for V moduli of some good objects2. This is the case for V the configuration space of n points on the affine plane. In this case, the fundamental group of V is the n-string Braid group Bn. We can choose canonical base point x, 6, which are actually on the boundary of the spaces (i.e., tangential base point of Deligne [7]), such that the image of the each canonical generator of Bn by a £ GQ can be written down in terms of those of 7Ti°p(Poioo(Q'^) simply by substitutions (note that 7ri°p(Poioo(C)>fr) is a free group of two generators x, y, that a(x),a(y) is an element of the profinite completion, and substituting x, y in cr(x), cr(y) to some elements in another profinite group makes sense). This was proved by Ihara and the author [17], where the insight comes from Drinfeld's result on the action of the Grothendieck-Teichmiiller group on the pro-unipotent completed braid groups [9]. Here we will describe some generalization of this, i.e., for V deformation space of rational singularity of surfaces (the above case corresponds to the singularity of An_i~type). Then, using a tight relation (see Theorem 1.3) between the deformation of E7 singularity and the moduli stack M3 of genusthree curves (see also Looijenga [21]), we extend the result to V = M 3 (see Theorem 3.1). Note that this conjecture for curves has been settled. The genus zero case was solved by H. Nakamura (see his integrated paper [25]). Recently, A. Tamagawa [33] settled this for affine curves (also finite field case), and S. Mochizuki [23][24] for the proper case (he proved a stronger p-adic version). After the author's talk on the genus three case in Luminy, H. Nakamura [27] succeeded in describing the Galois action on mapping class groups with any genus g by P 1 — {0,1, 00}. In the talk at Luminy, the author depended o n P 1 — {0, ± 1 , 00} rather than the desirable P 1 — {0,1, 00}. Now the result of this manuscript involves only P 1 — {0,1, 00}, and this improvement comes from the stimulation by the Nakamura's success. The author is thankful to him for informing him of his beautiful result.
182
Makoto Matsumoto
We shall exchange the order of the explanation: explain first about E7 and M3, and next about Galois actions. This is because the former is purely geometric and independent of the mess of the Galois actions, and it seems better for the readability.
§1. E7 and M 3 1.1. Fundamental group of moduli stack Let Mg^n denote the moduli stack of the genus g n-ordered pointed curves [8] over. The coarse moduli scheme of M^ n , equipped with C-topology, is analytically isomorphic to the quotient of the Teichmiiller space by the mapping class group. Then, this analytic space has a trivial fundamental group. However, there is Oda's theory of the fundamental group of analytic stacks [28]. Roughly speaking, considering the fundamental group of the stacks corresponds to considering the orbifold, and ^n{Mg^n ®Q C) is proved to be isomorphic to the mapping class group F ^ n of the genus g oriented surface fixing n distinguished points. Moreover, the monodromy representation of the universal fiber M9in+i —>• Mg^n, i.e., the outer monodromy action of the fundamental group of the base space on that of a fiber, coincides with the canonical morphism Tg^n —» Out7r*op(E^>n), where E^>n is the n-punctured genus g oriented surface. For n = 0, we simply write Mg = M9JQ, Tg = F^o, a n d T,g = S^o1.2. Humphries generators and Art in group of i^V-type It is known that the mapping class group Tg is generated by the following 2# + 1 Dehn twists [13], which are called the Humphries generators.
The relation of these Dehn-twist generators was determined by Wajnryb. There are four types of relations (for the explicit description, see Wajnryb [34] and Birman's survey article [3]).
Galois group GQ, Singularity E7, and Moduli M3
183
(i) Braid relations: DDf = D'D, if D and Df have no intersection. DD'D = DD'D if D and i)' have one intersection point (the number of intersections is at most one among the above curves). (ii) Monocular (or chain) relation, which is supported by the five leftmost twists. (iii) Lantern relation. (iv) Commutativity with the hyperbolic involution. The relation (i) shows similarity with the braid group. We write an intersection diagram F of these 2g + 1 twists. Thus, F has 2g + 1 vertices corresponding to the 2g + 1 curves, and 2g edges corresponding to the intersection point. The group generated by 2g + 1 generators with relation (i) is called Artin group of type F, and denoted by A(T). In general, let G be any graph each of whose edge e is labeled with non negative integer rae. Then, the corresponding Artin group A(G) is defined as the group with the vertices as generators and the relations aba • • • = bab - • • m e +2
m e +2
for each pair a, b of vertices. Here, e is the edge between a and b, me is the number assigned for this edge, and if there is no edge then we put me = 0, i.e., a and b commute. In this terminology, the braid group of n strings is nothing but the Artin group of straight line with n — 1 vertices. Brieskorn [5] [6] proved the following profound theorem. If the graph is a Dynkin diagram of some semisimple Lie algebra over C, then the associated Artin group is realized as the fundamental group of a rational variety, denoted here by V(G) with G the type of Lie algebra. This V(G) has a moduli interpretation. It is the smooth fiber locus of the miniversal deformation space of the rational singularity corresponding to the Dynkin diagram. Actually, the Artin group is defined to describe the fundamental group of these varieties. The braid group appears for An-\ singularities, to which the corresponding rational variety is the configuration space of n points on a plane. Now, we consider the case g = 3. Then, the intersection diagram of Humphries generators coincides with E7 Dynkin diagram. So, there is a surjection from the Artin group of E7 type to the mapping class group of genus 3: A(E7) -» r3, by mapping canonical generators to canonical generators. Now, a natural question arises: does this morphism come from an algebraic map?
184
Makoto Matsumoto
The answer is yes (see Theorem 1.3 for the precise statement): there exists a natural family of two pointed genus three curves over V(E'7) such that the classifying map to M 3 obtained by forgetting two points induces A(E7) -» F 3 by taking the fundamental groups. The rational variety V(E?) can be described explicitly by using the Weyl group, and the Galois action on n^ 9(V(E?), x) can be described in terms of that on ^1^(^0100^)7 for some appropriate tangential base points x, b. Thus, by the functoriality of TT^P, we can describe the Galois action on IV This will be explained in §3.
1.3. Variety V(E7) is a miniversal deformation space of £7 singularity. Brieskorn solved conjecture of Grothendieck on the explicit construction by Lie group [5] of this deformation space. Here we need this construction. For the reference on the following facts, see Slodowy [31]. Let $ be a root system of a semisimple Lie algebra, i.e., one of An, Bn, C n , D n , EG, £7, Eg, F4, G2. This is a finite subset of a vector space U over Q. The set of reflections generates a finite group, called the Weyl group of type <$. For each root a G $, let Ha be the reflection hyperplane (i.e., the fixed divisor by the reflection along a) of the complexified vector space Uc := U <8>Q C. Let us define V($>) as the quotient of Uc — UaHa by the action of the Weyl group. This variety V(&) is known to be a smooth rational variety defined over Q by invariant theory: the quotient of the affine space by a finite group G is an affine space if and only if G is a reflection group, i.e. generated by (possibly nonorthogonal) reflections (see for example Theorem 4.2.5 in [32]). Theorem 1.1. (Brieskorn [5] [6]) The fundamental group ofV($) is isomorphic to the Artin group of type <1>. This isomorphism is given as follows. Take the fundamental Weyl chamber C in U^ as the base point of V($>) (since the chamber is simply connected, this makes sense). For each fundamental root a, take a path in Uc — Ha from this chamber to the reflection image of C with respect to a, which goes around the divisor Ha counterclockwise. Then, taking the quotient by the Weyl group, we obtain a closed path in V($>). This is the corresponding generator o/7rJop(y(),C) to the generator associated with a in the Artin group The simplest example is An type, which yields a usual braid group. In this case, U is the set of vectors (xi, #2, • • • ? #n+i) with ]T^ Xi — 0. If we denote by e?; the z-th unit vector ( 0 , . . . , 0 , 1 , 0 , . . . , 0), then $ = {e; — ej\l
Galois group GQ, Singularity £7, and Moduli M3
185
inner product with any fundamental root is positive, i.e., (#i,.. with xi > X2 > - - > xn+\. The reflection associated with e^ — ej is the exchange of X{ and Xj, and the Weyl group is the symmetric group 5 n + i. The reflection hyperplane is the locus with xi = Xj. Thus, V(An) is (U — the locus with Xi = Xj for some i ^ j) divided by Sn+i, i.e., the configuration space of unordered n + 1 points. Its fundamental group is the braid group, i.e., A(An). It is easy to see that the path described above corresponds to the standard generator of the braid group. In §3, we describe the Galois action on the profinite fundamental group of some of these varieties. Each V($) has a natural meaning as a moduli of singularities as follows. We don't use the universality in this manuscript, so we let the reader consult the reference for the definition of universality of these moduli spaces. (For the case of Bn, Cn, F4, G2, see [31].) Theorem 1.2. [5] Let $ be a root system of A,C,E type. Then, the germ of U divided by the Weyl group near the origin is a miniversal deformation space of the rational singularity of$ type. That is, there is a family of curves on U, such that the fiber at the origin is a curve of $>-type singularities, satisfying the universal property for the deformation. The quotient ofUaHa by the Weyl group is the locus with singular fibers. By the universality of M^ n , we have a classification morphism from V($) to Mg^n for suitable g, n. If $ = A n _i, then it is known that the deformation of the singularity is given by the equation y2 = xn + aixn~1 + a,2Xn~2 + • • • -f a n , i.e., the hyperelliptic curves. Here (a x ,... ,a n ) is the parameter of kn/Sn. Thus, if (61,..., bn) is the parameter of A n , then a's are the fundamental symmetric polynomials of 6's. After compactification of the fibers, we have V(An) -> Mg with n = 2g-\-2. The image is the hyperelliptic locus of Mg, and it can be easily checked that the generators of 7Ti(V(.An)) are mapped to the Dehn twists along the following circles.
This group homomorphism is surjective if and only if g < 2. For g = 3, we can consider an E7 analogue.
186
Makoto Matsumoto
Theorem 1.3. The deformation of singularities of E7 type gives a family of two pointed curves with genus three. By forgetting the two points, we have a classification morphism V(E7) —> M 3 . Then, the corresponding group homomorphism ni(V(E7)) —> Ts is given by mapping the generators of the Artin group to the Humphries generators ofT%. Proof. The E7 singularity is given by the equation y(y2 — x3) = 0. It is known that a smooth deformation of this singularity becomes a twopointed curve with genus three. See for example [31], or, consider a small deformation of the one punctured cusp y2 — x3 = 0 to an elliptic curve. This yields y(y2 — x3+a) — 0, which is the three points normal intersection of onepunctured projective line and one-punctured elliptic curve. By deforming the three points, we obtain two pointed genus three curves. By forgetting the punctures, we have a smooth family F —>• V(E7) of genus three curves. Now we have the classifying morphism V(E7) —> M3. We shall show that the n 1 gives the canonical surjection A(E7) —> F 3 . Since F —> V(E7) is a smooth family of genus 3 curves, by the universality of Mg^i —>• Mg, we have the following pullback diagram:
V(E7)
. Mg.
This yields a compatible monodromy map
where Tg is the genus g mapping class group. Since the lower map is injective (Dehn-Nielsen's Theorem), it is enough to show that the image of each generator by the monodromy
in(V(E7)) ^ Outfapg)) coincides with prescribed Dehn twists. First, in the miniversal deformation of singularities, it is known that the codimension one component of the singular locus corresponds to onepinched Riemann surface, and the corresponding monodromy is a Dehn
Galois group GQ, Singularity E7, and Moduli M3
187
twist. Thus, we may assume each generator of the Artin group is mapped to a Dehn twist. Moreover, this Dehn twist is not trivial, even if we pass to the monodromy in the first cohomology [31]. Next, we use Ishida's result [19]: Let C, C be two Dehn twists in Tg, g > 2, along the curves with same notation. We choose the curves from their homotopy classes so that the number of the intersections is the minimum. Then, Ishida's result is: (i) C and C" commute if and only if C D C = 0, (ii) C and C" satisfy the braid relation CC'C = C'CC if and only if C u C " =one point, (iii) C and C satisfy no relation, i.e., generate a free group, otherwise. Now we shall complete the proof. We put the names Co, Ci, G2, C3, C4, C5, Ce for the generators of A(E?) and give the same names to the closed curves in E^, such that the corresponding Dehn twists are the images of Gi's in T3.
-o
C,
C
c,
C
C5
Q
Then, up to homotopy equivalence and diffeomorphisms, there are only finite possibilities of C\ in T,g: the one which does not separate E p , or, the one which separates E 9 into two pieces with genus #1, g
Next, we consider the image of C2. The tubular neighborhood of C±, C2, respectively, is a tube. Since C\ and C2 intersect at exactly one point, the tubular neighborhood of C\ U C2 in E^ is the two tubes glued along a square neighborhood of the crossing point. This is homeomorphic to a torus
188
Makoto Matsumoto
minus a closed disk. Then, the whole surface is a union of this torus and the complement surface, glued at the boundary homeomorphic to a circle. If we redraw the picture along this decomposition, we may assume C2 is as follows, up to a homeomorphism.
We shall consider C3. C3 intersects with C2 at one point, and does not intersect with C\. Thus, the tubular neighborhood of C\ U C2 U C3 is isomorphic to a genus one surface with two closed boundaries. There are two kinds of possibilities: these two closed boundaries are connected in the complement of the tubular neighborhood in £ p , or not connected. If not connected, then we can't choose C4, which intersects at one point in C3 and has no intersection with the others. Thus, the two boundaries are connected in the complement, so we may assume that C3 is just as expected. Similarly, we can prove that C;, i = 1,2,3,4,5,6 are like the following picture.
Now, Co intersects with only C4. If we cut Ci,C2,C3,C5,C6, then we obtain a tube with several disks removed. Since Co does not intersect with these curves, we conclude that the only possibility is DQ or its mirror image Df0. This completes the proof. 0 §2. Tangential morphisms First we will give a rough intuitive explanation on the notion of tangential morphisms, which includes as a zero dimensional case tangential base points. Let V* = Spec A be an affine variety over a field k C C, and let £1,..., t\ be regular functions on V*. Let Di,...,Di denote the zero divisor of ti,...,ti, respectively, and assume that these are normal crossing. Put Z := Dx H D2 n • •. n Dt and V := V* - Dx U D2 U • • • U Dh Then, a
Galois group GQ, Singularity £7, and Moduli M3
189
tangential morphism Z —r V is a morphism from Z to F , obtained by "infinitesimally small perturbation" of Z c-> V* "in the direction" specified by We regard Z, F* as complex analytic spaces via k C C. Then Z is the locus in V* defined by £1 = t2 = • • • = ti = 0. We perturb Z to obtain Z', which is defined by t\ = t2 = • • • = t\ = e, with infinitesimally small positive real number e. Since D^s are normal crossing, Z' is homeomorphic to Z, and we have Z ^ Z' -> V C V* and 7r*oP(Z) 2 T T ^ Z ' ) -> 7r*op(V). Less trivial is the algebraic structure, which is explained in the next section. In the case that Z is a point Spec k, we have a section -K\ p(Specfc,Spec k) —> 7Tj g(V, Z), which is nothing but a fc-rational tangential base point. In §2, we need only the characteristic zero case. However, the results in 2.1-2.3, using Galois categories are valid for positive characteristics. We treat these cases in parallel for future reference. 2.1. Tangential morphisms. Let k be a field of characteristic p > 0. We denote by N7 the set of natural numbers coprime with p. Thus, if k has the characteristic zero, then N = N'. Let V be a geometrically connected variety over k. We denote by Vfet the category of finite etale covers of V. We mainly deal with the case V = V* — D, where V* is a variety and D a divisor of V*. If p > 0, we consider V ^ , the category of finite covers of V* that are etale outside D and tamely ramified at D, When it cannot cause confusion, we denote Vf£ simply by Vfet. Vfet is a Galois category. For fiber functors F, F1 : Vfet —> {Finite Sets}, we denote Tr^V^i71, Ff) the set of isomorphisms from the functor F to F1. This constitutes a groupoid. If F = Ff, this is nothing but the algebraic fundamental group. We shall deal with the following case. Definition 2.1. Let V be a Zariski open subscheme of a geometrically connected variety V* over k. Let U — Spec A be an affine open subscheme of V*. We assume that there are regular functions £1, t2, -.., t\ on J7, such that (1) U = 0 (1 < i < I) give normal crossing divisors (each denoted by D{) on U and (2) V f) U is the complement of these divisors in U. Let Z be the intersection of Di, i — 1,2,...,/. The above datum (V, V*, U, £1,..., ti, Z) is called a tangential morphism from Z to V, denoted by r : Z -7 V. (I.e., r : Dx n • • • Pi Dt -r U - (Di U • • • U A ) C V\) This tangential morphism is called fc-rational. Let L be an algebraic extension field of A:. A tangential morphism to V®L is called an Irrational tangential morphism.
190
Makoto Matsumoto
If Z is a point Specfc,then r : Z —7 V is called k-rational tangential (base) point A tangential morphism induces an exact functor from Vfet to Zfet, and hence a morphism between fundamental groupoids, as follows. Let U be the completion of U along Z, i.e. Spec A := Spec lim«_nGN where / z = ( i i , . . . , i i ) i s the ideal defining Z. Let us denote by Z{t) := Z{t\,..., t{\ the spectrum of
where £*' denotes a compatible system of JV-th roots of U € A. Z is identified with the closed subscheme of Z{t} defined by t\ = • • • = t\ = 0. When Z is Specfc,by the regularity of V, Z{t} = Spec &{£i,..., £j}, which is the ring of Puiseux series over A;. There is a natural morphism Z{t} ->> V* obtained as Z{t} -> U -+ V*. The image of this map is contained in any Zariski open neighborhood of Z in F*. Let W be a finite etale cover of V. Then we take the normal closure of V* with respect to W, and call it W*. In a neighborhood of Z in V*, the possibly ramified locus is t\ • • • t\ = 0 . We pullback W* along Z{t} —> V*, and denote the fiber product by W*|z{t}- Take the normal closure of Z{t} with respect to the generic fiber of the fiber product, and denote it N(W\z{t1,...M})' Then this is finite etale over Z{£i,.. .,fy}. This follows from Abhyankar's Lemma for Z\t\,... ,t{\ —> U, since the ramification is preserved by completion. Then we obtain an etale cover of Z by base change, which is denoted by rJet(W). We have constructed a functor r^et : Vfet —>• Zfet. This functor preserves the fiber product, since the generic fiber of W\z{tlt...,ti} d ° e s n ° t change after taking normalization. By uniqueness of normalization, taking fiber products commutes with normalization. This functor maps direct sum to direct sum and epimorphisms to epimorphism, for the same reason. Thus, by SGA1 [11], Tfet is an exact functor between Galois categories, which yields a morphism between groupoids TTI(T) : iri(Z]F,Ff) —» T T I ( V ; T ( F ) , T ( F ' ) ) , where F,F' are any fiber functors of Zfet, and r(F) is the fiber functor F o Tfet : Vfet —> {Finite Sets}. Now we proved: Proposition 2.1. For a tangential morphism r : Z —r V', we have an associated exact functor r^et : Vfet —> Zfet, which induces a homomorphism of groupoids 7n(r) : 7n(Z;F,F') -> ^{V\ Assume Z = Spec A:. In this case, r is called a k-rational tangential point. This yields a group homomorphism ?ri(Spec k,x) —>- 7ri(V,r(x)),
Galois group GQ, Singularity £7, and Moduli M3
191
where x : Specfi -> Spec k is any geometric point. This is a section to the structural morphism TTI(V, T(X)) -> 7Ti(Specfc,x). It is easy to check this at Galois category level. This is nothing but a tangential base point considered in [7] [15]. This yields an action of G^ = 7Ti(Specfc,x) on 7Ti(V; r(x), F) for any fiber functor F from left by a € G^ : 7 6 7ri(V;r(x),F) 4 7 0 7Ti(r)(cr~1), and similarly an action from left on Ki{V; F, T(X)) by 7 I-» TTI(T)(
7r1(V;r(x),rf(x))
is the inverse image of the unit element by ni(V; r(x), T'(X)) —> ?ri(Spec h, x). Here, f is the fc-rational base point, which is obtained by extending the coefficient in the construction of Tfet. By 7 •->> 7ri(r/)(cr)77Ti(r)(cr)~1, 7Ti(Specfc,x) acts on 7TI(V;T(X)^T/(X)). If ; r = T , then this is an automorphism of KI(V,T(X)). This representation Gal(k/k) —> Aut7ri(F, r(x)) is what we want to study. 2.2. Homotopy between tangential morphisms. Let F, G : C -> V be two exact functors from a Galois category to another. We call an isomorphic natural transformation from F t o G a homotopy from F to G. If the target category V is {Finite sets}, then F and G are fiber functors, and a homotopy between fiber functors is nothing but an element of 7Ti(C; F, G). This is an analogy to that a path is a homotopy between two points. Let T,T' : Z —rV be two tangential morphisms, which comes from Z{t} — Z{h, ...,*/}-»• V*, Z{s} = Z{su ..., si) -> V*, respectively. Definition 2.2. An infinitesimal homotopy from r to rf is an isomorphism Z{t} —> Z{s}, commuting with the morphisms to V*, and mapping the divisors (ti) to (si) for each z. Proposition 2.2. An infinitesimal homotopy yields a homotopy from rjet t0
T
'f*ef
Proof. Let W be a finite etale cover of V. The normalizations on Z{t} and on Z{s} are canonically isomorphic, since they coincide outside the normal crossing divisor. Thus, we have a functorial isomorphism Txet(W) —> Tf*et(W), which yields a homotopy. 0 This proposition says that the tangential morphisms essentially depend only on the choice of the basis of normal cotangent bundle of Z in V*. For example, t{ »-> ^(1 — tj)~1 does not change the morphism. This proposition is useful when we compare two distinct tangential base points which are mapped each other by an automorphism of V.
192
Makoto Matsumoto
Another interesting example is Q{t} -» FQ1OO and Q{-£} -» Poioo- These two are not infinitesimally homotopic. If we add e2ni^N for all iV, then we have an infinitesimal homotopy Q°6{*} -> Q*b{-t} by t^N -> e ^ e 1 ^ . This expresses the fact that a small half circle from 01 to 10 yields a path, on which GQC*& acts trivially. This path realizes the infinitesimal homotopy. By taking the conjugate by this path, we may identify the action of GQab on these two Q0'6-rational tangential base points. If we change the base field to Q"6, then this is true for any Q°6{C*} with £ a root of unity. We can regard this as Spec Qlb homotopically moves along an infinitesimally small circle around zero. A similar thing can be said about Q{t} and Q{a£}, with a a rational number. There exists a homotopy if we "restrict" to 2.3. Composition of tangential morphisms. Let us consider the diagram V
Here, r and rf are tangential morphisms which are composable in the following sense. The vertical inclusions are open immersions, and the horizontal arrows are immersions. There is an open affine subscheme U of V**, on which there are functions t\,..., U, S\,..., s m . (1) (V, F**, [/, * i , . . . , U, si,..., 5 m , Z') gives a tangential morphism Z1 —r V, which is symbolically denoted by rrf. Thus, Zf is defined by ti = • • • = ti = si = • • • = sm = 0 in [/, and V C\ U is the complement of t\.. .tiSi.. .Sm = 0 in U. (2) (V, V*,U H V*,ti,...,thZ) gives a tangential morphism r : Z -r V. (3) Z* is the closed subscheme of U defined by ti = ... = ti = 0.
(Z,Z*,Z*,si,...,
sm, Z') gives a tangential morphism r' : Z1 —r Z.
Proposition 2.3. There is a canonical homotopy between two functors rTfet • Vfet -»• Z'fet.
Galois group GQ, Singularity E?, and Moduli M3
193
Proof. Let W be a finite etale cover of V. rjfet is defined by the following diagram () -N(W\Z{t}) *W
Z{t}
•V*
where • denotes the fiber product, and B3 denotes: to take the fiber product and take normalization of the base scheme at the generic fiber (denoted by N()). It is easy to see that this double-square diagram extends to N(r}et(W)) D •Z*{t}
Z*
By uniqueness of normalization, we have a unique morphism N(W\z{t}) ~^ N(W\z+{t}) which is a pullback along Z{t} ->- Z*{t}. Then, by pulling back along Z -» Z{t}, we have r*fet(W). N(rfet(W)) is the pullback of N(W\z*{t}) if restricted to Z, and hence on all Z*. Next we consider N(T/et(W)\z.{t}) D
Z'
z*
and
(TT%tW
—+N(Tfet(W)\Z.{ttS}) D
Z'
•Z'{t,s}
There exists a canonical morphism from the upper diagram to the lower one extending the pullback Z* —> Z*{t} at the right end. This is because when restricted to the locus t / 0 of Zf{t}, E3 is just a pullback, and the normalization is unique. Now we have a canonical morphism r?et o T fet(W) -* (TT')*fet(W) w n i c n is a pullback of the identity Z1 -> Z1, hence an isomorphism. §
Makoto Matsumoto
194
2.4. GAGA. In this section, we assume A; to be a subfield of C and denote by k the algebraic closure in C. 2.4-1° Tangential morphism without completion. In the construction of r^et in 2.1, the completion of U along Z is actually unnecessary. For a finite etale cover W of V, we obtain a finite etale cover r^et{W) of Z as in the following diagram: -N(W\Ul/N)
ul/N
W*
U
is defined by
where the right down arrow U —>• A1 is given by (£i,...,fy), A* —>• A1 is given by ti *->• sf1 with N a common multiple of the ramification index of each divisor corresponding to U = 0, and O is the origin of the affine space. This gives the same rjfet, since by Abhyankar's Lemma N(W) is etale over and Z -> UI/N factors through Z{t}. 2.4-2. Analytification of a tangential point. For an algebraic variety V over fc, we denote by V the base extension to k C C, and by Van the complex analytic variety corresponding to V. It is known that the category of unramified finite covers of Van is canonically equivalent to the category of finite etale covers of V (SGA1 [11]). For a point Q of Van, there is the associated fiber functor FQ : Vf™t —>• {Finite sets}. For another point i?, the homotopical lift of a path in Van from Q to R yields an isomorphic natural transformation from FQ to FR. This gives which becomes an isomorphism if we complete the left set to 7r1op(Van; Q, R)/N, where N runs over the finite index normal subgroups of \
Galois group GQ, Singularity £7, and Moduli M3
195
In the case of a geometric tangential point, there is no corresponding topological point on Van. So, we content ourselves with giving an isomorphism between the fiber functor coming from a tangential point and another fiber functor coming from a geometric point. Let 9 : Spec k —7 V be a geometric tangential point. Thus, we have an immersion F c F * , open affine subset U C V*, with V Pi U = U — D where D is the union of the zeros of ti,... ,t\ G F([/, OJJ), which are normally crossing. Let P be the point on U defined by £1 = ... = £/ = 0. Since U = 0 are normal crossing, (£1,... ,£/) gives a local coordinate at P, and induces an analytic isomorphism from an open neighborhood Mp of P in Uan to an open ball. Fix a point Q in this neighborhood, and identify Q — (ci,... ,cj) by these coordinates. Let Afp^i/N C U*?N be the inverse image of Np C Uan. This is again an open ball. Fix a compatible system of c\/N G fc, 1 < i < /, for N G N. Put QN := (cx , . . . , Cj ) G Np^i/N- Then, we can specify an isomorphic natural transformation from r^et to FQ, as follows. Let V,V*,U be as in 2.1, If be a finite etale cover of V, and Then, the finite set FQ(W) is the fiber of W -> V above Q G F , which bijects to FQN(N(W\UI/N)) by C/ v ^ ^ U. Since JV(W|^ 1/N ) on ^ ^ is etale, the inverse image of MP^/N is isomorphic to the disjoint union of copies of MpA/N. Now both FQ^(N(W\Ul/N)) and 9*fet = ^ ( ^ ( ^ 1 ^ , ^ ) ) biject to this set of the copies. It is not difficult to see that this defines a natural transformation 0 —> FQ. Now we have fixed a "path" from 9 to FQ. There is a canonical way to choose Q and QN- We take c» (1 < i < I) to be small enough positive real numbers. We choose the c/ to be all positive real. (This choice is then unique.) Let Qf = (c' l 7 ..., c[) be another such point. It is easy to see that the natural transformation FQ —> FQ< defined by a path from Q to Q' in the positive real region makes the following diagram commute. <9*
9*
(This is because the positive real path between Q and Qf lifts to that between QN and Q'N in Ui/N.) Definition 2.3. We identify these fiber functors FQ for any such Q, and denote it simply by 9an.
Makoto Matsumoto
196
2.4-3. Compatibility. We consider the case where k = k and Z' is a geometric point in 2.3. Let r : Z —r V be a tangential morphism, and 8 : Z' = Spec k —7 Z be a geometric tangential point, such that we have a composite tangential point T6 : Spec k —r V, as in 2.3. We denote by P the corresponding geometric point of V** to ti = • - • = ti = si = - • = sm = 0. We assume another such 9f : Spec k —r Z exists, so that we have a tangential morphism r6f. Then, we would like to understand the upper morphism in
n{op(Zan', 6an, e'an
- 7r[op{Van; (r6)an,
(TO')™
By definition of 0\ and the composability of rd\, we have a commutative diagram .V*
(unv*)t,i/N
Q QN
P = Spec k
•u*t,l/N
Here, (U n V*)I/N etc. is defined by
(unv*)til/N—^u D
z* uSjtA/N u
•
t
• N
S
•
S,t
N
N
P is defined by si = • • • = sm = t\ = • • • ti — 0. Q is a point in Z chosen to analytify #i, whose coordinates are ( s i , . . . , sm) = (ci,..., c m ). Since we fix the iV-th roots of c^'s, we have QN on Z* 1*N. Let W be a finite etale cover of V. Take AT as a common multiple of all ramification indices over F**. Then, 6* o r*(W) is identified with the fiber
Galois group GQ, Singularity £ 7 , and Moduli
197
of a suitable etale cover of Z*X,N on the point P . This is identified with the fiber on P in N{W\Ustl/N)
-> Ua,t,1/N, where
P = Spec k -> U8ytii/N
is defined by 5 = £ = 0. We denote the functor taking the fiber on P by Fp, which is identified with (T0)*. We fix Q e V**an, whose coordinate is small positive real numbers: (si,... , s
m
, t i , . ,.,ti)
= (ci,.. .iCm,di,..
.,d{).
Then we have -Q
QN
For #', we have a similar construction to obtain Fp> ,FQ» ,Fp,,FQ,. The coordinates for 0' is given by s[,..., s'm, but we take the common £1,..., £j and d i , . . . , d/. What we want to see is the following. Let 7 be a topological path from Q to Q' in Zan. Then, which morphism can be substituted for the question mark in the diagram below, in order to make the diagram commute? Fp
=
Fn
•-
FQI
=
-
F
Q
Fp>
FP..
We can regard FP,FQ,FP,FQ, respectively, as a functor to take a fiber on P,QN,P,QN € U8itii/N, respectively, in the etale map N(W\Ustl/N) -> an Us,t,i/N- Let Mp C U be the open ball neighborhood of P , and , be its inverse image by Usjyi/N ~^ U, as in 2.4.2. Then, P = P , QA / l/N
1/N
n
nx
',' / ;
/ l/AT
1/N
A/N
il/Nx
(Ci ,...,Cm , 0 , . . . , 0) and QN = {ci ,...,Crk , d / , . . . , dz 7 ) are on Afp,N- Since A/p,iv is simply connected, all the functors Fp, FQ and are identified. Similar things hold for 0'. Thus, the problem is reduced to obtaining FQ
^
^
FQ,
198
Makoto Matsumoto
We may consider these four functors as taking fibers in N(W\ut Ut,i/N o n t n e points RN,R'N,RN,R'N, where RN = ( c i , . . • , c m , 0 , . . . , 0 ) ,
RN = ( c i , . . . , c m , d /
1/N)
—>
,.. . , d / )
in the (s, t) coordinate, and
R'N = (ci, . - . , 4 , 0 , . . . , 0), JJ'N = (ci,..., 4 , dJ / N ,..., d)'N) in the (s',£) coordinate (note that N(W\ut l/N) —> £^,I/JV is etale on these points). What we need to do is to make a path from Q to Qf in F a n , which lifts in Utti/N to a path from RN to RfN, such that the square of paths RN
RN
*~ R'N
is homotopically trivial in Ut^/N for any N. Let Ze be the inverse image of {(£i,..., ti)\0 < t\ < d i , . . . , 0 < U < d{\ in C/nV* ( t l ^ ' t / } A'. Let 7 be the path from Q to Qf in Z C Z e . Then, we take a path 7' from <5 to Q' in F , which is homotopic to 7 in Z€ when composed with the paths Q -* Q, Q' -)• Q', given by (ti,...,tj) = (0, ...,0) -> (di,..., d/). Now the following is clear. Proposition 2.4. The above 7' zs £fte image 0/7 in 7r^ p (Z a ^0 a n ,0' a r 7-> ^ ( F ^ ; (r0)an,
{rO1)^.
In sum, the above homomorphism 7 •-> 7' is given by the translation of 7 from Z to outside, by perturbing t±,..., £/ (the parameters giving tangential morphism r : Z —r V) to small positive real, so that the perturbation is homotopically trivial in Ze. Proof. Take the inverse image of Z - Ze by (U n V*)tA/N -> U D V*. The connected component with ^ ,.. .,t] > 0 contains both RN and if^, and isomorphic to Z — Ze. If we add Z to this component, then it becomes isomorphic to Ze, and 7' lifts to a path from i?;v to R'N, which makes the square of paths homotopically trivial on (U fl V*)t^/N. 0 2.4-4- Identifiable tangential points. If two geometric tangential points have an infinitesimal homotopy, and the germs of the positive real regions
Galois group GQ, Singularity £7, and Moduli M3
199
coincide, then we can identify them as analytic tangential points, too. In this case, we say these two tangential points are identifiable, and identify through the canonical isomorphisms. 2.5. Example. 2.5.1. Projective line minus three points. For V = PQ1OO = SpecQ[£, 1/(1 £)], we take P 1 as V*. Six Q-rational tangential points are considered in Ihara [15] t, —t, 1 — £,£ — 1,l/t, — l/t, which were denoted by 01, Ooo, 10, ool, 00O, loo respectively. The symmetric group S3 acts on these points. Precisely saying, we need canonical infinitesimal homotopies. For example the image of 1 - t by 0 *+ 00 is 1 - l/t = (t - 1)/*, but Q{(t - l)/t} and Q{t — 1} are canonically infinitesimally homotopic by (t — l)/t K> (t — 1)/(1 -f (t - 1)) = (t - 1) - (t - I) 2 + (t - I) 3 , so we may identify these two. We write T01 : Q —r Pjioo f° r ^e corresponding Q-rational tangential base point, and T5I the geometric tangential point given by Q <->• Q. We have two generators x,y € 7Ti(Poioo'roi)- We have a path p € ^iCoiooJ roi^ rio)- An element a e GQ acts on p by p «->- rioi^proiia)'1 = r Pff
0
0
p
1
0
j>
0
2.5.2. Another example. Let a be a rational number. We consider the (s,i)-plane minus a divisor: V — Specfc[s, t] — D, where D is the union of the following five prime divisors: s = 0, t = 0, ( = 1, st = a, s(l — t) — 1.
Makoto Matsumoto
200
Z,
0
There are two horizontal Pjioo? namely, t = 0, t = 1, and one vertical Poioo> s = 0. We name these Zi,Z 2 ,Z 3 , respectively. For Z\ we choose t for the tangential morphism, for Z 2 we choose 1 — £, and for Z 3 we choose s. Then, we have three tangential morphisms r; : Z; —7 V. The composition of tangential morphisms Q—^PJICXD"^^ lsi Q{S^} f° r ri> Q{as, 1 — t} for r2 , and Q{£, s} for r 3 . The as for r 2 comes from that Z2 is isomorphic to PQICXD if w e t a ^e as as the coordinate. The image of V '- T01 = Q{0 -> rio = Q{1 - t} in Z 3 by r 3 is a homotopy from Q{s, t} to Q{s, 1 — t}. Thus, unless o 7^ 1, we have to fix a path (or homotopy) from the geometric tangential point Q{s, 1 — t} to the other Q{as, 1 — t} to patch Z 3 and Zi. This is done by a homotopy s^N K-> e-^°^a^N(as)l^N from Q{s} to Q{as}, named la. Here we choose the argument of log a to be ni if a is negative. r 2 (/ a ) is a homotopy Q{s, 1 — £} -> Q{as, 1 — t} compatible with Galois action. We write down Galois action on la. Let us define Xa(&) £ Z by aid1'") = e2WNxa(
. Now, set
V\
The actions of a € GQ on #1,2/1 and #3,2/3 are same as that of since T\ and r 3 are tangential morphism over Q: a : Xi 1—> x- ,yi
Galois group GQ, Singularity £7, and Moduli Ms
201
f<j(xi,yi)~lyf^fa(xi,yi) for i = 1,3. For i = 2, we need the action on Zo. We have cr: p \-> pfa(x, y) and a : la »->• laxXa^aK By substitution, we have
and 2/2 -> / a ( X 3 , W 3 ) - ^ a ( * V * ^ This example shows that topologically identical objects may give rise to different Galois actions. Even a scaling in the embedding of PQ1OO influences on the Galois action. §3. Galois action on Artin groups 3.1. Motivation Let V be a geometrically connected variety over the rational number field Q, and x be a Q-rational point of V. Then, we have Galois representations on the fundamental group of V: Pv,x : Gal(Q/Q) -> Autiti(V,x), where TTI(V, X) denote the profinite completion of the topological fundamental group 7r*op(y(C),x). We may use Q-rational tangential base point as x. To obtain an approximation of the Galois group, it is a natural idea to consider all V, and consider systems of elements of Aut7fi(V, x), one for each V, which commutes with any algebraic homomorphism V —> W defined over QWe may start with some restricted families of varieties and restricted families of morphisms. Grothendieck-Teichmiiller group defined by Drinfel'd [9] is the one obtained from M0,4 and M0,5, and some natural morphisms (see Ihara [16]). It would be also natural to consider the families of Mg^n with all g, n. One open problem is whether a new relation occurs or not if we consider g > 0, not only g = 0. For g = 2, it is easy to prove that the GrothendieckTeichmiiller group acts on M2, and hence no new relations will be obtained. So, the first nontrivial example is g — 3. The author knows nothing on this. The author also suggests the possibility to consider the family of V($) with root systems <$, with some (still ambiguous) suitable morphisms. One of the reasons is that these varieties have a moduli interpretation of the deformation spaces of singularities with finite group action [31].
202
Makoto Matsumoto
This idea comes from a trial to find a geometric meaning of an interesting research by P. Lochak and L. Schneps [20]. There, they treated the families of groups Arn, Brn, n > 3, with Brn the braid group of n-strands and Arn the subgroup of Brn consisting of braids that connects the first point with the first point, i.e., the inverse image of the stabilizer of the first element by the morphism Brn -> Sn. Their main result is: if we consider Arn, Brn rc > 3 and suitable group homomorphisms, then we can characterize Grothendieck-Teichmuller group as the automorphism group of a tower of these groups. (In fact, n — 3,4 suffices.) The author just notes that this somewhat strange group Arn is the fundamental group of V(Bn) = V(Cn). The author does not see what are the corresponding algebraic homomorphisms to those group homomorphisms considered in [20]. We shall write down the Galois action on 7tiV($), in terms of that on Pj l o o for $ = An, Bn, Cn, Dn or £ 7 . The author did not study £ 6 , £ 8 , F4 and G2- Since An case was treated in [17] and J5, C, D basically follow from this, we first treat the most nontrivial case £7. 3.2. Tangential base point on V(E7). 3.2.1. £7 root system. An £7 root system is constructed as follows. Consider seven vectors e2 — 63, e$ — e4i e4 — e^, e$ — e§, e& + 67, — -^ X^=i ei m ^ •> where e i , . . . , eg denote the unit vectors. Let U be the seven dimensional subspace of A8 spanned by these vectors. Then, these seven vectors are fundamental roots of £7 root system. Let ^(£7) denote U minus the reflection hyperplanes. If we use the coordinate Yll=2xiei + xs(ei + es) of U, then the corresponding Weyl Chamber C is — x6 < x7 < x6 < x5 < • • • < x2, x8 < —\{x2 + — - + X7) By a general result of Brieskorn [6], the following are the generators of £7 Art in group. o
0
0
0
0
0
0
(i=2,3,-.,6)
- (x 2 +x,+...+x 7 )
Galois group GQ, Singularity E7, and Moduli M3
203
In this figure, the moves of the configuration of the points give a path in U(E7)(C) from C to T^C, where Ti is an element of the Weyl group corresponding to T{. After taking the quotient by the Weyl group, these give the desired elements of 7rJop(V(£V), C), with it Dynkin diagram as follows:
-o X 2
X
X
3
4
X 5
X
T
7
8
3.2.2. Blow up. Let A 7 1 denote seven dimensional affine space with coordinates t2, • • • 5^85 a n d A7a; that with coordinates x2, • • • >#8- We shall define A7^ —> A 7 X , which induces
U(E7) so that the upper morphism gives a Q-rational tangential point (U(E7), V*, Y, t2,..., tg, SpecQ), for a suitable open affine Y C V*. A7t —>• A7X is defined as follows:
X8 = - - ( £ 2 + t2^3 + ^2^3^4 4- M 3 M 5 4-
4- t2t3t4t5t6t7 + 2t2t3tAt5t6(t7
+ l)t 8 ).
These parameters are chosen so that on U(E7) we have the following inverse
204
Makoto Matsumoto
map: *2 = X2
t3 = x3/x2 £4 = x4/x3
t7 =
X7/XQ
ts = ~(x2 + x3 + • •. + x7 The removed divisors are Xi±Xj
=0
(2
7
V^ ±Xf -f 2xs = 0 (the number of minus is even.) i=2
in A7 x. The corresponding divisors in A7* are: t2t3 • • -*i(l ± * i+1 • • -tj) = 0 (2 < i < j < 7) 7
7
] T ± t 2 • • 'U = ^ t i=2
2
--U + 2*2 • -t6{l + t7)ts,
2=2
where the number of minus in the lower identity is even. We denote by D the set of the prime divisors contained in the above-defined locus in A7*. So, D contains the divisors t{ = 0, (2 < i < 8, i ^ 7). After removing these, there are no prime divisors on (£2, • • •, ^s) = (0,..., 0). This is easy to see for the former type of divisors. For the latter, let k be the minimum such that the minus sign is chosen in the left side of the second equation. If k < 5, then —12 - - • tk ± t2 • • • tk+i ± • • • = t2 • • • tk +1 2 - - • £fc+i H holds. Both sides can be divided by the first term (since we already removed U = 0). Thus we get —2 = tk+i x f(t), where f(t) is a polynomial in £fc+2, • - -,ts with constant term 2 or 0. These do not lie at the origin. If k = 6, then automatically the minus sign is chosen for t2 • • • t7. Then, the corresponding divisor after divided by t2 ••• t6 is -l-t7 = 1 +t7 + 2(1 +t7)t8, i.e., (l+* 7 )(l + *8) = 0. If k does not exist, then all signs are plus, and we have (1 + t7)t$ = 0. Since t{ — 0 (2 < i < 8) are normal crossing, we defined a tangential morphism SpecQ -r U(E7) by (U(E7), V*,Y,t 2 ,...,t 8 ,SpecQ) with Y the complement in F* of D - {U = 0|2 < i < 8, i^ 7}. Note that t7 = 0 is not a removed divisor, and thus we need some modification for the definition of tangential morphism. However, this is easy: just do not take t? , but simply t7 in the definition of the associated functor
Galois group GQ, Singularity £7, and Moduli M3
205
between the etale site in 2.1. Then it is easy to see that (£1,..., t6, ct7, £g) for any c G Q gives the same tangential morphism. We denote this tangential point by 0 , as origin. Now we have seven tangential morphisms fi{ (2 < i < 8), which correspond to (£1,..., ti-i, £i+i,..., tg). It is not difficult to show that this gives a tangential morphism from A1 — {0, ±1} to U(E7) for //», 1 < i < 6, from A1 - {±1} to U(E7) for /i 7 , and from A1 - {-1,0} to U(E7) for /x8. This can be seen as follows. For each /x^, tj = 0 holds for every 2 < jf < 8 except one j , namely, j = i. Prime divisors which can intersect with such one dimensional locus are: of the first type, 1 ± t3; = 0 (3 < j < 7) and of the second type, 2 < k < 5, — 2 = £fc+i/(£), where /(£) is a polynomial in tk+2, • - • ^s with constant term 0 or 2, or, (t7 + l)(£s +1) w e s a w Jus^ above. For the former, surely 1 ± tj is removed in the codomain of 11j . For the second, if the constant term of f(t) is zero, then this does not intersect with the locus t2 = • • • = i/-i = t/+i = • • • = t$ = 0, and if the constant term is two, then this intersects at tj = — 1. Thus, in any way, fij defines a tangential morphism with codomain described above. Note also that, in the above investigation, we noticed that ^- — 1 = 0 and U = 0 (i ^ j) are all the divisors on (0,0,..., 0,1,0,.. .0). (The troublesome divisors are —2 = tk+\ x /(£), where f(t) is a polynomial in £fc+2>... ,^8 with constant term 2 or 0. These may pass through (0,0,.. .,0, —1,0,.. .,0), but not through (0,0,.. .,0,1,0,.. .,0).) Thus, we can define a tangential point (£2, • • •, tj-u tj - 1, fy+i,..., ts) on U(E7). For t7, all the divisors on (0,0,..., 0, -1,0) are t7 + 1 = 0, U = 0 (1 < i < 8, i ^ 7). Again a trouble may lie in —2 = £fc+i x /(£), but this divisor exists for k < 5. Thus we have a tangential point (t 2 ,..., te,t7 + 1, £s)3.3. Pj l o o and P j ± l o o . We consider a double cover 4 A1 - { 0 , 1 } given by t »->• t2. Then, TTJ^A 1 — {0, ±1}, 01) has three generators as shown below.
It is easy to see that u = x2, v = y, w = x lyx in the inclusion ^ {0,±l},01) -> Trf^A1 - {0, l},01). Let p denote the path from 01 to 10
206
Makoto Matsumoto
on the (0,1) interval of A1 — {0, ±1}. An element a G GQ acts on this path by p ^ pgaiu, v, w). Here, <7a(u, v, w) is a proword. In TTI(A1 — {0,1}), we have ga(u,v1w) = fff{x,y). Thus, ^ ( f 2 , ^ " 1 ^ ) = fA^v) f o r anY t w 0 elements £, r/ in any profinite group. We also define elements of groupoids in the figure below.
-1
P
'
P
0
/
We use the following notation. For an odd number x, we denote by rx the following path. 0 o
Here, the point goes around (x~ l)/2 times counterclockwise. By passing to the profinite completion, we define rx for % £ Z x . 3.4. Galois action on each generator. 3.4-1- Ti (i = 2,3,4,5). Let i be a number 2 < i < 5. Then, roughly, r; is realized by moving ^ + i in the figure below, i.e., rp in the previous section. (See 3.2.1 for a global picture. If U+i moves as below, then £i+i,... ,^7 move proportionally, and the ordering of the points xi and Xi+\ is reversed and the other orderings are stable.) 0 o
>-
Let /ii+i,t be the tangential morphism given by (£2?^3> • • • >£t?U+2, • • • ?^s) from A 1 — {0, ± 1 } to U(E7). Then, /ii + i^(rp) is a topological path on U(E-j), with tj = e for j / i+1, and ^ + i moves along rp. Since e is positive small real number, £7, # 6 , . . . ,x;_|_2 moves around the small positive real number and don't change their order. Only Xi+i moves to the right to X{, and other Xj (j < i) does not move. x% moves but does not go beyond — (#2 H 1- XT)/2. This gives Tj. The tangential morphism J / * + M m a P s t n e starting (tangential) point (£i+i) of rp on A 1 — {0, ± 1 } to the tangential point (£2, H, t±, £ 5 , . . . , t%) on U(Et), i.e., the origin 0 . It maps the terminal (tangential) point (U+i — 1) tO (£2, £3, . . . , ti, ^_|_i — 1, ti+2, • • • , ^8)-
Galois group GQ, Singularity E7, and Moduli M3
207
If we contract the fundamental Weyl chamber (since it is simply connected), then the image of this path is surely T;+I after U(E7) divided by the Weyl group. However, to obtain the correct Galois action, we need to connect the end (tangential) point of this path to the tangential base point TiO. (Remember that Ti is the element of the Weyl group corresponding tO Ti.)
Ti exchanges Xi and £»+!. Let us assume 2 < i < 4 for a while. Then the coordinates giving TiO is s 2 , ^ 3 ,..., s 8 , with s2 = t2, S3 = £3,..., Si = = £i+i*t+2jSi+3 = U+3,--<>S8 = ^8- (This can tiU+i,Si+i = l/ti+i,si+2 be calculated by exchanging Xi+i and Xi in the defining identities of £'s. For i = 5, sg 7^ t$, and we shall consider this case soon later.) We denote by £*i+i,s the tangential morphism from A1 — {0,±l} to U(E?) given by (52, S3,..., Si, Si_|_27 • • • 1 ss)- The end tangential point of /x^+i^(rp) is given by (t2, ^3, • • •, U, *i+i - 1, ti+2, • • •, is). Since Si - ^(1 + (ti+1 - 1)), 1 - s i + i - {ti+1 - 1)/(1 + {ti+1 - 1)), s i + 2 = ^ + 2 (1 H- (£»+! - 1)), 5 i + 3 = ^ + 3 , . . . , 5 8 = t8, this tangential point is identifiable with (s 2 , s 3 ,. • •, Si, 1 — Si+i, Si+2> • • • ? $8)- Now we move 5i + i from 1 — e to e, then we return to the tangential point given by (52,53,..., s 8 ). Thus, the path r» from the tangential point (ti) to (si) is obtained by the composition of the following three paths: U+\: e to 1 — e (on the interval (0,1)), 1 —etol + e (rotate counterclockwise), Si+i:l — e to e (on the interval (0,1)). Here, the end point of the first path is identified with the starting point of the last path via infinitesimal homotopy. Then, n = ^+ M (p~ 1 )/i;+i,t(r)/^+ M (p). Thus, the image of this element by a G GQ is /ii+i,a(&r(ifc, v, w)~1p~1)(jLi+i^t(rx{a))iii+1^t(pg(7(u,
v, w)).
We define & by & := fj,i+ij(x). This path connects the fundamental Weyl chamber to another chamber, and can be considered as an element of 7n(F(£ 7 )), where V(E7) = U(Er)/W(E7). One can show that
A difficulty in the proof is in the jump along Weyl chambers. The element £i in the Artin group yields an element of the Weyl group, denoted by JQ, which is given explicitly by (xi,...,
x$) H^ (x 2 ,..., Xi, - x i + i , -Xi + 2 , •. •, - x 6 , ±x 7 , x$)
208
Makoto Matsumoto
Here, ± of x-j is chosen so that the number of minus is even. The tangential point XiO is given by (*1? t2, ...,*», -ti+i,ti+2, • • •, ±£7, *s)- H e r e *7 = 0 does not ramify, so ± does not affect. To show the last identity i^rf^i — Hi+ij(w), for example, we have to show that the move of £;+i in the next figure -1
0 0
coincide with rf, after identifying by X» the Weyl chamber containing XiO with that containing O. But, explicitly Xi exchanges v and the above path, and hence both give rf. Since rf (
Now, &+1 commutes with both & and r?. Since fa(x,y) is in the commutator, we have fa{Z>i,T?) = f<j{r)i,Ti)i where m = f i f ^ . This path r/f is given by moving £j+i along x at the same time moving ti+2 along x1. Thus only a:i+i moves and other Xj are fixed. We have a formula r]i = Ti+iTi+2 • • • T4T5(T6T7)T5T4 • • • Ti+2n+i. This can be proved by chasing the Weyl chambers. For i = 5, we need a small modification, since the s-coordinate is then 1 6 = t^" , 57 = t6t7, but = -{x2 4- • • • x 7 4- 2x 8 )/2(x 5 + aj7) = (x6 4- x7)t8/(x5 -f
Here we need to check the existence of an infinitesimal homotopy between (£2, • • •, h, t6 - 1, £7, ts) and (s 2 , • • •, «5» 1 ~ 56? «7» s s)- The only difference is sg, and 58 = t 8 (l + «7)(1 4- (*6 - 1))/(1 + (14- (*6 " 1))*7) shows the infinitesimal homotopy. The remaining arguments follow in the same way. Thus, we have
for 2 < i < 5, where % = Ti+iTi+2
• • • T4T5T6T7T5T4
• • • Ti+2Ti+i
Galois group GQ, Singularity 2?7, and Moduli M3
209
3.4-2. TQ. TQ is the image of rp by /x7>t, similarly to the above. T 6 exchanges XQ and X7 and maps ^-coordinates to s-coordinates by §2 = ^2?---? 5 5 — £ 5 ,s 6 = £ 6 £ 7 ,s 7 = l / £ 7 , s 8 = ^8- Now the situation is completely analogous to the previous section, and we have T6 =
~-i>
and 7yt(v),
/JL7,t(w))-
A difference from the previous section is that we don't have a good £»+i for z = 6. Thus we need a direct computation. We have u = 1 in 7Ti(A1 —{±1}). /JL7J(V) = TQ. However, fi7^{w) requires some calculation. This is the path in U(E7) shown in the figure below. -(x + ... +x - x - x )/2 V
2
5
6
77
-x,
-(x + x3+ ... + x?) 12 This is actually {T7T$T7)2. This can be checked by a straightforward chasing of the Weyl Chambers where this path passes over. It is left to the reader. Note that (T 7 T8T 7 ) 2 commute with both r 7 and rs, by the braid relations. Thus, we have
Now TQ commutes with both r 2 and we have
(T 7 T8T 7 ) 2 ,
hence ga cancels out. Thus
3.4-3. T7 and r 8 . r 7 is the most complicated element. If we move t7 along r'p1 (see 3.3), then x^'s move as follows. -(x + ... +x 5 - xft- x
-(x + x + ... + x ) /2 v
2
3
iJ
To make this r 7 , we have to move x$ to the left. This can be done as follows. Let s be the image of t by T 7 . T 7 maps XQ \-t —x7 and X7 \-> —XQ.
210
Makoto Matsumoto
Thus we have s2 = t2,...,s5
= £ 5 ,s 6 = -x7/x5
= -t6t7,
s7 = x6/x7
=
l/t7, s$ = - 1 / 2 • (x2 + • • • + x 5 - x6 - x7 + 2xg)/(-x 6 - xy) = - ( t 8 + 1). Now, r7 is realized by r7 = Ii%i8(<x)v>7,8'(p''~1)v>7it(r'pl).
Here, fj,7jS> is the tangential morphism coming from (52, S3,..., s&, ~( and a is the following path:
/o
^
o
-1
^
0
\
That is, x~1p~1r/~1 in 3.3. The right most H>7,t{r'p') moves x2, -. • ,xs as in the figure in 3.4.2, ^7iS'(pf~1) changes the ratio, then /xg)S(a) unwind xs to get r7. For these three paths, the end tangential point is identifiable, via an infinitesimal homotopy, with the start tangential point of the next path. The end of the first path is (t2,..., ^6, —(£7 + 1), t$), and the start of the second path is (s2,..., sg, 57 + 1, ~( 5 8 + !))• These are identifiable. The next point is (s2, ...,57, — (s$-\-1)), which is compatible with both fi7^si and M8,s-
Now the Galois action is described as follows:
We shall calculate these three terms from the right. Now
where the exchange of w and v comes from the relation between p and p'. We have 1*7,t(v) = T l M7,tM = M7,t(p'-VV) = (r 7 r 8 r 7 ) 2 . Then »7,tWr'p')) = M7,t(rVp'- 1 r / xW-V ff
The middle term is 1*7,8*
^ we have
Galois group GQ, Singularity E7, and Moduli M3
211
The former is proved using the commutativity of r 6 with T7,r8, and the latter is proved by drawing a picture. Then we have
The first term is obtained as follows. (T [Ou J — X
p
J (T \
iP
P
)
Then we have and Thus,
Gathering all together, we have
Thus we get
Since (^rgry) 2 , rg, r 7 are commutative, go cancels out, and we have the following result:
To simplify3 this complicated expression, we need 2-cycle and 3-cycle defining relations of Grothendieck-Teichmiiller group [9] [15] [16]. Let x, y, z be elements of a profinite group with xyz = 1. Then, (I)if<,(x,y)f
=l
In the previous version of this manuscript, the author stopped the calculation here, and the above expression was the final form. The following simplification was given by L. Schneps, who got her insight from the comparison with Nakamura's result [27].
212
Makoto Matsumoto
(II):fa(z,x)zmf(r(y,z)ymU(x,y)xm =1 x 1 hold, where m = ^~ g Z. Set temporarily only in this paragraph x = T£, y = T^XT7TS. Then z = {xy)~x = T^T^2T^X and x = r^1yr7 hold. Put 2 c = (T7T8T7) . This is in the center of the group generated by r 7 and T$, and zc = r | holds. Now, the image of T7 becomes x-mfa(y,x)y-mT7cm
The last step used fa(x,z) = fa(x,TjC~l) = fa(x^r^). This follows from that f
3.5. The result on Ej and M3. In summary, we have proved the following: Theorem 3.1. a £ Gq acts on the profinite Artin group n^9(V(E7), as follows: for 2 < i < 5,
r
7 ^ /r 7 )
r
7
/o- ( r 8 5 r 7 ) >
and
where r 2 , . . . , rg are £/ie following generators of E7 Artin group,
O)
Galois group GQ, Singularity £7, and Moduli M3
213
-o X
and Tji =
If we take the image of O in V(E?) —> Ais, then we get the same action for r 2 ,..., T$, where each generator is a Dehn Twist as follows.
3.6. An,Bn,Cn,Dn
cases.
3.6.1. An. This case is exactly same as the action on profinite braid groups given in Ihara-Matsumoto [17]. Let e i , . . . , e n + i be the unit vectors of the n + 1-dimensional vector space. We consider the n-dimensional subspace U := {(xi,X2,...,xn+i)\xi
+ x2-{
h^n+i = 0}.
The root system is ei — ej (1 < i / j < n + 1). The fundamental roots are e\ — e^ e2 — ^ 3 , . . . , en — en+1. The Weyl chamber C is the set of vectors that has positive inner product with any of the fundamental roots, i.e., C = {(xi,..., £ n +i)|xi > x2 > x3 > " - > £ n +i}. There are n walls of this Weyl chamber, which are given by x\ > • • • > Xi = Xi+i > Xi+2 > • • • > #n+i (i — 1,2,..., n), corresponding to the z-th fundamental root. We denote by U(An) the complement of the union of the reflection hyperplanes. The corresponding Artin group is generated by 7^, 1 < i < n. o
o
o
o
X
X
X
x,
O
o
X
x,
Makoto Matsumoto
214
These are actually the standard generators of the braid group, after divided by the Weyl group isomorphic to the symmetric group 5 n +i as the permutation of coordinates. The elements rji are necessary to write down the Galois action. The "suitable" parameters are U := (xi—xn+i)/(xi-i—xn+i) (2 < i < n), ti := (xi — £ n +i). We see easily that Xi — xn+\ = t\t2 •••£*. Then, we get the same construction as in 3.2.2: C/(A n )c
-A n t -y*
U(An) and get a Q-rational tangential point ( t i , . . . , tn) on U(An). The tangential morphisms given by (£i,..., £», ti+2, • • • ? ^n) a r e easily checked to be from PQ1OO to U(An). Similar arguments to those in 3.4.1 show that the Galois action is given by
n ->
fa T?) (1 < i < n),
with r]n — 1. ^.^.^. 5 n ? C n . These cases can be reduced to the An case. The Bn root system is {±e; ± ej|l < i / j < n} U {zte*|l < i < n}. The Weyl group is same as that of C n , and hence V(Bn) = V(Cn). The vector space spanned by the roots minus the reflection planes is: U = {(^i,...,x n )\xi ± Xj ^ 0,Xi / 0}. The Weyl group is the group of permutation matrices with components 0, ± 1 , and it is isomorphic to the semidirect product (Z/2) n x Sn. By putting Si := x\, we see easily that the quotient l/(z/2)n i s { ( 5 l , . . . , S n ) | 5 i / ^ ( 1 < i / j < n),Si / 0}. This is nothing but the configuration space Fn(A1 — 0) of ordered n points on the affine line minus origin. Thus, by dividing by S n , we have V(Bn) = V(Cn) = configuration space of unordered n points on A1 — 0. The fundamental group of this space is the subgroup of the n + 1-strands braid group, which stabilizes one point after mapping to S n + i . It is notable that these are the groups used in [20] to characterize Grothendieck-Teichmuller group. The generators of the corresponding Art in group are r » ( l < i < n — 1) and rjn-i as in the following figure.
Galois group GQ, Singularity £7, and Moduli M3 o
o S
o n
h-1
••• ^ j
n-1
CL
jo
s
s
"O
O
i+1
i
...
215
o S,
...
O
l
If we take the parameters t\ — si, £2 = S2/S1, • • •> *n = s n /s n _i, then we have a Q-rational tangential point on U^l2^ — Uifa, and hence on y ( S n ) . A similar argument with An case shows that the generators are mapped by:
Vn-l H> r£<$
n -* Umrfr^Mmrf)
(1 < i < n -1).
3.0.3. D n . The root system is {±ei ± ej\l < i ^ j < n}. The vector space spanned by roots minus reflection planes is: U = {(xi,...,x n )\xi±Xj ^ 0}. The Weyl group is (Z/2) n ~ 1 x 5 n , which is the set of permutation matrices, with coefficients 0, ±1, where the number of —1 is even. It is easy to see that the quotient £/(z/2)'1 is {(^i,..., sn, t)\t2 = s\ • • • s n , Si ^ Sj}, by putting Si := Xi and t := x\ • • -xn. Then we divide this space by Sn. The result is {(ui,...,un,t)\t2 = un, xn -uxxn-x + u2xn-2 + (-l)nun has no multiple root}. This is a double cover of the configuration space of ordered n points, which ramifies only along the locus un = 0. Thus, we obtain n\(V(Dn) — {un = 0}) as the subgroup of index two of 7Ti(V(Bn)), which is the kernel of the composition Tt\(V{Bn)) -+ Z -> Z/2 (the left arrow maps ryn_i i-> 1, r» i-> 0 (1 < i < n — 1)). By van Kampen's theorem, /K\(V{Dn)) is its quotient by normal subgroup generated by the square of the path which goes around the divisor un = 0. Thus, T)n-i = 1 is the newly added relation, and 7Ti(V(Dn)) is generated by Ti,..., r n _i, r'n_l := ryn _irn _i7/n_i. It can be checked that r n _i and r'n_x commute, and the group is isomorphic to the Artin group of Dn type.
216
Makoto Matsumoto
x
«
t..
x
By the compatibility of the Galois action with an algebraic morphism, we have
ri ^ / „ ( * , T?)-1T?^UVi,
r?)
(1 < » < n - 2).
Note that r/» = r i + ir i + 2 • • • r n _ 2 r n _ir4_ 1 r n _2 • • • ri+2Tt+iAcknowledgment. The main idea of this research comes from the joint work with Y. Ihara, who constantly encourages me by giving invaluable comments. T. Oda gave me a motivation to study Galois actions on mapping class groups. As for the result related with the deformation of singularities, I am indebted to P. Slodowy, K. Saito, and J. Matsuzawa who taught me much when I was in RIMS. S. Morita, T. Kitano, A. Ishida and other topologists I met at the Riemann Surface workshop in Hokkaido University informed me of the topology of Dehn twists. A. Tamagawa helped me on the notion of tangential morphisms. H. Nakamura kindly informed me of his beautiful result on the Galois action on the mapping class groups. His stimulation enables me to get rid of FQ ±1OO in my result. L. Schneps kindly showed me the simplification of the expression in Theorem 3.1, using the two and three cycle relations of the Grothendieck-Teichmiiller group in a nice way. This is celebrating, because the original complicated expression is dramatically shortened to a familiar expression. The anonymous referee showed me a number of important improvements. I would also like to thank all my friends working on algebraic fundamental groups, in particular P. Lochak and L. Schneps, who inspire me constantly. Lastly, I would like to dedicate this research to the memory of categorist N. Yoneda, who motivated me to start mathematics when I was a student in a computer department. References [1] [2]
G.V. Belyi, On Galois extensions of a maximal cyclotomic field, Math USSR Izv. 14 (1980), 247-256. J. Birman, Braids, links, and mapping class groups, Ann. of Math. Studies 82, Princeton Univ. Press 1975.
Galois group C?Q, Singularity £7, and Moduli M3 [3] [4] [5] [6]
[7] [8] [9]
[10] [11] [12]
[13]
[14] [15] [16]
[17]
[18]
[19]
217
J. Birman, Mapping class groups of surfaces, in AMS Contemporary Math. 78 Braids (1988), 13-43. N. Bourbaki, Groupes et algebres de Lie, Ch. 2 et 3, Elements de Mathematique, 1972. E. Brieskorn, Singular elements of semisimple algebraic groups, in Actes Congres Intern. Math. (1970), t. 2, 279-284. E. Brieskorn, Die Fundamentalgruppe des Raumes der regularen Orbits einer endlichen komplexen Spiegelungsgruppe, Inventiones Math. 12 (1971), 57-61. P. Deligne, Le groupe fondamental de la droite projective moins trois points, in Galois groups overQ, Publ. MSRI 16 (1989), 79-298. P. Deligne and D. Mumford, The irreducibility of the space of curves of given genus, Publ. IHES 36 (1969), 75-109. V.G. Drinfel'd, On quasitriangular quasi-Hopf algebras and a group closely connected with Gal(Q/Q), Algebra i Analiz2 (1990), 114-148; English transl. Leningrad Math. J. 2 (1991), 829-860. K. Fujiwara, Theory of tubular neighborhood in etale topology, Duke Math. 80 (1995), 15-57. A. Grothendieck, Revetement Etales et Groupe Fondamental (SGA 1), Lecture Notes in Math. 224, Springer-Verlag, 1971. A. Grothendieck, Esquisse dyun programme, typescript (1984), Geometric Galois Actions, London Math. Soc. Lecture Note Series, Cambridge University Press, 1997. S. Humphries, Generators for the mapping class group, in Topology of Low-Dimensional Manifolds, Lecture Notes in Math. 722, SpringerVerlag 1979, 44-47. Y. Ihara, Profinite braid groups, Galois representations and Complex multiplications, Ann. Math. 123 (1986), 43-106. Y. Ihara, Braids, Galois groups, and some arithmetic functions, Proceedings of the ICM 90 (I), 1991, 99-120. Y. Ihara, On the embedding of Gal(Q/Q) into GT, in The Grothendieck Theory of Des sins d'Enfants, London Math. Soc. Lecture Note Series 200, Cambridge Univ. Press, 1994, pp. 289-305. Y. Ihara and M. Matsumoto, On Galois Actions on Profinite Completions of Braid Groups, in AMS Contemporary Math. 186 Recent Developments in the Inverse Galois Problem (1994), 173-200. Y. Ihara and H. Nakamura, On Deformation of Maximally Degenerate Stable Marked Curves and Oda's Problem, to appear in J. reine angew. Math. A. Ishida, The structure of subgroup of mapping class groups generated by two Dehn twists, preprint 1995 (in Japanese), preprint 1996
218
[20]
[21] [22] [23] [24] [25] [26]
[27]
[28]
[29]
[30] [31] [32] [33] [34]
Makoto Matsumoto (in English). P. Lochak and L. Schneps, The Grothendieck-Teichmiiller group and automorphisms of braid groups, in The Grothendieck Theory of Dessins d'Enfants, London Math. Soc. Lecture Note Series 200, Cambridge Univ. Press, 1994, pp. 323-358. E. Looijenga, Cohomology of Ms and M\, Contemporary Mathematics 150 (1993), 205-228. M. Matsumoto, Galois representations on profinite braid groups on curves, J. reine angew. Math. 474 (1996), 169-219. S. Mochizuki, The local pro-p Grothendieck conjecture for hyperbolic curves, RIMS preprint 1045 (1995). S. Mochizuki, The local pro-p anabelian geometry of curves, RIMS Preprint 1097 (1996), H. Nakamura, Galois rigidity of pure sphere braid groups and profinite calculus, J. Math. Sci. Univ. Tokyo 1 (1994), 71-136. H. Nakamura, Coupling of universal monodromy representations of Galois-Teichmuller modular groups, Math. Ann. 304 (1996), 99119. H. Nakamura, Galois representations in the profinite Teichmiiller modular groups, Geometric Galois Actions, London Math. Soc. Lecture Note Series, Cambridge University Press, 1997. T. Oda, Etale homotopy type of the moduli spaces of algebraic curves, preprint (1989), Geometric Galois Actions, London Math. Soc. Lecture Note Series, Cambridge University Press, 1997. T. Oda, The universal monodromy representations on the pro-nilpotent fundamental groups of algebraic curves, Mathematische Arbeitstagung (Neue Serie) 9-15 Juni 1993, Max-Planck-Institute preprint MPI/93-57. A. Shiho, Tangential map and Log scheme, in Hokkaido Univ. Tech. Rep. Ser. in Math. 47, Oct. 1996, 28-36 (in Japanese). P. Slodowy, Simple Singularities and Simple Algebraic Groups, Lecture Notes in Math. 815, Springer-Verlag 1980. T. A. Springer, Invariant Theory, Lecture Notes in Math. 585, Springer-Verlag 1977. A. Tamagawa, The Grothendieck conjecture for affine curves, RIMS preprint 1064 (1996), to appear in Comp. Math. B. Wajnryb, A simple presentation for the mapping class group of an orientable surface, Israel J. Math. 45 (1983), 157-174.
Department of Mathematics, Keio University 3-14-1 Hiyoshi Kohoku-ku 223 Japan e-mail: [email protected]
Monodromy of Iterated Integrals and Non-abelian Unipotent Periods Zdzislaw Wojtkowiak
Contents 0. 1. 2.
Introduction. Canonical connection with logarithmic singularities. The Gauss-Manin connection associated with the morphism Xd^W of cosimplicial schemes. 3. Homotopy relative tangential base points on P1(C) \ { a i , . . . , an_|_i}. 4. Generators of TTI(P 1(C) \ {ai,.. . ,a n + i},x). 5. Monodromy of iterated integrals on P1(C) \ { a i , . . . , a n +i}. 6. Calculations. 7. Configuration spaces. 8. The Drinfeld-Ihara Z/5-cycle relation. 9. Subgroups of the groups of automorphisms. 10. Torsors. 11. Torsors associated to non-abelian unipotent periods. 12. Torsors associated to the canonical unipotent connection with logarithmic singularities. 13. Partial information about GDR^1^) \ {0,1, oo}). 14. On the group GDR for pointed projective lines and for configuration spaces. A.I. Malcev completion. A.2. The torsor and the group corresponding to the Drinfeld -Ihara relation. §0. Introduction. 0.1. Let X be a smooth projective algebraic variety defined over a number field k. Let a : k ^ C be an embedding. We set Xc = X xC. Let X(C) be the set of C-points of Xc with its complex topology. There is a canonical isomorphism C - ^ HlR{X) ®C
220
Zdzislaw Wojtkowiak
between Betti (singular) cohomology and algebraic De Rham cohomology. The period matrix (pij) is defined by equations u^ = ^Pji&j, where {u^} and fa} are bases of H%R(X) and H%(X(C)). In this paper we study periods for fundamental groups. The main motivation of the paper is the construction considered in [D2]. We describe it briefly. First we recall that there is a canonical way of associating a class clB(Z) in H2Bp(X(Q)(p) := H%(X(Q) ®Q{(2iri)r) and a class clDR(Z) in H^R(X) with an algebraic cycle Z on X of codimension p. The class dirnZ CIB(Z) is a (2Txi) multiple of the Poincare dual of the homology class determined by Z. We also recall that T is a G-torsor if T is equipped with a free transitive action of G. Let S C T. We say that S is a subtorsor of T if there is a subgroup H of G such that 5 is an i7-torsor under the natural action of H. Let us assume that X is an abelian variety. Let v € G m (Q) act on Q(2TTZ) as multiplication by i/" 1 . Then the group GL(HQ(X(C))) X G m acts on HB(X(C)7n)(n) for any r, n and m because HB(X(C)m) can be expressed via H^(X(C)). Let G be the subgroup of GL(HlB(X(C))) x G m fixing all tensors of the form CIB(Z), Z an algebraic cycle on some X171. Let P be a functor from fc-algebras to sets such that an element of P{A) is an isomorphism p : HB(X(C)) 0 A —>• H^R(X) <& A mapping CIB{Z) 0 1 to CIDR(Z) 0 1 for all algebraic cycles Z on all Xn. The isomorphism Pcomp belongs to P(C). The functor P is represented by an algebraic variety over fc, denoted also by P. Let Gk '-= G x k be obtained from G by an extension of scalars. The group Gk(A) - the group of ^4-points of Gk - acts on P(A) by composition and P becomes a Gfc-torsor. It is a subtorsor of the GL(H^(X(C)) 0 A:)-torsor Iso{H^(X(Q) 0 k.H^X)). Let T be the smallest subtorsor defined over k of the torsor Iso(HB(X(C)) 0fc,iJ^ H (X)), which contains pComP as a C-point and let Q be the corresponding subgroup (defined over k) of GL(HB(X(C)) 0 k). Observe that T C P because both are subtorsors defined overfc,both contain pcomp as a C-point and T is the smallest one with these properties. Let Z(pComp) denote the k-Zariski closure of pComP in Iso(if^(X(C) 0 fc, H^R(X)) i.e. the smallest Zariski closed subset defined over fc, which contains pComp as a C-point. Then we have Z(pComp)
CTCP
and QcG.
In order to calculate the dimension of Z(pcomp) one needs to know the transcendence degree of the field generated by the periods p^ over k. In
Monodromy of iterated integrals
221
[D2] (Remark 1.8.) Deligne asks whether Z(pcomp) = P. He mentioned that Chudnowsky has shown it when X is an elliptic curve with complex multiplication. On the other hand to calculate T and Q seems to be an easier task. The requirement that T be a subtorsor of lso(H^(X(C)) ® k,H})R(X)) is very strong and usually only relatively weak information about the periods pij is necessary to calculate T and Q. We give an obvious example. If X = PQ is the projective line over Q and p c o m p : # 1 ( P Q ( C ) ) 0 C A H]yR(F^) ® C then in order to show that Z(pcomp) is isomorphic to PQ \ {0, oo} we must know that 2iri is transcendental. Observe that the group G m has no non-trivial algebraic subgroups other than the finite groups /i^ of roots of unity. Notice that for any rational number a / 0 the set of all roots of an equation xk — a = 0 is a //^-torsor and these are the only /x^-torsors. Hence already the fact that 2ni is not a fcth-root of a rational number for any k eN implies that T is isomorphic to P^ \ {0,oo} and g = Gm. In this paper we construct analogues of T and Q for fundamental groups. On the other hand we have no analogue of P and G. 0.2. Let us briefly discuss the content of the paper. Let us assume that X is a smooth quasi-projective geometrically connected scheme of finite type over a number field k. Let x be a fc-point of X. In [Wl] (see also [D4]) we defined affine connected pro-unipotent group schemes over k and Q respectively: irfR(X,x) - the algebraic De Rham fundamental group and 7rf (X(C),x) - the Betti fundamental group. We have also the inclusion (of Q-points into C-points) $ x : Trf (X(Q,x)(Q)
-> 7rfR(X,
such that the induced homomorphism on C-points
-> 7rP R (X,
is an isomorphism (see [Wl] Theorem 7.4.). The affine pro-algebraic scheme over k Iso := Iso(7rf (X(Q,ar) x
k,ir?R(X,x))
is an Aut(?rPR(X, x))-torsor. Let Z(
Z(
222
Zdzislaw Wojtkowiak
The reader can compare this question with the question of Deligne in [D2] (Remark 1.8) and Conjecture 1 in [An] which says that any polynomial relation with cofficients in Q between periods for the n t h cohomology of a smooth projective algebraic variety defined over Q comes from algebraic cycles (see also [G]). We shall denote by GBR{X) the image of GDR(<£X) in Ont(7rfR(X,x)). Our aim is to calculate the group QDR(X) or at least to get some information about this group. The calculation of the homomorphismV(C). We denote by 9X : in(V(C),x) -> w(V) the monodromy homomorphism at the base point x. (This is the monodromy homomorphism of all iterated integrals on V.) Let us take the vector 0i as a tangential base point. Then we have: Theorem A. Let So, Si and S^ be the elements O/TTI(P 1 (C)\{0, 1, oo}, ot) shown in the following figure (loops around 0, 1 and oo respectively, with So - Si - Soo = 1^. Then the monodromy homomorphism
is given by
•<(X,Y)
= (-niX) • {alliX, Z))~x • (-2ni)Z • aJ?(X, Z) • mX where Z=-X-Y
and aJ?(X, Y) € (n{V), n(V)).
Monodromy of iterated integrals
223
The element ctoi(X, Y) £ TT(V) is the same one which appears in [D5] and in [Dr]. In section 8 we give a proof that the element a}fl(X,Y) satisfies the Drinfeld-Ihara Z/5-cycle relation. Our proof is analogous to the proof in [12] and it is similar to the proof of the Z/2 and Z/3 relations in [D5]. It is different from the proof in [Dr]. In section 7 we discuss the monodromy of iterated integrals on configuration spaces. The main result (stated somewhat imprecisely) is: the monodromy representation of iterated integrals on a configuration space of n points in C is determined up to conjugacy by the monodromy homomorphism \ {0,
n(V).
Let us now describe the main results of the second part of the paper. One shows that there is an affine pro-algebraic pro-unipotent group scheme II(V) over Q such that the group of C-points of H(V) is ir(V). Let us set
T(C):={(x,y,z)e{AV)))3\ 3a G C*, x = aX, y
z ~ aZ, x • y • z = 0}
and
f(X) = aX,
f(Z)~aZ}.
(Here w means conjugation by an element of (n(V),ir(V)), whereas ~ means conjugation by an element of n(V).) One shows that T(C) (resp. Aut*(n(V))(C) ) is a set (resp. a group) of C-points of an affine proalgebraic scheme (resp. group scheme) over, which we denote by T (resp. by Aut*(lI(Vr)). The group Ant* (TL(V)) is a subgroup of the group of automorphisms of II(V). Moreover T is an Aut*(lI(V))-torsor. Let 0_4 : 7Ti(Vr(C), ~ot) —> TT(V) be the monodromy homomorphism from Theorem A. Observe that the triple
We define T{6_±) to be the smallest subtorsor defined over Q of T such that the triple (0_^(SO), 0-^CSi), O^S^)) is a C-point of T{6,). Let ) be the corresponding group. We have:
224
Zdzislaw Wojtkowiak
Theorem B . The affine pro-algebraic group schemes GDR{^Q \ {0,1, oo}) and G(6_±) are isomorphic. We calculate a quotient of G(6—±) explicitly. Let us set n' := (^(V), TT(V)), TT" := (T^TT') and n2(V) := ir(V)/ir". TT2(V) is also a group of Cpoints of an affine pro-unipotent pro-algebraic group scheme n2(Vr) denned over Q. We can also regard n2(V) as a Lie algebra. The elements X, Y and {YX)Xi~1Yi-1
: = ( . . . (Y,X)X)..
.X)Y)..
.Y)
i = 1 , 2 , 3 . ..J
=
1,2,3,... form a linear topological basis of n2(V) considered as a Lie algebra. Let u be an element of this Lie algebra. If the polynomials ex P( E fi,jxiYJ) and 1 + Yl Fn^rnXnYrn are equal then we set
n=l,m=l
Let us set 3a e C*, rr = a l , y « a7, z ~ aZ, x • y • z = 0} and Aut*(n 2 (V))(C):= {/ G Aut(7T20O)| 3a € C*, f(X) = aX, f(Y)*aY,
f{Z)~aZ).
As before there are affine pro-algebraic schemes T2 and Aut*(Il2(V)), whose sets of C-points are describe above. Aut*(Il2(V)) is a group scheme and T2 is an Aut*(Il2(Vr))-torsor. Let 6 : 7Ti(V, "oi) —>• n2(V) be the composition of 6>_, with the quotient map n(V) -> ir2{V). The triple (0(SO), 0(Si), 0{Soo))e T2(C). We define T(6) to be the smallest subtorsor defined over Q of T2 such that the triple (6{S0), fl(Si), 0(Soo)) is a C-point of T(0). Let be the corresponding subgroup of Aut*(n2(V)). Then we have: Theorem C. The affine pro-algebraic group scheme G{6) is isomorphic to a quotient of £DR(PQ \ {0,1, oo}). Let C : G m -> Aut*(n 2 (F)) be given by C(t)(X) = tX, C(t)(Y) = tY. Let H be the image of C. The next theorem gives a description of the torsor T(0) and the group G(0). Theorem D. %) The torsor T(0)(C) is given by
laX, (aY,exp(f; £ 1
fcl
ij2fc
'
M /
'
Monodromy of iterated integrals
225
E E
k=3 k odd
,
- C ( 2 * ) = r 2fc • (27ri) 2fc ,
and e ( * ) = 0
if C(*)eQ,
V 6i, fc _i € C,
cfc
e(Jfc) = 1 */
nj JTie corresponding group G(6)(C) is given by
{/ 6 Aut*(7r2(F)) I f(X) = tX,f(Y) =
fc-odd * > l , i > l
with t EC*, /?!,*_! E m7 Let G 6e t/ie smallest closed affine pro-algebraic subgroup o/Aut* defined over Q, which contains G(0) and H. Then G(C) = with t G C , A , ^ ! G C}.
{tY, exp
Corollary E. The group G{6) contains the group H if and only if all numbers C(2fc + 1) for k = 1,2,... are irrational. Let C : G m -> Aut*(n(F)) be given by C(t)(*) = *X, C(*)(y) = *V. Let % be the image of C. Let G be the smallest subgroup of Aut*II(Vr) defined over Q which contains G{6^) and %. Let T be a subtorsor of T defined over Q, whose corresponding group is G and such that the triple
is a C-point of T. Let Z be the Q-Zariski closure of the triple
in T. Then we have Z c
C T.
226
Zdzisiaw Wojtkowiak
Question. Is it true that Z = T{9^) = T?
In section 15 we state some conjectures relating the group GDR(X) to the Galois representations on fundamental groups. Deligne in [D4] considers the motivic Galois group of the tannakian category of mixed motives generated by the Tate motive Q(l). The group ^ D R O ^ Q \ {0, l,oo}) from this paper should be a quotient of the group considered by Deligne. Perhaps they are even equal. In this paper there is missing a motivic interpretation of GBR(X). We are also not discussing the mixed Hodge structures on the fundamental groups, though it would be very natural to do this here. We hope to study these topics in a future paper. Acknowledgments. I would like to thank very much Professor Deligne, who once showed me the one-form considered in Section 1 in the case of C \ {0,1}. Thanks are due to Professor Y. Ihara for showing me his proof of the 5-cycle relation, which helped us to find analogous one for unipotent periods. I would like to express thanks to Professor Hubbuck for his invitation to Aberdeen, where Section 8 was written and where in May 1993 I gave seminar talks on 5-cycle relation for unipotent periods. I would like to thank very much Professor Y. Ihara for his invitation to Kyoto, and Professors Oda and Matsumoto and Tamagawa for useful discussions and comments during my seminar talks. Thanks are due to Professor L. Lewin, who once invited me to write a chapter in the book on polylogarithms and suggested also including results on monodromy of iterated integrals. This encouraged me very much to continue to work on this subject (see preprints [W2] and [W3] of which some parts are included in the present paper). Finally, I would like to thank Leila Schneps and Pierre Lochak for their help in the editing of the paper and their encouragements, and the referees for their comments, remarks and interest. This paper is a revised version of [W5].
Monodromy of Iterated Integrals
227
§1. Canonical connection with logarithmic singularities. In this section we introduce the canonical connection with logarithmic singularities. The results of this section are well-known and are due to Aomoto, Chen, Deligne, Hain and others. Our approach is slightly different from the one in [HZ]. We consider the one-form of the canonical unipotent connection as a connection form of a principal G-bundle, where G is isomorphic to (a quotient of) TT®R(X,X). We found this approach useful studying functional equations of iterated integrals (see [W4]). It is also well adapted to study the monodromy. We also use it in work in progress concerning a generalization of the Zagier conjecture on polylogarithms to arbitrary iterated integrals.
1.0. Connection on a trivial principal G-bundle. Let G be a Lie group. Let us consider the principal G-bundle 1.0.1.
p:VxG->V.
Let 8 be the canonical one form on G i.e. the left invariant one-form with values in the Lie algebra & of G such that 6e(X) — X for X G g. Let pr : V x G —>• G be a projection. Then the one-form pr*0 is a connection form on the principal G-bundle 1.0.1. Let w G ^ ( V ) ® g be a one form on V with values in g. Let a) be a one-form on V x G vanishing on vertical vectors and such that i) &(v,e)(Y) — wv(pr*Y), where (v,e) G V x G, e is the neutral element of ii) ti(Vyg)((Rg)*Y) — a^{9~1)^{vye)0ir)'> where Rg is the right action of g on V xG. It follows from [L] (Theoreme, p. 68) that pr*9 + uj is the connection form of a certain connection on the principal G-bundle 1.0.1. If a; is integrable then pr*9 + CJ is also integrable. Abusing the notation we shall also denote the one-form pr*Q -j- UJ by UJ and we shall say that a; is a connection form on the principal G- bundle 1.0.1. 1.1. Let X be a smooth projective scheme of finite type over a field k of characteristic zero. Let D be a divisor with normal crossings in X and let
V = X \ D. Let be the differential algebra of global sections of the algebraic De Rham complex on X with logarithmic singularities along D. It follows from [Dl] Corollaire 3.2.14 that each element of A*(V) is closed and the natural map A*{V) -> H^R(V) is injective.
228
Zdzislaw Wojtkowiak
We shall denote by A2(A1(V)) the exterior product of the vector space A (V) with itself and by A\V) A Al(V) the image of A 2 ^ 1 ^ ) ) in A2(V). Let H(V) := (A^V))* and R(V) := (A^V) A Al{V))* be dual vector spaces. The map A2(AX(F)) -> A 1 ^ ) A A 1 ^ ) induces a map R(V) -> A2(H(V)). Let Lie (H(V)) be the free Lie algebra over jfc on H(V). Observe that R(V) is contained in degree 2 terms of Lie (H(V)). Let (R(V)) be the Lie ideal generated by R(V). We set X
and L(V) := lim (Lie (V)/rnLie (V)) . n
Definition 1.1.1. The Lie algebra L(V) equipped with the multiplication given by the Baker-Campbell-HausdorfF formula is a group. We shall denote this group by TT(V). Observe that TT(V) is a set of fc-points of an affine pro-algebraic prounipotent group defined over k. The Lie algebra of ft(V) can be identified with L(V). Observe that H(V) can be regarded as included in L(V). Let T[H(V)] be the tensor algebra over k on H(V) and let (R(V)) be the ideal of T[H(V)] generated by R(V). Let Q(V) := TlH(v)}/(R(V)) be the quotient algebra and let Q(V) be its completion with respect to the augmentation ideal / := ker (Q(V) ->fc),i.e. Q(V) := Urn (Q(V)/lnY n
Definition 1.1.2. We denote by P(V) the group of invertible elements in Q(V), whose constant term equals 1. Observe that the vector space H(V) can be regarded as included in the Lie algebra of P(V). We identify the elements of L(V) with Lie elements (which can be of infinite length) in P(V). The exponential series defines an injective homomorphism exp : n(V) -> P(V). The inverse of exp is defined on the subgroup exp(?r(Vr)) of P(V) and it is given by the formula logz = (z -1) - y2(z -1)2 + y3(z - 1 ) 3 - y^z -1)4 + . . . . We define a one form wy on V with values in the Lie algebra L(V) (and in the Lie algebra of P(V)) in the following way:
Monodromy of Iterated Integrals
229
Definition 1.1.3. The form uy corresponds to the identity homomorphism id^i(^) under the natural isomorphism A1(V)®H{V)
= Al{V) ® (
(see [D4] and [H] for the definition of the form uy). Lemma 1.2. (see [D4], [H]) The one-form ujy is integrable. Proof. The facts that Lie(V) is a quotient of a free Lie algebra by the ideal (R(V)) and that each element of A1(Vr) is closed imply integrability. Let us assume that k is the field of complex numbers C. Then V^(C) is a complex variety with the standard complex topology. Abusing the notation we shall denote it by V. Let us consider the principal bundles (1) V x P ( F ) - > V a n d
(2) V x ir(V) -> V.
We equip them with the connection given by the one-form ujy. Let us fix a path 7 on V from x to ZQ. We define a germ of flat sections of the trivial bundle (1) (resp. (2)) near ZQ\ Ay(z;x,/y) G P{V) (resp. Ly(z;x,/y) G TT(V)) to be a function of z moving in a simply connected neighbourhood of ZQ with values in P(V) (resp. TT(V)). We require it to take the value 1 (resp. 0) at x along the flat section over the path 7. Definition 1.3. homomorphism
Let x G V and let a G 7Ti(V,X) be a loop. We define a
-> P(V)
(resp. 6xy :
(l);x,a)
(resp. 6xy{a) :=
by the formula
and we call it the monodromy homomorphism of the form ujy at the base point x. (The homomorphism 6xy we shall also denote by 0X.) Proposition 1.4. Let x\,X2 G V and let 7 be a smooth path from x\ to X2> The path 7 gives an identification of TTI(V,x\) with TTI(V,X2)> Via this identification the monodromy homomorphisms 0Xly and 0X2y of the form ujy are conjugate. Proof. This is a property of a connection on a principal fibre bundle.
^
The next result is a special case of a result of Chen (see [Ch2], Theorem 2.1.1).
230
Zdzislaw Wojtkowiak
Proposition 1.5. If A1^) —> HpR(V) is an isomorphism then the monodromy homomorphism 6xy ' fti(V,x) —> TT(V) induces an isomorphism of the Malcev C-completion of TTI(V,X) into TT(V). Remark. If the map A1^) -> H\)R{V) is an isomorphism then Hl(V,C) has a Hodge type (1,1)- Examples of such varieties are Zariski open subsets of simply connected smooth project ive varieties. Let Xi (for i = 1, 2) be smooth, projective schemes of finite type over k. Let Di be divisors with normal crossings in Xi (for i = 1,2). Let Vi = Xi\Di (for i = 1,2) and let / : X\ —>• X2 be a morphism such that f~1(D2) — D\. Then / induces /* : Al{V2) -> A^Vi). Let /„ : Jf(Vi) -> if(F 2 ) be the dual map. This map induces group homomorphisms /* : P(Vi) -> P(V2) and /* : 7r(Vi) -> Lemma 1.6. We have {id 0 U){wVl) = (/* ® id)(u>v2).
The lemma follows from the definition of ujyi as i Corollary 1.7. We have (z; x, 7)) = KV2 (f(z); f(x),
/,(L V l (z; x, 7)) = LV2 (f(z); f(x), Let a>i,..., un be a basis of A1^). Let X i , . . . , X n be the dual basis of H(V). Let C{{Xi,... ,X n }} be the algebra of formal power series in non-commuting variables X\,..., Xn. Let C{{Xi,..., Xn}}* be the multiplicative group of these power series whose constant term is 1. Then the algebra Q(V) is a quotient of C{{Xi,... ,X n }} by the ideal generated by R(V). The group P(V) is a quotient group of C{{XU..., X n }}*. If a ( X i , . . . , X n ) is a formal power series in non-commuting variables X i , . . . , Xn, we shall also denote by a ( X i , . . . , Xn) its image in P(V). Proposition 1.8. Let 7 be a path on X from x to z. We have
Monodromy of Iterated Integrals
231
(The summation is over all non-commutative monomials in variables Xn, the interated integrals are calculated along the path 7.) llj
X\,...,
±j\f yZ\ X^'Y) -
Proof. We equip the principal bundle V x C{{Xi,... ,X n }}* ->• V with n
the connection given by a one-form ^ <*>* ® -X"t- Using the definitions in 1.0 we calculate the horizontal subspaces and then the system of differential equations for horizontal sections. We get that horizontal sections are given by z^{zl
+
y{{-if f
w.
WiXi
• -Xi)
Hence the point i) follows. Observe that exp : K(V) —> P(V) identifies UJV G Al{V) 0 Lie (n(V)) with uv G Al(V) ® Lie (P(V)). Hence the point ii) follows. 0
§2. The Gauss-Manin connection associated with the morphism _^ XdA[i] o f cosimplicial schemes.
XA[i]
2.1. We review here some constructions from [Wl]. Let / be a unit interval and let X be an arc-connected and locally arc-connected topological space. Let p : X1 -> X x X be given by p(u) = (CJ(0),O;(1)). This is a fibration. Applying the connected component functor to each fibre we get a local system of sets p : P -» X x X, such that the fibre over (x, x) G X x X is TTI (X, x) and the restriction to X x {x} is a universal covering space of X. We want to do an analogous construction in algebraic geometry. Let us notice that the inclusion of simplicial sets
induces a morphism of cosimplicial spaces
whose geometric realization is p : X1 —> X x X. We use this observation to construct an analogue of the local system p : P -» X x X in algebraic geometry. Let V be a smooth quasi-projective scheme over a field k of characteristic zero. The inclusion of simplicial sets 9A[1] °->> A[l] induces a morphism of cosimplicial schemes
232
Zdzislaw Wojtkowiak
Assume that k is the field of complex numbers C. As in the case of a morphism of algebraic varieties X -» 5, the cohomology of the fibres Hl(pm ~ 1 (x, ?/);C) form a local system of complex vector spaces. Imitating the construction of Katz and Oda (see [KO]) we equipped the sheaves Hl(tRplQ*,vA[1]y,vdA[1],) of the relative De Rham cohomology with the integrable connection dc such that the horizontal sections form a local system described above. (Rpl is the component-wise derived functor of pi and t is the functor which associates a total complex to a bicomplex.) H°(pm ~l{x, x),Q) is the ring of polynomial functions on the Malcev rational completion of the fundamental group TTI(V, X). The representation of 7Ti(F,x) x TTI(V,X) on the vector space H°(p* ~1(x,x);Q) is induced by(bijections of TTI(V,X)) given by (p(a, /3)(g) = a-g-f}-1. Therefore the sheaf
H°(tRplCl*vA[1])/(vdA[1]))
equipped with the connection dc is an analogue of the local system P —> X x X in algebraic geometry. This suggests the following definitions of the Betti and De Rham fundamental groups. We set nf(V,x)~SpecHo(p*-1(x,x);Q). If V is a smooth quasi-projective scheme over a field k of characteristic zero and x is afe-pointof V then we set:
n?R(V,x) := Spec H°DR(p' 2.2. 2.2.1.
^fax)).
Let V be such as in 1.1. We assume additionally that The inclusion of complexes (C-%A1(V)-^A1{V)
A AX(V) -> 0) *->
induces an isomorphism H° := H^BaviC-^A^V^A^V)
A Al{V) -> 0)) « H°DR{pm ~l{x,x)).
(The bar construction is chosen to be compatible with the inclusion into the total complex of the De Rham complex of pm ~ 1 (x, x). The isomorphism means that to calculate HpR(pm ~ 1 (x, x)) we need only Al{V) and AX{V) A A\V).)
Monodromy of Iterated Integrals
233
We recall that H(V) = (A^V))*, T[H(V)} is the tensor algebra on H{V), R(V) := (A^V) A A^V))* is contained in A2H(V) and Q(V) := T H v i ( )}/(R(V))Let us set
where / is the augmentation ideal of Q(V). Recall that Q(V)
:=U n
and P(V) C Q(V). 2.2.2. Notice that the symmetric algebra on Q(V)* is the algebra of polynomial functions on Q(V). Let us fix a basis u)\,..., un of A1 (V). Let Xi,..., Xn be the dual basis of H(V). Then T[H(V)] = C{XU..., Xn} is the polynomial algebra on non-commuting variables Xi,...,Xn. Let (X^ • ... • Xik)* denotes the linear form on T[H(V)} which to / in T[H(V)] associates its coefficient at the monomial X^ • ... • Xik. Let T* be the vector space generated by all {Xh -...-Xik)* including 1*. Observe that Q{V)* is a subspace of T*. We equipped T* with a shuffle multiplication and a Hopf algebra structure given b y ( X i , •...• X i p ) * • ( X i p + 1 .....
XipJq)* =
£
( X i a W •...•
Xia(p+q)Y
a€Sh(p,q)
and (Xh •... • XihY
-+ J2(Xh
. . . . . Xity
0 (Xil+1 . . . . . Xiky.
The shuffle
multiplication and the Hopf algebra structure on T* induce the analogous structures on Q(V)*. H° is also equipped with a shuffle multiplication and a Hopf algebra structure which on the form level are given by the same formulas. Lemma 2.2.3. There is an isomorphism of Hopf algebras
given i y w ^ ® . . . ® ujik -> (X^ • . . . •
Xik)*.
Proof. We identify the linear form (X^ • ... • Xik)* with the vector (X^ ® ... 0 Xik )* =Ljil®...®uJik. Observe that H° is a subspace of C 0 A1 (V) 0 (A^V) 0 ^ O O ) 0 (A1{V)®A1{V) O-A 1 ^)) © ••• = r*. Both algebras H° and Q(V)* have a natural grading. The part of H° in degree 2 is equal to ker(A 1 (^) 0 AX(V) -> A1^) A A x (y)). This is also the part of Q(F)* in degree 2. One checks that H° and Q(F)*are isomorphic in each degree.
234
Zdzislaw Wojtkowiak
Let Sym(T*) be the symmetric algebra on T*. Let Sh(V) be the ideal in Sym(T*) generated by the shuffle product relations
(xiY -... • xipY - (xip+1 •... • xip+q)* =
2 ^ ( \ ( 1 ) • • • • • xia{p+q))*.
Then Sym,(T*) divided by Sh(V) is isomorphic to T* equipped with shuffle multiplication. 2.2.4. This implies that the symmetric algebra on Q(V)* divided by the ideal Sh(V) flQ(V)* is isomorphic to Q(F)*equipped with the shuffle multiplication. Let Alg (TT(V)) be the algebra of polynomial functions on n(V). It follows from a theorem of R. Ree (see [R] Theorem 2.5) that a formal power series / e P(V) is in the image of exp : ir(V) -> P(V) C Q(V) if and only if its coefficients (i.e. the coefficients of some lifting to C{{Xi,..., Xn}}) satisfy shuffle product relations. Therefore 2.2.2 and 2.2.4 imply that exp : TT(V) -> P(V) induces an isomorphism of Hopf algebras 2.2.5
R : Q(F)*-^ Alg(n(V)).
Combining 2.2.1, Lemma 2.2.3 and 2.2.5 we get: Lemma 2.3. isomorphic.
The Hopf algebras H^R{pm~1{x,x))
and Alg(7r(V)) are
Corollary 2.4. The group schemes TT^R(V^X) and Spec (Alg (7r(V))) are isomorphic. We denote by u : 7rf>R(V,x) —> Spec (Alg (TT(V))) the isomorphism induced by isomorphisms 2.2.1, 2.2.3 and 2.2.5. 2.5.
Let k be the field of complex numbers C. We shall assume that
In section 1 we considered the principal 7r(F)-bundle equipped with the integrable connection given by toy We shall relate it to the objects described in 2.1. The group ir(V) acts on Alg(7r(V)) on the left by the formula a) 9(f)(x):=f(x.g)
b) g(f)(x) := f(g-lx)
We form the associated vector bundle V x Alg (n(V)) ztVx
n(V) x Alg (n(V)) -+ V 7T(V)
Monodromy of Iterated Integrals
235
and we equip it with the connection induced by uy. We shall denote this connection by u'v in the case a) and by fujy m the case b). Let tyVA[i])/(vd±[i]) k e the algebraic De Rham complex of smooth relative differentials on VAM (i.e. on each VAW»). Let H% (respectively x ^ ° ) be the restriction of the vector bundle to X x {x} (resp. {x} x X.) Both vector bundles H°(tRplQ*,vA[l])/,vdA[1])) are equipped with the induced connection, which we also denote by dcOur main result in this section is the following theorem. Theorem 2.6. Let V be as in 2.1. Then the algebraic vector bundles equipped with connections (H°x,dc)
and
{VxAlg(x(V))->V,ui'v)
are isomorphic in the category of algebraic vector bundles equipped with algebraic integrable connections. Remark. (xW°,dc) and (V x A\g(n(V)) —> V/uy)
are also isomorphic.
Proof. We shall calculate horizontal sections of the connection dc on % := #°(£i?p*£l? yA[1])/ , vaA[1] A We recall the construction of dc- Following Katz and Oda (see [KO]) the filtration {F^}i on £lyn+2 is defined by i W n + 2 := image(il^ +2 ® (pn)*fiVxV -> ^
+ 2
).
The filtrations {F^i for n = 0,1,2,... give a filtration Fi of QyA[1]. The connection dc is the connecting homomorphism of the functor H°(tRpl( )) for the short exact sequence of complexes on VrAt1l 0 _> Fl/F2 _+ F°/F2 _>. F°/Fi ^ o. On Vn+2 we have t)
Observe that Oy 0 fi^>n ® O V C Let us assume that 5^ u;^.. .®tc;im € ^(V)®771 is closed in the total complex of fi*(p* ~1(a:, x)). Its cohomology class c belongs to HpR(pm ~ 1 (x, x)), the fibre of W over (x, x) G V x V. We face two problems. We must extend c to a continuous section of % and we must show that this section is horizontal. Let T := Tot(n -> Ov ® Q*(Fn) O Oy) be the total complex of the bicomplex (p, g) -> Oy ®
236
Zdzislaw Wojtkowiak
top(Vq) ® Oy. The element u^ ® ... ® u^m can be interpreted as a global continuous section of T. One verifies that the class of the element
k+l<m
is a horizontal section of Uan. (Uan := U ®ov O^1). Hence we get: 2.7. Let (a,/?) € 7Ti(V X V,(x,a;)). Then the representation of TTI(V X V, (x,x)) on the vector space HjjR(pm ~1(x,x);Q) is given by (a, /3) : u)ix (8)... ® vim ->
0
^ , . . . , a;ifc )
k+l<m
By 2.2.5 the Hopf algebra Q(V)* is isomorphic to Alg (7r(y)). Hence we shall consider u>fv on the associated bundle V x Q(F)* —>• V. It follows from Proposition 1.9.i) that the monodromy representation is given by mi^x)
3 a : (Xt, •... • Xik)* -+ ^ ( X , , • . . . • Xit)
for the connection u'v, and by
for the connection 'uy. The isomorphism H°^>Q(V)* denned by wix ®.. . gives an isomorphism of monodromy representations of holomorphic vector bundles (H°x,dc)an and (V x Q(V)* -> F,o;^) an (resp. (xn°,dc)an and (V x Q(F)* -> V/a;v) an ) at the point xeV. (( ) an ) is an analytic object corresponding to an algebraic one ). Observe that both connections are regular. One can check this directly for the connection u'v. The connection dH(F,x)(C). Let a G TTI(V^X). There is a canonical map ni(V,x) —>
Monodromy of Iterated Integrals
237
TT^R(V,
x)(C) (the monodromy homomorphism) and it follows from 2.7 that the value of c on the image of a is equal to ^2 fa_1 u^,..., uJim. §3. Homotopy relative tangential base points on P 1 (C)\{ai, • • •, a n +i}. 3.1. In this section we define a tangential base point on the complex projective line minus a finite number of points. Definition 3.1.1. Let X = P1(C) \ {ax, ...,a n + i}. A tangential base point on X is a non-zero tangent vector v at one of the missing points a*. We shall denote by v (or by v) the tangential base point defined by the tangent vector v. If X = C\ {ai,..., an} we identify the tangent space at a^ with the vector space C and the tangent vector v at a* with the vector aix , where x = di+v. The complex projective line P 1 (C) has two canonical charts C —> P 1 (C), z -> [z : 1] and C —> PX(C), z —> [1 : z] which allow to identify tangent vectors to P 1 (C) with tangent vectors to C. Let A; be a subfield of C. We say that a tangential base point v at a^ is a tangential base fc-point if ai G k and a^ 4- v G k. Definition 3.1.2. Let X = PX(C) \ {au ...,a n +i} and let v be a tangential base point (at the missing point a*). A path on X starting from the tangential base point vis a smooth path in PX(C), satisfying the following conditions: i) the image of the open interval (0,1) is contained in X; ii) the tangent vector to the path at a^ is equal to v. We say that a path 7 ends at a tangential base point v if the path 7" 1 (7 -1 (£) := 7(1 — t)) starts from the tangential base point v. If a path 7 on P 1 (C) starts from v or ends at w or starts from v and ends at w and if 7((0,1)) C X then we say that 7 is a path on X. Definition 3.1.3. A homotopy between two paths 7 and 7' on X starting from a tangential base point v is a smooth homotopy H8 between the paths 7 and 7' on PX(C) satisfying the following conditions: i) # s ( ( 0 , l ) ) c X f o r s G [ 0 , l ] ; ii) the tangent vector of the path Hs at the starting point is v for s G [0,1], A homotopy between two paths 7 and 7' which end at a tangential base point v is a homotopy between the paths 7" 1 and j ' ~ l which start at v. According to this definition, the paths a, f3 on C \ {0} shown in the following figure are not homotopic.
238
Zdzislaw Wojtkowiak
w Definition 3.1.4. Let 7 be a path starting from a tangential base point v (a tangent vector at a^) and ending at the same tangential base point v. We say that 7 is null homotopic if there is a smooth homotopy Hs between 7 and the constant map into a^ such that: i) Hs is a path on X starting from v and ending at v for s < 1; ii) Hi is the constant map into a^. If 7 is a path on X we denote by [7] the homotopy class of 7. Suppose that a path (p on X ends at a tangential base point v and a path ip on X starts from v. We define a composition of homotopy classes [ip] • [0 be small. Let (fi-£ := y?|[o,i-e] and ^ e := ^|[e,i]- We can assume that <^(1 — e) = ip{e). We define the composition of paths ip • ? to be the composition ?/>e • ^ i _ e (with a necessary reparametrization and smoothing). While i\) • ? depends on £, its homotopy class is well defined and we set
n+1
Let X = X U |J (Ta. (P1(C))\{0}) be a sum of X and the tangent spaces i=l
minus zero at the missing points. We define a groupoid P over J x X i n the following way. If (x, y) G X x X then P^y is the set of homotopy classes of paths on X starting from x and ending at y. We set P = U,x y)exxx^^y We define a projection p : P -> X x X by p(Px,3/) = (#, 2/)- The partial composition of homotopy classes makes p : P —> X x X into a groupoid over X x X. The fibre over (x,x), TT^X^X) := p~1(x,x) is the fundamental group of X at a base point x G X. Observe that x can be a tangential base point v. 3.2. We shall construct functions Ax{z; v, 7) if v is a tangential base point. Let us set X = C\{ai,..., an}. Let XQ G C and let # : [0,1] 3 t —>> a; + t • (#0 ~~ a i) be an interval joining ai and XQ. Let 7 be a path from a^ to z £ X (not passing through any a^, fc = 1,.. .71) tangent to 5 in a*. We assume
Monodromy of Iterated Integrals
239
that in a small neighbourhood of a* the path 7 coincides with 5. We identify v — XQ - cii with a tangent vector to C in a^. Let u)\ — ^z^7? • • • > ^n — ztzan • We set
if ax / L J a,i
Let e £ im(S) be near a*. Let 7^ be a part of 7 from e to 2, and let S£ be a part of 6 from £ to :co- We set
-\, \
r
dt
f
:
A
= }™ I
dt
a i i tr(*,...,*)(*) w
and
:= / Jatn
Aa $(i,.. .,i,ak+i..
.,ai)(t)ujaa/+l
if a^-f 1 7^ z, and / > k + 1.
L e m m a 3.2.1. T/ie integrals Aaii#(aii /2/^'c, multivalued functions on X.
• • • ^fcX-z) exis^ and t/iey are ana-
Proof. Assume that at ^ i for t < I and ai+i = i. The function g(z) := f*. (jjai,..., uJai is analytic multivalued on X U {a^} and vanishes in a^. Hence the integral g\{z) := J^. ^(^)^z^: exists, the function g\(z) is analytic multivalued on X U {a;} and vanishes in a^ Hence by induction we get that A a i ,v(^i,..., an)(jz) exists and it is analytic, multivalued on X U {ai}. Assume now that at — i for t < I and a/+i 7^ i. Without loss of generality we can assume that a^ = 0 and XQ = 1. Observe that lim
240
Zdzislaw Wojtkowiak
where /% are rational numbers. The function zq(logz)p for q and p positive integers, is analytic multivalued on X, continuous on any small cone with a vertex in a; (0 in this case) and it vanishes in a^. The function j ^ - for j 7^ i is bounded on any sufficiently small neighbourhood of a*. Hence the integral lim im
f
is an analytic multivalued function on V, continuous and univalued on any small cone with a vertex in a* and it vanishes in a^. This also implies that the last integral exists. 0 We recall that X x , . . . , Xn are duals of 7^7, • • • > jz^-
Let us set
AxOz; ff,7) = 1 + $^(-l)*A o < | i r (ai,...,a*) (*)*<,, •... • Xai. Observe that Ax (2; #,7) depends on v, i.e. on XQ. Lemma 3.2.2. The map X 3 z -> (z,Ax(z;v,j))
e X x P(X)
is horizontal with respect to UJX • Proof. We have dA . a
Hence the functions (—l)fcAa<)j?(ai, • • • »«fc)(^) satisfy the system of differential equations defining horizontal sections of the principal bundle X x 0 P(X) —> X equipped with the connection LUXIt remains to define functions Ax(z; #,7) if a^ = 00 or all ai are different from 00. Let / : Y = C\{6i,... ,bn} -> X = P x (C)\{ai,... , a n + i } be a regular map of the form ^ ^ with det (
J ^ 0. Let y G Y and let
7 be a path on Y from 2/ to z. Then it follows from corollary 1.7 that ifyeY. We shall use this fact to /*(A y (z;y,7)) = Ax(f(zyj(y)J(j)) define Ax (z\ v, 7) where v is a tangent vector to P 1 (C) in a^ and 7 is a path from a,i to z, which is tangent to v. We set (3.2.2.1.)
A^(z;t;,7) := /•(Ay(/- 1 (z);/,- 1 (fl),/
Monodromy of Iterated Integrals
241
It is clear that Ax(z;v^) does not depend on the choice of / . Thus we have the following lemma. Lemma 3.2.3. Let X = P1(C) \ {ai,... ,a n +i} and let v be a tangential base point. Then the map
is horizontal with respect to UJX • We set Lx{z\ v, 7) := \ogAx{z\ v, 7). If we are dealing with only one space X we shall usually omit the subscript X and we shall write A(z;v, 7) and L(z;v,7), or A^(z;y) and L#(z;7), or even A${z) and L${z). The constructions from 3.1 and 3.2 are summarized in the following proposition. Proposition 3.2.4. Let v G X. The functions Ax{z\v,'y) depend only on the homotopy class ofj in Pvz .
and Lx(z;v^y)
§4. Generators of 7Ti(P 1 (C)\{ai,... ,a n + i},x). Let X = P 1 (C)\{ai,..., a n + i} and let x G X. Let vl be a tangent vector in a&. Then the loop around a& at the base point v*k is the following element 5^. of?ri(X, v*k) (see picture).
a i)
ii)
iii)
Now we shall describe how to choose generators of TTI (X, x). Let us choose tangent vector v{ G Tai (P^Q) \ {0} for each i = l , 2 , . . . , n + l . If x G X then we choose a family of paths F = {7i}"J"11 on X from x to each ^ such that any two paths do not intersect and no path self intersects. The indices are chosen in such a way that when we make a small circle around x in the opposite clockwise direction, starting from 71, we meet 72,73,..., 7n+iIf x = v{ then we choose a family of paths F = {7i}^J21 fr°m x t o e a c h Vi such that making a small circle around a\ in the opposite clockwise direction, starting from 72 we meet 73,... ,7 n +i. We associate the following element Si in TTI(X, X) to the path 7*. We move along 7i, we make a small loop in the opposite clockwise direction around
242
Zdzislaw Wojtkowiak
di and we return to x along 7" 1 . In the case ii) the element S\ is the loop
sv-x. iv) We can choose any order of missing points to construct the elements The following lemma is obvious. Lemma 4.1. The elements Si, S2, • • • > ^n+i are generators o/7Ti(X, X). We have 5 n +i •... • S2 * S\ — 1. Definition 4.2. The ordered sequence ( £ 1 , . . . , S^+i) of elements of TTI(X, X) obtained from the family of paths F as in i),...,iv) will be called a sequence of geometric generators of TTI(X, X) associated to F. §5. Monodromy of iterated integrals on P 1 (C)\{ai,..., a n +i} Let X = P 1 (C)\{ai,..., a n +i}. We want to compare functions Av(z) for different base points. Let £1, £2, #3 6 X and let ZQ € X. Let Si for i = 1,2,3 be a path on X from Xi to z$. Let us set 7^ := 5" 1 o J^. Proposition 5.1. Let us continue each function AXi(z) along Si to the point ZQ. There exist elements 0^.(7*^) G P(X) such that
for all z in a small neighbourhood of z0. The elements a%i.(/yij) satisfy the following relations
(Observe that the elements a^. (7^) depend on the choice of paths 7i?.) Proof. The existence of a** (7^) follows from the fact that the AXi(z)'s are horizontal sections. The fir^t two relations are obvious. The last relation follows from the equalities Ax.(z) • a** (7^) = AXj(z), AXj(z) • axi(
where Av(z) and Aw(z) are continued along 7.
0
Monodromy of Iterated Integrals
243
Proposition 5.2. Let v*k G TafcP1(C)\{0}. Let Sv-k be a loop around ak based at v*k G Tak (P1(C))\{0} (see figure at the beginning of section 4)- The monodromy of Av-k (z) along Sv-*k is given by
Proof. The monodromy of Av-k(km)(z) := Av~(fc, &,..., k)(z) along S^ is given by Sv-*k : A,* (*")(*) -> Av-k(km)(z) + £ K-h(km-l)(z)l=%±.
Let
7 be a path on X from v*k to z and let ai ^ fc. We must analytically continue the function Av-k (fcn, a i , . . . , ap)(z) along the composition 7 • 5 ^ . The contribution along Sv-k can be made arbitrarily small. However the function Av-k(km)(z) changes after the tour along Sv-k. This implies that the monodromy of Av-k (A:m, a i , . . . , ap)(z) along Sv-i is given by
(
Av^(fc
o
-\l
,
1=1
for a\ / k. The formula for the monodromy of Ay*k (z) along 5 ^ follows from this. 0 Let x G X. Let us choose v{ G T^P^QXJO} for i = 1,2,..., n + 1. Let ( 5 i , . . . , Sn+i) be a sequence of geometric generators of TTI(X, X) associated to F = {ji}^i (if x G X) or F = {7i}^ 2 1 (^ x *s a tangential base point, i.e. x = tJi), where F is a family of paths on X from x to t^ for i = 1,2,..., n + 1 or i = 2 , . . . , n + 1 as in section 4. Theorem 5.3. The monodromy of the function Ax(z) along the loop Sk is given by Sk : Ax{z) -»• Ax(z) • (al*^))'1
• e-2«iX* •
af{lk).
Proof. It follows from Proposition 5.1 that for z in the small neighbourhood of some 7fc(£). This equality is preserved after the monodromy transformation along S&, hence we have (*2)
{Ax(z))s» = {kv-k{z))s» •
af{lk),
244
Zdzislaw Wojtkowiak
where ( )Sk denotes the function ( ) after the monodromy transformation along Sy>k. Observe that (Av-k(z))Sk = (Av-k(z))Strk. It follows from Proposition 5.2 that (Av-k(z))s*=Av-h(z).e-2*iX*.
(*3)
Substituting (*3) in (*2) and then substituting (*i) for Av~k(z) yields the formula for (Ax(z))Sk. 0 Corollary 5.4. The monodromy of the function Lx(z) along the loop Sk is given by
Sk : Lx{z) -> Lx(z) • KHTfc))"1 • ( where af{^k) = \og{af{^k)). Proof. The corollary follows immediately from Proposition 1.8. ii).
0
Definition 5.5. Let x € X. Define a homomorphism ex,x : TTipT,*) -> P(X) (resp.^, x : 7n(X,x) -> n(X)) by the formula 1
*iX
af (7fc) (resp.
9XtX(Sk)
it is called the monodromy homomorphism of the form ux at the base point x. Observe that x can be a tangential base point. If x € X then the homomorphisms from Definitions 1.3 and 5.5 coincide. Let v,v' £ X and let 7 be a path on X from v to v'. If TTI(X, V) and 7Ti(X, v') are identified via the path 7, then the homomorphisms 6v,x and #v',x are conjugate. Proposition 5.6. Let
be a regular bijective map. Then for any v,w and any path S from v to w we have
G X , any path 7 from v to z
UAx(z;v,'Y))=AY(f(z);f(v)J(1)) and
Monodromy of Iterated Integrals
245
(notation: f(v) := f*(v) if v is a tangent vector). Proof. The proposition follows from the definitions of Kx{z\v^) for tangent vectors, and from Corollary 1.7.
and a^ 0
§6. Calculations. Let V = P 1 (C)\{0,1, oo}. The forms & and ^ form a basis of Ax{y). Let X := (&y and Y := ( ^ ) * be the dual basis of (A^V))*. Set Z := —X — Y. The group P(V) is the group of invertible power series in noncommuting variables X and Y with constant term equal to 1. Fix a path 71 = interval [0,1] from 01 to 10. It follows from Proposition 5.1 that along the path 71 we have
Let f(z) = l — z. It follows from Proposition 5.6 that
(We omit the arrow over vectors, when it does not cause confusion.) Proposition 5.1 implies that
Observe that /*(X) = Y and f*(Y) = X. Hence we get Deligne's formula (2)
al\{X,Y).al\{Y,X) = \.
(The proof of (2) given here essentially repeats that of Deligne (see[D5]).) Let us fix a path 7 ^ = interval [0, e] 4- arc from e to — e passing through (-i) • e + interval [-e, 00] from Ol to 00O where e > 0 (see figure). Yl
Let So (around 0), Si (around 1) and SQQ (around 00) be geometric generators of TTI(V, 01) associated to the family {71,700} ( s e e picture on page 5). Then we have So - Si - 5 ^ = 1.
246
Zdzislaw Wojtkowiak
Theorem 6.1. The monodromy of A^z) las:
is given by the following formu-
So : A ol (z) -> A ol (z) • e (3) 5i : A ol (z) -> Aol(*) • « ( X , r ) ) " 1 • e^2^Y • aJ?(X, F), Soo : A ol (z) -> A ol (z) • e~«iX • « ( X , Z))" 1 • e<-a*«>z • < ( X , Z) Proof. The formulas for the monodromy along So and Si follows from Theorem 5.3. The monodromy along S ^ needs some explanation. By Theorem 5.3 it is given by the formula 5oo : A ol (s) -> A o l ( z ) > ^ ° ( X , Y))- 1 -^- 2 ™^a%?(X,Y). By Proposition 5.1 a^°(X,Y) = a%£(X,Y) • a$°(X,Y). One calculates using 3.2.3 that a(Jf(X,F) = eniX. Let /(z) = ^ f j . We have = X and /*(y) = Z. It follows from Proposition 5.6 that oJ?(X, Z) = Hence we get the formula describing the monodromy along SQQ.
§
We recall that the Lie algebra L(V) is the completion of the free Lie algebra over C on two generators X and Y. The group n(V) is L(V) equipped with the multiplication given by the Baker-Campbell-Hausdorff formula. Let us set a(X,Y) := a$(X,Y) := log(oJ2(X,y)). It follows from Corollary 5.4 that the monodromy of L o l( z ) ^s gi v e n by the following formulas: (4) Si : L ol (z) -> L ol (z) • a(X, F ) " 1 • (-2?ri)y • c*(X, y), oo : LQ^(Z)
—>• ^ o i v / * v ^ v ^ - ' ^ v ^ ^ A)
' \—*'R%)Zi - o:(A, Z) • (7^^jA.
Observe that Theorem A (in the introduction) follows from the formulas (4). Let us compute the coefficients of a^(X,Y) and a(X,Y). If w is a monomial in X and Y we denote by a(w) the coefficient of agi(X, Y) at w. Let X be the first Lie basis element and let Y be the second element. We choose a basis of a free Lie algebra on X and Y as in [MKS] pages 324-325. If w is an element of this basis, let a(w) be the coefficient of w of a(X, Y). Observe that the coefficients a(Xn) and a(Yn) vanish. We assume that z approaches 1 in the formula (0). Then it follows from the formula (0), the formula (1) and the formula (2) that (5) a(XnY) = (-l)"C(n + 1),
a(YnX) = (-l) n + 1 C(n + 1)
Monodromy of Iterated Integrals
247
and a(XiYj)+a(YiXj) = 0. (If a; is a one-form then UJ1 := O;,CJ, ...,u> i-times.) Let us set ctij := a((y, X ) ^ * - 1 ^ - 1 ) , where t i ^ y * - ((... ((w, X)X).. )X)Y).. . ^ . C o m paring the coefficients at Y^X1 and X*y^' in a(X, Y) and in log(aJ;J(X, we get It follows from the equality a^Y^)+0,^X3) = 0 that aitj = ajyi. Observe { j j that we also have a{X Y ) + (-1)*+*a(Y X*) = 0. This last equality can also be obtained from formula (1.6.2) in [Chi]. Let us set nf := (ir(V),n(V)), *" := (TT',^) and TT2(V) := ^(V)/^ 7 . It follows from (3) that the monodromy homomorphism 6^ : TTI(V,01) —> is given by
(7)
SQ ->(Si -^(-27rz)y + [-2?rzy, a(X, Y)]
= (-2ni)Y+ £ (27ri)a i=O,j=O
Let us set F{z) =
(los(
^ z ) ) "'. The formula
(which can be proved by induction) implies that
(8)
a n + h m ={
L)
y>
71.TH.
6.2. Suppose that rti > 1 for i = 1,..., k — 1 and n^ > 2. Then we have oo
These numbers are values of the functions
)Sk
248
Zdzislaw Wojtkowiak
at z = 1 and ( s i , . . . , sk) = (n x ,..., nk). §7. Configuration spaces. and V = P ^ Q Y K , . . . , < + 1 } . If the Let V = V1(Q\{au...,an+1} sequences (a,x) := ( a i , . . . ,a n + i,x) and (a^x') := (a^,... ,a^ + 1 ,x / ) are close then the groups TTI(V, X) and TTI(V, a;') are canonically isomorphic. We shall study how the monodromy homomorphisms 9XiQr := 6xy &nd #x',a' : = 9x>y> from sections 1 and 5
depend on a and a'. If T is a topological space we set T? = {(tu . . . , £„) G T n | U ^ ^ if i ^ j } . Let Wn := C™. The space A1(TVn) of global holomorphic one-forms on Wn with logarithmic singularities is spanned by ujij — d ^~^ Z j for i,j e { 1 , 2 , . . . , n} and i < j . Let Xy = ( d ^ l f i ) * Xji = Xij and X« = 0. In A2(Wn) we have
b e their
formal duals. We set
Vki A CJ^ 4- cjjfc A LUki H- cJij A u ; ^ = 0
for z, j , A; different. These are the only relations between the 2-forms ujij (this result is due to V.I.Arnold). Dualizing the map A\A1(Wn))-+A1{Wn)AA1(Wn) we find that R(Wn) is generated by [Xij,Xik
4- Xjk] with i,j,k
different
and [Xij^Xki] with i , j , k,l different (see [C]). Let x = ( x i , . . . , x n , x n + i ) G W n +i be a base point. Let pi : W n + i -> Wn (i = 1 , . . . , n -h 1) be a projection p i ( z i , . . . , z n + 1 ) = ( z 1 ? . . . , zh . . . let
and let F(i, x) := p " 1 (x(i)) =
Monodromy of Iterated Integrals
249
(z means z is omitted). Let ki : V(i,x) —>> W n + i be given by ki(z) = (#1,..., Xi-i, z, Xi+i,..., x n +i). The inclusion ki induces
and
One calculates that (ki)*
/dzi-dzj\ \
Zi — Zj /
=
dz Z — Xj
and (ki)*
(dzi-dzj\ \ Z\ — Zj J
. , . , .
= 0 if I ^ % and j / i.
This implies that (ki)*(Xj) = Xij for j — 1,2,..., % - 1, i + 1,..., n + 1 where X/ is the formal dual of jz^~- It follows from the form of the generators of R(Wn) that the Lie subalgebra of Lie(Wn) generated by X^i,.. .Xij-i, Xiyi+i,... ,X i ? n + i is free and it is a Lie ideal of Lie(Wn). Hence the map (/?»)* is injective and its image, (ki)*(n(V(i, x))) is a normal subgroup of 7r(Wn+i). Let x = (a?i,...,a: n ,x n+ i) G Wn+i and x' = (xi,... , ^ , x ^ + 1 ) G W n +i. Let us set V := V(n + l,x) and V =: V(n + l,x'). We choose a family of non-intersecting paths 7 1 , . . . , 7 n , 7n+i in C from #1 to x[,..., xn to x^ and xn_|_i to x'nJtl. We shall identify 7Ti(V,a;n+i) and Ki(V,xfn+1) in the following way. Observe that 7 = (71,..., 7 n , 7 n +i) is a path in Wn+i from x to x'. The identification isomorphism 7. : 7ri(F,a;n+i) —> 7i"i(^',^n+i) i s the unique isomorphism making the following diagram commute ^
7Ti(W n+ i,x)
i 7#
is induced by the path 7 in a standard way). Proposition 7.1. After the identification of the fundamental groups of V = C\{xi,... ,xn} and V = C^x^,.. .,xfn} via 7, the monodromy homomorphisms 0Xu+ltv ' *i(V,xn+1)
-> ir(V) and 0x>n+1,v : ^i(V',x'n+1)
-+ n(V) = n(V)
are conjugated by an element of the group 7r(Wn+i). (The group (kn+i)*n(V) is a normal subgroup ofTt(Wn+i) so 7r(Wn-|_i) acts on K{V) by conjugation.)
250
Zdzislaw Wojtkowiak
Proof. The result is a consequence of the following commutative diagram: * ^ £
TT(V)
where c ^ (a;';a;,7) 1S a conjugation by the element Lwn+I(xf]x,j). It follows from Proposition 1.4 that the square (1) commutes. Corollary 1.7 implies that (2) and (3) commutes. The square (4) commutes by the construction. 0 Corollary 7.2. Let x = (xi,.. .,x n +i) € Wn+i. Se£ V(i) := V(i,x). Let dij be a geometric generator ofni(V(i),Xi), which is a loop around the point Xj. Let Aij be its image in 7Ti(Wn+i,a;). Then 0x,wn+1(Aij) is conjugate to (—27ri)Xij in the group 7r(Wn+i). Proof. It follows from Theorem 5.3 that ^Xi,y(i)(^ij) is conjugated to (—27ri)Xij in the group n(V(i)). Hence the corollary follows from Corollary 1.7. 0 Now we shall study the relation between the monodromy representations for the configuration spaces (P 1 (C)\{0, l,oo})? and ( P ^ C ^ O , l,oo})™. We shall use the Ihara result (see [II] The Injectivity Theorem (i)). Let us set Yn = (P^C))" and yn = (P^qXfO, l,oo})^" 3 . The group PGL2{C) acts diagonally on Yn and yn = Yn/PGL2(C). The identification is given by the map (0,1, oo, xux2,..., x n _ 3 ) -> (xi, x2,..., x n _ 3 ). Let ^ : W n _i -> 3^n be the composition of the map ( x i , . . . , Xfc_i, Xfc+i,..., x n ) —> (xi,.. .,Xfc_i,oo,Xfc+i,.. .,x n ) and the projection Yn —t yn. The map ipk induces (>fc)* : ff(W n _i) -> H(yn). Let us set X^- = {i/>k)*{xij) where X^ = ( ^.Z^ )* ^ ^(W^n-i)- (We use the same notation for Xij G if(W n _i) and its image in H(yn). Notice also that it can be shown by computation that Xij in H(yn) does not depend on the choice of ipk-)
Monodromy of Iterated Integrals
251
Let Aij € fl"i(Wn_i, x) be as in Corollary 7.2. We shall also denote the image of A^ in yn by A^. Corollary 7.3. The element 0yyn {Aij) is conjugate to {—2ni)Xij in 7r{yn). Proof. This follows from Corollary 7.2 and the commutative diagram ^x)
->
7T(W n _i)
0 Let Aut(7r(3^n)) be the group of algebraic automorphisms of a complex pro-algebraic group 7r(yn). Let Aut*(7r(3^n)) be a subgroup of Aut(7r(3^n)) denned in the following way: Aut*(7r(yn)) = {/ E Aut(7r(y n ))|3a/ € a (~ means "is conjugate to".) Let us set Tn{Q ={ipe Hom(7ri(;yn,y);7r(;yn))|3av € C* {A^ E 7Ti(yn,2/) are as in Corollary 7.3.). The group Aut(7r(Xi)) acts freely on Tn{C). We indicate arguments that the action is transitive. The homomorphism tp £ Tn{C) factors through the Malcev C-completion ofni{yn, y) and it becomes an isomorphismOut*(7r(J;n_i)) is injective for n > 5. Proof. Let Outl(7r(;yn)) := ker(Out*(7r(yn)) 4 C ) , where N{f) = af. The Lie algebra of Out*(7r(3^n)) is the Lie algebra of special derivations of L{yn) modulo inner derivations. The Lie version is proved in [II] section 4. The categories of complex unipotent algebraic groups and nilpotent Lie algebras of finite dimension over C are equivalent. The group is pro-unipotent. Hence the Lie version implies the result for and then also for Out*{7r{yn)).
252
Zdzislaw Wojtkowiak
The surjective homomorphisms (pn_j_i)* : /K\(ynjti^y) —> K\(yn,y') {Pn+i)* : ^(Xi+i) —t ^(iVn) induce a morphism of torsors
and
compatible with Out*(7r(3^n+1)) -> Out*(7r(yn)). Lemma 7.5. T/ie canonical morphism of torsors tn+1(C) —>• £n(C) is mjective for n > 4. This follows immediately from Proposition 7.4. Corollary 7.6. Tfte monodromy homomorphism 0yyn : 7Ti(^n,2/) —»Tr^n) zs determined (up to conjugacy by an element of 7r(yn)) by the homomorphism 9yiyA : 7riCy4,2/') -^ TTQ^). Proof. Observe that ^,3; n G t n (C) and ^',^4 is the image of 6yyn under the canonical morphism £n(C) -> £4(C). Let a := ( a i , . . . , a n , a n + i ) be a sequence of n + 1 different points in P 1 (C) and let Va := P 1 (C)\{ai,..., a n , a n + i } . The vector space H(Va) is spanned
by Xi := ( ^ - - i z t r r ) * * = 1,.. •, n. Let us set Xn+1 := - E? = 1 Xt. Let Afc denote a geometric generator of Va, which is a loop around a&. Let us set Ta(C) := {/ € Hom(7n(Vra,x);7r(Vra))|3a/ G C*,V^lfc
f(Ak)
Assume that a = (a^)^ 1 is such that ai = 0, a2 = 1, as = 00. The fibration Va —
>yn+2 —
>yn+l
(Va =
1
realizes ^(Va) as a normal subgroup of ^(^+2) ((fcn+2)*(Xj) = Hence the group 7r(3^n_|_2) acts on Ta(C); let U Q := T a (C)/7r(^ n+ 2). Observe that any 7r(J;n+2)-conjugate of Xi>n+2 is in the image of ir(Va). Hence the restriction map (kn+2)* : tn+2(C) -»• to(C) given by / —> /|7n(VOJa:) ^s defined. We set r a (C) := im
Monodromy of Iterated Integrals
253
Observe that the diagram (kn+2y
Ipr t4(C)
(fc 4 ).
——^
Tofifoo (C)
commutes where the map pr\ is induced by the inclusion Va ^ P 1 (C)\{0,1, oo}. The map (£4)* is bijective because y4 = P 1 (C)\{0, l,oo}. Lemma 7.5 implies that the map pr is injective. The map (fcn+2)* is surjective by the construction. Hence both maps, (kn+2)* and pr\ are injective. Therefore we have proved the following result. Proposition 7.7. i) The 7r(yn+2)-conJugacy class of the monodromy homomorphism 0Xiyn+2 : 7Ti(3^n+2,x) —>• 7r(3^n_}_2) is determined by its restriction to7n(Va,x'). ii) The fK{ynjt2)-conjugacy class of the monodromy homomorphism Oxya : determined by the monodromy homomorphism ,l,oo},x')
§8. The Drinfeld-Ihara ^/5-cycle relation. In this section we show that the element which describes the monodromy of all iterated integrals on P1(C) \ {0,1,00} satisfies the Drinfeld-Ihara relation. 8.1. Introduction We recall that Yn = (Fl(Q)"
and yn = (P^QUO, l,oo})"~ ) " ~ 3 . Let
a,b,c e PX(C) be three different points and let (paib,c(z) = Y^L ' f ^ c * Th map ^ 4 j 5 : Y5 —>yy5 given i bby The $4,5(£l,
induces a bijection
The group £5 acts on Y5 by permutations. The action of £5 on I5 induces an action of £5 on ^5- The map a : ^5 —>• ^5, (r(s, t) = f | ^ , i j corresponds to the permutation a of I5 given by
254
Zdzislaw Wojtkowiak
Observe that the points -y/S-1
-\/5
are fixed by a. The one-forms ^ , ^ 3 , f, f , j-^ j - ^ p ^ f generate A\y5) and ^3, Let 50,^1, To, Xi and AT be their formal duals. There are two non-trivial relations in A2(y5): ds 5
dt t
dt t
— A—+ —A
ds — dt ds — dt ds —+ — A —, s—t s—t s
and
ds dt dt ds — dt ds — dt ds 5—1 t — 1 t — 1 S— t S—t 5— 1 Elementary computations of linear algebra imply that the subspace of H(y^)®2 is generated by A / \
~
—
_|_ -\~
—
A / \
•
I T"
[S^Tj + ^ i V ]
i = 0,l;
[Ti,Si] + [TuN]
<= 0,l;
[5 0 ,Ti]
A / \
;
and [SUTO]
where [A,B] = A® B - B ® A. We recall that P(y$) is a multiplicative group of the algebra of formal power series in non-commuting variables 5o, Si, TQ, T\ and N divided by the ideal generated by i?(3^s). The principal fibre bundle y 5 x p(y6) -+ y5 we equipped with the integrable connection given by the one form dt rds-dt
\
s
-t
dt\ _ dt\ —) t) tJ
Ar
, ds ^
5
a
ds
t
5
-
I
where T^ = —To — T\ — N. More simply, we shall write u instead of 8.2. Integration of u
Monodromy of Iterated Integrals
255
Recall that on P 1 (C)\{0,1, 00} we have 8.2.0
A ^ ( z ) • a^(Y,Z)
= AlSo(z)
(see Proposition 5.1).
Please notice that we are considering a^(Y, Z) as a power series in Y and Z. Thus we will be able to substitute Y and Z in a<^)(Y1 Z) by other pairs of letters. We shall denote the generators X, Y and Z from the beginning of section 6 by To + N, T\ and T^. The monodromy of Ao^1(z) is given by: (around 00) : Ao^1(z) —> Ao^1(z) • e~27rlTo°, (around 1) : A ^ ( z ) -> A ^ ( z ) • a ^ ^ , ^ ) • e " 2 ^ 1 • (a^fT^Too))- 1 (see Theorem 5.3). It follows from the definition of horizontal sections starting from a tangential base point, and also from the identity dz z
dz z— 1
dz z
dz dz z— 1 z
°°
that asymptotically at 00 and 1, we have respectively
8.2.1
A -,(«) ~ e ^ i.e. lira (A -, (z) • e ~ ^ i T ) T ~ ) = 1 z—>oo V ° ° 1 V
y
/
and
i.e. lim Z
z>l
Let Pe = (e, 1 + e) 6 ^5 where e is a small positive real number. Let Aps((s,t); path) be a horizontal section of u; such that Ape(Pe) = 1. Let 7 be a path in y5 from P£ to cr(Pe) = (e, 1/e) which is constant (= e) on the first coordinate. Assuming 5 = constant (= e) we have
Hence for small positive e, we asymptotically have
loo
e = 0
256
Zdzislaw Wojtkowiak
It follows from 8.2.0, 8.2.1 and 8.2.2 and Proposition 5.1 that for small positive £, we have 8.2.3
P
.W«);7)
o
Let p = <J4(7) •
which follows from Corollary 1.7 implies that
where L = Ape(cr(Pe);7). Let
It follows from 8.2.3 that £=0
The factors e< XT ( A ~ f ))M=<^))
ande
(/>*)^
canbeput
together in the product 0\J(L) •... • L because T^ = cr^(Ti) commutes with cr*(Ti) = 5o and a*(Too) = Si. After the calculations we get
f1+£
dt
dt\
f1/£ dt
Repeating the same argument for Si, Si + Ti + iV, T\ and So and passing to the limit e —> 0 we get
where a = a ^ ( T i , Too). The last formula can be written in the form a(S1 + TX + N,So) • aiSuTj
• a(Too,S1 + T1 + N)- a(SQ,Si) • a(TuT^
Monodromy of Iterated Integrals
257
since a*(S0) = T ^ , a* (Si) = Si + Ti + W, a*(r 0 ) = N, ^ ( T i ) = S o and Let ^5 : Ct - • 3^5 be given by ^5(21,22,23,24) = ^4,5(21,22,23,24,00). r-^2-)
Let (A;A -• be formal duals of ( —£ J
J
\
z
i
z
j
ip5*(Ai2) = So
. Then we have
) ij
-Si-To-Ti-N,
ip5*(Au) = So, ^5*{^23) = So + TO + N, ^5.(^24) = Si, ^5.(^34) = - 5 0 - S i - N. Using Vi : C* ->• ^5 given by il>i(z2,z3,z4,z5) get
= $4t5(oo, z2,z3,z4,
z5) we
^1*^23) = 5 0 + T o + iV,
<M>l34) = - S o - Si - JV, ^.(^35) =
- ^ - ^ - ^ ,
^1*^45) = ^ .
Set X^ := tpet(Aij) s = 1,5; then X15 = To. Hence finally we get a formula 8.2.4
a(Xi3,Xi4)-a(X24,X25)-a(X35,X13)-a(Xu,X24)-a(X25,X35) :
= 1.
If we use $2,4 C^ —> 3^5 given by $2,4(0? s, 1, ^, 00) = (5, ^) and repeat the calculations in y5 we get the same formula as before, but the X^'s names of 5o, S i , . . . are now different and the resulting formula is: 8.2.5 a(X 15 , X12) • a(X 23 , X34) • a(X 45 , Xi 5 ) • a(X i 2 , X23) • a(X 3 This is exactly the formula which appears in [12, p. 106]. Proposition 8.3. For any permutation a of five letters we have
ii)
258
Zdzislaw Wojtkowiak
where a(ij) = o~(i)o~(j). Proof. This follows from 8.2.4, 8.2.5 and Corollary 1.7.
0
Remark. The formulas of Proposition 8.3 take place in the group If we apply log we obtain formulas in the group 7r(y5). In the sequel we shall work in the group TT^S)We finish this section with a formula from which the Deligne ^/3-cycle relation can be obtained. The proof is an imitation of Deligne's proof. Proposition 8.4. Let a := log a. In the group 7r(3^) we have i) a(X23, X25)(-niX23)a(X35, X23)(-7riX35)a(X25, X35)(-7riX25) = -7riX14c and ii) a(X23, X2b)(niX23)a(X36i X23)(niX35)a(X25, X35)(7riX2b) = -7riX 14 . Proof. We prove only i) since ii) is similar. Let 5(x\,x2,x3,X4i,xr)) = (#i, £5, £ 2 , £4, £3). Then the induced map a : y$ -* 3^5 is given by a(s, t) =
( ¥ • ^T. ¥ )
and
°2M = fe I^)-
Let P
~ = ^ 1 - ') and P+ =
(r, 1 + r) where r is positive and small. Let Q_ = (—r, 1 — r) and Q+ = (—r, 1-hr). Let 7 be a path from P+ = (r, 1-hr) to(r, 1 + re i(v?+7r)) and let S" be a path [0, TT] 3 y? —> (—r, 1 -h re l ^ +7r ^). Let us consider the composition p = a(jf)oa(Sf)oa2(j)oa2(S)o1foS'oa(-f)oa{S)oa2(j/)oa2(Sf)o1oS. If we integrate the form u along this path and pass to the limit as r —> 0 we obtain the square of the left hand side of the equality in i). Let a be a loop in the opposite clockwise direction around (0,0) in the plane P = {(5, t ) e C 2 \ as + fit = 0}. The integration of the form u along a gives (-27rz)(So + iV + To) = (-27ri)X23. In the model of Y*/PGL2(C) in which the subspace {(xi,x2,x3,x±,x§) \ x\ = X4} of (P1 (C))5 degenerates to a point (for example for $2,5(0, s, 1, 00, t) = (5, £)), the path p is homotopic to a loop around one of the points (0,0), (1,1) or (00,00) in the plane passing through the corresponding point (0,0), (1,1) or (00,00) (the point (1,1) in the case of the model $2,5)- Hence the square of the left hand side 0 of the expression i) is also (—2ni) • X14. Corollary 8.5. For any permutation a of five letters 1,2,3,4,5 we have formulas i1) and ii1), which are obtained from formulas i) and ii) by replacing indices 1,2,3,4,5 by a (I), o~(2), 0£<7(i))*=i,...,5. The
Monodromy of Iterated Integrals
259
induced map a : J^ —> y$ satisfies (a* ® id)u = (id ® <J*)U>, which implies formulas i') and ii'). 0 Remark 1. We have X23 + X2*> + X35 = X14 in the Lie algebra Lie(^5 ). If we set X14 = 0 then the formulas i) and ii) reduce to the Deligne formula (see [D5]). Remark 2. Let f(z) = jz^- Let us return to the notation from section 6. We have f+(X) = Y and /*(T) = Z. Hence we get
(We consider the element a ^ as a power series in Y and Z.) Therefore we have Let p = [0,1] be a path from 01 to 10. The element a^(X,Y) is the result of integration from Ol to 10. The element considered by Ihara is the result of the Galois action on the path p composed with the path p~l. Both elements satisfy the Z/2, Z/3 and Z/5 cycle relations. The analogy is more striking when we consider torsors corresponding to the homomorphism %i : ^ ( P ^ C ) \ {0,l,oo},0l) -> ^(P^C) \ {0,1, oo}) and the associated groups (see Appendix A2).
§9. Subgroups of the groups of automorphisms. 9.0. We summarize here the notation which will be used in the rest of the paper. Let A; be a field of characteristic zero. We say that X is an algebraic variety defined over k if X is an algebraic scheme over Specfc. If A is a fc-algebra, we set XA := X x Spec A. We denote by X(A) the set of A-points of Specfc
X. Observe that X(A) = XA{A). We say that G is an affine algebraic (resp. pro-algebraic) group defined over fc, if G is an affine algebraic (resp. pro-algebraic) group scheme over Spec k. Let L be a nilpotent (resp. pro-nilpotent) Lie algebra over a field k of characteristic zero. We equip the Lie algebra L with a multiplication given by the Baker-Campbell-Hausdorff formula. We denote by n the obtained group. The group n is the group of fc-points in a connected affine algebraic unipotent (resp. pro-algebraic pro-unipotent) group scheme over A;, which we denote by II. Let Rbe &fc-algebra.We denote by II(i?) the group of itpoints of II. Let us set UR := U x SpecR. Observe that U(R) = UR(R). Speck
260
Zdzislaw Wojtkowiak
Let V be as in section 1.1. We shall assume that ^41(V) —> HpR(V) is an isomorphism. In 1.1 we defined the group TT(V). It is easy to see that TT(V) is the group of fc-points of a connected affine pro-unipotent pro-algebraic group scheme over k. We denote by II(V) the corresponding group scheme. (II(y) is Spec of the algebra of k-valued functions on n(V).) Let V = PQ \ {0,1, oo}. We recall that Lie(Vr) is the free Lie algebra over Q on X = ( ^ ) * and Y = ( ^ j ) * and that Z := -X - Y. We recall that L(V) is the completion of Lie(Vr) with respect to the filtration induced by the lower central series. Let us set V := [L(V),L(V)], L" := [L'.L1] and L2(V) := L(V)/L"We denote by ^ ( F ) the group scheme over SpecQ corresponding to L2(V). Let us set n2(V) = U2(V)(C). Notation.
Let w G L2{V).
If /(X, Y) = 1 4-
^2
a
We set wXlY^ n 7n
nmX Y
: = ( . . . (w,X).
..X)Y)
...Y).
is a power series, then we set
n=l,m=l
(wJ(X,Y)):=w+ n = l , 77i=l
9.0.1. Observe that (wXiY^)XaYb = wXi+aY^b in L2(V). This implies that the elements X, Y and (Y,X)Xi~1Yj-1 for i = 1,2,3,..., j = 1, 2 , 3 , . . . form a linear topological basis of L2(V). Let F2(Lie(V)) be the Lie ideal of L(V) topologically generated by Lie brackets in X and F , which contain X at least twice. Let us set L2(V) := L(V)/F2(L(V)). We denote by II 2 the corresponding group scheme over SpecQ. To simplify notation let us set TT2 :— II2 (C). If we replace X by Y in the above definition then the corresponding group of C-points we denote by w2. The elements X, Y and (YX)Yi (resp. (YX)Xi) i = 0,1,2,... form a topological linear basis of n2 (resp. m2). We denote by G m the multiplicative group scheme Spec(Q[t, t" 1 ]). 9.1.
Let us set
Aut*(7r2(V)) := {/ G Aut(7r2(y))|3 af G C*,/(X) = afX, f(Y) » a / y, f(Z) - afZ} (« is a conjugation by an element of (TT 2 (F), TT2(V)) and ~ is a conjugation by an element of TT2(V)). Let p : n2(V) -> ZA72 be the natural projection. The map p induces p* : Aut*(7r2(V)) -> Aut(ti72). Let C : G m (C) ->• Aut(tu2) be given by C*(X) = ^X, Ct(Y) = tF. We shall investigate liftings of C to Aut*(7T2(V)).
Monodromy of Iterated Integrals
261
Proposition 9.1. %) Let $ : G m (C) -» Aut*(7r2(l0) be given by
$t(X) = tX oo
Y
J^
n=2
*
where
c
i,j € C.
t+i=n * > 1, j > 1
T/aen $ 25 a homomorphism. ii) All homomorphisms $ o/G m (C) m£o Aut*(7r2(V)) s^cft thatp*o$ = C and $t(X) = £X are of this form. Hi) All one-dimensional subgroups G of Aut*(^(V)) which are projected onto C(G m (C)) by the map p* and preserve the subgroup of 7r2(V) generated by X are of this form. iv) The group G is defined over a subfield k of C if and only if all coefficients Cij are in k. Proof. The point i) is a straightforward verification if one uses 9.0.1. To N-l
show point ii) we can assume that &t(Y) = (tY,exp( ^ n=2
+
£
fijtyXiY*)).
]T} ci,j(l ~ i+j=n
Comparing coefficients of $ t
and * , ( * t ( y ) ) we get fid(t) 4- tNfid(s) = f^s) + ^ / , , , - W . This implies fij(t) = Cij (I — tN). Hence the point ii) is proved. If a subgroup of Aut*(?r2(V)) is one dimensional then the coefficients fij are algebraic functions of t. They cannot be multivalued functions because then the dimension of the subgroup would be greater than 1. Hence fij are Laurent polynomials of t. Now the point iii) follows from the proof of ii). The last point is obvious. <) In §13 we shall also need results about the subgroups of Aut(7r 2 / r n+i 7r 2) and Aut(7r 2 ). Let C : G m ( C ) -> Aut(7r 2 / r27r 2) be given by C[(X) = tX,C[{Y) = tY. Let (p n )* : Aut(7r 2 / r n +27r2 ) -> Aut(7r 2 / r27r 2) and p* : Aut(7r2) —>• Aut(7r 2 / r27r 2) be induced by projections of 7r2/rn+2n2 and ?r2 Onto 7T 2 / r 2 7 r 2.
Corollary 9.2. All one dimensional subgroups G of Aut(7r2/rn+27r2) (resp. Aut(?r2)) which the map (pn)* (resp. p*) projects onto C'(G m (C)) are of the form G = {ft eAut(7T 2 / r n +27r2 ) (resp.Aut(7r 2 ))
= tX,
n(resp.oo)
ft(Y) = tY+
J2
°i(t - ti+1)((YX)Yi~1)
with
262
Zdzislaw Wojtkowiak
The group G is defined over a subfield kofCif
and only if all Ci 's are in k.
We shall denote the subgroup G of Aut(7T2/r»+27r2) (resp. Aut(?r 2 )) considered in the corollary by G(ci, C2,..., cn) (resp. G((ci)g 1 ) = G(ci, C2,..., cn,...)). Proposition 9.3. Suppose that for each n we have a point (c»j) i+j=n G Cn~1 and a sequence Z™,... , ZJ? of linear forms on Cn~1. Let Vn be a set of common zeros of Z™,..., ZJJ. Let us set L := {(Z™,..., ZJJ )n€N} o,nd c :=
i) Let G(c,L) := {/ G AnffaiV)) f(X) = tX, teC*, n=2
i+j=n
Then G(c,L) is a subgroup of Ant* ( Any subgroup G of Aut*(7T2(V)), which preserves the subgroup of 7T2(V) generated by X and whose projection into Aut(tu 2 ) is C(G m (C)) is of this form. Two subgroups G(c, L) and G(c', L1) coincide if and only if for each
ii)
Hi)
n ( * ; ) * £ £ +Vn = (dii)gi-h + VI The group G(c, L) is defined over the subfield k of Q if and only if for each n the affine space (cij) i+j=n 4- Vn is defined over k.
iv)
Proof. The point i) is a standard checking. Let G be as in the point ii). Let Gi := ker(p* : G —> Aut(o72)). Any element / € G\ is of the form f(X) = X, f(Y) = (y, exp( g n=2
£
f^X^)).
Observe that / ->
i+j=n t>i,i>i
N
((fij) i+j=n )n=2 3 N defines a homomorphism G\ -> J2 i
^
1
-^
1
' '
'
n=2
S
Ga(C) of
i+j=n
G\ into a direct sum of additive groups G a (C). Hence there are only linear relations between various / i / s . The group G is an extension of G m (C) by the pro-unipotent group Gi, hence there is a lifting C\ : G m (C) —> G of C. If we calculate the coefficients of C\t o f o Ci^1 we get that it can only be linear relations between the coefficients (/t,n-t)*=i,...,n-i- Hence for each n we have a finite number of linear forms Z",..., Z£ on C n - 1 such that
n—2
Monodromy of Iterated Integrals 0ij G C,V
V
n \
liWhn-l,p2,n-2,
263 • • •) = 0}.
Let L = {(/?,...,JJ n )nGi\r}- T n e group Ci(G m (C)) is as in Proposition 9.1.i) for some sequences (c™) z+j=n , n = 2 , 3 , , . . . Observe that G\ o i>
l,j>1
Ci(G m (C)) = G(c,L). Hence for dimensional reasons G — G(c,L). The points iii) and iv) are evident. <0 Let {ei)il1 be a sequence such that e» G {0,1} for each z. Let (c^)?^ be a sequence of complex numbers. For i = 1,2,...}, we set ^ i e ^ Z i ) = G ( C l , c 2 , . . . |e 1 ,e 2 , . . . ) : = { / = /t, ( e i A )£i € Aut(7r2)
= t •Y
with teC,fteC Replacing oo by n in the above definition gives a subgroup of Aut(7r2/rn+27r2), denoted by G ( c i , . . . , c n | e i , . . . , en). Let S((ei)f=1) be a number of €i equal to 1. Then dimG(ci,..., cn\eu ... ,en) = *((ei)? = i) + 1. Corollary 9.4. All subgroups of Aut(7r2/r«+27r2) (resp. Aut(7r2)), which the map (p n )* (resp. p*) projects onto C / (G m (C)) and which preserve the subgroup of ir2 generated by X are oftheform G ( c i , . . . , c n | e i , . . . ,e n ) (^res^. G((ci)g 1 |(ei)g : 1 )). The group is defined over a subfield k of C if and only if all numbers (1 — £i)ci are in k. We shall denote by Go the subgroup {ft € Aut(7r2/rn+27r2) ft(X) = tX, ft(Y) = tY,t e C*} of Aut(7r2/rn+27r2). map defined by 6(0) = 0,6(z) = 1 if z ± 0.
Let 6 : C -> {0,1} be a
Lemma 9.5. Le^ /o G Aut(7r2/rn+27r2) 6e siicft that fo(X) = aoX,fo(Y)
=
n
a0Y -f ^ ^ ( ( y X ) y * " 1 ) . Let G be the smallest closed algebraic subgroup i=l
of Aut(7r2/rn+27r2) such that Go C G and f0 G G. Then
Proof. Let Xt(^) = tX and xt(^) = tY and let fi = f0o (Xao)'1- Then = X and h(Y) = Y+J2 0SdQ1((YX)Yi-1). Let Gi C G be a subgroup consisting of h such that ft(X) = X, h(Y) = Y + fl
f3i((YX)Yi~l).
264
Zdzislaw Wojtkowiak
Assume that for some i, f3f ^ 0. Then G\ ^ {Id} because f\ e G\. The subgroup G\ is isomorphic to a subgroup of ( G a ( C ) ) n , hence it is given by a finite number of linear forms. Let k — xt ° h ° (Xt)" 1 - Then k(X)
= X, Jfe(y) = y + f2tif3i((YX)Yi~1).
Let Z(a?i,...,x n ) be one of
2=1
linear forms defining G i . Then l(t(31,t2p2,. • .,tnpn) = 0 implies / = 0. Hence the /?i, /?2? • • • ,j8n» if they are non-zero, they are linearly independent. Therefore G = G ( 0 , 0 , . . . , o|<J(/3j),..., <*(#1)). <> Corollary 9.6. Le£ G C Aut(7r 2/r"+27r2) be the smallest closed algebraic such that the subgroups Go and G ( c i , . . . , c n ) subgroup of Aut(n2/^+2^2) are contained in G. Then G =
G(0,...,0\5(c1),5(c2),...,6(cn)).
Proof. One takes any element / of G(ci,..., cn) such that f(X) = aX and
§10. Torsors. Let G be a group. We say that a set T is a G-torsor if T is equipped with a free transitive action of G. We say that a subset 5 C T is a subtorsor of T if there is a subgroup H C G such that the natural action of H on S is free and transitive. Main example. Let Gi and G2 be two groups. Assume that G\ and G2 are isomorphic. Then the set of isomorphisms from G\ to G2, which we denote by Iso(Gi,G2), is an Aut(G2)-torsor. For any non-empty subset S C Iso(Gi, G2), the intersection of all subtorsors of Iso(Gi,G2), which contain 5, is a subtorsor of Iso(Gi,G2), which we denote by T(S). 10.1. Unipotent affine algebraic groups and torsors. L e m m a 10.1.1. Let G be an affine unipotent algebraic group over a field k of characteristic zero. Then there is an affine algebraic group Aut(G) over k such that for any k-algebra A we have
Let G\ and G2 be two affine unipotent algebraic groups over k. Then there is a smooth affine algebraic variety Iso(Gi,G2) over k, such that for any k-algebra A we have
Monodromy of Iterated Integrals
265
Proof. Let g be the Lie algebra of G. Let us equip g with the group law given by the Baker-Campbell-Hausdorff formula. The exponential map exp: g -> G is an isomorphism of affine algebraic groups. The group automorphisms of g coincide with the automorphisms of the Lie algebra g. One can easily give an ideal defining AutLie(s) in k[GL(g)]. One constructs Iso(Gi,G2) in a similar way. 0 We say that an affine algebraic variety T over A: is a G-torsor, if there is a morphism T x G —> T over Spec k, which defines a free transitive action o f G o n T (see [S] page 149). Let 5 be a closed subvariety of T and let if be a closed subgroup of G. We say that S is an i7-subtorsor or a subtorsor of T if S is an if-torsor under the natural action of H. Lemma 10.1.2. Let T be a G-torsor. Let T\ be an Hi-subtorsor of T and let T2 be an H2-subtorsor of T. Assume that T\ D T2 / 0. Then the intersection T\ D T2 is an H\ D H2-subtorsor of T. 10.1.3. Main example. Let G\ and G2 be two unipotent affine algebraic groups over k. Assume that there is an isomorphism (Gi)^ —> (^2)^ of algebraic groups. Then the algebraic variety Iso(Gi,G2) is an Aut(G2)torsor, if we equip Iso(Gi, G2) with the obvious action of Aut(C?2). Let k C C be a subfield of the field of complex numbers C. Let O : Gi(C) -> G2(C) be an isomorphism. Then 9 is a C-point of Iso(Gx, G 2 ). We denote by Z(Q) the A;-Zariski closure of © in Iso(Gi,G2) i.e. the smallest algebraic subset of Iso(Gi,G 2 ) defined over &, which contains 0 as a Cpoint. The unipotent affine algebraic group Gi is isomorphic as an algebraic variety over k to the affine space A™, hence 6 can be viewed as a C-point (9?;j)i
0
Definition-Proposition 10.1.4. LetT(Q) be the intersection of all subtorsors T defined over k o/Iso(Gi,G2), which contain © as a C-point. Then T(Q) is a G(0)-subtorsor oflso(GuG2) for some G(0) C Aut(G 2 ). Proof. The intersection of a family of algebraic varieties coincides with an intersection of a finite number of them. Hence it follows from Lemma 10.1.2
266
Zdzislaw Wojtkowiak
that T(Q) exists and it is unique. The group is also unique because it is an intersection of the corresponding subgroups of Aut(G2). 0 Lemma 10.1.5. Let G be a unipotent affine algebraic group over k. Then Aut(G) is an extension of an algebraic subgroup o/Aut(G ab ) by a unipotent affine algebraic group. Hence the group G(©) is an extension of an algebraic subgroup of Aut(G ab) by a unipotent affine algebraic group. Proof. Let g be the Lie algebra of G and let (Pg); be the filtration of g by the lower central series. Any automorphism of the Lie algebra g preserves the filtration and the induced automorphism of I y / r i + 1 g is determined by the automorphism of g ab = F Bfr2g- Hence AutLie(#) is an extension of a closed subgroup of GL(g3h) by a unipotent group. The lemma follows from the identification of Aut(G) with AutLie(s) by the exponential map exp : g -^ G. 0 Lemma 10.1.6. Assume that G(0) is an extension ofGm (orG such that H1(Gsl(k/k),G)= 0) by a unipotent affine algebraic group N. Then T(Q) has a k-point. Proof. It follows from [S] Proposition 4.1 that ^(Gal^/fc),TV) = 0. It follows from [S] Proposition 2.2 and the assumption of the lemma that = 0. Prop. 1.1 of [S] implies that T(S)(k) / 0. 0 Let P(G) be a filtration of a group G by the lower central series. Let us set G (i) := G/ri+1G- The isomorphism 6 : Gi(C) -> G2(C) induces isomorphisms 9 ( i ) : G^(C) -> G^(C). Let k < i. The projections Gf -> G{k) for j = 1,2 induce pi:Iso(G^\G(^)-^lso(G[k\G{2k))
and p(2)l : Aut(G^) -> Aut(G^).
Lemma 10.1.7. We have
a) pUne^)) = r(e(fe)), Hi)
W
()
Proof. In the point i) ( ) means thefc-Zariskiclosure and we omit its proof because we do not need this fact later. Let us set p = p\ and p' = p(2)lk* Observe that the image pf(G(@^)) of the group G(@^) by the morphism p' is a closed subgroup of Aut(G;T ) defined over &. This implies that p(T(0W)) is a closed subvariety of lso(G[k\G(2k)) and a p'(G(e^))-torsor defined over k. This torsor contains 9 ^ as a C-point, so we have and
Monodromy of Iterated Integrals
267
Let P (resp. P1) be the projection p (resp. p') restricted to T ( 0 ^ ) (resp. G(0«)). Then P" 1 ^©***)) is a P / - 1 (G(9^)))-torsor defined over fc, which contain 0 « as a C-point. Hence we get p - 1 ( T ( e ^ ) ) = T(©W) and P ' - ^ G i e ' * ' ) ) = G(0W) This implies that p(T(0«)) = T(e<*>) and
10.2. AfRne pro-algebraic pro-unipotent groups and torsors. 10.2.1 We assume that G = limG^, where the groups G ^ are affine iinipotent algebraic groups over k and the morphisms GW —> G^) are also over k. We assume further that G^ = G/YiJtlG- Finally we assume that the Lie algebra g of G is finitely presented, i.e. that for i big enough the number of relations defining O/r*"1"1^, of degree less than i + 1 does not depend on i. 10.2.2. The condition that G is finitely presented implies that for i big enough the morphisms Aut(G^+1^) —>• Aut(G^) are surjective. We set Aut(G) :=limAut(G (i) ). Similarly, if Gi and G2 satisfy 10.2.1 and if there is an isomorphism (Gi)*. —> {G2)k of affine pro-algebraic pro-unipotent groups, then the morphisms Iso(Gi~ ,G2 ) —> Iso(Gi ,G^ ) a r e surjective for i big enough. We set Iso(Gi,G2) := limIso(Gi JGJ ). i
Observe that Iso(Gi,G2) is an Aut(G2)-torsor defined over k. 10.2.3. Examples of groups satisfying 10.2.1. 1) Let F b e a smooth algebraic variety over k. Let us set
(see §2 and [Wl]). Then G satisfies 10.2.1. 2) Let V be as in 1.1. Then the group scheme H(V) satisfies 10.2.1. Lemma 10.2.4. Let G be as in 10.2.1. The affine group scheme Aut(G) is an extension of a closed subgroup of GL(Gah) by an affine pro-unipotent pro-algebraic group. This lemma follows from Lemma 10.1.5 and the definition of Aut(G).
268
Zdzislaw Wojtkowiak
10.2.5 Let Gi and G2 satisfy 10.2.1. Let O : GX(C) -> G2(C) be an isomorphism. Then 6 = I i m 9 ( i \ where 9 ( i ) : G^(C) -> G^(C) are i
isomorphisms for all i. Let us set Z(@) := lnnZ(6 ( i ) ), T(0) := limT(9 (i )), G(9) := UrnG(6 w ). Then T(9) is a G(9)-torsor. §11. Torsors associated to non-abelian unipotent periods. Let X be a smooth quasi-projective algebraic variety defined over a number field k. Assume that X has a fc-point x. Let us fix an embedding k c-> C. In [Wl] §7 (see also 2.0 in this paper) we have constructed affine pro-algebraic pro-unipotent finitely presented group schemes 7rf (X(C),x) and ?rPR(X, x) over SpecQ and Spec k respectively. We set
Then we have:
irf(X(Q,x) =lim7r?(X(C),z)(n) and n? In [Wl] Proposition 7.5 we have also constructed a homomorphism (called Boain[Wl])
such that the induced map on C-points,
-+ 7T»R(X,
is an isomorphism. We have cpx = lim<^", where y?" : 7r^(X(C),a;)(n^(C) —>• n
x)(n\C) is induced by (px. For each n we have an Aut(7rPR(X, x)^)torsor Ison := Iso(7rf (X(C),x)W x Jb, 7rf^(X,x)( n )). Applying the construction from §10 to the isomorphismx) is a subgroup of Aut(?rPR(X, x)^). We set and TTP R (X,
Monodromy of Iterated Integrals
269
We have projections pn : Aut(7r? R (A»< n >) Aut(7rPR(X, x)(n) and p : Aut(7rP R(X,x)) -> Out(7rP R (X,z)). Definition 11.1. i) ii)
is the image of G((p%) in
The group Q{X,x)W
The group Q(X,x) is the image of G((px) in Out(?rPR(X, x)).
Let x and y be two fc-points of X. Let 7 be a path in X(C) from x to y. Then 7 induces an isomorphism c 7 : TTP R (XC,X) —» 7rPR(Xc,y). The induced isomorphism of outer automorphisms groups (c 7 ), : Out(7rP R (X c ,z)) -> Out(7rPR(Xc,)) does not depend on the choice of 7 and gives the canonical identification. We need this identification over Specfc. Let us consider the morphism 1 (x,y))} is p. . XA[i] _^ XdA[i] of cosimplicial spaces. Spec (H^R(p— a ?rPR(X, 2/)-torsor (see [Wl] Section 3). It follows from [S] Proposition 4.1 that any 7rPR(X, y)^-toTSOi is trivial, hence any 7rPR(X, ?/)-torsor is trivial (inverse limit of surjective maps of sets (A:-points) is always non empty). Any fc-point 77 of Spec ( ^ D R ^ * " 1 ^ ? / ) ) ) determines an isomorphism
The isomorphism is unique up to conjugation by elements of Hence the induced isomorphism (c,), : Out(7T?R(X,z)) ->
nfR(X,y).
Out{nfR(X,y))
is canonical, in other words it does not depend on the choice of a fc-point of Spec ( ^ ( p - " 1 ^ , ))). The isomorphisms (px and (py are related by the following commutative diagram
irf(X(C),x)(C)
^
7r?R
where c 7 and c7 are induced by the path 7. Observe that c7 = cv oconj(#), where 77 is a fc-point of Spec (H^R((X; x, y))j and conj(#) is a conjugation by an element g € 7rfR(X,x)(C).
This implies the following result.
270
Zdzislaw Wojtkowiak
Proposition 11.2. The groups Q(X,x)^ and Q(X,y)(n^ coincide under the canonical isomorphism (c^)*. The groups Q(X,x) and Q(X,y) coincide under the canonical isomorphism (c^)*. We shall denote these groups by and GDR(X) respectively.
§12. Torsors associated to the canonical unipotent connection with logarithmic singularities. 12.0. Let V be as in section 1.1. We shall assume that A1^) -» H]^R(V) is an isomorphism. We assume that V has a fc-point x. We recall from Corollary 2.4 that we have an isomorphism u : Trf^V, x) —> U(V) (U(V) := Spec(Alg7r(V))). Let u(C) :
and Let ^ a;^ 0 •. . 0 ^ f c be a cocycle representing a class c in if£ R (/9 # ~ 1(x, #)). (We recall from §2 that n?R(V,x) := Spec (flS R (p— ^ ^ x ) ) ) . ) We can The assume that all Uit are one-forms (see 2.2.1). Let a e TTI(V(C),X). value of c on bDR(a) is given by the following formula:
12.1.
c(bDR(a)):=Y
-->«>ik (see2.8).
The homomorphisms 6Q and 6^ are defined in the same way. We want to calculate G(<^x)-torsor T((px) associated to the comparison homomorphism
Lemma 12.2. Tfte G(ipx)-torsor T(tpx) is isomorphic to G(u(C) o ipx)torsor T(u(C) o ipx). Proof. It follows from the fact that u :
*.
TT^R(V^X)
-> Ii{V) is defined over
0
We shall relate the G(u(C) o (px)-toTSOT T(u(C) o cpx) to the monodromy homomorphism 9xy : 7Ti(V(C),:r) -> II(V)(C) of the form UJV from §1. Lemma 12.3. We have
Monodromy of Iterated Integrals
271
Proof. The isomorphism u is induced by Uix ®.. .<8>u)ik —> (X^ ®.. .(&Xik)* (see corollary 2.4.). The value of the class c on bDR(a) is given by 12.1. The value of £(* i f c , . . . , c ^ (see Definition 1.3 and Proposition 1.8). It follows from [Chi] (1.6.2) that It follows from the definition of the homomorphisms bDR and b^ that
Let R be a fc-algebra. Let iR : Trf (V(Q,ar)(Q) -> Trf ( ^ ( C ) , ^ ) ^ ) be the inclusion of Q-points into ii-points. We have ic ° bq = b^. We set We define a functor Xy on fc-algebras in the following way: IV(R) := {/ : TTI(F(C),X) -> II(Vr)(/2) | 3 an isomorphism / : **(V(Q,x)(R)
-> U(V)(R), fobBR = / } .
Observe that / is uniquely determined by /. The functor Xy is represented by an affine pro-algebraic scheme overfc,which we also denote by Xy. Moreover Xy is an AutII(V)-torsor. Observe that 6xy : TTI(V(C),Z) -> n(V)(C) is a C-point of Xy because 6xy = (u(C) o ipx) o b^. We denote by T{0xy) the smallest subtorsor defined over k of Xy such that 6xy G T(6xy)(C) (i.e. the intersection of all subtorsors of Xy defined over fc, which contain 6xy as a C-point). The corresponding group we denote by G(9xy). Lemma 12.4. The G(9xy)-torsor T(6xy) ipx)-torsor T(u(C) o ipx).
is isomorphic to the G(u(C) o
Proof. First we notice that the Aut(II(V)-torsor Xy is isomorphic to the Aut(II(Vr) torsor Iso(?rf (V(C),a;) x k,H(V)). The isomorphism is given by / -> / . The equality 8xy = (u(C) o ipx) o bB implies the lemma. 0 Corollary 12.5. Let V be a projective line over k minus a finite number of k-points. Let S be a loop on V(C) around a missing point. Then the element bDR(S) e K?R{V,V){C) is conjugated to s~2ni, where s e 7r?R(V,v)(k). Proof. It follows from Lemma 12.3 and Corollary 5.4.
0
12.6. Let V be a projective line P£ minus a finite number of fc-points. If x is a tangential base fc-point v then the groups ?rf and ?rfH are not defined in [Wl]. We set ir?(V(C),v) := (iri{V(Q>v)h -Malcev rational completion of 7ri(V(C), v). For the Q-algebra R, the map 6f : TTI(V(C), V) -> n?(V(C),v)(R) is the natural map ni(V(C),v) -> (ni(V(C),v))Q(R) (see
272
Zdzislaw Wojtkowiak
§A.l). The monodromy homomorphism 0$y • fti{V(C),v) —» Il(V)(C) is the Malcev C-completion (see §A.l), hence there is an isomorphism (p$ : ?rf (F(C), v){C) -> 11(7) (C) such that (fyob£ = 0€y. The G(0ity)-toT8OT T(9$y) is defined as above. We set also ?rfR(V, v) := H(V). The isomorphism u is the identity and the G(^y)-torsor T(6^y) is isomorphic to G(^)-torsor T{{p$). §13. Partial information about GDR(PQ \ {0, l,oo}). 13.1. Let V = PQ \ {0,1, oo} and let v be a Q-point of V or a tangential base Q-point of V. For any Q-algebra R we set: T(R):={(x,y,z)€(H(V)(R))3\ 3a eR*,
x = aX,
y « aY, z ~ aZ, x-y-z
= 0}
and
Auf (n(V))(ii):= aY,
f(Z)~aZ}.
The functors T and Aut*(II(V)) are represented by affine pro-algebraic schemes over Q, which we also denote by T and Aut*(II(Vr)). Aut*(II(Vr)) is an affine pro-algebraic group scheme, a subgroup of Aut(II(V)). Moreover T is an Aut*(II(Vr))-torsor. (Let (x, y, z) and (xf, y', zf) be in T(T^). The map x —> x', y —> y' can be extended to an automorphism of the Lie algebra, which is also an automorphism of the group U(V)(R). Hence the action is transitive.) If we replace the group U(V) in the above definitions by n 2 (F) (resp. II 2 ), then we obtain an Aut*(n2(Vr))-torsor T2 (resp. Aut*(II2)-torsor T 2 ). Let us fix generators 5Q, 5^ and S^ of TTI(V(C), V) which are loops around 0,1 and oo respectively, such that Sf0 • S[ • S'^ = 1. Lemma 13.1.1. The Aut* (U(V))-torsorT is a subtorsor of the Aut(U(V))torsor Xy. Proof. Let (x,y,z) G T(R). Let / : ni(V(C),v) -> U(V)(R) be given by f(S'Q) = x, f(S[) = y. The homomorphism 6f : KI(V(C),V) -> Trf (F(C), v)(R) is the Malcev ^-completion of TTI(V(C), V) (see §A.l), hence there is an isomorphism / : 7rf (V(C), v)(i?) -^ II(Vr)(i2) such that /ofeg = /. Therefore f £lv(R). 0 Let 0V : 7Ti(V(C),v) -> n(V)(C) be the monodromy homomorphism of the one form uv- Observe that the triple (0v(SfQ), 0v(S[), ^ ( S ^ ) ) £ T(C). Let Tf(9v) be the smallest subtorsor defined over Q of T such that the triple
Monodromy of Iterated Integrals
273
(01/OSo), ev(S[), OviS'n)) G T(C) is a C-point of T'(0V). Let G'(0V) be the corresponding group. We recall that in §12 we defined G(0v)-torsor T(6v). Lemma 13.1.2. The G(0v)-torsorT(0v) and the G'(0v)-torsorT'(0v) are equal if we identify T with a subtorsor of Xy via the map from Lemma 13.1.1. We shall use the same notation T(0V) and G(0V) for both torsors and both groups. 13.2. We recall that the monodromy homomorphism 0_A> : 7Ti(V(C), ~ot) —> Tl(V)(C) is given by 0^(SO) = (-2ni)X 1
• (-2ni)Y . <(X, y) 1
aJ?(X, Z)
where aJ?(X,F) 6 (n(V),ir{V)) and 5 0 ,5i,5oo are as in the figure in 0.2. The triple (0^(S o ),0_£(Si),0^(Soo)) ^ T(C). Hence we get the G(0^)torsor T(0__A.), a subtorsor of T. Proof of Theorem B from the introduction. Let u be a Q-point of V. It follows from Lemmas 12.2 and 12.4 that the group schemes G(tpv) and G(0V) are isomorphic. The isomorphism is induced by the isomorphism u : TTP^V, V) -¥ n(V). Hence the images of the groups G((pv) and G(0V) in the groups of outer automorphisms are also isomorphic. The homomorphisms 0V and 0_A, are conjugate; hence the images of groups G(0V) and G(0_^) in Out(II(Vr)) are equal. Observe that the map Aut*(II(Vr)) ->• O\it(U(V)) is injective. Hence fon(V) and G(0—±) are isomorphic. 0 13.3. Let 0 : 7n(V(C), ot) -> n 2 (V)(Q (resp.02 : TTI(V(Q, ot) ^ n 2 (C)) be the composition of 0_^ with the projection on n2(Vr)(C) (resp.n2(C)). The triple (0(5o),0(5i)1°0(5oo)) e T2(C) (resp. (02(5O),02(51),02(5OO)) € T 2 (C)). Repeating the construction from 13.1 we get a G(0)-torsor T(0) (resp. G(02)-torsor T(0 2 )), a subtorsor of Aut*(n 2 (^))-torsor T2 (resp. Aut*(n2(C))-torsor T 2 ). The projections p : U(V) -> n 2 (F) and p 2 : II(Vr) -^ n 2 induce homomorphisms of group schemes and (p2)* : Aut*(n(F))-> Aut*(n 2 ) and morphisms of torsors compatible with homomorphisms of group schemes P*:T^T2
and (p2)* : T-> T' 2 .
274
Zdzislaw Wojtkowiak
Lemma 13.3.1. We have
and
>)) = G(e2). This lemma follows from Lemma 10.1.7. Observe that the group G(9) is the image of the group G(Q—±). This implies Theorem C from the introduction. Corollary 13.3.2.
The group G(62) is isomorphic to a quotient of
13.4. Below we shall calculate the G(02)-torsor T(0 2 ). We have II2 = lim (n 2 /T n + 2 II 2 ). If in definitions of T and Aut*(II(V)) in 13.1 we replace the group U(V) by n 2 /r n + 2 II 2 , then we get an Aut*(n 2 /r n + 2 n 2 )-torsor T2. Let 6 n be the composition of 02 with the projection onto II 2 /r n + 2 II 2 (C) = n 2 (C)/r"+ 2 II 2 (C). The triple (e n (5 0 ),e n (5i),e n (Soo)) € 7^(C). By definition Tn := T(Qn) is the smallest subtorsor of T2 defined over Q which contains the triple as a C-point and Gn = G(On ) is the corresponding group, a subgroup defined over Q of A u t * ( n 2 / r n + 2 n 2 ) . Let p^ T2+ x and p^1 : A u t *(( n) 2 / r n + 1 + 2 n 2 ) -^ Aut*(n 2 /r n + 2 n(2 )) be induced by the projection n 2 / r n + 1 + 2 I I 2 -> n 2 / r n + 2 n 2 . It follows from Lemma 10.1.7 that
13.4.1.
PZ+1(Tn+1) = Tn
and P£ + 1 (G n + 1 ) = G n ,
We recall that by definition T(02) := limT n and G{62) := limG n . n
n
We recall that the group of Q-points (II 2 /r n+2 II 2 )(<2) is a Lie algebra overequipped with the multiplication given by the Baker-Campbell-HausdorfT formula. A linear basis of this Lie algebra is given by X, Y, (Y,X),..., (Y,X)Yn~1. The homomorphism 62 : 7Ti(F(C), ot) -> II2(C) is given by 92(S0) = {-2m)X and 0{Si) = {-2m)Y + § {2-Ki)C>{k){{YX)Yk-1) (see §6). k=2
0. Calculations of To and Go-
Monodromy of Iterated Integrals
275
We have 60(50) = (-2ni)X and 60(51) = (-2ni)Y. Observe that {2m) is not afc-throot of a rational number for any k = 1,2,3... . This implies that T0(C) - {aX,aY | a G C*} and G0(C) - {fa I fa(X) = aX, fa(Y) = aY I a G C*} C Aut(n 2 (C)/r 2 n 2 (C)). 1. Calculations of Xi and Gi. We have 6i(5 0 ) = {-2m')X and 0i(Si) = (-27ri)y. Hence we get aeC*} and Gl(C) = {/a I / a W = <*X, /a 0 0 = ^
I * G C*}.
2. Calculations of T2 and G 2 .
We have 0 2 (5 O ) = (-2ni)X, 6 2 (5i) = (-2ni)Y + (2ni)C(2)((YX)Y). Observe that (2TTZ)C(2) = ^(-2TTZ) 3 . This implies that T2(C) = { a l , a 7 + ^ a 3 ( ( y i ) y ) | a G C* } and G2(C) = {/a I fa(X) = OLX, fa(Y) = aY | a e C*}. 3. Calculations of T3 and G3. We have 6 3 (5 0 ) = (-2TTZ)X, 6 3 ( 5 I ) = (-27ri)y + (27ri)((2)((YX)Y) + +(27rz)C(3)((rX)F2). Assume that dimG 3 = dimT3 = 1. Then it follows from 13.4.1 and Corollary 9.2 that G3 = G(0,0,c3) where c3 e Q. As T3 is a one-dimensional variety we have T3(C) = {aX, aY + -^a3((YX)Y) + P3((YX)Y2) I a e C*, fteC, p(a,/?3) = 0} for some Laurent polynomial p(x,y) € Q[x, ^,2/]- Observe that to each value of a corresponds exactly one value /33(a), because T3 is a G3-torsor. Hence /33(a) = p(a) where p(x) 6 Q[x, i ] . If (aX1aY+^a3((YX)Y)^p(a)((YX)Y2) G T3(C), then ((aOX,(aOy+^(aO 3 ((yX)y)+(p(a)^ 4 + C 3 ^-^ 4 )a)((yX)y 2 )) G T3(C). Hence p(at) = p(a)t4 + c3(t — tA)a. Therefore p(x) = ax 4 + c3x. The equality (27rz)C(3) = a(—2TT2')4 + C3(-2TT2') implies a = 0 and c3 = — C(3). Therefore we get dimG 3 = 1 if and only if £(3) G Q. Therefore if C(3) ^ Q (one knows that £(3) is irrational) then dimG3 = 2, T3(C) = { a X , a F + ±a3((YX)Y)
+/? 3 ((FX)F 2 ) | a G C*, /% G C}
276
Zdzislaw Wojtkowiak
and G3 = (0,0,0 | 0,0,1). n. Calculations of Tn and Gn. We have 9 n (S 0 ) = {-2ni)X, n-l
k=2
Assume that Gn-\
= G(0,0, C3,0, C5,0, C7,... | 0,0,63,0,£5,0,67,...)
(see Corollary 9.4), where c\ = c2 = c 4 = . . . = c2k = . . . = 0, C2fc+i =
-C(2fc 4-1) and e2k+1 = 0 if C(2fc + 1) G Q, and c2fc+i = 0 and e2k+1 = 1 if -1) g Q. Then the torsor T n _i corresponding to the group G n _i is given by n-l
n-l
((1 - £fcHa + ek/3k) ((YX)Yk-1)
| a € C , VA; ft €
Assume that n = 2p. Then C(n)(27ri) = r n (-27ri) n + 1 , rn e Q. This equality, the fact that (2ni)n £ Q and the property 13.4.1 imply that Gn = G(0, 0, C3, 0, C5, . . . , Cn-i, 0 I 0, 0,ff3,0, . . . , £„_!, 0) and
+ ekfa) ((yX)y fc - 1 ) | a 6 C*, V ft G c } . fc=3,fc-odd
fe
Assume that n = 2p+ 1. Assume that dimG n = dimG n _i. Then G n - G(0,0, c 3 ,..., 0, c n | 0,0, e 3 ,0, e5 , • • •, 0,0)
(by Corollary 9.4)
Monodromy of Iterated Integrals
277
and n-l
Tn(C) = {aX, aY +
£
rkak+1((YX)Yk-1)
+
k=2,k-even n-l
((l-ek)cka + ekpk)((YX)Yk-1)+pn((YX)Yn-1)
]T
with
k=3,k-odd CKGC*;
Vfc,fteC; p{a,ez-faeb-fa
.. .,&) = 0}
for some polynomial p(x, y 3 ,2/5,..., yn) € Q[x, ±, 2/3, y 5 ,..., yn). Assume that for some a$ there are two different j3n. Then there is g in Gn(C) such that g(X) = X, g(Y) = Y + . . . + bn((YX)Yn-1) with bn ^ 0. Then it follows from the proof of Lemma 9.5 that Gn — G(0, 0, C3,0,..., 0, 0 | 0,0, £3,0,..., 0,1) and dimG n = dimG n _i + 1 . Hence for any a G C* there is a unique /3n corresponding to that a. Observe that /?„ is an algebraic function of a, /?3, /?5,... , which depends only on a. Hence /3n = p(a) for some p(x) G Q[rr, ^] • Choose any / G G n (C). Then f(X) = tX and n-l
f(Y) = tY+ Yl
((! " ^)C*C "
for some t G C*, 63,65,..., 6 n _ 2 G C. Acting by / on any element of Tn(C), we get n-l
{{crt)X, (at)Y+
Yl
rk(at)k^1((YX)Yk-1)
fc=2, fc-even
n-l
+
J2
(«(1 - ek)ck(t - t^1) + aekbk + (1
k
k=3,k-odd
+ ek(3ktk+1)((YX)Yk-1) + ( a C n ^ - r + ^ + ^ Therefore we get p(at) = cna(t - tn+1) + p(a)tn+1. This implies p(x) = ax n + 1 + c n x. The equality C(n)(-27ri) = a(2ni)n+1 cn(27ri) implies p(x) = cnx and cn = — C(^). Therefore we get dimG n = dimG n _i
if and only if C(n) G Q.
278
Zdzislaw Wojtkowiak
The final result is the following. Proposition 13.5. Let G = limG n . Then G(62) = G and G = n
G(0,0, c 3 ,0, c 5 ,0, c 7 , . . . , 0, c2jfe+i, 0 . . . | 0,0, e 3 ,0, e 5 , 0 . . . , 0, e2fc+i, 0 . . . ) , :
£2fc+i = 0 if and only if ((2k + 1) G Q, and then c2fc+i = — ((2k + 1). The torsorT(62)(C) is given by oo
{aX,aY+
J2
oo
k+l
k 1
rka ((YX)Y - )+
^
((
'M I a € C , for all Jk, /3fc e
where ((2k) = -r2k(27ri)2k, andc2k+i = —C(2*:H-1), ^2fc+i = 0 i/C(2fc+l) G = 0, Corollary 13.6. T/ie ^ro^p G contains the group H := {/a | fa(X) = aX, fa(Y) = ctY \ a e C*} if and only if all numbers (,(2k + 1) are irrational. Proof. It follows from Proposition 13.2 that all £(2fc + l) are irrational if and only if G = G(0,0,0,...,0,... | 0,0,1,0,1,0,.. .0,1,0,1,0,...) (all a = 0, e\ = 0, all 62k = 0 and all £2fc+i = 1 for fe = 1,...). Then we have H C G. 0,...). Then there exists f € G such that /(X) = a l and
where all X2A;+I / 0, Corollary 9.5 implies that G = G ( 0 , 0 . . . , 0 , . . . | 0 , 0 , 1 , 0 , 1 , . . . 0 , 1 , 0 , . . . ) (all a = 0, ex = e2fc = 0, e2fc+i = 1 for fe = 1,2,...). 0
Corollary 13.7. Le£ ^ 6e the smallest closed subgroup of Aut(?r2) defined over Q, containing G and H. Then Q = G(0,0..., 0,... | 0,0,1,0,1,0 ...) fa// Ci = 0 , e i = e2fc = 0, e2k+i
= 1 for k =
1,2,...).
Proof. The group Q contains / such that f(X) = X and
k=l
where all x2k+i ^ 0. Lemma 9.5 implies the corollary.
Monodromy of Iterated Integrals
279
13.8. Proof of Theorem D in the introduction. We recall from §6 that the monodromy homomorphism
is given by S o - • (-27rt)X, i=O,j=O
(see §6, formula (7)). It is an observation of Drinfeld that the numbers satisfy the equation
-1+
( n>0,m>l
k=2
(see [Dr]). Therefore we can describe the monodromy homomorphism 0 in the following way
k=2 ?+j=k
J
Let us temporarily write X" for the torsor given in Theorem D i) and Gf for the corresponding group. Then T(6) is a subtorsor of T' and the group G{6) is a closed subgroup of G' because (0(So),0(Si)) G T'(C). The image of the group G' in Aut(n(F)/r n+2 II(V r )) has the same dimension as the subgroup Gn of A u t ( l l 2 / r n + 2 n 2 ) . The group G(6) projects onto G n , hence Gf = G{0) andT' = T(0). The point iii) of Theorem D and Corollary E follow from Corollary 13.7 and Corollary 13.6 respectively. §14. The group G*DR(V) for pointed projective lines and for configuration spaces. In this section we shall translate the results about the monodromy representations from §§5-8 into results about the group GDR(V). Let V = P£\{ai,..., a n +i} and let v be a Appoint of V. We can assume that a\ = 0, a2 = 1 and a n +i = oo. We recall that Lie(V) is a free Lie algebra over k on Xi := (z^_za. )*, 2 = 1,2,..., n. Let us set Xn+\ = — ^^=1 Xi. Let R be a fc-algebra. Let us set
Aut* (U(V))(R) := {/ G Aut(U(V)R) | 3a/ G i?*, /(*i) = afXu f(X2) « a/X 2 , /(Jffc) ~ a/^ib * = 3, ...n + l}.
280
Zdzislaw Wojtkowiak
Here « means a conjugation by an element of U(V)(R) with the zero coefficients at X\ and X2 and ~ means conjugation by an element of U(V)(R). Similarly for yn = (P x (C)\{0, l,oo})"~ 3 we set Aut*(U(yn))(R)
:= {/ G Aut(n(y n )R)| 3a/ € JT, /(X i f i ) - a , * ^ - } .
(The elements Xij are as in §7.) These functors are representable by affine pro-algebraic group schemes over k and over Q, which we denote by Aut* (11(10) and Aut*(lI(Xi)) respectively. Let us set Out*(11(10) := image(Aut*(n(Vr)) -> Aut(n(F))/Inn(n(F))) and
Out*(n(yn)) := Aut*(n(yn))/inn(n(yn)). We shall define an Aut* (11(10 )-torsor T and the Aut*(n(^ n ))-torsor Tn as follows. Let S i , . . . , S n + i be geometric generators of TTI(V(C),V), i.e. loops around a i , . . . , an_|_i respectively. For any fc-algebra R we set T(R) := {/ €Hom(in(V{C),v);Il(V)(R)) \ 3af G IT, f(S1) = f(S2) w afX2, f(Sk) ~ a/Xfc * = 3,..., n + 1} and T"(i2) = {(f eHom(vi(yn{C),y);Il{yn)(R))
\ 3a,, £ R* such that
(The elements A^- are as in §7.) Proposition 14.2. We have GDR(V)
C Out*(R(V)).
Proof. Let us choose the vector 0i as a tangential base point. Let S\ be a loop around 0 as in the picture of §4. Let us choose a tangent vector Vk at each point a&. Assume that V2 = T^. Let F be a family of paths from the base point to V2, -. •, vn+i such that 72 is the interval [0,1]. Let Si, S2,..., S n +i be a sequence of geometric generators associated to F. Let 0_^ : TTI(V(C), "ot) —f II(y)(C) be the monodromy homomorphism. It follows from Corollary 5.4 and Theorem 6.1 that G T(C).
Monodromy of Iterated Integrals
281
Hence G(0^) C Aut*(n(F)). This implies that QDR{V) C Out*(n(F)) if we identify n^R(V,v) and II(F) by the homomorphism u from Corollary 2.4. 0 Observe that Tn is a Aut*(n(3^n))-torsor. Hence T n /Inn(n(^ n )) is an Out*(n(^ n ))-torsor. Proposition 14.3. n)c
Out* (n(yn)).
Proof. Corollary 7.3 implies that 0y^yn G T n (C). Hence the proposition follows. Proposition 14.4. For n > 4 we have isomorphisms
induced by the projection yn+\ —> yn.
Proof. The proposition follows from Proposition 7.4, Lemma 7.5 and Corollary 7.6. 0 Observe that 34 = PQ \ {0,1, oo}. Hence we have the following corollary. Corollary 14.5. For n > 4 we have GDR(yn) « GDR{FQ \ {0,1, oo}). The group H(V) is a normal subgroup of n(3^ n+2 ). Hence n(>?n+ 2) C Aut(n(F)). Proposition 7.7 implies the following result. Proposition 14.6. We have GDR(V)
- n(yn+2)/n(yn+2)«
gDR{v\i \ {o, 1, oo}).
§15. Conjectures. Let X be a smooth quasi-projective algebraic variety defined over a number field k and let a; be a fc-point or a tangential base Appoint. Denote by 7rf*(X xfc,x)i the pro-1 completion of the etale fundamental group o f X x L k
k
Let &x ' Gal(k/k) —> Out(7rf*(X x k,x)i) be the natural representation of k
the Galois group into the outer automorphisms of the pro-1 completion of the etale fundamental group.
282
Zdzislaw Wojtkowiak
The monodromy representations of iterated integrals, i.e. the homomorphismsTT? R (X,X)(C) have many properties analogous to the action of Gal(k/k) on etale fundamental groups. We give some examples. Example 1. Let X be the projective line minus a finite number of kpoints. Let 5 be a loop on X around a missing point. If a G Gal(k/k) then a(S) is conjugate to Sx^a\ where x is the cyclotomic character (see [13]). It follows from Corollary 12.5 that (px(S) is conjugate to an element s27™, where «€7r? H (-Y,*)(*). Example 2. Assume that X is a smooth projective variety and D is a divisor with normal crossings in X. Assume that X := X \ D is defined over a field k. Let S be a loop in X around an irreducible component of D. As before a(S) ~ Sx^ in the case of a Galois action. It follows from Corollary 7.2 that ipx(S) ~ s2nt, where s £ /K^R(X,x)(h) and X is a configuration space of n points in C (see §7). Example 3. The element aJ?(X, Y) satisfies Z/2, Z/3 and Z/5 cycle relations. The element considered by Ihara in [12] satisfies analogous relations, (see also Remark 2 at the end of §8.) Example 4. Chudnowsky has shown that the transcendence degree of the field generated by abelian periods is 2 if X is an elliptic curve with complex multiplication. The existence of a non-trivial endomorphism of X implies that the image of Gal(fc/fc) is contained in a two dimensional subgroup of
GL(Hlt(X x *),). In the second part of this paper we investigated the group GDR(X). Our results show that this group remains very close to the image of Gal(^/A:) in the group of outer automorphisms of the etale fundamental group of X. We shall formulate some conjectures to make this relation precise. Let 7rf*(X x k,x)i 0 Q be the rational Malcev completion of k
Trf*(X x k,x)t/rn+1Tr?(X k
x k,x),. k
Let $ £ : Gal(fe/fe) -> Out (Trf* (X x k.x)^
® Q) be the natural homomor-
AC
phism. Conjecture 15.1. There exists an affine algebraic subgroup Gn of the group Out (Trf*(X xfe,x)\n)®Q) such that $£(Gal(ifc/ik)) is open in Gn(Qz) for the non-archimedean metric ofQi. Let Lie(Gn) be a Lie algebra of the group Gn.
Monodromy of Iterated Integrals
283
We conjecture furthermore that the Lie algebras Lie(Gn) form a compatible system. We denote the resulting pro-Lie algebra by Lie(<&x(G&l(k/k))). (We hope that such a conjecture has already been made by someone else. We need it to compare the group GDR(X) with the image of Gal(fc/fc). We have formulated the conjecture in analogy with [D4, 8.14].) Let us assume that X is defined over Q. Conjecture 15.2. There is an isomorphism Lie($ x (Gal(Q/Q))) « Lie(gDR(X)) ® Q/. We recall that we have the isomorphism
Let Z((fx ) be thefc-Zariskiclosure of (pxn in ISO(TT?(X(C),z) (n) x fc;7rPR(X,x)(n)). We recall that T{ipxn') is the corresponding torsor (see §11). Question 15.3. Is it true that
Assume that X is a complement of a divisor with normal crossings in a smooth projective scheme of finite type over a number field k and that H})R(V) — Al(V). In this case the group QDR(X) can be considered as a subgroup of Out(npf)). Let CX : G m -> Aut group (n(X)) = Aut Lie algebra(£W) be a homomorphism such that t G Gm(k) acts on H(X) as a multiplication by t. Question 15.4. Does the group 5 D R ( ^ ) contain the image of Cx7> Corollary 15.5. Let V = P^ \ {0,1, oo} and let v = ot be a tangential base point. If Questions 15.3 and 15.4 have affirmative answers for V and v then the transcendence degree of the field Q(27ri)(£(3), C(5), • • • C(2& 4-1)) is k+1. The corollary follows from Theorems A, C and D and Lemma 10.1.3.
284
Zdzislaw Wojtkowiak
§A.l. Malcev completion. A1.0. Let G be a unipotent algebraic group over a field k of characteristic zero. The exponential map exp : Lie(G) —> G is an isomorphism of Lie(G) equipped with the multiplication given by the Baker-Campbell-Hausdorff formula with G. For a nilpotent Lie algebra L over k equipped with the multiplication given by the B.-C.-H. formula we have Aut groU p(£) = AutLie algebra(£)•
Let R be a fc-algebra. Then the same holds for a unipotent algebraic group scheme over R. A l . l . Let 7T be a finitely generated nilpotent group. Then there exists an affine unipotent algebraic group scheme TTQ over SpecQ and a homomorphism bq : n —> TTQ(Q) functorial with respect to homomorphisms ?r —> TT', which has the following universal property: Let GQ be a unipotent algebraic group over SpecQ. Let> Gq(Q) be a homomorphism. Then there is a unique morphism of unipotent algebraic groups
Gq such that on Q-points we have TTQ(Q) ^^ TTK{K), which is the composition of bq and the inclusion has the same universal property with respect to homomorphisms GK(K), where GK is a unipotent algebraic group scheme over K. Proof. It follows from Al.l that there is a unique / : TTQ(Q) -» GK(K) such that / o 6Q =
The group TTK together with the homomorphism &K : n —)• TTK(K) has the same universal property with respect to homomorphisms (p : n -> GK(K), where GK is a pro-unipotent pro-algebraic group scheme over K. We call the pair (TTK, 6x : TT —> TTK(K)) the Malcev K-completion of TT. It is unique up to a unique isomorphism. A1.4. It follows from [Ch2] (Theorems 2.1.1 and 2.7.2) and [Q] (Lemma 3.3 and (1.4)) that the pair (af *(V,v), bDR : ^ ( ^ ( C ) , v) -> *f *(V, v)(Q) (resp. (Trf (^(C),t;), b^ : m(V(C),v) -+ Trf (V(Q,v)(Q))) is the Malcev C-completion (resp. rational completion) of ?ri(y(C), v).
Monodromy of Iterated Integrals
285
§A.2. The torsor and the group corresponding to the DrinfeldIhara relation. A.2.0. In this section we describe the torsor and the group which correspond to Z/2, Z/3 and Z/5-relations. It is interesting to observe that the Z/3-relation for the group is different from the Z/3-relation in the Galois theory. Let us set X = P 1 (C) \ {0, l,oo}. Let So, Si and SQO be the generators of TTI(X, 01) (see the introduction, and more particularly the picture on p. 5). Observe that S^ • Si • So = 1. Notation. If a and b are elements of the group G we set ab := b • a • b"1. A.2.1. For any Q-algebra k, let us set: TG(k) := {/ : m(X, Of) -> 7r(X)(k)\3t G k* and a(X,Y) G nf(X)(k) with f(S0) = -tX f(Si) = a(X,Y).(-tY)-a(X,Y)-1 *oc) /(5oo) - ( - ^ r 1
• a(X, Z) • (-tZ) • a(X, Z)"1 • ( |
*3) ( - V ) -a{Y,Z) • (-i*Z) -a(Z, Jf) • (-^X) -a(X,y) = 0, *5) a(Xi 2 ,Xi 5 ) • a(X 34 ,X 23 ) • a(X 15 ,X 45 ) • a(X 23 ,X 12 ) • a(X 45 ,X 34 ) = where the last equality takes place in the group 7r(y^)(k). The condition *3 is equivalent to the condition *g) obtained by replacing A.2.2. We shall define an affine group scheme G over Q in the following way. For a Q-algebra k we set: G(k) := ig G Aut(7r(X)(A;))|3s G A;* and a(X,Y) G nf(X)(k) with g(X) = 5X **oo) g(Z) =
**3) a(y, Z) • a(Z, X) • a(X, y ) = 0 **5) a(Xi2,Xi5)
- a(X 34 ,X 23 ) • a(Xi 5 ,X 4 5 ) • a(X 2 3 ,X i 2 ) • a(X 4 5 ,X 3 4 ) = 0.
The last equality takes place in the group 7r(^
286
Zdzislaw Wojtkowiak
Lemma A.2.2.1. Let us suppose that a(X,Y) andA(X,Y) are inn'(X)(k) and s G A;*. Then there is a unique element (3(X,Y) G n'(X)(k) such that a(sX, (sY)^x'Y^)
• (3(X, Y) = A(X, Y).
Proof. From the equation one gets induction formulas for the coefficients of0(X,Y). 0 Proposition A.2.2.2. The set G(k) is a group. Proof. Let g(Y) = (sY)a^x^ and^i(F) = (tY)^x^Yl ThenagiOg(X,Y) = XiY a(tX, (tY)^ ^)-P(X, Y). It is immediate to check the conditions ** oo , **2 and **3 for agiOg(X,Y). To show that relation **5 holds, we must use the group 7r(3^5)- Let g G G(k). We shall extend g to an automorphism of ft(y5(k) in the following way. We set g(X12) = sX12, g(XM) - sX34, g(X23) = {sx23)«(x^x^x^x™\
g(X45) = g(X51) =
Applying the automorphism g of 7r(3^5)(fc) to the relation **5 for g we get the relation **5 for g\ o g. Assume that g\og = id. Then the element /3(X, Y) is uniquely determined by the equation
(see Lemma A.2.2.1). Applying gi to the equality a(X, Y) • a(Y, X) = 0, we get
a(-Y,, ((-X)^'^-1) •ftx,Yy1 = 0. s
s s
It follows from Lemma A.2.2.1 that
One proves the conditions **oo, **3 and **5 for (3(X, Y) similarly.
A.2.3. We shall show that TG is a G-torsor. Proposition A.2.3.1. The set TG(k) is a G(k)-torsor, i.e. the group G(k) acts freely and transitively on TG(k). If / G TG(k) and g G G(k) then the verification that go f G TG(k) is instantaneous. The action of G(k) on TG(k) is obviously free.
Monodromy of Iterated Integrals
287
Suppose that / G TG(k), /(Si) = (-tY)<x^ and h e TG(Jfe), ft(Si) = ). We are looking for g <E G(k) such that g o f = h. If s(y) = then a(X, Y) is determined uniquely by the equation a((pX, ( 0 Y ) a ( x ' r ) ) • a{X, Y) = b(X, Y)
A2.3.2
where (j) = | . Suppose that we have already shown the conditions **oo and **2 for a(X, y). Applying the automorphism g to the formula *3 for / we get a(<j)Y, (4>Z)0(Y>X)'MX>Z)) • (3(Y, X) - (3(X, Z) = 6(y, Z).
Replacing X, 7 by F and Z in A.2.3.2 and using Lemma A.2.2.1 we get /3(y, Z) • /3(Z, X) - /3(X, Y) = 0. The proofs of *oo, *2 and *5 are similar. 0 Proposition A.2.3.3. The functors TG : Q — algebras -* Sets and G : Q — algebras —>• Groups
are representable by an affine pro-scheme over Q and an affine pro-group scheme over Q. We denote these pro-schemes by TG and G respectively. The pro-scheme TG is a G-torsor. Proof. This follows from the definitions of the functors TG and G, Proposition A.2.2.2 and Proposition A.2.3.1. 0 One can show that the monodromy representation #_A : n\ (X, 01) -> TT(X)(C) is an element of TG(C). One takes t = 2ni and a(X, Y) = loga^i(X, Y). This implies the following result.
Corollary A.2.3.4.
References [A]
K. Aomoto, Special Values of Hyperlogarithms and Linear Difference Schemes, Illinois J. of Math. 34 (2) (1990), 191-216. [An] Y.Andre, G-Functions and Geometry, Aspects of Mathematics, F. Vieweg & Sohn Publ., 1989. [C] P.Cartier, Developpements recents sur les groupes de tresses, applications a la topologie et a l'algebre, Seminaire Bourbaki, 1989-90, no. 716, Asterisque 189-190 (1990), SMF Publ., 17-67.
288
Zdzislaw Wojtkowiak
[Chi] K. T. Chen, Algebra of iterated path integrals and fundamental groups, Transactions of the AMS 156, (1971), 359 - 379. [Ch2] K. T. Chen, Extent ion of C°° function algebra by integrals and Malcev completion of TTI, Adv. in Math. 23 (1977), 181 -210. [Dl]
P. Deligne, Theorie de Hodge II, Publ. Math. IHES 40, (1971), 5 58.
[D2]
P. Deligne, Hodge Cycles on Abelian Varieties, Lecture Notes in Math. 900, Springer-Verlag 1982, 9 - 100.
[D3]
P. Deligne, Equations differentielles a points singuliers Reguliers, Lecture Notes in Math. 163, Springer Verlag, 1970.
[D4]
P. Deligne, Le groupe fondamental de la droite projective moins trois points, in Galois Groups over Q, Math. Sc. Res. Ins. Publ. 16, 1989, 79 - 297.
[D5]
P. Deligne, letter to Bloch, 2.02.1984.
[Dr]
W. G. Drinfeld, On quasitriangular quasi-Hopf algebras and a group closely connected with Gal(Q/Q), Leningrad Math. J. 2 (4), (1991), 829 -860.
[G]
A.Grothendieck, On the De Rham cohomology of algebraic varieties, Math. Publ. I.H.E.S. 29 (1966), 95-103.
[H]
R.M.Hain, On a generalization of Hilbert's 21st problem, Ann. Scient. Ec. Norm. Sup. 4 e serie 19 ( 1986), 609-627.
[HZ]
R.M.Hain, S.Zucker, A guide to unipotent variations of mixed Hodge structure, Lecture Notes in Math. 1246, Springer Verlag, 92 - 106.
[KO] N. Katz and T. Oda, On the differentiation of De Rham cohomology classes with respect to parameters, J. Math. Kyoto Univ. 8 (1968), 199 - 213. [II]
Y. Ihara, Automorphisms of pure sphere braid groups and Galois representations, in The Grothendieck Festschrift, Volume II, Birkhauser, 1990, 353 - 373.
[12]
Y. Ihara, Braids, Galois groups, and some arithmetic functions, Proc. of the ICM 1990 (vol. I) (1991), 99 - 120.
[13]
Y. Ihara, Profinite braid groups, Galois representations and complex multiplications, Annals of Math. 123 (1986), 43 - 106.
[L]
A.Lichnerowicz, Theorie globale des connexions et des groupes d'holonomie, Edizioni Cremonese Roma, 1955.
[MKS] W. Magnus, A. Karrass, and D. Solitar, Combinatorial Group Theory, Pure and Applied Mathematics, XIII, Interscience Publ., 1966.
Monodromy of Iterated Integrals [R] [S]
[Wl]
[W2] [W3]
[W4] [W5]
289
R. Ree, Lie elements and an algebra associated with shuffles, Annals of Mathematics 68(2) (1958), 210 - 220. T. A. Springer, Galois cohomology of linear algebraic groups, in Algebraic Groups and Discontinuous Subgroups, Proc. of Symp. in Pure Math. IX, AMS (1966), 149 - 158. Z. Wojtkowiak, Cosimplicial objects in algebraic geometry, in Algebraic K-theory and Algebraic Topology, Kluwer Academic Publishers, 1993, pp. 287 - 327. Z. Wojtkowiak, Monodromy of polylogarithms and cosimplicial spaces, preprint I.H.E.S., 1991. Z. Wojtkowiak, A note on the monodromy representation of the canonical unipotent connection on P 1 (C)\{ai,..., a n }, preprint MPI, Bonn, 1990. Z. Wojtkowiak, Functional equations of iterated integrals with regular singularities, Nagoya Math. J. 142 (1996), 145-159. Z. Wojtkowiak, Non-abelian unipotent periods, monodromy of iterated integrals, RIMS-969, February 1994 preprint.
Universite de Nice-Sophia Antipolis Departement de Mathematiques Laboratoire Jean Alexandre Dieudonne U.R.A. au C.N.R.S., No 168 Pare Valrose - B.P.N° 71 06108 Nice Cedex 2, France
Research Institute for Mathematical Sciences Kyoto University Kitashirakawa, Sakyo-ku, Kyoto 606, Japan
Part IV. Universal Teichmiiller theory
The universal Ptolemy group and its completions Robert Penner Sur Pisomorphisme du groupe de Richard Thompson avec le groupe de Ptolemee Michel Imbert The universal Ptolemy-Teichmiiller groupoid Pierre Lochak and Leila Schneps
The Universal Ptolemy Group and Its Completions Robert C. Penner* §0. Introduction A new model of a universal Teichmiiller space was introduced in [P2], and a universal analogue of the mapping class groups, called the "universal Ptolemy group", was denned and studied. In our concentration on the geometry in [P2], we perhaps obfuscated the essentially easy algebraic and combinatorial arguments underlying this definition, and one goal here is to give a gentle survey of these aspects and a complete definition of the universal Ptolemy group. To be sure, the geometric side of the story provides the main calculational tools of the theory (and explains the terminology "Ptolemy group"), but we do not discuss these aspects here. To hopefully gain some perspective on the results in this paper, we next review the "universal (decorated) Teichmiiller theory" from [P2]. In this context, the appellate "universal" means that we seek certain infinitedimensional spaces (actually, we shall find infinite-dimensional symplectic Frechet manifolds supporting a group action) together with maps of each of the "classical (decorated) Teichmiiller spaces" into our universal object in such a way that classical combinatorial, topological, and geometric structures arise as the pull-back or restriction of corresponding (group invariant) structures on the universal objects. The classical "(decorated) Teichmiiller theory" may be described as follows. Fix a smooth oriented surface F* of genus g > 0 with s > 0 punctures or distinguished points, and consider the corresponding Teichmiiller space Tg consisting of all complete finite-area metrics of constant Gauss curvature -1 on Fg modulo push-forward by diffeomorphisms of Fg which are homotopic to the identity. The moduli space A4sg is likewise defined to be the collection of all such metrics modulo push-forward by all orientationpreserving diffeomorphisms of Fg. The mapping class group MCsg consists of all homotopy classes of orientation-preserving diffeomorphisms of Fg, and MCsg thus acts on Tgs with quotient Msg. In case s > 1, we may consider the decorated Teichmiiller space Tg which is by definition the total space of the R+-bundle Tg -» Tgs', where the fiber over a point of Tg is the set of all s-tuples of horocycles, one horocycle about each puncture of F*. MCsg evidently also acts on Tgs by simultaneously Partially supported by the National Science Foundation
294
Robert Penner
pushing-forward metrics and permuting horocycles, and the quotient of Tg by this MG*-action is the decorated moduli space Mg. The sequel does not actually require any real mastery of these classical constructions, and we have included the precise definitions here only for completeness (see [PI] for further details and references) and to describe the following classical square of Fg
fs '9
Msg
> Ts
r
'9
— > Msg
where each morphism in the square is just a corresponding forgetful map, so MCg acts transitively on the fibers of the vertical maps. We shall study here a corresponding universal square Tess
—>
Tess
Mod
—> Mod
relating our corresponding universal spaces and (equally to the point) define a countable group G, the "universal Ptolemy group", one of whose completions acts transitively on the fibers of the vertical maps. There are in fact several reasonable completions of G as we shall discuss. Indeed, it seems clear (cf. [LS] in this volume) that the Galois group of the algebraic closure of Q over Q (to be denoted here simply Galois) arises as a group of automorphisms of a suitable completion of G explaining our exposition of this material here. To abide by our stated goal to gently survey, we shall only highlight various aspects of the theory restricting attention to the righthand side Tess —>• Mod of the universal square for simplicity. This is already sufficient to describe the group G, which is our primary focus in this paper. We define the spaces Tess and Mod in §1 and then define and discuss the group G in §2 taking this opportunity to describe a few recent related results of Imbert-Kontsevich and Lochak-Schneps. §3 and §4 contain complete proofs of several new results; specifically, §3 describes a natural completion of G which acts transitively on the fibers of the map Tess -> Mod and improves our discussion of this in [P2;§4], while §4 describes some related transitivity results which give a new structure theorem for normalizers of Fuchsian groups and suggest still other completions of G as we shall discuss.
Universal Ptolemy Group
295
It is a pleasure to acknowledge helpful conversations on various aspects of this work with Bob Guralnick, Pierre Lochak, Leila Schneps, Vlad Sergiescu, and Dennis Sullivan. §1. Tesselations A tesselation r = {ji} is a countable and locally-finite family of geodesies 7»> i > 1> m the Poincare disk D so that each component of D — Ur is an ideal triangle in D. More explicitly, each ji is a bi-inifinite geodesic with well-defined endpoints on the circle S^ at infinity, the ji are pairwise disjoint and decompose D into triangles whose vertices lie at infinity, and this decomposition of D is furthermore locally-finite in D. We refer to a choice of edge ji together with a specification of orientation on it as a "distinguished oriented edge" or a "doe" for short, and we typically denote a tesselation r together with a doe by r1. As a further point of notation, we let r° C SIQ denote the collection of vertices at infinity of the triangles complementary to r, so r° is some countable dense subset of the circle at infinity. There is a standard classical tesselation r#, called the Farey tesselation and illustrated in Figure 1, defined by reflections about the sides of a fixed ideal triangle in D. Explicitly, we have labeled three of our vertices 0 = j , 1 = Y, oo = ~, and these determine a suitable corresponding triangle complementary to T* . We specify that the doe is the oriented edge running from Y to ^ (i.e., a diameter of D) to determine the standard Farey tesselation T' with doe. -2/3
-1/3
296
Robert Penner
As is well-known [Ra], iteratively labeling the ideal vertices r* of r* by , ^ i 4 £ ^ as illustrated in Figure 1 gives a bijective enumeration of the rationals. (Indeed, the problem of giving a one-to-one enumeration of the rationals was first solved in this manner by the mineralogist Farey without proofs, which were essentially immediately supplied by Cauchy.) Indeed, we say that §? "i? 5?—"i a r e the "first generation" of the enumeration, and each application of the addition law above increases the "generation" by one. For instance, ±^ and ± | are the generation two points. Thus, T® is canonically identified with the set of rational points in the circle at infinity which comes equipped with its Farey enumeration. Furthermore, the Farey enumeration of rf also determines a one-to-one enumeration of r* itself by Q — {+1,-1} as illustrated in Figure 1; namely, given a triangle t complementary to r*, one edge of £, call this edge e, separates the other two edges of t from the doe, and we label e by the Farey number of the vertex opposite e in t (letting 0 denote the doe of Farey by convention). We shall require this enumeration in §3 below. Now, if T' is any tesselation with doe, then there is a unique mapping fr' : r * -* T° defined iteratively as follows. The respective initial and terminal points of the doe of r^ (that is, the points j and ^) are mapped by fT' to the corresponding endpoints of the doe of r'. Next, map the vertex of the triangle to the right of the doe in r* to the vertex of the corresponding triangle to the right of the doe in r'; continue in this way a generation at a time to iteratively define the function / T /. Not only is fT> evidently a bijection between the dense sets T® and r°, but it is also order-preserving (for the counter-clockwise ordering on S^) by construction. It follows from elementary topology and one-dimensional dynamics that fTi interpolates a homeomorphism of 5 ^ which maps T* to r°; this homeomorphism of the circle is also denoted fT> and is called the characteristic mapping of r'. Formalizing part of the discussion above, we define 2
Tess' = {tesselations of D with doe}, Homeo+ = {orientation — preserving homeomorphisms of the circle S^}, so the assignment of characteristic mapping to a tesselation with doe induces a function Tess' —> Homeo+ T'
-> fr'
Conversely, given a homeomorphism / of the circle, we may define a corresponding tesselation f(rf) with doe as follows. Each edge 7 of the Farey tesselation is determined by a pair {rr, y} of points in the circle, there is an associated "image" geodesic f(j) in D spanned by {/(#), /()}, and we
Universal Ptolemy Group
297
take the "image" f(r') = {/(7) = 7 £ r . } , with doe given by the edge from /(y) to / ( ^ ) . It is straight-forward to check that f(rf) is actually a tesselation and that the assignment / — i > f(rr) is the inverse of the characteristic mapping. We have proved Theorem 1. The characteristic mapping r' — t > fTi induces a bisection between Tess1 and Homeo+.
There are immediately a myriad open questions, for instance, specify some family of orientation-preserving homeomorphisms of the circle and characterize the corresponding family of tesselations. Now, the Mobius group Mob (i.e., the group of orientation-preserving isometries of D, a group isomorphic to PSL(2, R)) acts on the left on both Tess' and Homeo+. We finally define the universal Teichmuller space Tess — Tess' IMob = Homeo+/Mdb, where the topology is induced as the quotient of the pull-back by the characteristic mapping of the compact-open topology on Homeo+. Since the Mobius group acts transitively on ordered triples of points in the circle, we may simply consider "normalized" representatives of Mobiusorbits of tesselations, namely, tesselations where the doe is the edge from j to ^ and the triangle to the right of it is the one spanned by j , y, ^; the corresponding "normalized" homeomorphisms of the circle evidently fix each of these points. In fact, it is easy to give global coordinates on Tess as follows. Given a tesselation r', let us specify an edge 7 of the Farey tesselation, so there is then a corresponding edge of r', namely the image 6 = fr'ii)- Insofar as T' is a tesselation, 5 separates two ideal triangles complementary to r, which together determine an ideal quadrilateral. One of the vertices of this quadrilateral, call it x, is distinguished in that 7 separates it from the triangle to the right of the doe, so we may enumerate the vertices starting from the distinguished one in counter-clockwise order around the circle as x,a,b,c. Recall that the cross ratio of these points is the value x —a
b— c
x —c
b—a
of x under the fractional linear transformation mapping a,b,c H-> 0,1,00, and this is a complete R_ -valued invariant of Mobius-orbits of ordered fourtuples of points in the circle.
298
Robert Penner
Assigning in this way a cross ratio of points in r° to each edge of Farey descends to a mapping E : Tess -> (R_)T* by Mobius invariance of cross ratios. Furthermore, since cross ratios are a complete invariant, we conclude that E is an injection. In fact, we have Theorem 2. Giving (R_)T* the product topology, E is an embedding onto a path connected open set, so Tess has naturally the structure of a Frechet space. In fact, it turns out that Tess also has a canonical formal symplectic structure as we shall amplify momentarily. Insofar as Tess has been identified with Homeo+/Mob, we can compare it with other such models, namely, with Bers' universal Teichmuller space HomeOqs/Mob, where Homeoqs denotes the space of quasi-symmetric homeomorphisms of the circle (that is, boundary values of quasi-conformal homeomorphisms of the disk D), and with the L2 universal Tecihmuller space Diff+/M6b, where Diff+ denotes the space of orientation-preserving diffeomorphisms of the circle (and the quotients are again by the left action of Mob). We therefore have the inclusions Diff+/Mbb -¥ HomeoqslMb'b -> Homeo+/Mob = Tess, so we are naturally generalizing the other well-known universal Teichmuller spaces; we mention that the spaces in this chain of inclusions are respectively a Hilbert, Banach, and Frechet space. In fact, these inclusions are also geometrically natural, for instance in the sense that our formal symplectic form on Tess pulls back under these inclusions to the Kirillov-Kostant twoform on Diff+/Mob. It follows that our formal symplectic form gives an honest (i.e., convergent) symplectic form on the subspace of all C*+e homeomorphisms of the circle. To complete our description of the right hand side of the universal square, we must still define our model Mod of a universal moduli space, and we shall be content here to describe this simply as a set. As before, we first define the set Mod1 to be the collection of all countable dense subsets of the circle together with the specification of three labeled points a, 6, c among the countable set. The Mobius group acts on the left on Mod', and we let Mod = Mod'/Mob denote the set of Mob orbits. As before, by taking the points a, 6, c to be Y? 1? ^? we may consider "normalized" countable dense subsets. The mapping r' \-+ r° which furthermore takes a, b, respectively, to the initial, terminal point of the doe and c to be the other vertex of the triangle
Universal Ptolemy Group
299
in T' to the right of the doe descends to a map Tess —> Mod,
which defines the morphism on the righthand side of the universal square. To see that this morphism is a surjection, fix some normalized countable dense set, and enumerate it xi,X2,x3,..., where we assume that £i,£2,#3 agree with y, y, ^. At each generation of the Farey enumeration, map the new vertices to the least index possible. Specifically, at a given generation of the Farey enumeration, the circle has been decomposed into afinitecollection of open intervals. In each interval, choose the Xi with i least among points lying in the interval and map the next generation of vertices of Farey to these points in each interval. This determines a tesselation r with r° the given countable dense set, as desired. In fact, the quotient topology on Mod is not Hausdorff. Indeed, let us take the rationals r^ and choose one irrational point s in the circle. Define a normalized sequence of tesselations r[ which agree with the Farey tesselation up to some generation z, include an extra vertex at s, and then continue exhausting all the rationals. Each of these tesselations has rf = T£ U {S} by construction yet the limit in i converges to the Farey tesselation, so the two normalized countable dense sets r° and r* U {s} cannot be separated. With a little more work, one similarly sees that the largest Hausdorff quotient of Mod is a singleton. §2. The Ptolemy Group We shall define the group G discussed in the introduction in three stages: groupoid, monoid, and finally group. To define the groupoid, suppose that r' is a tesselation with doe and that 7 € T is an edge of r. 7 separates two triangles in D — r, which together with 7 comprise an ideal quadrilateral of which 7 is a diagonal; let 7* denote the other diagonal of this quadrilateral. We may alter r to produce a new tesselation T7 = (rU{ 7 *})-{7}
as in Figure 2a. We say that r 7 arises from r by applying an "elementary move" along 7 G r. Notice that there is a canonical identification of r — {7} with r 7 — {7*}, so it makes sense to stipulate that if 7 is not the doe of r', then r 7 inherits the given doe of r ' as its doe to determine r 7 . On the other hand, if 7 is the doe of r ; , then we specify 7* as the distinguished edge of ^ taken with an orientation determined by the condition that 7, 7* (in this order) meet with a positive orientation in D. Thus, the elementary
300
Robert Penner
move on the doe has order four; see Figure 2b. For later use, two further sequences of moves are given in Figures 2c and 2d.
(a)
0-H-0-H-0 (b)
(c)
(d)
Figure 2 Define the groupoid Pt' as a category, where the objects are elements of Tess1 and a morphism from a' to rf is a finite sequence a1 = TQ, T{, . . . , r^ +1 = rf of tesselations with doe, where r/ +1 arises from r[ by an elementary move along some edge of r^, for i — 0, . . . , n . Notice that if there is a morphism between a1 and r', then <J° = r°. Passing to Mo6-orbits, we find a category, called the universal Ptolemy groupoid Pt, whose objects are given by elements of Tess and whose morphisms are given by Mo6-orbits of sequences of tesselations with doe differing by elementary moves as above. To pass to a monoid, observe that given any r' G Tess', the characteristic mapping establishes a bijection between the edges r* of Farey and the edges of r. To specify an elementary move along an edge of r', we might instead simply specify the corresponding edge of r^. Thus, if q £ r*, then we let q • r' denote the tesselation with doe gotten from r' by applying the
Universal Ptolemy Group
301
elementary move along the edge fT* (q) of r1 corresponding to q £ r* via the characteristic mapping. Define the free monoid M generated by r*, and inductively extend the action in the previous paragraph by setting ten ' • * 4Mi) * T' = qn • ( . . . • tf2 • tel • T') . . . ) , thereby denning an action of M on Tess'. Passing to Mo&-orbits, we find an action of M on Tess as well. This replacement of a groupoid by a monoid is a general categorical construction applicable whenever the morphisms from any object are a priori identified with some fixed set. To finally define the group, consider the sub-monoid K of M consisting of those words which act identically on r^, that is, whose corresponding groupoid elements begin and end at the Farey tesselation r^ with its fixed doe. For instance, the fourth power of the doe of Farey lies in if as in Figure 2b. The universal Ptolemy group is the quotient G = M/K, where we take the quotient of M generated by insertion and deletion of words in K. To see that G is actually a group, it remains only to verify that there are inverses, namely, that K is big enough to allow inverses in G. For instance, the third power of the doe corresponds to the inverse of the doe in G. In the remaining case of an edge other than the doe, return to the earlier notation of a quadrilateral with diagonals 7 and 7* as in Figure 2a; the word in M corresponding to the composition of the move on 7 with the move on 7* lies in K and provides the required inverse in G. This defines the group G, which we next identify with two other known groups. To this end, suppose that g £ G and consider applying the corresponding elementary moves to the Farey tesselation r^ with its doe to produce another tesselation g • r* with doe. There is a corresponding characteristic mapping fg = fg.T^ giving a representation G —> Homeo+ 9^
fg
of G which is faithful by definition (of the relations K). We mention here that there are also several other interesting essentially geometric representations of G (partly discussed in [P2;p. 180-181] and to be taken up further elsewhere). One can compute directly that in fact the homeomorphism fg actually lies in the subgroup PPSL(2, Z) of piecewise PSL(2, Z) homeomorphisms of the circle. Specifically, if A e PPS r L(2,Z), then there is a finite decomposition of the circle into intervals with rational endpoints, and on each such interval
302
Robert Penner
/, the function A restricts to an element A\j of PSL(2, Z). Of course, there are conditions on the restrictions A\j to guarantee that they combine to give a homeomorphism, and in fact, this homeomorphism is then automatically once continuously differentiate (cf. [P2; p. 206-207] and [P2;§3], where various properties of PPSL{2, Z) homeomorphisms are discussed). In a recent DEA memoire [II] and also in Imbert's contribution [12] to this volume, we find the following result: Theorem 3. [Imbert-Kontsevich] The assignment g \-t fg is an isomorphism between the groups G and PPSL(2,Z). In fact, there is yet another well-studied group (see [12] and the references therein for a survey) isomorphic to G = PPSL(2, Z), namely, the Thompson group Th defined to be the group of piecewise linear homeomorphisms of the circle where the "breakpoints" (in the piecewise structure) are required to be rational numbers of the form p/2n.
Figure 3 The connection between Th and our current discussion is as follows. Consider the dyadic tesselation r 2 defined a generation at a time as with Farey, but where at each generation, we take the (Euclidean) midpoint of the interval as the next vertex; see Figure 3. With the usual doe (from y to jjj) on r 2 , there is a characteristic mapping / : r'2 —> T£. This homeomorphism evidently conjugates Th to PPSX(2,Z) in Homeo+, and so we find that furthermore G = PPSL(2,Z) = Th. Using Theorem 3 and a known presentation of Th, [LS] have given a new
Universal Ptolemy Group
303
essentially geometric presentation of these groups as follows. Corollary 4. [Lochak-Schneps] The groups G, PPSL(2, Z) and Th admit a presentation with relations a\ /33, (a/3)5, [/?a/3, a2papa2], [fia/3, a2$a2pa$a201
a\
where [•, •] denotes commutator and a = the elementary move on the doe (3 — the first six consecutive frames of Figure 2c where • is the doe. In other words, the stated relations generate the sub-monoid K discussed above. To explain the presentation, observe that (3 just cyclically permutes the doe around the base triangle hence has order 3, while a as the move on the doe has order 4. The word a(3 generates the well-known pentagon relation and has order 5. The remaining two relations correspond respectively to the commutativity of elementary moves on disjoint but contiguous quadrilaterals, and quadrilaterals which are separated by one triangle (cf. Figure 2d). Another consequence of Theorem 3 is that the group G is simple (since Th is known to be so), and we therefore cannot simply take the profmite completion of G (there is none!) as a candidate completion to support the desired action of Galois. [LS] describes a certain extension of G by braid groups which does indeed support an action of Galois (where the pentagon relation in G corresponds to the pentagon relation in Galois). In the remaining sections of this paper, we briefly discuss other sensible completions of G which would seem suitable in various contexts but which are not yet known to support any action of Galois. §3. The Transitive Completion In analogy to the mapping class groups, we seek a completion of G which acts transitively on the fibers of the map Tess —y Mod. In fact, in [P2] we gave a a sort of completion (to be recalled below) of G, but it was a set-theoretic completion, not a group-theoretic one. We also proposed there a group-theoretic completion G, which was a "huge" group based on "nets of countable ordinals". We find here that the construction in [P2] is already essentially enough to give a suitable group-theoretic completion G of G based on a much smaller index set. We refer to G as the "transitive completion" since it has the
304
Robert Penner
required transitivity property. Furthermore, we go on to identify G with a natural group of homeomorphisms of the circle. To begin, we recall some material from [P2;§4.2]. We consider semiinfinite words w = - - - q2q\ where qi G r*, and we read these words from right to left as a corresponding sequence of elementary moves as before. Furthermore, using the identification of r* with Q — {+1, —1} discussed in §1, we regard it; as a sequence of rationals. We must impose two conditions, and the first one is: w = - - q2qi is stable if for each n > 1, there is some L > 1 so that if £ > L, then the Farey generation of the rational qi is at least n; that is, the generation of qt tends uniformly to infinity in £. Given a stable semi-infinite word w = • • • q2qi and r' G Tess1', recursively define r j + 1 = qj+i • rj, where T'Q = r', and let fj denote the characteristic mapping of rj, for j > 0. Let Pn C D denote the ideal polygon of 2 n + 1 sides whose vertices comprise the set P% of Farey points of generation at most n and let P* denote the set of geodesies in T* lying in Pn. Since w is stable, for each n > 1, there is some j n so that fj(e) = fjn(e) for each j > jn and each e G P*. Define
and let Qn C D denote the closed convex hull of Q^. By construction, Qn determines an ideal triangulation of Qn which agrees with the restriction of Tj to Qn for each j > j>n, and Qm C Qn for m1} = r°; that is, the finite sets {Qn} exhaust the countable set r°. We proved in [P2;§4] that stable and convergent semi-infinite words have a well-defined action on Tess' and Tess, and this action is transitive on the fibers of Tess -» Mod. This basic action and transitivity result will be simply assumed here. (Transitivity essentially follows from the Axiom of Choice and a finite transitivity result to the effect that finite sequences of elementary moves act transitively on triangulations with doe of a finite polygon). To define the completion G, first consider the monoid M consisting of all finite concatenations w = wn - - 'W2wi, where each wi is either a finite word in r* or a stable and convergent semi-infinite word in r*. In case vu is a
Universal Ptolemy Group
305
subword of w, where u is semi-infinite, we may think of the first letter of v (and notice that both finite and semi-infinite words have a well-defined first letter) as a limit ordinal for the sequence u. Notice that M is the smallest monoid containing both M and the set of stable convergent semi-infinite sequences. It follows from results already mentioned that there is a well-defined action of M on Tess' and Tess, and we as usual denote the action of w G M on T' G Tess' by w • r1'. Define
so if is a sub-monoid of M. Finally, define the transitive completion G of G to be the quotient G = M/K, which is defined to be the monoid of equivalence classes of elements of M, where the equivalence relation is generated by insertion and deletion of (finitely many) elements of K. Theorem 5. The transitive completion G is a group which acts transitively on the fibers of the map Tess -> Mod. Proof. To see that G is a group, it remains only to verify that K is large enough so that every class in M is invertible, and to this end, it suffices to show separately that the class of each element of G and of each stable and convergent semi-infinite word has an inverse. We have already argued above that elements of G have inverses relative to K C K, so suppose finally that w is a stable and convergent semi-infinite word. Since w • r* is a well-defined tesselation rf with doe and r° = r* by convergence, we may apply the transitivity result from [P2;§4] to produce another stable and convergent semi-infinite word u so that u • r' — r*; thus, the class of u is inverse to the class of w, as desired. Finally, transitivity of G follows again from [P2;§4]. q.e.d. It is worth pointing out that it follows from the proof the every equivalence class in G actually admits a representative w which is a single-letter word in M, that is, either w G G or w is a stable and convergent semi-infinite word (so, in a sense, [P2;§4] was the whole story). It is also worth pointing out that this representative may not always be the most desirable; for instance, if we wish to work equivariantly for a Fuchsian group F, then a useful representative w = wn • • • W2W1 takes each Wi to be a F-orbit of elementary moves on the universal cover. We shall discuss these possibilities further in the next section.
306
Robert Penner
A reasonable paradigm for the construction of G is to consider the monoid m consisting of all finite concatenations of words which are themselves either a finite sequence of natural numbers or a "stable" semi-infinite sequence • • • n2ni of naturals, where a sequence is "stable" if each natural number occurs in it only finitely often. We construct a sub-monoid k of m generated by n2 for each natural number n and all commutators. The quotient group g = rn/k is isomorphic to the operation of symmetric difference on the power set of the natural numbers. Insofar as G acts on Tess1, given w G G, we might consider the characteristic map fw — fw.ri of w - T^. We also define V = {feHomeo+:f(Q)=Q}, certainly a natural subgroup of Homeo+. Theorem 6. The assignment w — i >• fw establishes an isomorphism between the groups G and V. Proof. First consider an element w G G, so w is represented by wn • • • W2W1, where each Wi is either finite or stable and convergent semi-infinite. Convergence is used to verify that the characteristic map of each Wi indeed sendsto Q, hence functoriality of characteristic mappings finally proves that
fwev. On the other hand, suppose that / G V. There is a well-defined "image" /(r^) G Tess1 defined as before, so by transitivity of G, there is some word Wf with Wf - r* = f(rl). The assignment / 1—»• Wf induces the inverse homomorphism, as required. q.e.d. §4. Normalizers of Fuchsian Groups Though parts of what we discuss in this section apply in the more general setting of an arbitrary Fuchsian group, we shall restrict ourselves once and for all for simplicity to the case of a torsion-free finite-index subgroup F < PSX(2,Z). The surface Fr = D/F is a finite branched cover of the "modular curve" D/PSX(2, Z), which has the ramification points 0,1,00 of respective ramification indices 2,3,00, and the points of Fr lying over 00 G D/F5L(2,Z) are necessarily punctures of Fr since F < PSX(2,Z) is finite-index. In contrast, the points lying over 0,1 G D/P5L(2,Z) are smooth points of Fp since F is torsion-free. Furthermore, PSX(2, Z) leaves the Farey tesselation r* invariant (cf. [PI; §6]), so r* descends to an ideal triangulation Ap of Fp? that is, a decomposition of Fr m ^o ideal triangles. Indeed, the edges of Ap are naturally
Universal Ptolemy Group
307
identified with the set of F-orbits of edges of r*. More generally, if A is any ideal triangulation of Fr, then we may lift A to a tesselation A of D; F again acts on A, and the edges of A are naturally identified with the F-orbits of edges A. Moreover, for any ideal triangulation A of Fr, we have A0 = r°. An edge e of an ideal triangulation A of Fr is of one of the two following types. We may choose a lift e of e to D, and this lift is the common frontier of two triangles £1,^2 complementary to A in D; let q denote the quadrilateral spanned by these two triangles. It may be that £1 7^ £2 (so the projection is an embedding on the interior of q), and in this case, we say that e is locally separating. In the remaining case that t\ — ti-> q projects to a once-punctured monogon in Fr, and e is the geodesic connecting the puncture to the cusp on the boundary; in this case, we say that e is locally non-separating. Given a locally separating edge e of A, we may sensibly perform an elementary move as before on e in Fr to produce another triangulation of FT- In analogy to the construction of the universal Ptolemy group, let us specify some doe e of A in order to enumerate the edges of A by r*: Given a specified doe e G A, choose some lift e G A to get A' G Tessf, so the characteristic map r* -» A induces a surjection r* -> A. F-invariance of A shows that this enumeration is independent of the choice of e lying over e. (This enumeration depends only on the combinatorics of edges lying to the left/right of the doe in Fr.) In particular, the doe of Farey r* gives rise to a canonical doe of Ar, so every Fr comes equipped with a canonical ideal triangulation Ap with doe. We may specify, as before, a sequence of elementary moves on a triangulation A' with doe of Fr instead as a sequence of elements of r*. We must include also letters corresponding to elementary moves on locally nonseparating edges to achieve this a priori enumeration of A, and we simply think of these letters here and throughout as acting identically on triangulations with doe. Thus, the free monoid M on r* acts naturally on the set of triangulations with doe of Fp, and we may build the kernel KY corresponding to all sequences of moves which act identically on Ap to define the Ptolemy group Gp = M/KY of F. Lemma 7. GY is a group. Proof. Again we must simply prove the existence of inverses. To this end, recall from [Pl;Proposition 7.1] that finite sequences of elementary moves in Fp act transitively on ideal triangulations of Fp. Thus, given w G M, we may find u G M so that (uw) • Ap agrees with the tesselation (without doe) underlying Ap. We can finally find a word v G M in the letters a2 and (3
308
Robert Penner
of Corollary 4 above to arrange that (vuw) • (Ap) = Ap, so the class of vu is inverse to the class of w. q.e.d. In particular, if e € r* corresponds to a locally non-separating edge in Fp, then e lies in Kr, and if ei, e^ € r* lie over a common edge of Fr, then ei = e2 in Gr> In fact, we are more keenly interested here in another group, a certain extension of Gp by F to be discussed below. In case e\ and £2 are edges of A' lying over a locally separating edge e of A, then the elementary moves on e\ and £2 commute as in Figure 2d. Explicitly, we have hAe2)
ei
=
/e a (ei) e2
in G r ,
where, as before, /e;(e?) is the "image" of ej under the characteristic map /e, of ei-r^. Now, suppose that e is a locally separating edge in a triangulation A' of Fp with doe. Enumerate the edges e^, i > 1, of A lying over e regarded as elements of r*, and consider the formal semi-infinite word
Our immediate goal is to show that this is actually a stable and convergent semi-infinite word. Assuming this, the order ei,e2> • • • on the edges lying over e is immaterial according to the commutativity discussed in the previous paragraph; that is, different orders give rise to the same action on tesselations with doe. On the other hand, it is convenient in the sequel to actually specify a particular order as follows. Choose a fundamental domain E G D for F which is a connected ideal polygon triangulated by A, where we may assume that the doe lies in E and some fixed lift e\ of e lies in the interior of E. We say that E itself is "first generation" and also that the vertices of E are "first generation". Of course, the frontier edges of E are identified in pairs by F, and across each frontier edge of E there is therefore some other contiguous fundamental domain; these are the "second generation" fundamental domain, and their vertices taken together form the "second generation" of points in the circle. One continues in this way to associate a generation to each ideal point of A and to each F-orbit of E. Let us choose an enumeration E = E\, E2,... of the F-orbits of E so that the generation of 2?j+i is at least as great as the generation of Ei, for each i > 1. In the interior of each E^ there is a unique lift e^, and we define the T-move to be the semi-infinite word Wq='-
fester (e4) /e2ei (h) Hx (h) ^1,
Universal Ptolemy Group
309
where gGr* corresponds to e G A' as above. Lemma 8. For any edge gGr* and any T, w^ is a stable and convergent semi-infinite word. Proof. If the edge q corresponds to a locally non-separating edge, then w^ acts identically and is thus stable and convergent by convention. In case the edge q corresponds to a locally separating edge, then adopt the notation of the previous paragraph. Stability is clear just from the existence of the ordering above on fundamental domains by non-decreasing generation. Since e; lies in the interior of Ei by construction, there is always at least one triangle in Ei with e^ in its frontier and separated from the doe by e^. Thus, the set of points in the circle of generation at most n contains the set of Farey points of generation at most n by induction, so the word is convergent as well. q.e.d. It follows from [P2;§4] and Lemma 8 that for any edge q G r* and any Finvariant tesselation r' with doe of D, there is a well-defined corresponding tesselation with doe wvq • r', and we furthermore have its characteristic map fq,T"
Consider the normalizer NHomeo+(r) = {/ G Borneo^ : / F / " 1 - F} of F in Homeo+. It follows from the defining equation / F / " 1 = F that / necessarily maps the set of parabolic fixed points of F to itself, that is, / necessarily satisfies /(Q) = Q. It follows that / G V, so iVy(F) = NHomeo+ (F), and we let =
NHorneo+(T)
denote the normalizer of F in either group. Lemma 9. For any edge q G r*? any T, and any F-invariant tesselation r' with doe, we have f^T, G N(T). Proof. To see that /£ T , normalizes F, consider two copies of the universal cover of Fp, where F acts on each copy, but we take r' in the first copy and Wq - r' in the second copy. /£ T , induces a topological mapping on the universal covers which sends r' to w^ • r', and so / T r , conjugates F to itself. q.e.d. We next show that conversely any / G iV(F) can be expressed as the characteristic map of a finite composition wn • • • w2wi, where each wi is a
310
Robert Penner
F-move. The proof again depends upon [PI; Proposition 7.1] as in Lemma 7. One may think of this result as a kind of structure theorem for normalizers of Fuchsian groups. At the same time, these results together may be seen as a sort of converse to Nielsen's famous theorem that a homeomorphism of a surface Fr lifts to a mapping on D which extends continuously to a homeomorphism of S^ normalizing F. Given a F-move w^, where g 6 r*, we may regard the stable and convergent semi-infinite word w^ as acting on r[ itself. Letting M as usual denote the free monoid over T*, an element w — qn • • • <72
Define KY = {w E M : w • r^ = r*} and the F covering Ptolemy group Gr =
MjKv.
One sees that Gr is indeed a group arguing just as in the proof of Lemma 7. Furthermore, Gr acts on the set of F-invariant tesselations with doe as before, and we have an exact sequence 1 -> F -> G r -> G r -> 1. If w € Gr, then we let fw denote the characteristic map of w • r*. Theorem 10. The assignment w — i »• fw establishes an isomorphism between the groups Gp Proof. Using Lemma 9, the mapping is evidently an injective homomorphism. To see that it is surjective, choose a fundamental domain for F, and take its "image" under / as before, for any fixed / in N(T). This is again a fundamental domain for F, as one can check, and it comes equipped with an induced triangulation with doe. Now, as in the proof of Lemma 7, there is a sequence of F-moves taking Farey to its "image" under / and mapping the doe correctly, so surjectivity follows. Finally, use that / normalizes F to prove that this construction is independent of the choice of fundamental domain. q.e.d. Relating the constructions of this section to the transitive completion G discussed in §3, we have Corollary 11. For any V, the assignment M -> M induced by q H-> W^ descends to a canonical injection N(T) = Gr —> G.
Universal Ptolemy Group
311
As to completions, there is a homomorphism N(Tt) = GTl -> Gr2 = N(T2) whenever F2 < I \ defined as follows. Given q G r^ and any Fi-invariant tesselation rf with doe, there are induced triangulations with doe A^ and A'2, respectively, of Fr1 and Fp2. Let e G Ai correspond to q G r* and consider the full pre-image of e under the projection Fp2 —> -Frr> choose labels <# G r*, for i = 1,..., n corresponding to this family of arcs in Fr 2 , and define Q*->fqn-i -9291 fan) " • /
312
Robert Penner
[P2] —, Universal constructions in Teichmiiller theory, Adv. Math. 98 (1993), 143-215. [Ra] H. Rademacher, Lectures on Elementary Number Theory, Blaisdell, New York, 1964. [Su] D. P. Sullivan, "Relating the universalities of Milnor-Thurston, Feigenbaum and Ahlfors-Bers", in the Milnor Festschrift Topological Methods in Modern Mathematics, eds. L. Goldberg and A. Phillips, Publish or Perish (1993), 543-563. Departments of Mathematics and Physics/Astronomy University of Southern California Los Angeles, CA 90089
Sur l'isomorphisme du groupe de Richard Thompson avec le groupe de Ptolemee Michel Imbert
§0. Introduction. Le but de cet article est de donner une demonstration de Fobservation suivante de M. Kontsevitch: le groupe universel de Ptolemee G defini par R. Penner dans [Pel] est isomorphe au groupe de R. Thompson PI^S 1 ). Le groupe PL 2 (S 1 ) a ete decouvert par R.Thompson dans un contexte algebrique et ce sont ses notes manuscrites non publiees qui constituent la reference de base. Le groupe G est lui un ingredient important dans la theorie de Penner des espaces universels de Teichmuller, il possede par exemple des completions interessantes (voir Particle de Penner dans ce volume); d'autre part P. Lochak et L. Schneps ont etendu le groupoide universel de Ptolemee (a partir duquel R. Penner a construit G) en un nouveau groupoide qui possede une completion profinie sur laquelle le groupe de Galois absolu agit en tant que groupe d'automorphisnaes (voir leur article dans ce volume). Nous suivrons le plan suivant: dans une premiere partie nous rappellerons la definition de PL2(51) et quelques formes equivalentes de ce groupe, ce qui nous permettra de demontrer dans une deuxieme partie qu'il est isomorphe a G, et ceci de deux manieres. Dans tout Particle nous noterons Homeo+ le groupe des homeomorphismes du cercle preservant l'orientation, et Q(2) l'ensemble des nombres dyadiques de Tintervalle [0,1], ie les nombres de la forme ^ compris entre 0 et 1 avec q et n entiers naturels. Enfin nous identifierons souvent S1 avec R U {oo} et avec [0,1] ou 0 et 1 sont identifies. Je suis redevable a M. Kontsevitch pour la communication de ce resultat, et je remercie V. Sergiescu, par lequel j'en ai eu connaissance, de m'avoir encourage a le demontrer.
314
Michel Imbert
§1. Le groupe de R.Thompson §1.1. Definition, survey. Definition 1.1. PL^S 1 ) est le sous-groupe de Horrieo+ forme des elements qui sont affines par morceaux et tels que: 1. les points de coupure appartiennent a Q(2). 2. Pimage de 0=1 appartient a Q(2). 3. les derives de chaque application affine constituant un element du groupe sont des puissances de 2. Ce groupe, defini en 1965 par R. Thompson, ainsi que F le groupe des homeomorphismes de [0,1] lineaires par morceaux possedant les proprietes I et 3, fut introduit au debut par son auteur pour construire des groupes de presentation finie avec un probleme de mots non resoluble (voir [M,Th]). II s'est avere ensuite posseder des proprietes remarquables et il est apparu dans de nombreux contextes mathematiques. Voici une liste non exhaustive de ces proprietes: PL2(S'1) est un groupe infini, simple, de presentation finie. C'est un resultat de R. Thompson qui est demontre dans [CFP], article ou Ton trouvera les proprietes basiques de ce groupe. W. Thurston a trouve une interpretation de PL^S 1 ) liee aux structures projectives entieres par morceaux (voir plus bas). Dans [GhS] les auteurs classifient les actions de PILES'1) sur le cercle suivant la dynamique topologique de cette action, et donnent une description de ses representations dans DifFr(51) pour r> 2; d'autre part ils donnent une serie de resultats sur rhomologie de ce groupe (par exemple Hn(PL2(S1), Z) est de type fini pour tout n) ainsi que diverses relations avec les classes caracteristiques. Dans les annees 80, F est egalement apparu en theorie de l'homotopie et en homologie des groupes. En particulier, K.S. Brown et R. Geoghegan ont montre que les groupes F et PL^S 1 ) sont de type FPQQ (voir par exemple [Br]). §1.2. Le groupe modulaire par morceaux. Definition 1.2. PPSL2(Z) est le groupe des homeomorphismes preservant l'orientation de RU {oo} qui sont PSL^Z) par morceaux, avec des points de coupure rationnels. Theoreme 1.1. PPSL2(Z) est isomorphe a PL2(S'1).
Groupe de Thompson
315
Esquisse de la demonstration. Elle repose sur une bijection canonique que nous noterons i entre Qu {00} et Q(2) et qui est construite ainsi. On envoie d'abord 0/1 sur 0 = 1, 1/1 sur 1/4, 00 = 1/0 sur 1/2 et - 1 / 1 sur 3/4. Pour une fraction reduite a/b, appelons max(|a|, |6|) l'ordre de a/b. Alors si q € Q + est d'ordre n, q s'ecrit | ^ ou a/b et c/d sont d'ordre au plus n - 1 (voir [Ch] pages 6 a 8). Soit d\ — i(a/b) et d
est un isomorphisme de groupe.
0
§1.3. Couples d'arbres. Les elements de P I ^ S 1 ) peuvent etre parametres par des couples d'arbres binaires ordonnes munis d'une permutation cyclique des feuilles. Pour ceci la reference est [Br]. Definition 1.3. Un arbre binaire ordonne est un arbre S tel que: 1. S ait un sommet distingue ^0 (appele la racine) tel que si S n'est pas reduit a v0 alors v0 est de valence deux. 2. si v est un sommet de S de valence >2, alors il y a exactement deux aretes ev^\ et ev>r, appelees respectivement arete gauche et arete droite, qui contiennent v et qui ne sont pas contenues dans le chemin d'aretes allant de ^0 au. Les sommets de valence un sont appeles des feuilles et peuvent done etre ordonnes de gauche a droite. Definition 1.4. Un intervalle standard dyadique (i.s.d) est un intervalle ferine inclus dans [0,1] de la forme [^-,^r] ou a et n sont des entiers positifs ou nuls. Une partition de [0,1] est dyadique si et seulement si les intervalles de la partition sont des i.s.d. II existe un arbre infini dyadique note ^4^ dont les sommets sont les i.s.d, et une arete est une paire (/, J) d'i.s.d telle que si J est la moitie gauche (resp. droite) de / alors (7, J) est une arete gauche (resp. droite). Nous appelons ici un arbuste un arbre binaire ordonne fini identifie a un sousarbre binaire ordonne fini de ^loo (i.e. un sous-arbre fini de racine v0 dont les aretes gauches (resp. droites) sont des aretes gauches (resp. droites)
316
Michel Imbert
de Aoo). II y a une bijection canonique entre les arbustes et les partitions dyadiques de [0,1], les feuilles correspondants aux intervalles de la partition. Exemple.
Definition 1.5. Un diagramme arboricole est une paire ordonnee d'arbustes (iZ, S) qui ont le meme nombre de feuilles, munie d'une permutation cyclique des feuilles, cr, modulo la relation d'equivalence engendree de la maniere suivante. Soit (R, 5, a) un tel diagramme, / la n-ieme feuille de R et a(I) la feuille correspondante de S. Si on rajoute a R n'importe quel arbre binaire B ordonne a la racine / et si on rajoute a S le meme arbre a la racine cr(/), alors on dira que (i?,5,a) est equivalent a(J?UB,5uB,a') ou a' se deduit de a et de l'identite sur B. Si / £ PL2(5 1 ), alors il existe des partitions dyadiques P et Q telles que / soit affine sur les intervalles de P, et les envoient sur les intervalles de Q (voir lemme 2.2 de [CFP]). On peut done associer a / un diagramme arboricole {R, 5, a) ou R est l'arbuste associe a P et 5 celui qui est associe a Q. Si /(0) = 0 alors a=id. Notons DA l'ensemble des diagrammes arboricoles. Si (Q,R,a) EDA et (/?, T, r) €DA correspondent respectivement a / et g alors (Q,T, rcr) correspond a g o f. Theoreme 1.2. DA est un groupe isomorphe a PILES'1). On en trouve la preuve dans [CFP], ou les auteurs donnent aussi une presentation de DA en terme de trois generateurs et six relations, trouvee par R. Thompson (voir la figure suivante ou ces trois generateurs sont exprimes dans PPSL2 (Z), P I ^ S 1 ) et DA ).
Groupe de Thompson
PPSL
PL (S 1 )
(Z)
z A
1/2
z+1 1/4 0
id z
B
"!
*' Z " z
sur [0/1,1/0] sur [1/0,-1/1]
317
J
Ml 3/4
DA
CM CT = id
1
1 3/4 5/8 1/2
sur [-1/1,-1/2]
/
z sur [-1/2,0/1]
0
1/2
CJ = id
3/4 7/8 1
1 -1
3/4
1+z
1/2
\
2
3
2
J
C
0
1/2
3/4
I
CJ=(132)
§1.4. Couples de polygones triangules. Le groupe DA est isomorphe a un groupe que nous noterons DP qui consiste en des couples de polygones triangules ordonnes munis d'une permutation. II s'agit d'une construction de dualite classique, voir par exemple [STT]. Soit done R un arbuste. A chaque sommet on fait correspondre un triangle; en particulier on fait correspondre a la racine de l'arbuste v0 un triangle de base To. A une arete de R entre deux sommets on fait correspondre une instruction de collage des deux triangles correspondants le long d'une arete. On forme ainsi un polygone triangule P(i?). Grace a la propriete 2 de R, on observe que si T est un triangle de P(R) de valence >2, alors il y a exactement deux triangles ST,I et Sr,r adjacents a T et non contenus dans le chemin de triangles allant de To a T. Les triangles de valence un (i.e. avec deux cotes libres) correspondent aux feuilles de l'arbuste; ils sont dits externes. Ils sont ordonnes de gauche a droite.
318
Michel Imbert
Definition 1.6. DP est l'ensemble des diagrammes polygonaux constitues de couples de polygones triangules ordonnes avec le meme nombre de triangles externes, munis d'une permutation de ces triangles externes, et modulo une relation d'equivalence analogue a celle definie pour les diagrammes arboricoles. On obtient sans difficulty le theoreme suivant: Theoreme 1.3. DP est un groupe isomorphe a DA. Un fait important pour l'isomorphisme entre PL2(51 ) et le groupe universel de Ptolemee est le resultat suivant de transitivite : etant donnes deux polygones triangules a n cotes, il existe une serie finie de mouvements element air es permettant de passer de l'un a l'autre, ou un mouvement elementaire est un changement de diagonale a l'interieur d'un quadrilatere (voir [STT] et lemme 4.4 de [Pel]). §2. Le groupe universel de Ptolemee. §2.1. Tesselations. Le paragraphe 2.1 reprend quelques notions des deux articles de R. Penner cites en references. Definition 2.1. Une tesselation r est une collection denombrable et localement finie de geodesiques du demi-plan de Poincare % dont les regions complementaires sont des triangles ideaux de sommets appartenants a RU {oo}. r(°) denotera les sommets de r. Us sont denses dans Ru{oo}. r^ designe l'ensemble des triangles complementaires de r dans %. II faut considerer des couples (r, e) ou r est une tesselation et e est une geodesique distinguee des autres que Ton oriente. L'ensemble de ces couples est note TESS' par R. Penner. L'avantage de ces couples reside dans le fait qu'ils sont combinatoirement rigides, ie : si (T;, e^) pour i = 1,2 sont deux tels couples et x», yi sont les sommets de e^, alors il existe une unique bijection /
: r
~~^ r 2
i
Xi !->> X2
y\ ^ 2 / 2
et telle que si (x,y,z) engendre un triangle de r{ ^ alors (f(x),f(y),f(z)) engendre un triangle de r^ . / respecte l'ordre et determine une application caracteristique F : T\ —y T2
Groupe de Thompson
319
Definition 2.2. L'espace universel de Teichmiiller, TESS, du a R. Penner est defini comme les classes d'equivalence des orbites des couples (r, e) de TESS' modulo Faction de PSL2(R). II existe une tesselation particuliere, dite de Farey, qui joue un role important dans la theorie de Penner. On considere le triangle geodesique de sommets 0, 1, oo. La tesselation de Farey consiste en les images de ce triangle par PSL2(Z) . Le groupe modulaire est alors le sous-groupe de PSL2(M) qui preserve cette tesselation. On note r* cette tesselation, et son arete orientee notee e0 est celle qui va de 0 vers oo. Ce qui est remarquable, c'est qu'il y a une bijection entre T;' et Qu{oo} : si T G r* possede deux sommets assignes respectivement a a/b et c/d alors le troisieme sommet est assigne a |q^. Soit maintenant Q7 =\{—1,1}. Alors on a la bijection suivante:
e0
ou eq est definie ainsi: soit e E r* (e / e0 ), U la composante connexe de % \ {e} qui ne contient pas eo, et T l'unique triangle de r* DU ayant e dans sa frontiere. Un des sommets de T n'appartient pas a e; il est assigne a un rationnel q, et alors e = eq. Voir la figure suivante represent ant la tesselation de Farey.
320
Michel Imbert
Lemme 2.1. Homeo+ /PSL2(M) et TESS sont des espaces topologiques homeomorphes. Esquisse de la demonstration. Si g G Homeo+, on lui associe la tesselation (g(e) : e £ r*) ou si x et y sont les sommets de e alors #(e) est l'unique geodesique joignant g(x) hg(y), munie de l'arete orientee distinguee joignant a. #{°°}- D'autre part si (r, e) ETESS', on lui associe l'unique element de Homeo+ qui prolonge / : r*
—» r^0^.
<^
§2.2. Le groupe universel de Ptolemee et l'isomorphisme remarquable. Soit (ri,e) ETESS' et a une arete de TI; soit a' l'autre diagonale du quadrilatere forme par les deux triangles ayant a dans leur frontiere; enfin soit F : r* —>> ri l'application caracteristique et a = F(eqi). Alors r 2 = (ri\(a)) U (a;) est une autre tesselation. On ecrit r 2 = q\ • T\ et on dit que Ton a obtenu r2 par un gi-mouvement. II faut remarquer que r{ ' = r^0 . D'autre part on possede une bijection * i : n -> r2 a h-> a',
et qui vaut l'identite ailleurs. Si l'arete orientee distinguee de T\ est differente de a, alors elle sera aussi celle de r2; sinon a' devient l'arete distinguee, et on l'oriente de telle fagon que {a, a1) soit une base directe du plan. Soit M le monoide libre forme sur (Q/ et soit une action a gauche de M sur TESS donnee par: M x TESS -> TESS (Qi'"Qn = m, [ri]) ^ [m • n ] ou m-ri
= qi(- • • (gn_i(9n • 7i)) • • •)•
Ensuite, si m = ^i • • • q\ et n — q[ • • • ^ € M et si on pose 3>m = $ i o ^>2 • • • o 3>j etn = $i o $2 . •. o <£p, alors m et n sont dits equivalents si et seulement si pour tout [r] 6TESS, on a [ra-r] = [n-r] et les correspondances induites $ m et <£n sont identiques. La classe d'equivalence d e m G M est notee [777,]. Lemme 2.2. G, Vensemble des classes d'equivalence de M, muni de Voperation [m][n] = [mn] est un groupe, dit universel de Ptolemee, qui agit a gauche sur TESS. On trouve la demonstration dans [Pel].
Groupe de Thompson
321
Voici maintenant le theoreme qui motive cet article, initialement demontre dans [Im]. Theoreme 2.3. G est isomorphe a PPSL2(Z), et par consequent a PL^S 1 ). Demonstration. Demontrons d'abord l'existence d'un homomorphisme injectif de groupes entre G et PPSL 2(Z). Dans [Pel], R. Penner donne une representation fidele de G dans l'ensemble des applications de r* dans r*. J'ai remarque dans [Im] qu'en passant a l'ensemble des sommets ri ' cela fournissait un plongement de G dans Homeo+. Enfin dans son article dans ce volume, R. Penner resume cela ainsi: soit g G G, et fg : T* ' —>• {g-r*)^ = r* 1'application correspondante. Elle se prolonge de maniere unique en un element de Homeo+ qui preserve l'ensemble des nombres rationnels et que Ton note encore par fg. Alors h : G —> Homeo+ 9^
f9
est une representation fidele de G. Je demontre alors que l'image de G est incluse dans le sous-groupe PPSL2(Z) de Homeo+. Soit done q G Q^, qu'on suppose strictement positif, et Q le quadrilatere forme par les deux triangles possedant eq dans leur frontiere. II y a une unique geodesique eq> appartenant a la frontiere de Q qui est dans la composante de % \ Q contenant eo- Si q — c/cf est strictement superieur h q' = b/bf, alors on voit facilement que les sommets de Q sont dans l'ordre a/a', b/b\ c/d, d/d1 avec c = b + d,d = V + d' et b = a + d,V = a1 + d'. Dans la composante de % \ Q qui contient eo, y compris sur e^, on a Fq =id. Dans les trois autres composantes de 1-L\Q, Fq est une bijection respect ant la structure triangulaire d'une partie infinie de r* dans une autre partie infinie de r*. Comme Aut(r*) est le groupe PSL 2(Z), sur chacune de ces trois composantes, y compris sur leur frontiere avec Q, Fq est egal a un element de PSL2(Z). On en deduit que fq est une bijection de R U {CXD} qui se comporte comme un element de PSL^Z) sur chacun des intervalles [a/a',6/6'], [b/b',q], [q,d/d% [d/d',a/a% i.e. fq e PPSL2 (Z). On procede de maniere analogue si q est strictement inferieur a q'. Enfin si q est negatif la demonstration est egalement analogue. Pour prouver que notre homomorphisme est en fait un isomorphisme de groupes, il sufRt de trouver h' : PPSL2(Z) —> G tel que h o h' =id sur PPSL2(Z). Les elements de PPSL2(Z) sans points de coupure sont ceux de PSL2(Z); par contre il n'existe pas d'elements avec un ou deux points de coupure. Soit g e PPSL2(Z) avec au moins trois points de coupure. On lui associe r isi~1) — (^" 1 (e),e G r*). Soient r i , r 2 , . . . ,r p (p > 3) les p points de
322
Michel Imbert
coupure de g~l, et posons Si = g~l(ri) et n = max(ordre(si)) pour i G { 1 , . . . ,p}; soit Pn le poly gone geodesique regulier ouvert dont les sommets sont les rationnels d'ordre au plus n. (L'ordre d'une fraction reduite a/b est egal a max(|a|, |6|).) Les aretes frontieres de Pn et exterieures sont retrouvees intactes dans r{g~x) car Aut(r*) = PSL2 (Z). Par contre, les aretes situees a l'interieur de Pn peuvent etre remplacees par de nouvelles aretes. Grace au lemme 4.4 de [Pel] il existe m(g) G M tel que m(g) • r* = rig-1). Ondefinit ti{g) = [m(g)}. Si g G PSL2(Z), on ecrit g = (goh)oh,-1, goh et h~x possedent alors quatre points de coupure si h en possede quatre, et on definit h'(g) = [m(g o /i)m(ft~1)]. Soit maintenant g G PPSL 2 (Z). Alors hoh'(g) = / , l'unique bijection de rigidite qui envoie l'arete orientee distinguee de r(g~~l) sur eo et les triangles de r(g~1)^ sur ceux de ri . Or par construction g possede ces proprietes done h o hf =id. Dans la pratique, on peut calculer directement fq a partir de q; on calcule ainsi que: sur [0/1,1/1] sur [1/1,1/0] sur [1/0,-1/1] sur [-1/1,0/1]. h([0])2 vaut - l/z.
{
id
sur [1/0,0/1]
z-1
sur [2/1,1/0].
En utilisant ce theoreme et la presentation de PL 2 (5X ) deja citee au §1.3, P. Lochak et L. Schneps ont trouve une presentation remarquable de G en termes de deux generateurs et cinq relations (voir leur article dans ce volume). Ces deux generateurs sont [0] et [0][2/l][0]3[2/l][0] qui correspond au generateur C de la presentation precitee. Corollaire 2.4. L'homeomorphisme entre Homeo+ \ PSL2(R) et TESS est equivariant pour les actions respectives de PPSL2 (Z) et G. De plus PPSL2(Z) (et done G) agit sur Vespace universel de Bers Homeoqs\PSL2(R) oil HomeOqs est le sous-groupe de Homeo+ forme des elements qui sont quasi-symetriques. Demonstration. On a les actions suivantes: PPSL2(Z) x Borneo+ \ PSL2(R) -> Homeo+ \ PSL2(R)
Groupe de Thompson
323
ou [h] represente une classe a gauche; et G x TESS -> TESS
Comme, d'apres la preuve du theoreme 2.3, on a r{hog~x) = m(g)-T(h), on a bien l'equi variance. La deuxieme partie du corollaire resulte du fait suivant: les elements de PPSL2(Z) sont de classe C1 done quasi-symetriques. 0 On peut voir dans [Pel], a la fin du §1, l'interet de la derniere action du corollaire. Nous allons maintenant donner une deuxieme demonstration du theoreme 2.3, suggeree par V. Sergiescu, en demontrant que G est isomorphe au groupe DP forme de couples de polygones. Demonstration. Soit gGQ' strictement positif; on considere le -mouvement sur r*. Alors on a deja remarque qu'il existe n > 2 tel qu'il n'y ait rien de change a l'exterieur du polygone ouvert P n . Soit In le polygone triangule forme des geodesiques de r* reliant les rationnels positifs d'ordre inferieur ou egal a n; et soit Sn le polygone triangule forme des geodesiques de r* reliant les rationnels negatifs d'ordre inferieur ou egal a n. On associe a q Pelement (i?n , Q n , id) de DP, ou Rn et Qn sont construits ainsi : soit T un triangle de base, a et b deux de ses aretes; on obtient Rn en collant T avec In (resp Sn) grace a une instruction de collage de a avec l'arete (0/1,1/0) (resp. de b avec l'arete (0/1,1/0)); enfin on obtient Qn en remplagant In par q- In, le nouveau polygone triangule obtenu apres le g-mouvement. On procede de meme pour q strictement negatif, en remplagant Sn par q • Sn. Enfin si q = 0, on lui associe l'element (i?2,i?2, (1234)). On en deduit une application de M dans DP, et par definition de G et de son action sur les tesselations on a done construit un homomorphisme inject if de groupes entre G et DP. La surjectivite de cet homomorphisme decoule encore du resultat de transitivite finie cite a la fin du §1.4. Soit (P, Q, a) un couple de polygones triangules ordonnes et une permutation de ses triangles externes; alors en inject ant P dans un certain Rn en identifiant le triangle de base avec le triangle T, on sait que Q s'injecte dans le meme polygone que P mais avec une triangulation differente, ce qui permet d'appliquer le lemme de transitivite finie, et en repassant a la tesselation de Farey de voir que (P, Q, a) est 1'image d'un element de G. 0 Remarque. Par definition l'arbre de Farey consiste en deux exemplaires de l'arbre infini dyadique defini au §1.3 relies par une arete joignant les racines. On peut le construire egalement comme dual a la tesselation de Farey. Alors on demontre facilement que le groupe universel de Ptolemee G
324
Michel Imbert
est isomorphe au groupe des germes d'automorphismes a Tinfini de Tarbre de Farey. Cette action a 1'infini sur l'arbre de Farey est utilise dans [GrS] pour construire une extension acyclique du groupe de tresses a partir du groupe F mentionne au debut du §1.1. References [Br] [CFP]
[Ch] [GhS] [GrS] [Im] [M,Th]
[Pel] [STT]
K.S Brown, Finiteness properties of groups, J. Pure App. Alg. 44 (1987), 45-75. J.W. Cannon, W.J. Floyd, W.R. Parry, Notes on Richard Thompson's groups, Preprint (University of Minnesota, 1994), a paraitre dans Ens. Math. K. Chandrasekharan, Introduction to analytic number theory, Springer Verlag (1968). E. Ghys, V. Sergiescu, Sur un groupe remarquable de difTeomorphismes du cercle, Comm. Math. Helv. 62 (1987), 185-239. P. Greenberg, V. Sergiescu, An acyclic extension of the braid group, Comm. Math. Helv. 66 (1991), 109-138. M. Imbert, Constructions universelles dans la theorie des espaces de Teichmiiller, Memoire de DEA, Grenoble (1994). R. McKenzie, R.J. Thompson, An elementary construction of unsolvable word problems in group theory, Word problems, W. W. Boone, F.B Cannonito and R.C Lyndon, eds., Studies in Logic and the Foundation of Mathematics, Vol. 71, Amsterdam (1973), 457478. R. Penner, Universal constructions in Teichmiiller theory, Adv. in Math. 98 (1993), 143-215. D. Sleator, R. Tarjan, W. Thurston, Rotation distance, triangulations, and hyperbolic geometry, J. of the Amer. Math. Soc. (1988), 647-681.
The universal Ptolemy-Teichmiiller groupoid Pierre Lochak and Leila Schneps
Abstract We define the universal Ptolemy-Teichmiiller groupoid, a generalization of Penner's universal Ptolemy groupoid, on which the Grothendieck-Teichmiiller group - and thus also the absolute Galois group - acts naturally as automorphism group. The essential new ingredient added to the definition of the universal Ptolemy groupoid is the profinite local group of pure ribbon braids of each tesselation.
§0. Introduction The goal of this article is to give a completion (by braids) of Penner's Ptolemy group G such that there is a natural action of the GrothendieckTeichmiiller group (and a fortiori, the absolute Galois group Gal(Q/Q)) on it. This work was motivated on the one hand by the deep relation of the Ptolemy group - shown to be isomorphic to Richard Thompson's group and to the group of piecewise PSL2(Z)-transformations of the circle - and mapping class groups and the geometry of moduli spaces in general, most visibly in genus zero, and on the other by the presence of the remarkable pentagonal relation, stimulating the natural reflex of the authors to associate every pentagon appearing in nature to that of the Grothendieck-Teichmiiller group GT. The difficulties in defining a GT-action on G were the following. Firstly, the profinite version of GT which interests us (mainly by virtue of the fact that it contains the Galois group) acts on profinite groups, whereas via its isomorphism with Richard Thompson's group, G is known to be simple, and therefore its profinite completion is trivial. Furthermore, G contains no braids and GT naturally seems to introduce them into every situation where it appears. The undertaking therefore framed itself as follows: instead of restricting attention to G, is it possible to extend G by braids, in such a way that it is possible to define a GT-action on a profinite version of the extension, in such a way that the pentagonal relation of G reflects that of GT? The answer turned out to be nearly yes, namely in order to succeed
326
Pierre Lochak and Leila Schneps
it was necessary to use not the group, but the groupoid interpretation of G, in which its elements are considered as morphisms between marked tesselations (this groupoid is known as the Ptolemy groupoid), and to relax the condition of the Ptolemy groupoid stating that the group of morphisms from any marked tesselation to itself is trivial to a condition stating that the group of morphisms from any marked tesselation to itself is isomorphic to a certain ribbon braid group. It is the profinite completion of the ensuing braid-groupoid which admits a GT-action. In §1, we give the definition and presentation of the Ptolemy group, and its interpretation as a groupoid whose objects are marked tesselations of the Poincare disk. In §2, we recall the definitions and important properties of braid and mapping class groups, and their generalizations to ribbon braid and mapping class groups, which will be the braid groups used to extend the Ptolemy groupoid to the Ptolemy-Teichmiiller groupoid. §3 is devoted to the actual construction of the Ptolemy-Teichmiiller groupoid Poo, with a "physical" interpretation of the new, added groups of non-trivial morphisms from a tesselation to itself via braids of ribbons viewed as hanging from the intervals; at the end of the section we define the profinite completion of the groupoid 'Poo by simply taking the profinite completions of each of the local groups, while preserving the set of objects (i.e. marked tesselations) and the basic (Ptolemy) morphisms from one to another. §4 contains the main theorem of the article (theorem 4) explicitly a GT-action on the universal profinite Ptolemy-Teichmiiller groupoid 'PQO ; the role of the two pentagons appears in lemma 5. Finally, in §5 we give a very brief discussion of the relation between the situation considered here and the geometry and arithmetic of genus zero moduli spaces. This article was motivated by the idea of discovering a link between number theory and Penner's universal Ptolemy groupoid, an idea suggested to us by Bob Penner who had himself had conversation with Dennis Sullivan, and immediately seized upon by us because of the distant echo of the pentagonal defining relation of GT which could be heard (upon listening carefully) when considering Penner's sequence of ten moves giving a fundamental defining relation of the Ptolemy group. We recall with great pleasure the enthusiasm of our early discussions with him about the possible links between those relations; later communal discussions with Dennis Sullivan and Vlad Sergiescu were both enlightening and stimulating. We warmly thank all three.
Universal Ptolemy-Teichmiiller Groupoid
327
§1. The universal Ptolemy groupoid A groupoid is a category all of whose morphisms are isomorphisms. We begin by giving some basic definitions leading to the definition of the universal Ptolemy groupoid from [PI, 4.1] (see also [P2]). Identify the Poincare upper half-plane with the Poincare disk via the transformation (z — i)/(z + i). Traditionally, points on the Poincare disk are labeled by the corresponding upper half-plane, so that for instance the points - 1 , —z, 1 and i on the unit circle in C are labeled 0, 1, oo and —1, whereas the central point 0 G C is labeled i. In particular, PXR is wrapped once around the boundary of the disk, the rational numbers of course lie densely in it. We will be particularly concerned with these rational numbers. Let a marked tesselation be a maximal (i.e. triangulating) tesselation of the Poincare disk such that its vertices lie on the set of rational numbers on the boundary, equipped with a directed oriented edge. The standard marked tesselation is the dyadic tesselation T* with the marked edge from 0 to oo:
Figure 1. The standard dyadic tesselation T* with its oriented edge The elementary move on the oriented edge of a tesselation changes it from one diagonal of the unique quadrilateral containing it to the other by turning it counterclockwise; it is of order 4.
328
Pierre Lochak and Leila Schneps
Figure 2. The elementary move on the oriented edge An arrow-moving move on a marked tesselation is an operation on the tesselation which moves the oriented edge to another edge without changing the tesselation itself.
Figure 3. An arrow-moving move Definition: The universal Ptolemy groupoid is the groupoid defined as follows: Objects: The marked tesselations; Morphisms: Finite sequences (or chains) of elementary moves on the oriented edge and arrow-moving moves; Relations: The only morphism from an object to itself is the trivial one. Remark: The condition that the local groups of morphisms (i.e. groups of morphisms from an object to itself) are trivial implies that if T\ and T2 are two marked tesselations, then there is a unique morphism in the groupoid from Ti to T2.
Universal Ptolemy-Teichmiiller Groupoid
329
On any marked tesselation T, let a denote the elementary move on its oriented edge, as shown for the standard marked tesselation in Figure 2. Let the triangle to the "left" of the oriented edge of T denote the triangle lying to our left if we imagine ourselves to be lying face down on the tesselation along the oriented edge, with our head in the direction indicated by the arrow, and let (3 denote the move which sends the arrow counterclockwise to the next edge of the triangle to the left of the oriented edge, as in Figure 3. As noted by Penner, the universal Ptolemy groupoid can be given a group structure, simply because if we write a chain of a's and /?'s, it can be considered as a morphism on any given tesselation: at each point in the chain, the tesselation being acted on is the one resulting from application of all the previous moves. In particular, given a starting tesselation, a word in a and p uniquely determines a morphism in the groupoid, and conversely, every morphism in the groupoid can be given as a starting tesselation and a word in a and /?. Moreover, clearly, if a word in a and /? gives the trivial morphism from some tesselation to itself, then the same word will give the trivial morphisms from every tesselation in the Ptolemy groupoid to itself. We want to find exactly which words these are, namely to determine the relations in the group generated by a and (3 induced by the condition that the group of local morphisms of a tesselation is trivial. In other words, we need to determine the chains of a's and /3's which bring a marked tesselation to itself. Theorem 1. All words in a and j3 taking a given tesselation to itself are generated by the following words (where square brackets denotes the commutator): a 4 , /33, (a/3)5, [pap, a2 papa2],
[pap,
a20a2papa2(32a2].
Proof. In the contribution to this volume by M. Imbert, it is proved that Penner's group G is isomorphic to Richard Thompson's group. Therefore, it suffices to show that the group, let us call it G, defined by generators a and P and relations as in the statement of the theorem is isomorphic to Thompson's group. Let us give a presentation of Thompson's group which can be found on page 2 of Thompson's unpublished notes [T] (and under a different but recognizable notation, [CFP], lemma 5.2.). It is given by three generators, R, D and c\, and six relations, namely: [R'^^D^RDR'1] = 1, [R-1D,R2DR~2] = 1, a = DClR-lD, RDR^DcxR'1 = D2cxR-2D, Rc\ = (DciR'1)2 and c\ — 1. We define a homomorphism > from Thompson's group to G by setting(R) = a 2 /? 2 , <j>{D) = a3p and (ci) = /?. We
330
Pierre Lochak and Leila Schneps
need to check first that 0 is really a homomorphism, and second that it is invertible, so an isomorphism. To see that it is a homomorphism it suffices to compute the images of both sides of the six defining relations by (j). All are easily seen to hold in G. Indeed, the first two relations are the analogous commutator relations in G, and c\ = 1 is sent byto f33 = 1. The two sides of the relation cx = Dc1R~1D are sent to (3 and a3(3(3(3a2a3(3 = (3 and the two sides oiRci = (DdR~1)2 are sent to a2 and (a3f3f3f3a2)2 = a2. Finally, rewriting the remaining relation as D2c\R~2DRc~[1D~1RD~1R~1 = 1, we see that the left-hand side is sent by
This shows that cj> is a homomorphism; to show it is an isomorphism, it suffices to define (j)~1(a) = c\D~l and <j)~1(f3) = c\ (indeed, the relation c\ = Dc\R~1D shows that Thompson's group is generated by the two elements c\ and D). A Remark. The generators of this group can be identified with the corresponding moves on marked tesselations in the Ptolemy groupoid. Indeed, the words a4 and (33 clearly bring a marked tesselation back to itself and are therefore trivial; similarly (a/3)5 — 1 corresponds exactly to Penner's trivial sequence of 10 moves (cf. (c) on p. 179 of [PI]; note that (a/3) simultaneously moves the two diagonals of a pentagon whereas Penner moves one at a time). Finally, the two commutation relations in G imply that that elementary moves on the diagonals of two neighboring quadrilaterals commute, and elementary moves on the diagonals of two quadrilaterals separated only by a triangle; it is a remarkable fact that the commutation of all pairs of elementary moves taking place in disjoint quadrilaterals (Penner's relation (d) on p. 179 of [P]) are consequences of the relations in G. §2. Ribbon braids Let us recall the definitions of the Artin ribbon braid groups and the ribbon mapping class groups. First recall the definitions of the usual Artin braid and mapping class groups. The Artin braid group Bn for n > 1 is generated by <7i,...,2.
There is a natural surjection p : Bn -> Sn for n > 1 which induces a natural surjection p : M(0, n) —> Sn (for n = 4 we have the surjection p :
Universal Ptolemy-Teichmiiller Groupoid
331
B3/Z -> S3). The kernel of p (denoted by Kn in Bn and K(0, n) in M(0, n)) is known as the pure braid group resp. mapping class group. Both Kn and K(0,n) are generated by the elements Xij := o~j-i--ai+iafa^--a^\ for 1 < i < j < n. The center of Bn and of Kn is cyclic generated by the element ujn = £i2#i3#23 • °°xln • • 'Xn-i,n- The mapping class group M(0, n) (resp. the pure mapping class group if (0, n)) is the quotient of Bn (resp. if n ) by the following relations: (i) ujn = 1; (ii) Xiii+ia;i>t+2 • "Xi^x^i • • -x^i-i = 1; for 1 < i < n, where the indices are considered in Z/nZ. The Artin ribbon braid group i?* is a semi-direct product
it is generated by generators <7i,..., crn_! of the £ n factor and f 1,..., tn (all commuting) of the Z n factor, with the "semi-direct" relations given by: (Titia~l = £»+i for 1 < i < n — 1 (Jiti+io~~l = ti for 1 < i < n — 1
{(Ti,tj) — 1 for 1 < i < n - 1, j ^ i, i + 1 This group is visualized like the usual braid groups, except that the strands are replaced by flat ribbons, so that a twist on any one of them is non-trivial. The subgroup of B^ consisting of "flat" braids of the ribbons (i.e. without twists on the ribbons, cf. Figure 4) is canonically isomorphic to Bn, and from now on we identify Bn and its generators Oi with this subgroup of £*.
Figure 4. The flat braid
332
Pierre Lochak and Leila Schneps
We identify Z n with the abelian subgroup of i?* generated by a full (2?r) twist ti on each of the n ribbons (cf. Figure 5).
Figure 5. A full twist on a ribbon This visualization corresponds to the definition of #* as a semi-direct product ZnxBn given above. Since the twists ti commute with pure braids, the pure ribbon braid subgroup if* of £?* is just a direct product Z n x Kn. Let us define the ribbon mapping class group M*(0, n) to be B* modulo the following relations. Firstly, the center of J9* is generated by the element which also generates the center of ifn, namely ujn = X12X13X23 • • 'X\n • • *# n _i ?n , together with the subgroup Z n . The first relation we quotient out by is
Now, instead of using the usual sphere relations by which we quotient Bn to obtain M(0, n), we use the ribbon sphere relations: {\\') Xiyi+iXij+2 ' ' ' Xi,n%l,i ' ' ' %i,i-l =
tj.
The quotient of B* by the relations in (i') and (ii;) is the ribbon mapping class group M*(0, n). The surjection of Bn onto Sn extends to £?* by sending the subgroup Z n to 1, and the kernel of this surjection is the pure ribbon braid group K^. The surjection passes to M*(0, n) and its kernel in this group is denoted by i
Universal Ptolemy-Teichmiiller Groupoid
333
Let us give some admirable properties of the ribbon groups. (1) It is well-known that removing any strand gives a surjection from Kn onto if n _i, which induces a surjection from if (0, n) onto if (0, n— 1). There are analogous natural surjections from if* into K*_x and from K *(0, n) into if*(0, n — 1) obtained by removing one ribbon. Removing several ribbons thus gives surjections from K*(0,n) onto if*(0,ra) for m < n. (2) The braid obtained by holding two adjacent ribbons i and i + 1 firmly by their bottom ends and twisting them one full turn is equal to Uti+iXij+i. We denote this ribbon braid by U^+i. It is the same as the full twist on the single "wide" ribbon obtained by sewing the two adjacent ribbons together. The expression for the simultaneous full twist of several adjacent ribbons can easily be deduced from this one by induction. (3) There are natural injections Km into Kn for m < n given by dividing up the n strands into m adjacent packets (each of which can consist of one or more strands) called A±,... ,Am; the subgroup of Kn generated by the "flat" braids XA^AJ (as in Figure 4, considering each packet as a ribbon) is isomorphic to Km. Analogously, there are natural injections of K^ into if* for any division of the n ribbons into m adjacent packets. Each packet, considered as adjacent ribbons sewn together, forms a "wide ribbon", and the group if^ of braids on these wide ribbons is naturally a subgroup of Now, there is no such natural injection for the pure mapping class groups if (0, n), because the twist on a packet of strands is non-trivial whereas the twist on a single strand is trivial. This problem is eliminated for the ribbon mapping groups where the ribbons and the wide ribbons behave in the same way with respect to twists. Therefore we have such injections for the pure ribbon mapping class groups if *(0, m) ^^ if *(0, n); this point represents the major advantage of the use of ribbon braid groups with respect to ordinary braid groups.
§3. The universal Ptolemy-Teichmiiller groupoid The universal Ptolemy-Teichmiiller groupoid T^oo is a generalization of the universal Ptolemy groupoid in the sense that we add morphisms from a given marked tesselation to itself. The objects of Poo are those of the universal Ptolemy groupoid, namely marked braid tesselations; what we do
334
Pierre Lochak and Leila Schneps
here is to relax the condition stating that the local groups are all trivial to a condition defining the local groups as certain ribbon braid groups. To explain exactly what is going on, we visualize a marked tesselation T a little differently; we assume that there is a ribbon hanging from the interval on the circle delimited by each edge of the tesselation. To be precise, each edge of the tesselation actually divides the circle into two intervals, and we want to choose only one of them; we choose to hang the ribbon from the interval lying entirely on one side of the oriented edge. This makes sense for every edge except the oriented one, to which we associate the interval lying to the left of it in the sense explained earlier. Note that since the ribbons are determined by the edges of T, assuming their presence does not add anything to T; the point of adding them is that the nontrivial morphisms from T to itself which we are going to introduce correspond exactly to braiding them. Before continuing, we note that if one considers the infinite trivalent tree dual to the tesselation, then it comes to the same thing to attach a strand to each of its "ends" (the rationals) and consider the set of strands in a given interval as forming a ribbon, and this in turn is equivalent to attaching a strand to each vertex of the tree (uniquely associated to a rational). This idea, due to Greenberg and Sergiescu (cf. [GS]) was one of the starting points of this article. Consider thus from now on each marked tesselation T to come equipped with its ribbons. Each ribbon is automatically associated to an interval on the circle (delimited by an edge of T, a fortiori by two numbers in PXQ). Two ribbons of T are said to be disjoint if their intervals are disjoint except for at most one common endpoint. They are said to be neighbors if their associated intervals are disjoint except for exactly one common endpoint. Two ribbons of T are said to be adjacent if their intervals are delimited by two sides of a triangle of the tesselation; thus, adjacent ribbons are of course neighbors and neighbors are disjoint, but the converses are not necessarily true. Two neighboring ribbons of a given marked tesselation can always be made into adjacent ribbons of another tesselation which can be obtained from the first by a finite number of elementary moves, successively reducing to zero the (finite) number of edges coming out of the common endpoint of the two ribbons and lying inside the smallest polygon of the tesselation having as two of its edges those associated to the ribbons. On the left-hand side of Figure 6, we show two neighboring ribbons; the smallest polygon of the tesselation containing having their corresponding edges as edges is a
Universal Ptolemy-Teichmiiller Groupoid
335
quadrilateral and there is just one edge coming out of the common endpoint of the two ribbons and lying inside it (namely, its diagonal). Thus, after the elementary move on that diagonal, the two ribbons become adjacent (right-hand side), associated to two sides of the triangle A.
Figure 6. Adjacent and neighboring ribbons of a tesselation Definition: Let A and B be disjoint ribbons of a given marked tesselation T. Let t\ denote the full twist on A, oriented as in Figure 5, and let xTAB denote the flat braid of A and B shown in Figure 7, where the ribbon on the left-hand side passes in front of the right-hand one (whether the observer stands inside or outside the tesselation).
Figure 7. The braid xTAB
336
Pierre Lochak and Leila Schneps
The local group KT at a marked tesselation T is a group of morphisms from T to itself, essentially given by braiding the ribbons associated to intervals of T; it is defined as follows. Definition. Let KT be generated by the flat braids xTAB for all pairs of disjoint ribbons A and B of T and by the full twists tTA on each of these ribbons. Define the set of relations in the group KT to be the set of relations coming from the finite polygons of T containing the oriented edge, as follows. Let S be such a polygon, say with n sides, and let A\,..., An be the n ribbons associated to the intervals determined by the sides; they are pairwise disjoint (which would not be the case if the polygon S did not contain the oriented edge of T and thus lay entirely on one side of it). Let KT(S) denote the subgroup of KT generated by the twists t^. and the flat braids xTA. A.. Then the relations of KT are generated by all the relations induced by the assertion: For every S, the group KT(S) is isomorphic to the pure ribbon mapping class group K*(0,n). Definition: The universal Ptolemy-Teichmuller groupoid Voo is defined as follows: Objects:
Marked tesselations.
Morphisms: They are of two types: firstly, those of the universal Ptolemy groupoid, which act on marked tesselations as usual, and secondly, the groups Hom(T, T) ~ KT of morphisms from each marked tesselation to itself. Relations: The full set of relations in the universal Ptolemy-Teichmuller groupoid VQO is given by: (i) those of the universal Ptolemy groupoid (any sequence of elementary moves leading from a tesselation to itself is equal to 1); (ii) those of the local ribbon braid groups; (iii) commutativity of these two types of morphisms as in equation (1) below. The universal Ptolemy-Teichmuller groupoid contains the universal Ptolemy groupoid as a subgroupoid, because of (i). Let us explain (ii) and (iii) further. Firstly, from now on we use the term interval to denote an interval of the circle delimited by two rationals, and an interval ofT or equivalently,
Universal Ptolemy-Teichmiiller Groupoid
337
a ribbon of T to denote the interval delimited by an edge of the tesselation T, as before (the one on the opposite side from the oriented edge of T). Not every interval of the circle is an interval of T, of course, but every interval is the union of a finite number of intervals of T; we call such an interval a wide interval or a wide ribbon of T (obtained by sewing together a finite number of neighboring ribbons of T). Thus every interval of the circle is associated to a ribbon or a wide ribbon of T. Let A and B denote two disjoint intervals (recall that "disjoint" intervals may have one endpoint in common), equipped with ribbons. Let the braid X AB denote the usual flat braid (as in Figure 7); this twist can be applied to ribbons corresponding to any two disjoint intervals of the circle, without needing to refer to a specific tesselation. However, fixing a tesselation T, we see that the ribbons associated to the intervals A and B are either ribbons or wide ribbons of T, which implies that the braid XAB actually lies in KT for all T. We write xTAB when we want to consider the braid XAB Q>S an element of KT. Recall that given marked tesselations T' and T, there is a unique morphism 7 in the universal Ptolemy groupoid from V to T. This gives rise to a canonical isomorphism between KT and KT via KT = j~1KT'y. For all pairs of disjoint intervals A and B of the circle, this isomorphism has the property that *
=7~1*7
(1)
in the universal Ptolemy-Teichmiiller groupoid. This is what is meant by the commutation of the morphisms of the universal Ptolemy groupoid with braids in (iii) above. Proposition 2. Let T be a marked tesselation. (i) A set of generators for the group KT is given by the set of braids xTAB for all pairs of ribbons A and B corresponding to disjoint intervals of the circle, and the twists tTA on the wide ribbons of T corresponding to all intervals of the circle. The set of relations associated to this set of generators is independent of T. (ii) A smaller set of generators for KT is given by the twists tTA and xTAB where A and B are disjoint ribbons of T. A set of relations for KT associated to this set of generators is given by the relations between the generators of each KT(S) for finite polygons S of T. (iii) Another set of generators for KT is given by the set of twists t^ on
338
Pierre Lochak and Leila Schneps
all ribbons ofT and xTAB where A and B are pairs of adjacent or neighboring wide ribbons ofT. Indeed, if S a finite polygon ofT containing the oriented edge and the ribbons of S are the ribbons of T associated to the edges of S, then each subgroup KT(S) is generated by the twists on ribbons of S and the xTAB where A and B are adjacent or neighboring wide ribbons of S.
Proof, (i) and (ii) are immediate consequences of the definition of KT. For (iii), we start by showing the statement for KT{S) C KT. By definition, this subgroup is isomorphic to K*(0,n) where n is the number of edges of the polygon S. A set of generators for the pure Artin braid group Kn is given by the elements Xi...jj+i...k for all 1 < i < j < k < n; this braid denotes the flat braid of the "neighboring packets" of strands numbered i • • • j and j - f l • • • fc, which can be considered as ribbons; it looks like the one in Figure 4, with one or several strands in place of the ribbons. To see that this set really generates, it suffices to write each of the usual generators Xij of Kn in terms of these, which can be done via the formula x
ij —
x
i---j-l,jxi+l...j-lj
(draw the picture!) If j = i + 1, the usual x^ is itself a twist of neighboring packets, which consist of one strand each. The flat braids of neighboring packets Xi...jj+i...k also generate the quotient if (0, n) of Kn, so adding in the full twists on ribbons, we have a set of generators for if*(0,n). Since this group is isomorphic to KT(S), this proves the statement of (iii) for the groups KT(S). It follows immediately for KT since the group KT is generated by the subgroups KT(S) as S runs over all the finite polygons of T containing the oriented edge of T. A Let us now describe the "profinite completion" of the universal PtolemyTeichmuller groupoid VOQ. We need the following: Lemma 3. (i) If S and R are finite polygons containing the oriented edge of a given marked tesselation T\ and S lies inside R, then the subgroup KT(S) C KT is contained in KT(R). (ii) For n > 4, let Sn denote the 2n-gon in the standard marked dyadic tesselation T* obtained from S4, the basic quadrilateral containing the oriented edge Ooo, by successively dividing every edge into two. Then
KT' =
\jKT'(Sn), n>2
Universal Ptolemy-Teichmiiller Groupoid
339
where this union of groups is given by the natural inclusion of KT (Sn) into KT (£ n +i) induced by the inclusion of Sn in 5n_j-i, as in (i). (Hi) Let KT*(Sn) denote the profinite completion of KT*(Sn) forn > 2. Then the inclusion of Sn into S n +i induces a natural inclusion of KT (Sn) intoKT\Sn+1). Proof. Part (i) follows from the existence of injections if*(0, m) —> i^*(0, n) for m < n, sending ribbons in i^*(0, m) to wide ribbons in if*(0,n), cf. property (3) in §2. Part (ii) is a corollary of this, since every polygon of T* lies inside Sn for some sufficiently large n. Finally, (iii) is a consequence of the fact that like all braid and mapping class groups, the KT* (Sn) inject into their profinite completions. A Let us define the profinite local group at T* by
n>2
For any pair of disjoint intervals A, B of the circle, there exists n such that xT^B lies in iir T *(5 n ), since every rational is a vertex of some Sn. Since KT* (Sn) injects into its profinite completion, the braid xT^B lies in KT (Sn). The group KT is topologically described by the same set of generators and relations as KT . We define the profinite local group at any marked tesselation T to be the one obtained from KT* as in equation (1), i.e. by conjugating by the unique morphism 7 in the universal Ptolemy groupoid which takes T* to T. Definition: Let the profinite completion V^ of Poo be the groupoid defined as follows: Objects:
Marked tesselations T.
Morphisms: All the morphisms of the universal Ptolemy groupoid, together with the local groups Hom(T, T) = Kj, at each T. Relations: (i) those of the universal Ptolemy groupoid; (ii) those of the local profinite braid groups; (iii) commutativity of these two types of morphisms as in equation (1).
340
Pierre Lochak and Leila Schneps
§4. GT and the automorphism group of VOQ Definition: The automorphism group Aut°('Poo) of the completed Ptolemy groupoid Poo is defined to be the set of automorphisms of the groupoid P ^ which act trivially on the set of objects. The goal of this article is to show that V^, considered as a completion of the universal Ptolemy groupoid, has the property that the GrothendieckTeichmuller groupoid lies in its automoprhism group Aut°('Poo) (cf. §0). To prove this, we begin by recalling the definition of GT (cf. the survey [S] for references and details). Definition: Let GT be the monoid of pairs (A, / ) G Z* x F^ satisfying the three following relations, the first two of which take place in the profinite completion F2 of the free group on two generators F2, and the third in the profinite completion K(Q,5) of the pure mapping class group: (I) /(y, *)/(*, y) = l; (II) f(z,x)zmf(y,z)ymf(x,y)xm
= 1, where m = ±(A- 1) and z = {xy)'1]
(III) /(X34,X45)/(x5i,3;i2)/(a;23)^34)/(^45^5l)/(^12,^23) = 1.
Under a suitable multiplication law, this set forms a monoid. The group GT is defined to be the group of invertible elements of the monoid GT. Drinfel'd and Ihara showed that the group GT contains the absolute Galois group Gal(Q/Q) as a subgroup in a natural way (again, cf. [S] for all relevant references). In particular, whenever GT acts on an object, this object becomes equipped with a Galois action, indicating a - sometimes quite unexpected - link with number theory, which was one of the main motivations behind this article. Theorem 4. There is an injection GT <-•» Aut°('Poo ). Proof. Let F = (A, / ) e GT. Then we let F act trivially on the set of objects of VOQ. The proof is outlined as follows: first we define the action of F to be trivial on arrow-moving morphisms, next we give its definition on the morphisms of the universal Ptolemy groupoid (considered as a subgroupoid of T^oo) and prove in proposition 5 that the relations of the universal Ptolemy groupoid are respected, andfinallywe define the action on a set of generators of the profinite local group KT - using lemma 5 to show that the action on the generators is well-defined - and prove in lemma 6 that the action extends to an automorphism of KT'. This takes care of two of the three types of morphisms in VOQ\ to conclude, we show that the commutation
Universal Ptolemy-Teichmuller Groupoid
341
relations of equation (1) are respected. Let us proceed to the definition of the action of F on the morphisms of Arrow-moving morphisms. F acts trivially on these. Elementary morphisms. The oriented edge of a marked tesselation T determines two pairs of adjacent edges forming a quadrilateral called the basic quadrilateral of T; we call the ribbons hanging from the intervals delimited by these four edges XT, YT, %T and WT respectively, going around counterclockwise from the point of the arrow. Let ar denote the elementary move on the oriented edge of T. We set F(aT) = aT • f(xTXTYT,xlTZT). The profinite braid /(^x T y T ' x ?T^T) *s a in KT\ so F((XT) is a morphism of Poo.
mor
(2)
P m s m from T to itself, lying
Figure 8. The basic quadrilateral of a marked tesselation Lemma 5. This action of GT respects all the relations of the universal Ptolemy groupoid. Remark. The lemma can be restated by saying that every element of GT is a groupoid-isorphism from the universal Ptolemy subgroupoid of Poo to its image. This shows that GT respects the first of the three types of relation in Poo.
342
Pierre Lochak and Leila Schneps
Proof. The point is to check that if F = (A, / ) is an element of GT, then its action on any two finite sequences of elementary moves leading from T to T' is the same, since in the universal Ptolemy groupoid two such sequences give the same morphism. It is equivalent to check that F respects the relations in the groupoid, and these relations were given explicitly in theorem 1. So we simply need to compute the action of F on the five relations in a and /? given in the statement of theorem 1. Here we index the moves according to the tesselation they are acting on, so that the group can be recovered from the groupoid by dropping all indices. The relation /33 = 1 is trivially satisfied since F fixes /?. It is also clear that F respects the two commutation relations since they involve braids on disjoint ribbons (or at worst, one braid is made of ribbons all of which are contained in a single ribbon of the other braid) and such braids commute in the local braid groups. As for the pentagon relation (a/?)5 = 1, or equivan d five repeated alency (/?a)5 = 1, we have F(/3a) = (3af{xxY^Yz)i applications of this map, together with the use of equation (1) to push all the factors of /3a to the left, leave us with exactly the famous pentagon relation (III) defining GT, equal to 1. Let us check the remaining relation, a4 = 1. To start with, fix a marked tesselation To, so that OLT0 and /3T0 are the moves shown in figures 2 and 3. Then by equation (2), F(aTo) = aTo f (^^YT^ xYrQ ZTQ) - L e t Tu T2 and T3 denote the tesselations obtained from TQ via ay 0 , a^ &T0 and (XT2 OLTX &T0 (i.e. a, a2 and a 3 in the group), and write cti — a^., so that
asa2aiao = 1. Note that for the four tesselations T;, we have
where Xi, Yi, Zi and Wi are the ribbons attached to the four intervals of the each tesselation Ti shown in figure 8. Applying F to this relation, we obtain F(a4) = F(a3a2aia0)
= a3a2aia0
=
= a4 = 1,
since / satisfies f(x,y) = f(y,x)~l of the relation a4 — 1.
by relation (I) of GT. This takes care A
Universal Ptolemy-Teichmiiller Groupoid
343
Braids: Let us define the action of GT on the local groups KT. We begin by defining an action on a set of generators of KT. By (iii) of proposition 2, a set of (topological) generators of KT is given by the braids xTAB where A and B are neighboring wide ribbons of T, i.e. (finite) unions of neighboring ribbons corresponding to neighboring edges of a finite polygon of T containing the oriented edge. An example is shown in figure 9.
Figure 9. Neighboring intervals corresponding to wide ribbons of T To define the action of F on all the xTAB for neighboring wide ribbons A and B of T, it suffices to define it only on the xrAB for adjacent (possibly wide) ribbons A and B (recall that adjacent ribbons are ribbons attached to intervals corresponding to two edges of a triangle of T) for all tesselations T; we can then extend it to pairs of neighboring wide ribbons by equation (1) and the fact that we know the action of F on morphisms of the universal Ptolemy groupoid. This works as follows. Firstly, if A and B are adjacent (possibly wide) ribbons of T we set V AH/
\
AD
)
'
\
)
If A and B are neighboring (possibly wide) ribbons of T, i.e. (unions of) intervals corresponding to edges of a finite polygon S of T, then we change T into another tesselation T" such that the intervals A and B are two edges of a triangle of T", via a finite number of elementary moves on T, all taking place inside S. By writing them down explicitly and using (2) and the action of GT on elementary moves, we can compute the explicit expression
344
Pierre Lochak and Leila Schneps
for F(x\B) as follows. Choose a finite sequence of marked tesselations T* = T 0 , . . . , T r such that (1) for i > 0, Ti is obtained from T^_i by one elementary morphism gi on some edge lying inside 5 (not an edge of 5); (2) each Ti contains the polygon 5 and is identical to T* outside of 5; (3) the ribbons corresponding to the intervals A and B are adjacent ribbons of T r , i.e. the intervals A and B are two sides of a triangle of Tr. Let 7 = gr o gr_i o • • • #i; then no matter what choices we make for T i , . . . , Tr and # i , . . . , # r , 7 is the unique morphism of the universal Ptolemy groupoid taking T* to Tr. By equation (1), we have ^f~lxAB^ = XAB. By repeated applications of equations (1) and (2), we find an element 77 G K*(0,s) C K^* such that ^(7) = 777; the fact that 77 is well-defined is a consequence of lemma 5. Thus, the action of F on X\B when A and B are neighboring wide ribbons is given by
F(xTAB) = T T V ^ B ) ^ = T ^ L O V
(4)
#; This action of GT on the generators of VQQ extends to a groupoid automorphism of V^. We must check that all the relations of the groupoid are respected. Lemma 6. The action defined above ofGT on the generators xTAB of K^ for all pairs of adjacent or neighboring clumps A and B of T* determines an automorphism of Proof. Let F G GT. Then the action of F on the generators of each of the groups K*(0, Sn) for n > 2 extends to an automorphism. Indeed, it is known (cf. [PS], chapter II) that GT is an automorphism group of the pure mapping class group i^(0,2 n ) in many ways, corresponding to the trivalent trees with 2 n edges; the action we consider here corresponds to the trivalent tree dual to the polygon Sn. Now, we have the exact sequence
and it is easily seen that the GT-action on K(0,2n) extends to an automorphism of K*(Q, Sn) simply by letting F = (A, / ) eGT act on each twist tTA by sending it to {t^)x. Now let us show that the automorphisms of each lf*(0, Sn) given by F eGT respect the natural inclusions ^ ( O , ^ ) <->• K*(0,Sn+1). Recall
Universal Ptolemy-Teichmiiller Groupoid
345
that S n +i is obtained from Sn by subdividing each interval of Sn into two. If A and B are neighboring wide ribbons of T, then xTAB lies in K*(0, Sn) if and only if A and B are actually supported on S n , i.e. correspond to intervals delimited by a finite number of neighboring edges of Sn. Supposing this is the case, then of course xTAB is also supported on Sn+i, so xTAB also lies in K*(0,Sn+i), as it should since K*(0,Sn) injects into if*(0, S n +i). Furthermore, in order to compute F(x'AB), we need to use a finite series of elementary morphisms as explained in the definition of the GT-action on braids, and they consist of moves on edges lying in 5 n , and are therefore the same whether xTAB is considered as lying in if*(0, Sn) or jfrT*(O, Sn+i), so that the expression of F(x'AB) is not dependent on n, i.e. F respects the injection £*(0,S n ) -> K*(0,Sn+1). Q Lemma 7. For A and B disjoint intervals of the circle and T and T' different marked tesselations, the commutativity relations xTAB — 'Y~1X'AB'Y of equation (1) are respected by the action of GT. Proof. Recall that 7 is a finite chain of morphisms in the universal Ptolemy groupoid taking T' to T. In the case where A and B are actually adjacent ribbons for T, the lemma follows immediately from the definition of the action of F on the braids xTAB. If A and B are not adjacent ribbons for T, it suffices to take a third tesselation T" such that they are adjacent ribbons for T" and then again use the definition of F on the braids xTAB and xTAB, by commuting them to Tn via an element of the universal Ptolemy groupoid.
0 Lemmas 5, 6 and 7 show that the action of GT respects all defining relations of the universal Ptolemy-Teichmiiller groupoid, and this concludes the proof of theorem 4. A §5. Relations with the ordered Teichmiiller groupoids Let us very briefly sketch the relationship between the universal PtolemyTeichmuller groupoid and the fundamental Teichmiiller groupoids of genus zero moduli space. Let A^o,n denote the moduli space of Riemann spheres with n ordered marked points. The Teichmiiller groupoids are the fundamental groupoids 7Ti(.Mo,n; #n) of the moduli spaces A1o,n for n > 4 of genus zero Riemann surfaces with n ordered marked points, on the set Bn of base points near infinity of maximal degeneration. The use of these groupoids
346
Pierre Lochak and Leila Schneps
was suggested in [D], page 847, and their structure was investigated in detail in [PS], chapters 1.2 and II. The set Bn is essentially described by isotopy classes of numbered trivalent trees with n leaves. Let us define ordered Teichmuller groupoids to be the fundamental groupoids of the moduli spaces Mo,n on a certain subset Cn of the base point set 6 n , so that the ordered Teichmiiller groupoids are subgroupoids of the full Teichmuller groupoids. We define the base point sets Cn of the ordered Teichmuller groupoids to be the set of base points near infinity in Moyn corresponding to trivalent trees whose n leaves are numbered in cyclic order 1,..., n. The set of associativity moves acts transitively on such trees, so that the paths of the ordered Teichmuller groupoids are of two types: the braids (local groups), i.e. the fundamental groups of M 0 ,n based at each base point, which are all isomorphic to if(0,n), and associativity moves going from one base point to another. The universal Ptolemy-Teichmuller groupoid covers all the ordered Teichmiiller groupoids for n > 4 in the sense that these groupoids naturally occur as quotients of subgroupoids in many ways. Indeed, for n > 4, choose a n-sided polygon S in any given tesselation T, and consider the set of elementary paths on T which act only on edges inside the polygon. In other words, consider the finite set of tesselations X" differing from T only inside the polygon S. Now consider the set of marked tesselations obtained from these by marking any chosen edge of T not in the interior of S, and the same edge on the other tesselations differing from T only inside S. This gives a subgroupoid of 'Poo on a finite number of base points. Now we quotient the local group at each tesselation by suppressing all the ribbons except those attached to intervals delimited by edges of the polygon S. The quotient of Kj> obtained in this way is exactly the pure mapping class group KT(S) on these ribbons, isomorphic to K*(0, n); thus we obtain the ordered Teichmuller groupoid as a quotient of Poo. It is shown in [SI] that there is a GT-action on the proflnite completion of the fundamental groupoid 7Ti(A^o,n;^n); this action fixes the objects of the groupoid, i.e. the elements of B n , so it restricts to an action of 7ti(Mo,n;Cn). The relation with the main theorem of this article is that our GT-action on 'PQQ passes to the quotient described here, and gives exactly the usual one on 7Ti(Mo,n;Cn).
Universal Ptolemy-Teichmiiller Groupoid
347
References. [CFP] J.W. Cannon, W.J. Floyd and W.R. Parry, Notes on Richard Thompson's groups, preprint (University of Minnesota, 1994), to appear in Enseignement Math. [D]
V.G. Drinfel'd, On quasitriangular quasi-Hopf algebras and a group closely connected with Gal(Q/Q), Leningrad Math. J. 2 (1991), 829860.
[GS] P. Greenberg and V. Sergiescu, An acyclic extension of the braid group, Comm. Math. Helv. 66 (1991), 109-138. [IS]
M. Imbert, Sur l'isomorphisme du groupe de Richard Thompson avec le groupe de Ptolemee, this volume.
[MS] G. Moore and N. Seiberg, Classical and Quantum Conformal Field Theory, Commun. Math. Phys. 123 (1989), 177-254. [PI]
R.C. Penner, Universal Constructions in Teichmiiller Theory, Adv. in Math. 98 (1993), 143-215.
[P2]
R.C. Penner, The universal Ptolemy group and its completions, this volume.
[PS]
Triangulations, courbes arithmetiques et theories des champs, ed. L. Schneps, issue to appear of Panoramas et Syntheses, Publ. SMF, 1997.
[S]
L. Schneps, On GT, a survey, Geometric Galois Actions, volume I.
[SI] [T]
, On the genus zero Teichmiiller tower, preprint. R. Thompson, unpublished handwritten notes.
URA 762 du CNRS, Ecole Normale Superieure, 45 rue d'Ulm, 75005 Paris UMR 741 du CNRS, Laboratoire de Mathematiques, Faculte des Sciences de Besangon, 25030 Besangon Cedex
Errata for Tame and stratified objects* Bernard Teissier I wish to thank Selma Kuhlmann for bringing to my attention some errors and omissions in the text. Page Page Page Page Page Page Page
234, line 15: a parenthesis is missing after {fj)j£j234, line -11: ....consists only of finite unions of intervals. 235, line 12: language 235, line -18: Gi(x),..., Gi(x) (and not Gp(x)). 235, line -17: k[x\,..., xm+i] 235, line -13: ....proved the model completeness of the structure... 235, line -12: ....adding the exponential ...
This article appeared in volume I of Geometric Galois Actions.