Absolute Anabelian Geometry

Download as pdf or txt
Download as pdf or txt
You are on page 1of 45

The Absolute Anabelian Geometry of

Hyperbolic Curves

Shinichi Mochizuki

Contents:
§0. Notations and Conventions
§1. Review of Anabelian Geometry
§1.1. The Anabelian Geometry of Number Fields
§1.2. The Anabelian Geometry of p-adic Local Fields
§1.3. The Anabelian Geometry of Hyperbolic Curves
§2. Reconstruction of the Logarithmic Special Fiber
Appendix: Terminology of Graph Theory

Introduction

Let XK be a hyperbolic curve (cf. §0 below) over a field K of characteristic 0.


Denote its algebraic fundamental group by ΠXK . Thus, we have a natural surjection

ΠXK  GK

of ΠXK onto the absolute Galois group GK of K.


When K is a finite extension of Q or Qp , and one holds GK fixed, then it is known
(cf. [Tama1], [Mzk6]; Theorem 1.3.4 of the present manuscript) that one may recover
the curve XK in a functorial fashion from ΠXK . This sort of result may be thought of
as a “relative result” (i.e., over GK ). Then the question naturally arises:

To what extent are the “absolute analogues” of this result valid — i.e., what
if one does not hold GK fixed?

If K is a number field, then it is still possible to recover XK from ΠXK (cf. Corollary
1.3.5), by applying the theorem of Neukirch-Uchida (cf. Theorem 1.1.3). On the other

1
2 SHINICHI MOCHIZUKI

hand, when K is a p-adic local field (i.e., a finite extension of Qp), the analogue of the
theorem of Neukirch-Uchida fails to hold, and indeed, it is the opinion of the author at
the time of writing that it is unlikely (in the p-adic local case) that one can recover XK
in general (i.e., in the fashion of Corollary 1.3.5) from ΠXK .
In the present manuscript, we begin by reviewing/surveying in §1 the anabelian
geometry of number fields, p-adic local fields, and hyperbolic curves from the point of
view of the goal of understanding to what extent the anabelian geometry of hyperbolic
curves over p-adic local fields can be made “absolute”. Our main result (Theorem
2.7), given in §2, states that when K is a p-adic local field, (although we may be unable
to recover XK itself) one may recover (in a functorial fashion) the special fiber of
XK , together with its natural log structure, in an absolute fashion, i.e., solely from
the isomorphism class of the profinite group ΠXK .

Acknowledgements: I would like to thank A. Tamagawa for the time that he so gen-
erously shared with me in numerous stimulating discussions, and especially for the
following: (i) informing me of the arguments used to prove Lemma 1.1.4 in §1.1; (ii) ex-
plaining to me the utility of a theorem of Raynaud in the context of §2 (cf. Lemma 2.4).
Also, I would like to thank F. Oort, as well as the referee, for various useful remarks.

Section 0: Notations and Conventions

Numbers:

We will denote by N the set of natural numbers, by which we mean the set of
integers n ≥ 0. A number field is defined to be a finite extension of the field of rational
numbers Q.

Topological Groups:

Let G be a Hausdorff topological group, and H ⊆ G a closed subgroup. Let us write


def
ZG (H) = {g ∈ G | g · h = h · g, ∀ h ∈ H}

for the centralizer of H in G;


def
NG (H) = {g ∈ G | g · H · g −1 = H}

for the normalizer of H in G; and



CG (H) = {g ∈ G | (g · H · g −1 ) H has finite index in H, g · H · g −1 }
def
ABSOLUTE ANABELIAN GEOMETRY 3

for the commensurator of H in G. Note that: (i) ZG (H), NG (H) and CG (H) are
subgroups of G; (ii) we have inclusions

H, ZG (H) ⊆ NG (H) ⊆ CG (H)

and (iii) H is normal in NG (H).


Note that ZG (H), NG (H) are always closed in G, while CG (H) is not necessarily
closed in G. Indeed, one may construct such an example as follows: Let

def

M = Zp
N

endowed with the product topology (of the various copies of Zp equipped with their usual
topology). Thus, M is a Hausdorff topological group. For n ∈ N, write F n (M) ⊆ M
for the sub-topological group given by the product of the copies of Zp indexed by
m ≥ n. Write AutF (M) for the set of automorphisms of the topological group M
that preserve the filtration F ∗ (M) on M. If α ∈ AutF (M), then for every n ∈ N,
α induces a continuous homomorphism αn : M/F n (M) → M/F n (M) which is clearly
surjective, hence an isomorphism (since M/F n (M) is profinite and topologically finitely
generated — cf. [FJ], Proposition 15.3). It thus follows that α induces an isomorphism

F n (M) → F n (M), hence that the inverse of α also lies in AutF (M). In particular, we
conclude that AutF (M) is a group. Equip AutF (M) with the coarsest topology for which
all of the homomorphisms AutF (M) → Aut(M/F n (M)) (where Aut(M/F n (M)) ∼ =
GLn (Zp ) is equipped with its usual topology) are continuous. Note that relative to
this topology, AutF (M) forms a Hausdorff topological group. Now define G to be the
semi-direct product of M with AutF (M) (so G is a Hausdorff topological group), and
H to be  
pn · Zp ⊆ Zp = M
n∈N N

(so H ⊆ G is a closed subgroup). Then CG (H) is not closed in G. For instance, if one

denotes by en ∈ N Zp the vector with a 1 in the n-th place and zeroes elsewhere, then
the limit A∞ (where
def
A∞ (en ) = en + en+1
def
for all n ∈ N) of the automorphisms Am ∈ CG (H) (where Am (en ) = en +en+1 if n ≤ m,
def
Am (en ) = en if n > m) is not contained in CG (H).

Definition 0.1.

(i) Let G be a profinite group. Then we shall say that G is slim if the centralizer
ZG (H) of any open subgroup H ⊆ G in G is trivial.
4 SHINICHI MOCHIZUKI

(ii) We shall say that a continuous homomorphism of profinite groups G → H is


relatively slim if the centralizer in H of the image of every open subgroup of G is trivial.

(iii) We shall say that a closed subgroup H ⊆ G of a profinite group G is commen-


surably (respectively, normally) terminal if the commensurator CG (H) (respectively,
normalizer NG (H)) is equal to H.

Remark 0.1.1. Thus, a profinite group G is slim if and only if the identity morphism
G → G is relatively slim. Moreover, if H ⊆ G is a closed subgroup such that the
inclusion H → G is relatively slim, then both H and G are slim.

Remark 0.1.2. It is a formal consequence of the definitions that:

commensurably terminal =⇒ normally terminal

and that (if H ⊆ G is a closed subgroup of a profinite group G, then):

H ⊆ G commensurably terminal, H slim =⇒


the inclusion H → G is relatively slim

Remark 0.1.3. It was pointed out to the author by F. Oort that a profinite group is
slim if and only if every open subgroup of G has trivial center. (Indeed, the necessity
of this condition is clear. Its sufficiency may be shown as follows: If H ⊆ G is an open
subgroup, then for any h ∈ ZG (H), let H  ⊆ G by the (necessarily open) subgroup
generated by H and h. Thus, h lies in the center of H  , which is trivial by assumption.)
This property of a profinite group — i.e., that every open subgroup be center-free —
was investigated in detail in [Naka1](cf., e.g., Corollaries 1.3.3, 1.3.4).

Curves:

Suppose that g ≥ 0 is an integer. Then a family of curves of genus g

X→S

is defined to be a smooth, proper, geometrically connected morphism X → S whose


geometric fibers are curves of genus g.
Suppose that g, r ≥ 0 are integers such that 2g − 2 + r > 0. We shall denote the
moduli stack of r-pointed stable curves of genus g (where we assume the points to be
unordered) by Mg,r (cf. [DM], [Knud] for an exposition of the theory of such curves;
strictly speaking, [Knud] treats the finite étale covering of Mg,r determined by ordering
ABSOLUTE ANABELIAN GEOMETRY 5

the marked points). The open substack Mg,r ⊆ Mg,r of smooth curves will be referred
to as the moduli stack of smooth r-pointed stable curves of genus g or, alternatively, as
the moduli stack of hyperbolic curves of type (g, r).
A family of hyperbolic curves of type (g, r)

X→S

is defined to be a morphism which factors X → Y → S as the composite of an open


immersion X → Y onto the complement Y \D of a relative divisor D ⊆ Y which is finite
étale over S of relative degree r, and a family Y → S of curves of genus g. One checks
easily that, if S is normal, then the pair (Y, D) is unique up to canonical isomorphism.
(Indeed, when S is the spectrum of a field, this fact is well-known from the elementary
theory of algebraic curves. Next, we consider an arbitrary connected normal S on which
a prime l is invertible (which, by Zariski localization, we may assume without loss of
generality). Denote by S  → S the finite étale covering parametrizing orderings of the
marked points and trivializations of the l-torsion points of the Jacobian of Y . Note that
S  → S is independent of the choice of (Y, D), since (by the normality of S), S  may be
constructed as the normalization of S in the function field of S  (which is independent
of the choice of (Y, D) since the restriction of (Y, D) to the generic point of S has
already been shown to be unique). Thus, the uniqueness of (Y, D) follows by considering
the classifying morphism (associated to (Y, D)) from S  to the finite étale covering of
(Mg,r )Z[ 1l ] parametrizing orderings of the marked points and trivializations of the l-
torsion points of the Jacobian [since this covering is well-known to be a scheme, for l
sufficiently large].) We shall refer to Y (respectively, D; D; D) as the compactification
(respectively, divisor at infinity; divisor of cusps; divisor of marked points) of X. A
family of hyperbolic curves X → S is defined to be a morphism X → S such that the
restriction of this morphism to each connected component of S is a family of hyperbolic
curves of type (g, r) for some integers (g, r) as above.

Section 1: Review of Anabelian Geometry

§1.1. The Anabelian Geometry of Number Fields

In this §, we review well-known anabelian (and related) properties of the Galois


groups of number fields and (mainly p-adic) local fields.

Let F be a number field. Fix an algebraic closure F of F and denote the resulting
absolute Galois group of F by GF . Let p be a (not necessarily nonarchimedean!) prime
of F . Write Gp ⊆ GF for the decomposition group (well-defined up to conjugacy)
6 SHINICHI MOCHIZUKI

associated to p and Fp for the completion of F at p. Then we have the following result
(cf. also Corollary 1.3.3 below for a generalization of the slimness of Gp , GF ):

Theorem 1.1.1. (Slimness and Commensurable Terminality) Suppose that p


is nonarchimedean. Then:

(i) The closed subgroup Gp ⊆ GF is commensurably terminal.

(ii) The inclusion Gp → GF is relatively slim. In particular, Gp , GF are slim.

Proof. Assertion (i) is a formal consequence of [NSW], Corollary 12.1.3. As for assertion
(ii), the slimness of Gp follows from local class field theory (cf., e.g., [Serre2]). (That is,
if σ ∈ Gp commutes with an open subgroup H ⊆ Gp , then σ induces the trivial action
on the abelianization H ab . But, by local class field theory, H ab may be identified with
the profinite completion of K × , where K is the finite extension of Fp determined by
H. Thus, σ acts trivially on all sufficiently large finite extensions K of Fp , so σ = 1,
as desired.) Relative slimness thus follows formally from the slimness of Gp and (i) (cf.
Remark 0.1.2).

Theorem 1.1.2. (Topologically Finitely Generated Closed Normal Sub-


groups) Every topologically finitely generated closed normal subgroup of GF is trivial.

Proof. This follows from [FJ], Theorem 15.10.

Theorem 1.1.3. (The Neukirch-Uchida Theorem on the Anabelian Nature


of Number Fields) Let F1 , F2 be number fields. Let F 1 (respectively, F 2 ) be an
algebraic closure of F1 (respectively, F2 ). Write Isom(F 2 /F2 , F 1 /F1 ) for the set of field

isomorphisms F 2 → F 1 that map F2 onto F1 . Then the natural map

Isom(F 2 /F2 , F 1 /F1 ) → Isom(Gal(F 1 /F1 ), Gal(F 2 /F2 ))

is bijective.

Proof. This is the content of [NSW], Theorem 12.2.1.

Remark 1.1.3.1. It is important to note, however, that the analogue of Theorem


1.1.3 for finite extensions of Qp is false (cf. [NSW], p. 674). Nevertheless, by considering
isomorphisms of Galois groups that preserve the higher ramification filtration, one may
obtain a partial analogue of this theorem for p-adic local fields (cf. [Mzk5]).
ABSOLUTE ANABELIAN GEOMETRY 7

Next, we would like to consider a situation that arises frequently in anabelian


geometry. Suppose that G is equal to GF or Gp (where we assume now that p is
nonarchimedean!), and that we are given an exact sequence of profinite groups:

1→Δ→Π→G→1

Suppose, moreover, that this sequence splits over some open subgroup of G, and that Δ
is topologically finitely generated. The following result was related to the author by A.
Tamagawa:

Lemma 1.1.4. (Intrinsic Characterization of Arithmetic Quotients)

(i) Suppose that G = GF . Let Π ⊆ Π be an open subgroup. Then the kernel of


the homomorphism Π → G may be characterized as the unique maximal closed normal
subgroup of Π which is topologically finitely generated.

(ii) Suppose that G = Gp . Assume further that for every open subgroup Π ⊆ Π,
def 
the abelianization (Δ )ab of Δ (where Δ = Π Δ) satisfies the following property:

(∗) The maximal torsion-free quotient (Δ )ab  Q of (Δ )ab on which the
def
action of G = Π /Δ (by conjugation) is trivial is a finitely generated free

Z-module.

Let Π ⊆ Π be an arbitrary open subgroup. Then:

[G : G ] · [Fp : Qp] = dimQp ((Π )ab ⊗Z Qp) − dimQl ((Π )ab ⊗Z Ql )

def  def
(where Δ = Δ Π ; G = Π /Δ ; p is the rational prime that p divides; and l is
any prime number distinct from p). (In fact, p may also be characterized as the unique
prime number for which the difference on the right is nonzero for infinitely many prime
numbers l.) In particular, the subgroup Δ ⊆ Π may be characterized as the intersection
of those open subgroups Π ⊆ Π such that:

[G : G ] = [Π : Π ]

(i.e., such that [G : G ] · [Fp : Qp] = [Π : Π ] · ([G : G] · [Fp : Qp ])).

Proof. Assertion (i) is a formal consequence of Theorem 1.1.2.


Now we turn to assertion (ii). Denote by K  the finite extension of Fp determined
by G . Then:
[G : G ] · [Fp : Qp] = [K  : Qp ]
8 SHINICHI MOCHIZUKI

Next, let us observe that we have an exact sequence

0 → Im(Δ ) → (Π )ab → (G )ab → 0


(where Im(Δ ) is the image of Δ in (Π )ab ) of finitely generated Z-modules. Note,
  
moreover, that we have a natural surjection Q  Im(Δ )/(torsion) (where Q is as
in (∗)). On the other hand, it follows formally from our assumption that Π  G
splits over an open subgroup of G (together with the well-known fact that the group
cohomology of a finite group is always annihilated by the order to the group) that this
natural surjection is, in fact, an isomorphism. In particular, it follows formally from (∗)
that:

dimQp ((Π )ab ⊗Z Qp )−dimQl ((Π )ab ⊗Z Ql ) = dimQp ((G )ab ⊗Z Qp )−dimQl ((G )ab ⊗Z Ql )

Thus, to complete the proof of Lemma 1.1.4, it suffices to prove that:

[K  : Qp ] = dimQp ((G )ab ⊗Z Qp ) − dimQl ((G )ab ⊗Z Ql )

But this is a formal consequence of local class field theory (cf., e.g., [Serre2]; §1.2 below),
i.e., the fact that (G )ab is isomorphic to the profinite completion of (K  )× .

Typically, in applications involving hyperbolic curves, one shows that the condition
(∗) of Lemma 1.1.4 is satisfied by applying the following:

Lemma 1.1.5. (Tate Modules of Semi-abelian Varieties) Let K be a finite


def
extension of Qp . Fix an algebraic closure K of K; write GK = Gal(K/K). Let A be a
semi-abelian variety over K. Denote the resulting (profinite) Tate module of A by:

def
T (A) = Hom(Q/Z, A(K))

Then the maximal torsion-free quotient T (A)  Q on which GK acts trivially is a



finitely generated free Z-module.

Proof. A semi-abelian variety is an extension of an abelian variety by a torus. Thus,


T (A) is the extension of the Tate module of an abelian variety by the Tate module of a
torus. Moreover, since (after restricting to some open subgroup of GK ) the Tate module

of a torus is isomorphic to the direct sum of a finite number of copies of Z(1), we thus
conclude that the image of the Tate module of the torus in Q is necessarily zero. In
particular, we may assume for the remainder of the proof without loss of generality that
A is an abelian variety.
ABSOLUTE ANABELIAN GEOMETRY 9

Now it follows from the theory of [FC](cf., in particular, [FC], Chapter III, Corollary
7.3), that T (A) fits into exact sequences (of GK -modules)

0 → Tgood → T (A) → Tcom → 0

0 → Ttrc → Tgood → T (B) → 0

where T (B) is the Tate module of an abelian variety B over K with potentially good
reduction; and Tcom = Mcom ⊗Z Z,  Ttrc = Mtrc ⊗Z Z(1)  for finitely generated free Z-
modules Mcom , Mtrc on which GK acts via a finite quotient. [Here, “com” (respectively,
“trc”) stands for “combinatorial” (respectively, “toric”).] It is thus evident that Ttrc
maps to 0 in Q. Moreover, by [Mzk4], Lemma 8.1 (the proof of which is valid for
arbitrary B, even though in loc. cit., this result is only stated in the case of a Jacobian),
and the Riemann Hypothesis for abelian varieties over finite fields (cf., e.g., [Mumf], p.
206), it follows that T (B) also maps to 0 in Q. Thus, we conclude that Q is equal to
the maximal torsion-free quotient of Tcom on which GK acts trivially. Since Z  is Z-flat,
however, this implies that Q is equal to the result of applying ⊗Z Z  to the maximal
torsion-free quotient of Mcom on which GK acts trivially. But this last quotient is
manifestly finitely generated and free over Z. This completes the proof.

§1.2. The Anabelian Geometry of p-adic Local Fields

In this §, we review certain well-known “group-theoretic” properties of Galois groups


of p-adic local fields, i.e., properties preserved by arbitrary isomorphisms between such
Galois groups.
For i = 1, 2, let pi be a prime number. Let Ki be a finite extension of Qpi . We
denote the ring of integers (respectively, maximal ideal; residue field) of Ki by OKi
(respectively, mKi ; ki ). Also, we assume that we have chosen an algebraic closures K i
of Ki and write
def
GKi = Gal(K i /Ki )

for the corresponding absolute Galois group of Ki . Thus, by local class field theory (cf.,
e.g., [Serre2]), we have a natural isomorphism

(Ki× )∧ → Gab
Ki

(where the “∧” denotes the profinite completion of an abelian group; “×” denotes the
group of units of a ring; and “ab” denotes the maximal abelian quotient of a topological
group). In particular, GabKi fits into an exact sequence

×
0 → OK → Gab 
i Ki → Z → 0
10 SHINICHI MOCHIZUKI

(arising from a similar exact sequence for (Ki× )∧ ). Moreover, we obtain natural inclu-
sions
ki× → OK
×
i
⊆ Ki× → GabKi

Ki× /OK
×
i
→ Z → Gab ×
Ki /Im(OKi )

(where “ → ” denotes the morphism induced by the valuation on Ki× ) by considering the
Teichmüller representatives of elements of ki× and the Frobenius element, respectively.
Also, in the following discussion we shall write:
×
def
μZ (K i ) = Hom(Q/Z, K i );
def ;
μZ (K i ) = μZ (K i ) ⊗Z Z
def
μQ/Z (K i ) = μZ (K i ) ⊗Z Q/Z

  def
(where Z  p ). Finally, we denote the cyclotomic character of GKi by:
= Z/Z


χi : GKi → Z

Proposition 1.2.1. (Invariants of Arbitrary Isomorphisms of Galois Groups


of Local Fields) Suppose that we are given an isomorphism of profinite groups:

α : GK1 → GK2

Then:

(i) We have: p1 = p2 . Thus, (in the remainder of this proposition and its proof)
def
we shall write p = p1 = p2 .

(ii) α induces an isomorphism IK1 → IK2 between the respective inertia sub-
groups of GK1 , GK2 .

(iii) The isomorphism αab : Gab K1 → GK2 induced by α preserves the images
ab
× × ×
Im(OKi ), Im(ki ), Im(Ki ) of the natural morphisms discussed above.
×
Ki /Im(OKi )
(iv) The morphism induced by α between the respective quotients Gab
preserves the respective Frobenius elements.

(v) [K1 : Qp] = [K2 : Qp ]; [k1 : Fp ] = [k2 : Fp ]. In particular, the ramification


indices of K1 , K2 over Qp coincide.

(vi) The morphisms induced by α on the abelianizations of the various open sub-
groups of the GKi induce an isomorphism

μQ/Z (K 1 ) → μQ/Z (K 2 )
ABSOLUTE ANABELIAN GEOMETRY 11

which is Galois-equivariant with respect to α. In particular, α preserves the cyclotomic


characters χi .

(vii) The morphism H 2 (K1 , μQ/Z (K 1 )) → H 2 (K2 , μQ/Z (K 2 )) induced by α (cf.
(vi)) preserves the “residue map”

H 2 (Ki , μQ/Z (K i )) → Q/Z

of local class field theory (cf. [Serre2], §1.1).

Proof. Property (i) follows by considering the ranks of Gab Ki over various Zl (cf. Lemma
1.1.4, (ii)). Property (iii) for Im(ki ) follows from the fact that Im(ki× ) may be recovered
×

as the prime-to-p torsion subgroup of Gab Ki . Property (v) follows for [Ki : Qp ] by
considering the Zp -rank of GKi (minus 1) and for [ki : Fp ] by considering the cardinality
ab

of Im(ki× ) (plus 1) — cf. (i), (iii). Property (ii) follows from the fact that whether or not
a finite extension is unramified may be determined group-theoretically by considering
×
the variation of the ramification index over Qp (cf. (v)). Property (iii) for Im(OK i
)
follows formally from (ii) (since this image is equal to the image in GKi of IKi ). Property
ab

(iv) follows by applying (iii) for Im(ki× ) to the various open subgroups of GKi that
correspond to unramified extensions of Ki and using the fact the Frobenius element
×
is the unique element that acts on ki (where k i denotes the algebraic closure of ki
induced by K i ) as multiplication by |k1 | = |k2 | [where, relative to our use of the term
×
“multiplication”, we think of the abelian group k i additively]. Here, we note that if Li
is a finite extension of Ki , then the inclusion
∼ × ∧ × ∧ ∼
Gab
Ki → (Ki ) → (Li ) → GLi
ab

may be reconstructed group-theoretically by considering the Verlagerung, or transfer,


map (cf. [Serre2], §2.4). Property (iii) for Im(Ki× ) follows formally from (iv). Property
(vi) follows formally from (iii). Finally, property (vii) follows (cf. the theory of the
Brauer group of a local field, as exposed, for instance, in [Serre2], §1) from the fact that

the morphism H 2 (Ki , μQ/Z (K i )) → Q/Z may be constructed as the composite of the
natural isomorphism
∼ ×
H 2 (Ki , μQ/Z (K i )) = H 2 (GKi , μQ/Z (K i )) → H 2 (GKi , K i )

— which is group-theoretic, by (iii) — with the inverse of the isomorphism


∼ ×
H 2 (Gal(Kiunr /Ki ), (Kiunr )× ) → H 2 (GKi , K i )

(where Kiunr denotes the maximal unramified extension of Ki ) — which is group-


theoretic, by (ii), (iii) — followed by the natural isomorphism
∼  Z) = Q/Z
H 2 (Gal(Kiunr /Ki ), (Kiunr )× ) → H 2 (Gal(Kiunr /Ki ), Z) = H 2 (Z,
12 SHINICHI MOCHIZUKI

— which is group-theoretic, by (ii), (iii), (iv).

Before proceeding, we observe that Proposition 1.2.1, (i), may be extended as


follows: Write 
def
AQ = |Spec(Z)| {∞}

(where “| − |” denotes the underlying set of a scheme) for the set of “all arithmetic
def
primes of Q”. If v ∈ AQ is equal to (0) ∈ |Spec(Z)| (respectively, ∞), set Gv = GQ
def
(respectively, Gv = Gal(C/R)). If v ∈ |Spec(Z)| ⊆ AQ is equal to the prime determined
def
by a prime number p, set Gv = GQp .

Proposition 1.2.2. (Intrinsicity of Arithmetic Types) For i = 1, 2, let vi ∈ AQ .


Suppose that Hi is an open subgroup of Gvi . Then H1 ∼
= H2 implies v1 = v2 .

Proof. Indeed, open subgroups of GQ may be distinguished by the fact that their
abelianizations fail to be topologically finitely generated. (Indeed, consider the abelian
extensions arising from adjoining roots of unity.) By contrast, abelianizations of open
subgroups of GR or GQp (cf. the above discussion) are topologically finitely generated.
Next, open subgroups of GR may be distinguished from those of any GQp by the fact
they are finite. The remainder of Proposition 1.2.2 follows from Proposition 1.2.1, (i).

Next, let us write Spec(OKi )log for the log scheme obtained by equipping the scheme
Spec(OKi ) with the log structure defined by the divisor V (mKi ). Thus, by pulling back
this log structure via the natural morphism Spec(ki ) → Spec(OKi ), we obtain a log
scheme Spec(ki )log , which we denote by

kilog

for short. Note that the “étale monoid” that defines the log structure on kilog “admits a
global chart” in the sense that it is defined by a single constant monoid (in the Zariski
topology of Spec(ki )) Mklog , which fits into a natural exact sequence (of monoids):
i

1 → ki× → Mklog → N → 0
i

Thus, the ki× -torsor Ui determined by considering the inverse image of 1 ∈ N in this
sequence may be identified with the ki× -torsor of uniformizers ∈ mKi considered modulo
m2Ki .
Next, let us write
GKi  Glog
ki
ABSOLUTE ANABELIAN GEOMETRY 13

for the quotient defined by the maximal tamely ramified extension Kitame of Ki . Thus,
Glog log
ki may also be thought of as the “logarithmic fundamental group” π1 (ki ) of the log
scheme kilog . Moreover, Glog
ki fits into a natural exact sequence:

1 → μZ (ki ) → Glog 
ki → Z → 1

where, just as in the case of Ki , we write:


def ×
μZ (k i ) = μZ (k i ) = Hom(Q/Z, k i );
def
μQ/Z (k i ) = μZ (k i ) ⊗Z Q/Z
The “abelianization” of this exact sequence yields an exact sequence:

1 → ki× → (Glog ab →1


→Z
ki )

Now we have the following:

Proposition 1.2.3.

(i) Any isomorphism α : GK1 → GK2 (as in Proposition 1.2.1) induces an isomor-

phism Glog log
k1 → Gk2 between the respective quotients.

(ii) There is a natural isomorphism of ki× -torsors between the torsor of uni-
formizers Ui discussed above and the H 1 (Z,  μ (k i )) = k × -torsor H 1 (Glog , μ (k i ))[1]
Z i ki Z
of elements of H 1 (Glog
ki , μ (k
Z i )) that map to the identity element in H 1
(μ (k
Z i ), μZ (k i )) =
HomZ (μZ (k i ), μZ (k i )). This isomorphism is defined by associating to a uniformizer
π ∈ Ui the μZ (k i )-torsor over kilog determined by the roots π 1/N , as N ranges (multi-
plicatively) over the positive integers prime to pi .

(iii) The profinite group Glog


ki is slim.

Proof. Property (i) follows from Proposition 1.2.1, (ii), together with the fact that
the quotient GKi  Glog ki may be identified with the quotient of GKi by the (unique)
maximal pro-p subgroup of IKi . Next, since any morphism of ki× -torsors is necessarily an
isomorphism, property (ii) follows by observing that the stated association of coverings
to uniformizers is indeed a morphism of ki× -torsors — a tautology, which may by verified
by considering the case N = qi − 1 (where qi is the cardinality of ki ), in which case
this tautology amounts to the computation: (ζ 1/N )qi = ζ · (ζ 1/N ) (for ζ ∈ ki× ). Finally,
property (iii) follows formally from the fact that the quotient Glog
ki /Im(μZ (k i )) is infinite
and acts faithfully on all open subgroups of the closed subgroup Im(μZ (k i )) ⊆ Glog ki .

In the following, let us denote by (kilog )∼ → kilog the “universal covering” of kilog
defined by the extension Kitame of Ki . Thus, Glogki acts naturally as the group of covering
log ∼ log
transformations of (ki ) → ki .
14 SHINICHI MOCHIZUKI

Proposition 1.2.4. (The “Grothendieck Conjecture” for the Logarithmic


Point) Suppose that we are given an isomorphism of profinite groups:

λ : Glog log
k1 → Gk2

Then:

(i) We have: |k1 | = |k2 |; p1 = p2 . Thus, (in the remainder of this proposition and
def
its proof) we shall write p = p1 = p2 .

(ii) λ preserves the subgroups Im(μZ (k i )) ⊆ Glog


ki as well as the Frobenius elements
log
in the quotients Gki /Im(μZ (k i )).

(iii) Assume further that the morphism


× ∼ ×
k 1 → k2
×
induced by λ (by thinking of k i as μQ/Z (k i )) arises from an isomorphism of fields

σ : k1 → k 2 . Then there exists a unique commutative diagram
(σ log )∼
(k1log )∼ −→ (k2log )∼
⏐ ⏐
⏐ ⏐

σlog
k1log −→ k2log
log ∼
of log schemes, compatible with the natural action of Glog
ki on (ki ) (for i = 1, 2), in
which the vertical morphisms are the natural morphisms, and the horizontal morphisms
are isomorphisms for which the morphisms on the underlying schemes are those in-
duced by σ.

Proof. Property (i) follows by observing that pi is the unique prime number such
that 1 plus the cardinality of the torsion subgroup of (Glog ki )
ab
— i.e., the cardinality
of ki — is equal to a power of pi . Property (ii) follows by thinking of the quotients
Glog log
ki /Im(μZ (k i )) as the quotients of Gki obtained by forming the quotient of (Gki )
log ab

by its torsion subgroup, and then using that the Frobenius element is the unique element
that acts on the abelian group Im(μZ (k i )) via multiplication by |k1 | = |k2 |. As for (iii),
the morphism σ log is the unique logarithmic extension of σ whose induced morphism

U1 → U2 is the morphism obtained (cf. Proposition 1.2.3, (ii)) by considering the
morphism induced by λ between the H 1 (Z,  μ (k i )) = k × -torsors H 1 (Glog , μ (k i ))[1]
Z i ki Z
(for i = 1, 2) — which are preserved by λ, by (ii). Note that here we also use (cf. (ii))
that the Frobenius element ∈ Z  is preserved, since this element is necessary to ensure
the compatibility of the identifications

 μ (k i )) = k ×
H 1 (Z, Z i
ABSOLUTE ANABELIAN GEOMETRY 15

(cf. Proposition 1.2.3, (ii)). The morphism (σlog )∼ is obtained by applying this con-
struction of “σ log ” to the various finite log étale coverings of kilog obtained by considering
various open subgroups of Glog ki . Here, the transition morphisms among coverings are
induced by the Verlagerung, as in the proof of Proposition 1.2.1. Finally, the uniqueness
of the lifting (σlog )∼ of σ log is a formal consequence of the fact that Glog ki is center-free
(cf. Proposition 1.2.3, (iii)).

§1.3. The Anabelian Geometry of Hyperbolic Curves

Characteristic Zero:

Let K be a field of characteristic 0 whose absolute Galois group is slim. Let X be


def
a hyperbolic curve of type (g, r) over S = Spec(K). Fix an algebraic closure K of K
def
and write s : Spec(K) → S; GK = Gal(K/K). Let x ∈ X(K) be a K-valued point of
def
X lying over s. Then, setting ΠX = π1 (X, x), we obtain an exact sequence

1 → ΔX → ΠX → GK → 1

which determines a well-defined continous homomorphism

GK → Out(ΔX )

to the group of outer automorphisms Out(ΔX ) of ΔX .

Lemma 1.3.1. (Slimness of Geometric and Arithmetic Fundamental


Groups) The profinite groups ΔX , ΠX are slim.

Proof. The slimness of ΠX is a formal consequence of the slimness of ΔX and our


assumption that GK is slim. Thus, it remains to prove that ΔX is slim. One approach
to proving this fact is given in [Naka1], Corollary 1.3.4. Another approach is the fol-
lowing: Let H ⊆ ΔX be an open normal subgroup for which the associated covering
def
XH → XK = X ×K K is such that XH is a curve of genus ≥ 2. Thus, H ab may be
thought of as the profinite Tate module associated to the generalized Jacobian of the
singular curve obtained from the unique smooth compactification of XH by identifying
the various cusps (i.e., points of the compactification not lying in XH ) to a single point.
In particular, if conjugation by an element δ ∈ ΔX induces the trivial action on H ab ,
then we conclude that the image of δ in ΔX /H induces the trivial action on the gen-
eralized Jacobian just discussed, hence on XH itself. But this implies that δ ∈ H. By
taking H to be sufficiently small, we thus conclude that δ = 1.

16 SHINICHI MOCHIZUKI

In particular, it follows formally from Lemma 1.3.1 that:

Corollary 1.3.2. (A Natural Exact Sequence) We have a natural exact sequence


of profinite groups:
1 → ΔX → Aut(ΔX ) → Out(ΔX ) → 1

In particular, by pulling back this exact sequence, one may recover the exact sequence
1 → ΔX → ΠX → GK → 1 entirely group-theoretically from the outer Galois represen-
tation GK → Out(ΔX ).

One example of the sort of “K” under consideration is the case of a “sub-p-adic
field”:

Corollary 1.3.3. (Slimness of Sub-p-adic Fields) The absolute Galois group of


a sub-p-adic field (i.e., a field isomorphic to a subfield of a finitely generated field
extension of Qp , for some prime number p) is slim.

Proof. This fact is implied by the argument of the proof of [Mzk6], Lemma 15.8.

In [Mzk6], the author (essentially) proved the following result (cf. [Mzk6], Theorem
A, for a stronger version that treats arbitrary dominant morphisms and open group
homomorphisms [i.e., which are not necessarily isomorphisms]):

Theorem 1.3.4. (“Sub-p-adic Profinite Grothendieck Conjecture”) Suppose


that K is a sub-p-adic field, and that X and Y are hyperbolic curves over K. Denote

by IsomK (X, Y ) the set of K-isomorphisms X → Y ; by IsomOut GK (ΔX , ΔY ) the set of
outer isomorphisms between the two profinite groups in parentheses that are compatible
with the respective outer actions of GK . Then the natural map

IsomK (X, Y ) → IsomOut


GK (ΔX , ΔY )

is bijective.

Thus, by combining Theorems 1.1.3; 1.3.4; Lemma 1.1.4, (i), we obtain the follow-
ing:

Corollary 1.3.5. (Absolute Grothendieck Conjecture over Number Fields)


Let K, L be number fields; X (respectively, Y ) a hyperbolic curve over K (re-

spectively, L). Denote by Isom(X, Y ) the set of scheme isomorphisms X → Y ; by
ABSOLUTE ANABELIAN GEOMETRY 17

IsomOut (ΠX , ΠY ) the set of outer isomorphisms between the two profinite groups in
parentheses. Then the natural map

Isom(X, Y ) → IsomOut (ΠX , ΠY )

is bijective.

Remark 1.3.5.1. Since the analogue of Theorem 1.1.3 in the p-adic local case is false,
it seems unlikely to the author at the time of writing that the analogue of Corollary
1.3.5 should hold over p-adic local fields.

Remark 1.3.5.2. In fact, the statement of Corollary 1.3.5 holds when K is taken
to be an arbitrary finitely generated field extension of Q. This follows by applying a
theorem of F. Pop (in place of the Neukirch-Uchida theorem) — cf. the discussion on
[NSW], p. 677, for more details.

One interesting result in the present context is the following, due to M. Matsumoto
(cf. [Mtmo], Theorems 2.1, 2.2):

Theorem 1.3.6. (Kernels of Outer Galois Representations) Let X be an


affine hyperbolic curve over a sub-complex field K — i.e., a field isomorphic to
a subfield of the field of complex numbers. Then the kernel of the resulting outer Galois
representation
GK → Out(ΔX )

is contained in the kernel of the natural homomorphism GK → GQ .

Remark 1.3.6.1. Thus, in particular, when K is a number field, the homomorphism


GK → Out(ΔX ) is injective. This injectivity was first proven by Belyi in the case of
hyperbolic curves of type (g, r) = (0, 3). It was then conjectured by Voevodskii to be
true for all (hyperbolic) (g, r) and proven by Voevodskii to be true for g = 1. Finally,
it was proven by Matsumoto to hold for all (g, r) such that r > 0. To the knowledge
of the author, the proper case remains open at the time of writing. We refer to the
discussion surrounding [Mtmo], Theorem 2.1, for more details on this history.

Remark 1.3.6.2. One interesting aspect of the homomorphism appearing in Theorem


1.3.6 is that it allows one to interpret Theorem 1.3.4 (when X = Y ) as a computation
of the centralizer of the image of this homomorphism GK → Out(ΔX ). This point of
view is surveyed in detail in [Naka2].
18 SHINICHI MOCHIZUKI

Next, we would like to discuss various properties of the inertia groups of the cusps
def
of a hyperbolic curve. For every cusp x of XK = X ×K K — i.e., point of the unique
smooth compactification of XK over K that does not lie in XK — we have an associated

inertia group (abstractly isomorphic to Z)

Ix ⊆ Δ X

(well-defined, up to conjugation). If l is any prime number, then let us denote the


maximal pro-l quotient of a profinite group by means of a superscript “(l)”. Thus, we
(l) (l)
also obtain an inertia group Ix ⊆ ΔX (abstractly isomorphic to Zl ).

Lemma 1.3.7. (Commensurable Terminality of Inertia Groups) The sub-


(l) (l)
groups Ix ⊆ ΔX , Ix ⊆ ΔX are commensurably terminal.

Proof. Indeed, let σ be an element of the commensurator. If the asserted commensu-


rable terminality is false, then by projecting to a finite quotient, we may assume that
we have a finite Galois covering
Z → XK
(of degree a power of l in the pro-l case), together with a cusp z of Z such that: (i) z
maps to x; (ii) z = z σ ; (iii) z, z σ have conjugate inertia groups in ΔZ . We may also
assume (by taking ΔZ ⊆ ΔX to be sufficiently small) that Z has genus ≥ 2 and admits
a cusp z  = z, z σ . Then it is easy to see that Z admits an infinite abelian (pro-l, in the
pro-l case) covering which is totally ramified at z, z  , but not at z σ . But this contradicts
property (iii).

Remark 1.3.7.1. As was pointed out to the author by the referee, in the case r ≥ 2,
the assertion of Lemma 1.3.7 is a formal consequence of Theorem B of [HR].

Now, let us assume that we are given two hyperbolic curves (Xi )Ki (for i = 1, 2),
each defined over a finite extension Ki of Qpi . Let us write qi for the cardinality of
the residue field of Ki . Choose basepoints for the (Xi )Ki and denote the resulting
fundamental groups by Π(Xi )Ki . Also, let us denote the unique proper smooth curve
over Ki that compactifies (Xi )Ki by (Yi )Ki . Suppose, moreover, that we are given an
isomorphism

αX : Π(X1 )K1 → Π(X2 )K2
of profinite groups.

Lemma 1.3.8. (Group-Theoreticity of Arithmetic Quotients) The isomor-


phism αX is necessarily compatible with the quotients Π(Xi )Ki  GKi .
ABSOLUTE ANABELIAN GEOMETRY 19

Proof. This follows formally from Lemmas 1.1.4, 1.1.5.

Thus, Lemma 1.3.8, Proposition 1.2.1, (v), imply that q1 = q2 .

Lemma 1.3.9. (Group-Theoreticity of the Cusps) The types (gi , ri ) of the


hyperbolic curves (Xi )Ki coincide. In particular, for any prime number l, αX maps
(l)
inertia groups of cusps in ΔX1 (respectively, ΔX1 ) to inertia groups of cusps in ΔX2
(l)
(respectively, ΔX2 ).

Proof. Whether or not ri = 0 may be determined by considering whether or not ΔXi


is free as a profinite group. When ri > 0, one may compute ri by considering the weight
— i.e., the number w such that the eigenvalues of the action are algebraic numbers of
archimedean absolute value qiw — of the action of the Frobenius element ∈ Gki (cf.
Proposition 1.2.1, (iv)) as follows: First, we observe that (as is well-known) the weights
of the action of Frobenius on Δab Xi ⊗ Ql (where l is a prime number distinct from p1 ,
p2 ) belong to the set {0, 1, 2 }. (Here, we compute weights by choosing a lifting of the
1

Frobenius element ∈ Gki to GKi . Since (as is well-known) the action of the inertia
subgroup of GKi on Δab Xi ⊗ Ql is quasi-unipotent (i.e., unipotent on an open subgroup of
this inertia subgroup), it follows immediately that the weights are independent of the
choice of lifting.) Now if M is a Ql -vector space on which Frobenius acts, let us write

M wt w

for the Ql -subspace of M on which Frobenius acts with eigenvalues of weight w. Then,
def
setting Ii = Ker(ΔabXi ⊗ Ql  ΔYi ⊗ Ql ), we have:
ab

ri − 1 = dimQl (Ii ) = dimQl (Iiwt 1 )


Xi ⊗ Ql )
= dimQl (Δab − dimQl (Δab
Yi ⊗ Ql )
wt 1 wt 1

Xi ⊗ Ql )
= dimQl (Δab − dimQl (Δab
Yi ⊗ Ql )
wt 1 wt 0

Xi ⊗ Ql )
= dimQl (Δab − dimQl (Δab
X i ⊗ Ql )
wt 1 wt 0

(where the fourth equality follows from the auto-duality (up to a Tate twist) of Δab Yi ⊗
Ql ; and the second and fifth equalities follow from the fact that Frobenius acts on Ii
with weight 1). On the other hand, the quantities appearing in the final line of this
sequence of equalities are all “group-theoretic”. Thus, we conclude that r1 = r2 . Since
Xi ⊗ Ql ) = 2gi − 1 + ri (respectively, = 2gi ) when ri > 0 (respectively, when
dimQl (Δab
ri = 0), this implies that g1 = g2 , as desired.
Finally, the statement concerning preservation of inertia groups follows formally
from the fact that “ri is group-theoretic” (by applying this fact to coverings of Xi ).
20 SHINICHI MOCHIZUKI

Indeed, let l be a prime number (possibly equal to p1 or p2 ). Since ri may be recovered


group-theoretically, given any finite étale coverings

Zi → Vi → Xi

such that Zi is Galois, of degree a power of l, over Vi , one may determine group-
theoretically whether or not Zi → Vi is “totally ramified at a single point of Zi and
unramified elsewhere”, since this condition is easily verified to be equivalent to the
equality:
rZi = deg(Zi /Vi ) · (rVi − 1) + 1
Moreover, the group-theoreticity of this condition extends immediately to the case of
pro-l coverings Zi → Vi . Thus, by Lemma 1.3.7, we conclude that the inertia groups of
cusps in (ΔXi )(l) (i.e., the maximal pro-l quotient of ΔXi ) may be characterized (group-
theoretically!) as the maximal subgroups of (ΔXi )(l) that correspond to (profinite)
coverings satisfying this condition.
In particular, (by Lemma 1.3.7) the set of cusps of Xi may be reconstructed (group-
theoretically!) as the set of (ΔXi )(l) -orbits (relative to the action via conjugation) of
such inertia groups in (ΔXi )(l) . Thus, by applying this observation to arbitrary finite
étale coverings of Xi , we recover the inertia subgroups of the cusps of ΔXi as the
subgroups that fix some cusp of the universal covering X
i → Xi of Xi determined by
the basepoint in question. This completes the proof.

Remark 1.3.9.1. As was pointed out to the author by the referee, the argument
given in the second paragraph of the proof of Lemma 1.3.9 may be replaced by the
more group-theoretic argument of [Tama2], Proposition 2.4.

Positive Characteristic:

For i = 1, 2, let ki be a finite field of characteristic p; Xi a hyperbolic curve over ki .


Choose a universal tamely ramified (i.e., at the punctures of Xi ) covering X
i → Xi of
Xi ; write
Πtame
def
= Gal(X
i /Xi )
Xi

for the corresponding fundamental groups. Thus, we obtain exact sequences:

1 → Δtame
Xi → Πtame
Xi → Gki → 1


i ). As is well-known, the
(where Gki is the absolute Galois group of ki determined by X
Frobenius element determines a natural isomorphism Z ∼
= Gki .

Lemma 1.3.10. (Slimness of Fundamental Groups) For i = 1, 2, the profinite


tame tame
groups ΔXi , ΠXi are slim.
ABSOLUTE ANABELIAN GEOMETRY 21

Proof. The slimness of ΔtameXi follows by exactly the same argument — i.e., by consid-
ering the action of Δtame
Xi on abelianizations of open subgroups — as that given in the
proof of Lemma 1.3.1. [Alternatively, as was pointed out to the author by the referee,
the slimness of Δtame
Xi follows from [Tama1], Proposition 1.11 (cf. Remark 0.1.3).]
By a similar argument, the slimness of Πtame
Xi follows formally from:

(i) the slimness of Δtame


Xi ;

(ii) some positive power of the character of Gki arising from the determinant of
the prime-to-p portion of (Δtame
Xi )
ab
coincides with some positive power of the
cyclotomic character.

 ∼
(Here, we note (ii) is sufficient to deal with both the “l-primary portion” of Z = Gki
 ∼
(for l = p) and the “p-primary portion” of Z = Gki .)

Remark 1.3.10.1. Property (ii) in the proof of Lemma 1.3.10 was substantially
simplified by a suggestion made to the author by the referee.

The following fundamental result is due to A. Tamagawa (cf. [Tama1], Theorem


4.3):

Theorem 1.3.11. (The Grothendieck Conjecture for Affine Hyperbolic


Curves over Finite Fields) Assume, for i = 1, 2, that Xi is affine. Then the natural
map
Isom(X
1 /X1 , X

2 /X2 ) → Isom(Πtame , Πtame )
X1 X2


1 →∼

(from scheme-theoretic isomorphisms X X2 lying over an isomorphism X1 → X2

to isomorphisms of profinite groups Πtame
X1 → Πtame
X2 ) is bijective.

Finally, we observe that, just as in the characteristic zero case, inertia groups of
def
cusps are commensurably terminal: If xi is a cusp of (Xi )ki = Xi ×ki k i , then we have
)
an associated inertia group (abstractly isomorphic to Z

Ixi ⊆ Δtame
Xi

(well-defined, up to conjugation). If l is any prime number distinct from p, then we also


(l)
obtain an inertia group Ixi ⊆ (Δtame
Xi )
(l)
(abstractly isomorphic to Zl ).

Lemma 1.3.12. (Commensurable Terminality of Inertia Groups) The


(l)
subgroups Ixi ⊆ (ΔXi ) , Ixi ⊆ Δtame
tame (l)
Xi are commensurably terminal.
22 SHINICHI MOCHIZUKI

Proof. The proof is entirely similar to that of Lemma 1.3.7. [Alternatively, as was
pointed out to the author by the referee, the assertion concerning Ixi ⊆ Δtame
Xi follows
(l)
from [Tama2], Lemma 2.1, while the assertion concerning Ixi ⊆ (Δtame
Xi )
(l)
follows, at

least in the case r ≥ 2, formally from Theorem B of [HR].]

Section 2: Reconstruction of the Logarithmic Special Fiber

For i = 1, 2, let Ki be a finite extension of Qpi (cf. §1.2), and suppose that we are
given a hyperbolic curve (Xi )Ki over Ki . Let us fix a K i -valued basepoint for (Xi )Ki
and denote the resulting fundamental group π1 ((Xi )Ki ) by Π(Xi )Ki . Suppose, moreover,

that we are given an isomorphism αX : Π(X1 )K1 → Π(X2 )K2 , which, by Lemma 1.3.8,
necessarily fits into a commutative diagram
αX
Π(X1 )K1 −→ Π(X2 )K2
⏐ ⏐
⏐ ⏐

α
GK1 −→
K
GK2

where the vertical morphisms are the natural ones, and the horizontal morphisms are
assumed to be isomorphisms. Note that by Proposition 1.2.1, (i), this already implies
def
that p1 = p2 ; set p = p1 = p2 . That such a diagram necessarily arises “geometrically”
follows from the main theorem of [Mzk6] (cf. Theorem 1.3.4) — if one assumes that αK

arises geometrically (i.e., from an isomorphism of fields K1 → K2 ). In this §, we would
like to investigate what one can say in general (i.e., without assuming that αK arises
geometrically) concerning this sort of commutative diagram. In some sense, all the
key arguments that we use here are already present in [Mzk4], except that there, these
arguments were applied to prove different theorems. Thus, in the following discussion,
we explain how the same arguments may be used to prove Theorem 2.7 below.
Let us denote the type of the hyperbolic curve (Xi )Ki by (gi , ri ). Also, we shall
denote the geometric fundamental group by

def
ΔXi = Ker(Π(Xi )Ki → GKi )

and the unique proper smooth curve over Ki that compactifies (Xi )Ki by (Yi )Ki .

Lemma 2.1. (Group-Theoreticity of Stability) (X1 )K1 has stable reduction if


and only if (X2 )K2 does.

Proof. This follows (essentially) from the well-known stable reduction criterion: That
is to say, (Xi )Ki has stable reduction if and only if the inertia subgroup of GKi acts
ABSOLUTE ANABELIAN GEOMETRY 23


Yi ⊗ Z and trivially on the (finite) set of conjugacy classes of inertia
unipotently on Δab
groups of cusps in ΔXi (i.e., the set of cusps of (Xi )Ki ⊗Ki K i — cf. Lemma 1.3.9).
[Note that this condition is group-theoretic, by Proposition 1.2.1, (ii).]

Now let us assume that (Xi )Ki has stable reduction over OKi . Denote the stable
model of (Xi )Ki over OKi by:

(Xi )OKi → Spec(OKi )

Here, in the case where ri > 0, we mean by the term “stable model” the complement of
the marked points in the unique stable pointed curve over OKi that extends the pointed
curve over Ki determined by (Xi )Ki . Then, by the theory of [Mzk4], §2, 8, there exists
a well-defined quotient
Π(Xi )Ki  Πadm
(Xi )Ki

whose finite quotients correspond to (subcoverings of) admissible coverings of the result
of base-changing (Xi )OKi to rings of integers of tamely ramified extensions of Ki . In
particular, we have a natural exact sequence:
log
1 → Δadm
Xi → Π(Xi )K → Gki → 1
adm
i

(where Δadm adm


Xi is defined so as to make the sequence exact). Moreover, Π(Xi )Ki itself
admits a natural quotient (cf. [Mzk4], §3)

Π(Xi )Ki  Πadm


(Xi )K  Π(Xi )K
et
i i

whose finite quotients correspond to coverings of (Xi )Ki that extend to finite étale
coverings of (Xi )OKi which are tamely ramified at the cusps. In particular, we have a
natural exact sequence:

Xi → Π(Xi )K → Gki → 1
1 → Δet et
i

(where Δet
Xi is defined so as to make the sequence exact).

Lemma 2.2. (Admissible and Étale Quotients)

(i) The profinite groups Π(Xi )Ki , Πadm adm


(Xi )K , and ΔXi are all slim.
i

(ii) The morphism αX is compatible with the quotients

Π(Xi )Ki  Πadm


(Xi )K  Π(Xi )K
et
i i

of Π(Xi )Ki .
24 SHINICHI MOCHIZUKI

Proof. We begin with assertion (i). The slimness of Π(Xi )Ki follows formally from
Theorem 1.1.1, (ii); Lemma 1.3.1. Next, let us assume that Δadm Xi has already been
shown to be slim. Then it follows formally from Proposition 1.2.3, (iii), that Πadm
(Xi )Ki is
slim.
Thus, to complete the proof of assertion (i), it suffices to show that Δadm Xi is slim.
We argue as in Lemma 1.3.1, i.e., we consider the covering associated to an open normal
subgroup H ⊆ Δadm Xi . To simplify notation, we assume that (Xi )ki is “that covering”;
then it suffices to show (cf. the proof of Lemma 1.3.1) that any automorphism (ki -linear)
σ of (Xi )ki which acts trivially on (Δadm
Xi )
ab
is itself trivial. We may assume without
loss of generality [i.e., by taking H to be sufficiently small] that (Xi )ki is sturdy (cf.
[Mzk4], Definition 1.1) — i.e., that the normalizations of all the geometric irreducible
components of (Xi )ki have genus ≥ 2. Then by [Mzk4], Proposition 1.3, it follows that
σ acts trivially on the set of geometric irreducible components of (Xi )ki . In particular,
σ acts on each such geometric irreducible component individually. Thus, by considering
the action of σ on the torsion points of the Jacobians of each these geometric irreducible
components of (Xi )ki (cf. the proof of Lemma 1.3.1), we conclude that σ acts as the
identity on each of these geometric irreducible components, as desired.
Next, we turn to assertion (ii). For Πadm
(Xi )Ki , this follows (essentially) from Propo-
sition 8.4 of [Mzk4] (together with Lemma 1.3.9, which shows that αX preserves the
pro-p inertia groups associated to the cusps in Π(Xi )Ki ). Of course, in [Mzk4], K1 = K2
and αK is the identity, but in fact, the only property of αK necessary for the proof of
[Mzk4], Proposition 8.4 — which is, in essence, a formal consequence of [Mzk4], Lemma
8.1 (concerning unramified quotients of the p-adic Tate module of a semi-abelian vari-
ety over a p-adic local field) — is that αK preserve the inertia and wild inertia groups
(which we know, by Proposition 1.2.1, (ii); Proposition 1.2.3, (i), of the present paper).
Similarly, the portion of assertion (ii) concerning Πet
(Xi )Ki follows (essentially) from
[Mzk4], Proposition 3.2, at least in the case ri = 0. That is to say, even though αK is
not necessarily the identity in the present discussion, the only properties of αK that are
necessary for the proof of [Mzk4], Proposition 3.2, are Proposition 1.2.3, (i); Proposition
1.2.4, (ii) (of the present paper).
Finally, to treat the case of arbitrary ri > 0, we argue as follows: Consider an
open normal subgroup H ⊆ Πadm (Xi )K . Then observe that there exists an open normal
i
subgroup H  ⊆ Πadm
(Xi )Ki such that the orders of the finite groups Π(Xi )Ki /H, Π(Xi )Ki /H
adm adm 

are relatively prime and such that the covering of (Xi )ki determined by H  is sturdy.
Moreover, [by the assumption of relative primeness] it follows easily that the covering
determined by H is “of étale type” (i.e., arises from a quotient of Πet (Xi )Ki ) if and only
if it becomes a covering “of étale type” after base-change via the covering determined
by H  . Thus, we conclude that we may assume without loss of generality that (Xi )ki is
sturdy.
ABSOLUTE ANABELIAN GEOMETRY 25

Now, let us observe that since (Xi )ki is sturdy, it admits admissible coverings “of
étale type” with arbitrarily large prime-to-p ramification at the cusps. Thus, [since the
condition that an admissible covering be “of étale type” amounts to the condition that
there be no ramification at the nodes or at the prime of OKi ] we conclude that the
covering of (Xi )ki determined by H is of étale type if and only if, for every open normal
H  ⊆ Πadm
(Xi )K which has ramification index divisible by the prime-to-p portion of the
i
order of the finite group Πadm(Xi )Ki /H at all of the cusps, it becomes a covering of étale
type after base-change by the covering determined by H  . [Indeed, the necessity
of this condition is clear; the sufficiency of this condition follows from the existence of
coverings of étale type with arbitrarily large prime-to-p ramification at the cusps. Also,
we note that by Lemma 1.3.9, this condition is preserved by αX .] On the other hand,
such base-changed coverings are [by construction] necessarily unramified at the cusps.
Thus, the issue of whether or not this base-changed covering is of étale type reduces
— by Lemma 1.3.9, which shows that αX preserves the quotient Π(Xi )Ki  Π(Yi )Ki
— to the “ri = 0” case, which has already been resolved. This completes the proof of
assertion (ii) in the case ri > 0.

Lemma 2.3. (Group-Theoreticity of Dual Semi-Graphs of the Special


Fiber) The morphism αX induces an isomorphism

αX,Γc : Γc(X1 )k → Γc(X2 )k
1 2

between the “dual semi-graphs with compact structure” (i.e., the usual dual graphs
Γ(Xi )ki , together with extra edges corresponding to the cusps — cf. the Appendix) of
the special fibers (Xi )ki of (Xi )OKi . Moreover, αX,Γc is functorial with respect to pas-
sage to finite étale coverings of the (Xi )Ki .

Proof. Indeed, if one forgets about the “compact structure”, then this is a formal
consequence of Lemma 1.3.9 [which shows that αX preserves the quotient Π(Xi )Ki 
Π(Yi )Ki ]; Lemma 2.2, (ii), and the theory [concerning the ri = 0 case] of [Mzk4], §1 – 5,
summarized in [Mzk4], Corollary 5.3. Even though αK is not necessarily the identity
in the present discussion, the only properties of αK that are necessary for the proof of
[Mzk4], Corollary 5.3 are Proposition 1.2.3, (i); Proposition 1.2.4, (ii) (of the present
paper). That is to say, the point is that the Frobenius element is preserved, which means
that the weight filtrations on l-adic cohomology (where l is a prime distinct from p) are,
as well.
The compatibility with the “compact structure” follows from the pro-l (where l = p)
portion of Lemma 1.3.9, together with the easily verified fact (cf. the proof of Lemma
1.3.7) that the inertia group of a cusp in Πadm
(Xi )Ki is contained (up to conjugacy) in the
decomposition group of a unique irreducible component of (Xi )ki .

26 SHINICHI MOCHIZUKI

Next, we would like to show that αX is necessarily “of degree 1”. This is essentially
the argument of [Mzk4], Lemma 9.1, but we present this argument in detail below since
we are working here under the assumption that αK is arbitrary. For simplicity, we
assume until further notice is given that:

(∗CSSN ) ri = 0 [i.e., the curves are compact] and, moreover, the special fiber
(Xi )ki of (Xi )OKi is singular and sturdy (cf. [Mzk4], Definition 1.1) — i.e.,
the normalizations of all the geometric irreducible components of (Xi )ki have
genus ≥ 2 — and has a noncontractible dual semi-graph Γc(Xi )k — i.e., this
i
semi-graph is not a tree.

(These conditions may always be achieved by replacing (Xi )Ki by a finite étale covering
of (Xi )Ki — cf. [Mzk4], Lemma 2.9; [Mzk4], the first two paragraphs of the proof of
Theorem 9.2.)
We begin by introducing some notation. Write:
def
Vi = Δab
Xi ;
def
Hi = H1sing (Γc(Xi )k , Z) = H1sing (Γ(Xi )ki , Z)
i

(where “H1sing ”
denotes the first singular homology group). Thus, by considering the
coverings of (Xi )OKi induced by unramified coverings of the graph Γ(Xi )ki , we obtain
natural (group-theoretic!) “combinatorial quotients”:


def
Vi  (Hi )Z = Hi ⊗ Z

Lemma 2.4. (Ordinary New Parts, after Raynaud) For a “sufficiently


large prime number l” (where “sufficiently large” depends only on p, gi ), and after
possibly replacing Ki by a finite unramified extension of Ki , there exists a cyclic étale
def
covering (Zi )OKi → (Xi )OKi of (Xi )OKi of degree l such that the “new part” Vinew =
(Zi )Ki /Δ(Xi)Ki of the abelianized geometric fundamental group of (Zi )Ki satisfies the
Δab ab

following:

(i) We have an exact sequence:


def
0 → Vimlt → (Vinew )Zp = Vinew ⊗Z Zp → Vietl → 0

— where Vietl is an unramified GKi -module, and Vimlt is the Cartier dual
of an unramified GKi -module.

(ii) The “combinatorial quotient” of Δab (Zi )Ki (arising from the coverings of
(Zi )OKi induced by unramified coverings of the dual semi-graph of the special
fiber of (Zi )OKi ) induces a nonzero quotient Vinew  (Hinew )Z of Vinew .
ABSOLUTE ANABELIAN GEOMETRY 27

(Xi)Ki → Δ(Zi )Ki is the injection induced by pull-back via (Zi )Ki →
Here, the injection Δab ab

(Xi )Ki and Poincaré duality (or, alternatively, by the “Verlagerung”).

Proof. Note that since both conditions (i), (ii) are group-theoretic, they may be realized
simultaneously for i = 1, 2. Now to satisfy condition (i), it suffices — cf., e.g., the
discussion in [Mzk4], §8, of “VG ”, “VGord ” — to choose the covering so that the “new
parts” of the Jacobians of the irreducible components of the special fiber of (Zi )OKi are
all ordinary. That this is possible for l sufficiently large is a consequence of a theorem of
Raynaud (as formulated, for instance, in [Tama1], Lemma 1.9). Then let us observe that,
so long as we choose the étale covering (Zi )OKi → (Xi )OKi so that it is nontrivial over
every irreducible component of (Xi )ki , condition (ii) is automatically satisfied: Indeed,
def
if we write hi = rkZ (Hi ) — so hi > 0 since Γc(Xi )k is assumed to be noncontractible —
i
then to assert that condition (ii) fails to hold — i.e., that there are “no new cycles in
the dual graph” — is to assert that we have an equality of Euler characteristics:
 
gZ,j + hi − 1 = l gX , j + hi − 1
j j

(where the first (respectively, second) sum is the sum of the genera of the irreducible
components of the geometric special fiber of (Zi )OKi (respectively, (Xi )OKi )). But, since

(gZ,j − 1) = l(gX, j − 1)
j j

  
we thus conclude that (l − 1) = j (l − 1) + hi (l − 1), hence that 1 = ( j 1) + hi
— which is absurd, since both the sum and hi are ≥ 1. This completes the proof.

Remark 2.4.1. The author would like to thank A. Tamagawa for explaining to him
the utility of Raynaud’s theorem in this sort of situation.

In the following discussion, to keep the notation simple, we shall replace (Xi )Ki by
some (Zi )Ki as in Lemma 2.4. Thus, Vinew is a GKi -quotient module of Vi . Moreover,
we have a surjection
Vinew  (Hinew )Z

such that the quotient (Hi )Z  (Hinew )Z is defined over Z, i.e., arises from a quo-
tient Hi  Hinew . (Indeed, this last assertion follows from the fact that the quotient
Hi  Hinew arises as the cokernel (modulo torsion) of the morphism induced on first sin-
gular cohomology modules by a finite (ramified) covering of graphs — i.e., the covering
induced on dual graphs by the covering (Zi )OKi → (Xi )OKi of Lemma 2.4.)
28 SHINICHI MOCHIZUKI

On the other hand, the cup product on group cohomology gives rise to a nondegen-
erate (group-theoretic!) pairing

Vi∨ ⊗Z Vi∨ ⊗Z μZ (K i ) → Mi = H 2 (ΔXi , μZ (K i )) (∼ 


def
= Z)

def  as H 1 (ΔX , Z)),


(where we think of Vi∨ = Hom(Vi , Z)  hence, by restriction to
i
new ∨ ∨
(Vi ) → Vi , a pairing

(Vinew )∨ ⊗Z (Vinew )∨ ⊗Z μZ (K i ) → Mi = H 2 (ΔXi , μZ (K i )) (∼ 


def
= Z)

which is still nondegenerate (over Q), since it arises from an ample line bundle — namely,
the restriction of the polarization determined by the theta divisor on the Jacobian of
(Xi )Ki to the “new part” of (Xi )Ki . This pairing determines an “isogeny” (i.e., a
morphism which is an isomorphism over Q):

(Vinew )∨ ⊗Z μZ (K i ) ⊗Z Mi∨ → Vinew

Thus, if we take the dual of the surjection discussed in the preceding paragraph, then
we obtain an inclusion

(Hinew )∨
Z
⊗ μZ (K i ) ⊗ Mi∨ → (Vinew )∨ ⊗ μZ (K i ) ⊗ Mi∨ → Vinew

which (as one sees, for instance, by applying the fact that μZ (K i )GKi = 0) maps into
the kernel of the surjection Vinew  (Hinew )Z .
Next, let us observe that the kernel Ni of the surjection of unramified GKi -modules
(i.e., Gki -modules)
Vietl  (Hinew )Zp
satisfies:
H 0 (Gki , Ni ⊗ Qp ) = H 1 (Gki , Ni ⊗ Qp ) = 0
(Indeed, Ni arises as a submodule of the module of p-power torsion points of an abelian
variety over ki , so the vanishing of these cohomology groups follows from the Riemann
Hypothesis for abelian varieties over finite fields (cf., e.g., [Mumf], p. 206), i.e., the
fact that (some power of) the Frobenius element of Gki acts on Ni with eigenvalues
which are algebraic numbers with complex absolute values equal to a nonzero rational
power of p.) In particular, we conclude that the above surjection admits a unique
GKi -equivariant splitting (Hinew )Zp → (Vietl )Qp . Similarly, (by taking Cartier duals)
the injection (Hinew )∨ ∨
Zp ⊗ μZp (K i ) ⊗ Mi → Vi
mlt
also admits a unique GKi -equivariant
splitting over Qp . Thus, by applying these splittings, we see that the GKi -action on
(Vinew )Zp determines a p-adic extension class

(ηi )Zp ∈ {(Hinew )∨


Qp }
⊗2
⊗Mi∨ ⊗(H 1 (Ki , μZ (K i ))/Hf1 (Ki , μZ (K i ))) = {(Hinew )∨
Qp }
⊗2
⊗Mi∨
ABSOLUTE ANABELIAN GEOMETRY 29

where (by Proposition 1.2.1, (vii)) H 1 (Ki , μZ (K i )) may be identified with (Ki× )∧ , and
we define
× ∼
⊆ (Ki× )∧ → H 1 (Ki , μZ (K i ))
def
Hf1 (Ki , μZ (K i )) = OK i


so the quotient group (H 1 (Ki , μZ (K i ))/Hf1 (Ki , μZ (K i ))) may be identified with Z.
Next, let us observe that the kernel Ni of (Vinew )Z  (Hinew )Z is an unramified
representation of GKi (since it arises from the module of prime-to-p-power torsion points
of a semi-abelian variety over ki ). Moreover, the injection of unramified GKi -modules

(Hinew )∨
Z
⊗ μZ (K i ) ⊗ Mi∨ → Ni

splits uniquely over Q, since (by the Riemann Hypothesis for abelian varieties over finite
fields — cf., e.g., [Mumf], p. 206) the Frobenius element of Gki acts on the smaller
module (respectively, quotient by this smaller module) with weight 1 (respectively, 12 ).
Thus, just as in the p-adic case, we may construct a prime-to-p-adic extension class
(ηi )Z from the GKi -action on (Vinew )Z , which, together with (ηi )Zp , yields an extension
class (cf. [FC], Chapter III, Corollary 7.3):

ηi ∈ {(Hinew )∨
Z
}⊗2 ⊗Mi∨ ⊗{H 1 (Ki , μZ (K i ))/Hf1 (Ki , μZ (K i ))} ⊗ Q = {(Hinew )∨
Z
}⊗2 ⊗Mi∨ ⊗ Q

That is to say, ηi may be thought of as a (group-theoretically reconstructible!) bilinear


form:
−, −i : (Hinew )⊗2 → (Mi∨ )Q = Mi∨ ⊗ Q
def
Z

Moreover:

Lemma 2.5. Assume that (Xi )OKi arises as some “(Zi )OKi ” as in Lemma 2.4.
Then:

(i) (Positive Rational Structures) The image of (Hinew )⊗2 under the morphism
(Hinew )⊗2 → (Mi∨ )Q forms a rank one Z-submodule of (Mi∨ )Q . Moreover, for any two
Z
nonzero elements a, b ∈ Hi , a, ai differs from b, bi by a factor in Q>0 (i.e., a positive
rational number). In particular, this image determines a “Q>0 -structure” on (Mi∨ )Q ,
i.e., a Q-rational structure on (Mi∨ )Q , together with a collection of generators of this
Q-rational structure that differ from one another by factors in Q>0 . Finally, this Q>0 -
structure is the same as the Q>0 -structure on Mi∨ determined by the first Chern class
of an ample line bundle on (Xi )Ki in Mi = H 2 (ΔXi , μZ (K i )).

(ii) (Preservation of Degree) The isomorphism



M1 = H 2 (ΔX1 , μZ (K 1 )) → H 2 (ΔX2 , μZ (K 2 )) = M2

induced by αX preserves the elements on both sides determined by the first Chern class
of a line bundle on (Xi )Ki of degree 1.
30 SHINICHI MOCHIZUKI

Proof. Indeed, assertion (i), follows formally from [FC], Chapter III, Corollary 7.3,
and Theorem 10.1, (iii) (by considering “new part” of the Jacobian of (Xi )Ki equipped
with the polarization induced by the theta polarization on the Jacobian).
As for assertion (ii), the elements in question are the unique elements that, on the
one hand, are rational and positive with respect to the structures discussed in assertion

(i), and, on the other hand, generate Mi as a Z-module.

Remark 2.5.1. Note that the conclusion of Lemma 2.5, (ii), is valid not just for
(Xi )Ki , but for any finite étale cover of the original (Xi )Ki , i.e., even if this cover does
not arise as some “(Zi )OKi ” as in Lemma 2.4. Indeed, this follows from the fact that
the crucial “Q>0 -structure” of Lemma 2.5, (i), is preserved by pull-back to such a cover,
which just multiplies the Chern class at issue in Lemma 2.5, (ii), by the degree of the
cover (an element of Q>0 !).

Remark 2.5.2. In the discussion of [Mzk4], §9, it was not necessary to be as careful
as we were in the discussion above in constructing the p-adic class (ηi )Zp (i.e., “μp ” in
the notation of loc. cit.). This is because in loc. cit., we were working over a single
p-adic base-field “K”. In this more restricted context, the extension class (ηi )Zp may be
extracted much more easily from Vi by simply forming the quotient by the submodule
of H 1 (Ki , Ker((Vi )Zp  (Hi )Zp )) generated by the elements which are “crystalline”,
or, more simply, of “geometric origin” (i.e., arise from OK -rational points of the for-
mal group associated to the p-divisible group determined — via “Tate’s theorem” (cf.
Theorem 4 of [Tate]) — by the GKi -module Ker((Vi )Zp  (Hi )Zp )). Unfortunately, the
author omitted a detailed discussion of this aspect of the argument in the discussion of
[Mzk4], §9.

Remark 2.5.3. Relative to Remark 2.5.2, we note nevertheless that even in the
discussion of [Mzk4], §9, it is still necessary to work (at least until one recovers the
“Q>0 -structure” — cf. Remark 2.5.1) with (Xi )Ki such that the dual graph of the
special fiber (Xi )ki is noncontractible. This minor technical point was omitted in the
discussion of [Mzk4], §9.

Next, let us write (Xilog )OKi for the log scheme obtained by equipping (Xi )OKi
with the log structure determined by the monoid of regular functions ∈ O(Xi )OK which
i
are invertible on the open subscheme (Xi )Ki ⊆ (Xi )OKi . Thus, in the terminology of
[Kato2], (Xilog )OKi is log regular. Also, let us write (Xilog )ki for the log scheme obtained
by equipping (Xi )ki with the log structure determined by restricting the log structure
of (Xilog )OKi . Thus, the quotient Π(Xi )Ki  Πadm (Xi )Ki determines a “universal admissible
covering”
(X
ilog )ki → (Xilog )ki
ABSOLUTE ANABELIAN GEOMETRY 31

of (Xilog )ki .
Now let us choose a connected component I
i of the ki -smooth locus (i.e., the com-
plement of the nodes) of (X
ilog )ki . Write Ii ⊆ (Xi )ki for the image of I
i in (Xi )ki .
Thus,
I
i → Ii
is a “tame universal covering” of Ii (i.e., a universal covering of the hyperbolic curve
Ii among those finite étale coverings that are tamely ramified at the “cusps” of this
hyperbolic curve). In the following discussion, we shall also assume, for simplicity,
that Ii is geometrically connected over ki (a condition that may always be achieved by
replacing Ki by a finite unramified extension of Ki ).
Now the Galois group ΠIi of this covering may also be thought of as the quotient

(Xi )Ki of Ii by its inertia group. In particular, since


of the decomposition group in Πadm
ΠIi is formed by taking the quotient by this inertia group, it follows that the surjection
log
(Xi )K  Gki induces a natural surjection
Πadm
i

ΠIi  Gki

whose kernel is the geometric (tame) fundamental group π1tame ((Ii )ki ) of Ii .

Finally, we observe that it makes sense to speak of I


1 and I
2 as corresponding
via αX . Indeed, by Lemma 2.3, αX induces an isomorphism between the pro-graphs
determined by the (X
ilog )ki . Thus, the I
i may be said to correspond via αX when the
vertices that they determine in these pro-graphs correspond. Moreover, when the I
i
correspond via αX , it follows (by considering the stabilizer of the vertex determined
by I
i ) that αX induces a bijection between the respective decomposition groups Di in

(Xi )Ki of Ii , as well as between the respective inertia subgroups of these decomposi-
Πadm
tion groups Di (which may be characterized group-theoretically as the centers of the

subgroups Di Ker(Πadm (Xi )Ki  Gki ), since π1
tame
((Ii )ki ) is center-free — cf. Lemma
1.3.10). Thus, in summary, αX induces a commutative diagram:

ΠI1 → ΠI2
⏐ ⏐
⏐ ⏐


Gk1 → Gk2

We are now ready (cf. [Mzk4], §7) to apply the main result of [Tama1]. This result
states that commutative diagrams as above are in natural bijective correspondence with
commutative diagrams ∼
I
1 → I
2
⏐ ⏐
⏐ ⏐


I1 → I2
32 SHINICHI MOCHIZUKI

lying over commutative diagrams



k1 → k2
⏐ ⏐
⏐ ⏐


k1 → k2

(cf. Theorem 1.3.11). In particular, these commutative diagrams induce an isomorphism

 ∼ ∼ ∼Z)
(Z =) Hc2 ((I1 )k 1 , μZ (k 1 )) → Hc2 ((I2 )k 2 , μZ (k 2 )) (=

(where “Hc2 ” denotes étale cohomology with compact supports — cf. [Milne], Chapter
III, Proposition 1.29; Remark 1.30) which maps the element “1” (i.e., the element
determined by the first Chern class of a line bundle of degree 1) on the left to the

element “1” on the right. (Indeed, this follows from the fact that the morphism I1 → I2
appearing in the above commutative diagram is an isomorphism, hence of degree 1.)

Note that the isomorphism μZ (k 1 ) → μZ (k 2 ) that we use here is that obtained from
the commutative diagram above, i.e., that provided by Theorem 1.3.11.

Lemma 2.6. (Compatibility of Isomorphisms Between Roots of Unity)


Assume that (Xi )OKi arises as some “(Zi )OKi ” as in Lemma 2.4. Then the following
diagram ∼
μZ (k 1 ) → μZ (k 2 )
⏐ ⏐
⏐ ⏐


μZ (K 1 ) → μZ (K 2 )
— in which the vertical morphisms are the natural ones (obtained by considering Te-
ichmüller representatives); the upper horizontal morphism is the morphism determined
by Theorem 1.3.11; and the lower horizontal morphism is the morphism determined by
Proposition 1.2.1, (vi) — commutes.

Proof. Indeed, the diagram in the statement of Lemma 2.6 induces a diagram:

Hc2 ((I1 )k 1 , μZ (k 1 )) → Hc2 ((I2 )k 2 , μZ (k 2 ))
⏐ ⏐
⏐ ⏐
. . . (∗1 )

Hc2 ((I1 )k 1 , μZ (K 1 )) → Hc2 ((I2 )k2 , μZ (K 2 ))
Moreover, we have a diagram

Hc2 ((I1 )k1 , μZ (K 1 )) → Hc2 ((I2 )k2 , μZ (K 2 ))
⏐ ⏐
⏐ ⏐
. . . (∗2 )

H 2 ((X1 )K 1 , μZ (K 1 )) → H 2 ((X2 )K 2 , μZ (K 2 ))
ABSOLUTE ANABELIAN GEOMETRY 33

where the horizontal morphisms are induced by αK (cf. Proposition 1.2.1, (vi)), and
the vertical morphisms are induced “group-theoretically” as follows: First, observe that
[since we continue to operate under the assumption (∗CSSN )]

H 2 ((Xi )K i , μZ (K i )) ∼
= H 2 (ΔXi , μZ (K i )) ∼ Xi , μZ (K i ))
= H 2 (Δadm

while

Hc2 ((Ii )ki , μZ (K i )) ∼


= H 2 ((I i )ki , μZ (K i )) ∼
= H 2 (π1 ((I i )k i ), μZ (K i ))

(where we write I i for the unique nonsingular compactification of Ii ). Moreover, since


we continue to operate under the assumption (∗CSSN ), it follows (cf. the discussion
of “Second Cohomology Groups” in the Appendix) that the natural “push-forward”
morphism in étale cohomology

(H 2 (π1 ((I i )ki ), μZ (K i )) ∼


=) Hc2 ((Ii )ki , μZ (K i ))
→ H 2 ((Xi )ki ⊗ki k i , μZ (K i )) (∼
= H 2 (Δet
Xi , μZ (K i )))

may be reconstructed group-theoretically (by using the various natural homomorphisms


π1 ((I i )ki ) → Δet
Xi [well-defined up to composite with an inner automorphism]). Thus,
the desired vertical morphisms of diagram (∗2 ) may be obtained by composing these
“push-forward” morphisms with the morphisms

Xi , μZ (K i )) → H (ΔXi , μZ (K i ))


H 2 (Δet 2 adm

induced by the surjections Δadm Xi  Δet


Xi [which are group-theoretic by Lemma 2.2,
(ii)]. In particular, we thus see that the vertical morphisms of diagram (∗2 ) are group-
theoretic, i.e., (in other words) diagram (∗2 ) commutes.
Now let us compose the above two diagrams (∗1 ), (∗2 ) to form a single diagram:

Hc2 ((I1 )k 1 , μZ (k 1 )) → Hc2 ((I2 )k 2 , μZ (k 2 ))
⏐ ⏐
⏐ ⏐
. . . (∗3 )

H 2 ((X1 )K 1 , μZ (K 1 )) → H 2 ((X2 )K 2 , μZ (K 2 ))

Note that this diagram (∗3 ) commutes, since, by Lemma 2.5, (ii) [applied to the lower
horizontal morphism of (∗3 )]; the discussion immediately preceding the present Lemma
2.6 [applied to the upper horizontal morphism of (∗3 )]; and the discussion of “Second
Cohomology Groups” in the Appendix [applied to the vertical morphisms of (∗3 )], all
of the morphisms of this diagram are compatible with the elements “1” determined
by the first Chern class of a line bundle of degree 1. But this implies that diagram
(∗1 ) commutes (since diagram (∗2 ) has already been shown to be commutative, and
all the arrows in both of these diagrams (∗1 ), (∗2 ) are isomorphisms between rank one
34 SHINICHI MOCHIZUKI

  -modules). On the other hand, since diagram (∗1 ) was obtained by applying
free Z
the functors Hc2 ((Ii )ki , −) (which are manifestly faithful, at least when restricted to
constant coefficient modules which are finitely generated and free over Z   — as in the
present situation) to the diagram appearing in the statement of Lemma 2.6, we thus
conclude that the diagram appearing in the statement of Lemma 2.6 is commutative,
as desired.

The significance of Lemma 2.6 from our point of view is the following: Lemma 2.6
implies that αK induces an isomorphism


αGlog : Glog log
k1 → Gk2
k

which satisfies the hypothesis of Proposition 1.2.4, (iii). Thus, we conclude from Propo-
sition 1.2.4, (iii), that αGlog arises geometrically. In particular, it follows that we may
k
apply [Mzk4], Theorem 7.2, to the commutative diagram

Πadm
(X1 )K1 → Πadm
(X2 )K2
⏐ ⏐
⏐ ⏐


Glog
k1 → Glog
k2

(where we note that Lemma 2.6 also implies — when translated into the terminology
of [Mzk4], §7 — that the “RT-degree” associated to this commutative diagram is 1, as
is necessary for the application of [Mzk4], Theorem 7.2). In particular, we conclude
that the above commutative diagram of fundamental groups arises geometrically from
a commutative diagram:

(X1log )k1 → (X2log )k2
⏐ ⏐
⏐ ⏐


k1log → k2log
Moreover, the isomorphism exhibited in the upper horizontal arrow of this commutative
diagram is easily seen to be functorial with respect to arbitrary finite étale coverings of
the (Xi )Ki (i.e., not just coverings that arise from finite étale coverings of the (Xi )OKi ).
Indeed, this functoriality follows formally from the uniqueness assertion in Proposition
1.2.4, (iii), and the fact that dominant (i.e., not just finite étale) morphisms between
proper hyperbolic curves in characteristic p may be distinguished by considering the
morphisms that they induce between the respective Jacobians, hence, in particular, by
the morphisms that they induce between the l-power torsion points (where l = p) of
the respective Jacobians. [Note also that the morphisms between log structures at the
nodes are uniquely determined by the morphisms between the various normal bundles
on the branches at the nodes — cf. [Mzk3], §3.7, 3.8, 3.10.] Thus, in summary:
ABSOLUTE ANABELIAN GEOMETRY 35

Theorem 2.7. (Group-Theoretic Reconstruction of the Logarithmic Special


Fiber of a p-adic Hyperbolic Curve) Let p be a prime number. For i = 1, 2, let Ki
be a finite extension of Qp , and (Xi )Ki a hyperbolic curve over Ki whose associated
pointed stable curve has stable reduction over OKi . Denote the resulting “stable model”
of (Xi )Ki over OKi by (Xi )OKi . Assume that we have chosen basepoints of the (Xi )Ki
(which thus induce basepoints of the Ki ). Then every isomorphism of profinite groups

Π(X1 )K1 → Π(X2 )K2 induces commutative diagrams:
∼ ∼
Π(X1 )K1 → Π(X2 )K2 Πadm
(X1 )K1 → Πadm
(X2 )K2
⏐ ⏐ ⏐ ⏐
⏐ ⏐ ⏐ ⏐

∼ ∼
GK1 → GK2 Glog
k1 → Glog
k2

Moreover, the latter commutative diagram (of admissible quotients Πadm


(Xi )K of the i

Π(Xi )Ki lying over the tame Galois groups Glog


ki of the Ki ) necessarily arises from unique
commutative diagrams of log schemes
∼ ∼
(X
1log )k1 → (X
1log )k1 (k1log )∼ → (k2log )∼
⏐ ⏐ ⏐ ⏐
⏐ ⏐ ⏐ ⏐

∼ ∼
(X1log )k1 → (X2log )k2 k1log → k2log
where the commutative diagram on the left lies over the commutative diagram on the
right (which is as in Proposition 1.2.4, (iii)). Here, we equip Spec(OKi ) (respectively,
Spec(ki ); (Xi )OKi ; (Xi )ki ) with the log structure determined by the closed point (re-
spectively, determined by restricting the log structure on Spec(OKi ); determined by the
monoid of functions invertible on the open subscheme (Xi )Ki ; determined by restricting
log
the log structure on (Xi )OKi ) and denote the resulting log scheme by OK i
(respectively,
log log log
ki ; (Xi )OKi ; (Xi )ki ); the vertical morphisms in the above commutative diagrams
of log schemes are the universal coverings induced by the various basepoints chosen.

Proof. First, note that the additional assumptions that were used in the course of
the above discussion — e.g., “(∗CSSN )”; the assumption that (Xi )OKi arise as some
“(Zi )OKi ” as in Lemma 2.4 — were applied only to show that the hypotheses of Propo-
sition 1.2.4, (iii) (and [Mzk4], Theorem 7.2) are satisfied. Moreover, we observe that
although [Mzk4], Theorem 7.2, is only stated in the proper singular case, it extends im-
mediately to the affine singular, as well as affine nonsingular, cases. Thus, (cf. Remark
2.5.1; the discussion surrounding “(∗CSSN )”) one concludes that — except when (Xi )Ki
is proper, with good reduction — one may reconstruct the logarithmic special fiber in a
functorial fashion (i.e., with respect to finite étale coverings of the (Xi )Ki ), as desired.
In the case that (Xi )Ki is proper, but has good reduction over OKi , we may still
reconstruct its logarithmic special fiber (despite the fact that [Tama1], Theorem 4.3,
36 SHINICHI MOCHIZUKI

is only stated in [Tama1] for affine hyperbolic curves!) by arguing as follows: First
of all, we observe that in the case of good reduction, the log structure of the special
fiber of the curve is obtained by simply pulling back the log structure of kilog . Thus,
it suffices to construct the (non-logarithmic, scheme-theoretic) special fiber. Next, we
observe that (after possibly enlarging Ki ) there exist — cf., e.g., [Mzk4], the first two
paragraphs of the proof of Theorem 9.2 — corresponding finite Galois étale coverings
(Zi )Ki → (Xi )Ki (for i = 1, 2), where (Zi )Ki is a hyperbolic curve over Ki with bad
stable reduction (Zi )OKi over OKi . Thus, by applying Theorem 2.7 to (Zi )Ki allows
us to reconstruct the logarithmic special fiber (Zilog )ki , together with the action of the
def
Galois group Gi = Gal((Zi )Ki /(Xi )Ki ). Note that irreducible components of (Zi )ki
that dominate (Xi )ki may be distinguished (group-theoretically!) by the fact that their
geometric fundamental groups map surjectively onto open subgroups of the geometric
fundamental group of (Xi )ki . Let us choose corresponding (closed, proper) irreducible
components
Ci ⊆ (Zi )ki

that dominate (hence surject onto) (Xi )ki . Denote the decomposition (respectively, in-
ertia) group associated to Ci by Di ⊆ Gi (respectively, Ii ⊆ Di ⊆ Gi ). Thus, Di /Ii
acts faithfully on Ci , and the order |Ii | of Ii is a power of p, equal to the degree of
inseparability of the function field of Ci over the function field of (Xi )ki . Then we may
reconstruct (Xi )ki as a finite flat quotient of Ci by considering the subsheaf

|I |
(OCii )Di ⊆ OCi

|I |
(i.e., the Di -invariants of the subalgebra OCii ⊆ OCi , where we use that |Ii | is a power
of p). By applying the functoriality with respect to finite étale coverings of (Xi )Ki
observed in the discussion immediately preceding the statement of Theorem 2.7, we
conclude that this construction of (Xi )ki is independent of the choice of (Zi )Ki , Ci , and
itself functorial with respect to finite étale coverings of (Xi )Ki .

This completes our reconstruction of the logarithmic special fibers of the (Xi )Ki , in
a fashion that is functorial with respect to finite étale coverings of the (Xi )Ki . Thus,
we conclude, in particular, (from this functoriality, applied to covering transformations;
the slimness of Lemma 2.2, (i)) that the morphism induced on admissible fundamental
groups by the isomorphism constructed between logarithmic special fibers coincides with
the original given morphism between admissible fundamental groups. This completes
the proof of Theorem 2.7.

Remark 2.7.1. Given data as in Theorem 2.7, one may consider the outer Galois
representation
GKi → Out(ΔXi )
ABSOLUTE ANABELIAN GEOMETRY 37

which is known to be injective if ri > 0 (cf. Theorem 1.3.6). Thus, at least in the case
ri > 0, it is natural to ask:

What is the commensurator of Im(GKi ) in Out(ΔXi )?

Although Theorem 2.7 does not give a complete explicit answer to this question, it
tells us that, at any rate, elements of this commensurator (which define isomorphisms
of the sort that are treated in Theorem 2.7) preserve the logarithmic special fiber. In
particular, (although one does not know whether or not elements of this commensurator
induce “self-isogenies” of Ki , i.e., are “geometric”) one obtains that elements of this
commensurator do induce “self-isogenies” of kilog . [Here, by a “self-isogeny of an object”,
we mean an isomorphism between two finite étale coverings of the object.] Moreover,
since it follows from Theorem A of [Mzk6] (cf. Theorem 1.3.4, Remark 1.3.6.2) that
the centralizer of Im(GKi ) in Out(ΔXi ) consists precisely of those (finitely many) au-
tomorphisms that arise geometrically (i.e., from automorphisms of (Xi )Ki ), it follows
that a “self-isogeny” of GKi induced by an element of this commensurator corresponds
to (up to finitely many well-understood possibilities) an essentially unique element of
this commensurator. This motivates the point of view that:

The “self-isogenies” of GKi defined by elements of this commensurator —


which we shall refer to as quasi-conformal self-isogenies of GKi — are
natural objects to study in their own right.

The reason for the choice of the terminology “quasi-conformal” is that those self-
isogenies that are “of geometric origin” — i.e., “conformal” — are (by the main theo-
rem of [Mzk5]) precisely those which preserve the higher ramification filtration, which
is closely related to the “canonical p-adic metric” on the local field in question. Thus,
quasi-conformal self-isogenies do not preserve the “metric (or conformal) structure” but
do preserve the “logarithmic special fiber” which one may think of as a sort of p-adic
analogue of the “topological type” of the objects in question.

Remark 2.7.2. Note that isomorphisms



k1log → k2log

(such as those arising from “quasi-conformal isomorphisms” GK1 → GK2 as in Theorem
2.7) need not be “geometric” from the point of view of characteristic zero (i.e., induced

by an isomorphism of fields K1 → K2 ). For instance, such an isomorphism might take
the section of the log structure corresponding to p to some multiple of this section by
a root of unity (a situation which could never occur if the isomorphism arose from an

isomorphism K1 → K2 ). Whether or not, however, this sort of phenomenon actually
38 SHINICHI MOCHIZUKI

takes place in the case of “quasi-conformal isomorphisms” as in Theorem 2.7 is not clear
to the author at the time of writing.

Remark 2.7.3. The theory of the present § prompts the question:


Do isomorphisms Π(X1 )K1 → Π(X2 )K2 as in Theorem 2.7 only preserve the
logarithmic special fiber or do they preserve other information as well con-
cerning the liftings (Xi )Ki of the respective logarithmic special fibers?

Although the author is unable to give a complete answer to this question at the time of
writing, it does appear that when the lifting in question is in some sense “canonical”,
then this canonicality is preserved by isomorphisms as in Theorem 2.7. In a future paper,
we hope to discuss this sort of phenomenon — which may be observed, for instance, in
the following cases:

(1) Serre-Tate canonical liftings;

(2) “arithmetic hyperbolic curves”, i.e., hyperbolic curves isogenous to


a Shimura curve [that is to say, curves which admit a finite étale
covering which is isomorphic to a finite étale covering of some Shimura
curve];

(3) canonical liftings in the sense of “p-adic Teichmüller theory” (cf.


[Mzk1], [Mzk2])

— in more detail. Perhaps this phenomenon should be regarded as a natural extension


of the phenomenon of preservation of the logarithmic fiber in the sense that canonical
liftings are, in some sense, liftings that are “defined over F1 ” — i.e., a hypothetical (but,
of course, fictional!) absolute field of constants sitting inside Zp .

Appendix: Terminology of Graph Theory

The Notion of a Semi-Graph:

We shall refer to as a semi-graph Γ the following collection of data:

(1) a set V — whose elements we refer to as “vertices”;


(2) a set E — whose elements we refer to as “edges” — each of whose elements

e is a set of cardinality 2 satisfying the property “e = e ∈ E =⇒ e e = ∅”;
(3) a collection ζ of maps ζe [one for each edge e] — which we refer to as

the “coincidence maps” — such that ζe : e → V {V} [where we note that
 
V {V} = ∅ since V ∈/ V] is a map from the set e to the set V {V}.
ABSOLUTE ANABELIAN GEOMETRY 39


We shall refer to the subset ζe−1 (V) ⊆ e [i.e., the inverse image of the subset V ⊆ V {V}
of elements = V] as the verticial portion of an edge e; to the restriction of ζe to the
verticial portion of e as the verticial restriction of ζe ; and to the cardinality of the
verticial portion of e as the verticial cardinality of e. A graph Γ is a semi-graph Γ for
which every e ∈ E has verticial cardinality precisely 2. We will say that a graph or semi-
graph is finite if its sets of vertices and edges are finite. We shall refer to an element
b ∈ e as a branch of the edge e.
Let Γ = {V, E, ζ} be a semi-graph. If e ∈ E is an edge of Γ of verticial cardinality 2
whose image via ζe consists of two (not necessarily distinct) elements v1 , v2 of V, then
we shall say that e joins v1 to v2 . If v is any vertex in the image of ζe , then we shall
say that e meets or abuts to v. Thus, an edge of a graph always abuts to at least one
vertex, while an edge of a semi-graph may abut to no vertices at all.
By thinking of vertices as points and edges as line segments that join points to
points or are “open” at one or both ends, we may think of semi-graphs as defining
topological spaces. Thus, it makes sense to speak of a semi-graph as being contractible
(in the sense of algebraic topology). Such a semi-graph will be referred to as a tree.
Finally, a morphism between semi-graphs

Γ = {V, E, ζ} → Γ = {V  , E  , ζ  }

is a collection of maps V → V  ; E → E  ; and for each e ∈ E mapping to e , an injection


e → e — all of which are compatible with the verticial restrictions of the respective
coincidence maps. Thus, here, we allow an edge that abuts to no (respectively, precisely
one) vertex to map to an edge that abuts to any number ≥ 0 (respectively, ≥ 1) of
vertices.

Semi-Graphs of Profinite Groups:

We shall refer to the following data G:

(i) a finite semi-graph Γ;


(ii) for each vertex v of Γ, a profinite group Gv ;
(iii) for each edge e of Γ, a profinite group Ge , together with, for each branch
b ∈ e mapping to a vertex v via ζe , a continuous homomorphism b∗ : Ge → Gv

as a semi-graph of profinite groups. When Γ is a graph, we shall refer to this data G as


a graph of profinite groups.
Suppose that we are given a semi-graph of profinite groups G. Then to G, one
may associate (in a natural, functorial fashion) a profinite group ΠG — namely, the
40 SHINICHI MOCHIZUKI

profinite analogue of the well-known construction of the fundamental group associated


to a (semi-)graph of groups (cf. [Serre1], I, §5.1). One way to describe this profinite
analogue is via Galois categories (cf., e.g., [SGA1] for an exposition of the theory of
Galois categories) as follows: If H is an arbitrary profinite group, then let us write

B(H)

for the category of finite sets with continuous H-action. Then H may be recovered as the
“fundamental group” of the Galois category B(H). Also, we note that any continuous
homomorphism of profinite groups H → H  determines a natural functor B(H  ) → B(H)
[in the reverse direction]. In particular, the homomorphisms b∗ determine functors:

def def
b∗ : Cv = B(Gv ) → Ce = B(Ge )

At any rate, to define ΠG , it suffices to define a Galois category CG [which we would


like to think of as “B(ΠG )”]. We define CG to be the category of collections of data:

{Sv ; αe }

Here, “v” (respectively, “e”) ranges over the vertices (respectively, edges of verticial
cardinality 2) of Γ; Sv is an object of Cv ; and αe : b∗1 (Sv1 ) ∼
= b∗2 (Sv2 ) (where e = {b1 , b2 };
def
vi = ζe (bi ), for i = 1, 2) is an isomorphism in Ce . Morphisms between such collections
of data are defined in the evident way. One then verifies easily that this category CG is
indeed a Galois category, as desired.

Pointed Stable Curves:

Let k be an algebraically closed field. Let g, r ≥ 0 be integers such that 2g−2+r > 0.
Let (X → Spec(k), D ⊆ X) be an r-pointed stable curve of genus g (where D ⊆ X is
the divisor of marked points) over k, and set:

def
X = X\D

Also, let us write X  ⊆ X for the complement of the nodes in X.


Next, we define the dual graph
ΓX

of X. The set of vertices of ΓX is the set of irreducible components of X. To avoid


confusion, we shall write Xv for the irreducible component corresponding to a vertex
v. The set of edges of ΓX is the set of nodes of X. To avoid confusion, we shall write
νe for the node corresponding to an edge e. The node νe has two branches νe [1] and
νe [2], each of which lies on some well-defined irreducible component of X. We take
ABSOLUTE ANABELIAN GEOMETRY 41

def
e = {νe [1], νe [2]} [so e is of verticial cardinality 2], and let ζe be the map that sends
νe [i] to the irreducible component on which the branch νe [i] lies.
def 
Let us write Xv = Xv X  ⊆ X, and (for i = 1, 2) Xν e [i] for the scheme-theoretic
intersection with X  of the completion of the branch νe [i] at the node νe . Thus, Xν e [i]
is noncanonically isomorphic to Spec(k[[t]][t−1 ]) (where t is an indeterminate).
In the following discussion, we would like to fix (k-linear) isomorphisms:

Xν e [1] ∼
= Xν e [2]

via which we shall identify Xν e [1] with Xν e [2] and denote the resulting object by Xe . In
def
particular, we have natural morphisms Xe → Xv 1 , Xe → Xv 2 (where vi = ζe (νe [i]),
for i = 1, 2). One verifies immediately that the induced morphism on tame algebraic
fundamental groups [i.e., corresponding to coverings which are tamely ramified at the
nodes and marked points] is independent of the choice of isomorphism. Thus, the dual
graph ΓX , together with the result of applying “π1tame (−)” [more precisely: the tame
algebraic fundamental group functor, together with choices of basepoints for the various
Xv , Xe , and choices of paths to relate these basepoints via the natural morphisms
Xe → Xv i discussed above] to the data {Xv ; Xe ; Xe → Xv } determines a graph of
profinite groups GX associated to the stable curve X.
When considering the case of a curve with marked points (i.e., r > 0), it is useful
to consider the following slightly modified “data with compact structure”: Let us denote
by ΓcX the semi-graph obtained from ΓX by appending to ΓX , for each marked point
x ∈ X, the following:

def
an edge ex = {x}, where x ∈ ex is sent by ζex to the vertex vx corresponding
to the irreducible component of X that contains x.

We shall refer to the new edges “ex ” that were added to ΓX to form ΓcX as the marked
edges of ΓcX and to ΓcX itself as the dual graph with compact structure associated to X.
If, moreover, we associate to ex the scheme Xx (which is noncanonically isomorphic
to Spec(k[[t]][t−1 ])) obtained by removing x from the completion of X at x, and apply
“π1tame (−)” to the natural morphism Xx → Xv x , then we obtain a natural semi-graph
of profinite groups GX c
with underlying semi-graph ΓcX . Moreover, one checks easily that
when k is of characteristic 0, the profinite group associated (as described above) to GX
or GXc
is isomorphic to “Π g,r ”, i.e., the profinite completion of the fundamental group of
a Riemann surface of genus g with r points removed.
In fact, if we take k = C, and we think of the Xv as Riemann surfaces and of the
Xe ,
Xx as “copies of the circle S1”, then we see that this construction corresponds quite
geometrically to gluing Riemann surfaces with boundary along copies of the circle.
42 SHINICHI MOCHIZUKI

Finally, let us observe the following (valid for k of arbitrary characteristic): The
profinite group associated to GX or GX c
may be identified with the admissible funda-
mental group
Πadm
X

of (X, D) (cf. [Mzk4], §2, 8; [Mzk3], §3). Moreover, if we define semi-graphs of groups
c,et
GXet
, GX by taking the profinite group at a vertex v to be the tame fundamental group
of Xv [i.e., instead of Xv ] and the profinite group at all of the edges to be trivial, then
the resulting profinite group may be identified with the fundamental group

Πet
X

associated to the Galois category of finite étale coverings of X which are tamely ramified
at the cusps. In particular, we have a natural surjection ΠadmX  Πet
X.

Second Cohomology Groups:

We continue with the notation of the above discussion. Let l be a prime number dis-
tinct from the characteristic of k. In the following discussion, we shall use the notation
“H j (−)” (respectively, “Hcj (−)”) to denote the j-th étale cohomology module (respec-
tively, j-th étale cohomology module with compact supports — cf. [Milne], Chapter III,
Proposition 1.29; Remark 1.30) with coefficients in Zl .
If v is a vertex of ΓX , write X v for the normalization of the corresponding irre-
ducible component of X, so Xv = X  ×X X v . Then we have a natural “push-forward”
isomorphism

Hc2 (Xv ) → H 2 (X v )
between free Zl -modules of rank 1. Moreover, the natural morphisms X v → X (as v
ranges over the vertices of ΓX ) determine natural restriction morphisms whose direct
sum is easily verified to be an isomorphism:


H 2 (X) → H 2 (X v )
v

On the other hand, the composite of the natural “push-forward” morphism Hc2 (Xv ) →
H 2 (X) with a restriction morphism H 2 (X) → H 2 (X w ), where w = v, is easily verified
to be zero. [Indeed, this follows by considering the Chern class of the line bundle on
X associated to a closed point of Xv : This Chern class is the image of a generator of
Hc2 (Xv ) and, moreover, vanishes upon restriction to X w since the restriction to X w of
this line bundle is clearly trivial.] In particular, we conclude that the direct sum of the
natural “push-forward” morphisms Hc2 (Xv ) → H 2 (X) yields an isomorphism
 ∼
Hc2 (Xv ) → H 2 (X)
v
ABSOLUTE ANABELIAN GEOMETRY 43


which, if we identify Hc2 (Xv ), H 2 (X v ) via the isomorphism Hc2 (Xv ) → H 2 (X v ), is
∼ 
inverse to the isomorphism H 2 (X) → v H (X v ).
2

Put another way, if we identify the cohomology modules Hc2 (Xv ), H 2 (X v ) via

the isomorphism Hc2 (Xv ) → H 2 (X v ), then the natural “push-forward” morphism
Hc2 (Xv ) → H 2 (X) may be recovered as the composite
 ∼
H 2 (X v ) → H 2 (X v ) → H 2 (X)
v

of the natural inclusion of a direct summand with the inverse of the isomorphism
∼ 
H 2 (X) → v H (X v ) determined by the restriction morphisms.
2

Finally, let us observe that if X is sturdy (cf. [Mzk4], Definition 1.1) — i.e., every
X v has genus ≥ 2 — then the natural morphisms from profinite group cohomology to
étale cohomology give rise to a commutative diagram

H 2 (π1tame (X) = π1 (X), Zl ) −→ H 2 (X)


⏐ ⏐
⏐ ⏐

H 2 (π1tame (X v ) = π1 (X v ), Zl ) −→ H 2 (X v )

in which the horizontal morphisms are isomorphisms [since arbitrary étale cohomology
classes with finite l-power torsion coefficients vanish upon restriction to some finite étale
covering of X or X v ].

References

[FC] G. Faltings and C.-L. Chai, Degenerations of Abelian Varieties, Springer-Verlag


(1990).
[FJ] M. Fried and M. Jarden, Field Arithmetic, Springer-Verlag (1986).
[HR] W. Herfort and L. Ribes, Torsion elements and centralizers in free products of
profinite groups, J. Reine Angew. Math. 358 (1985), pp. 155-161.
[Kato1] K. Kato, Logarithmic Structures of Fontaine-Illusie, Proceedings of the First JAMI
Conference, Johns Hopkins Univ. Press (1990), pp. 191-224.
[Kato2] K. Kato, Toric Singularities, Amer. J. Math. 116 (1994), pp. 1073-1099.
[Knud] F. F. Knudsen, The Projectivity of the Moduli Space of Stable Curves, II, Math.
Scand. 52 (1983), 161-199.
[Mtmo] M. Matsumoto, Galois representations on profinite braid groups on curves, J. Reine
Angew. Math. 474 (1996), pp. 169-219.
44 SHINICHI MOCHIZUKI

[Milne] J. S. Milne, Étale Cohomology, Princeton Mathematical Series 33, Princeton Uni-
versity Press (1980).
[Mzk1] S. Mochizuki, A Theory of Ordinary p-adic Curves, Publ. of RIMS 32 (1996), pp.
957-1151.
[Mzk2] S. Mochizuki, Foundations of p-adic Teichmüller Theory, AMS/IP Studies in
Advanced Mathematics 11, American Mathematical Society/International Press
(1999).
[Mzk3] S. Mochizuki, The Geometry of the Compactification of the Hurwitz Scheme, Publ.
of RIMS 31 (1995), pp. 355-441.
[Mzk4] S. Mochizuki, The Profinite Grothendieck Conjecture for Closed Hyperbolic Curves
over Number Fields, J. Math. Sci. Univ. Tokyo 3 (1996), pp. 571-627.
[Mzk5] S. Mochizuki, A Version of the Grothendieck Conjecture for p-adic Local Fields,
The International Journal of Math. 8 (1997), pp. 499-506.
[Mzk6] S. Mochizuki, The Local Pro-p Anabelian Geometry of Curves, Invent. Math. 138
(1999), pp. 319-423.
[Mumf] D. Mumford, Abelian Varieties, Oxford Univ. Press (1974).
[Naka1] H. Nakamura, Galois rigidity of pure sphere braid groups and profinite calculus, J.
Math. Sci. Univ. Tokyo 1 (1994), pp. 71-136.
[Naka2] H. Nakamura, Galois rigidity of profinite fundamental groups [translation of Sugaku
47 (1995), pp. 1-17], Sugaku Expositions 10 (1997), pp. 195-215.
[NTM] H. Nakamura, A. Tamagawa, and S. Mochizuki, The Grothendieck Conjecture on
the Fundamental Groups of Algebraic Curves, Sugaku Expositions 14 (2001), pp.
31-53.
[NSW] J. Neukirch, A. Schmidt, K. Wingberg, Cohomology of number fields, Grundlehren
der Mathematischen Wissenschaften 323, Springer-Verlag (2000).
[Serre1] J.-P. Serre, Trees, Springer-Verlag (1980).
[Serre2] J.-P. Serre, Local Class Field Theory in Algebraic Number Theory, ed. J.W.S. Cas-
sels and A. Fröhlich, Academic Press (1967).
[SGA1] Revêtement étales et groupe fondamental, Séminaire de Géometrie Algébrique du
Bois Marie 1960-1961 (SGA1), dirigé par A. Grothendieck, augmenté de deux ex-
posés de M. Raynaud, Lecture Notes in Mathematics 224, Springer-Verlag (1971).
[Tama1] A. Tamagawa, The Grothendieck Conjecture for Affine Curves, Compositio Math.
109 (1997), pp. 135-194.
[Tama2] A. Tamagawa, On the Fundamental Groups of Curves over Algebraically Closed
Fields of Characteristic > 0, International Mathematics Research Notices 16 (1999),
pp. 853-873.
ABSOLUTE ANABELIAN GEOMETRY 45

[Tate] J. Tate, p-Divisible Groups, Proceedings of a Conference on Local Fields, Drieber-


gen, Springer Verlag (1967), pp. 158-183.

  


  

   

  


  
E-mail address: [email protected]

You might also like