ON THE COMBINATORIAL CUSPIDALIZATION OF HYPERBOLIC CURVES

Shinichi Mochizuki
In this paper, we continue our study of the pro- fundamental groups of conguration spaces associated to a hyperbolic curve, where is either the set of all prime numbers or a set consisting of a single prime number, begun in an earlier paper. Our main result may be regarded either as a combinatorial, partially bijective generalization of an injectivity theorem due to Matsumoto or as a generalization to arbitrary hyperbolic curves of injectivity and bijectivity results for genus zero curves due to Nakamura and Harbater-Schneps. More precisely, we show that if one restricts ones attention to outer automorphisms of such a pro- fundamental group of the conguration space associated to a(n) ane (respectively, proper) hyperbolic curve which are compatible with certain ber subgroups [i.e., groups that arise as kernels of the various natural projections of a conguration space to lower-dimensional conguration spaces] as well as with certain cuspidal inertia subgroups, then, as one lowers the dimension of the conguration space under consideration from n + 1 to n 1 (respectively, n 2), there is a natural injection between the resulting groups of such outer automorphisms, which is a bijection if n 4. The key tool in the proof is a combinatorial version of the Grothendieck Conjecture proven in an earlier paper by the author, which we apply to construct certain canonical sections.

Contents: Introduction 0. Notations and Conventions 1. Generalities and Injectivity for Tripods 2. Injectivity for Degenerating Ane Curves 3. Conditional Surjectivity for Ane Curves 4. The General Pronite Case 5. The Discrete Case

2000 Mathematical Subject Classication. Primary 14H30; Secondary 14H10. Typeset by AMS-TEX 1

SHINICHI MOCHIZUKI

Introduction Topological Motivation: From a classical topological point of view, one way to understand the starting point of the theory of the present paper is via the Dehn-Nielsen-Baer theorem [cf., e.g., [13], Theorem 2.9.B] to the eect that if X is a topological surface of type (g, r ) [i.e., the complement of r distinct points in a compact oriented topological surface of genus g ], then every automorphism of its [usual topological] fundamental group top 1 (X ) that stabilizes the conjugacy classes of the inertia groups arising from the r missing points arises from a homeomorphism X : X X . For n 1, let us write Xn for the complement of the diagonals in the direct product of n copies of X . Then one important consequence of the Dehn-NielsenBaer theorem, from the point of view of the present paper [cf., e.g., the proof of top (Xn ). Corollary 5.1, (ii)], is that extends to a compatible automorphism of 1 Indeed, this follows immediately from the fact that X induces a homeomorphism Xn : Xn Xn . Note, moreover, that such an argument is not possible if one only knows that X is a homotopy equivalence. That is to say, although a homotopy equivalence X X is, for instance, if r = 0, necessarily surjective, it is not necessarily injective. This possible failure of injectivity means that it is not necessarily the case that such a homotopy equivalence X X induces a homotopy equivalence Xn Xn . Put another way, one group-theoretic approach to understanding the DehnNielsen-Baer theorem is to think of this theorem as a solution to the existence portion of the following problem: The Discrete Combinatorial Cuspidalization Problem (DCCP): top Does there exist a natural functorial way to reconstruct 1 (Xn ) from top (X )? Is such a reconstruction unique? 1 At a more philosophical level, since the key property of interest of X is its injectivity i.e., the fact that it separates points one may think of this problem as the problem of reconstructing the points of X , equipped with their natural topology, top (X ). Formulated in this way, this problem group-theoretically from the group 1 takes on a somewhat anabelian avor. That is to say, one may think of it as a sort of problem in discrete combinatorial anabelian geometry. Anabelian Motivation: The author was also motivated in the development of the theory of the present paper by the following naive question that often occurs in anabelian geometry. Let X be a hyperbolic curve over a perfect eld k ; U X a nonempty open subscheme of X . Write 1 () for the etale fundamental group of a scheme.

COMBINATORIAL CUSPIDALIZATION

Naive Anabelian Cuspidalization Problem (NACP): Does there exist a natural functorial group-theoretic way to reconstruct 1 (U ) from 1 (X )? Is such a reconstruction unique? For n 1, write Xn for the n-th conguration space associated to X [i.e., the open subscheme of the product of n copies of X over k obtained by removing the diagonals cf. [24], Denition 2.1, (i)]. Thus, one has a natural projection morphism Xn+1 Xn , obtained by forgetting the factor labeled n + 1. One may think of this morphism Xn+1 Xn as parametrizing a sort of universal family of curves obtained by removing an eective divisor of degree n from X . Thus, consideration of the above NACP ultimately leads one to consider the following problem. Universal Anabelian Cuspidalization Problem (UACP): Does there exist a natural functorial group-theoretic way to reconstruct 1 (Xn ) from 1 (X )? Is such a reconstruction unique? The UACP was solved for proper X over nite elds in [21], when n = 2, and in [7], when n 3. Moreover, when k is a nite extension of Qp [i.e., the eld of p-adic numbers for some prime number p], it is shown in [22], Corollary 1.11, (iii), that the solution of the UACP for n = 3 when X is proper or for n = 2 when X is ane is precisely the obstacle to verifying the absolute p-adic version of Grothendieck Conjecture i.e., roughly speaking, realizing the functorial reconstruction of X from 1 (X ). Here, we recall that for such a p-adic k , the absolute Galois group Gk of k admits automorphisms that do not arise from scheme theory [cf. [30], the Closing Remark preceding Theorem 12.2.7]. Thus, the expectation inherent in this absolute p-adic version of Grothendieck Conjecture is that somehow the property of being coupled [i.e., within 1 (X )] with the geometric fundamental group 1 (X k k ) [where k is an algebraic closure of k ] has the property of rigidifying Gk . This sort of result is obtained, for instance, in [21], Corollary 2.3, for X of Belyi type. Put another way, if one thinks of the ring structure of k which, by class eld theory, may be thought of as a structure on the various abelianizations of the open subgroups of Gk as a certain structure on Gk which is not necessarily preserved by automorphisms of Gk [cf. the theory of [15]], then this expectation may be regarded as amounting to the idea that this ring structure on Gk is somehow encoded in the gap that lies between 1 (Xn ) and 1 (X ). This is precisely the idea that lay behind the development of theory of [22], 1. By comparison to the NACP, the UACP is closer to the DCCP discussed above. In particular, consideration of the UACP in this context ultimately leads one to the following question. Suppose further that is a set of prime numbers which is either of cardinality one or equal to the set of all prime numbers, and that k is () for the maximal an algebraically closed eld of characteristic zero. Write 1 pro- quotient of 1 (). Note that [unlike the case for more general k ] in this

SHINICHI MOCHIZUKI

case, 1 (Xn ), 1 (X ) are independent of the moduli of X [cf., e.g., [24], Proposition def (Xn ). 2.2, (v)]. Thus, in this context, it is natural to write n = 1

Pronite Combinatorial Cuspidalization Problem (PCCP): Does there exist a natural functorial group-theoretic way to reconstruct n from 1 ? Is such a reconstruction unique? Here, it is important to note that although the PCCP is entirely independent of k [and hence, in particular, of any Galois group actions], an armative answer to PCCP implies an armative answer to UACP [and hence to NACP]. That is to say: Despite the apparently purely combinatorial nature of the PCCP, our discussion above of ring structures on Gk suggests that there is quite substantial arithmetic content in the PCCP. This anabelian approach to understanding the arithmetic content of the apparently combinatorial PCCP is interesting in light of the point of view of research on the Grothendieck-Teichm uller group [cf., e.g., [5]] which is also concerned with issues similar to the PCCP [cf. the OPCCP below] and their relationship to arithmetic, but from a somewhat dierent point of view [cf. the discussion of Canonical Splittings and Cuspidalization below for more on this topic]. From a more concrete point of view motivated by the goal of proving Grothendieck Conjecture-style results to the eect that 1 () is fully faithful [cf. Remark 4.1.4] one way to think of the PCCP is as follows. Out-version of the PCCP (OPCCP): Does there exist a natural subgroup Out (n ) Out(n ) of the group of outer automorphisms of the pronite group n such that there exists a natural homomorphism Out (n ) Out (n1 ) [hence, by composition, a natural homomorphism Out (n ) Out (1 )] which is bijective? From the point of view of the DCCP, one natural approach to dening Out is to consider the condition of quasi-speciality as is done by many authors [cf. Remarks 4.1.2, 4.2.1], i.e., a condition to the eect that the conjugacy classes of certain inertia subgroups are preserved. In the theory of the present paper, we take a slightly dierent, but related approach. That is to say, we consider the condition of FC-admissibility, which, at rst glance, appears weaker than the condition of quasi-speciality, but is, in fact, almost equivalent to the condition of quasi-speciality [cf. Proposition 1.3, (vii), for more details]. The apparently weaker nature of FCadmissibility renders FC-admissibility easier to verify and hence easier to work with in the development of theory. By adopting this condition of FC-admissibility, we are able to show that a certain natural homomorphism Out (n ) Out (n1 )

COMBINATORIAL CUSPIDALIZATION

as in the OPCCP is bijective if n 5, injective if n 3 when X is arbitrary, and injective if n 2 when X is ane [cf. Theorem A below]. Main Result: Our main result is the following [cf. Corollary 1.10, Theorem 4.1 for more details]. For more on the relation of this result to earlier work ([10], [29], [32]) in the pro-l case, we refer to Remark 4.1.2; for more on the relation of this result to earlier work ([14], [26], [5]) in the pronite case, we refer to Remarks 4.1.3, 4.2.1. Theorem A. (Partial Pronite Combinatorial Cuspidalization) Let U S be a hyperbolic curve of type (g, r ) [cf. 0] over S = Spec(k), where k is an algebraically closed eld of characteristic zero. Fix a set of prime numbers which is either of cardinality one or equal to the set of all prime numbers. For integers n 1, write Un for the n-th conguration space associated to U [i.e., the open subscheme of the product of n copies of U over k obtained by removing the diagonals cf. [24], Denition 2.1, (i)];
(Un ) n = 1 def

for the maximal pro- quotient of the fundamental group of Un ; OutFC (n ) Out(n ) for the subgroup of FC-admissible [cf. Denition 1.1, (ii), for a detailed definition; Proposition 1.3, (vii), for the relationship to quasi-speciality] outer automorphisms i.e., that satisfy certain conditions concerning the ber subgroups of n [cf. [24], Denition 2.3, (iii)] and the cuspidal inertia groups of certain subquotients of these ber subgroups. If U is ane, then set n0 = 2; if def U is proper over k , then set n0 = 3. Then: (i) The natural homomorphism OutFC (n ) OutFC (n1 ) induced by the projection obtained by forgetting the factor labeled n is injective if n n0 and bijective if n 5. (ii) By permuting the various factors of Un , one obtains a natural inclusion Sn Out(n ) of the symmetric group on n letters into Out(n ) whose image commutes with OutFC (n ) if n n0 and normalizes OutFC (n ) if r = 0 and n = 2.
def

SHINICHI MOCHIZUKI

(iii) Write tripod for the maximal pro- quotient of the fundamental group of a tripod [i.e., the projective line minus three points] over k ; OutFC (n )cusp OutFC (n ) for the subgroup of outer automorphisms which determine outer auto1 [obtained by forgetting the factors of Un with morphisms of the quotient n labels > 1] that induce the identity permutation of the set of conjugacy classes of cuspidal inertia groups of 1 . Let n n0 ; x a cusp of the geometric generic ber of the morphism Un1 Un2 [which we think of as the projection obtained by forgetting the factor labeled n 1], where we take U0 = Spec(k). Then x determines, up to n -conjugacy, an isomorph Ex n of tripod . Furthermore, this n -conjugacy class is stabilized by any OutFC (n )cusp ; the commensurator and centralizer of Ex in n satisfy the relation Cn (Ex ) = Zn (Ex ) Ex . In particular, one obtains a natural outer homomorphism OutFC (n )cusp OutFC (tripod ) associated to the cusp x. Here, we note in passing that, by combining the group-theoreticity of the isomorph of the tripod fundamental group given in Theorem A, (iii), with the injectivity of Theorem A, (i), one obtains an alternative proof of [14], Theorem 2.2 cf. Remark 4.1.3. In 1, we discuss various generalities concerning etale fundamental groups of conguration spaces, including Theorem A, (iii) [cf. Corollary 1.10]. Also, we prove a certain special case of the injectivity of Theorem A, (i), in the case of a tripod [i.e., a projective line minus three points] cf. Corollary 1.12, (ii). In 2, we generalize this injectivity result to the case of degenerating ane curves [cf. Corollary 2.3, (ii)]. In 3, we show that similar techniques allow one to obtain a corresponding surjectivity result [cf. Corollary 3.3], under certain conditions, for ane curves with two moving cusps. In 4, we combine the results shown in 1, 2, 3 to prove the remaining portion of Theorem A [cf. Theorem 4.1] and discuss how the theory of the present paper is related to earlier work [cf. Corollary 4.2; Remarks 4.1.2, 4.1.3, 4.2.1]. Finally, in 5, we observe that a somewhat stronger analogue of Theorem 4.1 can be shown for the corresponding discrete [i.e., usual topological] fundamental groups [cf. Corollary 5.1]. Canonical Splittings and Cuspidalization: We continue to use the notation of the discussion of the PCCP. In some sense, the fundamental issue involved in the PCCP is the issue of how to bridge the gap between 2 and 1 1 . Here, we recall that there is a natural surjection 2 1 1 . If we consider bers over 1 , then the fundamental issue may be regarded as the def 1 ) [where the surjection is issue of bridging the gap between 2/1 = Ker(2 the surjection obtained by projection to the rst factor; thus, the projection to the 1 ] and 1 [i.e., relative to the surjection second factor yields a surjection 2/1 2 /1 1 ].
def

COMBINATORIAL CUSPIDALIZATION

If one thinks of 2/1 as 1 (X \{ }) for some closed point X (k ), then there 1 . On the other hand, suppose that is no natural splitting of the surjection 2/1 X is an ane hyperbolic curve, and one takes X \{ } to be the pointed stable log

curve Z log [over, say, a log scheme S log obtained by equipping S = Spec(k) with the pro-fs log structure determined by the monoid Q0 of nonnegative rational numbers together with the zero map Q0 k cf. 0] obtained as the limit x, where x is a cusp of X . Thus, Z consists of two irreducible components, E and F , where F may be identied with the canonical compactication of X [so X F is an open subscheme], E is a copy of the projective line joined to F at a single node , and the marked points of Z consist of the points = of F \X and the two marked points = of E . Write UE E , (X =) UF F for the open subschemes obtained as the complement of the nodes and cusps; Y log for the pointed stable log curve obtained from Z log by forgetting the marked point of E Z determined by the limit of [so we obtain a natural map Z log Y log ; X may be identied with the complement of the marked points of Y ]. Thus, by working with logarithmic fundamental groups [cf. 0], one may identify the surjection 2/1 1 with the log log surjection 1 (Z ) 1 (Y ) = 1 (X ). Then the technical starting point of the theory of the present paper may be seen in the following observation: The natural outer homomorphism
1 = 1 (X ) (UF ) (UF Z Z log ) 1 (Z log ) = 2/1 = 1 = 1 determines a canonical splitting of the surjection 1 (Z log ) = 2/1 1 (Y log ) (X ) = 1 . = 1

def

Put another way, from the point of view of semi-graphs of anabelioids determined by pointed stable curves [cf. the theory of [20]], this canonical splitting is the splitting determined by the verticial subgroup (1 (UF ) (Z log ) = 2/1 =) F 1 corresponding to the irreducible component F Z . From this point of view, one sees immediately that 2/1 is generated by F and the verticial subgroup (1 (UE ) =) E 2/1 determined by E . Thus: The study of automorphisms of 2/1 that preserve E , F , are compatible with the projection 2/1 1 [which induces an isomorphism F 1 ], and induce the identity on 1 may be reduced to the study of automorphisms of E . Moreover, by the combinatorial version of the Grothendieck Conjecture i.e., combGC of [20], it follows that one sucient condition for the preservation of [the conjugacy classes of] E , F is the compatibility of the automorphisms of 2/1 under consideration with the outer action of the inertia group that arises from the degeneration x. On the other hand, since this inertia group is none other than the inertia group of the cusp x in 1 , and the automorphisms of 2/1 under consideration arise from automorphisms of 2 , hence are compatible with the outer action of 1 on 2/1 determined by the natural exact sequence

SHINICHI MOCHIZUKI

1 2/1 2 1 1, it thus follows that the automorphisms of 2/1 that we are interested in do indeed preserve [the conjugacy classes of] E , F , hence are relatively easy to analyze. Thus, in a word: The theory of the present paper may be regarded as an interesting application of the combGC of [20]. This state of aairs is notable for a number of reasons which we shall discuss below but in particular since at the time of writing, the author is not aware of any other applications of Grothendieck Conjecture-type results.

canonical splitting

In light of the central importance of the canonical splitting determined by the combGC in the theory of the present paper, it is interesting to compare the approach of the present paper with the approaches of other authors. To this end, let us rst observe that since the canonical splitting was originally constructed via scheme theory, it stands to reason that if, instead of working with arbitrary automorphisms as in the OPCCP, one restricts ones attention to automorphisms that arise from scheme theory, then one does not need to apply the combGC. This, in eect, is the situation of [14]. That is to say: The canonical splitting determined by the combGC takes the place of i.e., may be thought of as a sort of combinatorial substitute for the property of arising from scheme theory.

COMBINATORIAL CUSPIDALIZATION

Here, it is important to note that it is precisely in situations motivated by problems in anabelian geometry that one must contend with arbitrary automorphisms that do not necessarily arise from scheme theory. As was discussed above, it was this sort of situation i.e., the issue of studying the extent to which the ring structure of the base eld is somehow group-theoretically encoded in the gap that lies between n and 1 that motivated the author to develop the theory of the present paper. Next, we observe that the canonical splitting determined by the combGC is not necessary in the theory of [5], precisely because the automorphisms studied in [5] are assumed to satisfy a certain symmetry condition [cf. Remark 4.2.1, (iii)]. This symmetry condition is suciently strong to eliminate the need for reconstructing the canonical splitting via the combGC. Here, it is interesting to note that this symmetry condition that occurs in the theory of the Grothendieck-Teichm uller group is motivated by the goal of approximating the absolute Galois group GQ of Q via group theory. On the other hand, in situations motivated by anabelian geometry for instance, involving hyperbolic curves of arbitrary genus such symmetry properties are typically unavailable. That is to say, although both the point of view of the theory of the Grothendieck-Teichm uller group, on the one hand, and the absolute anabelian point of view of the present paper, on the other, have the common goal of unraveling deep arithmetic properties of arithmetic elds [such as Q, Qp ] via their absolute Galois groups, these two points of view may be regarded as going in opposite directions in the sense that: Whereas the former point of view starts with the rational number eld Q as a given and has as its goal the explicit construction and documentation of group-theoretic conditions [on Out(1 ), when (g, r ) = (0, 3)] that approximate GQ , the latter point of view starts with the ring structure of Qp as an unknown and has as its goal the study of the extent to which the ring structure on GQp may be recovered from an arbitrary group-theoretic situation which is not subject to any restricting conditions. Finally, we conclude by observing that, in fact, the idea of applying anabelian results to construct canonical splittings that are of use in solving various cuspidalization problems i.e.,

Grothendieck Conjecture-type result

canonical splitting

application to cuspidalization

is not so surprising, in light of the following earlier developments [all of which relate to the rst ; the second and third [i.e., (A2), (A3)] of which relate to the second ]:

10

SHINICHI MOCHIZUKI

(A1) Outer Actions on Center-free Groups: If 1 H E J 1 is an exact sequence of groups, and H is center-free, then E may be recovered J i.e., as the pullfrom the induced outer action of J on H as H back via the resulting homomorphism J Out(H ) of the natural exact sequence 1 H Aut(H ) Out(H ) 1 [cf. 0]. That is to say, the center-freeness of H which may be thought of as the most primitive example, i.e., as a sort of degenerate version, of the property of being anabelian gives rise to a sort of anabelian semi-simplicity in the form of the isomorphism E H J . This anabelian semi-simplicity contrasts sharply with the situation that occurs when H fails to be centerfree, in which case there are many possible isomorphism classes for the extension E . Perhaps the simplest example of this phenomenon namely, the extensions 1 p Z Z Z/pZ 1 and 1 p Z (p Z) (Z/pZ) Z/pZ 1 [where p is a prime number] suggests strongly that this phenomenon of anabelian semi-simplicity has substantial arithmetic content [cf., e.g., the discussion of [19], Remark 1.5.1] i.e., it is as if, by working with center-free groups [such as free or pro- free groups], one is aorded with canonical splittings of the analogue of the extension 1 p Z Z Z/pZ 1! (A2) Elliptic and Belyi cuspidalizations [cf. [22], 3]: In this theory one constructs cuspidalizations of a hyperbolic curve X by interpreting either a multiplication by n endomorphism of an elliptic curve or a Belyi map to a projective line minus three points as, roughly speaking, an open immersion Y X of a nite etale covering Y X of X . This diagram X Y X may be thought of as a sort of canonical section; moreover, this canonical section is constructed group-theoretically in loc. cit. precisely by applying the main [anabelian] result of [16]. (A3) Cuspidalization over Finite Fields: Anabelian results such as the main result of [16] have often been referred to as versions of the Tate Conjecture [concerning abelian varieties] for hyperbolic curves. Over nite elds, the Tate Conjecture is closely related to the Riemann hypothesis for abelian varieties over nite elds, which is, in turn, closely related to various semi-simplicity properties of the Tate module [cf. the theory of [25]]. Moreover, such semi-simplicity properties arising from the Riemann hypothesis for abelian varieties play a key role i.e., in the form of canonical splittings via weights in the construction of cuspidalizations over nite elds in [21], [7]. (A4) The Mono-anabelian Theory of [23]: If one thinks of canonical splittings as canonical liftings, then the idea of applying anabelian geometry to construct canonical liftings permeates the theory of [23] [cf., especially, the discussion of the Introduction to [23]].
out out

COMBINATORIAL CUSPIDALIZATION

11

Acknowledgements: The material presented in this paper was stimulated by the work of Makoto Matsumoto [i.e., [14]], as well as the work of Marco Boggi on the congruence subgroup problem for hyperbolic curves. Also, I would like to thank Akio Tamagawa for helpful discussions concerning the material presented in this paper and for informing me of the references discussed in Remarks 4.1.2, 4.2.1.

Section 0: Notations and Conventions Topological Groups: If G is a center-free topological group, then we have a natural exact sequence 1 G Aut(G) Out(G) 1 where Aut(G) denotes the group of automorphisms of the topological group G; the injective [since G is center-free!] homomorphism G Aut(G) is obtained by letting G act on G by inner automorphisms; Out(G) is dened so as to render the sequence exact. If J Out(G) is a homomorphism of groups, then we shall write G
out out

J = Aut(G) Out(G) J

def

for the outer semi-direct product of J with G. Thus, we have a natural exact sequence: 1 G G J J 1. If H G is a closed subgroup of a topological group G, then we shall use the notation ZG (H ), NG (H ), CG (H ) to denote, respectively, the centralizer, the normalizer, and commensurator of H in G [cf., e.g., [20], 0]. If H = NG (H ) (respectively, H = CG (H )), then we shall say that H is normally terminal (respectively, commensurably terminal) in G. Log Schemes: When a scheme appears in a diagram of log schemes, the scheme is to be understood as a log scheme equipped with the trivial log structure. If X log is a log scheme, then we shall denote its interior i.e., the largest open subscheme over which the log structure is trivial by UX . Fiber products of (pro-)fs log schemes are to be understood as ber products taken in the category of (pro-)fs log schemes. The Etale Fundamental Group of a Log Scheme: Throughout the present paper, we shall often consider the etale fundamental group of a connected fs noetherian log scheme [cf. [11]; [6], Appendix B], which

12

SHINICHI MOCHIZUKI

we shall denote 1 (); we shall denote the maximal pro- quotient of 1 () (). The theory of the 1 () of a connected fs noetherian log scheme by 1 extends immediately to connected pro-fs noetherian log schemes; thus, we shall apply this routine extension in the present paper without further mention. Recall that if X log is a log regular, connected log scheme of characteristic zero [i.e., there exists a morphism X Spec(Q)], then the log purity theorem of Fujiwara-Kato asserts that there is a natural isomorphism 1 (X log ) 1 (UX ) [cf., e.g., [11]; [17], Theorem B].
log Let S be a log regular log scheme such that S = Spec(R ), where R is a complete noetherian local ring of characteristic zero with algebraically closed residue eld k . Write K for the quotient eld of R . Let K be a maximal algebraic extension of K among those algebraic extensions that are unramied def over R . Write R K for the integral closure of R in K ; S = Spec(R). Then by considering the integral closure of R in the various nite extensions of K in K , one obtains a log structure on S such that the resulting log scheme S log may be thought of as a pro-fs log scheme corresponding to a projective system of log regular log schemes in which the transition morphisms are [by the log purity theorem] nite Kummer log etale. Write k for the residue eld of R [so k = k ]; def def log slog = Spec(k ) S S ; slog = Spec(k) S S log .

Next, let
log log X S

be a proper, log smooth morphism; write

log log for the result of base-changing via the morphisms S log S , slog S , log slog S . Then by [33], Th eor` eme 2.2, (a) [in the case where S is a trait]; [6], Corollary 1 [for the general case], we have a natural specialization isomorlog log phism 1 (X s ) 1 (X ). We shall also refer to the composite isomorphism log log 1 (X s ) 1 (X ) 1 (UX ) [where the second isomorphism arises from the log purity theorem] as the specialization isomorphism. By applying these speciallog log S to the various log ization isomorphisms to the result of base-changing X regular log schemes that appear in the projective system [discussed above] associated to the pro-fs log scheme S log , we thus obtain specialization isomorphisms log ) 1 (X log ) 1 (UX ) 1 (Xs

for X log S log . Here, we note that if K is any algebraic closure of K , and the log log S to US is a log conguration space associated to some restriction of X

COMBINATORIAL CUSPIDALIZATION

13

family of hyperbolic curves over US [cf. [24], Denition 2.1, (i)], then we have a natural isomorphism 1 (UX ) 1 (UX K K ) [cf. [24], Proposition 2.2, (iii)]. We shall also refer to the composite isomorphism log ) 1 (UX K K ) as the specialization isomorphism. 1 (Xs Curves: We shall use the terms hyperbolic curve, cusp, stable log curve, and smooth log curve as they are dened in [20], 0. Thus, the interior of a smooth log curve over a scheme determines a family of hyperbolic curves over the scheme. A smooth log curve or family of hyperbolic curves of type (0, 3) will be referred to as a tripod. We shall use the terms n-th conguration space and n-th log conguration space as they are dened in [24], Denition 2.1, (i). If g , r are positive integers such that log 2g 2 + r > 0, then we shall write Mg,r for the moduli stack Mg,r of pointed stable curves of type (g, r ) over [the ring of rational integers] Z equipped with the log structure determined by the divisor at innity. Here, we assume the marking log sections of the pointed stable curves to be ordered. The interior of Mg,r will be denoted Mg,r .

Section 1: Generalities and Injectivity for Tripods In the present 1, we begin by discussing various generalities concerning the various log conguration spaces associated to a hyperbolic curve. This discussion leads naturally to a proof of a certain special case [cf. Corollary 1.12, (ii)] of our main result [cf. Theorem 4.1 below] for tripods [cf. 0]. Let S = Spec(k), where k is an algebraically closed eld of characteristic zero, and X log S a smooth log curve of type (g, r ) [cf. 0]. Fix a set of prime numbers which is either of cardinality one or equal to the set of all prime numbers. Denition 1.1. Let n 1 be an integer.
def def

log (i) Write Xn for the n-th log conguration space associated to [the family of log hyperbolic curves determined by] X log [cf. 0]; X0 = S . We shall think of the log as labeled by the indices 1, . . . , n. Write factors of Xn log log log log log Xn Xn 1 . . . Xm . . . X2 X1

14

SHINICHI MOCHIZUKI

for the projections obtained by forgetting, successively, the factors labeled by indices > m [as m ranges over the positive integers n]. Write
log (Xn ) n = 1 log [cf. for the maximal pro- quotient of the fundamental group of the log scheme Xn 0; the discussion preceding [24], Denition 2.1, (i)]. Thus, we obtain a sequence of surjections n n1 ... m ... 2 1 def

which we shall refer to as standard. If we write Km = Ker(n {1}, then we obtain a ltration of subgroups

def

m ), 0 =

def

{1} = Kn Kn1 . . . Km . . . K2 K1 K0 = n which we shall refer to as the standard ber ltration on n . Also, for nonnegative integers a b n, we shall write b/a = Ka /Kb so we obtain a natural injection b/a n /Kb = b . Thus, if m is a positive integer n, then we shall refer to m/m1 as a standard-adjacent subquotient of n . The standard-adjacent subquotient m/m1 may be naturally identied with the maximal pro- quotient of the etale fundamental group of the geometric generic ber of the morphism on interiors UXm UXm1 . Since this geometric generic ber is a hyperbolic curve of type (g, r + m 1), it makes sense to speak of the cuspidal inertia groups each of which is [noncanonically!] isomorphic to the maximal pro- quotient Z of Z of a standard-adjacent subquotient. (ii) Let : n n be an automorphism of the topological group n . Let us say that is C-admissible [i.e., cusp-admissible] if (Ka ) = Ka for every subgroup appearing in the standard ber ltration, and, moreover, induces a bijection of the collection of cuspidal inertia groups contained in each standard-adjacent subquotient of the standard ber ltration. Let us say that is F-admissible [i.e., ber-admissible] if (H ) = H for every ber subgroup H n [cf. [24], Denition 2.3, (iii), as well as Remark 1.1.2 below]. Let us say that is FC-admissible [i.e., ber-cusp-admissible] if is F-admissible and C-admissible. If : n n is an FC-admissible automorphism, then let us say that is a DFC-admissible [i.e., diagonal-ber-cusp-admissible] if 1 induces the same automorphism of 1 relative to the various quotients n by ber subgroups of co-length 1 [cf. [24], Denition 2.3, (iii)]. If : n n is a DFC-admissible automorphism, then let us say that is an IFC-admissible automorphism [i.e., identity-ber-cusp-admissible] if induces the identity automorphism of 1 relative to the various quotients n 1 by ber subgroups of co-length 1. Write Aut(n ) for the group of automorphisms of the topological group n ; AutIFC (n ) AutDFC (n ) AutFC (n ) AutF (n ) Aut(n ) Inn(n )
def

COMBINATORIAL CUSPIDALIZATION

15

for the subgroups of F-admissible, FC-admissible, DFC-admissible, IFC-admissible, and inner automorphisms; OutFC (n ) = AutFC (n )/Inn(n ) OutF (n ) = AutF (n )/Inn(n ) Out(n ) for the corresponding outer automorphisms. Thus, we obtain a natural exact sequence 1 AutIFC (n ) AutDFC (n ) Aut(1 ) induced by the standard surjection n (iii) Write n n for the intersection of the various ber subgroups of co-length 1. Thus, we obtain a natural inclusion n AutIFC (n ) induced by the inclusion n n Inn(n ) Aut(n ) [cf. Remark 1.1.1 below].
log , one obtains a natural inclusion (iv) By permuting the various factors of Xn def def

1 of (i).

Sn Out(n ) of the symmetric group on n letters into Out(n ). We shall refer to the elements of the image of this inclusion as the permutation outer automorphisms of n , and to elements of Aut(n ) that lift permutation outer automorphisms as permutation automorphisms of n . Write OutFCP (n ) OutFC (n ) for the subgroup of outer automorphisms that commute with the permutation outer automorphisms. (v) We shall append the superscript cusp to the various groups of FCadmissible [outer] automorphisms discussed in (ii), (iv) to denote the subgroup of FC-admissible [outer] automorphisms that determine [via the standard surjec1 of (i)] an [outer] automorphism of 1 that induces the identity tion n permutation of the set of conjugacy classes of cuspidal inertia groups of 1 . = n . Suppose (vi) When (g, r ) = (0, 3), we shall write tripod = 1 , tripod n that (g, r ) = (0, 3), and that the cusps of X log are labeled a, b, c. Here, we regard the symbols {a, b, c, 1, 2, . . . , n} as equipped with the ordering a < b < c < 1 < 2 < . . . < n. Then, as is well-known, there is a natural isomorphism over k
log (M0,n+3 )k Xn log def def

where we write (M0,n+3 )k for the moduli scheme over k of pointed stable curves of type (0, n + 3), equipped with its natural log structure [cf. 0]. [Here, we assume

log

16

SHINICHI MOCHIZUKI

the marking sections of the pointed stable curves to be ordered.] In particular, there log is a natural action of the symmetric group on n + 3 letters on (M0,n+3 )k , hence log . We shall denote this symmetric group regarded as a group acting also on Xn log on Xn by SM n+3 . In particular, we obtain a natural homomorphism
tripod ) SM n+3 Out(n

the elements of whose image we shall refer to as outer modular symmetries. [Thus, the permutation outer automorphisms are the outer modular symmetries that occur as elements of the image of the inclusion Sn SM n+3 obtained by considering permutations of the subset {1, . . . , n} {a, b, c, 1, . . . , n}.] We shall refer to ele) that lift outer modular symmetries as modular symmetries ments of Aut(tripod n tripod . Write of n OutFCS (tripod ) OutFC (tripod ) n n for the subgroup of elements that commute with the outer modular symmetries; OutFC (tripod )S OutFC (tripod ) n n ) OutFC (tripod ) via the hofor the inverse image of the subgroup OutFCS (tripod 1 1 tripod FC FC tripod ) Out (1 ) induced by the standard surjection momorphism Out (n tripod tripod n 1 of (i). Thus, we have inclusions ) OutFC (tripod )S OutFC (tripod )cusp OutFCS (tripod n n n and an equality OutFCS (tripod ) = OutFC (tripod )S . Here, the second displayed 1 1 inclusion follows by considering the induced permutations of the conjugacy classes , in light of the fact that S3 is center-free. of the cuspidal inertia groups of tripod 1 Remark 1.1.1. We recall in passing that, in the notation of Denition 1.1, n is slim [cf. [24], Proposition 2.2, (ii)]. In particular, we have a natural isomorphism n Inn(n ). Remark 1.1.2. We recall in passing that, in the notation of Denition 1.1, when (g, r ) {(0, 3); (1, 1)}, it holds that for any Aut(n ) and any ber subgroup H n , (H ) is a ber subgroup of n [though it is not necessarily the case that (H ) = H !]. Indeed, this follows from [24], Corollary 6.3. Remark 1.1.3. If Aut(n ) satises the condition that (Ka ) = Ka for a = 1, . . . , n, then often e.g., in situations where there is a suciently nontrivial Galois action involved it is possible to verify the C-admissibility of by applying [20], Corollary 2.7, (i), which allows one to conclude group-theoretic cuspidality from l-cyclotomic full-ness. Remark 1.1.4. In the context of Denition 1.1, (vi), we observe that if, for instance, n = 2, then one veries immediately that the outer modular symmetry

COMBINATORIAL CUSPIDALIZATION
def

17

determined by the permutation = (a b)(c 1) yields an example of a C-admissible element of Out(tripod ) [since conjugation by preserves the set of transpositions 2 {(a 2), (b 2), (c 2), (1 2)}] which is not F-admissible [since conjugation by switches the transpositions (c 2), (1 2) cf. the argument of the nal portion of Remark 1.1.5 below]. On the other hand, whereas every element of Out(tripod ) is F1 tripod is a free pro- group] examples of admissible, it is easy to construct [since 1 tripod ) which are not C-admissible. Thus, in general, neither of elements of Out(1 the two properties of C- and F-admissibility implies the other. Remark 1.1.5. Let OutFC (n )cusp . Then observe that necessarily induces the identity permutation on the set of conjugacy classes of cuspidal inertia groups of every standard-adjacent subquotient of n [i.e., not just 1 ]. Indeed, by applying the interpretation of the various b/a as ba s for appropriate X log [cf. [24], Proposition 2.4, (i)], we reduce immediately to the case n = 2. But then the cuspidal inertia group 2/1 associated to the unique new cusp that appears may be characterized by the property that it is contained in 2 [which, in light of the F-admissibility of , is clearly preserved by ]. Proposition 1.2. (First Properties of Admissibility) In the notation of Denition 1.1, (ii), let Aut(n ). Then: (i) Suppose that (n ) = n . Then there exists a permutation automorphism Aut(n ) such that is F-admissible. In particular, if is C-admissible, then it follows that is FC-admissible. m be the quotient of n by a (ii) Suppose that AutFC (n ). Let : n ber subgroup of co-length m n [cf. [24], Denition 2.3, (iii)]. Then induces, relative to , an element AutFC (m ). If, moreover, AutDFC (n ) (respectively, AutIFC (n )), then AutDFC (m ) (respectively, AutIFC (m )). (iii) Suppose that AutFC (n ). Then there exist AutDFC (n ), Inn(n ) such that = . Proof. First, we consider assertion (i). Since (n ) = n , it follows that induces an automorphism of the quotient n 1 . . . 1 [i.e., onto the direct product of n copies of 1 ] determined by the various ber subgroups of co-length 1. Moreover, by [24], Corollary 3.4, this automorphism of 1 . . . 1 is necessarily compatible with the direct product decomposition of this group, up to some permutation of the factors. Thus, by replacing by for some permutation automorphism , we may assume that the induced automorphism of 1 . . . 1 stabilizes each of the direct factors. Now let us observe that this stabilization of the direct factors is sucient to imply that (H ) = H for any ber subgroup H n . Indeed, without loss of generality, we may assume [by possibly re-ordering the indices] that H = Ka for some Ka as in Denition 1.1, (i). By applying the same argument to 1 , it suces to verify that (Ka ) Ka . Thus, let us suppose that (Ka ) Kb for some b < a, but (Ka ) Kb+1 . On the other hand, the image of (Ka ) in

18

SHINICHI MOCHIZUKI

b+1/b = Kb /Kb+1 is normal, closed, topologically nitely generated, and of innite index [since, in light of the stabilization of direct factors observed above, this image 1 ]. Thus, by [24], Theorem 1.5 maps to {1} via the natural projection Kb /Kb+1 i.e., essentially the theorem of Lubotzky-Melnikov-van den Dries we conclude that this image is trivial, a contradiction. This contradiction completes the proof of assertion (i). Assertion (ii) is immediate from the denitions. Next, we consider assertion (iii). For positive integers m n, write m : n 1 for the quotient of n by the ber subgroup whose co-prole is equal to {m} [cf. [24], Denition 2.3, (iii)]. Thus, by assertion (ii), we obtain various m = m Aut(1 ), with images [m ] Out(1 ). Then let us observe that to complete the proof of assertion (iii), it suces to verify the following claim: [m ] Out(1 ) is independent of m. To verify this claim, we reason as follows: By applying assertion (ii) to the sur2 for which Ker() has co-prole {1, m} for m = 1, we reduce jection : n immediately to the case where n = 2. Then observe that it follows immediately from the uniqueness of a cusp associated to a given cuspidal inertia group [cf. [20], Proposition 1.2, (i)] that the decomposition groups 2 [all of which are 2 -conjugate to one another] associated to the diagonal divisor in X2 may be reconstructed as the normalizers of the various cuspidal inertia groups of 2/1 that lie in 2 . In particular, it follows immediately that induces a bijection of the collection of decomposition groups of 2 associated to the diagonal divisor in X2 [all of which are 2 -conjugate to one another]. Thus, the automorphism of 1 1 induced by relative to the quotient (1 , 2 ) : 2 1 1 maps the diagonal 1 1 1 [which is the image of a decomposition group associated to the diagonal divisor in X2 ] to some (1 1 )-conjugate of the diagonal 1 1 1 . But then it follows formally that [1 ] = [2 ]. This completes the proof of the claim, and hence of assertion (iii). Proposition 1.3. (Decomposition and Inertia Groups) Let n 1. Write Dn for the set of irreducible divisors contained in the complement of the interior log ; Xn \UXn of Xn I D n for the inertia and decomposition groups, well-dened [as a pair] up to n log log Xn conjugacy, associated to Dn ; log : Xn 1 for the projection obtained log log log for the projection obtained by forgetting the factor labeled n; : Xn X1 by forgetting the factors with labels = n; : n n1 , : n 1 for the surjections determined by log , log . Also, we recall the notation Z() (), N() (), C() () reviewed in 0. Then: (i) Dn may be decomposed as a union of two disjoint subsets
hor Dn = Dn ver Dn def

COMBINATORIAL CUSPIDALIZATION

19

hor where Dn is the set of divisors which are horizontal with respect to log [i.e., ver ver is the set of divisors Dn the cusps of the geometric generic ber of log ]; Dn hor which are vertical with respect to log [so n 2, and n ( ) Dn1 for Dn ].

(ii) Let n 2; Dn1 . Then the log structure on X log determines on the ber (Xn ) of log over the generic point of a structure of pointed stable curve; (Xn ) consists of precisely two irreducible components [which may be thought ver of as elements of Dn ] joined by a single node . One of these two irreducible ver components, which we shall denote F Dn , maps isomorphically to X1 = X ver , maps to a cusp of X1 = X via via ; the other, which we shall denote E Dn . (iii) In the situation of (ii), let {F , E }; suppose that the various conjugacy classes have been chosen so that (D ) = D . Write
1 n, = (I ) n ; def

DI = D
def

n, n, ;

= D

def

n/n1

and F E n/n1 for the decomposition group of in n/n1 . Then: (a) induces an isomorphism F 1 ; (b) maps E onto a cuspidal inertia group of 1 ; (c) , are commensurably terminal in n/n1 ; (d) induces an isomorphism I I ; (e) the inclusions I , n, I induce an isomorphism I DI ; (f ) D = Cn, ( ); (g) I = Zn, ( ). (iv) In the situation of (ii), let AutFC (n ); {F , E , }; , Dn1 . hor [Thus, we obtain primed versions F , E Dn , , corresponding to of the data constructed in (ii), (iii) for .] Suppose that the automorphism of n1 induced via by stabilizes I n1 (respectively, maps I n1 to I n1 ). Then maps the n/n1 -conjugacy (respectively, n -conjugacy) class of to itself (respectively, to the n -conjugacy class of ). If {F , E } [so {F , E }], then a similar statement holds with , replaced by DI , I D or I , I . (v) The assignment I determines an injection of Dn into the set of n -conjugacy classes of subgroups of n that are isomorphic to the maximal pro- quotient Z of Z. (vi) Every OutFC (n )cusp stabilizes the n -conjugacy class of the inertia group I , for Dn .
Dn (vii) Write Pn for the product X k . . . k X of n copies of X over k ; Dn for the subset consisting of the strict transforms in Xn of the various irreducible divisors in the complement of the image of the natural open immersion UXn Pn ;

OutQS (n ) Out(n ) where QS stands for quasi-special for the subgroup of outer automor . Then phisms that stabilize the conjugacy class of each inertia group I , for Dn QS FC cusp . Out (n ) = Out (n )

20

SHINICHI MOCHIZUKI

Proof. We apply induction on n. Thus, in the following, we may assume that Proposition 1.3 has been veried for smaller n than the n under consideration. Assertion (i) is immediate from the denitions. Assertion (ii) follows from the log log log well-known geometry of Xn , Xn 1 , by thinking of Xn1 as a certain moduli space of pointed stable curves and log as the tautological pointed stable curve over this moduli space. Next, we consider assertion (iii). First, we observe that by applying the specialization isomorphisms [cf. 0] associated to the restriction log log Xn of log : Xn 1 to the completion of Xn1 along the generic point of , we conclude that the pointed stable curve structure on (Xn ) [cf. assertion (ii)] determines a semi-graph of anabelioids of pro- PSC-type as discussed in [20], Denition 1.1, (i) [cf. also the discussion of [18], Appendix] whose associated PSCfundamental group may be identied with n/n1 . From this point of view, forms a verticial subgroup [cf. [20], Denition 1.1, (ii)]; forms a(n) [nodal] edge-like subgroup [cf. [20], Denition 1.1, (ii)]. In particular, is center-free [cf., e.g., [20], Remark 1.1.3]. Now (a), (b) follow from the description of F , E given in assertion (ii); (c) follows from [20], Proposition 1.2, (ii). To verify (d), observe that by general considerations, the inertia group I is isomorphic to some quotient of Z ; on the other hand, by the induction hypothesis, I is isomorphic to Z [cf. assertion (v) for n 1]; thus, since (Xn ) is reduced at its two generic points [which correspond to F , E ], it follows that the homomorphism (Z ) I I ( = Z ) is surjective, hence an isomorphism. Now (e) follows immediately from (d); (f) follows from (c), (d), and (e); since, as observed above, I is abelian, (g) follows from (d), (e), (f), and the fact that is center-free. This completes the proof of assertion (iii). Next, we observe that since induces a bijection of the collection of cuspidal inertia groups n/n1 [a fact which renders it possible to apply the theory of [20] in the noncuspidal case], assertion (iv) for , follows immediately from I [20], Corollary 2.7, (iii); assertion (iv) for DI , D or I , I follows from assertion (iv) for , by applying (f), (g) of assertion (iii). Next, we consider assertions (v), (vi). When n = 1, assertions (v), (vi) follow, respectively, from the uniqueness of a cusp associated to a given cuspidal inertia group [cf. [20], Proposition 1.2, (i)], and the fact that OutFC (n )cusp . Thus, we may assume that n 2. The fact that stabilizes the conjugacy classes of the hor follows immediately from the fact that is C-admissible [cf. also I for Dn ver Remark 1.1.5]. Now let Dn , Dn1 be as in assertion (iii). By the induction hypothesis, I is isomorphic to Z and determines a n1 -conjugacy class that is distinct from the n1 -conjugacy classes of the I() of elements of Dn1 that are = ; moreover, the outer automorphism OutFC (n1 )cusp induced by via stabilizes the conjugacy class of I . In particular, by (d) of assertion (iii), it follows hor may be that I is isomorphic to Z , hence that the I() of elements of Dn ver distinguished from those of Dn by the property that they lie in n/n1 = Ker( ) and from one another by [20], Proposition 1.2, (i). Thus, to complete the proof of assertions (v), (vi), it suces to verify assertions (v), (vi) with Dn replaced by the subset {F , E } Dn . But then assertion (vi) follows from the respd case of assertion (iv); moreover, by the non-respd case of assertion (iv), if IE , IF are n -conjugate, then they are n/n1 -conjugate.

COMBINATORIAL CUSPIDALIZATION

21

Thus, to complete the proof of assertion (v), it suces to derive a contradiction under the assumption that IE = IF 1 , where n/n1 . Note that by (e) of assertion (iii), this assumption implies that IE commutes with E , F 1 . Next, observe that by projecting to the various maximal pro-l quotients for some l , we may assume without loss of generality that = {l}. Then one veries immediately that the images of E , F in the abelianization ab n/n1 of n/n1 ab generate n/n1 , hence [since n/n1 is a pro-l group cf., e.g., [31], Proposition 7.7.2] that n/n1 is generated by E and any single n/n1 -conjugate of F . Thus, in summary, we conclude that IE commutes with n/n1 , i.e., that the outer action of I on n/n1 is trivial. On the other hand, since the nodal curve (Xn ) is not smooth, we obtain a contradiction, for instance, from [20], Proposition 2.6. This completes the proof of assertion (v). Finally, we consider assertion (vii). The fact that OutFC (n )cusp OutQS (n ) follows immediately from assertion (vi). Next, let us observe that by applying Zariski-Nagata purity [i.e., the classical non-logarithmic version of the log purity theorem discussed in 0] to the product of n copies of UX over k , it follows that the subgroup n n is topologically normally generated by the I , for the Dn that arise as strict transforms of the various diagonals in Pn . Thus, the fact that OutQS (n ) OutFC (n )cusp follows immediately from the denition of OutQS () and Proposition 1.2, (i). This completes the proof of assertion (vii).

Remark 1.3.1. The theory of inertia and decomposition groups such as those discussed in Proposition 1.3 is developed in greater detail in [22], 1. For i = 1, 2, write
log log : X2 X1 prlog i

for the projection to the factor labeled i, pri : X2 X1 for the underlying morphism of schemes, and pi : 2 1 for the surjection induced by prlog i . Denition 1.4. Let x X (k ) be a cusp of X log .

log determines on the ber (X2 )x of the (i) Observe that the log structure on X2 morphism pr1 : X2 X1 over x a structure of pointed stable curve, which consists of two irreducible components, one of which which we shall denote Fx maps isomorphically to X via pr2 : X2 X1 = X , the other of which which we shall denote Ex maps to the point x X (k ) via pr2 ; Fx , Ex are joined at a single node x [cf. Proposition 1.3, (ii)]. Let us refer to Fx as the major cuspidal component at x, to Ex as the minor cuspidal component at x, and to x as the nexus at x. Thus, the complement in Fx (respectively, Ex ) of the nodes and cusps [relative to the pointed stable curve structure on (X2 )x ] of Fx (respectively, Ex ) which we shall refer to as the interior UFx of Fx (respectively, UEx of Ex ) determines a hyperbolic curve UFx (respectively, tripod UEx ). Moreover, pr2 induces [compatible] isomorphisms UFx UX , Fx X .

22

SHINICHI MOCHIZUKI

(ii) As discussed in Proposition 1.3, (iii), and its proof, the major and minor cuspidal components at x, together with the nexus at x, determine [conjugacy classes of] verticial and edge-like subgroups [cf. [20], Denition 1.1, (ii)] Fx , Ex , x 2/1 which we shall refer to, respectively, as major verticial, minor verticial, and nexus subgroups. Thus, [cf. Proposition 1.3, (iii), (a), (b)] the morphism p2 : 2 1 determines an isomorphism F x 1 i.e., the major verticial subgroups may be thought of as dening sections of the projection p2 : 2 1 ; p2 maps Ex onto a cuspidal inertia group of 1 associated to x. For suitable choices within the various conjugacy classes involved, we have natural inclusions Ex [inside 2/1 ]. Proposition 1.5. (First Properties of Major and Minor Verticial Subgroups) In the notation of Denition 1.4: (i) x , Fx , and Ex are commensurably terminal in 2/1 . (ii) Suppose that one xes x 2/1 among its various 2/1 -conjugates. Then the condition that there exist inclusions x Ex ; x F x x F x

completely determines Ex and Fx among their various 2/1 -conjugates. (iii) In the notation of (ii), the compatible inclusions x Ex 2/1 , x Fx 2/1 determine an isomorphism lim Ex x Fx 2 /1

where the inductive limit is taken in the category of pro- groups. Proof. Assertion (i) follows from [20], Proposition 1.2, (ii) [cf. Proposition 1.3, (iii), (c)]. Assertion (ii) follows from the fact that every nodal edge-like subgroup is contained in precisely two verticial subgroups [cf. [20], Proposition 1.5, (i)]. Assertion (iii) may be thought of as a consequence of the van Kampen Theorem in elementary algebraic topology. At a more combinatorial level, one may reason as follows: It follows immediately from the simple structure of the dual graph of the pointed stable curves considered in Denition 1.4 that there is a natural equivalence of categories [arising from the parenthesized inductive system in the statement of assertion (iii)] between

COMBINATORIAL CUSPIDALIZATION

23

(a) the category of nite sets E with continuous 2/1 -action [and 2/1 equivariant morphisms] and (b) the category of nite sets equipped with continuous actions of Fx , Ex which restrict to the same action on x Fx , x Ex [and Fx -, Ex -equivariant morphisms]. The isomorphism between 2/1 and the inductive limit of the parenthesized inductive system of assertion (iii) now follows formally from this equivalence of categories.

Remark 1.5.1. The technique of van Kampen-style gluing of fundamental groups that appears in Proposition 1.5, (iii), will play an important role in the present paper. Similar methods involving isomorphs of the fundamental group of a tripod [cf. Corollary 1.10, (iii), below; Theorem A, (iii), of the Introduction] may be seen in the arguments of [27], [28]. Proposition 1.6. (Inertia Groups and Symmetry) In the notation of the discussion preceding Denition 1.4, write 1\2 = Ker(p2 : 2
def

1 )

1 )]. Thus, each cusp of the family of hyperbolic curves [cf. 2/1 = Ker(p1 : 2 pr2 |UX2 : UX2 UX1 gives rise to a well-dened, up to 1\2 -conjugacy, cuspidal inertia group 1\2 . Then: (i) Write for diagonal divisor in X2 . Let I D be a pair of inertia and decomposition groups associated to . Then: (a) the cuspidal inertia groups 1\2 corresponding to the cusp determined by are contained in 2/1 and coincide with the cuspidal inertia groups 2/1 corre2 = 1\2 sponding to the cusp determined by , as well as with the 2 -conjugates of I ; (b) either p1 or p2 determines [the nal nontrivial arrow in] an exact sequence 1 I D 1 1; (c) we have D = C2 (I ). (ii) Let x X1 (k ) = X (k ) be a cusp of X log . Let us think of x, Fx as elements ver of D1 , D2 , respectively [cf. Proposition 1.3, (i)]. Then: (a) the major cuspidal component Fx at x is equal to the closure in X2 of the divisor of UX2 determined by 1 pr 1 (x); (b) Ix = Dx ; (c) IFx is a cuspidal inertia group 1\2 associated to the cusp UFx of the family of hyperbolic curves pr2 |UX2 : UX2 UX1 ; (d) DFx = DI Fx ; 1\2 = IFx ; (f ) DFx = C2 (DFx ). (e) DFx (iii) Let be a non-inner permutation automorphism of 2 , AutFC (2 ). Then = 1 AutFC (2 ). Proof. The content of (a), (b) of assertion (i) follows immediately from the denitions involved; (c) follows immediately from (b), together with the fact that I is
def

24

SHINICHI MOCHIZUKI

commensurably terminal in either 2/1 or 1\2 [cf. [20], Proposition 1.2, (i)]. Next, we consider assertion (ii). First, let us observe that (a), (b) are immediate from the denitions; (c) follows immediately from the denitions and (a); (d) follows immediately from (b) [cf. Proposition 1.3, (iii)]. To verify (e), let us rst observe log log that it follows immediately from the geometry of the morphism prlog 2 : X2 X1 that p2 (IFx ) = {1}; thus, (e) follows [in light of (d)] from Proposition 1.3, (iii), (a), (e). Finally, since Ix is commensurably terminal in 1 [cf. [20], Proposition 1.2, (ii)], (f) follows immediately from (d) and Proposition 1.3, (iii), (d), (e), (f). This completes the proof of assertion (ii). Finally, we consider assertion (iii). It is immediate from the denitions that Aut(2 ) is F-admissible. Moreover, it follows immediately from Proposition 1.2, (iii), together with the C-admissibility of , that induces a bijection of the collection of cuspidal inertial groups of the 1 . Thus, it suces to verify that induces a bijection of the quotient p1 : 2 collection of cuspidal inertial groups of 2/1 , i.e., that induces a bijection of the collection of cuspidal inertial groups of 1\2 . But in light of assertions (i) and (ii), (c), this follows immediately from the FC-admissibility of and Proposition 1.3, (vi). This completes the proof of assertion (iii). Proposition 1.7. (Inertia and Decomposition Groups of Minor Cuspidal Components) In the notation of Proposition 1.6, suppose further that x X1 (k ) = X (k ) is a cusp of X log . Let us think of x, Ex as elements of ver D1 , D2 , respectively [cf. Proposition 1.3, (i)]. Then: (a) DEx = DI Ex ; (b) 1\2 = {1}; (c) DEx = C2 (DEx ); (d) for any open subgroup J Ex , IEx Z2 (J ) = IEx ; (e) DEx = C2 (Ex ). Proof. First, we observe that the equality of (a) (respectively, (c)) follows by a similar argument to the argument applied to prove Proposition 1.6, (ii), (d) (respectively, 1.6, (ii), (f)); (b) follows immediately from the geometric fact that the inverse image via pr2 : X2 X1 of the closed point x contains the divisor Ex with multiplicity one. Next, let us consider (d). First, let us observe that, in the notation of Proposition 1.6, (i), the diagonal divisor intersects Ex transversely; in particular, [for appropriate choices of conjugates] we have I Ex . Thus, Z2 (J ) Z2 (J I ) C2 (I ) = D [cf. Proposition 1.6, (i), (c)]. On the other hand, note that p2 (Ex ) is a cuspidal inertia group i.e., Ix of 1 associated to x [cf. Proposition 1.3, (iii), (b)], hence commensurably terminal in 1 [cf. [20], Proposition 1.2, (ii)]. Thus, the inclusion Z2 (J ) D implies [for appropriate choices of conjugates] that p1 (Z2 (J )) = p2 (Z2 (J )) Ix , so the desired equality Z2 (J ) = IEx follows immediately from Proposition 1.3, (iii), (e), (f), together with the fact that Ex is slim [cf. Remark 1.1.1]. This completes the proof of (d). Now it follows immediately from (d) that C2 (Ex ) N2 (IEx ). Thus, in light of (a), we conclude from Proposition 1.3, (iii), (e), that C2 (Ex ) C2 (DEx ), so (e) follows immediately from (c). For i, j {1, 2, 3} such that i < j , write
log log prlog ij : X3 X2

COMBINATORIAL CUSPIDALIZATION

25

log for the projection to the factors labeled i and j of X3 which we think of as log ; prij : X3 X2 corresponding, respectively, to the factors labeled 1 and 2 of X2 for the underlying morphism of schemes; and pij : 3 2 for the surjection induced by prlog ij . Also, for i {1, 2, 3}, write log log : X3 X1 prlog i log for the projection to the factor labeled i of X3 ; pri : X3 X1 for the underlying morphism of schemes; pi : 3 1 for the surjection induced by prlog i .

Denition 1.8. Write U = UX ; V U k U for the diagonal [so we have a natural isomorphism V U ]; V log for the log scheme obtained by equipping log [where we recall that we have a V with the log structure pulled back from X2 log be a tripod over k . natural immersion U k U X2 ]. Let P
log log (i) The morphism of log schemes prlog 12 : X3 X2 determines a structure of family of pointed stable curves on the restriction X3 |V V of pr12 to V . Moreover, X3 |V consists of precisely two irreducible components FV , EV which we refer to, EV respectively, as major cuspidal and minor cuspidal. Here, the intersection FV is a node V : V X3 |V ; either pr13 or pr23 induces an isomorphism FV V k X over V ; the natural projection EV V is a P1 -bundle; the three sections of EV V given by V and the two cusps of X3 |V V that intersect EV determine a unique isomorphism EV V k P over V [i.e., such that the three sections of EV V correspond to the cusps of the tripod, which we think of as being labeled by these three sections]. Write (V k UP =) W EV for the open subscheme given by the complement of these three sections; W log for the log scheme obtained by log via the natural inclusion equipping W with the log structure pulled back from X3 W EV X3 |V X3 . Thus, we obtain a natural morphism of log schemes W log V log .

def

(ii) For x U (k ), denote the bers relative to pr1 over x by means of a subscript x; write Y log Spec(k) for the smooth log curve determined by the hyperbolic curve U \{x}, y Y (k ) for the cusp determined by x. Thus, we have a natural log )x Y2log [cf. [24], Remark 2.1.2]; this isomorphism allows one isomorphism (X3 to identify 3/1 with the 2 associated to Y log [cf. [24], Proposition 2.4, (i)]. log Relative to this isomorphism (X3 )x Y2log , FV |x , EV |x may be identied with the irreducible components Fy , Ey of Denition 1.4, (i), applied to Y log , y [in place of X log , x]. In particular, we obtain major and minor verticial subgroups FV 3/2 , EV 3/2 [i.e., corresponding to the Fy , Ey of Denition 1.4, (ii)]. Proposition 1.9. (Minor Cuspidal Components in Three-Dimensional Conguration Spaces) In the notation of Denition 1.8, let us think of V , W as hor ver , D3 , respectively, and suppose that p12 (DW ) = DV [cf. Proposielements of D2 tion 1.3, (i), (iii)]. Then:

26
def

SHINICHI MOCHIZUKI

(i) Write JW = ZDW (EV ). Then: (a) p12 induces an isomorphism JW DV ; (b) the inclusions JW DW , EV DW induce an isomorphism JW EV DW ; (c) p1 determines natural exact sequences 1 IW JW 1 1, 1 IV DV 1 1, which are compatible with the isomorphisms IW IV , JW DV induced by p12 . (ii) For any open subgroup J EV , we have: Z3 (J ) = JW . (iii) We have: C3 (EV ) = DW . Proof. Since EV = tripod is center-free [cf. Remark 1.1.1], assertion (i) follows immediately from the isomorphism of log schemes W log V log k UP induced by the isomorphism of schemes W V k UP and the morphism of natural log schemes W log V log [cf. Denition 1.8, (i)]. Next, we consider assertion (ii). Since p1 induces a surjection JW 1 , and it is immediate that JW Z3 (J ), it suces to verify that JW 3/1 = Z3/1 (J ). But this follows from 3/1 = Z3 (J ) Proposition 1.7, (d) [cf. the discussion of Denition 1.8, (ii)]. In a similar vein, since p1 induces a surjection DW 1 , and it is immediate that DW C3 (EV ), in 3/1 = C3/1 (EV ). order to verify assertion (iii), it suces to verify that DW But this follows from Proposition 1.7, (e). This completes the proof of Proposition 1.9. Corollary 1.10. (Outer Actions on Minor Verticial Subgroups) Suppose that n 2. Then the subquotient n1/n2 of n may be regarded [cf. [24], Proposition 2.4, (i)] as the pro- fundamental group i.e., 1 of the geometric log log generic ber Z log of the morphism Xn 1 Xn2 [which we think of as the projection obtained by forgetting the factor labeled n 1]; the subquotient n/n2 may then be thought of [cf. [24], Proposition 2.4, (i)] as the pro- fundamental group of 2-nd log conguration space i.e., 2 associated to Z log . In particular, any cusp x of Z log determines, up to n/n2 -conjugacy, a minor verticial subgroup i.e., an isomorph of tripod Ex n/n1 . Then: (i) Any AutFC (n )cusp [cf. Denition 1.1, (v)] stabilizes the n/n2 conjugacy class of Ex . (ii) The commensurator and centralizer of Ex in n satisfy the relation Cn (Ex ) = Zn (Ex ) Ex . In particular, for any open subgroup J Ex , we have Zn (J ) = Zn (Ex ). (iii) By applying (i), (ii), one obtains a natural homomorphism OutFC (n )cusp OutFC (Ex ) and hence a natural outer homomorphism OutFC (n )cusp OutFC (tripod ), associated to the cusp x of Z log . Proof. In light of the superscript cusp and the FC-admissibility of [cf. Remark 1.1.5], assertion (i) follows immediately from the respd portion of Proposition 1.3,

COMBINATORIAL CUSPIDALIZATION

27

(iv). Next, we consider assertion (ii). First, let us recall that Ex is commensurably terminal in n/n1 [cf. Proposition 1.5, (i)]. On the other hand, it is immediate from the denitions that Cn (Ex ) Nn (Cn/n1 (Ex )). Thus, we conclude that Cn (Ex ) = Nn (Ex ). In particular, to complete the proof of assertion (ii), it suces [since Ex is slim cf. Remark 1.1.1] to verify that (*) the natural outer action of Nn (Ex ) on Ex is trivial. Now let j {1, . . . , n 1} be the smallest element m {1, . . . , n 1} such that x log log Xm corresponds to a cusp of the geometric generic ber of the morphism Xm 1 [which we think of as the projection obtained by forgetting the factor labeled m].
log [Here, we write X0 = Spec(k).] Now if j = 1, then by applying the projection 2 determined by the factors labeled 1, n, we conclude that (*) follows from n Propositions 1.3, (iii), (e); 1.7, (a), (e). In a similar vein, if j 2, then by applying 3 determined by the factors labeled j 1, j , n, we conclude the projection n that (*) follows from Proposition 1.9, (i), (b); 1.9, (iii). This completes the proof of assertion (ii). def

Finally, we observe that assertion (iii) follows immediately from assertions (i), (ii), by choosing some isomorphism Ex tripod [which is determined only up to composition with an element of AutFC (tripod )] that is compatible with the cuspidal inertia groups. That is to say, if AutFC (n )cusp , then by assertion (i), 0 ((Ex )) = Ex for some n -inner automorphism 0 of n . Since 0 is uniquely determined up to composition with an element of Nn (Ex ), it follows from assertion (ii) that the outer automorphism 1 OutFC (Ex ) determined by 0 is uniquely determined by . Moreover, one veries immediately that the assignment 1 determines a homomorphism OutFC (n )cusp OutFC (Ex ), hence an outer homomorphism OutFC (n )cusp OutFC (tripod ), as desired. Denition 1.11. (i) In the situation of Denition 1.1, (vi), let us write OutFC (tripod ) and OutFC (tripod )
+ def

= OutFCS (tripod ) = OutFC (tripod )S OutFC (tripod )

for the subgroup given by the image of OutFC (tripod )S via the natural homo2 morphism OutFC (tripod ) OutFC (tripod ) induced by the standard surjection 2 1 tripod tripod 1 . 2 (ii) Now let us return to the case of arbitrary (g, r ); suppose that n 2. Then let us write OutFC (n ) + OutFC (n ) OutFC (n )cusp for the subsets [which are not necessarily subgroups!] given by the unions of the respective inverse images of OutFC (Ex ) + OutFC (Ex ) OutFC (Ex ) via

28

SHINICHI MOCHIZUKI

the natural homomorphism OutFC (n )cusp OutFC (Ex ) associated in Corollary 1.10, (iii), to a cusp x [as in loc. cit.], as x ranges over all cusps as in loc. cit. Remark 1.11.1. It is shown in [5] [cf. Corollary 4.2, (i), (ii), below; Remark 4.2.1 below; [5], 0.1, Main Theorem, (b)] that OutFC (tripod ) + may be identied with the Grothendieck-Teichm uller group. Thus, one may think of the FC + set Out (n ) of Denition 1.11, (ii), as the set of outer automorphisms of Grothendieck-Teichm uller type. Corollary 1.12. Then: (Injectivity for Tripods) Suppose that X log is a tripod.

(i) The natural inclusion 2 AutIFC (2 ) is an isomorphism. (ii) The natural homomorphism OutFC (2 ) OutFC (1 ) induced by p1 : 2 1 is injective.

(iii) We have: OutFCP (2 ) = OutFC (2 ). Proof. First, we observe that assertion (ii) follows formally from assertion (i) and Proposition 1.2, (iii). Next, we observe that assertion (iii) follows formally from assertion (ii) and Propositions 1.2, (iii); 1.6, (iii). Thus, to complete the proof of Corollary 1.12, it suces to verify assertion (i). To this end, let AutIFC (2 ). Let us assign the cusps of X log the labels a, b, c. Note that the labels of the cusps of X log induce labels a, b, c for three of the cusps of the geometric generic ber of the morphism UX2 UX1 determined by pr1 ; assign the fourth cusp of this geometric generic ber the label . Since AutIFC (2 ), it follows that induces [relative to p1 or p2 ] the identity permutation of the conjugacy classes of cuspidal inertia groups of 1 . Since cuspidal inertia groups associated to may be characterized by the property that they are contained in 2 , we thus conclude that induces the identity permutation of the conjugacy classes of cuspidal inertia groups of 2/1 . Now let us x a cuspidal inertia group Ia 2/1 associated to the cusp labeled a. Thus, (Ia ) = Ia 1 , for some 2/1 . Since AutIFC (2 ), and Ja = p2 (Ia ) is normally terminal in 1 [cf. [20], Proposition 1.2, (ii)], it thus follows that p2 ( ) Ja , so [by replacing by an appropriate element Ia ] we may 1\2 = 2 . Thus, by replacing assume without loss of generality that 2/1 by the composite of with a 2 -inner automorphism, we may assume without loss of generality that (Ia ) = Ia . By [20], Proposition 1.5, (i), it follows that there exists a unique [i.e., among its 2/1 -conjugates] major verticial subgroup Fb at b (respectively, Fc at c) such that Ia Fb (respectively, Ia Fc ). By the nonrespd portion of Proposition 1.3, (iv) [which is applicable since AutIFC (2 )! cf. Remark 1.13.2 below], we thus conclude that (Fb ) = Fb , (Fc ) = Fc .
def

COMBINATORIAL CUSPIDALIZATION

29

Since AutIFC (2 ), and p2 induces isomorphisms Fb 1 , Fc 1 [cf. Denition 1.4, (ii)], we thus conclude that is the identity on Fb , Fc . On the other hand, it follows immediately for instance, by considering the well-known geometry of loops around cusps of the complex plane with three points removed [cf. Lemma 1.13; Fig. 1 below] that 2/1 is topologically generated by Fb , Fc . Thus, we conclude that induces the the identity on 2/1 . But since the extension 1 2/1 2 1 1 induced by p1 may be constructed naturally from the 1 cf. 0; Remark 1.1.1], resulting outer action of 1 on 2/1 [i.e., as 2/1 we thus conclude that is the identity. This completes the proof of assertion (i).
out

F b b a

F c c

Fig. 1: The geometry of a tripod equipped with a fourth cusp . The following result is well-known. Lemma 1.13. (Topological Generation by Loops around Cusps) In the notation of the proof of Corollary 1.12, the compatible inclusions Ia Fb 2/1 , Ia Fc 2/1 determine an isomorphism lim Fb Ia Fc 2 /1

30

SHINICHI MOCHIZUKI

where the inductive limit is taken in the category of pro- groups. In particular, 2/1 is topologically generated by Fb , Fc . Proof. In the following, we shall denote the usual topological fundamental group top (). We may assume without loss of generality that k is the eld C of by 1 complex numbers. Then, as is well-known, the topology of a stable curve may be understood from the point of view of pants decompositions [cf., e.g., [1], Chapter 2] as the result of collapsing various partition curves on a hyperbolic Riemann surface to points [which form the nodes of the stable curve]. In particular, in the case of interest, one obtains that Fb 2/1 , Fc 2/1 may be described in the following fashion: Write V for the Riemann surface obtained by removing the points {0, 3, 3} from the complex plane C. Write D+ (respectively, D ) for the intersection with V of the open disc of radius 3 centered at 1 (respectively, 1). Note that V is equipped with a holomorphic automorphism : V V given by multiplication by 1; (D+ ) = D , (D ) = D+ . Let us think of 3, 0, 3 as corresponding, respectively, to the cusps b, a, c. Then we may think of 2/1 as the top (V ) and of Fb 2/1 as corresponding, at least up to pro- completion of 1 top top (D ) 1 (V ). By transport of 2/1 -conjugacy, to the pro- completion of 1 structure via , we then obtain that we may think of Fc 2/1 as corresponding, top top (D+ ) 1 (V ). As at least up to 2/1 -conjugacy, to the pro- completion of 1 in the proof of Corollary 1.12, we may rigidify the various conjugacy indeterminacies top top top by taking the basepoints of 1 (V ), 1 (D+ ), and 1 (D ) to be the point i C and taking Ia 2/1 to correspond to the subgroup topologically generated by the top element of 1 (V ) determined by the circle a of radius 1 centered at a [i.e., 0], oriented counterclockwise [so a D+ D ]. Thus, if one takes b (respectively, c ) to be a loop in V , oriented counterclockwise, given by a slight deformation of the path obtained by traveling from i to b (respectively, c) and then back to i along the line segment from i to b (respectively, c), then b D , c D+ . Moreover, as is well-known from the van Kampen theorem in elementary algebraic topology [cf. also the more combinatorial point of view discussed in the proof of Proposition top top D ) is naturally isomorphic to the inductive limit, (V ) = 1 (D+ 1.5, (iii)], 1 in the category of groups, of the diagram
top 1 (D )

top 1 (D+

D )

top 1 (D+ )

top top D ) (D ) is generated by a and b , 1 (D+ where we observe that 1 top is generated by a , and 1 (D+ ) is generated by a and c . Thus, Lemma 1.13 follows by passing to pro- completions.

Remark 1.13.1. In the notation of Corollary 1.12 and its proof, we observe that the isomorphism of Lemma 1.13 suggests that it may be possible to verify that the natural injection OutFC (2 ) OutFC (1 ) of Corollary 1.12, (ii), is surjective [hence an isomorphism] via the following argument: Let 1 AutFC (1 ). Then it suces to verify that 1 arises [via p1 ] from

COMBINATORIAL CUSPIDALIZATION

31

an element of AutFC (2 ). Fix a rigidied triple F b Ia F c

as in the proof of Corollary 1.12. Let us assume, for simplicity, that 1 (Ja ) = Ja [where we recall that Ja = p2 (Ia )]. Next, let us observe that p2 induces isomor phisms Fb 1 , Fc 1 which coincide on Ia Fb , Ia Fc . Thus, it follows formally from the isomorphism of Lemma 1.13 that there exists a unique automorphism 2/1 of 2/1 that is compatible, relative to p2 , with the automorphism 1 of 1 . In particular, 2/1 constitutes a natural candidate for [the restriction to 2/1 of] a lifting of 1 to AutFC (2 ). On the other hand, unfortunately, it is not clear whether or not 2/1 , constructed in this way, stabilizes the 2/1 -conjugacy class of the cuspidal inertia groups associated to the cusp . In particular, this argument alone is not sucient to construct a lifting of 1 to AutFC (2 ) from 2/1 . Remark 1.13.2. Another [perhaps more fundamental!] problem with the approach proposed in Remark 1.13.1 is the following. If one already knows that 1 AutFC (1 ) arises [via p1 ] from some 2 AutFC (2 ), then one wishes for the explicit construction of 2/1 that is applied to give rise to the outer automorphism of 2/1 obtained by restricting 2 to 2/1 . For instance, if 1 is inner, then it arises from a 2 AutFC (2 ) which is inner. Moreover, in order to pass from the 2/1 constructed from an arbitrary 1 AutFC (1 ) by applying the natural isomor1 [cf. 0; Remark 1.1.1], it is of crucial importance for the phism 2 2/1 2/1 to be a homomorphism which yields the restriction explicit construction 1 to 2/1 of an inner lifting to AutFC (2 ) when applied to an inner 1 . On the other hand, if 1 is a non-trivial inner automorphism of 1 , then [as is easily veried] corresponding to the there do not exist cuspidal inertia groups Jb , Jc tripod 1 cusps labeled b, c such that 1 (Ja ) = Ja , 1 (Jb ) = Jb , 1 (Jc ) = Jc . In particular, in the case of such an arbitrary inner 1 , one may not apply the non-respd portion of Proposition 1.3, (iv), to conclude that the 2/1 -conjugacy classes of major and minor verticial subgroups or nexus subgroups of 2/1 are preserved by an inner lifting 2 . Instead, one may only apply the respd portion of Proposition 1.3, (iv), to conclude that the 2 -conjugacy classes of such subgroups are preserved by 2 which is insucient for the execution of the construction of Remark 1.13.1 [i.e., of the proof of Corollary 1.12]. Corollary 1.14. (Modular Symmetries of Tripods) Suppose that X log is a tripod. Let n 2. Then: (i) The outer modular symmetries Out(n ) normalize OutFC (n )cusp . If, moreover, the natural homomorphism OutFC (m ) OutFC (m1 ) induced by m1 is injective for all integers m such that the standard surjection m FCP 2 m n, then we have Out (n ) OutFC (n )S = OutFCS (n ). (ii) Let x be as in Corollary 1.10. Write : n 1 for the standard surjection. Then there exists an outer modular symmetry Out(n ) such that the
out

32

SHINICHI MOCHIZUKI

restriction of : n 1 to Ex n determines an outer isomorphism Ex 1 that is independent of the choice of Ex among its n -conjugates. (iii) Suppose that we are in the situation of (ii). Let OutFC (n )cusp ; |Ex OutFC (Ex ) the result of applying the displayed homomorphism of Corollary
FC 1.10, (iii), to ; = 1 OutFC (n )cusp [cf. (i)]; (1 )cusp 1 Out the outer automorphism of 1 induced by via . [Thus, = whenever OutFCS (n ).] Then |Ex and 1 are compatible with the outer isomorphism Ex 1 of (ii). In particular, if |Ex OutFC (Ex )S , then OutFC (n )S . def

(iv) We have: OutFCS (n ) OutFC (n )

Proof. First, we consider assertion (i). We apply induction on n. First, let us log log observe that relative to the natural isomorphism Xn (M0,n+3 )k [cf. Deni tion 1.1, (vi)], the divisors of Xn that belong to Dn [cf. Proposition 1.3, (vii)] log are precisely the divisors at innity of (M0,n+3 )k whose generic points parametrize stable curves of genus zero with precisely two components, one of which contains precisely two cusps. [Indeed, this follows immediately from the well-known log log geometry of (M0,n+3 )k .] In particular, the automorphisms of (M0,n+3 )k arising from the permutations of the ordering of the cusps permute the divisors that be . Thus, we conclude that the outer modular symmetries Out(n ) long to Dn normalize OutQS (n ) = OutFC (n )cusp [cf. Proposition 1.3, (vii)]. Now let Out(n ) be an outer modular symmetry that arises from a permutation of the subset {a, b, c, 1, 2, . . . , n 1} {a, b, c, 1, 2, . . . , n 1, n} [cf. the notation of OutFC (n )S OutQS (n ) [cf. PropoDenition 1.1, (vi)]; OutFCP (n ) sition 1.3, (vii)]; = 1 OutQS (n ). Then since is compatible with the standard surjection n n1 , it follows from the induction hypothesis that , map to the same element OutQS (n1 ) via the natural homomorphism OutQS (n ) OutQS (n1 ) induced by this surjection. Thus, we conclude from the injectivity condition in the statement of assertion (i) [cf. also Proposition 1.3, (vii)] that = . Since the group of all permutations of the set {a, b, c, 1, 2, . . . , n 1, n} is generated by the subgroups of permutations of the subsets {a, b, c, 1, 2, . . . , n 1} {a, b, c, 1, 2, . . . , n 1, n} and {1, 2, . . . , n 1, n} {a, b, c, 1, 2, . . . , n 1, n}, we thus conclude that OutFCS (n ). This completes the proof that OutFCP (n ) OutFC (n )S OutFCS (n ); the opposite inclusion follows immediately from the denitions. This completes the proof of assertion (i). In light of Corollary 1.10, (ii), assertions (ii) and (iii) follow immediately log log from the denitions and the well-known geometry of Xn [i.e., (M0,n+3 )k ]. Finally, we consider assertion (iv). By assertion (iii), it follows that the image of the restriction OutFCS (n ) OutFC (Ex ) to OutFCS (n ) of the natural homo2 , morphism of Corollary 1.10, (iii), lies in OutFC (Ex ) . Write : n : 2 1 [so = ] for the standard surjections. Then the existence of the factorization = ( ) : n 2 1 which is compatiFCS (n ) implies that the image of the homomorphism ble with elements of Out
def

COMBINATORIAL CUSPIDALIZATION

33

OutFCS (n ) OutFC (Ex ) in fact lies in OutFC (Ex ) + . This implies the desired inclusion OutFCS (n ) OutFC (n ) + and hence completes the proof of assertion (iv).

Section 2: Injectivity for Degenerating Ane Curves In the present 2, we generalize [cf. Corollary 2.3, (ii)] the injectivity asserted in Corollary 1.12, (ii), to the case of arbitrary X log such that UX is ane, by considering what happens when we allow X log to degenerate. Let k = k be as in 1; def R = k [[t]] i.e., the ring of power series with coecients in k ; K the quotient eld of R ; def K an algebraic closure of K ; = Spec(K ); R the integral closure of R in K ; def def log , S log the log schemes obtained by equipping S = Spec(R ), S = S Spec(R), respectively, with the log structures determined by the nonzero regular functions; def log = Spec(k ) S S ; slog log def log s = Spec(k) S S . Here, we wish to think of k as the residue eld of R. Next, let
log log S X def

be a stable log curve of type (g, r ) [whose restriction to US is a smooth log curve];
log log S log S S log X log = X log log log X = X S log s slog s ; def log log log Xs = X S log s slog def def

log log log for the result of base-changing via the morphisms S log S , slog S , s log log S . Thus, we are in a situation as discussed in 0. By ordering the cusps of X , log log Mg,r . If n is a positive integer, then we obtain a classifying [1-]morphism S

by pulling back the natural [1-]morphism Mg,r+n Mg,r obtained by forgetting the last n points via this classifying morphism, we thus obtain a log conguration space log log Xn S i.e., whose restriction to US is a log conguration space as in [24], Denition 2.1, (i). We shall write
log S log ; Xn log log Xn s s ; log Xn,s slog

log

log

34

SHINICHI MOCHIZUKI

log log log for the result of base-changing Xn to S log , slog . Thus, we may S , or s log S log for arbitrary n. Also, we may apply the apply the discussion of 0 to Xn theory of 1 by taking log def log = Xn S Xn, log to be the Xn S of 1; this results in a n of the form log n = 1 (Xn, ) def

to which we may apply the specialization isomorphisms discussed in 0. For i = 1, 2, write

for the projection to the factor labeled i, pri : X2 X1 for the underlying morphism of schemes, and pi : 2 1 for the surjection induced by prlog i . Denition 2.1. Let iX {1, 2}. Suppose that Xs is singular and has iX irreducible components, one of which we shall denote T ; if iX = 2, then we shall write Q for the other irreducible component of Xs . Write UT T (respectively, [when iX = 2] UQ Q) for the complement in T (respectively, [when iX = 2] log Q) of the nodes and cusps of Xs relative to the log structure of Xs . Suppose log further that UT is a tripod. Let x X (S ) be a cusp of X whose restriction xs Xs (s) X (k ) to s lies in T ( Xs ) [cf. Remark 2.1.1 below].
log (i) Observe that the log structure on X2 determines on the ber (X2 )xs of the morphism pr1 : X2 X1 (= X ) over xs X (k ) a structure of pointed stable curve, which consists of iX + 1 irreducible components, iX of which which we map isomorphically to T Xs and [when and [when iX = 2] Q shall denote T iX = 2] Q Xs , respectively, via pr2 : X2 X1 = X , the (iX + 1)-th of which x maps to the point xs Xs (s) via pr . Let us refer to which we shall denote E 2 x as T and [when iX = 2] Q as the sub-major cuspidal components at xs and to E (respectively, the sub-minor cuspidal component at xs . Thus, the complement in T [when iX = 2] Q; Ex ) of the nodes and cusps [relative to the pointed stable curve (respectively, [when iX = 2] Q ; E x ) which we shall structure on (X2 )xs ] of T refer to as the interior UT of T (respectively, [when iX = 2] UQ of Q; UE x of Ex ) determines a tripod UT (respectively, [when iX = 2] hyperbolic curve UQ ; tripod UE x ). Moreover, pr2 induces isomorphisms UT UT , [when iX = 2] UQ UQ ; we have a diagram [cf. also Fig. 2 below]

x E

is to be omitted if iX = 1; we refer to the unique node x where the nal Q of (X2 )xs that lies over xs Xs (s) [via pr2 ] as the sub-nexus at xs and to each of the remaining [one or two] nodes x of (X2 )xs as the internal nodes at x.
log , we obtain (ii) On the other hand, by applying Denition 1.4 to Xn, major and minor cuspidal components at x [i.e., the restriction x X ( ) of x to

COMBINATORIAL CUSPIDALIZATION

35

], as well as a nexus at x which we shall denote Fx , Ex (X2 )x , x . Write F x , E x , x for the closures of Fx , Ex , x in (X2 )x = X2 X1 S [where the ber product is taken with respect to the morphisms pr1 : X2 X1 , x : S X1 = X ]. F x , [when iX = 2] Q F x, E x E x , x x . Write Thus, we have T UF x F x ; UE x E x
def

for the open subschemes given by the complements of the closures of the nodes and cusps of Fx , Ex . Thus, UE x is a family of tripods over S ; pr2 determines an open immersion UF x X whose image is the complement of the cusps of X [relative to the log structure of X log ]. (iii) Write T T for the normalization of T ; T log for the log scheme obtained by equipping T with the log structure determined by the closed points of T that map to points of T \UT . Thus, UT is a tripod over s; we have a natural isomorphism log s for the n-th log conguration (T ) UT UT ( T Xs ). Write Tn space associated to UT [cf. 0]. Thus, we have a natural commutative diagram T2 X2,s pr pr
1

Xs

where, by abuse of notation, we write pri : T2 T1 = T for the projection to the factor labeled i [for i = 1, 2]; we write pri : X2,s X1,s = Xs for the restriction to the bers over s of pri : X2 X1 [for i = 1, 2]; the horizontal arrows restrict to immersions on UT2 , UT ; the lower horizontal arrow is compatible with the natural isomorphism (T ) UT UT ( T Xs ). Write (T2 )xs for the ber of pr1 : T2 T1 over the point xs , where, by abuse of notation, we write xs for the point T (s) determined by xs Xs (s). Then (T2 )xs has precisely two (X2 )x x (X2 )x , T irreducible components which map isomorphically to E s s so (T2 )xs may be thought of as consisting of a diagram x E x T

via the natural morphism T2 X2,s . By abuse of notation, we shall also use for the corresponding irreducible components of (T2 )x . Write x , T the notation E s
log tripod = 1 (Tn ). n def

(iv) By applying the specialization isomorphisms [cf. 0] associated to the log log log log restriction of prlog S the 1 : X2 X1 to the result of base-changing via S completion of X1 = X along the cusp of X determined by x, we conclude that the pointed stable curve structure on (X2 )xs [cf. (i)] determines a semi-graph of

36

SHINICHI MOCHIZUKI

anabelioids of pro- PSC-type as discussed in [20], Denition 1.1, (i) [cf. also the discussion of [18], Appendix] whose associated PSC-fundamental group may be identied with 2/1 . In particular, we obtain [conjugacy classes of] subgroups [cf. [20], Denition 1.1, (ii)] T , Q , E x , x , x 2/1 [where Q is to be omitted if iX = 1] corresponding to the sub-major and sub-minor cuspidal components, as well as to the sub-nexus and the internal node(s) which we shall refer to as sub-major verticial, sub-minor verticial, sub-nexus, and internal nodal, respectively. In a similar [but simpler] vein, by applying the specialization isomorphisms [cf. 0] associated to X log S log , we obtain [conjugacy classes of] subgroups T , Q 1 [where Q is to be omitted if iX = 1] such that the morphism p2 : 2 1 determines isomorphisms T T ;

Q Q

[where the second isomorphism is to be omitted if iX = 1] i.e., the sub-major verticial subgroups may be thought of as dening sections of the projection p2 : 2 1 over T , [when iX = 2] Q . On the other hand, p2 maps E x onto a cuspidal inertia group of 1 associated to x; in particular, p2 (E x ) is abelian. Finally, we observe that for suitable choices within the various conjugacy classes involved, we have natural inclusions E x x T x Q

[where Q is to be omitted if iX = 1] inside 2/1 .

Fx , Ex , x 2/1 associated to Fx , Ex , x [cf. (ii)] such that p2 determines an isomorphism Fx 1 . For suitable choices within the various conjugacy classes involved, we have natural inclusions Ex x Fx ; [inside 2/1 ], as well as natural inclusions T , Q F x

induced by the natural immersions UT UF x , UQ UF x [where Q , UQ UF x are to be omitted if iX = 1] by applying the isomorphisms
log 1 ((UF x X X log ) S s) 1 (Xs ) 1 (X log ) 1 (UF x X X log )

COMBINATORIAL CUSPIDALIZATION

37

[arising from the log purity theorem and the specialization isomorphism for X log (UF x X X log ) 1 (UFx ) Fx [the S log ], together with the isomorphisms 1 rst of which arises from the log purity theorem]. In a similar [but simpler] vein, we have equalities [of 2/1 -conjugacy classes of subgroups of 2/1 ] E x = Ex ; x = x induced by the natural immersion UE x UE x by applying the isomorphism 1 (UE x S s) 1 (UE x ) [arising from the log purity theorem and the specialization isomorphism for the smooth log curve determined, up to unique isomorphism, by the (UE x ) 1 (UEx ) family of tripods UE x S ], together with the isomorphisms 1 Ex [the rst of which arises from the log purity theorem]. (vi) One veries immediately that the natural commutative diagram of (iii) determines a natural morphism of exact sequences of pronite groups 1 tripod 2 /1 2 /1 tripod 2 2 tripod 1 1 1

where the vertical arrows are injective outer homomorphisms; the image of the vertical morphism on the right is equal to T . By abuse of notation, we shall write tripod (respectively, tripod ; tripod ) for the subgroup, well-dened up to 2 1 2 /1 2/1 - (respectively, 2 -; 1 -) conjugacy, determined by the image of the left-hand (respectively, middle; right-hand) vertical arrow. Thus, for suitable choices within the various conjugacy classes involved, we have natural inclusions E x , [inside 2/1 ]. Remark 2.1.1. One veries immediately that data as in Denition 2.1 exists for arbitrary (g, r ) such that (g, r ) = (0, 3) and r 1. Moreover, the case iX = 1 corresponds precisely to the case where (g, r ) = (1, 1). Proposition 2.2. (First Properties of Sub-major and Sub-minor Verticial Subgroups) In the notation of Denition 2.1:
tripod are com(i) T , [when iX = 2] Q , E x , x , x , Fx , Ex , x , 2/1 mensurably terminal in 2/1 ; T , [when iX = 2] Q are commensurably terminal in 1 .

T ,

tripod 2/1

(ii) Suppose that one xes x 2/1 among its various 2/1 -conjugates. Then the condition that there exist inclusions/equalities x Ex ; E x = Ex ; x = F x x T
tripod E x , T 2 /1

38

SHINICHI MOCHIZUKI

tripod completely determines Ex , among their var , F x , E x , and 2/1 x , T ious 2/1 -conjugates. tripod (iii) In the notation of (ii), the compatible inclusions x 2/1 , x E tripod 2/1 , x Ex 2/1 , x Fx 2/1 , determine isomorx T phisms tripod lim x 2 /1 x T E

lim Ex x Fx 2/1 where the inductive limits are taken in the category of pro- groups. Proof. Assertion (i) follows from [20], Proposition 1.2, (ii). Assertion (ii) follows from the fact that every nodal edge-like subgroup is contained in precisely two verticial subgroups [cf. [20], Proposition 1.5, (i)], together with the fact that tripod is topologically generated by E x , T [cf. assertion (iii)]. Assertion (iii) 2 /1 follows by a similar argument to the argument applied in the proof of Proposition 1.5, (iii).

tripod
2/1

.. x * x .. Ex .. T

.. x

.. Q

Fig. 2: A degenerating ane curve equipped with an extra cusp .

COMBINATORIAL CUSPIDALIZATION

39

Corollary 2.3. (Injectivity for Non-tripod Degenerating Ane Curves) In the notation of Denition 2.1 [cf. also Denition 1.1; Remark 2.1.1]: (i) The natural inclusion 2 AutIFC (2 ) is an isomorphism. (ii) The natural homomorphism OutFC (2 ) OutFC (1 ) induced by p1 : 2 1 is injective.

(iii) We have: OutFCP (2 ) = OutFC (2 ). Proof. First, we observe that assertion (ii) follows formally from assertion (i) and Proposition 1.2, (iii). Next, we observe that assertion (iii) follows formally from assertion (ii) and Propositions 1.2, (iii); 1.6, (iii). Thus, to complete the proof of Corollary 2.3, it suces to verify assertion (i). To this end, let AutIFC (2 ). Let us x some x 2/1 among its various 2/1 -conjugates; let Ex , , x , T tripod F x , E be as in Proposition 2.2, (ii). x , and 2/1 Since AutIFC (2 ), it follows that induces [relative to p1 or p2 ] an automorphism of 1 that stabilizes every cuspidal inertia group of 1 . Thus, by the non-respd portion of Proposition 1.3, (iv), we conclude that stabilizes the 2/1 -conjugacy classes of x = x . In particular, (x ) = x , Fx , Ex = E 1 IFC x , for some 2/1 . Since Aut (2 ), and p2 (x ) is a cuspidal inertia group of 1 associated to x, hence normally terminal in 1 [cf. [20], Proposition 1.2, (ii)], it thus follows that p2 ( ) p2 (x ), so [by replacing by an appropriate element x ] we may assume without loss of generality that 2 /1 1\2 = 2 . Thus, by replacing by the composite of with a 2 -inner automorphism, we may assume without loss of generality that (x ) = x . By Proposition 2.2, (ii), we thus conclude that (Fx ) = Fx , (Ex ) = Ex . Since AutIFC (2 ), and p2 induces an isomorphism Fx 1 [cf. Denition 2.1, (v)], we thus conclude that restricts to the identity on Fx . In particular, it follows tripod is topologically that stabilizes and restricts to the identity on T . Since 2/1 generated by E x = Ex , T [cf. Proposition 2.2, (iii)], we thus conclude that tripod tripod (2/1 ) = 2/1 . Now since AutIFC (2 ), and tripod is normally terminal in 2/1 [cf. Propo2 /1 sition 2.2, (i)], we thus conclude from the commutative diagram of Denition 2.1, tripod (vi) [i.e., by applying the natural isomorphism tripod 2 2 /1
out

tripod cf. 1

0; Remark 1.1.1], that the automorphism of tripod induced by arises from an 2 /1 automorphism tripod Aut(tripod ), which is easily veried to be F-admissible 2 [cf. Proposition 1.2, (i)]. Next, observe that since E x is normally terminal in 2/1 [cf. Proposition 2.2, (i)], it follows immediately from [20], Proposition 1.5, (i), that every cuspidal inertia group of 2/1 that is contained in E x and 2/1 conjugate to a cuspidal inertia group associated to a cusp of UE x is, in fact, equal

40

SHINICHI MOCHIZUKI

to a cuspidal inertia group associated to a cusp of UE x . Since is C-admissible, IFC tripod and Aut (2 ) restricts to the identity on T , we thus conclude that is IFC-admissible, i.e., tripod AutIFC (tripod ). 2 On the other hand, by Corollary 1.12, (i), it follows that tripod lies in the AutIFC (tripod ) [where we write tripod image of the natural inclusion tripod 2 2 2 for the analogue of 2 for tripod ]. In particular, we conclude that induces 2 an inner automorphism of tripod , which 2/1 . Since restricts to the identity on T is center-free [cf. Remark 1.1.1] and normally terminal in tripod [cf. Proposition 2 /1 2.2, (i)], it thus follows that restricts to the identity on tripod 2/1 , hence also on E x = Ex . Since 2/1 is topologically generated by Ex , Fx [cf. Proposition 2.2, (iii)], we thus conclude that restricts to the identity on 2/1 , hence [by applying 1 cf. 0; Remark 1.1.1] that is the the natural isomorphism 2 2/1 identity. This completes the proof of assertion (i). Before proceeding, we recall the following well-known result. Lemma 2.4. (FC-Admissible Permutations of Cusps) There exist elFC ements Out (n ) that induce, relative to the standard surjection n 1 , arbitrary permutations of the set of conjugacy classes of cuspidal inertia groups of 1 [i.e., the set of cusps of X log ]. Proof. One way to verify Lemma 2.4 is by thinking of n as the pro- completion of the topological fundamental group of the n-th conguration space associated to [i.e., the complement of the various diagonals in the product of n copies of] a topological surface X of type (g, r ) [cf. the theory of [24], 7]. Then it is easy to construct a homeomorphism of X that induces an arbitrary permutation of the cusps; one then veries immediately that such a homeomorphism induces a homeomorphism of the n-th conguration space associated to X that gives rise to an element OutFC (n ) satisfying the conditions in the statement of Lemma 2.4. Alternatively, one may give a more log scheme-theoretic proof by means of the objects introduced in the discussion preceding Denition 2.1 as follows. If r 1, then there is nothing to show. Thus, we suppose that r 2. Then [by applying the specialization isomorphisms of 0] it suces to verify the existence of autolog morphisms of Xs over slog that induce arbitrary transpositions [i.e., permutations that switch two elements and leave the remaining elements xed] of the set of cusps log log of Xs . If (g, r ) = (0, 3) [i.e., Xs is a tripod], then the existence of such autolog log morphisms of Xs [over s ] follows immediately from the well-known structure of tripods. Thus, we may assume that (g, r ) = (0, 3). This assumption implies [cf. Remark 2.1.1] that we may suppose that we are in the situation of Denition 2.1, log . and that precisely two of the cusps of the tripod UT arise from cusps a, b of Xs Then [by the case where (g, r ) = (0, 3), which has already been veried] UT admits an automorphism [over s] that switches the two cusps of UT corresponding to a, b and leaves the remaining cusp of UT xed. Moreover, one veries immediately
out

COMBINATORIAL CUSPIDALIZATION

41

log that such an automorphism of UT extends to an automorphism of Xs [over slog ] that switches a and b and restricts to the identity on Q [hence leaves the remaining log xed]. This completes the proof of Lemma 2.4. cusps of Xs

Section 3: Conditional Surjectivity for Ane Curves In the present 3, we prove a certain special case [cf. Corollary 3.3] of the surjectivity portion of our main result [cf. Theorem 4.1 below] for ane hyperbolic curves. The key observation is that the technical obstacles observed, relative to verifying surjectivity, in Remarks 1.13.1, 1.13.2 may be circumvented if one replaces 1 by 3 2 and works with the subset + of Denition 1.11, (ii). 2 We return to the notation of 1 [cf. especially the notation of Denition 1.4 and of the discussion preceding Denition 1.8]. Denition 3.1. Let x X (k ) be a cusp of X log . Write x X2 (k ) for the nexus x [cf. Denition 1.4, (i)].
log (i) Observe that the log structure on X3 determines on the ber (X3 )x of the morphism pr12 : X3 X2 over the point x X2 (k ) a structure of pointed stable curve, which consists of three irreducible components. Of these three irreducible components, there is a unique irreducible component F x which we shall refer to as the quasi-major cuspidal component of (X3 )x that maps isomorphically to X via pr3 : X3 X1 = X ; there is a unique irreducible component Lx which we shall refer to as the link cuspidal component of (X3 )x that intersects F x at a single point; there is a unique irreducible component E x which we shall refer to as the quasi-minor cuspidal component of (X3 )x that intersects Lx at a single point. [Thus, Lx , E x map to the point x X (k ) via pr3 .] The complement in F x (respectively, Lx ; E x ) of the nodes and cusps [relative to the pointed stable curve structure on (X3 )x ] of F x (respectively, Lx ; E x ) which we shall refer to as the interior UF x of F x (respectively, ULx of Lx ; UE x of E x ) determines a hyperbolic curve UF x (respectively, tripod ULx ; tripod UE x ). Moreover, pr3 induces

isomorphisms UF x UX , F x X . (ii) By applying the specialization isomorphisms [cf. 0] associated to the log log restriction of prlog X2 to the completion of X2 along x, we conclude 12 : X3 that the pointed stable curve structure on (X3 )x [cf. (i)] determines a semi-graph of anabelioids of pro- PSC-type as discussed in [20], Denition 1.1, (i) [cf. also the discussion of [18], Appendix] whose associated PSC-fundamental group may be identied with 3/2 . In particular, the quasi-major, link, and quasi-minor cuspidal components determine [conjugacy classes of] verticial subgroups [cf. [20], Denition 1.1, (ii)] F x , Lx , E x 3/2

42

SHINICHI MOCHIZUKI

which we shall refer to as quasi-major, link, and quasi-minor, respectively. Thus, the morphism p3 : 3 1 determines an isomorphism F x 1 i.e., the quasi-major verticial subgroups may be thought of as dening sections of the projection p3 : 3 1 . On the other hand, p3 maps Lx , E x onto cuspidal inertia groups of 1 associated to x; in particular, p3 (Lx ), p3 (Ex ) are abelian. Finally, let us refer to the node x E x Lx (respectively, x Lx F x ) of (X2 )x as the x-minor-nexus (respectively, x-major-nexus) [of (X3 )x ] so [cf. Fig. 3 below] Ex x Lx x F x and to the [nodal] edge-like subgroup [cf. [20], Denition 1.1, (ii)] x 3/2 (respectively, 3/2 )
x

determined up to conjugacy by x (respectively, x ) as an x-minor-nexus (respectively, x-major-nexus) subgroup. Thus, for suitable choices within the various conjugacy classes involved, we have natural inclusions E x [inside 3/2 ]. (iii) We shall refer to B = E x
def

Lx

F x

Lx

(respectively, B = Lx

def

F x)

as the -bridge (respectively, -bridge) of (X3 )x . If the various choices within conjugacy classes are made so that the natural inclusions of (ii) hold, then we shall refer to the subgroup [well-dened up to 3/2 -conjugacy] B 3 /2 (respectively, B 3/2 )

topologically generated by E x and Lx (respectively, by Lx and F x ) as the -bridge subgroup (respectively, -bridge subgroup). (iv) Recall the subgroups IFx DFx 2 (respectively, IEx DEx 2 ) of Proposition 1.6 (respectively, 1.7). By applying the specialization isomorphisms of 0 rst over the completion of Fx (respectively, Ex ) along x, and then over the completion of X2 along the generic point of UFx (respectively, UEx ), we conclude that the outer action of DFx (respectively, DEx ) on 3/2 stabilizes the 3/2 -conjugacy classes of E x , x , and B (respectively, of B , , and F x ). Since, morex

over, E x , x , and B (respectively, of B , , and F x ) are commensurably

terminal in 3/2 [cf. Proposition 3.2, (i), below], it follows that this outer action

COMBINATORIAL CUSPIDALIZATION

43

determines outer actions of DFx (respectively, DEx ) on E x , x , and B (respectively, of B , , and F x ), whose restriction to IFx (respectively, IEx ) is trivial [cf. the theory of specialization isomorphisms reviewed in 0]. Thus, we obtain outer actions of DFx /IFx Fx (respectively, DEx /IEx Ex ) on E x , x , and B (respectively, of B , , and F x ). Since the irreducible component of X3 |UFx (respectively, X3 |UEx ) [where | is taken with respect to pr12 : X3 X2 ] determined by E x (respectively, F x ) descends from UFx (respectively, UEx ) to k i.e., is naturally isomorphic to UFx k E x (respectively, UEx k F x ) we thus conclude that the outer action of Fx (respectively, Ex ) on E x (respectively, on F x ) is trivial. (v) On the other hand, the outer action of Fx on B may be made more explicit, as follows. Write xlog = X log X x. Recall that the geometric bers of log log log over points of UX may be regarded as 2-nd log conguprlog 1 : X3 X1 = X ration spaces associated to the smooth log curves determined by the corresponding log log log [cf. [24], Remark 2.1.2]. In a similar way, even bers of prlog 1 : X2 X1 = X log log though the ber (X2 )xlog of pr1 over xlog is a non-smooth stable log curve, we log log may think of the ber (X3 )xlog of prlog as the 2-nd log conguration 12 over x log space associated to (X2 )xlog i.e., in the sense that it may be obtained as the log log pull-back of the [1-]morphism Mg,r+3 Mg,r+1 [determined by forgetting the last two sections] via the classifying [1-]morphism xlog Mg,r+1 . If we forget the various log structures involved, then it follows from this point of view that the natural inclusion X Fx (X2 )x ts into a natural commutative diagram X2 (X3 )x pr pr
1 x x

def

log

12

(X2 )x

where [by abuse of notation] we use the notation pr12 to denote the appropriate restriction of pr12 . Now one veries immediately [cf. Denition 2.1, (vi)] that this commutative diagram determines a natural morphism of exact sequences of pronite groups 1 2 /1 3 /2 2 1 1

3/1

2/1

where the vertical arrows are injective outer homomorphisms; the image of the vertical morphism on the left is equal to B ; the image of the vertical morphism on the right is equal to Fx . In particular, this commutative diagram of pronite groups allows one to identify the outer action of Fx on B with the outer action of 1 on 2/1 .

44

SHINICHI MOCHIZUKI

(vi) In a similar vein, the outer action of Ex on B may be made more explicit, as follows. Write T log for the smooth log curve over k determined by the log tripod Ex ; Tn for the corresponding n-th log conguration space [where n 1 is an tripod def log = 1 (Tn ). Then just as in (v), we obtain a natural commutative integer]; n diagram T2 (X3 )x pr pr
1 12

(X2 )x

where we use the notation pr12 as in (v). Moreover, just as in (v) [cf. also Denition 2.1, (vi)], this commutative diagram determines a natural morphism of exact sequences of pronite groups 1 tripod 2 /1 3 /2 tripod 2 3/1 tripod 1 2/1 1

where the vertical arrows are injective outer homomorphisms; the image of the vertical morphism on the left is equal to B ; the image of the vertical morphism on the right is equal to Ex . In particular, this commutative diagram of pronite groups allows one to identify the outer action of Ex on B with the outer action of tripod on tripod 1 2/1 . Proposition 3.2. (First Properties of Quasi-Major, Link, and QuasiMinor Verticial Subgroups) In the notation of Denition 3.1: (i) x , , E x , Lx , F x , B , and B , are commensurably terminal x in 3/2 . (ii) Suppose that one xes x 3/2 (respectively, 3/2 ) among its x various 3/2 -conjugates. Then the condition that there exist inclusions x E x ;
x

x Lx ;
x

x B F x )
x

(respectively, B ;

Lx ;

completely determines E x , Lx , B , and B (respectively, B , B , Lx , and F x ) among their various 3/2 -conjugates. (iii) In the notation of (ii), the compatible inclusions x E x B 3/2 , x Lx B 3/2 , Lx B 3/2 , F x B
x x

COMBINATORIAL CUSPIDALIZATION

45

3/2 , determine isomorphisms lim lim lim lim E x x Lx E x x B Lx B

B 3/2 B 3 /2

F x F x

where the inductive limits are taken in the category of pro- groups. (iv) The operation of restriction to the various subgroups involved determines a bijection between the set of outer automorphisms of 3/2 that stabilize the 3/2 -conjugacy classes of x , , E x , Lx , F x , B , and B
x

and the set of pairs Out(B ), Out(B ) such that: (a) (respectively, ) stabilizes the B - (respectively, B -) conjugacy classes of E x , x , Lx , and (respectively, of x , Lx , , and F x ); (b) and induce [cf. (a); (i)] the same element Out(Lx ).
x x

Proof. Assertions (i), (ii), (iii) follow from precisely the same arguments applied to prove assertions (i), (ii), and (iii) of Proposition 1.5. In light of assertions (i), (ii), (iii), assertion (iv) follows, in a straightforward manner, from the fact that Lx is center-free [cf. Remark 1.1.1], together with the fact every nodal edge-like subgroup is contained in precisely two verticial subgroups [cf. [20], Proposition 1.5, (i); [20], Proposition 1.2, (i)], which one applies, when verifying (a) for (respectively, ), rst to (respectively, x ), and then to x (respectively,
x

).
x

Corollary 3.3. (Conditional Surjectivity for Ane Curves) Suppose that X log is of type (g, r ), where r 1. Then OutFC (2 ) + OutFC (2 ) is contained in the image of the natural homomorphism OutFC (3 ) OutFC (2 ) induced by p12 : 3 2 .

Proof. Let 2 OutFC (2 ) + ; 2 AutFC (2 ) an automorphism that lifts 2 . To complete the proof of Corollary 3.3, it suces to construct an 3 AutFC (3 )

46

SHINICHI MOCHIZUKI

that lifts 2 . Write x X (k ) for the cusp that exhibits 2 as an element of OutFC (2 ) + [cf. Denition 1.11, (ii)].

B E x *1 x B
Fig. 3: An ane curve equipped with two extra cusps 1 , 2 . (x is the cusp that corresponds to x) Next, let us x x , Ex , Fx 2/1 as in Proposition 1.5, (ii). By the nonrespd portion of Proposition 1.3, (iv), we may assume without loss of generality that 2 stabilizes x , Ex , and Fx . Write 2/1 = 2 |2/1 AutFC (2/1 ),
FC FC E (Ex ), F (Fx ) for the respective 2/1 = 2 |Ex Aut 2/1 = 2 |Fx Aut FC E (Ex ) restrictions of 2 to 2/1 , Ex , Fx ; 2/1 OutFC (2/1 ), 2 /1 Out + def def def

L x

F x

*2

F 2 /1

Out

FC

(Fx ) for the resulting outer automorphisms.

Ex , Next, let us recall the outer isomorphisms 2/1 B , tripod 1 tripod B implicit [cf. Propositions 1.5, (i); 3.2, (i)] in the natural mor2 /1 phisms of exact sequences of Denition 3.1, (v), (vi). Here, we note that it follows from the denitions that in fact, we have an equality tripod = Ex [i.e., with1 out any indeterminacy with respect to composition with an inner automorphism]. E By conjugating 2/1 , 2 /1 , respectively, by the rst two of these outer isomortripod phisms, we thus obtain elements 3/2 OutFC (B ), 1 OutFC (tripod ) 1 +

COMBINATORIAL CUSPIDALIZATION

47

tripod together with a particular lifting tripod AutFC (tripod ) of 1 . By the def1 1 tripod + ) [cf. Denition 1.11, (i)], it follows that 1 lifts inition of OutFC (tripod 1 tripod tripod S FC to a unique [cf. Corollary 1.12, (ii)] element 2 Out (2 ) . Write tripod tripod tripod FC |tripod determined by the lift2/1 Out (2/1 ) for the restriction 2 ing 3/2 OutFC (B ) isomorphism tripod B . 2 /1
2/1

tripod ; 1

tripod for the result of conjugating 2 by the outer /1

Next, let us observe that since 2/1 stabilizes x Ex [where we note that, from the point of view of Ex , the subgroup x is the cuspidal inertia group associated to one of the cusps of the tripod UEx ], it follows from the non-respd tripod of tripod portion of Proposition 1.3, (iv), applied to the outer automorphism 2 2 [cf. also the lifting tripod ], that stabilizes the -conjugacy classes of E x , B 1 3 /2 Lx , x , hence [cf. Proposition 3.2, (i)] induces elements 3/2 OutFC (E x ), 3/2 OutFC (Lx ). Moreover, it follows from Proposition 1.2, (iii), in the case of 3/2 , and from Corollaries 1.12, (ii), (iii); 1.14, (i), (iii), in the case of 3/2 [where we note that from the point of view of the situation of Corollary 1.14, (iii), Lx that corresponds to the minor cuspidal component, while E x corresponds to the major cuspidal component], that, for any outer isomorphisms tripod E x , 1 tripod Lx that arise scheme-theoretically [i.e., from isomorphisms of k -schemes 1 UT UE x , UT ULx ], the result of conjugating 3/2 , 3/2 , respectively, by these outer isomorphisms yields elements OutFC (tripod ) both of which are equal to 1 tripod tripod . [Here, we note that it is of crucial importance that we know that 1 1 tripod tripod FC FC Out (1 ) i.e., not just Out (1 )! since this symmetry of tripod 1 allows one to ignore the issue of precisely which cusp is sent to which by the various scheme-theoretic isomorphisms of tripods that appear.] In particular, tripod it follows from the denition of 3/2 and 1 that the restriction of 3/2 to Lx [cf. Proposition 3.2, (i)] is equal to 3/2 . Thus, it makes sense to glue 3/2 OutFC (B ), 3/2 OutFC (B ) along Lx so as to obtain an element 3/2 OutFC (3/2 ) as in Proposition 3.2, (iv), that restricts to 3/2 on B and to 3/2 on B . Next, we consider the extent to which 3/2 is compatible, relative to 2/1 , with the natural outer action of 2/1 on 3/2 . In particular, let us consider the following assertion: () 3/2 OutFC (3/2 ) is compatible, relative to 2/1 , with the natural outer actions of Ex ( 2/1 ) and Fx ( 2/1 ) on 3/2 . Now I claim that to complete the proof of Corollary 3.3, it suces to verify (). Indeed, since 2/1 is topologically generated by Ex , Fx [cf. Proposition 1.5,
L E L E L L
x

48

SHINICHI MOCHIZUKI

(iii)], it follows from () that 3/2 OutFC (3/2 ) is compatible, relative to 2/1 , with the natural outer action of 2/1 . Thus, by applying the natural isomorphism 2/1 [cf. 0; Remark 1.1.1], we conclude that 3/2 , 2/1 determine 3 /1 3 /2 an element 3/1 Out(3/1 ). It is immediate from the construction of 3/1 that 3/1 is C-admissible. Since 3/1 preserves the conjugacy class of inertia groups associated to the diagonal divisor in the geometric generic ber of pr1 : X3 X1 [cf. the argument applied in the proof of Proposition 1.3, (vii)], it follows from Proposition 1.2, (i), that 3/1 is FC-admissible, i.e., 3/1 OutFC (3/1 ). Next, let us write 1 . Since the 1 OutFC (1 ) for the automorphism induced by 3 via p1 : 3 natural homomorphism OutFC (3/1 ) OutFC (2/1 ) is injective by Corollary 2.3, (ii), we thus conclude [from the fact that 2/1 is manifestly compatible, relative to 1 , with the natural outer action of 1 on 2/1 ] that 3/1 is compatible, relative to 1 , with the natural outer action of 1 on 3/1 . In particular, by applying the 1 [cf. 0; Remark 1.1.1], we conclude that natural isomorphism 3 3/1 3/1 , 1 determine an element 3 OutFC (3 ) [cf. Proposition 1.2, (i)] that lifts 2 , as desired. This completes the proof of the claim. p13 Finally, we proceed to verify the assertion (). To this end, let us observe that : 3 2 (respectively, p23 : 3 2 ) induces a surjection 1 : 3/2 2/1 (respectively, 2 : 3/2 2/1 )
out out

whose kernel is topologically normally generated by the cuspidal inertia groups in 3/2 that correspond to the cusp parametrized by the factor labeled 2 (relog . That is to say, 1 (respectively, 2 ) corresponds to the spectively, 1) of X3 operation of forgetting the cusp parametrized by the factor labeled 2 (respectively, log 1) of X3 . Note that 1 (respectively, 2 ) induces isomorphisms E x Ex , F x Fx (respectively, Lx Ex , F x Fx , B 2/1 ). In the following, if () is an element of 3/1 , then let us write () Aut(3/2 ) for the automorphism induced by conjugation by (). Next, let us x , B , B , Lx , and F x as in the respd portion of Proposition 3.2, (ii). Here, we may assume without loss of generality that 2 ( ) = x . x Now let 2/1 Ex 2/1 ; 3/1 3/1 a lifting of 2/1 . Note that 3/1 stabilizes the 3/2 -conjugacy classes of B , , and F x [cf. the discussion of Denition 3.1, (iv)]. In particular, by replacing 3/1 by the product of 3/1 with an appropriate element of 3/2 , we may assume without loss of generality that 3/1 stabilizes the subgroups B , , and F x [cf. Proposition 3.2, (ii)]. Next, let us observe that [since p23 induces the natural surjection 2/1 1 ; the kernel of this surjection contains 2/1 Ex ] 3/1 induces, relative to 2 , an inner automorphism of 2/1 . Since 2 is surjective, it thus follows that there exists a 3/2 such that 3/1 induces, relative to 2 , the identity automorphism of 2/1 . On the other hand, since 2 ( ) = x is normally terminal in 2/1 [cf. Proposition 1.5, (i)], it follows that 2 ( ) x . In particular, by replacing 3/1 by the product of 3/1 with an appropriate element of , we may assume without loss of generality that:
x x x x x

COMBINATORIAL CUSPIDALIZATION

49

(a) 3/1 stabilizes the subgroups B , , and F x ; (b) 3/1 induces, relative to x 2 , the identity automorphism of 2/1 . We shall refer to a lifting 3/1 of 2/1 that satises these conditions (a), (b) as 2 -admissible. Now let 2/1 = 2/1 (2/1 ) 2/1 ; 3/1 , 3/1 3/1 2 -admissible liftings of 2/1 , 2/1 ; 3/2 Aut(3/2 ) an automorphism that gives rise to 3/2 . Since [by construction] 3/2 stabilizes the 3/2 -conjugacy classes of the subgroups B , ,
x

def

and F x [cf. Proposition 3.2, (iv)], we may assume without loss of generality [cf. Proposition 3.2, (ii)] that 3/2 stabilizes the subgroups B , , and F x . Now x to verify that 3/2 is compatible, relative to 2/1 , with the natural outer action of Ex [cf. ()], it suces to verify that:
1 (E ) We have: 3/1 = 3/2 3/1 3 /2 .

Next, let us recall from Denition 3.1, (iv), that 3/1 , 3/1 induce the trivial outer automorphism on F x ; in particular, the equality of (E ) holds over F x , up to composition with an F x -inner automorphism. Moreover, by the construction of 3/2 , it follows from Denition 3.1, (vi), that the equality of (E ) holds over B , up to composition with an B -inner automorphism. Since 3/2 , 3/1 , and 3/1 all stabilize [which is normally terminal in 3/2 cf. Proposition 3.2, (i)], we thus conclude that the equality of (E ) holds up to composition with some Aut(3/2 ) that stabilizes the subgroups B , , and F x , and, moreover,
x x

restricts to [possibly distinct!] -inner automorphisms over B [hence over Lx ] and F x . [That is to say, is a sort of abstract pronite analogue of a Dehn twist!] On the other hand, since 3/1 , 3/1 induce, relative to 2 , the identity automorphism of 2/1 , it follows that induces, relative to 2 , the identity automorphism of 2/1 . Since 2 induces isomorphisms of center-free [cf. Remark 1.1.1] pronite groups Lx Ex , F x Fx , we thus conclude that is the identity automorphism. This completes the proof of (E ). In a similar vein, let us x x , B , B , E x , and Lx as in the non-respd portion of Proposition 3.2, (ii). Here, we may assume without loss of generality that 1 ( x ) = x . Now let 2/1 Fx 2/1 ; 3/1 3/1 a lifting of 2/1 . Note that 3/1 stabilizes the 3/2 -conjugacy classes of E x , x , and B [cf. the discussion of Denition 3.1, (iv)]. In particular, by replacing 3/1 by the product of 3/1 with an appropriate element of 3/2 , we may assume without loss of generality that 3/1 stabilizes the subgroups E x , x , and B [cf. Proposition 3.2, (ii)]. Next, let us observe that [since 1 arises from p13 ] 3/1 induces, relative to 1 , an inner automorphism of 2/1 . Since 1 is surjective, it thus follows that there exists a 3/2 such that 3/1 induces, relative to 1 , the identity automorphism of 2/1 . On the other hand, since 1 ( x ) = x is normally terminal in 2/1 [cf. Proposition 1.5, (i)], it follows that 1 ( ) x . In particular, by replacing 3/1 by the product of 3/1 with an appropriate element of x , we may assume without loss of generality that: (a) 3/1 stabilizes the subgroups E x , x , and B ; (b)
x

50

SHINICHI MOCHIZUKI

3/1 induces, relative to 1 , the identity automorphism of 2/1 . We shall refer to a lifting 3/1 of 2/1 that satises these conditions (a), (b) as 1 -admissible. Now let 2/1 = 2/1 (2/1 ) 2/1 ; 3/1 , 3/1 3/1 1 -admissible liftings of 2/1 , 2/1 ; 3/2 Aut(3/2 ) an automorphism that gives rise to 3/2 . Since [by construction] 3/2 stabilizes the 3/2 -conjugacy classes of the subgroups E x , x , and B [cf. Proposition 3.2, (iv)], we may assume without loss of generality [cf. Proposition 3.2, (ii)] that 3/2 stabilizes the subgroups E x , x , and B . Now to verify that 3/2 is compatible, relative to 2/1 , with the natural outer action of Fx [cf. ()], it suces to verify that:
1 (F ) We have: 3/1 = 3/2 3/1 3 /2 . def

Next, let us recall from Denition 3.1, (iv), that 3/1 , 3/1 induce the trivial outer automorphism on E x ; in particular, the equality of (F ) holds over E x , up to composition with an E x -inner automorphism. Moreover, by the construction of 3/2 , it follows from Denition 3.1, (v), that the equality of (F ) holds over B , up to composition with an B -inner automorphism. Since 3/2 , 3/1 , and 3/1 all stabilize x [which is normally terminal in 3/2 cf. Proposition 3.2, (i)], we thus conclude that the equality of (F ) holds up to composition with some Aut(3/2 ) that stabilizes the subgroups E x , x , and B , and, moreover, restricts to [possibly distinct!] x -inner automorphisms over E x and B . [That is to say, is a sort of abstract pronite analogue of a Dehn twist!] On the other hand, since 3/1 , 3/1 induce, relative to 1 , the identity automorphism of 2/1 , it follows that induces, relative to 1 , the identity automorphism of 2/1 . Since 1 induces isomorphisms of center-free [cf. Remark 1.1.1] pronite groups E x Ex , F x Fx , we thus conclude that is the identity automorphism. This completes the proof of (F ), and hence of Corollary 3.3. Corollary 3.4. (Tautological Validity of , +) Suppose that X log is of type (g, r ), where r 0. Then: (i) We have: OutFCP (3 )cusp OutFC (3 ) . (ii) We have: OutFCP (4 )cusp OutFC (4 )
FC +

(iii) Suppose that r 1. Then OutFC (3 ) + contains the inverse image of Out (2 ) via the natural homomorphism OutFC (3 ) OutFC (2 ) induced by p12 . Proof. Assertion (i) follows immediately from the denitions, by observing that in the situation of Denition 1.8 and Proposition 1.9, the action of the group of permutations [i.e., automorphisms of the set {1, 2, 3}] on X3 preserves the subscheme W X3 of Denition 1.8, (i), and induces the automorphisms of W = V k UP

COMBINATORIAL CUSPIDALIZATION

51

given by permuting [over V ] the three cusps of UP . Assertion (ii) follows from 2 that appears in assertions (i) and (iii) by taking the surjection p12 : 3 3/1 . Thus, it remains to verassertion (iii) to be the standard surjection 4/1 ify assertion (iii). To this end, let us assume that we have been given an element 3 OutFC (3 ) that maps to an element 2 OutFC (2 ) , and that we are in the situation of Denition 3.1, with x X (k ) taken to be the cusp that exhibits 2 as an element of OutFC (2 ) . Let 2 AutFC (2 ), 3 AutFC (3 ) be elements that induce, respectively, 2 , 3 ; also, we suppose that 3 lifts 2 . By Propositions 1.3, (iv) [the respd portion]; 1.7, (a), we may assume without loss of generality that 2 stabilizes the subgroups (tripod =) Ex , IEx , and DEx of 2 , and that 2 1 tripod tripod FC induces an element 1 Out (1 ) = OutFC (Ex ) . Thus, it follows from the non-respd portion of Proposition 1.3, (iv), that 3 stabilizes the 3/2 conjugacy classes of B , F x [cf. the discussion of Denition 3.1, (iv), (vi)]. In
tripod tripod OutFC (tripod )S that lifts 1 [cf. particular, 3 induces an element 2 2 Denition 3.1, (vi)]. log log Now write X2 (X ) for the cusp of X2 [relative to prlog 1 : X2 X1 ] that corresponds to the cusp x X (k ). Thus, determines by restricting to the log log log geometric generic ber of prlog a minor verticial subgroup 1 : X3 X1 = X E 3/2 . Moreover, since the restriction of the section : X X2 to x X (k ) determines a cusp of UEx , it follows that [for suitable choices within the various tripod 3/2 -conjugacy classes] E B , and that this subgroup E of B = 2 /1

forms a minor verticial subgroup tripod at of tripod E 2/1 . In particular, we conclude

tripod that arises from a (iii), coincides relative to any isomorphism tripod 1 E OutFC (E ) + . k -isomorphism UE UT with tripod OutFC (tripod ) + =

Thus, by Denition 1.11, (ii), we conclude that 3 Out (3 ) , as desired. This completes the proof of assertion (iii), and hence of Corollary 3.4.
FC +

Section 4: The General Pronite Case In the present 4, we derive the main result [cf. Theorem 4.1] of the present paper from the various partial results obtained in 1, 2, 3. Theorem 4.1. (Partial Pronite Combinatorial Cuspidalization) Let X log S be a smooth log curve of type (g, r ) [cf. 0] over S = Spec(k), where k is an algebraically closed eld of characteristic zero. Fix a set of prime numbers

52

SHINICHI MOCHIZUKI

which is either of cardinality one or equal to the set of all prime numbers. For n a log for the n-th log conguration space associated nonnegative integer, write Xn
log to X log [cf. [24], Denition 2.1, (i)], where we take X0 = Spec(k); log (Xn ) n = 1 def def

for the maximal pro- quotient of the fundamental group of the log scheme log [cf. 0; the discussion preceding [24], Denition 2.1, (i)]; Xn OutFC (n ) Out(n ) for the subgroup of outer automorphisms that satisfy the following conditions (1), (2) [cf. Denition 1.1, (ii)]: (1) (H ) = H for every ber subgroup H n [cf. Remark 1.1.2; [24], Denition 2.3, (iii)]. (2) For m a nonnegative integer n, write Km n for the ber subgroup that arises as the kernel of the projection obtained by forgetting the factors of Xn with labels > m. Then induces a bijection of the collection of conjugacy classes of cuspidal inertia groups contained in each Km1 /Km [where m = 1, . . . , n] associated to the various cusps of the log log Xm geometric generic ber of the projection Xm 1 obtained by forgetting the factor labeled m. [Here, we regard the map m = n /Km n /Km1 of quotients of as the homomorphism that arises = m1 n by forgetting, successively, the factors with labels > m and the factors with labels > m 1.] If the interior UX of X log is ane [i.e., r 1], then set n0 = 2; if the interior def UX of X log is proper over k [i.e., r = 0], then set n0 = 3. Then: (i) The natural homomorphism OutFC (n ) OutFC (n1 ) induced by the projection obtained by forgetting the factor labeled n is injective if n n0 and bijective if n 5. (ii) The image of the natural homomorphism OutFC (n ) OutFC (n1 ) of (i) contains the following two subsets [cf. Denition 1.11]: (a) OutFC (n1 ) + , when n 2 [a set which is well-dened and nonempty only if (g, r ) = (0, 3) or n 1 n0 ]; (b) the inverse image in OutFC (n1 ) via the natural homomorphism OutFC (n1 ) OutFC (n2 ) of OutFC (n2 ) , when n 3 [a set which is well-dened and nonempty only if either (g, r ) = (0, 3) or n 2 n0 ]. (iii) Let OutFC (n ) OutFC (n1 ) be as in (i), where n n0 . Let Out(n ) be an outer automorphism that satises the following properties: (a) for
def

COMBINATORIAL CUSPIDALIZATION

53

every ber subgroup H n , (H ) is a ber subgroup; (b) (Kn1 ) = Kn1 ; (c) induces a bijection of the collection of conjugacy classes of cuspidal inertia groups contained in Kn1 ; (d) the outer automorphism Out(n1 ) determined by [cf. (b)] normalizes (respectively, commutes with) OutFC (n1 ). Then normalizes (respectively, commutes with) OutFC (n ).
log (iv) By permuting the various factors of Xn , one obtains a natural inclusion

Sn Out(n ) of the symmetric group on n letters into Out(n ) whose image commutes with OutFC (n ) if n n0 and normalizes OutFC (n ) if r = 0 and n = 2. Proof. First, we consider the injectivity portion of assertion (i). Consider the natural isomorphisms n Kn2
out

n2 ;

n1 (Kn2 /Kn1 )

out

n2

[cf. 0; Remark 1.1.1], together with the interpretation of n/n2 = Kn2 Kn2 /Kn1 = n1/n2 as the 2 1 [i.e., the projection that arises by forgetting the factor labeled 2] associated to an X log of type (g, r + n 2) [cf. [24], Proposition 2.4, (i)]. [Here, we note that one veries easily that this interpretation is compatible with the denition of the various OutFC ()s involved.] Now the above natural isomorphisms allow one to reduce the injectivity portion of assertion (i) to the case n = 2, r 1, which follows immediately from Corollaries 1.12, (ii); 2.3, (ii) [cf. also Remark 2.1.1]. This completes the proof of the injectivity portion of assertion (i). Next, we consider assertion (iii). Let OutFC (n ). Write for the image of in OutFC (n1 ); = 1 ; = ( )1 . Then it follows immediately from property (a) that is F-admissible and from properties (b), (c), (d) that is C-admissible. Thus, OutFC (n ). If, moreover, it holds that = , then it follows from the injectivity portion of assertion (i) that = . This completes the proof of assertion (iii). Next, we consider assertion (iv). When n = 2, assertion (iv) follows immediately from Proposition 1.6, (iii); Corollaries 1.12, (iii); 2.3, (iii) [cf. also Remark 2.1.1]. Note that when n 3, by applying the natural isomorphism n Kn2
out def def

n2

[cf. 0; Remark 1.1.1], together with the interpretation of n/n2 = Kn2 as the 2 associated to an X log of type (g, r + n 2) [cf. [24], Proposition 2.4, (i)], we thus conclude from assertion (iv) for n = 2 [whose proof has already been completed] that OutFC (n ) commutes with the permutation outer automorphism Out(n ) that arises from the permutation ((n 1) n) of {1, 2, . . . , n} [i.e., the permutation that switches n and n 1 and xes all other elements of {1, 2, . . . , n}].

54

SHINICHI MOCHIZUKI

Now we apply induction on n. When UX is ane, let us observe that [by the induction hypothesis] every permutation outer automorphism Out(n ) that arises from a permutation of {1, 2, . . . , n} that xes n satises the properties (a), (b), (c), (d) of assertion (iii) in the respd case. Thus, when UX is ane, the induction step [i.e., the derivation of assertion (iv) for n from assertion (iv) for n 1] follows from assertion (iii), together with the fact that the permutation group of {1, 2, . . . , n} is generated by ((n 1) n) and the subgroup of permutations that x n. If UX is proper and n 4, then the induction step [i.e., the derivation of assertion (iv) for n from assertion (iv) for n 1] follows by a similar argument. Thus, it remains to verify the induction step when UX is proper and n = 3. To this end, let us rst observe that, as discussed above, OutFC (3 ) commutes with [the permutation outer automorphism that arises from the permutation of {1, 2, 3} given by] (23). Moreover, by applying assertion (iii) in the non-respd case to [the permutation outer automorphism that arises from the permutation of {1, 2, 3} given by] (12), we conclude that (12) normalizes OutFC (3 ). Thus, by conjugating by (12), we conclude that OutFC (3 ) commutes with (13). Now since the group of permutations of {1, 2, 3} is generated by (12), (13), we conclude that OutFC (3 ) commutes with all permutation outer automorphisms. This completes the proof of assertion (iv). Next, we consider assertion (ii). First, let us observe that when (g, r ) = (0, 3) and n = 2, assertion (ii) for the subset of (a) is a tautology [cf. Denition 1.11, (i)]; when (g, r ) = (0, 3) and n = 3, assertion (ii) for the subset of (b) may be reduced, in light of the inclusion OutFC (2 )S OutFC (2 ) + [cf. Corollaries 1.12, (ii), (iii); 1.14, (i), (iv)], to assertion (ii) for the subset of (a) when n = 3. Next, let us observe that when n 4, by the denition of [cf. Denition 1.11, (ii)], every element OutFC (n1/n4 ) [where we recall that n1/n4 is the 3 associated to an X log of type (g, r +n 4)] that is induced, relative to the inclusion n1/n4 n1 , by an element OutFC (n1 ) of the subset of (b) maps, via the natural homomorphism OutFC (n1/n4 ) OutFC (n2/n4 ) [obtained by forgetting the factor labeled n 1], to an element of OutFC (n2/n4 ) , hence, by Corollary 3.4, (iii), is contained in OutFC (n1/n4 ) + ; but, by the denition of + [cf. Denition 1.11, (ii)], this implies that every element of the subset of (b) is contained in OutFC (n1 ) + . Thus, to complete the proof of assertion (ii), it suces to verify assertion (ii) for the subset of (a) in the case of n 3. On the other hand, when n 3, by applying the natural isomorphisms n n/n3
out out

n3 ,

n1 n1/n3 n3 [cf. the proof of the injectivity portion of assertion (i)], together with the injectivity portion of assertion (i) [which is necessary in order to conclude the compatibility of liftings, relative to the natural homomorphism OutFC (n/n3 ) OutFC (n1/n3 ), with the respective outer actions of n3 ], to complete the proof of assertion (ii), we conclude that it suces to verify assertion (ii) for the subset of (a) in the case of n = 3. But this is precisely the content of Corollary 3.3. This completes the proof of assertion (ii). Finally, we consider the surjectivity [i.e., bijectivity] portion of assertion (i) for n 5. First, let us observe that by Lemma 2.4, to complete the proof of

COMBINATORIAL CUSPIDALIZATION

55

assertion (i), it suces to verify that the image of the natural homomorphism OutFC (n ) OutFC (n1 ) of assertion (i) contains the subset OutFC (n1 )cusp OutFC (n1 ). Next, let us observe that by assertion (iv) and Remark 1.1.5, every element OutFC (n1/n5 ) [where we recall that n1/n5 is the 4 associated to an X log of type (g, r + n 5)] that is induced, relative to the inclusion n1/n5 n1 , by an element OutFC (n1 )cusp is contained in OutFCP (n1/n5 )cusp , hence, by Corollary 3.4, (ii), in OutFC (n1/n5 ) + . But this implies that OutFC (n1 )cusp = OutFC (n1 ) + [cf. Denition 1.11, (ii)]. Thus, in summary, to complete the proof of assertion (i), it suces to verify that the image of the natural homomorphism OutFC (n ) OutFC (n1 ) of assertion (i) contains the subset OutFC (n1 ) + OutFC (n1 ). But this follows from assertion (ii) [cf. the subset of (a)]. This completes the proof of assertion (i). Remark 4.1.1. The argument applied to verify Theorem 4.1, (iv), in the proper case suggests that even if one cannot verify the injectivity of the homomorphism OutFC (2 ) OutFC (1 ) in the proper case, it may be possible to verify the injectivity of the homomorphism OutFC (3 ) OutFC (1 ) [i.e., induced by the projection obtained by forgetting the factors labeled 2, 3] in the proper case. Remark 4.1.2. In the pro-l case [i.e., the case where is of cardinality one], a number of results related to Theorem 4.1, (i), have been obtained by various authors. (i) In [10], Theorem 1 [cf. also [8], which is discussed further in Remark 4.2.1, (ii), below], a similar injectivity result to that of Theorem 4.1, (i), is obtained in the pro-l case for outer automorphisms satisfying certain conditions i.e., the conditions ( 1), ( 2) of [10], Theorem 1. It is immediate [cf. Proposition 1.3, (vii)] that outer automorphisms lying in the kernel of the homomorphism in question which satisfy these conditions ( 1), ( 2) are FC-admissible. Thus, [at least when the condition of hyperbolicity 2g 2 + r > 0 is satised] [10], Theorem 1, may be obtained as a consequence of Theorem 4.1, (i). (ii) In [29], a ltered pro-l injectivity result [cf. [29], Theorem 4.3] is obtained ( n) ( n) for a certain ltration on a subgroup g,r Out(n ) [where g,r is as in [29], (2.11) except with r and n reversed!]. It follows immediately from the conditions ( n) used to dene g,r [cf. [29], (2.10), (2.11)] that
QS n) (n ) = OutFC (n )cusp ( g,r = Out

[cf. Proposition 1.3, (vii)]. In particular, the injectivity of Theorem 4.1, (i), in the pro-l case may also be thought of as yielding a new proof of the injectivity that holds as a consequence of the ltered injectivity of [29], Theorem 4.3. (iii) In the context of (ii), graded pro-l surjectivity results are obtained in [32]. Related results may be found in [9].

56

SHINICHI MOCHIZUKI

Remark 4.1.3. The injectivity of the restriction of the homomorphism of Theorem 4.1, (i), to an image of Galois OutFC (n ) that arises from scheme theory is precisely the content of [14], Theorem 2.2. Indeed, it was precisely the goal of attaining a more abstract, combinatorial understanding of the theory of [14] that motivated the author to develop the theory of the present paper. Also, we observe that the remaining portion of [14], Theorem 2.2 involving related outer actions on tripod follows immediately from the existence of the natural outer homomorphism of Corollary 1.10, (iii). Remark 4.1.4. (i) Observe that the various n that arise from dierent X log s of the same type (g, r ) are always isomorphic, in a fashion that is compatible with the various ber subgroups and cuspidal inertia groups of subquotients. Indeed, this follows immediately [cf. the various specialization isomorphisms discussed in 0] from the well-known fact [cf., [3]] that the moduli stack Mg,r [cf. 0] is smooth, proper, and geometrically connected over Z. (ii) Although we have formulated Theorem 4.1, (i), in terms of outer automorphisms, it is a routine exercise in light of the observation of (i) to reformulate Theorem 4.1, (i), in terms of outer isomorphisms, as is often of interest in applications to anabelian geometry. Remark 4.1.5. In [7], a group-theoretic construction is given for the geometrically pro-l arithmetic fundamental groups of conguration spaces of arbitrary dimension from the geometrically pro-l arithmetic fundamental group of a proper hyperbolic curve over a nite eld. This construction is performed by considering various Lie versions of these arithmetic fundamental groups of conguration spaces of arbitrary dimension. On the other hand, by applying the injectivity portion of Theorem 4.1, (i) [cf. the argument involving given in the proof of Theorem 4.1, (ii)], one may simplify the argument of [7]: That is to say, instead of working with Lie versions of geometrically pro-l arithmetic fundamental groups of conguration spaces of arbitrary dimension [associated to a proper hyperbolic curve over a nite eld], one may instead restrict oneself to working with Lie versions of geometrically pro-l arithmetic fundamental groups of two-dimensional conguration spaces [associated to a [not necessarily proper] hyperbolic curve over a nite eld]. [We leave the routine details to the interested reader.] This reduction to the case of Lie algebras associated to two-dimensional conguration spaces results in a substantial reduction of the book-keeping involved. The following result allows one to relate the theory of the present paper to the work of Nakamura and Harbater-Schneps [cf. [26], [5]]. Corollary 4.2. (Partial Pronite Combinatorial Cuspidalization for Tripods) In the notation of Theorem 4.1: Suppose further that X log is a tripod. Then, for n 1:
out

COMBINATORIAL CUSPIDALIZATION

57

(i) We have: OutFC (n )S = OutFCS (n ) = OutFC (n ) if n = 1; OutFC (n )S = OutFCS (n ) OutFC (n ) if n 2 [cf. Denitions 1.1, (vi); 1.11, (i), (ii)]. (ii) The natural homomorphism OutFCS (n ) OutFCS (n1 ) induced by the projection obtained by forgetting the factor labeled n is injective if n 2 and bijective if n 3. Proof. First, we consider assertion (i). When n = 1, assertion (i) follows immediately from Denitions 1.1, (vi); 1.11, (i). Thus, we may assume that n 2. Then the fact that OutFC (n )S = OutFCS (n ) follows formally from Corollary 1.14, (i); Theorem 4.1, (i), (iv). The fact that OutFCS (n ) OutFC (n ) + follows from Corollary 1.14, (iv). This completes the proof of assertion (i). Now the injectivity portion of assertion (ii) follows from the injectivity portion of Theorem 4.1, (i); in light of this injectivity, the bijectivity portion of assertion (ii) follows from assertion (i) and Theorem 4.1, (ii) [cf. the subset of (a)]. This completes the proof of assertion (ii) and hence of Corollary 4.2. Remark 4.2.1. (i) Suppose that we are in the situation of Corollary 4.2, and that is the set of all prime numbers. Then various injectivity and bijectivity results are obtained by Nakamura and Harbater-Schneps in [26], [5] concerning the subgroup Outn+3 Out(n ) [where n 1]. This subgroup is dened in [5], 0.1, Denition, by means of two conditions (i) [i.e., quasi-speciality], (ii) [i.e., symmetry]. From the point of view of the theory of the present paper, these two conditions amount to the condition on Out(n ) that OutQS (n ), and, moreover, commutes with all of the outer symmetry permutations i.e., Outn+3 = OutFCS (n ) [cf. Proposition 1.3, (vii)]. (ii) In [5], it is shown that the natural homomorphism Outn+3 Outn+2
+

OutFC (n )cusp

OutFC (n )cusp

58

SHINICHI MOCHIZUKI

is injective if n 2 and bijective if n 3 [cf. [5], 0.1, Corollary]. The injectivity portion of this result of [5] is derived [cf. [5], Proposition 8] from the injectivity obtained in [26], Lemma 3.2.2, and may be regarded as a pronite version of an earlier pro-l result due to Ihara [cf. [8]] cf. the discussion of [5], 0.2. On the other hand, unlike the case with [5], the approach of [8] allows one to treat, in essence, the full group OutQS (n ) [i.e., not just OutFCS (n ) = Outn+3 ] in the pro-l case. In light of the discussion of (i), the proofs given in the present paper of Theorem 4.1, (i), and Corollary 4.2, (ii), may be regarded as alternate proofs of these results of [8] and [5]. (iii) The strong symmetry assumption imposed on elements of OutFCS (n ) suggests that there is a substantial gap between injectivity or bijectivity results for OutFCS (n ) and injectivity or bijectivity results for OutFC (n ). This gap accounts for the lack of the need to invoke such results as the combinatorial version of the Grothendieck Conjecture [i.e., [20], Corollary 2.7, (iii)] in the proofs of [26], [5].

Section 5: The Discrete Case In the present 5, we discuss a discrete analogue [cf. Corollary 5.1] of Theorem 4.1. One important aspect of this discrete analogue is that it is a relatively easy consequence of the well-known theorem of Dehn-Nielsen-Baer [cf., e.g., [13], Theorem 2.9.B], together with the injectivity asserted in Theorem 4.1, (i), that the discrete analogue of the homomorphism of Theorem 4.1, (i), is surjective.
top () to denote the [usual] topological In the following, we use the notation 1 fundamental group of the connected topological space in parentheses.

Corollary 5.1. (Partial Discrete Combinatorial Cuspidalization) Let X be a topological surface of type (g, r ) [i.e., the complement of r distinct points in a compact oriented topological surface of genus g ]. For integers n 1, write Xn for the complement of the diagonals in the direct product of n copies of X ;
top (Xn ) n = 1 def

for the [usual topological] fundamental group of Xn ; n for the pronite completion of n ; OutFC (n ) Out(n ) (respectively, OutF (n ) Out(n )) for the subgroup of outer automorphisms that satisfy the following condition(s) (1), (2) (respectively, (1)): (1) (H ) = H for every ber subgroup H n [cf. [24], Denition 7.2, (ii); [24], Corollary 7.4].

COMBINATORIAL CUSPIDALIZATION

59

(2) For m a nonnegative integer n, write Km n for the ber subgroup that arises as the kernel of the projection obtained by forgetting the factors def of Xn with labels by > m; b/a = Ka /Kb for a, b {0, 1, . . . , n} such that a b. Then induces a bijection of the collection of conjugacy classes of cuspidal inertia groups contained in each m/m1 [where m = 1, . . . , n] associated to the various cusps of the topological surfaces Xm1 obtained by forgetting that arise as bers of the projection Xm the factor labeled m. [Here, we regard the map m = n /n/m n /n/m1 = m1 of quotients of n as the homomorphism that arises by forgetting, successively, the factors with labels > m and the factors with labels > m 1.] We refer to Denition 5.2 below for more details on the notion of an inertia group. If r 1 i.e., X is non-compact then set n0 = 2; if r = 0 i.e., X is def compact then set n0 = 3. Then: (i) The natural homomorphisms n n ; OutF (n ) OutF (n )
def

are injective for n 1. Here, the injectivity of the rst homomorphism is equivalent to the assertion that n is residually nite. (ii) The natural homomorphism OutFC (n ) OutFC (n1 ) induced by the projection obtained by forgetting the factor labeled n is bijective if n n0 and surjective if n = 2. (iii) Let OutFC (n ) OutFC (n1 ) be as in (ii), n n0 . Let Out(n ) be an outer automorphism that satises the following properties: (a) for every ber subgroup H n , (H ) is a ber subgroup; (b) (Kn1 ) = Kn1 ; (c) induces a bijection of the collection of conjugacy classes of cuspidal inertia groups contained in Kn1 ; (d) the outer automorphism Out(n1 ) determined by [cf. (b)] normalizes (respectively, commutes with) OutFC (n1 ). Then normalizes (respectively, commutes with) OutFC (n ).
log , one obtains a natural inclusion (iv) By permuting the various factors of Xn

Sn Out(n ) of the symmetric group on n letters into Out(n ) whose image commutes with OutFC (n ) if n n0 and normalizes OutFC (n ) if r = 0 and n = 2. Proof. In the following, we shall write AutFC (n ) = Aut(n ) Out(n ) OutFC (n ) AutF (n ) = Aut(n ) Out(n ) OutF (n )
def def

60

SHINICHI MOCHIZUKI

for n 1. Now let us consider assertion (i). The fact that n is residually nite is well-known [cf., e.g., [24], Proposition 7.1, (ii)]. Thus, it remains to verify the injectivity of the natural homomorphism OutF (n ) OutF (n ). When n = 1, the injectivity of the natural homomorphism Out(1 ) Out(1 ) is the content of [2], Lemma 3.2.1, when X is non-compact; when X is compact, the injectivity of this homomorphism is implicit in the proofs of [4], Theorems 1, 3. This completes the proof of assertion (i) when n = 1. Now assertion (i) for arbitrary n follows by applying induction on n, together with the natural isomorphism n K 1
out

[cf. 0; Remark 1.1.1] and the evident discrete analogue of the interpretation of n/1 = K1 given in [24], Proposition 2.4, (i), which allows one to apply the induction hypothesis to K1 [as well as to 1 ]. Indeed, if AutF (n ) induces an inner automorphism of n , then the automorphism 1 AutF (1 ) determined by induces an inner automorphism of 1 . Thus, by the induction hypothesis, 1 is inner, so by replacing with the composite of with an appropriate inner automorphism, we may assume that 1 is the identity. Then induces an automorphism K AutF (K1 ) which is compatible with the outer action of 1 on K1 . Moreover, K arises [relative to the inclusion K1 n n ] from conjugation by an element n whose image in 1 induces [by conjugation] the identity automorphism of 1 ( 1 ), hence also the identity automorphism of 1 . Since 1 is center-free [cf. Remark 1.1.1], we thus conclude that lies in the closure of the image of K1 in n [which is naturally isomorphic to the pronite completion of K1 cf. [24], Proposition 7.1, (i); [24], Proposition 2.2, (i)]. Thus, by applying the induction hypothesis to K1 , we conclude that K is inner, hence [by applying the natural isomorphism n K1 proof of assertion (i).
out

1 ] that is inner. This completes the

Next, we consider assertion (ii). First, let us recall that by the well-known theorem of Dehn-Nielsen-Baer [cf., e.g., [13], Theorem 2.9.B] every automorphism AutFC (1 ) arises from a homeomorphism [or even a dieomorphism!] X : X X . Since X then induces a homeomorphism Xn Xn for every n 1, we thus obtain elements n Aut(n ) that [as is easily veried] belong to AutFC (n ) and lift [relative, say, to the projection n 1 determined by the factor labeled 1]. In particular, the corresponding natural homomorphisms OutFC (n ) OutFC (1 ) are surjective for n 1. Next, let us observe that the injectivity of OutFC (n ) OutFC (n1 ) for n n0 follows formally from the injectivity of OutFC (n ) OutFC (n ) [cf. assertion (i)] and the injectivity of Theorem 4.1, (i). In light of the surjectivity of OutFC (n ) OutFC (1 ), we thus conclude that if X is non-compact [so n0 = 2], then OutFC (n ) OutFC (n1 ) is bijective for n 2. This completes the proof of assertion (ii) for non-compact X . Next, let us consider the case where X is compact. Then one may verify the surjectivity of OutFC (n ) OutFC (n1 ) for n 3 by arguing as follows. Let

COMBINATORIAL CUSPIDALIZATION

61

AutFC (n1 ), where we think of n1 as n /n/n1 = n /Kn1 . Then determines automorphisms K AutFC (K1 /Kn1 ), 1 AutFC (1 ) [where we think of 1 as n /n/1 = n /K1 ] which are compatible with the natural outer action of 1 on K1 /Kn1 . Then by applying assertion (ii) in the non-compact case [whose proof has already been completed] to K1 , we conclude that OutFC (K1 ) OutFC (K1 /Kn1 ) is bijective. Let K AutFC (K1 ) be a lifting of K . Note that the injectivity of OutFC (K1 ) OutFC (K1 /Kn1 ) [together with the compatibility of 1 , K with the natural outer action of 1 on K1 /Kn1 ] implies that 1 , K are compatible with the natural outer action of 1 on K1 . Thus, by applying the 1 [cf. 0; Remark 1.1.1], we conclude that natural isomorphism n K1 K , 1 determine an automorphism Aut(n ) which [as is easily veried, in light of the residual niteness of assertion (i), by applying Proposition 1.2, (i), (iii), to n ] belongs to AutFC (n ). This completes the proof of the surjectivity of OutFC (n ) OutFC (n1 ) for n 3, and hence of assertion (ii). The proof of assertion (iii) as a consequence of assertion (ii) is entirely similar to the proof of Theorem 4.1, (iii) [as a consequence of Theorem 4.1, (i)]. Finally, we consider assertion (iv). When r = 0 and n = 2, assertion (iv) follows immediately from the evident discrete analogue of Proposition 1.6, (i), (a). Thus, it remains to verify that OutFC (n ) Out(n ) commutes with the image of Sn when n n0 . To this end, let Out(n ) be an element of the image of Sn ; OutFC (n ); = 1 Out(n ). Then one veries immediately that OutF (n ). Moreover, by Theorem 4.1, (iv), the images of and in OutF (n ) coincide. Thus, the fact that = follows from the injectivity of OutF (n ) OutF (n ) [cf. assertion (i)]. This completes the proof of assertion (iv). Remark 5.1.1. There is a partial overlap between the content of Corollary 5.1 above and Theorems 1, 2 of [12]. Denition 5.2. Let n 2 be an integer.
def out

(i) Write R for the underlying topological space of the topological eld of real numbers; 2 R2 = R R for the unit circle; n Rn = R . . . R [i.e., the product of n copies of R] for the image of the embedding 2 R2 Rn obtained by taking the rst n 2 coordinates to be zero. (ii) Let N be a connected topological manifold of dimension n; M N a connected submanifold of dimension n 2; P = N \M. Thus, for each point x M, there exists an open neighborhood U N of x in U , together with an open immersion U Rn that maps x to the origin of Rn , contains n in its image, and induces an open immersion U M Rn2 ( Rn ) [where we think of Rn2 as the subspace of Rn whose last two coordinates are zero]. In particular, we obtain an immersion n P N ; write
top IM 1 (P ) def

62

SHINICHI MOCHIZUKI

top top for the image of the homomorphism (Z =) 1 (n ) 1 (P ) induced by this top (P )immersion n P ( N ). One veries easily that IM is well-dened up to 1 conjugacy and independent of the choice of x, U , and the open immersion U Rn . top We shall refer to IM as the inertia group associated to M in 1 (P ).

Corollary 5.3. (Quasi-Speciality) In the situation of Corollary 5.1: Suppose that X is obtained as the complement of r points i.e., cusps of a compact oriented topological surface Z . Write Pn for the product Z . . . Z of n copies of Z ; D n for the set of connected submanifolds of codimension 2 of Pn given by the 1 n(n 1) diagonals and the n r bers of cusps via the n projection maps Pn Z . 2 For each D n , write def = Pn \ Pn Xn
=

where the union ranges over elements

= of D n;

I n for the inertia group [well-dened up to n -conjugacy] determined by the sub Xn [where we note that Xn = Xn \( Xn )]. Write manifold Xn OutQS (n ) Out(n ) where QS stands for quasi-special [cf. Proposition 1.3, (vii)] for the subgroup of outer automorphisms that stabilize the conjugacy class of each inertia group I , for Dn ; OutFC (n )cusp OutFC (n ) for the subgroup of outer automorphisms that induce, via the surjection n 1 obtained by forgetting the factors with labels > 1, outer automorphisms of 1 that stabilize each of the conjugacy classes of the inertia groups of the cusps. Then: (i) We have: OutQS (n ) = OutFC (n )cusp . (ii) The natural homomorphism of Corollary 5.1, (ii), restricts to a homomorphism OutQS (n ) OutQS (n1 ) which is bijective if n n0 [where n0 is as in Corollary 5.1] and surjective if n = 2. Proof. First, we consider assertion (i). We begin by observing that it follows immediately from the denitions [together with well-known facts concerning the relationship between topological and etale fundamental groups] that pronite completion induces a homomorphism OutQS (n ) OutQS (n ) OutF (n ) [cf. Proposition 1.3, (vii)]. Thus, it follows immediately from the residual niteness of Corollary 5.1, (i), that OutQS (n ) OutF (n ). In particular, the fact that OutQS (n ) OutFC (n )cusp follows immediately from the denition of OutQS () [cf. the

COMBINATORIAL CUSPIDALIZATION

63

proof of Proposition 1.3, (vii)]. Now it remains to verify that OutFC (n )cusp OutQS (n ). To this end, let us rst observe that if X is compact, then every I [where D 1 obtained by forgetting n ] lies in the kernel of the surjection n the factors with labels > 1; in particular, [by thinking of Ker(n 1 ) as a n1 that arises for some topological surface of type (g, 1)] we conclude that it suces to verify the inclusion OutFC (n )cusp OutQS (n ) for non-compact X . Thus, let us suppose that X is non-compact. Then by Corollary 5.1, (ii), we have a bijection OutFC (n )cusp OutFC (1 )cusp i.e., [cf. the proof of Corollary 5.1, (ii)] every element OutFC (n )cusp arises from a homeomorphism X : X X . Moreover, it follows immediately from the superscript cusp that this homeomorphism extends to a homeomorphism Z : Z Z that xes each of the cusps. In particular, Z induces compatible self homeomorphisms of Xn Xn Pn for each D n . Thus, it follows immediately QS from the denitions that Out (n ). This completes the proof of assertion (i). Finally, assertion (ii) follows immediately from assertion (i) and Corollary 5.1, (ii).

Remark 5.3.1. Suppose that (g, r ) = (0, 3). Then the injectivity portion of Corollary 5.3, (ii), is [essentially] the content of [8], 1.2, The Injectivity Theorem (i). By applying this injectivity, together with a classical result of Nielsen to the eect that OutQS (1 ) = {1} [cf. [8], 6.1; here, the element of OutQS (1 ) corresponding to 1 is the automorphism induced by complex conjugation], one obtains that OutQS (n ) = {1} for all n 2 [cf. [8], 1.2, The Vanishing Theorem].

Bibliography

[1] W. Abiko, The Real Analytic Theory of Teichm uller Space, Lecture Notes in Mathematics 820, Springer, 1980. [2] Y. Andr e, On a Geometric Description of Gal(Qp /Qp ) and a p-adic Avatar of GT , Duke Math. J. 119 (2003), pp. 1-39. [3] P. Deligne and D. Mumford, The Irreducibility of the Moduli Space of Curves of Given Genus, IHES Publ. Math. 36 (1969), pp. 75-109. [4] Grossman, On the residual niteness of certain mapping class groups, J. London Math. Soc. 2 (1974), pp. 160-164. [5] D. Harbater, L. Schneps, Fundamental groups of moduli and the GrothendieckTeichm uller group, Proceedings of the American Mathematical Society 352 (2000), pp. 3117-3148. [6] Y. Hoshi, The exactness of the log homotopy sequence, preprint, to appear in Hiroshima Math. J.

64

SHINICHI MOCHIZUKI

[7] Y. Hoshi, Absolute anabelian cuspidalizations of conguration spaces over nite elds, preprint, to appear in Publ. of RIMS. [8] Y. Ihara, Automorphisms of pure sphere braid groups and Galois groups in The Grothendieck Festschrift, Vol. II, Progress in Mathematics 87, Birkh auser (1990), pp. 353-373. [9] Y. Ihara, On the stable derivation algebra associated with some braid groups, Israel J. Math. 80 (1992), pp. 135-153. [10] Y. Ihara, M. Kaneko, Pro-l pure braid groups of Riemann surfaces and Galois representations, Osaka J. Math. 29 (1992), pp. 1-19. [11] L. Illusie, An Overview of the Work of K. Fujiwara, K. Kato and C. Nakayama on Logarithmic Etale Cohomology, in Cohomologies p-adiques et applications arithm etiques (II), P. Berthelot, J.-M. Fontaine, L. Illusie, K. Kato, M. Rapoport, eds, Asterisque 279 (2002), pp. 271-322. [12] E. Irmak, N. Ivanov, J. D. McCarthy, Automorphisms of surface braid groups, preprint (arXiv:math.GT/0306069v1 3 Jun 2003). [13] N. Ivanov, Mapping class groups, in Handbook of geometric topology, NorthHolland, Amsterdam, 2002, pp. 523-633. [14] M. Matsumoto, Galois representations on pronite braid groups on curves, J. Reine Angew. Math. 474 (1996), pp. 169-219. [15] S. Mochizuki, A Version of the Grothendieck Conjecture for p-adic Local Fields, The International Journal of Math. 8 (1997), pp. 499-506. [16] S. Mochizuki, The Local Pro-p Anabelian Geometry of Curves, Invent. Math. 138 (1999), pp. 319-423. [17] S. Mochizuki, Extending Families of Curves over Log Regular Schemes, J. reine angew. Math. 511 (1999), pp. 43-71. [18] S. Mochizuki, The Absolute Anabelian Geometry of Hyperbolic Curves, Galois Theory and Modular Forms, Kluwer Academic Publishers (2003), pp. 77-122. [19] S. Mochizuki, Semi-graphs of Anabelioids, Publ. Res. Inst. Math. Sci. 42 (2006), pp. 221-322. [20] S. Mochizuki, A combinatorial version of the Grothendieck conjecture, Tohoku Math. J. 59 (2007), pp. 455-479. [21] S. Mochizuki, Absolute anabelian cuspidalizations of proper hyperbolic curves, J. Math. Kyoto Univ. 47 (2007), pp. 451-539. [22] S. Mochizuki, Topics in Absolute Anabelian Geometry II: Decomposition Groups, RIMS Preprint 1625 (March 2008). [23] S. Mochizuki, Topics in Absolute Anabelian Geometry III: Global Reconstruction Algorithms, RIMS Preprint 1626 (March 2008). [24] S. Mochizuki, A. Tamagawa, The algebraic and anabelian geometry of conguration spaces, Hokkaido Math. J. 37 (2008), pp. 75-131.

COMBINATORIAL CUSPIDALIZATION

65

[25] D. Mumford, Abelian Varieties, Oxford Univ. Press (1974). [26] H. Nakamura, Galois rigidity of pure sphere braid groups and pronite calculus, J. Math. Sci. Univ. Tokyo 1 (1994), pp. 71-136. [27] H. Nakamura, Coupling of universal monodromy representations of GaloisTeichm uller modular groups , Math. Ann. 304 (1996), pp. 99-119. [28] H. Nakamura, Limits of Galois representations in fundamental groups along maximal degeneration of marked curves, I, Amer. J. Math. 121 (1999), pp. 315-358. [29] H. Nakamura, N. Takao, R. Ueno, Some stability properties of Teichm uller modular function elds with pro-l weight structures, Math. Ann. 302 (1995), pp. 197-213. [30] J. Neukirch, A. Schmidt, K. Wingberg, Cohomology of number elds, Grundlehren der Mathematischen Wissenschaften 323, Springer-Verlag (2000). [31] L. Ribes and P. Zaleskii, Pronite Groups, Ergebnisse der Mathematik und ihrer Grenzgebiete 3, Springer-Verlag (2000). [32] H. Tsunogai, The stable derivation algebras for higher genera, Israel J. Math. 136 (2003), pp. 221-250. [33] I. Vidal, Contributions a ` la cohomologie etale des sch emas et des log-sch emas, Th ese, U. Paris-Sud (2001).

Research Institute for Mathematical Sciences Kyoto University Kyoto 606-8502 Japan motizuki@kurims.kyoto-u.ac.jp