Gal Ios Modular Forms

Galois Theory and Modular Forms
Developments in Mathematics
VOLUME 11
Series Editor:
Krishnaswami Alladi, University ofFlorida, U.S.A.
Aims and Scope
Developments in Mathematics is a book series publishing
(i) Proceedings of Conferences dealing with the latest research advances,
(ii) Research Monographs, and
(iii) Contributed Volumes focussing on certain areas of special interest
Editors of conference proceedings are urged to include a few survey papers for wider
appeal. Research monographs, which could be used as texts or references for graduate level courses, would also be
suitable for the series. Contributed volumes are those where various authors either write papers or chapters in an
organized volume devoted to a topic of special/current interest or importance. A contributed volume could deal with a
classical topic, which is once again in the limelight owing to new developments.
Edited by
Ki-ichiro Hashimoto
Waseda University
Tokyo, Japan
Katsuya Miyake
Tokyo Metropolitan University
Tokyo, Japan
Hiroaki Nakamura
Okayama University
Okayama, Japan
Kluwer Academic PubUshers

Boston! Dordrechtl London
Distributors for North, Central and South America:
K1uwer Academic Publishers
10 I Philip Drive
Assinippi Park
Norwell, Massachusetts 02061 USA
Telephone (781) 871-6600
Fax (781) 871-6528
E-Mail <[email protected]>
Distributors for all other countries:

K1uwer Academic Publishers Group
Post Office Box 322
3300 AH Dordrecht, THE NETHERLANDS
Telephone 31 786576000
Fax 31 786576474
E-Mail <[email protected]>
tt Electronic Services <http://www.wkap.nl>
Library of Congress Cataloging-in-Publication
Hashimoto, Ki-ichirol Miyake, Katsuya! Nakamura, Hiroaki

ISBN-13:978-1-4613-7960-7 e- ISBN-13:978-1-4613-0249-0
DO I: 10.1007/978-1-4613-0249-0
Copyright 2004 by K1uwer Academic Publishers

Softcover reprint of the hardcover 1st edition 2004
All rights reserved. No part ofthis pUblication may be reproduced, stored in a retrieval system
or transmitted in any form or by any means, electronic, mechanical, photo-copying,
microfilming, recording, or otherwise, without the prior written permission of the publisher, with
the exception of any material supplied specifically for the purpose of being entered and executed
on a computer system, for exclusive use by the purchaser ofthe work.
Permissions for books published in the USA: permj 55 j on5@wkap com
Permissions for books published in Europe: [email protected]
Printed on acid-free paper.
Contents
Preface ix
Part I. Arithmetic geometry 1
The arithmetic of Weierstrass points on modular curves Xo(P) 3

Scott AHLGREN
Semistable abelian varieties with small division fields 13

Armand BRUMER and Kenneth KRAMER
Q-curves with rational i-invariants and jacobian surfaces of GL2-type 39

Ki-ichiro HASHIMOTO
Points defined over cyclic quartic extensions on an elliptic curve and

generalized Kummer surfaces 65
Masato KUWATA
The absolute anabelian geometry of hyperbolic curves 77

Shinichi MOCHIZUKI
Part II. Galois groups and Galois extensions 123
Regular Galois realizations of PSL2(P2) over Q(T) 125

Takehito SHIINA
Middle convolution and Galois realizations 143

Michael DETTWEILER
VI GALOIS THEORY AND MODULAR FORMS
On the essential dimension of p-groups 159

Arne LEDET
Explicit constructions of generic polynomials for some elementary groups 173

Yii,ichi RIKTlNA
On dihedral extensions and Frobenius extensions 195

Masafumi IMAOKA and Yasuhiro KISHI
On the non-existence of certain Galois extensions 221

Hyunsuk MOON
Frobenius modules and Galois groups 233

B. Heinrich MATZAT
Part III. Algebraic number theory 269
On quadratic number fields each having an unramified extension

which properly contains the Hilbert class field of its genus field 271
Ken YAMAMURA
Distribution of units of an algebraic number field 287

Yoshiyuki KITAOKA
On capitulation problem for 3-manifolds 305

Masanori MORISHITA
On the Iwasawa j.L-invariant of the cyclotomic Zp-extension of certain

quartic fields 315
Tsuyoshi ITOH
Part IV. Modular forms and arithmetic functions 327
Quasimodular solutions of a differential equation of hypergeometric type 329

Masanobu KANEKO and Masao KOIKE
Special values of the standard zeta functions 337

Hidenori KATSURADA
CONTENTS vii
p-adic properties of values of the modular j-function 357

Ken ONO and Matthew A. PAPANIKOLAS
Thompson series and Ramanujan's identities 367

Masao KOIKE
Generalized Rademacher functions and some congruence properties 375

Hiroaki NAKAMURA
Preface
This volume is an outgrowth of the research project "The Inverse Ga-

lois Problem and its Application to Number Theory" which was carried
out in three academic years from 1999 to 2001 with the support of the
Grant-in-Aid for Scientific Research (B) (1) No.11440013. In September,
2001, an international conference "Galois Theory and Modular Forms"
was held at Tokyo Metropolitan University after some preparatory work-
shops and symposia in previous years. The title of this book came from
that of the conference, and the authors were participants of those meet-
ings. All of the articles here were critically refereed by experts. Some of
these articles give well prepared surveys on branches of research areas,
and many articles aim to bear the latest research results accompanied
with carefully written expository introductions.
When we started our re~earch project, we picked up three areas to
investigate under the key word "Galois groups"; namely, "generic poly-
nomials" to be applied to number theory, "Galois coverings of algebraic
curves" to study new type of representations of absolute Galois groups,
and explicitly described "Shimura varieties" to understand well the Ga-
lois structures of some interesting polynomials including Brumer's sextic
for the alternating group of degree 5. The topics of the articles in this
volume are widely spread as a result. At a first glance, some readers
may think this book somewhat unfocussed. The editors, however, be-
lieve it is of interest to present this collection of articles because they
discuss those topics each of which could trace its source back to a spring
of advanced ideas on Galois theory.
On the Inverse Galois Problem, Hilbert gave a remarkable result on
the symmetric groups over the rational number field at the beginning
of the last century. Then E. Noether proposed the so called "Noether's
Problem" which became one of basic problems of modern algebra. If
we go back to the 19th century, it was Abel who found new Abelian
equations other than cyclotomic ones with moduli and values at division
points of periods of elliptic functions. He was also interested in charac-
terizing Abelian polynomials which were algebraically solvable. Then,
x GALOIS THEORY AND MODULAR FORMS
for example, Kronecker formulated the Kronecker-Weber Theorem, and

proposed 'the fondest dream in his youth' to determine all of the Abelian
extensions of the rational number field, and those of imaginary quadratic
fields by special values of the exponential function, and elliptic, and ellip-
tic modular functions. The theory of complex multiplications of elliptic
curves was certainly a starting point towards the later vast studies of
automorphic forms and of Shimura varieties. The reader will find sev-
eral articles in this book which concern with the beautiful theory and
examples explaining/indicating mysterious phenomena arising in various
arithmetic functions and varieties related to modular forms.
In the 1920's, algebraic number theory succeeded in establishing class
field theory, and, as a result, opened the door to non-Abelian worlds.
These were there behind the work of Hilbert and the abstractions of
E. Noether stated above. Later in the latter half of the previous cen-
tury, when the classification of finite simple groups was realized, we see
a quite new development of the Inverse Galois Problem. For example,
Thompson gave a stimulating plan for finite simple groups and their
automorphism groups. Matzat is certainly one of the most active math-
ematicians who took this course. The editors are very happy to include
his substantial article in the book, which establishes a foundation for
the recent active studies of the differential Galois theory.
Investigation of structures of Galois groups is also an important
stream in the above history of the Inverse Galois Problem. In 1953,
Iwasawa showed that the commutator subgroup of the Galois group of
the maximal solvable extension over the rational number field is a free
pro-finite-solvable group with a countable number of generators. Then
Shafarevich's conjecture claims that the commutator subgroup of the
Galois group of the algebraic closure of the rational number field should
be a free profinite group with a countable number of generators. Re-
cently, our knowledge on structures of profinite fundamental groups has
been vastly progressed under the name of "anabelian geometry". The
reader will find an excellent paper by S. Mochizuki, where a new frontier
has been pioneered for our understanding the arithmetic fundamental
groups of curves over p-adic fields.
Classification of Galois extensions is another important branch of the
Inverse Galois Problem. Here let us mention a problem which is con-
cerned with "generic polynomials" for finite cyclic groups. When the
local class field theory was established by the global class field theory,
E. Noether immediately pointed out that the converse should be the
right way; namely, the global class field theory should be constructed on
the basis of the much simpler local class field theory. As a result of both
theories, we know that every Abelian extension of a local completion
PREFACE xi
field can be obtained by an Abelian extension of the base global field.
This has not, however, been proved yet in any simple and direct man-
ners. One way may be to utilize "generic polynomials" for finite cyclic
groups over some small algebraic number fields. It should be noted that
there exist no generic polynomials for cyclic group of order 8 over the
rational number field.
The editors would like to express their hearty gratitude to all the par-
ticipants and the speakers at our meetings, to the authors of articles here
and the referees behind them, for their intimate cooperation at various
stages of our modest project, all of which have been indispensable for
collecting resulted fruits successfully in the present proceedings volume.
Special thanks should also go to Dr. Yasuhiro Kishi and Mr. Masafumi
Imaoka for their efforts in arranging the articles of the volume in the
publisher'S format.
Katsuya Miyake,
Ki-ichiro Hashimoto,
Hiroaki Nakamura
Part I
Arithmetic geometry
Galois Theory and Modular Forms pp 3-12
K. Hashimoto, K. Miyake, and H. Nakamura, Editors
2003 Kluwer Academic Publishers
THE ARITHMETIC
OF WEIERSTRASS POINTS
ON MODULAR CURVES Xo(p)
Scott AHLGREN*
Department of Mathematics,
University of Illinois,
Urbana, Illinois 61801
ahlgren~math.uiuc.edu
1. Introduction
The purpose of this paper is to describe some recent results regarding
the arithmetic properties of Weierstrass points on modular curves Xo(P)
for primes p. We begin with some generalities; most of these can be
found, for example, in the book of Farkas and Kra [F-K]. Suppose that
X is a compact Riemann surface of genus g ~ 2. If r is a positive
integer, then let '}iT{X) denote the space of holomorphic r-differentials
on X. Each '}iT (X) is a finite-dimensional vector space over C; we denote
its dimension by dr{X). A point Q E X is called an r- Weierstrass point
if there exists a non-zero differential W E '}iT(X) such that
ordQw ~ dT{X).
Suppose that Q E X and that {WI, ... ,Wd,.(X)} is a basis of '}iT(X)

which is adapted to the point Qj in other words we have
o= ordQ"'l < ordQw2 < ... < ordQwdr(X)'
Then we define the r- Weierstmss weight of Q as

d,.(X)
wtT(Q):= L (ordQwj - j + 1) j (1.1)
j=l
The author thanks the National Science Foundation for its support through grants DMS
01-96443 and DMS 01-34577.
[received: June 13, 2002; accepted in revised form: October 16, 2002]
4 S. AHLGREN
this definition does not depend on the particular basis chosen. Two basic
facts are that
wtr(Q) > 0 -:::::> Q is an r-Weierstrass point
and that
L wtr(Q) = dr(X)(g - 1)(2r - 1 + dr(X)). (1.2)
QeX
In this paper we will be exclusively concerned with modular curves;

these are a class of Riemann surfaces which play an important role in
number theory. If N is a positive integer, then we denote by ro(N) the
usual congruence subgroup
Let lHI denote the upper half- plane of complex numbers, and define the
modular curve Xo(N) as the compactification of the quotient Yo(N) :=
ro(N)\lHI. Part of the importance of these curves arises from the fact
that the curve Xo(N) is the moduli space for pairs (E, C), where E is
an elliptic curve and C is a cyclic subgroup of order N.
1-Weierstrass points (or, more simply, Weierstrass points) on modular
curves have been studied by a number of authors. Works of Atkin [AtJ
and Lehner-Newman [L-NJ, for example, determine some conditions un-
der which the cusp at 00 is or is not a Weierstrass point (it remains an
open problem to classify those N for which 00 is a Weierstrass point).
Weierstrass points also play an important role in the work of Lehner-
Newman [L-NJ and Ogg [OlJ which determines the complete list of N
for which Xo(N) is hyperelliptic.
For the curves Xo(P) with p prime, it is known by work of Atkin and
Ogg (see [021) that 00 is not a Weierstrass point. Moreover, Ogg [02J
proved that if p is prime and if Q is a Q-rational Weierstrass point on
Xo(P) , then the reduction Q of Q modulo p is supersingular (Le. the
underlying elliptic curve is supersingular).
In the remainder of this paper (which is purely expository) our goal
will be to describe recent results of the author with K. Ono and M.
Papanikolas [A-O], [A-P] which develop the relationship between Weier-
strass points on Xo(P) and supersingular points in characteristic p. As we
shall see, the result which is obtained for I-Weierstrass points is quite
satisfactory, while the situation for higher Weierstrass points remains
somewhat more nebulous.
Weierstrass points 5
2. Weierstrass points on Xo(p) and supersingular

j- invariants
If p ~ 5 is prime, then the genus of the modular curve Xo(P) is given
by
1
(P -13)/12, if P == 1 (mod 12),
(p - 5)/12, if p == 5 (mod 12),
9p:= (p _ 7)/12, if p == 7 (mod 12),
(p + 1)/12, if P == 11 (mod 12).
We assume throughout that p ~ 23; this ensures that gp ~ 2 (if 9p < 2
then there are no r- Weierstrass points for any r). Since 00 is not a
Weierstrass point (from which it follows that the conjugate cusp at 0 is
not a Weierstrass point) and dl(XO(P = gp, we see from (1.2) that
L wtl(Q) = g: - gpo (2.1)

QEYo(P)
Throughout, we agree that q:= e211'iz, and we denote by j(z) the usual
elliptic modular function on SL2(Z); we have
j(z) = q-l + 744 + 196884q + ....
To investigate the arithmetic of Weierstrass points on Xo(P), we study

the polynomials
Fp(x):= II (x - j(Qwt 1(Q). (2.2)

QEYo(p)
Here j(Q) means jeT), where T E IHI is any point which corresponds to
Q under the standard identification. By (2.1), we see that the degree of
Fp( x) is g: - gpo We define the supersingular polynomial in characteristic
pby
II
(x - j(E))
E/Fp
8upersingular
(here the product runs over supersingular isomorphism classes of elliptic

curves over the algebraic closure of the finite field lFp ). Then Sp(x) E
lFp[x], the degree of Sp(x) is 9p + 1, and Sp(x) splits completely over lFp2
(the paper of Kaneko and Zagier [K-Z] describes this polynomial from a
number of perspectives).
The main result of [A-D) is the following.
6 S. AHLGREN
Theorem 1. lfp is prime, then Fp(x) has p-integral rational coefficients

and satisfies
Fp(x) == 8 p(x)9P(9P-l) (mod p).
We now give a brief description of the important elements of the proof
of Theorem 1. The starting point is the classical fact that the Weierstrass
weight of a point Q E Xo(p) is determined by the order of vanishing of a
certain differential form at Q. Let 82 (f 0 (p)) denote the space of weight
two cusp forms on fo(p), and let {iI, ... , fgp} be any basis for this space.
We then define the Wronskian modular form
iI(z) h(z)
ff(z) f2(z)
and we define Wp(z) as that scalar multiple of Wp(iI, .. . ,fgp)(z) whose

leading coefficient equals 1. In this way, Wp(z) is a cusp form of weight
gp(gp + 1) on fo(p) which is independent of the choice of iI,, f9p;
moreover, in view of the bijection between the spaces 1{l(XO(P)) and
S2(fo(p)), we have the following fundamental fact:
div(Wp(z)(dz)~gp(gP+1)) = L wtl(Q)Q (2.3)

QEYo(p)
Rohrlich [R] gave a precise description of Wp(z) (mod p) as a modular

form on SL2(Z). To describe his result requires the introduction of some
notation. For even integers k 2: 4, let Bk denote the usual Bernoulli
number, and define the Eisenstein series Ek(z) (a modular form of weight
k for SL2(Z)) by
Also, define the '..lsual delta function by

E (z)3
A(Z)= 4 1';8E (z)2 =q-24q2+252q3+ ... ;
6
A(z) is the unique normalized cusp form of weight 12 on SL2(Z), Finally,

if p ~ 23 is prime, then define rp E {O, 6, 8, 14} by rp := p + 1 - 12gp.
With this notation (and the agreement that Eo = 1), Rohrlich obtains
the following.
Theorem 2 (Rohrlich). If p ~ 23 is prime then Wp(z) has p-integral

rational coefficients and satisfies
Wp == Llgp(9p+1)/2E;Ef~(gp-l)/2 (mod p).
In addition to Rohrlich's result, a certain multiplicative construction is

crucial for the proof of Theorem 1. In particular, the following is proved
in [A-O]. We recall that if fez) is a function of the upper half-plane, k
is an integer, and, = (~ ~) E GLt(lR), then
(flk')(Z):= (ad-bc)~(cz+d)-kf (:;:~).

Lemma 2.1 ([A-O, Lemma 3.3]). Suppose that p ~ 23 is prime, and let
Wp(z) be defined by the product
II Wp(z) Igp (gp+1) A,

Aero(p)\ SL2(Z)
normalized to have leading coefficient equal to 1. Then Wp(z) is a cusp

form of weight gp(gp + 1)(P + 1) on SL2(Z). Moreover, Wp(z) has p-
integral rational coefficients and satisfies
Wp(z) == Wp(z)2 (mod p).
Remark 1. This result holds more generally for modular forms which
are eigenforms of the Fricke involution; see Proposition 5.2 of [A-P] for
a precise statement.
Remark 2. Since....!.his is a multiplicative construction, the divisor of
the modular form Wp(z) retains much of the information present in the
divisor of the original form Wp(z) (some care must of course be taken
to recover this information, particularly at the elliptic fixed points of
SL2(Z)). In view of (2.3), the construction is particularly useful for this
reason.
It follows from the von Staudt-Claussen congruences that Ep-l ==
1 (mod p). Using this fact together with Theorem 2 and Lemma 2.1, we
conclude that
Wp(z) == Ep_l(Z)g~-gpLl(z)gp(gp+1) E rp (z)2 gpE14(Z)gp(gp-l) (mod p).
(2.4)
Moreover, the modular forms on either side of (2.4) have the same weight.
8 S. AHLGREN
It is a well-known result of Deligne (see, for example, [K-Z] or [Sw])

that
II (x - j(T)) (mod p). (2.5)
TESL2(Z)\18I
Ep-l(T)=O
By (2.2), (2.3), and Remark 2 above, we see that the values of the j-
function at the zeros of Wp(z) correspond roughly to the roots of the
polynomial Fp(x). However, by (2.4) we see that many of these j-values
should also correspond roughly to the j-zeros of E p - 1 ; by (2.5) these are
just the supersingular j-invariants in characteristic p. Theorem 1 follows
from these considerations; of course it takes quite a bit of work to make
all of this precise.
3. Higher Weierstrass points on Xo (p)

Many of the ingredients (Rohrlich's theorem, for example) which con-
spire to give such a clean result in the case of 1-Weierstrass points rely
in turn on the isomorphism between Hl(XO(p)) and S2(rO(p)). The
absence of such an isomorphism for the spaces Hr(xo(P)) when r ~ 2
causes much complication in these cases. For example, when r ~ 2, it is
not clear whether or not 00 is an r-Weierstrass point. Moreover, the p-
integrality (or non p-integrality) of the analogue of the polynomial Fp(x)
becomes an issue. However, under a suitable assumption (which we shall
discuss in detail below), we obtain in (A-P] an analogue of Theorem 1
for r-Weierstrass points on Xo(p) when 2 ~ r ~ 5.
Suppose that p 2: 23 is prime and that r ~ 2. Then, in analogy with
(2.2), we define the polynomial
F?)(x):= II (x - j(Q))wtr(Q).
QEYo(p)
We note, using (1.2) and the fact that dr(Xo(p)) = (2r - 1)(gp - 1) for
r ~ 2, that FJr) has degree (2r - 1)2(gp - 1)2gp - 2wtr (oo). Also, we
define
SP(X)' if p == 1 (mod 12),

s*(x):= ( Sp(x)jx, if p == 5 (mod 12),
p Sp(x)j(x - 1728), if p == 7 (mod 12),
Sp(x)jx(x - 1728), if p == 11 (mod 12).
VVeierstruss points 9
In every case, S;(x) is actually a polynomial. Further, we define the

polynomial
1, if p == 1 (mod 12),
xr i(2r-l)2(gp-l)(gp-2)1, if p == 5 (mod 12),
!p,r(x):= (x _1728)~(2r-l)2(gp-l)(gp-2), if p == 7 (mod 12),
xr ~ (2r-l)2(gp--l)(gp-2)l
X (x - 1728) ~(2r-l)2(gp-l)(9p-2), if p == 11 (mod 12).
We obtain an analogue of Theorem 1 when 2 ::; r ::; 5 under the
assumption that the space '}-{r(xo(P)) is good at Pi the precise definition
(as well as an explanation of the assumption that r ::; 5) will be given
below.
Theorem 3. Suppose that p is prime and that 2 ::; r ::; 5. Suppose that
'}-{r(xo(P)) is good at p. Then F~r) (x) has mtional p-integml coefficients,
and there exists a polynomial H(x) E lFp[x] such that
FJr)(x) == H(x) . !p,r(x) . S;(x)(2r-l)2(gp-l)(9p-2) (mod p).
Remark. In view of the definitions of S;(x) and !p,r(X), Theorem 3

shows that FJr) (x) (mod p) is in every case divisible by a large power of
the full supersingular polynomial Sp(x). Computations suggest that the
powers of x and x - 1728 present in the definition of !p,r (x) may in fact
be replaced by larger powers of these monomials.
To define what it means to be "good" requires some further notation.
If r ~ 2 then let S2r(rO(p)) be the subspace of S2r(rO(p)) which corre-
sponds to '}-{r(xo(P)) under the standard identification of meromorphic
r-differentials on Xo(P) with meromorphic modular forms of weight 2r
on ro(P)' Let d:= dr(Xo(P)). Then we see that Hr(xo(P)) has a basis of
the form {Wl,W2, ,Wd}, where for each i we have Wi = h(z)(dzr, and
!i(Z) E S2r(rO(p)) has rational Fourier coefficients. In order to deter-
mine such a basis uniquely, we stipulate further that the modular forms
!i(Z) have expansions at 00 of the form
h(t:) = qrl+ O(qrl+l),
J2(z) = {2 + O(qr2+1) ,
10 S. AHLGREN
where r = rl < r2 < ... < r'd, and the coefficient of qTi in fJ is zero if
i =1= j. Then we call 'H,T(XO(p)) good at p if the modular forms h,, fd
have p-integral Fourier coefficients.
The question of whether or not a given space HT(XO(P)) is good at
p is an interesting one. Computations uncover a few "bad" primes, but
suggest that most such spa~es are indeed good. For example, the only
prime p < 800 for which 'H,2(XO(p)) fails to be good at p is p = 373.
Using W. Stein's modular forms package in MAGMA, we have carried
out the following computations.
Proposition 3.1.
(1) 'H,2(XO(p)) is good at p for 23 ~ p < 800 with the exception of

p = 373.
(2) 1{3(XO(p)) is good at p for 23 ::; p < 800 with the exception of
p = 373,643.
(3) 1{4(Xo(p)) is good at p for 23 ~ p < 800.
(4) 1{5(Xo(p)) is good at p for 23 ::; p < 400 with the exception of
p = 379.
The method of proof of Theorem 3 follows the rough outline of the

proof of Theorem 1, although there are additional difficulties which are
not present in the case r = 1. The assumption that 'H,T(XO(p)) is good
at p ensures p-integrality of the relevant Wronskian in this case (and
therefore the p-integrality of the polynomial F?)). We again use a mul-
tiplicative construction as in Lemma 2.1. However, we lack Rohrlich's
Theorem 2 when r ~ 2; it is therefore necessary to develop some ana-
logues of results of Serre regarding connections between the reductions
modulo p of spaces of modular forms on ro(p) and those on SL2(Z).
These results are strong enough to obtain a statement as in Theorem 3
only when r ~ 5 (this assumption ensures that there are no "oldforms"
of weight 2r on ro(p)).
4. Examples
We give some examples of the phenomena described by Theorems 1
and 3. As an example of Theorem 1, we consider the case when p = 31.
Here gp = 2, and
S31(X) = (x + 8)(x + 27)(x + 29).

Moreover, we find that

F31(X) = x6 -155973949688664Oxs - 61597026997169960170773x4
+ 57008654146386200663274528x3
-2414976566344468731192819276342x 2
- 1333724012279865296657027561330208x
-938882052398785241500705894914022689
== (x + 8)2(x + 27)2(x + 29)2 (mod 31).
Next we consider the case of Theorem 3 corresponding to p = 53 and
r = 2. Here 9p = 4 and wt2(00) = O. We have
S53(X) = x(x + 3)(x + 7)(x 2 + 50x + 39) = X SS3(X),
and
FJ~)(x) == SS3(X)S4 . x l8 . HI (x) (mod 53),
where HI (x) is the square of a polynomial of degree 18.
Finally we consider the case of Theorem 3 corresponding to p = 73
and r = 2. Here 9p = 5, wt2( 00) = 1, and
S73(X) = S73(X) = (x + 17)(x + 64)(x 2 + 57x + 8)(x 2 + 68x + 9).
We find that
Fi~){x) == S73{X)108 . H2{X) (mod 73),
where H2{X) is the square of a polynomial of degree 35.
5. Some questions
In closing, we mention a few natural questions related to the results
described above.
(1) Is it possible to find a proof of Theorem 1 (or of Theorem 3) which
relies on the underlying geometry of the curve Xo(p)?
(2) Are there interesting Galois-theoretic properties of the polynomials
Fp(x) and FJT)(X)?
(3) Is it possible to find a description of those p and r for which 00 is
an r- Weierstrass point on Xo(p)?
(4) Is it possible to give a description of those p and r for which
'J-lT(Xo(p fails to be good at p? Are there infinitely many such p
for any given r?
12 S. AHLGREN
(5) Is there a more natural description of what it means to be "good"?

(6) Is it possible to obtain a more precise description of the polyno-
mials H(x) which appear in Theorem 3? Do the roots of these
polynomials contain interesting arithmetic information?
(7) In the same vein, can the polynomial S;(x) in Theorem 3 always
be replaced by the full supersingular polynomial Sp(x)?
Acknowledgments
The author wishes to thank K. Hashimoto, K. Miyake, and H. Naka-
mura, as well as others involved in the organization of the Saga and
TMU conferences on Galois Theory and Modular Forms, for all of their
efforts towards running two successful meetings.
References
[A-O] S. Ahlgren and K. Ono, Weierstrass points on Xo(P) and supersingular j-
invariants, Math. Ann. 325 (2003),355-368.
[A-P] S. Ahlgren and M. Papanikolas, Higher Weierstrass points on Xo(P), Trans.
Amer. Math. Soc. 355 (2003), 1521-1535.
[At] A. O. L. Atkin, Weierstrass points at cusps of Xo(N), Ann. of Math. 85
(1967), 42-45.
[F-K] H. M. Farkas and I. Kra, "Riemann surfaces," Springer-Verlag, New York,
1992.
[K-Z] M. Kaneko and D. Zagier, Supersingular j-invariants, hyperyeometric series,
and Atkin's orthogonal polynomials, in "Computational perspectives on num-
ber theory (Chicago, IL,1995)," Amer. Math. Soc., Providence, RI, 97-126,
1998.
[L-N] J. Lehner and M. Newman, Weierstrass points on ro(N), Ann. of Math. 79
(1964), 360-368.
[01] A. Ogg, Hyperelliptic modular curves, Bull. Soc. Math. France 102 (1974),
449-462.
[02] A. Ogg, On the Weierstrass points of Xo(N), Illinois J. Math. 22 (1978),
31-35.
[R] D. Rohrlich, Weierstrass points and modular forms, Illinois J. Math. 29
(1985), 134-141.
[Sw] H. P. F. Swinnerton-Dyer, On f-adic representations and congruenc.s Jor mod-
ular forms, in "Modular functions of one variable, III," 1-55, Lecture Notes
in Mathematics, 350. Springer-Verlag, Berlin.
SEMISTABLE ABELIAN VARIETIES

WITH SMALL DIVISION FIELDS
Armand BRUMER
Department of Mathematics
Fordham University,
Bronx, NY 10458
brume~fordham.edu
Kenneth KRAMER
Queens College (CUNY)
Flushing, NY 11367
kramerlilforbin.qc.edu
1. Introduction
The conjecture of Shimura-Taniyama-Weil, now proved through the
work of Wiles and disciples, is only part of the Langlands program.
Based on a comparison of the local factors ([And], [Serl]), it also predicts
that the L-series of an abelian surface defined over Q should be the
L-series of a Heeke eigen cusp form of weight 2 on a suitable group
commensurable with Sp4(Z). The only decisive examples are related
to lifts of automorphic representations of proper subgroups of SP4, for
example the beautiful work of Yoshida ([Yos], [BSP]).
We do not know how to construct ''non-trivial'' weight 2 forms for
groups such as the paramodular group or the Hecke subgroup ro(P) in
genus 2. In fact, it seems that not a single example is known which is
not a lift. Not even a conjectural dimension formula for weight 2 cusp
forms has yet been proposed ([Ibu], [Has]).
It seems interesting to consider existence and non-existence on the
arithmetic side. If the special fiber Ap of the Neron model of the abelian
variety A has no unipotent part, we say that A has semistable reduction
This author was supported in part by a grant from the City University of New York PSC-
CUNY Research Award Program.
[received: June 7, 2002 ; accepted in revised form: October, 16, 2002]
14 A. BRUMER AND K. KRAMER
at p. In that case, the exponent fp(A) of p in the conductor of A is

the toroidal dimension of Ap. For example, the modular variety Jo(P)
has conductor pd, where d is its dimension. To ensure that the putative
automorphic form not be a lift, we could study surfaces of prime conduc-
tor. This guarantees that we are not dealing, for instance, with a surface
isogenous to a product of elliptic curves, the Weil restriction of scalars
of an elliptic curve defined over a quadratic number field or a surface
with non-trivial endomorphisms. Serre's conjecture [Ser4] implies that
such a surface should be Q-isogenous to the Jacobian of a curve also
defined over Q. This led us to try to understand curves C of genus 2
whose Jacobian variety J = J(C) has prime conductor p. A minimal
model for C has bad reduction at p, but also may have primes of bad
reduction where J has good reduction. We call such primes, at which
J reduces to the product of 2 elliptic curves, primes of mild reduction.
Because of mild primes, we could not immediately rule out the possibil-
ity that all six Weierstrass points of C be rational, that is Q(J[2]) = Q.
In the present investigation, we find that Gal(Q(J[2])/Q) cannot even
be nilpotent, as we shall explain below.
One expects that Gal(Q(A[])/Q) should be, in general, as large as is
compatible with the endomorphisms of A and the Weil pairing. When
End A = Z, this has been proved for f sufficiently large by Serre [
Ser5], in all odd dimensions as well as in dimensions 2 and 6. Note that
Q(A[t']) always contains the cyclotomic field Q(JLi), whether or not A is
principally polarized. As examples of the results obtained here, we have:
Proposition 2.4. Let A/Q be a semistable abelian variety with good
reduction at a prime t' ~ 19 and set H = Gal(Q(A[t'])/Q(JLi))' Then t'
annihilates the maximal abelian quotient H / H'. So H is an t'-group if
it is nilpotent.
Theorem 1.1. Suppose that A/Q is semistable, with bad reduction only
at p, and that H is nilpotent. If t' = 2 or 3, then fp(A) = dimA, so the
reduction at p is totally toroidal. Moreover p == 1 mod 8 if f = 2 and
p == 1 mod 3 if f = 3.
Consider elliptic curves of prime conductor p with some rational t'-
torsion. For t' = 2, Neumann [Ne] and Setzer [Set] found that such
elliptic curves exist if and only if p = n 2 + 64 or p = 17. For odd
t', Miyawaki [Mi] showed that the well-known examples with (P.,p) =
(3,19), (3,37) or (5,11) are the only ones. Under the stronger assumption
that A has prime conductor, Theorem 1.1 implies:
Corollary 1.2. Let A/Q be an abelian variety of prime conductor p
such that H is nilpotent. If P = 2, then A is a Neumann-Setzer elliptic
curve. If P = 3, then p = 19 or 37 and A is a Miyawaki elliptic curve.
Semistable abelian varieties 15
Let C be the Q-isogeny class of an abelian variety A with semistable

bad reduction at p. We denote by ~ A the group of connected components
of the special fiber of its Neron model at p. We say that A is I-maximal
if ordt I~A(iFp)1 is maximal among varieties in C. A major role will be
played by the collection
(1)
of I-maximal varieties. This notion depends on the choice of p, which
will be clear from the context.
We show that Q(A[IJ) = Q(1L2t) for all A E Vt(C) if C is the isogeny
class of a product of elliptic curves of conductor p with some rational
I-torsion, cf. Proposition 5.3. The following is a converse when I = 2 or
3.
Theorem 1.3. Let I = 2 or 3 and let C be a Q-isogeny class of semistable
abelian varieties with bad reduction only at p. If Q(A[IJ) ~ Q(1L2t) for
all A E Vt(C), then C is the isogeny class of a product of Neumann-Setzer
( = 2) or Miyawaki ( = 3) curves of conductor p.
A brief outline of the paper follows. With the help of class field
theory and the conductor bounds of Fontaine, we obtain in section 2 the
information we require about the group H = Gal(Q(A[I])jQ(lLt)).
Denote by '1f't(A) the Tate module of A and by Vv the decomposition
group in GIQ of a place v over p. In section 3, we recall the definition
of the Vv-submodules of'1f't(A) introduced by Grothendieck [Gro]. Now
suppose A is semistable with bad reduction only at p and that H is
an I-group. Let a denote the dimension of the abelian variety in the
special fiber of A at p. We use the Grothendieck modules together
with f-maximality and a strange lifting lemma to construct a pure Zt-
submodule WA of ll't(A) ofrank 2a. Although WA is canonically only a
Vv-module, we obtain the rather surprising result that it is a GIQ-module
under suitable hypotheses. Then, in section 4, we use the formal group
associated to A at I and Raynaud's theory of IF-vector space schemes to
prove that WA = 0, so that a = 0 and A is totally toroidal at p. The
proof of Theorem 1.1 may be found at the end of this section.
A generalization of Theorem 1.3 is proved in section 5, using the
theorem of Faltings on Tate's conjecture. We then give two examples
for which H is an I-group, namely, Jo(31) with I = 2 and Jo(41) with
I = 5. Since they clearly are not isogenous to a product of elliptic curves,
any attempt to weaken the hypotheses of Theorem 1.3 will require some
care.
Fontaine [Fo] proved that there are no abelian varieties of positive
dimension with everywhere good reduction over Q. We showed, in [BK2] ,
that no semistable abelian variety <;:!xists with bad reduction exactly at

one prime p :S 7 and Schoof [Scho] improved this result by different
methods.
In our next paper, using other techniques, we plan to study the ex-
istence of abelian surfaces AjQ of conductor p in the spirit of [BKl].
When A is principally polarized, G = Gal(Q(A[2])jQ) is a subgroup of
GSp4(lF 2) :::: 8 6 , the symmetric group on 6 letters. As an application
of Theorem 1.1, we will show that G can only be Sn for 3 :S n :S 6,
Sn x Zj2 for n = 3,4, or the wreath product S3 I Zj2, with each of these
possibilities occurring infinitely often under the Schinzel hypothesis [SS].
Our criteria will show that p = 277 is the smallest prime for which
there is a curve of genus 2 whose Jacobian has prime conductor. One
such curve is given by
y2 + y = x 5 _ 2x 3 + 2x2 - x.
We thus propose, as a challenge to our automorphic friends, a verification
that all modular forms of weight 2 for smaller prime level are "lifts", but
that at least one interesting eigenform exists for 277.
This paper concentrates OIl one of the results presented by the first
author at this Conference. He is grateful to K. Hashimoto and K. Miyake
for their kind invitation and for their support both moral and financial.
Other results, more influenced by the work of Hashimoto and Ibukiyama,
will be published later. Thanks to H. Nakamura for helping run the
Conference so pleasantly and seamlessly.
2. Controlling the i-division field

Let S be a set of rational primes not containing the prime f nor the
archimedean prime 00. We shall say that a finite extension LjQ is (f, S)-
controlled if it has the following properties:
(L1) L is Galois over Q and contains J-t;
(L2) LjQ is unramified outside each place over SU {f,oo};
(L3) the ramification degree for each place of Lover S divides f;
(L4) the higher ramification groups Viu ) are trivial for u > 1/(f - 1),
where VA is the decomposition group at a place .x over f in L.
If S = {p} consists of one prime, we say LjQ is (f,p)-controlled. Thanks
to Grothendieck [Gro, 2.5] and Fontaine [Fo, Thm. A], the f-division
field of a semistable abelian variety AjQ with good reduction at f is
(f, S)-controlled when S contains the primes of bad reduction. See also
[BK2, 3].
Since the upper ramification numbering behaves well with respect to
quotients, a subfield of L containing P,l and Galois over Q also is (f, S)-
controlled. Similarly, property (L4) is inherited by the compositum of
finitely many Galois extensions that enjoy it. Properties (L1) and (L2)
certainly are preserved under composition of (f, S)-controlled extensions.
By uniqueness of the tamely ramified extension of degree f over the
unramified closure of Qp, property (L3) is preserved under composition.
It is helpful to highlight the impact of (L4) on the following local
situation, a variant of which was treated in [BK2, Lem. 6].
Lemma 2.1. Let E ;2 F ;2 Ql(f..tl) ;2 Ql be a tower of local fields such
that EIQl is Galois and EIQl(f..tl) is an abelian i-extension. The higher
ramification groups Gal(EIQl)(u) are trivial for all u > 1/(i - 1) if and
only if each of the abelian conductor exponents HEI F) and f(FIQl(f..tl))
is at most 2.
Proof. First consider a general Galois extension oflocal fields ElK and
put H = Gal(EI K). Assume only that F is the fixed field of a normal
subgroup N <I H and let fI = HIN. Write ({'ElK for the Herbrand
function [Ser2, IV, 3] from the lower to upper ramification numbering
of H. Let us verify that Hx = 1 if and only if both N x and fIxl are
trivial, where x' = (('EIF(X). If Hx = 1, then N x = Hx n N = 1 and by
Herbrand's theorem [Ser2, IV, 3, Lem. 5], we also have fIxl = HxNIN =
1. Conversely, if fIx' = 1, then Hx ~ N. It follows that Hx = NXI so if
N x also is trivial, we may conclude that Hx = 1. This observation will
be used below with K = Ql(J.tl)'
Now assume all the given hypotheses and let D = Gal(EIQi). Since
the tame ramification degree of the extension EIQi is f - 1, we have
(('EI<rb(I) = IDII/IDol = 1/(f -1). Furthermore, Dx = Hx for all x ~ 1
because H is the i-Sylow subgroup of D. Then we have the following
chain of equivalences:
D(u) = 1 for all u > 1/(f - 1) <=> D1+ = 1 for all f > 0
<=> H1+ = 1 for all f > 0
<=> N1+f = 1 and fI1+f = 1 for all f > 0,
using our opening observation and the fact that (('EIF(l) = 1 for the last
equivalence.
The conductor exponent of an abelian extension ElK is given by
f(EIK) = I{)EIK(C) + 1, where C is the largest integer such that He'# 1.
When H is an f-group, we have (('EIK(I) = 1. For an abelian f-extension
ElK we therefore find that f(E/K) ~ 2 if and only if H1+ = 1 for all
f > O. Applying this fact tv the abelian extensions ElF and F IQl(f..tl) ,
we may conclude that D(u) = 1 for all u > 1j( - 1) if and only if both
HEj F) ~ 2 and f(FjQi(J.Li)) ~ 2. 0
We use below the fact, proved in [MMJ, that QU.Li) has class number
one exactly when ~ 19.
Lemma 2.2. Buppose M is (, B)-controlled and unramified outside ,
with ~ 19. If M is abelian over F = Q(JLu), then M = F.
Proof. To analyze the exten5ion Mj F, we introduce the usual notation
of class field theory. 'Let A~ be the idele group of F and write FX for
the image of FX on the diagonal. Let Fv be the completion of F at a
place v and put Uv for the connected component of the identity in Fvx .
If v is non-archimedean, set U~m) = {x E Uv I vex -1) ~ m}.
Observe that there is a unique prime A over in F and let [ be a
prime of M over A. By (L4), we have Gal(MdQ)(u) = 1 for all u >
1/( - 1). For brevity, write fA for the abelian conductor exponent of
the local extension Md F>... Since M / F is unramified outside i, there is
a surjection A;/N - Gal(M/F), where N is the norm subgroup
N = (Uib.l x II Uv ) . FX.
v;;6>'
Let Vp be the ilJJage of the global units of F in r = u>..jUif>.l. In our

case Uv ::: ex for archimedean places, so we have the exact sequence
0--+ r /Vp --+ A~/N --+ A;/(CIT Uv ) . FX) --+ 0, (1)

v
in which the last quotient is isomorphic to the ideal class group of F and
is trivial by assumption.
The group u>.jUi1) is generated by the images of global cyclotomic
units when is odd and is trivial when = 2. Since ui1) jUlIA) is an
-group, we find that r /Vp is an i-group and so Gal(Mj F) is an i-group
by exact sequence (1). Then f>.. ::; 2 by Lemma 2.1. The group ui 1) jUi2 )
is generated by the image of a global primitive th root of unity when
is odd or by the image of i when = 2. Thus r jVp and a fortiori
Gal(M/ F) is trivial. 0
Remark. Under the hypothesis that Gal(Mj F) is an abelian i-group,
similar reasoning shows that M = F when is a regular prime.
Lemma 2.3. Buppose that ~ 19 and that L is an (, B)-controlled
extension containing F = Q(JL2i)' Let E be the maximal subfield
Semistable abelian varieiies 19
of L abelian over F. Then Gal( E / F) is annihilated by f. and

dimlFt Gal(E/ F) ::; s, where s is the number of primes over S in F.
Proof. For each prime Pj over pin F, choose a prime s.}3'j over Pj in L.
Since p does not ramify in F /Q, the inertia group Ij = Iq3j is contained
in H = Gal(L/ F). Let N be the subgroup of H generated by I j for
j = 1, ... , s and by the commutator subgroup H'. The previous Lemma
shows that N = H. By (L3), each Ij has order dividing f., so Gal(E/ F)
is annihilated by i and its IFt-dimension is at most s. 0
Remark The bound on dimlFl Gal(E/ F) above can sometimes be sharp-

ened by a class field theoretic analysis similar to that of Lemma 2.2.
Define
rs = II Uv/U~,
VESF
where SF is the set of primes over S in F. Recall that ,A denotes the

prime over i in F and let U be the image in r s of those global units f.
in F such that f. == 1 mod ,A2. Then Gal(Ej F) is a quotient of rsju.
Proposition 2.4. Let AjQ be a semistable abelian variety with good
reduction at a prime i $ 19 and set H = Gal(Q(A[i])jQ(l-'t)). Then i
annihilates the maximal abelian quotient H / H'. If H is nilpotent, then
H is an i-group.
Proof. If i is odd, or if i = 2 and i E Q(A[]), let L = Q(A[]). Other-
wise, let L = Q(i,A[2]), in which case, Gal(LjQ(i)) ~ H. Then L is an
(i, S)-controlled extension and our result follows from Lemma 2.3. Note
that if H is nilpotent, it is the product of all its Sylow subgroups, so it
must be an f.-group. 0
Lemma 2.5. Suppose L is an (f.,p)-controlled extension containing F =

Q(I.t2i) and f. is a regular prime. Put G = Gal(L/Q) and assume H =
Gal(L/ F) is an f.-group. Let u be a generator for the inertia group I~
of a prime s.}3' over p in L and suppose f E G restricts to a generator of
Gal(F/Q). Then the conjugates of u by powers of f generate H, while
u and f generate G. Ifp is unramified in L, then L = F.
Proof. Note that I~ is cyclic of order dividing i by (L3). Let N be
the subgroup of H generated by H', Hi and the conjugates of u by
powers of f. Then N is normal in G and corresponds to an (f.,p)-
controlled extension M/ F, unramified outside f., such that Gal(M/ F)
is an elementary abelian i-group. So N = H by the Remark after
Lemma 2.2. Our claim now follows from Burnside's lemma. 0
Next, we give a description of the maximal (2,p)-controlled 2-

extension of Q. Since we do not need this result, we leave the proof
to the reader as an exercise in class field theory. A general study of
extensions of number fields with wild ramification of bounded depth can
be found in recent work of Hajir and Maire [HM].
Proposition 2.6. Consider K = Q( A) and let n be the 2-part of the
class number of K. If p == 3 mod 4, define M = K(i). If P == 1 mod 4,
let M be the 2-part of the ray class field of conductor 2 over K. Then
Gal(M/K) is cyclic of order 2n and Gal(M/Q) is dihedral. Moreover,
M is (2, p) -controlled and contains every (2, p) -controlled 2-extension of
Q.
Finally, we summarize for later use some elementary facts about pure
submodules unrelated to the rest of this section.
Lemma 2.7. Suppose X and Y are pure submodules of a free ZI.-module
T of finite rank. Write X = (X + fT)lfT for the projection of X to
TifT. Then:
(i) XnY~XnY andX+Y=X+Y;
(ii) X n Y is pure;
(iii) if X = 0, then X = 0;
(iv) if X n Y = 0, then X +Y is pure and a direct sum.
Let T' be a free ZI.-module (;f finite rank and suppose there is a perfect
pairing e : TxT' ~ 1['1. (1-) . Then X.l + y.l ~ (X n y).l, with equality
ifX.l+y.l is pure.
Proof. Property (i) is clear and does not use purity. The natural injec-
tion
T/(X n Y) <--> TIX EB T/Y
implies (ii). By Nakayama's Lemma and the isomorphism X ~ X/(X n
T) = XlfX we have (iii). For (iv), suppose fz = x + y for some z E T,
x E X and y E Y. Then the coset x = -y in TifT is an element
of X n Y = 0. Hence x = y = 0. By purity of X and Y, we may
write x = Xl and Y = Yl for some Xl E X and Yl E Y. Since T is
torsion-free, we then have z = Xl + Yl E X + Y. Hence T/(X + Y) is
torsion-free and (iv) is verified. Given the perfect pairing e, we clearly
have X.l + y.l ~ (XnY).l. Equality holds if Xl. + yl. is pure because
both sides have the same rank. 0
Semistable abelkn varieties 21
3. Decomposition of the Tate module

Suppose A/Q is an abelian variety with good reduction at i and
semistable bad reduction at p. Grothendieck described certain submod-
ules of the Tate module '1['l(A) that are Galois modules for a decomposi-
tion group over p in GQ. A convenient summary of the information we
need also appears in [Ed, 2J. In this section, we use our assumptions
that the i-division field of A is small and that A has good reduction
outside p to create GQ-submodules of'1['l(A) from these Grothendieck
modules. First we establish the relevant notation.
Let Ap denote the special fiber of the Neron model of A at p. Its
connected component ~ admits a decomposition
O--+T--+~--+8--+0
in which T is a torus and B an abelian variety defined over lFp Setting
dim T = t and dimB = a, we have t + a = dimA. Write ~A = Ap/Ag
for the group of connected components. Denote the perfect pairing on
the Tate modules of A and its dual abelian variety A by
(1)
Let Loo = Q(A[iooJ) be the i-division tower of A and set Goo =

Gal(Loo/Q). Since we may be moving among abelian varieties Q-
isogenous to A, it is important to note that Loo only depends on the
Q-isogeny class of A. Clearly GQ acts on '1['l(A) through Goo. Fix an
embedding /, : Loo --+ Qp and let v be the corresponding valuation. Write
Vv ;2 Iv for its decomposition and inertia groups in Goo. For 9 E Goo,
let gv denote the valuation corresponding to /, 0 g-1.
As in [Gro, 2.5], we define Ml = Ml(A, v) to be the submodule of
'1['l(A) fixed by Iv and M2 = M2(A, v) to be the submodule of'1['l(A)
orthogonal to M 1 (A, v) under the eoo -pairing. Clearly 9 E Goo acts
by g(Mj(A,v)) = Mj(A,gv) for j = 1,2. Thus Ml and M2 are V v -
modules.
The Igusa-Grothendieck theorem [Gro, Thm. 2.5] asserts that if A is
semistable at p, then M2 ~ MI. Further, the successive quotients in
the decomposition
(2)
are torsion-free Zl-modules. Both MI and M2 are modules for Vv/Iv ~
GlFp. We may identify M2 ~ '1['l(T) and Md M2 ~ '1['l(8), so the Zt-
ranks of M2 and Md M2 are t and 2a respectively. Using the eoo -
pairing, one can see that (g -1)('1['l(A)) ~ M2 for all 9 E Iv.
Let '1['l(<p) denote the map of Tate modules induced by a Q-isogeny
<p : A --+ A' and set Mj = Mj(A', v) for j = 1,2. Then '1['l(<p)(Mj ) ~
Mj. We can also show that 1r(<p)-l(Mj) ~ Mj by using the existence

of a quasi-inverse isogeny <p' : A' ---T A such that <p' 0 <p (resp. <p 0 <p')
is multiplication by m on A (resp. A'), where m is the exponent of the
kernel of <po Indeed, if 1I'(<p) (x) E Mj, then mx E 1I'(<p')(Mj) ~ M j .
But 1I'(A)/Mj is torsion-free, so x E Mj.
The following result, mildly strengthening [BK2, Lem. 3], describes a
more subtle effect of isogeny. Write Min) and M~n) for the projections
of Ml and M2 to the nth layer 1ri(A)/f111ri(A) ~ A[in]. Write IXli for
the order of the i-primary subgroup of a finite abelian group X. See (1)
for our definition of i-maximal abelian varieties.
Lemma 3.1. Suppose AjQ has good reduction at f and semistable bad
reduction at p. Let <p : A ---T A' be a Q-isogeny whose kernel K, has
exponent fn. Then
I~AI(lFp)li . 1K,/(K,rlMin))1 = I~A(lFp)li . lK,nM~n)l.

In particular, if A is i-maximal and K, is a Galois submodule of A[fn]
contained in Min), then K, n M~n) = O.
Proof. The main argument is sketched in [BK2, Lem. 3] for n = 1 and

carries over with obvious modifications. As a simplication to the formula
given there, we may use the equality I~A(lFp)1 = I~ A(lFp)l, arising from
the existence of a pairing
(3)
defined by Grothendieck and shown to be non-degenerate for semistable

abelian varieties by various authors. For this pairing and its history, see
~~. 0
Corollary 3.2. If A has semistable bad reduction at an odd prime p
and is 2-maximal for p, then Q(A[2]) =1= Q. If, in addition, Q(A[2]) is
unramified outside 2 and 00, then Q(A[2]) = Q(i).
Proof. If Q(A[2]) = Q, then M~l) is a non-trivial Galois module, in

contradiction to the last assertion of the previous lemma. When Q(A[2])
is unramified outside 2 and 00, property (L4) shows that the discriminant
d L / Q of L = Q(A[2], i) satisfies IdL/QI1/[L:1Qj < 4. Odlyzko's lower bounds
for root discriminants give [L : Q] ~ 5. It follows from Lemma 2.5 that
L = Q(i), whence Q(A[2]) = Q(i). 0
Recall that there is a dictionary (see [Sch, 2.5]) between Goo -
submodules X of T = 1I'(A) and abelian varieties Q-isogenous to A.
If tnT ~ X then K. = X/tnT ~ A [Rn] is the kernel of an isogeny

<p : A --. A' = A/ K.. Write <p' for the quasi-inverse isogeny, such that
<p' 0 :p is multiplication by Rn on A and <p 0 <p' is multiplication by Rn on
A'. Then 'lr(<p') provides a Galois isomorphism of'lr(A') onto X. In
particular, X/RX ~ 'lr(A')/R'lr(A') ~ A'[R].
Lemma 3.3. Let A/Q be an abelian variety and let W be a Z-submodule
of T = 'lr( A) stabilized by a subgroup 1i of Goo. Define Wn = W + tnT
and let S be the set of integers n ~ 0 such that Wn is stabilized by Goo.
Let Co be a collection of abelian varieties containing at least those that
correspond to the Goo-submodules Wn for n E S. Define
N = {g E Goo Ig acts by homothety on A'[R] for all A' E Co}.
If1i and N generate GOCH then Goo stabilizes W.
Proof. Since Wo = T, we have 0 E S. Suppose n E S. Then Wn is sta-
bilized by Goo, so Wn corresponds to A' = A/K.n , where K.n = Wn/tnT.
By assumption, A' is in Co, so the elements of N act as homotheties on
A'[R] ~ Wn/RWn. For w E W, consider a coset w + RWn in Wn/RWn .
If g ENacts on this coset by multiplication by a, then g(w) + RWn =
aw + RWn and therefore
g(w) E aw + lWn = aw + l(W + rT) ~ W + tn+1T = Wn+1'
Hence N stabilizes Wn +1' Given that Goo is generated by 1i and Nand
that 'H already stabilizes W, we find that Goo stabilizes Wn +1' Therefore
n + 1 is in S and it follows that Goo stabilizes Wn for all n ~ O. Hence
Goo stabilizes W. 0
We now explain the special assumptions on which our study of abelian
varieties with small i-division fields rests. Suppose A/Q is semistable,
with bad reduction only at p and that H = Gal(Q(A[l])/QU." is an
i-group. Let C denote the Q-isogeny class of A. Thanks to the theorem
of Faltings [Fa, Satz 6] and its extension to unpolarized abelian varieties
by Zarhin [Zar, Thm. 1], the set of isomorphism classes in C is finite. Put
L for the compositum of the i-division fields of the varieties in C and,
if necessary, adjoin i = A when e = 2. Then L contains F = Q(J.t2)
and L is an (R,p)-controlled extension, as defined in section 2.
Once and for all, we r.boose a topological generator q for the cyclic
pro-l group Iv C Goo. There exists an element T E Goo whose restriction
to F generates Gal(F/Q). When e is odd, we choose T to have order
R- 1, using the fact that H is an R-group. When R = 2, we choose T to
be the generator of some inertia group at an archimedean place of L oo ,
i. e. a complex conjugation. In view of Lemma 2.5, the restrictions of (1
and 7 to L generate Gal(LjQ) when is a regular prime. If = 2 or 3

then 7 2 = 1. However, our methods only require that the action of 7 on
1l'(A) satisfy a polynomial of degree at most 2, so they also apply, for
example, if A is of GL 2 -type
Basic Assumptions.
(C1) A/Q is semistable, with bad reduction only at Pi
(C2) is a regular prime al:d H = Gal(Q(A[])/Q(J.)) is an i-group;
(C3) the action of 7 on 1l'(A) satisfies a polynomial of degree at most
2.
These conditions depend only on the Q-isogeny class C of A. Indeed,
Q(B[oo)) = Loo is the same for all B in C. Furthermore, H is an i-group
if and only if Gal(Q(B[i])/Q(Jl.)) is an i-group, since Gal(Loo/Q(B[i]))
is pro-. Finally, the minimal polynomial of 7 can be read from 1l'(A)
Qt
We will show,cnder (Cl) and (C2), that the following Z-submodules
of 1l'(A) are in fact Goo-modules:
WA = n
i~O
7 i (M 1 (A,v))
Proposition 3.4. Suppose A/Q satisfies (C1) and (C2). Then WA and
YA are stabilized by Goo. Furthermore WA is a pure submodule of1I'e(A).
Proof. Let 'fi be the closed subgroup of Goo generated by a and 7.
Clearly, 7 acts on WA and YA, in view of the definitions (4). Since
a is the identity map on M 1 (A, v), it acts as the identity on WA. As
noted above, (a - 1)(1I'e(A)) ~ M2(A,v), so a acts on YA. Hence WA
and YA are modules for 'fi.
Let L be the compositum of the i-division fields of all the abelian
varieties Q-isogenous to A and, if necessary when i = 2, also adjoin i.
We wish to apply Lemma 4.6, with T = 1l'(A), 9 = Goo and W = WA or
YA. Then N certainly contains Gal(Loo/L). According to Lemma 2.5,
the restrictions of a and 7 to L generate Gal(L/Q). Therefore 'fi and N
generate Goo. It follows that WA and YA are Goo-modules.
The purity of M1(A, v) implies the purity of WA as a Z-submodule
of 1l'(A) by Lemma 2.7. 0
Recall from (1) the definition of an i-maximal abelian variety. If X
is a Ze-submodule of 1I'e(A) , we write X = (X + i!1I'e(A))/f1l'(A) for the
projection of X to 1I'e(A))/f1I'(A) ~ A[f].
Lemma 3.5. Let A be an abelian variety satisfying (Ct) and (C2). Put
Nit = .A/ft (A, v) and define
I'\, = n
j?:.O
rj(.Alit).
If A is i-maximal, then I'\, n M2 = 0.

Proof. By Lemma 2.5, the restrictions of u and r generate
Gal(Q(A[i))/Q). But u acts trivially on Ml and therefore on 1'\" while
r acts on I'\, from the definition. Hence I'\, is a Galois invariant subspace
of A[i] contained in Ntl and we conclude by Lemma 3.1. 0
Corollary 3.6. If A satisfies (et) and (C2), then there are at least 2
primes over p in F = Q(Jt2l)' In addition, if l = 2, then p == 1 mod 8.
Proof. We may assume that A is i-maximal. Suppose l is odd and there
is only one prime over p in F. Then the decomposition group 1Jv projects
onto the group Gal(F/Q) of order l - 1, with pro-l kernel. Hence we
could have chosen r in 1Jv and so I'\, in the lemma is simply MI. Therefore
M2 = I'\, n M2 = O. This implies that the toroidal dimension of the bad
fiber of A is 0, contradicting bad reduction at p.
For l = 2, we showed in [BK2, Prop. 5] that p == 1 mod 4, whence
there are 2 primes over pin F = Q(i). Suppose p == 5 mod 8 and let </J
be a Frobenius element of 1Jv , in the sense that </J induces the pth power
map on the residue field. Then 1Jv = (</J, u) and </Ju</J-I = up. Since
<p( \"'2) = -\"'2, the restrictions of r, </J and u to E = Q(i, \"'2, yIP) clearly
generate Gal(E/Q). But E is the maximal elementary 2-extension of Q
unramified outside {2,p, oo}, so Goo = (r, </J, u) by Burnside's Lemma.
Hence'Dv has index 2, so is normal in Goo.
Now we may write rur- 1 = xuY for some x, y E Z2. Checking the
action of both sides of this equation on a generator for '1r2(Jt), we find
that x = 0, so r normalizes Iv. Hence MI is stabilized by r and we get
a contradiction as in the previous case. Therefore p == 1 mod 8. 0
From hereon in, we impose the additional assumption (03). Since the
dual abelian variety A is isogenous to A, it also satisfies (03). Write Mj
for the projection of Mj = Mj(A, v) to .A[l]. In the perfect pairing
A[i] x A[i] --+ Jtl (5)

induced by the eoo-pairing (1), we have (MI).l = M~. Recall that t
and a respectively denote the toroidal and abelian dimension of the bad
fiber of A at p.
Proposition 3.7. Suppose A satisfies (C1), (C2), (C3) and is t-

maximal. Then WA and YA are pure Ze-submodule of1l'e(A) of rank 2a
and 2t respectively, stabilized by Goo. Under the eoo-pairing, Wi = Y A
and Y} = WA- We have a direct sum decomposition 1l'e(A) = WA EB YA.
Furthermore, the natural map WA ~ MI(A,v)/M2(A,v) is an isomor-
phism of Ze-modules.
Proof. Hypothesis (C3) implies that T2(M2) ~ M2 + T(M2), so
YA = LT j (M2) = M2 + T(M2)'
j~O
Furthermore, T2(M 2) ~ M2 + T(M2) and so T2(Ml) ;? Ml n T(Ml)

by the lFe-vector space duality (5). Hence the Galois invariant subspace
/'\, of Lemma 3.5 reduces to /'\, = MI n T(Md and /'\, n M2 = 0 by that
lemma. Since M2 ~ MI, we have
Therefore, YA = M2 + T(M:2) is pure and a direct sum, of rank 2t, by

Lemma 2.7(iv). Similarly, YA is pure in 1l'e(.A), so we have
Y A = LTj(M~)
j~O
= LTj(Mr) =
j~O
(n Tj(MI)l
j~O
= wi
Hence WA is pure of rank 2a. In view of Lemma 2.7(i) and (6), we have
tvA n M2 ~ (Ml n T(Ml)) n M2 = T(Ml) n M2 = O.
Hence W A + M 2 is pure ofr ank t + 2a and a direct sum. By the obvious
inclusion and equality of ranks, we have WA + M2 = Ml. It follows
that
WA ~ (WA +M2)/M2 = MI/M2.
By (6) and Lemma 2.7(iv), we also find that T(Ml) + M2 is a pure
submodule of 1l'e(A) and a direct sum. Equality of ranks implies that
1l'e(A) = T(Ml)EBM2 and so 1l'e(A) = T(WA)EBT(M2)EBM2 = WAEBYA,
as claimed. Finally, we have already observed in Proposition 3.4 that
WA and YA are Goo-modules. 0
4. The .e-adic formal group and the proof of

theorem 1.1
Suppose that A is an abelian variety of dimension d defined over Qe,
with good reduction modulo t. The kernel of reduction has the structure
of a formal group of height h and dimension d, with d :$ h :$ 2d. We
put FA for this formal group, suppressing the dependence on i. The
G~-module C = FA[.eOO ] is isomorphic to (Qt/Zt)h as an abelian group.
Furthermore, C is a connected i-divisible group over Zt while A[ioo1/C
is etale.
Set Vt(A) = '1l'tCA) Zt Qt and VtCFA) = '1l'tCFA) Zt Qt for the
Tate vector spaces of A and FA. Clearly the inertia group inside G~
acts trivially on Vt(A)/VtCFA), as this quotient corresponds to the Tate
vector space of the reduction of A over IFt. In the special case of ordinary
reduction (i.e. h = d), we have
(1)
under the Wei! pairing Vl(A) x Vt(.A) .- Qt(1) , where A denotes the
dual abelian variety of A. (See for example [CG, p. 1541.)
Lemma 4.1. Let A be an abelian variety of dimension d defined over
Qt with good reduction modulo t. If the tame ramification degree of the
extension Qt(A[i])/Qt divides - 1, then A is ordinary.
Proof. We extend the base to the ring of integers R in the unramified
closure K = Q~r to obtain an algebraically closed residue field. Ac-
cording to Raynaud (see [Gru, Thm. 4.4]), the finite group scheme A[i1
admits a composition series whose simple constituents are IF-vector space
schemes for varying finite fields IF of characteristic t. Such a constituent
M is a cyclic IF-module, determined as follows. By simplicity, the wild
ramification subgroup of GK must act trivially on M, so the action of
GK factors through a quotient Gal(EIK), such that E ~ K(A[]) and
ElK is tamely ramified. Suppose [E: K] = r. Then IF = lFt(J.tr), with
the action of a primitive rth root of unity induced by the action of a
generator for Gal(EI K).
In our case, r divides .e - 1, whence IF = lFt and M therefore is a
group scheme of order.e. These are classified by Oort-Tate. Thus each
simple constituent Mj of A[i] has the form Mj ~ G!j,bj ' with ajbj = i
in the standard notation [Gru, Theorem 2.1]. Since the valuation v of
R is unramified, we have v(aj) E {O,1}. Then G!j,bj is etale (resp.
connected) if v(aj) = 0 Crespo 1). Let no Crespo nl) be the nUiH~er of aj
such that v(aj) = 0 (resp. 1).
The exponent of the different of an affine group scheme g may be
defined as follows [Gru, p. 62]. Suppose g = spec(2L), where 2L is an R-
algebra. Let nilR be the module of Kahler differentials of 21 and define
o(Q) to be the length of the R-module s*ni/R = ni/R ~ R, where the
tensor product is formed over the zero-section 8 : 2L.- R.
Since the f-divisible group arising from f-power torsion on A has di-
mension d, we find that 8(Al]) = dv(f) = d for example by [Gru,
Prop. 3.4]. Furthermore, 8(G a . b.) = v(aj) by [Gru, Prop. 2.3]. But 8
J' J
behaves like an additive Euler characteristic on short exact sequences.
Hence d = Lj v(aj) = n1. Note that the simple constituents that be-
long to the connected group scheme FA [f] are precisely those for which
v(aj) = 1. By definition of the height h of FA we have
fh = IFA[f] I = IT IG~j,bjl =fnl.

v(aj )=1
Hence h = n1 = d as claimed. o
We now return to the global situation: A is an abelian variety de-
fined over Q and we impose the basic assumptions (C1) and (C2) of
section 3. Preserving previous notation, we set Loo = Q(A[foo]) and
Goo = Gal(Loo/Q). As usual, we have a topological generator a for
the inertia group Iv inside Goo at a place v over p and we have an
element T E Goo of order [F : Q), where F = Q(JL2t) , such that the
restriction of T generates Cal(F/Q). According to Lemma 2.5, the re-
strictions of (J and T to the f-division field of A generate its Galois
group. By fixing an embedding of Loo to Qt, we get a natural D v-
isomorphism 1l't(A) ---+ 1l't(A~), where A~ is obtained by base exten-
sion to Qt. We are interested in elements of 1l't(A) whose image lands
in 1l'e(FA) ~ 1l't(A~).
Lemma 4.2. Let A/Q be an abelian variety satisfying (C1) and (C2).
Suppose X is a Goo-submodule of1l'(A) with trivial action by the inertia
group Iv. Let K 00 be the fixed field of the kernel of the representation of
Goo afforded by X. Then Koo is unramified outside E and totally ramified
over E. Furthermore X ~ 1I't(FA).
Proof. By assumption, Koo is unramified at v. But then Koo/Q is un-
ramified at all places over p because it is Galois.
It is convenient to pass to the compositum KooF. Let). be the place
of KooF over f determined by our choice of embedding Loo ---+ Qt. Since
the prime over f in F is totally ramified, we may verify that Koo is
totally ramified over f by showing that the inertia group IF,).. inside
Hoo = Gal(KooF/ F) is equal to Hoo. Suppose on the contrary that
IF,).. is a proper subgroup of Hoo. Condition (C2) implies that Hoo is
pro-f, so there exists a subgroup of index f in Hoo , containing IF,)... The
corresponding fixed field is a cyclic extension of F of degee f unrami-
fied everywhere, whose existence contradicts our assumption that f is a
regular prime.
Put x(n) = (X + F'll"t(A))/F'll"t(A) ~ A [f!n] for the nth layer of
X. Then we may represent elements of X in the form x = limxn with
+--
Xn E x(n) compatible under the projection maps. Since f! is totally

ramified in K oo , the reduction xn modulo>. is an element of A(lFt ).
Suppose the f!-part of the order of A(lFt ) is f!m. For all n ;:::: m + 1, it
follows that xn-m = imxn = O. Hence Xn' is in the kernel of reduction
for all layers n' = n - m ;:::: 1 and we have X S;;; 'll"t(FA), as claimed. 0
Theorem 4.3. Let i be a regular prime and let A/Q be a semi-
stable abelian variety with bad reduction only at p, such that H =
Gal(Q(A[i])/Q(ILt)) is ani-group. Assume that the action ofr on'll"t(A)
satisfies a polynomial of degree at most 2. Then A has ordinary reduction
modulo i and totally toroidal reduction modulo p.
Proof. We may assume that among the members of its Q-isogeny class
C, the variety A is i-maximal; that is, A E Vt(C) in the notation of
(1). The dual abelian varieties A and A are Q-isogenous and A also
is i-maximal because I~ A(lFp ) I = I~ A (lFp ) I by the perfect pairing (3).
In view of Proposition 3.7, we have pure submodules WA of'll"t(A) and
WA of 'll"t(A), of rank 2a and stabilized by Goo. Furthermore, the Weil
pairing induces a perfect pairing
(2)
Because Iv acts trivially on WA, we find that WA S;;; '1l't(FA) by Lem-
ma 4.2. Similarly, WA ~ 'll't(F,A)' Under the assumption that H is an
i-group, Lemma 4.1 implies that A has ordinary reduction modulo i,
so the Wei! pairing on WA x WA is trivial by (1). This contradicts the
perfect pairing (2) unless WA = WA = O. Hence a = 0 and A is totally
toroidal at p. 0
Proof of Theorem 1.1. Assume that A/Q is semistable, with bad reduc-
tion only at p, that i = 2 or 3 and that H = Gal(Q(A[f!])/Q(ILt)) is
nilpotent. Then H is in fact an f!-group by Proposition 2.4. We have
p == 1 mod 8 when f! = 2 and p == 1 mod 3 when f! = 3 by Corollary 3.6.
Since r2 = 1 when f! = 2 or 3, we may conclude by Theorem 4.3. 0
5. Small i-division fields

As building blocks for semistable abelian varieties with small f!-
division fields and bad reduction at only one prime p, we consider el-
liptic curves E /Q of conductor p such that the mod-f! representation
PE,l : GQ ~ Aut(E[]) ~ GL2(lFl) is not surjective. Let D..E denote
the minimal discriminant of E and recall that ordp(LlE) = -ordp(jE),

where jE is the j-invariant. If e ~ 7 or ordp(jE) 0 mod e, then E
admits a Q-isogeny of degree e by [Ser3, Prop. 21J. However, if f ~ 7
and ILlEI is an gth power, E admits an isogeny of degree f, for example
by [BK1, Prop. 9.2J. It then follows, as in [Ser3, p. 307], that the isogeny
class of E contains a curve with a rational point of order f. The only
such examples for odd f occur in the well-known cases (f,p) = (3,19),
(3,37) and (5,11), referred to below as Miyawaki curves [Mi]. For e= 2,
according to Neumann [Ne] or Setzer [Set], we have p = 17 or p = u 2 +64.
l
In the latter case, there is a unique isogeny class consisting of the curves
X3 + U ~ 1x 2 - X, Ll = p,
y2 + xy = (1)
x3+ u ~ 1x2 + 4x + u, Ll = _p2,
where the sign of u is chosen to guarantee that u == 1 mod 4.

Since the order of the connected component group of a semistable
elliptic curve C/Qp is ICPc(lFp )I = ordp(Llc), one easily verifies that each
Neumann-Setzer I)r Miyawaki isogeny class contains a unique curve Et
with maximal connected component group. Its discriminant satisfies
ord (Ll ) = {4 if (f,p) = (2,17), (2)

pte otherwise.
From the Tate parametrization over Qp, we find that Q(EtfeJ) is unram-
ified over p and so Q(EdfJ) = F = Q(1'2t) by Lemma 2.5 and Corol-
lary 3.2. According to [MOl. Et is the strong Weil curve in its isogeny
class.
Write E = Et, suppressing the dependence on f and p, which should
be clear in context. Recall that the foo-division tower Loo is the same for
all curves in the isogeny class of E. Let Goo = Gal(Loo/Q) and observe
that Gal(L oo / F) is pro-e. As usual, a E Goo denotes a topological
generator of the inertia group at a place v over the bad prime p and
r E Goo denotes an element of order [F : Q] whose restriction to F
generates Gal(F/Q).
The Grothendieck module Ml (E, v) = M 2 (E, v) is a pure Zt-module
ofrank 1 in 1I'l(E) and we have 1I'l(E) = YE = M2(E, v) Ell r(M2(E, v))
by Proposition 3.7. Choose a generator P for M2(E, v). With respect
to the generating set {P, r(P)} for 1I'l(E), we obtain a matrix represen-
tation
The Tate parametrization of E over Qp shows that
PE(U) = (~ ~), (3)
where s equals ordp(~t) up to multiplication by a unit in Zl, so s ==

omod f. Since the eigenvalues of rare 1 and w = x(r), where X is the
cyclotomic character, we have
0
PEer) = ( 1 1+w .
-w) (4)
The following lemma will be used to study abelian varieties Q-

isogenous to products of Et for fixed f and p. It is sufficient for our
purposes to state it for abelian varieties A, B defined over Q. The repre-
sentation PB of Gal(Q(B[fOO])/Q) afforded by 'JI'l(B) naturally extends
to the completed group ring
At(B) = Zl[[Gal(Q(B[t)O])/Q)]].
Lemma 5.1. Let cp : B -+ A be a Q-is0geny of abelian varieties. For
9iven 91,92 in Gal(Q(B[fOO])/Q), assume there exists "( E Al(B) such
that PB (91) = PB (92 + f "(). Then the restrictions of 91 and 92 to A [l]
are equal.
Proof. Given a E A[f], we may find bE B of f-power order, say f!", such
that cp(b) = a. Since 91 acts on b through the representation PB modulo
fn, we have 91(b) = g2(b) + l"(b). Then
g1 (a) = g1 (cp(b)) = cp(g1 (b)) = cp(g2(b) + l "Y(b)) = g2(a) + "Y(f a) = g2(a),
because 'I' commutes with Galois and fa = O. o
Proposition 5.2. Suppose the abelian variety A is Q-isogenous to a
product of Neumann-Setzer or Miyawaki curves for fixed land p. Let
G = Gal(Q(A[f])/Q).
(a) Iff = 2, then G is a quotient of Z/2 x Z/2.
(b) If l = 3, then G is a quotient of S3'
(c) If l = 5, then H = Gal(Q(A[5])/Q(JL5)) is abelian of exponent
dividing 5 and rank at most 2. Furthermore, f- 2hf2 = h- 1 for all
hEH.
Proof. Since A is isogenous to Ed, where d = dimA and E = Et, we
have Q(AW)O]) = Q(Ed[fOO]) = Loo. Furthermore, 1f'l(Ed) ~ Eef 1f'l(E)
affords the representation PEd = Eef PE
Suppose i = 2 or 3, so that PEer) = (? 6). Then PE((1r-r(1-1) = s 1,

where 1 is the identity in M2(Zt). Therefore PEd((1r-r(1-1) = s 1, where
1 now denotes the identity in M 2d (Zt). By Lemma 5.1, the restrictions
0- and 7' of (1 and r to G therefore satisfy 0-7' = 7'0-- 1. But 0- and
7' generate G by Lemma 2.5 and o-t = 1 by property (L3) of (i, S)-
controlled extensions. Hence G is a quotient of 71./2 x Z /2 if i = 2 or a
quotient of 83 if i = 3.
For i = 5, we find that PE((1r 2 - r 2(1-1) = (1 + w)s 1 E M2(Z5), so
PEd((1r 2 - r 2 (1-1) = (1 + w)s 1 E M2d(Z5).
Now 0-7'2 = 7'20-- 1 by Lemma 5.1. Acccording to Lemma 2.5, the sub-
group H of G is generated by the conjugates of 0- under the action of 7'.
Since 7'2 normalizes 0-, to prove that H is abelian, it suffices to show that
0- commutes with o-T = 7'-10-7'. The reader may verify by judicious use
of PE((1-1) and PEer) that the image of PE on the group ring Z5[[G oo ]]
contains all 2 x 2 matrices congruent to a modulo s. One can also check
that
PE((1T ( 1 - (1(1T) = -S2w (~ 2(1_~W)).
Hence PE((1T ( 1 - (1 (1T) belongs to 5 PE(Z5[[Goo lJ) and so PEd((1T (1-(1 (1T)
belongs to 5PEd(Z5[[Goo ]]). It follows from Lemma 5.1 that o-T and 0- do
commute. We may conclude that H is abelian of exponent dividing 5
and rank at most 2, generated by 0- and o-T. Furthermore, the action of
conjugation by 7'2 on H is given by inversion. 0
Proposition 5.3. Let C be the Q-isogeny class of Ed, where E is an el-
liptic curve of conductor p with some rational i-torsion. Then Q(A[]) =
Q(JL2t) for all abelian varieties A in Vt(C).
Proof. We may assume that E = Et is i-maximal. Put
M2 = M2(E d, v) = EfJt M2(E, v).

In view of (3), we have (1r(z) = r(z) + s z for all z E M2.
Suppose A E C and let A be the corresponding isogeny.
The basic assumptions (Cl), (C2) of section 3 clearly are satisfied by
Ed. Condition (C3) holds because PEd = (fJPE is a sum of d copies of a
fixed representation of GL 2 -type. Since these conditions are isogeny
invariant, they also hold for A. Suppose, in addition, that A is -
maximal. Then Proposition 3.7 implies that 1I't(A) = M~ E9 r(M~),
where M2 = M2(A,v) = M1(A,v).
By construction, (1 acts trivially on M 2. To examine the action of (1
on r(M 2), recall from the start of section 3 that M2 contains <p(M2)
with finite index, say n. Given z' E M~, we may therefore find z E M2
such that nz' = cp(z). It follows that
nO'r(z') = cp(O'r(z)) = cp(r(z) + s z) = nr(z') + ns z'.
But 1I'1!(A) is torsion-free, so O'r(z') = r(z') + s z'. Since s == 0 mod i, we
find that 0' acts trivially on the first layer r(M~) of r(M~) and therefore
on all of A[i] = M~ ED r(M 2). Hence Q(A[i]) = Q(#J.2) by Lemma 2.5
and Corollary 3.2. . 0
Lemma 5.4. Suppose that C is a Q-isogeny class of abelian varieties of
dimension d satisfying (C1), (C2), (C3) and that Q(A[i]) ~ Q(#J.2) for
each A in Vi(C). Then there exists a Z-submodule X oflI'(A) of rank
2 and stabilized by Goo such that lI'(A) ~ Xd as Goo-modules.
Proof. Theorem 4.3 implies that A has totally toroidal reduction modulo
p, so M2 = M2(A,v) = MI(A,v). For all A E v(e) , we have M2 n
r(M2) = 0 by (6) and lI'(A) = YA = M2 6:) r(M2) by Proposition 3.7.
Choose a set of free generators ml, ... , md for M2' For j = 1, ... , d,
define xU) to be Z-span of mj and r(mj). Each xU) is a pure sub-
module of lI'(A) of rank 2 and lI'l(A) = 6:)1=IXU).
We wish to use Lemma 3.3 to show that X = xU) is a Goo-module.
Take 'H. to be the grou:) generated by r, so X certainly is an 'H.-module by
(C3). Suppose Xn = X +i nlI'l! (A) is a Goo-module and let cp : A -+ A' be
the Q-isogeny whose kernel is K = (X +.enlI'(A))/.enlI'(A) ~ Z/.en6:)Z/in.
Write M~n) for the projection of M2 to lI'(A)/.enlI'(A) ~ A[.en]. It is
clear that Kn M~n) is isomorphic to one copy of Z/in, generated by the
coset of mj' Hence IK n M~n)1 =.en and we have I~AI(iFp)l = I~A(iFp)11!
by Lemma 3.1. Therefore A' also is i-maximal. Then by assumption,
Q(A'[iD ~ F = Q(#J.2)' Hence the subgroup N defined in Lemma 3.3
contains Gal(Loo/F). But Gal(Loo/F) and 'H. certainly generate Goo.
We may conclude from Lemma 3.3 that X is a Goo-module.
The following standard argument now shows that the xU) 's are iso-
morphic as Goo-modules. Reasoning as above, we find that for each
j =P 1, the Z-submodule of lI'(A) of rank 2 spanned by ml + mj and
r(m! + mj) also is stabilized by Goo. But then the matrix representa-
tion of Goo afforded by the Z-span of ml and r(ml) must be identical
to the representation afforded by the Z-span of mj and r( mj). It now
follows that there is a Goo-isomorphism lI'(A) ~ Xd, where X is the
Z-module spanned by m and r(m) for any choice of mE M2 such that
m iM2. 0
For the proof of our final result, we need to observe that i-maximality
is preserved for certain products of abelian varieties. Note that A =
B x C satifies (C1), (C2), (C3) of section 3 if and only if each factor

does.
Lemma 5.5. Suppose A sa-tisfies (C1), (C2), (C3). If A = B x C is
the product of f-maximal abelian varieties, then A also is i-maximal.
Proof. To show A is i-maximal, it suffices to consider Q-isogenies r.p :
A -+ A' of i-power degree. Set", = Kerr.p and assume", has exponent
n. Theorem 5.4 implies that A has totally toroidal reduction at p, so
Ml(A,v) = M2(A,v) and similarly for Band C. Moreover, inside
11'(A) = 11'(B) x 11'(C), we have Ml(A,v) = Ml(B,v) x Ml(C,V).
Write M~n) for projection to 11'dln 11' and let
"'0 = ",n.M~n)(A,v)
= {(b,c) E "'I b E M2(B,v), c E M2(C,V)} C A[fn].
In view of (6), we have
- (n) - (n) - (n) - (n)
M2 (B, v) n T(M2 (B, v)) = M2 (C, v) n T(M2 (C, v)) = 0,
whence "'OnT("'O) = O. But", is a Galois module, so it contains "'O+T("'O)
and therefore 1"'1 ~ l"'oIIT("'o)1 = 1"'01 2 It follows from Lemma 3.1 that
I<pA(lFp)i ~ i<PA'(lFp)i. Hence A is f-maximal. 0
Theorem 5.6. Let C be a Q-isogeny class of semistable abelian varieties
of dimension d with bad reduction only at p. Assume f is a regular prime
and Q(A[]) ~ Q(JL2) for every A in V(C). Suppose the action of T on
11'(A) satisfies a polynomial of degree at most 2. Then = 2, 3 or 5
and there exists an elliptic curve E of conductor p with some rational
l-torsion such that C is the Q-isogeny class of Ed.
Proof. We may apply Lemma 5.4 to A E V(C) to obtain a Z-submodule
X of 11'(A) of rank 2 stabilized by Goo, such that 11'(A) ::: X d is a
Goo-isomorphism. According to the Tate conjecture, proved by Faltings
([Sch]), we have
EndQ(A) Z ::: EndG oo 'll'(A).

But the commutant EndGoo 11'(A) clearly contains Md(Z). It follows
that the Zl-rank of EndQ(A) Zl is at least cP.
Suppose A is simple, so that D = EndQ(A) Q is a division algebra.
We have shown above that [D : QJ ~ d,2. But D acts on the space of
invariant holomorphic differentials n1 (A), so
d = dimQOl(A) = [D: Q]dim D Ol(A) ~ d2 .
Hence d = 1 and A is a Neumann-Setzer (f = 2) or Miyawaki (f = 3 or

5) elliptic curve.
If A' E Vi(C) is not Q-simple, we may find proper abelian subvarieties
B' and C', such that A' is Q-isogenous to B' x C'. Certainly B' and
C' inherit properties (C1), (C2), (C3) from A'. Choose B and C to be
i-maximal in the isogeny class of B' and C' respectively. Then A' is
isogenous to A = B x C and Lemma 5.5 shows that A is in Vi(C). By
assumption, the i-division field of A therefore is contained in Q(P2i) and
so a fortiori the same is true of the i-division fields of B and C. Thus
we may complete the proof by induction. 0
Corollary 5.7. Let C be a Q-isogeny class of semistable abelian varieties
of dimesion d with bad reduction only at p and suppose Gal(Q(A[2])/Q)
is abelian for all A E V2 (C) . Then C is the isogeny class of Ed, where E
is a Neumann-Setzer curve of conductor p.
Proof. Theorem 1.1 tells us that A is totally toroidal, so for A E Vi(C),
we have A[2] = .M2 $ T(.M2) by (6) and equality of dimension. Since (J'
acts trivially on M2 and the restrictions of (J' and T to Q(A[2j) commute
by assumption, we find that (J' acts trivially on A[2]. It follows that p
is unramified in Q(A[2j), whence Q(A[2]) = Q(i) by Corollary 3.2. Our
claim is now implied by Theorem 5.6. 0
The following two examples show that the i-division field of a simple
semistable abelian variety of dimension greater than 1 with one place
of bad reduction may be All i-group. Compare with Theorem 5.6 and
Theorem 1.3.
From a model for the modular curve Xo(41) given by Weber [Web],
one finds the minimal model
y2 + (x 4 + 2x 3 - x)y = -3x 6 -16x5 - 29x4 -14x3 + 13x2 + 9x - 4.
Its Jacobian Jo(41) hl:iS real multiplications by the maximal order in the
totally real cubic field of discriminant 148 and bad reduction only at
41, where it is totally toroidal. The 2-division field Lo of Jo(41) can be
seen to be cyclic over K = Q( J-41) and dihedral over Q, as we expect
from Proposition 2.6. Compare with Proposition 5.2(a). Since the class
number of K is 8, the maximal (2,41)-controlled extension has degree
32 over Q. However [Lo : Q] = 16.
A minimal model for Xo(31) is given by
y2 + (x 3 + x 2 + 2x + 1)y = _x 4 - 2x3 - 3x2 - 2x-1.
Its Jacobian variety A = Jo(31) admits real multiplications by the ring
of integers of Q( VS) and is the full Eisenstein quotient at 5, cf. [Maj.
For the endomorphism 1[' = y'5, we have A[1['j = 'll/5'll 6:) J.t5 and so
H = Gal(Q(A[5])/Q(J.l.5)) is an elementary abelian 5-group. One can
check that its IF5-rank is 3, in contrast to Proposition 5.2(c). Using the
Remark after Lemma 2.3, we find that Q(A[5]) is the maximal (5,31)-
controller! extension abelian over Q(J.t5)'
Acknowledgments
The first author wishes to express his indebtedness for the generous
hospitality, patience, friendship and stimulation afforded him by Profes-
sor Hashimoto and his students at Waseda University. He also thanks
Professors K. Miyake of Tokyo Metropolitan University, T. Ibukiyama
of Osaka University and H. Saito of Kyoto University, their institutions
and colleagues. Together, they made it possible to spend a wonderful
month in Japan, beyond the week of this Conference.
References
[And] A. N. Andrianov, "Quadratic forms and Hecke operators," Grundlehren der
Math., 286. Springer-Verlag, 1987.
[BSP] S. B6cherer and R. Schulze-Pillot, Siegel modular forms and theta series at-
tached to quaternion algebras, Nagoya Math. J. 121 (1991),35-96.
[BK1] A. Brumer and K. Kramer, The rank of elliptic curves, Duke Math. J., 44
(1977), 715-743.
[BK2] A. Brumer and K. Kramer, Non-existence of certain semistable abelian vari-
eties, Manuscripta Math. 106 (2001), 291-304.
[CG] J. Coates and R. Greenberg, Kummer theory for abelian varieties over local
fields, Invent. Math. 124 (1996), 129-174.
[Ed] B. Edixhoven, On the prime to p-part of the group of connected components
of Neron models, Compositio Math. 97 (1995), 29-49.
[Fa] G. Faltings, Endlichkeitssiitze fur abelsche Varietiiten uber ZahlkiJrpern, In-
vent. Math. 73 (1983), 349-366.
[Fo] J.-M. Fontaine, Il n'y a pas de variete abelienne sur Z, Invent. Math. 81
(1985), 515-538.
[Gro] A. Grothendieck, Modeles de Neron et monodromie, in "Groupes de mon-
odromie en geometrie algebrique. I," Lecture Notes in Mathematics, 288.
Springer-Verlag, N<3w York, 1973.
[Gru] F. Grunewald, Some facts from the theory of group schemes, in "Rational
Points (G. Faltings and G. Wiistholz, ed.)," Aspects of Mathematics, E6. F.
Vieweg & Sohn, Braunschweig, 1984.
[HMJ F. Hajir and C. Maire, Extensions of number fields with wild ramification of
bounded depth, Inter. Math. Res. Notices (2002), 667-696.
[Has] K.-I. Hashimoto, The dimension of the spaces of cusp forms on Siegel upper
half-plane of degree two, I, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 30 (1983),
403-488.
[Ibu] T. Ibukiyama, On relations of dimension of automorphic forms of Sp(2,R)
and its compact twist Sp(2), I, Adv. Studies in Pure Math. 7 (1985), 7-29.
[MM] J. M. Masley and H. L. Montgomery, Cyclotomic fields with unique factor-
ization, J. Reine Ang. Math. 286/287 (1976), 248-256.
[Ma] B. Mazur, Modular curves and the Eisenstein ideal, Publ. Math. I.H.E.S. 47
(1976), 33-186.
[Me] W. McCallum, D'uality theorems for Neron models, Duke Math. J. 53 (1986),
1093-1124.
[MO] J.-F. Mestre and J. Oesterle, Courbes de Weil semi-stables de discriminant
une puissance m-ieme, ,T. Reine Ang. Math. 400 (1989), 171-184.
[Mi] I. Miyawaki, Elliptic curves of prime power conductor with Q-mtional pOints
of finite order, Osaka Math J., 10 (1973), 309-323.
[Ne] O. Neumann, Elliptische Kurven mit vorgeschriebenem Reduktionsverhalten.
II, Math. Nach. 56 (1973), 269-280.
[Seh] N. Schappacher, Tate's conjecture on the endomorphisms of abelian varieties,
in "Rational Points (G. Faltings and G. Wiistholz, ed.)," Aspects of Mathe-
matics, E6. F. Vieweg & Sohn, Braunschweig, 1984.
[Scho] R. Schoof, Semistable abelian varieties over Q, (2001), preprint.
[Serl] J.-P. Serre, Facte'lJ,rs locaux des fonctions zeta des varietes algebriques, Sem.
DPP, expose 19 (1969-19;'0).
[Ser2] J.-P. Serre, "Local Fields," Lecture Not,'S in Mathematics, 67. Springer-
Verlag, New York, 1979.
[Ser3] J.-P. Serre, Proprietes galoisiennes des points d'ordre fini des courbes ellip-
tiques, Invent. Math. 15 (1972), 259-331.
[Ser4] J.-P. Serre, Sur les representations modulaires de degre 2 de Gal(Q/Q), Duke
Math. J. 54 (1987),179-230.
[Ser5] J.-P. Serre, Oeuvres. Collected papers. IV (1986),1-55, Springer-Verlag, 2000.
[Set] C. B. Setzer, Elliptic curves of prime conductor, J. Lond. Math. Soc. 10
(1975), 367-378.
ISS] Sehinzel, A. and Sierpinski, W., Sur certaines hypotheses concernant les nom-
bres premiers, Acta Arithmetica 4 (1958), 185-208.
[Web] H.-J. Weber, Hyperelliptic simple factors of Jo(N) with dimension at least 3,
Exper. Math. J. 6 (1997), 273-287.
[Yos] H. Yoshida, On Siegel modular forms obtained from theta seri"s J. Reine
Ang. Math. 352 (1984), 184-219.
[Zar] Yu. G. Zarhin, A finiteness theorem for unpolarized abelian varieties over
number fields with prescribed places of bad reduction, Invent. Math. 79 (1985),
309-321.
Q-CURVES
WITH RATIONAL j-INVARIANTS
AND JACOBIAN SURFACES
OF GL 2-TYPE
Ki-ichiro HASHIMOTO
Department of Mathematical Sciences
Waseda University,
3-4-1, Okubo Shinjuku-ku, Tokyo,
169-8555 Japan
[email protected]
Abstract We study elliptic curves over quadratic fields with rational i-invariants
regarded as Q-curves, in connection with jacobian surfaces of genus two
curves over Q. We discuss their minimality as Q-curves, and the classi-
fication, as well as the Neben type characters of the associated modular
forms. This can be described as the sign change phenomenon by quar-
tic twists of curves over Q. We also study their 2-fold covers by genus
vi -
two curves. Among others, we construct a parametric family {~f
genus two curves over Q covering minimal Q-curves over Q( 123)
with i-invariant j. We find in the twists of {G(j)} concrete equations
of curves over Q whose jacobians are isogenous over an extension of Q
to Shimura's abelian surfaces Aj attached to normalized eigen forms
f E S2(N, (~), whose Fourier coefficients belong to Q(v'=I) , in all
known non eM cases i.e., N = 37, 65, 104, 157, 397, and 877.
o. Introduction
Let r 1(N) be the congruence subgroup
r1 (N) = { (~ ~) E SL2(Z) I c == 0, a, d == 1 (mod N) } ,
and let 82 (r 1(N)) be the space of cusp forms of weight 2 with respect
to r 1(N). To e~l.ch normalized Hecke eigen form f E 8 2(r 1(N) ), one can
[received: July 1, 2002; accepted in revised form: December 10, 2002J

40 K.-J. HASHIMOTO
attach as in Theorem 7.14, [Sh1J an abelian variety AI over Q obtained

as a Q-simple factor of J 1 (N), the jacobian variety of the modular curve
X1(N) . AI is called the Shimura's abelian variety associated with J.
Let J(z) = L:~=1 al(n)qn (q = e21r v'-IZ) be its Fourier expansion and
let K I = Q ({aI (n) }) be the field generated by the Fourier coefficients of
J. Put 9 := [KI : QJ. Then AI is of dimension g, and is characterized,
up to Q-isogeny, by the property EndQ(AI) Q = KI where EndQ(AI)
denotes the ring of endomorph isms of A I which are defined over Q, and
the action of a f (p) E K f corresponds to that of the Hecke operator T (p).
It then follows that the L-function of Aj/Q is given by
II L(UiJ,S),
9
L(AJlQ,s) =
i=l
where {0'1' ... ,0'd} denote the set of embeddings of K I into C, and
00
uJ = I:uaf(n)qn.
n=l
In general an abelian variety A over Q is called of GL2-type if EndQ (A)

Q is a commutative field of degree equal to dim A. The so called modu-
larity conjecture claims that any such A/Q should be isogenous over Q
to Af for some N and J E S2(r 1 (N)) (c.f. [Ri3], [HHM]).
In this paper we are concerned, not to the proof of this conjecture,
but to study the following closely related problems in the case 9 = 2 :
How many, for a given quadratic field K, abelian surfaces are there
over Q which are of GL2-type with EndQ(A) Q ~ K ?
Constuct as many curves of genus 2 over Q as possible, whose
jacobian surfaces are of GL2-type, with EndQ(Jac(C)) Q ~ K.
Of special interest to us is the following problem:
To each eigp.n form f E S2(rl(N)) such that [Kf : QJ = 2, find a
curve C/Q such that Jac(C) is isogenous over Q to Af.
Here of course we require to find an isogeny to be defined over a field
as small as possible. Also note that the existence of such a curve C
over Q is not a priori clear. Moreover, if f has nontrivial Neben type
character, one cannot in general expect that the isogeny Jac(C) --+ AI
is defined over Q, since the Euler factors of L-functions of these have
different shapes.
These problems have been studied in our previous work [HaJ in the
case K f = Q( A) and f is an eigen form with quadratic Neben type
Q-curves with j E Q and jacobian surfaces of GL2-type 41
character. We discussed with an explicitly constructed family, several

arithmetic properties of minimal elliptic Q-curves of degree 5 over real
quadratic fields and jacobian surfaces of GL2-type over Q associated
to each other. And we found in this family the solutions to the third
problem for all known cases such that Kf = Q( A).
The present paper treats the case Kf = Q( A), and obtains the
similar results as [HaJ. Thus we are interested in those Af which decom-
pose over Q( v'N) into the product of elliptic curve B f with itself which
has rational j-invariant. One may call Bf a Q-curve of "degree one", so
that the study of such objects, as presented here, could have been done
before [HaJ. We should note that the modularlity for such Q-curves has
been completely settled (see [BeDT]).
As for the first of the above problems, we shall establish the com-
plete classification of e-minimal Q-curves (e = 1) of degree one over
quadratic fields K, in the sense that the isogeny f.,Lu : E _ u E is also
defined over K and satisfies
u f.,Lu0f.,Lu = exid. (e = 1).

In the case e = +1, the question is easily answered: All (+1)-minimal
Q-curves of degree one over K are obtained as the twist ETate(j)6 by
some 0 E QX of the Tate's model
E Tate ( ,). y2 Xy X3 36 X 1
J. + = - j _ 1728 - j - 1728
We call an abelian surface A/Q to be of fake GL2-type, if EndQ(A)

Q ~ QEElQ, while EndK(A)Q ~ M2(Q) for a quadratic field K. Then
we shall prove, for a quadratic field K = Q( rm), the following
Theorem 0.1. For each j E Q(j =1= 0, 1728), the set of K-isomorphism
classes of (+l)-minimal Q-curves of degree one over K with j-invariant
j is in one to one correspondence with QX 18K where 8K := (Kx)2 n
QX. Moreover, to each class of 0 E QX 18K one can assign a Q-
isogeny class of abelian surfaces of fake GL2-type over Q represented
by E Tate (j)6 X ETate(j)m6.
We shall show that abelian surfaces of fake GL2-type over Q occur
as jacobian surfaces of algebraic curves, and they play imortant roles in
studying abelian surfaces A/Q of GL2-type over Q such that EndQ(A)
Q~Q(A).
The problem of (-1 )-minimal Q-curves over K is more difficult, and
to classify them one needs to study the quartic twists of elliptic curves
over Q. We note that if E is a (-1 )-minimal Q-curve of degree 1 over
42 K.-I. HASHIMOTO
K then ResK/Q(E) is an abelian surface over Q of GL2-type such that
(c.f. [Ri3]).
Let C(K) be the set of quartic cyclic extensions LIQ which contain K.
We shall prove
Theorem 0.2. Suppose that j E Q and that K := Q( vii -
17281) is
a real quadratic field with C(K) "# 0. Then for each 8 E KX satisfying
NK/Q(8) = c21j - 17281 (c E QX), the twisted curve E Tate (j)6 is a
( -1) -minimal Q-curve over K, so that
In particular, Res K / Q (ETate(j)6) is an abelian surface over Q of GL2-

type. Moreover, the map
E Tate (j)6 f-+ NK/Q(8) SK
induces a bijection from the set of K -isomorphism classes of (-1)-

minimal Q-curves of degree lover K to QX ISK.
As for the second and third questions listed above, our strategy is,
as in [Hal, to construct a generic family of genus two curves which are
double covers of elliptic curves with j-invariants j. For this purpose, we
study in full detail the action of involutive automorphisms on the space
of sextic polynomials. By this result we obtain the most general family
of split curves of genus two, from which we are able to find the one
parameter family C(j) which is defined by the equation y2 = F(Xij),
F(Xjj) := X 6 + 144X5 - (j -1728)X4 + 224(j -1728)X3

- (j - 1728)2X2 + 144(j - 1728)2 X + (j - 1728)3.
One of our main results, which has several interesting applications, is
stated as:
Theorem O.S. There exists a morphism 1/J : C(j) -+ E(j) of degree
two which is defined over Q( Jj - 1728), where E(j) is a twist of the
Tate's generic family of elliptic curves and is defined by
E(j): 3W2 = 4(j - 1728)U3 - 3jU + j, j(E(j)) = j.
We can show in particular that for all known cases of f E 82 (r1 (N))
such that Kf = Q( R), the solutions to the above problem are in our
family C(j).
Q-curves with j E Q and jacobian surfaces of GL 2 -type 43
This paper is organized as follows: In Sect.!, we introduce some basic

definitions such as Q-curves of degree one over quadratic fields K, their
minimality, and the group SK which plays an important role in our
theory. In Sect.2 we discuss some general theory of quadratic extensions
K (V6) of K which are cyclic quartic fields over Q. To each K (V6) we
associate a quartic twists of E(j), which will be shown to be a (-1)-
minimal Q-curves over K. We also discuss the other case where K ( V6)
is a (2, 2)-extension of Q, and associate to them the twists of E(j)j K, to
obtain abelian surfaces over Q of fake GL2-type. In Sect.3, we describe
one of our main results which gives a family C(j) of genus two curves that
are the double covers of elliptic curves E(j), and their jacobian varieties
are of fake GL2-type. The construction of such family is described in
the final Sect.6. In SectA we shall give concrete examples of C(j) with
quadratic twists whose jacobians are isogenous over cyclic quartic fields
to Shimura's abelian surfaces A f attached to cusp forms f of quadratic
Neben type character of level N = 37,65,104,157,397, and 877. In
Sect.5, we shall give another example of curve of genus two over Q
whose jacobian is isogenous over Q to A f for N = 65.
1. Minimal Q-curves of degree 1 over quadratic

fields
Let K = Q( rm) be a quadratic field and let j E K be such that
j =1= 0,12 3 . As in [HaJ we denote by K(j) the set of K-isomorphism
classes of elliptic curves E over K with j(E) = j. Then for j =1= 0,12 3 ,
the set K(j) is seen to be a homogeneous space over KX, where the
action of K X is given by the quadratic twist. Namely each 6 E K X
defines a bijection E I-t E V6 := E8 from K(j) to itself. Here if E is
defined by a Weierstrass equation
E: y2 = X 3 + AX + B (A,B E K),
then E8 can be defined by
E8: 8y2 = X 3 + AX + B. (1)
In what follows, we always assume that our elliptic curves are defined
over a fixed quadratic field K = Q( rm), and have no complex multipli-
cation. In such a case we have
Recall that a Q-curve E over K is an elliptic curve defined over K such

that there exists an isogeny flu : E - u E, wher a denotes the nontrivial
44 K.-l. HASHIMOTO
automorphism of KjQ. It is called of degree d, if d is the minimum of

deg(J.lu). Then d is a positive square free integer.
Definition 1. A Q-curve E over a quadratic field K is called minimal if
J.lu is also defined over K. It is called e-minimal ifu J.lu0J.lu = edxid. (e =
1).
Our aim is to study and classify all e-minimal Q-curves over K of
degree 1. This means that J.lu is an isomorphism, so that E and uE
have the same j-invariant, i.e. j(E) E Q. For such j, we choose as a
standard model of each class in &K(j) the following curve ETate(j) which
is defined over Q(j) = Q :
.,.,
E,ate(J'): y
2
+ Xy = X 3 - 36 X 1 (2)
j - 1728 - j - 1728
One has
j(ETate(j)) = j,
for the j-invariant and the discriminant of ETate(j). By a simple change

of coordinates X, Y, one sees that ETate(j) is isomorphic over Q to the
following model:
6(j - 1728)y2 = 4(j - 1728)X3 - 3jX + j. (3)
ETate (j) is obviously a (+ 1)-minimal Q-curve over K, with J.lu = id. For
8E K X put Eo := ETate(j) --18. We have an isomorphism
(x, y) I--t (x, y/V8),

which is defined over K( --18). By our assumption that Eo has no complex
multiplication, we see that the K-isomorphisms Eo ---t u Eo are given by
Hence Eo j K is a minimal Q-curve if and only if --18 j..;uJ E K X, or

equivalently
NK/Q(8) = !t5' (u8)2 E (KX)2.
We introduce a subgroup SK of QX defined as
SK := (Kx)2 n QX. (4)
From K = Q(v'm) we have SK = ((QX)2,m) = (Qx)2 Um(QX)2. The
above argument gives the following
Lemma 2. For 0 E KX, the curve Eo := ETate(j) -18 is a minimal

Q-curve over K if and only ifNK/Q(o) E SK.
To study the sign of the minimal Q-curve Eo, and the structure of the
associated abelian surface ResK/Q(E(j)o), we distinguish the two cases
according as 0 E K x \ Q x or 0 E Q x, each of which will be studied in
the followings sections.
2. Sign change of Tate curves by quartic twists

Here we study the case 0 E K x \ QX. We first recall some elementary
facts in Galois theory. For a quadratic field K, we denote by C(K) the
set of quartic cyclic extensions L/Q containing K.
Lemma 3. (Serre [Ser, Theorem(1.2.1)]) Let K = Q( vm)
be a quad-
ratic field. Then the quadratic extension K( -18), (0 E KX) of K belongs
to C(K) if and only iJNK/Q(O) = c2m for some c E QX.
In particular, we have C(K) ::j:. 0 if and only if m E NK/Q(K X), or
equivalently -1 E NK/Q(K X). It follows that K must be real quadratic.
Moreover, if L = K( -18) E C(K) then a : -18 1-+ vm/-18 gives a gener-
ator of the Galois group Gal(L/Q). Note also that one has &2 (-18) =
--18.
Suppose that K = Q( vm)
satisfies the above condition, and choose
00 E K X with NK/Q(oo) = m. Then the set C(K) is parametrized as
Corollary 4. For any c E QX the field Lc := K( #0) belongs to C(K),
and C 1-+ Lc induces a bijection
QX/SK ~ C(K).
Proof. Since NK/Q(coo) = c2m we see from Lemma 3 that Lc :=
K( #0) belongs to C(K). Conversely, for any member L E C(K) we
can find as above 0 E KX so that L = K(V6) and NK/Q(O) = c2m,
with c E QX. Then we see that a := ocoo-1 satisfies NK/Q(a) = 1,
hence by Hilbert's theorem 90, a = u({3)/{3 = NK/Q({3)/{32, so that
0= cOoNK/Q({3){3-2. It follows that
L = K(V6) = K( y'CIOO) ,
Finally we observe that
LCl = LC2 <===> CIC2- 1 E QX n (KX)2 = SK.
This completes the proof. 0
46 K.-l. HASHIMOTO
We now recall an argumfmt of [HaJ on the sign change of minimal

Q-curves. Let E be a minimal Q-curve over a real quadratic field K =
Q( rm), with an isogeny J-i,: E -> u E satisfying UJ-i,u0J-i,u = Ed, where
E = lj deg(J-i,u) = d (d > 0) is a positive square free integer, and a is
the non-trivial automorphism of K/Q.
We as::'l.Ime that C(K) 1- 0, and consider the isomorphism
y
v: E -> Eo (x, y) f-+ (x, ../6),
over L, where Eo = E 0 V8 / K denotes a quadratic twist of E / K. Define

J-i,; := u v0J-i,uov-1. Then J-i,; is an isogeny Eo -> u Eo which is defined
over K. We have the following commutative diagram
(x, y) E E = ETate(j) --+
1.1
Eo 3 (x,y/../6) = (x*,y*)
Jl.O'=idJ lJl.~
(x, y) E E = ETate(j) Eo 3 (x,y/..fUJ) = (x*, ~y*).
It then follows that
u J-i,;0J-i,; = C,.2 110U J-i,u ou v-I) (17 VOJ-i,u ov- I )

= u2VO (uJ-i,u0J-i,u) ov- I
= Ed (U 2voV -I)
= -cd xid.
This proves the following

Proposition 5. ([Hal) Assumption being as above, Eo/ K is again a
minimal Q-curve over K with an isogeny J-i,; : Eo -> u Eo defined over
K satisfying u J-i,;0/1; = -Ed xid.
We shall now apply this to E(j) with d = 1.
Theorem 2.1. Suppose that j E Q and that K := Q( vlj - 17281) is
a real quadratic field with C(K) 1- 0. Then for each 8 E K X satisfying
NK/Q(8) = c2 1j -17281 (c E QX), the twisted curve E(j)o is a (-1)-
minimal Q-curve over K, so that
(5)
In particular, ResK/Q(E(j)o) is an abelian surface over Q ofGL2-type.

The last assertion (5) follows from a result of Ribet [Ri3J.
Q-curues with j E Q and jacobian surfaces of GL2-type 47
Next we study the case 8 E QX. If 8 E Q x n(Kx)2 = SK then E6 is K-

isomorphic to E = ETate(j). So assume that 8 E QX\SK. In such a case
E6 is again defined over Q, and the isomorphism 11 : E = ETate(j) --+ E6
is defined over a quadratic field Q( V&) different from K. It follows that
K( V&)/Q is a (2, 2)-extension and one can take if E Gal(K( V&)/Q such
that iflK = u. Then one sees from the diagram below that E6 is a
(+l)-minimal Q-curve over K.
E = ETate(j) ~
::;L(j) ~
We note here that the two curves E6, Em6, which are not isomorphic over
Q, become isomorphic over K. Therefore it is meaningful to consider
an abelian surface E6XEm6 as a model over Q of ResK/Q(E(j)6).
Definition 6. An abelian surface A/Q is called of fake GL2-type over
Q, if the Q-algebra of endomorphisms of A defined over Q is
EndQ(A) Q ~ Q $ Q,
while EndK(A) Q ~ M2(Q) for a quadratic field K.
Thus an abelian surface AIQ is of fake GL2-type over Q if it is Q-
isogenous to a product El x E2 of elliptic curves over Q which are K-
isogenous but not Q-isogenous to each other. We shall show in the
next paragraph that abelian surfaces of fake GL2-type over Q occur
as jacobian sufaces of algebraic curves, and they play imortant roles in
studying abelian surfaces AI Q of GL2-type over Q such that EndQ (A)
Q~Q(A).
3. Genus two curves with jacobian of fake

GL2 -type
Let C(j)/Q(j) be a family of curves of genus two defined by the
equation y2 = F(Xjj),
F(Xjj) := X 6 + 144X5 - (j -1728)X4 + 224(j -1728) y3 (6)
- (j - 1728)2 X2 + 144(j - 1728)2 X + (j - 1728)3.
We shall describe the construction of such family in the final section.
Our first result is stated as follows:
Theorem 3.1. There exists a morphism 'I/J : C(j) --+ E(j) of degree two
which is defined over Q( Jj - 1728), where E(j) is a twist of ETate(j)
48 K.-I. HASHIMOTO
corresponding to the quadratic extension Q( J2(j - 1728)) of Q(j) and

is defined by
E(j): 3W 2 = 4(j - 1728)U3 - 3jU + j. (7)
Proof. We put, for simplicity, m := j - 1728. The curve E(j) is easily
seen to be Q(j)-isomorphic to a quadratic twist ETate(jhm of ETate(j).
We observe that a covering map 'lj; : C(j) ~ E(j), which is defined over
Q( y'm), is given by
'lj;*(U) = (y'm - 72)X2 + 2y'm( y'm + 24)X + m( y'm - 72)

2y'm(X + y'm)2 '
-48Y
'lj;*(W) = (X + y'm)3'
o
One can easily check the following identity for the pullback of invariant
i-forms:
'lj;*(~) = _2..;md: +2X~X. (8)
Let (J" be the generator of the Galois group of the quadratic extension
Q(y'm)/Q. Then 17'lj; is also a morphism of degree 2 from C(j) to E(j),
and we have
17'lj;*(~) = 2..;md: + 2X~X. (9)
Since 'lj;*eii{) and 17'lj;*cii{) are linearly independent, the morphism
q> := ('lj;, u'lj;) : C(j) ~ E(j)xU EU), P f-+ ('lj;(P) , u'lj;(P))
induces an isogeny
q> : Jac C(j) ~ E(j) x UE(j) = E(j) x E(j),
which is defined over Q( v'rri). Thus we have seen that E(j) is a (+)-
minimal Q-curve of degree one over Q( v'rri), the isogeny J.L17 : EU) ~
17 E(j) = E(j) being the identity map. It follows that all endomorphisms
of Jac C(j) are defined over Q( y'm), and in fact the Q-algebra of endo-
morphisms of Jac C(j) defined over Q( y'm) is
(10)
Theorem 3.2. The jacobwn variety Jac C(j) is of fake GL2-type over
Q(j). Namely the Q-algebra of endomorphisms ofJacC(j) defined over
Q(j) is described as
EndQ(j)(Jac CU)) Q ~ Q EEl Q. (11)
Proof. The assertion means that Jac C(j) is isogenous over Q(j) to a
product of two elliptic curves which are not Q(j)-isogenous but are
Q( Jrn)-isogenous to each other. This indicates the existence of non-
trivial morphisms
f : G(j) -+ E(j),
which are defined over Q(j), where E(j)m denotes the quadratic twists
of E(j) by Jrn. In fact f is given by f(P) = 'IjJ(P) EeE(j) CT'IjJ(P). A
computation shows that I is written explicitly as f((X, Y)) = (U, W),
with
'*(U) = 3m2 + 6mX2 - 192X3 - X4
j (3m + X2)2
r(W) = 6(m 2 -72mX - mX2 + 8X3)y
(3m+X2)3
Similarly the morphism 1m : G(j) -+ E(j)m is given by
Im(P) = 'ljJm(P) EeE(j)m CT'ljJm(P),
'ljJm(P) := (U, W/v'ffi) , 'IjJ(P) = (U, W).
Then after some computation one obtains fm((X, Y)) = (U, W), with
r (U) = _m2 - 192mX + 6mX2 + 3X4
m (m+3X2)2
r (W) = 6(Sm - mX - 72X2 + X3)y
m (m+ 3X2)3
One can now easily check the equalities
r(dU) = 4 XdX
W Y
f m* (dU)
W
= -4 dX
my.
Since r(~) and f~(~) are linearly independent, it follows that the
morphism (/,Im), which is defined over Q, induces an isogeny from
JacG(j) to E(j)xE(j)m. 0
Theorem 3.2 can be proved also by the following argument. Using
(10) we identify EndQ(v'ffl)(Jac G(j)) Q with M2(Q). Then, from
004.> = (oo'IjJ, 0") we see that the action of u on EndQ(vm) (Jac G(j))Q cor-
responds to the inner automorphism Int(J), J = (~ ~) on M2(Q).
50 K.-J. HASHIMOTO
It follows that EndQ(j)(JacC(j)) Q9 Q corresponds to the subalgebra

consisting of ME M 2 (Q) such that MJ = JM. Hence we have
EndQ(j) (Jac C(j 0 Q !~:q (~ !) I a, bE Q} ~ Q EB Q.

Remark 7. It is interesting to observe that the sextic polynomial
F(X;j) in (6) is irreducible over Q(j), and its splitting field is a regu-
lar extension of Q whose Galois group over Q(j) is isomorphic to 84,
the symmetric group of degree 4. This fact can be proved by computing
several resolvent polynomials in the roots of F(X;j).
4. Shimura's abelian surfaces

Let N the discriminant of a real quadratic field K = Q( .IN), and let
X(*) = (~) be the Dirichlet character associated to KIQ. Let ro(N)
be the modular group
and denote by S2(N, X) the space of cusp forms of weight 2 and Neben
type character X( *). We assume that there exists a Hecke eigen form
f E S2(N, X) such that the field K f = Q( {an}) generated by the Fourier
coefficients of f is an (imaginary) quadratic field. Then A f is decom-
posed over K into the product
(isogenous over K = Q(.JN) ),
where B I I K is called the Shimura's elliptic curve. It has the following

remarkable properties:
(i) BilK has good reduction at all finite primes of K.
(ii) BIlK is a minimal Q-curve, and its degree (= Min deg(p".) is
the square free part of D(KIIQ) :
p".IK : BI ~ "'BI'
We assume that K I = Q( R). This condition is equivalent to

j (BI) E Q. Up to isogeny, six such B I of non eM type are known; their
levels are N = 37, 65,104,157,397,877, and the dimensions of the corre-
sponding spaces S2(N, X) are 2,4,12,12,32,72, respectively. The defin-
ing equations for B I I K were studied by Shiota [Shi] , Pinch [Pi], and Cre-
mona [Cr]. On the other hand, elliptic curves over real quadratic fileds
with everywhere good reduction has been studied by Comalada [Co],

Pinch [Pi], Cremona [Cr], Kagawa [Ka], Kida [Ki] and others. The K-
isomorphism classes of known such curves over our K = Q (../N) are as
in the following table (c.f. [Co], [Cr) , [KiD.
Table 1a
N I Name I j E(j)6 I K-isogeny I
37 37A 2""' 537 A = 37 + 6v'37 5:A-B
37 37B 21~2113 537B = -29139211(37 + 6v37) 5:B-A
65 65A 173 56SA = -717(65 + v65) 2:A-G
65 65B 173 565B = -7-17(65 + 7v'65) 2:B-D
65 65G 257" 065C = -2773257(65 + 8v'65) 2:G-A
65 65D 2573 065D = -773257(1105 + 137v65) 2:D-B
877 877A 2""'.73 0877A = 57(8149877 + 241326v'877)
104 104A 2 0104A = -2(26 + 5v'26) 5:A-G
104 104B 2 5104B = -(26 + 5y'26) 5:B-D
104 104C _2ti 7193 5104C = 219199719(26 + 5v26) 5:C-A
104 104D -2719" 0104D = 19199719(26 + 5v'26) 5:D-B
157 157A -133 0157A = 513(17157 + 213v'157)
397 397A 11 3 0397A = -11(173397 + 3447v397)
The columns after N and the names are their j-invariants and the
models over K which are obtained by twisting E(j) with v'J. It turns
out that each of them is isomorphic to Shimura's elliptic curve B J for
some f E S2(N, X). Also it is important to observe that their j-invariants
are all perfect cubes, and that the real quadratic fields Q(../N) coincide
with Q( vlj - 123 1), as indicated in the following table:
Table Ib
I Name I j
37A 2u 2"37 37 -..
37B 2""'.211 3 229"139"37 29- 0.139 -0.211.37-"
65A 173 57"13 20.5 -3.7- 0.13- 3.17"_
65B 17" 57"13 20.5 -3.7- 0.13- 3.17"
65G 257" 57"1373" 2.5 '''7 '13 -"73 -0257"
65D 257" 57"1373" 25-"7 -013 -"73 -025~
877A 2""'7" 25"877 5 -0.7" .877-"
104A 2" -2 13 -2 .. 13-"
104B 2" -2 13 -2 .. 13-..
104G -2"719" -2 1319"199" -2 '3.13- 3.19- 0 .199-".719"
104D -2"-719" -2 1319"199" -2 -"13-"19-"199-"-719"
157A -133 -5"157 -20.5-0.130.157 -3
397A 11" -397 -211397 -"
52 K.-J. HASHIMOTO
We shall now state our second main results Theorem 4.1, and Theo-
rem 4.2, which describe Shimura's abelian surfaces as jacobian of curves
belonging to the family CU) with quadratic twists.
Theorem 4.1. Let N = 37,65,877, and f E S2(N,X) be an eigenform
with Kf = Q( A). The'Y/, Af is isogenous over LN to the jacobian
of the following curve C(j) 0 Vr defined over Q, where LN is a cyclic
quartic field given as L N = Q (../ON) with
8N = 2( -37 + V37), 2(65 + )65), -2(877 + 29v'877)

respectively, which satisfies K = Q( v'N) ~ LN ~ Q( (N).
Table 2a
C(j)@vr C: ry2 = f(X)
37 C(212)@V-I _y2 = X 6 + 18Xs _ 37X4
+1036X 3 - 1369X 2 + 24642X + 50653
37 C(2122113) @ V29.139.211 29139211y2 = 4031X 6 + 18X 5 - 149147X 4
+1036X 3 - 5518439X 2 + 24642X + 204182243
Table 2b
N CU) @vr C: ry2 = f(X)
65 C(173) @ V-2717 -2717y2 _ (X2 + 2X + 65)
x (7 X4 + 130 X 3 - 1170 X2 + 8450 X + 29575)
-2773-257y2 = (X2 -16X + 65)
65 C(2573 ) Q9 V-2773257 x (511 X4 + 8320 X 3 + 66690 X2
+ 540800 X + 2158975)
877 C(2 12 73 ) Q9 V-57 -57y2 = 5X6 + 18X 5 - 4385X 4 + 24556X 3
-3845645 X2 + 13844322 X + 3372630665
Proof. The polynomials f(X) listed above are obtained from F(Xjj)
by a minor change of variable X over Q. Therefore the assertions are
immediate consequences of Theorem 3.1, except for the fields over which
the isogenies are defined. Denote the pure generator of LN over Q( v'N)
by JtJN. Then we see from the argument of Sect.2 that, to prove the
last assertion it suffices to check that, up to a square factor, the quotient
ONA/ON coincides with the twisting factor r of C(j). The quartic Dirich-
let character 1/J, which corresponds to the extension L N / Q is given as
Q-curves with j E Q and jacobian surfaces of GL 2 -type 53
follows. Note that 5 (resp. 2) is a primitive root modulo 37 (resp. 5, 13

and 877). Thus for N = 37,877 the pair {'lj;1} of Dirichlet characters
of order 4 defined mod N is unique, and we have
N = 37: 'lj;(P) = (R)k for p == 5k (mod 37)

N = 877: 'lj;(P) = (R)k for p == 2k (mod 877).
LN is then generated over Q by the so called Gaussian period. Namely
we have LN = Q(7]j) for
(N-l)/4 -1
"'" ;-g4i+j (0 ::; j ::; 3) (12)
7]j := L- ~
i=O
where 9 E ZjNZ is a fixed primitive root modulo N. The irreducible

equation G4(Xj N) over Q for 7]j is known (c.f. [H, pp.489-494]j see also
[BE]), and is easily computed for each N. Indeed we have
G4(Xj 37) = X4 + X 3 + 5X 2 + 7X + 49
G4(Xj877) = X4 + X 3 + 1l0X2 + 3234X + 16317,
from which we obtain
L37 = Q ( V2(-37 + V37)), L877 = Q ( V -2(877 + 29V877)) .

It follows that for N = 37 (resp. N = 877), LN = Q( J'IN) with 837 =
2V37(1 - V37) (resp 8877 = -2V877(29 + v877)), so that we observe
837 = -(7 - V37)2.(37 + 6V37)

= -(7 - V37)2 x 837A,
837B = -29139211 'x 837A,
and
385 + v877 2
8877A = -57( 12 ) x 8877.
Thus we see that the factors -1, 29139211, and -57 are exactly the
twisting factors r of C(212), C(212 .2U 3 ), C(2 12 73 ), respectively. Hence
we can conclude that AI is isogenous to JacC(j) .jT/Q over LN.
On the other hand, since (Zj65Z)X ~ (Zj5Z)X x (ZjI3Z)X we see
that there are six cyclic subgroups of (lrder 4 of the Dirichlet characters
defined mod 65. They are divided into three classes according as the
quadratic subfields corresponding to their subgroups of order 2, which
54 K.-J. HASHIMOTO
are Q( J5), Q( Vl3) and Q( V65). Among the two which correspond to
Q( V65), we choose the following one generated by :
N = 65: 1jJ(P) = (H)i- k for p == 2i (mod 5), p == 2k (mod 13).
We have then Ker(1jJ) = (2) (~ (Z/65Z)X) so that the corresponding
quartic field L65 is generated over Q by
12
17:= L (~.
i=1
The equation for 17 over Q is easily computed to be X4 - X3 - 24X2 +

4X + 16, and solving this we obtain
17 = ~ ((1 + v65) + V130 + 2v65) .

It follows that L65 = Q( v%5) with 665 = 2V65(1 + 2V65), and we have
865A = -717(65 + v'65) = -2.7.17.2- 2 X 865.
Thus we see that the factor -2717 coincides with the twisting factor r
of C(173 ). It follows that Af is isogenous to JacC(j) .jT/Q over LN
This completes the proof of Theorem 4.1. 0
The L-function of C(j)/Q or JacC(j)/Q is defined by
L(C(j)/Q, 8) = L(JacC(j)/Q, 8) := II L (C(j)/Q,p-s)-1,

p
p
where for good primes p the Euler p-factor is given by
Here Ap, Bp are real numbers which are determined by the congruence
zeta function of C(j)p := C(j) Fp as
Z(C(j)p/Fp,u):= exp(I: Nmu m)

m=1 m
(1 - Q pu)(1 - apu)(1 - ,8pu)(1 -73pu)
=
(1 - u)(1 - pu)
(1 - Apu + pu2)(1 - Bpu + pu2)
(13)
- (1 - u)(1 - pu)
so that Ap = ap + Qp, Bp = {3p + f3p. We note that ap, {3p are complex
numbers satisfying
apQp = {3pf3p = p, lapl = l{3pl = pl/2.
By the isogeny theorem of Faltings [Fa], the above results imply the
following equality for C := CU) Vr
L(C/Q, 8) = L(f 'I/J, s)L(f 'I/J, s)
= L((BJlk) 'I/J, 8)
= L((AJlQ) 'I/J, s),
which can also be expressed as
Ap(P) = aj(p), Bp(P) = a/(p). (14)
We tabulate some of these quantities for small p.
Table 3a
p a,(pj Ap, Bp 'I/J{p)( ~l) a,(p) Ap, Bp 'I/J(P)( -WS )

3 -1 -1,-1 +1 -2A 2,-2 A
11 -3 -3,-3 +1 -2A 2,-2 A
17 -2A 2,-2 A 0 *,* 0
19 -6A 6,-6 -A 6A 6,-6 A
23 -4A 4,-4 -y'-I -6y'-I 6,-6 -y'-I
29 4Fi' 4,-4 -Fi' 6 -6,-6 -1
31 0 0,0 Fi' -6Fi' 6,-6 -Fi'
37 -1-6A *,* * 6 -6,-6 -1
41 -3 3,3 -1 8A 8,-8 -A
43 6A 6,-6 -A 6A 6,-6 -A
47 3 -3,-3 -1 -8 -8,-8 +1
Table 3b
p a,(p) Ap, Bp 'I/J (p)( """!5 ) p a,(p) Ap, Bp 'I/J{p)( --:5)

3 1 -1,-1 -1 11 -4A 4,-4 A
13 0 0,0 A 17 -2A 2,-2 A
19 -6A 6,-6 A 23 0 0,0 -1
29 6 -6,-6 -1 31 -3 3,3 -1
37 3 -3,-3 -1 41 6 -6,-6 -1
43 1 -1,-1 -1 47 4A 4,-4 A
56 K.-J. HASHIMOTO
Table 4
N CU) Vr CU) Vr: ry2 = f(X)
104 C(26) J13 13y2 = X6 - 109 X 4 + 40 X 3
+ 3635X 2 - 408 X - 42047
13-19.199.719y2 = 3781 X 6 + 18X 5
104 C(-2 6 719 3) Jl319199719 +98306 X4 - 728 X 3 - 2555956 X2
+ 12168 X - 66454856
2513y2 = 5X 6 + 144X 5
157 C(-2197) J2513 +785X 4 - 35168X 3 -123245X 2
+ 3549456 X - 19349465
2.11y2 = X6 + 144X 5
397 C(113) JH1 +397 X4 - 88928 X 3 - 157609 X2
+ 22695696 X - 62570773
Theorem 4.2. Let N = 104,157,397, and f E S2(N, X) be an eigenform

with Kf = Q( A). Then Af is isogenous over Q( A)LN to the
jacobian of the following curve C(j) Vr defined over Q, where LN is
a cyclic quartic field given as L N = Q (J(;N) with
ON = -(26 + 5V26), 2(-157 + 11 V157), -2(397 + 19V397)
respectively, which satisfies K = Q(..fFi) ~ LN ~ Q( (2N).
Proof. As in Theorem 4.1, the polynomials f(X) listed above are ob-
tained from F(Xjj) by changing variable X linearly over Q. Therefore
it suffices to check the assertions for the fields over which the isogenies
are defined. This is again equivalent to show that up to a square factor,
the quotient of two 0' s ON AION coincides with the twisting factor r of
C(j).
First note that the quartic Dirichlet character 1jJ, which corresponds
to the extension LN /Q is given as follows. For N = 157,397 the pair
{ 1jJ1} of Dirichlet characters of order 4 defined mod N is unique, and
using 20 as a primitive root modulo N we have
N = 157, 397: 1jJ(p) = (A)k for p == 20k (mod N).
LN is generated over Q by the Gaussian period TJj (see (12)) whose
irreducible equation G4(Xj N) over Q is
G4(Xj 157) = X4 + X 3 + 20X 2 - 206X + 517
G4(Xj 397) = X4 + X 3 + 50X 2 - 918X + 3069.
From this we obtain LN = Q(../lN) with
8157 = 2(-157 + UV157), 8397 = -2(397 + 19V397).

It follows that for N = 157 (resp. N = 397), we have
and
2
( 20+ V397 )
8397A = 211 6 x 8397A.
This proves the assertion for N = 157,397.

Finally suppose that N = 104 = 813. Then there is no Dirichlet
character 1/J defined modulo N such that 1/J2 corresponds to Q( VN).
Hence we have to raise the level to 2N = 208:
(Z/208Z)X ~ (Z/16Z)X x(Z/13Z)X ~ {1}x(5)x(2),
on which there are 8 candidates for 1/J satisfying the above condition.
They are determined by the values at -1,5, 2:
We choose 1/J for which (c1,c2,c3) = (1,1, -1). Then we have Ker(1/J) =
(5,23) c (Zj208Z)X so that the corresponding quartic field L208 is gen-
erated over Q by
3 5
17 := LL (5i 23J
i=O j=O
It is not difficult to see that the equation for 17 over Q is X 4 + 52X 2 + 26,
and solving this we obtain 17 = J
-26 - 5V26. It follows that L104 =
Q( J(;'iQ4) with 8104 = -26 - 5V26, and we have
8104A(V26)2 = -413(26 + 5V26) = -13.22 X 826,
8 104C(V26)2 = -13.19.199.719.22 X 826.
Thus we see that the factor -13, -1319199719 coincide with the twist-
ing factors r of C(26) C( -26 7193). It follows that AI is isogenous to
JacC(j) ..;r/Q over LN. This completes the proof of Theorem 4.1. 0
58 K.-I. HASHIMOTO
The assertion of Theorem 4.2 is slightly weaker then Theorem 4.1; in

fact it is equivalent to the following equality (0 = CU) JT)
L(C/Q(V-I),s) = L(f0'ljJ,s)L(/0'ljJ,s)
-1 - -1
x L(f 'ljJ (-), s)L(f 0 'ljJ 0 (-), s)
* *-1
= L((BJlk) 1/;, s)L((BJlk) 1/; (-), s)
*-1
= L((Af/Q) 1/;, s)L((At/Q) 1/; (-), s)
*
which can also be expressed as
We tabulate the quantities {Ap, B p, af(p)} for small p.
Table 5a
P af(p) Ap, Bp 1/J(P)( ~) p af(p) Ap, Bp 1/J(P)(~)

3 2 2,-2 -1 5 -2A 2,-2 A
7 -3A -3,-3 -A 11 0 *,* 0
13 0 0,0 --A 17 4A 4,-4 A
19 -2 2,-2 -1 23 6 6,-6 1
29 -3 -3,-3 1 31 -4 4,-4 -1
37 -9 9,9 -1 41 8A 8,-8 A
43 0 0,0 1 47 6 6,-6 -1
Table 5b
P a/(p) Ap, Bp 'IjJ(P)(i> a/(p) Ap, Bp 'IjJ(P)(~)

3 -A -1,-1 +1 -2 2,-2 -1
7 -3A -3,-3 A -3A -3,-3 A
11 2 2,-2 -1 6 6,-6 +1
17 3 -3,-3 -1 3 -3,-3 -1
19
23 -6 0,0
6,-6
-1
+1
-5A
0,0
5,5 -A
+1
29 6A 6,-6 A 4A 4,-4 A
31
37
3
0,0
-3,-3
A
-1
-3
0,0
3,3
+1
-1
41 lOA 10, -10 -A 2A 2,-2 A
43 9A 9,9 A 9A -9,-9 -A
47 7A -7,-7 -A 6 6,-6 +1
5. Another example for N = 65

In general the field over which the isogeny JacC(j) ~ AJ is defined
is bigger than Q when the Neben type character X of f is nontrivial.
Indeed the assertion is equivalent from the isogeny theorem of Faltings
[Fa] to the equality of their L-functions, i.e.
(1 - Apu + pu 2 )(1 - Bpu + pu2) =

(1 - aj(p)u + X(p)pu2)(1- bf(P)u + X(p)pu2)
for almost all primes p, where Ap, Bp are real numbers as defined in (13).
In very rare cases, however, it happens that the isogeny can be. defined
over smaller field. In this paragraph we shall present an example of a
curve C /Q such that JacC is isogenous to Aj over Q, for N = 65.
Theorem 5.1. Let C65/Q be a curve of genus two defined over Q by
the equation
C65: y2 = -5X(X + 13)(5X + 1)(5X2 + 2X + 13). (16)
Then C6 5 has an involution defined over the quadratic field K := Q( V65)
13 13V65)
p: (X, Y) 1-+ (X + 5X' (1 + 25X3 )Y
and the quotient curve C65 / (p) is an elliptic curve over K defined by
E65: W2 = U(U - 65)(U + 2 - 2V65). (17)
60 K.-J. HASHIMOTO
Furthermore, E6S/k is a (-I)-minimal Q-curve with j(E6S) = 173 and

has a trivial conductor over K. Finally J ace is is0genous to A f' f E
8 2 (65, x) over Q.
Proof. Putting
13
U:=X+ 5X' W:= ( 1 + 13V65)
25X3 Y,
we see in the same way as in Theorem 3.1 that the equation (16) be-
comes (17). Then we transform (17) into Weierstrass equation W2 =
4U 3 - g2U - g3, with
544(33 - V65) 1792(1105 - 49V65)

92 = 3 ' g3 = 27 (18)
Now we observe that for>. := (1 + V65)/8 we have

(192 = 33 + v'65 = ).4 (193 = 1105 + 49V65 = >.6.
92 33 - v'65 ' 93 1105 - 49V65
Hence we obtain an isomorphism over K
Moreover, we have
o
To our best knowledge, the curve C65 /Q is the unique example for
which there exists an isogeny over Q from JacC to A f with f E
82(fo(N), X) an eigenform with nontrivial Neben type character X.
6. Construction of C(j)/Q(j)
Here we shall describe the construction of the family {C(j)} given
in (6). We start from the following well known
Lemma 8. For a curve C / K of genus two to form a double cover
of some elliptic curve, it is necessary and sufficient that C has non
hyperelliptic involution.
The curve C can be defined by an equation y2 = F(X) with a separa-
ble polynomial F(X) E K[X] of degree 5 or 6. Then a non hyperelliptic
Q-curues with j E Q and jacobian surfaces of GL2-type 61
involution p of C induces an involutive automorphism p of the rational

function field K (X). Since
_p = ( a b
d): X ~ _p(X) = aX+b ,
c cX+d
we can write
p: K(X) -+ K(X), -(X) = aX +b.

p eX-a
with a, b, e E K, m:= a2 + be =f O. It is easy to see that p can be

extended to an automorphism of C of order 2 (resp. order 4) if and only
if
(m = a 2 + be). (19)
with e = +1 (resp. e = -1). To obtain such F(X) is nothing but

to obtain the eigen vectors of ws(p), where Ws is the symmetric tensor
representation of degree 6, on the space of binary sextic forms.
Lemma 9. The characteristic polynomial of ws(p) is (X - m 3 )4(X +

m 3 ). Moreover, the polynomials {Fj+)(X) I 1 :$ j :$ 4} (resp.
{Fj-)(X) I 1 :$ j :$ 3}) tabulated below in Table 6 (resp. Table 7),
form a basis of the eigenspace corresponding to the eigenvalue m 3 (resp.
_m 3 ).
Table 6
Fl+) (X) (b 2 + 2abX + 2a 2x 2 + bcX2) X (b4 + 4ab3X + 5a 2b2x 2 _ b3c)(2
+2a3bX 3 - 2ab 2cX 3 + a4X4 + a2bcX4 + b2c2X4)
FJ+l(X) -ab5 _ + b5 cX _ lOa3b3X2 + 5ab4c)(2 _ lOa4b2 X 3
5a2b4X
+lOa2b3cX3 _ 5a 5 bX4 + lOa3b2cX4 + 8a4bcX 5 + 3a2b2~ X 5
+ b3c3X 5 + a5 cx6
FJ+) (X) a 2b4 + 4a 3b3X _ 2ab4c)( + 6a 4b2X2 _ 8a2b3cX 2 + b4c2 X 2
+4a5 bX 3 _ 12a3b2 cX 3 + 4ab3c2X 3 + 2a6 X 4 _ 5a4bcX4
+9a2 b2c2X4 + b3c3X 4 _ 2a5 cX 5 + 4a3~ X 5 + a4~ X 6
Fl+ ' (x) (-ab+ 2bcX + acX2) x (a 2 b2 + 3a3bX - ab 2c)(
+3a4 X2 _ 2a 2 bcX2 + b2c2 X 2 _ 3a3cX3 + abc2x 3 + a2 c2X4)
62 K.-I. HASHIMOTO
1able 7
Fi-l(x) (-b - 2aX + CX2) X(b 4 + 4ab3 X + 7a 2b2X2 + b3cX 2
+6a3bX 3 + 2ab 2cX 3 + 3a4X4 + 3a 2bcX4 + b2c2X4)
FJ-l(X) (-b - 2aX + CX2)X( -ab4) - 3a 2 b3X + b4 cX - 4a 3b2 X 2
+2ab3cX 2 _. 2a 4bX 3 + 3a 2 b2 cX3 + b3 c2 X 3 _ as X4)
FJ-l(X) (-b - 2aX + CX2) x (a 2b3 + 2a 3b2X - 2ab 3cX
+2a4 bX2 _ 3a2b2cX2 + b3c2X2 _ 4a3bcX 3 - a4 cX4)
From the above result, we obtain a family of genus two curves with 7
free parameters a, b, e, p, q, r, s, which covers elliptic curves by degree
two.
C: y2 = pFf+) (X) + qFJ+) (X) + rFJ+) (X) + sFJ+)(X). (20)
Furthermore, we see from (19) that the non hyperelliptic involution p of

C is given by
i: (X, Y) f---+
aX +b mymy) '
( eX _ a' (eX _ a)3 (m = a2 + be).
It follows that the quotient curve C / (i) is of genus one, and has a model
over the quadratic extension Q(a, b, e, p, q, r, s, ym) of
Q(a, b, e, p, q, r, s), which is defined by the following equation:
C / (i): w2 = (-2a + 2v1m + eU) x G(U), (21)
G(U) = (2a 3b3p - 6ab4 cp + 6a2 b3eq - 2b4 e2q - 2a3b2cr + 6ab3e2r
- 6a 2 b2 e2 s + 2b3e3s) + (6a 4 b2 p - 9a 2 b3ep - 3b4 e2 p + 13a3b2 eq
+ ab3e2 q - 4a4 ber + 9a2 b2 e2 r + b3e3r - 9a3bc2 s + 3ab2 e3s)U
+ (6a 5 bp + 8a4 beq + 3a2 b2 e2 q + b3e3q - 2a5 cr + 4a 3bc2 r
- 3a4 e2 s + 3a2 be3s)U 2 + (2a 6 p + 3a4 bcp + 3a2 b2 e2 p
+ b3e3p + a5 eq + a4 e2 r + a3e3s)U 3
We note that the RHS cf the above equation is linear in p, q, r, s. The
j-invariant of C / (/,) is written in the form
j(C/(i)) = jo(a, b, e,p, q, r, s) + vlmjl(a, b, e,p, q, r, s),
with jO,jl E Q(a,b,c,p,q,r,s). Now we look for a specialization of our
free parameters a, b, e, p, q, r, s which satisfies jl(a,b,e,p,q,r,s) =
Q-curues with j E Q and jacobian surfaces of GL 2 -type 63
O. Although to solve this question in its full generality is difficult , it

is rather easy to find a sufficient condition, since the number of the
parameter is big enough.
Proposition 10. The equation jl (a, b, c, p, q, r, s) = 0 is satisfied by
a = 0, r = -bp/c, s = 7bq/(9c). Then F(X) in (20) becomes
F(X) = 9b3p + 9b2cqX - 9b2cpX2 + 14bc2qX3 - 9bc2pX4 (22)

+ 9c3 qX5 + 9c3 pXB
and the j-invariant of C / (t) is given by
j(C/(t) = 1728(12~ +cq2) E Q(b,c,p,q). (23)

cq
Making the transformation
20736b qX
C f-+ (j _ 1728)q2 ' X f-+ 144p
we now obtain the family C(j) and covering map of degree 2: C(j) -+
E(j).
References
[BCDT] C. Breuil, B. Conrad, F. Diamond and R. Taylor, On the modularlityof
elliptic curves over Q, J. Amer. Math. Soc. 14 (2001), 843-939.
[BE] B. C. Berndt and R. J. Evans, The Determination of Gauss Sums, Bulletin
Amer. Math. Soc. 5 (1981), 107-129.
[Co] S. Comalada, Elliptic curves with trivial conductor over quadratic fields,
Pacific J. Math. 144 (1990), 233-258.
[Cr] J. E. Cremona, Modular symbols for rl(N) and elliptic curves with every-
where good reduction, Math. Proc. Camb. Phil. Soc. 111 (1992), 199-218.
[Fa] G. Faltings, Endlichkeitssiitze fUr abelsche Varietiiten ii.ber Zahlkorpem,
Invent. Math. 13 (1983), 349-366; Erratum: 15 (1984), 381.
[HsJ Y. Hasegawa, Q-curves over quadratic fields, Manuscripta Math. 94 (1997),
347-364.
[HHM] Y. Hasegawa, K.Hashimoto and F.Momose, Modularity conjecture for Q-
curves and QM-curves, International J. Math. 10 (1999), 1011-1036.
[HaJ K. Hashimoto, Q-curves of degree 5 and jacobian surfaces of GL2-type,
Manuscripta Math. 98 (1999), 165-182.
[H] H. Hasse, "Vorlesungen iiber Zahlentheorie, zweite Auf.," Springer-Verlag,
Berlin, 1964.
[Ig] J. Igusa, Arithmetic variety of moduli for gunus two, Ann. of Math. 12
(1960),612-649.
64 K.-J. HASHIMOTO
[Ka] T. Kagawa, Elliptic curves with everywhere good reduction over quadmtic
fields, dissertation, Waseda Univ., 1998.
[Ki] M. Kida, Computing Elliptic curves having good reduction everywhere over
quadmtic fields, preprint (1999).
[Pi] R. G. E. Pinch, Elliptic curves with good reduction away from 3, Math.
Proc. Camb. Phil. Soc. 101 (1987), 451-459.
[Py] E. Pyle, Abelian varieties over Q with large endomorphism algebms and
their simple components over Q, dissertation, Univ. of California at Berke-
ley, 1995.
[Ril] K. Ribet, Galois representations attached to eigenforms with Nebentypus,
in "Modular Functions of One Variable V," 17-52, Lecture Notes in Math-
ematics, 601. Springer-Verlag, Berlin/Heidelberg/New York, 1977.
[Ri2] K. Ribet, Endomorphism algebms of abelian varieties attached to newfarms
of weight 2, in "Seminar on Number Theory, Paris 1979-80," 263-276,
Progress in Math., 12. Birkhiiuser, Boston, Mass., 1981.
[Ri3] K. Ribet, Abelian varieties over Q and modular forms, in "1992 Proceed-
ings of KAIST Mathematics Workshop," Korea Advanced Institute of Sci-
ence and Technology, Taejon, 1992, 53-79.
[Ser] J.-P. Serre, "Topics in Galois Theory," Research Notes in Mathematics, 1.
Jones and Bartlett Pub!., 1992.
[Se] B. Setzer, Elliptic curves with good reduction everywhere over quadmtic
fields and having mtional j-invariant, Illinois J. Math. 25 (1981), no 2,
233-245.
[Shl] G. Shimura, "Introduction to the arithmetic theory of automorphic func-
tions," Publications of the Mathematical Society of Japan, 11. Iwanami
Shoten, Publishers, Tokyo, 1971.
lSi] J. H. Silverman, "The Arithmetic of Elliptic Curves," Graduate Texts in
Mathematics, 106. Springer-Verlag, 1986.
[Shi] K. Shiota, On explicit models of Shimum's elliptic curves, J. Math. Soc.
Japan 38 (1986),649-659.
[Wi] A. Wiles, Modular elliptic curves and Fermat's Last Theorem, Ann. of
Math. 141 (1995), 443-551.
POINTS DEFINED
OVER CYCLIC QUARTIC EXTENSIONS
ON AN ELLIPTIC CURVE
AND
GENERALIZED KUMMER SURFACES
Masato KUWATA
Kanagawa Institute of Technology
1030 Shimo-Ogino
Atsugi-shi, Kanagawa,
243-0292 Japan
kuwata~gen. kanagawa-it.ac.jp
1. Introduction
Let E be an elliptic curve over a number field k. By the Mordell-Weil
theorem the group E(K) of K-rational points on E, where Klk is a finite
extension of k, is a finitely generated abelian group. We fix Elk once
and for all, and we study the behavior of the rank of the group E(K) as
K varies through a certain family. We are particularly interested in the
family :Fk(G) of all Galois extensions K/k whose Galois group Gal(K/k)
is isomorphic to a prescribed finite group G. In this article we focus on
the case G = Z/4Z.
One case that has been well studied is the case where G = '1./2'1.. If an
elliptic curve Elk is given by the Weierstrass equation y2 = x 3 +Ax+B,
and d is a nonzero element of k, the quadratic twist of E by d, denoted
by Ed, is given by the equation dy2 = x 3 + Ax + B. Since we have the
relation rankE(k(JlD) = rankE(k) + rank Ed(k), studying the behavior
of the rank of E(k( v'd)), as k( v'd) varies through :Fk(Z/2Z), is equivalent
to studying the family {Ed(k) IdE k Xl(kX)2}. In this case it is very
easy to find values of d such that rankEd(k) is positive. Indeed, take
any integer m in k, and write m 3 + Am + B = dl 2 , then for almost all
m, the point (m, l) E Ed(k) is of infinite order.
[received: July, 12; accepted in revised form: August, 15, 2002]

66 M. KUWATA
If k is the field of rational numbers Q, our problem is conjecturally

equivalent to the vanishing of the quadratic twists of the L-function of
E (see [5] for more detail). Let L(E, s) = L:~=1 ann- s be the L-function
of E/Q, and let X : Gal(Q/Q) ~ ex be a Dirichlet character of order 2.
The twist of L(E,s) by X is given by L(E,s,X) = L:~1 x(n)ann- s In
this case we have a functional equation that relates L(E,2 - s, X) and
L(E, s, X), and this functional equation forces L(E, 1, X) to vanish for at
least one half of quadratic characters.
When X runs through Dirichlet characters of order n, there is no a
priori reason that the twist L(E, s, X) vanishes for infinitely many X's.
However, numerical experiments seem to suggest that L(E, s, X) vanishes
quite often when the order of the character is small (see [2]).
If n is greater than 2, we do not dispose of an obvious method to find
a point on E defined over some cyclic extension of degree n. It turns out
that we have a higher dimensional analogue of Ed that is a variety over
k whose k-rational point corresponds to a point on E defined over some
cyclic extension K/k of degree n. When G = 1-/31-, 1-/41- or 1-/61-,
this variety is a surface which belongs to the class of surfaces called
generalized Kummer surfaces. Earlier, we obtained some results for the
case G = 7L/37L by studying this surface (see [4]). When G = 7L/47L, it
turns out that we have a stronger result:
Theorem 1.1. Let E be an elliptic curve over a number field k. Then

there exist infinitely many cyclic quartic extensions K / k such that
rankE(K) is strictly greater than rankE(K2), where K2 is the unique
intermediate quadratic extension of k in K.
Corollary 1.2. Let Fk,G = UKE.rk(G) K be the compositum of all the

Galois extensions of k whose Galois group is isomorphic to G. Then the
quotient group E(Fk,z/4Z)/ E(Fk,z/2Z) is not finitely generated.
To prove our theorem we find a k-rational curve contained in the

generalized Kummer surface in question, and write down its equation
explicitly.
The author would like to thank Professor Hershy Kisilevsky and
Dr. Jack Fearnley for useful conversations. He would also thank Profes-
sors Katsuya Miyake and Ki-ichiro Hashimoto for giving me an opportu-
nity to present an earlier version of this paper at the conference "Galois
theory and Modular forms" at the Tokyo Metropolitan University.
Generalized Kummer surfaces 67
2. A generalized Kummer surface

Let E be an elliptic curve over k. Consider the automorphism p of
Ex E given by
p : ExE --+ ExE
(P,Q) t---+ (Q,-P).
It is easy to see that p is of order 4, and p2 is the multiplication-by-( -1)
map. Let X be the quotient of Ex E by the group (p) generated by
p, and let X be its minimal non-singular model obtained by blowing-
ups. X is a generalized Kummer surface (cf. Katsura [3], Bertin [1]). In
particular X is a K3 surface.
The set of fixed points of p consists of the four points of the form
(T, T), where T is a 2-torsion point. The set of fixed points of p2 consists
of the sixteen points of the form (T, S), where T and S are 2-torsion
points. Thus p acts freely away from these sixteen points. We denote by
XO the open set of X obtained by removing the image of these sixteen
points in X.
Lemma 2.1. If the equivalence classes [P, Q] is in the set of k-rational
points XO(k), then (P, Q) satisfies one of the following:
(1) P and Q are defined over k.
(2) P and Q are both defined over some quadratic extension k(Vd)/k.
1fT E Gal(k(Vd)/k) is the generator, then we have T(P) = -P
and T(Q) = -Q.
(3) P and Q are both defined over some cyclic quartic extension K/k.
If we choose a suitable generator 0' E Gal(K/k), then we have
O'(P) = Q and O'(Q) = -Po
Proof. Let P and Q be points on E(k), and suppose that [P, QJ E XO(k).
If 0' is an element of Gal(k/k), then (O'(P) , O'(Q)) is one of the follow-
ing four pairs: (P, Q), p(P, Q) = (Q, -P), p2(p, Q) = (-P, -Q) or
p3(p, Q) = (_Q, P). Note that these four pairs are distinct since [P, Q]
is in Xo. We can therefore define a map 'I/J : Gal(k / k) -+ (p). Since
the automorphism p is defined over k, it commutes with any element of
Gal(k/k). Thus, if O'l(P,Q) = pi(P,Q) and 0'2(P,Q) = pi(P,Q), then
0'1(0'2(P,Q)) = O'l(pi(P,Q)) = pi(pi(p,Q)) = pi+j(P,Q). This shows
that the map 1/J is a homomorphism.
Let K be the Galois extension of k corresponding to ker 1/J via Ga-
lois theory. Then Gal(K/k) is isomorphism to a subgroup of {p}. If
Gal(K/k) = {id}, then K = k, and both P and Q are defined over k.
This is the case (1).
68 M. KUWATA
If Gal(K/k) ~ (p2), then K is Lt, quadratic extension of k. Let r E

Gal(k/k) be an element whose image in Gal(K/k) generates Gal(K/k).
Then 'IjJ(r) = (p2). This shows that r(P,Q) = p2(p,Q) = (-P,-Q).
This is the case (2).
If Gal(K/k) ~ (p), then K is a cyclic quartic extension. Let (J E
Gal(k/k) be an element whose image in Gal(K/k) maps to p by 'IjJ.
Then we have (J(P, Q) = (Q, -P). This implies that Q = (J(P) and
(J2(p) = (J(Q) = -Po This is the case (3). D
Conversely, if we have a point P defined over some cyclic quartic

extension of k, we have the following.
Lemma 2.2. Let P be a point whose field of definition is a cyclic quartic
extension K/k. Let (J be a generator of Gal(K/k). Define P' = P -
(J2(P). Then [P',(J(P')] is in X(k). Furthermore, if E does not have a
k-rational2-torsion point, then [P',(J(P')] is in XO(k).
Proof. First, note that P' is not O. Otherwise, we have P = (J2(p),
which is a contradiction to the fact that the field of definition of P' is
K. We see easily that (J2(p') = -P', which implies that (J(P', (J(P')) =
((J(P'),a 2(p')) = (a(P'),-P') = p(P',aCP')). This shows that
[P', (J(P')] is in X(k).
Suppose that [P', (J(P')] is not in XO(k). Then P' is a 2-torsion point
of E. Moreover, P' is defined over the intermediate quadratic extension
K2 between K and k, since we have (J2(p') = -P' = P'. Then we see
that either P' itself or P' +(J(P') is a k-rational2-torsion point of E. D
3. Equation of the surface X

In this section we will write down an equation of X in order to study
X in detail. We first fix an equation of E:
E : y2 = x 3 + Ax + B. (1)
For simplicity of exposition, we assume AB 0:/: 0 throughout. Let k(E x

E) be the function field of E x E. Then k(E x E) may be written as
k(Xl,X2,Yl,Y2), where (Xl,Yl) and (X2,Y2) both satisfy the equation (1).
The automorphism induced by p on k(E x E), also denoted by p, satisfies
Let Y be the quotient surface E x E/(p2) = E x E/{1}, which is a

singular model of the Kummer surface associated with Ex E.
Lemma 3.1. The function field k(Y) of the quotient surface Y is the
subfield of k(E x E) generated by Xl, X2 and y = yl/Y2, which satisfy
the equation
(2)
Furthermore, the automorphism p acts on keY) by
1
p(XI) = X2, p(X2) = Xl! p(y) = --.
Y
Proof. It is easy to see that the elements Xl, X2 and Y = yl/Y2 are fixed
by the automorphism p2 and thus belong to keY). Also, it is easy to see
that Xl, X2 and y satisfy the equation (2). We thus need to show that
these three elements generate keY). To see this it suffices to show that
the degree of extension [k(E x E) : keY)] is 2. The element YI is a root
of the quadratic equation in T with coefficients in keY):
T2 - x~ - AXI - B = O.
Thus, we have [k(E x E) : keY)] $ 2. The fact that YI and Y2 are
independent in k(E x E) implies that k(E x E) i= keY). Thus, we
conclude [k(E x E) : keY)] = 2. The action of p on Y is given by
p(y) = p(YI) = ~ = _~.

p(Y2) -Yl Y
o
Proposition 3.2. The function field k(X) of the quotient surface X =
Ex Ej(p) is the subfield of k(E x E) generated by
which satisfy the equation
e,
Proof. First, we can easily verify that the elements 'f/ and t are fixed
by the automorphism p, and thus belong to k(X). It is easy to see that
k(e, 'f/, y) = k(xl, X2, y) = keY). This shows that keY) is a quadratic
extension of k(e, 'f/, t) obtained by adding y satisfying y2 - ty - 1 = O.
e,
To obtain a relation among 'f/ and t, we first express the equation
e,
(2) in terms of 'f/ and y, and th~'l we eliminate Y using the relation
y2 _ ty - 1 = O. 0
70 M. KUWATA
4. Rational curves on the surface X

In this section we show that there exist infinitely many parametrized
curves defined over k on the surface X. This allows us to show that
there are infinitely many different cyclic quartic extensions such that
the Mordell-Wei! group over it increases from that over k.
Looking at the equation (2), we notice that the surface Y = E x
E/{1} can be regarded as a family of cubic curves in the Xlx2-plane
parametrized by y. The projective model of this plane curve defined over
key) intersects with the line at infinity at three points. These are not
defined over key), but one of them is defined over the extension k( ~).
Writing u = ~, we obtain a plane cubic curve defined over k(u) given
by
(4)
with a k(u)-rational point at infinity. The tangent line at this point
intersects with the third point P (see Figure 1), which is again k(u)-
rational.
(y2IJ :1:0)
Figure 1. Plane cubic curve (x~ + AX2 + B)y2 = x~ + AX1 + B.

(Here, 011,012 and 013 a.re the three roots of x 3 + Ax + B = 0.)
A simple calculation shows that the coordinates of P are given by
( ) _ (-B(U4+u2+1) -B(U4 +U2 +1)) (5)

Xl,X2 - A(u2 + 1) , Au2(u2 + 1) .
In other words the triple cover Y' of Y defined by the equation (4)
has a k-rational parametrized curve. The automorphism of p on Y can
be lifted to the automorphism of Y' by defining

p': (Xl,X2,U) ~ (X2,Xl,-1ju).
It is easy to see that the parametrized curve obtained above is stable
under the action of p. We have a commutative diagram:
y' ---+ Y
pll lp
y' ---+ y
Let X, be the quotient Y'j(p). The diagram above induce the diagram:
y' ---+ Y
1
x'--+x
1
Proposition 4.1. The function field of the quotient X' = Y' j (p) is the
subfield of k(E x E) generated by
1
s = u --,
u
which satisfy the equation
- s(s2 + 3)e + 3e11 + 113

+ A(s~ + 1)2(s2 + 4)11 + Bs(s2 + 1)2(s2 + 3)(s2 + 4) = O. (6)
The covering map X' -+ X is given by (e, 11, s) ~ (e, 11, s3 + 3s).
Proof. The proof goes similarly to that of Proposition 3.2. The last part
is a consequence of the simple calculation
s3 = (u - ~r = u3 - :3 -3( u -~) =y- ~ - 3s = t - 3s.
o
We think of the equation (6) as that of a cubic curve in e17-plane over
the function field k(s). This curve has two rational points, one at infinity
and the other coming from P on Y. Calculations show that the latter
has coordinates
(c ) = (-B(S2+ 3 ) -BS(S2+ 3 ))
.. ,11 A' A .
72 M. KUWATA
Choosing the point at infinity (1 : s : 0) as the origin, we regard X' as

the elliptic curve over k( s).
Proposition 4.2. The equation (6) is tronsformed to the Weierstross
form
y2 = x 3 _ 48A2(s2 + 4)2 X
-16(s2 + 4)3 ((4A 3 + 27B2)s2(s2 + 3)2 + 8A3 ) (7)
by the tronsformation
4(s2 + 4) ( )
X = (s~ _ 7]) As(s2 + 2)e + A(2s2 + 1)7] + 3Bs(s2 + 1)(s2 + 3) ,
y = 12s(s2 + 4) (3B(s2 + 3) (s(s2 + 2)e - 2e7] - S7]2)

(se - 7])2
+ 2A(s2 + 1f(s2 + 4)(A1] + Bs(s2 + 3))).
The point (e,1]) :-= (_B(s2 + 3)/A, _Bs(s2 + 3)/A) is tronsformed to
(X, Y) = (12 ((A 3 + 9B2)(s2 + 3)2 - A 3 ),
36B~: + 3) ((A3 + 6B2)(s2 + 3)2 _ A3),) (8)
which is of infinite order.

Proof. Converting to the Weierstrass form is standard and straight for-
ward. The elliptic surface given by (7) has the following types of singular
fibers under the condition AB =f:. O.
10 at s2 + 4 = 0,
12 at s(s2 + 3) = 0,
II at (4A3 + 27B2)8'2(s2 + 3)2 + 16A3 = O.
If the point given by (8) is a torsion point, its specialization at each
fiber is a torsion point of the same order. Since a fiber of type 10 is
isomorphic to Ga x ('1./2'1.)2 as an algebraic group, its torsion points are
of order 2. Since the Y -coordinate of the point (8) is not 0, it is not a
point of order 2, and thus it is of infinite order. 0
A k(s)-rational point on the elliptic curve (7) is one-to-one correspon-
dence with a k(s)-rational point (~(s), 1](s)) on the plane curve (6), which
in turn gives a parametrized curve (e(s), 1](8), s3 + 3s) on the surface X.
Since Proposition 4.2 gives us infinitely many k(s)-rational points on (7),
we have infinitely many k-rational parametrized curve on X.
Generalized Kummer surfacts 73
5. Points on E defined over cyclic quartic

extensions
We have seen that the surface X has a k-rationally parametrized curve
(c t) _ (-B(S2 + 3) _Bs(s2 + 3) 3 3)
.. ,1], - A ' A's + s (9)
with parameter s. This gives a point on E defined over a quartic exten-

sion of k(s).
Lemma 5.1. Let v be a root of the quadratic equation in T with k(u)-
coefficients:
A(u 2 + 1)T2 + A 3Bu4(u 2 + If + B 3(u 4 + u 2 + 1)3 = O.
Then k(v) is a cyclic quartic extension of k(s), and the elliptic curve E
has four k( v) -rational points
(x,y) = (
-B(U4 +U 2 +1)
A(u2 + 1) , A(u2 + 1)
v) ,
4 2
( -B(U + u + 1) v )
Au 2(u 2 + 1) , Au3(u 2 + 1) .
Proof. We have seen that the point on X given by (9) comes from the
point P on the Kummer surface Y. We obtain four points in the lemma
by taking the preimage of P in E x E. The arguments in 2 show that
the field of definition of these points, k( v), is a cyclic quartic extension
of k(s). This may be seen directly as v satisfies the following quartic
equation in T with k[s]-coefficients
A2(s2 + 4)T4
+ AB(82 + 4)(84 + 38 2 + 1)(A3(82 + 4) + B2(s2 + 3)3)T2
+ B2(A3(s2 + 4) + B2(s2 + 3)3)2 = O. (10)
o
Proof of Theorem 1.1. First we show that the points obtained in Lem-
ma 5.1 are of infinite order. If we let
A3 Bu4 (u 2 + 1? + B 3(u 4 + u 2 + 1)3
d(u) = - A(u2 + 1) ,
then the points in Lemma 5.1 may be considered as k(u)-rational point

on the twisted curve
Ed(u) : d(u)y2 = x 3 + Ax + B.
74 M. KUWATA
If we regard this elliptic curve over k(u) as an elliptic surface over k,

then the only singular fibers it has are of Kodaira type Io. This implies
that the only possible torsion k(u)-rational points are points of order 2.
However, our points are clearly not of order 2 since the y-coordinates
are not O.
Since k(v)/k(u)-trace of the k(v)-points in Lemma 5.1 are 0, we see
that the intersection between the subgroup generated by these k(v)-
points and the group E(k(u)) is {O}. Thus we have rankE(k(v)) >
rankE(k(u)).
By specializing s to different values of k, we obtain infinitely many
points on E defined over the splitting field of (10). By Hilbert's irre-
ducibility theorem these are cyclic quartic extensions for infinitely many
values of s. Also, by a theor~m of Silverman [6] on the specialization of
a family of elliptic curves, points on an elliptic curve obtained by spe-
cializing a point of infinite order is once again of infinite order except
for a finite number of values. This completes the proof. 0
Proof of Corollary 1.2. Since points on E whose field of definition are

different quartic extensions are clearly independent, the assertion follows
immediate from the theorem. 0
Remark 5.2. The discriminant of the equation (10) is given by
6. Elliptic curve with 2-torsion

If the elliptic curve E in question has a 2-torsion point, then the
surface X has more k-rational curves on it. In this section we show
explicit results for the case where E has three k-rationaI2-torsion points.
Other cases can be worked out similarly.
Let us fix the equation of E as
E : y2 = (x - c) (x - d) (x + c + d).
Then the equation (3) of X becomes
-te + 3e7] + 7]3 - (c 2 + cd + d 2 )(t2 + 4)7] + cd(c + d)t(t 2 + 4) = o.

We regard this as a cubic curve in the ~7]-plane over k(t). We have
several obvious k(t)-rational points, for example,
Pl = (2c, ct), P2 = (2d, dt), P3 = (-2c - 2d, -ct - dt).

These correspond to 2-torsion points on E x E. The tangent line at PI

intersects with the curve at Q = (-c, et). This gives four points on E
given by
9c t ) ( t)) '
3et
V{3c
8" ((c+ 2d)2 - t2+4
2 2
( -2c + 2Jt2TI' 1- Jt2TI
(
c 3ct
3c (
-2 - 2Jt2TI' 8" (c + 2d) -
2 9c2t
t2 + 4)
2
( t)) .
1 + Jt2TI
These are defined over the cyclic quartic extension that is the splitting
field of the equation
T4 + 3c(t2 + 4)((2c + d)(c - d)t 2 - (c + 2d)2)T2

+ 9c2(t 2 + 4)( (2c + d)(c - d)t 2 - (c + 2d)2)2 = O.
By considering tangent lines at other points, we obtain more points
defined over similar cyclic quartic extensions.
Remark 6.1. Choosing PI = (2c, et) as the origin, the equation of X
may be converted to the Weierstrass form.
y2 = X 3 _ + cd + ~)(t2 + 4)X
48(C
+ 16(t2 + 4)3((c - d)2(c + 2d)2(2c + d)2t 2 + 8(c2 + cd + d2))3.
The rank of the Mordell-Wei! group of this elliptic curve over k(t) is 2
or 3 depending on whether or not E has complex multiplication. If E
does not have complex multiplication, then the Mordell-Weil group is
generated by the images of P2 and P3'
References
[1] J. Bertin, Reseaux de Kummer et surfaces de K3, Invent. Math. 93 (1988), 267-
284.
[2] J. Fearnley, Vanishing and non-vanishing of L-series of elliptic curves twisted by
Dirichlet characters, Ph.D. thesis, Concordia University, 2001.
[3] T. Katsura, Generalized Ktl.mmer surfaces and their unirationality in character-
istic p, J. Fac. Sci., Univ. Tokyo, Sect. IA 34 (1987), 1-41.
[4] M. Kuwata, Points defined over cyclic cubic extensions on an elliptic curve and
generalized Kummer surfaces, preprint.
[5] D. E. Rohrlich, The vanishing of certain Rankin-Selberg convolutions, in "Au-
tomorphic forms and analytic number theory (Montreal, PQ, 1989)," Univ.
Montreal, Montreal, QC, 123-133, 1990.
76 M. KUWATA
[6J J. H. Silverman, Heights and the specialization map for families of abelian vari-
eties, J. Reine Angew. Math. 342 (1983), 197-251.
THE ABSOLUTE
ANABELIAN GEOMETRY OF
HYPERBOLIC CURVES
Shinichi MOCHIZUKI
Research Institute for Mathematical Sciences,
Kyoto University,
Kyoto 606-8502 Japan
motizukilkurims.kyoto-u.ac.jp
Contents:
Introduction
o. 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. O below) over a field K of char-
acteristic O. Denote its algebraic fundamental group by IIxK. Thus, we
have a natural surjection
of I1XK 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. [Tarnal], [Mzk6]; Theorem 1.3.4 of the present
manuscript) that one may recover the curve XK in a functorial fashion
from I1XK. This sort of result may be thought of as a "relative result"
(i.e., over GK). Then the question naturally arises:
[received: May 26, 2002; accepted in revised form: February 20, 2003]
78 S. MOCHIZUKI
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 IIxK
(cf. Corollary 1.3.5), by applying the theorem of Neukirch-Uchida (cf.
Theorem 1.1.3). On the other 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 IIxK'
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
IIxK'
o. Notations and Conventions

Numbers:
We will denote by N the set of natural numbers, by which we mean the
set of integers n ~ O. 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
ZG(H) ~f {g E Gig h = h . g, V h E H}
for the centralizer of H in G;
NG(H) ~f {g E Gig. H . g-l = H}

for the normalizer of H in Gj and
GG(H) ~f {g E G I (g. H . g-l) n H has finite index in H, 9 . H . g-l}
for the commensurator of H in G. Note that: (i) ZG(H), NG(H) and

GG(H) are subgroups of Gj (ii) we have inclusions
H, Za(H) ~ Na(H) ~ Ga(H)
Absolute anabelian geometry 79
and (iii) H is normal in Nc(H).

Note that Zc(H), Nc(H) are always closed in G, while Gc(H) is not
necessarily closed in G. Indeed, one may construct such an example as
follows: Let
M ~frr 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 E N, write pn(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 P*(M) on M. If a E AutF(M), then for every n E N, a in-
duces a continuous homomorphism an: M/pn(M) -+ M/pn(M) which
is clearly surjective, hence an isomorphism (since M / pn (M) is profinite
and topologically finitely generated - cf. [FJ] , Proposition 15.3). It
thus follows that a induces an isomorphism pn(M) ~ pn(M), hence
that the inverse of a 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/pn(M)) (where
Aut(M/pn(M)) e:! GLn(Zp) is equipped with its usual topology) are
continuous. Note that relative to this topology, AutF(M) forms a Haus-
dorff 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
II pn. Zp ~ II Zp = M
nE]\j ]\j
(so H ~ G is a closed subgroup). Then GG(H) is not closed in G. For

instance, if one denotes by en E TIN Zp the vector with a 1 in the n-th
place and zeroes elsewhere, then the limit Aoo (where
for all n E N) of the automorphisms Am E GG(H) (where Am(en ) ~f

en + en+l if n ~ m, Am(en ) ~f en if n > m) is not contained in Gc(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 Gis
trivial.
(ii) We shall say that a continuous homomorphism of profinite groups
G -+ H is relatively slim if the {'(mtralizer in H of the image of every
open subgroup of G is trivial.
80 S. MOCHIZUKI
(iii) We shall say that a closed subgroup H ~ G of a profinite group G is

commensurably (respectively, normally) terminal if the commensurator
GG(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 Hand 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
hE ZG(H), let H' ~ G by the (necessarily open) subgroup generated by
Hand 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 [Nakai] (cf., e.g., Corollaries
1.3.3, 1.3.4).
Curves:
Suppose that 9 ? 0 is an integer. Then a family of curves of genus 9
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 > O. 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. [DMJ, [Knud] for an
exposition of the theory of such curves; strictly speaking, [Knud] treats
the finite etale covering of Mg,r determined by ordering 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
9 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 com-
posite of an open immersion X ~ Y onto the complement Y\D of a
relative divisor D ~ Y which is finite etale 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 isomor-
phism. (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 norma! S on which a prime 1 is invertible (which,
by Zariski localization, we may assume without loss of generality). De-
note by S' ~ S the finite etale 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 Sin
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 etale
covering of (Mg,r )Z[fJ 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 1 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.
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 de-
note the resulting absolute Galois group of F by GF. Let p be a (not
necessarily nonarchimedear.!) prime of F. Write G p ~ GF for the de-
composition group (well-defined up to conjugacy) 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 , G F):
Theorem 1.1.1 (Slimness and Commensurable Terminality).

Suppose that p is nonarchimedean. Then:
(i) The closed subgroup G p ~ G F is commensurably terminal.
82 S. MOCHIZUKI
(ii) The inclusion Gp ~ Gp is relatively slim. In particular, Gp , Gp

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 a E Gp commutes with an open subgroup
H ~ Gp , then a induces the trivial action on the abelianization Hab.
But, by local class field theory, Hab may be identified with the profinite
completion of K X , where K is the finite extension of Fp determined by
H. Thus, a acts trivially on all sufficiently large finite extensions K of
Fp, so a = 1, as desired.) Relative slimness thus follows formally from
the slimness of Gp and (i) (cf. Remark 0.1.2). 0
Theorem 1.1.2 (Topologically Finitely Generated Closed Nor-
mal Subgroups). Every topologically finitely generated closed normal
subgroup of Gp is trivial.
Proof. This follows from [FJ], Theorem 15.10. o
Theorem 1.1.3 (The Neukirch-Uchida Theorem on the An-
abelian Nature of Number Fields). Let F l , F2 be number fields.
Let Fl (respectively, F2) be an algebraic closure of Fl (respectively, F2).
Write Isom(F2/F2,FdFd for the set of field isomorphisms F2 ~ Fl
that map F2 onto Fl. Then the natural map
Isom(F2/ F2, Fd F l ) -+ Isom(Gal(Ft/ H), Gal(F2/ F2))
is bijective.
Proof. This is the content of [NSW], Theorem 12.2.1. o
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]).
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 l' is nonarchimedean!), and that we are given an exact
sequence of profinite groups:
1-~-+TI-G-l
Suppose, moreover, that this sequence splits over some open subgroup
of G, and that ~ is topologically finitely generated. The following result
was related to the author by A. Tamagawa:
Lemma 1.1.4 (Intrinsic Characterization of Arithmetic Quo-

tients). (i) Suppose that G = GF. Let IT' ~ IT be an open subgroup.
Then the kernel of the homomorphism IT' -+ G may be characterized as
the unique maximal closed normal subgroup of IT' which is topologically
finitely generated.
(ii) Suppose that G = Gp Assume further that for every open subgroup
n
II" ~ II, the abelianization (L\,,)ab of L\" (where L\" ~f II" L\) satisfies
the following property:
(*) The maximal torsion-free quotient (L\,,)ab -1+ Q" of
(~")ab on which the action of G" ~f II" I L\" (by conjuga-
tion) is trivial is a finitely generated free Z-module.
Let II' ~ II be an arbitrary open subgroup. Then:
[G: G'] . [Fp : Qp] = dimQp ((II,)ab z Qp) - diml(b ((IT,)ab z Ql)
n
(where L\' ~f L\ II'; G' ~f II'I L\'; p is the rational prime that p diVides;
and 1 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 L\ ~ II may be characterized as the intersection of those
open subgroups II' ~ II such that:
[G : G'] = [II : II']
(i.e., such that [G : G'] . [Fp : Qp] = [II : II'] . ([G : G] . [Fp : QpJ)).
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]
Next, let us observe that we have an exact sequence
0-+ Im(L\') -+ (II')ab -+ (G,)ab -+ 0
(where Im(L\') is the image of L\' in (II')&b) of finitely generated 1.-

modules. Note, moreover, that we have a natural surjection Q' -1+
Im(L\')/(torsion) (where Q' is as in (*. On the other hand, it fol-
lows formally from our assumption that II' -1+ G' splits over an open
subgroup of G' (together with the well-known fact that the group coho-
mology of a finite group is always annihilated by the order to the group)
84 S. MOCHIZUKI
that this natural surjection is, in fact, an isomorphism. In particular, it

follows formally from (*) that:
dimQp ((ll,)ab z Qp) - dimlQj ((ll,)ab z Ql)

= dimQp (( G,)ab z Qp) - diffilQj (( G,)ab z Ql)
Thus, to complete the proof of Lemma 1.1.4, it suffices to prove that:
[K' : Qp] = dimlQp (( G,)ab z Qp) - dimlQj (( 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 profi-
nite completion of (K') x 0
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 extension of Qp. Fix an algebraic closure K of K; write
GK ~ Gal(KjK). Let A be a semi-abelian variety over K. Denote the
resulting (profinite) Tate module of A by:
clef -
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 G K) the Tate module of a torus is isomorphic
to the direct sum of a finite number of copies of Z(l), 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.
Now it follows from the th.30ry of [FC] (cf., in particular, [FC], Chap-
ter III, Corollary 7.3), that T(A) fits into exact sequences (of GK-
modules)
o---+ Tgood ---+ T (A) ---+ Teom ---+ 0
o ---+ Tt1c ---+ Tgoocl ---+ T(B) ---+ 0
where T(B) is the Tate module of an abelian variety B over K with
potentially good reduction; and Teom = Meom z Z, nrc = Mtre z
Z(1) for finitely generated free Z-modules Meom , M tre on which GK
acts via a finite quotient. [Here, "com" (respectively, "trc") stands for
"combinatorial" (respectively, "toric").] It is thus evident that 1trc maps
to 0 in Q. Moreover, by [Mzk4], Lemma 8.1 (the proof of which is valid
for arbitrary B, even though in loco 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 G K acts trivially. Since Z is Z-
fiat, however, this implies that Q is equal to the result of applying zZ
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.
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, mKt; k i ). Also, we assume that we
have chosen an algebraiC closures Ki of Ki and write
def -
GKt = Gal(KdKi)
for the corresponding absolute Galois group of Ki' Thus, by local class
field theory (cf., e.g., [Serre2]), we have a natural isomorphism
(Kt)" .::+ G~
(where the "1\" denotes the profinite completion of an abelian group;
"x" denotes the group of units of a ring; and "ab" denotes the maximal
abelian quotient of a topological group). In particular, G~ fits into an
exact sequence
(arising from a similar exact sequence for (Kt)"). Moreover, we obtain

natural inclusions
(where" .::+ " denotes the morphism induced by the valuation on Kt)
by considering the Teichmiiller representatives of elements of kt and
86 S. MOCHIZUKI
the Frobenius element, respectively. Also, in the following discussion we

shall write:
~ def ~
(where'll' = 'll/'llp). Finally, we denote the cyclotomic character of GK;
by:
Proposition 1.2.1 (Invariants of Arbitrary Isomorphisms of Ga-

lois Groups of Local Fields). Suppose that we are given an isomor-
phism of profinite groups:
Then:
(i) We have: Pi = P2. Thus, (in the remainder of this proposition and
its proof) we shall write p ~f Pi = P2.
(ii) a induces an isomorphism IKl ~ IK2 between the respective inertia
subgroups ofGK!, GK2'
(iii) The isomorphism aab : GW 1 ~ GW
2 induced by a preserves the
images Im( O~), Im( kt), Im( Kt) of the natural morphisms discussed
above.
(iv) The morphism induced bya between the respective quotients
G~/Im(O~) preserves the respective Frobenius elements.
(v) [Kl : Qp] = [K2 : Qp]; [kl : lFp ] = [k2 : lFp ]. In particular, the
ramification indices of K 1 , K2 over Qp coincide.
(vi) The morphisms induced by a on the abelianizations of the various
open subgroups of the G Ki induce an isomorphism
J,LQ/Z (K 1) ~ J,LQ/Z (K 2)
which is Galois-equivariant with respect to a. In particular, a preserves
the cyclotomic characters Xi.
(vii) The morphism H2(K1 ,J,LQ/z(Kl)) ~ H 2(K2,J,LQ/z(K2)) induced
by a (cf. (vi)) preserves the "residue map"
2 - ~
H (Ki , J,LQ/z(K i )) -t Q/'ll
of local class field theory (cf. fSerre2), 1.1).
Proof. Property (i) follows by considering the ranks of G~ over various
Zl (cf. Lemma 1.1.4, (ii)). Property (iii) for Im(k;) follows from the
fact that Im(kt) may be recovered as the prime-to-p torsion subgroup of

G~. Property (v) follows for [Ki : QpJ by considering the Zp-rank of G~
(minus 1) and for [ki : lFp] by considering the cardinality ofIm(kf) (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(O~) follows formally from (ii) (since this image
is equal to the image in G~ of IKJ. Property (iv) follows by applying
(iii) for Im(kt) 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 k; (where ki denotes the algebraic
closure of ki induced by K i ) as multiplication by lkll = Ik21 [where,
relative to our use of the term "multiplication", we think of the abelian
group k; additivelyJ. Here, we note that if Li is a finite extension of K i ,
then the inclusion .
G~ ~ (Kt)'\ ~ (L:>" ~ Gi~

may be reconstructed group-theoretically by considering the Ver-
lagerung, or transfer, map (cf. [Serre2], 2.4). Property (iii) for Im(Kt)
follows formally from (iv). Property (vi) follows formally from (iii). Fi-
nally, 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 mor-
phism H2(Ki,J..tQ./z(K i ~ Q/Z may be constructed as the composite
of the natural isomorphism
2 -
H (Ki,l-'Q/z(Ki = H 2(GKpl-'Qjz(K
-
i
'"
-+
2 -x
H (GKi,Ki )
- which is group-theoretic, by (iii) - with the inverse of the isomor-
phism
H2(Gal(Krr /Ki)' (Krr)x) ~ H 2 (GKi,K;)
(where Kfnr denotes the maximal unramified extension of Ki) - which
is group-theoretic, by (ii), (iii) - followed by the natural isomorphism
H2(Gal(Krnr / Ki), (Krr)x) ~ H2(Gal(Krr / K i ), Z) = H2(i, Z)

=Q/Z
- which is group-theoretic, by (ii), (iii), (iv). o
Before proceeding, we observe that Proposition 1.2.1, (i), may be
extended as follows: Write
2lQ ~f ISpec(Z) I U {oo}

88 S. MOCHIZUKI
(where "I - I" denotes the underlying set of a scheme) for the set of
"all arithmetic primes of Q". If v E 2tQ is equal to (0) E ISpec(Z)I
(respectively, (0), set Gv ~f GQ (respectively, Gv ~f Gal(CjlR)). If
v E ISpec(/l)I ~ 2lQ is equal to the prime determined by a prime number
c' Jf
p, set C v = CQp.
Proposition 1.2.2 (Intrinsicity of Arithmetic Types). For i =
1,2, let Vi E 2tQ. Suppose that Hi is an open subgroup of G Vi ' Then
Hi e:! H2 implies Vi = V2.
Proof. Indeed, open subgroups of GQ may be distinguished by the fact

that their abelianizations fail to be topologically finitely generated. (In-
deed, consider the abelian extensions arising from adjoining roots of
unity.) By contrast, abelianizations of open subgroups of GIR or GQp (cf.
the above discussion) are topologically finitely generated. Next, open
subgroups of GIR 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). 0
Next, let us write Spec(OKi)log for the log scheme obtained by equip-
ping the scheme Spec(OKi) with the log structure defined by the divisor
V(mK i ). Thus, by pulling back this log structure via the natural mor-
phism Spec(ki ) ~ Spec(OKi)' we obtain a log scheme Spec(kdog , which
we denote by
klog t
for short. Note that the "etale monoid" that defines the log structure
on k~Og "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
t
into a natural exact sequence (of monoids):
1~ kt ~ Mkiog ~ N ~ 0
t
Thus, the kt -torsor Ui determined by considering the inverse image of

1 EN in this sequence may be identified with the kt-torsor of uniJormiz-
ers E mKi considered modulo mki
Next, let us write
C Ki -#
C log
ki
for the quotient defined by the maximal tamely ramified extension Klame
of Ki. Thus, G~~g may also be thought of as the "logarithmic fundamental
group" 71'1 (k~og) of the log scheme k~Og. Moreover, c~~g fits into a natural
exact sequence:
where, just as in the case of Ki, we write:
I-'Z(ki ) = I-'Z,(ki) = Hom(Q/Z, -x
- - def
ki )j
- def -
I-'Q/Z(ki) = I-'Z(ki ) z Q/Z
The "abelianization" of this exact sequence yields an exact sequence:
1 -+ k~, -+ (dog)ab
ki
-+ Z-+ 1
Now we have the following:
Proposition 1.2.3. (i) Any isomorphism a: GKI ~ GK2 (as in Propo-
sition 1.2.1) induces an isomorphism Gl~g ~ G~g between the respective
quotients.
kr
(ii) There is a natural isomorphism of -torsors between the torsor of
uniformizers Ui discussed above and the HI (Z, 1-'1. (ki = kix -torsor
HI(G~g'l-'z(ki[ll of elements of HI(G~g'l-'z(ki that map to the iden-
tity element in HI(l-'z(ki),1-'7.(ki = H0ffil,,(l-'z(ki),l-'z(ki )). This iso-
morphism is defined by associating to a uniformizer 7r E Ui the I-'Z(ki)-
torsor over k!og determined by the roots 7r1/N, as N ranges (multiplica-
tively) over the positive integers prime to Pi.
(iii) The profinite group G~~g is slim.
Proof. Property (i) follows from Proposition 1.2.1, (ii), together with
the fact that the quotient GKi - G~g may be identified with the quo-
tient of GKi by the (unique) maximal pro-p subgroup of 1Ki' Next,
since any morphism of kr-torsors is necessarily an isomorphism, prop-
erty (ii) follows by observing that the stated association of coverings to
uniformizers is indeed a morphism of kt-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 com-
putation: ((l/N)qi = ( . ((l/N) (for ( E kr>.
Finally, property (iii)
follows formally from the fact that the quotient G~~g /Im(l-'z (ki is in-
finite and acts faithfully on all open subgroups of the closed subgroup
Im(l-'z(ki ~ G~~g. 0
In the following, let us denote by (k!og)", -+ k!og the "uni"prsal cov-

ering" of k!og defined by the extension Ki ame of Ki. Thus, G~g acts
naturally as the group of covering transformations of (k!og)", -+ k!og.
Proposition 1.2.4 (The "Grothendieck Conjecture" for the Log-
arithmic Point). Suppose that we are given an isomorphism of profi-
nite groups:
\ G10g '" G10g
1\. kl -+ k2
90 S. MOCHIZUKI
Then:
(i) We have: Ikil = Ik21i PI = P2. Thus, (in the remainder of this
proposition and its proof) we shall write p ~f PI = P2.
(ii) A preserves the subgroups Im(JLz(ki )) ~ G~g as well as the Frobenius
elements in the quotients G~~g lIm (J),z (k i ) ) .
(iii) Assume further that the morphism
induced by A (by thinking of k; as JlQ/Z (k i )) arises from an isomor-

phism of fields (j : kI ~ k2. Then there exists a unique commutative
diagram
of log schemes, compatible with the natural action of G~g on (k!og) . . . (for
i = 1,2), in which the vertical morphisms are the natural morphisms,
and the horizontal morphisms are isomorphisms for which the mor-
phisms on the underlying schemes are those induced by (j.
Proof. Property (i) follows by observing that Pi is the unique prime num-
ber such that 1 plus the cardinality of the torsion subgroup of (G~~g)ab
- i.e., the cardinality of k i - is equal to a power of Pi. Property (ii)
follows by thinking of the quotients G~g jlm(Jlz(ki )) as the quotients of
G~~g obtained by forming the quotient of (G~~g)ab by its torsion subgroup,
and then using that the Frobenius element is the unique element that
acts on the abelian group Im(Jlz(ki )) via multiplication by Ikil = Ik2 1.
As for (iii), the morphism o-log is the unique logarithmic extension of 0-
whose induced morphism UI ~ U2 is the morphism obtained (cf. Propo-
sition 1.2.3, (ii)) by considering the morphism induced by A between the
Hl{i,Jlz(ki )) = kr-torsors Hl(G~~g,Jlz(ki))[11 (for i = 1,2) - which
are preserved by A, by (ii). Note that here we also use (cf' (ii)) that
the Frobenius element E Z is preserved, since this element is necessary
to ensure the compatibility of the identifications
(cf. Proposition 1.2.3, (ii)). The morphism (o-log) . . . is obtained by apply-

ing this construction of "o-Iog" to the various finite log etale coverings of
k!Og obtained by considering various open subgroups of G~g. 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
(qlog) '" of qlog is a formal consequence of the fact that Gi~g is center-free
(cf. Proposition 1.2.3, (iii. 0
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 a hyperbolic curve of type (g, r) over S ~f Spec(K). Fix an
- - def-
algebraic closure K of K and write s: Spec(K) -+ Sj GK = Gal(K/K).
Let x E X(K) be a K-valued point of X lying over s. Then, setting
IIx ~f 7rl (X, x), we obtain an exact sequence
1 -+ Ilx -+ IIx -+ GK -+ 1
which determines a well-defined continous homomorphism
GK -+ Out(llx)
to the group of outer automorphisms Out(llx) of Ilx.
Lemma 1.3.1 (Slimness of Geometric and Arithmetic Funda-
mental Groups). The profinite groups Ilx, IIx are slim.
Proof. The slimness of IIx is a formal consequence of the slimness of Ax
and our assumption that GK is slim. Thus, it remains to prove that Ilx
is slim. One approach to proving this fact is given in [NakaI], Corollary
1.3.4. Another approach is the following: Let H ~ Ax be an open
normal subgroup for which the associated covering XH -+ Xl< ~f X x K K
is such that XH is a curve of genus ~ 2. Thus, Hab 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 (Le., points of the compactification
not lying in XH) to a single point. In particular, if conjugation by an
element 6 E Ilx induces the trivial action on Hab, then we conclude that
the image of 8 in Ilx / H induces the trivial action on the generalized
Jacobian just discussed, hence on XH itself. But this implies that 8 E H.
By taking H to be sufficiently small, we thus conclude that 6 = 1. 0
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 -+ Ilx -+ Aut(llx) -+ Out(llx) -+ 1
92 s. MOCHIZUKI
In particular, by pulling back this exact sequence, one may recover the
exact sequence 1 - ~x - IIx - GK - 1 entirely group-theoretically
from the outer Galois representation G K - Out( ~x).
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 Ga-

lois 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. 0
In [Mzk6], the author (essentially) proved the following result (cf.
[Mzk6], Theorem A, for a stronger version that treats arbitrary dom-
inant morphisms and open group homomorphisms [Le., which are not
necessarily isomorphisms]):
Theorem 1.3.4 ("Sub-p-adic Profinite Grothendieck Conjec-
ture"). 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 Isomg~t (~x , ~ y) the set of outer isomor-
phisms between the two profinite groups in parentheses that are compat-
ible with the respective outer actions of GK. Then the natural map
IsomK(X, Y) - Isomg~(~x,~y)
is bijective.
Thus, by combining The0rems 1.1.3; 1.3.4; Lemma 1.1.4, (i), we ob-
tain the following:
Corollary 1.3.5 (Absolute Grothendieck Conjecture over Num-
ber Fields). Let K, L be number fields; X (respectively, Y) a hy-
perbolic curve over K (respectively, L). Denote by Isom(X, Y) the set
of scheme isomorphisms X ~ Y; by IsomOut(IIx, IIy) the set of outer
isomorphisms between the two profinite groups in parentheses. Then the
natural map
Isom(X, Y) _ IsomOut(IIx,IIy)
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 ofQ. 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(Ax)
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 G K --+ Out( Ax) is injective. This injectivity was first
proven by Belyi in the case of hyperbolic curves of type (9, 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 > O. To the knowledge ofthe
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 ap-
pearing 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(Ax). This point of view is surveyed in
detail in [Naka2].
Next, we would like to discuss various properties of the inertia groups
= X x K -K
of the cusps of a hyperbolic curve. For every cusp x of X K def
- i.e., point of the unique smooth compactification of X K over K that
does not lie in X K - we have an associated inertia group (abstractly
isomorphic to Z)
Ix ~ Ax
(well-defined, up to conjugation). If 1 is any prime number, then let
us denote the maximal pro-l quotient of a profinite group by means of
a superscript "(1)". Thus, we also obtain an inertia group I~l) ~ A~
(abstractly isomorphic to Zl).
94 S. MOCHIZUKI
Lemma 1.3.7 (Commensurable Terminality of Inertia Groups).

The subgroups I~l) ~ ~~, Ix ~ ~x are commensurably terminal.
Proof. Indeed, let (7 be an element of the commensurator. If the as-
serted commensurable 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 1 in the pro-l case), together with a cusp z of
Z such that: (i) z maps to x; (ii) z =1= zU; (iii) z, ZU have conjugate
inertia groups in ~z. We may also assume (by taking ~z ~ ~x to be
sufficiently small) that Z has genus ~ 2 and admits a cusp z' =1= z, ZU.
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 ZU. But
this contradicts property (iii). D
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 [HRJ.
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 K i . Choose basepoints for the
(Xi)Ki and denote the resulting fundamental groups by IIexi)Ki' Also,
let us denote the unique proper smooth curve over Ki that compactifies
(Xi)Ki by (Yi)K i Suppose, moreover, that we are given an isomorphism
ax : IIeX1)Kl ~ II(x2)K2
of profinite groups.
Lemma 1.3.8 (Group-Theoreticity of Arithmetic Quotients).
The isomorphism ax is necessarily compatible with the quotients
IIexi)Ki ~ GKi .
Proof. This follows formally from Lemmas 1.1.4, 1.1.5. 0
Thus, Lemma 1.3.8, Proposition 1.2.1, (v), imply that ql = 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 I, ax maps inertia groups of cusps in ~Xl (respectively,
~~) to inertia groups of cusps in ~X2 (respectively, ~ ~2)'
Proof. Whether or not ri = 0 may be determined by considering whether
or not ~Xi 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 q"t -
of the action of the Frobenius element E Gkj (cf. Proposition 1.2.1, (iv
as follows: First, we observe that (as is well-known) the weights of the
action of Frobenius on A~ Ql (where 1 is a prime number distinct
from Pb P2) belong to the set {O, 1, n.
(Here, we compute weights by
choosing a lifting of the Frobenius element E Gki to GKi' Since (as
is well-known) the action of the inertia subgroup of GKt on A~ Ql
is quasi-unipotent (Le., 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
for the Ql-subspace of M on which Frobenius acts with eigenvalues of

weight w. Then, setting Ii ~f Ker(~~ Ql -- ~~ Ql), we have:
ri - 1 = diII1Q! (Ii) = dimQz (Ir 1)

= dimQz (A~ Ql)wt 1- diIDQz (A~~ Ql)wt 1
= diIDQz (A~ Ql)wt 1 - diIDQz (A~ Ql)wt 0

= diII1Q! (A~ Ql)wt 1 - diIDQz (A~ Ql)wt 0
(where the fourth equality follows from the auto-duality (up to a Tate
twist) of ~~~ 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 diIDQz (A~
Ql) = 2gi - 1 + ri (respectively, = 2gi ) when ri > 0 (respectively, when
ri = 0), this implies that gl = 92, as desired.
Finally, the statement concerning preservation of inertia groups fol-
lows formally from the fact that "ri is group-theoretic" (by applying this
fact to coverings of Xi). Indeed, let 1 be a prime number (possibly equal
to P1 or P2). Since ri may be recovered group-theoretically, given any
finite etale coverings
Zi ---. Vi ---. Xi
such that Zi is Galois, of degree a power ofl, 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 equalitv:
96 S. MOCHIZUKI
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 (.6.Xi)(I) (i.e., the maximal
prOol quotient of .6.xi ) may be characterized (group-theoretically!) as the
maximal subgroups of (.6.xi)(I) that correspond to (profinite) coverings
satisfying this condition.
In particular, (by Lemma 1.3.7) the set of cusps of Xi may be re-
constructed (group-theoretically!) as the set of (.6. xi )(lL orbits (relative
to the action via conjugation) of such inertia groups in (.6.Xi)(l). Thus,
by applying this observation to arbitrary finite etale coverings of Xi, we
recover the inertia subgroups of the cusps of .6.x; as the subgroups that
fix some cusp of the universal covering Xi ~ Xi of Xi determined by
the basepoint in question. This completes the proof. 0
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 [Tama2J,
Proposition 2.4.
Positive Characteristic:
For i = 1,2, let ki be a finite field of characteristic p; Xi a hyperbolic
curve over k i . Choose a universal tamely ramified (Le., at the punctures
of Xi) covering Xi ~ Xi of Xi; write
II~:ne ~f Gal(Xd Xi)

for the corresponding fundamental groups. Thus, we obtain exact se-
quences:
1 ~ .6.~e ~ II~:ne ~ Gk; ~ 1
(where Gki is the absolute Galois group of ki determined by Xi). As
is well-known, the Frobenius element determines a natural isomorphism
Z 9E Gki .
Lemma 1.3.10 (Slimness of Fundamental Groups). For i = 1,2,
the profinite groups .6.tame IItame are slim.
Xi ' Xi
Proof. The slimness of .6.~:ne follows by exactly the same argument -
i.e., by considering the action of .6.~:ne on abelianizations of open sub-
groups - 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 .6. ~:ne
follows from [TamaIJ, Proposition 1.11 (cf. Remark 0.1.3).J
By a similar argument, the slimness of II~:ne follows formally from:
(i) the slimness of .6.tame.
Xi '
(ii) some positive power of the character of Gkt arising from the deter-
minant of the prime-to-p portion of (~~a;ne)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 1 i= p) and the lip-primary portion" of Z ~ Gki') 0
Remark 1.3.10.1. Property (ii) in the proof of Lemma 1.3.10 was sub-
stantially simplified by a suggestion made to the author by the referee.
The following fundamental result is due to A. Tamagawa (cf. [Tamal],

Theorem 4.3):
Theorem 1.3.11 (The Grothendieck Conjecture for Affine Hy-
perbolic Curves over Finite Fields). Assume, for i = 1,2, that Xi
is affine. Then the natural map
Isom(i\/XI ,X2/X2) -+ Isom(II~e,II~~e)
(from scheme-theoretic isomorphisms Xl ~ X2 lying over an isomor-

phism Xl ~ X2 to isomorphisms of profinite groups rr~~e ~ rr~e) is
bijective.
Finally, we observe that, just as in the characteristic zero case, inertia
groups of cusps are commensurably terminal: If Xi is a cusp of (Xi)kt ~f
Xi x ki ki' then we have an associated inertia group (abstractly isomorphic
to Z')
IXi ~ ~~e
(well-defined, up to conjugation). If 1 is any prime number distinct from
p, then we also obtain an inertia group I~~) S;;; (~~;ne)(l) (abstractly
isomorphic to Zl).
Lemma 1.3.12 (Commensurable Terminality of Inertia Groups).
The subgroups [(1) C
Xi -
(~tame)(l)
Xi
[ C ~ tame
'Xi - Xi
are commensurably termi-
nal.
Proof. The proof is entirely similar to that of Lemma 1.3.7. [Alterna-
tively, as was pointed out to the author by the referee, the assertion
concerning [Xi ~ 6.~e follows from [Tama2], Lemma 2.1, while the as-
sertion concerning I~~) ~ (~~e)(l) follows, at least in the case r 2: 2,
formally from Theorem B' of [HR].] 0
98 S. MOCHIZUKI
2. Reconstruction of the Logarithmic Special

Fiber
For i = 1,2, let Ki be ajinite extension of QPi (cf. 1.2), and suppose
that we are given a hyperbolic curve (Xi)K i over K i . Let us fix a K i-
valued basepoint for (Xi)K i and denote the resulting fundamental group
7l"l((Xi)KJ by II(xi)Ki ' Suppose, moreover, that we are given an isomor-
phism ax : II(Xl)K 1 ~ II(X2)K2' which, by Lemma 1.3.8, necessarily fits
into a commutative diagram
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 that Pl = P2; set P ~ Pl = P2. That
such a diagram necessarily arises "geometrically" follows from the main
theorem of [Mzk6] (cf. Theorem 1.3.4) - if one assumes that aK arises
geometrically (i.e., from an isomorphism of fields Kl ~ K2)' In this ,
we would like to investigate what one can say in general (i.e., without
assuming that aK arises geometrically) concerning this sort of commu-
tative diagram. In some sense, all the key arguments that we use here
are already present in (Mzk4j, 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)K; by (gi, ri). Also,
we shall denote the geometric fundamental group by
~x; ~f Ker(II(x;)Ki - GK;)

and the unique proper smooth curve over Ki that compactijies (Xi)K;
by (1'i)Ki'
Lemma 2.1 (Group-Theoreticity of Stability). (Xl)Kl 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)K i has stable reduction if and only if the
inertia subgroup of G Ki acts unipotently on dy~ Z' and trivially on the
(finite) set of conjugacy classes of inertia groups of cusps in dx; (i.e.,
the set of cusps of (Xi)Ki K; Ki - cf. Lemma 1.3.9). [Note that this
condition is group-theoretic, by Proposition 1.2.1, (ii).] 0
Absolute an abelian geometry 99
Now let us assume that (Xi)Kt has stable reduction over OK. Denote
the stable model of (Xi)Ki over OKi by:
(Xi)OK' ~ 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
II(Xi)Ki - II~jKt
whose finite quotients correspond to (subcoverings of) admissible cover-
ings of the result of base-changing (Xi)OK. to rings of integers of tamely
ramified extensions of Ki. In particular, we have a natural exact se-
quence:
1 ~ ~adm ~ IIadm ~ G10g ~ 1
Xi (Xt}Ki ki
(where ~~m is defined so as to make the sequence exact). Moreover,
II~jKi itself admits a natural quotient (cf. [Mzk4], 3)
II _ IIadm _ net
(Xt}Ki (Xi)Ki (Xi)Ki
whose finite quotients correspond to coverings of (Xi)Ki that extend to
finite etale coverings of (Xi)OK. which are tamely ramified at the cusps.
In particular, we have a natural exact sequence:
(where ~;ii is defined so as to make the sequence exact).
Lemma 2.2 (Admissible and Etale Quotients). (i) The profinite

groups n(Xi)Ki ' II~jKi' and ~~m are all slim.
(ii) The morphism ax is compatible with the quotients
n _nadm _net
(Xi)Ki (Xi)Kt (Xt}K i
ojn(Xt)Kt '
Proof. We begin with assertion {i}. The slimness of II(Xi)Kt follows
formally from Theorem 1.1.1, (ii); Lemma 1.3.1. Next, let us assume
that ~ ~m has already been shown to be slim. Then it follows formally
from Proposition 1.2.3, (iii), that n~jKt is slim.
100 S. MOCHIZUKI
Thus, to complete the proof of fhsertion (i), it suffices to show that

~~m is slim. We argue as in Lemma 1.3.1, i.e., we consider the covering
associated to an open normal subgroup H ~ ~ ~m. To simplify notation,
we assume that (Xihi is "that covering"; then it suffices to show (cf.
the proof of Lemma 1.3.1) that any automorphism (ki-linear) a of (Xi)ki
which acts trivially on (~x~m )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)k i have genus;:::: 2.
Then by [Mzk4], Proposition 1.3, it follows that a acts trivially on the
set of geometric irreducible components of (Xi)k i . In particular, a acts
on each such geometric irreducible component individually. Thus, by
considering the action of a on the torsion points of the Jacobians of
each these geometric irreducible components of (Xi)k i (cf. the proof of
Lemma 1.3.1), we conclude that a acts as the identity on each of these
geometric irreducible components, as desired.
Next, we turn to assertion (ii). For IT(1jKi' this follows (essentially)
from Proposition 8.4 of [Mzk4] (together with Lemma 1.3.9, which shows
that ax preserves the pro-p inertia groups associated to the cusps in
IT(Xi)KJ Of course, in [Mzk4], Kl = K2 and aK is the identity, but in
fact, the only property of aK necessary for the proof of [Mzk4], Proposi-
tion 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 variety over a p-adic local field) - is that aK preserve the in-
ertia 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 IT(~;)Ki follows (es-
sentially) from [Mzk4J, Proposition 3.2, at least in the case ri = O. That
is to say, even though aK is not necessarily the identity in the present
discussion, the only properties of aK 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 ~ ITf1-':)Ki. Then observe that
there exists an open normal subgroup H' ~ II(i':)Ki such that the orders
of the finite groups ITf1-':)K/H, ITf1-':)K/H' are relatively prime and such
that the covering of (Xi)k i determined by H' is sturdy. Moreover, [by
the assumption of relative primeness] it follows easily that the covering
determined by H is "of etale type" (i.e., arises from a quotient of IT(t)Ki )
if and only if it becomes a covering "of etale type" after base-change via
the covering determined by H'. Thus, we conclude that we may assume
without loss of generality that (Xi)kt is sturdy.
Now, let us observe that since (Xi)kt is sturdy, it admits admissible
coverings "of etale type" with arbitrarily larye prime-to-p ramification
at the cusps. Thus, [since the condition that an admissible covering be
"of etale type" amounts to the condition that there be no ramification at
the nodes or at the prime of OKt] we conclude that the covering of (Xi)k t
determined by H is of etale type if and only if, for every open normal
H" ~ II(t)Ki which has ramification index divisible by the prime-to-p
portion of the order of the finite group ITad(Xmi ) Ki / H at all of the cusps,
it becomes a covering of etale 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
etale 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 ax.]
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 etale type reduces - by Lemma 1.3.9,
which shows that ax preserves the quotientIT(Xi)Ki - IT(l'i)Ki - to
the "ri = 0" case, which has already been resolved. This completes the
proof of assertion (ii) in the case ri > O. 0
Lemma 2.3 (Group-Theoreticity of Dual Semi-Graphs of the
Special Fiber). The morphism ax induces an isomorphism
a x,r c re
(Xl),.:l
~re(X2)k2
between the "dual semi-graphs with compact structure" (i. e., the usual
dual graphs r (Xi)ki' together with extra edges corresponding to the cusps
- cf the Appendix) of the special fibers (Xi)ki of (Xi)OKi. Moreover,
aX,rc is functorial with reS1)ect to passage to finite etale coverings of the
(Xi)Ki
Proof. Indeed, if one foryets about the "compact structure", then this is a
formal consequence of Lemma 1.3.9 [which shows that ax preserves the
quotient II(XdKi - I1(Yt)KJ; Lemma 2.2, (ii), and the theory [concerning
the ri = 0 case) of [Mzk4), 1 - 5, summarized in [Mzk4), Corollary 5.3.
Even though elK is not necessarily the identity in the present discussion,
the only properties of elK that are necessary for the proof of [Mzk4],
Corollary 5.3 are Proposition 1.2.3, (i)j Proposition 1.2.4, (ii) (of the
present paper). That is to say, the point is that the Probenius element is
preserved, which means that the weight filtrations on l-adic cohomology
(where 1 is a prime distinct from p) are, as well.
102 S. MOCHIZUKI
The compatibility with the "compact structure" follows from the pro-l
(where 1 =1= 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
Ki is contained (up to conjugacy) in the decomposition group of a
dm)
rra(X
unique irreducible component of (Xi)ki' 0
Next, we would like to show that ax is necessarily "of degree 1". This
is essentially the argument of [Mzk4J, Lemma 9.1, but we present this
argument in detail below since we are working here under the assumption
that aK 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)OK', is singular and sturdy (cf.
[Mzk4J, 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 f(Xi)ki - i.e.,
this semi-graph is not a tree.
(These conditions may always be achieved by replacing (Xi)Ki by a finite
etale covering of (Xi)K i - cf. [Mzk4], Lemma 2.9; [Mzk4], the first two
paragraphs of the proof of Theorem 9.2.)
We begin by introducing some notation. Write:
TJ: ~f Aab.
Vi - ~Xi'
H def Hsing (r C '71)
i = 1 (X;)k i ' /U = Hsing
1
(
r(Xi)k i '
Z)
(where "H~ing" denotes the first singular homology group). Thus, by

considering the coverings of (Xi)OK, induced by unramified coverings of
the graph r(Xi)k", we obtain natural (group-theoretic.0 "combinatorial
quotients":
def ~
Vi - (Hi)Z = Hi Z
Lemma 2.4 (Ordinary New Parts, after Raynaud). For a "suf-
ficiently large prime number l" (where "sufficiently large" depends only
on p, gi), and after possibly replacing Ki by a finite unramified exten-
sion of K i , there exists a cyclic etale covering (Zi)OK ---+ (Xi)OK, of
1;f A ab
.
(X)o Ki 01 degree 1 such that the "new part" v;new (Zi)Ki /A (Xi)Kj
ab
~ l
of the abelianized geometric fundamental group of (Zi)Ki satisfies the

following:
(i) We have an exact sequence:
o ---+ Vim1t ---+ (Vinew)zp ~f Vinew z Zp ---+ Viet1 -> 0
- where Viet! is an unramifted GKi -module, and Vimlt is the

Cartier dual of an unramified GKi-module.
(ii) The "combinatorial quotient" of ~(~i)Ki (arising from the cover-
ings of (Zi)OK, induced by unramified coverings of the dual semi-
, induces a nonzero quotient
graph of the special fiber of (Zi)OK.J
ltjnew _ (H~ew)~ oJltjnew
z z Z l'
Here, the injection ~(t)Ki '-+ ~(~i)Ki is the injection induced by pull-
back via (Zi)Ki ~ (Xi)Ki and Poincare 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 [Mzk4J, 8, of "Vo", "Voord"
- 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 ordi-
nary. That this is possible for I sufficiently large is a consequence of
a theorem of Raynaud (as formulated, for instance, in [Tarnal], Lemma
1.9). Then let us observe that, so long as we choose the etale covering
(Zi)OK , ~ (Xi)OK , so that it is nontrivial over every irreducible com-
ponent of (Xi)k i , condition (ii) is automatically satisfied: Indeed, if we
write hi ~f rkz(Hi ) - so hi > 0 since rc(x.) is assumed to be non con-
ki
tractible - then to assert that condition (ii) fails to hold - i.e., that
there are "no lIP.W cycles in the dual graph" - is to assert that we have
an equality of Euler characteristics:
(L gZ,j) + hi - 1 = 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)OK .
(respectively, (XdOK.))' But, since

L (gZ,j -1) = L l(gx,j -1)
j j
we thus conclude that (l - 1) = {Lj (l - 1)} + hi(l - 1), hence that

1 = (L: j 1) + hi - which is absurd, since both the sum and hi are? 1.
Remark 2.4.1. The author would like to thank A. Tamagawa for ex-
plaining to him the utility of Raynaud's theorem in this sort of situation.
104 S. MOCHIZUKI
In the following discussion, to keep the notation simple, we shall re-

place (Xi)Ki by some (Zi)K, as in Lemma 2.4. Thus, Vinew is a GKi-
quotient module of Vi. Moreover, we have a surjection
such that the quotient (Hi)Z -- (HrW)z is defined over Z, Le., arises
from a quotient Hi -- Hrw. (Indeed, this last assertion follows from
the fact that the quotient Hi -- Hr w arises as the cokernel (modulo
torsion) of the morphism induced on first singular cohomology modules
by a finite (ramified) covering of graphs - i.e., the covering induced on
dual graphs by the covering (Zi)ch -+ (Xi)OK of Lemma 2.4.)
On the other hand, the cup product on gro~p cohomology gives rise
to a nondegenerate (group-theoretic!) pairing
def ~ ~
(where we think ofViv = Hom(Vi,Z) as Hl(~Xi'Z)), hence, by restric-
tion to (Vinew) v ~ Vi v , a pairing
(Vinew)v z (Vinew)v z JLz(K i ) -+ Mi ~f H2(~Xi,JLz(Ki)) (~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" (Le., a morphism which is an iso-
morphism over Q):
(Vinew)v z JLZ(Ki) z M iv ~ Vinew
Thus, if we take the dual of the surjection discussed in the preceding

paragraph, then we obtain an inclusion
(HrW)~ JLZ(Ki) M iv ~ (Vinew)v JLZ(Ki) M/ ~ Vinew
which (as one sees, for instance, by applying the fact that JLZ(Ki)GK i =
0) maps into the kernel of the surjection Vinew -- (HrW)z.
Next, let us observe that the kernel Ni of the surjection of unramified
GK;-modules (Le., Gki-modules)
Vietl - - (HrW)zp
satisfies:
(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 G Ki -equivariant splitting (HrW)zp ~ (l!ietl)iQp. Similarly, (by
taking Cartier duals) the injection (HrW)~/9I'zp(Ki)MiV ~ l!imlt also
admits a unique GKi-equivariant splitting over Qp. Thus, by applying
these splittings, we see that the GKi-action on (l!inew)zp determines a
p-adic extension class
(r/i)zp E {(HrW)~ }2 M iv (HI (Ki' I'z (Ki))/H} (Ki' I'z(Ki)))
= {(HrW)~}2 M/
where (by Proposition 1.2.1, (vii)) Hl(Ki ,l'z(Ki )) may be identified
with (KtY', and we define
Hj(Ki,l'z(K i )) ~f 0Ki ~ (Kt)" ~ H 1 (Ki ,l'z(Ki ))
so the quotient group (Hl(Ki' I'z(Ki))/Hj(Ki , I'Z(Ki))) may be iden-
tified with Z.
Next, let us observe that the kernel NI of Cl!inew)z, - (HrW)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 G Ki -modules
(Hrw)'i, I'z,(Ki) M/ ~ 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 Gkt acts on the smaller module (respectively, quotient by this
smaller module) with weight 1 (respectively, !). Thus, just as in the
p-adic case, we may construct a prime-to-p-adic extension class (rli)z'
from the GKi-action on (l!inew)z" which, together with (TJi)Zp, yields an
extension class (cf. [FC], Chapter III, Corollary 7.3):
TJi E {(HrW)~}2 M iv {H 1 (Ki , I'Z(Ki))/H} (Ki' I'z(Ki))} Q
= {(Hrw)~}2 M/ Q
That is to say, TJi may be thought of as a (group-theoretically recon-
structible.0 bilinear form:
(-, -)i : (Hr W )f2 ---. (MnQ ~f M/ Q
106 S. MOCHIZUKI
Moreover:
Lemma 2.5. Assume that (Xi)c'K arises as some "(ZdOK" as in
Lemma 2.4. T h e n : ' ,
(i) (Positive Rational Structures) The image of (Hrw)2 under
the morphism (HrW)~2 -+ (MnQ forms a rank one Z-submodule of
(MnQ' Moreover, for any two nonzero elements a, b E Hi, (a, a)i dif-
fers from (b,b)i by a factor in Q>o (i.e., a positive rational number).
In particular, this image determines a i(ho-structure" on (MnQ, i.e.,
a Q-rational structure on (MiV)Q' together with a collection of genera-
tors of this Q-rational structure that differ from one another by factors
in Q>o. Finally, this Q>o-structure is the same as the Q>o-structure
on M/ determined by the first Chern class of an ample line bundle on
(Xi)Ki in Mi = H2(DoXi,J-tz(Ki)).
(ii) (Preservation of Degree) The isomorphism
induced by ax preserves the elements on both sides determined by the

first Chern class of a line bundle on (XdKi of degree 1.
Proof. Indeed, assertion (i), follows formally from [FCl, 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. 0
Remark 2.5.1. Note that the conclusion of Lemma 2.5, (ii), is valid
not just for (Xi)Ki , but for any finite etale cover of the original (Xi)Ki'
i.e., even if this cover does not arise as some "(Zi)OK" , as in Lemma
2.4. Indeed, this follows from the fact that the crucial i~Qbo-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>o!).
Remark 2.5.2. In the discussion of [Mzk4J, 9, it was not necessary
to be as careful as we were in the discussion above in constructing the
p-adic class (T]i)Zp (Le., "/-Lp" in the notation of loco cit.). This is because
in loco cit., we were working over a single p-adic base-field "K". In this
more restricted context, the extension class (T]i)Zp may be extracted much
more easily from Vi by simply forming the quotient by the submodule
of Hl(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 formal group associated to the p-divisible group
determined - via "Tate's theorem" (cf. Theorem 4 of [Tate]) - by the
GKi-module Ker((Vih~p - (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 "(lbo-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 (XlOg)OKi for the log scheme obtained by equip-
ping (Xi)OK. with the log structure determined by the monoid of reg-
E O(Xi)c:) which are invertible on the open subscheme
ular functions
Ki
(Xi)Kt ~ (.:ti)OK.' Thus, in the terminology of [Kato2j, (X;,log)OK' is

log regular. Also, let us write (X10g)ki for the log scheme obtained by
equipping (Xi)ki with the log structure determined by restricting the log
structure of (X10g)oK" Thus, the quotient II(xi)K. - IIad(xm)Ki deter-
mines a "universal admissible covering"
-log)
(X (Xlog)
i ki- i ki
of (XlOg)kt'
Now let us choose a connected component Ii of the ki-smooth locus
(Le., the complement of the nodes) of (XlOg)ki' Write Ii ~ (Xi)ki for the
image of Ii in (Xi)k i Thus,
is a ''tame universal covering" of Ii (i.e., a universal covering of the

hyperbolic curve Ii among those finite etale coverings that are tamely
ramified at the "cusps" of this hyperbolic curve). In the following dis-
cussion, 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 IIzt of this covering may also be thought of as
the quotient of the decomposition group in n(1-jKi of Ii by its inertia
group. In particular, since nIi is formed by taking the quotient by this
)Ki - G~g
inertia group, it follows that the surjection nad(Xm i
induces a
108 S. MOCHIZUKI
natural surjection
IIIi ---++ Gk i
whose kernel is the geometric (tame) fundamental group 7l'iame ((Ii )k) of
Ii
Finally, we observe that it makes sense to speak of Ii and I2 as cor-
responding via ax. Indeed, by Lemma 2.3, ax induces an isomorphism
between the pro-graphs determined by the (XfOg)ki' Thus, the Ii may
be said to correspond via ax when the vertices that they determine in
these pro-graphs correspond. Moreover, when the Ii correspond via a~,
it follows (by considering the stabilizer of the vertex determined by Ii)
that ax induces ~ bijection between the respective decomposition groups
Vi in II(~JKi of Ii, as well as between the respective inertia subgroups
of these decomposition groups Vi (which may be characterized group-
n
theoretically as the centers of the subgroups Vi Ker(n~jKi ---++ Gki)'
since 7l'iame ((Ii )k) is center-free - cf. Lemma 1.3.10). Thus, in sum-
mary, ax induces a commutative diagram:
We are now ready (cf. [Mzk4], 7) to apply the main result of [Tarnal].
This result states that commutative diagrams as above are in natural
bijective correspondence with commutative diagrams
Il '"
--+ I2
1 '"
1
Ii --+ I2
lying over commutative diagrams
kl '"
--+ k2
1 '"
1
kl --+ k2
(cf. Theorem 1.3.11). In particular, these commutative diagrams induce
an isomorphism
(where "H;" denotes etale cohomology with compact supports - cf.

[Milne], Chapter III, Proposition 1.29; Remark 1.30) which maps the
element "I" (Le., the element determined by the first Chern class of
a line bundle of degree 1) on the left to the element "I" on the right.
(Indeed, this follows from the fact that the morphism It ~ I2 appearing
in the above commutative diagram is an isomorphism, hence of degree
1.) Note that the isomorphism I-'z,(k t ) ~ I-'z,(k2) 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)OK_, arises as some U(Zi)OK_"
, as in Lemma
2.4. Then the following diagmm
I-'Z!(k!)
I
.J..
I-'Z,(Kl)
- in which the vertical morphisms are the natural ones (obtained by con-
sidering Teichmiiller representatives); the upper horizontal morphism is
the morphism determined by Theorem 1.3.11; and the lower horizon-
tal 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:
H;CCI!)kl'I-'Z,(k 1)) H;( (I2)kY I-'z,(k2))

2 - 2 -
He C(Idk1 , I-'z,(Kl)) '"
-+ He ((I2)k2' I-'z,(K 2))
Moreover, we have a diagram
where the horizontal morphisms are induced by aK (cf. Proposition 1.2.1,

(vi)), and the vertical morphisms are induced "group-theoretically" as
follows: First, observe that [since we continue to opemte under the as-
sumption (*CSSN)]
while
110 S. MOCHIZUKI
(where we write Ii for the unique llonsingular compactification of Ii).

Moreover, since we continue to operate under the assumption (*CSSN), it
follows (cf. the discussion of "Second Cohomology Groups" in the Ap-
pendix) that the natural "push-forward" morphism in etale cohomology
(H2(7rl(~Iih;), J.tz,(Ki)) ~) H;((Iih~i' J.tZ,(Ki))

2 - - ,....,2et -
-+ H ((Xi)ki ki ki' J.tz,(K i )) (= H (Ll Xi ' J.tz,(Ki)))
may be reconstructed group-theoretically (by using the various natural
homomorphisms 7r1 ((Ii)i.;J -04 Ll)ii [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
H2(Ll3li , J.z,(K i )) -+ H2(Ll~m, J.z,(K i ))
induced by the surjections ~~m - Ll)ii [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:
H;((I2 ):kr J.tz,(k 2 ))
H2((X2) K2' J.tz,(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 mor-
phism of (*3)]; and the discussion of "Second Cohomology Groups" in
the Appendix [applied to the vertical morphisms of (*3)], all of the mor-
phisms of this diagram are compatible with the elements "I" 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 free Z/-modules). On the other hand,
since diagram (*1) was obtained by applying the functors H;((Iih i , - )
(which are manifestly faithful, at least when restricted to constant coef-
ficient 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. D
The significance of Lemma 2.6 from our point of view is the following:
Lemma 2.6 implies that aK induces an isomorphism
log '" Clog
C kl
o:dkog : -+ k2
which satisfies the hypothesis of Proposition 1.2.4, (iii). Thus, we con-

clude from Proposition 1.2.4, (iii), that adog arises geometrically. In
k
particular, it follows that we may apply {Mzk4j, Theorem 7.2, to the
commutative diagram
'"
-+
(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:
'"
-+
'"
-+
Moreover, the isomorphism exhibited in the upper horizontal arrow of

this commutative diagram is easily seen to be functorial with respect
to arbitrary finite etale coverings of the (Xi)Kt (Le., not just coverings
that arise from finite etale coverings of the (Xi)OKt)' Indeed, this func-
toriality follows formally from the uniqueness assertion in Proposition
1.2.4, (iii), and the fact that dominant (i.e., not just finite etale) mor-
phisms between proper hyperbolic curves in characteristic p may be dis-
tinguished by considering the morphisms that they induce between the
respective Jacobians, hence, in particular, by the morphisms that they
induce between the I-power torsion points (where 1 =F p) of the respective
Jacobians. [Note also that the morphisms between log structures at the
nodes are uniquely determined by the morphisms between tl:" various
normal bundles on the branches at the nodes - cf. [Mzk3], 3.7, 3.8,
3.10.] Thus, in summary:
Theorem 2.7 (Group-Theoretic Reconstruction of the Logarith-
mic Special Fiber of a p-adic Hyperbolic Curve). Let p be a prime
number. Fori = 1,2, let Ki be afi.'f1tte extension ofQpl and (Xi)Ki a hy-
perbolic curve over Ki whose associated pointed stable curve has stable
112 S. MOCHIZUKI
reduction over OKi . Denote the resulting "stable model" of (Xi)Ki over
OKi by (Xi)OK' , Assume that we have chosen basepoints of the (Xi)Ki
(which thus induce basepoints of the Ki). Then every isomorphism of
profinite groups II(Xl)K 1 2:; II(x2)K2 induces commutative diagrams:
Moreover, the latter commutative diagram (of admissible quotients

II(~JKi of the II(x;)K; lying over the tame Galois groups G~g of the K i)
necessarily arises from unique commutative diagrams of log schemes
where the commutative diagram on the left lies over the commutative dia-
gram on the right (which is as in Proposition 1.2.4, (iii)). Here, we equip
Spec(OKJ (respectively, Spec(ki ); (Xi)OKo; , (Xi)k) with the log structure
determined by the closed point (respectively, determined by restricting
the log structure on Spec( OKJ; determined by the monoid of functions
invertible on the open subscheme (Xi)Ki; determined by restricting the
log structure on (Xi)OK) and denote the resulting log scheme by O~;
(respectively, k!og; (Xl og )oKi ; (Xl og )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)OKo, arise as some "(Zi)OK" , as in Lemma 2.4 - were applied
only to show that the hypotheses of Proposition 1.2.4, (iii) (and [Mzk4j,
Theorem 7.2) are satisfied. Moreover, we observe that although [Mzk4J,
Theorem 7.2, is only stated in the proper singular case, it extends imme-
diately 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 (XdK; is proper, with good reduction - one may
reconstruct the logarithmic special fiber in a functorial fashion (Le., with
respect to finite etale coverings of the (Xi)K;), 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
[Tarnal], Theorem 4.3, is only stated in [Tarnal] 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 ob-
tained by simply pulling back the log structure of k~Og. Thus, it suffices to
construct the (non-logarithmic, scheme-theoretic) special fiber. Next, we
observe that (after possibly enlarging K i ) there exist - cf., e.g., [Mzk4],
the first two paragraphs of the proof of Theorem 9.2 - corresponding
finite Galois etale coverings (Zi)K; -+ (Xi)K; (for i = 1,2), where (Zi)K;
is a hyperbolic curve over Ki with bad stable reduction (Zi)OK', over OK;.
Thus, by applying Theorem 2.7 to (Zi)K; allows us to reconstruct the
logarithmic special fiber (Z;Og)ki' together with the action of the Galois
group C i ~f Gal((Zi)K;/(Xi)KJ. Note that irreducible components of
(Zi)k; that dominate (Xi)k; 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)k Let
us choose corresponding (closed, proper) irreducible components
that dominate (hence surject onto) (Xi) k; . Denote the decomposition

(respectively, inertia) group associated to Gi by Di ~ C i (respectively,
Ii ~ Di ~ Cd. Thus, Dd Ii acts faithfUlly on Gi , and the order IIil of Ii
is a power of p, equal to the degree of inseparability of the function field
of Gi over the function field of (Xi)ki' Then we may reconstruct (Xi)k;
as a finite flat quotient of Gi by considering the sub sheaf
(Le., the Di-invariants of the sub algebra Og;1 ~ OC;, where we use that
IIil is a power ofp). By applying the functoriality with respect to finite
etale coverings of (Xi)K; observed in the discussion immediately preced-
ing the statement of Theorem 2.7, we conclude that this construction
of (Xi)k; is independent of the choice of (Zi)Ki , Gi , and itself functorial
with respect to finite etale coverings of (Xi)K;.
This completes our reconstruction of the logarithmic special fibers of
the (Xi) K;, in a fashion that is functOrial with respect to finite etale cov-
erings of the (Xi)K;. Thus, we conclude, in particular, (from this func-
toriality, 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 coin-
cides with the original given morphism between admissible fundamental
groups. This completes the proof of Theorem 2.7. 0
114 S. MOCHIZUKI
Remark 2.7.1. Given data as in Theorem 2.7, one may consider the
outer Galois representation
GK; - Out(b.x;)
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(GK;) in Out(b.x;)?
Although Theorem 2.7 does not give a complete explicit answer to this
question, it tells us that, at any rate, elements of this commensura-
tor (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 K i , i.e., are "geometric") one obtains that elements
of this commensurator do induce "self-isogenies" of k!og. [Here, by a
"self-isogeny of an object", we mean an isomorphism between two finite
etale 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(GK;) in Out(b.x;) consists precisely of those (finitely many) auto-
morphisms that arise geometrically (i.e., from automorphisms of (Xi)K i ),
it follows that a "self-isogeny" of G Ki induced by an element of this com-
mensurator corresponds to (up to finitely many well-understood pos-
sibilities) an essentially unique element of this commensurator. This
motivates the point of view that:
The "self-isogenies" of GKi defined by elements of this com-
mensurator - which we shall refer to as quasi-conformal
self-isogenies of GK, - 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 theorem 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
klog '" k10g
1 - 2
(such as those arising from "quasi-conformal isomorphisms" GKI ~ 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 Kl ~ K 2).

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 Kl ~ K2). Whether or not, however, this sort
of phenomenon actually takes place in the case of "quasi-conformal iso-
morphisms" as in Theorem 2.7 is not clear to the author at the time of
writing.
Remark 2.7.S. The theory of the present prompts the question:
Do isomorphisms II(Xt)Kl ~ II(x2)K as in Theorem 2.7
only preserve the logarithmic special fiter or do they preserve
other information as well concerning 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 licanonical", then this canonicality is preserved by iso-
morphisms 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 liftingsj
(2) liarithmetic hyperbolic curves", i.e., hyperbolic curves isogenous to
a Shimura curve [that is to say, curves which admit a finite etale
covering which is isomorphic to a finite etale covering of some
Shimura curve]j
(3) canonical liftings in the sense of 'p-adic Teichmiiller 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 lidefined over IP\" - i.e., a hypothetical (but, of course, fictional!)
absolute field of constants sitting inside Zp.
Acknowledgements
I would like to thank A. Tamagawa for the time that he so generously
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 1.1j (ii) explaining to me the utility of a theorem of Raynaud in
116 S. MOCHIZUKI
the context of 2 (cf. Lemma 2.4). Also, I would like to thank F. Dort,
as well as the referee, for various useful remarks.
Appendix: Terminology of Graph Theory

The Notion of a Semi-Graph:
We shall refer to as a semi-graph r the following collection of data:
(1) a set V - whose elements we refer to as "vertices";
(2) a set - whose elements we refer to as "edges" - each of whose
elements e is a set of cardinality 2 satisfying the property "e # e' E
===} ene' = 0";
(3) a collection ( of maps (e [one for each edge e]- which we refer to
as the "coincidence maps" - such that (e : e -+ V U{V} [where
we note that V n{V} = 0 since V rf. V] is a map from the set e to
the set V U{V}.
We shall refer to the subset (;l(V) ~ e [Le., the inverse image of the
subset V ~ V U{V} of elements # V] as the verticial portion of an edge
e; to the restriction of (e to the verticial portion of e as the verticial
restriction of (e; and to the cardinality of the verticial portion of e as
the verticial cardinality of e. A graph r is a semi-graph r 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 bEe as a branch of the edge e.
Let r = {V, , (} be a semi-graph. If e E is an edge of r of verticial
cardinality 2 whose image via (e consists of two (not necessarily distinct)
elements VI, V2 of V, then we shall say that e joins VI to V2. If V is any
vertex in the image of (e, 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
r = {V,,O -+ r' = {V',',('}

is a collection of maps V -+ V'; -+ '; 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 Proftnite Groups:
We shall refer to the following data G:
(i) a finite semi-graph r;
(ii) for each vertex v of r, a profinite group G v ;
(iii) for each edge e of r, a profinite group Ge , together with, for each
branch bEe mapping to a vertex v via (e, a continuous homomor-
phism b* : Ge -+ Gv
as a semi-graph of profinite groups. When r 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
IIG - namely, the profinite analogue of the well-known construction
of the fundamental group associated to a (semi-)graph of groups (cf.
[Serrel], I, 5.l). One way to describe this profinite analogue is via
Galois categories (cf., e.g., [SGAl] 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:
b* : Cv ~f B(Gv ) -+ Ce ~f B(Ge )
At any rate, to define IIG, it suffices to define a Galois category CG

[which we would like to think of as "B(IIG)"]. We define CG to be the
category of collections of data:
Here, "v" (respectively, "e") ranges over the vertices (respectively, edges
of verticial cardinality 2) of r; Sv is an object of Cv ; and Ce : bHSVl) ~
b2(SV2) (where e = {bI, b2 }; Vi ~f (e(b i ), 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.
118 S. MOCHIZUKI
Pointed Stable Curves:

Let k be an algebraically closed field. Let g, r ~ 0 be integers such
that 2g - 2 + r > O. 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:
X~f X\D
Also, let us write X' ~ X for the complement of the nodes in X.
Next, we define the dual graph
rx
of X. The set of vertices of r x 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 r x is the set of nodes
of X. To avoid confusion, we shall write Ve for the node corresponding
to an edge e. The node Ve has two branches ve [!] and ve [2], each of
which lies on some well-defined irreducible component of X. We take
e ~f {lie [1], lie [2]} [so e is of verticial cardinality 2J, and let (e be the map
that sends lIe[i] to the irreducible component on which the branch ve[iJ
lies.
n
Let us write X~ ~f Xv X' ~ X, and (for i = 1, 2) X~e[i] for the
scheme-theoretic intersection with X' of the completion of the branch
vetil at the node Ve. Thus, X'lie [']t is noncanonically isomorphic to
Spec(k[[t]][C I ]) (where t is an indeterminate).
In the following discussion, we would like to fix (k-linear) isomor-
phisms:
X'lie [I] '"
= X'lie [2]
via which we shall identify X~e[l] with X~e[2] and denote the resulting
object by X~. In particular, we have natural morphisms X~ - X~l'
X~ - X~2 (where Vi ~f (e(ve[i]), for i = 1,2). One verifies immediately
that the induced morphism on tame algebraic fundamental groups [Le.,
corresponding to coverings which are tamely ramified at the nodes and
marked points] is independent of the choice of isomorphism. Thus, the
dual graph rx, together with the result of applying "1l'iame ( -)" [more
precisely: the tame algebraic fundamental group functor, together with
choices of basepoints for the various X~, X~, and choices of paths to
relate these basepoints via the natural morphisms X~ - X~j discussed
above] to the data {X~;X~;X~ _ X~} determines a graph of profinite
groups 9x associated to the stable curve X.
When considering the case of a curve with marked points (Le., r > 0),
it is useful to consider the following slightly modified "data with compact
structure": Let us denote by r~ the semi-graph obtained from rx by

appending to r x, for each marked point x E X, the following:
an edge ex ~f {x}, where x E ex is sent by (e", 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 r x to form
r3c as the marked edges of r3c and to r3c itself as the dual graph with
compact structure associated to X.
If, moreover, we associate to ex the scheme X~ (which is noncanon-
ically isomorphic to Spec(k[[t]][rl])) obtained by removing x from the
completion of X at x, and apply "7riame ( -)" to the natural morphism
X~ -+ X~"" then we obtain a natural semi-graph of profinite groups Q,k
with underlying semi-graph r3c. Moreover, one checks easily that when k
is of characteristic 0, the profinite group associated (as described above)
to Qx or Q,k is isomorphic to "fig,r", Le., 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 X~ as Riemann surfaces
and of the X;, X~ as "copies of the circle b, then we see that this
construction corresponds quite geometrically to gluing Riemann surfaces
with boundary along copies of the circle.
Finally, let us observe the following (valid for k of arbitrary charac-
teristic): The profinite group associated to Qx or Q,k may be identified
with the admissible fundamental group
II~m
of (X, D) (cf. [Mzk4], 2, 8; [Mzk3], 3). Moreover, if we define semi-

graphs of groups Q~, Q~et by taking the profinite group at a vertex
v to be the tame fundamental group of Xv [Le., instead of X~] and the
profinite group at all of the edges to be trivial, then the resulting profinite
group may be identified with the fundamental group
II~
associated to the Galois category of finite etale coverings of X which are
tamely ramified at the cusps. In particular, we have a natural surjection
II~m _ II~.
Second Cohomology Groups:

We continue with the notation of the above discussion. Let 1 be a
prime number distinct from the characteristic of k. In the following
120 S. MOCHIZUKI
discussion, we shall use the notation "Hi (-)" (respectively, "Ht (-)")
to denote the j-th etale cohomology module (respectively, j-th etale co-
homology module with compact supports - cf. [Milne], Chapter III,
Proposition 1.29; Remark 1.30) with coefficients in 'Lz.
If v is a vertex of r x, write X v for the normalization of the corre-
sponding irreducible component of X, so X~ = X' x xXv. Then we have
a natural ''push-forward'' isomorphism
H;(X~) ~ H2(Xv)
between free 'lz-modules of rank 1. Moreover, the natural morphisms

X v -> X (as v ranges over the vertices of r x) determine natural restric-
tion morphisms whose direct sum is easily verified to be an isomorphism:
H2(X) ~ EB H2(Xv)
v
On the other hand, the composite of the natural "push-forward" mor-

phism H;(X~) -> H2(X) with a restriction morphism H2(X) ->
H2(Xw), where w i= 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 X~: This Chern class is the image of a generator of
H;(X~) and, moreover, vanishes upon restriction to Xw since the re-
striction to X w of this line bundle is clearly trivial.] In particular, we
conclude that the direct sum of the natural "push-forward" morphisms
H;(X~) -> H2(X) yields an isomorphism
EB H;(X~) ~ H2(X)
v
which, if we identify H;(X~), H2(Xv) via the isomorphism H;(X~) ~

H2(Xv), is inverse to the isomorphism H2(X) ~ E9v H2(Xv).
Put another way, if we identify the cohomology modules H;(X~),
H2(Xv) via the isomorphism H;(X~) ~ H2(Xv), then the natural
"push-forward" morphism H;(X~) -> H2(X) may be recovered as the
composite
v
of the natural inclusion of 8, direct summand with the inverse of the
isomorphism H2(X) ~ EDv H2(Xv) determined by the restriction mor-
phisms.
Finally, let us observe that if X is sturdy (cf. [Mzk4], Definition 1.1) -
i.e., every Xv has genus ~ 2 - then the natural morphisms from profi-
nite group cohomology to etale cohomology give rise to a commutative
diagram
in which the horizontal morphisms are isomorphisms [since arbitrary

etale cohomology classes with finite i-power torsion coefficients vanish
upon restriction to some finite etale covering of X or XvJ.
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.
[HRJ W. Herfort and L. Ribes, Torsion elements and centmlizers in free products
0/ profinite groups, J. Reine Angew. Math. 358 (1985), 155-161.
[Katol] K. Kato, Logarithmic Structures of Fontaine-Illusie, in "Proceedings of the
First JAMI Conference," 191-224, Johns Hopkins Univ. Press, 1990.
[Kato2J K. Kato, Toric Singularities, Amer. J. Math. 116 (1994), 1073-1099.
[KnudJ 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 bmid groups on curves,
J. Reine Angew. Math. 474 (1996), 169-219.
[Milne) J. S. Milne, "Etale Cohomology," Princeton Mathematical Series, 33.
Princeton University Press, 1980.
[MzklJ S. Mochizuki, A Theory of Ordinary p-adic Curves, Publ. of RIMS 32
(1996),957-1151.
[Mzk2J S. Mochizuki, "Foundations of p-adic Teichmillier Theory," AMSjIP
Studies in Advanced Mathematics, 11. American Mathematical Soci-
ety jInternational Press, 1999.
[Mzk3J S. Mochizuki, The Geometry of the Compactification of the Hurwitz
Scheme, Publ. of RIMS 31 (1995), 355-441.
[Mzk4) S. Mochizuki, The Profinite Grothendieck Conjecture for Closed HyperboliC
Curves over Number Fields, J. Math. Sci. Univ. Tokyo 3 (1996), 571-627.
[Mzk5) S. Mochizuki, A Version 0/ the Grothendieck Conjecture for p-adic Local
Fields, The International Journal of Math. 8 (1997), 499-506.
[Mzk6) S. Mochizuki, The Local Pro-p Anabelian Geometry of Curves, Invent.
Math. 138 (1999), 319-423.
[Mumf] D. Mumford, "Abelian Varieties," Oxford Univ. Press, 1974.
[NakaI) H. Nakamura, Galois rigidity of pure sphere bmid groups and profinite
calculus, J. Math. Sci. Univ. Tokyo 1 (1994), 71-136.
[Naka2) H. Nakamura, Galois rigidit't) 'J/ profinite fundamental groups [translation
of Sugaku 47 (1995), 1-17J, Sugaku Expositions 10 (1997), 195-215.
122 S. MOCHIZUKI
[NTMJ H. Nakamura, A. Tamagawa, and S. Mochizuki, The Grothendieck Conjec-

ture on the Fundamental Groups of Algebraic Curves, Sugaku Expositions
14 (2001), 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 (J. W.
S. Cassels and A. Frohlich, ed.)," 128-161, Academic Press, 1967.
[SGA1) "Revetement etales et groupe fondamental," Seminaire de Geometrie
Algebrique du Bois Marie 1960-1961 (SGAl), dirige par A. Grothendieck,
augmente de deux exposes de M. Raynaud, Lecture Notes in Mathematics,
224. Springer-Verlag, 1971.
[Tarnal) A. Tamagawa, The Grothendieck Conjecture for Affine Curves, Compositio
Math. 109 (1997), 135-194.
[Tama2) A. Tamagawa, On the Fundamental Groups of Curves over Algebraically
Closed Fields of Characteristic > 0, International Mathematics Research
Notices 16 (1999), 853-873.
[Tate) J. Tate, p-Divisible Groups, in "Proceedings of a Conference on Local
Fields," 158-183, Driebergen, Springer-Verlag, 1967.
Part II
Galois groups and Galois extensions

REGULAR GALOIS REALIZATIONS

OF PSL2(p2) OVER Q(T)
Takehito SHIINA
Mathematical Institute
Tohoku University,
Sendai Miyagi, 980-8578, Japan
s99m 17(Dmath. tohoku.ac.jp
History
The inverse Galois problem which asks whether every finite group
occurs as Galois group over Q is an unsolved problem originally proposed
by Hilbert [8] more than 100 years ago. The first approach succeeded in
the region of class field theory in the middle of the 20th century: Scholz
and Reichardt solved the problem affirmatively for nilpotent groups [15]
and Safarevic extended their result to solvable groups [18]. In contrast,
few Galois realizations were known for non-solvable groups until the
1970's except for the symmetric groups Sn and the alternating groups
An, which had already been realized by Hilbert. The first attack of
non-solvable groups was carried out in the 1970's and many important
results have been obtained for such groups from then on. For example,
using the p-adic Galois representation attached to cusp forms of weight
24, Ribet [17] realized the projective special linear group PSL2(P2) as
Galois group over Q in the case 144169 is non-square modulo p. Recently,
this result was widely extended by Reverter-Vila [16] and Dieulefait-Vila
[3]. Moreover, studying the Jacobian of a plane curve of genus 2, Mestre
[13] showed that the field generated by the p-division points realizes
PSL2(P2) regularly over Q for p == 2 (mod 5). Here a group G is said to
be realized regularly over Q if there exists a Galois extension of a rational
function field Q(T) = Q(tl, ... , t r ), regular over Q, with Galois group
G. If G occurs regularly over Q, then Hilbert's irreducibility theorem
assures that there exist infinitely many linearly disjoint number fields
[received: May 31,2002; accepted in revised form: March 1, 2003]

126 T. SHIINA
with Galois group G. So it is natural to ask the following more difficult

question.
Problem (Regular Inverse Galois Problem). Does every finite
group occur regularly over Q?
With regard to this problem, the most effective approach that has
been known is the rigidity method, which was established by Belyl, Fried,
Matzat, Shih and Thompson. This method is based on the theory of
Galois coverings of Riemann surfaces. By means of the rigidity method,
one can realize a finite group G regularly over Q if G satisfies certain
group theoretical conditions. Many variations of sufficient conditions for
a given finite group to apply the rigidity method are known and many
positive answers for non-solvable groups including 25 sporadic simple
groups have been obtained. Here we make a list of the projective special
linear groups PSL2(pn) which have been realized regularly over Q.
(i) n = 1
p= 2,3,5 (Hilbert, 1892)
(+) = -1, 31 E {2,3, 7} (Shih, 1974)
(!) =-1 (Malle, 1991)
(ii) n = 2
p= 2,3 (Hilbert, 1892)
(!) =-1 (Feit, 1984 / Mestre, 1987)
p=5 (Przywara, 1988)
(iii) n = 3
p=2 (Matzat, 1987)
The purpose of this paper is to show the following:

Main Theorem. If p =f=. 1 (mod 24), then PSL2(P2) occurs regularly
over Q.
This is an extension of our preceding result that PSL2(P2) occurs
regularly over Q for p == 3 (mod 8) [21]. We consider its covering
group P~L2(p2) instead of PSL2(p2), for avoiding some complexity in
the process of calculation. The key of the proof was the existence of
a certain isolated braid orbit to which we applied Matzat's braid orbit
theorem ([10], [12]). By changing some numerical conditions, we will
realize PSL2(p2) for the remaining primes p == 5 (mod 12) in a similar
way to [21].
Remark. After obtaining the result above, we were informed that Det-
tweiler realized the projective symplectic groups PSP2n (p2) regularly
Regular Galois realizations of PSL2 (p2) over Q(T) 127
over Q for odd primes p . 1 (mod 24) and integers n ~ 1. His strategy
is to verify the existence of a rigid braid orbit in PSP2n (P2) by apply-
ing the middle convolution functor to certain tuples of GL2(P2) (see his
paper [2] in this volume).
The content of this paper is organized as follows. The first two sec-
tions aim to present known results with examples of expository nature:
In Section 1, we formulate a system of definitions and notations to state
the basic rigidity theorem and in Section 2, we review the braid orbit
theorem which is the main tool of our proof of Main Theorem. In Sec-
tion 3, some group-theoretic properties of PEL 2(P2) are collected, and
the use of the braid orbit theorem will be exhibited in some easy cases.
In Section 4, we construct a certain nice braid orbit in PEL2(P2), and
settle the proof of Main Theorem.
1. Basic Rigidity Theorem

A finite unramified Galois covering of the punctured projective line
'P: X ~ ]pI '{PI , ... ,Pr}
gives a baslc example of the rigidity method. ~The meromorphic func-

tion field N = C(X) of the compactification X of X is a finite Galois
extension of the rational function field iC = C(]pl) = C(t) and its Ga-
lois group is isomorphic to the covering transformation group of 'P. Let
):11, ... ,):Ir be the corresponding prime divisors of N IC to the ramifica-
tion points PI, ... , Pr of 'P. Then N IiC is unramified outside {):II, ... , ):Ir}
and descends to Q. Namely, if PI, ... ,Pr are algebraic points over Q,
then there exists a unique Galois extension N of /C = Q(t) unramified
outside {):II, ... ,):Ir} such that NC = NCJ!C = N. Moreover, NI/C de-
scends to the number field k generated by the coefficients of the defining
polynomial of N I/C. We are interested in the smallest field of definition
k of NliC.
Example 1.1 (Sn-covering). The branched covering

xn
'Po: ]pI 3 x f--+ = t E ]pI
nx-n+1
is unramified outside {O, 1,00}. The corresponding function fields are
i, = C(x) and R:, = C(t) and the corresponding ramified prime divisors
):11,):12,):13 of iC/c are decomposed in i, '1B follows:
):13 = m
..... 3n - 1 cu
~l. (1.1)
128 T. SHIINA
Here qJl, qJ2, qJ3 and 9t are the prime divisors of llC corresponding to
0,1,00 and n~l, respectively. Similarly, .01,"" .on-2 correspond to
n- 2 distinct roots of
{xn - (nx - n + l)}/(x - 1)2 = x n- 2 + 2x n- 3 + ... + (n - 1).
Let 'P: X - pI be the Galois closure of the covering 'Po. Then the
meromorphic function field fi = C(X) coincides with the splitting field
of
f(x) = xn - t(nx - n + 1) E C(t)[x]
and the Galois group of fi / JC is represented as a permutation group
on n distinct roots of f(x). A fundamental fact of the theory of al-
gebraic function fields indicates that the prime divisor decompositions
(1.1) express the permutation types of the inertia subgroups correspond-
ing to the ramification points. Namely, generators 0"1,0"2,0"3 of the inertia
subgroups of qJl, qJ2, qJ3 (these groups are cyclic) are respectively an n-
cycle, a transposition and an (n-1 )-cycle. Since these cycles generate
the symmetric group Sn, the Galois group of fi / R:, is isomorphic to Sn.
Moreover, since f(x) has rational coefficients, the smallest field of def-
inition of fj / JC is Q. In other words, the splitting field N of f (x) over
Q(t) is a regular Galois extension of Q(t) with Galois group Sn.
Let PI, ... , Pr be r distinct algebraic points on p1 and
<P : U - pI" {P1, ... , Pr }

the universal covering. The covering transformation group of cP is iso-
morphic to the fundamental group r r := (/'1, ... , /'r I/'1 ... /'r = 1) of
the punctured projective line. Here /'i denotes the homotopy class of a
sufficiently small closed path counterclockwise around Pi. The property
of universal coverings and Galois theory yield a one-to-one correspon-
dence among the following objects.
The finite unramified Galois coverings of p1"{P1,'" ,Pr}.
The normal subgroups of r r of finite index.
The finite Galois extensions of Q(t) unramified outside {PI, ... , Pr}.
Therefore the maximal Galois extension Mr of Q(t) unramified outside
{PI, ... , Pr} has Galois group isomorphic to the profinite completion of
rr:
G(Mr/Q(t)) = rr := ~ rr/N ,
N
where the limit is taken over all normal subgroups N <l r r of finite index.
The meromorphic function field C/Q(t) corresponding to an unramified
Regular Galois realizations ojPSL2(p2) over Q(T) 129
Galois covering <p: X ----+ ]p'l'{Pl, ... ,Pr} is a subfield of Mr. Then we
have the canonical surjective homomorphism
rr = ('1'1, .. ,'Yr 1'1'1 'Yr = If ----+ G(C/Q(t)) =: G.
Let ai E G be the image of'Yi and Ci the conjugacy class of ai. Then
al, ... , a r generate G and satisfy al ... ar = 1. The class vector (Cl , ... ,
Cr) is called the ramification structure of C/Q(t).
Let G be a finite group with trivial center and C = (C1 , , Cr ) a
class vector of G. Consider the set of all generating r-systems of C:
I
~(C) := {(a1, ... , a r ) ai E Ci, al ... a r = 1, (al, ... , a r ) = G}.
The group G ~ Inn(G) acts on ~(C) by conjugation; denote by [(7'] =
[al, ... ,ar ] the Inn(G)-orbit of (7' = (al, ... ,ar ) E ~(C). The class
vector C is said to be rigid if G acts transitively on ~(C), or briefly if
1~(C)/Inn(G)1 = 1. In general, a conjugacy class C of a finite group
G is said to be rational if em = C for any integer m prime to IGI,
or equivalently if all irreducible characters of G take integral values at
C. We say the class vector C rational when each Ci is rational. The
following is known as a basic theorem in the rigidity method.
Theorem 1.2 (Basic Rigidity Theorem). Let G be a finite group
with trivial center and C = (Cl, ... , Cr ) a class vector of G. If C is
rational and rigid, then there exists a regular Galois extension N /Q(t)
with Galois group G and ramification structure C.
See [10] or [22] for proofs.
Remark 1.3. The regular Sn-extension N /Q(t) in Example 1.1 pos-
sesses the ramification structure C = (2A, (n-I)A, nA), where mA de-
notes the conjugacy class of m-cycles in Sn. Since each conjugacy class
in Sn is rational, the class vector C is rational. Moreover, one can verify
the rigidity of C by a brief calculation (see for example [19]).
Example 1.4. The projective general semi-linear group prL2(pn) is
the automorphism group of PSL2(pn). We take a class vector C =
(2C, 4B, lOA) of prL2(p2), where we follow the notation in ATLAS [1].
Feit [5] proved that C is rational and rigid if p == 2 (mod 5). Let
N/Q(t) be a regular Galois extension with Galois group PfL2(P2) and
ramification structure C. The fixed subfield K by the normal subgroup
PSL2(P2) of index 4 is an algebraic function field of genus O. Since the
prime divisor corresponding to 2C splits completely in K/Q(t), the sub-
field K is a rational function field over Q. Therefore the group PSL2(P2)
occurs regularly over Q if P == 2 (mod 5).
130 T. SHIINA
2. Braid Orbit Theorem
The projective special semi-linear group P1:L2(p2) is one of the normal
subgroups ofPrL2(p2) of index 2. Przywara [14] showed that there exists
a regular Galois extension of Q(t) with Galois group G = P1:L2(25) and
ramifica~icn structure C = (2A, 2C, 2D, 12A). Here the class vector C is
rational but non-rigid. Actually, one verifies that
11:(C)/Inn(G)I = 12.
Przywara realized P1:L2(25) by means of the braid orbit theorem which
had been proposed by Matzat [12] and can be applied to such non-rigid
class vectors. Matzat considered the Galois coverings of !PI . . . . {P1, ... , Pr}
with transcendental ramification points P1, ... , Pr and the braid actions
on 1:(C)/Inn(G) associated to such Galois coverings. We sketch this
theorem restricting to the principal case r = 4.
Let M be the maximal Galois extension of Q( u) unramified outside
{O, 1, oo} and M a maximal regular Galois extension of M (t) unramified
outside {0,1,00,u}, where v and t are transcendental over Q. Their
Galois groups are
and
G(M/ M(t)) = (i1, i2, i3,i41 il'Y2i3i4 = 1)~ =: r.
The fields M(t) and M are Galois over Q(u, t) and their Galois groups
split as follows:
G(M(t)/Ql(u, t
~ r ~ rQ (2.1)
and
G(M/Q(u, t) ~ r ~ G(M(t)/Q(u, t), (2.2)
where rlQl := G(Q/Q) is the absolute Galois group of Q.
Let G be a finite group with trivial center and C = (01 ,02,03,04)
a rational class vector of G. For a given u = (0'1,0'2,0'3,0'4) E 1:(C),
r
consider a surjective homomorphism 'l/Jcr: - - - t G defined by mapping
ii to O'i. The fixed subfield
Ncr := MKer (1f!a)
is a regular Galois extension of M(t) with Galois group G and
ramification structure C. We define an action of G(M(t)/Q(u, t)) on
1:(C)/Inn(G) by extending the Galois action G(M(t)/Q(u, t)) rv in r
the semi-direct product (2.2) to G via 'l/Jcr. Let
Llcr := {<5 E G(M(t)/Q(u, t I rule = [un
be the stabilizer of [0'] E E(C)/Inn(G) and

Ktr(t) := M(t)~C7
the fixed subfield of 6. tr . From (2.1), the action of G(M(t)/Q(u, t
splits into two sorts of actions; r acts on E(C)/Inn(G) and rQ acts on
(E(C)/Inn(G/r. The former is called the braid action because r is the
normal subgroup of the Hurwitz braid group H4 of index 2. Denote a
r-orbit of [0'] E E(C)/Inn(G) by B(O') and call it a braid orbit.
Lemma 2.1. Under the preceding conditions, we have the following.
(i) There exists a regular Galois extension N tr / Ktr(t) with Galois group
G and ramification structure C.
(ii) K tr is a finite extension of ktr(u) := K tr n Q(u), whose degree equals
the length of B(O').
(iii) ktr is a finite extension of Q, whose degree equals the length of the
rQ-orbit of B(O').
Proof. (i) For any 5 E G(M/Ktr(t, there exists T E G such that
0'6 = O'T by the definition of K tr . This means the Galois group G =
G(Ntr / Ktr(t is invariant under 5 and so is N tr . Hence N tr is a Galois
extension of Ktr(t) and each element of G(Ntr/Ktr(t acts as an inner
automorphism on G. Let Z be the centralizer of G in G(Ntr / Ktr(t,
then
G(Ntr/Ktr(t = G Z.
Since G has a trivial center, the right-hand side is a direct product.
Therefore the fixed subfield N tr of Z is a regular Galois extension of
Ktr(t) with N tr . M(t) = N tr .
(ii) Let L be the fixed subfield of the stabilizer of B(O'), then L is con-
tained in Q(u, t) nKtr(t) = ktr(u, t). Furthermore, let lCtr(t) be the fixed
subfield of r tr := r n 6. tr in M(t). Here r = G(M(t)/Q(u, t and
G(M(t)/L) act transitively on B(O') and the stabilizers of [0'] are r tr
and 6. tr , respectively. Hence
IB(O')I = [Ktr : L] = [lCtr(t) : Q(u, t)]
= [Ktr(t) . Q(u, t) : Q(u, t)]
= [Ktr(t): Ktr(t) nQ(u,t)]
= [Ktr(t) : ktr(u, t)].
This shows L = ktr(u, t) and the degree of K tr (t)/ktr (u, t) equals IB(O')I.
(iii) The Galois group G(M(t)/Q(u, tacts transitively on the rQ-orbit
of B(O') and the stabilizer of B(CT) is G(M(t)/ktr(u, t. Hence the
degree of ktr(u, t)/Q(u, t) equals the length of the rQ-orbit of B(O'). 0
132 T. SHIINA
The precedine; lemma shows that G occurs regularly over Q if a braid

orbit B(u) is invariant under the action of rQ and the corresponding
field of definition Ku is a rational function field. Matzat [12] stated a
sufficient condition for Ku to be a rational function field over Q. Let
1r'B be the permutation representation of r on a braid orbit B = B(cr)
and Ci the number of cycles in 1r'B(-Yi). The genus of the braid orbit B
is defined by
1 3
g(B) := 1 -IBI + "2 ~)IBI- Ci).
i=l
This coincides with the genus of the algebraic function field Ku I ku

Furthermore, consider the following oddness condition.
(0) There exists i such that 1r'B(-Yi) has an odd number of l-cycles for
a length l.
We demand this condition for the existence of ku-rational points on Ku

Now the braid orbit theorem is stated as follows.
Theorem 2.2 (Braid Orbit Theorem). Let G be a finite group with
trivial center and C = (Gl' G2, G3, G4) a rational class vector ofG. Sup-
pose ~(C)/Inn(G) contains a rQ-invariant braid orbit B = B(cr) which
has genus g(B) = 0 and satisfies the oddness condition (0). Then the
field of definition Ku is a rational function field over Q and hence G
occurs regularly over Q.
See the original paper [12] or a comprehensive book [10]. To apply

this theorem to a given finite group G and its class vector C, one needs
to calculate the action of r on ~(C)/Inn(G) explicitly. The action of r
on f in the semi-direct product G(M/Q(u, t)) ~ f ~ r is well-known
by the fundamental relations of the Hurwitz braid group. Namely, the
action on ~(C)/Inn(G) reads as follows:
[0'1,0'2,0'3,0'4Pl = [O'f20'3, 0'2, 0'3, 0':20'3],

[0'1, 0'2, 0'3, 0'4P2 = [0'1,0'~30'1,0'3,0':10'3], (2.3)
[0'1,0'2, 0'3, 0'4P3 = [0'1, 0'2,O'gl0'2, O'r10'2j.
Two braid orbits Band B' have the same length and genus if they are
conjugate to each other under the action of rQ. Hence a braid orbit B is
rQ-invariant if there exist no other braid orbits of the same length and
genus in ~(C)/Inn(G).
Regular Galois realizations of PSL2(P2) over Q(T) 133
3. Groups PEL2(p2)
The projective special linear group PSL2(P2) has few rational con-
jugacy classes. So we begin with the group P~L2(P2), which contains
PSL2(P2) as normal subgroup of degree 2. In this section, we concretely
calculate braid orbits to apply the braid orbit. theorem for small p. The
p-Frobenius automorphism lFp2 3 S ~ 8 := sP E lFp2 induces an auto-
morphism of PSL2(P2):
cp: PSL2(P2) 3 (~ ~) ~ (! !) E PSL2(P2).
The projective special semi-linear group P~L2(P2) =: G is defined by the

semi-direct product PSL2(P2) ><I (cp). The conjugacy classes 2C,2D,pA,
pB in G are defined as the classes of
respectively, where r E lFp2 is any non-square element. Since these con-

jugacy classes are rational, the class vectors C := (2C,2D,pA,pA) and
C' := (2C,2D,pA,pB) are rational. In the following examples, we use
the algebraic computation software GAP [6].
Example 3.1 (p = 3). The groups PSL2(9) and P~L2(9) are isomorphic
to A6 and 86, respectively. The classes in P~L2(9) are identified with
the following classes in 8 6 :
2C = [(12)], 2D = [(1 2)(34)(56)],
3A = [(1 23)], 3B = [(123)(456)].
Here ~(C') divides into two Inn(G)-orbits which are represented by

0' = ((1 2), (13)(24)(56), (245), (1 23)(456)),
T = ((1 2), (13)(24)(56), (253), (143)(256)).
Since these orbits are fixed by 'Y1 E r and permuted by 'Y2, 'Y3 E r, the
permutation types of 'Yb 'Y2, 'Y3 are (1)2, (2), (2), respectively. Hence the
genus of B':= {O',T} is
1
1 - 2 + -(0 + 1 + 1) = 0
2
and there exists a regular Galois extension of Q(u, t) with Galois group
PEL2(9) ~ 86 and ramification structure (2C, 2D, 3A, 3B).
134 T. SHIINA
Remark 3.2. The B6 -extension N u / Ku (t) with ramification structure

(2C, 2D, 3A, 3B) is concretely constructed as follows. Let L be the fixed
subfield of a subgroup B5 = {a E B6Ia(6) = 6} and Pl,P2,l)3,P4 the
corresponding prime divisors of Ku(t) to the ramification points 0,1,00
and u. They are decomposed in L as follows:
Pl = q3rq34, P2 = q3, 4'3 = nrn3, P4 = n~,

where the subscripts of q3i and .oj coincide with their degrees as divisors.
The Riemann-Hurwitz genus formula shows that the genus of L equals
o. Let x E L be a generator of the divisor q3 1 n 1l , then there exist monic
polynomials is, 14, 92, 93 E Ku [Xl such that
.03
q33
.or = (f3(X)), nt
q34 (
= (f4 x)), .or = (93(X)),
where the subscripts of !i and 9j coincide with their degrees as poly-

nomials. We may assume that 92(X) = X2 + 2uX + a, a E K u , by
changing x if necessary. These divisors satisfy
(t -1) = q3 = ( fa(x)2 ) ,
DrD3 93 (x)
and
(t _ u) =
DrD3
n~ = (993(X)
2(X)3 ) .
Hence there exist certain constants Cl, C2, C3 E Ku such that
Since each polynomial is monic, we have Cl = C2 = C3 =: c. From this

equation, we obtain a system of equations in 11 undetermined coefficients
of is, !4, 92, 93 and the unknown constant c. Solving these equations, we
have
fa(X) = X 3 + 3uX 2 + 3u(1 + v)X + u(1 + v)3,

!4(X) = X4 + 6uX 3 + {12u 2 + 3u(1 + v)2}X2
+ 16u2(1 + v)X + 6u 2(1 + v)3,
92(X) = X2 + 2uX + u(1 + v)2,
-C93(X) = 4u(1 + v)2 X 3 + 9u 2(1 + v)2 X2
+ 6u 2(1 + v)4X + u 2 (1 + v)6,
Regular Galois realizations o!PSL2(p2) over Q(T) 135
where v denotes a square root of 1 - u. Thus the field of definition KtT

equals Q( y'l-u). The sextic field L/ KtT(t) is generated by a root of the
polynomial F(t, X) := X2 !4(X) - cg3(X) t E Q( y'l-u, t)[X] and the
S6-extension NtT is given as the Galois closure of L. The discriminant
of F(X) is computed as
Disc(F) = _2 13 36u lO v(1 + v)23t (1 - t)3(u - t)4.
Therefore the fixed subfield of A6 in NtT/Q( v"1=U, t) is
L' = Q ( .,fl-u, V-2vl(~: V)t) ,

and the specialization F(s2/(s2 - 2v(1 + v, X) is a polynomial with
Galois group A6 over Q(v, s).
Example 3.3 (p = 5). For G = PEL2(25), we have
IE(C)/Inn(G)! = 4 and IE(C')/Inn(G)I = 2.
In either case, the action of r is transitive and the oddness condition
(0) is satisfied. Actually, the permutation types of "Yl, /'2, /'3 are (1 )(3),
(1)(3), (1)(3) on E(C)/Inn(G) and (1)2, (2), (2) on E(C')/Inn(G), re-
spectively. In the first case, the genus of B := E(C)/Inn(G) is
1
1 - 4 + "2 (2 + 2 + 2) = 0
and the field of definition KtT /Q( u) is a rational function field whose
Galois closure has Galois group A4 for each [0'] E B. In the latter case,
the genus of B' := E(C')/Inn(G) is calculated in the same way as the
case p = 3. Consequently, we obtain two types of Galois realization of
PEL2(25).
Example 3.4 (p = 7). For G = PEL2(49), we have
IE(C)jInn(G)1 = 4 and IE(C')/Inn(G)I = 8.

In either case, the action of r is transitive. The permutation types of
/'1,/'2,/'3 are (1)(3), (1)(3), (1)(3) on E(C)/Inn(G) and (1)4(2)2, (4)2,
(4)2 on E(C')/Inn(G), respectively. In the first case, the genus of B :=
E(C)/Inn(G) is calculated in the same way as the case p = 5. In the
latter case, while the genus of B' := E(C')/Inn(G) is
1
1- 8 + 2"{2 + 6 + 6) = 0,
136 T. SHIINA
the oddness condition (0) is not satisfied. Thus the braid orbit theorem
can be applied only to B.
Example 3.5 (p = 11,13). For C' = (2C, 2D,pA,pB), the cases p = 11
and p = 13 are similar. In either case, the number of Inn(G)-orbits is
1I:(C')jInn(G)1 = 18
and r acts intransitively on I:(C')jInn(G). Namely, I:(C')jInn(G) di-

vides into two braid orbits B~ of length 2 and Bi6 of length 16. The
shorter orbit B~ is a unique braid orbit of length 2 in I:(C')jInn(G) and
r acts on it in the same way as the case p = 3. On the other hand, for
C = (2C, 2D,pA,pA), we have
1I:(C)jInn(G)I = {1224 for p = 11,

for p = 13.
Here r acts transitively on B := I:(C)jInn(G). For p = 11, the permu-

tation types of all "Ii'S are (1)2(5)2 and hence g(B) = 1 and for p = 13,
the permutation types of all "Ii'S are (1)3(7)l and hence g(B) = 4. Thus
the braid orbit theorem can be applied only to B~.
Example 3.6 (p = 17,19). For C = (2C,2D,pA,pA), the cases p = 17
and p = 19 are similar. In either case, the number of Inn(G)-orbits is
1I:(C)jInn(G) I = 40
and r acts intransitively on I:(C)jInn(G). Namely, I:(C)jInn(G) divides
into two braid orbits B4 of length 4 and B36 of length 36. Since the
shorter orbit B4 is a unique braid orbit of length 4 in I:(C)jInn(G) and
r acts on it in the same way as the cases p = 5 and p = 7, the braid
orbit theorem can be applied to this orbit.
4. Proof of Main Theorem

In the examples of the previous section, we found isolated braid orbits
of length 2 for p = 3,5,11,13 and those of length 4 for p = 5,7,17,19.
These facts are generalized as follows:
Proposition 4.1. Let C = (2C,2D,pA,pA) and C' = (2C,2D,pA,pB)
be the prescribed class vectors of G = PI:L2(p2).
(i) lfp == 3 (mod 8), then E(C')jInn(G) contains a unique braid orbit
of length 2 and the permutation types of "11, "12, "13 E r on it are (1)2, (2),
(2), respectively.
Regular Galois realizations of PSL2 (P2) over Q(T) 137
(ii) lfp == 5 (mod 12), then E(C)/Inn(G) contains a unique braid orbit
of length 4 and the permutation types of /1, /2, /3 E r on it are (1)(3),
(1)(3), (1)(3), respectively.
By applying the braid orbit theorem to these orbits, we can realize
PSL2(P2) regularly over Q as follows:
Proof of Main Theorem. By the braid orbit theorem, there exists a reg-
ular Galois extension N of Q(T) = Q(u, t) with Galois group PEL2(P2)
and ramification structure C = (20,2D,pA,pA) (in the case p == 5
(mod 12 or C' = (20,2D,pA,pB) (in the case p == 3 (mod 8. The
fixed subfield L by the normal subgroup PSL2(P2) is a quadratic ex-
tension of Q(T). Here the two ramification points corresponding to pA
or pB are unramified at L/Q(T) since these classes are contained in
PSL2(P2). Therefore, the quadratic extension L is a rational function
field over Q, say L = Q(T'). Thus we obtain a regular extension N/Q(T')
with Galois group PSL2(P2). 0
Since the first part of Proposition 4.1 has been proved in [21], we will
show the remaining part. Hereafter we assume that p == 5 (mod 12), in
other words, lFp2 = lFp( v'3). Let P+ and P- be the following subgroups
ofG:
which are the centralizers of (6 (1

i) and ~), respectively. For C =
(20, 2D,pA,pA), the Inn(G)-orbits in E(C) are parametrized as follows:
Lemma 4.2. Each Inn(G)-orbit in E(C) possesses the following repre-
sentative element O'(s,t,u,v) = (O"i, 0"2' O"g,0"4) with
0" * -
1 -
(s(2u +- uv'3
sv)v'3
-uv'3 ) *
s - uv'3 <p, 0"2 =
(t + uv'3s +-t)
t - -t uVS <p,
0"3* = (10 1)
1 ' *
0"4=
1 0)
( -2+vVS 1 .
Here 4-3v 2 and 4-3u 2 (4-3v 2 ) are square in lFp with u, v E lFp, (u, v) =/:
(0,0). Moreover, s, t E IFp are the distinct roots of X2 - 3uvX + (3u 2 -
1) E lFp[X).
Proof. For each 0' = (O"I, 0"2, 0"3, 0"4) E ~(C), we must show there exists
T E G such that O'T = O'(s,t,u,v). We may assume that 0"3 = O"g by taking
138 T. SHIINA
a suitable conjugate. Then there exists r E P+ such that O'r E P-.

Rewrite each aT as ai and put
where C2, C3 E lFpV3 := {nV31 n E lFp} and d2 , d3 E lFp. Solve the

equation a1 a2a3a4 = 1; we have
C1 = 8 + uv'3, C2 = -uv'3, C3 = (2u - 8v)V3,

d1 = t+uV3, d2 = -t, d3 =t - s,
with 3uv = 8 + t, (u, v) =J (0,0) and b = -2 + vV3. Since b is square in

lFp2, its norm 4 - 3v2 must be square in lFp. Substitute Ci,c4 and b into
the equation det 0'1 = 1; we have 8t = 3u2 - 1. Finally, the reducibility
of X 2 - 3uvX + (3u 2 - 1) implies that 4 - 3u2 (4 - 3v2 ) is square in
~. 0
Conversely, the 4-system O'(s,t,u,v) as in the lemma belongs to ~(C) for

any (s, t, u, v) E lFt satisfying
4 - 3v 2 E IF;2, 4 - 3u2 (4 - 3v2 ) E IF;2,

(4.1)
(u,v) =J (0,0),8 + t = 3uv, 8t = 3u 2 -1.
Actually, if v =J 0, then D:ckson's theorem ([4J, see also [7J for a modern
description) shows that {ai, an generates PSL2(p2) and hence O'(s,t,u,v)
generates G. On the other hand, if v = 0, then {ai, an generates the
subgroup H := PSL2(P). By Dickson's classification of the subgroups of
PSL 2 (q) in [4], if a maximal subgroup of PSL2(P2) contains PSL2(p), it
must be isomorphic to PGL2(p), Since I(H, HO'i) : HI > 2, we conclude
that O'(s,t,u,v) generates G.
Remark 4.3. Under the notation in the previous lemma, [O'(s,t,u,v)J =
[O'(SI,tl,ul,VI)J if and only if
(8', t', u', v') = (8, t, U, v), (8, t, -u, -v), (-8, -t, -u, v), (-8, -t, U, -v).
By elementary number theory, there exist (p - 2 + ) /2 chokes for
(8, t, U, v) E lFt satisfying (4.1) and uv =
up to 1, where :=
(-1)Cp-l)/2. In the case uv i= 0, there exist (p - 2 + )/4 choices for
v E IF; with 4 - 3v 2 E IF;2 and (p - 2 + )/4 choices for u E IF; with
4 - 3u2 (4 - 3v2 ) E F;2 up to 1. For such u, v E IF;, there are two
choices for distinct roots of X 2 - 3uvX + (3u 2 - 1). Consequently, the

cardinality of E(C)/Inn(G) is
IE(C)/Inn(G)1 = p - ~ + + ( P - ~ + ) 2 .2= (p + 12 - 4 .
Lemma 4.4. Suppose [0'] E E(C)/Inn(G) be a fixed point of ,3, then
[uj = [e:u~ s-=-:1a)~, (-~~~J3 s+~J3)~' GD, (!2 ~)]

with su =f 0 and s2 + 3u2 = 1. Aside from these trivial fixed points, ,
has no non-trivial fixed points on E(C)/Inn(G).
Proof. Take the representative element 0' = of [0'] E
(0"1, 0"2, 0"3, 0"4)
E(C)/Inn(G) as in Lemma 4.2. If v = 0, then it is easily checked that
[O'ps = [0']. Suppose v =f 0 and [O'p~ = [0'], then (0"3,0"4) = PSL2(P2)
and (O"iO"S0"4)4, 0"~O"S0"4)4, 0"3, 0"4) = O'T with T E G. Since PSL2(P2) has
trivial centralizer, T must he trivial and hence (0"30"4)4 commutes with
0"1. Here
(0"30"4)40"1 = (: sx-:yJ3) cp, 0"1(0"30"4)4 = (: Sy-:xJ3) cp

with
x = 1 - 3v 2 - (2v - 3v3 )V3, Y = 1 - 9v 2 + 9v4 + (2v - 3v3 )V3.
So we have sx-uyv'3 = (sy-uxv'3), which means x = y. If x = -y,

then
o= 2 - 12v2 + 9v4
and hence 3v 2 (4 - 3v 2 ) = (2 - 3v 2 )2, which is a contradiction. Further
if x = y, then
0= x - y = (2 - 3v 2 )(-2 + vV3)vV3
and hence v 2 = 2/3. This contradicts that 4 - 3v 2 = 2 is square in lFp

Therefore, each fixed point of,~ is already a fixed point of 0 ,3.
Lemma 4.5. There exist no common fixed points of
E(C)/Inn(G).
,3 and ,t on
,3
Proof. Let [0'] be the fixed point of determined in the previous lemma.
Suppose [0'] be also a fixed point of ,t,
then (O"i0"20"S)4, 0"2,0"3, O"i0"20"S)4) =
O'T with T E G. Since T commutes with 0"3 and (0"20"3)4 T -l commutes
140 T. SHIINA
with 0'4, we have T E P+ and (0'20'3)4 T -1 E P-. This indicates that

(0'20'3)4 is of the form as (; :). By a calculation, we have
(0'20'3)4 = (s4-18S2u2+9U4:4SU(S2_3U2)V3 :).
Since su(s2 - 3u2) i= 0, this is a contradiction. o

Proof of (ii) of Theorem 4.1. We begin with the following element of
~(C)/Inn(G):
[0'4] == [(1+v'
v'33 I-v'
-v'33) (2+v'13 -2+v'
-2)3 (10 11) ' (-2+v'
r..p,
13 r..p, ~)] .
Denote by [O'i] (i = 1,2,3) the image of [0'4] by Ii, then
[C+ov'3 2-=-~) r..p, C~~ _1~1v'3) r..p, GD, (-2!v'3 ~)],

c;:
[O'd ==
[0'2] == [~ 1-=-~) r..p, ~ (-1~2v'3 1+1v'3) GD, (!2 ~)],

r..p,
[0'3] == [( -~ ~) r..p, (=~ D GD, (-2!v'3

r..p, ~)].
These elements construct a braid orbit B4 = {[O'l], [0'2], [0'3], [0' 4]} on
which 11,/2, 13 act as the following permutations:
[O'iPl = [O'Pl(i)l, P1 = 7T'B4(/t} = (124),
[O'iP2 = [O'P2(i)], P2 = 7T'B4(/2) = (234),
[O'iP3 = [O'P3(i)], P3 = 7T'B4(/3) = (1 43).
Let B ~ ~(C)/Inn(G) be a braid orbit of length 4. Lemma 4.4 indicates
that 13 has no transpositions nor 4-cycles under the permutation repre-
sentation of ron B. Thus 7T'B(/3) must be the identity or a 3-cycle and
hence 13 has at least one fixed point [0'] E B. Such [0'] is determined in
Lemma 4.4 and its (J1)-orbit must have length 3 by Lemma 4.5. Similar
to the proof of Lemma 4.5, (0'20'3)3 is of the form as (; : )cp. Since
( 0'20'3 ) 3 _- (-s3+9su2+3(s2U-U3)V3 *) cp,

* *
we have
-i + 9su 2 = 1, s2u - u 3 = O.
The unique solution of these equations is (s, u) = (1/2,1/2) up to 1.
Thus [0'] coincides with [0'2] E B 4. This means B = B4 is a unique braid
orbit of length 4. 0
Regular Galois realizations of PSL2 (p2) over Q(T) 141
Acknowledgments
The author wishes to thank Professors Ki-ichiro Hashimoto, Katsuya
Miyake and Hiroaki Nakamura for their invitation to the symposium
"Galois Theory and Modular Forms". The author also wishes to thank
Professor Toyofumi Takahashi for his continuing encouragement.
References
[1] J. H. Conwayet al., "Atlas of finite groups," Clarendon Press, Oxford, 1985.
[2] M. Dettweiler, Middle convolution and Galois realizations, to appear in this
volume.
[3] L. Dieulefait and N. Vila, Projective linear groups as Galois groups over Q via
modular representations, J. Symbolic Computation 30 (2000), 799-810.
[4] L. E. Dickson, "Linear groups with an exposition of the Galois field theory,"
Teubner, Leibzig, 1901.
[5] W. Feit, Rigidity of Aut{PSL2 (P2)), p == 2 (mod 5), p f 2, in "Proceedings of
the Rutgers group theory year, 1983-1984 (New Brunswick, N. J., 1983-1984),"
351-356, Cambridge Univ. Press, Cambridge, 1984.
[6] The GAP Group, "GAP - Groups, Algorithms, and Programming, Version 4.2,"
Aachen, St. Andrews, 1999.
[7] D. Gorenstein, "Finite groups," Harper and Row, New York-Evanston-London,
1968.
[8] D. Hilbert, Ueber die Irreduzibilitiit ganzer mtionaler FUnktionen mit ganzzahli-
gen KoejJizit:nten, J. Reine Angew. Math. 110 (1892), 104-129.
[9J G. Malle, Genus zero translates of three point ramified Galois extensions,
Manuscripta Math. 71 (1991) 97-111.
[10] G. Malle and B. H. Matzat, "Inverse Galois theory," Springer-Verlag, Berlin,
1999.
[l1J B. H. Matzat, "Konstruktive Galoistheorie," Lecture Notes in Mathematics,
1284. Springer-Verlag, Berlin, 1987.
[12J B. H. Matzat, Zopfe und Galoissche Gruppen, J. Reine Angew. Math. 420
(1991),99-159.
[13] J.-F. Mestre, Courbes hyperelliptiques d multiplications reelles, C. R. Acad. Sci.
Paris, 301 (1988), 721-724.
[14J B. Przywara, Die Operation der Hurwitzschen Zopfgruppe auf den Erzeugenden-
systemklassen endlicher Gruppen, Diplomarbeit, Karlsruhe, 1988.
[15] H. Reichardt, Konstruktion von Zahlkorpern mit gegebener Galoisgruppe von
Primzahlpotenzordnung, J. Reine Angew. Math. 111 (1937), 1-5.
[16] A. Reverter and N. Vila, Some projective linear groups over finite fields as Galois
groups over Q, Contemp. Math. 186 (1995), 51-63.
[17] K. A. Ribet, On l-adic representations attached to modular forms, Invent. Math.
28 (1975), 245-275.
142 ~ SHIINA
[18] 1. R. Safarevic, Construction of fields of algebraic numbers with given solvable

Galois group, Izv. Akad. Nauk SSSR 18 (1954), 525-578 (Amer. Math. Trans!.
4 (1956), 185-237).
[19] J.-P. Serre, "Topics in Galois theory," Jones and Bartlett, Boston, 1992.
[20] K-y. Shih, On the construction of Galois extensions of function fields and number
fields, lvlath. Ann. 207 (1974),99-120.
[21] T. Shiina, Rigid braid orbits related to PSL2(p2) and some simple groups, to
appear in Tohoku Math. J.
[22] H. Volklein, "Groups as Galois groups," Cambridge University Press, Cam-
bridge, 1996.
MIDDLE CONVOLUTION
AND GALOIS REALIZATIONS
Michael DETTWEILER
1WR, Universitiit Heidelberg,
1m Neuenheimer Feld 368,
69120 Heidelberg, Germany
michael.dettweiler~iwr.uni-heidelberg.de
Abstract The theory of the middle convolution is combined with the theory of
curves on Hurwitz spaces. This leads to the following theorem: The
projective symplectic groups PSP2n (lFp2) occur Q-regularly as Galois
groups over Q(t) if P is an odd prime 1 mod 24.
1. Introduction
The usual approach of realizing a finite group Q-regularly as Galois
group over Q(t) is via rigidity or, more generally, by detecting rational
points on the underlying moduli spaces (Hurwitz spaces) using braid
groups, see [12], [15] and the references therein. Once the geometry of
the Hurwitz spaces becomes more complicated it can be useful to look
for rational curves lying on the Hurwitz spaces, instead of studying the
whole Hurwitz space, see [3], [4].
In [6], [16], the authors give (different) purely algebraic constructions
of Katz' middle convolution functor MCx (see [11]). In [6] the resulting
analogon of MCx is denoted MC)., in [16] it is called the braid companion
(BC-) functor. As a first application, many families of quasi-simple
groups of Lie type were realized regularly as Galois groups over Q(t),
see [5], [6], [17].
The basic idea is the following: Choose a subgroup H ::; GLn(JFq) for
which the braid group action on tuples can easily be determined (e.g.,
a subgroup of the scalars or a dihedral group, viewed as subgroup of
GL2(lFq)) and the available braid group criteria can be applied. Ap-
plying successively the operations of middle convolution and suitable
[received: October 23, 2002; accepted in revised form: March 21, 2003]
144 M. DETTWEILER
multiplication with scalars one can realize groups in higher dimensions

regularly as Galois groups over Q(t).
It is the aim of this work to demonstrate how to combine the approach
of [3], [4] with the theory of middle convolution. In fact, T. Shiina
[13], [14] recently succeeded in solving the inverse Galois problem for
PSL2(p2) for p '= 1 mod 24 by using the classical braid orbit theorem
due to Matzat, see [12], Chap. III. We show that using our methods one
can extend his result to the higher rank Lie-type groups PSP2n(P2):
Theorem 1. The projective symplectic groups PSP2n (IFp2) occur Q-
regularly as Galois groups over Q(t) if P is an odd prime ~ 1 mod 24.
For n = 1 this reduces to the above mentioned results of Shiina ([13],
[14]). For other Galois realizations of symplectic groups, see [7], [12],
[17] and the references therein.
The plan of the paper is as follows: In Section 2 the definition of the
middle convolution functor MC>.., as given in [7], is recalled. Also the
main properties of MC>. are restated. In Section 3 we state a result of
[8] (Thm. 5), saying that the field of linear moduli is preserved under
application of MC-l (compare to [17], Thm. 4.4., where a similar result
is proved for the BC-operation). In Sections 4 - 6 the theory of Hurwitz
spaces and curves lying on them is revisited. Finally, Thm. 1 is proved
in Section 7 (Thm. 11).
2. Convolution
In the following we recall the definition of the middle convolution
functor as in [6], [7]:
For (AI, ... ,Ar ) E GLn(Kt and>' E KX, consider the following
matrices Bk E GLnr(K) for .Ie = 1, ... , r :
In o o
In
>..(Ak-l - In) >"Ak (Ak+l - In)
In
o o
where Bk is the identity matrix outside the k-th block row (In E GLn(K)
denotes the identity matrix). We call C>..(AI, ... ,Ar ) = (BI, ... ,Br )
the (multiplicative) convolution of (AI,'" ,Ar) with >.. There are the
following (left) (B l , ... , Br}-invariant subspaces
o
o
ICk = ker(Ak - 1n) (Arth entry), k = 1, ... , r,
o
o
and
nker(Bi - lnr ).
r
C:=
i=l
Let IC = EBk=llCk and m the dimension of the K-vector space Knr j(1C +
C). Choose an isomorphism I betw~en Knr L(IC + C) and Km.
The tuple MC>.(A1, ... ,Ar ) = (Bll ... ,Br ) E GLm(K)r which is in-
duced by the action of (Bl' ... , Br) on Km (via I) is called the middle
convolution of (AI, ... ,Ar) with A.
The main properties of the middle convolution are:
Theorem 2. Let (Ab'" ,Ar) E GLn(K)r such that Ai 'f: In for i =
1, ... ,r. Let further A E K X and MC>.(Ab'" ,Ar) = (Bb'" ,Br) E
GLm(Kt
a) If A =1= 1, then
m = dim(Knr j(1C + C))
r
=Lrk(Ak -In) - (n - rk(AA l Ar -In)).
k=l
b) Assume that (AI, ... , Ar) is an absolutely irreducible subgroup of

GLn(K) and r > 1. If Al,A2 E K with A = AI' A2, then
MC>'l (MC>'2(A l , ... , Ar)) E GLm(Kt and
MC>.(A 1 , ,Ar) ~ MC,xl(MC>'2(A l , ... ,Ar )),
where ~ means simultaneous conjugation by an element of GLm(K).

Moreover,
MCl(A1, ... ,Ar ) ~ (A1, ... ,Ar ).
c) If (AI, ... , Art is an a..bsolutely irreducible subgroup of GLn(K) and
r > 1, then (Bl, ... , Br) is an absolutely irreducible subgroup of
GLm(K).
d) Let Br = (Qb"" Qr-l) be the abstract Artin braid group, where
the generators Ql,'" , Qr-l of Br act in the following way on tuples
146 M. DETTWEILER
(gl, ... ,gr) E Cr , where C is any group:
Qi(g1, ... ,gr) = (g1, ... ,gi-1, gigi+1gi 1,gi, gi+2, ... ,gr),
i = 1, ... ,r - 1. (1)
Then
e) The convolution Me),. respects hermitian forms. Moreover, if

(A 1, ... , Ar) respects an orthogonal (resp. symplectic) form then
MC_ 1(Al, ... ,Ar) respects a symplectic (resp. orthogonal) form (the
characteristic of K is assumed to be =f. 2).
Proof. a)-e) follow analogously to [6], Lemma 2.7, Thm. 3.5, Cor. 3.6,
Thm. 5.1, Cor. 5.10 (in this order). 0
The following lemma is already used implicitly in [6J and will be useful
below (the proof follows from Lemma 4.1, loco cit.):
Lemma 3. Let).. E K X , (A 1 , ... , A r ), Ak E GLn(K), such that Ak =f. In,

(AI, ... ,Ar) is absolutely irreducible and r > 1.
a) Let Jk,l $ .. . $Jk,bk be the Jordan canonical form of Ak. where Jk,i (1 ::;
i ::; bk) is the Jordan block with eigenvalue ak,i of size nk,i' Then the
Jordan canonical form of ih is determined by the Jordan canonicaljorm
of Ak in the following way: Every Jk,i contributes a Jordan block A,i of
Bk with eigenvalue o'k,i and size mk,i, where o'k,i, mk,i are as follows:
One has o'k,i = )..ak,i and if ak,i =f. 1,)..-1 (resp. ak,i = 1 or ak,i = )..-1),
then mk,i = nk,i (resp. mk,i = nk,i - 1, or mk,i = nk,i + 1). The other
Jordan blocks of Bk are blocks of length one having an eigenvalue equal
to 1.
b) Let Aoo := A1 ... A r , Boo := Bl'" Br and J oo ,1 $ ... $ Joo,b oo the
Jordan canonical form of A oo , where Joo,i (1 ::; i ::; boo) is the Jordan
bloc:k with eigenvalue aoo,i of size noo,i' Then the Jordan canonical form
of Boo is determined by the Jordan canonical form of Aoo in the following
way: Every Joo,i contributes a Jordan block Joo,i of Boo with eigenvalue
o'oo,i and size moo,i, where o'oo,i, moo,i are as follows:
One has o'oo,i = )..aoo,i and if aoo,i "# 1, )..-1 (resp. aoo,i = 1 or aoo,i =
)..-1), then moo,i = noo,i (resp. moo,i = noo,i+1, ormoo,i = noo,i-1). The
other Jordan blocks of Boo are blocks of length one having an eigenvalue
equal to )...
3. Convolution and fields of moduli

Let Po := {Pb, Pr} ~ C, lPo I = r, Po E C, Po ~ Po and k be a
subfield of C such that Po C k.
Let 'I/J be the Galois cover of Y := ]pI (C) \ (Po U {oo}) associated to
the kernel of a homomorphism p: 71'l(Y,PO) - GLn(1Fq). Let 'Yb ,'YrH
be generators of 71'1 (Y, PO) which are chosen in the following way: The
element 'Yi is the homotopy class of a simple loop in C which encircles
the point Pi in counterclockwise direction and none of the other points
Pj, j #- i. (A simple loop in a topological space M is by definition a closed
path in M which is injective on ]0,1[.) We further assume, that 'YrH :=
('Y1 ... 'Yr)-l is the homotopy class of a simple loop in C, encircling the
set Po in clockwise direction.
If Di = p('Yi), i = 1, ... ,r, then P and 'I/J are uniquely determined by
the matrices Di. Therefore, we call'I/J the Galois cover of Y associated
to ('YI, ... ,'Yr) and(D1,,Dr ) EGLn{lFqt
Definition 4. Let 'I/J be the Galois cover of Y associated to ('Y1, ... , 'Yr)
and (DJ., ... , Dr) E GLn{lFq)r. Then k is called a field of moduli (resp. a
field of linear moduli) for 'I/J, iffor any K, E Gk := Gal(k/k) there exists an
element a E (Db . .. , Dr) (resp. an element a E NGLn(lFq ) ((Db . .. , Dr))
such that 'l/Jit (K, acting via base change, see [10)) is the Galois cover
associated to (,b ... , 'Yr) and (D?, ... , D?).
Let at, ... , a r be generators of 71'1 (]PI (C) \ (Po U {oo} ),po) which go
counter-clockwise around PI, ... , Pr as indicated in the following picture:
Next, let Pr+1 E C\ (PO U {Po} ) and {3t, ... , (3r generators of 71'1 (]PI (C) \
(PO U {oo} ), Pr+ 1) which go counter-clockwise around PI, ... , Pr as indi-
cated in the following picture:
Po

148 M. DETTWEILER
Theorem 5. Let 4> be the Galois cover of Y associated to (al, ... , ar )

and (AI, ... ,Ar) E GLn(lFq)r (q odd) and let ~ be the Galois cover of
Y associated to (i3l, ... ,i3r) and MC-I(A l , ... ,Ar) = (El, ... ,Er ) E
GLm (IF q r. Suppose that (A I, ... , Ar) is an absolutely irreducible sub-
group of GLn(lFq). If k is a field of linear moduli for 4>, then k is also a
field of linear moduli for ~.
Proof. This is proved in [8] in greater generality. o

Let 'I/J be the Galois cover of Y = pI (C) \ (Po U {oo}) associated to
("Y1, ... ,'/'r) and (Dl, ... ,Dr ) E GLn(JFqY and G:= (Dl, ... ,Dr ). Set
Pr+! := 00 and let Cl , ... , Cr +l be the conjugacy classes of the images
of the elements '/'1, .. , '/'r+ 1 in G. Let
[G, POO, C] := [G, Po U {Pr+1 = oo}, (Cp1 := Cl, ... , Cpr+1 := Cr+1 )]
be the ramification type of 'I/J (see [15], Def.'s 4.31 and 2.12, for the
definition of a ramification type of a cover). Then [G,POO,CJ is said
to be k-rational, if for each Pi E POO and for each /'i, E Gk one has
CK,(Pi) = C;;:, where m is an integer such that /'i,-l((n) = (~.
The following corollary is similar to [17], Cor. 4.6 (b):
Corollary 6. Let cp, , be as in Theorem 5 and assume that k is a field

of l~near mq,duli for_ cp. Let Jurth~r G1 , , Gr be the conjugacy classes
of Bl'"'' Br in (B l , ... , B r ), Cr+! the conjugacy class of Br+! .-
(EI ... Br )-1 and
S := {g E NGLm(Fq)( (E 1 , ... , Er}) I g(Gi ) = Gi , i = 1, ... , r + 1}.

If ~ has a k-rational ramification type and
then k is a field of moduli fo>" .
Proof. By the above theorem, k is also a field of linear moduli for ~. By

the branch cycle argument (describing the the action of Gk on the con-
jugacy classes via the cyclotomic character, see [12J, Chap. 1.4.2, or [15J,
L. 2.8), the k-rationality of t1.1e ramification type and the assumption on
S, every /'i, E Gk acts as an inner automorphism of the cover. 0
4. Configuration spaces
The standard reference for configuration spaces and braid groups is
the book of [1], where one can find the basic facts on fibrations of con-
figuration spaces and presentations of braid groups given below, see also
[9].
The following lemma, which will be useful below, is a special case of
the lemma of Schonfiies, see [2], Cor. 2.2.6:
Lemma 7. If Ei : [0,1] -+ pI, i = 1,2, are two simple loops which are
piecewise smooth, then there exists an orientation-preserving homeomor-
phism a : pI -+ pI, such that E2 = a 0 E I .
Let r E N with r ~ 3. Let further Po := {PI, ... ,Pr} ~ pI be the base
point of
Ur := {P ~ ]pl(e) I IPI = r}
and (Po, Po) the base point of
U(r + 1) := ((P,p) E Ur x pl(e) I P ~ P}.
Set 'Hr := 7r1(Ur,PO) (the Hurwitz braid group) and 'H(r+l) := 7r1(U(r+
1), (Po,po. It is well known that the group 'Hr is generated by elements
HI"'" Hr-I subject to the relations
[Hi, H j ] = 1, Ii - il > 1,
and
HI ... Hr-2Hr-IHr-IHr-2 ... HI = 1.
We call HI"'" Hr-I a system of standard generators of 'Hr.
A system of standard generators can be constructed in the following

way: There exists a simple loop E : [0,1] -+ pl(e) with initial point Po
such that there exist < tl < ... < tr < 1 with E(ti) = Pi, i = 1, ... ,r,
and such that E is smooth on ]ti' ti+d, i = 1, ... , r - 1. By Lemma
7, there exists an orientation preserving homeomorphism a : pI(e) -+
pI(e) which maps 1R U {oo} to S := im(E) such that
a(i) = Pi, i = 0, ... , r.
We can define Hi to be the homotopy class of the path
Hi(t) := {a(l), ... ,a(i - l),Pi(t),Pi+I(t), a(i + 2), ... , a(r)},
where
A,(t)._ ((2i+1)-e1l'At) (t).- ((2i+1)+e1l'At)
.- a ,PHI- a
A,
P1 2 2
150 M. DETTWEILER
for t E [0,1] (compare to [4], Section 1.1 and [15], Chap. 10):
The first projl"ction prl : U(r + 1) ~ Ur is a locally trivial fibration.

The long exact sequence of homotopy groups, associated to a locally
trivial fibration, leads then to an exact sequence
1 ~ 11'1 (Fr , (Po,po)) ~ H(r + 1) ~ Hr ~ 1, (2)
where Fr denotes the fibre over Po. Using the construction of the paths
HI (t), ... , H r-l (t) one can easily see that there exists a set of generators
11, .. "r of 11'1 (Fr , (Po, Po)) such that 11' .. Ir = 1 and
(11iIi """r'
iI) ( -1 )
= ,1"""i-1"i"Yi+l'i "i"i+2"""r,
i = 1, ... , r - 1, (3)
where Hi denotes the homotopy class of the path (Hi(t),po).
5. Hurwitz spaces
The standard references for the content of this section are [10] and
[15].
Let G be a finite group and C := (C1,"" Or) a tuple of conjugacy
classes of G. Let
Nr(G) := {(gl, ... ,gr) E Gr I gl'" gr = 1, G = (gl, ... ,gr)}/Inn(G),
Inn( G) acting componentwise. Define
and
Denote by 1ir (G) the Hurwitz space which parametrizes, up to inner

automorphisms, the Galois covers of the r-punctured projective line with
group of deck transformations isomorphic to G.
The map ~ : 1ir (G) -. Ur1 which associates the set of branch points
to the equivalence class of a cover, is an unramified covering map. By
covering theory, ~ induces an action of the fundamental group 1ir =
7rl(Ur,PO) on ~-I(PO). Thi:3 action can be made explicit in the follow-
ing way: Using the same arguments as Fried and Volklein [10], Sections
1.1-1.3, one can construct a bijection between ~-I(PO) and Nr(G) (by
sending the class of the cover, corresponding to the kernel of a homomor-
phism p : 7r1 (]PI (C) \ Po,po) -. G, to [gl,"" Or] := [pC ')'1), ... 'P(')'r))).
It follows then (analogously to [10]) from Equation 3 that the action of
1ir on ~-I(PO) == Nr(G) is given by
Hi([gl, ... ,Or]) = [Ob ... ,Oi-l,OiOi+10;:1,Oi,0i+2,'" ,Or],

i = 1, ... ,r - 1. (4)
We define 1ir (G,C) to be the union of components of 'Hr(G), which

corresponds to N(G,C) via covering theory. The spaces 'Hr(G,C) are
called Hurwitz spaces.
6. Curves on Hurwitz spaces

Let X ~ p2 be a reduced plane curve defined over a subfield k of C.
Let ql1 q2, q3 E p2 (k) be non-collinear points such that the connecting
line q2q3 is not contained in X. A line through ql which intersects the
line q2q3 in a point b is denoted Lb. Such a line is called exceptional if it
intersects X in a singularity or if it is tangent to X. We always assume
that Lq2 is an exceptional line, that Lq3 is not an exceptional line and
that q3 'i. X. Let L := Lbl U ... U Lb8 be the union of all exceptional lines
and SeX) := q2q3 \ {bI, ... ,bs }.
152 M. DETTWEILER
Let further III : JIll2 \ {qd ~ q2q3 (resp. II2 : JIll2 \ {q2} ~ L q3 ) be
the central projection with center ql (resp. q2) and 7 : Lq3 ~ JIlll be a
k- isomorphism.
Suppose that generically, a line Lb intersects the curve X in r points
(not counting multiplicities). The maps
ll1(r + 1) : JIll2 \ X U L ~ U(r + 1), P ~ (70 II2(LI1t (p) n X), 70 II2(P))
and
give rise to a commutative diagram
JIll2 \ X uL
1
U(r + 1)
whose rows are locally trivial fibrations (see [4], Section 2) and the first
vertical arrow maps fibres to fibres. Let :F := Lq3 \ {Lq3 n (X u L)}. The
long exact sequences of homotopy groups, associated to locally trivial
fibrations, lead then to a commutative diagram
1 ~ 7rl (F, q3) ~ 7rl(JIll2 \ X u L, q3) ~
1 1
1 ~ 7rl(Fr, (Po,po) ~ 1t(r + 1) ~ 1
(5)
where (Po, po) (resp. Po) is the image of q3 inU(r+1) (resp. Ur ). Since in
Diagram 5, the left vertical arrow is surjective, one can choose generators
';:i't, ... ,1'r of 7rl (F, q3) such that 1'1, ... ,1'r map to generators ,'1, ... , "Ir
of 7rl (Fr , (Po, po which are chosen in a way that Equation 3 holds. Let
W* : 7rl (8(X) , qg) --+ 1frbe the right vertical arrow in Diagram 5. Let
ObI' . ,Obs E 7rl (8(X), qs) be homotopy classes of simple loops encircling
the points bb ... ,bs (respectively) and let Rbi be the image of Obi under
W*. We call the elements RbI"'" ~s the braids of X with respect to
qb q2,q3
By a suitable change of coordinates, one can assume that ql = [1,0,0],
q2 = [0,1,0] and one can take
r : Lq3 --+ ]pI, [x, Y, z] ~ [x, z]. (6)

Under these assumptions it is shown in [3], [4] how one can compute the
expression of the braids
Rbk = ~k(HI"" ,Hr-d E 1fr

as words in the standard generators HI, ... ,Hr - l . The following example
will be used below:
Example 8. Let X S;;;; r 2 be the plane curve which is the union of the
curve defined by the affine equation
(y - x2)X(x - 1) (x - (r - 4)) = 0
and the line at infinity (given by the equation z = 0). Let ql =
[1,0,0], q2 = [0,1,0] and q3 = [-1, 1] = (-1, !, !).
Let T be as in Equa-
tion 6, Po = {P17 ... ,Pr} := r(X n L q3 ), where PI < P2'" < Pr-l (with
respect to the real ordering) and Pr = 00. This ordering gives rise to a
system of standard generators HI, . .. ,Hr - l of 1fr as described in Sec-
tion 4 (consider a homeomorphism 0' : r 1 --+ ]pI which maps lR U {oo}
to lR U {oo}, such that a(i) = Pi, i = 1, ... , r - 1 and a(r) = 00, com-
pare also to [4], Section 1). Let bi = (-1, Yi), i = 1, ... ,8 - 1 where
we assume that Yl < Y2 < ... < Ys-l and let bs = [0,1,0] (s = r - 2).
Choose generators ObI"'" t5bs_1 of 7rl (8(X), Q3), where t5b~ encircles bi
counter-clockwise as indicated in the following picture:
154 M. DETTWEILER
Set SeX) := im(w) and let 'H(X) := im(W*). It is clear that if W maps
SeX) injectively to SeX), then the elements RbI"'" Rb. generate the
image of 1Tl (S(X), Po) in 'Hr (under the homomorphism which is induced
by inclusion of paths). If ~ : 'Hr(G) -+ Ur is as in the last section, then
it follows from Equation 3 and Diagram 5 that Rbk E 'H(X) acts (as a
word in HI"'" Hr- l ) according to Equation 4 on Nr(G) == ~-l(PO).
By covering theory, the orbits of 'H(X) on N r(G) correspond then to the
irreducible components of ~-l(S(X)). The genus of such a component
is given as the number g below:
An 'H(X)-orbit 0 ~ Nr(G) is said to be of genus g if
1 s
g = 1 -101 + 22)101 - ni),
i=l
where ni is the cardinality of the cycles occurring in the disjoint cycle

decomposition of Rbi' acting on O.
Since X is defined over k, Gk = Gal(k/k) permutes the irreducible
components of ~-I(S(X)), inducing an action of Gk on the set of orbits
of 'H(X) on Nr(G).
Criterion 9. Let X ~ JP>2 be a reduced plane curve over k ~ C and let
~I"" ,Rb. E 'Hr be the braids of X with respect to qb Q2, Q3 E JP>2(k).
Assume that r ~ 3. Let G be a finite group and C = (Cl , ... , Cr) a
tuple of conjugacy classes of G such that the set N (G, C) is non-empty.
Then there exists a point Po = {PI, ... ,Pr} E SeX) and a Galois cover
cf> of JP>1(e) \ Po which has k as a field of moduli and ramification type
[G, Po, C], CPi = Ci , if the following conditions are satisfied:
a) The map \lI : SeX) -+ SeX) is injective.
b) There exists a Gk-stable orbit 0 of'H(X) on R(G,C) of genus O.
c) There exists i E {I, ... , s} with the following property: In the cycle
decomposition of Rbi on 0 there is one cycle length with odd multiplicity
and the number of braids Rbj , j = 1, ... ,s, with the same cycle decom-
position as Rbi' is odd.
Proof. Under the assumptions, the component of ~-I(S(X)) which be-
longs to 0 is a rational curve with a dense set of k-rational points. Every
such point corresponds to a cover with ramification type [G, P, e), P E
SeX), and field of moduli k, by [10], Thm. 1. 0
Remark 10. The Gk-invariance in Criterion 9 of an orbit 0 c R(G,C)
is assured, if there exists a point Qo in SeX) such that [G, Qo, CJ is a
k-rational ramification type and differs from the other orbits of H(X) on
R(G,C) by its length and/or its genus, compare to [4J, Section 3.
7. Middle convolution applied to subgroups of

GL 2 (lFq ).
Theorem 11. The projective symplectic groups PSP2n(lFp2) occur Q-
regularly as Galois groups over Q(t) if P is an odd prime 1 mod 24.
Proof We prove that the claim in two steps:
i) p = 5 mod 12 : Let H ~ D12, where D12 denotes the dihedral
group of order 24. For r ~ 5, consider the class vector
Cr := (2a, 2b, 2a, 2b, 2c, ... ,2c),
where 2a, 2b are the two conjugacy classes of non-central involutions of
H and 2c denotes the conjugacy class of the central involution, occurring
r-4-times. Let X and Rbi' i = 1, ... ,s be as in Example 8. Since the last
r - 4 conjugacy classes consist of central elements, the only elements in
{Rbi' ... '~s} which act non-trivially on N(H, Cr ) are ~i = H1H2H1,
~2 = H and Rbs (use Example 8). One can compute that N(H,Cr )
has cardinality 16, that 1i(X) acts transitively onN(H,Cr ) and that the
cycle decompositions of Rbp Rb2' Rbs are (2)8, (6)2(2)2, (6)1(3)2(2)1(1)2
(respectively). Using Remark 10, one easily checks that the conditions of
Criterion 9 are satisfied for X and Cr. Thus there exists a point Po E S(X)
and a Galois cover of r1(C) \Po which has ramification type [H,Po,Cr ]
and Q as a field of moduli.
Since 12 I (P2 - 1), we embed H in the general orthogonal group
Got (IFp2). Let r := r - 1. Let 01, ... ,0", be as in Section 3. Then 4J is
the Galois cover associated to (01, ... ,0",) and a tuple (All"" A",) E
GOt(lFp2)"', where Ai E Ci . Applying the middle convolution MC-1 to
(A 1 , , A",), one obtains a tuple of matrices
- - r-
(BlI"" Br) E SP2n(lFp2) ,n = r - 3,
by Thm. 2, a) and e). Moreover, ih, ... , B4 are transvections, Bs, ... , Br
are double transvections and
B1 B", =-1
by Lemma 3. By Thm. 2 c), the group (B1,"" B r ) is an irreducible
subgroup of GL2n(lFp2). Analogously to [6], Section 8, one concludes
that, under the assumptions on p,
(B1,"" B",) = SP2n(lFp2)
(using Thm. 2, b)).
Let 61, ... ,6", denote the conjugacy classes of BlI ... , B", and 6H1
the class of (B1 ... B", )-1 = -1.
156 M. DETTWEILER
We want to show that 61 = 63 : Let (A~, ... , A~, -1, ... , -1) E
(2a, 2b, 2a, 2b, 2c, ... ,2c) such that A3 = A~ and A~ = A~ -1 A~A~. It
is easy to check that
Ql1Q2Ql(A~, ... ,A~,-I, ... ,-I) = (A~, ... ,A~,-I, ... ,-I),
where Q1, Q2 are the first two generators of the abstract Artin braid
group, acting via Equation 1. Let
MC>.(A~, ... ,A~,-l, ... ,-l) = (EL ... ,E~).

The normalizer of SP2n(JFp2) in GL2n(JFp2) is the group of symplectic
similitudes GSP2n(JFp2). By Thm. 2 d), there exists an element B E
GSP2n (IFp2) such that
Q11Q2Ql (Hi, ... ,H~) = (B~, ... ,H~)B.

Since Q11Q2Q1 leaves the fourth element unchanged, B can be as-
sumed to be contained in SP2n (IFp2) (every element 1/ E GSP2n (IFp2) \
Z (GSP2n (IFp2 )) . SP2n (IFp2) permutes the two classes of transvections in
SP2n(lFp2)). It follows now from Thm. 2 d) that (\ = 63 .
Let 4> be the Galois cover associated to (/31, ... , /3i') and (HI, ... , Ei' ).
Now, since the classes 61 , ... , 6H1 are rational (in the sense that = 6t
6i for l coprime to ISP2n(lFp2)1) and 61 = 63 , 4> has a Q-rational rami-
fication type.
Again, since every element 1/ E GSP2n(lFp2) \Z(GSP2n (lFp2)) ,SP2n(lFp2)
permutes the two classes of transvections in SP2n (IFp2), one has
Z(GL2n(lFp2)) ,SP2n(lFp2) = {g E NGL2n(JFp2) ( (HI, ... ,Bi') Ig(Gi ) = Gi }.

It follows from the last equality and Cor. 6 that the cover 4> associ-
ated to '!pI,'" ,/3i') and (ih, ... , Bi') has Q as a field of ~oduli. Let
E1 , .. , Ei' be the images of E1 , ... ,Ei' in PSP2n(lFp2) and 4> be the Ga-
lois cover corresponding to the kernel of the homomorphism
71"1(,1((:) \ PO,Pr+1) ---t PSP2n(Fp2), /3i f-+ Bi, i = 1, ... ,r.

Of course, Q is also a field of moduli of 4>. Since the groups PSP2n(Fp2)
have trivial center, 4> is defined over Q, see [15], Prop. 3.6. This proves
the claim.
ii) P == 3 mod 8 : The claim follows analogously to i), using the non-
central classes 2a, 2b and the central class 2c of Dg, the class vector Cr :=
(2a, 2b, 2a, 2b, 2c, ... ,2c) and the curve X as above: One checks that
'H(X) acts transitively on N(H,Cr ) and that the cycle decompositions

of Rbll Rb2' Rbs are (2)4, (4)2, (1)2(4)1(2)1 (respectively). Proceeding
as above, one obtains the result for p == 3 mod 8. 0
Acknowledgments
The author would like to thank Professors K. Hashimoto, K. Miyake
and H. Nakamura for the invitation to the conference Galois Theory
and Modular Forms (held at the Metropolitan University of Tokyo),
T. Shiina for sending his papers and Professor B. H. Matzat, S. Reiter
and the referee for valuable comments.
References
[1) J. S. Birman, "Braids, Links and Mapping Class Groups," Princeton University
Press, Princeton, 1974.
[2] E. Bloch, "A First Course in Geometric Topology and Differential Geometry,"
Birkhauser, Boston, 1997.
[3] M. Dettweiler, Kurven auf Hurwitzraumen und ihre Anwendungen in der Ga-
loistheorie, Dissertation, Erlangen, 1999.
[4) M. Dettweiler, Plane curves and curves on Hurwitz spaces, IWR-Preprint (2001-
06).
[5] M. Dettweiler and S. Reiter, On rigid tuples in linear groups of odd dimension,
J. Algebra 222 (1999), 550-560.
[6] M. Dettweiler and S. Reiter, An algorithm of Katz and its application to the
inverse Galois problem, J. Symb. Compo 30 (2000), 761-798.
[7) M. Dettweiler and S. Reiter, Monodromy of FUchsian systems, in preparation.
[8) M. Dettweiler and S. Wewers, Hurwitz spaces and Shimum varieties, in prepa-
ration.
[9) E. R. Fadell and S. Y. Husseini, "Geometry and Topology of Configuration
Spaces," Springer Verlag, Heidelberg, 2001.
[10) M. Fried and H. VOlklein, The inverse Galois problem and mtional points on
moduli spaces, Math. Ann. 290 (1991), 771-800.
[11) N. Katz, "Rigid local systems," Princeton University Press, Princeton, 1996.
[12) G. Malle and B. H. Matzat, "Inverse Galois theory," Springer Verlag, Berlin,
1999.
[13) T. Shiina, Rigid bmid orbits related to PSL2(P2) and some simple groups,
preprint (2002).
[14) T. Shiina, Regular Galois realizations of PSL2(P2) over Q(T) , to appear in this
volume.
[15] H. V6lklein, "Groups as Galois groups," Cambridge Univ. Press, Cambridge,
1996.
158 M. DETTWEILER
[16] H. VOlklein, The braid group and linear rigidity, Geom. Dedicata 84 (2001),
135-150.
[17] H. Volklein, A transformation principle for covers ofpl, J. Reine Angew. Math.
534 (2001), 155-168 .
ON THE ESSENTIAL DIMENSION

OF p-GROUPS
Arne LEDET
Department 0/ Mathematics and Statistic
Texas Tech University
Lubbock, TX 79409-1042
aledet~math.ttu.edu
Abstract We improve the known bounds on the essential dimension of p-groups

over (large) fields of characteristic p.
1. Introduction
In [Le] , the author provided a construction of generic polynomials
with few parameters for p-groups over ground fields of characteristic p,
thus giving an (unstated) upper bound on essential dimensions in this
case. In the present paper, we improve this bound further, by proving
Theorem 1. Let K be a field of characteristic p -:f:. 0, and let
be an extension of p-groups, where A is elementary Abelian and maps

into the Frattini subgroup ~(E) of E. Assume that IAI :5 IKI. Then
Here, the essential dimension edK G of a finite group G over a field K

is defined as follows, cf. [BR]: Given a Galois extension MIL with
Gal(MI L) ~ G and L ;2 K (which we will refer to as a G-extension
over K), we consider all 'intermediate' G-extensions mil:
[received: May 17, 2002; accepted in revised form: October, 16, 20021
160 A. LEDET
The essential dimension of MIL over K, written edK (MIL) , is then

defined as the minimal possible transcendence degree of [ (or m) over K.
Since m can be chosen to be finitely generated, this essential dimension
is clearly finite, and in fact less than IGI: Let m = K({oB}O"EG), where
{o-B}O" is a transcendence basis with 2:0" (TO = 1.
Since edK(M/L) is always less than IGI, we can now define the essen-
tial dimension of the group Gover K, written edK G, as the maximal
value assumed by edK(MIL), when MIL runs through all G-extensions
over K.
Thus, the essential dimension gives an upper bound for the num-
ber of algebraically independent parameters needed to describe a G-
extension MIL over K. Here, the extension MIL is considered 'de-
scribed' by an intermediate extension mil as above, since the Galois
theoretical structure of MIL is completely given by the fact that it is
obtained from m/[ by extending the scalars from [ to L.
Example. The only group G that has essential dimension 0 is the

trivial group, since essential dimension 0 must mean that every G-
extension MIL over K must be described by an intermediate G-
extension m/[, where [ is algebraic over K. And clearly, if G =1= 1 we
can produce MIL for which this fails.
Over the field Q of rational numbers, the cyclic groups C2 and C3 , of
order 2 and 3, both have essential dimension 1. This is obvious for C2: A
quadratic extension has the form L( Ja) / L, and we can let m = Q( va)
and [ = Q(a). For C3, it requires a little more work: First, we find
x, y E M such that x and yare linearly independent over L, and such
that x, y and -(x+y) are conjugate over L. We can then let m = Q(x/y)
and [= Q(xly)Cs.
Again over Q, the cyclic groups of order 4 and 5 both have essential
dimension 2. We will not prove this here, but only observe that they can
easily be proven to have essential dimension at most 2. To get equality,
it is then necessary to invoke a result implicit in both [BR, Thm. 6.2]
and (JLY, Ch. 8]: If edK G = 1, then G is isomorphic to a subgroup
of PGL2(K).
In [BR1, Buhler and Reichstein prove a number of results about essen-

tial dimensions, including some lower bounds. In general, lower bounds
are substantially harder to obtain than upper bounds, and this paper
will be concerned only with upper bounds. (It may be of interest to
note that in many cases, the only lower bound known is the one ob-
tained by proving that the group is not a subgroup of PGL2(K), i.e., a
lower bound of 2.)
Theorem 1 is proved in section 2 below.
In terms of essential dimensions, the results of [Le1 amount to proving
the inequality
edKE:::; edKG+ 1
in the case IAI = p, and combining it with
Lemma 2. Let K be a field of characteristic p # 0, with IKI ~ pn.
Then
edK(C;) = 1,
where C; is the elementary Abelian group of order pn.
In characteristic 0, it is proved in [BR] that the essential dimension
of C; is at least n, with equality when the ground field contains the
nth roots of unity. Intuitively, this makes sense: A C;-extension is the
composite of n linearly disjoint Cp-extensions, and we clearly need at
least one parameter to describe a Cp-extension. It may therefore come
as a small surprise that we need only one parameter in characteristic p
(provided that ground field is large enough). At least it initially surprised
the author, and led to the realization that p-groups in characteristic p
have 'small' essential dimensions.
We will prove Lemma 2 in section 2 below, but refer the reader also
to [JLY]. Clearly, Theorem 1 and Lemma 2 provide a very low bound for
the essential dimension of a given p-group over 'sufficiently large' fields.
In most cases, this bound will be lower than the one obtained in [Le],
since we are guaranteed the existence of an A of order at least p2 if ib(E)
is non-cyclic, by [Hu, III.7.5 Hilfssatz].
Example. Let n E N, and consider the group
p
pc a 1 + pd
G n = {( l+ b)1 a,b,c,d E Zipn}
of 2 x 2 matrices over Z/pn, where p is an odd prime. It is a p-Sylow
subgroup of GL 2 (Z/pn), and has order p4n-3. Using Theorem 1 and
Lemma 2, we find that
162 A. LEDET
provided IKI ~ p4. See section 3 be;low for details.

Example. Let G be an Abelian p-group of exponent pn, minimally
generated by e elements. Then edK G ~ n if IKI ~ pe.
Remark. Both Theorem 1 and Lemma 2 requires the ground field K to
be 'big enough'. However, for an arbitrary ground field K and a finite
group G we have
edK G ~ edK(t) G ~ edK G + 1,

where t is an indeterminate. Here, the first inequality is clear from the
definition of essential dimension: We cannot need more algebraically
independent parameters to describe G-extensions over a larger field. The
second is also quite easy: If M j L is a G-extension over K, we either
have that L j K is algebraic and hence that edK (M j L) = 0, or that
M j L is a G-extension over K (t). Since edK G is the maximum of all
possible edK(MjL)'s, the first case does not contribute. And in the
second, we have edK(MjL) ~ edK(t) (MjL) + 1, simply because we can
consider t as an extra parameter.
Thus we get a bound for the essential dimension also in the case where
K is finite of characteristic p, and G is a p-group that is 'too large' to
allow a direct application of Theorem 1 and Lemma 2.
For instance, C; has essential dimension 2 over a ground field K of
order less than pn. (Here, it is clear that the essential dimension is at
most 2. To prove that it cannot be 1, it is enough to see that a p-Sylow
subgroup of PGL 2 (K) is elementary Abelian of order IKI.)
And of course: Unless the r,roup is cyclic, it does not occur as a Galois
group over a finite (or absolutely algebraic) field in prime characteris-
tic at all. So in that case, there is no loss in considering the ground
field lB'p(t). And a cyclic p-group is covered by our results, as well as by
the classical theory of Witt vectors, cf. [Wi2].
Remark. It seems a reasonable conjecture that the cyclic group Cpn of
order pn has essential dimension n over any field of characteristic p. Since
one of the results of [BR] is that the essential dimension of a subgroup
is less than or equal to that of the entire group, this would provide a
lower bound for the essential dimensions of p-groups in characteristic p
in terms of the exponent of the group, i.e., the maximal order of an
element. For instance, the group Gn from the Example above contains
an element of order pn, and would therefore have essential dimension
exactly n.
Unfortunately, it is at this point only a conjecture. Also, while it
would allow us to determine the essential dimension in many cases, such
as Gn or Abelian p-groups, there are p-groups for which the essential
dimension is strictly greater than the bound thus given. For instance,
the Heisenberg group of order p3, when p is odd, i.e., the non-Abelian
group of order p3 and exponent p: The exponent provides a trivial lower
bound of 1 for the essential dimension, but it is easily seen that the
essential dimension cannot be 1.
2. Proof of Theorem 1
Let the assumptions be as in Theorem 1. We first prove that any
G-extension M / Lover K can be extended to an E-extension. This will
establish the inequality
since any intermediate E-extension gives rise to an intermediate G-

extension of the same transcendence degree.'"
Next, we prove that any E-extension F / L containing M / L arises in
the manner established in the first part, and note that this description
involves only one parameter. This, then, gives us
Part one
Let M / L be a G-extension over K. Also, let c E Z2 (G, A) be a factor
system representing the extension
l-+A-+E-+G-+l.
This extension makes A into a IFp[GJ-module in the usual way (by con-
jugation in E), and we will need an injective IFp[G]-homomorphism
c.p: A -+ M+:
Let c.p': A <---+ K+ be an injective group homomorphism, and let
(U~)O'EG be a normal basis for M / L. Then we can let
c.p(a) = L c.p'(u-1a)ue, a E A.
O'EG
We also have an induced homomorphism

c.p2: H2(G,A) -+ H2(G,M+).
Note that we cannot simply say 'G is a homomorphic image of E, therefore edK G $ edK E.'
No such result is available (at this time), al1J in the more general setting of algebraic groups
it is known to be untrue, cf. [ReJ.
164 A. LEDET
By, e.g., [We, Cor. 3-1-4] the cohomology group H2(G, M+) is trivial,
and so there exists a map a: G - M with
'Vu, T E G: <P(CU,T) = au + uaT - aUT.
Let
H(X) = II (X - cp(b)) E L[X].
beA
This is a vectorial polynomial, as defined in [Ab] , in that H(x + y) =
H(x) + H(y). Now,
H(au ) + uH(aT) - H(aUT ) = H(CP(CU,T)) = 0,
meaning that u 1-+ H (au) is a crossed homomorphism. Since, again by
[We, Cor. 3-1-4], we also have Hl(G,M+) = 0, there therefore exists
an wE M with
'Vu E G: uw - w = H(a u).
Let B be a root of H(X) -w. We claim that M(B)jL is an E-extension:
The vectorial polynomial H(X) has the form
XP
n
+ dn-1XPn-l + ... + dlXP + doX,
where do =f:. O. It follows that H(X) - w is separable. Next, a conjugate
of B (over L) must be a root of H(X) - uw for some u E G, i.e., have
the form 0 + au + cp(b) for abE A. Therefore M(O)jL is Galois.
Extend (j E G to jj E Gal( M (0) j L). This extension is then given by
jj: 0 1-+ 0 + au + cp(bu )
for some bu E A.
Clearly, Gal(M(B)jM) can be identified with a subgroup B of A
through T 1-+ cp-l(TB - 0). We let b E B act on B by b: 01-+ B + <p(b).
Then
(ub) 0 jj = jj 0 b,
giving the proper action of G on B in
1 - B - Gal(M(B)jL) - G - 1.
Also,
jjf = [CU,T + (bu + ubT - bUT )]UT,
allowing us to embed Gal(M(B)jL) into E. Since Gal(M(B)jL) maps
onto G, and A ~ cIl(E), this is possible only if Gal(M(B)jL) = E.
Ergo: M(O)jL is an E-extension, and we may pick bu = O.
We note that replacing w by w + t for atE L will not change the
nature of the extension.
Part two
Let again MIL be a G-extension over K, and consider an E-extension
FIL containing MIL. 'Ve claim that F = M(Ot) for some tEL, when
Ot is a root of H(X) - (w + t), and w is obtained as above:
Since F1M is an A-extension, and <p can be considered as a crossed
homomorphism A -+ F+, we can find 0' E F with
'Va E A: a(O') = 0' + <pea).

Clearly, then, F = M(O'). Also, from u 0 a = (ua) 0 u we get
a(uO') = uO' + <pea),
and hence
uO' - 0' EM.
Let bO' = uO' - 0'. Then
bO' + ubT = ufO' - 0' = cO',T7fTO' - 0' = bUT + <p( cO',T),
i.e.,
<p( cO', T) = bO' + ubT - bUT
We already have one set of splitting factors for c, namely the au's. Then,
(J 1-+ bO' - au is a crossed homomorphism G -+ M+, and therefore
bO' = au + ((Jx - x)
for some x E M, or
u(O' - x) = (0' - x) + au.
We replace 0' by 0" = 0' - x. Then F = M(O") and
(jO" = 0" + au.

This of course means that
uH(O") - H(O") = H(uO") - H(O") = H(aO').
Let w" = H (0") EM. Then uw" - w" = H (au), and hence
w" =w + t
for some tEL. Ergo: F = M(Ot).
The details of why edK E ~ edK G + 1 are now clear: We can find
<p, w and the au's inside any intermediate extension mit of MIL, and
166 A. LEDET
get the intermediate E-extension m(t,8t )/[(t), where the transcendence

degree is obviously at most 1 greater than for miL
This completes the proof of Theorem 1. 0
Remark. The fact that an embedding problem as in Theorem 1 always
has a solution was first proved by Witt in [Wil, 1936]. This result does
not depend on the size of the fields involved, since we can simply restrict
ourselves to the case IAI = p, applied repeatedly. In fact, Witt's result is
that an embedding problem given by a G-extension M I K and a group
extension
1 - <'P(E) - E - G - 1
is always solvable, if K has prime characteristic p and the groups E
and G are p-groups. This result is an obvious consequence of the argu-
ment given in Part One above, and indeed the argument given here is
basically a generalization of the proof in [Will.
Proof of Lemma 2
Let K be a field in prime characteristic p of order at least pn, and
let M / L be a C;-extension over K. We let c.p: C; <-t K+ be an injec-
tive group homomorphism, and consider it as a crossed homomorphism
Gal( MIL) - M+. It is then principal, i.e., there exists ~ E M with
V(j E C;: (j~ = ~ + c.p((j) ,
and therefore in particular M = L(~). We now let m = K(~) and
[ = mCp. Clearly, m (or l) has transcendence degree at most lover K,
meaning that edK(C;) ~ 1. On the other hand, let M = K(t) for an
indeterminate t, and define a C;-action on M by (jt = t + c.p((j). Since
there are no non-trivial algeb::-aic extensions of K inside M, we must
necessarily have edK(M/Mcp) = 1, and therefore edK(C;i') = 1. 0
Remark. One of the interesting things about the essential dimension
for a group G over a field K is that it gives a lower bound for the number
of parameters needed in a generic polynomial for Gover K, cf. [JLY].
Here, a generic polynomial for a finite group G over a field K is a
monic polynomial P(t,X) E K(t)[X], where t = (tl,"" tn) and X are
indeterminates, such that
(a) P(t, X) has Galois group Gover K(t); and
(b) whenever MIL is a G-extension over K, M is the splitting field
over L of a specialization P(a, X) of P(t, X) over L.
It is not clear whether the construction above can be made to include
a generic polynomial in the general setting. However, if we restrict our
attention a little, requiring the successive group extensions - with the

exception of the first one-to be central, i.e., with A ~ Z(E), we can
produce a generic polynomial as well. For simplicity, we assume K to
be 'large':
(1) If we are considering an elementary Abelian group we em- C;,
bed C; into K+ by an injective homomorphism <p: C; ~ K+, and let
H(X) = ITuECp-(X - <p(o")). If MIL is then a C;-extension over K, we
e e
have M = L(e) for a with O'(e) = + cp(O'), and we conclude that
H(X) - t E K(t)[X] is generic, with the single parameter t. (This is
essentially the proof of Lemma 2.) Also, the element pn - I generates a e
normal basis for MIL: We know that (ei)f:ol is a basis for MIL, and
that pn_l
(e + <p(0'))pn-1 = L (_I)i<p(0')pn-1- i i ,e
i=O
since (pni-l) == (_I)i (mod p). The matrix transforming (ei)i into
(O'epn-1)u therefore has rows of the form
[<p(O')pn_l, _<p(0')pn_ 2, ... , -<p(0'), IJ,
meaning that it is basically a Vandermonde matrix (up to a few signs
and some swapping of columns), and so has non-zero determinant. This
proves that (O'epn-1)u is a basis for MIL. In particular, epn - I has non-
zero trace,t and epn - 1 I TrM/L(epn-l) has trace 1.
(2) If we are considering a group extension as in Theorem 1, with G
elementary Abelian, we can proceed as in the proof above. We have a
normal basis from step (1), and to get the various elements, i.e., split-
ting factors etc., we just need to make the results HI (G, M+) = 0 and
H2(G,M+) explicit:
Let a E M have trace 1. If 'IjJ: G -4 M+ is a crossed homomorphism,
we can let (3 = - L:UEG 'IjJ(O')O'a to get 'IjJ(0') = 0'{3 - (3, and if c: G x
G -4 M+ is a factor system, we get CU,T = au + O'aT - aUT by letting
au = L:pEG cu,pO'pa.
Note that these results allow us to construct all necessary quantities
in a purely 'mechanical' way, i.e., one that carries over homomorphic ally
under the kind of specialization used with generic polynomials. A new
element of trace 1 can then be found for the next step by taking such
elements in MIL and in F / M, and multiplying them.
tIn fact, pn - 1 is the smallest positive exponent k, for which ~k has non-zero trace. This
is most easily seen by considering the trace of ~k as a Newton power sum, since most of the
coefficients of H(X) - t are zero. In this ~ay, we show that ~k has trace 0 for 0 :::; k < pn -1,
and get an expression for the traCJ of ~p -1.
168 A. LEDET
(3) In the case of a (central) group extension as in Theorem 1, we

proceed as in (2). This time, we do not have a normal basis at our
disposal, but by assumption we will not need it.
In the general case, the problem lies exactly in the need for G-
equivariant embeddings. Depending on the size of K, it may not be
necessary to find actual normal bases, but it is required to produce the
G-equivariant embeddings in an algorithmic way at each step.
Example. To illustrate the construction of generic polynomials using
the procedure described above, we consider the cyclic group Cp2 of or-
der p2 over the field IFp'
Let K be a field in characteristic p, and let M / K be a Cp 2-extension.
There is then an intermediate Cp-extension L / K. Also, let a denote a
generator for the Galois group Gal(M/K) = Cp 2.
The group extension is
1 - Cp - Cp 2 - Cp - 1,
and we choose cp: (JP 1-+ 1 as our injection of Cp into IFp.
There is then, cf. the well-known Artin-Schreier theory, an element
a E K, such that L = K((}) for a () with a(} = (}+1 and H((}) = (}P-() = a.
The trace of (}p-l is -1, and we can pick _(}p-l as our element of
trace 1.
If we let a ' denote the restriction of (J to L, the 'obvious' choice of
factor system is given by picking a i as the pre-image of a'i for 0 ~ i < p.
To get splitting factors, we then only need to find au, E L, and the
only requirement for au, is that it should have trace 1. Thus, we can let
au, = _()p-l.
We now seek w E L with aw - w = H(au ') = _H((}p-I). Since
(}P = () + a, this means
aw - w = _[(}p(p-I) - (}p-I] = _[(() + a)p-l - (}p-I]
= _ L
p-I (
p ~ 1) aP-I-i(}i + (}p-I
i=O 2
p-2
= - L (- a )p-I-i (}i
i=O
Clearly,
(~)
Call the matrix A, and note that A is independent of the specific Cp2-
extension M / K. Then, if A -1 has ith row [ail, ai2, ... , ai,p- d, we get
p-l p-1
O'Xi-l - Xi-l = O'(I: aijBj) - I: aijBj = Bi - 1,
j=l j=l
and thus
p-2
W = - L(-a)p-l-i xi .
i=O
It follows that M = L(~) for a ~ with ~p - ~ = w + b, bE K.

To get a generic polynomial, we consider the two 'parameters' a and b
as indeterminates s and t, to get B and ~ defined over lFp(s, t). The
e
minimal polynomial for over lFp(s, t) is then generic, since we can define
a homomorphism from lFp[s, t] to K by S 1-+ a, t 1-+ b, and extend it to
a map lFp[s, t, B,~] -+ M. Thus, the corresponding specialization of the
minimal polynomial is well-defined and has splitting field Mover K.
In the simplest special case, p = 2, we get O'W -w = a, and so w = aB.
e
The minimal polynomial for over K is then
P(X) = X4 + (1 + a)X2 + aX + (a 3 + ab + b2).

It follows that
is generic for C4 over lF2.

We note that this is the same polynomial we would get from using
Witt vectors: In terms of Witt vectors, M = K(x, y), where O'(x, y) =
(x,y) + (1,0) = (x + 1,x + y). Also, with (u,v) = p(x,y) = (x 2,y2)_
(x, y) = (x 2 - x, y2 - Y + x 3 - x 2), the minimal polynomial for y over K
is
X4 + (1 + u)X2 + uX + (u 3 + uv + v 2),
and in the case of a 'general' Witt vector, i.e., M = lFp(x, y) with x
and y indeterminates, the fixed field is MC4 = lFp(u, v).
3. The matrix group

We recall that
Gn = {C;; 1: i Pd) a,b,c,d E

Zjpn} ,
170 A.LEDET
and that p is an odd prime. For n = 1, this is just a cyclic group of

order p, and the essential dimension is known to be 1.
Let therefore n > 1. We first note that Gn is then minimally generated
by the three matrices
A = C6 p ~), B = Gi), c = (~ ~).

This becomes clear, once we notice that
For n > 2, this means that the kernel of the natural map Gn - Gn- 1 ,
{C ;!~~la 1!::~ld) Ia,b,c,d EZjpn} ,

i.e.,
Kn =
is contained in <p(G n ). Also, a simple computation shows that it is
elementary Abelian of order p4. Therefore,
edK Gn ::; edK Gn-l + 1, n > 2,

or
edK Gn ::; edK G2 + (n - 2).
For n = 2, we have that 1<p(G2 )1 = p2. Moreover, BP E <p(G 2 ) and
BCB-1C- 1 = C; p ?p) 1 EG 2~ <P(G2).

Since BP and BCB-1C- 1 are not powers of each other, we must have
<P(G2) ~ C;,
and so
which implies that

edK Gn ::; n.
Remark. A simple calculation shows that
Z(Gn) = {(~:-f~ f:~~) I a,b,c E Zjpn} ,

and hence that the subgroup
is central of exponent p, and contained in Kn. Since Kn/Zn is normal

in Gn / Zn of order p, it mu.st be central as well. Thus, by replacing the
single group extension
1 -+ Kn -t Gn -+ G n- l -+ 1, n > 2,
with two extensions
and
1 -+ Zn -+ Gn -+ Gn/Zn -+ 1,
we get Gn 'built up' through central extensions. Consequently, by the
Remark in section 2, it is possible to construct generic polynomials by
means of these extensions. And assuming that IKI ~ p3, we get for the
number of parameters that
#(parameters for G2 ) = 2
and
#(parameters for Gn ) :::; #(parameters for Gn-d + 2,
i.e., a generic polynomial for Gn can be constructed with 2n - 2 param-
eters.
Remark. The other Example, of an Abelian p-group, can be treated by
means of Witt vectors in a way very similar to the proof of Lemma 2:
If A is an Abelian p-group of exponent pn, generated by e elements,
and K is a field of characteristic p and order at least pe, we can find an
injective group homomorphism cp: A ~ Wn(K)+, where Wn denotes the
ring of n-dimensional Witt vectors, cf. [Wi2]. In fact, if al,"" ae E K
are linearly independent over lFp, the Witt vectors (at, 0, ... ,0), ... ,
(a e, 0, ... ,0) will generate a subgroup of Wn(K)+ isomorphic to C;n.
Now, let M/ L be an A-extension over K. Then cp can be considered as
a crossed homomorphism from A into Wn(M)+, where A acts on Wn(M)
coordinate-wise. The cohomology group RI(A, Wn(M)+) is trivial, as
can be easily shown in general in the same way as for n = 1 (done in
section 2 above), and so there exists a 8 E Wn(M) with a8 = 8 + cp(a)
for all a. This means that M = L(8) (with 8 considered as a set on n
elements), and letting m = K (8) we see that edK A :::; n.
Moreover, by one of the results in [Le] we have: If t = (to, ... , tn-d
is a set of n indeterminates, and we let A act on K (t) by at = t +
cp(a), the fixed field K(t)A is a rational (purely transcendental) extension
of K. Consequently, if K is infinite, we get a generic polynomial with
172 A. LEDET
n parameters: If K(t)A = K(s), with s = (so, ... , Sn-l), any monic

polynomial P(s, X) E K(s)[X] with splitting field K(t) will be generic,
cf. [JLY, Prop. 1.1.5].
References
[Ab] S. S. Abhyankar, Galois embeddings for linear groups, Trans. Amer. Math.
Soc. 352 (2000), 3881-3912.
[BR] J. Buhler and Z. Reichstein, On the essential dimension of a finite group,
Compositio Mathematica 106 (1997), 159-179.
[Hu] B. Huppert, "Endliche Gruppen I," Grundlehren der mathematischen Wis-
senschaften, 134. Springer-Verlag, 1967.
[JLY] C. U. Jensen, A. Ledet and N. Yui, "Generic Polynomials," MSRI Publica-
tions, 45. Cambridge University Press, 2002.
[Le] A. Ledet, On p-group in characteristic p, preprint, 200l.
[Re] Z. Reichstein, On the notion of essential dimension for algebraic groups,
Transform. Groups 5 (2000), 265-304.
[We] E. Weiss, "Cohomology of Groups," Pure and applied mathematics, 34. Aca-
demic Press, New York, 19G9.
[Will E. Witt, Konstruktion von galoisschen Korpem der Charakteristik p zu
vorgegebener Gruppe der Ordnung pI, J. Reine Angew. Math. 174 (1936),
237-245.
[Wi2] E. Witt, Zyklische Korper und Algebren der Charakteristik p vom Grad pn, J.
Reine Angew. Math. 116 (1937), 126-140.
EXPLICIT CONSTRUCTIONS
OF GENERIC POLYNOMIALS
FOR SOME ELEMENTARY GROUPS
YUichi RIKUNA
Department of Mathematical Sciences,
Waseda University,
9-4-1 Ohkubo, Shinjuku-ku, Tokyo
169-8555, Japan
rikuna4)gm.math.waseda.ac.jp
Introduction
For a finite group G and a field k, we call a G-Galois extension over
k by G/ k-extension. Whether a G/ k-extension exists or not is the first
version of inverse Galois problem. Especially the case when k = Q,
the rational number field, plays an important role in the study of the
absolute Galois Group of Q. By many mathematicians, the existence
of G/Q-extensions has been shown for a lot of finite groups G by now
(cf. Malle-Matzat [14), Serre [19), etc.)
Recently the next stage of inverse Galois problem, so-called construc-
tive inverse Galois problem, has begun to be studied. If there exists a
G / k-extension, the problem asks for constructing all G-extensions over
the base field k concretely. It, however, is extremely difficult for us to
construct a general theory for this problem, because this contains the
construction problem of non-commutative class field theory. Therefore
constructive inverse Galois problem has been studied only for relatively
small or simple-structured groups until now. But we should remark here
that even study for the cyclic groups brings considerable progresses for
algebraic number theory.
Here we study explicit constructions of generic polynomiaL for some
simple-structured groups. A k-generic G-polynomial F is defined as a
G-polynomial over a rational function field k(t1, ... , t n ) such that every
G-extension L/ K with K ::> k and #K = 00 is obtained from F by a spe-
cialization of parameters tl, ... , tn. To construct such polynomials, we
[received: Jui. 03, 2002; accepted in revised form: Oct. 14, 2002]
174 Y. RIKUNA
compute an explicit positive answer of linear Noether(-Liiroth)'s prob-

lem. This method is indicated by Kemper [8; 9] and Kemper-Mattig [11].
G. Kemper also gave explicit expressions of generic polynomials for the
generalized quaternion groups in his thesis [8]. By now, this method has
been applied for some other important groups, for example, the cyclic
groups, the modular 2-groups, the quasi(semi)-dihedral groups, ... , etc.
(cf. Hashimoto-Rikuna [6], Jensen-Ledet-Yui [7], and Rikuna [16; 18]).
In this paper we construct, among others, generic polynomials for the
dihedral groups with even degree as a new result by using the above
method. These calculation is based on some results about constructing
cyclic polynomials from linear fractional transformations developed by
Miyake [15], Hashimoto-Miyake [5], and Rikuna [17]. We also study
two ways of descending for generic polynomials, namely, "arithmetic
descent" and "geometric descent". The former way makes the base field
k smaller. The latter way gives a generic polynomials for a subgroup of
G, which is a concrete example of the descent-generic property, which is
defined in Kemper [10]. Finally we give generic polynomials for almost
all finite groups of order up to sixteen by using data of corresponding
Noether's extensions.
Before beginning the following sections, the author wishes to express
his deep appreciation for the referee of this paper who gave him very
careful advices and very warm encouragements.
1. Generic polynomials derived from linear

Noether's problem
Let k be a field. Put k(x) := k(Xl,'" ,xn ), the rational function field
with n variables Xl, ... ,Xn, and let Autkk(x) be its k-automorphism
group. For GLn(k), the n-dimensional general linear group over k, define
an injective homomorphism GLn(k) - - Autkk(x) by (aij) f---+ (Xj ~
L:~=1 aijXi, 1 ::; j ::; n). We denote its image by Aut~neark(x) and we
call an element in Aut~neark(x) a linear k-automorphism of the field
k(x).
Let G be a finite subgroup of Aut~neark(x). We denote by k(xP
the fixed field of k(x) under G. The extension k(x)/k(x)G is G-Galois.
Hence there exists a finite G-stable subset RG of k(x) such that k(x) =
k(x)G(RG). (We call a subset R C k(x) G-stable if R satisfies {a(x) I
X E R, a E G} = R.) This implies that
XERC j
Explicit constructions of generic polynomials 175
is a defining polynomial of the extension k( z) / k( z)G. Our main purpose

is to construct a k-generic G-polynomial from fRG(X), Therefore the
following question is essential for us:
Is the fixed field k(z)G rational, i.e., purely transcendental, over k ?
This problem is called linear Noether(-Liiroth}'s problem for G/k(z).
We now assume that this problem has a positive answer, that is, there is
a k-isomorphism G : k(t) ~ k(z)G, where k(t) := k(tb"" t n ) is the
rational function field with n variables tI, ... , tn. Then "(/ is extended
naturally to an isomorphism between polynomial rings k(z)G[X] ~
k(t)[X], and
FG(t;X):= "(/URa(X)) = Lc;I(cj)xj E k(t)[X] (1.1)

j
has the Galois group Gover k(t).

Example 1. Let N ;::: 2 be an integer and (N is a primitive N-th
root of unity. We define an one-dimensional linear Q( (N )-automorphism
o E Aut~(~~)Q((N)(Xl) by o(xt) := (NX1, and we put G:= (o). This
group G is isomorphic to CN, the cyclic group of order N. In this case,
we can choose tl t---+ xi'" as a Q((N)-isomorphism G : Q((N)(tl) ~
Q((N)(xd G and {(~Xl 11 :s; j :s; N} as RG. Then we have fRG(X) =
XN - xi'" and FG(tl; X) = XN - tl E Q(tl)[X), Kummer theory shows
that this is a Q((N )-generic CN-polynomial.
In general, we can show the following result by using normal basis
theorem:
Theorem 1.1 (theorem 7 in Kemper-Mattig [11]). The polynomial
FG(t; X) E k(t)[X] defined by (1.1) is a k-generic G-polynomial.
To construct the k-generic G-polynomial FG(t; X), we have to
1 construct explicitly an affirmative answer of the corresponding lin-
ear Noether's problem,
2 find a suitable RG and compute the corresponding defining poly-
nomial fRG(X),
3 calculate the coefficients c/(Cj) E k(t).
Remark 2. The separability of the extension k(z)jk(z)G implies that
there exists an element x E k(z) such that RG can be chosen as OrbG(x),
the single G-orbit of x. The polynomial fRa (X) is irreducible over k(z)G
under this selection, hence FG(t; X) is also irreducible over k(t). (See the
176 Y. RIKUNA
next example.) The definition of generic polynomials does not require

their irreducibility. It is, however, important to construct irreducible
generic polynomials in the viewpoint of applicability to algebraic number
theory.
Example 3. Put G:= (al,a2) < Aut~nearQ(Xl,X2)' where
and
Then G is isomorphic to C2 x C2 and we can choose a Q-isomorphism

>a: Q(tl,t2) --- Q(X},X2)a by
x2 + x2 t (XIX2)2
tl 1--+ I 2 and 2 1--+ 2 2.
XIX2 Xl + X2
We take Rl := OrbG(xd = {Xl, X2} and R2 := OrbG(xl + X2) U
OrbO(Xl - X2) = {(Xl + X2), (Xl - X2)}, and hence both fRl (X) and
fR2(X) are defining polynomials of the extension Q(XI,X2)/Q(XI,X2)G.
Therefore we obtain that >"C/URI (X)) and >"C/(fR2(X are Q-generic
C2 x C2-polynomials. The former is irreducible over Q(t}, t2), but the
latter is not. Indeed, we have
{
>"(/(fRl (X = X4 - trt2X2 + trt~,
>"C/(fR2(X = (X2 - (tl + 2)tlt2)(X2 - (tl - 2)tlt2)'
(Naturally, (X2 - tl)(X 2 - t2) E Q(tl, t2)[X] is a Q-generic C2 x C2-

polynomial. )
Remark 4. For the non-linear cases, their defining polynomials have
not always generic. Whether these polynomials are generic or not is
very difficult problem. Only a few examples is known (e.g., Hashimoto [
3; 4], Ledet [12; 13], Jensen-Ledet-Yui [7], etc.)
2. Arithmetic descent
Keep the notation above. In this section we also suppose that linear
Noether's problem for G/k(z) has a positive answer, that is, there exists
a k-isomorphism </>0 : k(t) --- k(z)G. And we choose a finite subset
Ro of k(z) so that k(z) = k(z)G(Ro). Then fRa(X) E k(z)G[X] is
a defining polynomial of the G-extension k(z)/k(z)G, and Fa(t;X) E
k(t)[X] is a k-generic G-polynomial.
Let p : key) --- k(z) to be a linear k-isomorphism between rational
function fields, i.e., p is a k-isomorphism which is defined by (Yi 1-+
Ef=l O-ijXi, 1 :S j :S n) with some (aij) E GLn(k). And put
G' := {p-l 0 at 0 p I at E G} < Aut~neark(y).

This group is isomorphic to G. We fix a subfield k' of k which satisfies
that G' < Aut~peark'(y). Then the extension k'(y)/k'(y)G' is G'-Galois.
Here we suppose that
k'(y)G' = (p-l 0 G)(k'(t)). (Assumption 1)
This implies that G' := p-l OGlk'(t) is a k'-isomorphism between k'(t)

and k'(y)G' , namely, linear Noether's problem for G' /k'(y) has a positive
answer G"
To obtain a k'-generic G'-polynomial from FG(tjX), we accept the
following assumption to make use of IRG(X):
(Assumption 2)
Then RG' := p-l(RG) is G'-stable and generates k'(y) over k'(y)G'.
Hence we have
fR G , (X) = IT (X - y) = IT (X - p-l(x))
yep-l(Ra) xeRG
=LP-l(Cj)Xj E k'(y)G'[X]
j
is a defining polynomial of the G'-extension k'(y)/k'(y)G'. Moreover,

we obtain that
FGI(tjX) = "(}(fRG,(X)) = L)G~ op-l)(Cj)Xj
j
= LG1 (Cj)xj = FG(tjX).

j
From the above observation, we have:

Proposition 5. Under (Assumption 1 and 2), a k-generic G-polynomial
FG (tj X) has the coefficients in k' (t) and it is k' -generic for G.
3. Geometric descent
This section is independent of the previous section but continued from
section 1. Kemper [10] shows that a k-generic G-polynomial FG(tj X) E
k(t)[X] is also descent-generic over k, that is, for every subgroup H of
178 Y. RIKUNA
G, every H-extension M/ K with K :::> k and #K = 00 is obtained from

Fa(t; X) by a specialization of parameters t. This implies that Fa(tj X)
defined by our construction is also k-generic for every subgroup of G.
Here we study how to obtain a k-generic H-polynomial from Fa(tj X)
for a subgroup II of G. We assume that k(x)H is k-rational. Then
there exists an k-isomorphism cPH : k(u) --+ k(x)H, where k(u) :=
k(Ul, ... ,un) is the rational function field with n variables Ul, ... ,Un.
Define 7r : k(x)a --+ k(x)H to be the inclusion, and put <l> := cPi/07rocPa.
We note that the k-rational map <l> describes a relation between the
generators {cPa(tl), ... , <pa(tn)} and {<PH(Ul), ... , <PH (Un)}. There exists
a H-stable subset RH eRa such that k(x) = k(x)H(RH). Then
9RH(X):= II (X - x) E k(x)H[X]
xERH
is a defining polynomial of the H-extension k(x)/k(x)H, and
is a k-generic H-polynomial. Hence we obtain a relation between two

polynomials Fa(t;X) and FH(U;X),
Theorem 3.1. For a k-generic H-polynomial FH(UjX) E k(u)[X] ob-
tained by the above construction, we have
Fa(<l>(tl),"" <l>(tn ); X) = FH(U; X)cPi/ ( II (X - X)) ,

xERa\RH
that is, <l>(Fa(tjX)) contains FH(UjX) as a factor. Especially if RH =

Ra then we have FH(Uj X) = Fa(<l>(td, ... , <l>(tn);X).
Proof From k(x)a c k(x)H, we have 7r(Cj) = Cj for every 1 ~ j ~ n

and
~(Fa(t; X)) = ~(<Pc/URa(X))) = (<pl/ 0 7r 0 <pa)(<p"(/URa(X)))
= <Pl/(7rURa(X))) = <Pl/URa(X)),
We obtain that FH(UjX) divides cPl/URa(X)), because 9RH(X) divides

7r(fRc;(X)) = JRa(X) in k(x)H[X]. 0
Example 6. Take G = (a1, a2) < Aut~earQ(Xl' X2) as example 3.

The group G ~ C2 X C2 has three non-trivial subgroups isomorphic to
as the corresponding answers of linear Noether's problems for them.

They have the following relations:
<PG(tI) = <PH1 (uI) + 9H1 (uI)-I,
{
<PG(t2) = (<PH2(VI) + 2)<pH2(V2)2 = (<PH12(WI) - 2) <PH12 (W2)2 .
<PH2(VI) <PH12(WI)
Put Ra := OrbG(xI) = {XI, X2}, RH1 := OrbH1 (Xl) = {XI},
RH2 := OrbH2(xd = {XI,X2}, and RH12 := OrbH12(xd = {Xl, -X2},
and hence these are generating sets of the corresponding extensions
Q(XI,X2)/Q(XI,X2)G, ... , Q(XI,X2)/Q(XI,X2)H12. Then we obtain that
FG(tl, t2; X) = X4 - trt2X2 + trt~,
FH1(UI,U2iX) = X2 - (u~ + 1)u2,
FH2(VI, V2i X) = X2 - (VI + 2)V2X + (VI + 2)v~,
FH12(WI,W2iX) = X2 - (WI - 2)W2X - (WI - 2)w~,
are the corresponding Q-generic polynomials. And they have the follow-
ing relations:
FG(UI + Ui-!,U2iX) = FH1(UI,U2iX) x (X 2 - (u12 + 1)u2),
FG(VI, (VI + 2)vllv~; X) = FH2(VI, V2; X)
x (X 2 + (VI + 2)V2X + (VI + 2)v~),
Fa(w}, (WI - 2)wIIW~iX) = FH12(WI,W2iX)
x (X2 + (WI - 2)W2X - (WI - 2)w~).
4. Construction of generic even-dihedral

polynomials
In this section we apply the above construction to the even-dihedral
groups. Let N ~ 4 be an even integer, k be a field whose characteristic
180 Y. RIKUNA
does not divide N, and assume that k contains (, a primitive N-th

root of unity in a fixed algebraic closure of k. We define two linear
k-automorphisms of k(x) = k(XI' X2) by
and
They satisfy aN = (32 = 1 and (3-1a(3 = a-I, hence G := (a, (3) <
Aut~near k( Xl, X2) is isomorphic to V2N, the dihedral group of order 2N.
Our aim is to find a k' -generic V2N-polynomial, where k' is a subfield of
k contains w := ( + (-1.
We first show that linear Noether's problem for Glk(xl,x2) has an
affirmative answer. Let S <J G be the scalar subgroup, i.e.,
S:= {a E G I a(xl/x2) = xl/x2}.
Then GIS acts naturally on k(xl/x2), and k(xl/x2)lk(xdx2)G/s is a

GIS-extension. Because the transcendental degree of k(xl/x2)G/S over
k is one, there exists an GIS-invariant Bl E k(xl/x2)G/S such that
k(xl/x2)G/S = k(Bl) by Liiroth's theorem.
Lemma 7 (section 1.1 in Kemper [9]). If B2 E k(Xl' X2)G is a quo-
tient fig of homogeneous G-invariants f, 9 E k[Xl' X2]G with Ideg f -
deggl = #8, then we have k(Xl,X2)G = k(xl!x2)G/S(B2) = k(B 1 ,B2).
By using this lemma, we define a k- isomorphism <PG by
and
We give two pairs of (B I ,B2) explicitly in the following two subsections.
The former choice, which has very simple form, is not applicable for
arithmetic descent. But the latter choice matches our arithmetic descent
method for k'.
4.1 Trivial construction

From a(xdx2) = (2(xdx2) and (3(xl/x2) = x2/xl, one can easily
show that S = {I, aN/2} and that we can choose Bl and B2 by
And we choose Be:= OrbG(xd = {(i X1 ,(-ix2 11 $ j $ N}. Then we

have
N
fRo(X) = II(X - (i X1 )(X - (-i X2 )
i=l
= X2N - (x{\' + xr)X N + (X1X2)N E k(X1, X2)G[X].
By using two relations X1X2 = (h()2 and xf + xr = ()i N- 2)/2():/2, we

obtain
FG(t1,t2;X) = <Pc/(fRo(X))
= X2N - tiN-2)/2t:/2XN + (t1 t 2)N E k(tb t2)[X]
as a k-generic V2N-polynomial. But we cannot apply our arithmetic
descent method, which is stated in section 2, to this polynomial.
4.2 Non-trivial construction

For our purpose, we replace the previous choice for ()1! (h, and RG by
another.
Lemma 8. We can choose two generators fh and 82 of k(X1, X2)G over
k as follows:
and
Proof. Clearly 82 satisfies the condition in lemma 7. Therefore it is

sufficient to show that 81 generates k(xl/x2) over k(xl/x2)G/S. We
have k()l) c k(xl/x2)G/S and [k(xl/x2) : k(xl/x2)G/S] = N, hence
[k(xdx2) : k()l)] ~ N. On the other hand, XdX2 is a root of a equa-
tion XN - (_1)N/2(2lh + w2 - 4)()1 1 XN/2 + 1 = 0 of degree N, hence
[k(xl/x2) : k()l)] :5 N. This shows the assertion. 0
We choose RG := OrbG(x1 + X2) = {(iX1 + C-iX2 I 1 :5 j :5 N}.

Then we have
N
fRo(X) = II (X - (Ci Xl + C-iX 2)),
i=1
and the next lemma calculates the explicit expansion of this.
182 Y. RIKUNA
Lemma 9 (lemma 2 in Hashimoto-Rikuna [6]). We put B(N,j) :=

(N-:i-l)
)-1
+ (N-:i)
)'
then we have
N
II (X - (iXl + (-i X2 ))
i=l
(N-2)/2
= XN + L B(N,j)(-XlX2)jX N- 2j - (xf/2 - (_X2)N/2)2.
j=l
The definition (4.1) irn plies

-XlX2 = (h82 and (Xf/2 - (_X2)N/2)2 = (w 2 _ 4)8iN- 2)/28f/2.
Hence we obtain another k-generic V2N-polynornial:
Theorem 4.1. A polynomial
FG(tl,t2jX)
= c/URo(X))
(N-2)/2
= XN + L B(N,j)(tlt2)j X N- 2j - (w 2 - 4)tlN- 2)/2 t f/ 2 (4.2)
j=l
in k(tl, t2)[X] is a k-generic V2N-polynomial.
4.3 Arithmetic descent

We define a linear k-isornorphisrn p: k(Yl,Y2) ~ k(XbX2) by
p(Yd = Xl + X2 and p(Y2) = (Xl + (-lX2, (4.3)
then we have
(p-l 0 a 0 p)(Yl) = Y2, { (p-l 0 (3 0 p)(yt) = Yl,
{
(p-l 0 a 0 p)(Y2) = -Yl + WY2, and (p-l 0 (3 0 p)(Y2) = WYl - Y2.
Hence we have G' = (p-1 0 a 0 p, p-l 0 (3 0 p) < Aut~peark'(Yb Y2).
Lemma 10. The linear k-isomorphism p defined by (4.3) satisfies (As-
sumption 1), that is, p satisfies k'(Yl, Y2)G' = p-l(G(k'(tl, t2)))'
!
Proof. We have
-1 -1 (w 2 - 4)2(Yf - WYIY2 + y~)N/2
(p 0 G)(tl) = p (81 ) = Y1 ~ (Y2)N/2 _ (Yl _ (-lY2)N/2)2'
;- )N/2 ( ;--1 )N/2)2
(p-l 0 G)(t2) = -1(82) = Yl - . . Y2 - Y1 -.... Y2 ,
P (w2 - 4)2(Yf - WYIY2 + y~)(N-2)/2
and we obtain ((Yl - (Y2)N/2 + (Yl - (-lY2)N/2)2 E k'[Yl, Y2] by using

(Yl - (Y2)N + (Yl - (-lY2)N
N
= L) -1)i (~) ((i + (-i)yf-i 1A E k'[Yl, Y2].
j=O J
We put
TJnu := ((Yl - (Y2)N/2 - (Yl - (-lY2)N/2)2,
{
TJde := (w 2 - 4)2(y~ - WYIY2 + y~)(N-2)/2,
so that p-l(82) = TJnu/TJde. Then TJnu,TJde E k'(Yl,Y2)G'. And they are
coprime in k'[Yl, Y2]. Hence p-l(82) satisfies the condition in lemma 7.
We can show that k'(yl/Y2)G' = k'(p-l(fh)) by the same way as the proof
oflemma 8. Therefore we obtain that k'(yl. Y2)G' = k'(p-l(fh),p-l((h)).
o
By the definition of RG, we have p-l(Ra) = p-l(OrbG(xl + X2)) =
OrbG'(Yl) C k'(yl. Y2). Hence (Assumption 2) is also satisfied.
Therefore we obtain our main result from proposition 5.
Theorem 4.2. The k-generic V2N-polynomial FG(tl' t2; X) obtained
by (4.2) is also a k' -generic V2N -polynomial.
4.4 Geometric descent

Put H' := (p-l 0 a 0 p) < G'. Then H' is cyclic of order N. Here we
consider a relation between the k'-generic V2N-polynomial Fb(t!, t2; X)
(= FG(tb t2; X)) and the k'-generic eN-polynomial FH'(ul, U2;X) which
is obtained by Hashimoto-Rikuna [6].
From Hashimoto-Rikuna [6, section 3], we can define a k'-isomorphism
I
H' : k'(Ul, U2) - k'(Yl, Y2)H' by
(-l(Yl - (Y2)N/2 - ((Yl - (-lY2)N/2
H'(Ut) = ,
(Yl - (Y2)N/2 - (Yl - (-lY2)N/2
((Yl - (Y2)N/2 - (Yl - (-lY2)N/2)2
H'(U2) = (w2 _ 4)2(Yt _ WYIY2 + y~)(N-2)/2 '
as an answer of linear Noether's problem for H' /k'(Yl, Y2). In this case
we obtain that
and
that is, <P(tl) = Ur-WUl +1 and <P(t2) = U2. Since Orh(a) (Xl +X2) = RG,
we can choose RH' := RG', and hence fR H , (X) = fRo, (X). Therefore
theorem 3.1 implies the following result:
184 Y. RIKUNA
Proposition 11. For the k'-genenc V2N-polynomial FG,(tl,t2;X) ob-

tained above,
FGI(u~ - WUI + 1, U2; X) E k'(UI, U2)[X]
is a k' -generic eN -polynomial.
5. Generic polynomials for small groups

In this section we display generic polynomials for groups of order up
to sixteen obtained by our construction with characteristic zero. We
use computer manipulations with GAP (GAP [1]) and Mathematica
(Wolfram [22]). We denote a primitive n-th root of unity by (n and put
Wn := (n + (;;:-1.
Order 2
There is one group (isomorphism class) of order two.
2/1 C2 over Q
action { Xl I---t X2
X2 I---t Xl
generating set RG {Xl, X2}
answer of linear Noether's problem <pc/(t) = {fh, 02, ... }
_ {(XI+X2)2 ~}
- XIX2' Xl +X2
generic polynomial X~ - tlX + tl
The first row means the isomorphism class of G indexed by the ordering
in Thomas-Wood [21] and the base field k. The second and third row
express a generator of G < Aut~near k( x) and a generating set RG of
the extension k(x)/k(x)G respectively. The fourth and fifth row show
an explicit answer of linear Noether's problem for G/k(x) and the cor-
responding k-generic G-polynomial respectively.
Remark 12. Assume that the group G has trivial scalar subgroup, i.e.,
k(xI/x2)/k(xI/x2)G has the same Galois group as k(XI,X2)/k(XI,X2)G.
Then FG(tl, l;X) is also a k-generic G-polynomial if k(xI/x2)G = k(OI)
(see Kemper-Mattig [11, theorem 7]). With the above construction, we
obtain that X2_tlt2X +tlt E Q(tl, t2)[X] is a Q-generic C2-polynomial.
Hence X2 - tlX + tl = (X2 - tlt2X + tlt)lt2=1 is also Q-generic for C2
from this observation.
Order 3
There is one group of order three.
3/1 C3 over Q
act. { Xl 1---+ X2
X2 1---+ -Xl - X2
Ra {Xl, -Xl - X2, X2}
ans. {XI-3X1X2-X2 X1X2(X1 +X2& }
X1X2(X1 +X2) , X~+X1X2+X2
poly. X" - (tr + 3t! + 9)X + tr + 3t! + 9
note See remark 12.
Shanks' "the simplest cubic" (cf. Shanks [20]),
X3 - t!X2 - (tl + 3)X - 1
is well-known as a Q-generic C3-polynomial.
Order 4
There are two groups of order four.
4/1 C4 over Q
poly. X4 + 4(t~ + 1)t2X~ + 4(t~ + 1)t~
note This is obtained in section 4.
4/2 C2 x C2 over Q
act. { Xl 1---+ X2 { Xl 1---+ -X2
X2 1---+ Xl X21---+ -Xl
Ra {XI,X2}
ans. {X1+X2 (XJX2)2}
X1X2 ' X,+X 2
poly. X4 - trt2X~ + trt~
Order 5
There is one group of order five.
5/1 Cs over Q(ws)

poly. XO-5tIX4+1O(wstl-1)X;J+lOws(tl +1)X~-5(tl-WS)X-1
note This is given by Hashimoto-Miyake [5J and Rikuna [17J.
Order 6
There are two groups of order six.
6/1 Ca over Q
poly. Xu +6(t~ -tl + 1)t2X4 +9(tf -tl + l?t~X;.! + 3(tr -tl + l);.!t~
186 Y. RIKUNA
1
6/2 6 3, the symmetric group of degree 3, ~ 'D6 over Q
act. { Xl f-----t X2 Xl f-----t X2
X2 f-----t -Xl - X2 X2 f-----t XI
Rc {Xl,Xl + X2,X2}
ans. {(X~+X1X2+X~);j X1X2(Xl +X2J }
Xix~(Xl +X2)2 , X2+X1X2+X:;
poly. X--:> - tlX + tl
note This is obtained in section 4. (And see remark 12.)
By a geometric descent tl f-----t tr+3tl +9, we have a Q-generic
C3-polynomial displayed above.
Order 7
There is one group of order seven.
7/1 C7 over Q(W7)

poly. X7 - 7t l XO + 21(W7t l - 1)XO - 35((w -1)tl - W7)X4
-35(w?-I)(tl +1)X3+21(w7tl-wf+l)X2-7(tl-w7)X-l
note This is given by Hashimoto-Miyake [5] and Rikuna [17].
Order 8
There are five groups of order eight.
It is well-known that there is no Q-generic Cs-polynomial (cf. Jensen-
Ledet-Yui [7], etc.)
8/1 Cs over Q(ws) = Q( v2)

poly. XS + LJ=l {G=D r (~?)}( tr - v'2tl + 1 )t2)j XS- 2j
+2(tr - v'2tl + 1)3t~
8/1 Cs over Q( v-I)

act. { Xl f-----t X2
X2 f-----t A - Xl
Rc {Xl, \7- 1x I, X2, V-lx2}
ans. {XI-V-IX~ (xlx2)4 }
X1X2 ' x +x;;
poly. XS - (tt + 4V-l)ttt2X4 + (tt + 4V-1)ttt~
note This polynomial comes from a Q( V-1 )-generic Mod 16 -
polynomial, where Mod16 is the modular 2-group of order
16, in Rikuna [16, section 3] by a geometric descent.
8/1 C8 over Q({-2)

poly. FQV16 ((t~ + y-2tl + 1) "I + 2, t2jX),
where FQ'D16(tl,t2jX)
"_ X8 _ 24(tl-2)2t~ X 4 _ 32(tl-l)gt~ X 2 _ 8(tl-I)5t~
.- tl h t?
note We have that FQ'D16 (tb t2j X) is a Q( y -2)-generic Q'D 16 -
polynomial, where Q'D16 is the quasi-dihedral group of order
16, from Rikuna [18],
8/2 C2 x C4 over Q( y-l)

act. { Xl 1----+ -lx2V. .{ Xl 1----+ X2
X2 1----+ V-lxl Xl X2 1----+
RG {XI, y-1XI, X2, y-2X2}

ans. {Xt+X2 xi+xi}
X1 X2 'x +X'l
X8 _ (tj-2)"t2 X4 + (ti- 2)"t2
poly. t~-4t?+2 (t~-4t2+2):l
note We can also get a Q-generic C2 x C4-polynomial. See the next
remark.
Remark 13. Assume that FGl (tj X) and FG2(Uj Y) are irreducible
generic polynomials for GI and G2 over an infinite field k obtained by lin-
ear Noether's problem. And let TI and T2 he roots of there polynomials
respectively. Then there exists a constant c E k such that the minimal
polynomial of TI +cr2 over k(t, u) is a k-generic GI x G2-polynomial. But
the polynomial obtained by this construction has terribly complicated
expression in general.
8/3 C2 x C2 X C2 over Q
act. {Xl 1----+ -Xl {X2 1----+ -X2 {X3 1----+ -X3
RG {XI, X2, X3}
ans. it' ;t, x~+x~+x!
{X~ x~ x"+X!+x:t}
poly. X8 _ 4{tl +t2 +1 f 4 tg X6 + 2(h +t2+1)"(3tl'+3t2+2ht2+2tl +2t2+3)taX4

t1+t2+1 (tht 2+1)2
_ 4(h +t2+1)~(tf+t~-2tN+8t~t2+8ht~-2t~-2t1+8tlt2+1)t~ X2
(t~+t~+1)3
+ (h +t2+1)4(t~+t~-ht2-t1-2t2+1)2ta
(t2+t~+1)4
8/4 'D8 over Q

poly. X 4 + 4tlt2X:.I + 4tlt~
By a geometric descent tl 1----+ tr + 1, we have a Q-generic
C4-polynomial displayed above.
188 Y. RIKUNA
8/5 Qs, the generalized quaternion group of order eight,

over Q(A)
act. {Xl ~ V-lXI {Xl ~ -X2
X2 ~ -V-lX2 X2~XI
Rc {XI, J-IXl' X2, V-1X2}

ans. {(xt+X2)~ (xi X2 )2 }
(XIX2)2 , Xi+X2
poly. XS _ tJ t 2 X4
ti-4
+ J.!i-4
~
L2
Remark 14. This is equivalent to a Q( A)-generic Qs-polynomial

obtained by Kemper [8]. Some Q-generic Qs-polynomials are also given
by Kemper [8; 9], Grabner [2], etc.
Order 9
There are two groups of order nine.
9/1 Cg over Q(wg)
poly. X\J - 9t l XIS + 36(w9tl - I)XI - 84((w~ - l)tl - wg)Xtl
+126(wg - 1)(tl - Wg -1)X s + 126(wg + 1)(tl + I)X4
-84(wg - 1)((wg + 1)tl - 1)X3 + 36(w9tl - w + 1)X2
-9(tl - wg)X - 1
9/2 C3 x C3 over Q(3)
act. {Xl ~ (3XI {X2 ~ (3-1X2
Rc {(~XI + (1X2 11 ~ i,j ~ 3}
ans. {3~}
x~' X,+X~
poly. Xg - 3(tl+1) t2 X6 + 3(tl+1)~(tt-7h+l)t2X3 _ 3(h+W t 2

h ~ h
note We can also get a Q-generic C3 x C3-polynomial. See re-
mark 13.
Order 10
There are two groups of order ten.
10/1 ClO over Q(WlO)
poly. XlO + 2:;=1 {(~~D + CU?)
}((tr - WlOtl + 1)t2)j x 1O- 2j
-(wIo - 4)(tl - wlOtl + 1)4t~
10/2 'D1O over Q(ws)
poly. X:J - (3ws - 1)X4 -- (5ws - 3)X;5 - (3ws - 2)X~ + t1
note This polynomial is given by Hashimoto-Miyake [5].
Order 11
There is one group of order eleven.
11/1 Cl l over Q(Wll)
poly. XU -lltlXlU + 55(Wlltl -l)Xl:I -165((wtl -l)tl -Wl1)XII
+330(Wll (Wfl - 2)tl - Wfl + 1)X7
-462((wtl - 3Wfl + l)tl - Wfl + 2Wll)X6
-462(wtl - 3Wrl + l)(tl + 1)X5
+330(Wll (Wrl - 2)tl - Wtl
+3wfl - 1)X4 - 165((wfl - l)tl - Wfl
+2Wll)X3 + 55(Wlltl - Wfl + 1)X2 - !l(tl - wu)X - 1
Order 12
There are five groups of order twelve.
12/1 Cl 2 over Q(WI2) = Q( v3)
poly. X l 2 + Lj=de/~[) + e~j J)}((t~ - V3tl + 1)t2)iXI2-2i
+(tl - vatl + 1)5t~
12/2 C6 x C2 over Q((6)

act. {Xl t--+ (6Xl {X2 t--+ -X2
Rc {(Axi X2 11 ~ i < 6}
ans. ;t,;t
{X~ X~}
poly. X12 _ ruaXlO + 15~2X8 _ 2(h+lO)t2X6 _ 15(2t\-1)t2X 4

tl tl tl tl
6(5tl +1)t~ X2 + (tl-l)2tg
tS t~
note We can also get a Q-generic C6 x C2-polynomial. See re-
mark 13.
12/3 1)12 over Q

poly. XO + 6tl t2X4 + 9tft~X~ + 3tft~
12/4 2!4, the alternating group of degree four, over Q

poly. X4 _ 9f;X2 _ 8X + 9T( 12(tr- ta+ 27 )T2
2 T?'
where Tl := tf - t~ - 9t~ - 27t2 - 54
and T2 := tf - 3tlt~ + 2t~ - 9tlt2 + 9t~ - 27tl + 27t2 + 27.
note This polynomial is introduced in Jensen-Ledet-Yui [7].
190 Y. RIKUNA
12/5 Q12, the quaternion type group of order twelve, over

Q((6)
act. { Xl 1-----+ (6Xl { Xl 1-----+ -X2
X2 1-----+ ,-1
6 X2 X2 1-----+ Xl
RG {(~Xl,(~X211 < i < 6}
ans. (XIX2' x~+x~
J
{XY-X)23 xlx2(xl- x 2)
poly. +
X12 _ (tf+4)"t2 X6 (t1+W t2
t~ t~
note G. Kemper also gave a Q(w6)-generic Q12-polynomial by us-
ing dimension four version of our method in Kemper [8; 9J.
Order 13
There is one group of order thirteen.
13/1 CI3 over Q(WI3)

poly. XiS -13tlXi~ + 78(W13t l -l)X ll - 286((wI3 -l)tl -W13)X lU
+715(W13(W~3 - 2)tl - W~3 + 1)X 9
-1287((wt3 - 3Wt3 + l)tl - W~3 + 2W13)X8
+1716(w~3 - 3W13 -1)((wt3 -l)tl - W13)X 7
+1716w13(wt3 - 4wf3 + 3)(tl + 1)X6
-1287((wt3 - 3Wf3 + l)tl - Wf3 + 4W~3 - 3W13)X5
+715(W13(wt3 - 2)tl - wt3 + 3Wt3 - 1)X4
-286((Wf3 - l)tl -- W~3 + 2WI3)X 3 + 78(W13tl - wt3 + 1)X2
-13(tl - W13)X - 1
Order 14
There are two groups of order fourteen.
14/1 C14 over Q(W14)

poly. X 14 + L:j=l {(?~n + C4; J) }((ti - w14tl + 1)t2)j X14-2j
-(Wf4 - 4)(tl - w14tl + 1)6t~
14/2 1)14 over Q(W7)

poly. XI + (3w - 5W7 - 2)XO + (27w - 19w7 - 14)XO
+(80w? - 65w7 - 45)X4 + (117w? - 94w7 - 65)X3
+(65w? - 52w7 - 36)X2 + tl
note This polynomial is given by Hashimoto-Miyake [5].
Order 15
There is one group of order fifteen.
15/1 Cl 5 over Q(WI5)

poly. XH'-15tIX14+105(WI5tl _1)Xlll_455((wf5 -l)tl -WI5)XU
+1365(WI5(W~5 - 2)tl - W~5 + 1)Xll
-3003w15((W15 + W15 - 4)tI - W~5 + 2)XIO
+5005((wf5 - 2W15 -1)tl - Wf5 - W~5 + 4W15)X9
-6435(w~5 - W15 - 1)(tl - W15 - 1)X8
-6435(w~5 - W15 - 1)(tl + 1)X7
+5005(WI~ - W15 - 1)((W15 + 1)tl - 1)X6
-3003((W15 + W~5 - 4W15)tl - Wf5 + 2W15 + 1)X5
+1365w15((W~5 - 2)tl - W~5 - W15 + 4)X4
-455((W~5 -l)tl - W?5 + 2W15)X3 + 105(W15t1 - W?5 + 1)X2
-15(tl - W15)X - 1
Order 16
There are fourteen groups of order sixteen. Type 16/9 and 16/13
have no faithful linear representation of dimension two. So the author
has not obtained these generic polynomials yet.
16/1 CI6 over Q(WI6)

poly. XI6 + 2:;=1 {(~5~l) + (HS; J) }((t~ - WI6t l + 1)t2)j XI6 2j
-(W?6 - 4)(ti - WI6tl + 1)9t~
16/1 C16 over Q( (s)

poly. FMod32 ((stf - 2,t2;X)
note From Rikuna [18], we have that
FMo d32 (tl, t2i X) := X16 - (tt - 4)ttt2XS - (tt - 4)(t1t2)2
is a Q((s)-generic Mod32-polynomial.
16/1 C16 over Q(1I16 := (16 - (lS1)

poly. J
P.Q'D32 ~(t~+1I16tl-1)
IIf+4 + 2, t 2; Xl
note From Rikuna [18], we have that
FQ'D32 (tt, t2i X)
'= X16 _ 48(tl-1)2t~ X 12 + 592(tt _l)4t~ X 8 _ 128(tl-1)5t~ X6
. tl t~ t~
_140S(tl-1)6t~ X4 _ 512(tl-1)7t~ X2 _ 32(tl-1)9t~
tf tf tt
is a Q(lII6)-generic QV32-polynomial.
192 Y. RIKUNA
16/2 Cs X C2 over Q((s)

act. {Xl f-----+ (SXI {X2 f-----+ -X2
Rc {(gXI X2 11 S; i < 8}
ans. {X? X?}
it'it
Xl6 _ lliXl4 + 2S~2X12 _ 56:2XlO _ 2(tl-;S)t~xs
poly.
tl tl tl tl
_ 56(tl +l)t~ X6 _ 2S(5tl-l)tg X4 _ S(7tl +l)t~ X2 + (tl-I)2t~
t5 t5 ti t~
16/3 C4 x C4 over Q( .../--1)

act. {Xl f-----+ y'-IXI {X2 f-----+ - y' -lx2
Rc {v'=fXI + VCPX2 11 < i,j < 4}
ans. 13~}
x2 ' XI+X 2
poly. Xl6 _ 4(tl+1)2t2 X12 + 2(tl+1)~(3tl-62h+3)t~xs
tl tf
_ 4(tl +1)4(tr+30h +l)t~ X 4 + (tr-I)4t~
t~ t;{
note We can also obtain a Q-generic C4 x C4-polynomial. See re-
mark 13.
116/41 C4 x C2 X C2 over Q
note See remark 13.
116/61 'Ds x C2 over Q

116/71 Qs x C2 over Q
16/8 S7iI, the eleventh transitive subgroup of 6s, over
Q(A)
act. { Xl f-----+ -lXIv'. { Xl f-----+ X2 {X2 f-----+ -X2
X2 f-----+ V-I X 2 X 2 f-----+ X I
Rc
ans. 1(X1lX2f2 , x!+x~
{XI, y'-lXI, X2, y'- lx 2}
x~+x~ X?+x~}
poly. XS - tr
t -2
2 X4 Jl t2 2
+ (t]-2)
Q'D16 over Q( -2)
See Cs over Q( -2).
16/11 ModI6 over Q( vi-I)

poly. X~ + (tf - 4)t2X4 - (tf - 4)t~
note This polynomial is given by Rikuna [16].
16/12 1'16 over Q(ws) = Q( v'2)

poly. X~ + 8(tlt2)XO + 20(tlt2? X4 + 16(tlt2);j X~ + 2ttt~
16/14 Q16 over Q( (s)

poly. X16 _ !!i-4
W22XS+~ t _44
note By the same way as Q8 and Q16, we have this polynomial.
See also remark 14.
Acknowledgments
The author would like to thank the organizers of the two conferences
in Saga and Tokyo who gave such a good opportunity of speaking and a
discussion for him. And he also expresses gratitude to Armand Brumer,
Ki-ichiro Hashimoto, Gregor Kemper, Arne Ledet, B. Heinrich Matzat,
Katsuya Miyake, and many participants in the conferences who obtained
a lot of kind advices. The author is a Research Fellow of the Japan
Society for the Promotion of Science, and this study was supported by
Grant-in-Aid for JSPS Fellows.
References
[1] The GAP Group, GAP- Groups, Algorithms, and Progmmming, Version 4.3,
2002, http://www.gap-system.org.
[2] W. Grabner, Minimalbasis der Quaternionengruppe, Monatshefte f. Math. und
Physik 41 (1934), 78-84.
[3] K. Hashimoto, Inverse Galois problems related to dihedml groups (Japanese),
Algebraic number theory and related topics (Japanese) (Kyoto, 2000), Suri-
kaisekikenkyfisho KokYUroku No. 1154 (2000), 125-136.
[4] K. Hashimoto, On Brumer's family of RM-curves of genus two, Tohoku Math.
J. 52 (2000), no. 4, 475-488.
[5] K. Hashimoto and K. Miyake, Inverse Galois problem for dihedml groups, in
"Number theory and its applications (S. Kanemitsu and K. Gyory, ed.)," 165-
181, Developments in Mathematics, 2. Kluwer Acad. Pub!., Doredrecht, 1999.
[6J K. Hashimoto and Y. Rikuna, On generic families of cyclic polynomials with
even degree, Manuscripta Math. 101 (2002), 283-288.
[7] C. Jensen, A. Ledet, and N. Yui, "Generic polynomials, constructive aspects of
the inverse Galois problem," Mathematical Sciences Research Institute Publi-
cations, Cambridge, 2003.
194 Y. RIKUNA
[8] G. Kemper, Das Noethersche Problem und generische Polynome, dissertation,

Universitat Heidelberg, 1994.
[9] G. Kemper, A consructive approach to Noether's problem, Manuscripta Math.
90 (1996), 343-363.
[10] G. Kemper, Generic polynomzals are descent-generic, Manuscripta Math. 105
(200.l), 139--141.
[11] G. Kemper and E. Mattig, Generic polynomials with few parameters, J. Symbolic
Compo 30 (2000), 843-857.
[12] A. Ledet, Generic and explicit realisation of small p-groups, J. Symbolic Compo
30 (2000),859--865.
[13] A. Ledet, Generic extensions and generic polynomials, J. Symbolic Compo 30
(2000), 867-872.
[14] G. Malle and B. H. Matzat, "Inverse Galois theory," Springer Monographs in
Mathematics, Springer-Verlag, Berlin, 1999.
[15] K. Miyake, Linear Fractional Transformations and Cyclic Polynomials, Alge-
braic number theory (Hapcheon/Saga, 1996). Adv. Stud. Contemp. Math. (Pu-
san) 1, (1999), 137-142.
[16] Y. Rikuna, On generic polynomials for the modular 2-groups, Proc. Japan Acad.
Ser. A Mathe. Sci. 78 (2002;, no. 3, 33-35.
[17] Y. Rikuna, On simple familips of cyclic polynomials, Proc. Amer. Math. Soc.
130 (2002), no 8, 2215-2218.
[18] Y. Rikuna, On generic quasi-dihedral polynomials, preprint.
[19] J.-P. Serre, "Topics in Galois Theory," Research Notes in Mathematics, 1. Jones
and Bartlett Publ., 1992.
[20] D. Shanks, The simplest cubic fields, Math. Compo 28 (1974), 1137-1152.
[21] A. D. Thomas and G. V. Wood, "Group tables," Shiva Publishing Limited,
1980,
[22] S. Wolfram, "The Mathematica Book, 4th ed.," Wolfram Media/Cambridge
University Press, 1999.
ON DIHEDRAL EXTENSIONS AND

FROBENIUS EXTENSIONS
Masafumi IMAOKA
and Yasuhiro KISHI*
Tokyo Metropolitan University,
Minami-ohsawa 1-1 Hachioji-shi, Tokyo,
192-0997 Japan
imaokamClcomp.metro-u.ac.jp
ykishiClcomp.metro-u.ac.jp
Introduction
The two quadratic fields Q( va) and Q( J -3d) are known to be closely
related. A. Scholz [Scl gave the "Spiegelung" relation between the 3-
ranks of their ideal class groups. In Y. Kishi [Kill, we gave a precise
Spiegelung relation by a constructive approach: every cubic polynomial
which generates a cyclic cubic extension E of Q( va) unramified outside
3 over Q(va) with Gal(E/Q) ~ Sa is constructed by making use of
an element of the associated field Q( J -3d); here S3 is the symmetric
group on three symbols. In 1972, on the other hand, G. Gras [Gd]
showed that all S3-extensions of Q containing a fixed quadratic field
Q( va) correspond to some elements of Q( J -3d) (see Example 3.4 (1)).
In this paper, we extend it for an arbitrary odd prime p over an algebraic
number field k with k n Q( () = Q, where ( is a primitive p-th root of
unity (see Section 2).
For an odd prime p and for a quadratic extension Ld k of algebraic
number fields with Ll n Q(() = Q, we define an extension L2 as follows:
L2 is the proper subextension of Ll()/k(w) other than k() and Ll(W),
w = (+(-1. Let us call L2 the associated field with Ll. The main purpose
of this paper is to show the following relation between Ll and L2: all
Dp-extensions (resp. certain types of Fp-extensions) of k containing Ll
(resp. L2) correspond to some elements of L2 (resp. L 1 ), where Dp and
*This author was supported by JSPS Res6.:::ch Fellowships for Young Scientists.
(received: Jun. 25, 2002; accepted in revised form: Oct. 15, 2002)
196 M. IMAOKA AND Y. KISHI
Fp denote the dihedral group of order 2p and the Frobenius group of

order p(p - 1), respectively.
Before we prove our main theorem, we need a classical result which
may be regarded as the cyclic version of the main theorem.
Theorem (Theorem 1.2). Let K be a field with char(K) i- p. Let M(K)
be a set of elements of K (() with certain properties (to be specified later).
(1) For every "( E M(K), the field K(()( {/"(t(K) is a cyclic extension
of K of degree pm, where t(K) is a certain element of the group ring
Z[Gal(K(()IK)]. (Hence K(()( {/"(t(K) contains only one cyclic exten-
sion of K of degree p. We denote it by F(K,,).)
(2) Let E be a cyclic extension of K of degree p. Then there exists an
element"( E M(K) such that E = F(K,,).
Main Theorem (Theorem 2.1). Let 7r be a permutation of the set
{1,2}. Let M(L 7r (1) be a set of elements of L7r(2) with certain properties
(to be specified later), and let F(L 7r (1), 5) with 5 E M(L 7r (1) be the field
defined as in the above theorem. Moreover, let G1 = Dp and G2 = Fp.
(1) For every J E M(L7r (l), F(L 7r (1) , 5) is a G7r (1)-extension of k con-
taining L 7r (1)'
(2) Let E be a Galois extension of k and contains L7r(l)' Suppose that
the Galois group of the extension Elk is isomorphic to G7r(1) whose
generators have some special relations if 7r is not trivial. Then there
exists an element 8 E M(L 7r (1) such that E = F(L7r (1),8).
The case where 7r is trivial and k = Ql in this theorem was shown by
Gras [Gr1, Section 3].
Moreover, we give two results (Theorem 1.5 and Proposition 2.5) con-
cerning the generator of E. They are very effective for constructing
polynomials of degree p which generate Cp, Dp or Fp-extensions.
We treat the cyclic case in Section 1 and the dihedral and the F'robe-
nius cases in Section 2. In Section 3, we give some examples for p = 3
and 5. Finally, we give in an appendix, using a cohomological approach,
a relation between the p-rank of the congruent ideal class group modulo
a power of p of a quadratic field L1 and the p-rank of the absolute ideal
class group of the associated cyclic field L2 with L1 of degree p - 1.
Its relation is given by Gras' result [Gr2, Theorem 7.11] also (see Re-
mark 4.2).
We list here those symbols which will be used throughout this article.
Let Q denote the field of rational numbers and Z denote the ring of
rational integers.
For an extension L I K, denote the norm map and the trace map of
LIK by NL/K and by TrL/K' respectively. For a Galois extension LIK,
Dihedral extensions and Probenius extensions 197
we denote the Galois group of Lj K by Gal(Lj K). For u, r E Gal(Lj K),
the action of the product ur on L is as usual, namely,
(a E L).
For an integer n, let en denote the cyclic group of order n. For a
prime p, let Dp and Fp denote the dihedral group of order 2p and the
Frobenius group of order p(p - 1), respectively:
Dp = (u, I, I uP = 1,2 = 1, 1,-1U I, = u- 1),
Fp = (u, I, I uP = I,p-l = 1, 1,-1U I, = u a ),
where a is a primitive root modulo p.
1. Preliminaries (Cyclic case)

The aim of this section is to construct the polynomials with cyclic
groups of prime degree as Galois groups using classical method. This is
a review of [Co, Chapter 5].
Let p be a fixed prime and ( be a primitive p-th root of unity. Let K
be a field with char(K) :/= p. Fix a generator r of Gal(K()j K), and take
an element l(r) E Z so that we have C = (l(T). Put m := [K() : K]
and define an element t(K) of a group ring Z[Gal(K()jK)] by
L ril(r)m-l-i.
m-l
t(K) :=
i=O
We define a natural surjection : Z - ZjpZ. We use the same nota-

tion for the induced map Z[Gal(K()jK)] - (ZjpZ)[Gal(K()jK)].
Moreover, let T(K) denote the following subset of Z[Gal(K)/ K)]:
T(K) := {t'CK) E Z[Gal(K()jK)] I

(t'(K)) = n(t(K)) for some n E (ZjpZ)X}.
The element t(K) depends on r and l(r), but the set T(K) does not.
Moreover we easily see
K()( {j,t(K)) = K()( {j,tl(K))

for every, E K) and for every t'CK) E T(K).
We first introduce a result of H. Miki.
Proposition 1.1 ([Mik, Corollary of Proposition 2]). Let the no-
tation be as above. Put F := K () ({iY) with y E K () and y K ()P .
Then the following three statements are equivalent:
(a) F j K is cyclicj
(b) yT-I(T) E K(()Pj
(c) F = K(()( {/yt(K)).

We define subsets M(K) and JJ(K) of K(()X as follows:
M(K) := b E K(()X l,t(K) K(()P},

JJ(K) := K(()X \ M(K).
By this proposition, we obtain the following:
Theorem 1.2. Let the notation be as above. Then the following state-
ments hold.
(1) For every, E M(K), the field K(()( {/,t(K)) is a cyclic extension
of K of degree pm. (Hence K(()( {i,t(K)) contains only one cyclic ex-
tension of K of degree p. We denote it by F(K, ,).)
(2) If E is a cyclic extension of K of degree p, then there exists an
element, E M(K) such that E = F(K,,).
Proof. (1) Note that the order of r is equal to m and l(r)m-1 is divisible
by p. Then in lFp[Gal(K(()j K)], we see that
>(t(K)(r -l(r))) = >(rm -l(r)m) = O.
Hence we have ,t(K)(T-I(T)) E K(Op. I f , E M(K), therefore,
K(()( {i,t(K))jK is a cyclic extension by Proposition 1.1. It is clear
that K(()( {i,t(K))jK is of degree pm.
(2) Since E(O is a Kummer extension of K(O, there exists an element
, E K(() \ K(()P such that E(() = K(()( {i,t(K)). Then we see that
, E M(K) and E = F(K, ,). 0
We now state a criterion for two fields F(K"t} and F(K, ,2) with
,1, ,2 E M(K) to coincide with each other.
Proposition 1.3. For elements ,1, ,2 E M(K), the following state-
ments are equivalent:
(a) F(K"t} = F(K, ,2)j
(b) ,f112 E JJ( K) for some n E Z \ pZ.
Proof. By the definition of F(K, 'i) (i = 1,2), we have F(K, ,1) -
F(K, ,2) if and only if K(()(at} = K(O(a2) with af = ,;(K) (i = 1,2).
By Kummer theory, it is equivalent to ("'(f(K))n h~(K) = ('Yr/'Y2)t(K) E

K(()P for some n E Z \ pZ. Therefore the statement follows from the
definition of N(K). 0
Corollary 1.4. For every p E Gal(K(()/ K) and'Y E M(K), we have
F(K,'Y) = F(K,'YP).
Proof. For 'Y E M(K), we have 'Yt(K)(T-l(T E K(()P as we have seen in
the proof of Theorem 1.2 (1). Then
(",(l(T)i /'YTt)t(K) = 'Yt(K)(l(T)CTi)
= 'Y-t(K)(T-I(T(T i - 1 +T i - 2l(T)+"+TI(T)i-2+I(T)i-l)
E K(()P
for every i, 0 ::; i ::; m - 1. Therefore we have 'Yl(T)i /'YTi E N(K).

By Proposition 1.3, we see that F(K, 'Y) = F(K, 'Y P) for every p E
Gal(K(()/ K). 0
Next, we give a generator of a cyclic extension of K of degree p by
making use of elements of K(().
Theorem 1.5. Suppose'Y E M(K) and put E = F(K,'Y). Then we
have
E = K ( TrE()/E( {/'Ytl(K))
for every t'CK) E T(K).

Proof. It follows from Theorem 1.2 that E( () / K is cyclic of degree pm.
We extend T to an automorphism of E(()/ E. We put 0 = f/'Ytl(K) with
t'CK) E T(K). Then, as we saw in the proof of Corollary 1.4, we have
OTi_l(T)i E K(()X for every i, 0::; i ::; m - 1. On the other hand, 1, 0,
Ol(T) , Ol(T)2, "', ol(T)m-l are linearly independent over K((). Therefore
so are I ,0, 0 T, 0T2 , " ' , 0 T m - 1 . Hence t h e trace 0 f 0,
TrE(,)/E (0 ) = 0 + 0 + 0 + ... + 0 T m - 1 ,
T T2
does not belong to K ((). Since E / K is a cyclic extension of prime degree

p, we see E = K(TrE()/E(O)). The proof is completed. 0
By this theorem, we can construct a polynomial of which E is the
minimal splitting field over K. (See Pr()position 1.10 and Examples 3.1,
3.2 and 3.3.) Indeed, we give the polynomial in the case [K(() : K] = 2
as follows.
Definition 1.6. For an odd integer s, we define polynomials gs(X; a)

inductively:
gl(Xja) := X,
(5-1)/2
g5(X; a) := Xs - ~ aj G)9S-2 (X; a), j (s ~ 3),
J=l
where (;) is the binomial coefficient:
( ~) = s (s - 1) .... ~ 8 - j + 1) .
J J.
Remark 1.7. By the definition of gs(Xj a), the coefficient of xn is equal
to 0 when n is even.
Lemma 1.8. Let s be an odd integer and let c(s, n) denote the coefficient
of xn of 9s(Xj a). Then
c(s, s - 2g) = (-a)9_-

8
s - 2g
(s -9-1) 9
(05 9 5 -2-).
8-1
(1.1)
Proof. It is clear that
c(s, n) = {1'0, if n = s,
if n> s.
By the definition of gs(X; a), we immediately have
(s-1)/2
C(8, n) =- L aj (~)C(8 - 2j, n)
j=l J
(s-n)/2
=- L aj (~)C(8 - 2j, n) (1.2)
j=l J
for n < s.
Both sides of (1.1) are equal to 1 for 8 = 1. Assume that the relation
(1.1) holds for all s, s:S 1 - 2, and for all g, O:S 9 :S (s - 1)/2. We will
show that the relation (1.1) holds for 8 = I and for aUg, 05 9 5 (l-1)/2,
by mathematical induction on g.
Dihedral extensions and Frobenius extensions 201
It is clear that (1.1) holds in case 9 = O. Assume that (1.1) holds for
all g, 0 ~ 9 ~ k. Then by (1.2) and the assumptions,
c(l, 1 - 2(k + 1)) = c(l, 1 - 2k - 2)
(1-(1-2k-2j2
=- ~ aj (l.)C(l-2j , l-2k-2)
3=1 \.j
k+1 . (l)
=- 2:a3 . c(l-2j, l-2j-2(k+1-j))
j=l J
k+1 .fl) .
= _ ~a3 \j (_a)k+1-3
3=1
x 1-
2j (l - 2j - (k + 1 - j) - 1)
l- 2j - 2(k + 1 - j) k + 1- j
= (_1)kak+l ~(-1)i(l- 2j) (~) (l - k - 2 -. j).

1 - 2k - 2 ~ 3=1
J k +1- J
Now we need the following equation.
Lemma 1.9. For all t, 0:5 t :5 k, we have
I:
i=k-t+1
(-1)i (l _2j) (l.) (l ~ k ~ ~ -. j)
J + J
= (_1)k-t+l (t + 1)(k - t + 1) (1 ) (l - 2k - 2 + t).
k+1 k-t+1 t+1
Proof. A straightforward calculation, by mathematical induction on t,
gives this equation. 0
Take t = k in Lemma 1.9. Then we have
~(-1)i(l- 2')
~ J
(1) (1-k+1-j
j
k - 2- j) = _ (k (1) (1-k+1
+ 1) . 1
k+1
k - 2)
1
=_1(1-k-2)
3=1
k +1 .
(1- + 2)
Therefore
(_1)k+1 ak+l k-
c(l, 1 - 2(k + 1)) = 1 _ 2k _ 2 1 k 1
= (_a)k+1 I (1- (k+ 1) -1).

l-2(k+1) k+1
The proof of Lemma 1.8 is completed. 0

Proposition 1.10. (cf. [Co, Chapter 5, Exercise 6].) Let the notation
be as in Theorem 1.5. Suppose that K contains w = (+(-1 and does not
contain (. Let a E E(() be a root of XP - ,(p+1)/2(r7')(p-l)/2 = 0, , E
M(K). Then'IrE()/E(a) generates E over K: E = K('IrE()/E(a)).
Moreover, the polynomial
!p(X;,)
(p - kk-1)XP- 2k _ N(,)(p-l)/2'Ir(,)
(p-1)/2
= L (-N(r))k_P-
k::::O p - 2k
is the minimal polynomial of'IrE()/E(a) over K, where N = NK()/K

and 'Ir = 'IrK()/K'
Proof. By the assumptions, we have m = [K() : K] = 2 and l(r) = -1;
and hence t(K) = r - 1. Since
c/>( p ; 1 + P ; 1r) = p ; 1 c/>( r - 1),
we have, from Theorem 1.5,

E == K('IrE()/E(a)). (1.3)
We extend r to an automorphism of E(()/E. Let us show that
a a7' = N(r). (1.4)
Since
(a. a7')P = ,(p+l)/2(f7')(p-1)/2 . (fT)(p+1)/2,(p-1)/2 = (f. ,T)P = N(,)P,
we have aaT = (iN(,) for some i, 0 ~ i ~ p-1. Note that aa7' E E.
Since neither E nor K contains (, it must be i = 0, that is, aaT = N(f).
By (1.4), therefore, we have a relation
(8-1)/2
(a + a7')S = as + (aT)S + L N(f)i (~) (a s- 2i + (a7')s-2i) (1.5)
i=1 J
for an odd integer s. On the other hand, we have
a + aT = 'IrE)/E(a),
aP + (a7')P = ,(p+l)/2(r7')(p-l)/2 + ,(p-l)/2(r7')(p+1)/2
= ,(p-l)/2 (,7') (p-1)/2(r + ,7')
= N(f)(p-1)/2Tr(,).
Put 8 = P in the relation (1.5)j then we see that
(p-l)/2
TrE)/E(a)P - ~ N(r)j (~) (aP- 2j + (a T )p-2j)
- N(r)(p-l)/2Tr(,) = o. (1.6)
Next we will show that
(1.7)
for all odd integers 8. It is clear that the equation (1.7) holds in case
8 = 1. By the definition of 9s(Xj a) and the relation (1.5), we see that
9s(TrE()/E(a)j N(r))
(s-I)/2
= TrE()/E(a)S - ~
3=1
N(r)j G.) 9s-2j (TrE()/E(a)j N(,))
= as + (aT)S
(s-I)/2
+ ~ N(,)j G.) (a s- 2j + (aT)s-2j - 9s-2j(TrE()/E(a)j N(,)).
3=1
Hence if the equation (1.7) holds in all cases of odd integers less than 8,
then we have 9s(Tr E)/E(a)j N(r)) = as + (aT)s.
We define a polynomial !p(Xj,) as follows:
!p(Xj,) := 9p(X; N(,)) - N(r)(p-l)/2Tr(r).
Then !p(Xj,) has a root TrE()/E(a). Indeed, by (1.6) and (1.7), we

have
!p(TrE()/E(a)j ,)
= TrE()/E(a)P
(p-1)/2
- ~ N(,)j (~)9P-2j(TrE()/E(a)j N(r)) - N(r)(p-l)/2Tr(,)
= TrE()/E(a)P
(p-l)/2
- L N(r)j (~) (aP- 2j + (a T)p-2j) - N(r)(p-1)/2Tr(r)
j=1 J
= o.
Hence by the condition (1.3), fp( Xi "() is the minimal polynomial of

'frE)/E(a) over K. Therefore the statement of the proposition follows
from Lemma 1.8. 0
2. The cases Dp and Fp

In this section, we construct Dp-polynomials and Fp-polynomials by
utilizing the results in Section 1.
Let p be an odd prime. Let ( be a primitive p-th root of unity and put
w := (+ (-1. Let k be an algebraic number field and Ll be a quadratic
extensions of k which are linearly disjoint with Q(() over Q. Then
there exists a unique proper subextension of the bicyclic biquadratic
extension Ll(()/k(w) other than k(() and Ll(W). We denote it by L2
(see Figure 1). L2 is called the associated field with L 1 Moreover, we
define m1 and m2 by
m1 := 2 (= [L1 : kJ = [L2(() : L2]),

m2 := P -1 (= [L2 : kJ = [Ll(() : L 1 ]),
respectively.
Figure 1.
For an algebraic number field L, let teL) and M(L) be the same
notation as in Section 1. For an element "( E M(L), let the definition of
the field F(L, "() be as in Theorem 1.2 (1). Our main theorem states
Theorem 2.1. Let 7r be either one of permutations of two elements

{1,2}. Fix a generator p of Gal(L71"(l) (()/ L71"(2)) , and take an element
l(p) E Z so that we have (P = (l(p).
(1) For every 8 E M(L7I"(1)) nL7I"(2) , F(L7I"(1),8) is a Galois extension ofk

of degree pm7l"(1) containing L7I"(1)' Furthermore, let u and" be generators
of Gal(F(L7I"(1), 8)/k) which satisfy the following two relations:
(i) "IL"'(l) = pIL"'(1);

(ii) uP = "m,..(l) = 1.
Then we have
(iii) t- 1 Ut = ul(p).
(2) Let E be a Galois extension of k of degree pm7l"(1) and contains L7I"(1)'

Let u be a k-automorphism of E of order p. Suppose that the Galois
group of the extension Elk is
Gal(E/k) = (0', I uP =
l, l,m,..(l) = 1, ,,-lUl, = u1(p)}
with "IL"'(l} = pIL,..(l)' Then there exists an element 8 E M(L7I"(1)nL7I"(2)
such that E = F(L7I"(1),8).
First we prove a general proposition.
Proposition 2.2. Let p be a prime and K be a field with char(K) =/: p.
Put m := [K(() : K]. Let L be the minimal splitting field of XP -0 (0 E
K\KP) over K. And let (p} and (u) denoie the subgroups ofGal(L/K)
corresponding to K( if<i) and K((), respectively. Then we have
Gal(L/ K) = (u, p I uP = pm = 1, p- 1up = u b),
where b is an element of (Z/pZ) x with (P = (b.
Proof. Let us write G = Gal(L/ K) for simplicity. Since
(I(u}l, I(p}/) = (p, m) = 1
and since (u} is normal in G, we have
G = (u, p).
Since (u} is a normal subgroup of G, we have p- 1up = u b for some

b, 1 ~ b ~ p-1.
Now take c and d be integers satisfying
( {Yli)U = (C {Yli (1 ~ c ~ p -1),
(P = (d (1 ~ d ~ P -1).
Since we have (U = ( and ( VIa)P = VIa, we see that

(~)p-lUp = (~)UP = ((CVIa)P = (cd{Ya.
On the other hand, we have
( {Ya)P-1UP = ({Ya)U b = (cb {Ya.

Hence we must have (cd = (cb. Since (c, p) = 1, we have b == d (mod p),
and hence (P = (b. The proof is completed. D
By this proposition, we immediately have the following corollary.
Corollary 2.3. In the situation of Proposition 2.2, we have
if m = 1,
if m = 2,
if m = p-1.
Proof of Theorem 2.1. (1) Suppose 8 E M(L7l'(1) n L7l'(2)' Let 7 is the

generator of Gal(L7l'(1) ()/ L7l'(1)' Since L7l'(1) n L7l'(2) = k, the restriction
map 7I L7I"(2) is an automorphism of L7l'(2)/k. Then we have 8t(L7I"(1) E
L7l'(2)' Put
E' := L7l'(1)(()( {!8 t (L7I"(1)) = L7l'(2)(()( {!8 t (L7I"(1));
then E' is normal over L7l'(2) of degree pm7l'(1)' Let a' and ~' be generators
of Gal(E'IL7l'(1)(()) and Gal(E'IL7l'(2)({!8t(L7I"(1))), respectively. Then
by Proposition 2.2, the Galois group Gal(E' I L7l'(2) is
Gal(E'/L7l'(2) = (a',~' I a'P = t' m 7l"(1) = 1, ",-la',,' = a'b) ,

where b is an element of (Z/pZ)X with (~I = (b. From the natural
restriction map, we see that Gal(E'IL7l'(2)({!8t(L7I"(1)))) is isomorphic to
Gal(L7l'(1) (()I L7l'(2)), and ~'IL1f(l)) is a generator of Gal(L7l'(1) (()I L7l'(2)'
Then we have p = (~'IL1f(l)())C for some integer c with (c, m7l'(1)) =
1. Therefore we have (P = (~/C = (b C , and hence bC == l(p) (mod p).
Moreover, a' and ,,'c generate Gal(E' / L7l'(2) and satisfy
(,,'C)-la' ~'C = a,bc = a,l(p).
Now let us show that E' is normal over k. Let v be a k-isomorphism

from E' to the algebraic closure of E'. Then (E')II is the minimal split-
ting field of X P _(8t(L7I"(1)))1I over (L7l'(1) (())II. Since L7l'(1) (()Ik is abelian,
Dihedml extensions and Probenius extensions 207
we have (L7r(l)(())11 = L7r(l)(() and (8t (Lor(l)1I = (811 )t(Lor(l. Therefore

we have
(E')II = L 7f (1)()(V8 I1t (L or(l) = F(L7f(1),811 )().

Then by Corollary 1.4, we have (E')II = E'. Hence E' is normal over
k. Since (L7r(1)11 = L7r(1) and F(L 7r (1), 8) is the unique cyclic subexten-
sion of order p of E'/L7r (1), we see (F(L7r (1),8))1I = F(L7r (1),c5). There-
fore F(L7f(1),8) is normal over k. From the natural restriction map,
we see that Gal(E'/L7r (2) is isomorphic to Gal(F(L7f(1),8)/k). Then
F(L 7r (1),8) is normal over k of degree pm7r(1). Put 0' := O"IF(Lor (1),6) and
/, := (/"IF(L or (1),6)Y; then we see that 0' and t generate Gal(F(L7r (1), 8)/k)
and satisfy the relations
['IL"(1) = (t'IL".(1)C = pIL".(1)'

uP = [,m"'(1) = 1,
/,-10'/, = O'l(p).
The last relation does not depend on the choice of u and p. Indeed, if
we change 0' to u i satisfying the condition (ii), then we have
/,-10'il, = (1,- 1UI,)i = (O'l(p)i = (ui)l(p),
and if we change p to pi, then we have
[,i ILor (1) = pi IL".(1)

and
I, -i O'/,i = ul(p)j = O'I(pJ).
It is also clear that I, is determined uniquely by the choice of p.
(2) Let E be a Galois extension of k of degree pm7r(1) which contains
L 7r (1). Since E/L7r (1) is cyclic of order p, there exists 'Y E M(L7r (1)
such that E = F(L7f (1) , 'Y) and E() = L 7f (1)()(a) with aP = 'Yt(Lor(1 ,
by Theorem 1.2. From the natural restriction map, we see that
Gal(E()/ L 7f (2) is isomorphic to Gal(E/k); namely, the elements 0"
and /,' which satisfy O"IE = 0' and /,'IE = /, generate Gal(E()/ L 7r (2).
Then we may assume that aul = (cr. Denote the fixed field of (1,') by
E 1 Put f3 := N E (c,)/E1 (a) EEl and 8 := N L"'(1) (C,)/L".(2) ('Y) E L7f (2); then
we have
f3P = N E (c,)/E 1 (a P ) = N E (c,)/E1 C'Y t (L"'(1)
= N E (c,)/E1('Y)t(L"'(1 = N L "'(1) (C,)/L".(2) ('Y)t(L"'(1 = 8t (L"'(1.
Since
IT IT
m 7r (l)-l m 7r (l)-l
= ((l(p)-S nylS = (3 (l(p)8 1(p)-S = (3(m 7r (l)
8=0 8=0
and m 7r (l) '= 0 (mod p), we have (3(11 i= (3. Then we get (3 L7r(l)(()
and E1 = L7r(2) ((3). Therefore we have E(() = L 7r (l)(()((3), and hence
E = F(L7r(l) , 8). This completes the proof of Theorem 2.1. 0
Remarks 2.4. (1) When 7r is the identity in Theorem 2.1, we see that
every Dp-extension of k containing L1 is given as F(L1' 8) for some
8 E M(L1) n L2. This was shown in the case k = Q by G. Gras [Gr1,
Section 3].
(2) When 7r is not trivial in Theorem 2.1, we see that a certain Fp-
extension of k containing L2 is given as F(L2' 8) for some 8 E M(L2)nL1;
in this case, gep.8rators (J' and t of the Galois group Gal(F(L2,8))/k)
satisfying the conditions (i) and (ii) of Theorem 2.1 must also satisfy the
condition (iii) of Theorem 2.1. In general, however, there are <pep - 1)
primitive roots modulo p, where <p is the Euler function. Hence a natural
question arises: Are there Fp-extensions E of k containing L2 whose
Galois group have generators (J' and t satisfying the conditions (i), (ii)
of Theorem 2.1, and
(iii') t -1(J't = (J'b,
where b is anyone of primitive roots modulo p except l(p)?
Assume that p satisfies p == 1 (mod 4). Then there exists a quadratic
extension of k other than L1 which is contained in L1 (() but is not
contained in k( O. We denote it by L~. Let T be a generator of the
Galois group Gal(L2(O/ L2)' Then we have (T = (-1 and L~ is the fixed
field of (T p). For 8' E M (L 2) n L 1, therefore, E = F (L2' 8') contains
L2 and the Galois group Gal(E/k) has generators (J' and t satisfying the
conditions
(i) tlL2 = (Tp) IL2 = p1L2'
(ii) (J'P = tp- 1 = 1,
and
(iii") t- 1 (J't = (J'-l(p)
because
We note that -l(p) is also a primitive root modulo p.

The authors have not yet verified whether or not there exists an Fp-
extension other than these. However, it seems that there exist such
Fp-extensions because the p-rank of the ideal class group of Ll relates
only to the p-rank of the x* -component (x* is a suitable character of
Gal(LI ()/k)) of the ideal class group of L2, as we will study in Section 4.
Using Theorem 1.5 we have, for a E M(L7I"(1) n L7I"(2) ,
E:= F(L7I"(1) , a) = L7I"(1)(lJI),
where
w = TrE(t;;)/E( \Iat'(L7I"(1).
One of our aims is to construct a polynomial of degree p with coefficients
in k which generates E. As we see the following proposition, it can be
constructed by making use of w.
Proposition 2.5. Let the notation be as in Theorem 2.1. For E a
M(L7I"(1) n L7I"(2) , we put E = F(L7I"(I) , a). For t'(L7I"(I) E T(L7I"(I), let
-a
(3 E E() be a root of X P t '(L7I"(1 = 0 and put lJI:= Tr E(t;;)/EC(3). Then
we have
(2.1)
Moreover, the minimal polynomial of W over k is irreducible over L1I'(1)
and its minimal splitting field over k coincides with E.
Proof. For the first assertion it is sufficient to show [k(W) : k] =
[L1I'(1)(W) : L1I'(1)1. Let T be a generator of Gal(E()/ E) of degree m1l'(2).
Then we have W = E:~2)-1 (3Ti. Since at'(L7I"(1 E L1I'(2), L1I'(2) ((3)/ L1I'(2)
is of degree p and E( () / L 7r (2) ((3) is cyclic of degree m1l'(I). We extend
p to an automorphism of E(()/L1I'(2)((3). Since E(() = E L1I'(2) and
En L1I'(2) = k, the map
# : Gal(E()/k) --+ Gal(E/k) x Gal(L7I"(2)jk)
defined by #(0') := (alE, 0'I L7I"(2) is an isomorphism. Then #(r) -
(1, rIL!1 2 ) and #(p) = (piE, 1). Therefore rand p commute with each
other. Then we get
L L L L
m7l"(2)-1 ') P m7l"(2)-1 , m7l"(2)-1 , m7l"(2)-1
wP = ( (3T' = (3T'P = (3PT' = (3Ti = W,
i=O i=O i=O i=O
and hence W is contained in the fixed field E(p) of (p) of E. Since

k = L~)1)' we have
(L7I"(1) (w) : L7I"(1)] $ (k(lJI) : k] $ (E(p) : k] $ (L1I'(1) (lJI) : L1I'(I)]

Therefore we have [k(\lI) : k] = [L7r(1)(\lI) : L7r(1)).

The second assertion is obvious from (2.1) together with the fact that
the order of every proper normal subgroup of Dp or Fp is divisible by
p. 0
Note that we have the following diagram (Figure 2) by this proposi-
tion. When 7r is not trivial in Proposition 2.5, by using Proposition 1.10,
we have the following corollary as a construction of Fp-polynomials.
Figure 2.
L7r(1) (\lI)
m1r~
k(\lI) p, normal
p, not normal L7r(1)
k~l)
Corollary 2.6. Let the notation be as in Theorem 2.1. Let fp(X; 6) be
a polynomial defined in Proposition 1.10, where N = NLl/k and Tr =
TrLl/ k. Then for 8 E M(L 2) nL 1, the minimal splitting field of fp(X; 8)
over k is an Fp-extension of k containing L 2. Conversely, every Galois
extension E of k containing L2 with the Galois group
Gal(Ejk) = (a, ~ IaP = i P- 1 = 1, i- 1ai = al(p))
with ilL2 = PIL2 is given as the minimal splitting field of fp(X; 8) over k
for some 8 E M(L2) n L 1.
In [Ki2) and [Ki3), by utilizing this corollary, Y. Kishi gave infinitely
many cyclic fields of degree p - 1 containing Q(w) whose class numbers
are divisible by p for each odd prime p.
Theorem 2.7. Let k be the rational number field Q and L1 be a real
quadratic field. Then the associated field L2 with L1 is an imaginary
cyclic field of degree p - 1. Assume that there exist a unit IE of L1 with
IE ~ Ll which satisfies the conditions either
TrL1/1Q>(1E) == 0 (mod p2)

or
Then the minimal splitting field of fp(X; 0) over Q is an unmmified

cyclic extension of L2 of degree p. Hence the class number of L2 is
divisible by p.
3. Examples
Let the notation be as in Section 1 and suppose that ( . K. First
we construct Cp-polynomials over K for p = 3 or 5 explicitly. For 1 E
M(K), we put E := F(K, I)' For t'CK) E T(K), let hi(X; t'CK), I) (1 ::;
i ::; 3) in the following examples denote the minimal polynomial of W :=
TrE((,)/E( {//t'(K) over K.
Example 3.1. (cf. [Gr1, p.182], [Ch, Proposition 2], [Co, Chapter 5,
Examples (1), p.253].) The case p = 3. We have [K(() : K] = 2.
Since 2 is a primitive root modulo 3, we may take the generator r of
Gal(K(()/K) with (T = (2. Then we have t(K) = 2 + r, and hence
h 1 (X;t(K)'/) = X 3 - 3NCr)X - N(/)Tr(/)

for, E M(K) where N = NK(,)/K and Tr = TrK((,)/K' This polynomial
is immediately obtained from Proposition 1.10.
Example 3.2. (cf. [Co, Chapter 5, Examples (3), p.254].) The case
p = 5 and [K(() : K] = 2. Since 4 is of order 2 modulo 5, we may take
the generator r of Gal(K(()/K) with (T = (4. Therefore t(K) = 4 + r.
Put t'CK) = 3 + 2r; then we have t'CK) E T(K) because t'CK) _
2t(K) (mod 5). Hence by Proposition 1.10, we have
h2(Xjt'(K),/) = X 5 - 5NCr)X3 + 5NCr)2X - N(,)2Tr(/)
for, E M(K) where N = NK()/K and Tr = TrK()/K.
Example 3.3. (cf. [Ha, p.433], [Gr1, p.182], [Co, Chapter 5, Examples
(2), p.253].) The case p = 5 and [K(() : K] = 4. Since 2 is a primitive
root modulo 5, we may take the generator r of Gal(K(()j K) with (T =
(2. Therefore t(K) = 8 + 4r + 2r2 + r3. Put t'CK) = 3 + 4r + 2r2 + r3;
then we have t'(K) E T(K) because t(K) == t'CK) (mod 5). Hence we
have
h3(X; t ' (K), I)
= X5 - lONK(()/KCr)X 3 - 5NK(()/K (/)NK(w)/K (TrK(()/K(w) Cr)X 2
+ 5NK()/KCr){NK(()/K(/) - NK(w)/K(TrK()/K(w) (/1+T)}X
- N K((,)/K( I)NK(w)/KCfrK()/ K(w) (/ 2+T
for'Y E M(K).
Next we give examples for the dihedral and Frobenius cases. From
now on, let the notation be as in Section 2.
Example 3.4. Let k be an algebraic number field. Put L1 := k( Vd),
dE k. Then we have
k(J-3d), if p = 3,
{
L2 = k ( y'r---1O-d-+-2-d-J5=5') , if P = 5.
(1) Take K = L 1r (l) in Example 3.1. Then by Proposition 2.5, we see
that every S3( = D3 = F 3 )-extension of k containing L1r (I) is given by the
roots of the polynomial hI (X; t(L1r (I), 6) for some 6 E M(L 1r (1) n L 1r (2) ,
where N = N L7r (2)/k and Tr = Tr L7r (2)/k'
(2) Take K = L2 in Example 3.2. Since p = 5 == 1 (mod 4), as we
have seen in Remark 2.4, there exist two quadratic extensions L1 and
L~ = k( V5d) of k associated with L2. On the other hand, there exist
two primitive roots modulo 5. By Proposition 2.5, therefore, we see
that every F 5-extension of k containing L2 is given by the roots of the
polynomial h2(X; t'(L2), 6) for some 6 E M(L2) nL1 or 6 E M(L2) nL~,
where N = N L1 / k , Tr = TrLl/k or N = NL~/k' Tr = TrL~/k'
(3) Take K = Ll in Example 3.3. Then by Proposition 2.5, we see
that every D 5-extension of k containing Ll is given by the roots of the
polynomial h3(X; t'(Ll), 6) for some 6 E M(L 1 ) n L2, where NK((,)/K =
N L2 / k , TrK((,)/K = TrL2/k and NK(w)/KTrK((,)/K(w) = N k(w)/k Tr L2/k(w)'
Example 3.5. H. Hasse gave a polynomial which generates an unram-
ified cyclic quintic extension E of L1 = Q( J-47) and a Ds-extension of
Q (see [Ha, p. 434]). This polynomial is given as our h3 (- X; t' (L 1 ), 61 )
with ( = e21rA/ s and
.,= ~ (2_VS)v'_47+ 25 -;lVS(_C'))'

It has integral coefficients, but 81 is not an element of L2 = Q( J -47 ( -
(-1 )). Take
82 := N L1 ((,)/L2(81)
= ~ (424 - 189VS - 105 -247 VS v'-47{ ( - C')) E 1.".
Then we have 62 = -(61)2, and hence

h3(X;t'(L 1),62 ) = X5 -lOX3 - 5855X 2 -1368890X - 88365871
generates the same E over Q by Proposition 1.3.

On the basis of Example 3.4 (1), one of the authors, M. Imaoka [1m],
will give a family of S3-polynomials with negative discriminants whose
roots are fundamental units of the cubic fields. We will state his results.
We note the fact that for every & E Q("';-3d), there exists a rational
number m E Q such that the minimal splitting field of hI (X; Q( Vd), &)
coincides with the minimal splitting field of hI (X; Q( Vd), 8') with
8' = ~ +(~(m - 5) +~) V-3d.

Put X = (Y -1)/3 in hl(X; Q(Vd) , &') and multiply it by 33. Then we
have
fd,m(Y) := y3 - 3y2 - (A(m) - 3)Y - 1,
where
A(m) := d(2m - 1)2 + 27.
4
Then the discriminant of fd,m(Y) is d(2m -1)2A(m)2.
Assume that d is a square free negative integer with d == 1 (mod 4)
and m is a positive integer. Then A(m) is an rational integer. Moreover,
we assume that A(m) is cube-free. Then fd,m(Y) is irreducible over Q
and has only one real root. We denote it by 8 and put Kd,m := Q(8).
Theorem 3.6. Let the notation and the assumptions be as above. Then
8 is the fundamental unit of Kd,m except for (d, m) = (-3,3), (-11,1),
(-23,1), (-31,1). For exceptional cases, we have
8 = {e~,m' if (d, m) = (-3,3), (-11,1),

ed,m, if (d, m) = (-23,1), (-31,1),
and
X3 +3X2 +3X -1, if (d,m) = (-3,3),
( X3_X2 -X-1 if (d, m) = (-11,1),
9d,m(X) = X 3 -X -1, , if (d, m) = (-23,1),
X3+X -1, if (d, m) = (-31,1),
where ed,m is the fundamental unit of Kd,m and 9d,m(X) is the minimal
polynomial of ed,m over Q.
By using of this theorem, we have
Theorem 3.7. Under the same situation as that in Theorem 3.6, we
assume in addition that A(m) is square-free. Then there are exactly
nine pairs (d, m) such that the clas.:; number of Kd,m is one: their pairs
are (d, m) = (-3,1), (-7,1), (-15,1), (-19,1), (-23,1), (-31,1),
(-35,1), (-39,1), (-47,1).
4. Appendix
Let p be a fixed odd prime and ( be a primitive p-th root of unity.
For an algebraic number field K and an integral divisor f of K, we
denote the congruent ideal class group modulo f of K by ClK(f). For
simplicity, we denote ClK if f = 1; that is, ClK is the absolute ideal class
group of K. Moreover SK denote the set of prime divisors lying over p
in K, and GK,SK (resp. Gw'SK) denote the Galois group of the maximal
extension (resp. maximal abelian extension) of K unramified outside SK
over K.
For a character X of a finite abelian group G with (#G,p) = 1 and
an lFp[G]-module M, we denote the x-component of M by M&; that is,
Mfj = exM where
1 '~x(a)a
ex = #G " -1 .
(TEG
Let L1 be a quadratic field with L1 n Q( () = Q and L2 be the associ-

ated field with L 1 (For the meaning of "associated field," see Section 2.)
Let Li be the Galois group of L 1 (()/Q, and X be the character of Li cor-
responding to L1. Let A (C Li) be the Galois group of L1 (() / L 1. Let
w be the Teichmliller character of A and put X* := wX- l . Then X*
corresponds to L 2
We define positive integers p and pi by
p:= dimlFp(GL~,SL)pGL~,SL)'
,. x'
p := dlmlFp ( ClL2/pCIL2) ~ .
We note that by Class Field Theory we have
for sufficiently large n.

The aim of this appendix is to prove the following theorem:
Theorem 4.1 (M. Kurihara). Let the notation and the assumptions
be as above. Then we have
pi + 8 + 1, if P ~ 5,
p- {
- pi + 8 + c + 1] + 1, if P = 3,
where
8:= {a,1, if L1 is real,

if L1 is imaginary;
e:= {a,
-1,
if
if
['+ll E pCIL2'
['+ll ~ pCIL2;
TJ:= {a,1, if #8L2 = 1,

if #8L2 = 2,
and '+l E 8L2'

Remark 4.2. A result of G. Gras [Gr2, Theorem 7.11] immediately
leads to this theorem. However, we will give an another proof using
cohomological approach. The first version of this paper had an erroneous
choice of characters. The authors wish to express their gratitude to
Professor G. Gras for pointing out the error which led to improvement
of this appendix.
Proof of Theorem 4.1. First we introduce the following J. Tate's theo-
rem on which Theorem 4.1 is based.
Theorem 4.3 ([Tal], [Ta2, Theorem 2.2]). Let K be an algebraic
number field and M be a finite GK,sK-module. Assume that the order of
M is a power of p. Then the following equation holds:
#HO(GK,SK,.M). #H 2 (GK,SK,M) = (-tl=MGa1(C/R))Tl . (_1_)T2,

#H1(GK,SK,M) #M #M
(4.1)
where r1 and r2 denote the number of real and imaginary embeddings of
K into C, respectively.
Let us apply Theorem 4.3 to the case K = L1 and M = ZlpZ.
Then the Galois group GLl,SLl acts trivially on ZlpZ, and the following
relations hold:
HO(GLl,SLl ,ZlpZ) =ZlpZ,

H 1(GLl,SLl' ZlpZ) =Homconti(GLl,sLl' ZlpZ)
(cf. [Se, Chapter I, 2.3]). Then we have
#HO(GLl,SLl ,ZlpZ) = p, (4.2)

#H 1 (GLl,SLl ,ZlpZ) = #(ZlpZ)e p = pp. (4.3)
From now on, we calculate #H'2(GLl,SLl,ZlpZ). Since the order of

d is relatively prime to p and ZlpZ is isomorphic to the group J.lP of the
p-th roots of unity as a GLl (,;),SLI (~) -module, we have
H 2(GL l ,SLl' ZjpZ) = H 2(GLI(),SLl,)' ZjpZ) A

= (H2(GLl(),SLl(~)'f-tP) f-t~(-l))A, (4.4)
where f-t~-1 satisfies f-tp f-t~-1 = ZjpZ.

Denote the ring of integers of L 1 (() by Ch l ( ) . Then we have
1
Hq (GLl (),SLI )' f-tp) = H:t (Spec OLI () [:pJ, f-tp) (4.5)
for each q ~ 0, where HZt is an etale cohomology. For simplicity, we

denote
1
HZt(Spec Ch l ()[-],*) =: Hq(*).
p
For an etale cohomology, the following lemma is well-known.
Lemma 4.4. Let Gm be the multiplicative group. Then we have an exact
sequence:
o ----? Hl(G m )jpHl(Gm ) ----? H2(f-tp) ----? H2(G m )[PJ ----? O. (4.6)
Proof. We have an exact sequence:
[P]
o--+ J-Lp --+ Gm --+ Gm --+ 0,
where [P] denotes the endomorphism x ~ x p . From this, we have the
long exact sequence
Hl(G m ) Ji Hl(G m ) --+ H2(f-tp) ----? H2(G m ) Ji H2(G m ),

and hence we get the exact sequence (4.6). o
We go back to the proof of Theorem 4.1. We have
Hl(G m ) ~ PiC(Chl()[~]) ~ ClLl()j([~]1 ~ E SLl()) =: ClLl(),SLl,)

(cf. [Mi12, Chapter II, Propcsition 2.1]). Moreover we have
H2(G m ) ~ Br(Ch l ()[!])

p
(cf. [Mill, Chapter IV]). By Class Field Theory, we have
H2(G m )[p] ~ ker(Br(Ll(()) --+ EEl Br(Ll(()v))[P]
VESLlW
= ker( EB Z/pZ ~ Z/pZ),
VESLl,)
where ~ is the map (a v) 1-+ ~VESLIW av. Then by Lemma 4.4, we have
an exact sequence
0---+ CIL1((),SLIW/pCIL1((),SLIW ---+ H2(J..tp )

E
---+ ker( EB Z/pZ ---+ Z/pZ) ---+ O.
VESL1(~)
Then the sequence
o ---+ ((CIL1((),SL1(~/pCILl((),SLlW) 0 J.t:-1)il ---+ (H 2(J.tp) 0 J.t:-1)il

---+ (ker( EB Z/pZ....s Z/pZ) 0 J.t:-1)A ---+ 0 (4.7)
VESLIW
is also exact.
Now in general for an IFp[~]-module M, we have
p-2
(MJ.t:-1)A = $(MtJ.t:-1)A = M':%.J.t:- 1 = (MXEfJMf)J.t:- 1.
i=O
Therefore we have
CIL1((),SLl({/pCILl((),SLl(~) 0 J.t:-1)A = T 0 J.t:- 1,

(ker( EfJ Z/pZ....s Z/pZ) 0 J.t:-1)A = U 0 J.t:-l,
VESLl(~)
where
From this, together with (4.4), (4.5), and the exact sequence (4.7), we
have
#H 2(GLl,SLl ,Z/pZ) = #(H 2(J.tp) 0 J.t:-1)A

=#(T 0 t..t:- 1 ). #(U 0 t..t:- 1)
= #T #U. (4.8)
.
Let us calculate #T. Smce x*
(C4J!((),sQw/pC~(),sQ(~).& = 0, we have
Let S,J,l be a prime divisor lying over pin L2: S,J,l E 8L2'
The case [qJ] E pCIL2' It follows from CIL2, 8L 2 = CIL2/([qJ]) that
CIL2,8L2/pCIL2,sL2 = CIL2/pCIL2' Hence we have
x* ,
#T = #(ClL2/pClL2) b. = pP .
The case [S,J,l] pClL2 . Since
is exact and
if p ~ 5,
if p = 3,
we have
if p ~ 5,
if p = 3.
Next, we calculate #U. Since p is totally ramified in Q()/Q, we have

#8L2 = 1 or 2 and
E' *
U ~ (ker( EB Z/pZ ---+ Z/pZ))1 '
vE L 2
where E' is the map (a v ) ~ :EvESL2 avo

E'
The case #8L2 = 1. Since ker(Z/pZ ---+ Z/pZ) = 0, we have
#U = #(O)x.* = 1.
The caset::.#8L2 = 2. Let r be the generator of~. In this case,
we have r(S,J,l) = S,J,l' with S,J,l, S,J,l' E 8L2' S,J,l ::I S,J,l'. Therefore r acts on
(ker(Z/pZ$Z/pZ ~ Z/pZ))f by multiplication by -1. Then we have
(ker( EB Z/pZ -E' Z/pZ))X* = (ZjpZ)X* =

~~
{o ';j
~~,
'7J
if P ~ 5,
if p = 3,
and hence
#U = {1,
p,
if p ~ 5,
if P = 3.
For the right side hand of (4.1), we can get

# (Z/pZ) Gal(C/IR) # (Z/pZ)
# (Z/pZ) = #(Z/pZ) = 1,
1 1
#(Z/pZ) = p'
From this, together with (4.1), (4.2), (4.3), and (4.8), we have
p. #T #U =
pP
{I,
p-l,
if Ll is real,
if Ll is imaginary.
Hence by
PPI, if P ~ 5 or [~] E pCIL2'
{
#T = pp'-l, if p = 3 and [~] f/. pCIL2 j
#U = {I, p,
if p ~ 5 or #8L2 = 1,
if p = 3 and #8L2 = 2,
as we saw above, we obtain the desired conclusion. o
Acknowledgments
The authors would like to thank Professor Ryotaro Okazaki whose
suggestions led to the improvement of this paper. They would also like
to thank Professor Masato Kurihara for his suggestions and comments
on Section 4.
References
[Ch] R. J. Chapman, Automorphism polynomials in cyclic cubic extensions, J.
Number Theory 61 (1996), 283-291.
[Co] H. Cohen, "Advanced topics in computational number theory," Graduate
Texts in Mathematics, 193. Springer-Verlag, New York, 2000.
[Grl] G. Gras, Extensions abeliennes non ramifiees de degre premier d'un corps
quadratique, Bull. Soc. Math. France 100 (1972), 177-193.
[Gr2] G. Gras, TMoremes de refiexion, J. TMor. Nombres Bordeaux 10 (1998),
399-499.
[Ha] H. Hasse, Uber den Klassenkorper %'Um quadratischen Zahlkorper mit der
Diskriminante -47, Acta Arith. 9 (1964), 419-434.
[1m] M. Imaoka, Fundamental units in a family of cubic fields with negative dis-
criminants, preprint.
[Kil] Y. Kishi, A constructive approuch to Spiegelung relations between 3-ranks of

absolute ideal class groups and congruent ones modulo (3)2 in quadratic fields,
J. Number Theory 83 (2000), 1-49.
[Ki2] Y. Kishi, Imaginary cyclic fields of degree p - 1 whose relative class numbers
are divisible by p, Proc. Japan Acad. 77 A (2001), 55-58.
[Ki3] Y. Kishi, On imaginary cyclic fields of degree p - 1 which have p-ranks of the
ideal class groups greater than one, preprint.
[Mik] H. Miki, On Zp-extensions of complete p-adic power series fields and function
fields, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 21 (1974),377-393.
[Mill] J. S. Milne, "Etale cohomology," Princeton Mathematical Series, 33. Prince-
ton University Press, Princeton, N.J., 1980.
[Mil2] J. S. Milne, "Arithmetic duality theorems," Perspectives in Mathematics, 1.
Academic Press, Inc., Boston, MA, 1986.
[Sc] A. Scholz, Uber die Beziehung der Klassenzahl quadratischer Korper zueinan-
der, J. Reine Angew. Math. 166 (1932), 201-203.
[Se] J.-P. Serre, "Galois cohomology," Translated from the French by Patrick Ion
and revised by the author. Corrected reprint of the 1997 English edition.
Springer Monographs in Mathematics, Springer-Verlag, Berlin, 2002.
[Tal] J. Tate, Duality theorems in Galois cohomology over number fields, in "Proc.
Internat. Congr. Mathematicians (Stockholm, 1962)," 288-295, Inst. Mittag-
Leffler, Djursholm, 1963.
[Ta2] J. Tate, On the conjectures of Birch and Swinnerton-Dyer and a geometric
analog, in "Seminaire Bourbaki, Vol. 9," Exp. No. 306, 415-440, Soc. Math.
France, Paris, 1995.
ON THE NON-EXISTENCE
OF CERTAIN GALOIS EXTENSIONS
Hyunsuk MOON*
Graduate School of Mathematics
Kyushu University 33
F'ukuoka 812-8581, Japan
moontOmath. kyushu-u .ac.jp
Introduction
In this article, we give a survey of my results on the non-existence
and finiteness of certain Galois extensions of the rational number field
Q with prescribed ramification. The detail has been (will be) published
in [8], [9], [10], [11], [12].
First, we begin with Serre's conjecture, which is our motivation, and
explain how this conjecture implies the non-existence and finiteness of
certain Galois extensions of Q. Second, we recall what is known on
the non-existence together with the main ingredients in the proofs, and
sketch the proof of some of my results presented at the Conference.
Third, we state a generalized finiteness problem and summarize my re-
sults on this problem.
1. Serre's conjecture
Let GQ be the absolute Galois group Gal(Q/Q) and iFp an algebraic
closure of the finite field lFp of p elements. Serre ([14]) conjectured that
every odd and irreducible continuous Galois representation p : GQ --+
GL2(iFp) would be isomorphic to the representaion PI associated to a
mod p cuspidal eigenform f. More strongly, Serre also conjectured that
such an eigenform exists of a certain "minimal" type (N (p), k(p), c(p)).
Now we recall the definition of N(p) and k(p). Consider a representa-
tion p : GQ --+ GL(V) ~ GL2(iFp), where V is a two-dimensional vector
space over iFp. Let K/Q be a finite Galois extension such that p factors
The author was supported by the JSPS Postdoctoral Fellowship for Foreign Researchers.
[received: May 26, 2002; accepted: Oct. 14, 2002)
222 H. MOON
through Gal(K/Q). Then the Artin conductor N(p) outside p is defined

as follows:
N(p) := II In(l,p) ,
l=h
where I runs through the prime numbers different from p, and for each
l,
n(l,p) := t, ~
(Go G,) dim.,(VjV G,),
where VGi is the fixed part of V by the ith ramification subgroup G i

of the decomposition group of a prime of K lying above l. Hence the
level N(p) reflects the ramification away from p. For example, n(l, p) = 0
means that p is unramified at land n(l, p) = 1 means p is tamely ramified
at l.
The weight k(p) depends only on the restriction of p to some inertia
subgroup Ip C GIQ at p. Hence the weight reflects the ramification at p.
Let X be the mod p cyclotomic character. If p is wildly ramified at p,
then its restriction to Ip has the following form:
pilp ~ (~ ;a) with * i- 0, (1.1)
where ~ denotes the equivalence relation of representations of Ipo Take

the integers 0: and {3 (uniquely) so that 0 ~ 0: ~ P - 2 and 1 ~ {3 ~ p - 1.
We set a = min(o:,{3), b = max(o:,{3), and define
k(P):={1+ p a+b+ p -1 if {3-~=1 and px-aisnotfinite,

1 + pa + b otherwIse.
For other cases, see [4], [14] (if P = 2, we adopt the definition in [4]
rather than the original one of Serre [14]). According to the definition,
we have k(p) ::; p2 - 1. After twisting p by a power of the cyclotomic
char~cter, we can reduce the weight so that k(p) ::; p 1:.1; indeed if we
put k(p) := k(p X-a) with 0: as above, then we have k(p) ~ p + 1.
This is a very strong conjecture, and there are many consequences of
it. In particular, we are interested in the following ones:
For p = 2,3,5,7, there are no Galois representations p : GIQ ~ GL2(fp)
which are odd, irreducible and unramified outside p, since cusp forms
for SL 2(Z) do not exist for k < 12 .
For a given integer N, there exist only finitely many Galois represen-
tations p : GIQ ~ GL2(fp) up to isomorphism which are odd, irreducible
and N(p)IN, since normalized eigenforms for rl(N) are finite in number
if their weights are bounded.
In the following sections 2 and 3, we will treat the relating subjects

to the above assertions respectively.
2. The non-existence
We recall what is known about the above-mentioned non-existence
implied by Serre's conjecture. Tate ([16]) showed the non-existence in
the case of p = 2, and Serre also remarked that this can be extended
to the case of p = 3 by the same method. Recently Brueggeman ([2])
showed the non-existence in the case of p = 5 assuming the Generalized
Riemann Hypothesis (GRH). All these results are proved basically by
the same principle. That is by comparing two esimates for discriminants
of opposite directions. We let KjQ be the corresponding field to Ker(p).
Using class field theory, we estimate from above the discriminant of the
field K. On the other hand, we have a lower bound of the discriminant
by Odlyzko. Then the non-existence follows from the contradiction of
these two inequalities. This method is vital. For example, using this
method, Fontaine ([5]) showed the non-existence of abelian schemes over
Z. Along this line, we show the non-existence of some representations
having non-trivial Artin conductor N(p) outside p ([10)):
Theorem 1. (1) There exist no non-trivial Galois representations p :
GQ -+ GL2(iF2) which are semisimple and have N(p) dividing 3.
(2) Assume the GRH. Let p : GQ -+ GL2(iF3) be a non-trivial semisimple
representation with N(p) dividing 2. Then p ~ 16:) X3 or X3 6:) X3, where
X3 is the mod 3 cyclotomic chamcter.
By using a discriminant estimate, it is impossible (even under the

GRH) to prove the conjecture of Serre for p = 7 with N(p) = 1 (cf. [2]).
However, for even mod 7 representations, we can show:
Theorem 2. Assume the GRH. Let p: GQ -+ GL2(iF7) be a semisimple

even representation which is unmmified outside 7. Then p ~ X~ 6:) X~
with a + b = even, where X7 is the mod 7 cyclotomic chamcter.
Here, we give a sketch of the proof of Theorem 1(2). Our proof is
divided into two parts. If Im(p) is solvable, we use a structure theorem
on solvable subgroups of GL2(iFp) and class field theory. Then we find
that our representation is abelian. If Im(p) is non-solvable, we compare
two different estimations of the discriminant of the kernel field of p to
deduce contradiction.
Let p : GQ -+ GL2(iF3) with N(p) I 2. We first note that N(p) = 2
means that we allow tame ramification at 2. Let KjQ, be the fixed field
by Ker(p) and G = Im(p) its Galois group.
224 H. MOON
Solvable case:
Suppose G is solvable. When G is a reducible subgroup of GL2(1F3),
simisimplicity implies that the non-trivial representations are precisely
those listed in Theorem 1(2).
When G is an irreducible subgroup, we use the follwing Lemma ([15]):
Lemma 3. Let G be an irreducible solvable subgroup ofGL2(iFp ) and H
its maximal abelian normal subgroup. Let A be a normal subgroup of G
such that AIH is a maximal abelian normal subgroup of GIH. Then G
is either of the following types:
1 ~ H ~ G ~ Z/2 ~ 1, He (iF;)2 (G1)
or
1~A ~ G ~ G ~ 1, G C 53 (:::: SL2(1F2)),
c if; ,
(G2)
1 ~ H ~ A ~ A ~ 1, H A c (Z/2)2.
Furthermore, if G is maximal, then all the inclusions are the equalities.
If p = 2, only type (Gt) is possible. Also, G acts on A by conjugation,
and G injects into Aut(A), since A is its own centmlizer in GIH.
Apply this Lemma to our G. If G is of type (G 1), then K is an abelian
extension of degree prime to 3 over the quadratic field F = Q(;=3).
This field F has class number 1 and 0F,3 = 1F3 x (pro-3 group), kF,2 =
1F~ , OF :::: Z/6. Here OF is the ring of integers of F, OF,p := OF z Zp
and kF,p is the maximal prime-to-p quotient of 0F,p. Hence, by class
field theory, there is no abelian extension K/ F of degree prime to 3.
Thus K = F = Q(A).
Next, we suppose that G is of type (G2). The quotient group G of
G is contained in 53. We consider each of the cases G = 53, 03, 02
and trivial. For example, if G = 0 3 , then we have just one C3-extension
of Q, that is Q((g + (i 1 ). This field does not have a (Z/2)2-extension
unramified outside 3. Indeed, such an extension would produce a 2-
extension of Q((g) unramified outside 3, which does not exist. Since
G ~ Aut(A), it is impossible that G = 03. Hence G is abelian .
Non-solvable case:
First, we introduce two estimates for discriminants of opposite direc-
tions:
Lemma 4 (Tate, [16]). Let K/Q be a finite Galois extension of de-
gree n with Galois group G. Suppose that there is an embedding p :
G ~ GL2(iFp). Then the p-part d~~Q of the discriminant dKjQ of K/Q
satisfies the following estimates:
(i) If KjQ is at most tamely ramified, then
\d~~Q\ < pn.

(ii) If KjQ has wild ramification of degree pm, m ~ 1, then
/d~~Q\ ~ pn(2+1/p-I/(p-l)pm-l). (2.1)
On the other hand, we have

Lemma 5 (Odlyzko bound, [13]). Let K be an algebraic number
field of degree n, with rl real conjugate fields and 2r2 complex conjugate
fields. Then
! log \dK\ > {' + log 411' + 1 - 8.31730;n-~ if rl = n, (2.2)

n - , + log 411' - 6.860404n-'3 if rl = O.
Assume the GRH. Then

11 d . 1 8 r17r 211"2(,x + ~13) 1611'2(1 + ~)
11: og\ KI ~ 1+ og 11"+ 712 - (logn)2 - (logn)3(1 + (lo~:)2)2'
(2.3)
where A = 1.0517997903, f3 = 0.9689461463. Note that these bounds are
monotone increasing in n if either rl = n or O.
We assume G = Gal(K/Q) is non-solvable. Then IGI ~ 60. From
(2.2) for n = 60, we know that IdKI1/n > 14.3050. If K is tamely
ramified at 3, by Lemma 4 we have /dK\I/n < 32 = 6. Contradiction.
If K has wild ramification, we consider the composite map p : GQ -+
GL2(iFp) -+ PGL2(iFp)' We let Gbe the image of G and L/Q be the field
corresponding to Ker(p). Then Gis also non-solvable. Suppose LjQ has
wild ramification of index pm. Then by [3], we have
Lemma 6. Let G be a subgroup ofPGL2(iFp)' If its p-Sylow subgroups

have order pm, m ~ 1, and are not normal in G, then G is isomorphic
to either PGL2(lFpm) or PSL2(lFpm).
Apply this Lemma to our G c PGL2(iF3). Note that we have m ~ 2

because G is solvable if m == 1.
From (2.1), we have
\d \l/n {21.6169 if m = 2,
(2.4)
L < 25.9605 if m ~ 3.
226 H. MOON
We assume the GRH. Then from (2.3), we have
19.5441 if n ~ 360 = IPSL2(lFg)l,

IdLl l / n {
> 22.5793 if n ~ 1440 = 4 x IPSL2(lFg)l,
29.7470 if n ~ 9828 = IPSL2(lF27 )1
We have contradiction if m ~ 3. For the case m = 2, if p is odd, we
can show easily that the order of G is greater than or equal to 4 x
IPSL 2 (lFg)1 = 1440 (cf. [2J, 3). Thus we have contradiction for the case
m = 2. If p is even, since p =f:. 2, K is a totally real or eM field, and L
is totally real. From (2.2) we have
IdLl l / n > 51.6187 if n ~ 360.

This contradicts (2.4).
3. The Finiteness Problem

3.1 Problem
It will be natural to generalize the second assertion of Section 1:
(FP). Let K be an algebraic number field of finite degree over Q and
GK be its absolute Galois group Gal( K / K). Fix an integer d ~ 1 and a
nonzero integral ideal N of K. Then do there exist only finitely many iso-
morphism classes of continuous semisimple representations p : GK ~
GLiiFp ) whose Artin conductor N(p) outside p divides N?
(Here, N (p) is a suitable generalization of that of 1 ([8]), but we use the
same notation.) This problem, together with the proof of some special
cases (Th. 7 below), was presented in [8]. Khare ([6]) also considers the
same problem independently (He states this as a conjecture).
The case of d = 1 of (FP) follows from global class field theory, i.e.,
given an ideal N of K, there exist only finitely many abelian extensions
of K with conductor dividing N. It also follows from this that there
are only finitely many isomorphism classes of continuous representations
p : GK - t GLd(iFp) with N(p)IN which surject onto GLd(lFq) for some
finite field lFq with q elements. The case of d = 2, K = Q, odd, as we
said, follows from Serre's conjecture. The case of d ~ 3, K = Q, "odd" in
some sense is related to a conjecture of Ash and Sinnott ([1]), which is an
analogue of Serre's conjecture formulated by using cohomology classes
instead of modular forms. This (FP) is reduced to the case K = Q by
the induction of representations ([8], 4).
Now we give some remarks to explain why the conditions in the Prob-
lem are necessary:
Remarks. (1) Without any restriction on ramification outside p, we can-

not expect the finiteness. For example, the set of isomorphism classes
of p : GQ --+ GL1(iFp) unramified outside 1(=1= p) is infinite, since we
have, for each n ~ 1, the representation Pn : GQ --+ Z/lnz ~ iF;
corresponding to the n-th layer of the cyclotomic Zl-extension of Q.
(2) If we replace iFp by a finite field lFpm, the finiteness follows from
the Hermite-Minkowski Theorem saying that there exist only finitely
many algebraic number fields which are of a given degree and unramified
outside a given set of primes.
(3) The assumption of semisimplicity is necessary. In fact, there may be
infinitely many (mutually unisomorphic) non-semisimple representations
of a finite group G.
We can consider such non-existence and finiteness also in the context
of inverse Galois problem. Then our Finiteness Problem is equivalent to
the following:
(FP'). Given K, d, p, N, do there exist only finitely many Galois exten-
sions L/ K such that its Galois group Gal(L/ K) embeds semisimply in
GLd(iFp) and the prime-to-p part of the different 1)~/K of L/K divides
N?
For comparison, we mention the following two things:
Theorem (Hermite-Minkowski). Given any constant N > 0, there
exist only finitely many algebraic number fields L/Q such that the dis-
criminant IdLI :5 N.
Class field tower. For "large" N, there exist infinitely many finite
Galois extensions L/Q such that IdLI1/[L:Q] :5 N.
Note that in such an example of infinite family of L/Q coming from
an infinite class field tower, their Galois groups cannot all be embedded
into GLd(iFp) with fixed p and d.
3.2 Results
In this section, we survey my results on (FP) with a brief sketch of
the proofs. First, we discuss the Finiteness Problem in the case where
K = Q, N = 1 and, p and d are small. Let Xp,d be the set of isomorphism
classes of continuous semisimple representations P : GQ - GLd(iFp)
unramified outside p, and X:'d the subset of Xp,d consisting of those p's
of which the field corresponding to the kernel is totally real (in other
words, pis unramified also at 00). Then we have ([8]):
Theorem 7. (1) The set Xr,4 is finite. Under GRH, the sets X2,4 and
Xr,s are finite.
228 H. MOON
(2) Under GRH, the set Xf,4 is finite.

(3) The sets X2,2, X3,2, Xr,2 are finite. Under GRH, the sets X 5,2, Xf,2'
Xfl ,2' Xf3 ,2 are finite.
The proof of this Theorem is based on the same principle as in 2
(except that we use an asymptotic lower bound of discriminants (the
limit of Lemma 5 as n -- (0)). The key point of the proof is an extension
of a part of Tate's estimation (2, Lemma 4) in terms of the invariant
"p-length" of its Galois group:
Lemma 8. Let K/Q be a finite Galois extension of degree n with Galois
group G. Suppose that then is an embedding p : G '---+ GLd(iFp). Then
the p-part of the discriminant of K /Q divides pen where
rlog d"l
c < rlog d"l + 1 + 2
2 p-l
Here, r a'" denotes the integer satisfying a ::; r a'" < a + 1.

Further, if we compare this bound with Lemma 5, we can bound
the order of image of representations; two estimates contradicts if n ~
100. Hence there exists no totally real Galois extension L/Q unramified
outside 2 such that Gal(L/Q) ::::: GL4(lF2m), SL4(lF2m) or PSL4(lF2m) for
any m, because these groups have order > 100.
Next result is on the Finiteness Problem for general K, d, p, N under
the condition that their images are solvable ([11]):
Theorem 9. Given an integer d 1 and a nonzero integral ideal N
~
of K, there exist only finitely many isomorphism classes of continuous
semisimple representations p : GK --+ GLdOFp) with solvable image and
with N(p) dividing N.
The finiteness statement holds true also for classical Artin repre-
sentations, i.e., if we replace iFp by the complex number field C and
N(p) by the usual Artin conductor. Theorem 9 is proved by combining
the Hermite-Minkowski theorem on discriminants and class field theory
(finiteness of abelian extensions of bounded conductor) with a theorem
of Mal'cev and Kolchin ([15]) on the structure of finite solvable sub-
groups of GLd(iFp). Also, in [11], the Finiteness Problem is reduced to a
special case in which the image of p is a finite simple group of Lie type in
characteristic p. This is based on a theorem of Larsen and Pink ([7]) on
the structure of finite subgroups of GLd(iFp). Furthermore, these results
hold also for function fields K in one variable over a finite field, under
the reasonable condition that there are no constant field extensions.
Now we consider quantitative and effective versions of (FP). For mod

p representations, to give a quantitative result is to give an explicite
upper bound for the numb~r of representations and to give an effective
result is to give an upper bound of the image of representations. This
problem for the general case seems very difficult. So, we consider the set
of n-dimensional monomial mod p representations of GQ. with bounded
conductor. We say that a representation p: GQ -+ GLn{lFp) is monomial
if it is of the form p = Ind~x, i.e. if it is induced from a character
X : GK -+ iF; of the absolute Galois group G K of an algebraic number
field K of degree n over Q. From the construction together with the
Hermite-Minkowski theorem and the finiteness of ray class groups, it
follows easily that this set is finite. In the following theorem, (i) is a
quantitative result and (ii) is an effective result:
Theorem 10. Fix positive integers nand N. Consider n-dimensional

monomial mod p Galois representations p : GQ --+ GLn{iFp) with
N{p)/N.
(i) The number of isomorphism classes of such p's is bounded by
2n2+n+1 (ILl) (2
1l"n
!
+2 np
n n-1N)n 2n-1N n
p .
(ii) The order of the image of such a p is bounded by
_2n_(_n+_l_)"":;(1:-1_.1....:;)_n , n2 n(2n-l)Nn2
2 n.n P .
1l"n
A sharper estimate is given in [9]. The outline of the proof is: First,
we bound the discriminant of K and the conductor of X when the con-
ductor of p = Ind~ X is given. We give an upper bound of the number
of algebraic number fields K of degree n and discriminant (outside p)
dividing D in terms of n, p and D. For a given K, we give an upper
bound for the number of characters X of GK with a given Artin conduc-
tor N. Combining these results together, we obtain the above Theorem
(i). Finally, we deduce the estimate of the order of the image of Ind~x
from that of the image of X by group theoretic arguments.
Recently, we refined Tate's bound in terms of the reduced Serre weight
k{p) of 1 ([12]).
Lemma 11. Suppose p : Gco.r, -+ GL2(iFp) is wildly ramified, with 0,

/3 as in (1,1). Let k = k(p) be the reduced Serre weight of p. Put
d := (a,/3,p - 1) = (a,k - 1,p - 1). Let pm be the wild ramification
230 H. MOON
index of the kernel field K/Qp Then we have
k-1 k-1+d
1+ -- - if 2 ~ k
Vp('DK/Qp) = { P -l (p - l)~m
~ p,
2 + (p _ l)p - (p -l)pm if k = p + 1.
Then, using this Lemma, we can prove that there exist no continuous
irreducible representations p : GQ --+ GL2(iFp) unramified outside p for
p ~ 19 and with small Serre weight k. For example, there exist no
such representations in the cases p = 17 (k = 2,3,5,7), and p = 19
(k = 2,3,5,7).
References
[1] A. Ash and W. Sinnott, An analogue of Serre's conjecture for Galois representa-
tions and Hecke eigenclasses in the mod-p cohomology ofGL(n,Z), Duke Math.
J. 105 (2000), 1-24.
[2] S. Brueggeman, The nonexistence of certain Galois extensions unramified out-
side 5, J. Number Theory 75 (1999), 47-52.
[3] L. E. Dickson, "Linear Groups," Dover, New York, 1958.
[4] B. Edixhoven, Serre's Conjectures, in "Modular forms and Fermat's Last The-
orem," 209-242, Springer-Verlag, 1997.
[5] J.-M. Fontaine, Il n'y a pas de variete abelienne sur Z, Invent. Math. 81 (1985),
515-538.
[6] C. Khare, Conjectures on finiteness ofmodp Galois representations, J. Ramanu-
jan Math. Soc. 15 (2000), 23-42.
[7] M. J. Larsen and R. Pink, Finite subgroups of algebmic groups, preprint (1998).
[8] H. Moon, Finiteness results on certain mod p Galois representations, J. Number
Theory 84 (2000), 156-165.
[9) H. Moon, The number of monomial mod p Galois representations with bounded
conductor, Tohoku Math. J. 55 (2003), 89-98.
[10] H. Moon, The non-existence of certain mod p Galois representations, to appear
in Bulletin of Korean Math. Soc.
[11] H. Moon and Y. Taguchi, Mod p Galois representations of solvable image, Proc.
Amer. Math. Soc. 129 (2001), 2529-2534.
[12] H. Moon and Y. Taguchi, Refinement of Tate's discriminant bound and non-
existence theorems for mod p Galois representations, to appear.
[13] G. Poitou, Sur les petits discriminants, in "Seminaire Delange-Pisot-Poitou, 18e
annee: (1976/77), Theorie des nombres, Fasc. 1," Exp. No.6, Secretariat Math.,
Paris, 1977.
[14] J.-P. Serre, Sur les representations modulaires de degre 2 de Gal(Q/Q), Duke
Math. J. 54 (1987), 179-230.
[15J D. A. Suprunenko, "Matrix Groups," American Mathematical Society, Provi-
dence, R.I., 1976.
[16J J. Tate, The non-existence of certain Galois extensions ofQ unramified outside
2, Contemp. Math. 174 (1994), 153-156.
FROBENIUS MODULES
AND GALOIS GROUPS
B. Heinrich MATZAT
Interdiszipliniires Zentrum
fur Wissenschaftliches Rechnen (IWR)
der Universitiit Heidelberg,
1m Neuenheimer Feld 968
69120 Heidelberg, Germany
[email protected]
Contents:
Introduction
I. Finite Frobenius Modules
1. The Solution Field of a Finite Frobenius Module
2. Cyclic Frobenius Modules
3. The Associated ID-Module
4. The Galois Group of a Finite Frobenius Module
5. Effective Finite Frobenius Modules
II. p-adic Frobenius Modules
6. Integral p-adic Frobenius Modules
7. The Associated D-Module
8. The Solution Field of an Integral DF-Module
9. Frobenius Automorphism Groups
10. Reduction of p-adic DF-Modules
Introduction
In these notes some basic facts on Frobenius modules are collected.
Frobenius modules are finite-dimensional vector spaces over fields with
a Frobenius endomorphism </>, provided with an injective </>-semilinear
Frobenius operator ~.
The first part is devoted to Frobenius modules M over fields F of pos-
itive characteristic. It is shown that for any such Frobenius module M
[received: September 11, 2002; accepted in revised form: February 25, 2003]
234 B. H. MATZAT
there exists an extension field E containing a full system of -invariant

solutions. The solution field E is a finite Galois extension of F, and
any finite Galois extension of F can be obtained in this way. Further in
case F is a field with an iterative derivation, M becomes in a natural
way a differential module, whose Picard-Vessiot field coincides with the
Frobenius solution field (in the case of an algebraically closed field of
constants). This correspondence in mind we obtain an upper bound for
the Galois group given by the smallest connected linear algebraic group
in which a representing matrix of is contained. Since lower bounds
can be obtained by specialization in analogy to the classical Dedekind
criterion, this technique promises to be a useful tool for the construction
of finite Galois extension with given (connected) Galois group in positive
characteristic.
In the second part, we study Frobenius modules over p-adic Frobe-
nius fields and p-adic Frobenius rings, respectively. The latter are called
here integral p-adic Frobenius modules (in contrast to "bounded p-adic
Frobenius modules" as in [13] or [8]). It is shown that integral Frobe-
nius modules over p-adic number fields F always have a solution field
but that these in general are not algebraic over F in contrast to the
previous case. If the base field F is a differential field, again any integral
Frobenius module Mover F is a differential module, and a solution field
of the Frobenius module contains the corresponding PV-extension. On
the other hand after an unramified extension of the field of constants a
solution field of the differential module contains a full system of solutions
of the underlying Frobenius module, proving the existence of Frobenius
solution fields. The automorphism group over F of such a Frobenius
solution field is a p-adic analytic group, in which the subgroup of ele-
ments commuting with the Frobenius operator, the so-called Frobenius
automorphism group, is dense (at least in the case of connected groups).
Finally, it is shown that the residue module of an integral p-adic Frobe-
nius module (obtained by reducing modulo the valuation ideal of F)
becomes a Frobenius module in positive characteristic as studied in the
first part. Further, the Galois group of the reduced module is a subgroup
of the quotient of the Frobenius Galois group by its principal congru-
ence subgroup (at least in the connected case). This essentially proves
Conjecture 8.5 in [9J in the case of Frobenius modules.
Partial results are contained in Taguchi [16J in the characteristic p
case and in Tsuzuki [17J in the p-adic case (see also the recent Bourbaki
article of Colmez [4]).
I. Finite Frobenius Modules

1. The Solution Field of a Finite Frobenius
Module
1.1. Let F be a field containing the finite field lFq of q = pd elements.
Such a field together with the Frobenius endomorphism
<pq : F -+ F, a 1-+ aq (1)

is called here a finite Frobenius field for short (because the field fixed
by <pq is finite). Correspondingly a finite-dimensional F-vector space M
with an injective <pq-semilinear map
<I?q : M -+ M, x 1-+ <I?q(x) (2)

is called a finite Frobenius module (over F). The set of fixed points of
the Frobenius endomorphism <I?q on M
(3)
is called the solution space of (M,<I?q). Further (M,<I?q) is said to be
trivial if Sol4>(M) contains an F-basis of M. Obviously for a Frobenius
field (E,<p:) with E ~ F and Frobenius endomorphism <P: extending <pq
the tensor product ME := EFM in a natural way becomes a Frobenius
module over E. By abuse of notation
(4)
is called the solution space of Mover E and M is trivial over E if
Sol~(M) contains an E-basis of ME.
1.2. The following theorem is proved for iterative differential fields

F in [9], Prop. 5.6 and [8], Thm. 2.18, respectively, but it holds more
generally for any finite Frobenius field (compare [16], Lemma 3.2 for a
categorical treatment).
Theorem 1.1. Let (M, <I?q) be a Frobenius module over a finite Frobe-
nius field (F, <pq).
(a) For every Frobenius extension field (E, <P:) the solution space
Sol~(M) is an IFq-vector space of dimension
dimlF'q (Sol~(M)) :5 dimE(E F M) = dimF(.M). (5)

236 B. H. MATZAT
(b) There exists a finite Frobeniu8 field extension E / F such that

dimlFq (Sol~(M)) = dimF(M). (6)
(c) The minimal Frobenius field extension E / F admitting a full so-
lut~c(/' space is unique (inside a given algebraic closure of F) and
Galois over F.
Proof. Obviously Sol~(M) is an lFq-vector space. We assume that
dimJFq(Sol~(M)) > m := dimF(M). Then there a exists a minimal
number n of elements Xi E Sol~(.i\1) which are linearly independent
over IFq but linearly dependent over E. This implies a nontrivial rela-
n n
tion L aiXi = 0 over E with an = 1. Because of L </>q(ai)xi = 0 with
i=l i=l
</>q(an ) = 1 the difference of the two equations gives a nontrivial relation
of Xl,X2, ... ,Xn-l over E in contrast to the choice of n.
In order to prove (b), we choose a basis B = {b 1 , b2, ... , bm } of M.
m
Then by 4.>q(bj ) = L dijb i with dij E F we obtain a representing matrix
i=l
m
D:= DB(4.>q) E Fmxm of 4.>q with respect to B. Now let By = LbiYi
i=l
with y = (Yl,"" Ym)tr E Em be a solution of Mover E. Then
By = 4.>q(By) = 4.>q(B)</>q(y) = B . D </>q(Y) (7)
holds with </>q(Y) = (Yf,"" y'/n)tr. This equation defines a system of
algebraic equations Y - Dyq = 0 over F for Yi, where the Jacobian
matrix equals the identity matrix. Hence by the Theorem of Bezout this
system has qn different solutions (with multiplicity 1) in an algebraic
closure F of F. The field E / F generated by these solutions has finite
degree over F and admits a full system of solutions.
Obviously E b minimal inside F and does not depend on the chosen
basis B. Since all the zeroes of y - Dyq = 0 are simple, E / F is separable
and normal, hence Galois. 0
Corollary 1.2. Every Frobenius module over iFp and more generally
over any separably closed field containing IFp is trivial.
1.3. The unique Galois extension E in Theorem 1.1(c) is called the

solution field of (M,4.>q) and a fundamental system Y E GLm(E) of
solutions y E Em is called a fundamental solution matrix. It is char-
acterized by the property </>:(Y) = D-ly. From this observation we
obtain the following characterization of solution fields.
Corollary 1.3. The field E is the solution field of the finite Probenius
module (M,4>q) over F if and only if the following holds:
(a) There exists a matrix Y = (Yij)f,j=l E GLm(E) with </>:(Y) -

D- 1 y.
(b) E / F is generated by the coefficients Yij of Y, i. e., E = F(Y}.

By a change of basis B 1--+ B . 0 (and Y 1--+ C- 1 y respectively)
with 0 E GLm(F), the matrix D = DB(4)q} E GLm(F) transforms
into 0- 1D</>q(O). Such matrices are called Probenius equivalent in the
following. Frobenius equivalent matrices define the same solution field.
2. Cyclic Frobenius Modules

2.1. A Frobenius module (M,4>q) of dimension m over a Frobenius
field (F, </>q) is called cyclic if it possesses a q>q-basis, i.e., an element
bE M such that {b, 4>q(b), ... , 4>~-I(b)} forms a basis of M.
Theorem 2.1. Let (F, </Jq) be a finite Probenius field and (M, q>q) a
Frobenius module over F with dimF(M) = m. Assume F contains more
than ( q; ) elements, then M is cyclic.
m
Proof. Let E / P be a solution field of M and x = By = L biYi E
i=1
Sol~(M), i.e., y = D</>q(Y) = Dyq.
( q; )
Since F possesses more than
elements, there exists an F -linear combination Z = t
~=1
CiYi
in normal position with em = 1 (compare [6], Prop. 3.7.22). Then the

Buchberger algorithm applied with the variables Yl, ... ,Ym-l, z in lex-
icographical ordering leads to a polynomial equation f (z) = 0 where
f(X) E F[X] is monic separable and additive of degree qm , i.e., I(X) =
m
L ai Xqi with am = 1.
i=O
Now let ZI, ,Zm be a basis of the IFq-vector space of zeroes of I in
F. Then the Moore determinant
z~
Zm )
= det(Z) (8)
qm-l
Zm
238 B. H. MATZAT
is different from zero ([5], Lemma 1.3.3). All Zj are F-linear combi-
zt.
nations of the Yi. Because of >q(Y) = D- 1 Y for the given solution
matrix Y the same holds for the powers Hence there exists a matrix
C E GLm(F) such that Z = CY. From this equation we obtain
D = Y <pq(y)-l = 0- 1 Z<pq(Zr 1<pq(C) = C- 1D<pq(C) (9)
with
0 1 0 0
jj-l= (10)
0
0 0 1
-aD -am-l
where the entries ai are the coefficients of the additive polynomial f (X).
Thus D is Frobenius equivalent to D. The image of the basis 13 :=
BC- 1 = {b- 1 , ... , -bm } is
~q(13) = BC- 1CD>q(C)-1 = 13D
- a1 - - a m -1 - 1-
= (b2 - - bl,." ,bm - - - bl, - - bm ). (11)
ao ao ao
Hence b := b1 defines a ~q-basis of M. 0
Another proof for Theorem 2.1 with F infinite can be found in [16],
Lemma 3.3. The polynomial f(X) E F[X] in the proof only depends on
the q-basis {b} of M and ~s called the characteristic polynomial fb of
M with respect to b. Obviously, the splitting field of fb over F coincides
with the solution field of (M", ~q).
2.2. Given a cyclic Frobenius module with ~q-basis {b} and with
corresponding fundamental solution matrix Z as in the proof of Theorem
2.1 (with respect to the basis 13 with b1 = b, b2 = ~q(bd+~ b1 , , bm =
~q(bm-1) + a';;~l b1), we obtain the characteristic polynomial fb in the
following way (compare [5], Lemma 1.3.6).
Corollary 2.2. Let (M, ~q) be a cyclic Probenius module with ~q-basis
{b}. Then the characteristic polynomial of M with respect to b equals
z~
Zm )
m-l
z~
(12)
denotes the corresponding fundamental solution matrix.
3. The Associated ID-Module

3.1. In this section let F ~ IFq' q = pd, be a field with an iterative
derivation a F = (a~)keN' This is a field with a family of maps a1k ) :
F ...... F with aC:) = idF and the following properties for a, b E F and
i,j,k EN:
ac;)(a + b) = ac;)(a) + ac;)(b), (13)

ac;) (a b) = L a~) (a) a~) (b), (14)
i+j=k
a(i) a(j) _
F F -
( i +ij F
) a(i+j)
' (15)
a series of subfields Fz := n
Such a field sometimes is called an ID-field for short ([8], 1.1). It defines
Ker(ar) whose intersection K is the field
j<l
of constants of F, i.e.,
K = nFz
leN
= {a E F I a~)(a) = 0 for k > a}. (16)
On F we have a natural Frobenius endomorphism p : F ...... F, a 1--+ aP

In case Fz = ~(F) we call (F, aF' p) an ID-field with Frobenius struc-
ture or an IDF-field for short. Obviously, ID-fields which are function
fields of one variable over a perfect field are always IDF-fields (compare
[9], 2.1 or [8J, 1.5).
3.2. A vector space M over an ID-field (F, aF) with a family of maps
aM = (aft) is called an iterative differential module (or ID-module
keN
for short) if aM fulfills (13), (15) and for a E F, x E M
a~)(a. x) = L a~)(a)aW(x) (17)

i+j=k
for the scalar multiplication (instead of (14)). Now we want to show that
any finite Frobenius module over an IDF-field in a natural way becomes
an ID-module.
Proposition 3.1. Let (M, <Pq), q = pd, be a finite Frobenius module
over an IDF -field (F, a F,p) .
240 B. H. MATZAT
(a) There exists a unique iterative derivation oM = (O~)kEN on M

such that
~~(M) = n
Ker(o~).
j<dl
(18)
(b) If CPz : M -+ M denotes the F -linear extension of the Fd(I+1) -linear

embedding <PI : ~~+1(M) -+ ~~(M) we have
ott) = 'Po 0 . 0 'Pl-l 0 oe;) 0 'Pl':l 0 0 'POl for k < pdl, (19)
where at) is defined on the coefficients with respect to an Fdl-basis

of~~(M).
Proof. First let oM

be an iterative derivation with the property in (a).
Then with respect to a basis B = {bl, ... ,bm } of ~~(M) for any x E M
we have
m m
o~)(x) = o~)(~= biYi) = LbiO~)(Yi) for k < pdl. (20)
i=l i=l
This shows the uniqueness and (b). On the other hand the formula in
(b) defines an iterative derivation oM on M with the property ~~(M) =
nKer(o~)
j<dl
(compare [9], Prop. 5.1 or [8], Thm. 2.8). 0
The ID-module (M, aM' ~q) in Proposition 3.1 is called the ID-module
associated to (M, ~q), and the triple (M, aM' ~q) is an ID-module with
Frobenius structure or an IDF-module for short.
3.3. For an ID-module (M,oM) over an ID-field (F,o}) the solution

space is defined by
Sol(M):= n
l~kEN
Ker(o~). (21)
0E extending of'
If E / F is a field extension with an iterative derivation
the tensor product ME := E F M becomes an ID-module over E by
o~~(ax) = L o~)(a) oW(x). (22)

i+j=k
As in Section 1.1 the solution space of ME

SOlE(M) := Sol(E M) (23)
F'robenius modules and Galois groups 241
is called the solution space of Mover E. This is a K-vector space of di-

mension dimK(SoIE(M)) $; dimF(M). It is well known that there exists
a smallest ID-field extension ElF such that dimK(SoIE(M)) = dimF(M)
called an iterative Picard- Vessiot field (IPV-field) for M (compare [9],
Lemma 3.4 or [8], Thm. 3.4).
In case (M, 8M) is an ID-module associated to a finite Frobenius mod-
ule (M, ~q), the corresponding solution spaces are related in the follow-
ing way (compare [8], Thm. 3.12 in the special case d = 1).
Theorem 3.2. Let (F, 8;', q) be an IDF -field with algebraically closed
field of constants K and (M, 8 M,~ q) an IDF -module over F with q = pd.
Then the following holds:
(a) The solution field E of the Frobenius module (M, ~q) is an IPV-
field E of (M, 8M) and
SOIE(M) = K lFq Sol~(M). (24)
(b) The IPV-field E of the ID-module (M,8M) is a solution field of
(M, ~q) and the solution space of the Frobenius module (M, ~q) can
be obtained as the lFq-vector space of ~q-invariants in SOIE(M),
i.e.,
(25)
Proof. Let B = {b1 , .. ,bm } be a basis of the Frobenius module (M, ~q),
let E be the solution field of (M, ~q) and By E Sol~(M). Then by
definition D-ly = q{Y) belongs to q(E)m = ET, where D = DB(~q)
is the matrix representing ~q with respect to B. By induction we obtain
~-l(D-l) ... q{D-1)D-1y = ~(y) E ~(E)m = E:Jl. (26)
Hence By belongs to SOIE(M) by the criterion [8], Thm. 2.11(2). This
implies that K lFq Sol~(M) ~ SOIE(M). Equality holds since by the
proof of Theorem 1.1 (a) a basis of Sol~ (M) remains linearly independent
over E.
Now let E be an IPV-field of the ID-module (M, 8M) with basis B
and Y E GLm(E) a fundamental solution matrix. Then B = BY is a
K-basis of SOIE(M). Because of
~q(BE) = ~q(BY)q(E) = BDq{Y)Ed = By- 1 Dq(Y)Ed (27)
with y-l DE GLm(Ed) and q(Y) E GLm(Ed) we obtain by induction
for alII EN
(28)
242 B. H. MATZAT
This implies
m
n<p~(M) = EBKb; = SolE(M). (29)
lEN i=l
Hence <Pq acts on SolE(M) and SolE(M) is a Frobenius module over K

which is trivial by Corollary 1.2. 0
4. The Galois Group of a Finite Frobenius

Module
4.1. Let (M, <Pq) be a finite Frobenius module of dimension mover
a Frobenius field (F, <pq) and let E be the solution field of (M, <P q) over
F. By Theorem 1.1 (c), E / F is a finite Galois extension whose Galois
group is called the Galois group Gal~(M) of (M, <p). In the case F is an
ID-field with algebraically closed field of constants, this group Gal~(M)
further coincides with the ID-Galois group GalID(M) of the associated
ID-module (M,OM*) (compare [9], 4.1 or [8], 3, respectively).
First we note that Gal~(M) has a faithful linear representation in
GLm(IFq). This observation is based on
Proposition 4.1. Let (M, <P q) be a finite Frobenius module over a finite
Frobenius field (F, <pq) with solution field E and let Y, Y E GL m (E) be two
fundamental solution matrices of (M, <Pq). Then there exists a matrix
C E GLm(lFq) such that Y = YC.
Proof. Obviously there exists aCE GLm(E) with Y = YC. Then

from Y = D<pq(Y) and Y = Dq(Y) we obtain C = q( C) implying
C E GLm(lFq). 0
Corollary 4.2. The Galois group Gal~ (M) of a finite Frobenius module
(M,<Pq) of dimension m is isomorphic to a subgroup ofGLm(lFq).
Proof. Let E be a solution field of (M, <Pq) and Y E GLm(E) be a fun-
damental solution matrix. Then for any, E Gal~(M) = Gal(Ej F) the
matrix ,(Y) again is a fundamental solution matrix of (M, <Pq). Hence by
Proposition 4.1 there exists a matrix C-y E GLm(IFq) with ,(Y) '-' YC-y.
Then
(30)
defines a linear representation which is faithful because of E = F(Y). 0
4.2. From the representing matrix D = DB(41q) of the Frobenius

operator of a finite Frobenius module (M,41q) we obtain bounds for the
Galois group Gal4>(M) in a similar way as in the case of ID-modules.
Theorem 4.3. Let (M,41q) be a finite Probenius module over a finite
Frobenius field (F, 4l q ). Let Q be a reduced connected linear group defined
over IFq. Assume that there exists a basis B such that DB(41q) E Q(F).
Then
(31)
Proof. The proof relies on the surjectivity of the Lang isogeny
1T : Q(Fsep) ~ Q(Fsep), (xij)0=1 f-t q(Xij)(Xij)-l (32)

([15], Thm. 4.4.17). This proves the existence of a matrix Y E Q{Fsep)
with q{y)y- 1 = D-l. By Corollary 1.3, E := F{Y) is a solution field
of (M,4>q). Now for any"'( E Gal4>(M) = Gal(E/ F) the representing
matrix Gf = y-l"'((y) in Corollary 4.2 belongs to Q(E) and to GLm{lFq),
so Gf is an element of Q(IFq). 0
In the case q = p, i.e., d = 1, Theorem 4.3 already is contained in [9],

Prop. 5.4(3) or [8J Cor. 5.3, respectively.
4.3. Lower bounds for the Galois group Gal4>(M) can be obtained by
an analogue of the Dedckind criterion for linear groups. This is based
on
Proposition 4.4. Let (M,41q) be a finite Probenius module over a finite
field F ~ lFq with [F: IFq] =: f and D = DB{4>q) a representing matrix
of 4>q with respect to a basis B of M. Then
Gal4>{M) ~< 15 >, where 15:= Dq(D). {-l{D). (33)
Proof. Let Y E GLm{iFp) be a fundamental solution matrix and E :=

F(Y) the solution field of (M,4>g). Then F = E<</ for = { From
Y = D>q(Y) we derive Y = D>(Y). Since > generates Gal(E/ F),
there exists a matrix G</> E GLm(lFq) with >(Y) = YG</>. By the above
G</> = y- 1 fj-l Y is conjugate to fj-l inside GLm(E). 0
The proof of Proposition 4.4 shows more precisely that the linear
group r(Gal4>(M :::; GLm(iFp) given by (31) contains a subgroup H
which inside GLm(E) is conjugate to (fj). This is the key of the following
"Linear Dedekind Criterion" .
244 B. H. MATZAT
Theorem 4.5. Let (M, ~q) be a finite Probenius module over a global
field (F, q) with F ~ IFq. Let OQ :::; F be a valuation ring with valuation
ideal PQ and with residue field FQ := OQ/PQ :::; iFp . Assume D =
DB{~q) belongs to GLm{OQ) with residue matrixDQ E GLm{FQ). Then
Gal~{M) :::; GLm{lFq) contains a matrix CQ in GLm{FQ) conjugate to
~ t
DQ := DQcpq(DQ)'" cPq Q- (DQ)
1
with fQ := [FQ : IFqJ. (34)
Proof. Let E/ F be a solution field of (M, ~q) and Y E GLm(E) a funda-
mental solution matrix. Then from q(Y) = D-Iy with D E GLm(OQ)
we obtain Yij E Oij for all r,oefficients Yij of Y and all valuation ideals
Pij in E extending PQ and in addition det(Y) E O~. Hence the reduc-
tion Yij of Y is well defined and belongs to the linear group GLm(Eij)
over the residue field Eij := Oij/Pij. Over Eij the reduced equation
Yij = DQq(Yij) holds. Hence Eij contains the solution field EQ of the
reduced Frobenius module (MQ' ~Q) whose representing matrix of ~Q
with respect to the reduced basis coincides with DQ. From Eij = FQ(Yij)
finally follows Eij = EQ. .
Since FQ is a finite field, the Galois extension EQ/ FQ is cyclic and
its Galois group is generated by the power CPQ = CPtQ of the Frobenius
automorphism. By Proposition 4.4 the matrix CQ defined by Q(Yij) =
YijCQ fulfills CQ = Y<i 1DQIyQ. Now let E Gal{E/ F) be a pre-image
of Q which is trivial on the decomposition field Tij of Pij/PQ. Then
there exists a matrix C4J E GLm{lFq) with (Y) = YC4J. Reducing
modulo Pij we obtain Q{Yij) = YijC4J from which C4J = CQ follows.
This proves the theorem. 0
From the uniqueness of C4J in the above proof we obtain the unique-
ness of the lift of Q inside Gal{E/Tij)' This implies that under the
condition D E GLm(OQ) all extensions of valuation ideals Pij ~ PQ are
unramified in E / F.
4.4. The following simple example shows how Theorems 4.3 and 4.5
can be combined to determine the Galois group of a finite Frobenius
module. Let M = FbI ffi Fb2 be a 2-dimensional Frobenius module
over F = lFp{t) with D = DB{~p) = (~ t~1 ). Obviously D is an
element of the connected Borel subgroup B2(F) :::; SL 2(F). Hence by
Theorem 4.2 Gal~(M) :::; B2{lFp). The latter is a group of order p(P-1).
Reducing modulo PQ = (t -1) we obtain DQ = (~ i) E SL2{lFp) of
order p, reducing modulo (t - w) with a generating element w of IF'; we

obtain the element DQ = (~ w: 1 ) E SL2(lF'p) of order divisible by
(p - 1). This implies Gal4>(M) ~ B2(lF'p).

A more substantial example is elaborated in Malle [7], giving equations
for G2 (IFq )-extensions over IFq ( t) .
5. Effective Finite Frobenius Modules

5.1. Next we are interested in the question for which finite Frobenius
modules the upper bound for the Galois group given in Theorem 4.3
is sharp. First we show that the assumption that the Galois group is
connected is necessary.
Proposition 5.1. Let (M, ~q) be a finite Frobenius module over a finite
Frobenius field (F, q) with F 2: iFp . Assume there exists a linear alge-
braic group g defined over IFq with D(~q) E g(F) and Gal4>(M) = g(lF'q).
Then g is connected.
Proof. Let ElF be a solution field of (M,~q) and Y E GLm(E) a fun-

damental solution matrix, i.e., Y = D:(Y). By a theorem of Borel-
Serre [1] and Platonov (see [18], Lemma 10.10), the connected compo-
nent gO of g has a finite supplement rt in g. Hence there exist matrices
Do E gO(F) and Co E rt(F) with D = DoCo. Since rt is finite we obtain
rt(F) = rt(iFp). Now as in the proof of Theorem 4.3 the surje~ivity of
the Lang isogeny proves the existence of a matrix C E GLm(IFp) with
COl = q(C)C-1. We obtain
C- 1y = C- 1D:(Y) = C- 1DoCo:(Y) = C- 1DoC:(C-1y). (35)
Because of C-IDoC E gO(F)C ~ gO(F), Theorem 4.3 shows that

Gal4>(M) :::; gO(IFq ). 0
5.2. Now let (M, ~q) be a finite Frobenius module over an arbitrary
finite Frobenius field (F, q). Then (M, ~q) is called an effective Frobe-
nius module if there exists a reduced connected linear algebraic group g
defined over IFq such that the representing matrix D = DB(~q) belongs
to g(F) (for a suitable basis B of M) and in addition Gal4>(M) = g(lF'q).
Correspondingly a finite Galois extension ElF is called an effective Ga-
lois extension if E is the solution field of an effective Frobenius module
over F.
The following theorem shows the existence of effective Galois exten-
sions in some interesting C9Ses:
246 B. H. MATZAT
Theorem 5.2 (Nori). Let 9 be a semisimple simply connected linear

algebraic group defined over IF q. Then there exists an effective Galois
extension E/IFq(t) with Gal(E'/IFq(t)) ~ g(IFq) which is unramified at all
finite places (i. e., outside (r 1)).
The p.i.'C0f follows from [10] or [2], Ch. 14.3, respectively.
5.3. It is an interesting question if for a given field F all Galois ex-

tensions E / F with a given Galois group are effective. A positive answer
could help to construct generic additive polynomials for these groups
(in the sense of Saltman [14]) as characteristic polynomials of Frobenius
modules. But here up to now we can only present a conjecture.
Conjecture 5.3. Let (F, q) be a finite Frobenius field of cohomologi-
cal dimension cd( F) ~ 2 and E j F a finite Galois extension. Assume
Gal(Ej F) is isomorphic to the group of Fq-rational points of a semisim-
ple simply connected linear algebraic group. Then E j F is an effective
Galois extension.
The conjecture would give an analogue for finite Frobenius modules
of the following known result for ID-modules ([8]], Ch. 5.2): Every ID-
module over an ID-field F with cd(F) ~ 1 and with connected ID-Galois
group 9 is effective. The proof of the last result is based on the Theorem
of Springer and Steinberg stating HI (F, 9F) = o.
II. p-adic Frobenius Modules

6. Integral p-adic Frobenius Modules
6.1. In the second part of these notes we study Frobenius modules
over p-adic Frobenius fields. Let us first fix our notation. Zq denotes the
Witt ring W(Fq) of the finite field Fq of q = pd elements and by Qq its
quotient field. This is the unique unramified extension of degree d of the
field of p-adic numbers Qp = Quot(W(lFp)). The union 1I'p:= U
Qpd
I~dEN
of all these fields is the maximal unramified algebraic extension of Qp. Its
completion Tp coincides with the quotient field of the Witt ring W(iFp)'
For each d the field 1I'p possesses a unique Frobenius automorphism
<pq = pd whose reduction q to the residue field iFp gives the q-power
Frobenius automorphism. It 1S continuous with respect to the p-adic
valuation and generates Gal(1I'p/Qq). It follows that the fixed field of
the unique continuous extension q of q to j p still remains Qq.
Frobenius modules and Galo'is groups 247
The algebraic closure of Qp is denoted by Qp = Q~lg and its com-

pletion, the p-adic universe, as usual by Cpo The unique Frobenius au-
tomorphism 4>p_ on '!E can be extended (in a non-unique way) to an
automorphism 4>p of Qp/Qp and further it can be continuously extended
to Cpo
In the following a complete non-archimedian field F containing Qq
with a continuous Frobenius endomorphism 4>[ : F - F reducing to
the q-power Frobenius endomorphism is called a p-adic Frobenius field.
Examples of such Frobenius fields are (Qqf' 4>q), (Tp, q) and (Cp, q)
where the first two of them additionally are discrete valued. Notice
that a subfield L of a p-adic Frobenius field (F,4>f) with the restricted
Frobenius endomorphism 4>: IL may fail to be a Frobenius field (in the
case 4>: (L) i L).
6.2. Now let (F,4>f) be any p-adic Frobenius field with F ~ Qq.
Then a finite-dimensional F-vector space with an injective 4>[-semilinear
map ~: : M - M is called a p-adic Probenius module. The solution
spaces Sol~(M) and Sol~(M) := Sol~(E 0F M) for a p-adic Frobenius
extension field E / F are defined exactly as in the finite case in Section 1.1,
in particular M is called trivial over E if Solk(M) contains a basis of
the E-vector space ME := E F M.
Proposition 6.1. Let (M, ~f) be a p-adic Probenius module over a
p-adic Probenius field (F, 4>f) with F ~ Qp. Then the following holds:
(a) For all p-adic Frobenius extension fields E / F the solution space
Solk(M) of Mover E is a vector space over the field of 4>:-
invariants E4> := E<4>:> with
dimE(Solk(M)) S dimF(M) =: m. (36)
(b) Equality holds in (a) if and only if there exists a matrix Y E

GLm(E) such that Y = DB(~f)4>:(Y) with the representing ma-
trix DB(~n of~[ with respect to a basis B of M.
(c) Any two matrices Y, Y E GLm(E) with the properly in (b) only
differ by a matrix C E GL m (E4, i.e., Y= Y . C.
The proofs are exactly the same as in the finite case (compare Thm. 1.1
and Prop. 4.1) with IF q substituted by E4>.
A Frobenius field (E, 4>:) ~ (F, m~') containing a fundamental solution
matrix Y as in Proposition 6.1(b) is called a solution field of (M, ~f),
248 B. H. MATZAT
if ElF is generated by the entries Yij of Y. The p-adic Frobenius field

(E, ~) obtained by the completion of E and continuous extension of
</>~ is called a complete solution field. In contrast to the finite case the
existence of (complete) solution fields is not obvious.
6.3. A p-adic Frobenius module (M, q,f) is called an integral Frobe-

nius module (or an etale Frobenius module according to [17], Prop. 3.2.1)
if it contains a f-stable Op-Iattice A of full rank, i.e., with rank(A) =
dimp(M), and
Op: (A) = A. (37)
The corresponding lattice A is then called a Frobenius lattice.
Theorem 6.2. Let (M, :) be an integral p-adic Frobenius module over
a p-adic Frobenius field (F, <1':) with [F : Qq] < 00. Then (M, f) has
a solution field E inside Cp and unramified over F.
Proof. Let FT denote the intersection Fn1['p and let E be the completion
of F PT 1['p with the induced Frobenius endomorphism </>:. Then F 1FT
is totally ramified of finite degree and ElF is unramified. As before,
l~t Op be the valuation ring of F, rOp = Pp its valuation ideal and
F = Op IPp its residue field which is finite.
By assumption M contains a Frobenius lattice A with basis Band
with D := DBC:) E GLm(Op). Hence the residue matrix D of D
belongs to GLm(F). The surjectivity of the Lang isogeny 11' : GLmCiFp) -+
GLm(iFp) already used in Section 4.2 gives a matrix Do E GLm(iFp) with
D- I = q(Do)Dol. In fact, the entries of Do belong to some finite
extension Fol F. Thus there exists a Frobenius field (Fo, </>:0) unramified
and of finite degree over (F, f) and a matrix Do E GLm(Opo) such that
D- I = :0(Do)(1 + rGO)DOI with Go E o~xm. (38)
Now we want to refine the resulting congruence D- I = :O(Do)Do l

(mod r) modulo higher powers of r. The next such approximation step
with DI = 1 + rHI and :(r) = er would lead to the congruence
1 + rCo =q(DI)(1 + r 2GI )D1I

= (1 + </>q(rHI))(1 - rHI )
= 1 + erq(HI) - rHI (mod r2). (39)
Since the reduced equation C}o = eq(iil ) - iiI with the q-power Frobe-
nius endomorphism q over ib has a solution matrix HI over a finite
Probenius modules and Galois groups 249
extension I'd Fo there exists a Frobenius field (Fl' cf{l) unramified and
of finite degree over (Fo,</{O) and a matrix Dl = J + rHl E GLm(Fl)
such that
D- l = <1>:0 (DO)q{l (Dd(J + r 2GI)D1l DOl with Gl E OF{xm.
(40)
Thus by induction we obtal!I a tower of unramified field extensions F :::;
Fa:::; Fl :::; .,. :::; Fl inside E and matrices Dl E GLm(Fl) such that
D- l == :O(Do)'" <l>f1(D,)Dl- l ... DOl (mod rl+ l ). (41)
Because of Dl = I + rl Hl E GLm(.E), the product Do'" Dl converges

in GLm(E). Hence there exists a matrix Y E GLm(E) with D-l =
1>f(y)y-l. By Proposition 6.1(b), Y is a fundamental solution matrix
of (M,cpn. 0
Since any p-adic Frobenius module over (Qp, q) is already defined
over a finite extension F /Qp, from Theorem 6.2 immediately follows
Corollary 6.3. Every integral p-adic F'robenius module over the F'robe-
nius field (Qp, p) becomes trivial over Cpo
6.4. Now we want to show by examples that integral p-adic Frobenius

modules over p-adic number fields, i.e., finite extensions of Qp, may have
transcendental solutions. For this purpose set F := Qp and M := Fb
with Frobenius endomorphism given by cp:(b) = J1.b and J1. E Z;' Then
by Theorem 6.2, the solution field E/ F is generated by an element y E Tp
which satisfies the functional equation <t>ff (y) = J1.-ly. Since <t>ff is an
automorphism of E / F, the field extension E / F is algebraic if and only
if J1. is a root of unity.
7. The Associated D-Module

7.1. Let (K, <t>q) be a discrete valued p-adic Frobenius field with
K ~ Qq. Then on the rational function field K(t), the Gauss norm
II . II with II t II = 1 dennes a valuation extending I . I on K. In the
following, this will again be denoted by I I. The completion F = K(t)
of K(t) with respect to the Gauss valuation becomes a Frobenius field
-
by defining <t>: (t) = t q and by extending 1>:
continuously. As before
the valuation ring and the valuation ideal of F are denoted by OF and
PF, respectively. Then the residue field I' := OF/PF is isomorphic to
the field of rational functions over the residue field K = OK /PK, i.e.,
F = K(t).
250 B. H. MATZAT
The p-adic Frobenius field F = K(t) with the derivation at

obtained
from at = -it
by continuous extension in addition is a differential field
(D-field) with field of constants K. In the following an arbitrary p-adic
Frobenius field (F, f) with a derivation OF is called a p-adic differential
field with Frobenius structuTe or a p-adic DF -field for short if the D-
structure and the Frobenius structure are related by the formula
l F F . of(f(t)) ( )
UF 0 q = ZF . q 0 OF wIth ZF = f(OF(t)) E PF 42
t. For the powers of q:= :

for a non-constant t E F. Obviously ZF is independent of the choice of
we derive
OF 0 ~ = ZF>q(ZF) ... ~-1 (ZF )>~ 0 OF for 1 S lEN (43)

implying
OF(~(OF)) ~ orlOF for r E PK with Irl = IZFI. (44)
Then the quotient field Fz := ~(F) of Oz := ~(OF) is something like a

"congruence constant field". In particular, (F, OF, f)
is an example of a "rational" p-adic DF-field with F} = K(t ql ).--
= (K(t), lit, :)
7.2. Next we want to show that on an integral Frobenius module M

over a p-adic DF-field in a natural way a derivation OM can be defined
which is compatible with the Frobenius action in the above sense.
Proposition 7.1. Let (F, OF, >f) be a discrete valued p-adic DF-field
with F 2: Qq and let (M, ipf) be an integral p-adic Frobenius module
over F with Frobenius lattice A.
(a) For 1 S 1 E N the Frobenius image Az := ip~(A) of A is an Oz-
lattice inside A containing an OF-basis of A.
(b) The OZ+1 -linear embeddings <Pz : AZ+1 ~ Az define a projective
system (Az, <PZ)ZEN.
(c) Any <PI in (b) can be extended uniquely to an OF-isomorphism
CPI : A = OF Ol+l AI+l -- OF Ol Al = A.
Proof. We only have to verify (a): Al is an Ol-lattice because of ip~(ax) =
~(a)ip~(x) for a E F and x E M and it contains a basis of A since
OF(CPf(A)) = A by (37). D
7.3. Now we are able to define a differential structure on M in a way

very similar to the finite case (compare [17J, Thm. 3.3.2, for fields of
power series).
Theorem 7.2. Let (M, <.pf) be an integral p-adic Frobenius module with
Frobenius lattice A over a discrete valued p-adic DF -field (F, OF, c/Jf) with
F~Qq.
(a) Then there exists a unique derivation OM on M with congruence

solution lattices AI, i. e.,
oM(Al) == 0 (mod riA) for 1 ~ lEN. (45)
(b) This derivation is characterized by the congruences
OM == $0 0" '0$1-1 OOFO$~\ 0" '0$0 1 (mod rIA) for 1 ~ lEN.

(46)
Proof. First we assume the existence of a derivation OM on M and A
with congruence solution lattices Al = <.P~(A) as in (a). Then for any
basis Bl = {bil,"" bl m } of Al we get oM(blj ) == 0 (mod rZ A). Hence for
m
x = Lajblj EM with aj E F we obtain
j=1
m m
OM(X) = I)oF(aj)b1j + ajoM(blj ) == LOF(aj)blj (mod rIA). (47)
j=l j=l
This shows the uniqueness of OM and the formula in (b).

Now we want to use thp. congruences in (b) for the definition of OM.
For this purpose we let D = DB(<.P q ) E GLm(OF) denote the represent-
ing matrix of <.P q = <.P: with respect to a basis B = {bl' ... , bm } of A. By
induction we see that Bl := BDo'" DZ-1 with Dk = c/J~(D) E GLm(Ok)
is an Ol-basis of AI' In view of the congruences in (b) we now define
8z(x) := BDo'" DI- 10F(Di-\ ... Do1y) E A (48)
m
for x = LbjYj = By E A. Because of OF (D1- 1) == 0 (mod rl) the
j=l
coefficients of 8t (x) converge in F, hence
OA(X) := lim (81 (x E A (49)
1...... 00
is well defined. It is easy to verify that the map OA : A -- A is additive

with oA(ax) = of(a)x + aOA(x) for a E OF. Hence OA is a derivation
252 B. H. MATZAT
on A with congruence solution lattices Al by construction. Finally, we

obtain the desired derivation OM on M = F OF A by setting OM :=
OF idA + idF OA. 0
For later use we state the explicit congruence formulae for OM found
in the proof as a corollary.
Corollary 7.3. Let B = {b1 , ... , bm } be a basis of the Probenius lattice
A in Theorem 7.2, D := DB(CP q ) E GLm(OF) and Dl := ~(D) for
1 EN. Then the oM-derivative of x = By E A has the property
oM(By) = BDo'" DI-lOF(D~\ '" Daly) (mod r l ). (50)
The differential module (M, OM) constructed in Theorem 7.2 is called
the D-module associated to (M, <pn
and the triple (M, OM, <P:) is a p-
adic D-module with Probenius structure or a p-adic DF -module for short.
7.4. From Theorem 7.2, we also can derive a criterion for the solu-
tions of the D-module (M, OM) associated to an integral p-adic Frobenius
module. As in Section 3.3 for any extension D-field E of F the solution
space of Mover E is denoted by
SolE(M) := {a x E E F M I oE(a) x + a OM (x) = o}. (51)

It is a vector space over the field of constants KE of (E,OE) with
dimKE(SolE(M)) ~ dimF(M). Correspondingly for the Frobenius lat-
tice A in M we define
SOlE(A) := {a x E OE OF A I oE(a) x + a OM(X) = O} (52)

which is an OKE-Iattice of full rank in SolE(M). In the following, we
use the notations ME for E F M and AE for OE OF A.
Proposition 7.4. Let (M, OM, :) be an integral p-adic DF -module with
Frobenius lattice A as in Theorem 7.2 and Corollary 7.3. Then for any
D-field extension E of F the following statements are equivalent.
= By E SolE(A),
(a) x
(b) OE(y) == AIY (mod r l+ l ) for lEN with
Al := of(Do ''' DI)(Do'" DI)-l,
(c) OE(Y) = Ay with A = lim (AI) E Or;xm.

l--too
Proof. As in the proof of Theorem 7.2 let Bl = BDo'" Dl- 1 be an Ol-
basis of AI. Then any x = By E AE with yEO"! can be written as
x = BIYI with Yl = (Do'" DI_1)-ly EO,,!. Now x = By belongs

to SoIE(A) if and only if 8E(YI) == 0 (mod rl) for all lEN. The last
congruence is equivalent to
8E(Y) = 8E(Do'" DI-1YI) == 8p(Do'" Dl-dYl = AI-1Y (mod r1).

(53)
From Dz E GLm(Oz) follows
Al = 8p(Do'" Dz)(Do '" Dl)-l == AI- 1 (mod r l ). (54)
Hence A := lim (AI) exists in Or;xm and the congruences in (b) are
Z-oo
equivalent to the one equation 8E(Y) = Ay. 0
Proposition 7.4 further implies the defining property of a (>, 8)-
module (compare for example [17J or [4]).
Corollary 7.5. For an integral p-adic DF-module (M, 8M , ~n, we have
8M o~: = zp~: 08M with Zp as in (42). (55)

Proof. The definition of Az gives
AzDo = 8F(Do ... DI)(Do ... DZ)-l Do
= 8p(Do) + Do8F(>:(Do '" DI-d)>:(Do'" DI_d- 1
= 8p(Do) + zpDo>:(8p(Do'" DI-I)(Do '" DI_d- 1 )
= 8p(Do) + zpDo>: (AI-I) (56)
leading to the matrix identity
(57)
But then for the basis B of M we obtain
8M(~:(B)) = 8M(BDo) = -BADo + B8p(Do)

= -zpBDo>:(A) = zp~:(-BA) = zPCb:(8M(B)). (58)
o
7.5.
-
Now we want to apply Proposition 7.4 to two examples. First
let F = Qp(t) and M = Fb the 1-dimensional Frobenius module with
Cb:(b) = tb. Then from DB(Cb:) = (t) we obtain
AI=t1 (1 +p.+. ... +pI) and A = (l-p)t

1 (59)
254 B. H. MATZAT
Thus the solution field E of the differential module M is generated by

an element y with aE(Y) = (l!P)t Y' hence by Y = t 12P = P-~t-I.
For the second example we use the same base field F and M = FbI EB
Fb2 with DB(iP:) = (~ ~). This leads to
A= (~ g) with g = l)itpJ-I E F.
jEN
(60)
Hence the fundamental solution matrix Y of M has the form
Y=(~ {) with f = 2)pJ (j. F.

jEN
(61)
Since the differential Galois group of the solution field E / F is a subgroup

of the additive group, f has to be transcendental over F.
8. The Solution Field of an Integral DF-Module

8.1. As before let F be a discrete valued p-adic DF-field with field
of constants K and (M, cpn an integral Frobenius module over F. By
Theorem 7.2, the Frobenius endomorphism cprinduces a natural deriva-
tion aM on M providing M with the structure of a D-module over F.
Then differential Galois theory proves the existence of a Picard- Vessiot
extension (PV-extension) E / F for (M, OM) over F := K K F, where K
denotes an algebraic closure of K (see for example [12]). By definition E
is a solution field of (M, OM) over F. We will see that the PV-extension
E / F is already defined over a finite extension of constants of F.
Proposition 8.1. Let (M, OM) be a D-module over aD-field (F, OF)
with field of constants K. Then (M,OM) has a solution field (E,OE)
with field of constants KE of finite degree over K. Moreover, using
FE := KEK F, the field extension E/ FE is a PV-extension of (M, OM)
over FE.
Proof. The prooffollows from the construction of a PV-ring R and a PV-
field E as presented in [12] in the case of an algebraically closed field
of constants. There the field of constants KE of E (and R) is finitely
generated and algebraic over K, hence finite. 0
By Proposition 8.1 the D-module (M, aM) associated to an integral
p-adic Frobenius module (M, CP:) has a solution field E which is a PV-
extension over a finite extension of constants. In contrast to the finite
case, we cannot expect that its solution space SOIE(M) always contains
solutions of the original Frobenius module: We have already seen in

Section 6.4 that the solution field of a "constant" Frobenius module
over F may contain transcendental elements over F.
8.2. In this section we study the relation between the solution spaces
of a DF -module with respect to its F'robenius and differential structure
and prove the existence of F'robenius solution fields.
Theorem 8.2. Let (M, iPf) be an integral p-adic Frobenius module over
a discrete valued p-adic DF-field (F, OF, 4>f) with field of constants K ~
Qq.
(a) Assume an F -field (E,4>:) is a solution field of (M, iPf)
over F
with solution space Sol~ (M) . Then E is a PV -extension for the
associated D-module (M, OM) over some constant field extension
FE := KE K F, and the corresponding solution spaces are related
by
SOlE(M) = KE Ktf> Sol~(M) (62)
E
where Kt is the fixed field of KE with respect to 4>:.

(b) Assume a PV-extension E / FE with FE = KE K F is a solution
field of (M, OM). Then there exists an unramified extension K / KE
inside Cp such that the Frobenius module (M, iPq) becomes trivial
over 13 := K K E. Moreover the solution spaces over P := K K F
are related by
(63)
Proof. Let A be a F'robenius lattice of M, B = {b 1, ... , bm } an OF-

basis of A and D = DB(iPf) E GLm(OF) the representing matrix of iPf
with respect to B. Then by Proposition 6.1 there exists a fundamental
solution matrix Y over E, i.e., Y E GLm(E) with Y = D4>:(Y). By
induction it follows that Y = Do DI-1Yi with Dk = 4>~(D) and Yi =
4>~(Y) E Glm(EI). Hence the congruences
OE(Y) == OF (Do ... DI-dYi = AIY (mod rl) (64)
with Al = OF (Do ... DI-1)(Do DI_1)-1 define a derivation OE on E

extending OF with OE(Y) = AY for A = lim (AI) as in Proposition 7.4.
1..... 00
Thus Y is a fundamental solution matrix of (M, OM) over F. By Propo-
sition 6.1 the solution space of (M, OM) is given by
SolE(M) = KE Etf> Sol~(M) (65)

256 B. H. MATZAT
where gl> = K~ because of Sol~(M) ~ SoIE(M).

To prove (b) we first observe that equation (55) in Corollary 7.5 or
its matrix version (57), respectively, implies that 8M(~q(SoIE(M))) = 0
from which we conclude that ~q(SoIE(M)) ~ SoIE(M). Thus the KE-
vector S~3:;e SoIE(M) is a Frobenius module over the field of constants
KE. By Theorem 6.2, (SoIE(M), ~q) possesses an unramified solution
field KIKE inside Cpo Hence (M,4>q) becomes trivial over E, and the
solution spaces are related by formula (63). 0
8.3. Next we are going to prove without using PV-theory that an

integral p-adic Frobenius module (M, ~n over a rational p-adic DF-field
(F, 8F, >n, where the field of constants K has finite degree over Qq, has
a solution field whose field of constants is un.!].@ifi5:.d over K. In this case
(F, 8F, >:)is the field of analytic elements (K (t), 8t ,>n
-with coefficients
in K (compare [3], Def. 2.1.3). Further let MK S K((t)) denote the
field of meromorphic functions on the open disk 'OK = {a E K I lal < 1}
which becomes a DF-field by continuous extension of 8t and >: ([3],
Ch.2.5).
Proposition 8.3. Let (M, 8M, 4>n
be an integral p-adic DF-module
over a rational p-adic DF-field (F, 8F, >n
with field of constants K of
finite degree over Qq. Then the Frobenius module (M, 4>:)
has a solution
DF-field E whose field of constants KE is unramified over K.
Proof. Let A be a Frobenius lattice in M, B = {bI , ... , bm } an OF-basis
of A and D := DB(~n E GLm(OF) the representing matrix of the
Frobenius endomorphism with respect to B. By a linear transformation
of t we can assume that t = 0 is regular for D, i.e., D(O) E GLm(OK).
Then by Theorem 6.2 there exists an unramified extension LI K and a
matrix C E GLm(OL) such that
Do := C- I D>~(C) =I (mod tOL). (66)
Hence there exists a Go E Or xm with Do = 1+ tGo. This leads to
Dl := >~(Do) = I + >~(t)>~(Go) = I + >~(t)Gl E GLm(OL). (67)

Thus the sequence of matrices Zl := Do'" Dl E GLm(OL) converges in
GLm(ML) with limit Z := lim (Zl). Defining EL by EL := FL(Z) where
1--'00
FL := F@K L, the field EL is a subfield of ML with field of constants L
and hence a DF-field.
By construction we obtain Z = DO:L(Z). For Y := CZ then follows
Y = CDO:L(Z) = CC- 1 D~(C):L(Z) = D:L(Y). (68)
Hence Y E GLm(E) is a fundamental solution matrix of (M, ~p over

L. The subfield E := F(Y) of EL is a solution DF-field of (M, ~q ) over
F whose field of constants KE :::; L is unramified over K. 0
8.4. Finally we want to generalize the solutions of (M, ~n con-

structed in Proposition 8.3 to solutions convergent on the generic disc.
-
For this purpose the completion of KE in Proposition 8.3 is again de-
-
noted by K. Further let F(z) be the completion of the field of rational
functions over F with Izl = 1 with respect to the Gauss norm and
M} :5 Fz - t) the field 0/ merom orphic functions on the generic disc
-
'D} = {u E F(z)llu - tl < I} (69)
with coefficients in F. Then the map
TF: F ~ M},/(t) f-+ Lk\lJf(f)(z - t)k (70)
-
keN'
......
identifies the valued D-field (F, OF) = (K(t),Ot) with the subfield
(F*,OF*) .- (TF(F),8z ) of M}, where the D-structure is translated
by
(71)
(compare [3], Prop. 2.5.1). Now from Theorem 8.3 we can derive:
Corollary 8.4. With assumptions as in Proposition 8.3 for F := FE
the Frobenius module (M, ~f) has a solution field inside the field of
meromorphic functions M} on the generic disc with coefficients in F.
Proof. By continuous extension we obtain an injection TE : E ~ M} of
the solution field E :::; Kt into M} extending TF. This translates the
equation Y = Dr/>: (Y) into
Y* := TE(Y) = TF(D)TE(r/>:(Y = TF(D)r/>fCY*) (72)
with E* := TE(E). Hence E* = F*(Y*) is a solution field in M} of

(M, ~n over F* (compare [3], Prop. 4.6.3). 0
9. Frobenius Automorphism' Groups

9.1. As at the beginning of the second part, let (F, r/>q) be a p-adic
Frobenius field and (M, ~n an ir.t'Jgral p-adic Frobenius module over F
with Frobenius lattice A and basis B of A. Assuming the existence of the
258 B. H. MATZAT
solution field (E,1>f) of M (as for example shown in Theorem 6.2) we

define the Frobenius automorphism group of M (or of E / P respectively)
by
Aut~(M):= (r E Aut(E/P) 1,,01>: 1>:
= o,,}. (73)
This group acts in a natural manner on ME = M FE and on Solk(M)
via ,,(x y) = x ,,(y). In case the fundamental matrix Y belongs to
GLm(OE) (as in Theorem 6.2), the intersection of Solk(M) and AE =
A op OE is an O~-lattice with basis BY called the lattice of integral
solutions of M in E:
Solk(A) := Solk(M) n AE. (74)
With these assumptions we define the Probenius automorphism group of
A to be the subgroup of Aut~CM) stabilizing AE:
Aut~(A) := b E Aut~(M) I ,,(AE) ~ AE}. (75)
Proposition 9.1. Let (M, ~:) be an integral p-adic Frobenius module
of dimension m with Probenius lattice A over a p-adic Frobenius field
(P,1>f) with solution field (E,1>f) and SCI> = F<P.
(a) Aut~(M) is a reduced Zariski-closed subgroup of GLm(FrI.
(b) Aut~(A) is a Zariski-closed subgroup ofGLm(Ot) and is Zariski-

dense in Aut~(M) in case the latter group is connected.
Proof. For (a), let Y E GLm(E) be a fundamental solution matrix of
M, i.e., Y = D1>:(Y) with D = DB(~:)' Then for any " E Aut~(M)
(76)
Thus ,,(Y) is again a fundamental solution matrix of M. By Proposi-
tion 6.1, there exists a C, E GLmCErI = GLmCprI with ,,(Y) = YC,.
Hence
(77)
defines an injective homomorphism because of E = F(Y). As in the
case of D-modules Aut~(M) coincides with
Aut~(M) = {C E GLm(P<t I fCYC) = 0 for all f E P} (78)
where
P = {f E F<t>[GLmJ 1 fey) = O} <1 p<t>[GLmJ (79)
denotes the ideal of Y = (Yij )0=1 inside the coordinate ring of GL m over
p<t> (compare [12], Observations 4.2). Therefore Aut~(M) as a subgroup
of GLm(FCP) is defined by algebraic equations and hence is Zariski-c1osed

in GLm(FCP) and reduced.
In (b) M has a lattice of integral solutions Sol~(A). We obtain
(SO)
Thus Aut~(A) too is Zariski-c1osed in GLm(O~). If moreover Aut~(M)
is connected then Aut~(A) is Zariski-dense in Aut~(M) by [11], Lem-
ma 3.2. 0
From Proposition 9.1 follows that Aut~(M) in addition is a p-adic
analytic group and Aut~(A) is an open compact subgroup of Aut~(M)
(using [11], Ch. 3.1). It remains to answer the question under what con-
ditions F already is the fixed field of Aut~(M) or Aut~(A), respectively.
9.2. Now we assume that (F,q) is a p-adic DF-field as in 7 and

(M, aM, q,:) is an integral DF-module over F. Then by Proposition S.l,
a solution field E of M is a Picard-Vessiot extension over a constant
field extension FEI F. Hen,~e without loss of generality we can assume
that ElF is a PV-extension with differential Galois group Galn (M) =
Galn(EI F). This then is a group 9(K) of K-rational points of a linear
algebraic group 9 defined over the common field of constants K of F and
E. Now in analogy to the last section we define the OF-automorphism
group of M by
Aut~(M) := b E Galn(M) 1,0 : : o,}
= (Sl)
and the DF-automorphism group of A by
Autt(A) := b E Autt(M) I,(AE) ~ AE} (S2)
in case a lattice of integral solutions exists.
Proposition 9.2. Let (M, aM, q,f) be an integral p-adic OF-module
over the OF -field (F, OF, :) with field of constants K, lattice A, solution
field (E, :) and differential Galois group GaIn (M) ~ 9 (K) .
(a) Suppose 9 is defined over the fixed field KCP of :IK then
Autt(M) ~ 9(KCP) (S3)

is a locally compact p-adic analytic group.
(b) Suppose 9 is defined over the ring of integers O~ of KCP then
AuttCA) ~ 9CO~) (S4)

260 B. H. MATZAT
is a compact subgroup of9(K4.

(c) In case 9 is connected, both groups Autt(M) and Autt(A) are
Zariski-dense in Q(K).
Proof. The proofs of (a) and (b) immediately follow from Proposition 9.1
and the remark following it. It only remains to verify (c), namely that
Q(K4 and Q(O~) are Zariski-dense in Q(K) in case Q is connected. But
this is a consequence of [11], Thm. 2.2 and Lemma 3.2, respectively. 0
By Proposition 9.2 (c) in the case of connected groups the fixed fields
of Autt(M) and Autt(A) equal the base field F. Hence it seems to be
appropriate to call these groups DF-Galois group of M or of A, respec-
tively, and to denote them hy Galb(M) and Galb(A).
9.3. Finally as in the case of finite Frobenius modules for a DF-

module M we want to give an upper bound for GalD(M) (and hence for
Galb(M) and Galb(A), too) only depending on the representing matrix
of the Frobenius endomorphism of M.
Theorem 9.3. Let (M, OM, 4!f) be an integral p-adic DF -module defin-
ing a PV -extension over a p-adic DF -field (F, OF, cf>f) and A a Frobenius
lattice in M. Let 9 be a reduced connected linear group defined over the
field of constants K of F. Assume there exists a basis B of A such that
D = DB (cpf) belongs to the group Q(OF) of OF -rational points of Q.
Then
Gab(M) ::; Q(K). (85)
Proof. The matrices Al = of(Do''' DI)(Do'" DI)-l in Proposition 7.4
belong to the Lie algebra LieF(Q) of Q over F since they are images of
the logarithmic derivative
(86)
Now LieF(Q) is complete with respect to the p-adic topology induced

from F. Hence by the proof of the Proposition 7.4 the matrices Al con-
verge to a limit A E LieF(Q) defining OM. But this implies GalD(M) ::.;
Q(K) according to [12], Cor. 4.3. 0
9.4. With Theorem 9.3 and Proposition 9.2 we obtain the following
Frobenius Galois groups for the examples studied in Section 7.5. In the
first case M = Fb with DB(4!%) = (t), the solution field ElF is cyclic
of order p - 1. Thus
Gal(EI F) = Galb(M) = Galb(A) ~ Zp-l. (87)
F'robenius modules and Galois groups 261
In the second example M = Fb l E9 Fb2 with DB (~:) = (~ i), the

differential Galois group is the additive group. This leads to
GaID(E/ F) = Gal~(M) ~ Qp and Gal~(A) ~ Zp. (88)
10. Reduction of p-adic DF-Modules

As before let (M, aM, ~n be an integral p-adic DF-module
10.1.
with Frobenius lattice A over a DF-field (F, OF, n
and with field of
constants K ~ Qq. Then the residue field F := OF/,PF of the valuation
ring OF (with respect to the Gauss norm) is a field of characteristic p.
Assuming OF is an ID-ring, i.e.,
for kEN, (89)
F is an IDF-field with respect to the reduced Frobenius action q and

the induced iterative derivation o~ defined by
o~)(a + 'PF) := ~!oi(a) + PF for kEN. (90)
Moreover the field of iterative differential constants K of F coincides

with the residue field OK /'PK'
Now let B = {bl, ... , bm } be a OF-basis of A and D = DB(~n E
m
GLm(OF) the representing matrix of ~r Then A = EBOFbi reduces
i=l
to the F -vector space
m
M := A/'PFA = EBFbi (91)
i=l
with the reduced basis 13 = {bl' ... ,bm } and with a q-semilinear Frobe-
nius action given by
F
~q:M-M, B- f-+ B- . D(~q
-
), (92)
where jj(~:) E GLm(F) denotes the matrix obtained from D(~:) by

reducing modulo 'PF. Hence (M, ~q) is a Frobenius module over F.
According to Proposition 3.1, the Frobenius endomorphism ~q defines a
canonical iterative derivation o~ on M such that (M, o~, ~q) becomes
262 B. H. MATZAT
an IDF-module over F, called the residue module of (M, aM' iI>n. The
next proposition shows that any finite Frobenius module can be obtained
as a residue module of an integral p-adic Frobenius module, i.e., every
finite Frobenius module can be lifted to an integral p-adic Frobenius
module.
Proposition 10.1. Let (F, aF, n be a discrete valued p-adic DF -field
with F 2:: Qq, valuation ring OF and valuation ideal PF with respect to
the Gauss norm. If OF is an ID -ring then the residue field F = OF /PF
is an IDF -field with the induced iterative derivation a~. Let (M, a~, ~q)
be an IDF -module over F. Then there exists an integral p-adic DF-
module (M, aM, iI>n over F and a Frobenius lattice A ~ M such that
the residue module A/PFA is isomorphic to M as an IDF -module.
Proof. Let B be a basis of Ai and D := DB(~q) E GLm(F) be the ma-
trix of ~q with respect to B. Then there exists a matrix D E GLm(OF)
with residue matrix D. On a free OF-module A ofrank m with basis B
we can define a :-semilinear Frobenius action by iI>{(B) := BD and
we can extend it :-semilinearly to the F-vector space M := F OF A.
By Theorem 7.2, iI>: defines a canonical derivation aM on M providing
M with the structure of a integral p-adic DF-module (M, aM, iI>{). Ob-
viously the residue module of (M, aM, iI>{) equals (M, a~, ~{) since by
~roposition 3.1 the iterative derivation a~ is uniquely determined by
iI>q. 0
10.2. Next we want to study the behavior of the Frobenius automor-
-
phism groups under reduction. For simplicity we restrict ourselves to the
case of a rational p-adic DF-field F = K(t). Then the valuation ring OF
is automatically an ID-ring (for aF = at). Further by Proposition 8.3
we can assume without loss of generality that the solution field E has
no new constants, i.e., E / F is a PV-extension.
Theorem 10.2. Let (M, aM, iI>{) be an integral p-adic DF -module with
Frobenius lattice A over a discrete valued rational p-adic DF -field
(F, aF, f) with field of constants K. Assume that the solution field E / F
has no new constants and that the differential Galois group GaID(M) is
connected and defined over ot. Then the residue module (M, a~, ~ q)
of M is a finite Frobenius module over the residue field F= OF/PF
whose Galois group is bounded by
Gal;j;(M) :s; Ga1b(A)j Ga1b(Ah (93)

where Gal~(Ah denotes the principal congruence subgroup of the p-adic

analytic group Gal~(A).
Proof. For the proof we follow the construction of the PV-ring of M
over OF instead of F. Let B = {bl,"" bm } be a basis of A and D =
DB(~:) E GLr. (OF) the matrix of~: with respect to B. Then we have
~:(B) = BD and oM(B) = -BA (94)

with the matrix A E eJ pxm constructed in Proposition 7.4(c). The
ring U:= eJF[GLmJ = eJF[xij,det(xij)-lJr,j=l becomes a DF-ring with
Frobenius endomorphism f and derivation 8r; defined by
f(X) = D-1X and 8r;(X) = AX for X = (Xij)f,j=l' (95)

Further U is an ID-ring by Corollary 7.5 because of
O~(X) = 8f,(Do ... Dl-1Xl) == o~(Do ... Dl-1)Xl (mod rl) (96)
with Xl = Dt=-\ .. . DOIX E (f)I(U). Now let Q <J U be a maximal
ideal stable under the action of f and of Or; and let P <J U be a maxi-
mal differential ideal containing the "DF-ideal" Q. Then the eJF-algebra
R := U/ P is an "integral PV-ring" with quotient field E (up to differen-
tial isomorphism). By assumption E is the solution field of the Frobenius
module (M, ~:), too. Hence there exists a fundamental solution matrix
Y E GLm(R) with
OE(Y) = AY and :(Y) = D-1y. (97)

Thus the ring R of integral differentially finite elements in E is closed
by the action of the induced Frobenius : of f (i.e., the maximal dif-
ferential ideal P already is a DF-ideal) and an ID-ideal. By assumption
Galo(M) = Galo(E/F) is connected, hence by Proposition 9.2 Gal~(A)
is Zariski-dense in GalD(M). Its matrix representation on the solution
space is given by
(98)
where ,eY) = YC-y.
Now let S := F OF R be the residue ring of Rover F, i.e., S =
F[Yij, det(Yij)Jf,j=l with Yij = lYij E S. Because of DE GLm(eJF), the
residue matrix.e belo,Egs to GLm(F) and d~nes the reduced Frobenius
endomorphism q on S whose restriction to F coincides with the q-power
Frobenius endomorphism. The derivation OE restricted to R defines
264 B. H. MATZAT
an iterative derivation O, on 8 as in Section 10.1 compatible with the

Frobenius structure, i.e., (8, o,, q) is an IDF-ring. Hence 8 can be
obtained as a quotient of F[GLml by an IDF-ideal Q:
(99)
The Frobenius automorphism group
is a finite group since F<P = lFq. Because ~ = YG\ = YO,), for

'Y E Gal~(A), the restriction map
- cp
: GalD(A) --+ Aut ~--
(SjF),C,), I--' c')'
-
(101)
is a group homomorphism with kernel Gal~(Ah. It is surjective since
the fixed field of 8 by Aut~ (81 F) can also be obtained as the reduced
fixed field of R by Aut~(A) which equals F. This proves
(102)
Now let P <J F[GL ml be a maximal ID-ideal containing Q. Then R :=

F[GLmJI P is an IPV-ring for M with Frobenius structure and with
Gal~(M) = Gal~(Rj F) = n E GLm(JFq)

I ::Y(P) ~ P}
: : ; fy E GLm(lFq) I ;:;;(Q) ~ Q} = Aut~(8IF)
~ Gal~(A)1 Gal~(Ak (103)
o
There remains the question under which conditions equality can be
proved. This will be answered in the next section in some special cases.
10.3. Nori's examples in Theorem 5.2 are Galois extensions over JFq(t)
unramified outside the denominator of (t) with unipotently generated
Galois group Q(lFq). By construction these belong to effective Frobenius
modules (M, ~q) over F = lFq(t) with DB(~q) E Q(lFq[t]) with respect
to a suitable basis B of M. (compare [9], Cor. 7.6 or [8], Cor. 7.16,
respectively). For modules of this type the following holds.
Proposition 10.3. Let Q be a reduced connected linear group defined
over Zq with the property that 1i(lFq) is a proper subgroup oJQ(JFq) Jor all
proper connected subgroups 1i of g. Let (F,8];;) = (K(t),8t) be a field

with iterative derivation and finite field of constants K ;::: lFq Assume
(M, ~q) is a Frobenius module over F with
for a suitable basis B of M. Then there exists an effective integral p-adic

DF-module (M, 8M, ~n over F = K(t), where Quot(W(K)) =5 K =5 Tp
with differential Galois group
GalD(M) ~ 9(K) and Gal~(A) ~ g(Zq) (105)

for the corresponding Frobenius lattice A in M. The residue module
of (M,8M'~:) coincides with (M,8~,iq), and the Galois groups are
related by
(106)
let K be a quotient field of W(K) -

Proof. For the proof we almost only have to collect earlier results. First
- and F := K(t) the rational p-adic
DF-field of analytic elements with 8F = 8t . Then we can find a matrix
D E g(K[tj) with residue matrix D. Using D as representing matrix
of a Frobenius operator of a free OF-module A, we obtain a Frobenius
lattice and by M := F OF A an integral p-adic Frobenius module over
F together with a canonical derivation 8M by Theorem 7.2. By Propo-
-
sition 8.3 M has a solution field E whose field of constants KE is a
subfield of Tp. Now we substitute K by the completion KE of KE and
F by K(t). Then ElF is a PV-extension with GalD(EIF) =5 g(K) by
Theorem 9.3. __
From D E 9(K[t]) it follows by Proposition 7.4 that in the induced
-
differential equation 8E(Y) = Ay the matrix A E LieF(g) has entries
belonging to K (t). Hence ElF has at most one singular point and
therefore a connected differential Galois group GalD(EIF), which is of
the form 1i(K) with 1i an algebraic subgroup of g. Then by The-
orem 10.2 the Galois group of the residue module M is a subgroup
of 1i(lFq ) = 'H.(K)I1i(Kh. This implies 1i(lFq ) = g(JFq ) and thus
1i(K) = g(K) by assumption. In particular, (M, ~n is an effective
Frobenius module over F. 0
10.4. An elementary numerical example is given by the additive

-
group already treated in Sections 7.5 and 9.4. Here the field F = Q..e(t)
of analytic elements over Qp reduces to the rational function field F =
266 B. H. MATZAT
lFp(t), and the integral DF-module (M,<f:J%,8M ) over F with DB(<f:J%) =

(~ ~) and Gal~(A) = Zp has the residue module (ii, ~p, 8~) with
DB(~P) = (~ ~) and
Gal$(M) ~ lFp ~ Zp/pZp ~ Gal~(A)/Gal~(Ak (107)
References
[1] A. Borel and J.-P. Serre, Theoremes de finitude en cohomologie galoisienne,
Comment. Math. Helvet. 39 (1964), 111-164.
[2] J.-B. Bost et al., "Courbes Semi-stables et Groupe Fondamental en Goometrie
Algebrique," Birkhauser, Boston, 2000.
[3] G. Christol, "Modules differentiels et equations differentielles p-adiques,"
Queen's Paper in Pure and Applied Mathematics, 66. Queen's University,
Kingston, 1983.
[4] P. Colmez, Les conjectures de monodromie p-adiques, Asterisque, (to appear).
[5] D. Goss, "Basic Structure of Function Field Arithmetic," Springer-Verlag,
Berlin, 1996.
[6] M. Kreuzer and L. Robbiano, "Computational Commutative Algebra 1,"
Springer-Verlag, Berlin, 2000.
[7] G. Malle, Explicit realization of the Dickson groups G2(q) as Galois groups,
Pacific J. Math., (to appear).
[8J B. H. Matzat, "Differential Galois Theory in Positive Characteristic," IWR-
Preprint 2001-35.
[9J B. H. Matzat and M. van der Put, Iterative differential equations and the Ab-
hyankar conjecture, J. reine angew. Math. 257 (2003), 1-52.
[lOJ M. V. Nori, Unramified coverings of the affine line in positive characteristic,
in "Algebraic Geometry and its Applications (C. L. Bajaj, ed.)," 209-212,
Springer-Verlag, 1994.
[l1J V. Platonov and A. Rapinchuk, "Algebraic Groups and Number Theory," Aca-
demic Press, Boston, 1994.
[12J M. van der Put, Galois theory of differential equations, algebraic groups and Lie
algebras, J. Symb. Comput. 28 (1999), 441-472.
[13J M. van der Put, Bounded p-adic differential equations, in "Circurnspice, Various
Papers in and Around Mathematics in Honor of Arnoud van Rooij," Subfaculteit
Wiskunde, Katholieke Universiteit Nijmegen, 2001.
[14J D. J. Saltman, Generic Galois extensions and problems in field theory, Adv.
Math. 43 (1982), 250-283.
[15] T. A. Springer, "Linear Algebraic Groups," Birkhiiuser, Boston, 1998.
[16] Y. Taguchi, Regular singularity of Drinfeld modules, Internat. J. of Math. 5
(1994), 595-608.
[17] N. Tsuzuki, The overconvergence of morphisms of etale r.p-V-spaces on a local
field, Compositio math. 103 (1996), 227-239.
[18] B. A. F. Wehrfritz, "Infinite Linear Groups," Springer-Verlag, Berlin, 1973.
Part III
Algebraic number theory

ON QUADRATIC NUMBER FIELDS

EACH HAVING AN UNRAMIFIED
EXTENSION WHICH PROPERLY
CONTAINS THE HILBERT CLASS FIELD
OF ITS GENUS FIELD*
Ken YAMAMURA
National Defense Academy,
Hashirimizu Yokosuka 299-8686, Japan
yamamuraC!lcc.nda.ac.jp
1. Introduction
The purpose of this article is to describe simple ways to construct
quadratic number fields each having an unramified extension which prop-
erly contains the Hilbert class field of its genus field (in the wide sense).
The motivation of this study is the author's observation that under the
Generalized Riemann Hypothesis (GRH), for most quadratic number
fields of small conductors, their maximal unramified extensions coincide
with the Hilbert class fields of their genus fields. More precisely, under
GRH, among the 305 imaginary quadratic number fields with discrimi-
nants larger than -1000, at most 16 fields are exceptional [39], [40], and
among the 1690 real quadratic number fields with discriminants less than
or equal to 5565, only 4 fields are exceptional [41]. Thus, the following
problem arises naturally:
Problem. Characterize the quadratic number fields K haVing an un-

ramified extension which properly contains the Hilbert class field of the
genus field of K.
*This article is based on the author's talk in the conference "Galois theory and modular
forms" held at Saga University (December 14-17, 1999). It was long before the decision of
the publication of the proceedings and some parts were already published. Therefore, we
slightly change some parts and add many new numerical data.
[received: Jul. 03, 2002; accepted in revised form: Oct. 14, 2002]
272 K. YAMAMURA
Before describing our basic ideas, we introduce some notations. We

denote by K = Q( Vd) a quadratic number field with discriminant d.
Kur denotes its maximal unramified extension and Kgen its genus field
in the wide sense, that is, the maximal unramified abelian extension of
K which is abelian over Q. Note that we refer also to the unramifiedness
of the infinite primes and therefore in the case where K is real, this is not
the usual one which admits the ramification of the infinite primes. For
a number field F, we denote by the sequence F = Fo ~ Fl ~ F2 ~ ...
the class field tower of F, that is, Fn+1 is the Hilbert class field of
Fn. The smallest number l with Fl+l = Fl is called the length of the
tower. In these notations, the above problem is to characterize K with
Kur ~ (Kgenh.
We have the following inclusions:
Kur ~ Koo ~ K3 ~ K2 ~ (Kgenh ~ Kl ~ Kgen ~ K,
where Koo = UnKn is the top of the class field tower of K, that is, the
maximal unramified solvable Galois extension of K. These inclusions
lead us to the following three basic ideas.
(1) If Kur ~ K oo , that is, if K has an unramified nonsolvable Galois
extension, then Kur ~ (Kgenh.
(2) If K3 1= K2, that is, if the length of the class field tower of K is at
least three, then KUT ~ (Kgenh.
(3) If K2 ~ (Kgenh, then Kur ~ (Kgenk
On the other hand, the composition is also a basic idea. For some
quadratic number fields K, there exists a Galois extension L of Q such
that the composhum K L is an unramified extension of K not contained
in (Kgenh. If F is a quadratic number field with Fur ~ (Fgenh, then
any quadratic number field K with Kgen ~ F has the same property
Kur ~ (Kgenh. Therefore the existence of such one field implies that a
positive rate of the quadratic number fields have the same property.
Now for simpler description, we introduce the new terminology
"quadratic-ramified" .
Definition. Let F be an algebraic number field of finite degree. We
call an extension E of F quadratic-ramified (over F) if there exists a
quadratic extension K / F such that the compositum K E / K is unrami-
fied at all finite primes.
Note that an extension E / F is quadratic-ramified if and only if L / F is
so, where L is the normal closure of E over F. We will study quadratic-
ramified normal extensions of the rational number field Q.
Unramified extensions 273
The organization of this article is as follows. In Section 2, we describe

the simplest way to construct quadratic number fields K with Kur ;f
Kgen. This way is the one to construct quadratic-ramified Sn-extensions
of Q. In Section 3, we review some facts on the structure of Galois
groups of quadratic-ramified normal extensions of Q. In Section 4, we
describe quadratic number fields with class field towers of length f; 3
and give a simple way to obtain imaginary quadratic number fields with
class field towers of length at least three.
2. Unramified An-extensions of quadratic

number fields
In this section we will describe the simplest way to construct quadratic
number fields K with Kur ~ Kgen.
First, we review the following general fact.
Lemma 1 ([16). See also (27), (37), (38).). Let E be an algebraic

number field of degree n ~ 3. If the discriminant dE of E is a fundamen-
tal quadratic di~criminant, that is, a discriminant of a quadratic number
field, then the normal closure M of E is an Sn -extension of Q and is
unramified at all finite primes over its quadratic subfield Q( y'([E).
Sketch of Proof. If dE is a fundamental quadratic discriminant, then for

its any rational prime divisor p, the prime p has exactly one ramified
divisor in E and its ramification index is two. Then for any prime ideal
'.P of M above p, its inertia group I~ is generated by a transposition if we
consider Gal(M/Q) as a (transitive) subgroup of Sn. By the Cebotarev
monodromy theorem, Gal(M/Q) is generated by all I~ for '.P I dE, and
therefore Gal(M/Q) 9;! Sn, because any transitive subgroup of Sn gen-
erated by transpositions is Sn (see Proposition 1 below).
The unramifiedness follows from I~ nAn = {l}. 0
By this lemma from an algebraic number field of degree n f; 3 whose

discriminant dE is a fundamental quadratic discriminant, we obtain
a quadratic number field Q( y'([E) having an unramified (at all finite
primes) An-extension. Therefore if E is totally real or dE < 0, and
n ~ 4, then M is an unramified extension of Q( v'di) not contained
in (Q( v'di)genh. (Note that since Q( v'di)gen and M are linearly
disjoint over Q( v'di), we have Gal((Q( v'di)genhM/Q( v'di)gen) 9;!
Gal(M/Q( JdE)) 9;! An. Moreover, for any K =1= Q( JdE), we have
Gal(KM/ K) 9;! Gal(M/Q) 9;! Sn and if furthermore Kgen ::> Q( JdE),
then K M / K is unramified at all finite primes.) By constructing (in-
274 K. YAMAMURA
finitely many) number fields with squarefree discriminants t, Y. Ya-

mamoto and K. Uchida independently proved the following [33], [34J,
[35J.
Theorem 1. For each natural number n ~ 3, there exist infinitely many
real and infinitely many imaginary quadratic number fields having an
unramified (at all finite primes) An -extension (Sn -extension).
Therefore, we conclude that to find such a field E is the simplest way
to obtain K with desired property. (For n ~ 5, this gives also the easiest
way to obtain examples of unramified nonsolvable Galois extensions of
number fields.) In fact, probabilistic argument leads the author to the
conjecture stating that at least 60% of algebraic number fields of each
degree n ~ 3 have discriminants that are fundamental quadratic and
the conjecture stating that for at least 40% of monic polynomials with
rational integer coefficients of each degree n ~ 3, their root generates an
algebraic number field with squarefree discriminant [38]. Therefore we
may expect that it is easy to find such fields unless we require that the
fields are totally real. We exhibit below a list of negative and totally real
square free discriminants of degree n for 4 ~ n ~ 16 that are smallest
in absolute value among known ones. (Here, for simplicity, we call the
discriminant of a number field totally real if the field is totally real.)
They are really known to be smallest for n ~ 7 and for n = 8, 9 in the
totally real case ([7], [8], [10], [25J, [26], [6], [13], [32]). Most of the others
are due to D. Simon [31] t.
Table 1
n negative totally real
4 -283 1957
5 -4511 24217
6 -14731 592661
7 -184607 20134393
8 -4296211 483345053
9 -109880167 9685993193
10 -215067767 513087549389
11 -5781612911 30733206894581
12 -172922533711 1272819520434029
13 -5894076408919 62557040066579369
14 -9977823237443 6013825446222379157
15 -336785064838151 363145034778499996657
16 -13627506010977623 75341762604432007766521
tThey do not note the square freeness of discriminants, we can easily verify it.
tD. Simon kindly gave the author his large data [31J few of which can be found at his
homepage. The author expresses gratitude to him.
3. Galois groups of quadratic-ramified normal

extensions
In the previous section we describe a way to obtain quadratic-ramified
8 n -extensions of Q. In this section we review some facts on the struc-
ture of Galois groups of quadratic-ramified normal extensions of Q as
primitive permutation groups, and give examples of such extensions.
Our study is based on the following.
Lemma 2. Let F be an algebraic number field with no nontrivial exten-
sion unramified at all finite primes (e.g. the rational number field Q, any
imaginary quadratic number field with class number one, any cyclotomic
field with class number one, etc.). Then for any quadratic-ramified nor-
mal extension L of F, its Galois group G is generated by elements of
order two.
Proof. Let H be the subgroup of G generated by the inertia groups of
the prime ideals of L which are ramified over F and M the corresponding
subextension of LIF. Then MIF is unramified at all finite primes and
therefore M must coincide with F. Hence H coincides with G. Since
LI F is quadratic-ramified, any inertia group of a ramified prime ideal is
a cyclic group of order two. 0
Let ElF be an extension of degree nand L its normal closure over F.
Then the Galois group of L I F can be considered as a transitive subgroup
of 8n , the symmetric group of degree n. Now, we consider primitive
subgroups of 8 n generated by elements of order two. An element of
order two of 8n is a transposition or a product of disjoint transpositions.
We know the following facts on such subgroups.
Proposition 1. If a transitive subgroup of Sn is generated by transpo-
sitions, this subgroup coincides with Sn.
Proposition 2 ([18, Theorem 3]). Let G be a primitive subgroup
of Sn other than Sn and An. If G contains a product of two disjoint
transpositions, then one of the following holds:
(1) n = 5 and G ~ D5, the dihedral group of order ten.
(2) n = 6 and G ~ PSL(2, 5) ~ A5, the alternating group of degree
five.
(3) n = 6 and G ~ PGL(2, 5) ~ 85, the symmetric group of degree
five.
(4) n = 7 and G ~ PSL(2, 7).
276 K. YAMAMURA
(5) n = 8 and G ~ Hol(Ci)' the holomorph of ci, the elementary

abelian 2-groups of order eight.
Moreover, all these subgroups except for those of (3) are generated by
products of two disjoint transpositions. (The exceptional groups contain
odd permutations . The information for these groups is useless for our
study.)
As described in the previous section, Yamamoto and Uchida proved

the existence of infinitely many quadratic-ramified Sn-extensions of Q
for each n ~ 3. All such extensions are unramified at all finite primes
over their quadratic subfields. (We note that for n ~ 4 there exist
quadratic-ramified Sn-extensions of Q that are not unramified at all fi-
nite primes over their quadratic subfields.) Moreover, the author studied
the distribution of algebraic number fields of degree n ;:;; 3 whose normal
closures are quadratic-ramified Sn-extensions of Q that are unramified
at all finite primes over their quadratic subfields, and posed a conjecture
for its density.
T. Kondo proved the following.
Theorem 2 ([18, Theorem 1]). Let E be an algebraic number field of
degree n ~ 3 whose discriminant dE is the square of an odd squarefree
rational integer & and M its normal closure. We put G = Gal(MjQ)
and consider it as a transitive subgroup of Sn. If G is primitive, then
one of the following holds:
(a) G ~ An.
(b) n = 5 and G ~ D5.
(c) n = 6 and G ~ PSL(2, 5) ~ A5.
(d) n = 7 and G ~ PSL(2, 7).
(e) n =8 and G ~ Hol(Ci).
Moreover, in the cases (b)-(e), M is quadratic-ramified and for any
quadratic number field K whose discriminant is divisible by <5, M K / K is
unramified at all finite primes and furthermore M K / K is a G -extension
unless K c M in the case (b).
Note. The oddness of dE is not essential for the quadratic-ramified-ness.
However, we cannot weaken this to that ffE is fundamental quadratic.
They are missing in [18, Theorem 3]. Since Kondo considers [18, Theorem 3] as a lemma
for [18, Theorem 1] (Theorem 2 below), he forgets to list.
In fact, for some G in the cases (b)-(e), there exist G-extensions with dE
even and v'iIE fundamental that are not quadratic-ramified. Neverthe-
less, this restriction is meaningful, because once one knows the Galois
groups in the cases (b)-(e), the quadratic-ramified-ness follows from the
cubefreeness of dE.
Sketch of Proof. By the assumption dE = &2, for any prime ideal of M
dividing dE, its inertia group is generated by a 3-cycle or a product of
two disjoint t~anspositions. Any primitive group containing a 3-cycle is
An. If G does not contain a 3-cycle, only the other cases can occur by
Proposition 2, because G ~ An from dE = &2. 0
Thus, the problem is the existence of quadratic-ramified G-extensions
ofQ.
Note that in the case (b) where G ~ D s, M has exactly one quadratic
subfield and is unramified over it. Therefore the class number of the
unique quadratic subfield of M is divisible by five. Conversely, we ob-
tain a quadratic-ramified Ds-extension of Q from a quadratic number
field with class number divisible by five. More generally, Y. Yamamoto
[35J proved the existence of infinitely many real and infinitely many
imaginary quadratic-ramified Dm-extensions of Q for general m ~ 3
by constructing real and inaginary quadratic number fields whose class
groups contain an element of order m.
T. Kondo [IJ, [17J, [18J verified that there exist many quadratic-
ramified As-extensions of Q. They are realized by specializations of sex-
tic polynomial constructed by A. Brumer (as splitting fields). A. Brumer
constructed the following family of curves of genus 2 with three param-
eters b, c, d whose Jacobian varieties have real multiplication:
y2 = f(x; b, c, d) :=
x 6 + 2cx s + (c2 + 2c + 2 - bd)x4 + (2c 2 + 2c + 2 - 2bd + b - 4d)x3
+ (c 2 + 4c + 5 - bd + 3b)x 2 + (2c + 6 + 3b)x + (b + 1).
(Until now his result seems not to have been published yet, though
his result has been often cited. K. Hashimoto [9J reconstructed this
family.) K. Anai and T. Kondo [IJ, [17J verified that the Galois group of
f(x; b, c, d) (over the field Q(b, c, d)) is isomorphic to As. (It is expected
that f(x; b, c, d) is parametric for As-extensions of Q.) T. Kondo [17],
[18J observed that many specializations of b, c, d to rational integers yield
polynomials whose splitting fields are quadratic-ramified As-extensions
of Q. We give below some examples of quadratic-ramified As-extensions
of Q with small discriminant &2. (Imaginary ones in Table 2 and real
ones in Table 3.) Each such an extension M is given by the coefficients of
278 K. YAMAMURA
a sextic polynomial X 6 + a1 X 5 + ... + a5X + a6 whose minimal splitting

field is M.
Table 2
al a2 a3 a4 as a6 6
-2 1 -3 2 4 1 653 (prime)
-2 4 0 -7 4 1 1061 (prime)
-1 1 -7 -7 -1 1 1381 (prime)
-3 1 -1 3 -1 -1 1507 = 11 . 137
-1 -3 0 4 0 1 1553 (prime)
Table 3
al a2 a3 a4 as a6 0
-3 -7 16 10 -12 1 5567 = 19 . 293
-3 -5 16 2 -8 1 8311 (prime)
-3 -11 15 51 29 1 10463 (prime)
-1 -19 -:8 26 22 -3 10687 (prime)
-1 -20 21 17 -21 4 10931 = 17643
Now we give an important remark. When we search for quadratic-

ramified A5-extensions of Q, it is much better to consult a list of sextic
fields of type A5 than to consult a list of quintic fields of type A5. (Here,
we mean by "of type A 5 " that the normal closure is an A5-extension of
Q.) In fact, quadratic-ramified fields are rare in the latter list, however,
are often found in the former. (Compare the table in [2] and the database
due to Klliners and Malle [14]. Among the 308 (non-isomorphic) quintic
fields of signature (1,2) and type A5 with discriminants ; 20832 , only
8 are quadratic-ramified. On the other hand, among (known) 211 (non-
isomorphic) sextic fields of signature (2,2) and type A5 with discrimi-
nants ; 215242, at least 70 are quadratic-ramified. (Check is omitted
for fields with even discriminants.)) Therefore Theorem 2 is meaningful
for our study. Why does such a difference occur? Before explaining the
reason, we first note that in most lists of algebraic number fields, fields
are ordered by absolute values of discriminants and we briefly review
the relationship between discriminants and ramifications. Let E be an
algebraic number field with discriminant dE. We know that for a prime
ideal of p of E, we have p I dE if and only if p is ramified and that
NE/QI(p)e-l i dE, where e is the ramification index of p. Therefore the
factorization of IdEI is closely related to the ramification in E. Let M be
an A5-extension ofQ and E5 (resp. E 6 ) its quintic (resp. sextic) subfield.
If M is quadratic-ramified and p is an odd prime number ramified in M,
then the factorization of pin E5 is 12121 or 221. (The former means that
p has exactly three prime divisors, their degrees are all one, and their
ramification indices are 2,2,1, respectively.) In both cases, p2 II dE5'

and the factorization of pin E6 is 121211 and 12122, respectively and
therefore p2 II dE6' On the other hand, if M is not quadratic-ramified
but dE5 is cubefree, any prime divisor q of dE5 without factorization of
the above types is neither 2 nor 3, and its factorization in E5 is 13 11
or 13 2. (In fact, such factorizations can occur.) In these cases, the fac-
torization of q in E6 is 13 13 and we have q4 II dEs' Similar argument
works for the other possible types of factorization. (All possible types of
factorization of ramified primes in quintic fields of type A5 are tabulated
in [3, p. 46, Table 3.1].) Thus, quadratic-ramified fields often appear in
a list of sextic fields of type A5. The above consideration leads us to the
conjecture that among sextic fields of type A 5 , quadratic-ramified fields
have a density > 0.5.
The author verified that there exist many quadratic-ramified
PSL(2,7)-extensions of Q. They are realized by specializations of the
following septic polynomial with two parameters a and A whose Ga-
lois group over Q(a,A) is isomorphic to PSL(2,7) constructed by S. E.
LaMacchina [19]:
x 7 + (-6a + 2)x6 + (8a 2 + 4a - 3)x5 + (-14a 2 + 6a - 2 - A)x4+
(-8a 3 + 6a 2 - 4a + 2 + A)x3 + (8a 3 + 16a2 )x 2 + (8a 3 - 12a2 )x - 8a3 .
The author observed that many specializations of a and A to rational
integers yield polynomials whose splitting fields are quadratic-ramified
PSL(2,7)-extensions of Q. We give below some examples of quadratic-
ramified PSL(2, 7)-extensions of Q. (Imaginary ones in Table 4 and real
a septic polynomial X 7 + alX 6 + ... + a6X + a7 whose minimal splitting
field is M.
Table 4
al a2 as a4 a5 a6 a7 8
-1 -3 0 4 -1 -2 1 1417 - 13109
-3 2 4 -3 -2 -1 -1 2741 (prime)
0 -2 -4 -1 5 5 1 3803 (prime)
-1 -3 -1 2 4 4 1 3983 = 7569
-2 0 4 -3 -4 2 1 4157 (prime)
Table 5
al a2 a3 a4 a5 as a7 8
0 -13 -8 36 27 -12 1 46517 = 181 . 257
-3 -7 15 6 -11 -3 1 62185 = 5 12437
-1 -11 13 ~4 -35 11 -1 81109 = 7 . 11587
0 -22 0 112 -4e -82 1 110849 (prime)
-3 -8 25 8 -28 -15 -1 122693 (prime)
280 K. YAMAMURA
We have the following observation. Among the list of septic fields

of type PSL(2, 7) in the database due to Kliiners and Malle, quadratic-
ramified ones are often found. More precisely, among known 222 (non-
isomorphic but possibly arithmetically equivalent) septic fields of type
PSL(2,7) of signature (3,2) with discriminants ~ 705492 , at least 103
fields are quadratic-ramified, and among known 136 totally real septic
fields of type PSL(2, 7) with discriminants;;; 45316552 , at least 36 fields
are quadratic-ramified.
By the database due to Kliiners and Malle we verified the existence
of quadratic-ramified Hol( C~)-extensions of Q. We give below some
examples of such extensions of Q. (Imaginary ones in Table 6 and real
an octic polynomial X 8 + alX 7 + ... + a7X + as whose minimal splitting
field is M.
Table 6
al a2 a3 a4 as a6 a7 as 8
-4 8 -11 12 -10 6 -3 2 5717 (prime)
-2 0 0 3 -2 0 -1 2 6037 (prime)
-3 5 -4 0 3 0 -3 2 6971 (prime)
-2 2 -3 5 -4 4 -2 1 8317 (prime)
0 0 -3 -1 2 6 2 1 10055 = 5 . 2011
Table 7
al a2 a3 a4 as a6 a7 as 8
-2 -17 15 53 -1 -37 -15 -1 640663 (prime)
-1 -21 23 126 -113 -241 126 1 749971 (prime)
-2 -17 40 30 -78 -14 35 1 855659 = 7251487
3 -11 -34 31 108 -8 -96 -25 1018757 = 953 . 1069
0 -12 5 31 -12 -14 8 -1 1038325 = 5 . 41 . 1013
We have the following observation. Among the list of octic fields

of type Hol( C~) in the database due to Kliiners and Malle, quadratic-
ramified ones are often found. More precisely, among known 234 (non-
isomorphic) totally imaginary octic fields of type Hol(C~) with discrimi-
nants ;;; 4202570441522 , at least 47 fields are quadratic-ramified, among
known 134 (non-isomorphic) oetic fields of type Hol(C~) of signature
(4,2) with discriminants ~ 9820075692 , at least 41 fields are quadratic-
ramified, and among known 596 totally real octic fields of type Hol( C~)
with discriminants ~ 66507458412162 , at least 169 fields are quadratic-
ramified. We note that there exist many quadratic-ramified octic fields
of type Hol( C~) whose discriminants are divisible by the fourth power
of an odd prime. For example, the splitting field of the polynomial
X8 + 4X7 - 408X6 - 2158X s + 43577X 4 + 282422X3 - 505558X2 +

89217X - 10321 is an imaginary quadratic-ramified Hol(C~)-extension
of Q, in which only the rational prime 5717 is ramified. (The factoriza-
tion of the prime 5717 in any octic field defined by this polynomial is
121222.) By this example we can see that the primitive subgroups of 88
of order 1344 (unique up to conjugacy and all isomorphic to Hol(C~))
are generated by products of four disjoint transpositions.
Thus, also for G = As ,PSL(2, 7),Hol(C~), we expect that there exist
infinitely many real and infinitely many imaginary quadratic-ramified
G-extensions of Q, and its proof is our next subject. By our observa-
tion of the rate of quadratic-ramified G-extensions in the database, it is
expected that for these groups G, among G-extensions of Q, quadratic-
ramified ones have a positive density. Since Hoi (C~) is an (group) ex-
tension of PSL(2, 7) by C~, for this group, we can treat the problem also
as an embedding problem.
To conclude this section, we give examples of some other quadratic-
ramified nonsolvable Galois extensions of Q found by consulting Kliiner
and Malle's database.
The splitting field of the polynomial X8 - X7 - 29X6 + 111Xs -
139X4 + 37X3 + 32X2 - lOX - 1 is a real quadratic-ramified PGL(2, 7)-
extension of Q, in which only the rational prime 107509 is ramified.
(The factorization of the prime 107509 in any octic field defined by this
polynomial is 12121211.) By this example we can see that the primitive
subgroups of 88 of order 336 (unique up to conjugacy and all isomorphic
to PGL(2, 7)) are generated by products of three disjoint transpositions.
The splitting field of the polynomial X10 - 4X9 - 11OX8 + 660X7 -
131X6 - 2311Xs + 30X4 + 2453X3 + 1401X2 + 261X + 14 is a real
quadratic-ramified PSL(2,9)-extension of Q, in which only the ratio-
nal prime 354317 is ramified. (The factorization of the prime 354317
in any field of degree lOdefined by this polynomial is 112222.) Note
that PSL(2, 9) ~ A6. By this example we can see that the primitive
subgroup of 810 of order 360 (unique up to conjugacy and all isomor-
phic to PSL(2,9)) are generated by products of four disjoint transposi-
tions. Like as search for quadratic-ramified As-extensions, in order to
find quadratic-ramified A6-extensions, it is much better to consult list
of fields of degree 10 than to consult list of fields of degree 6.
The splitting field of the polynomial' Xll - 4X 1o - 25X9 + 81X8 +
237X7 - 562X6 -lOlOXs + 1574X4 + 1805X3 -1586X2 - 847X + 579
is a real quadratic-ramified PSL(2, ll)-extension of Q, in which only the
'This polynomial is found also in [12).

282 K. YAMAMURA
rational prime 74843 is ramified. (The factorization of the prime 74843

in any field of degree 11 defined by this polynomial is 12 12 122 2.) By
this example we can see that the primitive subgroups of 811 of order 660
(unique up to conjugacy and all isomorphic to PSL(2, 11)) are generated
by products of four disjoint transpositions.
The splitting field of the polynomial X l5 + 4X l4 + 17X 13 + 59X12 +
148Xl l + 295XlO + 406X9 + 75X8 - 132X7 - 514X6 - 436X5 - 124X4 +
145X3 + 106X2 + 20X -1 is an imaginary quadratic-ramified PSL( 4,2)-
extension of Q, in which only the rational prime 12037 is ramified. (The
factorization of the prime 12037 in any field of degree 15 defined by
this polynomial is 1222 42 .) By this example we can see that the prim-
itive subgroups of SIS of order 20160 (unique up to conjugacy and all
isomorphic to PSL(4, 2)) are generated by products of six disjoint trans-
positions.
4. Class field towers of quadratic number fields

Here we discuss class field towers of quadratic number fields.
First we review sufficient conditions for the infiniteness. Let p be
a prime number. For a number field F, we denote by the sequence
F = Fci p) ~ F{P) ~ ... ~ FJP) ~ F~~l ... the p-class field tower of
F, i.e., F~~l is the Hilbert p-class field of FJp). If Cl(p)(K), the p-
part of the class group Cl(K) of K, has sufficiently large p-rank, then
[K~) : K] = 00: For odd p, if the p-rank of Cl(p) (K) is at least three,
then [K~) : K] = 00 (Venkov-Koch-Schoof [15J, [28]). If K is imaginary
(resp. real) and the 2-rank of Cl(2)(K) is at least five (resp. six), then
[K~) : K] = 00. Since the 2-rank of Cl(2)(K) is t or t -1, where t is the
number of prime divisors of d, we can easily construct K with infinite
2-class field tower. However, for odd p, though it is conjectured that
there exist infinitely many real and infinitely many imaginary quadratic
number fields K whose p-class groups have rank ~ 3, it is difficult to
construct such K. The infiniteness of such fields has been proven only
for p = 3. For p ~ 5, only few examples of such imaginary fields are
known for p = 5,7,11.
Next we consider smaller rank cases. For odd p, if the p-rank of
Cl(p)(K) is at least two, then the length of p-class field tower of K is at
least two. In this case we have
K2 ~ K~P)(Kgenh ~ (Kgenh.
In fact, since Kgen and K~p) are linearly disjoint over K,

Gal(K~p) Kgen/ Kgen) ~ Gal(K~p) / K) is not abelian. Therefore, if K2 =
(Kgenh, the odd part ofCI(K) is cyclic. We note that the Cohen-Lenstra
heuristics [5] state that the probability that an imaginary quadratic num-
ber field has class group whose odd part is cyclic is about 0.97757 and
this conjecture matchs well numerical data. Thus, the following prob-
lems arise.
Problem 1. Characterize the quadratic number fields K with
K2 = (Kgenh =1= Kl.
Problem 2. Characterize the quadratic number fields K with
(2) - (K )(2)...J. K(2)
K2 - gen l " " l '
We conclude this article by giving a simple way to obtain imaginary
quadratic number fields with class field towers of length at least three.
First, we repeat the idea described in [40]. Let F be a real real
quadratic number field with CI(2)(F) ~ C2m and CI~)(F) ~ C2m+1 for
some m ~ 1, where Cl~)(F) denotes the 2-part of the narrow class
group CI+(F) of F. Then the narrow Hilbert 2-class field FI~r of F
is a dihedral eM-field: Gal(FI~ar/Q) ~ D 2m+1. For simplicity, we put
(2) (2) ,
M = Fl ,nar and M+ = Fl . Now we assume that
(*) the relative class number h-(M) of M is greater than one.
Then from F we can construct a family of infinite imaginary quadratic
number fields whose class field towers are of length at least three. In
fact, let d' be any negative fundamental discriminant prime to d(F)
and put d = d' . d( F). Then K = Q( v'd) is an imaginary quadratic
number field and the length of the class field tower of K is at least
three. Now the Hilbert class field Ml of M is normal over Q and H :=
Gal(Mt/Q) is a solvable group with H" =f {I}. This implies Kg =1= K2,
because the compositum K Ml is an unramified Galois extension of K
with Gal(KMt/K) ~ H. For since H' corresponds (by Galois theory) to
the maximal abelian subfield of Ml and since Ml / F is unramified at all
finite primes, this field is the genus field Fg of F (in the narrow sense).
By the assumption on F, we have F ~ Fg ~ M+. On the other hand,
since 2 f heM) by [39, Lemma 9], we have Cl+(M+) ~ C2 x Cl(M+).
Hence the condition (*) h-(M) > 1 implies that Mt/M+ is not abelian,
neither is Mt/ Fg Thus, H" =f {I} and Kg =1= K2.
Now we characterize such real real quadratic number fields F in de-
tail. For Cl(2)(F) ~ C2m and CI~)(F) ~ C2m+1 for some m ~ 1, it is
necessary and sufficient that d(F) is of the form 8p or pq, where p and
284 K. YAMAMURA
q are distinct prime numbers with p == q == 1 (mod 4), and the norm
of the fundamental unit E of F is 1. We know necessary and sufficient
conditions for m = 1, and those for m = 2 (see [36]).
We know that there exist only finitely many normal CM-fields with
relative class number one [24] and all dihedral CM-fields with relative
class number one have already determined by S. Louboutin et al [20],
[21], [22].
For m = 1 the above is summarized in [40]. Here we summarize the
above for m = 2 as follows.
Proposition 3. Let p and q are distinct prime numbers satisfying the
following conditions:
(i) p == 1, q t 3 (mod 4).

(ii) (p/q)4 = (q/p)4 = 1.
(iii) The fundamental unit of Q( Jpq) has norm 1.
(iv) If q = 2, p 1= 257. Otherwise, {p, q} 1= {5, 101}, {5, 181}, {13,53},
{13,61}.
Then for any negative discriminant d' prime to pq, the imaginary quad-
ratic number field Q(.,fjXjO!) has class field tower of length at least three.
Remark. For the pairs {p, q} excluded in (iii), F1,nar is a dihedral CM-
field of degree 16 with relative class number one.
Example. Let F = Q()5 . 401). Then Cl F ~ C4, and Cl+(F) ~ C8.
We take d' = -3, that is, let K = Q( (-3)5401). Then d(K) =
-6015 and Cl(K) S:! V4 x Cg By the results of Kisilevsky in [11], we have
K':) = K~2) and Gal(K~2) / K) ~ Ds. Thus, the length of the 2-class
field tower of K is two, the length of the 3-class field tower of K is one,
the length of the I-class field tower of K is zero for any odd I 1= 3, but
the length of the class field tower of K is at least three.
References
[1] H. Anai and T. Kondo, A family of sextic polynomials with Galois
group As - computation of splitting fields and Galois groups, in "Proceed-
ings of AC95 (Tokyo, 1995)," 1996; electoronical; ftp:/ /tnt.math.metro-u.ac.
jp/pub/ac95/anaLtar.gz.
[2] J. Basmaji and I. Kiming, A table of As-fields, in "On Artin's conjecture for odd
2-dimensional representations (G. Frey, ed.)," Lecture Notes in Mathematics,
1585. 37-46, 122-141, Springer-Verlag, New York and Berlin, 1994.
[3] J. P. Buhler, "Icosahedral Galois representations," Lecture Notes in Mathemat-

ics, 654. Springer-Verlag, New York and Berlin, 1978.
[4] G. Butler and J. McKay, The transitive groups of degree up to eleven, Comm.
Algebra 11 (1983), 863-911.
[5] H. Cohen and H. W. Lenstra, Jr., Heuristics on class groups of number fields,
in "Number Theory, Noordwijkerhout, 1983 (H. Jager, ed.)," Lecture Notes in
Mathematics, 1068. 33-62, Springer-Verlag, Berlin and New York, 1984.
[6] F. Diaz y Diaz, Valuers minima du discriminant des corps de degre 7 ayant une
seule place reelle, C. R. Acad. Sci. Paris Ser. I Math. 296 (1983), 137-139.
[7] H. J. Godwin, Real quartic fields with small discriminant, J. London Math. Soc.
31 (1956), 478-485.
[8] H. J. Godwin, On quartic fields of signature one with small discriminant, Quart.
J. Math. Oxford (2) 8 (1957), 214-222.
[9] K. Hashimoto, On Brumer's family of RM-curves of genus two, Tohoku Math.
J. (2) 52 (2000), 475-488.
[10] J. Hunter, The minimum discriminants of quintic fields, Proc. Glasgow Math.
Assoc. 5 (1957), 57-67.
[11] H. Kisilevsky, Number fields with class number congruent to 4 mod 8 and
Hilbert's Theorem 94, J. Number Theory 8 (1976), 271-279.
[12] J. Kliiners, A Polynomial with Galois Group SL2(1l), J. Symbolic Comput. 30
(2000), 733-737.
[13] J. Kliiners and G. Malle, A Database for field extension of the rationals, LMS
J. Comput. Math. 4 (2001), 182-196.
[14] J. Kliiners and G. Malle, A Database for Polynomials over the Rationals, avail-
able from http://www.iwr.uni-heidelberg.deFJuergen.Klueners/datbase.ps.
[15] H. Koch and B. B. Venkov, The p-tower of class fields for an imaginary quadratic
field, Modules and representations, Zap. Nauen. Sem. Leningrad. Otdel. Mat.
Inst. Steklov. (LOMI) 46 (1974), 5-13, 140, (Russian); English translation, J.
Soviet Math. 9 (1978), 291-199.
[16] T. Kondo, Algebraic number fields with the discriminant equal to that of a
quadratic number field, J. Math. Soc. Japan 47 (1995), 31-36.
[17] T. Kondo, Some family of sextic polynomials and their Galois groups, in "Proc.
twelfth symp. on algebraic combinatorial theory (M. Kitazume, ed.)," 165-176,
1995, (Japanese).
[18] T. Kondo, Some examples of unramified extensions over quadratic number fields,
Sci. Rep. Tokyo Woman's Christian Univ., Nos. 121-124 (1997), 1399-1410.
[19] S. E. LaMacchina, Polynomials with Galois group PSL(2, 7), Comm. Algebra 8
(1980), 983-992.
[20] Y. Lefeuvre, Corps diedraux a multiplication comlexe principaux, Ann. Inst.
Fourier (Grenoble) 50 (2000),67-103.
[21] Y. Lefeuvre and S. Louboutin, The class number one problem for dihedral CM-
fields, in "Algebraic number theory and Diophantine analysis (Graz, 1998),"
249-275, de Gruyter, Berlin, 2000.
[22] S. Louboutin, The class number one problem for the dihedral and dicyclic CM-
fields, Colloq. Math. 80 (1999), 259-265.
286 K. YAMAMURA
[23] J. Nakagawa, On the Galois group of a number field with square free discrimi-
nant, Comm. Math. Univ. Santi Pauli 37 (1988), 95-98.
[24] A. M. Odlyzko Some analytic estimates of class numbers and discriminants,
Invent. Math. 29 (1975), 275-286.
[25] M. Olivier, Corps sextiques primitifs, Ann. Inst. Fourier (Grenoble) 40 (1990),
757-767.
[26] M. Pohst, The minimum discriminant of seventh degree totally real algebraic
number field, in "Number Theory and Algebra (H. Zassenhaus, ed.)," 235-240,
Academic Press, New York, 1977.
[27] A. Scholz, Aufgabe 208, Jahresber. Deutsch. Math.-Verein. 45 (1935), 110;
Losung der Aufgabe 208, ibid. 47 (1937), 47.
[28] R. Schoof, Infinite class field towers of quadratic fields, J. reine angew. Math.
372 (1987), 209-220.
[29] D. Simon, Equations dans les corps de nombres et discriminants minimeaux,
These, Universite Bordeaux I, 1998.
[30) D. Simon, Petits Discriminants de Polynomes Irreductibles, http://www.math.
r
unicaen.fr simon/maths/TableSmallDisc.html.
[31) D. Simon, davantagepolynomes.gp.gz, On demand.
[32) K. Takeuchi, Totaly real algebraic number fields of degree 9 with small discrim-
inant, Saitama Math. J. 17 (1999), 63-85.
[33] K. Uchida, Unramified extensions of quadratic number fields, I, T6hoku Math.
J. 23 (1970), 138-140.
[34) K. Uchida, Unramified extensions of quadratic number fields, II, Tohoku Math.
J. 23 (1970), 220-224.
[35] Y. Yamamoto, On unramified Galois extensions of quadratic number fields, Os-
aka J. Math. 7 (1970), 57-76.
[36] Y. Yamamoto, Divisibility by 16 of class numbers of quadratic fields whose 2-
class groups are cyclic, Osaka J. Math. 21 (1984), 1-22.
[37] K. Yamamura, On unramified Galois extensions of real quadratic number fields,
Osaka J. Math. 23 (1986), 471-486
[38] K. Yamamura, Some analogue of Hilbert's irreducibility theorem and the distri-
bution of algebraic number fields, J. Fac. Sci. Univ. Tokyo Sect. IA 38 (1991),
99-135.
[39) K. Yamamura, Maximal unramified extensions of imaginary quadratic number
fields of small conductors, J. Theor. Nombres Bordeaux 9 (1997),405-448.
[40] K. Yamamura, Maximal unramified extensions of imaginary quadratic number
fields of small conductors. II, J. Theor. Nombres Bordeaux 13 (2001), 633-649.
[41) K. Yamamura, Maximal unramified extensions of real quadratic number fields of
small conductors, in preparation.
DISTRIBUTION OF UNITS
OF AN ALGEBRAIC NUMBER FIELD
Yoshiyuki KITAOKA *
Meijo University,
Tenpaku, Nagoya,
468-8502, Japan
kitaoka~ccmfs.meijo-u.ac.jp
Let F be an algebraic number field with maximal order OF and let of

be the group of all units of F. We assume that of is an infinite group and
we are interested in their distribution. For an integral ideal n of F, we set
E(n) := {u mod n I u E of} (c (oF/n)X) and I(n) := [(oF/n)X : E(n)],
which is equal to the extension degree of the ramified part of the ray class
field corresponding to the ideal n over F. Let K be a Galois extension
of the rational number field Q, which contains the field F, and fix an
element", of Gal(K/Q). We consider a primitive integral polynomial g(x)
such that {9(11) I E of} is a finite group with order 151 , Here we choose
such a polynomial g(x) of minimal degree. Then hex) := (x d - l)/g(x)
is in Z[x] for d := [(77) : (77) n Gal(K/ F)]. Next we take a maximal
natural number 150 such that 6~i19(P) = 1 holds for every E of and
any extension p of 77. Let p be a prime number and ., a prime ideal of F
lying above p such that there is a prime ideal ~ of K with p = ~ n F
whose Frobenius automorphism is 77. Then R(P) := 80h(p)/81 is an integer
dividing 1(.,) and we conjecture that the set of such prime ideals with
1(.,) = f(p) has a positive density K.(77). We will give the conjectural
density K.(77) by the Euler product and study its positivity. Degrees of
prime ideals in question are greater than or equal to 1 and so our density
is a modified form of a natural density.
Partially supported by Grant-in-Aid for Scientific Research (C), The Ministry of Education,
Science, Sports and Culture.
[received: Feb. 21, 2002; accepted in revised form: Nov. 5, 2002]
288 Y. KITAOKA
Notations. For an algebraic nUTI1ber field L, OL, or

denote the ring
of integers and the group of units in L, respectively. dL stands for the
discriminant of L. For an automorphism p of L, we denote the action of p
to a E L by a P and for a polynomial f (x) = anx n + an -1 x n - 1+ ... + ao E
Z[x] we ~f't, af(p) := rr~=o(aak)pk. (n denotes an nth primitive root of
unity. The number of elements of a set S is denoted by is. (a) means
the group generated by a.
Hereafter throughout the paper, we fix finite extension fields F C K
of the rational number field Q and assume that K is a Galois extension
of Q, and that of is an infinite group. Note that we do not assume that
K is a Galois closure of F over Q. We fix an arbitrary automorphism
"., E Gal(K/Q) once and for all. Kn denotes the finite extension field of
K adjoined by all nth roots of all units of K. For an integral ideal n of
F, we set
E(n):= {u mod n I u E of} C (OF/n)X,

I(n) := [(OF/n)X : E(n)].
1.
In this section, we formulate the conjecture.
Lemma 1. Let g(x) be a polynomial in Z[x] such that
W1(g(X)) := {g(T/) I E of}
is a finite group. We fix a primitive polynomial g(x) of minimal de-
gree among them. Then it divides x d - 1 in Z[x] for d := [(".,) : (".,) n
Gal(K/F)).
Proof. By virtue of ".,d E Gal(K/ F), W1(X d - 1) = {1} is clear and
let g(x) be that as in the assertion. Take an integer a and integral
polynomials q(x), r(x) so that a(xd -1) = q(x)g(x)+r(x) and deg r(x) <
degg(x). The assumption #oF = 00 implies degg(x) 2 1. For E of'
we have r(T/) = a(T/ C 1)-q(T/)g(T/) = (g(T/))-q(T/) and hence W 1(r(x)) is a
finite group. The minimality of degg(x) implies r(x) = 0 and then the
primitiveness of both x d - 1 and g(x) yields that a divides q(x). Thus
g(x) divides x d - 1. 0
Hereafter the polynomial g(x) means that defined in Lemma 1 and
put, for d:= [(".,) : (".,) n Gal(K/ F)]
h(x) := (x d -1)/g(x) (E Z[x]), &1:= #W1(g(X)).

Distribution of units 289
Remark. Let Fo be the Galois closure of F over Q. The polynomial
g(x), d = [(7]) : (7]) n Gal(K/ F)] , 81 and hence hex) depend only on Fo
and 7]lFo'
The explicit form of g(x) is complicated, but when K = F is a Galois
extension of Q, it will be given at the end of this section.
Let ~ be a prime ideal of K whose Frobenius automorphism
((K/Q)/~) is 7]j then we say a prime number and a prime ideal of
F lying below ~ correspond to 7] if p f 2dK. By ramification theory, we
know [(7]) : (7]) n Gal(K/F)] = deg(~ n F).
Lemma 2. If a prime number p (f 2dK) corresponds to 7], then 81 divides
h(p).
Proof. Take a unit f. E o~ such that f.9 ('1) is a primitive 81 th root of unity,
and let ~ be a prime ideal of K lying above p such that ((K/Q)/~) = 7],
and we put p = ~ n F. For a generator Q E OF of (OF/P)X, we put
f. == aa mod P (a E Z). Then we have
1= f. 019 ('1) == i 19(P) mod'.J3 == aao 19(p) mod ~,

which implies 1 == a ao19(P) mod p. Since d := [(7]) : (7]) n Gal(K/F)] =
degp, there is an integer b such that a81 g(p) = (pd - l)b and hence
a81 = h(p )b. We have only to show (81 , b) = 1. Suppose that q is a prime
number dividing (81 , b); then we have
f.019(P)/q == Qao19(p)/q == Q(pd-1jb/q == 1 mod p,
which implies f.9 ('1)0l/q == 1 mod~. Since f.9 ('1)01/q is a qth root of unity
in K, if it is not the unity, then ~ I (q - 1 holds and p = q divides 2dK,
which is a contradiction. Hence we have f. 9 ('1)01/q = 1, which contradicts
the definition of 81 . 0
Lemma 3. Let m be a natural number, and let p and p (I p) be a prime
number and a prime ideal of F corresponding to 7]. Then we have
~
mh(p) II( )
p <==:}
{m I 81g(p) and
f. 019 (p)/m== 1 mod p for 'V f. E o~.
Proof. The assertion follows from
mh(p) mh(p) pd - 1 3
~ II(p) <==:} ~-n = #E(p) for n E Z (d:= degp)
<==:} n#E(p) = 81g(p)/m for 3 n E Z

<==:} m I 81g(p) and f. 019 (p)/m == 1 mod p for 'Vf. E o~.
o
290 Y. KITAOKA
Lemma 4. Let m be a natuml number and p a prime ideal of F corre-

sponding to "1 and we take a prime ideal $m of Km lying above p such
that p := ((Km/Q)/$m) is identical with "1 on K. If P t m moreover,
then we have
where p is a prime number lying below p.

Proof. Since 'f/li19(p) ( E OF) is an mth root of unity in Km, the
assumption p t 2mdK and the previous lemma yield
mh(p) II(p) {:::=} {m If19(P) and

01 'f/ 19(p) = 1 for v E of'
noting (n == 1 mod $m implies n I poo for a natural number n. Here the

condition m I 01g(P) follows from the other condition with = 1 and we
complete the proof. 0
We put for a natural number m
Hm("1) := {p E Gal(Km/Q) I PIK = "1 and yi19(p) = 1 for v E of}'
Note that (~g(p) = 1 for P E Hm("1)' By what we have proved,

Proposition 1. The following conditions (i),(ii),(iii) are equivalent for
a natuml number m.
(i) #Hm("1) = [Km : KJ.
(ii) For every P E Gal(Km/Q) which satisfies PIK = "1 we have
~li19(P) = 1 for v E OF'

(iii) Let p and p (I p) be a prime number and a prime ideal of F cor-
responding to "1; if p t 2mdK, then we have
mh(p)j01 II(p).
Lemma 5. If 0' E Gal(Km/ F) satisfies O'IK"1 = "1O'IK, then we have

0'-1 Hm("1)O' = Hm("1).
Proof. Let P be an automorphism in Hm("1) and put p' = O'-lpO'j then
PIK = (O'-lPO')IK = O'IK- l 7]O'IK = 7] is obvious. For E of' we put
-1
1J!j<7 = (1J!j, where ( is an mth root of unity. Then we have
1J!j619(pl) = 1J!ju- 161 9(P)U = (61g(p)u 1J!j6 19(p)u = 61g(p)U( ,\yi19(p)<7 =
1 and hence p' E Hm ("l). 0
Lemma 6. Let KTJ be the fixed subfield of K by ("l). For U E
Gal(Km/KTJ), we have uHm("l)u- 1 = Hm("l).
Proof. Let p be in Hm("l)j then PIK = "l, and the assumption yields
O'IK = ''It = PIK t for some integer t. Hence upu-1lK = "l is clear. For
t
E of' we put 1J!jU = (\liP with an mth root ( of unity. Therefore
we have \li619(upq-1) = \li61u 9(p)U- 1 = 61g(pU-1( \lipt619(P)u-1) =
619(PU- 1 (\li chg(p)i u - 1 = 1. Thus we have shown upu-l E Hm ("l).
o
Proposition 2. Let m be a natural number. Denoting a prime ideal of
F (resp. KTJ) by P (resp. PTJ) and a prime number lying below them by
p, we have
#{p I p :$ x, p f 2mdK, mh(p)/al I J(p), and p corresponds to "l }
= [{u E Gal(K/F) I O'''l = "lO'} : Gal(K/F) n ("l)r 1
x
#{
PTJ I ideal ~m of Km lying above PTJ'
p:$ x, pf 2mdK, Km/KTJ)/~m) E Hm("l) for some}
and degpTJ = 1 .
Proof. By Lemma 4, the left-hand side is equal, denoting a prime ideal
of Km lying above P by ~m. to
#{p I p :$ x, p f 2mdK and Km/Q)/~m) E Hm("l) for 3~m}'
Suppose that for a prime ideal P of F, there is a prime ideal ~m of
Km such that ~m n F = p and Km/Q)/~m) E Hm("l)j if a prime
ideal ~~ of Km also satisfies ~~ n F = p and Km/Q)/~~) E Hm("l),
then we have ~~ n K = (~m n K)U for u E Gal(K/ F) with O'''l = "lu,
since Frobenius automorphisms of ~m n K and ~~ n K are equal to TJ.
If, conversely a prime ideal ~~ of Km satisfies ~~ n K = (~m n K)U
-1
for 0' E Gal(K/F) with O'''l = TJU, then ~~u = ~~ for some K. E
Gal(Km/K) and then Km/Q)/~~) = (K.u)-lKm/Q)/~m)(K.u) E
Hm(TJ) by Lemma 5,6. Hence possible prime ideals ~~ n K of K lying
above P are transformed by 0' E Gal(K/ F) with U"l = "lu and therefore
the above is equal to
[{u E Gal(K/F) I u"l = "lu} : Gal(K/F) n ("l)t 1

x # {~I p:$ x, p f 2mdK and Km/Q)/~m) E Hm("l) }
for 3~m lying above ~ .
292 Y. KITAOKA
Here s,p denotes a prime ideal of K. Put Pt) := s,pm n Kt) = s,p n Kt);
then since the decomposition group of s,p is ('1]), degpt) = 1 and
((Km/Q)/s,pm) = ((Km/Kt))/s,pm) hold and s,p is the unique prime ideal
of K lying above Pt). Therefore we have the assertion. 0
Now Chebotarev's density theorem yields
Theorem 1. Let m be a natural number. By denoting a prime ideal of
F by I' and a prime number lying below it by p, the density of the set
{pI Pf 2mdK, mh(p)/ol 11(1'), and I' corresponds to '1] }
is equal to
[{a E Gal(K/F) 10''1] = '1]0'} : Gal(K/F) n ('1])r 1 [~=7~~]'

Hence we propose the following
Conjecture. Setting, for a natural number 0
"'('1];0) := [{a E Gal(K/F) 10''1] = '1]0'}: Gal(K/F) n ('1])]-1
x f
m=l
J-l(m)#H8m('1]) ,
[K8m: Kt)]
and denoting a prime number and a prime ideal of F by p, I' (I' I p), we
have
#{1' I p::; x, Pf2dK, 1(1') = 0Ih(p)I/01, and l' corresponds to'1]}
'" "'('1]; o)Li(x).
Here J-l( m) denotes the Mobius function. We note that the condition
p < x is used instead of N F /QJp) < x, and so this is a modification of the
usual natural density. This conjecture is a generalization of [CKY], [K]
and hence the conjecture for a real quadratic field K = F is true under
the Generalized Riemann Hypothesis with 0 being 00 in the next section.
When '1] E Gal(K/F), we know degp = [('1]) : ('1]) n Gal(K/F)] = 1 and
hence both ((K/Q)/s,p) = ((K/F)/'iJ) and h(x) = 1 hold. Then the
conjecture is true under G.R.H. by a result of [L]. The situation of [1]
is as follows: Let K / F be a finite Galois extension and let C be a
union of conjugacy classes of Gal(K/ F), and W is a finitely generated
subgroup of FX of finite rank (2: 1) modulo its torsion subgroup, and
k is an integer (> 0). In [L], under G.R.H. it is shown the density of
the set M(F, K, C, W, k) of prime ideals I' of F exists. Here I' is in
M(F, K, C, W, k) if and only if for a prime ideal s,p of K lying above 1',
the Frobenius automorphism Kj F)jl,p) is in C, ordp(w) = 0 for all

wE Wand the index [(OFjp)X : {w mod I' I w E W}] divides k. So the
index is bounded. On the contrary, our case allows that the index J(p)
tends to the infinity.
Proposition 3. Suppose that K = F is a Galois extension of Q; then
the polynomial g(x) is given as follows:
(R) The case where F is real.
(Rl) g(x) = x d- 1 + x d- 2 + ... + 1 if Gal(FjQ) = (ry).
(R2) g(x) = x d - 1 otherwise.
(I) The case where F is imaginary.
We denote the complex conjugation by J.
(11) g(x) = X d- 1 + x d- 2 + ... + 1 if [Gal(FjQ) : (ry)] = 2 and
J fj. (ry).
(12) g(x) = x d - 1 if [Gal(FjQ) : (ry)] > 2 and J fj. (ry).
(13) g(x) = x d/ 2- 1 +Xd / 2 - 2 + ... + 1 if Gal(FjQ) = (ry).
(14) The case of J E (ry) =F Gal(FjQ).
(i) If there is an element u E Gal(FjQ) such that
JuJ-1u- 1 fj. (ry), then g(x) = x d -1.
(ii) If JuJ-1u- 1 E (ry) holds Jor every element u E
Gal(F/Q), theng(x) = (xd/2-l)(Eixai-Eixbi), where
integers ai, bi are defined by
(JuJ-1u- 1 I u E Gal(FjQ)}
= {rya i 10::; ai < dj2} U {rybiJ I 0 ::; bi < dj2}.
In case of (14), g(x) = x d/ 2 - 1 if the maximal real subfield oj F is a
Galois extension over Q.
Proof. Let us define the condition (Al) for an integral polynomial G(x)
by that W1(G(x := {EG(T/) lEE o~} is a finite group. Let i.pl,i.p2,"
be all embeddings of F into the field of complex numbers and as usual
we suppose that F'Pi is in the field of real numbers for 1 ::; i ::; rl and
i.prl +r2+ i = i.prl +i' We remark that rl or r2 = 0, and the relation
(0)
is the unique one for 10 E o~. We note that a unit 10 is a root of unity
if IEul = 1 for all (1 E Gal(FjQ). Suppose that an integral polynomial
294 Y. KITAOKA
G(x) = a(n)xn+a(n-l)xn-l+ .. +a(O) satisfies the condition (AI). We

assume that (a(n), a(n - 1), ,a(O)) = 1, and that G(x) is of minimal
degree. We show that G(x) is equal to the polynomial g(x) up to sign
in the assertion. By the condition (AI), we have
n
(1) La(i) log 11/ i l = 0 for v E o~.
i=Q
We can see easily that in cases except (14), the polynomial g(x) in the
assertion satisfies the condition (AI). The condition (1) is not valid for
any integral polynomial of lower degree than g(x) by the remark above.
Hence the cases except (14) are over. In the rest, we assume J E ('f]) =I-
Gal(F/Q). Hence d = the order of 'f] ~ [F : QJ/2 is clear. We suppose
that n < d and put T := {'f]i I a( i) =I- 0, 0 ~ i ~ n} = Tl U T2 U T 2J.
Here the union is disjoint and T n Tl J = 0. For simplicity, for t E T we
let aCt) stand for a(i) if t = 'f]i (0 ~ i ~ n); then by the assumption on
G(x), we have
for v E o~. We note that a(t) =I- 0 if and only if t E T. If Tl is not

empty or aCt) + a(tJ) =I- 0 for some t E T2, then this relation should be
the relation (0) and hence #Tl + #T2 = [F : Q]j2 holds. On the other
hand, #Tl + 2#T2 = #T ~ n + 1 ~ d ~ [F : Q]j2 = #Tl + #T2 yields
#T2 = 0, and #Tl = d. Then Tl C ('f]) implies Tl = ('f]). By virtue of
Tn T1J = 0 and J E ('f]), it is a contradiction. Hence we have Tl = 0
and aCt) = -a(tJ) for t E T2. Exchanging t and tJ if necessary, we may
assume aCt) > 0 for "It E T2J, and
-G(1/) = IT (tJ / t)a(t) ,
tET2
is a root of unity. Now, we have for u E Gal(F/Q)
L a(t) (log ltJul -log Itul) = O.

tET2
This is not the relation (0) and so it should be a trivial equation and
we can make pairs (tt, t2) E T2 x T2 such that a(tl) = a(t2) and tlJu =
t2uJ, and so we get JuJ-1u- 1 = tl1t2 E (Tf), and the case (i) is over.
Suppose H := (JuJ-1u- 1 I u E Gal(F/Q)) C (Tf); then the relation
t2 = tlJuJ-1u- 1 shows that H acts on the set T2, keeping aCt). Hence
there is a subset A C (TJ) such that
-G(T/) = II (II thJ /th)a(t).

teA heH
It is easy to see that for u E Gal(FjQ) and K := JuJ-1u- 1 E H
(II thJ jth)'U = II (thK'UJ jth'U) = II (th'UJ jth'U)

heH heH heH
is of absolute value 1 and hence each factor IheH(thJ /th) of cG(T/)

is a root of unity. Hence the minimality of degG(x) implies cG(T/) =
ilheH (thJ / th) for some tEA. Hence, putting
tH = {TJAt 10:::; Ai < d/2} U {TJBjJ I 0:::; Bi < d/2}

we have -G(x) = (x d/ 2 -- 1)(:Ei x At - :Ei x Bt ). The minimality of
degG(x) implies Ai or Bi = 0 for some i, which implies t or tJ E H and
hence =FG(T/) = llheH(hJ /h) holds and we can replace tH by H in the
above equation and G(x) is equal to g(x) in the assertion. The last
remark follows from H = {I} if (J) is a normal subgroup. 0
2.
First we show
Theorem 2. Let R be an algebraic number field and E a subgroup of o~.
Suppose that a primitive wth root (w of unity and l, ... ,r generate E,
and {l' ... ,r} is a free basis of E / ((w). Let A be a square-free integer
such that a prime number p divides A if p divides 2dR or if for some
a E R \ E and some E E, a P = holds. Set Rm = R( { l{YE} I E E),
where {l{YE} means all mth roots of . For m = ab with a I AOO and
(b, A) = 1, we have
[Rm : R] = [Rb : R][Ra : R] = br cp(b)[Ra : R] and Ra n ~ = R,

where cp is Euler's junction.
We need several lemmas.
Lemma 7. Setting L = R((mw), we have [L( '\Yl) : L] = b[L(~) : L].
Proof Since [L( '\Yl) : L] = [L(~)( fl) :

L( ~)][L(~) : L], we
have only to show that the polynomial a.: -1 is irreducible over L( %).
Assume that it is reducible; then by virtue of 2 f b, there are a prime
296 Y. KITAOKA
divisor p of b and a E L(~) satisfying a P = l. Since L c L( a) C

L(~) and L(~) I L is a Kummer extension, L( a) I L is abelian. If
a . L holds then x P - 1 is irreducible and hence we have a contradiction
p = [L(a) : L] I [L(~) : L] I a. Thus a E L holds and then R C R(a) C
L = R((mw), and p f A implies that x P - l is irreducible over R and
so R( a) IRis a Galois extension of degree p, because L IRis abelian.
It implies that for a pth primitive root (p, (pa E R and so (pala E R
holds. In the sequence R c R((p) c R(a), we have [R((p) : R] I p-1 and
[R(a) : R] = p. Thus [R((p) : R] = 1, i.e., (p E R occurs. It contradicts
pfdR. 0
Lemma 8. [L( '\Yl, yI2) ; L] = b2[L(~, ~) : L].
Proof. By virtue of
[L( 'V"!, yIf2) : L] = [L( 'V"!, yt2)(~) : L( 'V"!, yt2)]
x [L( 'V"!, yt2) : L( 'V"!)][L( 'V"!) : L],
we have only to show

(i) x b - 2 is irreducible over L( '{Yl, ~),
(ii) [L( '{Yl, ~) : L( '{Yl)] = [L(~, ~) : L( ~)].
Suppose that x b - 2 is reducible over L( '{Yl, ~); then there exist a

prime factor p of b and a E L( '{Yl, ~) satisfying a P = 2. Therefore
the inclusion L( '{Yl) C L( '{Yl, a) C L( '{Yl, ~) implies [L( '{Yl, a) :
L( '{Yl)] I a. On the other hand, [L( '{Yl, a) : L( '{Yl)] I p holds by
aP = 2 ELand (p E L. The assumption (a, b) = 1 yields [L( '\Yl, a) :
L( '{Yi)] = 1, and so a E L( '{Yi). From the property a P(= 2) E L
follows that there is an integer n satisfying al ~ E L. Set pnlm = die
with (e, d) = 1; then we have
(al ~)CP = ~/t E R

and from p f A follows that there is no element j3 in R satisfying j3P =
~/t, in particular (al ~)C . R. Thus we have
R C R((al ~)C) C R(al ~) C L = R((mw),

which yields that R((al ~)C) is a Galois extension of R of degree p,
since any pth root of (al ~)CP = ~/t is not in R. Thus we have
(p E R((al ~)C), and then [R(p) : R] I p - 1 implies [R(p) :
R] = 1. Since p is odd and (p E R, p divides dR, which contradicts
p I b. The ass'ertion (i) has been proven. To show (ii), noting that
[L( '{Yl) : L(~)] = b, [L(~,~) : L(~)] I a and (a,b) = 1, we
have L( \Yl) n L( {!'i, ~) = L( {!'i) and hence L( {!'i) n L(~) =

(L( '{Yl) n L( {!'i, yI2)) n L( yI2) = L( \Yl) n L( yI2). Thus we obtain
[L( ~, yt2) : L( ~)J = [L( yt2) : L( yt2) n L( ~)J

= [L( yt2) : L( yt2) n L( ~)J = [L(~, yt2) : L( ~)].
o
Lemma 9. Let 3:::; t :::; r, and suppose that
[L( '{YEl,'" , yET) : L] = bT[L(~, ... , y'T) : L]
holds for T = t - 1. Then it holds for T = t.

Proof. Set M = L( '{il,'" , '\YEt-d. Since
[L( ~, ... , '\YEt) : L] = [M( '\YEt) : MJ
x bt - I [L( ~, ... , V'Et-l) : L]
= [M ( ytft) (\Vt) : M ( ytft)][ M (ytft) : MJ
x bt-I[L(~,.. , V't-I) : L),
we have only to show

(i) x b - Et is irreducible over M( ytft),
(ii) [M( ytft) : M] = [L( ytl, ... , ytft) : L( ytl, ... , {yEt-I)].
Suppose that x b - Et is reducible over M( ytft); then there exist a prime

factor p of b and an element 0: E M( ytft) satisfying o:P = ft. Then
M c M(o:) c M( ytft) implies [M(o:) : MJ I a because M( efi)/M is
a Kummer extension. On the other hand, [M(o:) : M] I p holds by
(p E M. Thus 0: E M holds. Setting N := L( '{il,'" , '\Yt-2), we have
N c N(o:) c M = N( '\YEt-I) and then there is an integer n satisfying
0:/ o/t-l n EN, since o:P = Et EN. Writing pn/m = d/e with (e, d) = 1,
we have
by p f A. Setting ~-l := U4-1' we can extend {l,'" ,t-2'~-1}

to a basis of E/((w) and then the assumption of induction on t yields
that x P - E~_I is irreducible over L( '{il, ... , '\Yt-2)' Thus we have
[N( f/~-I) : NJ = p. On the other hand, f/~-l = (0:/ ~)C E N
contradicts it and so we have proved (i).
298 Y. KITAOKA
Secondly, since
[L( \YEl,'" , ~Et-I) : L( {IEl,'" , {lEt-I)] = bt - I
and
[L( {IEl, ... , ~) : L( {IEl, ... , {lEt-I)] I a
imply
L( {IEl, ... , ~) n M = L( {IEl, ... , ~) n L( yl, .. , , ~Et_t)

= L( {IEl,'" , {1ft-I),
we have
[M(~): M] = [L(~): L(~)nM]

= [L(~): (L(~) nL({IEl,'" ,~)) n M]
= [L(~) : L( {YEt) n L( {IEl,'" , {1ft-I)]
= [L( V'l," . , {1ft-I, ~) : L( V'l,'" , {1ft-I)]'
Thus the assertion (ii) has be~n proved. o
Lemma 10. [L( {1'1,'" , ~) : R] = cp(b)[R( VB) : R].
Proof. It is easy to see
[L( {IEl, ... , {1Er) : R]

= [R((mw)( V'l" .. , v'Er) : R( VB)][R( VB) : R]
= [R( V'E)((b) : R( V'E)][R( VB) : R]
= [Q((b) : Q((b) n R( VB)][R( V'E) : R].
Since dR( VB) I Aoo and (A,b) = 1, we have Q((b)nR( VB) = Q, which
completes the proof. 0
Now we have
[Rm : R] = [R( \IE) : R]

= [L( \YEl, ... , 'V'r) : R]
= br[L( {IEl,'" ,~) : R]
= cp(b)br[Ra : R] = [Rb : R][Ra : R],
which implies Ra n Rb = R. Thus we have completed the proof of the

theorem.
By virtue of theorem 2, we see for natural numbers 0 = 1'1'Y2, a, b
satisfying 1'1 I (2dK)OO, (1'2,2dK) = 1 and (bf'2,2adK) = 1,
#Hoab(TJ)
= #{p E Gal(Koab/Q) I PIK = TJ, 6~019(P) = 1 for v E o~}
PIK = TJ, "Yu:;t19(P) = 1 and }
= # { P E Gal(Koab/Q) "Y')b{;019(P) _ 1 v E x
-y - or of
= #{p E Gal(K,,),la/Q) I PIK = TJ, "Ylf/i19 (P) = 1 for v E O~}
x #{p E Gal(K,,),2b/Q) I PIK = TJ, "Y2ifi19 (p) = 1 for v E O~}
= #H")'la(TJ)#H")'2b(TJ),
which yields
[{O" E Gal(K/F) I O"TJ = TJO"} : Gal(K/F) n (TJ)]K(7]jO)

= ' " J1.(a)#H")'la(7]) x ' " J1.(b)#H")'2b(TJ)
L..J
al2dK [K 'K]
")'la 7J L..J
(b,2dK)=1 [K"Y2'
b' K ]
= '"' J1.(a)#H")'la(TJ) x
L..J [K K 1
II (1 _ #Hp(TJ) )
~r(K) (~) ,
al2dK ")'la: 7J pt2dK P <P\.Y
where r(K) := dimQ(oK z Q) and p := pllp(")'2)+1 for IIp (1'2) defined by

pllp ("Y2) 1I'Y2.
The evaluation of #Hp(TJ) is not easy. However, the following assures
the absolute convergence of K(TJj 0).
Proposition 4. For almost all prime numbers p we have
#Hp(TJ) < degg(x) .

pr(K)<p(p) - p(p - 1)
Proof. Let q be a natural number and U a subgroup of OK generated by

7J m ( E o~, m E Z). Denote by K' an extension field of K adjoined
by all qth roots of all elements of U. Any extension p of TJ induces an
automorphism of K', since TJ(U) = U. Let p E Aut(K') be an extension
of TJ. If (!fif19(P) = 1 for vuE U holds, then ~019(P) = 1 for v E
o~ holds clearly. Conversely suppose ~019(P) = 1 for v E O~j then
1 = ffi 019(P)pm = ffipm019(p) = ~7JT;,019(P) holds, because of <:19(P) = 1.
Hence ~019(P) = 1 for vf E o~ holds if and only if ~fii19(P) = 1 for
300 Y. KITAOKA
VUE U holds and then

#Hq(rJ) = [K' : Kr 1
[Kq:K]
x #{p E Aut(K') I PIK = rJ, ~/i/ilg(P) = 1 for Vu E U}.
Let W = ((w) be a set of all roots of unity in U and let {U1,'" ,us} C U
induce a basis of U jW, and define a unimodular matrix A = (aij) by
u"! = /,Bi
l'oW
IT u~ij.
J
j
Then it is easy to see g(A) = O. Let P E Aut(K') be an automorphism

of K' with PIK = rJ and set
/'P -
'oqw -
/,c
'oqw, V
q~u'p
"'l
_/,bi
- 'oqw
IT qr:i'i"7uaij
V ~J .
j
We note that c mod w is determined by (:1 = (~, and bi == Bi mod w is

clear. Putting At = (aij(t)), we can see easily by induction on t (~ 0)
~P
t "t-l
= (~k=O C
t-k-l " . b
LJj=1 jaij
(k)
II vtu'iau(t).
I.
Then setting g(x) = gnxn + ... + 90, we have
q~81g(p) _ /,81 L~=o gt Lt:1 Ct - k - 1 LJ=1 bjaij(k)
VUt - 'oqw ,
by virtue of 9(A) = O. Therefore {Jfil 19 (P) = 1 holds for Vu E U if and

only if
819(c) == 0 mod qw (in the case of u = (w),
n t-1 8
81 L 9t L ct- k - 1 L bjaij(k) == 0 mod qw (in the case of U = Ui).

t=O k=O j=l
Put
n t-l
9A(C):= L9tLCt - k- 1A k and b:= (b1,'" ,bs )
t=O k=O
Then we have shown

#{p E Aut(K') I PIK = rJ, .~/"il19(P) = 1 for Vu E U}
819(C) == 0 mod qw, }
= # { c mod qw, b mod qw 819A(C)tb == 0 mod qw
and the extension condition
where the extension condition means that the extension automorphism

p of rJ exists subject to the condition (*). Next, let us show that the min-
imal polynomial of A is equal to g(x). g(A) = 0 is known. Suppose that
G(x) is the minimal polynomial of A and we write G(x) = Ei Gi(X)Wi
where Gi(x) E Z[x] and {Wi} is a basis ofC over Qjthen G(A) = 0 implies
Gi(A) = 0 and hence {uGi(T}) I u E U} is a finite group and the minimal-
ity of degg(x) yields deg Gi(x) ~ degg(x) and so deg G(x) ~ degg(x).
Thus g(x) is the minimal polynomial of A. Under these let us prove the
proposition. Assume that q is a prime number p which does not divide
2d dK and that if u E U is a pth power in o~, then u is a pth power
in U. Then by virtue of theorem 2, we have [K' : K] = pS(p - 1) and
there exists an extension p of rJ defined by (*) for any integers e, bi which
satisfy (:1, = ,~, bi == Bi mod w. Suppose that p satisfies f/U6 19 (P) = 1
for Vu E U and let e, bi correspond to pj then gee) == 0 mod p and
gA(e)tb == 0 mod p hold, since (h I (2dKYX> and p f 2dK. We can take
a non-zero vector e E ZS such that e A == ee mod p because the min-
imal polynomial g(x) of A divides the characteristic polynomial. Then
e gA(e) == g'(e)e mod p yields e gA(e)tb == g'(e)e tb == 0 mod p. Since
g(x) divides x d -1 = g(x)h(x), g'ee) " 0 mod p holds by virtue of p f d.
Thus the equation g'(e)e tb == 0 mod p is non-trivial and then the num-
ber of extension p with a fixed e is less than or equal to ps-l, which
completes the proof. 0
Lemma 11. There exists the maximal natural number 80 such that
Proof. Since the condition #HN(rJ) = [KN : K] implies #Ha(rJ) = [Ka :

K] for every divisor a of N, it is easy to see for (a, b) = 1 #Hab(rJ) =
[Kab : K] holds if and only if #Ha(rJ) = [Ka : K] and #Hb(rJ) = [Kb : K]
hold. First suppose #Hq(rJ) = [Kq : K] for a prime number qj then we
have for r = r(K), qr(q - 1) = [Kq : K] = #Hq(rJ) ::; (degg(x))qr-l if
q does not divide some integer by theorem 2 and proposition 4. Thus
the condition #H6(rJ) = [K6 : K] yields that prime divisors of 8 are
bounded. Next, we suppose that #Hqn(rJ) = [Kqn : K] holds for an
integer n and a fixed prime number q. We apply Lemma 4 to m = qn.
Let I.l3 be a prime ideal of K such that I.l3 t 2qdK, ((K/Q)/I.l3) = ", and
g(p) ::/= 0 for a prime number p lying below 1.l3. I.l3 is independent of n. For
a prime ideal ~qn of Kqn lying above ~, we put p = ((Kqn,Q)/~qn).
Then Lemma 4 implies qn h(p)/O'1 I I(p) for p = ~ n F and hence by
Lemma 3 we have qn 1O'1g(p) and so nlogq::; log IOlg(p)l. D
302 Y. KITAOKA
Remark. If a prime number p divides 6o, then p divides 2d dK or

[L n QM : Mj, where L is a Z-lattice o~/{roots of unity in K} and M
is a submodule of L corresponding to (TJ t I t E Z, E o~).
To study the positivity of ~("'; 80), we need the following well-known
fact.
Lemma 12. Let S be a finite set and Ai (i = 1,2"" ,n) a subset of S.
For a subset A of S, we set P(A) = #A/#S. Then we have
n
P(U~IAi) = 2) _1)k+l L P(A i1 n n A ik )
k=1 l:Sil <i2<"-<ik:Sn
Proposition 5. Let 6 = ,1')'2 with,l I (2dK)OO, (,2,2dK) = 1. ~("'; 6)

is positive if and only if both #Hp("') =f fir(K)'{)(fi) holds forfi:= pVp(-Y2)+1
in case of p f 2dK and there is an automorphism P such that PIK"fl E
Hn ("') and PIKYlq is not in H'nq("') for every prime divisor q of 2dK.
Proof. By virtue of what we have shown, we have only to study
Put S := Gal(K2'Y1dK / KTJ) and A(q) := {p E S I PIKYlq E HYlq(",n for a

prime number q. Then for a square-free integer a, we see
nqlaA(q) = {p E S I PIK = "', 'Y~619(P) = 1 for '<I E 0;" and '<I q I a}

= {p E S I PIK'Yl a E H"Y1a(",n
Therefore we have
and using Lemma 12, we see
L J.L(a)#HYla (",)
al2dK [K'Yla: KTJ]
= [!:~l ~~l - #(UqI2dKA(q))/#S

= [K21'1dK : KTJrl {#{p E S I PIK = "', 'Y-Vi 19 (P) = 1 for 'V E o~}
- #{p E S I PIK = "', 'Y~819(P) = 1 for 'V E o~ holds for 3q I 2dK} } .

Thus we have shown the assertion. o

Theorem 3. K( 'fJj 00) is positive.
Proof. We define integers np by pnp 1/00 for prime divisors p of 2dK. We
have only to show that there exist a unit up EO;' and an extension Pp of 'fJ
such that [K( pnp+VU'P) : K] is a power of p and pnp+~19(Pp) =I- 1, since
fields K( pnp+VU'P) are linearly disjoint over K and hence the composite
p of Pp exists and p H6op('fJ) for any prime p 12dK. By the maximality
of 00, there exist a unit up E 0;' and an extension Pp of 'fJ such that
pn p+.vru;619 (PP) =I- 1. If up is a pth power of v E F, then pnW19(PP) =I- 1
and hence pn p+;,/i/19(PP) =I- 1. Repeating this, we may assume up is not a

pth power in F. Suppose up = vP for 3 v E K j then xP - up is irreducible
over F and it is reducible over K. Since K / F is a Galois extension, it
yields (p E K and therefore [K( pnp+.vru;) : K] is a power of p. If up is
not a pth power in K, then x pnp +1 - up is irreducible over K in the case
of p =I- 2 and hence we have [K( pnp+VU'P) : K] is a power of p. In case of
p = 2, it is clear. Thus we have completed the proof. 0
References
[CKY] Y-M. J. Chen, Y. Kitaoka and J. Yu, Distribution of units of real quadratic
number fields, Nagoya Math. J. 158 (2000), 167-184
[H] C. Hooley, On Artin's Conjecture, J. reine angew. Math. 225 (1967), 209-
220.
[IK] M. Ishikawa and Y. Kitaoka, On the distribution of units modulo prime ideals
in real quadratic fields, J. reine angew. Math. 494 (1998), 65-72.
[K] Y. Kitaoka, Distribution of units of a cubic field with negative discriminant,
J. of Number Theory 91 (2001),318-355.
[L) H. W. Lenstra, Jr., On Artin's conjecture and Euclid' algOrithm in global
fields, Inventiones math. 42 (1977), 201-224.
[M] K. Masima, On the distribution of units in the residue class field of real
quadratic fields and Artin's conjecture (in Japanese), RIMS Kokyuroku 1026
(1998), 156-166.
[R] H. Roskam, A quadratic analogue of Artin's conjecture on primitive roots,
J. of Number Theory 81 (2000), 93-109.
ON CAPITULATION PROBLEM
FOR 3-MANIFOLDS
Dedicated to Professor Yasu.o Morita
Masanori MORISHITA
Kanazawa University,
Kakuma-machi, Kanazawa, Ishikawa,
920-1192 Japan
morisita<Dkenroku.kanazawa-u.ac.jp
Introduction
The purpose of this paper is to investigate analogues for 3-manifolds
after the model of capitulation theorems for number fields. This will
give one of the foundational analogies between 3-dimensional topology
and algebraic number theory.
The capitulation problem deals with the structure of ideal classes
which capitulate in an extension of number fields, and it is formulated
as a purely group-theoretic problem on the transfer by means of the Artin
reciprocity map. In recent years, H. Suzuki ([Sul,2]) has established a
fundamental theorem on this group-theoretic version which includes the
three classical capitulation theorems, Hilbert's theorem 94, the principal
ideal theorem and Tannaka-Terada's theorem.
In this paper, we shall consider a similar problem for homology classes
in a covering of manifolds. Since the problem is again translated into a
similar problem on the group-transfer by means of the Hurewicz isomor-
phism, we obtain a topological analog of Suzuki's theorem (Theorem 1.3
below). For a Galois covering of 3-manifolds, using the spectral sequence
and Poincare duality, we obtain more precise results on capitulation,
analogous to the theorems by Iwasawa, Chevalley-Tate, and Suzuki for
number fields (Theorem 2.3, Corollaries 2.4, 2.6 below). We note finally
that some analogues of capitualtion theorems for coverings of topological
spaces were also studied in the recent Master thesis of K. Fuluta ([Fun.
[received: May 24, 2002; accepted in revised form: Oct. 18,2002]

306 M. MORISHITA
1. Capitulation problem
For a number field F,* we denote by Hp the ideal class group of F.
For an extension K/k of number fields, the capitulation problem deals
with the structure of the finite abelian group
CK/k := Ker(tK/k : Hk - - HK)

where tK/k is the natural homomorphism induced by the extension of
an ideal under the natural inclusion k - t K. For a useful survey on this
topic, we refer to [MiJ. A fundamental theorem is due to H. Suzuki.
Theorem 1.1 ([Su1,2]). Let klko be a finite cyclic extension of numer
fields of Galois group Go and let K I k be an unramified extension so that
K is abelian over ko. Then the number of Go-invariant ideal classes in
C K / k is divisible by the degree [K: kJ of Kover k.
We note that Suzuki's theorem includes three classical theorems,
Hilbert's theorem 94 ([Hi]) for ko = k and cyclic K/k, the principal
ideal theorem ([F]) for case that ko = k and K is the Hilbert class field
of k, and finally Tannaka-Terada's theorem ([T]) for case that K is the
genus field of k / ko
Since these capitulation theorems are translated into purely group the-
oretical statements by means of the Artin reciprocity map, we have their
analogues for a wide class of coverings of topological spaces by means of
the Hurewicz isomorphism. Let p : N - t M be a finite abelian cover-
ing of path-connected topological manifolds and let tN/M : Hl (M, Z) - t
Hl(N,Z) be the transfer map defined by tN/M([C]) = [P-l(C)]. We set
CN/ M := Ker(tN/M : Hl(M, Z) -t Hl(N, Z))
and we say that a subgroup A of Hl (M, Z) capitulates in N if A c
CN / M. Let Ma and Na be the maximal abelian coverings of M and N
respectively, and set r = Gal(NajM) and H = Gal(NajN). We write
g' for the commutator subgroup of a group 9 throughout this paper.
Consider the diagram
r/r'
f'V
Hl(M,Z) --~)
tN/M 1 f'V
1 Vr/H
Hl(N,Z) ) HIH' = H
*In this paper, we sha.1l mean by a number field a finite extension over the rational number
field Q, although the capitulation problem makes sense for any algebraic extensions over Q.
where the horizontal maps are Hurewicz isomorphisms and Vr / H is the

group-transfer. Basically, by applying the group-theoretical versions
of capitulation theorems for number fields to the above diagram, we
have analogous statements for CNIM However, for a topological setting
above, the groups r and HI H' are possibly infinite (cf. Examples 1.5
below), while r is assumed to be finite in the group-theoretic version
of Theorem 1.1 ([Su2), Th. 2) for the application to number fields. So,
we give here a bit generalized form of Theorem 2 in [Su2], of which was
informed by H. Suzuki, for our purpose.
Theorem 1.2. Let r be a finitely generated group such that r Ir' is
finite. Let H be a normal subgroup of r with finite index in rand
containing r'. Let u be an automorphism ofr such that u(H) cHand
that u induces the identity map on G = r I H. Then the order of the
subgroup
{-yr' E Ker(Vr / H) I u(r){-l E r'}
of the transfer kernel Ker(Vr / H) is divisible by #G = [r : H].
Proof. (due to H. Suzuki). Since r Ir' is finite, we have vr/H(r Ir') c
(HI H')tor where A tor denotes the torsion part of an abelian group A.
Take a positive integer m so that m is a multiple of the exponents of
r/r' and (HIH')tor. Set M:= HmH'. Then we see that M is a normal
subgroup of r with finite index and is contained in r'. Consider the
following commutative diagram:
Vr / H : r Ir' - - - - + ) (HI H')tor
I
~r/M)/(H/M) : (rIM)(rIM)'
1
) HIM.
Since MIH' n (HIH')tor = {O} by our choice of m, the right vertical

map is injective and so we find Ker(Vr / H ) = Ker(Vr/HmodM). Noting
that M is stable under the action of u, we can reduce our assertion to
the case r is finite. 0
Theorem 1.2 yields immediately the following topological counterpart
of Suzuki's theorem 1.1.
Theorem 1.3. Let M -+ Mo be a finite cyclic branched covering of
path-connected compact manifolds with Galois group Go and let N -+ M
be a finite covering so that the composite N -+ Mo is an abelian covering.
Assume that H1(M, Z) is finite. Then the number of Go-invariant 1st
308 M. MORISHITA
homology classes in CN / M is divisible by the degree [N : M] of N over

M.
Remark 1.4. In [M05], we introduced the notion of the genus covering
of an abelian covering of a link in the 3-sphere. However, this notion
is purely group-theoretically defined and so we can define the genus
covering of an abelian Mj1vIo to be the maximal abelian subcovering
of Ma over Mo. Then we have a topological Tannaka-Terada theorem:
any Go-invariant 1st homology class of M becomes trivial in the genus
covering of MjMo.
Example 1.5. Let M be a 3m-fold cyclic covering of 8 3 ramified over
the figure eight knot, m ~ 1. It was shown in [He2] that Hl(M, Z) is
finite and M has an abelian covering N of degree 4 such that H1 (N, Z) ~
Hl(M, Z) EB Z3 modulo 2-torsion. By Th. 1.3, #CN / M is divisible by 4.
Of course, if Hl (M, Z) is finite and Hl (N, Z) is free with rank ~ 1,
then the whole Hl(M, Z) capitulates in N. J. Hempel informed me of
the following such examples: Let M be the union of 2 copies of twisted
I-bundles (cf. [He1]) along a common boundary which is an oriented
compact surface E. Then we have a double covering N = E X 8 1 ~ M
with finite H1 (M, Z) and infinite and free H1 (N, Z). A particular case
is that M = p1#p1(connected sum) and N = 8 2 X 8 1.
2. Capitulation theorems for 3-manifolds

For coverings of 3-manifolds, we can derive more precise information
on the capitualtion problem using the Poincare duality and the coho-
mology spectral sequence. We find that a similar cohomology spectral
sequence for number fields leads a simple proof of a classical theorem
by Iwasawa ([Il]) on CK / k . This would deepen our underlying philos-
ophy, namely, analogies between 3-manifolds and number fields. Here
is a part of a basic dictionary of this analogy which we shall employ
in the following. Hereafter, we shall mean by a 3-manifold an oriented,
connected, closed 3-dimensional manifold, and we shall write Ok for the
ring of integers of a number field k.
3-manifold M ~ 83 ~ number field kj Q
knot, link C M ~ prime ideal(# 0), ideal C Ok

end infinite prime
surface E c M
E I--> BE
chain complex C.(M, Z) kX -+ E9 Z

p:primes
Hk = (ideal class group of k)
Hurewicz isomorphism U nramified classfield theory

H 1 (M, Z) ~ Gal(Ma jM) +---+ Hk ~ Gal(ka jk)
(Ma = max. abel cover of M) (k = Hilbert classfield of k)
a
(free part of) Ok

For more and precise analogies between knots and primes, 3-manifolds
and number fields, we refer to [Kj, [Mal,2j, [MolIV7j, [Raj, [Rl,2], [Sij,
[W].
Now, we shall firstly give a simple proof of the following well-known
theorem by Iwasawa, which relates CK / k with the Galois cohomology of
the unit group, by using cohomology spectral sequence.
Theorem 2.1 ([11]). For a finite unramified Galois extension Kjk of
number fields with Galois group G, we have
where OK is the unit group of the ring of integers of K.
Proof. Consider the Hochschild-Serre spectral sequence associated to the

Galois etale covering Spec(OK) -+ Spec(Ok)
E~,j = Hi(G, H~t(Spec(OK), G m )) ==> H!t j (Spec(Ok), G m )
with the associated differentials cf,.,j : E;,j -+ E;+r,j-r+1 and the filtra-
tion Hk = H!t(Spec(Ok), G m ) = Fo :J Fl = E;,oD. Then we see easily
that
Hk/Fl = E~l ~ ... ~ E~,l = Ker(c4,l) C Eg,l = no(G,HK) c Hk,
F1 -- El,o
00
IV
-
' "
-
E 21,0 -- P\X) .
Hl(G ,vK
o
When K/k is cyclic, the theorem by Chevalley and Tate on the Her-
brand quotient of the unit group ([AT], Ch 5, 3) yields the following
precise version of Hilbert's theorem 94.
310 M. MORISHITA
Theorem 2.2 (Chevalley-Tate). For a finite unramified cyclic exten-

sion K / k of number fields, we have
where HU denotes the Tate cohomology.

Now, let p : N ---t M be a finite Galois covering of 3-manifolds with
Galois group G. By the Poincare duality, we have
Then we can derive the following information on CN / M = Ker(p*) as in

our proof of Iwasawa's theorem 2.1 from the Hochschild-Serre spectral
sequence
E~,j = Hi(G,Hj(N,Z)) ==> Hi+j(M,Z).
Theorem 2.3. Notation being as above, we have the following exact

sequence
o ~ COk(~,l) ~ CN / M ~ Ker(d~,l) ~ 0
where ~,l : HO(G, Hl(N, Z)) ---t H2(G, Z) and d~,l : Hl(G, Hl(N,
Z)) ---t H3 (G, Z) are the differentials attached to the spectral sequence.
In particular, CN / M is a finite abelian group.
Proof. We let
-li,j .. Ei,j
Ur r
~ Ei+r,j-r+l
r
be the differentials associated to the spectral sequence and let

H2(M, Z) = Fo :> Fl :> F2 ...
be the filtration of H2(M,Z) such that Fi/Fi+l :::: E;';;-i. Then we see
that
(1) H2(M, Z)/ Fl = Ef(.l :::: ... :::: E~,2 :::: Ker(~,2) C E~,2 :::: Ker(~,2) C
E~,2 = HO(G, H2(N, Z)),
(2) FdF2 = E~~/:::: ... :::: Ej,l = Ker(d~,l) C E~,l = Hl(G,Hl(N,Z)),
(3) F2 = E~o :::: ... :::: Ei'o = Ker(d~,o)/Im(~,l) = H2(G, Z)/Im(~,l).
By (1), we have CN / M :::: Fl. Hence, by (2), (3), we get the assertion. 0
Recall that a 3-manifold M is called a rational homology 3-sphere if
Hi(M,Q) = Hi (S3,Q) for all i, which is equivalent to H 1 (M,Q) = 0
(or H1 (M,Z) being finite) by the Poincare duality. We see easily by

the Hochschild-Serre spectral sequence that M is a rational homology
3-sphere if a finite Galois covering space N of M is so. The following is
seen as a more precise analog of Suzuki's theorem for 3-manifolds.
Corollary 2.4. If N is a rational homology 3-sphere, then we have
CN / M ~ G/G' (non-canonical).
Hence, #CN / M is equal to the degree of the maximal abelian subcovering
of N over M.
Proof. For a rational homology 3-sphere N, Hl(N, Z) = Hom(H I (N, Z)
,Z) = 0 and so Ker( d~,l) = 0 in Th. 2.3. Hence by Th. 2.3 again,
CN / M ~ H2(G,Z) ~ Hom(GjG', QjZ). 0
Remark 2.5. Suppose that M and its maximal abelian covering Ma
are rational homology 3-spheres. Then Cor. 2.4 tells us that HI (M, Z)
capitulates in Ma and canllot capitulate already in a proper subcovering
N of Ma. In fact, if HI (M, Z) capitulates in a sub covering N of Ma, by
Cor. 2.4, we have [N : MJ = #CN / M = #HI (M, Z) = [Ma : MJ, hence
N = Ma. On the other hand, it can happen that all ideal classes of
a number field capitulate in a proper subextension of the Hilbert class
field (e.g, [12]).
Corollary 2.6. If M is a rational homology 3-sphere and G is cyclic,
we have the equality
#CN / M = [N: MJh(H2(N, Z))
where h(H2(N, Z)) := nf=o #Hi(G, H2(N, Z))(-I)i+l is an integer.

Proof. Since G is cyclic, we have H3(G, Z) = Hl(G, Z) = 0 and so
Ker(d~,I) = Hl(G, Hl(N, Z)) = HI(G, H 2(N, Z)) by the Poincare dual-
ity. Since HI(M,Z) = Hom(HI(M,Z),Z) = 0 by the Poincare duality,
a piece of the spectral sequence shows the injectivity of ~,I. Hence we
have the equality in our assertion by Th. 2.3. By Th. 1.3, #CN / M is
divisible by [N: MJ and so h(H2(N, Z)) is an integer. 0
Since H2(N, Z) is an analog of the unit group O~, Cor. 2.6 has a
flavour analogous to Th. 2.2 by Chevalley and Tate.
Example 2.7. Let M be the lens space L(m, n) which has the funda-
mental group ZjmZ. Hence, for any cyclic covering N - t M of degree d,
a divisor of m, we have CN / M ~ ZjdZ by Cor. 2.4. Suppose m is even.
Then M is the double cover of 8 3 ramified over the 2-bridge link B(m, n)
312 M. MORISHITA
([BZl, Ch.12). By [Mo5l, the genus covering M9 of Mover 8 3 is the

double cover L(m/2,n) of M and #H1(M,Z)Go = 2, Go = Gal(M/83 ).
Hence, by 1.4, we have CM9 / M = H1 (M, Z)Go.
Acknowledgments
I thank M. Furuta for informing me of the method to compute CN / M
by the Poincare duality and spectral sequence. Many thanks go to
J. Hempel and H. Suzuki for useful communications and to H. Sato for
pointing out a mistake in the first manuscript. I thank K. Hashimoto,
K. Miyake and H. Nakamura for giving me an opportunity to join the
conference "Galois theory and Modular forms" held at Tokyo Metropoli-
tan University, Sep. 25-29, 2001. This work was partially supported by
the Grants-in-Aid for Scientific Research (C) (No.13640014), Ministry
of Education, Culture, Sports, Science and Technology, Japan.
References
[AT] E. Artin, J. Tate, "Class field theory," 2nd edition. Advanced Book Classics.
Addison-Wesley Publishing Company, Advanced Book Program, Redwood
City, CA, 1990.
[BZ] G. Burde, H. Zieschang, "Knots," de Gruyter Studies in Math. 5, Walter de
Gruyter and Co., Berlin, 1985.
[F] Ph. Furtwangler, Beweis des Hauptidealsatzes fur Klassenkorper algebraischer
Zahlkorper, Abh. Math. Sem. Univ. Hamburg 7 (1930), 14-36.
[Fu] K. Fuluta, On Hilbert's theorem 94 for covering spaces, Master thesis at To-
hoku University (Japanese), 2002.
[Hi] D. Hilbert, Bericht: Die Theorie der algebraischen Zahlkorper, Jer. dt. Math.-
Ver.4 (1897),175-546; Gesam, Abh. 1., 63-363.
[He1] J. Hempel, "3-manifolds;' Ann. of Math. Studies, No. 86, Princeton Univ.
Press, 1976.
[He2] J. Hempel, Homology of coverings, Pacific J. Math. 112 (1984), 83-113.
[11] K. Iwasawa, A note on the group of units of an algebraic number field, J.
Math. pures et appl. (9) 35 (1956), 189-192.
[12] K. Iwasawa, A note on capitulation problem for number fields, Proc. Japan
Academy 65 (1989), 59-61; II, ibid, No.6, (1989), 183-186.
[K] M. Kapranov, Analogies between number fields and 3-manifolds, (unpublished
note).
[Mal] B. Mazur, unpublished mimeographed note, circa, 1965.
[Ma2] B. Mazur, The theme ofp-adic variation, Mathematics: frontiers and perspec-
tives, 433-459, Amer. Math. Soc., Providence, RI, (2000).
[Mi] K. Miyake, Algebraic investigation of Hilbert's theorem 94, the principal ideal
theorem and the capitulation problem, Expo. Math. 7 (1989), 289-346.
[Mol] M. Morishita, Milnor's &ink invariants attached to certain Galois groups over
Q, Proc. Japan Academy 76 (2000), 18-21.
[Mo2] M. Morishita, On certain analogies between knots and primes, J. reine und
angew. Math. 550 (2002), 141-167.
[M03] M. Morishita, Knots and primes, 3-manifolds and number fields, Algebraic
number theory and related topics (Japanese) (Kyoto, 2000, Dec.), RIMS re-
port, 1200, (2001), 103-115.
[M04] M. Morishita, Knots and primes, 3-manifolds and number fields, A confer-
ence report "Art of Low Dimensional Topology VII," Kansai Seminar House,
February, (2001),99-109 .
[Mo5] M. Morishita, A theory of genera for cyclic coverings of links, Proc. Japan
Academy 77 (2001), 115-118. .
[Mo6] M. Morishita, Analogies between knots and primes, 3-manifolds and number
fields, 2001 (submitted).
[Mo7] M. Morishita, Milnor invariants and Massey products for prime numbers,
Compositio Math. (2003), (to appear).
[Ra] N. Ramachandran, A note on arithmetic topology, C. R. Math. Acad. Sci.
Soc. R. Can. 23 (2001), 130-135.
[R1] A. Reznikov, Three-manifolds class field theory (Homology of coverings for a
nonvirtually bl-positive manifolti), Sel. math. New ser. 3 (1997), 361-399.
[R2] A. Reznikov, Embedded incompressible surfaces and homology of ramified cov-
erings of three-manifolds, Sel. math. New ser. 6 (2000), 1-39.
lSi] A. Sikora, Analogies between group actions on 3-manifolds and number fields,
(2001), (to appear).
[Su1] H. Suzuki, A generalization of Hilbert's theorem 94, Nagoya Math. J. 121
(1991), 161-169.
[Su2] H. Suzuki, On the capitulation problem, Class field theory-its centenary and
prospect (Tokyo, 1998), Adv. Stud. Pure Math., 30, Math. Soc. Japan, Tokyo,
(2001), 483-507.
[T] F. Terada, On a generalization of the principal ideal theorem, Tohoku Math.
J. (2) 1 (1950), 229-269.
[W] J.-L. Waldspurger, Entrelacements sur Spec(Z), Bull. Sc. math. 100 (1976),
113-139.
ON THE IWASAWA It-INVARIANT

OF THE CYCLOTOMIC Zp-EXTENSION
OF CERTAIN QUARTIC FIELDS
Tsuyoshi ITOH
Department of Mathematical Sciences,
School of Science and Engineering
Waseda University,
9-4-1 Okubo, Shinjuku, Tokyo,
169-8555, Japan
tSitoh<Daoni.waseda.jp
Abstract We give a criterion of vanishing of the Iwasawa j.t-invariant of the cyclo-

tomic :la-extension of certain quartic fields.
1. Introduction.
Let Q be the field of rational numbers and p a prime number. We
denote by Qoo the cyclotomic Zp-extension of Q, that is, the unique
Galois extension of Q whose Galois group is isomorphic to the additive
group of the p-adic integer ring Zp. Let Qn be the unique subfield of Qoo
whose degree over Q is pn. Moreover, for any algebraic number field k,
we put koo = kQoo and k n = kQn. The field koo is called the cyclotomic
Zp-extension of k.
Let A(kn ) be the Sylow p-subgroup of the ideal class group of kn .
For a finite set 8, we denote the order of 8 by 181. According to the
Iwasawa theory, there exist non-negative integers A, J.t and an integer v
which depend only on k and p such that
for all sufficiently large n.
[received: June 1, 2002; accepted in revised form: October, 29, 2002)

316 T.ITOH
The determination of the invariants A, f-l and 1/ for any given number
field k and any prime p is an important problem in the theory of Zp-
extensions. In particular, the following conjecture is well-known:
Conjecture (Iwasawa [6]). The f-l-invariant of the cyclotomic Zp-
extension vanishes for every number field k and every prime p.
Concerning this conjecture, the following result is known.
Theorem (Ferrero-Washington [4]). If k is an abelian extension
field over Q, then the f-l-invariant of the cyclotomic Zp-extension of k
vanishes for every prime p.
But the above conjecture is not completely solved until now. We shall
deal with the case where the base field is a certain non-Galois quartic
extension over Q. Let F be a real quadratic field which is embedded
in the field of real numbers. Taking an element a of F which satisfies
a > 0 and a' < 0, where a' is the conjugate of a, we put M = F( fo).
Then M is a non-Galois quartic field. Therefore it is to be questioned
whether the f-l-invariant of the cyclotomic Zp-extension of M vanishes or
not.
But in the case where p = 2, we can show the following result:
Proposition 1. The f-l-invariant of the cyclotomic Z2-extension of M
vanishes.
Proof. It is easy to see that the Galois closure M* of Mover Q is a
dihedral extension over Q of degree 8 and M* contains an imaginary
quadratic field k. By the theorem of Ferrero-Washington, the f-l-invariant
of the cyclotomic Z2-extension of k vanishes. Since M* / k is a quartic
extension, Iwasawa's theorem ([6, Theorem 3]) says that the f-l-invariant
of the cyclotomic Z2-extension of M* vanishes. Thus the proposition
follows. 0
Therefore, the case p > 2 is to be treated. In this paper, we shall
give a criterion for vanishing of the f-l-invariant of the cyclotomic Z3-
extension. Our method is based on the work of Kraft and Schoof [8]
about Greenberg's conjecture for real quadratic fields.
Theorem 2. Let F be a real quadratic number field and a an element
of F which satisfies the condition that a > 0 and a' < 0 (a' denotes
the conjugate of a). We put M = F(fo). Moreover, let Moo be the
cyclotomic Z3-extension of M and Mn the n-th layer of Moo/M. We
assume that every prime of F above 3 does not split in M. We fix a
generator I of Gal(MI/M). Let 'TIl = X h (1, Gl ) be the Stark-Shintani
unit of Ml which will be defined later.
On the Iwasawa {t-invariant of the cyclotomic Zp-extension 317
If 1}~-1 is not a cube in M 1 , the {t-invariant of the cyclotomic Zg-

extension of M vanishes.
The Stark-Shintani units are special ones in a certain abelian exten-
sion field of a real quadratic field. The general existence of these units
is not known (Stark-Shintani conjecture, see Stark [13] , Shintani [12]).
But in the case where we treat in this paper, the Stark-Shintani units
exist in every layer Mn of the cyclotomic Zp-extension of M and these
units have behave themselves like the cyclotomic units in a real abelian
field (see Nakagawa [9], [10]).
2. Proof of Theorem 2.
We begin by explaining our notation. For an algebraic number field k,
we denote by Ek the unit group of k and by A(k) the Sylow p-subgroup
of the ideal class group of k. Let K be a finite Galois extension of k.
We denote by Gal(K/k) the Galois group of Kover k and by NK/k the
norm mapping from K to k.
Let F and M be as in the previous section and p an odd prime. We
fix a generator 7 of the Galois group Gal(Moo / Foo). We know that 7
has order 2 and we shall abuse the same letter 7 as its restriction to Mn.
Since p is an odd prime, we can decompose A(Mn) by the action of 7.
We put
A;t = {c E A(Mn) I c.. = c}

A;; = {c E A(Mn) I c.. = c- l }.
Then we have A(Mn) = A;t E9 A;; and A;t ~ A(Fn).
For a Zp-module Y, the p-rank of Y denotes the dimension of Y/pY
as a vector space over the finite field lFp. It is known that the {t-invariant
of the cyclotomic Zp-extension of M vanishes if and only if the p-rank of
A(Mn) is bounded as n -+ 00 (cf. Washington [14, Proposition 13.23]).
Because F is a real quadratic field, the p-rank of A(Fn) is bounded as
n -+ 00 by the theorem of Ferrero-Washington. Thus we may restrict
our attention to A;;- .
Lemma 3. Let p be an odd prime, K an algebraic number field and L
a quadratic extension of K. We put Gal(Loo/Koo) = (7) and
A(Ln)- = {c E A(Ln) I c.. = c- l }.

We assume that every prime of K above p does not split in L. If
IA(Lo)-1 = 1, then IA(Ln)-1 = 1 for all n.
318 T.ITOH
Proof. We fix a positive integer n and we put G = Gal(Ln / L). Since

any element of G is commutative with r, we have A(Ln)G ~ A(Kn)G EB
(A(Ln)-)G. Then, by the genus formula and the assumption that every
prime of K above p does not split in L, we obtain the equation
I(A(Ln)-)GI = IA(Ln)~1 = IA(Lo)-I' (EKo : EKo n NKn/Ko(~;)).

IA(Kn) I (ELo : ELo n NLn/Lo(Ln))
By using the method which is given in the proof of Lemma 1 of Kida [
7], we can show that the following map:
EKo/(EKo n NKn/Ko(K;)) -+ ELo/(ELo n NLn/Lo(L~))

induced from the natural embedding is injective.
Considering the above fact, if IA(Lo)-1 = 1, we have I(A(Ln)-)GI = 1.
Therefore we have IA(Ln)-1 = 1, because both of A(Ln)- and G are ~
groups. 0
Next, we define the Stark-Shintani units (cf. Shintani [12], Naka-
gawa [9]). We fix a positive integer n. Let fn be the finite part of the
conductor of Mn/ F. We denote by H(fn) the narrow ray class group
modulo fn of F. Take a totally positive integer 1/ of F with the property
1/ + 1 E fn, and denote by the same letter 1/ the narrow ray class modulo
fn represented by the principal ideal (1/). For each element e of H(fn),
we define
a
where a. runs over all integral ideals of F belonging to the class e.
The Stark-Shintani ray c!ass invariant Xfn (c) is now defined by
Xfn(e) = exp((F(O,e) - (F(O,el/)).
Moreover, for the subgroup Gn of H(fn) corresponding to M n, we put
Xfn (e, G n ) = TIgEG n Xfn (cg).
From now on, we assume that every prime divisor of p in F does not
split in M.
By Theorem 2 of Shintani [12] and Theorem 1 of Nakagawa [10], the
ray class invariant Xfn (c, Gn ) is a unit of M n , satisfying the reciprocity
law
Xfn (e, GnyT(cl) = Xfn (ee', Gn )
for e, c' E H(fn), where a denotes the Artin map from H(fn) to
Gal(Mn / F). Then the group
en = (-l,Xfn (e,Gn)le E H(fn))
On the Iwasawa j.1-invariant of the cyclotomic Zp-extension 319
is a subgroup of EMn and it is stable under the action of Gal(Mnl F). If

there is a prime divisor p of p with the property p lfo, then we replace Co
by the group generated by -1 and X fo (c,GO)2 e , where e is the number
of such prime divisors p of p.
By Theorem 1 of Nakagawa [10], it is known that Cn is a subgroup of
E;;={uEEMn I uT=u- 1}
with finite index and the p-primary part of the index (E;; : Cn) cor-
responds to the order of A;. Let Bn be the Sylow p-subgroup of the
quotient group E;; ICn.
For any integers m 2: n 2: 0, we put r m,n = Gal(Mm/Mn).
Lemma 4. We have B~m,n ~ Bn for all m 2: n 2: O.
Proof. By Proposition 1.2 of Nakagawa [10], both of the Tate cohomol-

ogy groups HA (rm,n, Cm) and HAl (r m,n, Cm) vanish. Then we get the
following exact sequence:
o ~ c~m,n ~ E~- ~ (E:;;.jCm)rm,n ~ iIl(rm,n, Cm) = O.
Considering the fact that c~m,n = Cn, we obtain the lemma. 0
The above lemma says tbat the natural map Bn ~ Bm is injective for
all m 2: n 2: O. We put
and
E= limBn,
+--
where the projective limit is taken with respect to the dual map of the
natural injection Bn to Bm for m > n.
We define the action of"f E r n,O on f E Bn by "f(f)(u) = f(u'Y- 1 )
for all u E Bn- Then E is a module over the complete group ring
Zp[[Gal(Moo/M)JJ ~ Zp[[TJ]. (1 +T acts as a fixed topological generator
of Gal(Moo/M).)
Proposition 5. If the p-rank of Bo is equal to the p-rank of B 1, then
the j.1-invariant of the cyclotomic Zp-extension of M vanishes.
Proof. First, we show that E is a finitely generated Zp-module. The
proof is similar to that of Theorem 1 of Fukuda [5J.
From the assumption, we have Bo/pBo ~ BI/pBl' By Lemma 4, we
have E/wnE ~ Bn for all n, where Wn = (1 + T)pn - 1. Then we see
woE + pB ~ w1B + pE.
320 T.ITOH
Since the element WI / Wo is containt:)d in the unique maximal ideal of

Zp[[T]], we can show that woE is contained in pE by using Nakayama's
lemma.
From the above argument, we have
p-rank of 130 = p-rank of (B/woB + pB)
= p-rank of B/pB
= p-rank of B.
Thus the p-rank of B is finite.
By the structure theorem of finitely generated Zp[[T]]-modules (cf.
Washington [14, Proposition 13.19, Lemma 13.21]), there exist integers
A and l/ such that
IBnl = IB/wnBI = pAn+v
for all sufficiently large n. Since IA~I = IBnl and the theorem of Ferrero-
Washington, the p-index of the order of An increases linearly with respect
to n. Then the It-invariant of the cyclotomic Zp-extension of M vanishes
by Corollary 2.5 of Sands [11]. 0
Now, we shall finish the proof of Theorem 2. Consider the exact
sequence,
0-+ C1 (F:-)3/(En 3 -+ E 1 /(E1 )3 -+ E 1 /C 1 (En 3 -+ O.
Because El /(En 3 ~ (Z/3Z)3, if the 3-rank of C1(E1 )3/(En 3 is at
least 2, then the 3-rank of El /C1 (En 3 is at most 1 and therefore B1 is
cyclic or trivial. Then we shall calculate the 3-rank of G1 (En 3 /(En 3 ~
GI/(G1 n (En 3 ).
We put 1]1 = X h (1, G I ) and fix a generator 'Y of Gal(MI/M). Since it
is known that CI/{l} is generated by 1]1,1]7,r/ (cf. Nakagawa [10]),
we have only to examine the relations of these elements in Ct/(C1 n
(En 3 ). In fact, if 1]~-')' is not a cube in M 1 , we can conclude that the
3-rank of C1 (En 3/(E1)3 is at least 2. Then the group Bl is cyclic or
trivial. In this case, we conclude that the 3-rank of Bo is equal to the
3-rank of B1 from Lemma 3. Thus the theorem follows.
3. Computing examples.
In this section, we give examples of vanishing of the J.L-invariant of
the cyclotomic Z3-extension by computer. Let the notation be as in the
previous sections.
First, we shall state an algorithm for computing of the minimal poly-
nomial of the Stark-Shintani unit X h (c, Gl) in the first layer MI. The
On the Iwasawa J.t-invariant of the cyclotomic Zp-extension 321
method of computation is well-known (cf. Chapter 6 of Cohen [2J, Shin-

tani [12]).
We assume that approximate values of X h (c, Gd are known for all
c E H(h). Since X h (1, Gl) is a unit in M 1 , we see that
peT) =
CEH(h)/Gl
is a polynomial whose coefficients are all algebraic integers in F. More-

over, it is known that
Ip(Xh (1, G1)u(c))1 = 1
for every embedding p of Ml into C, which includes the non-trivial au-
tomorphism on F, and all C E H(ft). We note that [Ml : F] = 6 and we
write peT) = E~=o aiTi. From the above fact, we have
la~1 ~ (~).
for each i (Recall that a~ is the conjugate of ai).
Since ai is an algebraic integer, if we compute an approximate value
of ~ with sufficiently precision, we can determine the exact value of ai
as an element of F.
We know that Xh (1, G1)u(c) = Xh (e, (i1). Therefore if we compute
II (T X h (e, Gt})
CEH(h)/Gl
for all signs and find only one possibility of signs, we can determine the
minimal polynomial peT) of"'1 = X h (l,Gt) over F.
Using this, the minimal polynomial Q(T) of ",~-'Y over F is also de-
termined. We use this polynomial for checking whether the element is a
cube or not.
Now, we shall give some examples. By virtue of Lemma 3, we shall
treat the case that the relative class number IA(M)-I is divisible by 3.
We shall treat the following case. Let 1 be a prime which satisfies
l == 3 (mod 4). We put F = Q( v'3i). Assume that the class 1111mber of
F is one. Let p be the unique prime of F which divides 3. If the order
of the narrow ray class group modulo p of F is 4, there exists a unique
quadratic extension M of F which is contained in the narrow ray class
field modulo p of F and satisfies the assumption of Theorem 2. We shall
calculate the JL-invariant of the cyclrJtomic Z3-extension of M.
Now, we put F = Q( J1B57) s.r..d p be the unique prime of F which
divides 3. We can see that the order of the narrow ray class group modulo
322 T.ITOH
P of F is 4. Therefore, let M be the uniquely determined quadratic

extension of F given by the above. In fact,
M ~F ( 1847277637 + 87771412 ( 1 + ~) ) .
By using KASH [3], we know that IA(M)-I = 3.
We computed approximate values of rll for all (J E Gal(Md F) by
using the function bnrLl of PARI-GP [1J with a precision of 150 decimal
digits. The values are the following:
171 = 127829214079.2490851840966 ..
177 =9.018079848618284017837940
2
171 = 0.000152017669860072698846 ...
By using the above mentioned method, we can compute the minimal
polynomial of 17i-"Y over F, hence we can also obtain its minimal poly-
nomial over Q. The minimal polynomial R(T) of 17i-"Y over Q is the
following:
R(T) = T12 - 840898094422485T ll
+ 11919386639653012969153266T10
-707132118709586493882042723418T9
+2749828994632785471910474645701~
-5454060080455494682796432087649~
+6699174123340475549716147390896~
-5454060080455494682796432087649~
+2749828994632785471910474645701~
-707132118709586493882042723418T3
+ 11919386639653012969153266T2
- 840898094422485T + 1.
It is easy to see that the polynomial R(T3 ) is irreducible over Q. Then
17i-"Y is not a cube in };[l. Therefore, the IL-invariant of the cyclotomic
Z3-extension of M vani.shes.
Moreover, applying Lemma 3 or Theorem 2, we have the following:
Corollary 6. Let l be a prime which satisfies l == 3 (mod 4). We put
F = Q( J3l). Assume that the class number of F is one and the or-
der of the narrow ray class group modulo (9) of F is 12. Let p be the
On the Iwasawa ,IL-inval"iant of the cyclotomic Zp-extension 323
unique prime of F above 3 and M(l) the unique quadratic extension of

F which is contained in the narrow ray class field modulo p of F and
satisfies the assumption of Theorem 2. Then for all l < 3000 satisfying
the above assumption, the J-L-invariant of the cyclotomic Zp-extension of
M(l) vanishes.
By using KASH, we can see that IA(M(I))-I = 1 except for the case
that I = 619, 1051, 1447, 1627, 2203, 2851 for alII < 3000 which satisfies
the assumption of Corollary 6. We shall give the minimal polynomial of
7J}-'Y for the excepted primes. (The case that l = 619 is already given.)
l = 1051
T'~ - 2542278225978963204331'Ll + 11693936340008463751284329420826Tu
-64631785736335621783029081743312125896910q9
-12791541530839596503320566556018092597069gr8
-175108867006294104086391689987020369063953T7
-223622353665891680117787205751936565518592r6
-175108867006294104086391689987020369063953T5
-12791541530839596503320566556018092597069gr4
-64631785736335621783029081743312125896910Jfi
+11693936340008463751284329420826T2 -254227822597896320433T+1
l = 1447
T'~ - 113672385330963276T" + 2424591727568282369368909939351746T 'U
-11690028900495605759755163234034460q9
+ 19578446431728482350223635099756527Jfi
-2789235255264008522712121751034204~
+34405225733217127980434116042421916J6
-2789235255264008522712121751034204~
+19578446431728482350223635099756527~
-11690028900495605759755163234034460Jfi
+2424591727568282369368909939351746T2
-113672385330963276T + 1
l = 1627
T'" - 5808189473979278450890876516921'''
+113029613762306360676648702561906749056384157442]i
-457848413488674428847769170295321691687066260632~
+1036017453975092892149945980475303304708737299287.r8
-1831393653954697709002068259804080470768300873916r7
+ 184597568042557306529931456201499165805364806580gr6
-1831393653954697709002068259804080470768300873916r5
+1036017453975092892149945980475303304708737299287.f4
-457848413488674428847769170295321691687066260632Jfi
+113029613762306360676648702561906749056384157442~
-580818947397927845089087651692T+1
324 T.ITOH
l = 2203
25437120536557149090797783433T"
T1:1. -
+512568729156253582510784888374639282630950566634To
-647047102314751027582855558829266691466837654199141257718]9
+2712248390171009806228064106559952419431810864089521681749]2
-5568303814535042477675452041376418784078869885055159190545~
+6951003181274469803024778973305789028706067730390560267936~
-5568303814535042477675452041376418784078869885055159190545Ji
+2712248390171009806228064106559952419431810864089521681749~
-647047102314751027582855558829266691466837654199141257718~
+512568729156253582510784888374639282630950566634Ji
-25437120536557149090797783433T + 1
l = 2851
T12 - 193553313251949925314995306745996573Tll
+12388276406007257115062365245914601586652682576355778106Jfo
-37462885070807392959051492934251075271574908048541902516443664421583922f9
-111398835362485827371779955278897436481955831146807532805279059108108539]8
-219591672808453587575337446499053253583352535397425539746057326454167361]1
-255802047289847856767973574463044610817826190920616524617543293974661472]8
-219591672808453587575387446499053253583352535397425539746057326454167361?5
-111398835362485827371779955278897436481955831146807532805279059108108539]4
-37462885070807392959051492934251075271574908048541902516443664421583922r3
+12388276406007257115062365245914601586652682576355778106~
-193553313251949925314995306745996573T + 1
References
[1] C. Batut, K. Belabas, D. Bernardi, H. Cohen and H. Oliver, "User's guide to
Pari-GP."
[2] H. Cohen, "Advanced topics in computational number theory," Graduate Texts
in Mathematics, 193. Springer-Verlag, New York, 2000.
[3] M. Daberkow, C. Fieker, J. Kliiners, M. Pohst, K. Roegner, M. Schornig, and
K. Wildanger, KANT V4, J. Symbolic Comput. 24 (1997), 267-283.
[4J B. Ferrero and L. Washigton, The Iwasawa invariant j.Lp vanishes for abelian
number fields, Ann. of Math. 109 (1979), 377-395.
[5] T. Fukuda, Remarks on Zp-extensions of number fields, Proc. Japan Acad. Ser. A
70 (1994), 264-266.
[6] K. Iwasawa, On the j.L-invariants ofZI-extensions, in "Number theory, algebraic
geometry and commutative algebra, in honor of Yasuo Akizuki," Kinokuniya,
Tokyo, 1973, 1-11.
[7] Y. Kida, I-extensions of CM-fields and cyclotomic invariants, J. Number Theory
12 (1980), no. 4, 519-528.
[8] J. Kraft and R. Schoof, Computing Iwasawa modules of real quadratic number
fields, Compositio Math. 97 (1995), no. 1-2, 135-155.
[9] J. Nakagawa, On the Stark-Shintani conjecture and cyclotomic Zp-extensions of
class fields over real quadratic fields, J. Math. Soc. Japan 36 (1984), 577-588.
On the Iwasawa p-invariant of the cyclotomic 'lrextension 325
[101 J. Nakagawa, On the Stark-Shintani conjecture and cyclotomic 'l.p-extensions of
class fields over real quadratic fields. II, Tohoku Math. J. 36 (1984), 439-452.
[11] J. Sands, On small Iwasawa invariants and imaginary quadratic fields, Proc.
Amer. Math. Soc. 112 (1991), 671-684.
[12] T. Shintani, On certain ray class invariants of real quadratic fields, J. Math.
Soc. Japan 30 (1978), 139-167.
[13] H. M. Stark, L-functions at s = 1. III. Totally real fields and Hilbert's twelfth
problem, Advances in Math. 22 (1976), 64-84.
[14) L. Washington, "Introduction to cyclotomic fields. Second edition," Graduate
Texts in Mathematics, 83. Springer-Verlag, New York, 1997.
Part IV
Modular forms and arithmetic functions

QUASIMODULAR SOLUTIONS
OF A DIFFERENTIAL EQUATION
OF HYPERGEOMETRIC TYPE
Masanobu KANEKO
Graduate School of Mathematics,
Kyushu University 33, Fukuoka,
812-8581, Japan
mkaneko<omath.kyushu-u.ac.jp
Masao KOIKE
Graduate School of Mathematics,
Kyushu University, Ropponmat'IJ" Fukuoka,
810-8560, Japan
koike<Omath.kyushu-u.ac.jp
1. Introduction and Main Theorem

In our previous paper [2], we studied further the solutions of the fol-
lowing differential equation in the upper half-plane .fj which was origi-
nally found and studied in [4] in connection with the arithmetic of su-
persingular elliptic curves;
Here, k is an integer or half an integer, the symbol' denotes the differen-

tiation (27ri)-ld/dr = q. d/dq (q = e21Ti7'), and E2(r) is the "quasimod-
ular" Eisenstein series of weight 2 for the full modular group SL2 (Z):
00
E2(r) = 1- 24 I)Ld)qn.
n=l din
Let p :::::: 5 be a prime number and Fp _ 1 (r) be the solution of the

above differential equation for k = p - 1 which is modular on SL 2 (Z)
[received: June 4, 2002; accepted in revised form: July 31, 2002]

330 M. KANEKO AND M. KOIKE
(such a solution exists and is unique up to a scalar multiple). For any

zero TO in f) of the form Fp - 1 (T), the value of the j-function at TO is
algebraic and its reduction modulo (an extension of) p is a supersingular
j-invariant of characteristic p, and conversely, all the supersingular j-
invariant.;; I'tre obtained in this way from the single solution Fp -l(r) with
suitable choices of ro. This is the arithmetic connection that motivated
our study of the differential equation.
Various modular forms on SL 2 (Z) and its subgroups were obtained
in [2] as solutions to this differential equation, the groups depending
on the choice of k. Every modular solution is expressed in terms of a
hypergeometric polynomial in a suitable modular function (hence the
"hypergeometric type" in the title of the paper), also depending on the
choice of k. For instance, if k == 0,4 mod 12, we have a modular solution
E I!l k k - 4 k - 5 1728
4(r) 4 F( -12' -12' --6-; j(r))'
where
F( b . ) = ~ (a)n(b)71 Tn
(a)n = a( a + 1) ... (a + n - 1)
a, ,C, x ~ () ,'
n=O en n.
is the Gauss hypergeometric series (which becomes a polynomial when
a or b is a negative integer, which is the case here), E4(r) the Eisenstein
series of weight 4 on SL 2 (Z), and j(r) the elliptic modular invariant.
In addition to the modular solutions, quite remarkable was an occur-
rence of a quasimodular form, not of weight k as in the modular case but
of weight k + 1. In the present paper, we give another supply of examples
of quasimodular forms as solutions to an analogous differential equation
attached to the group ro(2), which is not contained in SL2(Z);
where
ro(2) = {(~ ~) E SL2(Z)1 C == 0 mod 2}.
(ro(2) is the triangular group "2A" in the notation of Conway-Norton
[1].)
Let
E2A(r) := (E2(r) + 2E2 (2r))/3 = 1 - 8q - 40q2 - 32q3 - ...
be the quasimodular form of weight 2 on rQ(2) which is the logarithmic

derivative of the form
A2A{r) := 1]{r)81](2r)8 =q- 8q2 + 12q3 + 64q4 - ...
Quasimodular forms 331
of weight 8 on rQ(2)i E2A(r) = ~2A(r)/~2A(r), an analogous situa-

tion in the SL 2 (Z) case where E2(r) is the logarithmic derivative of the
Ramanujan ~(r). Consider the following differential equation;
Solutions which are modular on the group ro(2) and its subgroups were
studied in [6, 7J. In particular, when k is a non-negative integer congru-
ent to 0 or 6 modulo 8, the equation (#)k has a one dimensional space
of solutions which are modular on the group ro(2) itself. We note here
that the equation (#)k has a characterization by the invariance of the
space of solutions under the action of rQ(2), similar to the previous case
for SL2(Z), owing to the fact that there is no holomorphic modular form
of weight 2 on r;)(2) (see [5J and [2, 5]). By a general theory of ordinary
differential equations, we see that the equation (#)k has a quasimodu-
lar solution (which, since its transformation under r -- -1/2r is also a
solution, inevitably gives a solution having log q term in the expansion
at q = 0) only when k is a positive integer congruent to 3 modulo 4.
In the following, we show there indeed exists a quasi modular solution
in this case and describe explicitly the solution in terms of a certain
orthogonal polynomials. First we need to n.evelop some notations. Put
= 1 + 24 L C~=
00
d) qn = 1 + 24q + 24q2 + 96q3 + ... ,

n=l din
d:odd
where
1
17(r) = q"24
II (1- qn) = q"24 -
00
1 25 49
q"24 - q"24 + qu
121
+ ...
n=l
is the Dedekind eta function. The functions C(r) and D(r) are mod-
ular forms of respective weights 2 and 4 on the group ro(2) (="2B")
and the graded ring of modular forms of integral weights on ro(2) is
generated by these C(r) and D(r). Recall that (see [3]) an element
of degree k in the graded ring q E::.l r), C (r), D( r) J, where the genera-
tors E2(r),C(r),D(r) have degrt:ts 2, 2, and 4 respectively, is referred
to as a quasimodular form of weight k (on ro(2)). Incidentally, the
graded ring of modular forms of integral weights on rQ(2) is generated

by three elements C(T)2 = (E4(T)+4E4(2T))/5, C(T)3-128C(T)D(T) =
(E6(T) + 8E6(2T))/9, and .6.2A(T) of respective weights 4,6,8, of which
C (T? and .6. 2A (T) generate freely the subring consisting forms of weight
being multiple of 4, and the whole space as a graded module is generated
over this ring by C(T)3 -128C(T)D(T).
Now define a sequence of polynomials Pn(x) (n = 0,1,2, ... ) by
where
An = 4 (4n + 1)(4n + 3).
n(n + 1)
First few examples are
The Pn (x) is even or odd polynomial according as n is even or odd. We

also define a second series of polynomials Qn(x) by the same recursion
(with different initial values):
a couple of examples being
The Qn (x) has opposite parity: It is even if n is odd and odd if n is

even.
Put G(T) = C(T)2 -128D(T) (= (4E4(2T) - E4(T))/3).
Theorem. Let k = 4n+3 (n =0,1,2, ... ). The following quasimodular
form of weight k + 1 on ro(2) is a solution of (#)k :
Remark. The appearance of the square root J .6.2A(T) in the formula

is superficial because of the parities of Pn(x) and Qn(x), that is, the
form is actually an element in Q[E2(T), C(T), D(T)], by noting .6. 2A(T) =
D(T)(C(T)2 - 64D(T)) and C'(T) = (E2(T)C(T) - C(T)2)/6 + 32D(T).
The form does not belong to (2).ro
2. Proof of Theorem
It is convenient to introduce the operator ilk defined by
ilk(f)(r) = I'(r) - ~E2A(r)/(r).
By the quasimodular property of E2 (r) or the fact that E2A (r) is the
logarithmic derivative of ~2A(r), we have the transformation formulas
and
From these we see that if f is modular of weight k on a subgroup of

r0(2), then ilk(f) is modular of weight k + 2 on the same group. If 1
and 9 have weights k and l, the Leibniz rule
holds. We sometimes drop the suffix of the operator ilk when the weights
of modular forms we consider are clear. With this operator, the equation
(#)k can be rewritten as
(use E~A(r) = (E2A(r)2 - C(r)2)/8).

Denote the form in the theorem by Fk (r). We first establish the
recurrence relation (note n = (k - 3)/4):
(1)
This is a consequence of the recursion of Pn and Qn, namely (we often

omit the variable Thereafter)
Now we prove by induction that the Fk(T) satisfies the equation (#')k.
We can check the cases k = 3 and 7 directly. Assume Fk-4 and Fk satisfy
(#')k-4 and (#')k respectively. Then by using (1) and the formulas
we have
Hence we find
The proof of the theorem therefore boils down to show the equation
For this we also proceed by induction. For k = 7 the equation is checked

directly. Assuming that this is valid for k, we have
and
k+4~ k+4
-S- GFk+4 + Ci}(Fk+4) = 16(k + 1) G((k + 2)GFk - SCi}(Fk))
1 (1 3
+ 2(k + 1) C -"2(k + 2)C Fk
+ (k + 2)Gt9(Fk) + 2Gt9(Fk) - 8Ct9 2 (Fk))
= (k + 2)(k + 4) (G2 _ C 4 )R
16(k+1) k
k+5
= - -4- An+16. 2A Fk'
Here we have used the (previous) induction assumption that Fk satisfies

(#') k and the relation G2 - C 4 = - 2566.2A. This completes our proof.
References
[1] J. Conway and S. Norton, Monstrous Moonshine, Bull. London Math. Soc. 11
(1979), 308-339.
(2] M. Kaneko and M. Koike, On modular forms arising from a differential equation
of hyperyeometric type, to appear in Ra.ma.nujan J.
[3] M. Kaneko and D. Zagier, A generalized Jacobi theta function and quasimodular
forms, in "The Moduli Space of Curves (Texel Island, 1994)," 165-172, Progress
in Math., 129. Birkhauser Boston, Boston, MA, 1995.
[4] M. Kaneko and D. Zagier, Supersingular j-invariants, Hypergeometric series, and
Atkin's orthogonal polynomials, in " Computational perspectives on number the-
ory (Chicago, IL, 1995)," 97-126, AMSjIP Studies in Advanced Mathematics, 7.
Amer. Math. Soc., Providence, RI, 1998.
[5] M. Koike, Modular forms on non-compact arithmetic triangle groups, preprint.
[6] M. Koike and T. Saijo, in preparation.
[7] T. Saijo, Modular forms and differential equations, Kyushu University Master's
thesis, 1999.
SPECIAL VALUES OF THE

STANDARD ZETA FUNCTIONS
Hidenori KATSURADA
Muroran Institute of Technology,
27-1 Mizumoto, Muromn,
050-8585, Japan
hidenori(Dmmm.muroran-it,ac.jp
1. Introduction
Let M and k be positive integers, and > a Dirichlet character modulo
M such that >(-1) = (_1)k. Let
L
00
I(z) = a(m) exp(271'imz)

m=l
be a normalized cuspidal Hecke eigenform of weight k and Nebentypus

4> belonging to ro(M). Then, for a Dirichlet character X modulo N, we
define the standard zeta function L(j, s, X) twisted by X as
L(j, s, X) = II{(1 - x(p)a p,Bpp-s-k+1) (1 - x(p)a;p-s-k+1)

p
x (1 - X(P),B;p-s-k+1)} -1,
where af,,Bp are complex numbers such that ap + ,Bp = a(p) and ap,Bp =
4>(P )pk- for each prime number p. The standard zeta function of a
modular form is an important subject in number theory. To explain this
briefly, from now on we assume that k is even and X( -1) = >( -1) = 1.
We then put
L*(I ) L(j,m,x)
,m,X = 7r k +2m < 1,1 >
for an odd positive integer m ~ k - 1, where < *, * > is the normalized
Petersson product. As is well known, these values are algebraic numbers
and their qualitative natures have been fully investigated by many people
[received: June 6, 2002; accepted in revised form: November, 01, 2002)

338 H. KATSURADA
(cf. [Stur], [Sh2], [B-S]). Furthermore, these values are related to many
other areas, especially to the Galois representations. As for this, see,
for examples, [D-H-I] and [D]. To investigate various problems related
to these values, it is important to compute these values exactly. Several
people have considered algorithms for computing these values and have
carried out the computations. Zagier [Z] gave an explicit formula ex-
pressing L*(j,m,x) in the case where M is a squarefree positive integer
congruent to 1 modulo 4, c/> is the Kronecker character (Af) correspond-
ing to the extension Q( VM)/Q, and X is trivial. Sturm [Stur] gave an
algorithm for computing these values for a general X. However, it seems
difficult to give exact values by a direct use of his method. Stopple [Sto]
gave an explicit formula expressing L*(j, m, X) in case M = 1 and X is
a quadratic character.
In this note, we give some formulas which seem useful for the com-
putation of L*(j, m, X) in the case where M = 1 or a prime number
congruent to 1 modulo 4, c/> = (Af), and X is not necessarily quadratic
character of prime conductor p. To state our main result more precisely,
let M, C/>, X and p be as above. For simplicity assume that M t= p. Let k
and I be even positive integers such that I :s; k-2. Then put s = -1/2 or
o according as 1 = 2 and X2 = 1, or not. Then for two positive integers
ml,m2 put
(ml,m2;l,s)
= L C2,l( (~~ ~~), s)G7- l (mlm2, r)x(r)T(X)

r 2 :::;4mt m 2
Here, for a semi-positive definite half-integral matrix A of degree 2,

C2,l(A, s) denotes (up to constant multiple) the A-th Fourier coefficient of
a certain Eisenstein series E 2,l(Z,Mp2,c/>X,s) (cf. Section 2), G7- l (u,v)
is the polynomial introduced by Zagier [Z], and T(X) is the Gauss sum
(cf. Section 3). Furthermore, put
t(m; l, s) = (p,p2 mj l, s) - pk-2(p, m; l, s).
Take a basis {h}t;'l of Sk(fo(M),c/ consisting of primitive forms. Write
L
00
h(z) = ai(m) exp(27rimz)

m=l
with ai(1) = 1. Then for any positive integer m prime to pM we have

dl
t(m; l, s) = dk,l,p,M L L*(fi, 1+ 2s - 1, x)~ai(M)ai(p)ai(m),
i=l
where dk,l,p,M is a rational number explicitly determined by k, l,p, M,
and Ci is a certain complex number with absolute value 1. In case M = p
we have a similar formula (cf. Theorem 4.1.) We note that an explicit
form of C2,I(A, I) is given (cf. Section 2.) Thus, by the above formula
combined with the trace formula of Hecke operators, we can compute
the norm NK,.,,(L*(f,m,x) for a primitive form f E Sk(ro(M),4J) in
the case where m is an odd integer 3 ~ m ~ k - 3, or m = 1 and X2 :f: 1.
Here K /;x. is the field generated over Q by all the eigenvalues of Hecke
operators relative to f and all the values of X (cf. Theorem 4.2.) We
can also compute NK,.,,(L*(f,O,X) in case X2 = I,X:f: 4J by the same
method, and by the functional equation due to Li [L], we can finally
compute NK,." (L*(f, I,X in this case. We note that L*(f, 1,4J) has
been obtained by a different method (cf. Proposition 4.4.)
The main tool of the proof of the above formula is the pull-back
formula of the Siegel Eisenstein series of degree 2 due to B6cherer and
Schmidt [B-S], Shimura [Sh2]. Such a formula has been used to study
a qualitative nature of the special values of the standard zeta function.
However, as far as the author knows, no one has used the formula to
give its exact values. In this note, we carry out such a computation.
This note is organized as follows. In Section 2, we review the Siegel
Eisenstein series of degree 2. In Section 3, we review the pull-back for-
mula of Eisenstein series following [B-S]. In Section 4, by using the pull-
back formula we give an explicit formula for L*(f,m,x). In Section 5,
we give some numerical examples, and discuss some related topics.
The author is grateful to the referee for helpful comments on the
original manuscript.
2. Eisenstein series
Let GSp;t(R) be the group of proper symplectic similitudes of degree
n, and Hn Siegel's upper half space of degree n. As usual we write
,(Z) = (AZ + B)(CZ + D)-1 and j("Z) = det(CZ + D) for, =
(~ ~) E GSp;t(R). We write Ilk,(z) = (det,)k/2j("z)-kl(,(z
for, E GSp;t(R) and I E COO(Hn ). We simply write II, as Ilk, if
there is no confusion. Let SPn(Z) be the Siegel modular group of degree
n. For a positive integer M, we denote by r~n) (M) (resp. ra (n) (M)) the
subgroup of Spn(Z) consisting of matrices whose lower left n x n block
(resp. upper right n x n block) is congruent to 0 modulo M. For a
Dirichlet character modulo M, we denote by (resp. /) the character
340 H.KATSURADA
of r~n)(M) (resp. ro(n)(M)) defined by (,) = 4>(detD) (resp. '("() =

4>( det A)) for , = (~ ~).
For a Dirichlet character 4> modulo M, we denote by Mk(r&n)(M),4
(resp. M k (r&n)(M),4) the space of holomorphic (resp. C oo _) modu-
lar forms of weight k and Nebentypus 4> belonging to r~n)(M), and by
Sk(r&n\M), 4 the subspace of Mk(r&n) (M), 4 consisting of cusp forms.
Furthermore, for a subgroup r of Spn(Z) we denote by roo the subgroup
of r consisting of matrices whose lower left n x n block is O. For a func-
tion j on Hn we write jC(Z) = j( -Z). Let dv denote the invariant
volume element on Hn given by dv = det(Im(Z))-n-l /\l:::;j:::;l:::;n (dXjl /\
dYjl). Here for Z E Hn we write Z = (Xjl) + i(Yjl) with real matri-
ces (Xjl) and (Yjl). For two Coo-modular forms j and 9 of weight k and
Nebentypus 4> belonging r&n) (M), we define the Petersson scalar product
< j,g >r~n)(M) by
< j,g >r(n)(M)=

o
r
Jr~n)(M)\Hn
j(Z)g(Z) det(Im(Z))kdv,
provided the integral converges. Furthermore, we define the normalized

Petersson scalar product
< j,g >= m(r~n)(M)-l < j,g >r~n)(M)'
where r&n\M) is the fundamental domain for Hn modulo r~n)(M), and

m(r~n)(M)) = fr~n)(M)\Hn dv. For a Dirichlet character 1/J we denote
by L(8,1/J) the Dirichlet L-function associated to 1/J. Let n, l and M
be positive integers. For a Dirichlet character 4> modulo M such that
4>( -1) = (_1)1, we define the Eisenstein series E~ ,1(Z; M, 4>, 8) by
[n/2]
E~,l (Z; M, 4>, 8) = detIm(Z)S L(l + 28,4 II L(2l + 48 - 2i, 4>2)
i=l
x L ' ("()j (,,(, Z) -llJ (,,(, Z) 1-25 .

"YEro (n) (M)oo \ro (n) (M)
We then define E~l(Z;M,4>,8)
, by
E~ ,I(Z; M, 4>, 8) = j(t, Z)-l E~ ,l(t(Z); M, 4>, 8),

Special values of the standard zeta junctions 341
where [, = ( 0 1: -1
Onn ) . Let 'H.n(Z) denote the set of half-integral
matrices of degree n over Z, and we denote by 'H.n(Zho (resp. 'H.n(Zo)
the subset of 'H.n(Z) consisting of positive definite (resp. semi-positive
definite) matrices. Then it is well known that E~ l(Zj M, 4>, s) belongs
to M,OO(r&n) (M), 4, and has the following Fourier 'expansion:
E~,l(X + iYjM,4>,s) = L Cn,l,s(A, Y,M,4exp(21l'itr(AX,

AE?-tn(Z)
where tr( *) denotes the trace of a matrix. In particular, if

E~,(ZjM,4>,s)
, . belongs to M,(r&n) (M), 4, it has the following Fourier
expanSIOn:
E~,l(Zj M, 4>, s) = L Cn,l(A, M, 4>, s) exp(21l'itr(AZ.

AE?-tn(Z)~o
From now on we exclusively consider the case n = 2. Let l be an even

positive integer. Let M > 1 be an integer, and let 4> be a Dirichlet
character modulo M such that 4>(-1) = 1. Then E 2,l(Zj M, 4>,0) belongs
to M,(rb2)(M), 4 in case 1 ~ 4. Furthermore E 2,2(Zj M,4>, 0) belongs to
M2(r~2)(M),4 if 4>2 =1= 1. We remark that E 22 (ZjM,4>,O) is neither
holomorphic nor nearly holomorphic in the se~se of [Sh2] if 4>2 = 1.
However, E 2,l(ZjM,4>,-1/2) belongs to M2(r&2) (M), 4 in this case.
Now to see the Fourier coefficient of the Eisenstein series, for an ele-
ment A = ( au/2 a12/2) E 'H.2 (Z ) , put e = eA = QeD (au, a12, a22).
a12 a22 I
For an element A E 'H.2(Z) such that rank A = 1 and for each prime
number p define a polynomial Fp(A, X) as
ordp(eA)
Fp(A,X) = L (pX)i.
i=l
For an element A = (au 2 a 12 /2) E 'H.2(Zho -4 det A = dAiA2

a12/ a22
with dA the fundamental discriminant of Q( V- det A) and i A a posi-
tive integer. Furthermore, let XA = () be the Kronecker character
corresponding to Q(V-detA)/Q. For a prime number p define a poly-
342 H. KATSURADA
nomial Fp(A, X) as
ordp(eA) ordp(fA)-i
Fp(A,X)= L (p2X)i L (p3X2)i
i=O j=O
ordp(eA) ordp(fA)-i-l
- XA(p)pX L (p2X)i L (p3X2)j.
i=O j=O
For a Dirichlet character 'ljJ, let m'lj; denote its conductor, and 'ljJCO) the
associated primitive character. Furthermore, let Bm,'Ij; be the m-th gen-
eralized Bernoulli number associated with 'ljJ, and let T('ljJ) be the Gauss
sum defined by
T('ljJ) = L 'ljJ(X)exp(27riX/L).
XmodL
Let 1 be an even positive integer, and s = 0 or -1/2. Let be a Dirichlet

character such that ( -1) = 1. Now assume that the triple (I, s, )
satisfies one of the following conditions (h-1),(h-2),(h-3):
(h-1) 1 ~ 4 and s = 0,
(h-2) 1 = 2, s = 0 and 2 i= 1,
(h-3) 1 = 2 and s = -1/2.
First assume that (l,s,) satisfies either the condition (h-l) or (h-2).
Then for A E 'H2(ZO put
I
c\l(A,O) = C2,l(Aj M, , 0)
TIplf Fp(A, (p)p-l)!dA!l-3/2
_ xf11- 3( -i)T((XA)(O))m~;~A)(O)
- OXBl-1,CtPXA)(O) I1pI M(l - pl-I(XA)(O) (p)) if A > 0
otherwise.
Next assume that (l,s,) satisfies the condition (h-3). Then for A E
'H2(ZO put
if A > 0
ifrankA=l
if A = O.
Special values oj the standard zeta junctions 343
Remark. Let be a primitive character with conductor m such that

( -1) = 1. If dA is prime to m, we have
c2,tCA,0) = (dA)XA(m)r() II Fp(A, (P)p-l)f~l-3ml-lBl_l,(t/>XA)(O)

P/JA
for 1 ~ 4. On the other hand, if 2 = 1, we have
- (A , 0) =
C2,1 II F.p(A, ,t.(P)P-I) IdA11-3/2/21-3
'P
3/2-1
A m(t/>XA)(O)
p/JA
x B1-l,(t/>XA)(O) II (1 - pl-l(XA)(O)(P)).
plM
Thus if 2 = 1 and dA is prime to m, we have
C2,l(A,0) = II Fp(A, (P)p-l)/11-3m3/2-1 B1-1,(t/>XA)(O).

PI/A
Theorem 2.1 (cf. [Ka1] , [Sh2]). Let M > 1 be an integer, and a
Dirichlet character modulo M such that ( -1) = 1. Let 1 be an even
positive integer, and s = 0 or -1/2. Assume that the triple (l, s, )
satisfies one of the conditions (h-l),(h-2),(h-3). Let
871"5/2 ( -1 )1/22171"31-3/2
A(l, s) = r(3/2) or r(l)2r(l-1/2)
according as 1 = 2 and s = -1/2, or 1 ~ 4 and s = 0, where r(*) is

Gamma function. Then we have
C2,I(AiM,,S) = A(l,s)C2,I(Ai M ,,S)
for S = 0 or -1/2.
3. Pull-back formula
From now on put ro(N) = r~1)(N), and in particular put r = r(l).
For an even positive integer 1 and non-negative integer II we define a
polynomial G?" (
u, v) in u, v by
G211 ( ) = ~(-1)~ (l + 211 - J.L - 2)! ~ 211-2~

I U, V ~ (211 _ 2J.L)!J.L! u v .
This type of polynomial was introduced by Zagier (cf. [Z], [I].) Let 1
be an even positive integer, and a Dirichlet character modulo M
344 H.KATSURADA
such that 4>( -1) = 1. For an even positive integer k 2: i and a prim-
itive Dirichlet character X modulo N, we define a function (ZI, Z2) =
2,k(ZI,Z2;i,M,4>,s) on HI x HI by
2,k(ZI, Z2; l, M, 4>, x, s)

= ~ ~ ~ (mi r /2) (YI 0) _)
L....J L....J L....J C2,I,s ( r /2 m2 ' 0 Y2 ' M, 4>x
ml =0 m2=0 r2:::;4ml m2
X Gr- 1(mIm2, r)x(r)r(x) exp(27rimIXI) exp(27rim2x2),

where we write ZI = Xl + iYI, Z2 = Xl + iY2. Assume that N2 divides
M. Then, 2,k(ZI, Z2; l, M, 4>, x, s) defines an element of Mr(ro(M), 4
when considered as a function of Zl or Z2, and hence it belongs to
Mr(ro(M),4 0 Mr(ro(M),4 (cf. [B-S, Sections 2 and 6J.) Thus
for an element f(ZI) E Sk(ro(M), 4, we can define
< f, 2,k( *, -Z2; i, M, 4>, x, B)) >ro(M)
when we consider 2,k(*,-z2jl,M,4>,x,)) as a function of Zl. Then by

[B-S, Theorem 3.1J, we have
Theorem 3.1. In addition to the notation and the assumption as above,
assume that M > 1, N2 divides M,4>2 = 1 and X( -1) = 1. Let f E
Sk(ro(M), 4 be a common eigenfunction of all Hecke operators. Then
we have
< f,2,k(*,-zjl,M,4>,x,)) >ro(M)
= /'i,1,k(S)Nk+l+2s-2 M I- k/ 2L(fIWM, i + 28 - 1, X)
X f/WM/T(M/N2)(z),
where
r(k + s -1/2)r(k + s -1)
(_1)1/2
/'i,1,k(S) = 2-3+2k-l+2s7rk-l-1 r(i + s)r(i + s -1/2)
r(k - 1)
x nz:~ (Jj/2)(k - 1 - s - Jj/2)'
T(M/N2) is the Hecke operator, and WM = (it ~1).

Now let 4> be as in Theorem 3.1, and assume that the triple (l, s, 4>X)
satisfies one of the conditions (h-1),(h-2),(h-3) in Section 2. Then we
t
where ql = exp(27rizl),q2 = exp(27riz2). Then 2,k(Z},Z2;l,M,c/>,X,s)
belon&.s to Mk(ro(M), c/ Mk(ro(M), c/, and, in particular if k > l,
then 2,k(Zl,Z2;I,M,c/>,X,s) belongs to Sk(ro(M),c/ Sk(ro(M),c/.
Furthermore, by Theorem 2.1 we have
2,k(Zl, Z2; I, M, <p, x, s) = A(I, S)2,k(Z}, Z2; I, M, <p, x, s)

From now on, for a Dirichlet character 'l/J modulo Mo we use the same
symbol 1/J to denote the character modulo M induced from 1/J if Mo di-
vides M. For a positive integer r put 6r = (~ ~), and let
Sk(ro(M), <p)(r) = U18r ; f E Sk(ro(M), <PH, and Sk(ro(M), <p)new the
space of new forms in Sk(ro(M), <p). Furthermore for a primitive form f
in Sk(ro(M),<p) let cf be the complex number such that flWM = cffC.
Let Af(m) be the eigenvalue of the Hecke operator T(m) for a positive
integer m. For an odd positive integer m ::; k - 1, let
A(f ) _ r(k - l)r(k + m - l)r(m + 1) LU, m, X)

, m, X - r(k _ m) 22k+2m- 47rk+2m < f, f >'
and
A(f LU, 0, X)
)
,O,X = r (k -1 ) 22k-37rk < f,f >'
We note that mC~ro(N) = 7r/3[r : roCN)]. Thus by Theorem 3.1 we
have
Theorem 3.2. Under the above notation and the assumption, we have
the following:
(1) Let c/> be a primitive quadratic character modulo pr with r ~ 1. Let
f be a primitive form in SkCro(PT),<p)new. Then
< f,2,kC*, -z; I,pr, <p, 1, s >

= 3[r : ro(pr)t 1p(1-k/2)r Acr, l + 28 -1,1) < f, f > ).,(pr)c,r(z).
346 H. KATSURADA
(2) Let cp be a primitive quadratic character modulo P, and f a prim-

itive form in Sk(rO(p),cp)new. Then we have
and
(3) Let Po = 1 or a prime number different from p. Let cp be a prim-

itive character modulo Po such that cp2 = 1, and f a primitive form in
Sk(rO(pO),cp)new. Then for a primitive character X modulo p, we have
- -. 2
< f,2,k(*,-z,l,p PO,cp,X,8)) >
= 3[r : ro(p2po)1-lpl+2sp~-k/2 A(rlop2, 1+ 28 - 1, X)
X < f18p2,f18p2 > c,).,(po)rI8p2(z),
- - 2
< fI8 p ,2,k(*,-z;l,p Po,cp,X,s)) >
= 3[r : ro(p2po)J-lpl+2sp~-k/2 A(rI8p , l + 2s - 1, X)
x < fl8p, fl8p > c,).,(po)rI8p(z),
and
- - 2
< fI 8p2 ,2,k(*,-z;l,p Po,cp,X,8)) >
= 3[r : ro(p2po)rlpl+2Sp~-k/2 A(r, 1+ 28 - 1, X)
x < f, f > c,).,(po)r(z).
4. Computation of L(j, l, X)
Let p be prime a number such that p == 1 mod 4. Let Po = 1, or a
prime number such that (EQ) = 1 and Po == 1 mod 4. Here (Pff) denotes
the Legendre symbol. In t&is section we give some formulas to compute
LU, m, X) for a primitive form f E Sk(ro(N), 1/J) in the following three
cases:
(1) N = p, 1/J = (~) , and X is trivial,
(2) N = p, 1/J = (~) , and X is a primitive character modulo p.

(3) N = Po, 1/J = (:0) , and X is a primitive character modulo p.
Put M = p,p2 or POp2 according as the case (1), (2), or (3). Let
k 2:: 4 be an even positive integer, and 1 a positive integer such that
1 :::; k - 2. Assume that the triple (1, s, 1/JxJ satisfies one of the conditions
(h-l),(h-2),(h-3) in Section 2. For two positive integers ml! m2 put
Furthermore, put
3-1(p + l)pk/2-1(l, m; 1, s) case (1)

{ 3- 1(p + 1)pl/2-l-2s(p, m; l, s) case (2)
t(m'l
, , s) :- 3-1t k/2-1 -k-l-2s+2
poPa P
X ((p,p 2m; l, s) - p k- 2(p, m; l, s)) case (3),
where tpo = Po + 1 or 1 according as Po is prime number or 1. Take
a basis {/dt;'1 of Sk(ro(N), 1/J)new consisting of primitive forms. Let
lilWN = Ci/f with constant Ci, and write
L
00
li{Z) = ai(m) exp(27l"imz)

m=1
with aiel) = 1. Using Theorem 3.2 we shall show

Theorem 4.1. Let the notation and the assumption be as above.
(1) In case (1), for any positive integer m we have
L A(fi, 1+ 2s - 1, I)Ciai(P)(ti(m).
dl
t(m; l, s) =
i=1
(2) In case (2), for any positive integer m we have

dl
t(m; l, s) = L A(/i, 1+ 2s - 1, 1/J)ai (p)ai (m).
i=1
348 H.KATSURADA
(3) In case (3), for any positive integer m prime to PPo we have
dl
t(m; l, 8) = L A(h, l + 28 - 1, x)~ai(po)ai(p)ai(m).
i=l
Sketch oj the proof. Here we treat only the case (3), and show how to
derive the formula from Theorem 3.2. As for the details, see [Kat2]. As
is well known,
Sk (ro (p2pO) , 'lj;)

= Sk(rO(PO) , 'lj;) $ Sk(rO(PO), 'lj;)(p) $ Sk(rO(PO), 'lj;)(p2)
..lSk (ro (Ppo) , 'lj;)new ..lSk(rO(Ppo), 'lj;)new(p) ..lSk (ro (P2pO), 'lj;)new.
Take bases {gi}t!l of Sk (ro(PPo), 'lj;)new, and {hi}t!l of Sk(rO(p2pO),

'lj;)new consisting of primitive forms. Then {Ii (i = 1,2, ... ,dl ), fil8p (i =
1,2, ... ,dd, fil8p2 (i = 1,2, ... , dd, gi (i = 1,2, ... , d2), gil8p (i =
1,2, ... , d2), hi (i = 1,2, ... , d3)} forms a basis of Sk (ro (p2pO) , 'lj;). Thus
we have
- 2
2,k(Zl,Z2;l,p po,'lj;,x,s)
d3 1 d2
= L c~J,0)hi(Zl)hj(Z2) + L L b~j,/3)9iI8p"(Zl)gjI8p13(Z2)
i,j==l Ct,/3==o i,j==l
2 dl
+ L L a~j,/3) fil 8p" (zdfj 18p13 (Z2)
Ct,/3==O i,j==l
WI
'th Cij ij ' Cij(Ct,/3) E C . Now we h ave f i ITXT
(0,0) , b(Ct,/3) YYp2pO = Ci fel s:
i 0p2,
fil8pIWp2pO = CifPl8p. Similarly, gilWpoP2 = C;g;18p with constant C;, and
hilWpop2 = d/hi with constant df. From this we have gil8plWp2pO = <;g;.
For a positive integer write Ai(m) = A/i(m), Ai(m)' = AgJm) , and
Ai (m)" = Ahi (m). By assumption, we have 'lj;(p) = 1. Thus Ai (P) is a
real number. Now by using the formula [Shl, (2.5)]' we can show that
we have
and
Then by a direct computation we have
- 2
e2,k(ZI, Z2; l,p Po, 'I/J, X, 8)
d3 d2
=L Cii hi(ZI)hi (Z2) + L bii { _p-k/2 Ai (P)9i (zI)9i (Z2)
i i=1
+ 9i (zI)9i Idp(Z2) + 9ildp(zI)9i(Z2) - p-k/2 Ai (P)9i Idp(ZI)9i Idp(Z2)}
dl
+L llii{p-l fildp2(zdfildp2(Z2) - p-k/2 Ai(P)fildp(ZI)fildp2(Z2)
;=1
+ !i(zl)!ildp2(Z2) - p-k/2 Ai (P)fi Idp 2 (zdfi Idp (Z2)
+ (1 + Ai(P)2p-k - p-2)fildp(zdfildp(Z2) - Ai (p)p- k/ 2fi(Z1)!ildp(Z2)
+ !ildp2(ZI)!i(Z2) - p- k/2Ai (P)!i Idp(Zl)fi(Z2) + p-l !i(Zl)!i(Z2)},
where
C;i = 3p- 1 (p + 1)-ltPc/pl+28p~-k/2 A(hi' 1 + 28 -1, X)t;'Ai(PO)",
b.. = 3 -1(p + 1)_lt_lPl+28p~-k/2 A(9i, 1+ 28 - 1, X)~Ai(PO)'

n p Po 1 -p-2 '
and
.. = 3 -1 (P + 1)-lrl (1 + p-l )pl+28p~-k/2 A(fi, 1 + 28 - 1, X)~Ai(po)
au p po (1_p-l)1+p-l)2_p-kAi(P)2) ,
where dq = Po + 1 or 1 according as Po is prime number or 1. Now let
=L
00
9i(Z) bi(m) exp(21l"imz) (i = 1,2, ... , d2)

m=1
and
=L
00
hi(Z) c;(m) exp(21l"imz) (i = 1,2, ... , da)

m=1
with bi (1) = c;(1) = 1. Put
L
00
ep (z2) = (p, m2; 1, 8) exp(21l"im2z2).

m2=1
We note that c;(pm) = 0 and bi(pm) = Ai(P)'bi(m) for any positive

integer m. Thus we have
dl
e p(z2) = L aii{pk/2 (1 - p-2)!i Id (Z2) -p (1 - p-l)Ai(P)!i(Z2)}
i=1
350 H. KATSURADA
d2
+ Lpk/2(1 - p- 2)bii gi (Z2).
i=l
We note that bi (p 2 m) = pk - 2 bi (m) for any positive integer m prime to
p (cf. [Miy, Theorem 4.6.17]). Thus the assertion follows by a direct
computation. 0
For a prime number q not dividing PPo put
[i/2)
, -- 13('Z + 1, q,'l , s ) --
13t+l ~(,C
~ t r - t 'cr-l )qr(k-l)t( q
i-2r'l
, ,S)
r=O
"
for i = 0,1, .... , d1 - 1, where iCr = ~)

r:\z-r): . We understand that iC-l =
O. For a Hecke eigenform f let K f be the field generated over Q by
all the eigenvalues of the Hecke operators relative to f. Furthermore,
for a character X let K f,x be the field generated over K f by all the
values of X. Put ef = [Kf : Q], and we denote by NKf(ex) the norm
of ex over Q for ex E K f. Similarly we define e f,x and N Kf,x. (ex) for
ex E Kf,x' Let {fi}~=l be the basis of Sk(ro(N), 'Ij;) as above, and write
Ki = K fi,x and ei = efi,x' Let cI>(X) be the characteristic polynomial
of T(q) on Sk(fo(N), 'l/J). We note that NK;(Ci) = 1 in cases (1),(2),(3).
FUrthermore we have NK;(ai(p)) = p(k-l)ei /2 in cases (1) and (2),and
NK;(ai(PO)) = p&k-l)ei/2. Thus by [G, Lemma 2.2] we obtain
Theorem 4.2. Let the notation and the assumption be as above. Let f
be a primitive form in Sk(ro(N), 'Ij;), and a(q) be the q-th Fourier coeffi-
cient of f. Assume that cI>'(a(q)) =1= O. Write cI>(X) = Lt~obdl-iXi,K =
Kf,x, and e = ef,x'
(1) In case (1) and (2), we have
N (A(f I + 2 _ 1 )) = (1-k)e/2 N (Et~ol E1~il f3dl_jbj_ia( q)i)

K , s ,X P K cp' (a( q)) .
(2) In case (3), if a(p) =1= 0, we have
N (A(f I + 2 _ 1 )) _ (1-k)e/2 N (Lt~ol L1~il f3dl-jb j _ia( q)i )

K , 8 , X - Po K a(p)cI>'(a(q)) .
Remark 1. For a Hecke eigenform f E Sk(fo(M), 'l/J),

L(f) L(j, 8,1)
,8,1 = TIpI M(1 _ a(p)2 p -s-k+1)'
and
L(I s .f,) = LU, s, 1/J)
, ,or ilp lM(1 _ p-S) .
Put for an odd positive integer l :$ k - 1
A(I l ) = r(k - 1)r(k + l - 1)r(l + 1) LU, l, X)
, ,x f(k -l) 22k+2l-47rk+2l < I, I>'
and
AU, 0, X) = r(k - 1) 22k-~~' ~ ~~ I>
for X = 1 or 1/J. Then Zagier [Z] gave a useful algorithm to compute
the exact value of AU, l, 1). Sturm [Stur] gives a general algorithm to
compute AU, l, X) for an eigenform I E Sk(ro(N),1/J) and a Dirichlet
character X. However, even in the case N = 1 and X is a quadratic char-
acter modulo a prime number p, one needs precise information about the
Fourier coefficients of primitive forms in Sk(ro(p2 to compute AU, l, X)
by using his method. Thus it seems difficult to carry out his method
directly for the computation of such a value. Such Fourier coefficients ap-
pear also in the course of our algorithms as seen above. However, they
eventually disappear in the final formula. We note that Stopple [Sto]
gave a formula similar to (3) in case f E Sk(SL2(Z and X is quadratic.
His formula involves some non-elementary character sums. On the other
hand, our formula does not contain such a non-elementary term.
Remark 2. By using the same method, we can give an exact value of
AU, l, 1) for 1 2:: 3 for a cuspidal Hecke eigen form of level 1.
Remark 3. The function t 2,k(Zl! Z2; k, M, 1/J, x, 0) does not belong
to Sk(ro(M),1/J) Sk(ro(M),1/J), but it belongs to Mk(ro(M),1/J)
Mk(ro(M),1/J). Thus, modifying the above method, we can compute
the value AU, k - 1, X). We remark that the numerator of AU, k - 1,1)
is related to the congruence between the Klingen-Eisenstein lifting of f
and another Siegel modular form. As for this, see [Ku] and [Miz].
Remark 4. As seen above, if X2 :f: 1, E2 2(Zj M, 1/Jx, 0) belongs to
M2(ro(M), 1/J). Thus we can use the same method to compute the value
AU, 1, X). On the other hand, if X2 = 1, E2,2(Z; M, 1/Jx, 0) does not
belong to M2(ro(M),1/JxJ, and thus we cannot give an exact value of
AU, 1, X) by a direct use of the above method. However, we can relate
the value AU, 1, X) to AU, 0, X) by using the functional equation. We
explain this in case (1), (3). First in case (1), put
RU, s, 1) = M(s+k-l)/2 7r -3/2(s+k-l)
x r((s + k - 1)/2)r((s + k)/2)r((s + 1)/2)LU, s, 1).
352 H.KATSURADA
Next in case (3), for f E Sk(SL 2 (Z)), and a quadratic character X
modulo M, put
R(j, s, X) = M 3(s+k-1)/211'-3/2(s+k-1)
x r((s +k - 1)/2)r((s + k)/2)r((s + 1)/2)L(j, s, X).
Then by Li [Ll, we have the following functional equation:
Proposition 4.3. (1) In case (1),
R(j,l - s, 1) = R(j, s, 1).
In particular
-
A(j, 1, 1) = p-1/2-A(j,O,l).
(2) In case (3), under the above notation and the assumption, we have
R(j,l - s, X) = R(j, s, X).

In particular, in case (3), we have
A(j, 1, X) = p-3/2 A(j, 0, X).

In case (2), the value A(j, 1,</ can be given by a different method
(cf. [Z].)
Proposition 4.4. In case (2), we have
A(j, 1, </ = 2/3(1 _ p-2).
5. Numerical examples and comments

By Theorem 4.2 combined with Proposition 4.3 we can compute the
values A(j, m, X). A subspace S of Sk(ro(N), 1/J) is called non-splitting
if it is spanned by all Galois conjugates of a primitive form in S. Take
a primitive form f of Sk(SL2(Z)). Assume that Sk(SL2(Z)) is non-
splitting. Then, NKf(A(j,l,(~))) is independent of f. Thus, in this
case, we write this value as L(k,l,p). Similarly, in case Sk(rO(p), (~))
is non-splitting, we define (k,l,p) as NKf(A.(j,l,l)) for a primitive
form f of Sk(rO(p), (~)). It is conjectured by Maeda that Sk(SL2(Z))
is non-splitting, and so far, this conjecture has been verified at least for
k ::; 2000 (cf. [H-MJ, [F-J]). We show some examples of L(k,l,p) for
various k, l,p. We have computed the values by using Mathematica.
(1) We compute L(12, l,p) for 1 ::; 1 ::; k - 3 and some prime numbers
p. The results are as follows:
1 L(12,1,5)
1 2'" . 3 . 7/5'v
3 214 .35 .7.2851/513
5 219 .35 7 1511599/516
7 219 . 38 . 73 . 521 . 295387/5 20
9 226 .310.72.110308273279/524
1 L(12, 1,13)
1 2'" . 3 . 5"' . 7 . 563/13'"
3 214 .35 .52 .7.41177. 1445419/1316
5 219 .3 5 .54 .7.299696968678699/1320
7 219.37.53.73.312.5479.306945156059/1323
9 226 . 310 . 55 . 72 . 547 10267 1634679978646831/13 26
l L(12,l,17)
1 2~' 3O5"72389/17'~
3 217 .35 .52 .7.477200018623/1714
5 220 . 35 54 . 7 . 23 . 29 . 997 . 46316422211/1720
7 222 . 38 53 . 73 . 167 . 11003 322079 . 915248119/1724
9 230.310.55.72.43.892028959.1604767911433/1727
We note that the value L(12, l,p) has been obtained by Stopple [StoJ in
the case where p = 5 and 1 = 1,3,5,7, or p = 13,17,29,37,41 and 1 = 1.
We note that relatively large prime numbers appear in the numerator of
L(k, 1, 5) contrary to the untwisted case in [D). At present, the author
does not know whether these primes relate to any arithmetic algebraic
geometry. We note that the numerator of AU, l, 1) is related to the order
of the Shafarevich-Tate group (cf. [D).)
(2) The values of the standard zeta functions at s = 1 are particu-
larly important. To explain this, let q be a prime number congruent
to 1 modulo 4~ and let OQ(y'q) be the ring of integers in Q(JQ). Let
Sk,k(SL2(OQ(y'q) be the space of cusp forms of weight (k, k) with re-
spect to SL2(OQ(y'q). Then Sk,k(SL2(OQ(y'q)) has the following de-
composition:
Sk,k(SL2(OQ(y'q) = Sk(SL2(Z)).lSk(rO(q), ;- ).lSk,k'

A A (q) 0
where Sk(SL2(Z)) (resp. Sk(rO(q), (!))) is the image of Sk(SL2(Z))

(resp. Sk(rO(q), (!) under the Doi-Naganuma map, and B2,k the
orthogonal complement of Sk(SL2(Z).l Sk(rO(q), (!) in
Sk,k(SL2(OQ(y'q)) with respect to the Petersson product. Take a prim-
itive form 9 E B2,k' and for an integral ideal 21 in Q(JQ) let c(21jg) be
the 2l-th Fourier coefficient of g. Let Kg be the field generated over Q
by all c(21jg)'s and Kt the subfield of Kg generated by c((p)jg) for all
rational primes p. We denote by D(Kg/ Kt) the relative discriminant
354 H.KATSURADA
of K g / Kt. Assume that SZ,k is non-splitting, that is, it is spanned by
all Galois conjugates of a primitive form in sZ ,k' Then N K+D(Kg/
9
K g+)
does not depend on g. Hence we write this as Dk,q' Then in [D-H-I] Doi,
Hida, and Ishii, among others, conjectured the following:
"Any prime factor of Dk,q divides either the numerator of
NKf(A(f, 1, (;))) for some primitive form f in Sk(SL 2(Z))
or the numerator of NKf(A(f, 1, 1)) for some primitive form
f in Sk(fo(q), (;))."
Doi, Hida and Ishii [D-H-I] computed an exact value of L(k,l,p),
and verified the above conjecture in some cases. Goto [G] computed
the value L(20, 1,5), and Hiraoka [Hir] computed the values L(22, 1, 5)
and L(24, 1, 5). Then, combining the result of [D-H-IJ, they verified the
conjecture for (k, q) = (20,5), (22, 5), and (24,5). Now let k = 12 and
q = 13. In this case, S?2,12 and Sk(fo(13),e;)) are non-splitting. Fur-
thermore, we have D 1213 = 13 . 563 6205151 and the numerator of
L(12, 1, 13) is 5.7.13 29 '.6205151. Thus it is expected that 563 appears
in the numerator of L(12, 1, 13). The example in (1) shows that this is
true.
Now we compute L(k, 1,5) for 16 S k S 38. As for the other numerical
examples, see [Sto].
First let k = 16,18,20,22,26. Then we have dimSk(SL2(Z)) = 1 and
we have the following table:
k L(k, 1,5)
16 21~347~11/514
18 213 . 34 . 7 . 11 . 13/513
20 215 . 34 . 7 . 11 . 13 . 977/5 18
22 215 . 35 . 72 . 13 . 17 . 71/5 18
26 213 .34 7 11 . 17 19337 1409/522
Next let k = 24,28,30,32,34,38. Then we have dimSk(SL2(Z)) = 2.

Furthermore, K h = K h is a real quadratic field, and thus we write this
as K. Then we have the following table:
k K L(k, 1,5)
24 Q( y144169) 229 .39 .74 .11 4 13171910954449/5~1/ 144169
28 Q(v'131139) 226.38.73.112.133.17.19.23.66876860429/544 131139
30 Q(v'51349) 226.310.74.112.13.17.19.23.3253.20017939/543.51349
32 Q( v'67273067) 226.310.74.11.132 17192337 157P1 /5 54 67273067
34 Q( v'4794919) 226 .3 10 .73 .11 3 .132 .19.23.29.191.3191p2 /5 51 4794919
38 Q( v'181.3491009) 228.38.74.112 .13.17.23.29.31P3 /5 60 1813491009
PI = 222142617427425679,
P2 = 121120620073,
P3 = 24539630352019799615221087.
Remark. By using the pull-back formula of Eisenstein series of degree 4,

1bukiyama's differential operators in [I], and an explicit formula for the
Siegel series in [Kat1], we can obtain exact values of standard zeta func-
tion of cuspidal Hecke eigenform of degree 2. As for this, see [Kat3].
References
[B-S] S. BOcherer and C. G. Schmidt, p-adic measures attached to Siegel modular
forms, Ann. Inst. Fourier 50 (2000), 1375-1443.
[D] N. Dummigan, Symmetric square L-functions and Shafarevich-Tate groups,
Exp. Math. 10 (2001), 383-400.
[D-H-I] K. Doi, H. Hida, and H. Ishii, Discriminant of Hecke fields and twisted
adjoint L-values for GL(2), Invent. Math. 134 (1998), 547-577.
[F-J) D. W. Farmer and K. James, The irreducibility of some level-l Hecke poly-
nomials, Math. Compo 71 (2002), 1263-1270
[G] K. Goto, A twisted adjoint L-values of an elliptic modular form, J. Number
Theory 73 (1998),34-46
[H-M] H. Hida and Y. Maeda, Non abelian base change for totally real fields, Pacific
J. Math. Olga Taussky-Todd Memorial Issue (1997), 189-218.
[Hir] Y. Hiraoka, Numerical calculation of twisted adjoint L-values attached to
modular forms, Exp. Math. 9 (2000), 67-73.
[I) T. Ibukiyama, On Differential opemtors on automorphic forms and invari-
ant pluri-harmonic polynomials, Comm. Math. Univ. St. Pauli 48 (1999),
103-118.
[Katl] H. Katsurada, An explicit formula for Siegel series, Amer. J. Math. 121
(1999), 415-452.
[Kat2] H. Katsurada, Special values of the twisted standard zeta function of an
elliptic modular form, preprint (2001).
[Kat3] H. Katsurada, Exact values of the standard zeta function of a Siegel modular
form of degree two, preprint (2001).
[Ku] N. Kurokawa, Congruence between Siegel modular forms of degree two, Proc.
Japan Acad. 55 (1979), 417-422.
[L] W. Li, L series of Rankin type and their functional equations, Math. Ann.
244 (1979), 135-166.
[Miy] T. Miyake, "Modular forms," Springer-Verlag, Berlin, 1989.
[Miz] S. Mizumoto, Congruences for eigenvalues of Hecke opemtors on Siegel mod-
ular forms of degree two, Math. Ann. 275 (1986), 149-161.
[Shl] G. Shimura, The special values of the zeta functions associated with cusp
forms, Comm. pure appl. Math. 29 (1976), 783-804.
356 H. KATSURADA
[Sh2] G. Shimura, "Arithmeticity in the theory of automorphic forms," Math-

ematical Surveys and Monographs, 82. American Mathematical Society,
Providence, 2000,
[Sto] J. Stopple, Special values of twisted symmetric square L-functions and the
trace formula, Compositio Math. 104 (1996),65-76.
[Stur] J. Sturm, Special values of zeta functions and Eisenstein series of half in-
tegral weight, Amer. J. Math. 102 (1980), 219-240.
[Z] D. Zagier, Modular forms whose Fourier coefficients involve zeta functions
of quadratic fields, in "Modular functions of one variable, VI (Proc. Second
Internat. Conf., Univ. Bonn, Bonn, 1976)," Lecture Notes in Mathematics,
627. Springer, Berlin, 105-169, 1977.
p-ADIC PROPERTIES OF VALUES

OF THE MODULAR j-FUNCTION*
Ken ONO
University of Wisconsin, Madison,
Wisconsin 53706 liSA
onolDmath.wisc.edu
Matthew A. PAPANIKOLAS
Brown University, Providence,
Rhode Island 02912 USA
mapIDmath.brown.edu
1. Introduction and Statement of Results

As usual, let q := e27riz and let j(z) be the classical modular function
L
00
j(z) = c(n)qn = q-l + 744 + 196884q + ....

n=-l
The values of j(z) and its coefficients play many important roles in math-
ematics. For example, its values generate class fields and its coefficients
appear as dimensions of a graded representation of the Monster via the
Moonshine phenomenon. In a recent paper, Kaneko [K] produced an
interesting connection between the values of j(z) at Heegner points, the
so-called singular moduli, and its coefficients. Using recent formulas of
Zagier, he systematically expresses the coefficients c(n) in terms of sin-
gular moduli. In this paper we want to illustrate some peculiar p-adic
The first author thanks the National Science Foundation, the Alfred P. Sloan Foundation,
the David and Lucile Packard Foundation, and an H. I. Romnes Fellowship for their gen-
erous research support. Both authors thank the referee for his/her helpful suggestions and
comments. The first author thanks Professors K. Hashimoto, K. Miyake and H. Nakamura
for their kind hospitality during the 1999 and 21)()1 meetings. These were two wonderful
conferences!
[received: February 6, 2002; accepted in revised from: May 16, 2002]
358 K. ONO AND M. A. PAPANIKOLAS
properties of certain values of the j-function in connection with class

numbers and the integer 720.
To motivate our first result, we begin by recalling some classical for-
mulas for values of the Riemann zeta-function at negative odd integers.
To state these formulas, we begin by recalling some standard facts and
notation. If k ~ 4 is even, then let Ek(Z) denote the usual weight k
Eisenstein series for the full modular group 82('1.)
As usual, Bk is the k-th Bernoulli number and O'k-l(n) = Edlndk-1,

and H Ir denotes the usual fundamental domain of the action of the
modular group r = 82('1.) on the upper half of the complex plane. If
k ~ 4 is even, then
((1 ~ k) = 60k - L eTordT(Ek(z))j(r). (1)

TEHjr
If r E H Ir, then eT is defined by
if r = i,
if r = e27ri / 3 , (2)
otherwise.
These formulas follow from Euler's formula for ((1 - k) and the Fourier
expansion for Ek(Z). For example, consider the case where k = 10. The
divisor of ElO (z) is supported by simple zeros at r = i and e27ri / 3 , and
so the fact that j(i) = 1728 and j(e 27ri / 3 ) = 0 confirms (1):
_2_ = -264 = 600 _ j(e 27ri / 3 ) _ j(i).

((-9) 3 2
Here we prove a p-adic analog of (1), where the values of the Kubota-
Leopoldt zeta-function at s = -1 are given by a p-adic limit of certain
precise expressions in the values of the j-function.
Theorem 1. If r E Hlr is a point for which j(r) E Q, and p S; 7 is a
prime for which
j(r) == {o
6
(mod p)
(mod 7)
if P ::; 5,
if p = 7,
p-adic properties of the values of j 359
then as p-adic numbers we have
~-.
2
= -- -
744.
hm
( n pG_l . (pn-a r
"" "" J
+ b))
(;(-1) p -1 n-++oo ~ ~ pa .
Note the similarity between (1) and the p-adic formulas appearing in
Theorem 1. On the left hand side we have a zeta-value, and on the
right hand side we have an expression involving values of the j-function.
In (1) these expressions are traces of j-values over the divisor of an
Eisenstein series, and in Theorem 1 these expressions are p-adic limits
(as n -+ +00) of the 'traces' of j-values over the images of r under the
upper triangular Mobius transformations with determinant pn.
There are some important differences. First of all, the formulas in
Theorem 1 only pertain to those zeta values at s = 1 - k = -1 (i.e.
where k = 2). This is a byproduct of the fact that Theorem 1 is a
statement about the arithmetic of p-adic modular forms of weight k = 2.
Furthermore, Theorem 1 does not appear to be a statement related to
the arithmetic of Eisenstein series. However, this is not true. The proof
of Theorem 1 requires the arithmetic of the Eisenstein series (at least
E4 and E6) in a vital way.
Using the fact that (;(-1) = -(p - 1)/12, Theorem 1 may be inter-
preted as a curious collection of p-adic formulas for the integer 720:
0= 720+ (P-l) n~oo (~P~'j (pn-; +b)). (3)
Example. To illustrate Theorem 1, let r = i. Since j(i) = 1728 == 0

(mod 2), let p = 2 in Theorem 1. Formula (3), which is equivalent to
Theorem 1, asserts that if Hn is defined by
then Hn tends to zero 2-adically. If n = 0, 1, ... ,5, then

Ho = 720 + j(i) = 2448 = 24.32 17,
Hl = 25 32 . 5 . 401,
H2 = 26 .32 5 167341927,
H3 = 27 .3 2 .131.89604100069763687,
H4 = 28 32 11 .... ,
Hs = 29 . 32 124909 .....
Combining similar arguments with recent work of the first author and
Bruinier and Kohnen [Th. 9, B-K-O], one obtains the following p-adic
formulas for class numbers H( -D) of quadratic orders with discriminant
-D. First we recall the notion of a Heegner point. A complex number
T of the form T = -b+~ 2a with abc
' , E Z , gcd(a , b, c) = 1 and
b2 - 4ac < 0 is known as a Heegner point, and its discriminant is the
integer d-r := b2 - 4ac.
Theorem 2. Suppose that - D < -4 is a fundamental discriminant of

an imaginary quadratic field, and let T E H/r be a Heegner point of
discriminant - D. If K = Q(j (T) ), then the following are true:
(1) If D == 3 (mod 8), then as 2-adic numbers we have
(2) If D == 1 (mod 3), then as 3-adic numbers we have
(3) If D == 2,3 (mod 5), then as 5-adic numbers we have
(4) If D == 1,2,4 (mod 7), then as 7-adic numbers we have
H(-D) = -120
1
lim TrK/Q "
- n-+oo "j
(n 7a_l (7n-aT+b)) .
~ ~ 7a
a=O b=O
Example. Here we illustrate Theorem 2 using the Heegner point T =

(3 + A) /2, using the classical evaluation j (T) = -153 . Since H( -7) =
1, Theorem 2 (3) implies that
tends to zero 5-adically IlS n --+ +00. For n = 0,1,2 and 3 we find that
10 = -3195 = _32 . 5 . 71,
11 = -32. 52.11 2 .41118261707401,
r, 3
12 = -3.0 . 5 . 1123 .... ,
13 = _32 . 54 . 109 .... .
Example. For another example, we consider the Heegner point T =
(1 + J-15)/2, using the evaluation (see [II.6, SiD
jeT) = -191025 - 85995v'5.

2
Now H( -15) = 2, and so Theorem 2 (4) implies that
I n := 240+ 'IrQ( "")/'1 (~ ~ j C"-~: +b) )

tends to zero 7-adically as n --+ +00. For n = 0,1 and 2 we find that
Jo = -190785 = -3572379,
J1 = -3 . 5 . 72 . 3678217 .... ,
J2 = _22 . 3 . 5 . 73 ...
In the next section we prove Theorem 1, and we sketch the proof of
Theorem 2.
2. Proof of Theorems 1 and 2

We begin this section with two essential facts. First of all, we con-
struct holomorphic modular forms with prescribed divisor whose Fourier
expansions are trivial modulo primes p ~ 7. For our purposes, it suffices
to observe that
L u3(n)qn,
00
E4(Z) = 1 + 240 (4)

n=1
00
E6(Z) = 1- 504 Lus(n)qn. (5)

n=1
Proposition 2.1. Suppose that T E H/r is a point for which jeT) E Q.

If p ~ 7 is a prime for which
jeT) == {o
6
(mod p)
(mod 7)
if p ~ 5,
if P = 7,
then there is a holomorphic integer weight modular form with p-integral

coefficients, say r,p(z), for which r,p(T) = 0 and
r,p(Z) == 1 (mod p).
Proof. D~fine r,p(z) by
r,p(z) := .6.(z)(j(z) - jeT)). (6)
Here .6.(z) is the classical cusp form
.6.(z) = q nO (1 _ n)24 = E4(Z)3 = E6(Z)2. (7)

n=l q j(z) j(z) - 1728
It follows that r,p(z) is a holomorphic modular form of weight 12 for
which r,p(T) = O.
If p ~ 5 is prime, then the desired congruence follows from (4), (6)
and (7). If p = 7, then (5), (6), (7) and the fact that 1728 == 6 (mod 7)
implies the desired congruel!ce. 0
Now we recall some important facts about Ramanujan's Theta-
operator in connection with the values of the j-function. Ramanujan's
Theta-operator is the differential operator defined by
e (t. a(n)qn) := t. na(n)qn (8)
It is a classical fact that if fez) = 2:~=h a(n)qn is a weight k meromor-

phic modular form on SL2(7l..), then
(9)
where j is a meromorphic modular form of weight k+2 on SL2(Z), Here
E2 denotes the weight 2 quasi-modular form
00
E 2 (z) = 1- 24 LO'l(n)qn.
n=l
Although 8 is simple to define, its arithmetic nature is deeper and
is dictated by the j appearing in (9). There is an explicit formula for
8(f) in terms of a natural sequence of modular functions jm(z). Let
jo(z) := 1, and for every positive integer m let jm(z) be the unique
modular function whose Fourier expansion is of the form
00
jm(z) = q-m + L em(n)qn,

n=l
which is also holomorphic on the upper half of the complex plane. Notice
that if m is a positive integer, then a convenient description of these
functions is given by using the normalized weight 0 Hecke operators
T(m):
jm(z) := (j(z) - 744 I T(m)) = L

ad=m,d>O
(j (az: b) - 744). (10)
O:::;b:::;d-l
The first few jm are:

jO(z) = 1,
i1(z) = j(z) - 744
= q-l + 196884q + ... ,
h(z) = j(z)2 - 1488j(z) + 159768
= q-2 + 42987520q + ... ,
jg(z) = j(z)3 - 2232j(z)2 + 1069956j(z) - 36866976
= q-3 + 2592899910q + ....
Each jm is a monic degree m polynomial in j with integer coefficients.

The conclusions of the next theorem may be found in [A-K-N, B-K-O]:
Theorem 2.2. For every point T E Hlr, define Hr(z) by
00
HT(z) := Ljn(T)qn.
n=O
(1) For every point T E Hlr, we have
HT(z) = El(z)E6(Z) . 1 .
A(z) j(z) - jeT)
In particular, HT is a weight 2 meromorphic modular form.
(2) If f = E~=h af(n)qn is a nonzero weight k meromorphic modular
form on 8L2(Z) for which af(h) = 1, then
9(f) = k~;f - f fa,
where fa is defined by
fa:= I: eTordT(f)HT(z).
TEH/r
364 K. DND AND M. A. PAPANIKDLAS
Corollary 2.3. If 7 E H/r, then
8\j(z) -.j(7)) = -HT(z) = -1- fjn(7)Qn.

J(z) - J(7) n=l
Therefore, the series appearing in Corollary 2.3 is a meromorphic

modular form of weight 2. The following theorem of Serre [Th. 7, Se] is
vital for all of the results in this paper.
Theorem 2.4. If P :::; 7 is prime and
n=O
is a p-adic modular form of weight k =1= 0, then

~'*(1 - k)
a(O) = p. lim a(pn).
2
n---+oo
Proof of Theorem 1. If s is a non-negative integer, then consider the
modular form
P ()._ ()p8. 8(j(z) - j(7)) (11)
T,S Z .- T,p Z .().().
JZ - J 7
By Proposition 2.1 and Corollary 2.3, PT,s(z) is a holomorphic modular

form for which
(12)
Furthermore, PT,s(z) has weight 2 + 12 . pS. By letting s ---t +00, we
find that HT(z) is a p-adic modular form of weight 2. The conclusion of
Theorem 1 now follows from (10) and Theorem 2.4. D
Sketch of the Proof of Theorem 2. Let 71,72, ... 7H(-D) E H/r be the
Heegner points of discriminant -D. Then let FD(Z) be the modular
function defined by
H(-D)
FD(Z):= II (j(z) - j(7t)). (13)
t=l
Gross and Zagier proved that if [Cor. 2.5, G-Z] 7 is a Heegner point
with discriminant fir, then
IdTI == 3 (mod 8) ==} j(7) == 0 (mod 215 ),
Ifir I == 1 (mod 3) ==} j(7) == 1728 (mod 36 ),
IdTI == 2,3 (mod 5) ==} j(7) == 0 (mod 53),
IdTI == 1,2,4 (mod 7) ==} j(7) == 1728 (mod 72 ).
Arguing as in the proof of Theorem 1, for every integer 8 the holomorphic

modular form QD,s(Z) defined by
8(FD(Z)) H(-D)
II
8
QD,s(Z):= FD(Z) . ert,p(z)P

t=l
satisfies the congruence
_ 8(FD(Z)) H(-D)
QD,s(Z) = F () = -
D Z
Hrt(z) L
t=l
(mod pS+l).
By letting 8- +00, we find that S};n<i)l) is a weight 2 p-adic modular

form. Since the numbers j(Tt) are algebraic integers which form a com-
plete set of Galois conjugates over 10, the result follows from Theorem
2.4. 0
Remark. It would be interesting to see a direct and elementary proof
of the p-adic convergence of the limits appearing in Theorems 1 and 2.
References
[A-K-N) T. Asai, M. Kaneko and H. Ninomiya, Zeros of certain modular functions
and an application, Comm. Math. Univ. St. Pauli 46 (1997) 93-101.
[B-K-O) J. Bruinier, W. Kohnen and K. Ono, The arithmetic of the values of modu-
lar functions and the divisors of modular forms, Compositio Mathematica,
accepted for publication.
[G-Z] B. Gross and D. Zagier, On singular moduli, J. reine angew. math. 355
(1985) 191-220.
[K) M. Kaneko, Traces of singular moduli and the Fourier coefficients of the
elliptic modular function j(T), in "Number theory (Ottawa, ON, 1996),"
173-176, CRM Proc. and Lecture Notes, 19. Amer. Math. Soc., Providence,
RI,1999.
[Se] J.-P. Serre, Formes modulaires et fonctions zeta p-adiques, in "Modular
functions of one variable. III (Willem Kuyk and J.-P. Serre, ed.)," 191-
268, Lecture Notes in Mathematics, 350. Springer-Verlag, Berlin-New York,
1973.
lSi] J. H. Silverman, "Advanced Topics in the Arithmetic of Elliptic Curves,"
Springer-Verlag, New York, 1994.
THOMPSON SERIES AND

RAMANUJAN'S IDENTITIES
Masao KOIKE
Graduate School of Mathematics
Kyushu University,
Pukuoka, 812-8581 Japan
[email protected]
Introduction
B. J. Birch published some manuscripts of Ramanujan which con-
tained, among other things, a list of identities involving Rogers-
Ramanujan functions
G(x) = 11 00 1
(1 - x5n +l )(1 - x 5n +4) ,
H(x) = 11 00 1
(1 - x5n+2)(1 - x5n+3) ,
and
IT (1 - xmn)
00
q(m) =
n=l
which is related to Dedekind 1] function
IT (1 - e2m
00
1](7) = e 2;!T l'iT).

n=l
A typical one, the identity (2) in [1], reads as follows:
[received: October 25, 2001; accepted in revised form: September 9, 2002J

368 M. KOIKE
These identities involve the functions G(x) and H(x) on one side and
the products of q( m) for various m on the other side. All of them except
one are already proved. (For the history of proofs, see [1], [2]). In [1],
Birch wrote that these identities of Ramanujan seemed too complicated
to guess, E"ren for one with Ramanujan's incredible instinct for formulae.
In this paper, we shall propose an idea which gives us an easier way to
find, but not to prove, some of such complicated identities, moreover
new identities which Ramanujan missed. The idea is to connect these
results with Thompson series. Thompson series Tg is the normalized
generator of a genus zero function field parametrized by the Monster
M, the largest sporadic finite simple group. In [3], Conway and Norton
gave all products for Tg+constant expressible in terms of TJ(kT) for var-
ious k. We observed that many products of TJ(kT) for various k which
are involved are common in the both identities. Therefore we try to
find many identities between Thompson series and Rogers-Ramanujan
functions directly by using computers.
1. Preliminaries
G (x) and H (x) can be expressed as theta functions by using the Jacobi
triple product identity. Put
g(T) and h(T) are modular funtions on rl(5) with some character. We
define
F;!s, = g(rT)g(sT) + h(rT)h(ST), for r + S == 0 (mod 5)

and
Fr-:s = g(rT)h(sT) - h(rT)g(sT), for r - S == 0 (mod 5).
Then the leading term of Fourier expansion of F/s ' Fr- s is x W , X l1~or
respectively. Put N = rs. N is called the level ~f Fr,~. Since rand S
satisfies r + S == 0 (mod 5) or r - S == 0 (mod 5), the level N satisfy
N == 1 or 4 (mod 5). Moreover we suppose that N is an order of any
element of the Monster. Then N is in the set
s= {4, 6, 9,11, 14, 16, 19, 21, 24, 26, 29, 31, 34, 36, 39, 41,} .
44, 46, 51, 54, 56, 59, 66, 69, 71, 84, 94, 104, 119
By observing the list of Ramanujan's identities in [IJ and Table 3 and

Table 4 in [3], we could make a guess that if N is in the set S, then there
exists an element 9 of order N in the Monster such that Thompson series

Tg+constant can be expressed as a product of some Rogers-Ramanujan
functions of level N.
We recall several notation from Conway and Norton [3]. For every
gEM, Thompson series T g :
Tg = x.- 1 + 0 + HI (g)x + H2(g)X 2 + ... ,

where Hn are representations of M and Hn(g) are character values for
every gEM.
Let N be a positive integer. Let e be a Hall divisor of N, namely,
(e, N/e) = 1. Then the Atkin-Lehner involution We = (: :e).
where a, b, c and d are integers and ade 2 - beN = e, is in the normalizer
of ro(N). We denote by N +e, f, the group obtained from ro(N) by
adjoining Atkin-Lehner involutions We, WI, . Especially we denote
by N + (resp. N + N) the group obtained from ro(N) by adjoining all
its Atkin-Lehner involutions (resp. by adjoining the particular Atkin-
Lehner involution WN, which is also called the Fricke involution). If
Thompson series Tg is the normalized generator of the modular function
field with respect to the Fuchsian group N + e, j, ... , Tg is also denoted
by TN+e,J,o oo
Then we notice that the above guess is not true for N = 11, 16, 84, 94
and 104. When N = 11, there exists ol!ly one Rogers-Ramanujan's
function FIll oflevel 11 which is equal to 1. When N = 16, there exists
two Roger-Ramanujan's functions FiB I' Ft2 of level 16. Then it holds
that (Fi61)2 = Ft2 and (F:2)6 = TSB.' Wh~n N = 84, 104 the Fuchsian
group associated ~ith TN A is not of the above form N + e, f, ... When
N = 94, there exist two Rogers-Ramanujan functions F9tl and F47 ,2 .
The leading term of Fourier expansion of F~ I (resp. F47 2) is x-H (resp.
5 ' ,
x- I2 ) , so there is no simple linear combination 19n + 5m = 12,which is
different from others.
So our conjecture is
Conjecture 1.1. If N is in the set
= {4, 6, 9,14,19,21,24,29,26,31,34,36, 39,},

41,44,46,51,54,56,59,66,69,71,119
then there exists an element 9 of order N in the Monster M such that
Thompson series Tg+constant can be expressed as a product of some
Rogers-Ramanujan functions of l01 el N.
370 M. KOIKE
2. Results
The more precise statement of Conjecturel.l is given by the following
list of formulae. We checked that the first fifty powers of x in the fol-
lowing formulae are correct. The first fifty coefficients of all Thompson
series are given in [4]. The s~'mbol acx tJ3 ... /c i d8 ... means the function
'f/(ar)cx'f/(br)!3 ... /'f/(cr)'Y'f/(dr)8 ....
Level Formulae
2 48
4 (F4+,1 )12 = TH + 24 = ~
6 (F3+,2) 12 = T6+6 + 12 = '[1'26!"!

2 12 3 12
9 ( +)6
F9,1 = T9+ + 6 = 93"
13
+ 2713"
93
+9
14 ( + )4
F14 ,1 = T14+14 + 4 = 14i:44
24 74
19 (Fi'g,1)3 = T19+ + 3
3 2 72 1221 2
21 ( +
F7,3)
6
= T21+ = ~ +~ - 2
(p,+ )2p,+ = To24+24 + 2 = 124.6.242

2 2
24 24,1 6,4
2.3 8 12
2 2
26 ( +
F 13,2)
4
= T26+ = ~
2213
+~
1226
- 2
29 (Fig,1)2 = T29+ + 2
31 (F31 ,1)3 = T31+
34 (F~ ,IFi"1 ,2)2 = T3H + 2
F-
36 ~ -7',
F- - 36+36
+ 1-
-
4 .9
1.36
9.4
Level Formulae
F+
39 3~,1
F 13 ,3
= T39+39 + 1 = ~
1.39
41 (F41 ,1)2 = nl+
+ + _ _ 24224
44 F44 ,I F 11 ,4 - n4+ + 1 - 1242 112442 - 1
46 F-
46,1 p,+
23,2 -- T.46+ -- '2."46
1.23 + 2 I:'23
2 . 46 + 1
51 F51,IFi~,3 = Tsl+
54 F~,IF27,2 = T54+ + 1
56 p,- p,+ -
56,1 8,7 -
1]
56+ -
- 2.4.14.28
T.7:8.56 -
1
59 Fit,1 = T 59+ + 1
66
F+
~~
F-
F+
F-
-- T.66+6,11,66 + 1 -- 2.3.22.33
T:6.iT."6li
11,6 66,1
69 F6't, 1 F2'3,3 = Ts9+ + 1
71 F 71 ,1 = T71 +
119 Fii9,lFi7 ,7 = T 1l 9+ +1
In the above list, Formulae for N = 4,6, 14,39,44,46,66 follow from

Ramanujan's identities in [1]. We prove the case N = 4 for example.
The identity (2) in [1] which is given in the Introduction is equivalent to
372 M. KOIKE
1
Dividing both sides by xI2 i3.nd taking 12-th power of both sides, we get
the proof.
There exist relations between Rogers-Ramanujan functions of the
same level as shown in the next list. Formulae for N = 6, 14, 21, 24
are contained in the list of Ramanujan's identities [1]. But there exist
more relations. Even in the case that the level N is not contained in the
set S in Conjecture 1.1, relations between Rogers-Ramanujan functions
of the same level exist. For example, let N = 94. The leading term of
Fourier expansion of the product F~ 1 F47 2 is x- 2 , so this may be equal
to the polynomial of T94+ of degree' 2, ~amely, T~4+ + T94+ - 1. We
checked that the first fifty powers of x are correct.
Level Formulae
6 F::2F6~1 = 1
14 Fi~,1 Fi,2 =1
16 (Fi6,1)2 = F S,2+
21 Ft3
, = F211,
24 FB:"3F~,1 = 1
26 Fit ,2 = F26I
,
34 (Ft,,1)3 (Fi7 2)3 _ (T )2 3T

(F- )3
17,2
+ (F-F)a
34,1
- 34+ + 34+
36 FIt ,3F it 2, Fi6 1, = T36+36
39 (Fig,l)3 = (T39+39? + 3T39+39 + 2
44 (F4tl)4 + (F~ ,4)4 = (T44+)3 + 4(T44+? + 8T44+ + 6
51 (Fsl,I)3 - (Fib)3 = (TS1+)2 - Ts1+ + 1

Level Formulae
66 Fi3,2 Fi;,3 = Ts6+6,1l,66

F-
F.+ -....Qd
84,1 - F-
12.7
84
F+ F-
-;!f1-
F214
= -2!d
Fi; 7
94 F9t1F47,2 = (T94+)2 + T94+ - 1
96 F96,1 = F4t.2 F3t3
104 Fit4 "1Fi3 8 = F522,
References
[1] B. J. Birch, A look back at Ramanujan's Notebooks, Math. Proc. Camb. Phil. Soc.
78 (1975), 73-79.
[2] A. J. F. Biagioli, A proof of some identities of Ramanujan using modular forms,
Glasgow Math. J. 31 (1989), 271-295.
[3] J. H. Conway and S. P. Norton, Monstrous Moonshine, Bull. London Math. Soc.
11 (1979), 308-309.
[4] J. Mckay and H. Strauss, The q-series of monstrous moonshine and the decom-
position of the head characters, Comm. in Algebra 18 (1990), 253-278.
GENERALIZED
RADEMACHER FUNCTIONS
AND
SOME CONGRUENCE PROPERTIES
Hiroaki NAKAMURA
Faculty of Science,
Okayama University, Okayama
700-8530, JAPAN
h-naka@math .okayama-u .ac.jp
Abstract In 1, we introduce the classical Dedekind sum and the Rademacher

function. Then, in 2, certain generalized Rademacher functions are
introduced as the Eichler-Shimura type period integrals of Eisenstein
series. In 3, we present a version of continued fraction algorithm
which computes efficiently the generalized Rademacher functions . In
4, we show a congruence formula connecting values of the generalized
Rademacher functions of weight 2 and weight k > 2. This formula will
be applied in a forthcoming paper [N4J.
1. The classical Dedekind sums

The classical Dedekind eta function
IT (1- qn)
00
1'/(r) = ql/24 (1.1)
n=l
on the upper half plane S) is one of the most beautiful objects in num-
ber theory. Since b.(r) = (21l'i)121'/24(r) is a cusp form of weight 12
for SL2(Z), it is clear that, for any A = (~~) E SL2(Z), the ra-
tio 1'/(Ar)/v crt d1'/(r) is a 24-th root of unity. The behavior of this
little ratio with respect to A E SL2(Z) looks delicate and mysteri-
[received: June 18, 2002; accepted in revised form: October 18, 2002J
376 H. NAKAMURA
ous. R. Dedekind, in his note [D] included in Gesammelte Mathe-

matische Werke of B. Riemann, studied the behavior with Riemann's
method. Outstanding is that he found an explicit integer valued func-
tion 'P : SL2(Z) - t Z controlling this ratio in the form
e 22~icp(A)7](T), (c = 0);
7](AT) =
{
e 22~icp(A) Jcrtd7](T), (c > 0).
(1.2)
Dedekind's formula for the function 'P (which, by definition, factors

through PSL 2 (Z) - t Z) reads as follows:
(A) = {~' d (c = 0),

'P at - 12s(a, c), (c> 0).
(1.3)
Here s(a, c), called now the Dedekind sum, is defined by

c-I . .
~ z az
s(a,c) = L..,.PI(-)H(-)
i=O c c
with PI (*) the "sawtooth" function:
PI(x) =
{oX - [x]- ~
(x E Z);
(x Z).
([x]: the greatest integer not exceeding x E R)

In [R], H. Rademacher intensively studied algebraic properties of the
function 'P. He derived, for example, the composition formula
(A, BE SL2(Z)), (1.4)
where, for any matrix S E SL2(Z), Cs denotes the lower left entry of S.
Following [KM], we shall call 'P the Rademacher ('P- )function.
One of the key ingredients that enabled Dedekind to find the above
result was to consider the modular transformation of not only 7](T) but
also of log7](T). In fact, Dedekind's key formula (for the case c > 0)
reads:
1 i a+ d
log7](AT) -log7](T) = -2 log CT + d + 7ri( 12c - s(a,c)). (1.5)
This left hand side may also be written in the form:
log7](AT) -log7](T) = -7ri l T

AT
E2(Z)dz, (1.6)
where
is the Eisenstein series of weight 2 (with specified conditional conver-

gence). In other words, the Rademacher function cp can be viewed almost
as the periods of the Eisenstein series. Unfortunately, E2(7') is only a
quasi-modular form and not a usual modular form. This causes the need
of the adjustment term ~ log CT~d of (1.5) which connects cp(A) and the
period. To gain a more satisfactory viewpoint, we would need a more
subtle treatment by using a non-holomorphic modification of E2. How-
ever, we do not enter this subtlety any more here, because this sort of
adjustment is rather exceptional phenomenon occurring in this original
case among generalizations of E2 discussed in the next section.
2. Generalized Rademacher functions

We wish to consider a generalization of the Rademacher function cp in-
troduced in 1 by considering cp as the case of weight 2 and level 1. With
regard to the appearance of E2 in (1.6) , we shall begin by introducing
the generalized Eisenstein series Ekx1,xz),s (k 2: 2, (Xl,X2) E (-kZ/Z)2)
given by
(2.1)
where Pk(x) : 1R -+ 1R is the k-th periodic Bernoulli function (whose

Ek
definition will be recalled in (3.4. It is well known that x ) is a holo-
morphic modular form of level N and weight k, except for the original
case E2 = E~O) (cf. [StD.
Let r(N) c SL2(Z) be the principal congruence subgroup of level
N 2: 1. For any pair of k 2: 2 and x = (Xl, X2) E (-kZ/Z)2 with
378 H. NAKAMURA
(k, x) 1= (2,0), we then consider the Eichler integral of Ekx ):
F~x)(r) = - (k ~ 2)! l ioo

(EkX)(U) + Pkr d ) (r - u)k- 2du
Pk(Xl) r k- 1
(2.2)
k (k-l)!"
This (indefinite) integral satisfies (d~ )k-l F~x) = E1x). Writing

j(A, r) := (cr + d) for A = (~~) E r(N), we then see that the dif-
ference
4>~x)(A)(T) := j(A ,T)k-2 F~X)(AT) - Ft)(T) (2 .3)
is a polynomial in T of degree k - 2. What we wish to look closely at here
is its "real part" Re4>1X)(A)(r) which is, by definition, the polynomial in
the "complex variable" r whose coefficients consist of the real parts of the
corresponding coefficients of 4>ix ) (A) (r). It is known that Re4>ix ) (A) (r)
has only rational coefficients.
The assignment A I-> Re4>ix)(A)(r) enjoys the l-cocycle property of
the form:
Re4>iX )(AB)(r) = j(B, T)k-2 Re4>iX ) (A)(Br) + Re4>iX ) (B)(T) (2.4)
on A, B E r(N).
To switch Re4>ix)(A)(r) into the form of l-cocycle for a left module,
let us introduce the space Sym k- 2(Q2) = Q[X, Y]deg=k-2 of the homo-
geneous polynomials in X, Y of degree k - 2 on which SL2(Z) acts on
the left (written p) by
p (~ ~). 4>(X, Y) := 4>(aX + cY, bX + dY).
Definition (2.5). Let (k, x) E Z~2 X (-1Z/Z)2 with (k, x) 1= (2,0).

We define the generalized Rademacher function
~~k): r(N) ~ Sym k - 2 (Q2) = Q[X, Y]deg=k-2
(A = (~ ~)~. ~ 4'l;+1,k-'-'\A)X- y k-2-_)

by
As expected from the above definition, (2.4) can be converted into the
following l-cocycle property of ~~k):
(2.6)
on A, B E r(N). But one can get a nicer result. In fact, G. Stevens [St]
showed how to interpret ~~k) as the periods of a certain real differential
form extended to the Borel-Serre compactification i:J of the upper half
plane 5). With his method, we can extend ~~k) canonically to a function
from SL2(Z) (or even from GL2(Q)+) to Sym k - 2(Q2). Moreover the case
of (k, x) = (2,0) may naturally be included in this unified construction
of <p~k) with regarding <p~2) = -cp(A)/12. Then, the above properties
(1.4) and (2.6) can be generalized to
<P~k)(AB) = <p~k)(A) + p(A).<p~l(B) + l8~=2sgn(cAcBcAB)' (2.7)
Here, sgn(*) E {1,0} denotes the signature of *, and the Kronecker

symbol 8~=2 is defined by 8~=2 = 1 when (k, x) = (2,0) and 8~=2 = 0
otherwise.
It will also be nice to have generalizations of the beautiful formula
(1.3) to the cases of our generalized Rademacher functions <p~k). Al-
though the author could not find a literature describing the explicit
form, by careful calculations one obtains the following formula of <p~k) :
PSL2(Z) -+ Symk - 2(Q2) = Q[X, Y]deg=k-2;
<P~~~'X2) ((~ ~)) =

b
-Pkfd Jod(tX + y)k- 2 dt, (c = 0);
1
- Pk x l) Jo~ (tX + y)k- 2 dt
Pk( ax t cx2 ) J~g (t(aX
c
+ cY) + bX + dy)k- 2 dt
k-2
+ 2: (-1r (k;2)xr(aX + ey)k-2-r s((:-;-)r,r+l) (a, c), (c> 0).
r=O 1, 2
(2.8)
where the last factor (called the generalized Dedekind sum) is defined
by
(k-l-r,r+l)(
s( ) a,e -
) _ L
c-l n
.rk-l-r
(x!+i) n
c
(
.rr+l X2 + a X!+i)
c
. (2.9)
Xl,X2 k-1-r r+1
i=O
380 H. NAKAMURA
The following distribution relation is an important property which con-

nects values of <p~k) (A) of different x's:
<p~k) (A) = n k- 2 I: <p~k) (A) (n ~ 1). (2.10)

YE~x
3. Continued fraction algorithm for ip~k)
The calculation of <p~k) (( ~ ~)) by means of (2.8) involve those of gen-

eralized Dedekind sums s~k-I-r,r+1) (a, e) of (2.9) which need lots of times
when e is big. Variants of continued fraction algorithms have been known
to improve this sort of inefficiency on computers. We shall exploit a
version of it well suited to calculations of our generalized Rademacher
functions. The implementation and various numerical tests have been
given by Y. Morimoto [Mol by using PARI-GP and Maple softwares.
Given a matrix A = (~~) E SL2(Z), choose an expression of ale by
a continued fraction of the form:
all 1
- = k o - - - ... -
e kl - k2- -kn
1
= ko - - - - - - - 1 - - -
kl --------
(3.1)
1
k2------
1
kn-I - t

wi th an integer sequence (ko, ki , ... , kn ). This expression may be ob-
tained by first putting ro = a, rl = e > and then defining integers
{kdi with the Euclidean method:
ro = kOrl - r2, Iril > Ir21 > 0,

rl = k I r 2 - r3, Ir21 > Ir 31 > 0,
(3.2)
rn-I = kn-Irn - rn+l, Irnl > Irn+11 = 1,
rn = knrn+1 - 0, Irn+11 = 1, rn+2 = 0,
where each k i (i = 0, ... ,n) is chosen to be either h/ri+1l (the greatest

r
integer not exceeding rdrHd or rdrHll (the least integer greater than
or equal to rdrH1). Let us write E = rn+1 E {1}.
Define also a sequence of matrices {Adi in SL2(Z) by
A-I = (~ ~), Ao = (;0 ~1), Al = (;0 ~1) (;1 ~1),
and put
(-1 S; is; n).
Then,
Lemma (3.3). Notations being as above, we have
(i) a = -cpn, C = -cqn;

(ii) 6:= (d - cqn-l)/C E Z;
''') (aC db) =c (ko1

(111 -1) 1 -1)
0 ... (kn 0 (10 6)
1 .
Proof. The Euclidean sequence (3.2) may be rephrased as
( a)
C
= (ko (kl (kn
1010"'10
-1) -1) -1) (rn+l)
O
The claim (i) follows immediately from this. Both A and cAn are in
SL2(Z) and have the same left column. Therefore, there exists 6 E Z
with A = cAn 00, i.e.,
The claims (ii) and (iii) follow from this simultaneously. o

Definition (3.4). The Bernoulli polynomials Bk(X) (k = 0,1,2, ...)
are defined by the generating function
and the Bernoulli number Bk is by definition Bk(1). We define the

periodic Bernoulli function Pk : R ~ IR by Pk(x) = Bk(X - [x)) except
382 H. NAKAMURA
for the case k = 1 and x E Z where we set P1(n) = 0 (n E Z). For

x = (Xl,X2) E Q2 and (a,b) E Z~o, define
JDa(Xl) JDb(X2)
f3a,b(X) := max{ a, I} . max{b, I} .
Proposition (3.5) (Continued fraction algorithm). Notations

being as above, put kn+l := 8. Then, for x = (Xl,X2) E Q2 and
A = (~~) E SL2(Z), we have
..!:') (A) = - t. ~ Pj +1,k-l-j (xA,)

.
(k ~ 2)
k 2 .
(Pi-IX + qi-IY)J(PiX + qiY ) -
X -J
n+I
- :Lf3k,o(xAi -d
i=O
(ki k-2
X Jo (-Pi-I t +Pi-2)X+(-Qi-I t +qi-2)Y) dt
8k =2
+ T(s gn(Q-IQO) + sgn(QOQI) + ... + sgn(Qn-IQn)).
JDrooj. Applying (2.7) iteratedly to (3.3)(iii), we obtain
Then, the formula (2.8) gives
(k) (1 8)
q>xA n( 0 1 ) = - f3k,o(xA n) Jo
kn +1
(tX
r + Y) k-2 dt,
(k) (ki
q>xAi-l (1
-1) 0 ) = - f3k ,o(xAi-d Jor (tX + Y)
ki
dt k-2
+ I:( -1)j (k ~ 2)
j=O J
x Xj(kX
t
+ y)k-2- j s(k-l-
XAi-l
j ,j+I)(k 1).
t,
Writing here (Yi,Zi) := xA i- 1 (i = 0, ... ,n) so that
we see that
(k-l-j,j+l)(k . 1) = Pk-I-j(Yi) Pj+1 (Zi + kiYi)
sxAi_l t, k - 1- j j +1
= (_I)k-I- j ,Bj+l,k-l-j(xA).
Therefore,
Noticing that p(Ai - l ) maps
{~xS
~ -Pi-IX - qi-I Y,
~ Pi-2 X + qi-2 Y,
~ -PiX -qi Y,
we conclude the proof of Proposition (3.5). o
4. Some congruence properties

In this section, we shall present a congruence formula (4.3) which
connects N2 values of ~~2)(A) (x E (~Z/Z)2) and the coefficients of
~ ~k) (A). The congruence formula will be applied to the study of a certain
measure function on the congruence kernel of SL2 which governs the
meta-abelian monodromy representation in the fundamental group of
the universal elliptic curve minus the origin section ([NI-4]). We refer
to the forthcoming paper [N4] as a main reference for details of such an
application.
Let ZN C Q denote the ring of fractional numbers with denominators
prime to N. For n ~ 1, write B~(X) := Bn(X) - Bn(O), and define
384 H. NAKAMURA
dn (resp. d~) to be the least common multiple of the denominators of

coefficients of the polynomial ~ Bn (X) (resp. ~ B~ (X) ). In the proof of
Proposition (4.3) below, we shall frequently make use of the following
well known properties:
N-l N
L
x=o
xn = _l-B~+l(N) == 0 (mod -d* z~),
n+1 n+l
(4.1)
N:- 1
(Pn(~ + u) - Pn(~)) == x n- 1 (Pl(~ + u) - Pl(~))
(mod ~ Z'tv) (4.2)
for x E 'I., N E N, u E Z'tv.

Proof. The formula (4.1) is quite popular, and its proof may be left to
readers. To prove (4.2), let {t} := t - [t] denote the fractional part of
t E ~, and put v := tv
+ u} - {,V} so that {u} = {v}. Then, v E Z'tv
and we compute
Nn-l x x
-n-(Pn(N + u) - Pn(N))
Nn-l x x
= -n- (Bn ({ N} + v) - Bn ({ N }))
Nn-l
= _
- n
[(( {~} + v)n _ ~({ ~} + v)n-l) _ ({ ~}n _ ~{~
N 2N N 2N
}n-l)]
Nn-l [ x x n X]
=-n- n{N}n-1V+Pn-2({N})+'2Qn-2({N}) .
Z
Here, the congruence is taken modulo Z~, and Pn-2, Qn-2 are poly-
nomials of degree n - 2 with coefficients in Z. Since ~Bn(X) is a poly-
nomial of the form ~xn - ~xn-l + ... , it follows that dn is a multiple of
LC M {2, n}. Therefore, the second and third terms of the above last side
t.
vanish modulo 'lL,N' Moreover the integer x' := {x / N} N is congruent
to x (mod N). Thus, the above last side continues to
== (x')n-l v == xn-1v = x n- l (Pl(~ + u) - Pl(~))'
o
Proposition (4.3). Let N, r, k be integers satisfying N 2: 1, k 2: 2
and 0 ~ r ~ k - 2, and, for s = 1, ... , k - 2, let es be the denominator of
k(f!:s)' Define Dk,r to be the least common multiple of the set
{ dj+ldk-l-j (r ~ j ~ k - 2), el, .. ,ek-2 }

d;+l d'k_r' d;+2 d'k_l_r' d;+l d'k-r+l' d;+2 d'k_r' d;+3 d'k_l_r .
Then, for A E r(N), we have the following congruence:
This congruence formula was first proved by the author in the special
case of r = o. The general case of the formula was then conjectured by
Y. Morimoto after numerical computations and evidences [Mo] .
(4.4) Proof of Proposition 4.3. Let A = (~~) E r(N). Since

(t g~3) ( ~ ~) = (a+~gN3 b+~9N3), it is sufficient to prove the formula
when a ~ C > O. In what follows, we shall assume this. Applying (2.7)
to the equation (~ -l) ( ~a ~b) = (~ ~), we find
+ ~ X ryk-2-r(k-2)(_ly Pk-I-r(xd Pr+1(X2)
e
L., r k-l-r r+l
r=O
- ~(ax + cy)'yk-2-" ~ 2)
a-I p ( X2+i) n (X2+i)
X ( _l)r ~ k-I-r Xl + C a rr+l a .
L., k- 1- r r +1 (4.4.0)
i=O
Let us first evaluate the left hand side of the congruence formula (4.3).
When k = 2, r = 0, the above (4.4.0) gives
(2) C b
<Px (A) = 2a P2(X2) - 2aP2(axl + CX2)
a-I X2 +i X2 +i }
+ { Pl (XI)Pl (X2) - ~ PI (Xl + c-a-)PI (-a-) .
386 H. NAKAMURA
Since a == 1, b == c == 0 (mod N), for any integers x, y, P2( ax cy ) = t

P2(N) . Thus,
(4.4.1)
where
N-I N-I
So:= 12 L L xk-2-.ryr(2caP2(~) - ;aP2(~))'
x=O y=O
N-1 N-1
S1 := 12 L L Xk-2-ryrpI(~)PI(~)'
x=O y=o
N-l N-I a-I X 1L + i +i 11
S2:= -12 L L Lxk-2-ryrPI(N +c7)P1(-a-)
x=O y=O i=O
Recalling the notation B;(X) := Bs(X) - Bs(O), by (4.1), we immedi-

ately see
s _ 6c B k- r- I(N) ( B;+3(N) B;+2(N) B;+1 (N))

o - ~ k - r -1 N2(r + 3) - N(r + 2) + -6':-'(r~+-1-:-)
_ 6b ( B k- r+1 (N) _ Bk_r(N) + B k- r - 1(N) ) B;+1 (N)
a N2(k-r+1) N(k-r) 6(k-r-1) r+1
= 0 ( d GCD{6, N}N '7J1 )
- mo D IIJN .
k,r
(Note also here that b == c == 0 (mod N).) Let us rewrite S2. According
JL+'
to the decomposition of PI(~) as cl1v + (~- ~), we may write
(4.4.2)
where
Noticing that .ia - 12 E lZ'N

2 and that cY+jf
a E Z'N as a = - 1, c ==
o (mod N), we may apply (4.2) to the part "x k- 2- r PI(-f, + cU;jf)"
of 8 2 to get the congruence
12N ')
82 == Tl
/I
+ T2 (mod 2d ZN (4.4.3)
k-r-I
where
N-l N-l a-I Nk-2-r

TI := - 12 LLL
x=O y=O i=O
k - 1- r
X y + Ni X ) r i 1
X ( Pk-I-r(N+ C aN )-Pk-I-r(N) Y (0:-2)
N-I a-I
= - 12 L L k- ~-
y=O i=O
r
X (Pk-I-r(C Y +aNi ) - Pk-I-r(O) yr( - ~),

N-IN-la-l .
T2:= -12 L L Lxk-2-rpl(~)yr(': -~)
x=O y=O i=O a
N-IN-l 1
= -12 L L x k- 2- r pl(~)yr(-2)'
x=O y=O
388 H. NAKAMURA
Combining the above 51, 5~, T2, we obtain
Meanwhile, since
1 y + Ni 1
L L k -1- r Pk - l -r(C
N-l a-I r
-12 a )y (-2)
y=O i=O
6 Pk-l-r(O) B;+1 (N)
= ak - r k - 1- r r+1
and
N-l a-I ( ) B* (N)
- 1 2 " " _ Pk-l-r(O)yr(i _~) = _6 Pk - 1- r 0 r+1 ,
LJLJ k-1-r a 2 k-1-r r+1
y=O i=O
it follows that
y + Ni
_
Tl = -12 L L k -1-r
N-l a-I1
'0
Pk-l-r(C
a
i 6N
)y - (mod -D ZN)'
a kr
r I
y= 0 z= '
Summing up the above discussions with (4.4.2-3), we may rewrite (4.4.1)

as
modulo (k;2) GCD.J6,N}N 71./N . Next, we shall consider the right hand
k,r
side of the statement formula. Reading the coefficients of xryk-2-r of
(4.4.0) for (Xl, X2) = (0,0), we get
Generalized Rademacher junctions and some congruence properties 389
where
As b == c == 0 (mod N), it is easy to see U1 == 0 (mod k

LC:: e1,
-;:2
.. ,
~k- 2 ZN)'
Moreover, taking the equality (~) (k j 2) = (k;2) (j~;r) into accounts, we
'1 h U .h d 1 12(k;21N '7/1 (
eaSl y see t at 3 vams es mo u 0 LCM{dj+ldk_1_j r+l:5i:5k-2}/UN agam

as c == 0 (mod N)). Now, comparing the distribution relations

N-l N-l
~~+l,k-l-r)(A) = N k- 2 L L ~~l,~k)-l-r)(A),
x=O y=o
N-l N-l N-l N-l
Pk(O) = N k- 2 L L Pk ( ~) = N k- 2 L L Pk(ax; cY ),
x=o y=o x=o y=o
we obtain
r
12Nk- 2 (k;2)(_l {N-IN-l x Y
U2 = (k -1- r)'r + 1)
(
L L r
Pk-l-r(N)P +1(N)
x=O y=o
~~~ r X cy ci i+1!t }
- ~ ~ ~a Pk-l-r(N + aN + ~)Pr+l(-a-)
= 12(k -
r
2) (-l r { Pk-l-r(O)
k-1-r r+1
Pr+l(O)
N-la-l p. (Bl. ciN) p (Hi)}

_ ""' ""' ar k-l-r a + a Nr r+l a . (4.4.5)
L...JL...J k-1-r r+1
y=O i=O
390 H. NAKAMURA
We shall decompose the second term of the above last side according to
(aNY y i (aNY Y r i N 1
- - Pr + 1 ( - + -) == -'- - Pr +1(-) + y - (mod -71N ),
r+1 aN a r+1 aN a dr +1
which fo~Lws from (4.2) and 0 ::; !R + < 1. We compute:
a-l N-l n (y+iN .) D (..1L)

""' ~ rk-l-r a (. (aNf rr+1 aN
L.JL.J k-l-r r+l
i=O y=O
= ~ a-(k-2-r) Pk-l-r(O) (aNr Pr+1 (!R)

L.J k-l-r r+l
y=O
p, (0) N-l r+1 1
== a-(k-2-r) k-l-r ~ (_y _ _ ~ r)
(k - 1 - r) (r + 1);;0 aN 2 y
N
(mod dr
1 )
d 7l N .
+1 k-r-1
Noticing that the coefficient of X in B r+2(X) is (_1)r+l(r+2)Br+l and
that Pr +1 (0) = 0 if r + 1 is odd, the above last side continues to
= a-(k-2-r) Pk-1-r(0) ( B;+2(N) _ B;+l (N)

k-l-r aN(r+l)(r+2) 2(r+l)
= Pk-1-r(0) (-ly+1 P (0) ( d ~71' )
- k - 1- r r+1 r+l mo 2Dk,r N
_ Pk-l-r(O) Pr+1(0)
- k-l-r r+l .
Here we also used a == 1 (mod N). Therefore, we conclude
U2 =- _12(k;2)(-IY ~~ D (Y+iN ) r~ ( d 6(k;2)N '711 )

k 1 L.J L.J rk-l-r CY mo D IiJN ,
- -r l= . 0 y= 0 a a kr
'
which coincides with (4.4.4). This completes the proof of Proposi-

tion (4.3). 0
Corollary (4.5). Let k ~ 2. Then, there exists an integer D(k) (de-

pending only on k) such that for every N ~ 1,
Proof. This is an immediate consequence of Proposition (4.2). We may

take D(k) so that
~'7J1
D(k) /UN
= ~ (k -
;So r
2) GCD{6Dk,r, N}N '/UN7J1 If1I
C -...e.
o
Example (4.6). Take a matrix A = Oii
-=-1~:6\) E r(1I2), and let us
examine our congruence formula (4.5) for k = 6, N = 11. In this case,
or any 0 <
_ r <4
_ 'r(4) GCD{6,N}11
12D6,r '/U11
7J1 = 11'7J1 N '7J1
/uu, hence 12D(6) /U11 may be
taken as lIZ!l Computation on RHS shows
~~6) (A)(X, Y)
= 6157810527168637 X4 1I7260782677249595 y X3
, 315 + 504
37381997569467617 y2 X2 5190578682530622937 y3 X
+ 36 + 2520
1930511018334372017 y4
+ 1260 '
while that on LHS shows
10
L (xY - yX)4~~ Jl.)(A)
x,y= O
11 ' 11
= 157339 X4 + 557102y X 3 _ 102157y2 X2

6 3
_ 230146 y3X 717205 y4
3 + 6 .
In both sides, each coefficient except for that of X2Y2 is prime to 11.
Now, their difference may be computed as:
L (xy - YX)4~~ JL)(A)

10
~~6)(A)(X, Y) -
x,y= O
11'11
= 12315621037816679 X4 117260782583656459 Y X3
630 + 504
37381997573145269 y2 X2 5190578682723945577 y3 X
+ 36 + 2520
1930511018183758967 y4
+ 1260
== 0 (mod lIZl1 ).
392 H. NAKAMURA
Exam~le (4.7). This example was suggested by Y. Morimoto. Let

A = (1~j~ ~~~~~) E r(81), and consider the case of weight 10. Then, for
o < r < 8 (8) GCD{6,81}81 Z' = 3Z'. The coefficient <1>(5,5)(A) of X4Y4
- - 'r 12DlO,r 3 3 0
in <1>~1O) (A) is computed as
1160735419039913093577749564892899519
<1>&5,5) (A) =
8
which is prime to 3, while
80
I: (48) x 4( _y)4<1>((~ ..!L)(A)
Sl ' Sl
= -8836456074579123550
x,y= O
which is also prime to 3. Now the difference of these two values is
&5,5) (A)- ' " (8)x4(_y)4(~ JL (A)

L 4 (Sl'Sl)
x,y
1160735419039913022886100968259911119
=-
8
which is divisible by 3 (but not divisible by 9).
Example (4.8). Let us consider the case of weight 4 for A =
(d~14 i3~6;9~) E r(128). Then, (;)GC~~6D128}128Z2
6,r
is 4Z2 for r = 0,2,
and 8Z2for r = 1. The computation shows
(4)(A) = 1236394022 X2 33706991732 XY 229732849464 y2
o 45 + 15 + 5 '
and
127
L
x ,y=O
(xY - yX)2~2~ ...lL..)
12S ' 12S
(A)
= 8085 4186 X2 _ 108~8380 XY _ 8016;720 y2 .

3
Both of them do not belong to 4Z~[X, Y] (because of the coefficients of

X2). Now the difference of these two polynomials is computed as
127
<1>(4) (A) - '"' (xY - yX)2(2~ ...lL.. (A)
o L (12S'12S)
x,y=O
= 23581232 X2 33761033632 XY 689599376992 y2
45 + 15 + 15
which belongs to 4Z~[X, Y] (actually to 16Z~[X, YD.

At the time of writing this paper, the author does not have an exam-
ple assuring whether our estimate of the modulus is best possible with
respect to 2-powers.
Acknowledgements
The author would like to thank Yasuhiko Morimoto for valuable dis-
cussions on the topic and useful calculations of numerical examples. He
is also very grateful to the referee for crucial comments which were very
helpful to improve the submitted version of this paper.
References
[D] R. Dedekind, Erliiuterungen zu zwei Pragmenten von Riemann, in "Riemann's
Gesammelte Math. Werke, 2nd edition," 466-472, 1892.
[KM] R. Kirby and P. Melvin, Dedekind sums, j.L-invariants and the signature cocy-
cle, Math. Ann. 299 (1994) , 231-267.
[Mo] Y. Morimoto, Arithmetic behavior of functions related with generalized
Dedekind sums (Japanese), Master Thesis, Tokyo Metropolitan University,
January 2002.
[NIJ H. Nakamura, On exterior Galois representations associated with open elliptic
curves, J. Math. Sci. Univ. Tokyo 2 (1995), 197-231.
[N2] H. Nakamura, Galois representations in the pro-l fundamental groups of punc-
tured elliptic curve (Japanese), in "RIMS Technical Report Series 884," 46-53,
1994.
[N3] H. Nakamura, Tangential base points and Eisenstein power series, in "Aspects
of Galois Theory (H.Voelkein, D.Harbater, P.Mueller, J.G.Thompson, eds.),"
202-217, London Math. Soc. Lecture Note Ser., 256. Cambridge Univ. Press,
Cambridge, 1999.
[N4] H. Nakamura, On exterior monodromy representations associated with affine
elliptic curves, in preparation.
[R] H. Rademacher, Zur Theorie der Modulfunktionen, J. Reine Angew. Math.
161 (1931), 312-336.
[RGJ H. Rademacher and E. Grosswald, "Dedekind sums," Carus Math. Mono-
graphs, 16. The Mathematical Association of America, Washington, D.C.,
1972.
[Scz] R. Sczech, Eisenstein cocycles for GL2Q and values of L-functions in real
quadratic fields , Comment. Math. Helvetici 61 (1992), 363-382.
[St] G. Stevens, The Eisenstein measure and real quadratic fields, in "Proceedings
of the International Conference on Number Theory (Quebec, PQ, 1987)," 887-
927, de Gruyter, Berlin, 1987.

Gal Ios Modular Forms

Uploaded by

Copyright:

Available Formats

Gal Ios Modular Forms

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gal Ios Modular Forms

Uploaded by

Copyright:

Available Formats

Galois Theory and Modular Forms

Krishnaswami Alladi, University ofFlorida, U.S.A.

Aims and Scope

Developments in Mathematics is a book series publishing

(i) Proceedings of Conferences dealing with the latest research advances,

(ii) Research Monographs, and

(iii) Contributed Volumes focussing on certain areas of special interest

Kluwer Academic PubUshers

Distributors for all other countries:

tt Electronic Services <http://www.wkap.nl>

Library of Congress Cataloging-in-Publication

Hashimoto, Ki-ichirol Miyake, Katsuya! Nakamura, Hiroaki

Copyright 2004 by K1uwer Academic Publishers

Part I. Arithmetic geometry 1

The arithmetic of Weierstrass points on modular curves Xo(P) 3

Semistable abelian varieties with small division fields 13

Q-curves with rational i-invariants and jacobian surfaces of GL2-type 39

Points defined over cyclic quartic extensions on an elliptic curve and

The absolute anabelian geometry of hyperbolic curves 77

Part II. Galois groups and Galois extensions 123

Regular Galois realizations of PSL2(P2) over Q(T) 125

Middle convolution and Galois realizations 143

On the essential dimension of p-groups 159

Explicit constructions of generic polynomials for some elementary groups 173

On dihedral extensions and Frobenius extensions 195

On the non-existence of certain Galois extensions 221

Frobenius modules and Galois groups 233

Part III. Algebraic number theory 269

On quadratic number fields each having an unramified extension

Distribution of units of an algebraic number field 287

On capitulation problem for 3-manifolds 305

On the Iwasawa j.L-invariant of the cyclotomic Zp-extension of certain

Part IV. Modular forms and arithmetic functions 327

Quasimodular solutions of a differential equation of hypergeometric type 329

Special values of the standard zeta functions 337

p-adic properties of values of the modular j-function 357

Thompson series and Ramanujan's identities 367

Generalized Rademacher functions and some congruence properties 375

This volume is an outgrowth of the research project "The Inverse Ga-

for example, Kronecker formulated the Kronecker-Weber Theorem, and

Suppose that Q E X and that {WI, ... ,Wd,.(X)} is a basis of '}iT(X)

o= ordQ"'l < ordQw2 < ... < ordQwdr(X)'

Then we define the r- Weierstmss weight of Q as

wtr(Q) > 0 -:::::> Q is an r-Weierstrass point

In this paper we will be exclusively concerned with modular curves;

2. Weierstrass points on Xo(p) and supersingular

L wtl(Q) = g: - gpo (2.1)

j(z) = q-l + 744 + 196884q + ....

To investigate the arithmetic of Weierstrass points on Xo(P), we study

Fp(x):= II (x - j(Qwt 1(Q). (2.2)

(here the product runs over supersingular isomorphism classes of elliptic

Theorem 1. lfp is prime, then Fp(x) has p-integral rational coefficients

and we define Wp(z) as that scalar multiple of Wp(iI, .. . ,fgp)(z) whose

div(Wp(z)(dz)~gp(gP+1)) = L wtl(Q)Q (2.3)

Rohrlich [R] gave a precise description of Wp(z) (mod p) as a modular

Also, define the '..lsual delta function by

A(z) is the unique normalized cusp form of weight 12 on SL2(Z), Finally,

Theorem 2 (Rohrlich). If p ~ 23 is prime then Wp(z) has p-integral

In addition to Rohrlich's result, a certain multiplicative construction is

(flk')(Z):= (ad-bc)~(cz+d)-kf (:;:~).