An Introduction To The Theory of Local Zeta Functions: Studies in Advanced Mathematics

AMS/IP
Studies in
Advanced
Mathematics
S.-T. Yau, Series Editor
An Introduction
to the Theory of
Local Zeta Functions
Jun-ichi Igusa
American Mathematical Society • International Press

Selected Titles in This Series
14 Jun-ichi Igusa, An Introduction to the Theory of Local Zeta Functions, 2000
13 Vasilios Alexiades and George Siopsis, Editors, Trends in Mathematical Physics,
1999
12 Sheng Gong, The Bieberbach Conjecture, 1999
11 Shinichi Mochizuki, Foundations of p-adic Teichmüller Theory, 1999
10 Duong H. Phong, Luc Vinet, and Shing-Tung Yau, Editors, Mirror Symmetry III,
1999
9 Shing-Tung Yau, Editor, Mirror Symmetry I, 1998
8 Jürgen Jost, Wilfrid Kendall, Umberto Mosco, Michael Röckner,
and Karl-Theodor Sturm, New Directions in Dirichlet Forms, 1998
7 D. A. Buell and J. T. Teitelbaum, Editors, Computational Perspectives on Number
Theory, 1998
6 Harold Levine, Partial Differential Equations, 1997
5 Qi-keng Lu, Stephen S.-T. Yau, and Anatoly Libgober, Editors, Singularities and
Complex Geometry, 1997
4 Vyjayanthi Chari and Ivan B. Penkov, Editors, Modular Interfaces: Modular Lie
Algebras, Quantum Groups, and Lie Superalgebras, 1997
3 Xia-Xi Ding and Tai-Ping Liu, Editors, Nonlinear Evolutionary Partial Differential
Equations, 1997
2.2 William H. Kazez, Editor, Geometric Topology, 1997
2.1 William H. Kazez, Editor, Geometric Topology, 1997
1 B. Greene and S.-T. Yau, Editors, Mirror Symmetry II, 1997
An Introduction
to the Theory
of Local Zeta Functions
https://doi.org/10.1090/amsip/014
AMS/IP
Studies in
Advanced
Mathematics
Volume 14
An Introduction
to the Theory
of Local Zeta Functions
Jun-ichi Igusa
American Mathematical Society • International Press

Shing-Tung Yau, Managing Editor
2000 Mathematics Subject Classification. Primary 11Sxx, 11S40, 11Mxx, 11Gxx, 14Gxx.
Library of Congress Cataloging-in-Publication Data

Igusa, Jun-ichi, 1924–
An introduction to the theory of local zeta functions / Jun-ichi Igusa.
p. cm. — (AMS/IP studies in advanced mathematics, ISSN 1089-3288 ; v. 14)
Includes bibliographical references and index.
ISBN 0-8218-2015-X (hard cover; alk. paper)
ISBN 978-0-8218-2907-3 (soft cover; alk. paper)
1. Functions, Zeta. I. Title. II. Series.
QA351 .I38 2000
515.56–dc21 99-087031
Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Requests for such
permission should be addressed to the Acquisitions Department, American Mathematical Society,
201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can also be made by
e-mail to [email protected].

c 2000 by the American Mathematical Society and International Press. All rights reserved.
Reprinted by the American Mathematical Society, 2007.
The American Mathematical Society and International Press retain all rights
except those granted to the United States Government.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
Visit the International Press home page at URL: http://www.intlpress.com/
10 9 8 7 6 5 4 3 2 1 12 11 10 09 08 07
Contents
1 Preliminaries 1
1.1 Review of some basic theorems . . . . . . . . . . . . . . . . . . . 1
1.2 Noetherian rings . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Hilbert’s theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Implicit function theorems and K-analytic manifolds 15

2.1 Implicit function theorem . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Implicit function theorem (non-archimedean case) . . . . . . . . 21
2.3 Weierstrass preparation theorem . . . . . . . . . . . . . . . . . . 24
2.4 K-analytic manifolds and differential forms . . . . . . . . . . . . 28
2.5 Critical sets and critical values . . . . . . . . . . . . . . . . . . . 32
3 Hironaka’s desingularization theorem 35

3.1 Monoidal transformations . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Hironaka’s desingularization theorem
(analytic form) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Desingularization of plane curves . . . . . . . . . . . . . . . . . . 40
4 Bernstein’s theory 45
4.1 Bernstein’s polynomial bf (s) . . . . . . . . . . . . . . . . . . . . . 45
4.2 Some properties of bf (s) . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Reduction of the proof . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 A general theorem on D-modules . . . . . . . . . . . . . . . . . . 52
4.5 Completion of the proof . . . . . . . . . . . . . . . . . . . . . . . 55
5 Archimedean local zeta functions 59

5.1 The group Ω(K × ) . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 Schwartz space S(K n ) . . . . . . . . . . . . . . . . . . . . . . . . 61
5.3 Local zeta function ZΦ (ω) . . . . . . . . . . . . . . . . . . . . . . 67
5.4 Complex power ω(f ) via desingularization . . . . . . . . . . . . . 73
5.5 An application . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6 Prehomogeneous vector spaces 83

6.1 Sato’s b-function b(s) . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.2 The Γ-function (a digression) . . . . . . . . . . . . . . . . . . . . 87
6.3 b(s) = bf (s) and the rationality of the zeros . . . . . . . . . . . . 91
vii
viii CONTENTS
7 Totally disconnected spaces and p-adic manifolds 97

7.1 Distributions in totally disconnected spaces . . . . . . . . . . . . 97
7.2 The case of homogeneous spaces . . . . . . . . . . . . . . . . . . 101
7.3 Structure of eigendistributions . . . . . . . . . . . . . . . . . . . 106
7.4 Integration on p-adic manifolds . . . . . . . . . . . . . . . . . . . 108
7.5 Serre’s theorem on compact p-adic manifolds . . . . . . . . . . . 113
7.6 Integration over the fibers . . . . . . . . . . . . . . . . . . . . . . 114
8 Local zeta functions (p-adic case) 117

8.1 Selfduality of K and some lemmas . . . . . . . . . . . . . . . . . 117
8.2 p-adic zeta function ZΦ (ω) . . . . . . . . . . . . . . . . . . . . . . 120
8.3 Weil’s functions FΦ (i) and FΦ∗ (i∗ ) . . . . . . . . . . . . . . . . . . 125
8.4 Relation of FΦ (i) and ZΦ (ω) . . . . . . . . . . . . . . . . . . . . 129
8.5 Poles of ω(f ) for a group invariant f . . . . . . . . . . . . . . . . 134
9 Some homogeneous polynomials 137

9.1 Quadratic forms and Witt’s theorem . . . . . . . . . . . . . . . . 137
9.2 Quadratic forms over finite fields . . . . . . . . . . . . . . . . . . 141
9.3 Classical groups over finite fields . . . . . . . . . . . . . . . . . . 145
9.4 Composition and Jordan algebras . . . . . . . . . . . . . . . . . . 149
9.5 Norm forms and Freudenthal quartics . . . . . . . . . . . . . . . 154
9.6 Gauss’ identity and its corollaries . . . . . . . . . . . . . . . . . . 160
10 Computation of Z(s) 163

10.1 Z(ω) in some simple cases . . . . . . . . . . . . . . . . . . . . . . 163
10.2 A p-adic stationary phase formula . . . . . . . . . . . . . . . . . 167
10.3 A key lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
10.4 Z(s) for a Freudenthal quartic . . . . . . . . . . . . . . . . . . . 178
10.5 Z(s) for the Gramian det(t xhx) . . . . . . . . . . . . . . . . . . . 184
10.6 An integration formula . . . . . . . . . . . . . . . . . . . . . . . . 188
10.7 Z(s) for det(t xhx) in product forms . . . . . . . . . . . . . . . . 193
11 Theorems of Denef and Meuser 199

11.1 Regular local rings . . . . . . . . . . . . . . . . . . . . . . . . . . 199
11.2 Geometric language . . . . . . . . . . . . . . . . . . . . . . . . . 202
(algebraic form) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
11.4 Weil’s zeta functions over finite fields . . . . . . . . . . . . . . . . 210
11.5 Degree of Z(s) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
11.6 The field Ke (a digression) . . . . . . . . . . . . . . . . . . . . . . 217
11.7 Functional equation of Z(s) . . . . . . . . . . . . . . . . . . . . . 221
Bibliography 227
Index 231
INTRODUCTION
Local zeta functions are relatively new mathematical objects. The first general
theorems were proved from 1968 to 1973. Since then, especially during the last
fifteen years, remarkable results have been obtained, allowing one to call the accu-
mulation a “theory.” Nevertheless, there remain several challenging problems whose
solution will make the theory much richer. The purpose of this book is to introduce
the readers to this theory. The book is written in such a way that it should be
appropriate for those who have mastered the “basic courses” taught in America for
first year graduate students. Assuming the reader has this background, nearly all
material will be explained with detailed definitions and proofs. There are, however,
two exceptions. We shall use Hironaka’s desingularization theorem and the func-
tional equations of Weil’s zeta functions over finite fields. We shall explain these
theorems by examples so that the readers can accept them with some understand-
ing. The references are given primarily to indicate our indebtedness to the authors
and not for the readers to consult.
Since local zeta functions are new, we shall define them briefly with details given
in the text. If k is a number field, any completion of k is called a local field. Every
local field K carries a Haar measure, and the rate of measure change under the
multiplication by a in K × = K\{0} defines its absolute value |a|K ; it is completed
by |0|K = 0. If now X = K n for some n ≥ 1, f is a K-valued non-constant
polynomial function on X, and Φ is in the Schwartz-Bruhat space S(x) of X, then

ZΦ (s) = |f (x)|sK Φ(x) dx, Re(s) > 0
X
is called a local zeta function. If Φ is the standard function on X, i.e., exp(−π t xx)
for K = R, exp(−2π t xx̄) for K = C, and the characteristic function of OK n
for a
p-adic field K with OK as its maximal compact subring, then we drop the subscript
Φ. Furthermore, we normalize the Haar measure dx on X so that Z(s) tends to
1 as s → 0. The set of ωs (·) = | · |sK for all s in C forms the identity component
ix
x INTRODUCTION
of the group Ω(K × ) of all continuous homomorphisms ω from K × to C× . By

replacing ωs by ω we get a more general local zeta function ZΦ (ω). In view of
the fact that |ω(·)| = ωσ(ω) (·) for a unique σ(ω) in R satisfying σ(ωs ) = Re(s), we
define the right-half plane in Ω(K × ) by σ(ω) > 0. Then the first general theorems
are as follows: ZΦ (ω), which is clearly holomorphic on the right-half plane, has a
meromorphic continuation to the whole Ω(K × ). Furthermore, in the p-adic case,
if πOK denotes the ideal of nonunits of OK , then ZΦ (ω) is a rational function of
t = ω(π). These results were obtained jointly by I. N. Bernstein and S. I. Gel’fand,
independently by M. F. Atiyah, then by a different method by Bernstein, and in
the p-adic case, by the author.
We shall now explain the motivation. In the archimedean case where K = R or
C, the general theorem was proposed as a problem by I. M. Gel’fand in 1954 and was
discussed for some well-selected f (x) in the first volume on generalized functions by
I. M. Gel’fand and G. E. Shilov. The solution of the problem implies the existence
of fundamental solutions for constant-coefficient differential equations. It appears
that this situation served as a motivation of the work by Atiyah, Bernstein, and S.
I. Gel’fand. On the other hand, we started differently. In the middle 60’s, A. Weil
showed that Siegel’s main theorem on quadratic forms is a Poisson formula. More
precisely, if kA denotes the adele group of k and ψ a nontrivial character of kA /k,
then in the special case where the base space is one dimensional and

FΦ∗ (i∗ ) = ψ(i∗ f (x))Φ(x)dx,
n
kA
in which i∗ is in kA , f (x) is a nondegenerate quadratic form on kn , Φ is in S(kA

n
), dx
n n n
is the Haar measure on kA normalized as vol(kA /k ) = 1, and FΦ (i) is the inverse
Fourier transform of FΦ∗ (i∗ ), then the Poisson formula takes the form

FΦ (i) = FΦ∗ (i∗ ).
i∈k i∗ ∈k
There is a condition n > 4 for the convergence of the series. Following Weil, we
call the RHS, the right-hand side, the Eisenstein-Siegel series, and the identity itself
with a modified LHS, the Siegel formula. Later, J. G. M. Mars proved the Siegel
formula for a certain cubic form. Toward the end of 60’s, we proved the Siegel
formula for the Pfaffian and determined all cases where the Siegel formula might
hold. However the proof of the convergence of the Eisenstein-Siegel series in general
became a serious difficulty. In order to overcome this obstacle, we introduced ZΦ (ω)
over K as above and showed that the general theorems on ZΦ (ω) can effectively be
used to examine the convergence problem. In fact, it was shown to be sufficient to
estimate Z(ω) for almost all p-adic completions K of k.
These were some of the developments up to the middle 70’s. Before we start an
explanation of later activities, we mention that the general theorems were proved
by using Hironaka’s theorem except for the second proof by Bernstein. In that
proof he used the following remarkable fact: If k◦ is any field of characteristic 0
and f (x) is in k◦ [x1 , . . . , xn ]\{0}, where x1 , . . . , xn are variables, then there exists
INTRODUCTION xi
a differential operator P with coefficients in k◦ [s, x1 , . . . , xn ], where s is another

variable, such that
P f (x)s+1 = bf (s)f (x)s
for a monic polynomial bf (s). The bf (s), which is reserved for the one with the
smallest degree, is called Bernstein’s polynomial of f (x). In the archimedean
case,
the above fact immediately implies the general theorem. In fact if bf (s) = (s + λ),
then
Γ(s + λ)−1 · ZΦ (s)
λ
becomes a holomorphic function on the whole s-plane. The proof via Hironaka’s
theorem shows that the poles of ZΦ (s) are negative rational numbers. On the other
hand, M. Sato developed his theory of prehomogeneous vector spaces in the middle
60’s. Suppose that G is a connected reductive algebraic subgroup of GLn (C) acting
transitively on the complement of an irreducible hypersurface f −1 (0) in Cn with
f (x) necessarily homogeneous of degree say d. Then without losing generality we
can normalize G and f (x) so that t G = Ḡ = G, f (x) is in R[x1 , . . . , xn ] and further
f (∂/∂x)f (x)s+1 = b(s)f (x)s
for a monic polynomial b(s) of degree d called Sato’s b-function. By definition bf (s)
is a factor
of b(s). Actually they are equal. It can be seen, e.g., as follows: If
b(s) = (s + λ), then Z(s) has the form
Γ(s + λ)
−ds
|f (x)|C exp(−2π xx̄) dx = (2π)
s t
·
Cn Γ(λ)
λ
for Re(s) > 0. This with the above results implies b(s) = bf (s). It also gives in the
prehomogeneous case another
proof to a general theorem of M. Kashiwara stating
that all λ’s in bf (s) = (s + λ) are positive rational numbers.
Now in the p-adic case, what we did after the middle 70’s was to compute Z(ω),
especially Z(s), for those f (x) which might give the Siegel formulas. In compiling
a list of Z(s), we gradually became interested in patterns appearing in the shape
of Z(s) as a rational function of t = ωs (π) = q −s , where q = card(OK /πOK ).
We therefore started a systematic computation of Z(s) for a larger class of f (x),
especially for those f (x) which appeared in Sato’s theory, hoping to find conjectures
on Z(s). It did not take too long to find the first conjecture stating that if f (x) is
a homogeneous polynomial in k[x1 , . . . , xn ]\k, where k is a number field as before,
then
degt (Z(s)) = −deg(f )
for almost all K. We might emphasize the fact that this conjecture was not sug-
gested by any existing theory, but it came from explicit computations. At any rate
the conjecture was investigated by D. Meuser and proved as stated by J. Denef in
the late 80’s. In a similar manner, the ever-increasing list of explicitly computed
Z(s) suggested a new type of functional equations satisfied by Z(s). This conjecture
also became a theorem by Meuser and Denef in the early 90’s. More precisely, the
xii INTRODUCTION
new functional equation was derived from the functional equations of Weil’s zeta
functions over finite fields proved by A. Grothendieck. We shall devote the last
chapter to a detailed explanation of their work.
We shall explain some problems on the denominator and the numerator of Z(s)
as a rational function of t. It is known for a general f (x) that except fora power of t
and the allowance of cancellation, the denominator of Z(s) is of the form (1−q −a tb )
for some positive integers a, b. Now in all known examples bf (−a/b) = 0, i.e., the
real parts of the poles of Z(s) are zeros of bf (s), and the order of each pole is at most
equal to the order of the corresponding zero. What it says is that bf (s) for some
hidden reason describes the poles of Z(s) also in the p-adic case. This is extremely
remarkable in view of the fact that bf (s) does not play any direct role in that case.
At any rate the problem is to convert the above experimental fact into a theorem.
In the two variable case, i.e., if n = 2, the problem was settled by F. Loeser. Also in
the prehomogeneous case, it was settled jointly by T. Kimura, F. Sato, and X.-W.
Zhu, except for the information on the orders of poles stated above. In the general
case, a solution seems to require a new theory.
Again, in the case of f (x) appearing in Sato’s theory, hence d = deg(f ) =
deg(bf ), if b > 1 in some factors 1 − q −a tb of the denominator of Z(s), then Denef’s
theorem suggests that Z(s) might have a nonconstant numerator. By going through
the list of Z(s), we notice that certain cubic polynomials in t of the same type appear
rather mysteriously in the numerators of Z(s) for those f (x) which do not have any
apparent similarity. No hint to solve this mystery can be found in the complex case
by the uniform simplicity of Z(s) mentioned above, and a similarly explicit and
general form of Z(s) is not known in the real case. At any rate, no conjecture of
any kind has been proposed on how to describe the numerator of Z(s). We might
finally make it clear that there are several important results, especially those by J.
Denef, which we did not mention in this book. The reader can find most of them
in Denef’s Bourbaki seminar talk [11] and our expository paper [31].
The author would like to thank Professor S.-T. Yau for kindly inviting him
to publish this book in the AMS-IP series. The author would also like to thank
Professor M. M. Robinson for her effort to bring the manuscript into this final form.
Finally, the author would like to gratefully acknowledge the invaluable assistance
by his wife, Yoshie, for providing ideal working conditions for the last fifty years.
https://doi.org/10.1090/amsip/014/01
Chapter 1
Preliminaries
1.1 Review of some basic theorems
We shall assume that the reader is familiar with definitions and basic theorems on
groups, rings, vector spaces, and modules in algebra. We shall review two theorems,
among others, which we shall use later.
We shall assume, unless otherwise stated, that all rings are associative. If A is
any ring with the unit element 1 = 0, then we shall denote by A× the group of units
of A. If Mn (A) is the ring of n × n matrices with entries in A, then Mn (A)× will
be denoted by GLn (A). Suppose that A is commutative. Then an element a of A
is called a zero divisor if ab = 0 for some b = 0 in A. If A has no zero divisor other
than 0, then it is called an integral domain; and an element a = 0 of A is called
irreducible if a is not in A× and if a = bc for b, c in A implies that either b or c is
in A× . An integral domain is called a unique factorization ring if every a = 0 in A
can be expressed uniquely, up to a permutation and elements of A× , as a product
of irreducible elements. The classical examples are the ring Z of integers and the
ring F [x] of polynomials in one variable x with coefficients in a field F . In general,
we have the following consequence of Gauss’ lemma:
If A is a unique factorization ring and x is a variable, then A[x] is also a unique
factorization ring.
We might give some explanation. If we denote by F the quotient field of A, then
every f (x) = 0 in F [x] can be written as f (x) = cf◦ (x) with c in F × and f◦ (x) = 0 in
A[x] such that its coefficients are relatively prime. We call such an f◦ (x) primitive.
According to the Gauss lemma, the product of primitive polynomials is primitive.
Therefore if f (x), g(x) are primitive and f (x) = g(x)h(x) with h(x) in F [x], then
necessarily h(x) is in A[x] and primitive. This implies the above statement. The
irreducible elements of A[x] are irreducible elements of A and primitive polynomials
which are irreducible in F [x]. At any rate, as a corollary we see that F [x1 , . . . , xn ],
where x1 , . . . , xn are variables, is a unique factorization ring.
We shall also use the following fact, which is sometimes called the “Principle of
the irrelevance of algebraic inequalities”:
Let F denote an infinite field and f (x), g1 (x), . . . , gt (x) elements of the polyno-
mial ring F [x] = F [x1 , . . . , xn ], in which g1 (x) = 0, . . . , gt (x) = 0. Suppose that
f (a) = 0 for every a = (a1 , . . . , an ) in F n satisfying g1 (a) = 0, . . . , gt (a) = 0. Then
f (x) = 0.
1
2 JUN-ICHI IGUSA
The proof is as follows. If we put h(x) = f (x)g1 (x) . . . gt (x), then by assumption,
h(a) = 0 for every a in F n . If we can show that this implies h(x) = 0, since F [x] is
an integral domain and g1 (x) = 0, . . . , gt (x) = 0, we will have f (x) = 0. Suppose
that h(x) = 0. Suppose further that n = 1. Then the number of zeros of h(x) in F
is at most equal to deg(h). Since F is an infinite field, we have a contradiction. We
shall therefore assume that n > 1 and apply an induction on n. If we write
h(x) = c0 (x ) + c1 (x )xn + . . .
with c0 (x ), c1 (x ), ... in F [x1 , . . . , xn−1 ], then ci (x ) = 0 for some i by h(x) = 0.
By induction, we can find a = (a1 , . . . , an−1 ) in F n−1 satisfying ci (a ) = 0. Then
h(a , xn ) = 0, hence h(a , an ) = 0 for some an in F . We thus have a contradiction.
We shall assume that the reader is familiar with topological spaces, their com-
pactness, continuous maps, etc., in general topology. We shall review some defini-
tions and theorems.
A topological space is a nonempty set X equipped with a family of its subsets,
called open, with the property that the family is closed under the taking of arbitrary
union and finite intersection. The family then contains X and the empty set ∅. If
Y is any nonempty subset of X, then Y will be considered as a topological space
by the induced topology. The family of open sets in Y consists of intersections of
Y and open sets in X. Complements of open sets are called closed. If A is any
subset of X, the intersection A of all closed sets containing A is called the closure
of A; it is the smallest closed set containing A. An open set containing a point is
called its neighborhood. A topological space is called a Hausdorff space if any two
distinct points have disjoint neighborhoods. A Hausdorff space X with the following
property is called compact: If X is the union of open sets in a family I, then X
is also the union of open sets in a suitable finite subfamily of I. Every nonempty
closed set in a compact space is compact. The Tychonoff theorem states that the
product space, i.e., the product set with the product topology, of any nonempty set
of compact spaces is compact. A Hausdorff space is called locally compact if every
point has a neighborhood with compact closure. The fields R and C of real and
complex numbers with the usual topology are locally compact. A Hausdorff space
is called normal if any two disjoint closed sets are contained in disjoint open sets.
Every compact space is normal. If ϕ is an R-valued function on any nonempty set
X satisfying ϕ(x) ≥ α for some α in R and for every x in X, we shall write ϕ ≥ α
and also α ≤ ϕ. If A, B are disjoint nonempty closed sets in a normal space X,
there exists an R-valued continuous function ϕ on X satisfying 0 ≤ ϕ ≤ 1 such that
ϕ = 0 on A, i.e., ϕ(x) = 0 for every x in A, and ϕ = 1 on B. This remarkable
theorem is called Urysohn’s lemma. We shall later use the following fact:
A Hausdorff space X is locally compact if and only if for every neighborhood U
of any point a of X there exist an R-valued continuous function ϕ on X satisfying
0 ≤ ϕ ≤ 1 and a compact subset C of U containing a such that ϕ = 0 on X\C and
ϕ(a) = 1.
We shall outline the proof. The if-part is clear. In fact, if Oϕ denotes the set of
all x in X where ϕ(x) > 0, then Oϕ is a neighborhood of a with its closure contained
in C, hence it is compact. Conversely, suppose that X is locally compact. Then
INTRODUCTION TO THE THEORY OF LOCAL ZETA FUNCTIONS 3
a has a neighborhood U◦ with compact closure. By replacing U by its intersection

with U◦ , we may assume that Ū is compact, hence normal. We choose disjoint open
subsets V and W of Ū respectively containing a and Ū \U . Then V is open in U ,
hence also in X. By Urysohn’s lemma there exists an R-valued continuous function
ϕ on Ū satisfying 0 ≤ ϕ ≤ 1 such that ϕ = 0 on Ū \V and ϕ(a) = 1. If we extend
ϕ to X as ϕ = 0 on X\V , then ϕ will have the required property with V̄ as C.
We shall assume that the reader is familiar with routine theorems in complex
analysis. We shall review a few basic facts, just to refresh reader’s memory. If

P (z) = cn (z − a)n
n≥0
is a power series with center a and coefficients c0 , c1 , . . . all in C, then there exists
0 ≤ r ≤ ∞, called the radius of convergence, such that P (z) is convergent (resp.
divergent) for every z in C satisfying |z − a| < r (resp. |z − a| > r). Furthermore r
is given by
1
r= 1 ,
lim supn→∞ |cn | n
which is often called Cauchy-Hadamard’s formula. As a consequence, if we define
a power series Q(z) by a termwise differentiation of P (z) as

Q(z) = ncn (z − a)n−1 ,
n>0
then Q(z) has the same radius of convergence as P (z). Furthermore, the C-valued
function P on the disc |z − a| < r defined by P (z) is differentiable at every point of
the disc and its derivative P is given by the function Q defined by Q(z). If D is any
nonempty open subset of C and a is a point of D, we put ∂D = D̄\D and denote
by dis(a, ∂D) the distance from a to ∂D. We have 0 < dis(a, ∂D) ≤ ∞. Suppose
now that f is a C-valued function on D with the following property: At every point
a of D there exists a power series P (z) with a radius of convergence r > 0 such
that f (z) = P (z) for every z in D satisfying |z − a| < r. Then we say that f is
holomorphic on D. The above remark on P and Q shows that if f is holomorphic
on D, then its derivative f exists and is holomorphic on D. We call P (z) the
Taylor expansion of f (z) at a. A standard criterion for f to be holomorphic on D
is that f is differentiable at every point of D. In fact “Cauchy’s theorem” holds for
such an f , hence f (z) can be expressed by “Cauchy’s integral formula,” and hence
it can be expanded into a Taylor series. This line of argument implies that the
above r satisfies r ≥ dis(a, ∂D), and this fact supplies a basis for the holomorphic
continuation of f . At any rate, the above criterion implies the following criterion,
sometimes called Morera’s theorem:
Let f denote a C-valued continuous function on D such that its integral along
any closed curve of finite length in D is 0. Then f is holomorphic on D.
In the application D usually has the property that it is connected and simply
connected. Then Morera’s theorem becomes the converse of Cauchy’s theorem
mentioned above. At any rate the proof is straightforward. We may assume that D
4 JUN-ICHI IGUSA
is connected. Choose a and z from D and join them by a curve C of finite length
in D. Then, by assumption, the integral of f along C is independent of the choice
of C, hence we may denote it by F (z). In this way we get a well-defined function
F on D with f as its derivative. Therefore F , hence also f , is holomorphic on D.
If a is a point of D and the function of z defined by (z − a)k f (z) becomes
holomorphic on D for some k > 0, then a is called a pole of f . In such a case,
f (z) can be expanded into a Laurent series at a, which is similar to P (z) above but
with finitely many terms cn (z − a)n for n < 0. Furthermore, the above mentioned
Cauchy’s integral formula implies that

1 f (z)
cn = dz
|z−a|=r◦ (z − a)
2πi n+1
for every n, in which 0 < r◦ < dis(a, ∂D) and the integral is from θ = 0 to 2π
after writing z − a = r◦ exp(iθ). We shall use meromorphic functions, meromorphic
continuation, etc. later on.
We shall finally review two theorems in calculus and one theorem in analysis.
The first one is called Gauss’ theorem, and it is as follows:
Suppose that D is a bounded open subset of R3 such that ∂D = D̄\D is piecewise
smooth; let f1 (x), f2 (x), f3 (x) denote continuously differentiable functions on D̄.
Then

∂f1 ∂f2 ∂f3
(f1 (x)dx2 dx3 +f2 (x)dx3 dx1 +f3 (x)dx1 dx2 ) = ( + + ) dx1 dx2 dx3 .
∂D D ∂x 1 ∂x2 ∂x3
Since D is assumed to be piecewise smooth, it has a “surface element” dσ. If

n = (n1 , n2 , n3 ) denotes the outer normal unit vector at a smooth point of D and
f · n = f1 n1 + f2 n2 + f3 n3 , then the integrand on the LHS can be replaced by
(f · n) dσ. The second one is called Fubini’s theorem, and it is as follows:
Suppose that −∞ ≤ a < b ≤ ∞, −∞ ≤ c < d ≤ ∞ and D consists of (x, y)
satisfying a < x < b, c < y < d; let f (x, y) denote a continuous function on D
which is absolutely integrable on D, i.e., the integral of |f (x, y)| over D is finite.
Then
b d d b
f (x, y)dxdy = f (x, y)dy dx = f (x, y)dx dy.
D a c c a
The third one is called Lebesgue’s theorem, and it is as follows:

Suppose that D is defined by a < x < b and {fi (x)}i is a sequence of continuous
functions on D which converges to a continuous function on D; suppose further that
there exists an absolutely integrable function φ(x) on D satisfying |fi (x)| ≤ φ(x) for
all i and all x in D. Then

lim fi (x) dx = ( lim fi (x)) dx.
i→∞ D D i→∞
We have formulated the above theorems rather restrictively so that the integrals
become those familiar in calculus. Actually, the third theorem is proved in analysis
for a convergent sequence of Lebesgue integrable functions. In our later applications

we more or less keep the restrictions on functions, but the domains of integration
will become multidimensional. We shall use Lebesgue’s theorem to shorten the
proof, i.e., to avoid longer and artificial argument.
1.2 Noetherian rings

We shall summarize basic theorems on noetherian rings. Although some of them
may be included in the first course on algebra, we shall give outlines of standard
proofs to all of them.
A noetherian ring A is a commutative ring with the unit element 1 in which
every ideal has a finite ideal basis. Equivalently, every strictly increasing sequence
of ideals is finite. In the following, we shall explicitly state it if this condition is
used. If a, b are ideals of A, then ab denotes the set of finite sums of ab for all a in
a and b in b; by definition ab is an ideal of A. If a is an ideal of A, then its root r(a)
consists of all a in A such that its image in the factor ring A/a is nilpotent, i.e., ae
is in a for some positive integer e depending on a. We see that r(a) is an ideal of
A, r(r(a)) = r(a), and the operation “r” is monotone, i.e., preserves the inclusion
relation, and commutes with the taking of finite intersection. Furthermore, if A is
a noetherian ring, hence r(a) has a finite ideal basis, a suitable power r(a)e of r(a)
is contained in a. We say that an ideal p of A is prime if A/p is an integral domain,
i.e., if p = A and ab in p, a not in p imply b in p. We say that an ideal q of A is
primary if A/q = 0 and every zero divisor of A/q is nilpotent, i.e., if q = A and
ab in q, a not in q imply b in r(q). In that case, we see that r(q) is a prime ideal.
Clearly every prime ideal is a primary ideal. If a is an ideal of A and E is a subset
of A, then we denote by a : E the set of all a in A such that ab is in a for all b in E.
We observe that a : E is an ideal of A. Furthermore, the operation “· : E” clearly
commutes with the taking of finite intersection. We shall often use the following
fact: namely, if an ideal a is not contained in r(q) for a primary ideal q, i.e., r(a) is
not contained in r(q), then q : a = q.
Lemma 1.2.1 Let a denote an ideal of a noetherian ring A which is different from
A and meet-irreducible in the sense that it cannot be expressed as an intersection
of two ideals of A both strictly containing a. Then a is a primary ideal.
Proof. We shall assume that a is not primary and derive a contradiction. Since
a = A is already assumed, we will have the situation that ab in a, a not in a,
and b not in r(a). Since the sequence a : b, a : b2 , . . . is increasing, we will have
a : bk = a : bk+1 for some positive integer k. Then both a1 = Abk +a and a2 = Aa+a
strictly contain a. If x is in a1 , then x = cbk + d for some c in A and d in a. If x is
also in a2 , since ab is in a, we see that bx is in a. Then c is in a : bk+1 = a : bk , hence
x is in a. Therefore a is the intersection of a1 and a2 , hence a is not meet-irreducible.
In a noetherian ring A every ideal a can be expressed as an intersection of a finite
number of meet-irreducible ideals. Otherwise, starting from a we can construct an
infinite sequence of strictly increasing ideals in A. On the other hand if q1 , . . . , qt are
primary ideals of A satisfying r(qi ) = p for all i and t > 0, then their intersection q is
6 JUN-ICHI IGUSA
a primary ideal and r(q) = p. Therefore, by Lemma 1 every ideal a can be expressed
as an intersection of primary ideals q1 , . . . , qt with distinct r(q1 ), . . . , r(qt ). If no
qi is redundant, then for the sake of simplicity we call such an expression a minimal
representation of a. We shall only use the following uniqueness theorem:
Theorem 1.2.1 If a is an ideal of a noetherian ring A and

a= qi
1≤i≤t
is a minimal representation of a, then the set {r(q1 ), . . . , r(qt )} is uniquely deter-

mined by a.
Proof. If t = 0, i.e., a = A, then the theorem holds trivially. Suppose that a has
another minimal representation as an intersection of primary ideals q1 , . . . , qs and
assume by induction that min(s, t) > 0. Since the situation is symmetric, we may
assume that p = r(qt ) is maximal among r(qi ), r(qj ) for all i, j. Then p is among
r(q1 ), . . . , r(qs ). Otherwise we will have

a : qt = qi = qj = a,
1≤i<t 1≤j≤s
hence qt is redundant. This is a contradiction. Therefore we may assume that

r(qs ) = p. Then we get

a : qt qs = qi = qj
1≤i<t 1≤j<s
We observe that they are minimal representations of the LHS. Therefore, by induc-
tion the two sets {r(q1 ), . . . , r(qt−1 )} and {r(q1 ), . . . , r(qs−1 )} are the same.
We shall use the operation S −1 by a multiplicative subset S of A, i.e., a multi-
plicatively closed subset containing 1, but only in the case where S is free from zero
divisors. In that case, S −1 A simply consists of a/s for all a in A and s in S with the
convention that a/s = a /s if and only if s a = sa . We convert S −1 A into a ring
as in the special case where A = Z and S is the set of positive integers. Then S −1 A
becomes a commutative ring with 1/1 as its unit element and A can be identified
with its subring under the correspondence a → a/1. If a is an ideal of A, then S −1 a
is defined as the set of a/s for all a in a and s in S. Then S −1 a becomes an ideal of
S −1 A. We observe that the operation a → S −1 a is monotone and commutes with
the taking of finite intersection. Furthermore, if a∗ is an ideal of S −1 A, then A ∩ a∗
is an ideal of A and a∗ = S −1 (A ∩ a∗ ) because a/s = (1/s)(a/1) for every a/s in a∗
with a = a/1 in A ∩ a∗ . Therefore if a∗1 , a∗2 , . . . form a strictly increasing sequence of
ideals of S −1 A, then A ∩ a∗1 , A ∩ a∗2 , . . . form a strictly increasing sequence of ideals
of A. This implies that S −1 A is noetherian if A is noetherian.
Lemma 1.2.2 Let q denote a primary ideal of A. If q and S are disjoint, then
S −1 q is a primary ideal of S −1 A, r(S −1 q) = S −1 r(q), and A ∩ S −1 q = q. If q and
S intersect, then S −1 q = S −1 A.
Proof. Since the second part is clear, we shall only prove the first part. If q and S
are disjoint, then clearly S −1 q = S −1 A and A ∩ S −1 q = q because r(q) and S are
also disjoint. Furthermore, if a/s for a in A and s in S is in S −1 q, then a is in q.
In fact, if a/s = a /s with a in q and s in S, then s a = sa is in q and s is not in
r(q), hence a is in q. This implies r(S −1 q) = S −1 r(q). We shall show that S −1 q is
a primary ideal of S −1 A. If (a/s)(a /s ) is in S −1 q with a/s not in S −1 q, then aa
is in q with a not in q, hence a /s is in S −1 r(q) = r(S −1 q).
Proposition 1.2.1 In Theorem 1.2.1 suppose that qi and S are disjoint for i ≤ r
but not for i > r. Then
S −1 a = S −1 qi
1≤i≤r
−1
is a minimal representation of S a in S −1 A.
Proof. This follows immediately from Lemma 1.2.2 except for the fact that none
of S −1 q1 , . . . , S −1 qr is redundant. If we denote by a the intersection of q1 , . . . , qr ,
then by Lemma 1.2.2 we get A ∩ S −1 a = a . Therefore, if, e.g., S −1 qr is redundant,
we will have
a = A ∩ S −1 a = qi ,
1≤i<r
hence qr becomes redundant in the minimal representation of a. This is a contra-

diction.
We say that an ideal m of A is maximal if m is maximal in the set of all ideals
of A different from A. In that case, Aa + m = A for every a in A not in m, hence
A/m is a field, and hence m is a prime ideal. We say that A is a local ring if the
set of all nonunits of A forms its maximal ideal. If p is a prime ideal of A such that
S = A\p is free from zero divisors, then Lemma 1.2.2 shows that S −1 A is a local
ring with S −1 p as its maximal ideal. At any rate, if A is a local ring with m as its
maximal ideal, then every element of 1 + m is a unit of A.
Theorem 1.2.2 Let m denote an ideal of A with the property that every element of
1+m is a unit of A and M a finitely generated A-module, i.e., M = Ax1 +. . .+Axn
for some x1 , . . . , xn in M , satisfying mM = M . Then M = 0. Furthermore, if A
is a noetherian ring, then the intersection of all powers of m is 0.
Proof. We shall prove the first part. Suppose that M = 0 and express M as in
the theorem with the smallest n necessarily positive. Then xn is in M = mM ,
hence xn = a1 x1 + . . . + an xn with ai in m for all i. This implies (1 − an )xn =
a1 x1 + . . . + an−1 xn−1 . Since 1 − an is a unit of A by assumption, we see that
xn is in Ax1 + . . . + Axn−1 , hence M = Ax1 + . . . + Axn−1 . This contradicts the
minimality of n.
We shall prove the second part. We denote the intersection of all powers of m
by n. Then n is an ideal of A, hence a finitely generated A-module. By the first
part we have only to show that mn = n. Since mn is contained in n, we shall show
that n is contained in mn. Let

mn = qi
1≤i≤t
8 JUN-ICHI IGUSA
denote a minimal representation of mn. Then mn is contained in qi , hence n is

contained in qi : m for all i. If m is not contained in r(qi ), then n is contained in
qi : m = qi . On the other hand if m is contained in r(qi ), then me is contained in qi
for some e, hence n is contained in qi . Therefore, n is contained in all qi , hence in
mn.
The second part, i.e., the fact that the intersection of all powers of m is 0, is due
to W. Krull [36]. We might mention that the first part, i.e., the fact that mM = M
implies M = 0, is sometimes called Nakayama’s lemma. It has the following useful
corollary:
Corollary 1.2.1 Let A denote a local ring with m as its maximal ideal and M a
finitely generated A-module; for every a, x respectively in A, M denote by ā, x̄
their images in A/m, M/mM and convert M/mM into a vector space over A/m as
āx̄ = ax. Then M = Ax1 + . . . + Axn if and only if x̄1 , . . . , x̄n span M/mM over
A/m.
Proof. Since the other part is clear, we shall prove the if-part. We observe that
M/mM = (A/m)x̄1 + . . . + (A/m)x̄n can be rewritten as M = (Ax1 + . . . + Axn ) +
mM . If we put N = M/(Ax1 + . . . + Axn ), then N is a finitely generated A-module
and mN = N , hence N = 0.
1.3 Hilbert’s theorems

It is a well-known historical fact that D. Hilbert put a period to the classical theory
of invariants by his monumental paper [19]. We shall explain three important
theorems in that paper with proof. We shall start with the following lemma:
Lemma 1.3.1 . Let A denote a noetherian ring, M a finitely generated A-module,

and N any A-submodule of M . Then N is also finitely generated.
Proof. If we express M as M = Ax1 + . . . + Axn , then the Lemma holds trivially

for n = 0. Therefore we shall assume that n > 0 and apply an induction on n. If
we denote by N the intersection of N and M = Ax1 + . . . + Axn−1 , then N is
finitely generated by induction. If we denote by a = (M + N ) : xn the set of all a
in A such that axn is in M + N, then a is an ideal of A. We choose a finite ideal
basis {a1 , . . . , at } for a and write ai xn = xi + yi with xi in M and yi in N for all
i. If we put N = Ay1 + . . . + Ayt , then we have only to show that N = N + N .
Since N + N is contained in N , we shall show that every element x of N is in
N + N . If we write x = x + axn with x in M and a in A, then a is in a, hence
a = b1 a1 + . . . + bt at for some b1 , . . . , bt in A. Then x − (b1 y1 + . . . + bt yt ) is in N ,
hence x is in N + N .
Lemma 1.3.2 If A is a noetherian ring and x is a variable, then the polynomial

ring A[x] is also a noetherian ring.
Proof. We take any ideal A of A[x] and show that A has a finite ideal basis. We
shall exclude the trivial case where A = 0 and denote by a the set of coefficients of
the highest powers of x which occur in elements of A. If axm , bxn are the highest
terms of elements of A and m ≥ n, then xm−n · bxn = bxm is also the higest term of
an element of A. Therefore if a, b are in a, then a + b is in a and ca for any c in A
is clearly in a. This shows that a is an ideal of A. We choose g1 (x), . . . , gs (x) from
A such that the coefficients of the highest powers of x which occur in them form an
ideal basis for a. By the same adjustment as above, we may assume that xn for some
n > 0 is the highest power of x which occurs in all of them. We denote by A the
intersection of A and A+Ax+. . .+Axn−1 and put A = A[x]g1 (x)+. . .+A[x]gs (x).
We shall show that A = A + A . Since A contains A + A , we have only to show
that A is contained in A + A . Take any f (x) from A and put deg(f ) = m. If
m < n, then f (x) is in A . Therefore we shall assume that m ≥ n and apply an
induction on m. By the choice of g1 (x), . . . , gs (x) we can find a1 , . . . , as in A such
that for
f (x) = f (x) − ai xm−n gi (x)
1≤i≤s
we have deg(f ) < m. Then by induction f (x), hence also f (x), is in A + A . The
rest of the proof is as follows.
Since A is an A-submodule of the finitely generated A-module A + Ax + . . . +
Ax n−1
, by Lemma 1.3.1 it is finitely generated. Therefore A = Af1 (x)+. . .+Afr (x)
for some f1 (x), . . . , fr (x) in A. This implies

A= A[x]fi (x) + A[x]gj (x).
1≤i≤r 1≤j≤s
If K is an arbitrary field, then 0 and K are the only ideals of K, hence it is a

noetherian ring. Therefore by using Lemmas 1.3.1 and 1.3.2 we get Hilbert’s basis
theorem, which we state as follows:
Theorem 1.3.1 Let A denote the polynomial ring K[x1 , . . . , xn ], where K is a

field and x1 , . . . , xn are variables, M a finitely generated A-module, and N any
A-submodule of M . Then N is also finitely generated. In particular, every ideal of
A has a finite ideal basis.
We shall next explain Hilbert’s Nullstellensatz. We shall start with its shallow
generalization for a better understanding. Since we shall not use this result, we
just outline its proof. Let A denote any commutative ring with the unit element.
Then a nilpotent element of A is clearly contained in every prime ideal of A. The
converse is true and the proof is quite simple. Consider the polynomial ring A[x] as
in Lemma 1.3.2 and take any a from the intersection of all prime ideals of A. Then
1 − ax is a unit of A[x]. Otherwise 1 − ax is contained in a maximal ideal, hence a
prime ideal, say, P of A[x]. This is clear if A, hence A[x], is a noetherian ring. In
the general case, we have only to use Zorn’s lemma. At any rate, if p denotes the
intersection of A and P, then p is a prime ideal of A, hence p contains a. Then P
contains (1 − ax) + ax = 1, a contradiction. Therefore (1 − ax)f (x) = 1 for some
f (x) in A[x], and this implies an+1 = 0 if deg(f ) = n. The above result implies
that for any ideal a of A, its root r(a) is the intersection of all prime ideals of A
10 JUN-ICHI IGUSA
which contain a. Hilbert’s Nullstellensatz is similar to the above statement. We

shall state and prove it after the following lemma:
Lemma 1.3.3 Let m denote any maximal ideal of the polynomial ring K[x] =
K[x1 , . . . , xn ]. Then the images of x1 , . . . , xn in K[x]/m are all algebraic over K.
Proof. We denote the image of xi in K[x]/m by xi , assume that they are not all
algebraic over K, and derive a contradiction. After a permutation we may assume
that yi = xi for i ≤ r are algebraically independent over K and xi for i > r are
algebraic over K(y), where y = (y1 , . . . , yr ). We choose d(y) = 0 from K[y] so
that if we denote d(y)xi for i > r by z1 , . . . , zs , then they become zeros of monic
polynomials with coefficients in K[y] of respective degrees say n1 , . . . , ns . Since
K[x ] = K[x1 , . . . , xn ] is a field, d(y)−1 is in K[x ], hence K[x ] = K[d(y)−1 , y, z],
where z = (z1 , . . . , zs ). Furthermore if we put N = n1 · · · ns and denote z1e1 · · · zses ,
where 0 ≤ ei < ni for all i, by w1 , . . . , wN , then we will have K[y, z] = K[y]w1 +
. . . + K[y]wN . On the other hand, for some y = (y1 , . . . , yr ) with yi algebraic over
K we have d(y ) = 0. We can take y from K r if K is infinite. If we denote by
p the kernel of the homomorphism K[y] → K[y ] defined by yi → yi , then p is a
prime ideal of K[y] not containing d(y) and p = 0 by r > 0, hence K[x ]p = K[x ].
We put S = K[y]\p, A = S −1 K[y], and M = S −1 K[y, z]. Then A is a local ring
with S −1 p as its maximal ideal and M = K[x ] = Aw1 + . . . + AwN . Furthermore,
(S −1 p)M = M . This implies the contradiction M = 0 by Theorem 1.2.2.
Now Hilbert’s Nullstellensatz is a consequence of Lemma 1.3.3 and the fact that
r(a) for any ideal a of K[x] is the intersection of all maximal ideals of K[x] which
contain a. The classical statement is as follows:
Theorem 1.3.2 Let f (x), f1 (x), . . . , fr (x) denote elements of the polynomial ring
K[x] = K[x1 , . . . , xn ] and Ω any algebraically closed extension of K such that
f (a) = 0 for every a = (a1 , . . . , an ) in Ωn satisfying fi (a) = 0 for all i. Then there
exists a positive integer e and a1 (x), . . . , ar (x) in K[x] such that

f (x)e = ai (x)fi (x).
1≤i≤r
Proof. We exclude the trivial case where f (x) = 0, introduce a new variable y, and
denote by a the ideal of K[x, y] generated by f1 (x), . . . , fr (x), 1 − f (x)y. Then
Lemma 1.3.3 and the assumption imply that a is not contained in any maximal
ideal of K[x, y], hence a = K[x, y], and hence

1= ai (x, y)fi (x) + a(x, y)(1 − f (x)y)
1≤i≤r
for some a1 (x, y), . . . , ar (x, y), a(x, y) in K[x, y]. If y e◦ is the highest power of y
which occurs in a1 (x, y), . . . , ar (x, y) and e = e◦ + 1, then by replacing y by 1/f (x),
we get the relation in the theorem with ai (x) = f (x)e ai (x, 1/f (x)) in K[x] for all i.
Finally, we shall explain Hilbert’s theorem on his characteristic functions.
Hilbert’s original proof depends on the theory of syzygies. Another proof by B. L.
van der Waerden depends on ideal theory. We shall explain the proof by M. Nagata
[44] which seems to be the simplest known proof. We shall start with the following
well-known lemma, in which N denotes the set of nonnegative integers:
Lemma 1.3.4 Let χ(t) denote a polynomial of degree d in one variable t with
coefficients in a field of characteristic 0 such that χ(r) is in Z for all large r in N.
Then χ(t) is necessarily of the form
t
t
χ(t) = ai , = t(t − 1) . . . (t − i + 1)/i!
i i
0≤i≤d
with a0 , a1 , . . . , ad in Z.

Proof. Since the highest degree term of ti is ti /i!, we can uniquely write χ(t) as in
the lemma with a0 , a1 , . . . , ad in the field. Since they are clearly in Z for d = 0, we
shall assume that d > 0 and apply an induction on d. If we put σ(t) = χ(t+1)−χ(t),
then deg(σ) = d − 1 and σ(r) is in Z for all large r. Therefore if we write
t
σ(t) = bi ,
i
0≤i<d
then b0 , b1 , . . . , bd−1 are in Z by induction. On the other hand

t
χ(t + 1) − χ(t) = ai+1 ,
i
0≤i<d
hence a1 = b0 , a2 = b1 , . . . , ad = bd−1 are in Z. Then

r
a0 = χ(r) − ai
i
0<i≤d
with χ(r) in Z for all large r shows that a0 is also in Z.

We take the polynomial ring A = K[x1 , . . . , xn ] as before and denote by Ar
its subspace consisting of homogeneous polynomials of degree r for all r. Then A
becomes the direct sum of A0 = K, A1 , A2 , . . . satisfying Ai Aj ⊂ Ai+j for all i, j.
Such an A, with A0 just a subring of A in general, is called a graded ring. If an
A-module M is a direct sum of its additive subgroups M0 , M1 , M2 , . . . satisfying
Ai Mj ⊂ Mi+j for all i, j, then it is called a graded A-module. If N is an A-
submodule of M such that it becomes the direct sum of Nr = N ∩ Mr for all r, then
automatically Ai Nj ⊂ Ni+j for all i, j; such an N is called a graded A-submodule
of M . In that case, the factor module M/N becomes a graded A-module with
(M/N )r = Mr /Nr for all r. We observe that A itself is a graded A-module; we call
its graded A-submodule a homogeneous ideal of A. If M is any graded A-module,
we put
Fr (M ) = Mi = M0 + M1 + . . . + Mr
i≤r
for all r. The Hilbert characteristic function, abbreviated as Hf, is the polynomial
χ(M, t) of t in the following theorem:
12 JUN-ICHI IGUSA
Theorem 1.3.3 If A is the polynomial ring K[x1 , . . . , xn ] and M is a finitely gen-

erated graded A-module, then there exists a polynomial χ(M, t) of degree at most n
satisfying
dimK (Fr (M )) = χ(M, r)
for all large r.
Proof. If χ(M, t) exists, then we say that M has an Hf. To be proved is that every
finitely generated graded A-module M has an Hf. We observe that if M has an
Hf and M # is a new graded A-module defined as (M # )r = Mr+r◦ for some fixed
r◦ and for all r, then M # also has an Hf and χ(M, t), χ(M # , t) have the same
degree. Furthermore, if M is a graded A-submodule of M and if both M and
M/M have Hf’s, then M has χ(M , t) + χ(M/M , t) as its Hf. In particular, if
deg(χ(M , t)), deg(χ(M/M , t)) ≤ n, then deg(χ(M, t)) ≤ n. After these remarks
we write M = Af1 + . . . + Afm . We may assume that each fi is in Mri for some ri .
If m = 0, hence M = 0, then M has 0 as its Hf. Furthermore, if we can show that
every M with m = 1 has an Hf, then by induction M = Af1 + . . . + Afm−1 and
M = M/M will have Hf’s, hence M has an Hf. Therefore, we have only to show
that M = Af1 has an Hf and deg(χ(M, t)) ≤ n.
If we denote by a the kernel of the A-homomorphism from A to M defined by
a → af1 , then a is a homogeneous ideal of A and (A/a)r becomes K-isomorphic
to Mr+r1 , i.e., (A/a)r is mapped K-linearly and bijectively to Mr+r1 , for all r.
Therefore, if A/a has an Hf, so does M. Furthermore, since

n+r
dimK (Fr (A/a)) ≤ dimK (Fr (A)) = ,
n
we will have deg(χ(M, t)) ≤ n. Consider the set Σ of all homogeneous ideals a of
A, different from A, with the property that A/a does not have an Hf. We have
only to derive a contradiction assuming that Σ is not empty. We choose any a
which is maximal in Σ. Then a1 is strictly contained in A1 for otherwise a becomes
A1 + A2 + . . . and A/a will have 1 as its Hf. Therefore, we can choose f from
A1 \a1 . Then the homogeneous ideal Af + a strictly contains a, hence A/(Af + a)
has an Hf, and hence σ(t) = χ(A/(Af + a), t) is defined. We observe that a : f is a
homogeneous ideal of A containing a and that the correspondence a → af gives rise
to a K-isomorphism from Ar /(a : f )r to (Af + a)r+1 /ar+1 for every r. A simple
computation of dimensions then shows that
dimK (Fr+1 (A/a)) − dimK (Fr (A/(a : f ))) = dimK (Fr+1 (A/(Af + a)))
for every r, and the RHS is equal to σ(r + 1) for all large r. If a : f = a, then
we see, as in the proof of Lemma 1.3.4, that A/a has an Hf, which is not the case.
Therefore a : f strictly contains a, hence A/(a : f ) has an Hf, and then A/a also
has an Hf. This is a contradiction.
We remark that if a is any homogeneous ideal of A = K[x1 , . . . , xn ], different
from 0, then in
t
χ(a, t) = ai
i
0≤i≤n
we always have an = 1. In fact a = 0 implies ar◦ = 0 for some r◦ . If f = 0 is in ar◦ ,

then Ar f ⊂ ar+r◦ ⊂ Ar+r◦ , hence
dimK (Ar ) = dimK (Ar f ) ≤ dimK (ar+r◦ ) ≤ dimK (Ar+r◦ )
for every r. This implies the above assertion.

Chapter 2
Implicit function theorems

and K-analytic manifolds
2.1 Implicit function theorem
We shall prove an implicit function theorem over an arbitrary complete field K
using calculus of limits. This method was discovered by A. Cauchy in the case
where K = C to prove general existence theorems. The fact that it can be applied
to prove similar theorems over an arbitrary complete field K was pointed out,
possibly for the first time, in [21]. At any rate, since Cauchy’s calculus of limits is
very seldom taught in any graduate course, we shall give all the details to the proof.
We shall denote by K a field with an absolute value | · |K . This means that | · |K
is an R-valued function on K satisfying the following conditions:
AV 1. |a|K ≥ 0 for all a in K and |a|K = 0 if and only if a = 0.
AV 2. |ab|K = |a|K |b|K for all a, b in K.
AV 3. |a + b|K ≤ |a|K + |b|K also for all a, b in K.
We shall exclude the trivial case where Im(K × ), the image of K × under | · |K , is
{1}. Then Im(K × ) contains a sequence {ri }i = {r1 , r2 , . . . } which tends to 0. If
for any b in K we take the family of subsets of K defined by |a − b|K < ri for
i = 1, 2, . . . as a base of open sets containing b, then K becomes a Hausdorff space.
Furthermore the algebraic operations in K are continuous. If a = (a1 , . . . , an ) is in
K n , we put
a = max(|a1 |K , . . . , |an |K ).
Then in K n with the product topology the family of subsets defined by aK < ri
for i = 1, 2, . . . forms a base of open sets containing 0. We shall later replace | · |K
by | · |ρK for some ρ > 0, e.g., in the case where K = C. If we do this, the new
absolute value ceases to satisfy AV 3 and yet it defines the same topology on K.
We shall assume in most cases that K is complete. This means that every
Cauchy sequence in K , i.e., a sequence satisfying Cauchy’s criterion of convergence,
is convergent. Explicitly stated, if a1 , a2 , . . . are elements of K and |ai − aj |K tends
to 0 as i, j tend to ∞, then there exists an element a of K, necessarily unique, such
that |ai − a|K tends to 0 as i tends to ∞. If K is any field with an absolute value
| · |K , the set of Cauchy sequences in K forms a commutative ring with the unit
element under the termwise addition and multiplication. In this ring the set of all
null sequences, i.e., sequences which tend to 0, forms a maximal ideal. We denote
15
16 JUN-ICHI IGUSA
the corresponding factor ring by K ∗ . If a∗ is an element of K ∗ represented by a

Cauchy sequence {ai }i , then the sequence {|ai |K }i in R has a limit which depends
only on a∗ . If we denote this limit by |a∗ |K ∗ , then | · |K ∗ gives an absolute value on
K ∗ and K ∗ becomes a complete field. If to every a in K we associate the element
of K ∗ represented by the sequence all terms of which are a, then we get an injective
homomorphism from K to K ∗ . If we identify K with its image in K ∗ , then | · |K ∗
restricts to | · |K on K and K is dense in K ∗ , i.e., K ∗ becomes the closure of K. The
field K ∗ is called the completion of K. The completion of Q by the usual absolute
value on Q is R. If p is a prime number, then every a in Q× can be written uniquely
as a = pe b for some e in Z and b in Q× with its denominator and numerator both
not divisible by p. If we put
|a|p = p−e , |0|p = 0,
then | · |p gives an absolute value on Q and the corresponding completion of Q is
the Hensel p-adic field Qp .
If A is a commutative ring with the unit element, in particular if A is a field
K, and if i = (i1 , . . . , in ) is in Nn , x = (x1 , . . . , xn ) where x1 , . . . , xn are variables,
and ci is in A for all i, then we shall write

c i xi = ci1 ...in xi11 · · · xinn .
i1 ,... ,in
The set A[[x1 , . . . , xn ]] of all such formal power series forms a commutative ring. If
K is a complete field, basic properties of sequences and infinite series for K =
R, which one learns in calculus, remain valid for K. If a series in K[[x]] =
K[[x1 , . . . , xn ]] is convergent at every a in K n satisfying a < r for some r > 0,
then it is called a convergent power series. The set Kx = Kx
1 , i. . . , xn of all
such
◦ i convergent power series forms a subring of K[[x]]. If for ci x in K[[x]] and

ci x in Rx we have |ci |K ≤ c◦i for all i, then we shall write

ci xi << c◦i xi

◦ i
and call ci x a dominant series for c i xi .

Lemma 2.1.1 A formal power series is a convergent power series if and only if it
has a dominant series.
Proof.
Since the if-part is straightforward, we shall

prove the only-if part. Suppose
that ci xi is a convergent power series, i.e., ci ai is convergent for every a in
K n satisfying ||a|| < r for some r > 0. Choose 0 < r◦ < r from Im(K × ). Then for
every a in K n satisfying ||a|| ≤ r◦ we have
|i|
|ci ai |K ≤ |ci |K r◦ , |i| = i1 + . . . + in .
|i|
Furthermore |ci |K r◦ tends to 0 as |i| tends to ∞, hence it is bounded by some
M > 0. It is then clear that
M
ci xi << |i|
xi .
r◦
We observe that if F (y) is in A[[y1 , . . . , ym ]] and fi (x) is in A[[x1 , . . . , xn ]] sat-

isfying fi (0) = 0 for 1 ≤ i ≤ m for some m, n, then F (f (x)) = F (f1 (x), . . . , fm (x))
is clearly in A[[x1 , . . . , xn ]]. If A = K is a complete field and F (y), fi (x) are
convergent power series for all i, then F (f (x)) is a convergent power series.
Lemma 2.1.2 If F (x, y) is an element of K[[x, y]] = K[[x, y1 , . . . , ym ]], it can be
written uniquely as

F (x, y) = F (x, 0) + Hk (x, y1 , . . . , yk )yk
1≤k≤m
with Hk (x, y1 , . . . , yk ) in K[[x, y1 , . . . , yk ]] for 1 ≤ k ≤ m. If further F (x, y) is a

convergent power series, then every Hk (x, y1 , . . . , yk ) is a convergent power series.
Proof. Firstly, the uniqueness is clear. Since the lemma holds trivially for m = 0,
we shall assume that m > 0 and apply an induction on m. We have
F (x, y) = F (x, y1 , . . . , ym−1 , 0) + Hm (x, y)ym
with Hm (x, y) in K[[x, y]]. Furthermore if F (x, y) << F ◦ (x, y) and
F ◦ (x, y) = F ◦ (x, y1 , . . . , ym−1 , 0) + Hm
◦
(x, y)ym ,
then F (x, y1 , . . . , ym−1 , 0) << F ◦ (x, y1 , . . . , ym−1 , 0) and Hm (x, y) << Hm
◦
(x, y).
Therefore we have only to apply an induction to F (x, y1 , . . . , ym−1 , 0).
As a consequence,
if G(x, y) is an element of K[[x, y]] (resp. Kx, y ) of the
form G(x, y) = cij xi y j with c00 = 0 and c0j = 0 for |j| = 1 and if z1 , . . . , zm are
variables, then
G(x, y + z) = G(x, y) + Hk (x, y, z)zk
1≤k≤m
with Hk (x, y, z) in K[[x, y, z]] (resp. Kx, y, z ) free from zk+1 , · · · , zm . Since

G(0, z) = H1 (0, 0, z)z1 + . . . + Hm (0, 0, z)zm , we have Hk (0, 0, 0) = 0 for all k.
All termwise partial derivatives of a convergent power series are convergent
power series. More precisely if

ci xi << c◦i xi ,
then
∂( ci xi )/∂xj << ∂( c◦i xi )/∂xj
for all j. In fact, we have |ij ci |K ≤ ij c◦i for every i and the RHS is a convergent
power series.
Lemma 2.1.3 Suppose that
M
i
f (x) = ci x << xi
r |i|
for some M, r > 0 and U is the neighborhood of 0 in K n defined by a < r;
denote by f the K-valued function on U defined by a → f (a). Then f = 0 implies
f (x) = 0. Furthermore partial derivatives of f are the functions on U obtained from
the corresponding termwise partial derivatives of f (x). In particular f is continuous.
18 JUN-ICHI IGUSA
Proof. We shall prove the first part, i.e., f = 0 implies f (x) = 0. This can be done
by using the principle of the irrelevance of algebraic inequalities in Chapter 1.1 or
directly as follows. Suppose first that n = 1 and

f (x) = c i xi , ck = 0
i≥k
for some k ≥ 0. If a is in K × and |a|K < r, then
f (a) i−k M |a|K

= ck + ci ai−k = 0, ci a ≤ .
a k K r k (r − |a|K )
i>k i>k
We know that Im(K × ) contains a null sequence in R. Therefore we can choose a

above so that the RHS becomes less than |ck |K . We then have a contradiction.
Suppose next that n > 1 and

f (x) = f (x , xn ) = fi (x )xin , fk (x ) = 0,
i≥k
in which fi (x ) are all in K[[x1 , . . . , xn−1 ]]. Then they are convergent at every a
in K n−1 satisfying a < r, where · is relative to K n−1 . Since fk (x ) = 0,
by induction we can find a such that fk (a ) = 0. Then f (a , xn ) = 0, hence
f (a , an ) = 0 for some an in K satisfying |an |K < r, and hence f = 0. This is a
contradiction.
We shall prove the second part. As in calculus, we may assume that n = 1.
Take a from K, h from K × satisfying |a|K + |h|K < r. Then we have
f (a + h) − f (a)
M r|h|K
− ici ai−1 ≤ .
h K (r − |a|K − |h|K )(r − |a|K )2
If |h|K tends to 0, the RHS, hence also the LHS, tends to 0.
Theorem 2.1.1 (i) Let K denote an arbitrary field and assume for some m, n that
every Fi (x, y) in F (x, y) = (F1 (x, y), . . . , Fm (x, y)) is in
K[[x, y]] = K[[x1 , . . . , xn , y1 , . . . , ym ]] satisfying Fi (0, 0) = 0 and further
∂(F1 , . . . , Fm )/∂(y1 , . . . , ym )(0, 0) = 0,
in which ∂(F1 , . . . , Fm )/∂(y1 , . . . , ym ) is the jacobian, i.e., the determinant of the

square matrix of degree m with ∂Fi /∂yj as its (i, j)-entry. Then there exists a unique
f (x) = (f1 (x), . . . , fm (x)) with every fi (x) in K[[x]] = K[[x1 , . . . , xn ]] satisfying
fi (0) = 0 and further F (x, f (x)) = 0, i.e., Fi (x, f (x)) = 0 for all i.
(ii) In the above situation, if K is a complete field and if every Fi (x, y) is a conver-
gent power series, then every fi (x) is also a convergent power series. Furthermore
if a is near 0 in K n , then f (a) is near 0 in K m and F (a, f (a)) = 0, and if (a, b) is
near (0, 0) in K n × K m and F (a, b) = 0, then b = f (a).
Proof of (i). If we write

Fi (x, y) = aij yj − Gi (x, y), Gi (x, y) = cijk xj y k
1≤j≤m |j|+|k|>0
with aij , cijk in K, in which ci0k = 0 for |k| = 1, then aij = (∂Fi /∂yj )(0, 0) for all
i, j. The square matrix a with aij as its (i, j)-entry is in GLm (K) by assumption.
We observe that f (x) satisfying F (x, f (x)) = 0 is not affected by any invertible
K-linear transformation of entries of F (x, y). Therefore, after multiplying a−1 to
F (x, y), regarded as a column vector, we may assume that aij = δij , i.e., aii = 1
and aij = 0 for i = j. After this adjustment we write

fi (x) = dij xj
|j|>0
with unknown coefficients dij in K. Since F (x, f (x)) = 0 can be rewritten as

fi (x) = cijk xj f (x)k
|j|+|k|>0
for all i, if we denote by fip (x) the homogeneous part of degree p in fi (x), then
F (x, f (x)) = 0 becomes equivalent to
 

fip (x) = cijk xj  fα pαβ (x) ,
1≤α≤m 1≤β≤kα
in which kα = |k| and

(∗) |j| + pαβ = p, pαβ > 0,
1≤α≤m 1≤β≤kα
in particular |j| + |k| ≤ p, for p = 1, 2, 3, . . . . Since ci0k = 0 for |k| = 1, if p = 1,

then
fi1 (x) = cij0 xj .
|j|=1
Furthermore, if p > 1, then in (∗) we have pαβ < p for cijk = 0. In fact if pαβ ≥ p,
hence pαβ = p, for some α, β, then j = 0, kα = 1, and kα = 0 for α = α. This
implies |k| = 1 and cijk = ci0k = 0. We have thus shown that fi1 (x) is as above and
fip (x) for p > 1 is determined by fi p (x) for p < p. Therefore fi (x) is uniquely
determined by an induction on p.
Proof of (ii). We start with an additional observation still in the case where K is
an arbitrary field. The above proof shows that dij = cij0 for |j| = 1 and that dij
for |j| = p > 1 is a polynomial in ci j k and di j for |j | + |k | ≤ p and |j | < p
with coefficients in N. Therefore, again by an induction on p, we see that
dij = Pij (ci j k ),

20 JUN-ICHI IGUSA
in which Pij is a polynomial in ci j k for |j | + |k | ≤ p with coefficients in N.

Suppose now that K is a complete field and Fi (x, y) is in Kx, y . Then by
Lemma 2.1.1 we will have
M
Gi (x, y) << xj y k
r |j|+|k|
|j|+|k|>0
M
<< p+q
(x1 + . . . + xn )p (y1 + . . . + ym )q
p+q>0
r

for some M, r > 0, in which (resp. ) indicates the restriction |k| > 1 for
j = 0 (resp. q > 1 for p = 0). If we denote the last power series by G◦i (x, y) and
put
Fi◦ (x, y) = yi − G◦i (x, y), G◦i (x, y) = c◦ijk xj y k ,
|j|+|k|>0
then |cijk |K ≤ c◦ijk for all i, j, k and by (i) we have a unique fi◦ (x) =
(f1 (x), . . . , fn (x)) in R[[x]] satisfying fi◦ (0) = 0 and F ◦ (x, f ◦ (x)) = 0. Further-
◦ ◦
more
fi◦ (x) = d◦ij xj , d◦ij = Pij (c◦i j k )
|j|>0
with the same polynomial Pij as before. Since the coefficients of Pij are in N,
therefore, we get
|dij |K = |Pij (ci j k )|K ≤ Pij (|ci j k |K ) ≤ Pij (c◦i j k ) = d◦ij
for all i, j. If, for a moment, we accept the fact that fi◦ (x) is in Rx , then by
definition fi (x) << fi◦ (x) for all i. Therefore the formal identity F (x, f (x)) = 0
implies F (a, f (a)) = 0 for all a near 0 in K n .
We shall show that if (a, b) is near (0, 0) in K n × K m and F (a, b) = 0, then
b = f (a). If y1 , . . . , ym

are other variables, then by the remark after Lemma 2.1.2
we can write

Gi (x, y) − Gi (x, y ) = Hij (x, y, y )(yj − yj )
1≤j≤m
with Hij (x, y, y ) in Kx, y, y satisfying Hij (0, 0, 0) = 0 for all i, j. Furthermore,

bi − fi (a) = Gi (a, b) − Gi (a, f (a)) = Hij (a, b, f (a))(bj − fj (a)),
1≤j≤m
hence
(δij − Hij (a, b, f (a)))(bj − fj (a)) = 0
1≤j≤m
for all i. Since Hij (a, b, b ) depends continuously on (a, b, b ) near (0, 0, 0) in K n ×
K m × K m by Lemma 2.1.3 and Hij (0, 0, 0) = 0, the coefficient-matrix is invertible
for (a, b) near (0, 0), hence b = f (a).
Finally, we shall show that fi◦ (x) is in Rx . If we put X = x1 + . . . + xn and

Y = y1 + . . . + ym , then we have
M
G◦i (x, y) = − M (1 + Y /r),
(1 − X/r)(1 − Y /r)
hence F ◦ (x, y) = 0 if and only if y1 = . . . = ym = Y /m and

M 1 Y MX
+ Y2− + = 0.
r2 mr m (r − X)
This equation in Y has a unique solution in RX which becomes 0 for X = 0.

This clearly implies that fi◦ (x) is in Rx .
Corollary 2.1.1 (i) If gi (x) in K[[x]], where K is an arbitrary field, satisfies

gi (0) = 0 for 1 ≤ i ≤ n and
∂(g1 , . . . , gn )/∂(x1 , . . . , xn )(0) = 0,
then there exists a unique f (x) = (f1 (x), . . . , fn (x)) with fi (x) in K[[x]] satisfying
fi (0) = 0 for all i and g(f (x)) = x. (ii) In the above situation, if K is a complete
field and gi (x) is in Kx , then fi (x) is also in Kx for all i. Furthermore if b
is near 0 in K n and a = g(b), then a is also near 0 in K n and b = f (a). Therefore
y = f (x) gives rise to a bicontinuous map from a small neighborhood of 0 in K n to
another.
This follows immediately from Theorem 2.1.1. We have only to take Fi (x, y) =
xi − gi (y) for 1 ≤ i ≤ m = n.
Corollary 2.1.2 If g(0) = 0 for g(x) in K[[x]], then 1/g(x) can be expressed
uniquely as an element of K[[x]]; if further g(x) is in Kx , then 1/g(x) is also
in Kx .
We may assume that g(0) = 1. Then we have only to apply Theorem 2.1.1 to
F (x, y) = g(x)(1 + y) − 1 for m = 1. We also mention that if char(K) does not
divide a positive integer m and if the m-th power map is surjective on K × , then
the m-th power map is also surjective on K[[x]]× . In fact if g(x) is any element
of K[[x]] with g(0) = 0, then am = g(0) for some a in K × . This time we apply
Theorem 2.1.1 to F (x, y) = (a + y)m − g(x).
2.2 Implicit function theorem (non-archimedean

case)
We say that an absolute value | · |K on a field K is non-archimedean if, instead of
AV 3, it satisfies the following stronger condition:
AV 3 . |a + b|K ≤ max(|a|K , |b|K ) for all a, b in K.
22 JUN-ICHI IGUSA
In such a case if we put

OK = {a ∈ K; |a|K ≤ 1},
×
then OK forms a subring of K and the group of units OK of OK is given by
×
OK = {a ∈ K; |a|K = 1}.
We shall assume that Im(K × ), the image of K × under | · |K , is discrete in R× . This
condition is equivalent to the ideal of nonunits of OK being principal, i.e., of the
form πOK for some π in OK . We shall sometimes write πK instead of π to avoid
×
any confusion. We keep in mind that π is unique up to a factor in OK , hence |π|K
does not depend on the choice of π. At any rate, if a, b are in K and a − b is in
π e OK for some e in Z, mostly for e > 0, then we shall write a ≡ b mod π e . If c is
×
in OK , then c ≡ 0 mod π means that c is in OK .
We know that a non-archimedean absolute value is characterized by the con-
dition that |n1|K ≤ 1 for all n in N. We might as well recall its simple proof. If
|n1|K ≤ 1 for all n in N, then
(|a + b|K )n ≤ (n + 1)max(|a|K , |b|K )n
for all a, b in K, and this implies AV 3 above as n → ∞. In particular Qp is
a complete non-archimedean field. This implies, e.g., by Proposition 11.6.1, that
every extension of Qp of finite degree is also such a field.
We shall assume that K is a complete non-archimedean field. Then a series
in K is convergent if and only if the sequence obtained by replacing each term by
its absolute value forms a null sequence in R. If A is a subring of K, then we
shall denote by Ax the intersection
of A[[x]] = A[[x1 , . . . , xn ]] and Kx =
Kx1 , . . . , xn . We say that f (x) = ci xi in K[[x]] is a special restricted power
series, abbreviated as SRP, if f (0) = 0, i.e., c0 = 0, and
ci ≡ 0 mod π |i|−1 , |i| = i1 + . . . + in
for all i = 0 in Nn . This clearly implies that f (x) is in OK [[x]]. Furthermore f (x)
n
is convergent at every a in OK . In fact, we have
|ci ai |K ≤ |ci |K ≤ (|π|K )|i|−1

and the RHS tends to 0 as |i| → ∞. Therefore f (a) = ci ai is convergent and

f (a) is in OK . We introduce the notation
P # (x) = πP (π −1 x) = πP (π −1 x1 , . . . , π −1 xn )
for every P (x) in K[[x]] satisfying P (0) = 0. Then clearly P (x) is an SRP in
x1 , . . . , xn if and only if P # (x) is in OK [[x]]. Furthermore if F (y) is an SRP in
y1 , . . . , ym and all entries of f (x) = (f1 (x), . . . , fm (x)) are SRP’s in x1 , . . . , xn ,
then F (f (x)) is also an SRP in x1 , . . . , xn . This can be proved, e.g., as follows.
If we put G(x) = F (f (x)), then G(0) = 0 and G# (x) = F # (f # (x)), in which
f # (x) = (f1# (x), . . . , fm
#
(x)). Since F # (y) is in OK [[y]] and all entries of f # (x) are
#
in OK [[x]], clearly G (x) is in OK [[x]], hence G(x) is an SRP in x1 , . . . , xn .
We shall now go back to Theorem 2.1.1. In (i) below the field K need not be
complete:
Theorem 2.2.1 (i) If Fi (x, y) is in OK [[x, y]] and Fi (0, 0) = 0 for all i and further
∂(F1 , . . . , Fm )/∂(y1 , . . . , ym )(0, 0) ≡ 0 mod π,
then every fi (x) in the unique solution f (x) = (f1 (x), . . . , fm (x)) of F (x, f (x)) = 0
satisfying fi (0) = 0 is in OK [[x]]. (ii) If every Fi (x, y) is an SRP in
x1 , . . . , xn , y1 , . . . , ym , then every fi (x) is an SRP in x1 , . . . , xn . Furthermore if a
is in OK n
, then f (a) is in OK m
and F (a, f (a)) = 0, and if (a, b) in OK n
× OKm
satisfies
F (a, b) = 0, then b = f (a).
Proof of (i). If, as in the proof of Theorem 2.1.1, we write

Fi (x, y) = aij yj − Gi (x, y), Gi (x, y) = cijk xj y k
1≤j≤m |j|+|k|>0
with aij , cijk in OK this time, in which ci0k = 0 for |k| = 1, then the square matrix
a with aij as its (i, j)-entry is in GLm (OK ) by assumption. Therefore, after the
normalization aij = δij by multiplying a−1 to F (x, y), the new cijk is still in OK
for all i, j, k. Since

fi (x) = dij xj , dij = Pij (ci j k ),
|j|>0
in which the coefficients of the polynomial Pij are in N, we see that fi (x) is in
OK [[x]] for all i.
Proof of (ii). We observe that the normalization aij = δij does not affect the
assumption that Fi (x, y) is an SRP in x1 , . . . , xn , y1 , . . . , ym for all i. Therefore
Fi# (x, y) = πFi (π −1 x, π −1 y) is in OK [[x, y]] and Fi# (0, 0) = 0 for all i, and further
∂(F1# , · · · , Fm
#
)/∂(y1 , · · · , ym )(0, 0) = 1.
Therefore if g(x) = (g1 (x), · · · , gm (x)), where gi (0) = 0, is the unique solution of
F # (x, g(x)) = 0, then every gi (x) is in OK [[x]] by what we have shown. On the
other hand, F (x, f (x)) = 0 implies F # (x, f # (x)) = 0, in which fi# (0) = 0 for all
i. Therefore by the uniqueness we get fi# (x) = gi (x), hence fi (x) is an SRP in
x1 , · · · , xn for all i. In particular if a is in OKn
, then f (a) is in OK m
and the formal
identity F (x, f (x)) = 0 implies F (a, f (a)) = 0.
We shall show that if (a, b) in OK n
× OK
m
satisfies F (a, b) = 0, then b = f (a). We
observe that the LHS of

Gi (x, y) − Gi (x, y ) = Hij (x, y, y )(yj − yj )
i≤j≤m
in the proof of Theorem 2.1.1 (ii) is an SRP in the entries of x, y, y , and it

starts from the degree 2 part. Therefore the remark after Lemma 2.1.2 shows
that π −1 Hij (x, y, y ) is also an SRP in the same variables. Now since F (a, b) = 0
implies
(δij − Hij (a, b, f (a)))(bj − fj (a)) = 0
1≤j≤m
24 JUN-ICHI IGUSA
for all i with the coefficient-matrix in GLm (K), in fact in GLm (OK ) because its
determinant is in 1 + πOK , we see that b = f (a).
Theorem 2.2.1 has a corollary similar to that of Theorem 2.1.1. We shall use
the same notation as in that corollary.
Corollary 2.2.1 (i) If gi (x) is in OK [[x]] and gi (0) = 0 for all i and further
∂(g1 , . . . , gn )/∂(x1 , . . . , xn ) ≡ 0 mod π,
then every fi (x) in the unique solution of g(f (x)) = x satisfying fi (0) = 0 is also
in OK [[x]]. (ii) If every gi (x) is an SRP in x1 , . . . , xn , then every fi (x) is also an
n
SRP in the same variables, and y = f (x) gives rise to a bicontinuous map from OK
to itself
2.3 Weierstrass preparation theorem

In general if f (x) is an element of A[[x]] = A[[x1 , . . . , xn ]], where A is any com-
mutative ring with the unit element, and if the homogeneous part fp (x) of f (x) of
degree p satisfies fp (x) = 0 for all p < r and for some r in N, then we shall write
f (x) = fr (x) + . . . . If further fr (x) = 0, then r and fr (x) are called respectively the
leading degree and the leading form of f (x). If gs (x) is the leading form of a similar
element g(x) of A[[x]] and if fr (x)gs (x) = 0, then it is the leading form of f (x)g(x).
This implies that if A is an integral domain, then A[[x]] is also an integral domain.
We shall explain an immediate consequence of Theorem 2.1.1. If F (x, y) is in
K[[x, y]] = K[[x1 , . . . , xn , y]] satisfying F (0, 0) = 0 and c = (∂F/∂y)(0, 0) = 0,
i.e., cy is the leading form of F (0, y), then there exists a unique f (x) in K[[x]]
satisfying f (0) = 0 and F (x, f (x)) = 0. If we put z = y − f (x), we will have
F (x, y) = F (x, f (x) + z) = zE0 (x, z) with E0 (x, z) in K[[x, z]]. This implies
F (x, y) = E(x, y)(y − f (x)),
in which E(x, y) = E0 (x, y − f (x)) is in K[[x, y]] and E(0, 0) = c. Furthermore

if F (x, y) is a convergent power series, then f (x) and E(x, y) are also convergent
power series. The Weierstrass preparation theorem is a generalization of this fact,
and it is as follows:
Theorem 2.3.1 If F (x, y) is in K[[x, y]] = K[[x1 , . . . , xn , y]] and has the property
that F (0, y) is different from 0 and has cy m for some m > 0 as its leading form,
then F (x, y) can be written uniquely as
F (x, y) = E(x, y)(y m + a1 (x)y m−1 + . . . + am (x)),
in which E(x, y) is in K[[x, y]] with E(0, 0) = 0 and a1 (x), . . . , am (x) are in K[[x]],
hence necessarily ai (0) = 0 for all i and E(0, 0) = c. Furthermore if F (x, y) is
a convergent power series, then a1 (x), . . . , am (x) and E(x, y) are also convergent
power series.
Proof. We shall first show that we indeed have ai (0) = 0 for all i and E(0, 0) = c.
Suppose that am−i (0) = 0 for all i < k and for some k < m. Then we will have
cy m + . . . = (E(0, 0) + . . . )(am−k (0)y k + . . . + a1 (0)y m−1 + y m ).
By comparing the coefficients of y k on both sides we get 0 = E(0, 0)am−k (0), hence
am−k (0) = 0. Therefore by induction we get ai (0) = 0 for all i. Then by comparing
the coefficients of y m on both sides we get c = E(0, 0).
After dividing F (x, y) and E(x, y) by c, we may assume that c = 1. Since E(x, y)
is a unit of K[[x, y]] by Corollary 2.1.2, we replace it by its inverse say H(x, y). Also
we replace F (x, y) by y m − G(x, y). Then the equation to be solved becomes
(y m − G(x, y))H(x, y) = y m + a1 (x)y m−1 + . . . + am (x).
We express G(x, y), H(x, y) as power series in y as

G(x, y) = gi (x)y i , H(x, y) = hi (x)y i
i≥0 i≥0
and denote the homogeneous parts of gi (x), hi (x) of degree j by gij (x), hij (x) for
all j. The condition that y m is the leading form of F (0, y) then becomes
gi0 (x) = gi (0) = 0
for all i ≤ m. Furthermore if we compare the coefficients of y k+m on both sides of

the above equation, we get

hk (x) − gk+m−i (x) hi (x) = δ0 (k)
0≤i≤k+m
for all k, and for each k this can be replaced by

hkl (x) − gk+m−i,l−j (x) hij (x) = δ0 (k) δ0 (l)
0≤i≤k+m 0≤j≤l
for all l. In the above, δ0 (k) is the function taking the value 1 at k = 0 and 0
elsewhere. If we incorporate the fact that gi0 (x) = 0 for all i ≤ m, then we finally
get

(∗) hkl (x) = δ0 (k)δ0 (l) + gk+mi,0 (x)hil (x)
0≤i<k

+ gk+m−i,l−j (x)hij (x)
0≤i≤k+m 0≤j<l
for all k, l. If now we introduce a function φ on N2 as φ(i, j) = αi + βj with α, β in

R, then the condition that both φ(k − 1, l) and φ(k + m, l − 1) are less than φ(k, l)
becomes 0 < αm < β. This condition is satisfied by α = 1, β = m + 1. We shall
use the so-normalized φ. We observe that (∗) permits us to determine hkl (x) by an
26 JUN-ICHI IGUSA
induction on φ(k, l) for all k, l starting with h00 (x) = 1. Hence H(x, y) and also
a1 (x), . . . , am (x) are uniquely dertermined. We have thus shown that a solution by
formal power series exists and is unique.
We shall show that if G(x, y) is a convergent power series, then the unique
H(x, y) is also a convergent power series. As we have seen in section 2.1, if we
choose M , r > 0 suitably, then we will have
M
G(x, y) << G◦ (x, y) = (x1 + . . . + xn )p y q ,
r p+q

in which the summation is over N2 and indicates the restriction q > m for
p = 0. We shall show that the unique H ◦ (x, y) for G◦ (x, y) gives a dominant series
for H(x, y). We shall first make H ◦ (x, y) explicit. If we write
y i
G◦ (x, y) = ◦
gij (x) ,
r
i,j≥0
we will have j
◦ x1 + . . . + xn
gij (x) =M
r
◦
with the exception that gi0 (x) = 0 for all i ≤ m. Therefore by (∗) we get
y i
H(x, y) << H ◦ (x, y) = h◦ij (x) ,
r
i,j≥0
in which
h◦kl (x) = δ0 (k) δ0 (l)

x1 + . . . + xn l−j
M
+ h◦il (x) + h◦ij (x)
rm r
0≤i<k 0≤i≤k+m 0≤j<l
for all k, l. If we put ρ = M/r m and X = (x1 + . . . + xn )/r, then we see by an

induction on φ(k, l) that h◦kl (x) = akl X l , in which

(∗∗) akl = δ0 (k)δ0 (l) + ρ( ail + aij )
0≤i<k 0≤i≤k+m 0≤j<l
for all k, l. If we put Y = y/r, then we have only to show that

H ◦ (x, y) = aij X j Y i
i,j≥0
is a convergent power series in X, Y .

We shall show that
0 < akl ≤ αk β l ,
in which α = 2ρ + 1, β = αm+1 + 1, for all k, l. Since φ(k, l) = 0 implies k = l = 0

and since a00 = 1 by (∗∗), we shall assume that φ(k, l) > 0 and apply an induction
on φ(k, l). Then by (∗∗) we will have

0 < akl ≤ ρ ( αi β l + αi β j )
0≤i<k 0≤i≤k+m 0≤j<l
1 k
= (α − 1)β l + (αk − α−m−1 )(β l − 1) < αk β l .
2

The induction is thus complete. Therefore the series aij X j Y i is convergent for
−1 −1
|X| < β and |Y | < α .
The rest of the proof is straightforward. If F (x, y) is a convergent power series,
so is G(x, y) = y m − F (x, y), hence also H(x, y) by what we have shown. Since
H(0, 0) = 1, the inverse E(x, y) of H(x, y) is a convergent power series. Also
a1 (x)y m−1 + · · · + am (x) = F (x, y)H(x, y) − y m
is a convergent power series. If we replace y by distinct ci in K with small |ci |K

for i = 1, . . . , m, we get m convergent power series b1 (x), . . . , bm (x), and they are
K-linear combinations of a1 (x), . . . , am (x). Since up to sign the determinant of
the coefficient-matrix is the product of ci − cj for all i < j, it is different from
0. Therefore a1 (x), . . . , am (x) become K-linear combinations of b1 (x), . . . , bm (x),
hence they are also convergent power series.
In the case where K = C the usual proof of Weierstrass’ preparation theorem is
by Cauchy’s integral formula. The proof by Cauchy’s calculus of limits is due to H.
Späth [53]. At any rate any monic polynomial
y m + a1 (x)y m−1 + . . . + am (x),
in which the coefficients a1 (x), . . . , am (x) are power series satisfying ai (0) = 0 for
all i, is called a Weierstrass polynomial.
Corollary 2.3.1 The ring Kx = Kx1 , . . . , xn of convergent power series is

a unique factorization ring.
Proof. We put An = Kx1 , . . . , xn and denote the leading degree of any f (x) in

An \{0} by ldeg(f ). Then f (x) = g(x)h(x) with f (x), g(x), h(x) in An \{0} implies
ldeg(f ) = ldeg(g) + ldeg(h) and f (x) is a unit of An if and only if ldeg(f ) = 0.
Therefore we see that every f (x) in An \{0} can be expressed as a product of
irreducible elements with the number of factors at most equal to ldeg(f ). The
problem is to show that the product-decomposition is unique up to a unit. We
observe that if f (x) is in A1 \{0}, then up to a unit f (x) is equal to xe1 with e =
ldeg(f ). This shows that A1 is a unique factorization ring. We shall therefore
assume that n > 1 and apply an induction on n. We take a(x), b(x), p(x) from
An \{0} such that p(x) is irreducible and divides the product a(x)b(x). We have
only to show that p(x) then divides either a(x) or b(x). In doing so we may apply
any automorphism to An
28 JUN-ICHI IGUSA
If g is an
element of GLn (K) with gij as its (i, j)-entry, then the correspondence
xi → yi = gij xj gives rise to a K-automorphism of An . If f (x) is any element
of An \{0} with fm (x) as its leading form, since every complete field is infinite, we
will have fm (a) = 0 for some a in K n \{0}. We can then find g in GLn (K) with
a as its last column. In fact if the k-th entry of a is different from 0, we can take
e1 , . . . , ek−1 , ek+1 , . . . , en , where e1 = t (1, 0, . . . , 0), etc., as the first, ..., the (n−1)-
th columns of g. If in f (y) = f (gx) we put xi = 0 for all i < n, then we get cxm n +. . .
with c = fm (a) = 0. By applying the above observation to a(x)b(x)p(x) as f (x),
we may assume that a(0, xn )b(0, xn )p(0, xn ) = 0. Then by Theorem 2.3.1 we may
further assume that a(x), b(x), p(x) are Weierstrass polynomials in Bn = An−1 [xn ].
We know by induction and by a consequence of Gauss’ lemma that Bn is a unique
factorization ring. Furthermore a Weierstrass polynomial in Bn is a unit of An if
and only if it is 1; that a Weierstrass polynomial is irreducible in An if and only if
it is irreducible in Bn . Therefore p(x) divides either a(x) or b(x).
We remark that if K is any infinite field, the above proof is applicable without
any change to K[[x]] = K[[x1 , . . . , xn ]], hence it is also a unique factorization ring.
We might mention that both K[[x]] and Kx are noetherian rings, and the idea
of the proof for K[[x]] is as follows. If A is any ideal of K[[x]], then the leading
forms of elements of A\{0} generate an ideal a of K[x]. If we choose a finite subset
I of A\{0} such that the set of leading forms of its elements forms an ideal basis
for a, then I forms an ideal basis for A.
2.4 K-analytic manifolds and differential forms

There are two ways to define real-analytic manifolds, one non-intrinsically by
“charts” and another intrinsically by “sheaves.” The fact is that if K is any com-
plete field, then K-analytic manifolds can be defined in the same way as in the case
where K = R. For the sake of completeness, we shall review some basic definitions.
If x = (x1 , . . . , xn ), where x1 , . . . , xn are variables or letters, then by the usual
practice x will also be considered as a variable point of K n . Suppose that U is
a nonempty open subset of K n and f : U → K is a map. If at every point
a = (a1 , . . . , an ) of U there exists an element fa (x) of Kx − a = Kx1 −
a1 , . . . , xn − an such that f (x) = fa (x) for any variable point x near a, then f
is called a K-analytic function on U . As we have seen in section 2.1, such an f is
differentiable and all its partial derivatives are K-analytic functions on U . Suppose
that U is as above and f : U → K m is a map. If every fi in f = (f1 , . . . , fm )
are K-analytic functions on U , then f is called a K-analytic map. Let X denote
a Hausdorff space and n a fixed nonnegative integer. Then a pair (U, φU ), where
U is a nonempty open subset of X and φU is a bicontinuous map from U to an
open subset φU (U ) of K n , is called a chart. Furthermore φU (x) = (x1 , . . . , xn ) for
a variable point x of U are called local coordinates of x. A set of charts {(U, φU )}
is called an atlas if the union of all U is X and for every U , U such that U ∩ U = ∅
the map
φU ◦ φ−1
U : φU (U ∩ U ) → φU (U ∩ U )

is K-analytic. Two atlases are considered to be equivalent if their union is also

an atlas. This is an equivalence relation and any equivalence class is called an n-

dimensional K-analytic structure on X. If {(U, φU )} is an atlas in the equivalence
class, we say that X is an n-dimensional K-analytic manifold defined by {(U, φU )}
or simply an n-dimensional K-analytic manifold, and we write n = dim(X). We
observe that every nonempty open subset U of K n is an n-dimensional K-analytic
manifold defined by the atlas consisting of one chart (U, φU ), in which φU is the
inclusion map U → K n .
Suppose that X, Y are K-analytic manifolds respectively defined by {(U, φU )},
{(V, ψV )} and f : X → Y is a map. If for every U , V such that U ∩ f −1 (V ) = ∅,
where f −1 (V ) is the preimage of V under f , the map
ψV ◦ f ◦ φ−1
U : φU (U ∩ f
−1
(V )) → K dim(Y )
is K-analytic, then f is called a K-analytic map. The K-analyticity of f defined

above does not depend on the choice of atlases. We shall not repeat this kind of
remark. If f : X → Y , g : Y → Z are K-analytic maps of K-analytic manifolds,
the composite map g ◦ f : X → Z is K-analytic. If f : X → K is a K-analytic
map, it is called a K-analytic function on X. If f is a K-analytic function on X
such that f (a) = 0 at every a in X, then Corollary 2.1.2 shows that 1/f is also a
K-analytic function on X. If (U, φU ) is a chart on X and φU (x) = (x1 , . . . , xn ) as
above, and f is a K-analytic function on X, then we shall denote ∂(f ◦ φ−1 U )/∂xi
simply by ∂f /∂xi for 1 ≤ i ≤ n.
If X, Y are K-analytic manifolds respectively defined by {(U, φU )}, {(V, ψV )},
then {(U × V, φU × ψV )}, where (φU × ψV )(x, y) = (φU (x), ψV (y)) for every (x, y)
in U × V , gives an atlas on the product space X × Y . Therefore X × Y becomes
a K-analytic manifold with dim(X × Y ) = dim(X) + dim(Y ). We call X × Y the
product manifold of X and Y . Suppose that X is a K-analytic manifold defined
by {(U, φU )} and Y is a nonempty open subset of X. If for every U = Y ∩ U = ∅
we put φU = φU U , the restriction of φU to U , then {(U , φU )} gives an atlas on
Y . Therefore Y becomes a K-analytic manifold with dim(Y ) = dim(X). We call
Y an open submanifold of X. Suppose that Y is a nonempty closed subset of an
n-dimensional K-analytic manifold X and 0 < p ≤ n such that an atlas {(U, φU )}
defining X can be chosen with the following property: If φU (x) = (x1 , . . . , xn )
and U = Y ∩ U = ∅, then there exist K-analytic functions F1 , . . . , Fp on U such
that firstly U becomes the set of all x in U satisfying F1 (x) = . . . = Fp (x) =
0 and secondly ∂(F1 , . . . , Fp )/∂(x1 , . . . , xp )(a) = 0 at every a in U . Then by
Corollary 2.1.1-(ii) the correspondence x → (F1 (x), . . . , Fp (x), xp+1 , . . . , xn ) gives
a K-bianalytic map from a neighborhood of a in U to its image in K n . It follows
from this fact that if we denote by V the intersection of such a neighborhood of a and
Y , and put ψV (x) = (xp+1 , . . . , xn ) for every x in V , then {(V, ψV )} for all V and
for each U above gives an atlas on Y . Therefore Y becomes a K-analytic manifold
with dim(Y ) = n − p. We call Y a closed submanifold of X, p = dim(X) − dim(Y )
the codimension of Y in X, and denote p by codimX (Y ).
We shall define K-analytic differential forms on a K-analytic manifold X. If U ,
V are neighborhoods of an arbitrary point
a of X and f , g are K-analytic functions
respectively on U, V such that f W = g W for some neighborhood W of a contained
30 JUN-ICHI IGUSA
in U , V , then we say that f , g are equivalent. If we denote by OX,a , or simply by Oa ,

the set of such equivalence classes, then Oa becomes a commutative ring containing
K. We shall use the same notation for f and its equivalence class. We observe that
f (a) is well defined for every f in Oa and Oa is a local ring with its maximal ideal
ma defined by f (a) = 0. Furthermore if (U, φU ) is a chart with U containing a and
φU (x) = (x1 , . . . , xn ) as before, then ma has x1 − x1 (a), . . . , xn − xn (a) as its ideal
basis, and Oa becomes isomorphic to Kx − x(a) . We shall use Oa to define the
tangent space Ta (X) of X at a. In order to make the definition accessible, we recall
the following fact in calculus.
If a = (a1 , . . . , an ) is a point of Rn and v = (v1 , . . . , vn ) is a vector in Rn , then
the derivative in the direction v of a differentiable function f at a is defined as
df (a + tv) ∂f
(0) = (a)vi .
dt ∂xi
1≤i≤n
If we denote the LHS by ∂f , then the operation ∂ is R-linear and
(∗) ∂(f g) = (∂f )g(a) + f (a)(∂g)
for all differentiable functions f, g at a. Furthermore the vector v can be recovered

from such a ∂ as v = (∂x1 , . . . , ∂xn ).
We now define Ta (X) as the vector space over K of K-linear maps ∂ : Oa → K
satisfying (∗) for all f, g in Oa and denote its dual space by Ωa (X). We recall that
for any vector space E over a field K, its dual space E ∗ is the vector space over K
of all K-linear maps v ∗ from E to K; we write v ∗ (v) = [v, v ∗ ] for every v, v ∗ in V ,
V ∗ . We shall show in the present case that Ωa (X) can be identified with the factor
space ma /m2a as [∂, f + m2a ] = ∂f for all ∂ in Ta (X) and f in ma . If for any f in
Oa we denote by (df )a the image of f − f (a) in ma /m2a , then ∂f clearly depends
only on ∂ and (df )a . Furthermore if φU (x) = (x1 , . . . , xn ), then (dx1 )a , . . . , (dxn )a
form a K-basis for ma /m2a and (df )a can be written uniquely as
∂f
(df )a = (a)(dxi )a .
∂xi
1≤i≤n
Therefore we have only to observe that the correspondence f → (∂f /∂xi )(a) defines
an element (∂/∂xi )a of Ta (X) and
∂ ∂
∂= ∂xi , [ , (dxj )a ] = δij
∂xi a ∂xi a
1≤i≤n
for all ∂ in Ta (X) and all i, j. We observe that if (df1 )a , . . . , (dfn )a for f1 , . . . , fn in
Oa are linearly independent, then by Corollary 2.1.1-(ii) there exists a neighborhood
V of a in U such that ψV = (f1 , . . . , fn ) gives a K-bianalytic map from V to an
open subset of K n . Therefore the addition of (V, ψV ) to the given atlas {(U, φU )}
on X will produce an equivalent atlas. We shall apply such a process whenever it
becomes necessary. We call (f1 , . . . , fn ) local coordinates of X around a.
We shall use the Grassmann or the exterior algebra of a vector space. We shall
briefly recall its definition. In general, if A is a vector space over an arbitrary field K
equipped with a K-bilinear multiplication A × A → A, then A is called a K-algebra.
We shall consider, for the time being, only associative K-algebras each with the unit
element. If E is a vector space over K, assumed to be finite dimensional, then the
exterior algebra (E) of E is the K-algebra generated by E with v 2 = 0 for every
v in E as its “defining relation.” A more precise definition is as follows. If E, E
are vector spaces over K, their tensor product E ⊗ E is the vector space over K
with a K-bilinear map (v, v ) → v ⊗ v from E × E to E ⊗ E such that E ⊗ E
is spanned by the image and dimK (E ⊗ E ) = dimK (E) dimK (E ). If we choose
K-bases for E, E , the set of formal products of their members forms a K-basis for
E ⊗ E . This fact can be used as a non-intrinsic definition of E ⊗ E . At any rate,
if we denote by T (E) the direct sum of K, E, E ⊗ E, ... and define a product in
T (E) by ⊗, then T (E) becomes a K-algebra, and it is called the tensor algebra of
E. IfI(E) denotes the two-sided ideal of T (E) generated by v ⊗ v for all v in E,
then (E) is the factor ring T (E)/I(E). If v1 , . . . , vp are elements of E, the image
of v
1 ⊗ . . . ⊗ vp in (E) is denoted by v1 ∧ . . . ∧ vp and the K-span of such elements
p
by (E). We have
v ∧ v = 0, v ∧ v + v ∧ v = 0
p
for every v, v in E. As a vectorspace (E) is the direct sum of (E) for 0 ≤ p ≤ n
if dimK (E) = n, hence dimK ( (E)) = 2n .
If now we take Ωa (X) as E and if φU (x) = (x1 , . . . , xn ) with U containing a,
then we get
p
(Ωa (X)) = K (dxi1 )a ∧ . . . ∧ (dxip )a ;
i1 <...<ip
we shall denote it by Ωpa (X) for 0 ≤ p ≤ n. We say that α is a differential form of

degree p on X if α(a) is in Ωpa (X) for every a in X. If we replace a by a variable
point x of U , then we write dxi , etc. instead of (dxi )x , etc. A differential form α of
degree p on X has a local expression

α(x) = fU,i1 ...ip (x) dxi1 ∧ . . . ∧ dxip ,
i1 <...<ip
in which fU,i1 ...ip are K-valued functions on U . If they are all K-analytic functions
on U for every U , we say that α is a K-analytic differential form of degree p on
X. In particular, if f is any K-analytic function on X, then df is a K-analytic
differential form of degree 1 on X.
If f : X → Y is a K-analytic map of K-analytic manifolds and β is a K-
analytic differential form of degree p on Y , then we get a similar differential form
(δf )∗ (β) on X as follows. If a is any point of X and f (a) = b, then a K-linear map
δa f : Ta (X) → Tb (Y ) is defined as (δa f )(∂)(g) = ∂(g ◦ f ) for all g in OY,b and its
dual map (δa f )∗ : Ωb (Y ) → Ωa (X) is given by
(δa f )∗ (g + m2b ) = g ◦ f + m2a

32 JUN-ICHI IGUSA
for all gin mb . We observe

that (δa f )∗ uniquely extends to a K-algebra homomor-
phism (Ωb (Y )) → (Ωa (X)). If we write δf instead of δx f , then we have

(δf )∗ ( g dg1 ∧ . . . ∧ dgp ) = (g ◦ f ) d(g1 ◦ f ) ∧ . . . ∧ d(gp ◦ f )
for all K-analytic functions g, g1 , . . . , gp on Y or on its open submanifolds. There-
fore if (U, φU ), (V, ψV ) are charts on X, Y with φU (x) = (x1 , . . . , xn ), ψV (y) =
(y1 , . . . , ym ) satisfying U = U ∩ f −1 (V ) = ∅ so that

β(y) = gV,j1 ...jp (y) dyj1 ∧ . . . ∧ dyjp
j1 <...<jp
with K-analytic functions gV,j1 ...jp on V , then we have

(δf )∗ (β)(x) = (gV,j1 ···jp ◦ f )(x) d(yj1 ◦ f ) ∧ · · · ∧ d(yjp ◦ f )
j1 <···<jp
on U . We shall sometimes write f ∗ instead of (δf )∗ . In the special case where

dim(X) = dim(Y ) = p = n, if we put gV = gV,1...n , then we simply have
∂(y1 , · · · , yn )
f ∗ (β)(x) = gV (f (x)) · dx1 ∧ · · · ∧ dxn .
∂(x1 , . . . , xn )
2.5 Critical sets and critical values

Let K denote any complete field and f : X → Y a K-analytic map of K-analytic
manifolds; for every a in X and b = f (a) let δa f : Ta (X) → Tb (Y ), (δa f )∗ : Ωb (Y ) →
Ωa (X) denote the corresponding dual K-linear maps. Then a is called a critical
point of f if δa f is not surjective, i.e., if (δa f )∗ is not injective. The set Cf of all
critical points of f is called the critical set of f . We shall consider the special case
where Y = K, hence f is a K-analytic function on X. In that case we see that a is
a critical point of f if and only if δa f = 0, i.e., (df )a = 0. If a is a critical point of f ,
then f (a) is called a critical value of f . We shall denote by Vf the set of all critical
values of f . The critical set Cf is an important geometric object associated with f .
In the following we shall examine Vf in the case where f is a polynomial function
on a vector space X over K. We shall start with a generalization of Ta (X).
Let L denote an extension, i.e., an extension field, of any field F , R a commu-
tative F -algebra, and θ an F -algebra homomorphism from R to L; let DerF (R, L)
denote the set of all F -linear maps ∂ from R to L satisfying
∂(ab) = (∂a)(θb) + (θa)(∂b)
for every a, b in R. Then DerF (R, L) forms a vector space over L. We observe that
Ta (X) can be written as DerK (Oa , K) with θϕ = ϕ(a) for every ϕ in Oa . If now
θ is injective and S is a multiplicative subset of R free from zero so that θ extends
uniquely to S −1 R, then every ∂ in DerF (R, L) extends also uniquely to an element
∂ # of DerF (S −1 R, L) as
a (∂a)(θb) − (θa)(∂b)
∂# =
b (θb)2
for every a in R and b in S. The correspondence ∂ → ∂ # gives an L-linear bijection

from DerF (R, L) to DerF (S −1 R, L). Therefore we shall denote ∂ # simply by ∂.
In particular if R is a finitely generated integral domain over F , i.e., of the form
F [x] = F [x1 , . . . , xn ] with F (x) as its quotient field, we get an L-linear bijection
from DerF (F [x], L) to DerF (F (x), L). We shall make DerF (F [x], L) explicit.
We introduce variables t1 , . . . , tn , put F [t] = F [t1 , . . . , tn ], and choose an ideal
basis I for the kernel of the F -algebra homomorphism F [t] → F [x] defined by
ti → xi for 1 ≤ i ≤ n. If f (t) is arbitrary in F [t], θx = a, i.e., θxi = ai for all i, and
∂ is in DerF (F [x], L), then
∂f
(∗) ∂f (x) = (a)vj ,
∂tj
1≤j≤n
in which vj = ∂xj for 1 ≤ j ≤ n. In particular

∂fi
∂fi (x) = (a) vj = 0
∂tj
1≤j≤n
for every fi (t) in I. Conversely if v1 , . . . , vn are elements of L satisfying the above

condition, then we can easily verify that ∂f (x) is well defined by (∗) and ∂ gives an
element of DerF (F [x], L).
Lemma 2.5.1 Let x1 , . . . , xn denote elements of an extension field L of a field F

such that x1 , . . . , xd are algebraically independent over F and F (x) = F (x1 , . . . , xn )
is separably algebraic over F (x ) = F (x1 , . . . , xd ); define DerF (F [x], L) by using the
inclusion map of F [x] in L as θ. Then the correspondence ∂ → (∂x1 , . . . , ∂xd ) gives
an L-linear bijection from DerF (F [x], L) to Ld .
Proof. Since F (x) is separably algebraic over F (x ), we can write F (x) = F (x , y)
with y satisfying f◦ (y) = 0 for a unique irreducible monic polynomial f◦ (t) in
F (x )[t], where t is a variable. Since y is a simple root of f◦ (t), we have (df◦ /dt)(y) =
0. By multiplying a common denominator of the coefficients, we can convert f◦ (t)
into a primitive polynomial f1 (x , t) in F [x ][t] = F [x , t]. Then by a consequence
of Gauss’ lemma the kernel of the F -algebra homomorphism F [x , t] → F [x , y]
defined by x → x , t → y is the principal ideal generated by f1 (x , t). Therefore
if we denote the coordinates on Ld+1 by v1 , . . . , vd+1 , then by our previous obser-
vation the correspondence ∂ → (∂x1 , . . . , ∂xd , ∂y) gives an L-linear bijection from
DerF (F [x , y], L) to the subspace of Ld+1 defined by
∂f1 ∂f1
(x , y) vi + (x , y) vd+1 = 0,
∂xi ∂t
1≤i≤d
in which (∂f1 /∂t)(x , y) = 0. Since F (x , y) = F (x), we also know that

DerF (F [x , y], L), DerF (F (x), L), DerF (F [x], L) all have the same dimension as
vector spaces over L. Therefore the correspondence ∂ → (∂x1 , . . . , ∂xd ) gives an
L-linear bijection from DerF (F [x], L) to Ld .
34 JUN-ICHI IGUSA
Theorem 2.5.1 Let F denote a field with char(F ) = 0 and f (t) any element of the
polynomial ring F [t] = F [t1 , . . . , tn ]; define Cf as the set of all a in F n satisfying
(∂f /∂t1 )(a) = . . . = (∂f /∂tn )(a) = 0 and Vf as the set of f (a) for all a in Cf .
Then Vf is a finite subset of F .
Proof. If L is any extension of F , then Cf , Vf can be defined relative to L and

they respectively contain the original Cf , Vf . Therefore by extending F we may
assume for our purpose that it is algebraically closed. We consider the ideal a of
F [t] generated by ∂f /∂t1 , . . . , ∂f /∂tn and take its minimal representation as an
intersection of primary ideals:
∂f
a= F [t] = qj .
∂ti
1≤i≤n 1≤j≤r
If r = 0, hence a = F [t], we see that Cf and Vf are empty sets. Therefore we shall
assume that r > 0, choose q = qj and put p = r(q). If we denote by xi the image of
ti in F [t]/p, then F [x] = F [x1 , . . . , xn ] is an integral domain and (∂f /∂ti )(x) = 0
for 1 ≤ i ≤ n. We shall show that x0 = f (x) is in F .
If x0 is not in F , since F is algebraically closed, it is transcendental over F .
Therefore we may assume after a permutation that x0 , x1 , . . . , xd for some d ≥ 0
are algebraically independent over F and F (x) = F (x0 , x) is algebraic, necessarily
separable by char(F ) = 0, over F (x0 , x1 , . . . , xd ). Then by Lemma 2.5.1 there exists
an element ∂ of DerF (F [x], F (x)) so that ∂x0 takes any preassigned value in F (x).
On the other hand x0 = f (x) implies
∂f
∂x0 = (x) ∂xi = 0.
∂ti
1≤i≤n
We thus have a contradiction.

We now take any point a = (a1 , . . . , an ) of Cf . Then t1 −a1 , . . . , tn −an generate
a maximal ideal m of F [t] which contains a. Since the product of q1 , . . . , qr is
contained in a, hence in m, some q = qj is contained in m. Then p = r(q) is contained
in m. Therefore, in the above notation, we get an F -algebra homomorphism F [x] →
F [a] = F as xi → ai for 1 ≤ i ≤ n. Since x0 = f (x) is in F , this implies x0 = f (a).
Since the number of x0 is at most equal to r, we get card(Vf ) ≤ r including the
case where r = 0.
In the above theorem if we drop the assumption that char(F ) = 0, then it
becomes false. In fact if char(F ) = p > 0 and f (t) = tp1 + . . . + tpn , then Cf = F n .
Therefore if F is algebraically closed, then Vf = F , which is infinite. On the other
hand, if f (t) is homogeneous, then Vf is either an empty set or {0} provided that p
does not divide deg(f ). This follows immediately from the classical Euler identity
∂f
ti = deg(f )f (t)
∂ti
1≤i≤n
valid for any homogeneous polynomial f (t).

Chapter 3
Hironaka’s desingularization
theorem
3.1 Monoidal transformations
Hironaka’s desingularization is achieved by successive monoidal transformations;
they have been known in algebraic geometry for a long time. We just mention
O. Zariski’s paper [62] which contains a rigorous definition of a general monoidal
transformation and its basic properties. We shall explain a monoidal transformation
with smooth center following A. Borel and J.-P. Serre [4]. We fix a complete field
K and start with a definition of the projective space Pn (K).
We regard two points of K n+1 \{0} to be equivalent if they differ by a scalar
factor in K × and denote the set of all equivalence classes by Pn (K). If t is a point
of Pn (K), therefore, it is represented by some (t1 , . . . , tn+1 ) in K n+1 \{0}, called
the homogeneous coordinates of t. The condition ti = 0 on t is independent of
the choice of its homogeneous coordinates and defines a subset Ui of Pn (K) for
1 ≤ i ≤ n + 1. If t is in Ui , then a map φ : Ui → K n is well defined as
t1 ti−1 ti+1 tn+1
φi (t) = ,··· , , ,··· , .
ti ti ti ti
We observe that φi is a bijection. We shall make the map
φj ◦ φ−1
i : φi (Ui ∩ Uj ) → φj (Uj ∩ Ui )
for i = j explicit. After a permutation we may assume that i = 1 and j = 2. Then

we will have
φ1 (U1 ∩ U2 ) = φ2 (U2 ∩ U1 ) = K × × K n−1 .
Furthermore if we put φ1 (t) = (u1 , . . . , un ), φ2 (t) = (v1 , . . . , vn ), then φ2 ◦ φ−1
1 :
K × × K n−1 → K × × K n−1 is given by
1 u2 un
(∗) (v1 , v2 , . . . , vn ) = , ,... , ,
u1 u1 u1
hence u1 = 1/v1 , u2 = v2 /v1 , . . . , un = vn /v1 . Therefore the map φ2 ◦ φ−1

1 is K-
bianalytic. In particular if we topologize Ui by the condition that φi is bicontinuous,
then Ui and Uj induce the same topology on Ui ∩ Uj . We call a subset U of Pn (K)
35
36 JUN-ICHI IGUSA
open if and only if U ∩ Ui is open in Ui for all i. Then Pn (K) becomes a Hausdorff
space. Furthermore, again by (∗), we see that {(Ui , fi )} gives an atlas on Pn (K),
hence Pn (K) becomes an n-dimensional K-analytic manifold. We keep in mind that
the map K n+1 \{0} → Pn (K) defined by (t1 , . . . , tn+1 ) → t is K-analytic, hence
continuous, and its restriction to the subset defined by max(|t1 |K , . . . , |tn+1 |K ) = 1
is surjective.
We now take an n-dimensional K-analytic manifold X and a closed submanifold
C with p = codimX (C) ≥ 2, and define the monoidal transformation f : X # → X
with center C. It will have the following properties: Firstly X # is also an n-
dimensional K-analytic manifold; secondly f is a K-analytic map which induces a
K-bianalytic map X # \f −1 (C) → X\C, where the preimage f −1 (C) of C under f
is a closed submanifold of X # of codimension 1, called the exceptional divisor of f ;
thirdly f −1 (a) for every a in C is a closed submanifold of X # which is K-bianalytic
to Pp−1 (K). In particular f is surjective. In the special case where C is a point f
is often called the quadratic transformation with center C.
We take an atlas {(U, φU )} on X with φU (x) = (x1 , . . . , xn ) such that if U ∩C =
∅, then it consists of all x in U satisfying x1 = . . . = xp = 0. We then define for each
U an n-dimensional K-analytic manifold U # equipped with a K-analyticsurjection
fU : U # → U and piece them together to get X # and f : X # → X as f U # = fU .
If U ∩ C = ∅, we simply take U # = U and fU = idU , the identity map of U . If
U ∩ C = ∅, then U # is the closed subset of U × Pp−1 (K) defined as follows: It
consists of all (x, t) satisfying xi tj − xj ti = 0 for 1 ≤ i < j ≤ p, in which (t1 , . . . , tp )
are the homogeneous coordinates of t. We put fU (x, t) = x. By definition if x
is not in C, then t has (x1 , . . . , xp ) as its homogeneous coordinates and fU−1 (x)
consists of the single point (x, t) for that t. On the other hand if x is in C, then
fU−1 (x) = x × Pp−1 (K). After this simple observation, we shall examine U # more
closely.
We take the open covering of Pp−1 (K) by Vi defined by ti = 0 and introduce
local coordinates of t in Vi as
t1 ti−1 ti+1 tp
(u1 , . . . , up−1 ) = ,... , , ,... ,
ti ti ti ti
for 1 ≤ i ≤ p. Then we can easily see that Ui# = U # ∩ (U × Vi ) is defined by
(∗∗) x j = xi u j (1 ≤ j < i), xj = xi uj−1 (i < j ≤ p).
Therefore if we denote by Wi the image of φU (U ) in K n−p+1 under the projection

(x1 , . . . , xn ) → (xi , xp+1 , . . . , xn ), then
φ# (x, t) = (xi , xp+1 , . . . , xn , u1 , . . . , up−1 )
gives a bicontinuous map from Ui# to an open subset φ# #

i (Ui ) of Wi × K
p−1
. We
# # −1
shall examine φj ◦ (φi ) for i = j. After a permutation we may assume that i = 1
and j = 2. Then we will have
× ×
φ# # #
1 (U1 ∩ U2 ) ⊂ W1 × K × K
p−2
, φ# # #
2 (U2 ∩ U1 ) ⊂ W2 × K × K
p−2
.
Furthermore if (u1 , . . . , up−1 ) and (v1 , . . . , vp−1 ) are the local coordinates of t re-
spectively in V1 and V2 , then
# −1
φ#
2 ◦ (φ1 ) : φ# # # # # #
1 (U1 ∩ U2 ) → φ2 (U2 ∩ U1 )
is given by
1 u2 up−1
(x1 , xp+1 , . . . , xn , u1 , u2 , . . . , up−1 ) → x1 u1 , xp+1 , . . . , xn , , ,... , .
u1 u1 u1
Therefore it is clearly K-analytic. We have thus shown that {(Ui# , fi# )} gives an
atlas on U # , hence it becomes an n-dimensional K-analytic manifold. Also we
see by (∗∗) that the map fU : U # → U is K-analytic. Furthermore fU−1 (U ∩ C)
becomes a closed submanifold of U # of codimension 1 because we see by (∗∗) that it
is defined in each Ui# by xi = 0. Finally fU maps U # \fU−1 (U ∩ C) K-bianalytically
to U \(U ∩ C) because we see again by (∗∗) that on the open subset of U defined
by xi = 0 the inverse of fU is given by
x1 xi−1 xi+1 xp
x → xi , xp+1 , . . . , xn , , . . . , , ,... ,
xi xi xi xi
for 1 ≤ i ≤ p.
We shall construct X # and f : X # → X out of the set {(U # , fU )} for all U in
the atlas {(U, fU )} on X. We take U , U with U = U ∩ U = ∅. If U ∩ C = ∅,
then fU , fU respectively map fU−1 (U ), fU−1
(U ) K-bianalytically to U . We shall
#
identify the above open subsets of U , (U ) by this bijection. If U ∩ C = ∅, then
#
we shall identify (x, t) in fU−1 (U ) with the (x, t) in fU−1

(U ). This identification
is based on the following fact: If φU (x) = (x1 , . . . , xn ), fU (x) = (x1 , . . . , xn ) and
(u1 , . . . , up−1 ) are the local coordinates of t in Vi , then
# −1 #
φ#
U ◦ (φU ) : φ# #
U (U ∩ (U × Vi )) → φU ((U ) ∩ (U × Vi ))
#
is given by
(xi , xp+1 , . . . , xn , u1 , . . . , up−1 ) → (xi , xp+1 , . . . , xn , u1 , . . . , up−1 ).
Since x1 , . . . , xn are K-analytic functions of (x1 , . . . , xn ), we see by (∗∗) that the
above map is K-analytic. Therefore if we define X # as the union of all U # and
topologize X # as in the case of Pn (K), then X # becomes a Hausdorff space and
the set of charts defined above gives an atlas on X # . In this way X # becomes an
n-dimensional K-analytic manifold. If finally we define f : X # → X as f U # = fU
for every U , then f becomes a K-analyic map with the properties stated in the
beginning.
We add the following remark for our later use. If K is any field with an absolute
value |·|K , then for any ε > 0 we define Iε as the set of all a in K satisfying |a|K ≤ ε.
By making ε smaller if necessary, we shall assume that ε = |a0 |K for some a0 in K × .
We observe that K is locally compact if and only if I1 is compact. This follows
from the fact that a → a−1 0 a maps Iε bicontinuously to I1 . If K is locally compact,
38 JUN-ICHI IGUSA
then K is complete. Furthermore all K-analytic manifolds are locally compact.

We recall that Pn (K) is a continuous image of the subset of K n+1 \{0} defined by
max(|t1 |K , . . . , |tn+1 |K ) = 1. Since this subset is compact, Pn (K) is also compact.
Therefore if f : X # → X is a monoidal transformation, then f is a proper map,
i.e., the preimage under f of any compact subset of X is a compact subset of X # .

(analytic form)
We shall explain Hironaka’s desingularization theorem in his fundamental paper
[20]. Actually, we shall do so only in the special case we need and without proof.
We start with some preliminary observations and definitions.
Let X denote an n-dimensional K-analytic manifold and E a closed submanifold
of X of codimension 1. Then at every point a of E there exist local coordinates
(x1 , . . . , xn ) of X around a such that E is defined locally around a by x1 = 0, i.e.,
E ∩ U for some neighborhood U of a is defined by x1 = 0. We call x1 = 0 or rather
x1 itself a local equation of E around a. We may assume that xi (a) = 0 for all i.
A characterization of a local equation f of E around a is that f is an element of
the local ring Oa of X at a satisfying f E = 0 and (df )a = 0. We shall show that
two local equations f and g of E around a differ by a unit of Oa . We have only
to show that g divides f in Oa . We may assume that g = x1 . We know that Oa is
isomorphic to Kx = Kx1 , . . . , xn and by Lemma 2.1.2 the image f (x) of f
in Kx can be written uniquely as f (x) = x1 f1 (x) + f2 (x ) with f1 (x) in Kx
and f2 (x ) in Kx2 , . . . , xn . Since f and x1 are local equations of E around a,
we see that the function f2 defined by f2 (x ) on some neighborhood of 0 in K n−1
is 0. Then by Lemma 2.1.3, we get f2 (x ) = 0, i.e., f is divisible by g in Oa .
Suppose now that we have a set of closed submanifolds Ei of X of codimension
1 for i in an index set I satisfying the following condition: At every point a of X,
if Ei1 , . . . , Eip are all the Ei containing a with respective local equations f1 , . . . , fp
around a, then (df1 )a , . . . , (dfp )a are linearly independent over K or, equivalently,
there exist local coordinates of X around a of the form (f1 , . . . , fp , fp+1 , . . . , fn ).
In other words all the Ei passing through a “meet transversally” at a. Then we say
that the set {Ei ; i ∈ I} has normal crossings. This clearly implies p ≤ n. Therefore
if we denote by N the nerve complex of {Ei ; i ∈ I} as defined by P. Alexandroff,
then the dimension of the simplicial complex N is at most n − 1. We recall that
a p-simplex of N means any set of p + 1 elements of {Ei ; i ∈ I} with nonempty
intersection.
We take f (x) from the polynomial ring K[x1 , . . . , xn ], which we regard as a
subring of Kx1 , . . . , xn . Then f (x) defines a K-analytic function f on X = K n .
We shall assume that f (x) is not in K, i.e., f is not a constant function on X. We
shall further assume that n > 1. We shall denote by f −1 (0) the preimage of {0}
under f , i.e., the set of zeros of f in X. Also we put
f −1 (0)sing = f −1 (0) ∩ Cf ,
in which Cf is the critical set of f . Then we have the following three possibilities:
Firstly f −1 (0) = ∅; secondly f −1 (0) = ∅ and f −1 (0)sing = ∅; thirdly f −1 (0)sing = ∅.
All these cases occur, e.g., for K = R. Simple respective examples are as follows:

f (x) = x2i + 1, x2i − 1, x2i − x2j (1 ≤ p ≤ n).
1≤i≤n 1≤i≤n 1≤i≤p p<j≤n
At any rate in the second case f −1 (0) becomes a closed submanifold of X of codi-
mension 1 with f as its local equation around every point of f −1 (0). In an over-
simplified manner we can say that Hironaka’s desingularization theorem or rather
its consequence gives a method to improve the third case. An exact statement,
including the trivial case where n = 1, is as follows:
Theorem 3.2.1 Let K denote a complete field with char(K) = 0 and f (x) any
element, not in K, of the polynomial ring K[x1 , . . . , xn ] for n ≥ 1; put X =
K n . Then there exist an n-dimensional K-analytic manifold Y , a finite set E =
{E} of closed submanifolds of Y of codimension 1 with a pair of positive integers
(NE , nE ) assigned to each E, and a K-analytic map h : Y → X satisfying the
following conditions: Firstly, h is the composite map of a finite number of monoidal
transformations each with a smooth center; secondly,

(f ◦ h)−1 (0) = E
E∈E
and h induces a K-bianalytic map
Y \h−1 (f −1 (0)sing ) → X\f −1 (0)sing ;
thirdly, at every point b of Y if E1 , . . . , Ep are all the E in E containing b with

respective local equations y1 , . . . , yp around b and (Ni , ni ) = (NE , nE ) for E = Ei ,
then there exist local coordinates of Y around b of the form (y1 , . . . , yp , yp+1 , . . . , yn )
such that

f ◦h=ε· yiNi , h∗ ( dxi ) = η · yini −1 · dyi
1≤i≤p 1≤i≤n 1≤i≤p 1≤i≤n
on some neighborhood of b, in which ε, η are units of the local ring Ob of Y at b. In

particular E has normal crossings.
We observe that if
f −1 (0)smooth = f −1 (0)\Cf
is not empty and if we denote by E the union of those E not contained in
h−1 (f −1 (0)sing ), then h gives rise to a K-bianalytic map of E \h−1 (f −1 (0)sing )
to f −1 (0)smooth . We call E the strict transform of f −1 (0) under h. The nerve
complex N (E) of E with the function E → (NE , nE ), called the numerical data, on
the set of its vertices is an important combinatorial object associated with f (x) or
rather f −1 (0)sing .
40 JUN-ICHI IGUSA
3.3 Desingularization of plane curves

We shall outline the classical desingularization of plane curves, i.e., the case n = 2
in Theorem 3.2.1, and then explain the details by examples. We take any field F
with char(F ) = 0 and a polynomial f (x, y) = 0 in two variables x, y with coefficients
in F such that
∂f ∂f
f (0) = (0) = (0) = 0;
∂x ∂y
we have denoted (0, 0) by 0. If fm (x, y) is the leading form of f (x, y) as an element
of F [[x, y]] necessarily for m ≥ 2, then, as we have seen in Chapter 2.3, we may
assume that f (0, y) = cy m + . . . with c in F × . We have shown there that f (x, y)
differs from a Weierstrass polynomial
Px (y) = y m + a1 (x)y m−1 + . . . + am (x)
in F [[x]][y] by a unit of F [[x, y]]. We shall assume, for the sake of simplicity, that
f (x, y) is irreducible in Ω[[x, y]] for an algebraically closed extension Ω of F . Then
Px (y) is irreducible in Ω((x))[y]. Therefore if η is a zero of Px (y), then L = Ω((x))(η)
is an extension of Ω((x)) of degree m. The fact is that if x1/m is any m-th root of
x, then L = Ω((x1/m )). One way to see this is as follows:
We observe that K = Ω((x)) is a complete non-archimedean field with OK =
Ω[[x]] and xOK as its maximal ideal. Therefore L is also a complete nonarchimedean
field and, since OK /xOK = Ω is algebraically closed, we will have OL = Ω[[πL ]]
and xOL = πL m
OL . This follows, e.g., from Proposition 11.6.1. Furthermore, as we
×
have remarked at the end of Chapter 2.1, the m-th power map from OL to itself is
surjective. Therefore x 1/m
is in OL and x 1/m
OL = πL OL , hence OL = Ω[[x1/m ]]
and L = Ω((x1/m )).
Since η is an element of OL , it becomes a power series in x1/m . We write this
“Puiseux series” as

η= a0i xi + a1i x(µ1 +i)/ν1 + a2i x(µ2 +i)/ν1 ν2 + . . .
0<i≤j0 0≤i≤j1 0≤i≤j2

(µg +i)/ν1 ···νg
+ agi x ,
i≥0
in which the exponents are strictly increasing, a10 a20 · · · ag0 = 0, µi , νi are relatively
prime positive integers with νi > 1 for 1 ≤ i ≤ g, and ν1 ν2 · · · νg = m. Furthermore
µ1 > ν1 . A basic fact is that the g pairs
(µ1 , ν1 ), (µ2 , ν2 ), . . . , (µg , νg )
depend only on the factor ring Ω[[x, y]]/Ω[[x, y]]f (x, y) and, to some extent, they
determine the ring. At any rate they are called the characteristic pairs of the factor
ring.
We shall now replace F by a complete field K and put X = K 2 . We shall
assume, for the sake of simplicity, that f −1 (0)sing = {0}. Then there exists a
unique shortest sequence of quadratic transformations such that their composition
h : Y → X has the properties stated in Theorem 3.2.1. We might mention that in

the present case f −1 (0) is not just {0}. At any rate we recall that the quadratic
transform X # of X with any point (a, b) of X as its center is covered, up to K-
bianalytic maps, by two copies of K 2 . More precisely, if (x, y) are the coordinates
on X, then X # is covered by X1 = K 2 with coordinates (x1 , y1 ) and X1 = K 2
with coordinates (x1 , y1 ) such that the restrictions of the quadratic transformation
X # → X to X1 , X1 are given by
(x − a, y − b) = (x1 , x1 y1 ) = (x1 y1 , y1 ).
Therefore the open subsets K × K × , K × × K respectively of X1 , X1 are identified as

(x1 , y1 ) = (1/y1 , x1 y1 ), (x1 , y1 ) = (x1 y1 , 1/x1 ). The exceptional curve of X # → X
has x1 , y1 as its local equations in X1 , X1 . If now we denote by E the set of all
exceptional curves on Y , numbered as E1 , . . . , ET by the order of their “creation”,
and the strict transform ET +1 of f −1 (0) under h, then T and (NI , nI ) = (NE , nE )
where E = EI for all I can be described by the characteristic pairs. Actually
(NI , nI ) = (1, 1) for I = T + 1 is clear.
We shall elaborate on the above statement. If a0 , a1 are relatively prime positive
integers, the Euclid algorithm to find their GCD, which is 1, gives rise to a sequence
k0 , k1 , . . . , kt in N as
a0 = k0 a1 + a2 , a1 = k1 a2 + a3 , ... , at−1 = kt−1 at + 1,
in which a1 > a2 > . . . > at = kt > 1 for some t > 0 and k1 , . . . , kt−1 > 0. We shall
write a0 /a1 = [k0 , k1 , . . . , kt ]. In this notation we introduce kij as
µi /νi − µi−1 = [ki0 , ki1 , . . . , kiti ],
where µ0 = 0, and put
Ii = k10 + k11 + . . . + kiti , mi = νi+1 · · · νg

for 1 ≤ i ≤ g. Then T = Ig = kij and
NIi NIi−1 µi
= + − µi−1 νi2 ,
mi mi−1 νi
µi
n Ii = (nIi−1 + − µi−1 )νi
νi
for 1 ≤ i ≤ g with the understanding that (NI0 , nI0 ) = (0, 1), hence
n I1 µ1 + ν1
= .
NI1 µ1 ν1 · · · νg
Furthermore nI /NI > nI1 /NI1 if I < I1 and nI /NI > nIi /NIi if I > Ii for 1 ≤ i ≤ g,
hence nI1 /NI1 is smaller than any other nI /NI .
The above-outlined desingularization of f −1 (0) is due to F. Enriques and O.
Chisini, and it is entirely classical. The exact values of (NI , nI ) for all I and the fact
that nI1 /NI1 is smaller than any other nI /NI can be found in [24]. The structure of
42 JUN-ICHI IGUSA
N (E) as a one-dimensional simplicial complex and a beautiful monotonic property

of the function I → nI /NI on the set of its vertices as well as relations of (NI , nI )
for neighboring vertices have later been discovered by L. Strauss [54]. In particular,
N (E) looks exactly like a tree with g branching vertices at EIi for 1 ≤ i ≤ g. In
the following we shall illustrate the situation in two examples.
Example 1. If we take y 2 −x3 as f (x, y), then f (x, y), ∂f /∂x, ∂f /∂y all vanish only
at 0. Furthermore f (x, y) is irreducible in Ω[[x, y]], f (x, y) itself is a Weierstrass
polynomial, and η = x3/2 is the Puiseaux series of η. Therefore (3, 2) is the only
characteristic pair and 3/2 = [1, 2], hence T = I1 = 3 and (N3 , n3 ) = (6, 5). We
shall verify these and other properties directly without any assumption on char(K).
We apply a quadratic transformation, abbreviated as QT, to X with 0 as its
center. Then in X1 , X1 we have
f (x, y) = x21 (y12 − x1 ) = (y1 )2 (1 − (x1 )3 y1 ),

dx ∧ dy = x1 dx1 ∧ dy1 = y1 dx1 ∧ dy1 .
Since the two curves in X1 with local equations y1 , 1−(x1 )3 y1 have normal crossings,
we apply a QT to X1 with 0 as its center. Then, by increasing the subscripts by 1,
in X2 , X2 we have
f (x, y) = x32 (x2 y22 − 1) = (x2 )2 (y2 )3 (y2 − x2 ),

dx ∧ dy = x22 dx2 ∧ dy2 = x2 (y2 )2 dx2 ∧ dy2 .
Since the two curves in X2 with local equations x2 , x2 y22 − 1 have normal crossings,
we apply a QT to X2 with 0 as its center. Then in X3 , X3 we have
f (x, y) = x63 y33 (y3 − 1) = (x3 )2 (y3 )6 (1 − x3 ),

dx ∧ dy = x43 y32 dx3 ∧ dy3 = x3 (y3 )4 dx3 ∧ dy3 .
We observe that the three curves in X3 with local equations x3 , y3 , y3 − 1 have

normal crossings and the three curves in X3 with local equations x3 , y3 , 1 − x3 also
have normal crossings. Therefore if we denote by Y the union of X1 , X2 , X3 , X3 and
define h on each one of them as the composition of X1 → X, X2 → X1 → X, X3 →
X2 → X1 → X, X3 → X2 → X1 → X, then h : Y → X gives a desingularization of
f −1 (0). We observe that a list of local equations for E1 , E2 , E3 , E4 is as follows:
E1 : y1 , x3 ; E2 : x2 , y3 ; E3 : x3 , y3 ;

3
E4 : 1 − (x1 ) y1 , x2 y22 − 1, y3 − 1, 1 − x3 .
Therefore (Ni , ni ) = (2, 2), (3, 3), (6, 5), (1, 1) for i = 1, 2, 3, 4 and N (E) has three
segments, i.e., 1-simplices, joining E3 to E1 , E2 , E4 .
Example 2. If we put
f (x, y) = y 4 − 2x3 y 2 − 4x6 y + x6 − x9 , η = x3/2 + x9/4 ,

then f (x, y) is a Weierstrass polynomial Px (y) and η 2 + x3 = x3/2 (2η + x3 ), hence
Px (η) = 0. Furthermore Ω((x))(η) contains x1/4 , hence Ω((x))(η) = Ω((x1/4 )), and
hence f (x, y) is irreducible in Ω[[x, y]]. By rewriting f (x, y) as f (x, y) = (x3 +

y 2 )2 − x3 (x3 + 2y)2 we can easily see that f −1 (0)sing consists of all (a, b) satisfying
a3 + b2 = a3 + 2b = 0, i.e., (0, 0) and (a, 2) where a3 = −4. If 2 is not a cube in
K, e.g., if K = Q2 , therefore, the condition f −1 (0)sing = {0} is satisfied. We can
avoid such an artificial condition if we agree to replace X by a small neighborhood
of 0. At any rate (3, 2), (9, 2) are the characteristic pairs and 3/2 = 9/2 − 3 = [1, 2],
hence T = I2 = 6 and (N3, n3 ) = (12, 5), (N6 , n6 ) = (30, 13). We shall verify these
and other properties directly.
We apply a QT to X with 0 as its center. Then in X1 , X1 we have
f (x, y) = x41 f1 (x1 , y1 ) = (y1 )4 f1 (x1 , y1 ),

dx ∧ dy = x1 dx1 ∧ dy1 = y1 dx1 ∧ dy1 ,
f1 (x, y) = (y 2 − x)2 − 4x3 y − x5 ,
f1 (x, y) = (x3 y − 1)2 − x9 y 5 − 4x6 y 3 .
We apply a QT to X1 with 0 as its center. Then in X2 , X2 we have
f (x, y) = x62 f2 (x2 , y2 ) = (x2 )4 (y2 )6 f2 (x2 , y2 ),

dx ∧ dy = x22 dx2 ∧ dy2 = x2 (y2 )2 dx2 ∧ dy2 ,
f2 (x, y) = (1 − xy 2 )2 − 4x2 y − x3 ,
f2 (x, y) = (x − y)2 − x5 y 3 − 4x3 y 2 .
4 12
f (x, y) = x12 6
3 y3 f3 (x3 , y3 ) = (x3 ) (y3 ) f3 (x3 , y3 ),
dx ∧ dy = x43 y32 dx3 ∧ dy3 = x3 (y3 )4 dx3 ∧ dy3 ,
f3 (x, y) = (1 − y)2 − x6 y 3 − 4x3 y 2 ,
f3 (x, y) = (x − 1)2 − x5 y 6 − 4x3 y 3 .
We apply a QT to X3 with (1, 0) as its center. Since (1, 0) in X3 and (0, 1) in X3
represent the same point, we could have applied a QT to X3 with (0, 1) as its center.
At any rate in X4 , X4 we have
f (x, y) = (1 + x4 )4 x14 12
4 y4 f4 (x4 , y4 )
= (1 + x4 y4 )4 (y4 )14 f4 (x4 , y4 ),
dx ∧ dy = (1 + x4 )x54 y44 dx4 ∧ dy4
= (1 + x4 y4 )(y4 )5 dx4 ∧ dy4 ,
f4 (x, y) = 1 − 4(1 + x)3 xy 3 − (1 + x)5 x4 y 6 ,
f4 (x, y) = x2 − 4(1 + xy)3 y − (1 + xy)5 y 4 .
44 JUN-ICHI IGUSA
f (x, y) = (1 + x25 y5 )4 x15 14

5 y5 f5 (x5 , y5 )
= (1 + x5 (y5 )2 )4 (y5 )15 f5 (x5 , y5 ),
dx ∧ dy = (1 + x25 y5 )x65 y55 dx5 ∧ dy5
= (1 + x5 (y5 )2 )(y5 )6 dx5 ∧ dy5 ,
f5 (x, y) = x − 4(1 + x2 y)3 y − (1 + x2 y)5 x3 y 4 ,
f5 (x, y) = x2 y − 4(1 + xy 2 )3 − (1 + xy 2 )5 y 3 .
f (x, y) = (1 + x36 y6 )4 x30 14

6 y6 f6 (x6 , y6 )
= (1 + (x6 )2 (y6 )3 )4 (x6 )15 (y6 )30 f6 (x6 , y6 ),
dx ∧ dy = (1 + x36 y6 )x12
6 y6 dx6 ∧ dy6
5
= (1 + (x6 )2 (y6 )3 )(x6 )6 (y6 )12 dx6 ∧ dy6 ,

f6 (x, y) = 1 − 4(1 + x3 y)3 y − (1 + x3 y)5 x6 y 4 ,
f6 (x, y) = x − 4(1 + x2 y 3 )3 − (1 + x2 y 3 )5 x3 y 6 .
If we denote by Y the union of X1 , X2 , X3 \ {(0, 1)}, X4 , X5 , X6 , X6 and define
h on each one of them as the composition of X1 → X, X2 → X1 → X, . . . ,
X6 → X5 → X4 → X3 → X2 → X1 → X, then h : Y → X gives a desingularization
of f −1 (0). In fact, a list of local equations for E1 , E2 , . . . , E7 is as follows:
E1 : y1 , 1 + x4 , 1 + x5 (y5 )2 , 1 + x36 y6 , 1 + (x6 )2 (y6 )3 ;

E2 : x2 , y3 ; E3 : x3 , y4 ; E4 : x4 , y6 ; E5 : y5 , x6 ;
E6 : x6 , y6 ; E7 : f1 , f2 , f3 , f4 , f5 , f6 , f6 ,
in which fi , fi are fi (xi , yi ), fi (xi , yi ) for all i. We can easily verify that E =
{E1 , E2 , . . . , E7 } has normal crossings. Therefore
(Ni , ni ) = (4, 2), (6, 3), (12, 5), (14, 6), (15, 7), (30, 13), (1, 1)
for i = 1, 2, . . . , 7 and N (E) has six segments joining E3 to E1 , E2 , E4 and E6 to

E4 , E5 , E7 .
Chapter 4
Bernstein’s theory
4.1 Bernstein’s polynomial bf (s)
We take a field K with char(K) = 0 and the polynomial ring K[x] =
K[x1 , . . . , xn ] for some n > 0. The formal differentiation ∂/∂xi in K[x] uniquely
extends to an element, also denoted by ∂/∂xi , of DerK (K(x), K(x)) for 1 ≤ i ≤ n.
We shall denote by Dn or simply by D the K-subalgebra of EndK (K(x)), the K-
algebra of all K-linear transformations in K(x), generated by the multiplication
by xi and ∂/∂xi for all i. Furthermore, by an abuse of notation, we shall write
D = K[x, ∂/∂x]. The 2n generators of D satisfy the following Heisenberg commu-
tation relation:
xi xj − xj xi = 0, (∂/∂xi )(∂/∂xj ) − (∂/∂xj )(∂/∂xi ) = 0,

xi (∂/∂xj ) − (∂/∂xj )xi + δij = 0
for 1 ≤ i, j ≤ n. Such a commutation relation indeed appeared in quantum me-

chanics, cf., e.g., H. Weyl [59], p. 83 and its footnote 20, and D is sometimes called
the Weyl algebra. We shall show that D is an associative K-algebra with the unit
element generated by xi , ∂/∂xi for 1 ≤ i ≤ n with the Heisenberg commutation
relation as its defining relation.
Lemma 4.1.1 Let E denote a 2n-dimensional vector space over K with a basis
ξ1 , . . . , ξn , η1 , . . . , ηn and T (E) the tensor algebra of E; let I(E) denote the two-
sided ideal of T (E) generated by
ξi ⊗ ξj − ξj ⊗ ξi , ηi ⊗ ηj − ηj ⊗ ηi , ξi ⊗ ηj − ηj ⊗ ξi + δij
for all i, j. Then the K-algebra homomorphism θ : T (E) → Dn defined by ξi → xi ,

ηi → ∂/∂xi for 1 ≤ i ≤ n gives rise to a K-algebra isomorphism from T (E)/I(E)
to Dn .
Proof. If ξ is any element of E, we denote its p-th power in T p (E) = E ⊗ . . . ⊗ E

by ξ ⊗p . Every element w of T (E) determines d = dw as the smallest d ≥ 0 such
that w is contained in the direct sum of T p (E) for p ≤ d. Now by an induction on
dw we see that w can be expressed in the form

w= cij ξ1⊗i1 ⊗ . . . ⊗ ξn⊗in ⊗ η1⊗j1 ⊗ . . . ⊗ ηn⊗jn + z
45
46 JUN-ICHI IGUSA
with cij in K, z in I(E), and i, j in Nn , and we get

j
θ(w) = cij xi (∂/∂x) , xi = xi11 . . . xinn , etc.
Since for a similar reason every element of D can be written in the above form,
the homomorphism θ is surjective. Therefore we have only to show that θ(w) = 0
implies cij = 0 for all i, j. Suppose otherwise and choose cij = 0 with the smallest
|j| = j1 + . . . + jn . Then
θ(w)xj = j! cij xi ,
i
in which j! = j1 ! . . . jn ! = 0 by the assumption that char(K) = 0. Therefore
LHS = 0 while RHS = 0. We thus have a contradiction.
We shall consider D-modules. In general if A is any associative K-algebra with
the unit element, then a vector space M over K is an A-module, more precisely a left
A-module, if and only if there exists a K-algebra homomorphism θ : A → EndK (M ).
In that case a · ϕ, or simply aϕ, is defined as θ(a)ϕ for every (a, ϕ) in A × M . We
observe that K(x) is a D-module and so is K[x] because it is stable, i.e., mapped to
itself, under the operations of D. We change our notation to introduce the kind of
D-modules which we shall closely examine. We replace the above K by K0 , denote
by s another variable in addition to x1 , . . . , xn , and put K = K0 (s). The Weyl
algebra D will be relative to this K. We take an element f (x) of K0 [x]\{0}, denote
by S the multiplicative subset of K0 [x] generated by f (x), and put
K[x]f = S −1 K[x] = K0 (s)[x1 , . . . , xn , 1/f (x)].
This can be converted into a D-module as follows.
In the notation of Lemma 4.1.1 we first convert K[x]f into a T (E)-module as
ξi · ϕ(x) = xi ϕ(x), ηi · ϕ(x) = ∂ϕ/∂xi + sϕ(x)(∂f /∂xi )/f (x)
for 1 ≤ i ≤ n. We observe that K[x]f is stable under the operations of T (E).
Therefore we have only to show that the 3n2 generators of the ideal I(E) annihilate
every ϕ(x) in K[x]f . This can be done by formal computations. We shall do it by
using the following obvious lemma:
Lemma 4.1.2 Let a0 , a1 , a2 , . . . denote a finite number of elements of any field F
with char(F ) = 0 satisfying a0 + a1 s + a2 s2 + . . . = 0 for infinitely many s in Z.
Then ai = 0 for all i.
Now the verification goes as follows: Let w denote any one of the above 3n2
elements, ϕ(x) any element of K[x]f and express it as ϕ0 (x)/d(s) with ϕ0 (x) in
K0 [s, x1 , . . . , xn , 1/f (x)] and d(s) in K0 [s]\{0}. Then we will have
a 0 + a 1 s + a2 s 2 + . . .
w · ϕ(x) =
d(s)
with a0 , a1 , a2 , . . . in K0 (x). On the other hand, if we replace s by any element of
Z other than the zeros of d(s) so that f (x)s becomes the actual s-th power of f (x),
then we will have
(ξi · ϕ(x))f (x)s = xi ϕ(x)f (x)s , (ηi · ϕ(x))f (x)s = ∂(ϕ(x)f (x)s )/∂xi
for 1 ≤ i ≤ n. Therefore a0 + a1 s + a2 s2 + . . . = 0 for all such s, hence ai = 0 for

all i, and hence w · ϕ(x) = 0. We have thus shown that K[x]f is a D-module. In
the above notation Bernstein’s theorem can be stated as follows:
Theorem 4.1.1 There exists an element P0 of Dn satisfying P0 · f (x) = 1.
This is a fundamental theorem in the theory of local zeta functions, and it was
proved by I. N. Bernstein [3]. Later in this chapter we shall explain his original
proof. In the following we shall make a few preliminary remarks. If we multiply an
element b(s) of K0 [s]\{0} to both sides of P0 · f (x) = 1, then we will have
(∗) P · f (x) = b(s),
in which P is a “polynomial” P (s, x, ∂/∂x) in s, x1 , . . . , xn , ∂/∂x1 , . . . , ∂/∂xn with

coefficients in K0 . If we replace s by any element of Z, then (∗) becomes
P (s, x, ∂/∂x)f (x)s+1 = b(s)f (x)s .
At any rate, the set of all such polynomials b(s) and 0 forms an ideal in K0 [s],
which is a principal ideal ring. A monic polynomial b(s) of the smallest degree is
a generator of this ideal, and it is uniquely determined by f (x) and K0 . We shall
denote it by bf (s) and call bf (s) the Bernstein polynomial of f (x). In the special
case where
f (x) = x21 + . . . + x2n ,

if ∆ denotes the Laplacian (∂/∂xi )2 , then we will have
∆ · f (x) = 4(s + 1)(s + n/2),
hence bf (s) is a factor of (s + 1)(s + n/2). We shall see later that they are actually
equal. This case is classical. In the middle 60’s M. Sato proved a similar statement
for a large class of f (x) in his theory of prehomogeneous vector spaces. We shall
explain a part of this theory in Chapter 6.
4.2 Some properties of bf (s)

Before we start with the proof of Theorem 4.1.1, we shall prove five elementary
properties of bf (s). All proofs are straightforward, and they could have been left as
exercises.
(i) If the Bernstein polynomial of f (x) exists for an extension K0 of K0 , then
it also exists for K0 , and they are equal.
In fact let bf (s) denote the Bernstein polynomial of f (x) relative to K0 and
P · f (x) = bf (s) for P in K0 [s, x, ∂/∂x]; choose a K0 -basis {wα } for K0 and write

P = P α wα , bf (s) = bα (s)wα ,
in which Pα is in K0 [s, x, ∂/∂x] and bα (s) is in K0 [s] for all α. Then we get
Pα · f (x) = bα (s) for all α with bα (s) = 0 for at least one α. Hence the Bernstein
48 JUN-ICHI IGUSA
polynomial bf (s) relative to K0 exists and clearly bf (s) = bf (s)c (s) for some c (s)
in K0 [s]. On the other hand, bα (s) = bf (s)cα (s) for some cα (s) in K0 [s] for all α.
By putting these together we get

cα (s)wα c (s) = 1.
Since c (s) = bf (s)/bf (s) is a monic polynomial, this implies c (s) = 1, hence
bf (s) = bf (s).
(ii) If the Bernstein polynomial of f (x) exists and if f (x ) is obtained from f (x)
by an invertible K0 -linear transformation x → x , then the Bernstein polynomial of
f (x ) also exists, and they are equal.
The proof is as follows: We denote by x, ∂/∂x, etc. the column vectors with xi ,
∂/∂xi , etc. as their i-th entries and put x = g −1 x for any g in GLn (K0 ). Then
∂/∂x = (t g −1 )∂/∂x . Therefore if we put f (x ) = f (x) = f (gx ) and assume that
P (s, x, ∂/∂x) · f (x) = bf (s), then we get
P (s, gx , (t g −1 )∂/∂x ) · f (x ) = bf (s).
Therefore bf (s) exists, and it divides bf (s). The situation is now symmetric and
bf (s) divides bf (s), hence they are equal.
(iii) If c is in K0× and if bf (s) exists, then bcf (s) also exists, and they are equal.
This follows immediately from the fact that each one of xi · ϕ, ∂/∂xi · ϕ for every
ϕ in K[x]f = K[x]cf relative to f (x) and cf (x) are equal.
(iv) If f (x) is in K0× , then bf (s) = 1 while if f (x) is in K0 [x]\K0 and bf (s)
exists, then it is divisible by s + 1.
Since the first part is clear, we shall prove the second part. Suppose that bf (s)
exists for some f (x) in K0 [x]\K0 and that P (s, x, ∂/∂x) · f (x) = bf (s). Then we
will have
P (s, x, ∂/∂x)f (x)s+1 = bf (s)f (x)s
for every s in Z, in particular for s = −1. We observe that for s = −1 the LHS is
in K0 [x] while the RHS is bf (−1)/f (x), and 1/f (x) is not in K0 [x]. This implies
bf (−1) = 0, hence bf (s) is divisible by s + 1.
(v) Suppose that f (x) is in K0 [x]\K0 and there is no point a in Ωn , where Ω is
any algebraically closed extension of K0 , satisfying
∂f ∂f
f (a) = (a) = . . . = (a) = 0.
∂x1 ∂xn
Then bf (s) = s + 1.
This can be proved as follows: By Hilbert’s Nullstellensatz there exist a0 (x),
a1 (x), . . . , an (x) in K0 [x] satisfying
∂f
a0 (x)f (x) + ai (x) = 1.
∂xi
1≤i≤n
Define P as
P = (s + 1)a0 (x) + ai (x) ∂/∂xi .
1≤i≤n
Then we have P · f (x) = s + 1. Since we have seen in (iv) that bf (s), if it exists, is
divisible by s + 1, we get bf (s) = s + 1.
If we take, e.g., C as K0 and f (x) from C[x]\C, then we see by (iv), (v) that
bf (s)/(s + 1) = 1 implies f −1 (0)sing = ∅. In a certain sense the size of bf (s)/(s + 1)
corresponds to the complexity of f −1 (0)sing . We might recall that to simplify
f −1 (0)sing or, more precisely, to replace f −1 (0) by the union of closed submanifolds
of codimension 1 with normal crossings is one of the main objectives of desingu-
larization. We shall determine bf (s) for an f (x) such that f −1 (0) itself has such a
simple structure. Namely, we shall prove the following statement:
If f (x) = xm1 . . . xn , where mi is in N for every i, then
1 mn

bf (s) = (s + j/mi )
1≤i≤n 1≤j≤mi
with the understanding that the i-th factor represents 1 for mi = 0.

Since the general case can be reduced to the case where n = 1 by using multi-
indices, we shall assume that n = 1 and we drop the subscripts, hence f (x) = xm ,
etc.; we shall write d/dx instead of ∂/∂x and exclude the trivial case where m = 0.
We see by an induction on p ≥ 0 that

(d/mdx)p · xq = ( (s + (q − j)/m))xq−p
0≤j<p
for every q in Z. Therefore if we put

b(s) = (s + (m − j)/m) = (s + j/m),
0≤j<m 1≤j≤m
then we get
(d/mdx)m · f (x) = b(s),
hence bf (s) divides b(s). On the other hand, if P is in K0 [s, x, d/dx] and P · f (x) =
bf (s), then by expressing P as

P = cij (s)xi (d/mdx)j
i,j≥0
with cij (s) in K0 [s] we also get

ci,m+i (s) (s + (m − j)/m) = bf (s).
i≥0 0≤j<m+i
This shows that b(s) divides bf (s), hence bf (s) = b(s).
4.3 Reduction of the proof

We shall translate Theorem 4.1.1 into an equivalent statement and proceed to its
proof. In the following lemma Φ(s)|s
→s−r for any function Φ(s) means Φ(s − r).
50 JUN-ICHI IGUSA
Lemma 4.3.1 If ϕ(s, x) is in K[x]f = K0 (s)[x, 1/f (x)] and P (s, x, ∂/∂x) is in
Dn , then for every r in Z we have
P (s, x, ∂/∂x) · f (x)−r ϕ(s, x) = f (x)−r (P (s + r, x, ∂/∂x) · ϕ(s + r, x))|s

→s−r .
Proof. We first observe that the formula is valid in the special cases where
P (s, x, ∂/∂x) = xi , ∂/∂xi for 1 ≤ i ≤ n. Furthermore if the formula is valid for
P1 , P2 in D = Dn , then it is valid for c1 P1 + c2 P2 for any c1 , c2 in K = K0 (s).
Therefore we have only to show that the formula is valid also for P3 = P1 P2 , and
the proof is as follows: If we put
ϕ (s, x) = P2 (s + r, x, ∂/∂x) · ϕ(s + r, x),
then
P3 (s, x, ∂/∂x) · f (x)−r ϕ(s, x) = P1 (s, x, ∂/∂x) · f (x)−r ϕ (s − r, x)

= f (x)−r {(P1 (s + r, x, ∂/∂x) · ϕ (s, x))|s
→s−r }
= f (x)−r {(P3 (s + r, x, ∂/∂x) · ϕ(s + r, x))|s
→s−r } .
Proposition 4.3.1 Theorem 4.1.1 holds if and only if the D-module K[x]f is
finitely generated.
Proof. We shall, for the sake of simplicity, write f, ϕ(s), P (s) instead of f (x), ϕ(s, x),
P (s, x, ∂/∂x). We then have
D · f −r+1 = D · (f · f −r ) = Df · f −r ⊂ D · f −r ,
D · f −r ⊃ K[x] · f −r = K[x]f −r
for r = 0, 1, 2, . . . , hence

D · 1 ⊂ D · f −1 ⊂ D · f −2 . . . , D · f −r = K[x]f .
r≥0
Suppose now that K[x]f is generated as a D-module by a finite subset S of K[x]f

and choose r large enough so that D · f −r+1 contains S. Then K[x]f = D · f −r+1 ,
hence f −r = P (s) · f −r+1 for some P (s) in D. Lemma 4.3.1 then shows that
P (s + r) · f = 1. Conversely suppose that P0 · f = 1 for some P0 = P (s) in D. Then
by Lemma 4.3.1 we get
P (s − 1) · 1 = f −1 , P (s − 2) · f −1 = f −2 , . . . ,
hence K[x]f = D · 1.
In general if K is an arbitrary field, A is a K-algebra, which is associative and
with the unit element, and M is an A-module, then the finite generation problem
of M as an A-module can sometimes be investigated by converting A and M into a
filtered K-algebra and a filtered A-module, and then passing to the corresponding
graded G(A)-module G(M ). In the present case, as we shall see, this method works
perfectly.
We have already used graded algebras and modules to prove the existence of
Hilbert’s functions in Chapter 1.3. If a K-algebra A contains an increasing sequence
of subspaces F0 (A), F1 (A), F2 (A), . . . with A as their union satisfying
Fi (A)Fj (A) ⊂ Fi+j (A)
for all i, j, then A is called a filtered K-algebra. If an A-module M contains an

increasing sequence of subspaces F0 (M ), F1 (M ), F2 (M ), . . . with M as their union
satisfying
Fi (A)Fj (M ) ⊂ Fi+j (M )
for all i, j, then M is called a filtered A-module. In particular A itself is a filtered
A-module. We put Fr (M ) = {0} for r < 0 and call {Fr (M )} a filtration of M . If
M is a graded A-module, where A is a graded K-algebra, and if we put Fr (M ) =
M0 + . . . + Mr , etc., as we have done in Chapter 1.3, then A becomes a filtered
K-algebra and M a filtered A-module. Conversely, if M is a filtered A-module,
where A is a filtered K-algebra, and if we put Gr (M ) = Fr (M )/Fr−1 (M ), etc.,
then the direct sum G(M ) of Gr (M ) for all r becomes a graded module over the
graded K-algebra G(A).
Proposition 4.3.2 If M is a filtered A-module such that the G(A)-module G(M )

is finitely generated, then the A-module M is finitely generated.
Proof. By assumption there exists a finite subset {ψ1 , . . . , ψk } of G(M ) such that
G(M ) = G(A)ψ1 + . . . + G(A)ψk . After expressing each ψi as a finite sum of
homogeneous elements, we may assume that ψi itself is homogeneous, i.e., in Gri (M )
for some ri in N. We shall show that if we choose ϕi from Fri (M ) with ψi as its
image in Gri (M ), then we will have
Fr (M ) = Fr−r1 (A)ϕ1 + . . . + Fr−rk (A)ϕk

for all r in N. That will imply M = Aϕi . If we denote the RHS by Fi (M ), then

clearly Fr (M ) ⊃ Fr (M ) for all r. Therefore we have only to show that Fr (M ) ⊂
Fr (M ) also for all r. Since F−1 (0) = 0, this is clear for r = −1. Therefore we shall

apply an induction on r assuming that Fr−1 (M ) ⊂ Fr−1 (M ), hence Fr−1 (M ) =

Fr−1 (M ), for some r ≥ 0. We take ϕ arbitrarily from Fr (M ) and denote its image
in Gr (M ) by ψ. Then we have ψ = b1 ψ1 + . . . + bk ψk for some bi in Gr−ri (A)
for all
i. If we choose ai from Fr−ri (A) with bi as its image in Gr−ri (A), then ϕ − ai ϕi
is in Fr−1 (M ). Therefore ϕ is in
Fr−1 (M ) + Fr (M ) = Fr−1

(M ) + Fr (M ) = Fr (M ).
The induction is complete and the proposition is proved.

In view of the above proposition, we shall convert D into a filtered K-algebra
and M = K[x]f into a filtered D-module. We have seen in the proof of Lemma
4.1.1 that every element of D can be uniquely expressed as

cij xi (∂/∂x)j , xi = xi11 · · · xinn , etc.
52 JUN-ICHI IGUSA
with cij in K for all i, j in Nn . We define Fr (D) by the condition that |i| + |j| ≤ r
for all r. The Heisenberg commutation relation implies that

xi (∂/∂x)j xi (∂/∂x)j = xi+i (∂/∂x)j+j + ci j xi (∂/∂x)j
for every i, i , j, j in Nn , in which i + i = (i1 + i1 , . . . , in + in ), etc. and

|i | + |j | < |i + i | + |j + j |. This not only shows that F defines a filtra-
tion of D but also the crucial fact that G(D) is K-isomorphic to the polynomial
ring K[x, y] = K[x1 , . . . , xn , y1 , . . . , yn ] in 2n variables under the correspondence
xi → xi , ∂/∂xi → yi for 1 ≤ i ≤ n.
We shall next define Fr (M ) for M = K[x]f so that M becomes a filtered D-
module. We choose α independently of r and define Fr (M ) as the subspace of
f −r K[x] consisting of all ϕ = f −r p, where p = p(x) is in K[x], such that
deg(ϕ) ≤ αr, i.e., deg(p) ≤ (deg(f ) + α)r.
Then the condition that {Fr (M )} forms an increasing sequence with M as its union
becomes α > 0. We observe that the condition Fi (D)Fj (M ) ⊂ Fi+j (M ) for all i,
j is equivalent to xi · Fr (M ), ∂/∂xi · Fr (M ) ⊂ Fr+1 (M ) for 1 ≤ i ≤ n and for all
r. Since xi · ϕ = xi ϕ, hence deg(xi · ϕ) = deg(ϕ) + 1, the first condition becomes
α ≥ 1. Since deg(∂/∂xi · ϕ) ≤ deg(ϕ) − 1 and ∂/∂xi · ϕ is in f −r−1 K[x] for every
ϕ in f −r K[x], there is no new condition from ∂/∂xi · Fr (M ) ⊂ Fr+1 (M ). We shall
therefore take α = 1. We have thus shown that if we define Fr (M ) as the subspace
of f −r K[x] consisting of all ϕ such that deg(ϕ) ≤ r, then M becomes a filtered
D-module.
We might remark that except when we regard K[x]f as a D-module, the field
K need not be K0 (s). It can be any field with char(K) = 0. Furthermore if D is
replaced by a general filtered K-algebra A, then char(K) need not be 0.
4.4 A general theorem on D-modules

We shall start with some general definitions and observations. We fix a K-algebra
A and take A-modules M , N . We say that a K-linear map α : M → N is an A-
homomorphism if α is A-linear, i.e., α(aϕ) = aα(ϕ) for every (a, ϕ) in A × M . We
take A-modules M , M , M and A-homomorphisms α : M → M , β : M → M .
We say that the sequence
(∗) 0 → M → M → M → 0
is exact in the category of A-modules or simply exact as A-modules if α is injective,

β is surjective, and Im(α) = Ker(β). If we identify M with its image under α, this
means that the factor module M/M becomes A-isomorphic to M under β. In the
case where A is a filtered K-algebra and M , M , M are filtered A-modules, (∗) is
called exact as filtered A-modules if it gives rise to a sequence
0 → Fr (M ) → Fr (M ) → Fr (M ) → 0
for every r, which is exact as K-modules, i.e., vector spaces over K. In that case
we clearly have
dimK (Fr (M )) = dimK (Fr (M )) + dimK (Fr (M ))
for all r. In the case where A is a graded K-algebra and M , M , M are graded
A-modules, (∗) is called exact as graded A-modules if it gives rise to a sequence

0 → Mr → Mr → Mr → 0
for every r, which is exact as K-modules. The fact is that G is an “exact functor”
in the following sense: If (∗) is exact as filtered A-modules, then the associated
sequence
0 → G(M ) → G(M ) → G(M ) → 0
is exact as graded G(A)-modules. In fact, if we identify Fr (M ) with its image in
Fr (M ), then the kernel of the K-homomorphism from Gr (M ) = Fr (M )/Fr−1 (M )
to Gr (M ) = Fr (M )/Fr−1 (M ), which is clearly surjective, is
(Fr (M ) + Fr−1 (M ))/Fr−1 (M ) = Fr (M )/(Fr (M ) ∩ Fr−1 (M ))

= Fr (M )/Fr−1 (M ) = Gr (M )
for all r. We observe that if A is a filtered K-algebra and M in the exact sequence
(∗) of A-modules is a filtered A-module, and if we put
Fr (M ) = Fr (M ) ∩ M , Fr (M ) = (Fr (M ) + M )/M
for all r, then M , M become filtered A-modules and (∗) is exact as filtered A-
modules.
Suppose now that M is a filtered A-module for a filtered K-algebra A and
dimK (Fr (M )) = (e/d!)r d + o(r d ), i.e., dimK (Fr (M ))/r d → e/d!,
as r tends to ∞ for some d ≥ 0, e > 0 in Z. Then we say that the filtration F is

of type (d, e) or F is a (d, e)-filtration. We observe that if M has such a filtration,
then necessarily M = 0. There are two immediate examples of such filtrations. We
take the polynomial ring K[x1 , . . . , xm ] as A and M , and regard M as a filtered
A-module with Fr (M ) consisting of all polynomials of degree at most r. Then
dimK (Fr (M )) = card{i ∈ Nm ; |i| ≤ r} = (r + m) . . . (r + 1)/m!
for all r in N, hence F gives an (m, 1)-filtration. Also the filtration of the D-module
K[x]f defined in section 4.3 is an (n, (deg(f ) + 1)n )-filtration. At any rate the
following lemma is clear by definition:
Lemma 4.4.1 Let 0 → M → M → M → 0 denote an exact sequence of filtered
A-modules such that the filtrations of M , M are of types (d , e ), (d , e ) respec-
tively. Then the filtration of M is of type (d, e) where d = max(d , d ) and e = e ,
e + e , e according as d > d , d = d , d < d .
54 JUN-ICHI IGUSA
We also need the fact that the converse of Proposition 4.3.2 holds with addi-
tional information. Suppose that A is a filtered K-algebra and M is an A-module
considered as filtered A-modules under two filtrations F , F . We say that F , F are
equivalent if there exist r0 , s0 in N satisfying

Fr (M ) ⊂ Fr+s 0
(M ), Fr (M ) ⊂ Fr+r0 (M )
for all r. This is clearly an equivalence relation. It follows from the definition
that if F is of type (d, e) and F is equivalent to F , then F is also of type (d, e).
We say that a filtration F of M is standard if G(M ) becomes a finitely generated
G(A)-module.
Proposition 4.4.1 If A is a filtered K-algebra and M is a finitely generated A-
module, then M always has a standard filtration which is unique up to equivalence.
Proof. Suppose that M is generated as an A-module by its finite subset {ϕ1 , . . . , ϕk }.
If we regard Ak as an A-module by the prescription
a · (a1 , . . . , ak ) = (aa1 , . . . , aak ),
then the correspondence (a1 , . . . , ak ) → a1 ϕ1 + ... + ak ϕk gives rise to a surjective
A-homomorphism Ak → M with kernel, say N . If we put

Fr (Ak ) = Fr (A)k , Fr (M ) = Fr (A)ϕi , Fr (N ) = Fr (Ak ) ∩ N,
1≤i≤k
then the sequence 0 → N → Ak → M → 0 becomes exact as filtered A-modules.

Since G is an exact functor, the G(A)-homomorphism G(Ak ) = G(A)k → G(M )
is surjective. Therefore the G(A)-module G(M ) is generated by k elements, hence
the above filtration F of M is standard.
We take an arbitrary standard filtration F of M and show that it is equivalent
to F . As we have seen in the proof of Proposition 4.3.2, we can find a finite subset
{φ1 , . . . , φl }, where each φj is contained in Frj (M ) for some rj in N, such that
Fr (M ) = Fr−r1 (A)φ1 + ... + Fr−rl (A)φl

for all r, hence M = Aφj . In particular we can write

ϕi = aij φj , φj = bji ϕi
1≤j≤l 1≤i≤k
with aij , bji in A for 1 ≤ i ≤ k, 1 ≤ j ≤ l. We may assume that aij , bji are all
contained in Fr0 (A) for some r0 in N. If we put
s0 = max(r0 + r1 , . . . , r0 + rl ),
then we will have

Fr (M ) = Fr (A)ϕi ⊂ Fr+r0 (A)φj ⊂ Fr+s 0
(M ),
1≤i≤k 1≤j≤l

Fr (M ) = Fr−rj (A)φj ⊂ Fr+r0 (A)ϕi = Fr+r0 (M )
1≤j≤l 1≤i≤k
for all r. Therefore F and F are equivalent.

If now A is a filtered K-algebra such that G(A) is K-isomorphic to a polynomial
ring K[x1 , . . . , xm ] for some m > 0, M is a finitely generated A-module different
from 0, and F is a standard filtration of M , then F is a (d, e)-filtration with d ≤ m.
This basic fact follows from Theorem 1.3.3 in view of the fact that

dimK (Fr (M )) = dimK (Gi (M )) = χ(G(M ), r)
0≤i≤r
with deg(χ(G(M ), t)) ≤ m for all large r. Furthermore d, e are independent of the
choice of F by Proposition 4.4.1. Therefore we shall denote them by d(M ), e(M ).
The following statement is the main theorem in Bernstein’s theory:
Theorem 4.4.1 If M is any D-module with a (d, e)-filtration where D = Dn , then
necessarily d ≥ n. Furthermore if d = n, then the length of any strictly increasing
sequence of D-submodules of M is at most equal to e. In particular, the D-module
M is finitely generated.
Since the D-module K[x]f has an (n, (deg(f ) + 1)n )-filtration, Theorem 4.4.1
implies that it is a finitely generated D-module, and that implies Theorem 4.1.1 by
Proposition 4.3.1. As for the proof of Theorem 4.4.1 we shall show at this point
only the fact that the first part implies the second part. We, therefore, assume that
d = n for M and take any finite strictly increasing sequence of D-submodules of M :
0 = L0 ⊂ L1 ⊂ . . . ⊂ Lk .
If we choose ϕi from Li \Li−1 and put Mi = Dϕ1 + . . . + Dϕi for 0 ≤ i ≤ k, then

we get a similar sequence:
0 = M0 ⊂ M1 ⊂ . . . ⊂ Mk .
We observe that Mi and Mi /Mi−1 are finitely generated D-modules different from
0 for 1 ≤ i ≤ k. Therefore, by the first part and Lemma 4.4.1 we get n ≤
d(Mi /Mi−1 ) ≤ d(Mi ) for 1 ≤ i ≤ k. Since Mi is an A-submodule of M which
has an (n, e)-filtration, we get d(Mi ) ≤ n, hence d(Mi ) = d(Mi /Mi−1 ) = n for
1 ≤ i ≤ n. Again by Lemma 4.4.1 we then get
e(Mi ) = e(Mi−1 ) + e(Mi /Mi−1 ) > e(Mi−1 )
for 1 < i ≤ k and e(M1 ) > 0 by definition, hence e(Mi ) ≥ i for 1 ≤ i ≤ k. Since
Mk is an A-submodule of M and d(Mk ) = n, we have e(Mk ) ≤ e, hence k ≤ e.
Therefore the first part of Theorem 4.4.1 indeed implies its second part.
4.5 Completion of the proof

We shall prove the first part of Theorem 4.4.1. We start with preliminary remarks
and two lemmas. Firstly if A is a filtered K-algebra with a polynomial ring as
G(A), M is a finitely generated A-module, and N is any A-submodule of M , then
56 JUN-ICHI IGUSA
N is also a finitely generated A-module. The proof goes as follows: We take a

standard filtration F of M and restrict F to N , i.e., we put Fr (N ) = Fr (M ) ∩ N
for all r. Then G(N ) becomes a G(A)-submodule of G(M ). Therefore G(N ) is a
finitely generated G(A)-module by Theorem 1.3.1, hence N is a finitely generated
A-module by Proposition 4.3.2. Secondly, let A denote a K-algebra and M an A-
module defined by a K-algebra homomorphism θ : A → EndK (M ); let σ denote
any element of the group Aut(A) of K-algebra automorphisms of A. Then we can
convert M into another A-module by using θ ◦ σ instead of θ. We shall denote the
new A-module by M σ ; we keep in mind that M and M σ differ only in the actions
of A. If A is a filtered K-algebra, we shall denote by AutF (A) the subgroup of
Aut(A) defined by the condition that σ keeps Fr (A) for every r. In the special case
where M is a finitely generated A-module such that G(A) is a polynomial ring, we
clearly have d(M σ ) = d(M ), etc. for every σ in AutF (A).
Lemma 4.5.1 Suppose that char(K) = 0, F is the filtration of D = Dn defined in
section 4.3, and σ is any element of AutF (D); regard x, ∂/∂x as column vectors.
Then
σx = ax + b(∂/∂x) + γ , σ(∂/∂x) = cx + d(∂/∂x) + γ ,
in which the 2n × 2n matrix with a, b, c, d as its entry matrices is in the symplectic
group Sp2n (K) and γ = t (t γ t γ ) is in K 2n . Conversely, for every such element of
Sp2n (K) × K 2n the above prescription defines an element σ of AutF (D).
Proof. Since σ keeps F1 (D), we have the above situation with a, b, c, d in Mn (K)
and γ in K 2n . Since σ is in Aut(D), it keeps the defining relation of D. Therefore
if we denote the new x, ∂/∂x by x∗ , (∂/∂x)∗ , then we have
x∗i x∗j − x∗j x∗i = 0, (∂/∂xi )∗ (∂/∂xj )∗ − (∂/∂xj )∗ (∂/∂xi )∗ = 0,
x∗i (∂/∂xj )∗ − (∂/∂xj )∗ x∗i + δij = 0

for 1 ≤ i, j ≤ n. By a straightforward calculation, we see that they are respectively
equivalent to a t b = b t a, c t d = d t c, a t d − b t c = 1n , the unit element of Mn (K).
Therefore the coefficient-matrix is in Sp2n (K) while there is no condition on γ. By
reversing the above argument we see that every element of Sp2n (K) × K 2n gives
rise to an element of AutF (D).
We shall later use two kinds of σ both keeping xi , ∂/∂xi for 1 ≤ i < n and
mapping xn , ∂/∂xn respectively to xn +γn , ∂/∂xn +γ2n and −∂/∂xn +γn , xn +γ2n ,
in which γn , γ2n are in K.
Lemma 4.5.2 Suppose that K is an algebraically closed noncountable field such as
C, t is a variable, and M is a K[t]-module different from 0 with at most countable
dimK (M ); denote by θ : K[t] → EndK (M ) the K-algebra homomorphism defining
M as a K[t]-module. Then there exists an element α of K such that θ(t − α) is not
a unit of EndK (M ).
Proof. We shall assume that θ(t − α) is a unit of EndK (M ) for every α in K
a contradiction. Take any p = p(t) = 0 from K[t] and factorize it as
and derive
p(t) = β · (t − α) with β in K × and α in K. This is possible because K is

algebraically closed. Then θ(p) = β · θ(t − α) is a unit of EndK (M ). Therefore
θ uniquely extends to a K-algebra homomorphism from K(t) to EndK (M ), which
we shall denote also by θ. We choose any ϕ = 0 from M , which is possible because
M = 0, and consider the K-linear map from K(t) to M defined by q → θ(q)ϕ. We
observe that the set {1/(t − α); α ∈ K} is linearly independent over K. Since K
is noncountable and dimK (M ) is at most countable, therefore, the above K-linear
map K(t) → M is not injective. Therefore θ(q)ϕ = 0 for some q = 0; but θ(q) is a
unit of EndK (M ), hence ϕ = 0, a contradiction.
We are ready to prove the first part of Theorem 4.4.1 stating that if M is a
D-module with a (d, e)-filtration, then d ≥ n. Since M = 0, it contains a finitely
generated D-module M = 0 and then d(M ) ≤ d. Therefore we have only to
show that if M = 0 is a fintely generated D-module, then d(M ) ≥ n. We keep in
mind that, since dimK (D) is countable and the D-module M is finitely generated,
dimK (M ) is at most countable. Since d(M ) ≥ 0, the above statement becomes
trivial for n = 0. Therefore we shall apply an induction on n assuming that n >
0. After tensorizing D and M over K by an extension of K, we may assume
that K is algebraically closed and noncountable. We shall denote the K-algebra
homomorphism D → EndK (M ) defining M as a D-module by θ. We shall assume
that d(M ) < n and derive a contradiction.
Put xn = t. Then by Lemma 4.5.2 there exists an element α of K such that
θ(t − α) is not a unit of EndK (M ). We may replace t − α by t and denote the kernel
and the cokernel of θ(t) respectively by M and M . If we put D = Dn−1 , then
θ(t) is a D -homomorphism, hence M and M are D -modules. Since θ(t) is not a
unit of EndK (M ), either M = 0, M = 0 or M = 0. We shall show that either
case will bring a contradiction.
Suppose first that M = 0, hence M = M/tM = 0. Since M is a finitely
generated D-module, it has a standard filtration F by Proposition 4.4.1, and it
gives rise to a standard filtration F of M as its image under M → M , i.e.,
as Fr (M ) = (Fr (M ) + tM )/tM for all r. We observe that Fr (M )/tFr−1 (M ) is
mapped surjectively to Fr (M ) under M → M and θ(t) is injective. Therefore we
get
dimK (Fr (M )) ≤ dimK (Fr (M )) − dimK (Fr−1 (M )) = O(r d(M )−1 ),
i.e., the LHS divided by r d(M )−1 is bounded, as r tends to ∞. Since M = 0, it

contains ψ = 0, which is in Fr0 (M ) for some r0 in N. We introduce a D -module
L and its standard filtration as L = D ψ and Fr (L ) = Fr (D )ψ for all r. Then
by induction we have n − 1 ≤ d(L ). On the other hand, by definition we have
Fr (L ) ⊂ Fr (D)ψ ⊂ Fr+r0 (M ), hence
dimK (Fr (L )) ≤ dimK (Fr+r0 (M )) = O(r d(M )−1 )
as r tends to ∞, and hence d(L ) ≤ d(M ) − 1. Since d(M ) < n by assumption, we

have the contradiction that n − 1 ≤ d(L ) < n − 1.
Suppose next that M = 0 and denote the union of Ker(θ(tm )) for all m > 0 by
N . Then N is a D -submodule of M and it is stable under θ(t). We shall show that
N is also stable under θ(∂/∂t). In doing so we shall use the following consequence
58 JUN-ICHI IGUSA
of the defining relation of D:
(∗∗) (∂/∂t)tm = tm (∂/∂t) + mtm−1
valid for all m in N. If now ϕ is in N , then tm ϕ = 0 for some m > 0. This implies
by (∗∗)
tm+1 ((∂/∂t)ϕ) = (∂/∂t)(tm+1 ϕ) − (m + 1)tm ϕ = 0.
Therefore (∂/∂t)ϕ is in N , hence N is stable under θ(∂/∂t). We have thus shown
that N is a D-submodule of M . By our remark in the beginning N is then a finitely
generated D-module, hence d(N ) is defined and d(N ) ≤ d(M ), and hence d(N ) < n
by assumption. Furthermore N = 0 because N contains Ker(θ(t)) = M = 0.
Therefore we can replace M by N , and we will have the situation that every element
ϕ of M satisfies tm ϕ = 0 for some m > 0 depending on ϕ. The rest of the proof is
as follows.
We shall show that Ker(θ(∂/∂t − α)) = 0 for all α in K. Suppose otherwise
and choose ϕ = 0 from the above kernel for some α in K; also choose the smallest
m > 0 satisfying tm ϕ = 0. Then (∂/∂t)ϕ = αϕ implies by (∗∗)
0 = (∂/∂t)(tm ϕ) = αtm ϕ + mtm−1 ϕ = mtm−1 ϕ.
Since char(K) = 0, this implies tm−1 ϕ = 0, a contradiction. We now take σ in

Lemma 4.5.1 inducing the identity on D and mapping t, ∂/∂t respectively to −∂/∂t,
t. Then M σ becomes a finitely generated D-module different from 0, d(M σ ) =
d(M ) < n, and Ker(θ ◦ σ(t − α)) = 0 for all α in K. We can replace M by M σ ,
and we get Ker(θ(t − α)) = 0 for all α in K. On the other hand, θ(t − α) is not a
unit of EndK (M ) for some α in K by Lemma 4.5.2. If we finally replace t − α by
t, then we get back to the previous case where M = 0. We have seen in that case
that the assumption d(M ) < n brings a contradiction.
Chapter 5
Archimedean local zeta

functions
5.1 The group Ω(K × )
We say that G is a topological group if G is a Hausdorff space and a group such
that the group operations, i.e., the maps G × G → G and G → G defined by
(g, g ) → gg and g → g −1 , are continuous. If G, H are topological groups, we shall
denote by Hom(G, H) the set of all continuous homomorphisms from G to H. We
observe that if G is compact and H has no compact subgroup other than 1, e.g.,
the additive group R, then Hom(G, H) becomes 1, in which 1 denotes the group
consisting of the unit element only. We further observe that if H is commutative,
then Hom(G, H) becomes a commutative group as (θθ )(g) = θ(g)θ (g) for every
θ, θ in Hom(G, H) and g in G. We shall determine Hom(K × , C× ) for K = R, C
after introducing some notation and proving a lemma.
In general, if K is any field with an absolute value | · |K , we shall denote by
K1× the subgroup of K × defined by |a|K = 1. In the case where K = R, C we
respectively put |a|K = |a|, |a|2 for every a in K. We might remark that we have
created a minor discrepancy by introducing the above | · |C but with the advantage
of making some formulas uniform for K = R, C and also for a p-adic field. At any
rate K1× = {±1} if K = R and K1× is the unit circle with center 0 in the complex
plane if K = C. Furthermore, if R× ×
+ denotes the subgroup of R defined by a > 0,
× × ×
then K /K1 becomes isomorphic to R+ under | · |K , and the extension splits, i.e.,
K × = R× ×
+ × K1 .
Lemma 5.1.1 Let I = (−δ, δ) for any δ > 0 denote an open interval in R and θ a
continuous map from I to R with the property that θ(x + y) = θ(x) + θ(y) for all
x, y in I satisfying |x| + |y| < δ and θ(−x) = −θ(x) for all x in I. Then θ(x) = ax
for some a in R and for all x in I. In particular, θ uniquely extends to an element
of Hom(R, R).
Proof. First of all we have θ(0) = 0 and if x1 , . . . , xn are in R satisfying |x1 | +

. . . + |xn | < δ, hence necessarily in I, then by an induction on n we see that
θ(x1 + . . . + xn ) = θ(x1 ) + . . . + θ(xn ). Therefore if m, n are in N and m, n > 1/δ,
then by applying the above remark to xi = 1/mn for 1 ≤ i ≤ n, we get θ(1/m) =
nθ(1/mn), hence mθ(1/m) = mnθ(1/mn). Since the RHS is symmetric in m and
59
60 JUN-ICHI IGUSA
n, it is also equal to nθ(1/n). Therefore, a = mθ(1/m) is independent of m in N

satisfying m > 1/δ. Furthermore if m, n > 0 are in N and n/m is in I, then
θ(n/m) = nθ(1/m) = a(n/m), θ(−n/m) = −θ(n/m) = a(−n/m).
We have thus shown that θ(x) = ax for all x in Q ∩ I, hence for all x in I by
continuity.
We shall determine Hom(R, C× × ×
1 ) and Hom(C1 , C1 ) by using Lemma 5.1.1 start-
ing with Hom(R, C1 ). If we restrict the homomorphism e : R → C×
×
1 defined by
x → e(x) = exp(2πix) to (−1/2, 1/2), then it has a unique inverse, say ψ, over
C× ×
1 \{−1}. If χ is an arbitrary element of Hom(R, C1 ), then I = (−δ, δ) for a small
δ > 0 and θ = ψ ◦ χ will have the property in Lemma 5.1.1, hence θ extends to
R and θ(x) = ax for a unique a in R. This implies χ(x) = e(ax) for all x in R.
The converse is obvious and Hom(R, C× 1 ) is isomorphic to R as χ → a. Similarly,
if χ is an element of Hom(C× × ×
1 , C1 ), then χ ◦ e becomes an element of Hom(R, C1 ),
hence χ(e(x)) = e(ax) for a unique a in R and for all x in R. Since χ ◦ e maps Z
to 1, we see that a is in Z, and χ(t) = ta for all t in C× 1 . The converse is obvious
and Hom(C× ×
1 , C1 ) is isomorphic to Z as χ → a. These are well-known examples in
Pontrjagin’s theory. We are ready to prove the following proposition.
Proposition 5.1.1 If K = R or C, then Ω(K × ) = Hom(K × , C× ) consists of all
ω such that
ω(a) = |a|sK (a/|a|)p ,
in which s is in C and p is in Z considered modulo 2 for K = R. Furthermore under
the correspondence ω → (s, p) the group Ω(K × ) is isomorphic to C × (Z/2Z) or
C × Z according as K = R or C.
Proof. If a is an arbitrary element of K × , then under its product expression a =
|a|(a/|a|) we have K × = R× × × ×
+ × K1 . Since Hom(K1 , R+ ) = 1, therefore, we see that
×
Ω(K ) is isomorphic to
Hom(R× × × × × ×
+ , R+ ) × Hom(R+ , C1 ) × Hom(K1 , C1 ).
We shall determine each factor. Firstly, if ω is in Hom(R× ×

+ , R+ ), then θ = log ◦ ω ◦
exp gives an element of Hom(R, R), hence ω(x) = x for a unique σ in R and for
σ
×
all x in R+ . Secondly, if ω is in Hom(R× ×
+ , C1 ), then θ = ω ◦ exp gives an element
of Hom(R, C1 ), hence ω(x) = x for a unique τ in R and for all x in R×
× iτ
+ . Thirdly,
if ω is in Hom(K1× , C× ×
1 ), then ω(t) = t with p in Z for all t in K1 . In the case
p
where K = R, since K1× = {±1}, we take p mod 2. Then in both cases p becomes
unique. We have thus shown that if ω is any element of Ω(K × ), then we can write
ω(a) = ω(|a|)ω(a/|a|) = |a|σ+iτ (a/|a|)p
for all a in K × with unique σ, τ in R, and p in Z/2Z or Z according as K = R or

C. If we put s = [K : R]−1 (σ + iτ ), then ω(a) takes the form in the proposition.
The converse is clear.
A remarkable fact is that Ω(K × ) is a union of two or countably many copies of
C for K = R or C. In particular, Ω(K × ) is a one-dimensional C-analytic manifold.
Therefore we can define C-analytic functions or simply holomorphic functions on any

nonempty open subset of Ω(K × ) and talk about their meromorphic continuation,
poles, etc. For such purposes we introduce some notation. If s is arbitrary in C,
then
ωs (a) = |a|sK
defines an element ωs of Ω(K × ), and for every ω in Ω(K × ) we will have
|ω(a)| = ωσ(ω) (a)
with σ(ω) in R. If we express ω as in Proposition 5.1.1, then σ(ω) =Re(s), the real
part of s. If σ is arbitrary in R, then we denote by Ωσ (K × ) the open subset of
Ω(K × ) defined by σ(ω) > σ. We shall sometimes denote by Cσ the open subset of
C defined similarly by Re(s) > σ. In this notation Ω0 (K × ) becomes a union of the
right-half plane C0 .
5.2 Schwartz space S(K n )

We start with the definition of S(K n ) for K = R, C. Since S(Cn ) is defined as
S(R2n ) after identifying C with R2 , we assume that K = R and put X = Rn . If ϕ
is any C-valued continuous function on a topological space, its uniform norm ϕ∞
is defined as the supremum of |ϕ(x)| for all x in the space. A C∞ -function Φ on
X, i.e., a C-valued function Φ on X having derivatives of arbitrarily high order,
is called a Schwartz function if P Φ∞ for every P in Dn = C[x, ∂/∂x] is finite.
We observe that the set S(X) of all Schwartz functions on X forms a vector space
over C and it is invariant under any invertible R-linear transformation in X. In the
verification we use the fact that Φ = P Φ∞ for every P in Dn gives a seminorm
on S(X). We recall that if E is a vector space over C, then a seminorm · on E
is defined by the following conditions:
0 ≤ ϕ < ∞, cϕ = |c| ϕ, ϕ + ϕ ≤ ϕ + ϕ
for all ϕ, ϕ in E and c in C. We observe that the finiteness of P Φ∞ for all P in

Dn is equivalent to the finiteness of
Φi,j = sup |xi (∂/∂x)j Φ(x)|

x∈X
for all i, j in Nn . The set of such seminorms is countable. We further observe that
Φ∞ = Φ0,0 and Φ∞ = 0 implies Φ = 0. The following lemma is known in
general topology:
Lemma 5.2.1 Let E denote a vector space over C and { · i ; i ∈ N} a set of

seminorms on E such that ϕi = 0 for all i implies ϕ = 0; define the distance
d(ϕ, ϕ ) in E as

d(ϕ, ϕ ) = (1/2i )ϕ − ϕ i /(1 + ϕ − ϕ i ).
i≥0
62 JUN-ICHI IGUSA
Then E becomes a metric space. Furthermore if {ϕi } is any sequence in E, then

ϕj − ϕk i tends to 0 as j, k tend to ∞ for every i if and only if d(ϕj , ϕk ) tends to
0 as j, k tend to ∞.
We shall outline the proof that E is a metric space, i.e., d(ϕ, ϕ ) = 0 if and only
if ϕ = ϕ , d(ϕ, ϕ ) = d(ϕ , ϕ), and d(ϕ, ϕ ) ≤ d(ϕ, ϕ ) + d(ϕ , ϕ ). The first two
properties are clear while the third property follows from the fact that the function
f (t) = t/(1 + t) for t ≥ 0 is monotone increasing, 0 ≤ f (t) < 1, and
(f (t) + f (t )) − f (t + t ) = tt (2 + t + t )/(1 + t)(1 + t )(1 + t + t ) ≥ 0.
In particular, S(X) is a metric space and the topology in S(X) is invariant under
any invertible R-linear transformation in X. The vector space S(X) with the so-
defined topology is called the Schwartz space of X. If {Φk } is a Cauchy sequence in
S(X), then {(∂/∂x)j Φk } forms a Cauchy sequence relative to the uniform norm for
every j in Nn . Therefore (∂/∂x)j Φk is uniformly convergent to Ψj , say, as k tends
to ∞ and if we put Ψ0 = Ψ, then (∂/∂x)j Ψ = Ψj for every j. This is well known
in calculus; it is proved by using the representation of a differentiable function as
an integral of its derivative and applying an elementary form of Lebesgue’s theorem
reviewed in Chapter 1.1. Furthermore, {Φk i,j } is a Cauchy sequence in R with
Ψi,j as its limit for every i, j. In particular, this shows that Ψ is an element
of S(X), hence S(X) is a complete metric space. We denote the topological dual
of S(X) by S(X) . In other words, S(X) is the subspace of the dual space of
S(X) consisting of its elements which are continuous functions on S(X), i.e., which
convert every null sequence in S(X) into a null sequence in C. An element of
S(X) is called a tempered distribution in X. We recall that a distribution in X was
introduced by L. Schwartz in [51]-1 by using the space D(X) of all C ∞ -functions
on X with compact support instead of S(X) and a tempered distribution in [51]-II
to discuss Fourier transformations.
We shall show, for our later use, the well-known fact that D(X) is dense in S(X).
We might start with A. Cauchy’s historical remark that the R-analytic function
exp(−1/x2 ) on R× completed by the value 0 at x = 0 becomes a C ∞ -function on
R with 0 as its Maclaurin expansion. We replace the above x by (1 − 4(t − a)2 )1/2
and put
θa (t) = exp(−1/(1 − 4(t − a)2 ))
for |t − a| < 1/2 and θa (t) = 0 for |t − a| ≥ 1/2. Then θa becomes a C ∞ -function
on R. Furthermore, if we put I = {0, ±1/2, ±1} and J = I ∪ {±3/2}, then

χ= θa / θb
a∈I b∈J
is a C ∞ -function on R such that χ(t) = 1 for |t| ≤ 1, χ(t) = 0 for |t| ≥ 2 and
χ(−t) = χ(t), 0 ≤ χ(t) ≤ 1 for all t in R. In particular, χ(t) becomes a C ∞ -
function of t2
. In the following lemma and also later r(x) denotes the distance from
0 to x, i.e., ( x2i )1/2 :
Lemma 5.2.2 Take any Φ from S(X) and put Φk (x) = χ(k−1 r(x))Φ(x) for k =
1, 2, . . . . Then {Φk } gives a sequence in D(X) which tends to Φ as k tends to ∞,
i.e., {Φk − Φ} is a null sequence in S(X).
Proof. By definition Φk is a C ∞ -function on X satisfying Φk (x) = 0 for r(x) ≥ 2k,

hence Φk is in D(X). We shall show that
Φk − Φi,j = xi (∂/∂x)j ((χ(k−1 r(x)) − 1)Φ(x))∞
tends to 0 as k tends to ∞ for all i, j. We observe that the RHS is at most equal to

(j!/j ! j !) (∂/∂x)j (χ(k−1 r(x)) − 1)Ψj (x)∞ ,

in which Ψj (x) = xi (∂/∂x)j Φ(x) and the summation is with respect to j in Nn
such that j = j − j is also in Nn . Since Ψj is in S(X), we have only to show that
(∗) lim (∂/∂x)α (χ(k−1 r(x)) − 1)Φ(x)∞ = 0

k→∞
for every α = (α1 , . . . , αn ) in Nn and Φ in S(X). We observe that if we put

t = k−1 r(x), u = r(x)−1 x, then

(∗∗) (∂/∂x)α (χ(t) − 1) = φα,j (t)uj r(x)−|α| ,
|j|≤|α|
in which φα,j is a C ∞ -function on R satisfying φα,j (t) = 0 for |t| ≤ 1. In fact,

(∗∗) holds for |α| = 0 with φ0,0 (t) = χ(t) − 1. Therefore we assume by induction
that (∗∗) holds for some α. If we denote by β the new element of Nn obtained by
increasing αi for some i by 1, then we will have

(∂/∂x)β (χ(t) − 1) = |j|≤|α| {(dφα,j /dt)tui +

φα,j (t)(ji u−1
i − (|α| + |j|)ui )}u r(x)
j −|β|
.
This implies that (∗∗) holds also for β, hence (∗∗) holds for all α. Now if |α| = 0,
then
(∂/∂x)α (χ(k−1 r(x)) − 1)Φ(x)∞ ≤ 2rΦ∞ · k−1
and if |α| > 0, then
 

LHS ≤  φα,j ∞  Φ∞ · k−|α| .
|j|≤|α|
Therefore we certainly have (∗).

As a consequence of Lemma 5.2.2, if T in S(X) has the property that T (ϕ) = 0
for every ϕ in D(X), then T = 0. An example of the Schwartz function is
Φ(x) = exp(−π · r(x)2 ).

64 JUN-ICHI IGUSA
In fact, Φ is an R-analytic function, hence a C ∞ -function, on X. Furthermore, for

every i, j in Nn we have
|xi (∂/∂x)j Φ(x)| ≤ const · r |i|+|j| exp(−πr 2 )|r=r(x) ,
and the RHS tends to 0 as r(x) tends to ∞. Incidentally, the factor π in π · r(x)2
makes the integral of Φ over X to become 1.
We go back to the topological dual S(X) of S(X). In general, let E denote a
vector space over C and E a subspace of the dual space of E; let {Tk } denote any
sequence in E with the property that for every ϕ in E a finite limit
T (ϕ) = lim Tk (ϕ)

k→∞
exists. Then ϕ → T (ϕ) clearly gives a C-linear map from E to C, i.e., T is an

element of the dual space of E. If such a T is always contained in E , then we say
that E is complete. We shall explain, after Gel’fand and Shilov [16], Appendix A,
a proof by M.S. Brodskii of the following basic theorem:
Theorem 5.2.1 The vector space S(X) of tempered distributions in X = Rn is

complete.
Proof. We take a sequence {Tk } in S(X) with the property that for every Φ in
S(X) a finite limit T (Φ) of {Tk (Φ)} as k → ∞ exists. To be proved is the fact that T
is continuous. We shall assume that T is not continuous and derive a contradiction
in three steps.
If T is not continuous, there exists a null sequence {Φp } in S(X) such that
{T (Φp )} is not a null sequence in C. By replacing {Φp } by a subsequence, we may
assume that |T (Φp )| ≥ c for all p and for some c > 0 independent of p. Since {Φp }
is a null sequence in S(X), we have Φp i,j → 0 as p → ∞ for every i, j in Nn .
Therefore for any given k, if we choose p = p(k) sufficiently large, we will have
Φp i,j ≤ 2−2k
for all i, j satisfying |i| + |j| ≤ k. If we put ϕk = 2k Φp , then we have
(∗) ϕk i,j ≤ 2−k (|i| + |j| ≤ k), |T (ϕk )| ≥ 2k c
for every k in N.
We shall next show that if for every k in N we suitably choose k , k ≥ k
depending on k and put ψk = ϕk , Sk = Tk , then we will have

(∗∗) |Sp (ψk )| ≤ 2p−k (0 ≤ p < k), |Sk (ψk )| > |Sk (ψp )| + k.
0≤p<k
Since Tq (ϕ0 ) → T (ϕ0 ) as q → ∞ and |T (ϕ0 )| ≥ c by (∗), if we put ψ0 = ϕ0 and

S0 = Tq for a large q, we will have |S0 (ψ0 )| > 0. Then ψ0 , S0 satisfy (∗∗) for k = 0.
Assume by induction that for some k ≥ 0 we have ψp , Sp for 0 ≤ p ≤ k satisfying
(∗∗). Since Sp (ϕq ) → 0, T (ϕq ) → ∞ as q → ∞, if we put ψk+1 = ϕq for a large

q > k, then we will have

|Sp (ψk+1 )| ≤ 2p−k−1 (0 ≤ p ≤ k), |T (ψk+1 )| > |T (ψp )| + k + 1.
0≤p≤k
Since Tq (ψp ) → T (ψp ) as q → ∞ for 0 ≤ p ≤ k + 1, if we put Sk+1 = Tq for a large

q > k, then we will have

|Sk+1 (ψk+1 )| > |Sk+1 (ψp )| + k + 1.
0≤p≤k
This completes the induction.

We shall show that  

Ψ = lim  ψp 
k→∞
0≤p≤k
exists in S(X). If for any i, j in Nn we choose k1 ≥ k0 ≥ |i| + |j|, then, since

ψk = ϕk for k ≥ k, by (∗) we have

ψk i,j ≤ 2−k ≤ 21−k0 ,
k0 ≤k≤k1 k0 ≤k≤k1
in which 21−k0 → 0 as k0 → ∞. Since S(X) is complete, this implies the existence

of Ψ above in S(X). Then by the continuity of Sk = Tk we get
 

Sk (Ψ) = lim  Sk (ψp ) .
k0 →∞
0≤p≤k0
On the other hand, for all k0 ≥ k by (∗∗) we have

Sk (ψp ) ≥ |Sk (ψk )| − |Sk (ψp )| − |Sk (ψp )|

0≤p≤k0 0≤p<k p>k

> k− 2k−p
= k − 1,
p>k
hence |Sk (Ψ)| ≥ k − 1. On the other hand
lim Sk (Ψ) = lim Tk (Ψ) = T (Ψ)

k→∞ k→∞
is finite. We thus have a contradiction.

We shall now explain Gel’fand and Shilov [16], Appendix 2, especially the impor-
tant “method of analytic continuation.” It is based on the completeness of S(X) .
In fact we can replace S(X) and S(X) respectively by a vector space E over C
and any complete subspace E of its dual space. Let U denote a subset (resp. open
subset) of C, which we assume to be nonempty, and s → Ts an E -valued function
66 JUN-ICHI IGUSA
on U . We say that Ts is continuous (resp. holomorphic) on U if Ts (ϕ) for every ϕ

in E is continuous (resp. holomorphic) on U . If C is a continuous curve of finite
length in C and if Ts is continuous on C, then

S(ϕ) = Ts (ϕ)ds
C
is defined for every ϕ in E and ϕ → S(ϕ) defines an element S of the dual space
of E. We shall show that S is contained in E . We parametrize C by a continuous
function s(t) for 0 ≤ t ≤ 1 and for every k in N we put

S0 = 0, Sk = Tsi (si − si−1 ) (k > 0),
1≤i≤k
in which si = s(i/k) for 0 ≤ i ≤ k. Then Sk is in E and Sk (ϕ) → S(ϕ) as k → ∞

for every ϕ in E. Therefore S is in E by the completeness of E . We shall write

S= Ts ds.
C
Suppose now that Ts is holomorphic on U , s0 is in C, and the punctured disc

{s ∈ C; 0 < |s − s0 | ≤ r} is contained in U for some r > 0. Then Ts (ϕ) for every ϕ
in E can be expanded into its Laurent series

Ts (ϕ) = ck (ϕ)(s − s0 )k ,
k∈Z
in which
1 Ts
ck = ds
2πi |s−s0 |=r (s − s0 )k+1

is in E for every k in Z. We call

Ts = ck (s − s0 )k
k∈Z
the Laurent expansion of Ts at s0 . We say that s0 is a pole of Ts if ck = 0 only for

a finite number of k < 0.
Proposition 5.2.1 Let E denote a vector space over C and E any complete sub-
space of its dual space; let U ⊂ U1 denote open subsets of C, in which U is nonempty
and every connected component of U1 intersects U ; finally let Ts denote an E -valued
holomorphic function on U such that for every ϕ in E the C-valued holomorphic
function Ts (ϕ) on U has a holomorphic continuation to U1 . Then ϕ → Ts (ϕ) defines
an element Ts of E for every s in U1 .
Proof. We may restrict s to any connected component of U1 , hence we may assume
that U1 is connected. If we put
Σ = {s ∈ U1 ; Ts ∈ E },
then U ⊂ Σ ⊂ U1 . We shall derive a contradiction assuming that Σ = U1 . If we

denote by V the largest open subset of C contained in Σ, then U ⊂ V ⊂ Σ. We
observe that if we denote by V̄ the closure of V in C, then (U1 \V )∩V̄ = ∅. Otherwise
U1 \V = U1 \V̄ , hence U1 \V is open in U1 and nonempty because it contains U1 \Σ.
This contradicts the assumption that U1 is connected. We can therefore choose s1
from (U1 \V )∩ V̄ . We denote by dis(s, ∂U1 ) the distance from any point s of C to the
boundary ∂U1 of U1 . Since the subsequent argument becomes simpler if ∂U1 = ∅,
we shall assume otherwise, i.e., dis(s, ∂U1 ) is finite for every s. We choose s0 from
V satisfying |s1 − s0 | < (1/2)dis(s1 , ∂U1 ). Since Ts is an E -valued holomorphic
function on V and s0 is in V , we can expand Ts for every s near s0 into a Laurent
series or rather a Taylor series as

Ts = ck (s − s0 )k
k∈N
with ck in E for all k in N. Furthermore, since Ts (ϕ) for every ϕ in E is holomorphic

on the open disc
W = {s ∈ C; |s − s0 | < dis(s0 , ∂U1 )},
we get

Ts (ϕ) = lim ( cp (s − s0 )p )(ϕ)
k→∞
0≤p≤k
for every s in W . Therefore Ts is in E for every s in W by the completeness of

E and s1 is in W . If s1 is not in W , there exists a point s2 of ∂U1 satisfying
|s1 − s0 | ≥ |s0 − s2 | and that brings the following contradiction:
dis(s1 , ∂U1 ) ≤ |s1 − s2 | ≤ |s1 − s0 | + |s0 − s2 | ≤ 2|s1 − s0 | < dis(s1 , ∂U1 ).
Since W is open, we see that s1 is in V , a contradiction.

We call the extended Ts in Proposition 5.2.1 the holomorphic continuation of
the original Ts to U1 . If further U1 ⊂ U2 ⊂ C, in which U2 is open in C and U2 \U1
is discrete in U2 , such that the extended Ts has every point of U2 \U1 as a pole, then
it is called the meromorphic continuation of the original Ts to U2 .
5.3 Local zeta function ZΦ (ω)

We shall start with two lemmas. We have stated them with proof for the sake of
completeness.
Lemma 5.3.1 Let (X, dx) denote a measure space, e.g., a nonempty open subset
of Rn equipped with the usual measure, U a nonempty open subset of C, and f a
C-valued measurable function on X × U , e.g., a continuous function if X is an open
subset of Rn , with the following properties: (i) If C is any compact subset of U ,
there exists an integrable function φC ≥ 0 on X satisfying
|f (x, s)| ≤ φC (x)

68 JUN-ICHI IGUSA
for all (x, s) in X × C; (ii) f (x, ·) is a holomorphic function on U for every x in

X. Then
F (s) = f (x, s) dx
X
defines a holomorphic function F on U .
Proof. We have only to show that the restriction of F to every open disc in U is
holomorphic. Therefore we may assume that U is connected and simply connected.
We shall use some theorems which we have recalled in Chapter 1.1. Firstly F is
a continuous function on U . In fact, if s0 is any point of U , then by Lebesgue’s
theorem we get

lim F (s) = lim f (x, s) dx
s→s0 s→s0 X

= ( lim f (x, s)) dx = f (x, s0 ) dx = F (s0 ).
X s→s0 X
Secondly if C is any closed curve of finite length in U , then by using theorems of

Fubini and Cauchy we get

F (s) ds = ( f (x, s) dx) ds = ( f (x, s) ds) dx = 0.
C C X X C
Therefore F is holomorphic on U by Morera’s theorem.

Lemma 5.3.2 Let V denote a nonempty open subset of X = Rn with a piecewise
smooth boundary ∂V, ϕ a continuously differentiable function on the closure V̄ of
V in X with polynomial growth, i.e.,
|ϕ(x)| ≤ M · max(1, r(x)m )

for some M, m ≥ 0 and for all x in V̄ , such that ϕ∂V = 0. Then for every Φ in
S(X) and for 1 ≤ i ≤ n we have

(∂ϕ/∂xi )Φ(x) dx = ϕ(x)(−∂Φ/∂xi ) dx.
V V
Proof. We take a large R ≥ 1 so that VR = {x ∈ V ; r(x) < R} is not empty, put

θ = ϕΦ · (−1)i−1 dx1 ∧ . . . ∧ dxi−1 ∧ dxi+1 ∧ . . . ∧ dxn ,
and apply Gauss’ theorem recalled in Chapter 1.1 to the integral of dθ over VR . If
we denote by WR the intersection of V and the sphere SR in X defined by r(x) = R,
then by using ϕ∂V = 0 we get

θ= (∂(ϕΦ)/∂xi ) dx.
WR VR
If we denote by dσ the surface element of SR , then we will have

|LHS| ≤ M · r(x)m+n |Φ(x)|r(x)−n dσ ≤ M · r m+n Φ∞ · Ωn R−1 ,
SR
in which Ωn denotes the area of S1 . If we take the limit as R → ∞, then we get

the formula in the lemma.
If we use Lemma 5.3.2 repeatedly, then we shall be talking about the shifting
of a differential operator P applied to one factor to another as its adjoint operator
P ∗ . We have x∗i = xi , (∂/∂xi )∗ = −∂/∂xi for 1 ≤ i ≤ n. We observe that if
Dn = K[x, ∂/∂x] is the Weyl algebra, where K is any field with char(K) = 0, then
the above prescription uniquely extends to a K-involution P → P ∗ of Dn , i.e., a
K-linear map from Dn to itself satisfying (P1 P2 )∗ = P2∗ P1∗ and (P ∗ )∗ = P . In fact,
in the notation of Lemma 4.1.1, the tensor algebra T (E) has a K-involution defined
by ξi∗ = ξi , ηi∗ = −ηi for 1 ≤ i ≤ n under which the ideal I(E) is stable. For
instance, we have
(ξi ⊗ ηj − ηj ⊗ ξi + δij )∗ = −ηj ⊗ ξi + ξi ⊗ ηj + δij
for all i, j. This implies the above assertion.

In the following theorem Γ(s) denotes the gamma function. We shall give a
self-contained exposition of the theory of Γ(s) in Chapter 6.2, which can be read
separately. For the time being, we shall only use the well-known facts that Γ(s) is
holomorphic on C0 , 1/Γ(s) is holomorphic on C, Γ(s + 1) = sΓ(s) for every s in
C\(−N), and Γ(1) = 1.
Theorem 5.3.1 We take f (x) arbitrarily from R[x1 , . . . , xn ]\R, define an open
subset V of X = Rn as V = {x ∈ X; f (x) > 0}, and for any s in the right-half
plane C0 and any Φ in S(X) we put

s
f+ (Φ) = f (x)s Φ(x) dx
V
s
with the understanding that f+ (Φ) = 0 if V = ∅. Then f+
s
becomes an S(X) -valued
holomorphic function on C0 . Furthermore, if we put

bf (s) = (s + λ), γf (s) = Γ(s + λ),
λ λ
s
in which bf (s) is the Bernstein polynomial of f (x), then f+ /γf (s) has a holomorphic
continuation to the whole C.
Proof. We shall prove the first part. If we put d = deg(f ) and
M1 = max(1, sup |f (x)|), M2 = max(1, sup |r(x)−d f (x)|),

r(x)≤1 r(x)≥1
then 1 ≤ M1 , M2 < ∞. We take any compact subset C of C0 and denote by σ0

the maximum of Re(s) for all s in C. Finally, we denote by χ1 (resp. χ2 ) the
characteristic function of {x ∈ X; r(x) ≤ 1} (resp. {x ∈ X; r(x) ≥ 1}) and put
φC (x) = M1σ0 Φ∞ · χ1 (x) + M2σ0 r dσ0 +n+1 Φ∞ · r(x)−n−1 χ2 (x)

70 JUN-ICHI IGUSA
for every x in X. Then for every (x, s) in V × C we have |f (x)s Φ(x)| ≤ φC (x) and

|f+
s
(Φ)| ≤ φC (x) dx ≤ φC (x) dx
V X
= M1 · Φ∞ + M2 · r dσ0 +n+1 Φ∞ ,
in which
M1 = M1σ0 Ωn /n, M2 = M2σ0 Ωn
s s
are both finite and independent of Φ. In particular, f+ (Φ) is defined and f+ converts
every null sequence in S(X) into a null sequence in C. Furthermore, by applying
Lemma 5.3.1 to this case with V , C0 , f (x)s Φ(x) respectively as X, U , f (x, s), we
see that f+ s
(Φ) is a holomorphic function on C0 . Therefore, f+ s
is an S(X) -valued
holomorphic function on C0 .
As for the second part, for the sake of simplicity, we shall write b(s), γ(s) instead
of bf (s), γf (s). By Bernstein’s theorem, i.e., Theorem 4.1.1, there exists an element
P (s) of R[s, x, ∂/∂x] satisfying
P (s)f (x)s+1 = b(s)f (x)s
for every x in V . More precisely f (x)s is well defined for every x in V and P (s) is
applied to f (x)s+1 as a differential operator. If we take s from Cσ for a large σ > 0
depending on the order of P (s), then Lemma 5.3.2 becomes repeatedly applicable
and we get

s
b(s)f+ (Φ) = s+1
(P (s)f (x) )Φ(x) dx = f (x)s+1 (P (s)∗ Φ(x)) dx.
V V
We observe that Φ1 = P (s)∗ Φ is in S(X). Therefore we can keep on applying the

above process and, after k-times application, we get

b(s) . . . b(s + k − 1)f+
s
(Φ) = f (x)s+k Φk (x) dx,
V
in which Φk = P (s + k − 1)∗ . . . P (s)∗ Φ is in S(X). In fact, Φk is a polynomial in

s with coefficients in S(X). Since
γ(s)b(s) . . . b(s + k − 1) = γ(s + k),
therefore, we get

γ(s)−1 f+
s
(Φ) = γ(s + k)−1 · f (x)s+k Φk (x) dx
V
under the assumption that s is in Cσ for a large σ > 0. We observe that the LHS
is holomorphic on C0 while the RHS is holomorphic on C−k . Since k is arbitrary in
N, the above relation implies that f+
s
(Φ)/γ(s) has a holomorphic continuation to C.
s
Therefore we see by Proposition 5.2.1 that f+ /γ(s) has a holomorphic continuation
to C.
We recall that an arbitrary element ω of Ω(R× ) has the form
ω(a) = |a|sR (a/|a|)p , |a|R = |a|,
in which s is in C and p = 0, 1. We observe that if ω is in Ω0 (R× ), i.e., if

σ(ω) =Re(s) > 0, then ω has a continuous extension to R as ω(0) = 0. We in-
troduce the local zeta function ZΦ (ω) of f (x) in R[x1 , . . . , xn ]\R for ω in Ω0 (R× )
and Φ in S(X) as
ZΦ (ω) = ω(f (x))Φ(x) dx.
X
The integral does not change even if we replace X by X\f −1 (0), which is the disjoint
union of {x ∈ X; f (x) > 0} and {x ∈ X; −f (x) > 0}. Therefore, if we introduce
s
f− (Φ) as (−f )s+ (Φ), we can write
s
ZΦ (ω) = ω(f )(Φ) = f+ (Φ) + (−1)p f−
s
(Φ).
Theorem 5.3.1 then shows that ZΦ (ω) has a meromorphic continuation to Ω(R× ).
Furthermore, since b−f (s) = bf (s), the poles of ZΦ (ω) on the s-plane are in the
union of −λ − N as −λ runs over the set of zeros of bf (s). The theorem also shows
that the order of a pole of ZΦ (ω) is at most equal to the order of the corresponding
zero of bf (s).
If now f (x) is in C[x1 , . . . , xn ]\C, then V = Cn \f −1 (0) is an analogue of {x ∈
R ; f (x) > 0}. However f (x)s is not well defined on V . With this situation in mind
n
we proceed as follows: Suppose that s1 , s2 are elements of C with s1 − s2 in Z and

a is arbitrary in C× ; denote by ā the complex conjugate of a and put
s = (s1 + s2 )/2, p = s 1 − s2 , i.e., s1 = s + p/2, s2 = s − p/2.
Then
as1 ās2 = |a|sC (a/|a|)p , |a|C = |a|2 = aā
is well defined. We shall apply the above observation to a = f (x) for x in V .
We also make the following observation. If we put uα =Re(xα ), vα =Im(xα ) so
that xα = uα + ivα , x̄α = uα − ivα for 1 ≤ α ≤ n, then Cn becomes R2n under
the correspondence (x1 , . . . , xn ) → (u1 , v1 , . . . , un , vn ). Furthermore, if ϕ is any
differentiable function on a nonempty open subset of Cn , then
∂ϕ/∂xα = (1/2)(∂ϕ/∂uα − i∂ϕ/∂vα ), ∂ϕ/∂ x̄α = (1/2)(∂ϕ/∂uα + i∂ϕ/∂vα )
and Cauchy-Riemann’s equations become ∂ϕ/∂ x̄α = 0 for 1 ≤ α ≤ n. Finally, we

have
dx1 ∧ dx̄1 ∧ . . . ∧ dxn ∧ dx̄n = (−2i)n du1 ∧ dv1 ∧ . . . ∧ dun ∧ dvn
and, accordingly, we shall use 2n -times the usual measure on R2n as the Haar
measure dx on Cn .
In the above notation, if for any element ϕ(x) of C(s1 )[x, 1/f (x)] we regard s1
as a complex variable, then
(∂/∂xi )(ϕ(x)f (x)s1 f¯(x̄)s2 ) = (∂/∂xi · ϕ(x))f (x)s1 f¯(x̄)s2

72 JUN-ICHI IGUSA
for every variable x in V and for 1 ≤ i ≤ n. Therefore Bernstein’s theorem becomes

applicable. Namely, there exists an element P (s) of C[s, x, ∂/∂x] satisfying
P (s1 )f (x)s1 +1 f¯(x̄)s2 = bf (s1 )f (x)s1 f¯(x̄)s2
for every x in V . In the following theorem γf (s) is as in Theorem 5.3.1

Theorem 5.3.2 If we take f (x) arbitrarily from C[x1 , . . . , xn ]\C, s from C0 , p
from Z, Φ from S(X), where X = Cn , and put V = X\f −1 (0), s1 = s + p/2, s2 =
s − p/2, then
f s1 f¯s2 (Φ) = f (x)s1 f¯(x̄)s2 Φ(x) dx
V

defines an S(X) -valued holomorphic function f s1 f¯s2 of s in C0 , and further
f s1 f¯s2 /γf (s1 ) has a holomorphic continuation to the whole C.
Proof. We apply an entirely similar argument to f s1 f¯s2 (Φ) as in the proof of
Theorem 5.3.1, and we get the first part. As for the second part, also as in the
proof of Theorem 5.3.1, we get

γf (s1 )−1 f s1 f¯s2 (Φ) = γf (s1 + k)−1 · f (x)s1 +k f¯(x̄)s2 Φk (x) dx
V
for Re(s1 ) sufficiently large, in which Φk = P (s1 + k − 1)∗ . . . P (s1 )∗ Φ is in S(X)

for all k in N. Since
s+k/2
f (x)s1 +k f¯(x̄)s2 = |f (x)|C (f (x)/|f (x)|)p+k
for every x in V , the RHS of the above relation is holomorphic on C−k/2 , and this
implies the second part.
Remark. We observe that if we put Q(s) = P̄ (s, x̄, ∂/∂ x̄) for P (s) = P (s, x, ∂/∂x),
then we have
Q(s2 )f (x)s1 f¯(x̄)s2 +1 = b̄f (s2 )f (x)s1 f¯(x̄)s2 ,
in which P̄ , b̄f are as f¯ the images of P , bf under the complex-conjugation applied
to their coefficients. In particular, if bf (s) is written as in Theorem 5.3.1, then

b̄f (s) = (s + λ̄),
λ
and this is the Bernstein polynomial of f¯(x). We define γ̄f similarly. More precisely,
we define γ̄f (s) as the product of Γ(s + λ̄). Then in the notation of the above proof
we have

γ̄f (s2 )−1 f s1 f¯s2 (Φ) = γ̄f (s2 + k)−1 · f (x)s1 f¯(x̄)s2 +k Ψk (x) dx,
V
in which Ψk = Q(s2 + k − 1)∗ . . . Q(s2 )∗ Φ is in S(X) for all k in N.

We recall that an arbitrary ω in Ω(C× ) has the form
ω(a) = |a|sC (a/|a|)p ,

in which s is in C and p is in Z. We introduce the local zeta function ZΦ (ω) of f (x)

for ω in Ω0 (C× ) and Φ in S(X) as

ZΦ (ω) = ω(f (x))Φ(x) dx.
X
If s1 , s2 are as above, i.e., as in Theorem 5.3.2, then we can write

ZΦ (ω) = ω(f )(Φ) = f s1 f¯s2 (Φ).
Therefore, ZΦ (ω) is holomorphic on Ω0 (ω), and it has a meromorphic continuation
to the whole Ω(C× ). Furthermore, the poles of ZΦ (ω) on the p-th s-plane are in the
union of −λ − p/2 − N as −λ runs over the set of zeros of bf (s) and the order of a
pole of ZΦ (ω) is at most equal to the order of the corresponding zero of bf (s). The
above remark shows that the poles of ZΦ (ω) on the p-th plane are also in the union
of −λ̄ + p/2 − N as −λ runs over the set of zeros of bf (s) with the same information
about their orders.
Finally, in the special case where Φ(x) = exp(−π t xx) for K = R and Φ(x) =
exp(−2π t xx̄) for K = C, we shall write Z(s) instead of ZΦ (ωs ), i.e., we put

Z(s) = |f (x)|sK Φ(x) dx.
Kn
We shall later explain the computation of Z(s) in some cases.
5.4 Complex power ω(f ) via desingularization

If K is R or C, f (x) is in K[x1 , . . . , xn ]\K for some n, and X = K n , then the
S(X) -valued holomorphic function ω(f ) on Ω0 (K × ), especially
ωs (f ) = |f |sK ,
is called the complex power of f (x). We have seen that it has a meromorphic
continuation to the whole Ω(K × ) with some information about its poles via the
Bernstein polynomial bf (s). We might repeat a short history about complex powers.
(This time with references.) In the Amsterdam Congress of 1954 I.M. Gel’fand
proposed the problem of their meromorphic continuation. His joint book with G.E.
Shilov [16] contains instructive discussions of complex powers in some special cases.
Also M. Sato’s theory of prehomogeneous vector spaces [48] contains complex powers
of some group invariants. In the general case the problem was settled jointly by I.N.
Bernstein and S.I. Gel’fand [2] and also by M.F. Atiyah [1] by the same method, i.e.,
by using Hironaka’s desingularization theorem. Bernstein later obtained another
solution in [3], which we have explained. We shall now explain their original solution
because it provides some additional information about the poles of complex powers.
We shall make use of the results already obtained to simplify the argument.
In general if ϕ is any, say, C-valued function on a topological space, we call the
closure of the set of all x where ϕ(x) = 0 the support of ϕ and denote it by Supp(ϕ).
We go back to section 5.2 and for any ε > 0 and x in X = Rn we put
ρε (x) = kε−n θ0 (r((2ε)−1 x)), i.e., ρε (x) = kε−n exp(−1/(1 − r(ε−1 x)2 ))
74 JUN-ICHI IGUSA
for r(x) < ε and ρε (x) = 0 for r(x) ≥ ε. We determine k > 0 as

k· exp(−1/(1 − r(x)2 )) dx = 1
r(x)<1
so that the integral of ρε over X becomes 1. We are using the same notation as
in L. Schwartz [51], I, p. 22. If ϕ is any C-valued continuous function on X with
compact support, then the convolution of ϕ and ρε is defined as

(ϕ ∗ ρε )(x) = ϕ(y)ρε (x − y) dy.
X
∞
We see immediately that ϕ ∗ ρε is a C -function on X which vanishes outside the
ε-neighborhood of Supp(ϕ), i.e., the union of open balls of radius ε centered at all
points of Supp(ϕ). Furthermore we have
ϕ ∗ ρε − ϕ∞ ≤ sup |ϕ(x) − ϕ(y)|.

r(x−y)≤ε
Since ϕ is uniformly continuous, the RHS tends to 0 as ε → 0. We keep in mind

that if ϕ ≥ 0, i.e., if ϕ(x) ≥ 0 for every x in X, then ϕ ∗ ρε ≥ 0. We shall use the
partition of unity only in the following form:
Lemma 5.4.1 Let X denote an n-dimensional R-analytic manifold and C a com-

pact subset of X which is covered by a family of open subsets Uα of X. Then there
exists an open subset Ω of X containing C and a finite set {pi ; i ∈ I} of C ∞ -
functions pi ≥ 0 on Ω with the closure of {x ∈ Ω; pi (x) > 0} taken in X contained
in some Uα such that
pi (x) = 1
i∈I
for every x in C.
Proof. In Chapter 1.1 we have recalled the following fact with proof: Let X denote
a locally compact space and U a neighborhood of any point a of X. Then there
exists a continuous function ϕa on X with compact support contained in U such
that 0 ≤ ϕa ≤ 1 and ϕa (a) = 1. If now {(U, φU )} is an atlas on X, by replacing
each U by many subsets with compact closure contained in U we may assume that
every Ū is compact and local coordinates are valid on Ū . Then by replacing Uα by
Uα ∩ U we may assume for every α that Ūα is compact and local coordinates are
valid on Ūα . We then construct ϕa for every a in C so that Supp(ϕa ) is contained
in some Uα . Since C is compact, we can find a finite set {ai ; i ∈ I} such that if
we put ϕi = ϕa for a = ai , then for every x in C we will have ϕi (x) > 0 for some i
in I. If we denote by ϕ the sum of all ϕi , then ϕ has a positive minimum η on C.
Furthermore if for every i we choose Uα containing Supp(ϕi ), denote it by Ui , and
replace Ūi by its bicontinuous image in Rn , then
δ = min{dis(Supp(ϕi ), ∂Ui )} > 0.

i∈I
We now choose ε > 0 so small that ε < δ and ϕ ∗ ρε − ϕ∞ < η, in which ϕ ∗ ρε is

defined as the sum of all ϕi ∗ ρε . Then Supp(ϕi ∗ ρε ) is contained in Ui for every i
and (ϕ ∗ ρε )(x) > 0 for every x in C. Therefore, we have only to put
Ω = {x ∈ X; (ϕ ∗ ρε )(x) > 0}, pi (x) = (ϕi ∗ ρε )(x)/(ϕ ∗ ρε )(x)
for every x in Ω.
We also need the fact that the n-fold tensor product over C of D(R) or, equiva-
lently by Lemma 5.2.2, of S(R) is dense in S(Rn ). This follows from the following
lemma which contains additional information.
Lemma 5.4.2 The set of functions of the form exp(−λ · r(x − a)2 ) for all λ > 0
and a in X = Rn spans a dense subspace G(X) of S(X).
Proof. We have remarked in section 5.2 that G(X) is contained in S(X). We take
Φ from S(X) and show that it can be approximated by an element of G(X). In
doing so we may assume that C = Supp(Φ) is compact. If for any ε > 0 we put
χε (x) = ε−n · exp(−π · r(ε−1 x)2 ),
then χε is in S(X), in fact in G(X), and its integral over X is 1. We shall first show
that for any i, j in Nn we have
lim Φ ∗ χε − Φi,j = 0.
ε→0
By |j|-times application of Lemma 5.3.2 we see that the convolution by χε and

(∂/∂x)j commute. By replacing (∂/∂x)j Φ by Φ, therefore, we may assume that
j = 0. If i = 0 also, without using the compactness of Supp(Φ), we can proceed as
follows: If we put
M = ( (∂Φ/∂xi )2 )1/2 ∞ ,
1≤i≤n
then by using the mean-value theorem in calculus and the Schwarz inequality we
get |Φ(x) − Φ(y)| ≤ M · r(x − y) for every x, y in X. This implies

|(Φ ∗ χε − Φ)(x)| ≤ M · r(x)χε (x) dx = M Ωn ε · r n exp(−πr 2 ) dr,
X r≥0
which tends to 0 as ε → 0. If m = |i| > 0, we may replace xi by r(x)m . We have

only to show that
lim |r(x)m (Φ ∗ χε − Φ)(x)| = 0
ε→0
uniformly in x. We choose a large r0 > 0 so that Φ(y) = 0 implies r(y) ≤ r0 .
Since the above proof takes care of the case where r(x) ≤ 2r0 , we shall assume that
r(x) ≥ 2r0 . If Φ(y) = 0, then r(x) ≤ 2 · r(x − y) and r(x − y) ≥ r0 . Since Φ(x) = 0,
therefore, if we put r1 = r0 ε−1 , then

|r(x)m (Φ ∗ χε − Φ)(x)| ≤ 2m Φ∞ · r(x)m χε (x)dx
r(x)≥r0

= (2ε) Ωn Φ∞ ·
m
r m+n−1 exp(−πr 2 ) dr,
r≥r1
76 JUN-ICHI IGUSA
which tends to 0 as ε → 0.
We shall next show that Φ ∗ χε for any fixed ε > 0 can be approximated by an
element of G(X). We subdivide X into small cubes with vertices in k−1 Zn for a
large k in N. We denote by {δi ; i ∈ Ik } the finite set of cubes which intersect C
and choose yi from C ∩ δi for every i in Ik . We observe that

Sk (x) = (kε)−n · Φ(yi ) exp(−πε−2 · r(x − yi )2 )
i∈Ik
is a Riemann sum for (Φ ∗ χε )(x) and that it is in G(X). We shall show that Sk
tends to Φ ∗ χε as k → ∞. We shall prove, more generally, that if ϕ is in S(X) and

Sk (x) = k−n · Φ(yi )ϕ(x − yi ),
i∈Ik
then we will have

lim Sk − Φ ∗ ϕi,j = 0
k→∞
for every i, j in Nn . Since (∂/∂x) Sk (x) is a Riemann sum for (Φ ∗ (∂/∂x)j ϕ)(x), we
j
may assume as before that j = 0. We may also replace xi by r(x)m where m = |i|.
We shall show that for any given η > 0 we can make |r(x)m (Sk − Φ ∗ ϕ)(x)| < η
for all x in X. If r1 ≥ r0 and r(x) ≥ 2r1 , then by using r(x) ≤ 2 · r(x − y) and
r(x − y) ≥ r1 for Φ(y) = 0 we get
|r(x)m (Sk − Φ ∗ ϕ)(x)| ≤ |r(x)m Sk (x)| + |r(x)m (Φ ∗ ϕ)(x)|
≤ 2m Φ∞ r m+1 ϕ∞ (card(Ik )k−n + µ(C))r1−1 ,
in which µ(C) denotes the total measure of C. Since card(Ik )k−n tends to µ(C) as
k → ∞, the RHS becomes less than η for a large r1 and for all large k. We shall
fix such an r1 and take x from the remaining part, i.e., from the compact ball in
X defined by r(x) ≤ 2r1 . Since Supp(Φ) is also compact, clearly Φ(y)ϕ(x − y) is
equicontinuous in y, i.e., uniformly continuous in y with the uniformity independent
of x. Therefore we will have (2r1 )m |(Sk − Φ ∗ ϕ)(x)| < η for all large k.
Remark. If Φ(x) for every x in X is a holomorphic function of s in a fixed nonempty
open subset V of C, then in Sk (x) in the above proof s appears only in Φ(yi ) for i
in Ik . Therefore Φ can be approximated by

ci (s)(ϕi1 ⊗ · · · ⊗ ϕin )
with ϕij in S(R) and ci holomorphic on V .

Theorem 5.4.1 Suppose that f (x), h : Y → X = K n , and E = {E}, where each
E is equipped with a pair of positive integers (NE , nE ), are as in Theorem 3.2.1.
Then the poles of the complex power ω(f ) on the p-th s-plane are contained in the
union of
−(1/NE )(nE + N) (K = R), −|p|/2 − (1/NE )(nE + N) (K = C)
for all E in E with their orders at most equal to the dimension of the nerve complex
N (E) of E increased by 1. Therefore the poles of ω(f ) are all negative rational
numbers and their orders are at most equal to n = dim(X).
Proof. We take ω from Ω0 (K × ). We already know that ω(f ) has a meromorphic

continuation to Ω(K × ), and we are interested in its poles on the p-th s-plane. We
know by Proposition 5.2.1 that every coefficient c of Laurent expansions of ω(f ) at
its poles is an element of S(X) and we are interested in whether or not c = 0. By
Lemma 5.2.2 this can be detected by c(Φ) for Φ in D(X). Therefore we have only
to examine the meromorphic continuation of ω(f )(Φ) for Φ in D(X), i.e., under
the assumption that Supp(Φ) is compact. Since the map h is proper, we see that
C = h−1 (Supp(Φ)) is also compact. We know by Theorem 3.2.1 that at every point
b of Y there exist local coordinates (y1 , . . . , yn ) of Y around b such that
N n −1
f ◦h=ε· yj j , h∗ dxk = η · yj j · dyk ,
j∈J 1≤k≤n j∈J 1≤k≤n
in which (Nj , nj ) = (NE , nE ) with J bijective to the set of all E containing b and
ε, η are units of the local ring Ob of Y at b. We choose a small neighborhood Ub of
b over which the above local coordinates are valid and ε±1 , η ±1 are all K-analytic.
We apply Lemma 5.4.1 to Y , C, {Ub } instead

of X, C, {Uα }. In that way we get a
partition of unity {pi ; i ∈ I} such that pi (y) = 1 for every y in C. This implies
n −1
ω(f )(Φ) = ω(yj )Nj |yj |Kj (ω(ε)|η|K (Φ ◦ h)pi ).
i∈I j∈J
We observe that Ψ(y) = ω(ε(y))|η(y)|K Φ(h(y))pi (y) can be considered as an ele-

ment of D(K n ) and for a fixed y it has a holomorphic continuation to the whole
s-plane. Furthermore, since the Bernstein polynomial of yj is s + 1, by the general
results in section 5.3 for n = 1 we see that for every j and s in C0
n −1 N s+nj −1
ω(yj )Nj |yj |Kj = |yj |Kj (yj /|yj |)Nj p
for yj in K × has a meromorphic continuation to the whole s-plane with its poles
defined by the condition that “Nj s + nj − 1 is in −1 − Nj |p|/2 − N,” i.e., s is in
−|p|/2 − (1/Nj )(nj + N), with the understanding that p = 0 for K = R. Finally,
for a similar reason as before, we can apply Lemma 5.4.2 and its remark to replace
the above Ψ by a tensor product of elements of S(K) without losing any pole. In
that way we get the description of the poles of ω(f ) as stated in the theorem.
5.5 An application
We shall give an application of Theorem 5.4.1 after recalling some basic facts on
Fourier transformations in S(X) and S(X) for X = Rn , n > 0. We reserve the
notation Φ for an arbitrary element of S(X), and we shall not repeat “for all Φ in
S(X)” all the time. We first observe that every C-valued continuous function ϕ on
X with polynomial growth gives rise to an element Tϕ of S(X) as

Tϕ (Φ) = ϕ(x)Φ(x) dx.
X
78 JUN-ICHI IGUSA
In fact, by assumption we have |ϕ(x)| ≤ M · max(1, r(x)m ) for some M, m ≥ 0.

Then by splitting X into two parts as r(x) ≤ 1 and r(x) ≥ 1 we get
|Tϕ (Φ)| ≤ M Ωn (n−1 Φ∞ + r m+n+1 Φ∞ ).
This implies that Tϕ converts every null sequence in S(X) into a null sequence in
C. The correspondence ϕ → Tϕ is clearly C-linear and Tϕ = 0 implies ϕ = 0. In
fact if ϕ = 0, then ϕ(a) = 0 for some a in X. Then Φ(x) = ρε (x − a), where ρε is
as in section 5.4, is in D(X) and Tϕ (Φ) = 0 if ε is small. We shall sometimes write
ϕ instead of Tϕ . We shall now define some operations on elements of S(X) in such
a way that they are compatible with the identification of ϕ and Tϕ .
If ϕ is any C ∞ -function on X such that all its derivatives have polynomial
growth, then Φ → ϕΦ gives a C-linear continuous map of S(X) to itself. Therefore
if T is in S(X) , then
(ϕT )(Φ) = T (ϕΦ)
defines an element of ϕT of S(X) . Similarly, if P is any element of C[x, ∂/∂x] and
P ∗ is its adjoint operator, then
(P T )(Φ) = T (P ∗ Φ)

defines an element P T of S(X)
. If now ϕ is any C-valued continuous integrable
function on X and [x, y] = xi yi for every x, y in X with xi , yi as their i-th
coordinates for 1 ≤ i ≤ n, then

(Fϕ)(x) = ϕ(y)e([x, y]) dy,
X
where e(t) = exp(2πit), defines a uniformly continuous function Fϕ or simply ϕ∗

on X satisfying
Fϕ∞ ≤ |ϕ(x)| dx = ϕ1 .
X
The correspondence ϕ → Fϕ is clearly C-linear and F is called a Fourier transfor-

mation. We observe that
Φ1 ≤ Ωn (n−1 Φ∞ + r n+1 Φ∞ ).
Furthermore by using Lemma 5.3.2 we get

∗ 1
xp Φ (x) = − · (∂Φ/∂yp ) e([x, y]) dy,
2πi X

∂Φ∗ /∂xp = 2πi · (yp Φ(y)) e([x, y]) dy
X
for 1 ≤ p ≤ n. This implies
xα (∂/∂x)β Φ∗ (x) = (−1)|α| (2πi)|β|−|α| ((∂/∂x)α xβ Φ)∗ (x),

hence
Φ∗ α,β ≤ (2π)|β|−|α| (∂/∂x)α xβ Φ1

≤ const · max(1, r n+1 )(∂/∂x)α xβ Φ∞
for every α, β in Nn . Therefore FΦ = Φ∗ is in S(X) and further F gives a continuous

map from S(X) to itself. On the other hand, the following formula:

exp(−πλx2 + 2ax) dx = λ−1/2 exp(a2 /πλ)
R
for λ > 0 and a in R is known in calculus. If we put
ϕ(x) = exp(−πλr(x)2 + 2[a, x])
for x in X and a in Cn , then by using a holomorphic continuation of the above

formula we get
ϕ∗ (x) = λ−n/2 · exp(−πλ−1 r(x)2 + 2[iλ−1 a, x] + (πλ)−1 r(a)2 ),
and this implies (ϕ∗ )∗ (x) = ϕ(−x) for every x in X. Since the C-span of the
above ϕ for all λ > 0 and a in X is dense in S(X) by Lemma 5.4.2 and since
F is continuous, we get (Φ∗ )∗ (x) = Φ(−x) for every x in X. In particular, F is
bicontinuous. Furthermore if ϕ is any continuous integrable function on X, then by
using Fubini’s theorem we get

∗
ϕ(x)Φ (x) dx = ϕ(x)Φ(y)e([x, y]) dxdy = ϕ∗ (x)Φ(x) dx.
X X×X X
Therefore if we define the Fourier transform FT = T ∗ of any T in S(X) , as
T ∗ (Φ) = T (Φ∗ ),
then T ∗ is in S(X) and Tϕ∗ = (Tϕ )∗ . We shall formulate its immediate consequence
as a proposition for our later use.
Proposition 5.5.1 If ϕ is a continuous integrable function on X = Rn with inte-
grable Fourier transform ϕ∗ , then the Fourier inversion formula
(ϕ∗ )∗ (x) = ϕ(−x)
holds for every x in X.

In fact, if we put ψ(x) = ϕ(−x), then we will have
T(ϕ∗ )∗ (Φ) = ((Tϕ )∗ )∗ (Φ) = Tϕ ((Φ∗ )∗ ) = Tψ (Φ),
hence (ϕ∗ )∗ = ψ.
We shall give two examples of T ∗ . Firstly
1∗ (Φ) = T1∗ (Φ) = T1 (Φ∗ ) = (Φ∗ )∗ (0) = Φ(0) = δ0 (Φ),

80 JUN-ICHI IGUSA
hence 1∗ = δ0 , the Dirac measure on

X supported by 0. If we express an element
f (x) of C[x1 , . . . , xn ]\C as f (x) = cα xα with cα in C for α in Nn and define an
element P of C[∂/∂x1 , . . . , ∂/∂xn ] as

P = (2πi)−|α| cα (∂/∂x)α ,
α

then for any T in S(X) we will have
(f T )∗ = P T ∗ .
In fact,

(f T )∗ (Φ) = cα T (xα Φ∗ ) = (−2πi)−|α| cα T ∗ ((∂/∂x)α Φ) = P T ∗ (Φ).
α α
We might remark that if f (x) is any element of R[x1 , . . . , xn ]\R, then the hypersur-
face f −1 (0) in X is of measure 0. This can be proved, e.g., by using the Weierstrass
preparation theorem. We are ready to explain an elegant proof by M.F. Atiyah [1]
of the following theorem:
Theorem 5.5.1 If f (x) is any element of C[x1 , . . . , xn ]\C, there exists an element
T of S(X) for X = Rn satisfying f T = 1.
Proof. We may assume that f (x) is in R[x1 , . . . , xn ]\R and f ≥ 0. In fact, if
(f f¯)S = 1 for some S in S(X) , then T = f¯S is in S(X) and f T = 1. By
s s
assumption f− = 0, hence ωs (f ) = f+ , which we shall denote by f s . If we put
V = {x ∈ X; f (x) > 0}, then by definition

s
f (Φ) = f (x)s Φ(x) dx
V
for s in C0 . We know that f has a meromorphic continuation to C. Let

s

fs = ck (s + 1)k
k∈Z
denote its Laurent expansion at −1 with ck in S(X) for all k. Since the poles
of f s are negative rational numbers by Theorem 5.4.1, we see that f s+1 = f f s is
holomorphic at s = −1. Therefore f ck = 0 for all k < 0 and

f s+1 = f c0 + (f ck )(s + 1)k .
k>0
Since X\V = f −1 (0) is of measure 0, by using Lebesgue’s theorem we get

s+1 s+1
lim f (Φ) = lim f (x) Φ(x) dx
s→−1 s→−1
V
= Φ(x) dx = Φ(x) dx = T1 (Φ),
V X
hence f c0 = T1 = 1. We have only to take c0 as T .

Corollary 5.5.1 If P is any element of C[∂/∂x1 , . . . , ∂/∂xn ]\C, then there exists
an elementary solution for the differential operator P , i.e., an element S of S(X)
for X = Rn satisfying P S = δ0 .
In fact, we can write

P = (2πi)−|α| cα (∂/∂x)α
α

with cα in C. If we put f (x) = cα xα , then by Theorem 5.5.1 we will have f T = 1

for some T in S(X) . If we put S = T ∗ , then S is in S(X) and
P S = (f T )∗ = 1∗ = δ0 .
A classical example of an elementary solution, sometimes called a “fundamental

solution,” is as follows: If ∆ denotes the Laplacian and ϕ is in D(X) for X = Rn ,
n = 2, then Poisson’s formula

∆· r(x − y)−n+2 ϕ(y) dy = −(n − 2) Ωn ϕ(x)
X
can be proved by using Gauss’ theorem. This shows that −1/(n − 2)Ωn r n−2 is
an elementary solution for ∆. The readers can learn the significance and further
examples of an elementary solution in Schwartz [51] and also in Gel’fand and Shilov
[16].
Chapter 6
Prehomogeneous vector
spaces
6.1 Sato’s b-function b(s)
We shall explain M. Sato’s theory of prehomogeneous vector spaces. More precisely,
we shall only explain regular prehomogeneous vector spaces up to their b-functions.
We start from the beginning: We say that a group G acts on a nonempty set X
if there exists a map G × X → X denoted by (g, x) → gx satisfying (gg )x =
g(g x), 1x = x for every g, g in G and x in X. In that case for every ξ in X
Gξ = {g ∈ G; gξ = ξ}, Gξ = {gξ; g ∈ G}
are called respectively the fixer of ξ in G and the G-orbit of ξ. We say that the
action of G on X is transitive and also X is a G-homogeneous space if X itself is a
G-orbit. In general, X becomes a disjoint union of G-orbits. We observe that Gξ
is a subgroup of G and the map G → X defined by g → gξ gives rise to a bijection
from the coset space G/Gξ to Gξ.
We shall now consider the case where X = Cn and G is a subgroup of GLn (C)
which is algebraic in the sense that it is the set of all common zeros of some poly-
nomials in the n2 entries of g with coefficients in C. We shall assume that the
closed topological subgroup G of GLn (C) is connected. If we regard elements of X
as column vectors, then G acts on X by matrix-multiplication. If X has a dense
G-orbit, then (G, X) is called a prehomogeneous vector space. If further there ex-
ists an irreducible polynomial f (x) in C[x1 , . . . , xn ] such that X\f −1 (0) becomes a
G-orbit, then the prehomogeneous vector space (G, X) is called regular.
Proposition 6.1.1 If (G, X) is a regular prehomogeneous vector space with G act-

ing transitively on X\f −1 (0), then f (x) is homogeneous and it is a relative G-
invariant in the sense that
f (gx) = ν(g)f (x)
for every g in G with ν in Ω(G) =Hom(G, C× ). Furthermore, if F (x) is any relative
G-invariant, then F (x) is either 0 or a power of f (x) up to a factor in C× .
Proof. Since X\f −1 (0) is a G-orbit, every g in G keeps f −1 (0) invariant, i.e.,
gf −1 (0) = f −1 (0). Therefore if, for a moment, we put fg (x) = f (gx) then
83
84 JUN-ICHI IGUSA
fg−1 (0) = f −1 (0). We observe that fg (x) is also irreducible. Therefore by Hilbert’s
Nullstellensatz, we see that fg (x) and f (x) divide each other, hence they differ by
a factor ν(g) in C× necessarily satisfying ν(gg ) = ν(g)ν(g ) for every g, g in G.
Furthermore ν(g) is a polynomial in the entries of g, hence ν is continuous. There-
fore ν is an element of Ω(G). If now F (x) is any relative G-invariant different from
0, then F −1 (0) is G-invariant. If F −1 (0) intersects the G-orbit X\f −1 (0), then
F −1 (0) will contain X\f −1 (0), which is dense in X, hence F (x) = 0. Since this is
not the case, F −1 (0) is contained in f −1 (0). Then by Hilbert’s Nullstellensatz F (x)
divides f (x)e for some positive integer e. Since f (x) is irreducible, this implies that
F (x) is a power of f (x) up to a factor in C× . Finally, write

f (x) = fi (x), fd (x) = 0
i≥d
with fi (x) homogeneous of degree i for all i ≥ d. Then f (gx) = ν(g)f (x) implies
fi (gx) = ν(g)fi (x) for every g in G and for all i. By what we have just shown, there
exists a positive integer e satisfying fd (x) = cf (x)e for some c in C× . By comparing
the degrees of both sides, we get c = e = 1 and fi (x) = 0 for all i > d.
We call f (x) in Proposition 6.1.1 a basic relative invariant of (G, X); it is unique
up to a factor in C× . We observe that if γ is an arbitrary element of GLn (C), then
(γGγ −1 , X) is also a regular prehomogeneous vector space with (γf )(x) = f (γ −1 x)
as its basic relative invariant. We say that (G, X) and (γGγ −1 , X) are equivalent.
This is clearly an equivalence relation. In the special case where γ, hence also γ −1 ,
is contained in the normalizer N (G) of G, i.e., if γGγ −1 = G, then f (γx) is another
basic relative invariant of (G, X), hence it differs from f (x) by a factor in C× .
Therefore, ν extends to N (G) as f (γx) = ν(γ)f (x) for every γ in N (G).
We shall assume from now on that G is reductive. This condition is known to
imply the existence of γ in GLn (C) such that γGγ −1 is invariant under g → t g −1 .
Therefore we shall simply assume that G satisfies this condition, i.e., t G = G. We
keep in mind that γGγ −1 also satisfies this condition if and only if t γγ is in N (G).
Corollary 6.1.1 If f (x) is a basic relative invariant of (G, X), then there exists
an element B(s) of C[s] satisfying
f (∂/∂x) · f (x) = B(s).
If B(s) = 0, then deg(B) ≤ deg (f ) and ν(t g) = ν(g) for every g in G. Furthermore,
B(s) depends, up to a factor in C× , only on the equivalence class of (G, X).
Proof. If deg(f ) = d, then by definition we can write
f (∂/∂x) · f (x) = φ0 (x) + φ1 (x)s + . . . + φd (x)sd
with φi (x) in C[x1 , . . . , xn , 1/f (x)] for 0 ≤ i ≤ d. We shall obtain some information
about the above expression assuming that it is different from 0. If we specialize s
to an element k of N, then we will have

f (∂/∂x)f (x)k+1 = φi (x)ki f (x)k .
0≤i≤d
If we denote the LHS by Fk (x), then in view of ∂/∂(gx) = t g −1 (∂/∂x) we have
Fk (gx) = χ(g)Fk (x), χ(g) = ν(t g)−1 ν(g)k+1
for every g in G. If we exclude at most d values of k, then Fk (x) = 0 and deg(Fk ) =

dk. Therefore, by Proposition 6.1.1 we see that Fk (x) = B(k)f (x)k for some B(k) in
C× . This implies χ(g) = ν(g)k , hence ν(t g) = ν(g) for every g in G. Furthermore,
φ0 (x) + φ1 (x)k + . . . + φd (x)kd = B(k) for d + 1 distinct k, in fact for infinitely many
k, in N, hence φi (x) is in C for 0 ≤ i ≤ d. Finally, if γ is any element of GLn (C)
such that g = t γγ is in N (G), then by replacing x in f (∂/∂x) · f (x) = B(s) by γ −1 x
we get
(γf )(∂/∂x) · (γf )(x) = ν(g)−1 B(s).
Therefore the B(s) for (γGγ −1 , X) is ν(t γγ)−1 B(s).
We shall obtain more precise information about B(s) after reviewing the fol-
lowing fact: If g is in Mn (C), we denote by ḡ the image of g under the complex-
conjunction applied to its entries. If G is any subset of Mn (C), we denote by Ḡ the
image of G under g → ḡ. If now G is a connected reductive algebraic subgroup of
GLn (C), then by replacing G by γGγ −1 for some γ in GLn (C) we can achieve not
only t G = G but also Ḡ = G. We refer to G.D. Mostow [43] for the details. In view
of this situation, we shall proceed simply assuming that t G = Ḡ = G.
We order the set of monomials X α for α = (α1 , . . . , αn ) in Nn lexicographically.
Since Ḡ = G and X̄ = X, we see that f¯(x) is also a basic relative invariant of
(G, X), hence f¯(x) differs from f (x) by a factor in C× . If we normalize f (x) by
the condition that the coefficient of the highest monomial in f (x) is 1, then we will
have f¯(x) = f (x), i.e., f (x) is in R[x1 , . . . , xn ]. This implies that B(s) is in R[s].
We shall prove two lemmas.

Lemma 6.1.1 If F (x) = cα xα in C[x1 , . . . , xn ] is homogeneous, then

F̄ (∂/∂x)F (x) = α! |cα |2 ,
α
in which α! = α1 ! · · · αn ! with the understanding that 0! = 1.

Proof. We have
F̄ (∂/∂x)F (x) = c̄α cβ (∂/∂x)α xβ ,
α,β
in which
(∂/∂x)α xβ = βi (βi − 1) · · · (βi − αi + 1)xβi i −αi
1≤i≤n
if βi ≥ αi for all i; otherwise it is 0. Since F (x) is homogeneous by assumption, we

have |α| = |β| = deg(F ), hence (∂/∂x)α xβ = 0 implies α = β and (∂/∂x)α xα = α!.
Lemma 6.1.2 We have
log k! = k log k(1 + o(1)),
i.e., log k!/k log k tends to 1, as k → ∞.

86 JUN-ICHI IGUSA
Proof. This is an immediate consequence of Stirling’s formula. In the above form

it can be proved as follows: Since log t is a monotone increasing function of t for
t > 0 and log t > 0 for t > 1, if k is in N and k > 1, then
k
log k! > log t dt = k log k − k + 1
1
on one hand and

k+1
log k! < log t dt = (k + 1) log(k + 1) − k − 2 log 2 + 1
2
on another. We have only to put these together.
Theorem 6.1.1 We have B(s) = b0 b(s), in which b0 > 0 and b(s) is a monic
polynomial of degree d = deg(f ) in R[s].
Proof. If xα is the highest monomial in f (x), then its coefficient is 1 by the above
normalization. Furthermore, xα(k+1) is the highest monomial in f (x)k+1 and

f (∂/∂x)k+1 f (x)k+1 = B(j)
0≤j≤k
for every k in N. Since f (x) is in R[x1 , . . . , xn ], by Lemma 6.1.1 we get

f (∂/∂x)k+1 f (x)k+1 ≥ (αi (k + 1))!.
1≤i≤n
Since |α| = d, by Lemma 6.1.2 we get

log (αi (k + 1))! = dk log k(1 + o(1)),
1≤i≤n
hence
B(j) ≥ exp{dk log k(1 + o(1))},
0≤j≤k
and hence

(∗) B(j)/kδk → ∞
0≤j≤k
as k → ∞ for any δ in R satisfying δ < d. In particular B(k) = 0 for every k

in N and if b0 sd0 is the highest term in B(s), then d0 > 0. Furthermore, since
B(k) = b0 kd0 (1 + o(1)) as k → ∞, we also have b0 > 0 and for any b1 > b0 there
exists k0 in N such that B(k) ≤ b1 kd0 for all k ≥ k0 . This implies

B(j) ≤ c · bk1 (k!)d0
0≤j≤k
for all k ≥ k0 , in which c is independent of k. Therefore, by using (∗) we get

k log k(d0 − δ + o(1)) → ∞ as k → ∞, hence d0 ≥ δ. Since δ is arbitrary in R
subject to δ < d, we get d0 ≥ d, hence d0 = d by d0 ≤ d.
The monic polynomial b(s) in Theorem 6.1.1 depends only on the equivalence
class of (G, X), and it is called Sato’s b-function of (G, X). If f (x) denotes the
basic relative invariant of (G, X), then we see that its Bernstein polynomial bf (s)
divides b(s). Actually, they are known to be equal. We shall give a proof to this
fact in section 6.3 after reviewing some major properties of the Γ-function.
6.2 The Γ-function (a digression)

A standard definition of the Γ-function Γ(s) is

Γ(s) = xs e−x d log x, Re(s) > 0.
x>0
By using, e.g., Lemma 5.3.1, we see that Γ(s) is a holomorphic function on C0 .

Furthermore, by integration by parts we see that
Γ(s + 1) = sΓ(s)
for Re(s) > 0. Since Γ(1) = 1 by definition, we conclude that Γ(s) has a mero-
morphic continuation to C with a pole of order 1 at every point of −N. These are
well-known elementary properties of Γ(s). Now by Weierstrass

s · (1 + s/n)e−s/n ,
where the product is for n = 1, 2, . . . , defines an entire function, i.e., a holomorphic

function on the whole C, with a zero of order 1 at every point of −N. Therefore,
the product of Γ(s) and the above entire function is an entire function. A basic
theorem on Γ(s) states that if C is the Euler constant defined by
C = lim (1 + 1/2 + . . . + 1/n − log n),

n→∞
then the product is exp(−Cs). Finally, if we put σ = Re(s), t = Im(s) so that

s = σ + it and restrict s as |σ| ≤ R for any R > 0, then
|Γ(s)| = (2π)1/2 |t|σ−1/2 exp(−(π/2)|t|)(1 + o(1))
with o(1) uniform in σ as |t| → ∞. Since the product-representation and the above
important asymptotic behavior of Γ(s) are seldom included in the basic course, we
shall give their proofs. We shall follow the presentation by H. Mellin [40] because
the proof by the Mellin transformation seems most appropriate.
Firstly if we express log n as the integral of 1/t from 1 to n and rewrite the n-th
term of the sequence defining C as

1/k − log n + 1/n = 1 − log n − 1/k ,
1≤k<n 1<k≤n
88 JUN-ICHI IGUSA
then we see that both {·} are positive and increase with n. Therefore, the sequence
is between 0 and 1, and monotone decreasing, hence the limit C exists. In the
following log(1 + x) is defined for |x| < 1 by its power series expansion taking
the value 0 at x = 0. We leave it as an exercise to show that if |x| ≤ 1/2, then
| log(1 + x) − x| ≤ |x|2 . Also we keep in mind that Re(log x) = log |x| is well defined
for every x in C× . We now start with

1/G(s) = eCs s · (1 + s/n)e−s/n .
n≥1
If |s| ≤ R for any R > 0, then
| log((1 + s/n)e−s/n )| ≤ (R/n)2
for all n ≥ 2R. Therefore, the above infinite product is absolutely and uniformly
convergent for |s| ≤ R, hence it defines an entire function of s with a zero of order
1 at every point of −N. Furthermore, if s is not in −N, then

G(s + 1)/G(s) = se−C · lim n/(s + n + 1) · exp 1/k − log n = s.
n→∞
1≤k≤n
This implies
G(s + 1) = sG(s)
for all s in C. Since 1/sG(s) takes the value 1 at s = 0, we get G(1) = 1. Further-
more,
π/G(s)G(1 − s) = −π/sG(s)G(−s) = πs · (1 − (s/n)2 ),
n≥1
and the RHS is known in complex analysis to represent sin(πs), hence
G(s)G(1 − s) = π/ sin(πs).
Since we have

(1 + 1/n)s e−s/n = lim (n + 1)s exp − ( 1/k)s = e−Cs ,
n→∞
n≥1 1≤k≤n
we can also write

(∗) G(s) = (1/s) (1 + 1/n)s /(1 + s/n).
n≥1
We shall prove the asymptotic formula for G(s). Since G(s) becomes its complex
conjugate under the complex conjugation of s and
G(s)G(−s) = −π/(s · sin(πs)),
by replacing s by it we get
|G(it)| = (2π)1/2 |t|−1/2 |eπt − e−πt |−1/2

= (2π)1/2 |t|−1/2 exp(−(π/2)|t|)(1 + o(t))
as |t| → ∞. In the general case we first observe that (∗) implies

G(σ + it)/G(it) = 1/(1 + σ/it) · (1 + 1/n)σ /(1 + σ/(n + it)),
n≥1
hence
|G(σ + it)/|t|σ G(it)| = |1 + 1/(n + it)|σ /|1 + σ/(n + it)|,
n≥0
and hence

log( |G(σ + it)/|t|σ G(it)| ) = Re σ log(1 + 1/(n + it)) − log(1 + σ/(n + it)) .
n≥0
If now |x|, |σx| < 1, then
σ log(1 + x) − log(1 + σx) = x2 P (x),
in which
P (x) = σ · (−1)k (1 − σ k )/(k + 1) · xk−1
k≥1
is absolutely convergent. Furthermore, if for |σ| ≤ R we replace x by 1/(n + it),

where n is in N and |t| ≥ R + 1, then |x| ≤ 1/(1 + R), |σx| ≤ 1/(1 + 1/R), hence
|P (x)| ≤ M for some M > 0 depending only on R. Therefore, we get

log( |G(σ + it)/|t|σ G(it)| ) = Re 1/(n + it)2 · P (1/(n + it)) ,
n≥0
in which
|RHS| ≤ M · 1/(n2 + t2 ) ≤ M t−2 (1 + π|t|/2),
n≥0
hence
|G(σ + it)| = |t|σ |G(it)|(1 + o(1))

= (2π)1/2 |t|σ−1/2 exp(−(π/2)|t|)(1 + o(1))
with o(1) uniform in σ as |t| → ∞.

In order to show that G(s) = Γ(s), we shall use the following consequence of
Fourier’s inversion formula:
Proposition 6.2.1 Let ϕ(x) denote a continuous function on R×
+ such that the
integral
φ(s) = xs ϕ(x) d log x
x>0
is absolutely convergent for some s = σ + it; assume that t → φ(σ + it) is an

integrable function on R. Then we can recover ϕ(x) from φ(s) as

1 −σ−it 1
ϕ(x) = φ(σ + it)x dt = φ(s)x−s ds.
2π R 2πi σ+iR
90 JUN-ICHI IGUSA
Furthermore, if t → φ(s) = φ(σ + it) is a continuous integrable function on R so

that ϕ(x) is defined by the second integral, then we can recover φ(s) by the first
integral provided that it is absolutely convergent.
Proof. If we put y = log x/2π, hence x = exp(2πy), and
ψ(y) = 2πxσ ϕ(x),
then ψ(y) becomes a continuous integrable function on R with its Fourier transform
ψ ∗ (t) = φ(σ + it) also integrable on R. Therefore by Proposition 5.5.1, we get
(ψ ∗ )∗ (−y) = ψ(y). This implies the above proposition.
Now G(s) is a meromorphic function on C with a pole of order 1 at every point
of −N. Furthermore, G(s + 1) = sG(s), G(1) = 1 imply
lim (s + k)G(s)x−s = (−x)k /k!

s→−k
for every x > 0 and k in N. The asymptotic formula for G(σ + it) as |t| → ∞
guarantees that G(σ + it) is an integrable function of t if σ is not in −N. We take
0 < σ < 1 and determine

1
ϕ(x) = G(s)x−s ds.
2πi σ+iR
We take n from N and R > 0, and consider the following path of integration:
σ − i ∞ → σ − iR → σ − n − iR → σ − n + iR → σ + iR → σ + i ∞.
The asymptotic formula again shows that the integrals along the two horizontal
paths tend to 0 as R → ∞. Therefore by Cauchy’s theorem we get

1
ϕ(x) = (−x)k /k! + G(s)x−s ds,
2πi σ−n+iR
0≤k<n
in which

G(s)x−s ds ≤ x−σ · |G(σ + it)| dt · xn (k − σ).

σ−n+iR R 1≤k≤n
Therefore, by taking the limit as n → ∞ we get

ϕ(x) = (−x)n /n! = e−x .
n≥0
We have chosen 0 < σ < 1. However, since G(s) is holomorphic on C0 , by the above
process of shifting the vertical line of integration we can replace σ by any positive
real number. Therefore, by Proposition 6.2.1 we get

G(s) = xs e−x d log x = Γ(s)
x>0
for every σ > 0 because the RHS is absolutely convergent.

6.3 b(s) = bf (s) and the rationality of the zeros

We go back to Sato’s b-function b(s) of (G, X). We have assumed or rather normal-
ized under the equivalence that t G = Ḡ = G and that the basic relative invariant
f (x) of (G, X) is in R[x1 , . . . , xn ]. We know by Theorem 6.1.1 that
f (∂/∂x) · f (x) = b0 b(s),
in which b0 > 0 and b(s) is a monic polynomial of degree d = deg(f ). By replacing
f (x) by (b0 )−1/2 f (x) we can further normalize b0 = 1 so that f (x) becomes unique
up to sign.
Theorem 6.3.1 If we normalize the basic relative invariant f (x) of (G, X) to sat-
isfy
f (∂/∂x) · f (x) = b(s) = (s + λ),
λ
then Z(s) can be computed explicitly, i.e.,

|f (x)|sC exp(−2π t xx̄) dx = (2π)−ds · (Γ(s + λ)/Γ(λ))
X λ
for Re(s) > 0, in which X = C and Γ(λ) = ∞ for all λ.

n
Proof. We recall that Z(s) is the LHS of the formula to be proved. If we write
|f (x)|sC = f (x)s f (x̄)s = b(s)−1 f (∂/∂x)f (x)s+1 f (x̄)s
and apply the same argument as in the proof of Theorem 5.3.2, then we get
b(s) Z(s) = (2π)d Z(s + 1)
for Re(s) > 0. Therefore, if we put

C(s) = (2π)ds Z(s) / Γ(s + λ),
λ
then C(s + 1) = C(s) for Re(s) > 0. Since Z(s) is a holomorphic function on C0 ,
we see that C(s) is a periodic holomorphic function on C with period 1. Therefore
C(s) becomes a meromorphic function of z = e(s) on C× and its Laurent expansion
at 0 can be written as

C(s) = ck z k = ck e(ks),
k∈Z k∈Z
in which
ck = exp(2πkt) · C(σ + it) e(−kσ) dσ.
R/Z
This implies

|ck | ≤ exp(2πkt) · (2π)dσ |Z(σ + it)| |Γ(σ + it + λ)| · dσ.
R/Z λ
92 JUN-ICHI IGUSA
We shall estimate |Z(s)| for s = σ + it in C0 . If in the integral defining Z(s) we

write
x = ru, r = r(x) = (t xx̄)1/2 , dx = 2n r 2n−1 drdu
and put
φ(s) = 2n−1 · |f (u)|sC du,
r(u)=1
then we have
Z(s) = (2π)−ds−n φ(s) Γ(ds + n).
Furthermore,
|φ(s)| ≤ 2 n−1
· |f (u)|σC du.
r(u)=1
If we denote by M the maximum of the RHS say for 1 ≤ σ ≤ 2, then we have
|Z(s)| ≤ (2π)−dσ−n M |Γ(ds + n)|
for 1 ≤ σ ≤ 2.
If now we incorporate the asymptotic formula
|Γ(σ + it)| = (2π)1/2 |t|σ−1/2 exp(−(π/2)|t|) (1 + o(1))
as |t| → ∞, in which o(1) is uniform in σ restricted to any finite interval, then we

get
|ck | ≤ M1 · |t|σ1 exp(2πkt) (1 + o(1))
with M1 independent of s and

σ1 = n − (1/2)(1 − d) − Re(λ).
λ
We have tacitly used the fact that

exp − (π/2) d|t| − |t + Im(λ)| ≤ exp (π/2) |Im(λ)| .
λ λ
At any rate, if k = 0, then for t = −sgn(k)τ and τ → ∞ we get ck = 0, hence

C(s) = c0 . We have thus shown that

Z(s) = (2π)−ds · Γ(s + λ) · c0
λ
for
Re(s) > 0. If we take the limit as s → 0, then Z(s) tends to 1, hence c0 =
Γ(λ)−1 . Since c0 = 0, we have Γ(λ)−1 = 0 for all λ.
Theorem 6.3.2 If b(s) is the Sato b-function of a regular prehomogeneous vector
space and bf (s) is the Bernstein polynomial of its basic relative invariant f (x), then
b(s) = bf (s). Furthermore, all its zeros are negative rational numbers, hence b(s)
is in Q[s].
Proof. Since bf (s) divides b(s), after changing the notation in Theorem 6.3.1 we
write
bf (s) = (s + λ), b(s) = bf (s) · (s + λ );
λ λ
also we put
γf (s) = Γ(s + λ), γ (s) = Γ(s + λ ).
λ λ
According to that theorem, Z(s) and γf (s)γ (s) differ by a holomorphic function on
C with no zeros. On the other hand we know by Theorem 5.3.2 that Z(s)/γf (s) is
a holomorphic function on C, hence γ (s) is also a holomorphic function on C. This
implies γ (s) = 1, hence b(s) = bf (s). We know furthermore, by Theorem 5.4.1,
that the poles of Z(s) are negative rational numbers. Therefore all zeros of b(s) are
negative rational numbers.
Actually the zeros of bf (s) are known to be negative rational numbers for an
arbitrary f (x) by M. Kashiwara [33]. We might mention that Theorem 6.3.1, in a
weaker form, is in [27]. We shall explain its counterpart in the real case:
Proposition 6.3.1 If f (x) is normalized as in Theorem 6.3.1 and further if every

monomial in f (x) is of the form xi1 . . . xid where i1 < . . . < id , then Z(s) can be
computed explicitly, i.e.,

|f (x)|sR exp(−π t xx) dx = π −ds/2 · Γ((s + λ)/2) Γ(λ/2)
X λ
for Re(s) > 0, in which X = Rn .
Proof. This can be proved in the same way as Theorem 6.3.1. We have
b(s)Z(s) = (2π)d Z(s + 2)
for Re(s) > 0, hence

Z(s) = π −ds/2 · Γ((s + λ)/2) · C(s),
λ
in which C(s) is a periodic holomorphic function on C with period 2. Therefore,

C(s) = ck e(ks/2), ck = (1/2) exp(πkt) · C(σ + it) e(−kσ/2) dσ.
k∈Z R/2Z
The above expression for ck implies, similarly as before, that
|ck | ≤ M1 |t|σ1 exp(πkt) (1 + o(1))
for some M1 > 0 and σ1 , which are independent of t, as |t| → ∞. Therefore, we get
k = 0, hence C(s) = c0 . Since Z(s) tends to 1 as s → 0, we necessarily
ck = 0 for
have c0 = Γ(λ/2)−1 .
94 JUN-ICHI IGUSA
We shall discuss a classical example from various viewpoints. We start with the
following general remark. If we replace the normalized f (x) by a0 f (x) for any a0
in R× , then we will have
f (∂/∂x) · f (x) = a20 b(s)

while Z(s) in Theorem 6.3.1 (resp. Proposition 6.3.1) will be multiplied by |a0 |sC
(resp. |a0 |sR ). Now we observe that (GLn (C), Mn (C)) is a regular prehomogeneous
vector space with det(x) as its basic relative invariant. Actually there is a notational
discrepancy. In fact, if we arrange the n columns of an element of Mn (C) in the
natural order to make up an element of “X” = CN for N = n2 , then we have to take
the isomorphic image of GLn (C) in GLN (C) as “G.” At any rate the normalization
condition t G = Ḡ = G is satisfied, the coefficients of f (x) = det(x) are in R, in fact
in Z, and further it satisfies the condition in Proposition 6.3.1. Therefore we will
have det(∂/∂x) · det(x) = b0 b(s) for some b0 > 0. We shall compute

Z(s) = | det(x)|sR exp(−π tr(t xx)) dx
X
directly, in which X = Mn (R) and tr(y) for any square matrix y denotes the trace
of y. We denote the above Z(s) by Zn (s) and show by an induction on n that

Zn (s) = π −ns/2 · Γ((s + k)/2) Γ(k/2) .
1≤k≤n
In doing so we shall use the well-known formula

Ωn /2 = π n/2 Γ(n/2)

in calculus, which becomes 1 = π 1/2 Γ(1/2) for n = 1. We have

Z1 (s) = π −(s+1)/2 Γ((s + 1)/2) = π −s/2 Γ((s + 1)/2) Γ(1/2).
If n > 1, we write x = (x1 x ) with x in Mn,n−1 (R) and put x1 = ru, where
r = r(x1 ), so that u is on the unit sphere and dx1 = r n−1 drdu. Then we get

Zn (s) = r s+n−1 exp(−πr 2 ) drdu | det(u x )|sR exp(−π tr(t x x )) dx .
Since the group of rotations in Rn acts transitively on the unit sphere, we can write
u = ge1 , where e1 = t (10 . . . 0), for some g in GLn (R) satisfying t gg = 1. In the
integral {·} above if we replace x by gx , then it becomes Zn−1 (s), hence
Zn (s) = Zn−1 (s) · (Ωn /2) π −(s+n)/2 Γ((s + n)/2).

If we apply the induction assumption to Zn−1 (s), we get the above expression for
Zn (s).
If we compare the above result with Proposition 6.3.1, then we get

(∗) det(∂/∂x) · det(x) = b(s) = (s + k)
1≤k≤n
in view of the following obvious fact. Namely if I, I are finite subsets of C and
φ(s) is a holomorphic function on C with no zeros satisfying

Γ(s + λ) = φ(s) · Γ(s + λ ),
λ∈I λ ∈I
then I = I . This is a roundabout way of computing b(s). At any rate by Theorem

6.3.1 we also have

| det(x)|sC exp(−2π t xx̄) dx = (2π)−ns · (Γ(s + k)/Γ(k)),
X 1≤k≤n
where X = Mn (C).
We might mention that (∗) follows from Capelli’s identity in invariant theory,
cf., e.g., H. Weyl [60]. It can also be proved, after R. Sasaki, as follows: If for
1 ≤ i1 < . . . < ik ≤ n, 1 ≤ j1 < . . . < jk ≤ n we put
Pi1 ...ik ,j1 ...jk = det(∂/∂xiα jβ )1≤α,β≤k
and denote by Xi1 ...ik ,j1 ...jk the coefficient of xi1 j1 . . . xik jk in det(x), then we will
have
Pi1 ...ik ,j1 ...jk · det(x) = (s + i) Xi1 ...ik ,j1 ...jk .
1≤i≤k
In fact, it is obvious for k = 1 and the general case is by an induction on k ending

up with (∗) for k = n. We might also mention that, as we shall see much later, an
entirely similar method of computation of Z(s) as above works in the p-adic case
for a polynomial such as det(x).
In the case where a connected algebraic subgroup G of GLn (C) is irreducible,
i.e., X = Cn and 0 are the only G-invariant subspaces of X, then G is reductive
by a theorem of E. Cartan. Furthermore, if G is transitive on X\f −1 (0) for some
polynomial f (x), then f (x) is necessarily a power of an irreducible polynomial, hence
we may assume that it is irreducible. At any rate, in such a case (G, X) is called
an irreducible regular prehomogeneous vector space. There is a classification theory
of prehomogeneous vector spaces by M. Sato and T. Kimura [49] which includes all
such (G, X). Furthermore, the problem of making b(s) explicit was investigated by
M. Kashiwara, T. Kimura, M. Muro, T. Oshima, I. Ozeki, M. Sato, and T. Yano.
In particular b(s) is now known for all irreducible regular prehomogeneous vector
spaces. We just mention the fundamental paper [50] by M. Sato and others, and
the valuable paper [34] by T. Kimura.
Chapter 7
Totally disconnected spaces

and p-adic manifolds
7.1 Distributions in totally disconnected spaces
We shall start with a review of some properties of totally disconnected spaces and
groups. We say that a Hausdorff space is totally disconnected if points are the only
connected subsets of X. If X is such a space, then every nonempty subset of X is
also totally disconnected. The following lemma, in a more general form, is in L. S.
Pontrjagin [45], p. 104:
Lemma 7.1.1 Let X denote a locally compact space. Then X is totally discon-
nected if and only if for every point a of X the set Ta of all compact open subsets
of X which contain a forms a base at a.
Proof. Since the if-part is clear, we shall prove the only-if part. We shall first show
that if X is a compact totally disconnected space, hence X is a member of Ta , and
if C denotes the intersection of all members of Ta , then C = {a}. Otherwise C is
not connected, hence it becomes a disjoint union of nonempty closed subsets F1 , F2
of C. Since C is compact, they are also compact, hence closed in X. We may
assume that F1 contains a. Since X is normal, F1 , F2 are respectively contained in
some disjoint open subsets G1 , G2 and the complement F of their union is compact.
We shall include the possibility that F is empty. Since C and F are disjoint and
since C is the intersection of all members of Ta , which are compact, there exists a
finite intersection A of members of Ta such that A and F are disjoint. Then A is a
member of Ta , and it becomes the disjoint union of Ai = A ∩ Gi for i = 1, 2. We
observe that A1 , A2 are compact open subsets of X and A1 contains a, hence it
is a member of Ta . On the other hand A1 and F2 are disjoint, hence A1 does not
contain C. This is a contradiction.
We are ready to prove the lemma. We take any neighborhood U of a and show
that U contains a member of Ta . Since X is locally compact, we may assume that
Ū is compact. Since Ū is totally disconnected and a is not in the compact subset
∂U = Ū \U , by the above result applied to Ū there exists a compact open subset A
of Ū which contains a such that A and ∂U are disjoint. Then A is contained in U
and it is a member of Ta .
We shall prove one more lemma here; it is in A. Weil [56], p. 19 and also in
Pontrjagin [45], pp. 149-150.
97
98 JUN-ICHI IGUSA
Lemma 7.1.2 Let G denote a locally compact group. Then G is totally discon-
nected if and only if the set of all compact open subgroups of G forms a base at its
unit element 1.
Proof. Since the if-part is clear, we shall again prove the only-if part. We take
any neighborhood U of 1. Then by Lemma 7.1.1 there exists a member A of T1
contained in U . We introduce a subset B of G as
B = {g ∈ G; gA ⊂ A}.
Then B contains 1 and, since A contains 1, B is contained in A. We shall show that

B is open. If g is arbitrary in B, then for every a in A, since A is a neighborhood
of ga, there exist neighborhoods Va , Wa of g, a respectively such that Va Wa is
contained in A. Since A is compact, we can cover A by a finite number of Wa and
if V is the intersection of the corresponding Va , then V is a neighborhood of g and
V A is contained in A. This shows that V is contained in B, hence B is open. We
shall show that B is also closed. If g is arbitrary in G\B, then ga is not in A for
some a in A. Since G\A is a neighborhood of ga, there exists a neighborhood V of
g such that V a is contained in G\A. This shows that V is contained in G\B, hence
G\B is open, and hence B is closed.
We have thus shown that B is a closed and open subset of the compact open
subset A, hence B is compact open, and hence B −1 is also compact open. Therefore,
N = B ∩ B −1 is compact open, and N consists of all g in G satisfying gA = A.
Therefore, N is also a subgroup of G contained in A, hence in U .
We shall now fix a locally compact totally disconnected space X and denote by
T (X) the set of all compact open subsets of X in which we include the empty set ∅.
We observe that T (X) is closed under the taking of finite union and intersection,
and difference. A C-valued function ϕ on X is called locally constant if every
point of X has a neighborhood U such that the restriction ϕ|U becomes a constant
function on U . A locally constant function is a counterpart of a C ∞ -function and
it is clearly continuous. The set D(X) of all locally constant functions on X with
compact support forms a vector space over C. We observe that if A is a member of
T (X), then its characteristic function χA is locally constant with A as its support.
Therefore, the C-span of the set of χA for all A in T (X) forms a subspace of D(X).
Actually, they coincide. In fact, if ϕ is in D(X), then the image I of X under ϕ is
a finite subset of C and
ϕ= αχϕ−1 (α) .
α∈I
−1
We observe that ϕ (α) for α = 0 in I is a member of T (X) and Supp(ϕ) is the
union of all such ϕ−1 (α). In particular, Supp(ϕ) = {x ∈ X; ϕ(x) = 0} for every ϕ
in D(X).
We take a finite subset F of T (X) and denote by DF the C-span of the set
of characteristic functions of all members of F . Then D(X) becomes the union or
rather the direct limit of DF for all F , hence D(X) has the direct limit topology
defined as follows: Every finite dimensional vector space over C, in particular DF
above, has the usual product topology of C. We define a subset of D(X) to be open
if and only if its intersection with DF is open in DF for all F . Let D(X) denote the
topological dual of D(X) with the so-defined topology. Then, since every C-linear
function on DF is continuous, we see that D(X) coincides with the dual space of
D(X). In particular, D(X) is complete. We call elements of D(X) distributions in
X.
We shall give another description of distributions in X. If T is in D(X) , for
every A in T (X) we put T (A) = T (χA ). Then we get a C-valued simply additive
function T on T (X) in the sense that if A is a necessarily finite disjoint union of
A1 , A2 , . . . in T (X), then T (A) = T (A1 ) + T (A2 ) + . . . . Conversely, suppose that
T is such a function. Express an arbitrary ϕ in D(X) as α1 χA1 + α2 χA2 + . . . with
α1 , α2 , . . . in C, in which A1 , A2 , . . . are disjoint members of T (X). Then we define
T (ϕ) as α1 T (A1 )+α2 T (A2 )+. . . . We shall show that T (ϕ) is well defined. Suppose
that β1 χB1 + β2 χB2 + . . . is a similar expression of ϕ, and put Cij = Ai ∩ Bj for
all i, j. Then we get
ϕ= αi χCij = βj χCij
i,j i,j
with αi = βj for Cij = ∅. Therefore, by the simple additivity of T we get

αi T (Ai ) = αi T (Cij ) = βj T (Cij ) = βj T (Bj ).
i i,j i,j j
Furthermore, if ϕ, ψ are arbitrary in D(X), by a similar argument as above we can

express ϕ, ψ as C-linear combinations of characteristic functions of the same set of
disjoint members of T (X). Then for any α, β in C we clearly have T (αϕ + βψ) =
αT (ϕ) + βT (ψ), hence T gives an element of D(X) . We have thus shown that
distributions in X and simply additive functions on T (X) can be identified.
We shall discuss extensions and restrictions of distributions. We express X as
a disjoint union of nonempty open and closed subsets Y and F , respectively. Then
both Y and F are locally compact totally disconnected spaces. Furthermore we
have an exact sequence of C-linear maps:
(∗) 0 → D(Y ) → D(X) → D(F ) → 0.
In detail, if ψ is an element of D(Y ), then the function ψ X defined as

ψ X Y = ψ, ψ X F = 0
is an element of D(X); and if ϕ is an element of D(X), then ϕ|F is an element of

D(F ). The point is that the above sequence defined by ψ → ψ X and ϕ → ϕ|F
is exact. Clearly, D(Y ) → D(X) is injective. Suppose that ϕ in D(X) has the
property that ϕ|F = 0. Then Supp(ϕ) is contained in Y , hence ϕ = (ϕ|Y )X .
This is the exactness at D(X). We shall show that D(X) → D(F ) is surjective.
We have only to show that if A0 is an arbitrary member of T (F ), there exists a
member A of T (X) satisfying χA |F = χA0 , i.e., A ∩ F = A0 . Since A0 is an open
subset of F , it can be expressed as an intersection of F and an open subset U
of X. Take x arbitrarily from A0 . Then by Lemma 7.1.1 we can find a compact
neighborhood Vx of x contained in U . Since A0 is compact, it can be covered by a
100 JUN-ICHI IGUSA
finite number of Vx . If we denote by A the union of such Vx , then A is in T (X) and

A0 = A ∩ F . Therefore, (∗) is indeed an exact sequence. Since every subspace of a
vector space over a field has a supplement, i.e., another subspace so that the vector
space becomes their direct sum, by dualizing (∗) we get a similar exact sequence.
We shall formulate it as a proposition.
Proposition 7.1.1 Let X denote a locally compact totally disconnected space and
express it as a disjoint union of nonempty open and closed subsets Y and F , re-
spectively. Then we have an exact sequence of C-linear maps:
0 → D(F ) → D(X) → D(Y ) → 0.
If T0 is any element of D(F ) , its image T0X in D(X) is defined as T0X (ϕ) = T0 (ϕ|F )
for every ϕ in D(X), and if T is any element of D(X) , its image T |Y in D(Y ) is
defined as (T |Y )(ψ) = T (ψ X ) for every ψ in D(Y ).
We might mention that Proposition 7.1.1 holds trivially if either Y or F becomes

∅. At any rate, we accept the fact that for every T in D(X) , there exists the
largest open subset O(T ) of X satisfying T |O(T ) = 0 and define the support of T
as Supp(T ) = X\O(T ). Then Proposition 7.1.1 implies the following corollary:
Corollary 7.1.1 If T is in D(X) and T = 0, then there exists a unique T0 in

D(Supp(T )) such that T = T0X .
The existence of O(T ) can be proved by a partition of unity, which is as follows:

Let C denote a compact subset of X which is covered by a family of open subsets
Uα of X. Then there exists a finite set {Ai ; i ∈ I} of mutually disjoint members
Ai of T (X) with each Ai contained in some Uα such that

χAi (x) = 1
i∈I
for every x in C. The proof is quite simple. At every x in C we can find by Lemma
7.1.1 a compact neighborhood Vx of x contained in some Uα . Since C is compact,
it can be covered by a finite number of Vx , say W1 , W2 , . . . . We have only to put
Ai = Wi \(W1 ∪ . . . ∪ Wi−1 )
for i = 1, 2, . . . . We shall prove the existence of O(T ). Define O(T ) as the union of
all open subsets Y of X such that T |Y = 0. We have only to show that T (ϕ) = 0
for every ϕ in D(X) with Supp(ϕ) contained in O(T ). If we apply the partition of
unity to C = Supp(ϕ) and Uα = Y above, then we get

ϕ= ϕχAi , T (ϕ) = T (ϕχAi ) = 0
i∈I i∈I
because Supp(ϕχAi ) is contained in Ai , hence in Wi , and hence in some Y .

7.2 The case of homogeneous spaces

We shall consider the case where a locally compact totally disconnected space X
is continuously acted upon by a locally compact totally disconnected group G. We
shall first show that Ω(G) = Hom(G, C× ) consists of all locally constant homomor-
phisms from G to C× . We have only to show that if ρ is any continuous homomor-
phism from G to C× , then ρ is locally constant. We choose a small neighborhood
of 1 in C× which does not contain any subgroup other than 1. We can, e.g., define
U as |a − 1| < 1. By Lemma 7.1.2 there exists a compact open subgroup H of G
such that ρ(H) is contained in U . By the choice of U we then have ρ(H) = 1. Since
ρ is a homomorphism, it takes the value ρ(g) on gH for every g in G.
After this remark we define the action of G on D(X) and D(X) in the usual
way, i.e., as
(g · ϕ)(x) = ϕ(g −1 x), (g · T )(ϕ) = T (g −1 · ϕ)
for every g in G, x in X, and ϕ, T respectively in D(X), D(X). We keep in mind
that g · χA = χgA for every g in G and A in T (X).
Lemma 7.2.1 Let T denote an element of D(X) satisfying
g · T = ρ(g)−1 T
with ρ(g) in C× for every g in G and assume that T = 0. Then necessarily ρ is in
Ω(G).
Proof. Since ρ clearly gives a homomorphism from G to C× , it will be enough to
show that ρ|N = 1 for some open subgroup N of G. Since T = 0, we have T (ϕ) = 0
for some ϕ in D(X), hence T (χA ) = 0 for some A = ∅ in T (X). As in the proof of
Lemma 7.1.2 we see that if B = {g ∈ G; gA ⊂ A}, then N = B ∩ B −1 is an open
subgroup of G and gA = A, i.e., g · χA = χA , for every g in N . This implies
ρ(g)T (χA ) = (g −1 · T )(χA ) = T (g · χA ) = T (χA ).
Since T (χA ) = 0, therefore, we get ρ(g) = 1 for every g in N .
We shall review some basic theorems about Haar measures on locally compact
groups and homogeneous spaces. We shall do so only in the totally disconnected
case but with proof. A measure µ on X is an element of D(X) satisfying µ(A) =
µ(χA ) ≥ 0 for every A in T (X). If ϕ is in D(X) we shall write

µ(ϕ) = ϕ(x)µ(x) = ϕ(x) dµ(x).
X X
By definition we have

ϕ(x)µ(x)≤ |ϕ(x)|µ(x) ≤ µ(Supp(ϕ))ϕ∞

X X

for every ϕ in D(X). In fact, if we express ϕ as a i χA i = a 1 χA 1 + a 2 χA 2 + . . .

with a1 , a2 , . . . in C, in which A1 , A2 , . . . are disjoint members of T (X), then

|µ(ϕ)| = | µ(Ai )ai | ≤ µ(Ai )|ai | = µ(|ϕ|) ≤ µ(Supp(ϕ)) max{|ai |}.
102 JUN-ICHI IGUSA
Therefore, if a sequence {ϕi } in D(X) has the property that µ(|ϕi − ϕj |) → 0 as

i, j → ∞, then {µ(ϕi )} forms a Cauchy sequence in C. In that case we say that ϕi
converges to Φ as i → ∞ and define its integral as the limit of µ(ϕi ):

Φ(x)µ(x) = lim ϕi (x)µ(x).
X i→∞ X
The above Φ is symbolic, but it can be interpreted as an integrable function on X.

We shall not give further details to this well-known extension process of µ except
for the following remark: If ϕ is a C-valued continuous function on X with compact
support, we can find a sequence {ϕi } in D(X) with Supp(ϕi ) contained in a fixed
member of T (X) satisfying ϕi −ϕ∞ → 0 as i → ∞. This implies ϕi −ϕj ∞ → 0,
hence µ(|ϕi − ϕj |) → 0 as i, j → ∞. In such a case we identify the above symbolic
Φ with ϕ.
We also mention that Fubini’s theorem is quite simple in the totally disconnected
case. If ν is a measure on a space Y similar to X and A, B are respectively in T (X),
T (Y ) and if ϕ = χA ⊗ χB , i.e., ϕ(x, y) = χA (x)χB (y) for every (x, y) in X × Y ,
then

ϕ(x, y)ν(y) µ(x) = ϕ(x, y)µ(x) ν(y)
X Y Y X
because both sides are equal to µ(A)ν(B). If we accept the fact that D(X × Y )
is the C-span of the set of functions of the form χA ⊗ χB , then the above formula
holds for every ϕ in D(X × Y ). Now we have seen that D(X × Y ) is the C-span of
the set of χC for all C in T (X × Y ) and we observe that every such C is a finite
union of its subsets of the form A × B. Therefore we have only to observe further
that any intersection of sets of the form A×B is of that form and that if A1 , . . . , An
are subsets of any set, the characteristic function of their union is the sum of the
characteristic function of the intersection of Ai1 , . . . , Aip with the sign (−1)p−1 for
all i1 < . . . < ip .
Now G continuously acts on G by left multiplication, i.e., as g0 · g = g0 g. A
Haar measure µG or simply µ on G is a G-invariant measure different from 0. We
shall show that it exists and is unique up to a factor in R× + . We observe that if
we have a finite number of members A, B, . . . of T (G), they can be expressed as
disjoint unions of cosets gN by one compact open subgroup N of G. In fact, by
Lemma 7.1.2 and by the compactness of A, B, . . . they can be expressed as finite
unions of gi Ni , where every Ni is a compact open subgroup of G. We can then take
as N any compact open subgroup of G contained in all Ni . After this remark, we
fix a compact open subgroup N0 of G, choose N for A and B = N0 , and put

µ(A) = card(A/N ) card(N0 /N ),
in which, e.g., card(A/N ) denotes the number of distinct cosets gN in A. Then

µ(A) is well defined, i.e., it is independent of the choice of N , and µ(gA) = µ(A)
for every g in G. Furthermore, µ is a simply additive function on T (G). Therefore,
µ is a Haar measure on G normalized as µ(N0 ) = 1. We observe that if T is any
G-invariant element of D(G) , the above argument shows that T (A) = T (N0 )µ(A)
for every A in T (G), i.e., T = T (N0 )µ, with T (N0 ) in C. If T is a Haar measure
on G, then T (N0 ) is in R×+.
If µ is a Haar measure on G and g is in G, then µ (A) = µ(Ag) also gives a Haar
measure on G, hence by the uniqueness we will have
µ(Ag) = ∆G (g) µ(A)
with ∆G (g) in R×+ independent of A. If we define a new continuous action of G on

−1
G as g0 · g = gg0 , then we can write g · µ = ∆G (g)µ. Therefore by Lemma 7.2.1 we
see that ∆G , called the module of G, gives a locally constant homomorphism from
G to R×+ . We shall show that

µ(A−1 ) = ∆G (g −1 ) µ(g)
A
for every A in T (G). We express A as a disjoint union of gi N , in which N is a

compact open subgroup of G such that ∆G |N = 1. Then A−1 becomes the disjoint
union of N gi−1 , hence

µ(A−1 ) = µ(N gi−1 ) = ∆G (gi−1 ) µ(N )
i i

= ∆G (g −1 ) µ(g) = ∆G (g −1 ) µ(g).
i gi N A
Since
χA (gg0−1 ) = χAg0 (g), χA (g −1 ) = χA−1 (g),
the above properties of µ can also be expressed as

−1
ϕ(gg0 ) µ(g) = ∆G (g0 ) ϕ(g) µ(g),
G G

ϕ(g −1 )∆G (g −1 ) µ(g) = ϕ(g) µ(g)
G G
for ϕ = χA , hence for every ϕ in D(G).

We take an arbitrary closed subgroup H of G, hence H need not be compact
or open, and consider the coset space G/H. Then p(g) = gH for every g in G
gives a surjection p : G → G/H. We call a subset of G/H open if its preimage
under p is open in G. The family of open subsets of G/H so defined is closed under
the taking of arbitrary union and finite intersection, hence it converts G/H into a
topological space. Furthermore p is open in the sense that it maps an open set to
an open set. If g1 H = g2 H for some g1 , g2 in G, then 1 is not in g1 Hg2−1 , hence
G\g1 Hg2−1 is a neighborhood of 1. By Lemma 7.1.2 there exists a compact open
subgroup N of G contained in G\g1 Hg2−1 . Then N and g1 Hg2−1 are disjoint, hence
p(N g1 ) and p(N g2 ) are disjoint, and hence G/H is a Hausdorff space. Furthermore,
p(N g) is a compact neighborhood of p(g) for every g in G and the set of p(N g)
for all compact open subgroups N of G forms a base of p(g). We could have
104 JUN-ICHI IGUSA
used gN instead of N g above. At any rate we have shown that G/H is a locally
compact totally disconnected space. We observe that G continuously acts on G/H
as g0 · p(g) = p(g0 g). The action is transitive and p is equivariant.
Since H is a closed subgroup of G, it is also a locally compact totally discon-
nected group. Therefore H has a Haar measure µH . If ϕ is in D(G), then ϕ(gh) as
a function of h is in D(H), hence

π(ϕ)(gH) = ϕ(gh) µH (h)
H
is defined for every g in G. In particular, if N is a compact open subgroup of G,

then for any g0 in G we can take χg0 N as ϕ, and we get

π(χg0 N )(gH) = χg0 N (gh) µH (h).
H
If gh is in g0 N , then g is in g0 N H. In that case, the RHS becomes µH (H ∩ N ),

hence
π(χg0 N ) = µH (H ∩ N )χg0 N H .
This formula implies that π gives an equivariant C-linear surjection from D(G) to
D(G/H). In the following proposition EX (ρ) denotes the vector space over C of all
T in D(X) satisfying g · T = ρ(g)−1 T for every g in G.
Proposition 7.2.1 Let G denote a locally compact totally disconnected group, H

a closed subgroup of G, and ρ an element of Ω(G). Then EG/H (ρ) = 0 if and only
if
ρ∆G H = ∆H .
In that case dimC (EG/H (ρ)) = 1 and EG/H (ρ) has a basis T0 defined by

T0 (π(ϕ)) = ρ(g)ϕ(g)µ(g)
G
for every ϕ in D(G).
Proof. This can be proved in the same way as the statement in italics on p. 45
in A. Weil [56]. Suppose that T = 0 is in EG/H (ρ) and define S in D(G) as
S(ϕ) = T (π(ϕ)) for every ϕ in D(G). Then S is in EG (ρ) and S1 (ϕ) = S(ρ−1 ϕ)
defines an element S1 of EG (1). Since S1 = 0 by T = 0, we have S1 = cµ for
some c in C× , hence T (π(ϕ)) = cµ(ρϕ). In particular, dimC (EG/H (ρ)) = 1. If
we take h0 arbitrarily from H and replace ϕ(g) by ϕ1 (g) = ϕ(gh−1
0 ), then we get
π(ϕ1 ) = ∆H (h0 )π(ϕ), hence
T (π(ϕ1 )) = ∆H (h0 )T (π(ϕ)), µ(ρϕ1 ) = ρ(h0 )∆G (h0 )µ(ρϕ).
This implies ∆H (h0 ) = ρ(h0 )∆G (h0 ), hence ρ∆G |H = ∆H .

We shall show that if ρ∆G |H = ∆H , then T0 (π(ϕ)) = µ(ρϕ) gives a well-defined
element T0 of D(G/H) . That will complete the proof because then T0 is in EG/H (ρ).
We have only to show therefore that π(ϕ) = 0 implies µ(ρϕ) = 0. Now π(ϕ) = 0
implies
ϕ(gh−1 )∆H (h−1 )µH (h) = ϕ(gh)µH (h) = 0
H H
for every g in G. If we take any θ from D(G), multiply (ρθ)(g) to the LHS of the
above equation and integrate over G, then by using Fubini’s theorem we get

−1
∆H (h ) (ρθ)(g) ϕ(gh−1 ) µ(g) µH (h) = 0,
H G
in which
(ρθ)(g) ϕ(gh−1 ) µ(g) = ∆G (h) (ρθ)(gh) ϕ(g) µ(g).
G G
Therefore, by using ∆H (h) = ρ(h)∆G (h) and Fubini’s theorem again we get

(ρϕ)(g) θ(gh) µH (h) µ(g) = 0.
G H
If we choose A from T (G) which contains Supp(ϕ), e.g., A = Supp(ϕ), and specialize
θ to any element of D(G) satisfying π(θ) = χAH , then we finally get µ(ρϕ) = 0.
We shall prove another proposition for our later use. As we have remarked
in Chapter 5.1, if G is a compact group, then Hom(G, R× + ) = 1, hence Ω(G) =
Hom(G, C× 1 ). If further G is commutative, then elements of Ω(G) are called char-
acters of G. We keep in mind that if G is a finite abelian group, then G is isomorphic
to a product of cyclic groups, hence Ω(G) and G have the same order.
Proposition 7.2.2 Let G denote a compact totally disconnected abelian group and
µG its Haar measure normalized as µG (G) = 1; define an inner product (ϕ, ϕ ) of
every ϕ, ϕ in D(G) as

(ϕ, ϕ ) = ϕ(g)ψ(g)µG (g),
G
in which ψ(g) is the complex conjugate of ϕ (g). Then Ω(G) forms an orthonormal
basis for D(G), i.e., (χ, χ ) = 1 or 0 according as χ = χ or χ = χ for every χ, χ
in Ω(G) and every ϕ in D(G) can be expressed as a finite sum

ϕ= cχ χ
χ∈Ω(G)
with cχ = (ϕ, χ) for every χ in Ω(G).
Proof. We shall first prove the orthonormality of Ω(G). Since (χ, χ) = 1 is clear,
we shall show that (χ, χ ) = 0 for χ = χ . Since the complex conjugate of χ(g) is
χ(g)−1 , we have only to show that I = (χ, 1) represents 0 for χ = 1. Now χ = 1
means χ(g0 ) = 1 for some g0 in G. If we replace χ(g) in the integral I by χ(g0 g),
then the new integral is still I because µG is a Haar measure and also equal to χ(g0 )I
because χ(g0 g) = χ(g0 )χ(g), hence I = 0. We shall show that Ω(G) is complete,
106 JUN-ICHI IGUSA
i.e., it spans D(G). Take ϕ arbitrarily from D(G). Since ϕ is locally constant, by
Lemma 7.1.2 for every g in G there exists a compact open subgroup Hg of G such
that ϕ|gHg becomes a constant function on gHg . Since G is compact, it can be
covered by a finite number of gHg . Let H denote any compact open subgroup of G
contained in all such Hg . Then ϕ becomes a C-valued function on the finite abelian
group G/H. We observe that the dimension over C of the vector space of all such
ϕ is equal to the order of G/H, which is equal to the order of Ω(G/H) considered
as a subgroup of Ω(G). The orthonormality implies that elements of Ω(G/H) are
linearly independent over C, hence they form a basis for the above vector space.
Therefore, ϕ can be expressed as a linear combination of elements of Ω(G/H) with
coefficients as stated.
We might mention that in the notation of Proposition 7.2.2 we have the following
Plancherel formula:
|ϕ(g)|2 µG (g) = |cχ |2 .
G χ∈Ω(G)
Proposition 7.2.2 will be used mostly as

(χ, 1) = χ(g)µG (g) = 0
G
for χ = 1. We shall refer to this fact as the orthogonality of characters.
7.3 Structure of eigendistributions

We start with Baire’s theorem in general topology. We say that a locally compact
space X is countable at ∞ if X can be expressed as a union of countably many, i.e.,
at most countably many, compact subsets F1 , F2 , . . . . The theorem then states that
at least one Fi contains a nonempty open subset of X. For the sake of completeness
we shall give a proof in the case where X is totally disconnected. Suppose otherwise
and choose A1 = ∅ from T (X). Assume by induction that A1 ⊃ . . . ⊃ Ai has been
chosen from T (X) such that Ai = ∅ and Ai is disjoint from F1 , . . . , Fi−1 for some
i ≥ 1. Since Ai \Fi is nonempty and open by assumption, by Lemma 7.1.1 it contains
Ai+1 = ∅ in T (X). Then Ai+1 is disjoint from F1 , . . . , Fi . Therefore, the induction
is complete. If now A∞ denotes the intersection of all Ai , then A∞ is nonempty
and it is disjoint from Fi for all i, hence with their union, which is X. We thus have
a contradiction.
We shall next recall a theorem of L. S. Pontrjagin. Let G denote a locally
compact group acting continuously and transitively on a locally compact space X;
let ξ denote any point of X and H the fixer of ξ in G. Then H is a closed subgroup
of G and the continuous map f : G → X defined by f (g) = gξ gives rise to an
equivariant continuous bijection f0 : G/H → X. The theorem states that if G
is countable at ∞, then f0 is bicontinuous. We shall again give a proof in the
case where G and X are totally disconnected. We have only to show that f0 , or,
equivalently, f is an open map. Namely if U is any nonempty open subset of G, then
f (U ) is open in X. An arbitrary point of f (U ) can be written as f (g) = gξ for some
g in U . Since g −1 U is a neighborhood of 1 = 1−1 1, by Lemma 7.1.2 we can find a

compact open subgroup N of G such that N −1 N is contained in g −1 U . Since G is
countable at ∞ by assumption, there exists a countable subset {gi ; i ∈ I} of G such
that G becomes the union of gi N , hence X becomes the union of f (gi N ) = gi f (N ),
for all i in I. Since the action of G on X is bicontinuous, every gi f (N ) is compact,
hence X is countable at ∞, and hence by Baire’s theorem gi f (N ) for some i contains
a nonempty open subset of X. Then f (N ) itself contains a nonempty open subset
say V of X. Take any point of V and express it as f (g0 ) = g0 ξ for some g0 in N .
Then W = gg0−1 V becomes a neighborhood of gξ = f (g) in X and by definition
W ⊂ gg0−1 f (N ) ⊂ gf (g0−1 N ) ⊂ gf (g −1 U ) = f (U ).
Therefore, f (U ) is open in X.
We have separated, for the sake of clarity, the following lemma from the next
theorem; the proof will be given only in the totally disconnected case.
Lemma 7.3.1 Let G denote a locally compact group acting continuously on a lo-
cally compact space X; assume that G is countable at ∞ and the number of G-orbits
in X is countable. Then there exists an open orbit, i.e., a G-orbit which is open
in X. Furthermore for every ξ in X the G-orbit Gξ is the unique open orbit in its
closure in X and if H is the fixer of ξ in G, then G/H is bicontinuous to Gξ under
gH → gξ.
Proof. If N is any compact open subgroup of G, since G is countable at ∞ by
assumption, there exists a countable subset {gi ; i ∈ I} of G such that G becomes
the union of gi N for all i in I. Also by assumption, we can choose a countable set
of representatives {ξj ; j ∈ J} of all G-orbits in X. Then X becomes the union of
countably many compact subsets gi N ξj for all i, j. Therefore, by Baire’s theorem
gi N ξj for some i, j contains a nonempty open subset of X. This implies that N ξj
contains a nonempty open subset say V of X. Then Gξj for that j becomes the
union of gV for all g in G, hence Gξj is an open orbit. If for any ξ in X we put
Y = Gξ, then G acts on the closure Ȳ of Y in X. Since Ȳ is locally compact and
the number of G-orbits in Ȳ is countable, we can apply the above observation to
Ȳ instead of X. We see that Ȳ contains an open orbit say Y0 . If Y = Y0 , then
they are disjoint, hence Y is contained in Ȳ \Y0 . Since Y0 is open in Ȳ , Ȳ \Y0 is
closed in Ȳ , hence in X, and hence Ȳ is contained in Ȳ \Y0 . This is a contradiction.
Therefore Y is the unique open orbit in Ȳ . The last part of the lemma follows from
Pontrjagin’s theorem.
We also make the following remark. Let G denote a locally compact totally
disconnected group acting continuously on a locally compact totally disconnected
space X. Then for every g in G, ϕ in D(X), and T in D(X) , we have
Supp(g · ϕ) = g · Supp(ϕ), Supp(g · T ) = g · Supp(T ).
Therefore, if T is in EX (ρ) for some ρ in Ω(G), then Supp(T ) is G-invariant. If now
X is a disjoint union of closed and open subsets F and Y , respectively, and if they
are G-invariant, then the C-linear maps in the dual exact sequences
0 → D(Y ) → D(X) → D(F ) → 0, 0 → D(F ) → D(X) → D(Y ) → 0
108 JUN-ICHI IGUSA
are all equivariant. For instance, if g is in G and T0 , T are respectively in D(F ) ,

D(X) , then
(g · T0 )X = g · T0X , (g · T )Y = g · (T Y ).
The verifications are all straightforward. In particular, T0X is in EX (ρ) if and only
if T0 is in EF (ρ), and T |Y is in EY (ρ) if T is in EX (ρ).
Theorem 7.3.1 Suppose that G is a locally compact totally disconnected group
which is countable at ∞ and that it acts continuously with countably many orbits
on a locally compact totally disconnected space X; suppose further that T is in
EX (ρ), i.e., T is an element of D(X) satisfying g · T = ρ(g)−1 T for every g in G,
and T = 0. Put F =Supp(T ) and write T = T0X for a unique T0 in D(F ) ; choose
ξ from any open orbit in F and denote by H the fixer of ξ in G. Then T0 is in
EF (ρ), hence T0 |Gξ is in EGξ (ρ), and T0 |Gξ = 0. Therefore

EGξ (ρ) = C(T0 Gξ), ρ∆G H = ∆H .
Furthermore, if µG , µH denote Haar measures on G, H and π the C-linear surjec-
tion from D(G) to D(Gξ) defined by

(π(ϕ))(gξ) = ϕ(gh) µH (h),
H
then

(T0 Gξ)(π(ϕ)) = c · ρ(g)ϕ(g) µG (g)
G
for every ϕ in D(G) with c in C× independent of ϕ.
Proof. In view of Lemma 7.3.1, the above remark, and Proposition 7.2.1, we have
only to show that T0 |Gξ = 0. Suppose that T0 |Gξ = 0 and choose an open subset
U of X satisfying U ∩ F = Gξ. If we put V = U ∪ O(T ), then V is open in X.
Furthermore, if ϕ is in D(X) with Supp(ϕ) contained in V , then

Supp(ϕF ) ⊂ V ∩ F = U ∩ F = Gξ,
hence T (ϕ) = T0X (ϕ) = T0 (ϕ|F ) = 0 by the assumption that T0 |Gξ = 0. Therefore,
T |V = 0, hence V is contained in O(T ), and hence U is contained in O(T ). This
implies that Gξ is empty, which is a contradiction.
We might mention that we have been motivated by A. Weil [58], Chapter 3,
Lemma 16. An example of locally compact totally disconnected spaces which are
not countable at ∞ is R but with discrete topology. We keep in mind that every
locally compact space which is separable, i.e., has a countable base of open sets, is
countable at ∞. We also keep in mind that the condition ρ∆G |H = ∆H will never
be satisfied if ρ|H is not R×
+ -valued.
7.4 Integration on p-adic manifolds

We take a complete field K with respect to a non-archimedean absolute value | · |K .
We have already used such a field in Chapter 2.2. In the same notation as at that
place we impose the following condition:
AV 4. The factor ring OK /πOK is a finite field Fq with q elements.

We call such a K a p-adic field and normalize | · |K once and for all as
|π|K = q −1
We observe that if we choose a subset R of OK which is mapped bijectively to Fq
under OK → OK /πOK , then K becomes the set of all series of the form

a= ai π i = π k (ak + ak+1 π + . . . )
i≥k
with ak , ak+1 , . . . in R for some k in Z depending on a. In particular, OK is the

set of all a above for k = 0, hence OK becomes bijective to the product R∞ under
the correspondence a → (a0 , a1 , . . . ). We recall that K is a topological field and
π i OK for all i in N form a base at 0. Therefore, if R∞ is equipped with the product
topology, then the bijection OK → R∞ becomes bicontinuous, hence OK is compact
by Tychonoff’s theorem. We could have said that OK is the inverse limit of the
sequence of finite rings OK /πOK , OK /π 2 OK . . . , hence OK is a compact ring. At
any rate, since π i OK are compact open subgroups of the additive group K for all i
in Z with K as their union, we see that K is a locally compact totally disconnected
group which is countable at ∞. This implies that K n is also such a group. We shall
n
normalize its Haar measure as µ(OK ) = 1 and denote it sometimes by µn and also
by dx. We observe that every K-analytic manifold X for such a K or simply a p-
adic manifold is locally compact and totally disconnected. In the following we shall
explain the process of associating a measure to any K-analytic differential form on
X of the highest degree. We shall start with an elementary divisor theorem.
We denote by GLn (OK ) the subgroup of GLn (K) consisting of all g in Mn (OK )
×
with det(g) in OK . If we denote by 1n the unit element of GLn (K), then 1n +
πMn (OK ) is a compact open normal subgroup of GLn (OK ) with a finite factor
group. Therefore, GLn (OK ) is a compact open subgroup of GLn (K). Furthermore,
it contains all permutation matrices, i.e., matrices of determinant ±1 with each row
containing 1 once and having 0 for the remaining n − 1 entries.
Lemma 7.4.1 Every element a of Mm,n (K) can be expressed as a = gdg , in which
g, g are respectively in GLm (OK ), GLn (OK ) and d is a diagonal matrix, i.e., the
(i, j)-entries are 0 for all i = j, such that its diagonal entries are powers of π with
increasing exponents in Z ∪ ∞ and with the understanding that π ∞ = 0.
Proof. After replacing a by t a if necessary, we may assume that m ≥ n. Since
we can take d = 0 if a = 0, we shall assume that a = 0. If π e u is an entry of
×
a with u in OK and with the smallest e, then after replacing a by (π e u)−1 a and
e
multiplying π u later, we may assume that a is in Mm,n (OK ) with 1 as one of its
entries. After multiplying permutation matrices from both sides, we may further
assume that a has 1 as its (1, 1)-entry. We express a by its entry matrices a11 = 1
and 1 × (n − 1), (m − 1) × 1, (m − 1) × (n − 1) matrices a12 , a21 , a22 . If we denote
by g1 the element of GLm (OK ) with 1, 0, a21 , 1m−1 as its entry matrices and by
g1 the element of GLn (OK ) with 1, a12 , 0, 1n−1 as its entry matrices, then we will
have
a = g1 a1 g1 ,
110 JUN-ICHI IGUSA
in which the entry matrices of a1 are 1, 0, 0, a∗ with a∗ = a22 − a21 a12 . We have
tacitly assumed that n > 1. If n = 1, then we simply have g1 = 1. At any rate, by
using elements of GLm (OK ), GLn (OK ) with entry matrices of the form 1, 0, 0, g ∗
with g ∗ respectively in GLm−1 (OK ), GLn−1 (OK ) we can simplify a∗ in the same
way as above, i.e., by an induction on n.
Lemma 7.4.2 We have
µn (gA) = | det(g)|K µn (A)
for every g in GLn (K) and A in T (K n ).

Proof. We shall use general observations in section 7.2, especially Lemma 7.2.1. If
we let GLn (K) act on K n by matrix-multiplication, then we will have
g −1 · µn = ρ(g) µn , i.e., µn (gA) = ρ(g) µn (A)
for every g in GLn (K) and A in T (K n ) with ρ in Hom(GLn (K), R× + ). If we write

g = γdγ with γ, γ in GLn (OK ) by Lemma 7.4.1, since ρ|GLn (OK ) = 1 by the
compactness of GLn (OK ), we get ρ(g) = ρ(d). We similarly have | det(g)|K =
| det(d)|K . If π e1 , . . . , π en are the diagonal entries of d, we choose e from N such
that e + ei ≥ 0 for all i. Then, by using card(OK /π e OK ) = q e = |π e |−1 K , we get
n

ρ(d) = µn (dOK ) = card(dOkn /π e dOK n
) card(OK n
/π e dOK n
)

= |π |K = | det(d)|K ,
ei
1≤i≤n
hence ρ(g) = | det(g)|K .

In the following lemma an SRP in x1 , . . . , xn is, as we have defined in Chapter
2.2, an element of K[[x1 , . . . , xn ]] with the coefficient ci of xi in π |i|−1 OK for all
i = 0 in Nn and c0 = 0.
Lemma 7.4.3 (i) Suppose that f (x) in OK [[x1 , . . . , xn ]] is convergent at some a

n
in OK and e > 0 is in N. Then

g(x) = π −e f (a + π e x) − f (a)
is an SRP in x1 , . . . , xn . (ii) If every fi (x) in f (x) = (f1 (x), . . . , fn (x)) is an SRP

in x1 , . . . , xn and
∂(f1 , . . . , fn )/∂(x1 , . . . , xn ) ≡ 0 mod π,

n
then the bicontinuous map from OK to itself defined by y = f (x) is measure pre-
serving.

Proof of (i). By assumption f (x) = ci xi with ci in OK for all i in Nn , hence

f (a + π e x) = f (a) + di (π e x)i
|i|≥1
with di in OK , and hence

g(x) = π ki di xi ,
|i|≥1
in which ki = e(|i| − 1) ≥ |i| − 1 for all i = 0. Therefore, g(x) is an SRP in

x1 , . . . , xn .
Proof of (ii). We have only to show that the image of a + π e OK
n
under y = f (x)
is f (a) + π OK for every a in OK and e > 0 in N because then
e n n
µn (f (a) + π e OK
n
) = µn (π e OK
n
) = µn (a + π e OK
n
),
hence y = f (x) is measure preserving. We know by Corollary 2.2.1 that the inverse
of y = f (x) has a similar form as y = f (x). Therefore, we have only to show
that f (a + π e b) is contained in f (a) + π e OK
n
for every b in OKn
. If we put g(x) =
−e
π (f (a + π x) − f (a)), then the entries of g(x) are all SRP’s in x1 , . . . , xn by
e
n n
(i), hence they are convergent at every b in OK . Therefore, g(b) is in OK , hence
e e n n
f (a + π b) is in f (a) + π OK for every b in OK .
We are ready to prove a change of variable formula in the p-adic case. The
formula is identical to the well-known formula in the archimedean case.

Proposition 7.4.1 If every fi (x) in f (x) = f1 (x), . . . , fn (x) is K-analytic
around some point a of K n and
∂(f1 , . . . , fn )/∂(x1 , . . . , xn )(a) = 0
so that y = f (x) gives a K-bianalytic map from a neighborhood U of a to a neigh-

borhood V of b = f (a), then
dy = |∂(f1 , . . . , fn )/∂(x1 , . . . , xn )|K dx
with the understanding that dx = µn |U and dy = µn |V .

Proof. We first observe that if the formula is valid on some small neighborhood
of every point of U , then it is valid on U . Therefore, in proving the formula we
can make U smaller if necessary. Also if the formula is valid for y = f (x) and for
a similar map z = g(y), then by the chain rule of the jacobian it is valid for the
composite map z = g(f (x)). Now the formula is valid for y = gx + a, in which g is
in GLn (K) and a is in K n , this by Lemma 7.4.2 and by the fact that µn is a Haar
measure on K n . Also the formula is valid by Lemma 7.4.3 if every fi (x) is an SRP
in x1 , . . . , xn of the form
fi (x) = xi + cij xj
|j|≥2
for 1 ≤ i ≤ n. Therefore, in the general case we may assume that fi (x) is of the
above form with cij in K for all i, j. Since every fi (x) is a convergent power series
by assumption, there exists e0 in N such that cij π e0 |j| tends to 0 as |j| → ∞ for
1 ≤ i ≤ n. Then there exists e1 in N such that
cij = π e1 · cij π e0 |j|

112 JUN-ICHI IGUSA
is in OK for all i, j. If we choose e ≥ 2e0 + e1 + 1 from N and put

gi (x) = π −e fi (π e x) = xi + cij xj ,
|j|≥2
we will have
cij = π kj · π |j|−1 cij ,
in which
kj = (e − e0 − 1)(|j| − 1) − e0 − e1 ≥ (e0 + e1 )(|j| − 2) ≥ 0
for all |j| ≥ 2. Therefore, as we have remarked, the formula is valid for x → g(x) =
(g1 (x), . . . , gn (x)). Since y = f (x) is the composition of x → π −e x, x → g(x), x →
π e x, the formula is valid for y = f (x).
After the above preparation, suppose that we have an n-dimensional p-adic
manifold X defined by an atlas {(U, φU )} and a K-analytic differential form α of
degree n on X; put φU (x) = (x1 , . . . , xn ) for every x in U . Then α|U has an
expression of the form
α(x) = fU (x) dx1 ∧ · · · ∧ dxn ,
in which fU is a K-analytic function on U . If A is any member of T (X) small
enough to be contained in U , then we define its measure µα (A) as

µα (A) = |fU (x)|K µn (φU (x))
A

= q −e · µn φU (fU−1 (π e OK
×
) ∩ A) .
e∈Z
We observe that the above series is convergent because fU (A) is a compact subset
of K, hence a subset of π −e0 OK for some large e0 in N, and then the summation
is restricted as e ≥ −e0 . The point is that if (U , φU ) is another chart and if A is
also contained in U , then we will have the same µα (A) relative to that chart. In
fact, if φU (x) = (x1 , . . . , xn ), then
fU (x) ∂(x1 , . . . , xn )/∂(x1 , . . . , xn ) = fU (x),
µn (φU (x)) = |∂(x1 , . . . , xn )/∂(x1 , . . . , xn )|K µn (φU (x))
for every x in U ∩ U , the first by definition and the second by Proposition 7.4.1.
Therefore, we indeed have

|fU (x)|K µn (φU (x)) = |fU (x)|K µn (φU (x)).
A A
If now A is arbitrary in T (X), we first express A as a disjoint union of small

A1 , A2 , . . . in T (X) such that every Ai is contained in some U above depending
on i, and then define µα (A) as µα (A) = µα (A1 ) + µα (A2 ) + . . . . In this way we
get a well-defined measure µα on X, i.e., a simply additive function µα on T (X)
such that µα (A) ≥ 0 for every A in T (X). The measure µα is also denoted by dµα ,
|α|K , etc. At any rate, by the general procedure explained in sections 7.1 and 7.2
the integral of any ϕ in D(X) by µα is also defined.
7.5 Serre’s theorem on compact p-adic manifolds

We shall explain a theorem of J.-P. Serre [52] which describes the structure of an ar-
bitrary compact p-adic manifold by its invariant. The proof is by p-adic integration
and short, and yet the result is very remarkable. We shall start with the following
lemma:
Lemma 7.5.1 Let K denote a p-adic field and X any compact n-dimensional K-
analytic manifold for n > 0. Then X is K-bianalytic to the disjoint union r · OK
n
n
of r copies of OK for some 0 < r < q.
Proof. We take an atlas {(U, φU )} on X. We may assume that every U is a compact

open subset of X. Since X is compact by assumption, it is covered by a finite number
of U ’s, say U1 , U2 , . . . . After replacing U1 , U2 , U3 , . . . by U1 , U2 \U1 , U3 \(U1 ∪
U2 ), . . . , we may assume that they are disjoint. We know that each φU (U ) for
U = U1 , U2 , . . . is a disjoint union of a finite number of subsets of K n of the form
a + π e OK n
for some a in K n and e in Z. We observe that OK n
is K-bianalytic to
a + π OK under x → y = a + π x. Therefore, X is K-bianalytic to r · OK
e n e n
for some
r > 0 in N. If now {c1 , c2 , . . . , cq } is a subset of OK which is mapped bijectively
to Fq under c → c mod π, then OK becomes the disjoint union of ci + πOK for
1 ≤ i ≤ q. Therefore, OK is K-bianalytic to q · OK , hence OK n n−1
= OK × OK is
K-bianalytic to q · OK for every n > 0. If we write the above r as r = (q − 1)i + r0
n
with i in N and 0 < r0 < q, then by an induction on i we see that X is K-bianalytic

to r0 · OK n
.
If X is any n-dimensional K-analytic manifold and α is a K-analytic differential
form on X of degree n, then for any chart (U, φU ) on X with φU (x) = (x1 , . . . , xn )
we can write
α(x) = fU (x) dx1 ∧ · · · ∧ dxn ,
in which fU is a K-analytic function on U . We say that α is a gauge form on X if fU
is K × -valued for every U . We are ready to state and prove the following theorem
of Serre.
Theorem 7.5.1 Suppose that K is a p-adic field, n > 0, and X is a compact n-

dimensional K-analytic manifold. Then X possesses a gauge form α and µα (X) is
of the form N/q m for some N, m in N with N > 0. If we define 0 < i(X) < q in
N as
µα (X) ≡ i(X) mod q − 1,
i.e., as µα (X) − i(X) in (q − 1) Z[1/q], then i(X) depends only on X and X is
K-bianalytic to i(X) · OKn
. Furthermore, X is K-bianalytic to another compact
n-dimensional K-analytic manifold Y if and only if i(X) = i(Y ).
Proof. If X, Y are n-dimensional K-analytic manifolds and f : X → Y is a K-

bianalytic map, and further if Y has a gauge form β, then clearly α = f ∗ (β) is a
gauge form on X. Furthermore, if X, Y are compact, then µα (X) = µβ (Y ). We
know by Lemma 7.5.1 that X is K-bianalytic to r · OKn
for some 0 < r < q. If we
114 JUN-ICHI IGUSA
take Y = r · OK
n
and define β on each OK n
as dx1 ∧ · · · ∧ dxn so that µβ = µn in our
notation, then β is a gauge form on Y , hence α = f ∗ (β) is a gauge form on X, and
µα (X) = µβ (r · OK
n
) = r · µn (OK
n
) = r.
If now α, α are any two gauge forms on X, then we can find an atlas {(U, φU )} on
X with φU (x) = (x1 , . . . , xn ) such that
α(x) = fU (x) dx1 ∧ · · · ∧ dxn , α (x) = fU (x) dx1 ∧ · · · ∧ dxn ,
in which fU , fU are K × -valued K-analytic functions on U . Since q Z is a discrete

subset of R, by subdividing U if necessary we may assume that

|fU (x)|K = q eU , |fU (x)|K = q eU
for all x in U , in which eU , eU are in Z and independent of x. We may, as before,

assume that the U ’s above are disjoint, hence finite. We then have

µα (X) = q eU · µn (φU (U )), µα (X) − µα (X) = (q eU − q eU )µn (φU (U )),
U U
in which µn (φU (U )) is a finite sum of elements of q Z . Therefore, µα (X) = N/q m as

stated in the theorem and µα (X) ≡ µα (X) mod q − 1. We have only to put these
together.
7.6 Integration over the fibers

Let X, Y denote K-analytic manifolds which are respectively of dimensions n, m
and f : X → Y a K-analytic map; let a denote an arbitrary point of X and put
b = f (a). Then f gives rise to a K-linear map f ∗ : Ωb (Y ) → Ωa (X) and its dual
map Ta (X) → Tb (Y ). We say that f is submersive if Ta (X) → Tb (Y ) is surjective,
i.e., Ωb (Y ) → Ωa (X) is injective, for every a. This clearly implies that n ≥ m. We
shall express the above condition in terms of local coordinates on X, Y . We choose
charts (U, φU ), (V, ψV ) on X, Y , respectively such that f (U ) is contained in V ,
and we put φU (x) = (x1 , . . . , xn ), ψV (y) = (y1 , . . . , ym ). We shall examine the
condition that f |U : U → V is submersive. We have

f ∗ (dyi ) = d(yi ◦ f ) = (∂(yi ◦ f )/∂xj ) dxj
1≤j≤n
for 1 ≤ i ≤ m. We denote by Jx the jacobian matrix with ∂yi /∂xj = ∂(yi ◦ f )/∂xj
as its (i, j)-entry for 1 ≤ i ≤ m, 1 ≤ j ≤ n. Then f |U is submersive if and only
if the m rows of Jx are linearly independent, i.e., rank(Jx ) = m, for every x in
U . In that case, after a permutation of x1 , . . . , xn and by making U smaller if
necessary, we may assume that the first m columns of Jx are linearly independent.
Then by the implicit function theorem y1 ◦ f, . . . , ym ◦ f , xm+1 , . . . , xn form local
coordinates on X at every point of U . Therefore, we may replace xi by yi ◦ f for
1 ≤ i ≤ m possibly after making U still smaller. This implies that every submersive
map is open. Furthermore, it gives a simple description of the fiber Xb = f −1 (b)
of f over an arbitrary point b of Y . In fact, if b is not in f (X), then Xb = ∅.
If b = f (a) for some a in X, we may assume that a is in the above U . If we put
ψV (b) = (b1 , . . . , bm ), then Xb ∩U is defined by xi −bi = 0 for 1 ≤ i ≤ m. Therefore,
if we put p = n − m, then Xb becomes a closed p-dimensional submanifold of X.
Now let α denote a K-analytic differential form of degree n on X and β a gauge
form on Y . Then, in the above notation, we can write
α(x) = A(x) dx1 ∧ · · · ∧ dxn , β(y) = B(y) dy1 ∧ · · · ∧ dym ,
in which A, B are K-analytic functions respectively on U, V and B is K × -valued.

If we take a K-analytic differential form γ of degree p on U , then it can be written
as
γ(x) = Ci1 ...ip (x) dxi1 ∧ · · · ∧ dxip ,
in which Ci1 ...ip for every 1 ≤ i1 < . . . < ip ≤ n is a K-analytic function on U .

Furthermore, γ satisfies the condition
α = f ∗ (β) ∧ γ
on U if and only if A = (B ◦ f )Cm+1,... ,n on U , and then

γ (Xb ∩ U ) = (A/(B ◦ f )) dxm+1 ∧ · · · ∧ dxn (Xb ∩ U )
for b = f (a). We observe that the LHS does not depend on the local expressions of
α, β and the RHS is independent of γ. Therefore, the restriction of γ to Xb depends
only on α, β, hence it gives rise to a K-analytic differential form θb of degree p on
the whole Xb with the above local expression. We take a variable point y of Y and
define θy on Xy as above if Xy = ∅ and θy = 0 otherwise. We shall write
θy = (α/f ∗ (β))y = α/f ∗ (β) = α/β.
We remark that if α is a gauge form on X, then θy = α/β is a gauge form on Xy

for every y in Y .
Theorem 7.6.1 Let X, Y denote p-adic manifolds, f : X → Y a submersive map,

and α, β gauge forms on X, Y respectively. Then for every ϕ in D(X) the C-valued
function Fϕ on Y defined by

Fϕ (y) = ϕ(x) µα/β (x)
Xy
is in D(Y ) and

ϕ(x) µα (x) = Fϕ (y) µβ (y).
X Y
116 JUN-ICHI IGUSA
Proof. Since Fϕ and both sides of the integration-formula to be proved depend C-

linearly on ϕ, we may assume that ϕ = χW for some W in T (X). By subdividing W
if necessary, we may further assume that W is arbitrarily small. Therefore, we shall
assume from the beginning that W is contained in the U for a chart (U, φU ) on X
which we have used to define α/β and φU (W ) = a+π e OK n
for some a = (a1 , . . . , an )
in K and e in N. We recall that
n
α(x) = A(x) dx1 ∧ · · · ∧ dxn , β(y) = B(y) dy1 ∧ · · · ∧ dym ,

and xi = yi ◦ f for 1 ≤ i ≤ m. We put x = (x1 , . . . , xm ), x = (xm+1 , . . . , xn )

and similarly a = (a1 , . . . , am ), a = (am+1 , . . . , an ). We shall, for the sake of

simplicity, use the same notation for x and φU (x) and also for y and x , hence
W = a + π e OK n
. We put W = a + π e OK m
, W = a + π e OK p
, where p = n − m.

Since α, β are gauge forms, we have A(a) = 0, B(a ) = 0 and we may assume that
|A(x)|K = |A(a)|K , |B(y)|K = |B(a )|K for every x, y respectively in W, W . Then
Fϕ (y) = 0 if y is not in W and

Fϕ (y) = |A(x)/B(x )|K µp (x ) = |A(a)/B(a )|K µp (W )
W
if y = x is in W . Therefore, Fϕ = Fϕ (a ) χW , hence Fϕ is in D(Y ). Furthermore,

the formula to be proved becomes

|A(a)|K µn (W ) = |A(a)/B(a )|K µp (W ) · |B(a )|K µm (W ),
which is a trivial identity.
Remark. In Theorem 7.6.1 the assumption that α is a gauge form on X is not
really necessary. If we drop that assumption, then Fϕ need not be in D(Y ), but it
is a C-valued continuous function on Y with compact support, and the integration-
formula holds. In fact, it holds for every C-valued continuous function ϕ on X with
compact support. This can easily be proved by using the fact already mentioned
in section 7.2 that such a function is the uniform limit of a sequence in D(X) with
support of each term contained in a fixed compact subset of X.
We shall explain, for our later use, a special case of θy = α/β. We take any
element f (x) of K[x1 , . . . , xn ]\K and denote by Cf the critical set of the K-
analytic map f : K n → K defined by f (x). We assume that Cf = K n , put
X = K n \Cf , Y = K, hence m = 1, and denote the restriction of f to X also
by f . Then the K-analytic map f : X → Y is clearly submersive. Furthermore,
dx1 ∧ · · · ∧ dxn is a gauge form on K n , hence its restriction α to X is a gauge form
on X. If we denote the coordinate on Y by y, then β = dy is a gauge form on Y .
We shall show that if a is any point of X, hence (∂f /∂xi )(a) = 0 for some i, and if
Ui is the open subset of X defined by ∂f /∂xi = 0, then θb |(Xb ∩ Ui ) is given by

(−1)i−1 dx1 ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn (∂f /∂xi )(Xb ∩ Ui )
for b = f (a). In fact, if for every x in Ui we put

γ(x) = (−1)i−1 dx1 ∧ · · · ∧ dxi−1 ∧ dxi+1 ∧ · · · ∧ dxn (∂f /∂xi ),
then γ is a K-analytic differential form of degree p = n − 1 on Ui satisfying α =
f ∗ (β) ∧ γ on Ui . This implies the above assertion.
Chapter 8
Local zeta functions

(p-adic case)
8.1 Selfduality of K and some lemmas
Some arithmetic properties of p-adic fields were proved elementarily before the
thesis of J. Tate [55] of 1950. He applied the fruitful method of Haar integration
uniformly to all completions of an algebraic number field and to its adele groups.
In an unpublished paper K. Iwasawa independently applied the same method to
the idele group. Both authors had the same objective, i.e., to give a transparent
proof to the functional equation of Hecke’s zeta function. In the following we shall
explain some of Tate’s results.
If G is a locally compact abelian group, then the group G∗ = Hom(G, C× 1)
is called the dual group or simply the dual of G. If G → G is a continuous
homomorphism from G to a similar group G , then we have the dual homomorphism
(G )∗ → G∗ . If g, g ∗ are elements of G, G∗ respectively, then g ∗ (g) will sometimes
be denoted by g, g ∗ . We shall denote by K a p-adic field as before and prove the
selfduality of K in the following form:
Proposition 8.1.1 There exists an element ψ of the dual K ∗ of K satisfying

ψ OK = 1, ψ π −1 OK = 1.
Furthermore if for every a in K we define an element ψa of K ∗ as ψa (x) = ψ(ax),

then θ(a) = ψa gives an isomorphism θ : K → K ∗ .
Proof. We shall first prove the existence of ψ. If G is a finite abelian group, then
as we have already explained G, G∗ have the same order. Therefore, if H is a
subgroup of G, then the injective homomorphism G∗ /(G/H)∗ → H ∗ defined by
g ∗ → g ∗ |H is surjective because they have the same order. In particular, if we put
Ge = π −e OK /OK , the homomorphism G∗e+1 → G∗e dual to the inclusion Ge → Ge+1
is surjective for every e in N. Therefore, if χ1 = 1 is given in G∗1 , we can find χe
in G∗e satisfying χe+1 |Ge = χe for every e > 0 in N. If now a is arbitrary in K
and ā is its image in K/OK , then ā is contained in Ge for all large e. We observe
that ψ(a) = χe (ā) is independent of e and defines an element ψ of K ∗ satisfying
ψ|OK = 1 and ψ|π −1 OK = 1.
117
118 JUN-ICHI IGUSA
We shall next show that θ : K → K ∗ , defined as θ(a)(x) = ψ(ax), is an iso-

morphism. If we let K × act on K, K ∗ as a0 · a = a0 a, (a0 · a∗ )(x) = a∗ (a0 x),
then θ becomes equivariant. If θ(a) = 1 for some a in K, then aK is contained in
Ker(ψ), hence a = 0. Therefore, θ is injective. We shall show that θ is surjective,
i.e., every a∗ in K ∗ is in Im(θ). We know by a general remark in Chapter 7.2 that
a∗ is locally constant, hence a∗ |π e OK = 1, i.e., π e · a∗ |OK = 1, for some e in N.
Since Im(θ) is K × -invariant, we may assume from the beginning that a∗ |OK = 1.
Then a∗ |Ge = a∗e is defined for every e in N. On the other hand θ gives rise to an
injective homomorphism OK /π e OK → G∗e , which is surjective because they have
the same order. Therefore, we get a sequence {ae } in OK , where each ae is unique
mod π e , such that θ(ae )|Ge = a∗e for every e in N. Since a∗e+1 |Ge = a∗e by definition,
we have ae+1 ≡ ae mod π e for every e. Therefore {ae } is a Cauchy sequence in OK ,
hence it has a limit a in OK and a∗ |Ge = θ(a)|Ge for all e, and hence a∗ = θ(a).
We might remark that in the special case where K = Qp an element such
as ψ in Proposition 8.1.1 can be explicitly defined. We observe that if a is in
Qp , then the coset a + Zp can be represented by an element a of Z[1/p] unique
mod Z and ep (a) = e(a ) defines an element ep of Q∗p satisfying ep |Zp = 1 and
ep (1/p) = e(1/p), hence ep |p−1 Zp = 1.
×
We observe that U = OK is a totally disconnected compact abelian group.
Therefore, again by a general remark in Chapter 7.2, if χ is any element of Ω(U ) =
U ∗ , then we will have χ|(1 + π e OK ) = 1 for some e > 0 in N. We shall denote by
e(χ) = e(χ−1 ) the smallest e as above. In the following lemma, since dx/|x|K is a
Haar measure on K × , its restriction du to the open subgroup U of K × is a Haar
measure on U .
Lemma 8.1.1 If we put

g(χ) = ψ(π −e(χ) u)χ(u)−1 du,
U
then g(1) = −q −1 and

|g(χ)|2 = q −e(χ)
for every χ = 1 in U ∗ .
Proof. Suppose first that χ = 1, hence e(χ) = 1. Then by expressing U as OK \πOK

and using the orthogonality of characters, we get g(1) = −q −1 . Suppose next that
χ = 1 and, for the sake of simplicity, put e = e(χ). Then, since |g(χ)|2 is the
product of g(χ) and its complex conjugate, we get

|g(χ)|2 = ψ(π −e (u − v))χ(u−1 v) du dv.
U U
If we replace u by uv in the integral by du and change the order of integration, then

we get
−1
RHS = χ(u) ψ(π −e (u − 1)v) dv du.
U U
Therefore, similarly to the previous case, we have

|g(χ)|2 = χ(u)−1 ψ(π −e (u − 1)v) dv du
U OK

−q −1 · χ(u)−1 ψ(π −e+1 (u − 1)v) dv du
U OK

= q −e − q −1 · χ(u)−1 du,
U
in which U = U ∩ (1 + π e−1 OK ). We have only to show that the above integral I

over U is 0. If e = 1, hence U = U , then I = 0 by the orthogonality of characters
of U . If e > 1, hence U = 1 + π e−1 OK , then

−e+1
I=q · χ(1 + π e−1 x)−1 dx.
OK
Since 2(e − 1) ≥ e by e > 1, x → χ(1 + π e−1 x)−1 is a character of OK different

from 1, hence I = 0 by the orthogonality of characters of OK .
Lemma 8.1.2 If e > 0 is in N, then

ψ(π −e u) = (1 − q −1 )−1 · g(χ)χ(u)
e(χ)=e
×
for every u in U = OK .
Proof. Since u → ψ(π −e u) gives a function on U/(1 + π e OK ), by Proposition 7.2.2

and the Plancherel formula we get

ψ(π −e u) = cχ χ(u), 1= |cχ |2 .
e(χ)≤e e(χ)≤e
If e(χ) = e, then we have

g(χ) = ψ(π −e u)χ(u)−1 du = (1 − q −1 )cχ .
U
Since e(χ) > 0, therefore, the formula holds for e = 1. Suppose that e > 1. Then
we have
card{χ; e(χ) = e} = card{U/(1 + π e OK )} − card{U/(1 + π e−1 OK )},
which is (1 − q −1 )2 q e , hence by Lemma 8.1.1 we get

|cχ |2 = (1 − q −1 )−2 · |g(χ)|2 = 1.
e(χ)=e e(χ)=e
Therefore, cχ = 0 for e(χ) < e, hence the formula also holds.

120 JUN-ICHI IGUSA
Corollary 8.1.1 If e is arbitrary in Z and χ is in U ∗ , then



 g(χ) e = e(χ), e > 0



−e −1
ψ(π u)χ(u) du = 1 − q −1 χ = 1, e≤0


U 


0 all other cases.
In the following we shall denote D(K n ) by S(K n ). Actually, there is a general
definition of S(X) for an arbitrary locally compact abelian group X by F. Bruhat
[6], and it specializes to the Schwartz space for X = Rn and to D(K n ) for X = K n .
At any rate we shall denote elements of S(X) for X = K n by Φ, Ψ etc., and put
[x, y] = x1 y1 + . . . + xn yn for x, y in X as in the archimedean case.
Lemma 8.1.3 We define the Fourier transform Fϕ, also denoted by ϕ∗ , of any
integrable function ϕ on X as

(Fϕ)(x) = ϕ(y)ψ([x, y]) dy.
X
Then FΦ = Φ∗ is in S(X) for every Φ in S(X) and further (Φ∗ )∗ (x) = Φ(−x) for
every x in X. In particular, the Fourier transformation F gives a C-linear bijection
from S(X) to itself.
Proof. Since S(X) is the C-span of χA for all A in T (X), we may assume that
Φ = χA . We may further assume that A = a + π e OK
n
for some a in X and e in Z.
We then have
(χA )∗ (x) = q −ne · ψ([a, x])χB (x), B = π −e OK

n
for every x in X. We observe that x → ψ([a, x]) is a locally constant function on

X. Since Supp((χA )∗ ) = B, therefore, (χA )∗ is an element of S(X). Furthermore,
we have ((χA )∗ )∗ (x) = χA (−x) for every x in X.
There is a general theorem stating that the Fourier transform of an integrable
function is continuous. We shall give a proof to this theorem in the special case we
need. If ϕ is an integrable function on X such that Φe = ϕχA for A = π −e OK n
is in
S(X) for every e in N, then

lim ϕ∗ − (Φe )∗ ∞ ≤ lim |ϕ(x)| dx = 0.
e→∞ e→∞ X\A
Therefore ϕ∗ , as a uniform limit of a sequence of continuous functions on X, is itself

continuous on X.
8.2 p-adic zeta function ZΦ (ω)

We shall first make Ω(K × ) explicit. Every element a of K × can be written uniquely,
×
but depending on the choice of π, as a = π e u with e in Z and u in OK . They
are called respectively the order and the angular component of a, abbreviated as
×
ord(a) and ac(a). Therefore K × is bicontinuously isomorphic to Z × OK under a →
× × × ∗ ×
(ord(a), ac(a)), hence Ω(K ) is isomorphic to C × (OK ) as ω → (ω(π), ω|OK ). In
× × × ∗
particular, Ω(K ) becomes the disjoint union of countable copies of C with (OK )
× ∗ ×
as its index set. We observe that (OK ) is the union of finite groups (OK /(1 +
π e OK ))∗ for e = 1, 2, . . . .
We define an element ωs of Ω(K × ) for every s in C as
ωs (a) = |a|sK = q −ord(a)s .
If, for every ω in Ω(K × ), we choose s from C satisfying ω(π) = q −s , then we can
write
ω(a) = ωs (a)χ(ac(a)),
×
in which χ = ω|OK . We keep in mind that the above s is not unique. In fact, the
correspondence s → t = q −s gives a bicontinuous isomorphism

C (2πi/ log q)Z → C× .
We observe that a C-valued function of t is holomorphic on C× if and only if it is

a holomorphic function on the s-plane C. At any rate, although s is not uniquely
determined by ω, its real part Re(s) depends only on ω. If we denote it by σ(ω),
then we will have
|ω(a)| = ωσ(ω) (a)
as in the archimedean case. As in that case, we define an open subset Ωσ (K × ) of

Ω(K × ) by σ(ω) > σ for any σ in R.
Lemma 8.2.1 Take a from K, e from Z, ω from Ω0 (K × ), and N, n > 0 from N,

×
and put t = ω(π), χ = ω|OK . Then


 (1 − q −1 )(q −n tN )e /(1 − q −n tN ) a ∈ π e OK ,



 χN = 1



ω(x) |x|K dx =
N n−1
q −e ω(a)N |a|n−1 a ∈π e OK ,


K
a+π e OK 
 χN U = 1





0 all other cases.
in which U = 1 + π e a−1 OK .
Proof. We denote the LHS by I. Suppose first that a is in π e OK . Then the

integrand is not an element of S(K). If we denote by χk the characteristic function
of π e OK \π k OK for k > e in N and put
Φk (x) = ω(x)N |x|n−1

K χk (x),
122 JUN-ICHI IGUSA
×
then Φk is in S(K). Furthermore, if we put U = OK as before, then

I = lim Φk (x) dx = lim ω(x)N |x|n−1
K dx
k→∞ K k→∞ πj U
e≤j<k

= lim (q −n tN )j · χ(u)N du.
k→∞ U
e≤j<k
If χN = 1, since |q −n tN | < 1 in view of |t| < 1, we get the first expression for I. On
the other hand, if χN = 1, by the orthogonality of characters of U we get I = 0.
Suppose next that a is not in π e OK so that U above becomes a subgroup of U .
Then we simply have
I = ω(a)N |a|nK · χ(u)N du,
U
hence we get the second expression for I if χN |U = 1 and I = 0 otherwise.

We shall make one more preparation. Let ω denote any element of Ω(K × ) and
u(y) a unit of the ring of convergent power series Ky1 , . . . , yn . Then, firstly, we
have
u(y) = u(0) 1 + ci y i
|i|>0
with u(0) in K × and ci in K for all i = 0 in Nn such that the series is convergent on
π k OK
n
for some k in N. Secondly, by making k larger if necessary, we may assume
×
that u(0)−1 u(y) − 1 is contained in π e(χ) OK for all y in π k OK
n
, where χ = ω|OK .
Then we will have
|u(y)|K = |u(0)|K , ω(u(y)) = ω(u(0))
also for all y in π k OK

n
.
Theorem 8.2.1 Assume that char(K) = 0 and let f (x) denote an arbitrary el-
ement of K[x1 , . . . , xn ]\K; take ω, Φ respectively from Ω0 (K × ), S(X), where
X = K n . Then
ω(f )(Φ) = ω(f (x))Φ(x) dx
X\f −1 (0)
defines an S(X) -valued holomorphic function ω(f ) on Ω0 (K × ), and it has a mer-

morphic continuation to the whole Ω(K × ) such that ω(f )(Φ) is a rational function
×
of t = ω(π) for each χ = ω|OK and Φ. Furthermore, if h : Y → X, E = {E}, and
(NE , nE ) for every E in E are as in Theorem 3.2.1, then

(1 − q −nE tNE ) · ω(f )
E∈E
× ∗
becomes holomorphic on the punctured t-plane C× for all χ in (OK ) .
Proof. In the above definition of ω(f )(Φ), since every ω in Ω0 (K × ) has a continuous
extension to K as ω(0) = 0, we could have used X as the domain of integration.
We shall fix χ and Φ, and use Lemma 5.3.1 to prove the first part. Choose any
compact subset C of Ω0 (K × ), put
σ0 = max(σ(ω)), A = Supp(Φ), M = max(1, sup |f (x)|K ),

ω∈C x∈A
and define φC as
φC = Φ∞ M σ0 χA ,
in which χA is the characteristic function of A. Then
|ω(f (x)) Φ(x)| ≤ φC (x)
for every x in X\f −1 (0) and ω in C, and the integral of φC over X\f −1 (0) is finite.
Since ω → ω(f (x)) for every x in X\f −1 (0) is a holomorphic function on Ω(K × ),
hence on Ω0 (K × ), by that lemma ω(f )(Φ) is holomorphic on Ω0 (K × ), and this
implies the first part.
As for the main part, at every point b of Y we can choose a chart (U, φU ) such
that U contains b, φU (y) = (y1 , . . . , yn ) and
N n −1
f ◦h=ε· yj j , h∗ dxk = η · yj j · dyk ,
j∈J 1≤k≤n j∈J 1≤k≤n
in which (Nj , nj ) = (NE , nE ) with J bijective to the set of all E containing b and
ε, η are units of the local ring Ob of Y at b with ε(y), η(y) having expansions similar
to the expansion of u(y) discussed before. Since h is proper and A = Supp(Φ) is
compact open, we see that B = h−1 (A) is in T (Y ). Therefore, we can express B
as a necessarily finite disjoint union of members Bα of T (Y ) such that each Bα is
contained in some U above. Since Φ is locally constant, after subdividing Bα we
may assume that (Φ ◦ h)|Bα = Φ(h(b)) and further that φU (Bα ) = c + π e OK n
for
−1 −1
some c = (c1 , . . . , cn ) in K and e in N. Since h : Y \(f ◦ h) (0) → X\f (0) is
n
K-bianalytic, we then have

ω(f )(Φ) = Φ(h(b)) ω(ε(b)) |η(b)|K · ω(yi )Ni |yi |nKi −1 dyi
α 1≤i≤n ci +π e OK
with the understanding that Ni = 0, ni = 1 for all i not in J. Actually, yi for i in

J is restricted by yi = 0, but it makes no difference, and by Lemma 8.2.1 the RHS
is a rational function of t = ω(π). Furthermore, the denominator of each term is
the product of 1 − q −nj tNj for all j in J multiplied possibly by a power of t, which
is holomorphic on C× .
The above proof shows, as in the archimedean case, that the orders of the poles of
ω(f ) are at most equal to the dimension of the nerve complex N (E) of E increased by
1. Furthermore, the real parts of the poles of ω(f ) on each s-plane C are among the
finite set {−nE /NE ; E ∈ E}. As in the archimedean case, we call ω(f ) the complex
power of f (x) and introduce the local zeta function ZΦ (ω) of f (x) as ω(f )(Φ). In
n
the special case where Φ is the characteristic function of OK , we shall write Z(ω)
124 JUN-ICHI IGUSA
instead of ZΦ (ω) and, changing the notation slightly, we shall write Z(s) instead of
Z(ωs ). If σ(ω), Re(s) > 0, therefore, we have

Z(ω) = ω(f (x)) dx, Z(s) = |f (x)|sK dx.
n
OK n
OK
We shall introduce the Poincaré series P (t) of a polynomial f (x) in

OK [x1 , . . . , xn ]\OK . We observe that if ξ, ξ are in OK
n
, i is in N, and ξ ≡ ξ ,

f (ξ) ≡ 0 mod π , then we also have f (ξ ) ≡ 0 mod π . Therefore, the number
i i
ci of ξ mod π i satisfying f (ξ) ≡ 0 mod π i is well defined. In order to get some

information about ci , we introduce the following power series

P (t) = ci (q −n t)i
i≥0

i
in a complex variable t. Since P (t) clearly has t as its dominant series, it is
convergent for |t| < 1. On the other hand, we can express P (t) by the Z(s) for
the above f (x) as follows. If we denote by f −1 (π i OK ) the preimage of π i OK under
f |OK
n
, then it becomes the union of ξ + π i OK n
for all ξ in OK n
satisfying f (ξ) ≡ 0
i
mod π , hence
µn (f −1 (π i OK )) = ci · µn (π i OKn
) = ci · q −ni ,
and hence
×
µn (f −1 (π i OK )) = µn (f −1 (π i OK \π i+1 OK )) = ci · q −ni − ci+1 · q −n(i+1)
for all i in N. Therefore, if t = ωs (π) = q −s for Re(s) > 0, i.e., |t| < 1, then we get

Z(s) = |f (x)|sK dx = |f (x)|sK dx
n \f −1 (0) ×
OK i≥0 f −1 (π i OK )

= (ci q −ni − ci+1 q −n(i+1) )ti = P (t) − t−1 (P (t) − 1).
i≥0
We have thus obtained the following theorem:

Theorem 8.2.2 Let f (x) denote an arbitrary element of OK [x1 , . . . xn ]\OK and
define its Poincaré series P (t) as

P (t) = card{ξ ∈ OK n
, mod π i ; f (ξ) ≡ 0 mod π i } · (q −n t)i .
i≥0
Then
P (t) = (1 − tZ(s))/(1 − t)
with s and t related as t = q −s for Re(s) > 0. In particular, P (t) is a rational
function of t.
We might mention that the Poincaré series of f (x) was introduced and its ra-
tionality was conjectured in the joint book by S. I. Borewicz and I. R. Šafarevič [5],
which is based on a course given by the second author at the Moscow University.
8.3 Weil’s functions FΦ (i) and FΦ∗ (i∗ )

We shall discuss Weil’s functions FΦ , FΦ∗ in the one-variable case; we shall only con-
sider the p-adic case. We shall start with a review of these functions in the general
form as A. Weil introduced them in [58]. This is just to give a good perspective and,
therefore, no details will be given. If X, G are locally compact abelian groups, dx
is a Haar measure on X, G∗ is the locally compact dual group of G, and f : X → G
is a continuous map, then FΦ∗ is the bounded uniformly continuous function on G∗
defined by
FΦ∗ (g ∗ ) = Φ(x) f (x), g ∗ dx
X
for every Φ in the Schwartz-Bruhat space S(X) of X. In order to proceed further,

Weil introduced the following condition:
Condition (A) |FΦ∗ (g ∗ )| is integrable on G∗ with respect to its Haar measure
∗
dg and the integral is convergent uniformly in Φ if it is restricted to a compact
subset of S(X).
He then showed the existence of a unique family of measures {µg ; g ∈ G}, each
µg supported by the fiber f −1 (g), such that the integral of Φ(x) over X by dx can
be expressed as an integral over f −1 (g) by µg followed by an integration over the
base space G by the dual measure dg of dg ∗ . In particular,

Φ(x)f (x), g ∗ dx = Φ(x) µg (x) g, g ∗ dg.
X G f −1 (g)
The function FΦ (g) on G is then defined as the above integral of Φ(x) over the fiber
f −1 (g) by µg (x), and it is continuous.
We shall now make the following specialization: X = K n for a p-adic field K,
G = K, and f : X → K is the K-analytic map defined by an arbitrary f (x) in
K[x1 , . . . , xn ]\K, and we shall examine FΦ , FΦ∗ with all details. As we have shown
in Proposition 8.1.1, the bicharacter ψ(ab) of K × K puts K into a duality with
itself. Therefore FΦ∗ becomes the function on K defined by

FΦ∗ (i∗ ) = Φ(x)ψ(i∗ f (x)) dx
X
for every Φ in S(X) and i∗ in K, in which dx = µn . We observe that FΦ∗ is bounded.

In fact,
FΦ∗ ∞ ≤ |Φ(x)| dx ≤ Φ∞ · µn (C),
X
in which C = Supp(Φ) is in T (X). Since f (C) is a compact subset of K, it is

contained in π −e OK for some e in N. We shall show that FΦ∗ |(i∗0 +π e OK ) is a constant
function on i∗0 + π e OK for every i∗0 in K. In other words, FΦ∗ is uniformly locally
constant, hence uniformly continuous. The proof is simple. If i∗ is in i∗0 + π e OK ,
then

∗ ∗ ∗ ∗

FΦ (i ) − FΦ (i0 ) = Φ(x) ψ(i∗0 f (x)) ψ((i∗ − i∗0 )f (x)) − 1 dx,
X
126 JUN-ICHI IGUSA
in which ψ((i∗ − i∗0 )f (x)) = 1 for every x in Supp(Φ), hence the integral is 0, i.e.,
FΦ∗ (i∗ ) = FΦ∗ (i∗0 ).
n
In the special case where Φ is the characteristic function of OK , we shall write
∗ ∗
F instead of FΦ . If further the coefficients of f (x) are in OK , then we will have
F ∗ |OK = 1. We shall examine F ∗ (i∗ ) for i∗ in K\OK . If we write i∗ = π −e u with
×
e = −ord(i∗ ) > 0 and u in OK , then

F ∗ (i∗ ) = ψ(i∗ f (x)) dx
n
ξ+π e OK
n,
ξ∈OK mod π e

= q −ne · ψ(i∗ f (ξ)).
n,
ξ∈OK mod πe
In the special case where K = Qp , ψ = ep , and i∗ is in Z[1/p]\Z, we have

F ∗ (p−e u) = p−ne · e(p−e uf (ξ)),
ξ∈Zn , mod pe
in which u is in Z\pZ. Such an expression is called an exponential sum or a gener-

alized Gaussian sum.
We have seen that FΦ∗ is a nice function on K. However, it is not always
integrable over K. Before we give such an example, we observe the following con-
sequence of the orthogonality of characters of OK . If i∗ is in K\OK , then

∗
ψ(i x1 x2 ) dx = ψ(i∗ x1 x2 ) dx1 dx2 = µ1 ((i∗ )−1 OK ) = |i∗ |−1
K .
2
OK OK OK
Therefore, if n = 2m and f (x) = x1 xm+1 + . . . + xm x2m , then
F ∗ (i∗ ) = max(1, |i∗ |K )−m
for every i∗ in K. This implies

|F ∗ (i∗ )| di∗ = 1 + |i∗ |−m ∗
K di = 1 + (1 − q
−1
) q −(m−1)e ,
×
K e>0 π −e OK e>0
and the above series is divergent if m = 1.

We have just seen that Condition (A) is not always satisfied even in the special
case where X = K n and f : X → K is given by f (x) in K[x1 , . . . , xn ]\K. We
shall, therefore, use the classical observation in Weil [58], pp. 12-13 to define FΦ .
In fact, we know, by Theorem 7.6.1 that there is a function Fϕ which has a similar
property as FΦ . We shall make this situation precise.
Let Cf and Vf = f (Cf ) denote the critical set and the set of critical values
of f ; for the sake of simplicity we put S = Vf . We have shown in Theorem 2.5.1
that if char(K) = 0, then S is finite. Furthermore, even if char(K) = 0, if f (x) is
homogeneous and char(K) does not divide deg(f ), then S = {0} unless deg(f ) = 1;
in that case S = ∅. At any rate, if Φ is in S(X), then for every i in K\S we define
µi on f −1 (i) as µα/β in Chapter 7.4, where α is the restriction of dx1 ∧ · · · ∧ dxn to

X\Cf and β = dy or rather di in this case, and put

FΦ (i) = Φ(x) µi (x).
f −1 (i)
Then FΦ becomes a locally constant function on K\S if S is closed in K, hence if

S is finite. In fact, if U is a compact neighborhood in K\S of any i0 in K\S, then
FΦ |U remains the same even if Φ is replaced by ΦχA for A = f −1 (U ). Therefore, we
may assume that Supp(Φ) is contained in f −1 (U ). Then by Theorem 7.6.1 applied
to f −1 (U ), U , and f |f −1 (U ) instead of X, Y , and f we see that FΦ is locally
constant on U , hence on K\S.
Lemma 8.3.1 If A is in T (X), then
lim µn (f −1 (π e OK ) ∩ A) = 0.
e→∞
Proof. This lemma can be proved elementarily by using the Weierstrass preparation
theorem. In the case where char(K) = 0, the case in which we shall be interested,
it can also be proved as follows. If we put
×
ce = µn (f −1 (π e OK ) ∩ A), t = q −s ,
then for Φ = χA and Re(s) > 0 we have

ZΦ (ωs ) = |f (x)|sK dx = c e te
A\f −1 (0) e∈Z
with ce = 0 only for a finitely many e < 0. By Theorem 8.2.1 we know that
ZΦ (ωs ) is a rational function of t with poles possibly at 0 and outside the unit disc.
Therefore by Cauchy-Hadamard’s formula for the radius of convergence of a power
series we get
lim sup |ce |1/e < 1,
e→∞
hence 0 ≤ ce ≤ r for some 0 < r < 1 and for all large e. We have only to observe,
e
finally, that

µn (f −1 (π e OK ) ∩ A) = ci ≤ r i = r e /(1 − r)
i≥e i≥e
for all large e with r /(1 − r) → 0 as e → ∞.

e
Lemma 8.3.2 Suppose that S is finite. Then we have

Φ(x) dx = FΦ (i) di
X K\S
for every Φ in S(X).

128 JUN-ICHI IGUSA
Proof. If we put A = Supp(Φ) and

De = (f −1 (i0 + π e OK ) ∩ A), Ee = (i0 + π e OK ),
i0 ∈S i0 ∈S
then these unions become disjoint for all large e. Furthermore, since A is in T (X),
by applying Lemma 8.3.1 to f − i0 instead of f for every i0 in S, we see that
µn (De ) → 0 as e → ∞. Therefore, we get

Φ(x) dx = lim Φ(x) dx = lim FΦ (i) di = FΦ (i) di.
X e→∞ X\De e→∞ K\Ee K\S
We have used Theorem 7.6.1 to see that the two integrals under the limit signs are
equal and the fact that the second limit exists because the first limit exists.
Theorem 8.3.1 Suppose that S = f (Cf ) is finite and Φ is arbitrary in S(X).

Then we have

FΦ∗ (i∗ ) = FΦ (i) ψ(ii∗ ) di,
K\S

FΦ (i) = lim FΦ∗ (i∗ ) ψ(−ii∗ ) di∗
e→∞ π −e OK
respectively for all i∗ in K and all i in K\S. Furthermore, the integral over π −e OK
becomes independent of e for all large e.
Proof. Since Φ1 (x) = Φ(x)ψ(i∗ f (x)) is in S(X), by applying Lemma 8.3.2 to

Φ1 instead of Φ we get the first formula. In the proof of the second formula we
replace i by i1 . Since i1 is not in S and FΦ is locally constant on K\S, if we take
e sufficiently large, then i1 + π e OK becomes disjoint from S and FΦ becomes a
constant function on i1 + π e OK . Then by using the first formula, Fubini’s theorem,
and the orthogonality of characters of π −e OK we get

FΦ∗ (i∗ )ψ(−i1 i∗ ) di∗ = FΦ (i) ψ((i − i1 )i∗ ) di∗ di
π −e OK K\S π −e OK

= qe · FΦ (i) di = FΦ (i1 ).
i1 +π e OK
We observe that FΦ (i) = 0 if i is not contained in f (Supp(Φ)). In the special

n
case where Φ is the characteristic function of OK , we shall write F instead of FΦ .
If further the coefficients of f (x) are in OK , then we will have F |(K\OK ) = 0. We
shall examine F (i) for i in OK \S. Since F is locally constant on OK \S, we have

F (i) = lim q ·
e
F (i ) di = lim q e · µn (f −1 (i + π e OK ) ∩ OK
n
)
e→∞ i+π e OK e→∞
= lim card{ξ ∈ OK
n
, mod π e ; f (ξ) ≡ i mod π e }/q (n−1)e
e→∞
with the expression under the limit sign independent of e for all large e. Such an
F (i) is called a local singular series.
We shall determine F in the case where f (x) = x1 xm+1 + . . . + xm x2m . We
have seen that F ∗ (i∗ ) = max(1, |i∗ |K )−m for every i∗ in K. Since Cf = {0}, hence
S = {0}, we take i from OK \{0}. Then by using Theorem 8.3.1 we can easily see
that
F (i) = (1 − q −m ) q −(m−1)k .
0≤k≤ord(i)
In particular, if m = 1, then F (i) → ∞ as i → 0. Therefore, the locally constant

function F on K × does not have a continuous extension to K.
8.4 Relation of FΦ (i) and ZΦ (ω)

We shall assume that Vf = f (Cf ) is contained in {0} and establish the relation of
FΦ (i) and ZΦ (ω). As we have remarked, the above condition is satisfied if f (x) is
homogeneous and char(K) does not divide deg(f ). At any rate, if ω is in Ω0 (K × ),
then by Lemma 8.3.2 or rather by its proof we get

ZΦ (ω) = lim Φ(x)ω(f (x)) dx
e→∞ X\f −1 (π e O )
K

= lim FΦ (i)ω(i) di = FΦ (i)ω(i) di.
e→∞ K\π e OK K×
We keep in mind that FΦ is a locally constant function on K × . In order to proceed

further, we need the following well-known lemma.
Lemma 8.4.1 Let m denote a positive integer not divisible by char(K). Then for
any e > ord(m) in N the m-th power map gives a surjection from 1 + π e OK to
1 + mπ e OK .
Proof. we shall use the formula

m
e
(1 + π x) m e
= 1 + mπ x + m−1 π (k−1)e xk .
k
1<k≤m
If e ≥ ord(m) and a is in OK , then it shows that (1 + π e a)m is contained in

1 + mπ e OK . If further e > ord(m) and b is in OK , then it shows that

P (x) = (mπ e )−1 (1 + π e x)m − (1 + mπ e b) + b
is an SRP in x. Therefore Corollary 2.2.1 shows that OK is mapped bijectively to

itself under x → P (x). Hence P (x) = b, i.e., (1 + π e x)m = 1 + mπ e b has a solution
a in OK .
130 JUN-ICHI IGUSA
Theorem 8.4.1 Assume that char(K) = 0 and Cf is contained in f −1 (0). Then

there exists e(Φ) > 0 in N depending on Φ such that ZΦ (ω) = 0 unless e(χ) ≤ e(Φ)
×
for χ = ω|OK . Furthermore, if ω is in Ω0 (K × ), then

ZΦ (ω) = FΦ (i) ω(i) di
K×
and if we put t = ω(π), then

FΦ (i) = ((1 − q −1 )|i|K )−1 · Rest=0 (ZΦ (ω) t−ord(i)−1 ) χ(ac(i))−1
e(χ)≤e(Φ)
for every i in K × , in which Rest=0 means the taking of the residue at t = 0.

Proof. We shall use the same notation as in the proof of Theorem 8.2.1 with
the following modification. First of all, by multiplying a power of π to φU in
(U, φU ), we shall assume that ε(b)−1 ε(y) is in OK [[y1 , . . . , yn ]] for all U . In the
case where i = f (h(b)) = 0, hence J = ∅ and f (h(y)) = ε(y), we shall assume that
y1 = ε(b)−1 ε(y) − 1 in φU (y) = (y1 , . . . , yn ). This is permissible because f −1 (i) is
disjoint from f −1 (0), hence from Cf by assumption, and h : Y → X is K-bianalytic
over X\Cf . We put
m0 = max ord(NE ); E ∈ E
and, by making U smaller, we shall assume that φU (U ) is contained in π m0 +1 OK
n
for
all U . This time, since χ is not fixed, we can only assume that |η(y)|K is constant
on U . We recall that
n −1
ZΦ (ω) = Φ(h(b)) |η(b)|K · ω(ε(y)) ω(yj )Nj |yj |Kj · dy
n
c+π e OK
α j∈J
for some e > m0 , in which c is contained in π m0 +1 Okn . We shall show that
e(Φ) = m0 + e
will then have the required property, i.e., we shall derive a contradiction assuming
that ZΦ (ω) = 0 for e(χ) > e(Φ). At any rate ZΦ (ω) = 0 implies
n −1
I= ω(ε(y)) ω(yj )Nj |yj |Kj · dy = 0
n
c+π e OK j∈J
for some c.
Suppose first that J = ∅. Then ε(y) = ε(b)(1 + y1 ), hence

I = q −(n−1)e ω(ε(b)) · χ(1 + y1 ) dy1 = 0.
c1 +π e OK
×
Since 1 + c1 is in 1 + π m0 +1 OK , hence in OK , this implies

χ(y1 ) dy1 = 0,
1+π e OK

hence χ(1 + π e OK ) = 1, i.e., e(χ) ≤ e contradicting e(χ) > e(Φ) ≥ e.
Suppose next that J = ∅. Then we express c + π e OK n
as a disjoint union of
−1
c +π OK observing that e(χ) > e(Φ) ≥ e. Since ε(b) ε(y) is in OK [[y1 , . . . , yn ]],
e(χ) n
we then have
ω(ε(y))(c + π e(χ) OKn
) = ω(ε(b))χ(ε(b)−1 ε(c ))
for every c . Since I = 0, therefore, we get

n −1
ω(yj )Nj |yj |Kj dyj = 0
cj +π e(χ) OK
for some c and j. If cj is in π e(χ) OK , then χNj = 1 by Lemma 8.2.1, hence χNj = 1
on 1 + π m0 +1 OK . Since ord(Nj ) ≤ m0 , we see by Lemma 8.4.1 that χ = 1 on
1 + π 2m0 +1 OK . This implies e(χ) ≤ 2m0 + 1 contradicting e(χ) > e(Φ) ≥ 2m0 + 1.
If cj is not contained in π e(χ) OK , then m0 + 1 ≤ ord(cj ) < e(χ) and χNj = 1 on
1 + π e(χ) (cj )−1 OK by Lemma 8.2.1, hence χNj = 1 on 1 + π e(χ)−m0 −1 OK . Since
e(χ)−m0 −1 > m0 ≥ ord(Nj ), we see by Lemma 8.4.1 that χ = 1 on 1+π e(χ)−1 OK
contradicting the definition of e(χ). We have thus proved the first part.
As for the second part, since the expression of ZΦ (ω) in terms of FΦ (i) has
already been proved, we shall prove the converse. Since FΦ is a locally constant
function on K × , for every e in Z, not the above e, the function u → FΦ (π e u) is in
×
D(OK ). Therefore, by Proposition 7.2.2 we can write

FΦ (π e u) = ce,χ χ(u)
× ∗
χ∈(OK )
with ce,χ in C. Then for every ω in Ω0 (K × ) we will have

ZΦ (ω) = FΦ (i) ω(i) di = (1 − q −1 ) ce,χ−1 (q −1 t)e .
×
e∈Z π e OK e∈Z
Therefore by the first part, ce,χ = 0 for e(χ) > e(Φ) and

(1 − q −1 )−1 q e · Rest=0 (ZΦ (ω)t−e−1 ) χ(u)−1 = ce,χ−1 χ(u)−1 ,
e(χ)≤e(Φ) e(χ)≤e(Φ)
which is FΦ (π e u). This completes the proof.

We remark that in the finite sum expression of FΦ (i) in Theorem 8.4.1 the
number of terms depends on Φ but not on i. The meaning of the theorem is that
it permits us to translate properties of ZΦ (ω) into the corresponding properties of
Weil’s functions FΦ (i) and FΦ∗ (i∗ ). We might mention that our first paper on local
zeta functions was written to develop such a theory uniformly not only for a p-adic
field K but also for R and C. At any rate, in the p-adic case the rationality of
ZΦ (ω) as stated in Theorem 8.2.1 and supplemented by Theorem 8.4.1 is equivalent
to a certain behavior of FΦ (i) at its “singular point 0” and also to that of FΦ∗ (i∗ ) at
its “singular point ∞”, which can be expressed by certain asymptotic expansions
respectively as |i|K → 0 and |i∗ |K → ∞. We refer the reader to [23] for the details.
132 JUN-ICHI IGUSA
In the following, we shall only use the property of ZΦ (ω) being meromorphic on
Ω(K × ) and examine Condition (A) for a fixed Φ. In doing so, we shall use the
following lemma:
Lemma 8.4.2 Let φ(t) denote a meromorphic function on |t| ≤ r, i.e., on an open
set containing the closed disc, for some r > 0. Then a finite limit
a = lim r e · Rest=0 (φ(t)t−e−1 )
e→∞
exists if and only if φ(t) − b/(1 − r−1 t) is holomorphic on 0 < |t| ≤ r for some b in
C, and in that case a = b.
Proof. We consider the vector space over C of all meromorphic functions φ(t) on
|t| ≤ r and denote the Laurent expansion of each φ(t) at 0 by

φ(t) = c e te , ce = Rest=0 (φ(t)t−e−1 ).
e∈Z
We introduce its subspaces En and Ec respectively with poles at most at 0 and with
r e ce having a finite limit as e → ∞, and put
E = C · 1/(1 − r −1 t) + En .
Then the first part of the lemma can be restated as Ec = E. If φ(t) is in En , then it
is convergent at r, hence {r e ce } is a null sequence. Therefore, if φ(t) − b/(1 − r −1 t)
is in En , then r e ce tends to b as e → ∞. Hence Ec = E also implies the second
part of the lemma. Since we have shown that E is contained in Ec , we have only
to show that Ec is contained in E.
We observe that En , Ec , E are all stable under the multiplication by t, hence
that they are C[t]-modules. If φ(t) is in Ec , then the sequence {r e ce } is bounded,
hence φ(t) is holomorphic on 0 < |t| < r, and hence

φ(t) ≡ cα,m /(1 − α−1 t)m mod En
|α|=r 1≤m≤mα
for some cα,m in C with cα,mα = 0. We shall show that mα ≤ 1 for all α. Suppose
that mα > 1 for some α. Then Ec contains

(1 − α−1 t)mα −2 · (1 − β −1 t)mβ · φ(t) ≡ ce te mod En ,
β=α e∈Z
in which
ce = (cα,mα (e + 1) + cα,mα −1 )α−e
for all e in N. This contradicts the definition of Ec because the sequence {r e ce } is
unbounded. We shall next show that mα = 0 for α = r. Suppose that mα = 1 for
some α = r. Then Ec contains

(1 − β −1 t) · φ(t) ≡ ce te mod En ,
β=α e∈Z
in which ce = cα,1 α−e for all e in N. This again contradicts the definition of Ec
because r e ce has no limit as e → ∞. Therefore φ(t) is in E.
Theorem 8.4.2 Assume that char(K) = 0 and Cf is contained in f −1 (0). Then

the following conditions on FΦ , FΦ∗ , and ZΦ are equivalent:
(1) FΦ (i) has a finite limit FΦ (0) as |i|K → 0;
(2) FΦ∗ is an integrable function on K;
(3) ZΦ (ω) for χ = 1 and (1 − q −1 t)ZΦ (ωs ) are holomorphic on 0 < |t| ≤ q.
Furthermore, in that case
Rest=q (ZΦ (ωs )) = (1 − q)FΦ (0)
and if q is the smallest absolute value of the poles of ZΦ (ω) on |t| > q and ε > 0,
σ
then
|FΦ∗ (i∗ )| ≤ γ(Φ) · max(1, |i∗ |K )−σ+ε
for all i∗ in K, in which γ(Φ) > 0 is independent of i∗ .
Proof. We have seen in section 8.1 that the Fourier transform of an integrable
function is continuous. Therefore (2) implies (1) by Theorem 8.3.1. In the notation
of the proof of the second part of Theorem 8.4.1 we see that (1) implies

−1 −1
lim ce,χ = (1 − q ) · lim FΦ (π e u) χ(u)−1 du = δ1 (χ) FΦ (0),
e→∞ e→∞ ×
OK
in which δ1 (χ) = 1 or 0 according as χ = 1 or χ = 1. Therefore, by applying Lemma

8.4.2 to φ(t) = ZΦ (ω) for r = q we see that
ZΦ (ω) − δ1 (χ)FΦ (0)(1 − q −1 )/(1 − q −1 t)
is holomorphic on 0 < |t| ≤ q. Hence (1) implies (3) and also the residue formula
for ZΦ (ωs ) at t = q. Finally, since the integral of |i∗ |−σ
K
0
over K\OK is convergent
for every σ0 > 1, we have only to show that (3) implies the estimate for |FΦ∗ (i∗ )| as
stated in the theorem. By assumption the RHS power series in
ZΦ (ω) − δ1 (χ)b(1 − q −1 )/(1 − q −1 t)

≡ (1 − q −1 ) (ce,χ−1 − δ1 (χ)b)(q −1 t)e mod C[t−1 ]
e≥0
is holomorphic on |t| < q for some b in C. Therefore, if we replace t in its e-th

σ
term by any r satisfying 0 < r < q σ , then we get a bounded sequence, i.e., we have
ce,χ−1 = δ1 (χ)b + O((qr −1 )e )
as e → ∞. On the other hand, by using Theorem 8.3.1 and Corollary 8.1.1 we get

FΦ∗ (π −e u) = (1 − q −1 ) ck,1 q −k + q −e · ce−e(χ),χ−1 q e(χ) g(χ) χ(u)
k≥e χ
×
for every e in Z and u in OK , in which the summation in χ is finite. Since e(1) = 1
−1
and g(1) = −q , the contribution from the above series and the term for χ = 1
is O(r −e ) while the contribution from each term for χ = 1 is also O(r −e ) both as
e → ∞. If ε > 0, then we can take r = q σ−ε . Therefore, we get
∗
FΦ∗ (i∗ ) = O(r ord(i ) ) = O(|i∗ |−σ+ε
K )
as |i∗ |K → ∞.
134 JUN-ICHI IGUSA
8.5 Poles of ω(f ) for a group invariant f

We shall go back to the complex power ω(f ) in Theorem 8.2.1 and obtain different
kinds of information on the poles of ω(f ) in the special case where f (x) is a relative
invariant. We start with a remark in the general case that if Y = X\f −1 (0), then
ω(f )|Y is holomorphic on the whole Ω(K × ). This can be proved in the same way
as the first part of Theorem 8.2.1.
Suppose that A =Supp(Φ) is contained in Y and put
m0 = min(1, |f (x)|K ), m1 = max(1, |f (x)|K ).

x∈A x∈A
Then 0 < m0 ≤ 1 ≤ m1 < ∞ and m0 ≤ |f (x)|K ≤ m1 for every x in A. Restrict ω

in Ω(K × ) as σ0 ≤ σ(ω) ≤ σ1 , in which σ0 ≤ σ1 and otherwise arbitrary in R. Then
we will have
|ω(f (x))| ≤ M = max(mσ0 0 , mσ1 1 )
for every x in A. Therefore, if we put φ = M Φ∞ χA , then |ω(f (x))Φ(x)| ≤ φ(x)
for every x in X and

φ(x) dx = M Φ∞ µn (A) < ∞.
X
Since ω → ω(f (x)) for every x in Y is a holomorphic function on Ω(K × ), we see

by Lemma 5.3.1 that ω(f )(Φ) is a holomorphic function on Ω(K × ), hence ω(f )|Y
is holomorphic on Ω(K × ).
We shall assume that ω is a variable point of Ω(K × ), i.e., t = ω(π) is a variable
in C× . If is a pole of ω(f ) in Ω(K × ) and

ω(f ) = ck (t − α)k
k∈Z
is its Laurent expansion at α = (π), then all ck are in S(X) with ck = 0 for
k < −n by Theorem 8.2.1 and with Supp(ck ) contained in f −1 (0) for k < 0 by
what we have just shown. We shall denote the order of the pole α by m = mα and
put T = c−m . Then 0 < m ≤ n, T = 0 and Supp(T ) is contained in f −1 (0). We
shall obtain more precise information about T in the case where f (x) is a relative
invariant.
We recall that GLn (K) is a locally compact totally disconnected group. In fact,
the compact open subgroups 1n + π e Mn (OK ) of GLn (K) for all e > 0 in N form a
base at its unit element 1n . Furthermore, if we denote the n × n diagonal matrix
with d1 , . . . , dn as its diagonal entries by diag{d1 , . . . , dn }, then we see by Lemma
7.4.1 that
GLn (K) = ∪ GLn (OK ) diag{π e1 , . . . , π en }GLn (OK ),
in which the union is taken for all increasing sequences (e1 , . . . , en ) in Z. Since all
double cosets above are compact subsets of GLn (K), it is countable at ∞. Conse-
quently, if G is any closed subgroup of GLn (K), then it is also a locally compact
totally disconnected group which is countable at ∞. We observe that any subgroup
of GLn (K) continuously acts on X by matrix-multiplication. We shall assume that

the above f (x) is a relative G-invariant, i.e., that f (gx) = ν(g)f (x) with ν(g) in
K × for every g in G. Then necessarily ν is in Hom(G, K × ). We let G act on S(X)
and S(X) in the usual way, i.e., as
(g · Φ)(x) = Φ(g −1 x), (g · S)(Φ) = S(g −1 · Φ)
respectively for every Φ in S(X) and S in S(X) . Then we have
g · ω(f ) = ρω (g)−1 ω(f ),
in which ρω (g) = ω(ν(g))| det(g)|K for every g in G. In fact, if ω is in Ω0 (K × ),

then by definition
(g · ω(f ))(Φ) = ω(f (x)) Φ(gx) dx.
Y
−1
If we replace x by g x, then by using Lemma 7.4.2, we get
(g · ω(f ))(Φ) = ω(ν(g))−1 | det(g)|−1

K ω(f )(Φ) = ρω (g)
−1
ω(f )(Φ).
This relation is preserved under holomorphic continuation.

If we incorporate the above information in the Laurent expansion of ω(f ) at its
pole , then we get

(g · ck )(t − α)k = ρω (g)−1 · ck (t − α)k
k≥−m k≥−m
× ×
for ω close to , i.e., ω|OK = |OK and t close to α. Since
ω(a) = (α−1 t)ord(a) (a)
for every a in K × , we have
ρω (g) = (1 + α−1 (t − α))ord(ν(g)) ρ (g)
for every g in G. Therefore, if we compare the coefficients of (t − α)−m on both

sides of the above relation, then we get
g · T = ρ (g)−1 T
for every g in G. In the notation of Chapter 7.2-7.3 this means that T = 0 is an

element of EX (ρ ). Therefore, if the number of G-orbits in X is countable, i.e., at
most countable, then we can apply Theorem 7.3.1 to T . It gives the structure of the
eigendistribution T as well as the following information about the pole of ω(f ):
Theorem 8.5.1 Suppose that a closed subgroup G of GLn (K) acts on X = K n
with countably many orbits and f (x) in K[x1 , . . . , xn ]\K is a relative G-invariant,
i.e., f (gx) = ν(g)f (x) with ν(g) in K × for every g in G; let denote any pole of
ω(f ) and
ω(f ) = ck (t − α)k , T = c−m = 0
k≥−m
136 JUN-ICHI IGUSA
its Laurent expansion at α = (π) where t = ω(π). Then Supp(T ) is contained in

f −1 (0). Furthermore, if ξ is a point of any open G-orbit in Supp(T ) and H is the
fixer of ξ in G, then

(ν(g)) = ∆H (g) (∆G (g) | det(g)|K )
for every g in H. In particular, ◦ ν is R×

+ -valued on H.
In the above theorem it is, of course, enough to assume that the number of
G-orbits in f −1 (0) is countable. The theorem can be applied to the case where f (x)
is a basic relative invariant of a regular prehomogeneous vector space. We might
mention that the theorem was proved in that case under the assumption that the
number of G-orbits in f −1 (0) is finite. In fact, our proof in [25] is basically the same
as the one we have just given. We might remark that the countability assumption
on the number of G-orbits is very strong. However, it has been reported by A.
Gyoja [18] that he succeeded in removing that restriction.
Chapter 9
Some homogeneous
polynomials
9.1 Quadratic forms and Witt’s theorem
We shall make ourselves familiar with those homogeneous polynomials for which we
shall later compute Z(s). We shall start with quadratic forms. We take an arbitrary
field K and consider vector spaces over K all assumed to be finite dimensional. If
X is such a vector space, as before we shall denote its dual space by X ∗ and put
[x, x∗ ] = x∗ (x) for every (x, x∗ ) in X × X ∗ . A quadratic form Q on X is a K-valued
function on X satisfying the following conditions:
(Q1) Q(λx) = λ2 Q(x) for every λ in K and x in X.
(Q2) Q(x, y) = Q(x + y) − Q(x) − Q(y) is K-bilinear on X × X.
It follows from the definition that Q(x, y) = Q(y, x), Q(x, x) = 2Q(x), and Q|Y for
any subspace Y of X is a quadratic form on Y . The value of Q on any K-linear
combination λ1 x1 + . . . + λn xn of x1 , . . . , xn in X can be determined by the formula

Q λ i xi = Q(xi )λ2i + Q(xi , xj )λi λj ,
1≤i≤n 1≤i≤n 1≤i<j≤n
which can easily be proved by an induction on n. If S is any subset of X, we denote

by S the K-span of S, i.e., the set of all K-linear combinations of elements of
S, and by S ⊥ the set of all x in X satisfying Q(x, y) = 0 for every y in S. By
definition they are both subspaces of X. Furthermore, we denote by Q−1 (0) the set
of all x in X satisfying Q(x) = 0. If X ⊥ = 0, then Q is called nondegenerate and
if Q−1 (0) = 0, then Q is called anisotropic. We observe that X ⊥ ∩ Q−1 (0) for any
quadratic form Q is a subspace of X. If this subspace is 0, we propose to call Q
reduced for the following reason. The set of all x0 in X satisfying Q(x + x0 ) = Q(x)
for every x in X forms a subspace of X and this subspace is X ⊥ ∩Q−1 (0). Therefore,
Q is reduced if and only if it does not come from a quadratic form Q0 on a factor
space X/X0 by a subspace X0 = 0 as Q(x) = Q0 (x + X0 ). At any rate, if Q is either
nondegenerate or anisotropic, then it is clearly reduced. If char(K) = 2, then in
view of Q(x, x) = 2Q(x) every reduced quadratic form is nondegenerate. We might
remark that if X = Ka, where char(K) = 2 and Q(a) = 0, then Q is anisotropic,
hence reduced, but degenerate, i.e., Q is not nondegenerate.
137
138 JUN-ICHI IGUSA
If Y , Y are subspaces of X such that Q(y, y ) = 0 for every y, y in Y , Y , then

they are called orthogonal. If Y1 , Y2 , . . . are mutually orthogonal subspaces of X
such that X becomes their direct sum, we write
X = Y1 ⊕ Y2 ⊕ · · ·
and call it an orthogonal direct sum. Suppose that Y is any subspace of X. Then
Q|Y is nondegenerate if and only if Y ∩ Y ⊥ = 0 and in that case
(∗) dimK (Y ⊥ ) = dimK (X) − dimK (Y ),
hence X = Y ⊕ Y ⊥ . One way to see (∗) is as follows. We take a K-basis say
{w1 , . . . , wn } for X
such that the first p elements form a K-basis for Y and identify
X with K n as x = λi wi → λ = t (λ1 · · · λn ). We denote by h the p × n matrix
with hij = Q(wi , wj ) as its (i, j)-entry for 1 ≤ i ≤ p, 1 ≤ j ≤ n and by h0 the
p × p submatrix of h composed of its first p columns. Then Q|Y is nondegenerate
if and only if det(h0 ) = 0 and x is in Y ⊥ if and only if hλ = 0. Since rank(h) =
p = dimK (Y ) and n = dimK (X), therefore we get (∗). We observe that, in the case
where Q|Y is nondegenerate, if Q is nondegenerate (resp. reduced), then Q|Y ⊥ is
nondegenerate (resp. reduced). We also remark that if Q|Y and Q|Y ⊥ are both
nondegenerate, then (Y ⊥ )⊥ = Y and the relation of Y and Y ⊥ is symmetric.
We shall prove Witt’s theorem in [61], which is fundamental in the theory of
quadratic forms. We shall start with three lemmas.
Lemma 9.1.1 Let Y , Y denote subspaces of a vector space X both different from
X. Then their union is also different from X, i.e., strictly smaller than X.
Proof. Suppose that X is the union of Y and Y . Then, by replacing Y , Y by
larger subspaces if necessary, we may assume that they are both of codimension 1
in X. We can then write X = Y + Ka for any a in X\Y . Also we can express Y
as the set of all x satisfying [x, a∗ ] = 0 for some a∗ = 0 in X ∗ . Since X is the union
of Y and Y , we see that y + λa for every y in Y and λ = 0 in K is in Y , hence
[y + λa, a∗ ] = [y, a∗ ] + [a, a∗ ]λ = 0.
We can take y = 0, and we get [a, a∗ ] = 0. Then [y + λa, a∗ ] = 0 for every y in Y
and λ in K, hence X is contained in Y . This is a contradiction.
Lemma 9.1.2 Suppose that Q is a reduced quadratic form on X and Q(a) = 0 for
some a = 0 in X. Then Q(a, b) = 1 and Q(b) = 0 for some b in X. Furthermore,
if we put X = a, b ⊥ , then Q|a, b is nondegenerate, hence Q|X is reduced and
X becomes the orthogonal direct sum
X = a, b ⊕ X .
Proof. Since Q is reduced and Q(a) = 0, a = 0, we see that a is not in X ⊥ . There-
fore, the K-linear map x → Q(a, x) from X to K is surjective, hence Q(a, b0 ) = 1
for some b0 in X. If we put b = b0 − Q(b0 )a, then Q(a, b) = 1 and Q(b) = 0.
Futhermore, if λ, µ, λ , µ are in K, then
Q(λa + µb, λ a + µ b) = λµ + µλ ,
hence a, b ∩ a, b ⊥ = 0. Therefore, Q|a, b is nondegenerate. The rest follows

from our previous remark.
If Q, Q are arbitrary quadratic forms on vector spaces X, X respectively, then
a K-linear injection g : X → X satisfying Q (gx) = Q(x) for every x in X is called
an isometry from X to X . If Q is reduced, the group of all isometries from X to
itself is called the orthogonal group of Q, and it will be denoted by O(Q).
Lemma 9.1.3 Let Q denote a nondegenerate quadratic form on X and X = a, b ⊕

X as in Lemma 9.1.2. Then every isometry g from Y = Kb + X to itself uniquely
extends to an element g ∗ of O(Q).
Proof. We shall only assume, until the last moment, that Q is reduced. Since b is in
Y ⊥ and gY = Y , we see that gb is in Y ⊥ . Therefore, if we put c = gb − Q(a, gb)b,
then c is in Y ⊥ and Q(a, c) = 0, hence c is in X ⊥ . Since Q(c) = Q(gb) = 0 and Q
is reduced, we get c = 0, hence gb = Q(a, gb)b with τ0 = Q(a, gb) = 0. If x is in X ,
we can write gx = (lx)b + σx with K-linear maps l : X → K and σ : X → X . If
σc = 0 for some c in X , then gx = 0 for x = −τ0−1 (lc )b + c , hence x = 0, and
hence c = 0. Therefore, σ is an injection, hence a bijection, and the condition of g
being an isometry simply becomes σ in O(Q|X ).
We now take λ0 , µ0 from K and x0 from X , put g ∗ a = λ0 a + µ0 b + x0 , and
require that g ∗ acts on Y as g. Then the condition on the K-linear map g ∗ : X → X
so defined to give an element of O(Q) becomes λ0 = 0 and
(λ0 µ0 + Q(x0 ))λ2 + (λ0 τ0 − 1)λµ + (λ0 (lx) + Q(x0 , σx))λ = 0
for all λ, µ in K and x in X . If we take λ = 1, µ = 0, x = 0, we get λ0 µ0 +Q(x0 ) = 0;

then if we put λ = µ = 1, x = 0, we get λ0 τ0 − 1 = 0; finally if we put λ = 1, we get
λ0 (lx) + Q(x0 , σx) = 0 for every x in X . Since the converse is clear, the condition
becomes λ0 = τ0−1 , µ0 = −τ0 Q(x0 ), and
Q(x0 , x) = −τ0−1 l(σ −1 x)
for every x in X . If we invoke the assumption that Q is nondegenerate, then Q|X

is nondegenerate, hence such an x0 exists and is unique.
Now the Witt theorem is the following generalization of Lemma 9.1.3. The proof
which we shall explain is due to C. Chevalley [7].
Theorem 9.1.1 Let Q denote a nondegenerate quadratic form on a vector space

X over an arbitrary field K and Y any subspace of X. Then every isometry g0 :
Y → X extends to an element g of O(Q).
Proof. The extendability of g0 is trivial if Y = 0 or Y = X. Therefore, we

shall assume that Y = 0, X and apply an induction on dimK (Y ). Let Z denote
any subspace of Y of codimension 1 in Y . Then by induction g0 |Z extends to an
element g1 of O(Q). If (g1 |Y )−1 g0 has an extension g2 , then g1 g2 gives an extension
of g0 . Since (g1 |Y )−1 g0 |Z = idZ , the identity map of Z, we may assume from the
beginning that g0 |Z = idZ . We may further assume that g0 = idY for otherwise
the extendability of g0 becomes trivial. We denote by Σ the set of all subspaces Z1
140 JUN-ICHI IGUSA
of X containing Z such that g0 extends to an isometry g1 : Y + Z1 → X satisfying

g1 |Z1 = idZ1 . The set Σ is not empty because Z is its member. Since g0 = idY ,
Y is not contained in Z1 , hence Z = Y ∩ Z1 , and hence Z1 is of codimension 1 in
Y + Z1 . We shall now assume that Z1 is maximal in Σ. After replacing Y , Z by
Y + Z1 , Z1 , we may then assume that Z is the only member of Σ, and we shall
examine the consequence of this assumption.
We write Y = Z + Ka necessarily with a in Y \Z and put b = g0 a. Then b is
not in Z and b = a because g0 = idY . Furthermore if we put Y = Z + Kb, then
the condition that g0 : Y → Y is an isometry becomes
(Q(a) − Q(b))λ2 + Q(a − b, z)λ = 0
for every λ in K and z in Z. This is equivalent to Q(a) = Q(b) and a−b in Z ⊥ . If now
H = (a−b)⊥ , then H is contained in Y ∪Y . Otherwise H contains c which is not in
Y ∪Y . Then we can easily verify that the prescription g1 : λa+µc+z → λb+µc+z
for every λ, µ in K and z in Z defines an isometry g1 from Y + Kc to Y + Kc
with the property that g1 |Y = g0 and g1 |Z1 = idZ1 for Z1 = Z + Kc, hence
Z1 = Z becomes a member of Σ. This contradicts the assumption. Therefore, H is
contained in Y ∪ Y , hence H becomes the union of H ∩ Y and H ∩ Y . This implies
by Lemma 9.1.1 that H is contained either in Y or in Y . Since the codimension of
H in X is at most 1 for any Q, this implies that either H = Y or H = Y , hence
either Q(a, a − b) = 0 or Q(b, a − b) = 0. Since Q(a) = Q(b), we then get both
Q(a, a − b) = 0 and Q(b, a − b) = 0, hence H = Y = Y . If we put a1 = a − b, then
a1 = 0 and Q(a1 ) = 0. Therefore by Lemma 9.1.2, we will have X = a1 , b1 ⊕ X
for some b1 satisfying Q(a1 , b1 ) = 1 and Q(b1 ) = 0 with X necessarily contained
in Y . Then g0 becomes an isometry from Y = Ka1 + X to itself. Since Q is
nondegenerate by assumption, we see by Lemma 9.1.3 that g0 uniquely extends to
an element of O(Q).
Theorem 9.1.2 If Q is a reduced quadratic form on a vector space X over an
arbitrary field K, then X contains 2p elements a1 , . . . , ap , b1 , . . . , bp satisfying
Q(ai ) = Q(bi ) = 0, Q(ai , bi ) = 1 for 1 ≤ i ≤ p such that it becomes an orthogonal
direct sum
X = a1 , b1 ⊕ · · · ⊕ ap , bp ⊕ X0
with Q|X0 anisotropic. Furthermore, if Q is nondegenerate, then up to an isometry
the anisotropic kernel X0 does not depend on the choice of such a Witt decomposition
of X.
Proof. If Q is reduced but not anisotropic, then by Lemma 9.1.2 we will have
X = a, b ⊕ X , in which Q(a) = Q(b) = 0, Q(a, b) = 1 and Q|X is reduced. If
Q|X is not anisotropic, we can apply Lemma 9.1.2 to Q|X . In that way we get a
decomposition of X as stated. If
X = a1 , b1 ⊕ · · · ⊕ aq , bq ⊕ X0
is a similar decomposition of X, we may assume by symmetry that p ≤ q. Then

the prescription g0 ai = ai , g0 bi = bi for 1 ≤ i ≤ p defines an isometry g0 from
H = a1 , b1 ⊕· · ·⊕ap , bp to H = a1 , b1 ⊕· · ·⊕ap , bp . If now Q is nondegenerate,

then g0 extends by Theorem 9.1.1 to an element g of O(Q), and
gX0 = g(H ⊥ ) = (gH)⊥ = ap+1 , bp+1 ⊕ · · · ⊕ aq , bq ⊕ X0 .
Since Q|X0 is anisotropic, so is Q|gX0 . Therefore q = p and gX0 = X0 .
We call the unique p in Theorem 9.1.2 the Witt index of Q. We observe that if
we put W = a1 , . . . , ap , then Q|W = 0. Such a subspace of X is called totally
isotropic. We shall show that W is maximal, i.e., it is not contained in a strictly
larger totally isotropic subspace. Any element x of X can be written uniquely as

x= (λi ai + µi bi ) + x0
1≤i≤p
with λi , µi in K for all i and x0 in X0 . We observe that W +Kx is totally isotropic if

and only if µi = 0 for all i and Q(x0 ) = 0, hence x0 = 0 because Q|X0 is anisotropic,
and hence x is in W . Furthermore, if W1 , W2 are any two maximal totally isotropic
subspaces of X and dimK (W1 ) ≤ dimK (W2 ), then any K-linear injection from
W1 to W2 is an isometry, hence it extends to an element g of O(Q) by Theorem
9.1.1. Then g −1 W2 is totally isotropic and contains W1 , hence W1 = g −1 W2 by
the maximality of W1 . Therefore O(Q) acts transitively on the set of all maximal
totally isotropic subspaces of X.
If {w1 , . . . , wp } is a K-basis for any totally isotropic subspace W of X, then we
can find another totally isotropic subspace W of X with a K-basis {w1 , . . . , wp }
satisfying Q(wi , wj ) = δij for 1 ≤ i, j ≤ p. This can be proved directly or simply
by embedding W in a maximal totally isotropic subspace of X in any Witt decom-
position of X. We call H = W + W a hyperbolic subspace of X. We observe that
Q|H is nondegenerate.
Remark. In the proof of Theorem 9.1.1 the nondegeneracy assumption on Q is used
only at the last stage where we have applied Lemma 9.1.3. Furthermore in the proof
of Lemma 9.1.3 the nondegeneracy assumption of Q is used only for the existence
of “x0 ” there. The fact is that if we just assume Q to be reduced, such an x0 may
not exist. In fact, Lemma 9.1.3 and Theorem 9.1.1 break down as the following
example shows. Let K denote any field with char(K) = 2, X a three-dimensional
vector space over K with a basis {a, b, c}, and
Q(αa + βb + γc) = αβ + γ 2
for every α, β, γ in K. Then we see that Q is a reduced quadratic form on X. Put
Y = Kb + Kc and define an isometry g : Y → Y as g(βb + γc) = (β + γ)b + γc.
Then there is no g ∗ in O(Q) which extends g. In fact, if such a g ∗ exists, then
g ∗ a = α0 a + β0 b + γ0 c for some α0 , β0 , γ0 in K. Since g ∗ X = X, we have α0 = 0
while α0 = Q(g ∗ a, g ∗ c) = Q(a, c) = 0. This is a contradiction.
9.2 Quadratic forms over finite fields

We shall denote by X a finite dimensional vector space over a field K and by Q a
reduced quadratic form on X. We shall assume that K is a finite field and make
142 JUN-ICHI IGUSA
some computation to prepare for section 9.3. As a side result, we shall show that
the anisotropic kernel X0 in Theorem 9.1.2 is unique up to an isometry. Actually,
although we have not included its proof in this book, the above uniqueness holds
for any K. At any rate, we shall start with some general observations without
assuming that K is finite.
In the case where char(K) = 2, there is a classical diagonalization of an arbi-
trary quadratic form Q on X stating that X has a K-basis {w1 , . . . , wn }, hence
dimK (X) = n, such that

Q xi w i = Q(wi )x2i
1≤i≤n 1≤i≤n
for every x1 , . . . , xn in K. If Q = 0 and {w1 , . . . , wn } is any K-basis for X, then

it trivially has the required property. If Q = 0, then Q(w1 ) = 0 for some w1 in
X. We observe that Q|Kw1 is then nondegenerate, hence X = Kw1 ⊕ (Kw1 )⊥ .
Therefore we have only to apply an induction on the dimension to Q|(Kw1 )⊥ to
find the remaining w2 , . . . , wn .
Suppose that K is arbitrary for a moment, take a K-basis {w1 , . . . , wn } for
X, and define a symmetric matrix h with hij = Q(wi , wj ) as its (i, j)-entry for
1 ≤ i, j ≤ n. Then Q is nondegenerate if and only if det(h) = 0. In that case
d(Q) = (−1)n(n−1)/2 det(h)
is called the discriminant of Q. If we use another K-basis for X, then h will be

replaced by ght g for some g in GLn (K), hence d(Q)(K × )2 is a well-defined element
of K × /(K × )2 . Furthermore, if X, X equipped with nondegenerate quadratic forms
Q, Q are isometric, then d(Q)(K × )2 = d(Q )(K × )2 and if X = H ⊕ H ⊥ for any
hyperbolic subspace H, then

d(Q)(K × )2 = d(QH ⊥ )(K × )2 .
If now K is a finite field with char(K) = 2, i.e., if K = Fq for an odd q, then

K × /(K × )2 is isomorphic to {±1}. If we denote by χ the character of K × which
gives rise to this isomorphism, then χ(d(Q)) depends only on the isometry class of
Q and, in fact, on that of its anisotropic kernel.
We recall that if K = Fq , where q is arbitrary, then K has L = Fqe as its unique,
hence normal, extension of degree e for every e > 0 in N and that the Galois group
of L over K is the cyclic group generated by x → xq . We shall denote the norm
homomorphism L× → K × by N = NL/K and show that N is surjective. Since
2
N (x) is the product of x, xq , xq , . . . , we have
N (x) = xE , E = (q e − 1)/(q − 1),
hence card(Ker(N )) ≤ E. Since card(Im(N )) ≤ card(K × ) = q − 1 and card(L× ) =

q e − 1, we have equalities at both places. In particular, N is surjective. We shall
restrict our attention to the case where e = 2. We observe that if we put N (0) = 0,
then Q = N gives a nondegenerate quadratic form on the vector space L over K.
In fact, if {1, ξ} is a K-basis for L, i.e., if ξ is an element of L\K, and if we express

any x in L as x = x1 + x2 ξ with x1 , x2 in K, then
Q(x) = N (x1 + x2 ξ) = x21 + ax1 x2 + bx22 ,
in which a = ξ + ξ q , b = ξ 1+q , and Q(x, y) = 2x1 y1 + a(x1 y2 + x2 y1 ) + 2bx2 y2 for

a similar element y = y1 + y2 ξ of L. Therefore, if Q(x, y) = 0 for every y in L,
then ax1 + 2bx2 = 2x1 + ax2 = 0. Since the determinant of the coefficient-matrix
is a2 − 4b = (ξ − ξ q )2 = 0, we get x = 0. In the following, L = Fq2 will always be
equipped with the quadratic form N .
Lemma 9.2.1 Let X denote a two-dimensional vector space over K = Fq and Q
any anisotropic quadratic form on X. Then X is isometric to L = Fq2 , hence Q is
nondegenerate.
Proof. If {w1 , w2 } is any K-basis for X, hence Q(w1 ) = 0, and
Q(x1 w1 + x2 w2 ) = Q(w1 )(x21 + ax1 x2 + bx22 )
for x1 , x2 in K, then a zero ξ of t2 − at + b is not in K. Otherwise, x = ξw1 − w2

is in X, Q(x) = 0, and yet x = 0, which is a contradiction. Therefore, L = K(ξ) =
K + Kξ. We observe that N (x1 + x2 ξ) = x21 + ax1 x2 + bx22 . As we have remarked,
we can write Q(w1 ) = N (α) with some α in L× . If we put
θ(x1 w1 + x2 w2 ) = α(x1 + x2 ξ),
since the multiplication by α is a K-linear bijection from L to itself, we see that θ

gives a K-linear bijection from X to L satisfying N (θ(x1 w1 + x2 w2 )) = Q(x1 w1 +
x2 w2 ) for every x1 , x2 in K. We have already remarked that N is nondegenerate.
Therefore Q is also nondegenerate.
Lemma 9.2.2 If there exists an anisotropic quadratic Q on a vector space X over
K = Fq , then necessarily dimK (X) ≤ 2.
Proof. Suppose that dimK (X) > 2 and take any two-dimensional subspace Y of
X. Then Q|Y is anisotropic, hence nondegenerate by Lemma 9.2.1, and hence
X = Y ⊕ Y ⊥ . Since Y ⊥ = 0, it contains y = 0, and then Q(y) = 0. Since
Q|Y is isometric to L = Fq2 and N is surjective, we can find x in Y satisfying
Q(x) = −Q(y). Then Q(x + y) = 0 and x + y = 0, which is a contradiction.
We are ready to prove the following theorem, in which Q−1 (i) for every i in K
denotes the set of all x in X satisfying Q(x) = i:
Theorem 9.2.1 Let Q denote a reduced quadratic form on a vector space X over
K = Fq , i.e., a quadratic form Q on X satisfying X ⊥ ∩ Q−1 (0) = 0. If dimK (X) =
2m, the anisotropic kernel of Q is either 0 or L = Fq2 , and if we respectively put
χ(Q) = ±1, then
 2m−1
 q + χ(Q) (q m − q m−1 ) i = 0
−1
card(Q (i)) =
 2m−1
q − χ(Q) q m−1 i = 0.
144 JUN-ICHI IGUSA
Furthermore, if q is odd, then χ(Q) = χ(d(Q)). If dimK (X) = 2m + 1, the

anisotropic kernel of Q is one-dimensional, hence of the form Kw. If q is odd,
there are also two cases separated by χ(2d(Q)) = χ(Q(w)), and
 2m
 q i=0
card(Q−1 (i)) =
 2m
q + χ(2d(Q)i)q m i = 0,
but if q is even, there is only one case and card(Q−1 (i)) = q 2m for every i.
Proof. If dimK (X) = 2m, then by Lemmas 9.2.1, 9.2.2 the anisotropic kernel has the
two possibilities stated in the theorem. If we can prove the formula for card(Q−1 (0)),
then it will show that the two cases are independent of the decomposition of X in
Theorem 9.1.2. Furthermore, if L = K(ξ) and ξ 2 − aξ + b = 0 as before, then with
respect to the K-basis {1, ξ} for L we have d(N ) = a2 − 4b, which is (ξ − ξ q )2 .
Therefore if q is odd, then d(N ) is not in (K × )2 , hence χ(d(N )) = −1, and hence
χ(Q) = χ(d(Q)). On the other hand, if dimK (X) = 2m + 1, then by Lemma 9.2.2
the anisotropic kernel is of the form Kw for some w in X and Q(w)(K × )2 clearly
determines its isometry class. Since χ(2d(Q)) = χ(Q(w)) if q is odd, what remains
to be proved is only the formula for card(Q−1 (i)).
First of all, in the notation of Theorem 9.1.2 we have

Q (xi ai + xi bi ) + x0 = t x x + Q(x0 )
1≤i≤p
for every x = t (x1 . . . xp ), x = t (x1 . . . xp ) in K p and x0 in X0 . If now Q(x) =
t
x x for x , x in K m and i is in K × , then Q(x) = i implies x = 0. The number
of such x is q m − 1 and for each x = 0 the number of x satisfying t x x − i = 0
is q m−1 , hence
card(Q−1 (i)) = (q m − 1)q m−1 = q 2m−1 − q m−1 .
If i = 0, we have only to add the number of x = 0 and x free in K m , hence
card(Q−1 (0)) = q 2m−1 + q m − q m−1 .
If Q(x) = t x x + N (ξ) for x , x in K m−1 and ξ in L, then card(Q−1 (0)) is the
sum of the numbers of solutions of N (ξ) = −t x x = 0 and = i in K × . Since
card(N −1 (0)) = 1 and card(N −1 (i)) = q + 1 for i in K × , we therefore get
card(Q−1 (0)) = (q 2m−3 + q m−1 − q m−2 ) + (q − 1)(q + 1)(q 2m−3 − q m−2 )

= q 2m−1 − q m + q m−1 .
As for card(Q−1 (i)) for i in K × , if N (α) = i, then Q−1 (1) is mapped bijectively to
Q−1 (i) under (x , x , ξ) → (ix , x , αξ). Therefore, card(Q−1 (i)) is independent of
i, hence (q − 1)card(Q−1 (i)) + card(Q−1 (0)) = q 2m , and hence
card(Q−1 (i)) = q 2m−1 + q m−1 .

In the above argument we have tacitly assumed that m > 1, but the formulas are
valid also for m = 1.
If Q(x) = t x x + ax20 for x , x in K m , x0 in K, and a is in K × , we add the
numbers of solutions in x , x of t x x = −ax20 for all x0 in K. In that way we get
card(Q−1 (0)) = q(q 2m−1 − q m−1 ) + q m = q 2m .
As for card(Q−1 (i)) for i in K × , we shall first assume that q is odd. We introduce
a quadratic form Q1 (x, x1 ) = t x x + ax20 − ix21 , in which x1 is a new variable, and
compute card(Q−1 1 (0)) directly and also as the sum of the numbers of solutions in
x of Q1 (x, x1 ) = 0 for all x1 in K. Since χ(d(Q1 )) = χ(ai), we then get
card(Q−1
1 (0)) = q
2m+1
+ χ(ai)(q m+1 − q m ) = q 2m + (q − 1)card(Q−1 (i)).
This implies
card(Q−1 (i)) = q 2m + χ(ai)q m ,
in which χ(ai) = χ(2d(Q)i). We shall next assume that q is even. Then the square
map gives an automorphism of K. Therefore, if we write a = b2 , i = j 2 with b in
K × , j in K, then −ax20 +i = (bx0 +j)2 . Since x0 → bx0 +j gives a bijection from K
to itself, if we put Q0 (x) = t x x +x20 , then card(Q−1 (i)) =card(Q−1
0 (0)). Therefore,
we get card(Q−1 (i)) = q 2m by the above result or from q·card(Q−1 (i)) = q 2m+1 for
all i in K.
9.3 Classical groups over finite fields

We shall define classical groups over an arbitrary field K and compute their orders
in the case where K is a finite field, i.e., K = Fq . This topic goes back to L. E.
Dickson [12]. We shall use the fact that if a finite group G acts transitively on a set
S and if H is the fixer in G of any point ξ of S, then the bijection G/H → S defined
by g → gξ implies card(G) = card(S)card(H). We shall also use the notation
[i] = 1 − q −i , [i]+ = 1 + q −i
for every i in N.
We observe that G = GLn (K) acts transitively on S = K n \{0} by matrix-
multiplication. If we express any element g of GLn (K) by its 1 × 1, 1 × (n − 1),
(n − 1) × 1, (n − 1) × (n − 1) entry matrices g11 , g12 , g21 , g22 , then the fixer H
of e1 = t (1 0 . . . 0) in G is defined by g11 = 1, g21 = 0 necessarily with g22 in
GLn−1 (K). Therefore, if K = Fq , since card(S) = q n [n], we get
card(GLn (K)) = q n [n] · q n−1 card(GLn−1 (K)).
We have tacitly assumed that n > 1, but if n = 1, then GL1 (K) = K × , hence
card(GL1 (K)) = q[1]. Therefore, by an induction on n we get
2
card(GLn (K)) = q n · [i].
1≤i≤n
146 JUN-ICHI IGUSA
Furthermore, g → det(g) gives a homomorphism from GLn (K) to K × and the

kernel is denoted by SLn (K). If d = diag{τ, 1, . . . , 1}, i.e., a diagonal matrix with
τ, 1, . . . , 1 as its diagonal entries, then det(d) = τ for every τ in K × . Therefore,
the above homomorphism is surjective, hence card(SLn (K)) =card(GLn (K))/q[1],
and hence
card(SLn (K)) = q n −1 ·
2
[i].
1<i≤n
We next take a reduced quadratic form Q on a vector space X over K and define
the orthogonal group O(Q) of Q as in section 9.1. We observe that if g : X → X
is any K-linear transformation satisfying Q(gx) = Q(x) for every x in X, then g
is necessarily an injection, hence a bijection. In fact, if gx = 0 for some x in X,
then Q(x, y) = Q(gx, gy) = 0 for every y in X and Q(x) = Q(gx) = 0, hence x is
in X ⊥ ∩ Q−1 (0), and hence x = 0 because Q is reduced. Therefore O(Q) consists
of all such g.
After this remark, we shall consider the special case where X = K 2m and

Q(x) = x2i−1 x2i = x1 x2 + x3 x4 + . . . ,
1≤i≤m
hence Q is nondegenerate. We shall write O2m (K) instead of O(Q). If we take

g = (w1 w2 . . . w2m ) from M2m (K), then the condition for g to be in G = O2m (K),
i.e., Q(gx) = Q(x) for every x in X = K 2m , becomes Q(wi ) = 0 for every i,
Q(w2i−1 , w2i ) = 1 for 1 ≤ i ≤ m, and Q(wi , wj ) = 0 for all other i < j. If we
denote by S the set of all (x, y) in X × X satisfying Q(x) = Q(y) = 0, Q(x, y) = 1,
then by Theorems 9.1.1, 9.1.2 we see that G acts transitively on S and the fixer H
in G of (e1 , e2 ) in S consists of all g above with w1 = e1 , w2 = e2 . In fact, if we
denote the 2 × 2, 2 × (2m − 2), (2m − 2) × 2, (2m − 2) × (2m − 2) entry matrices
of g by g11 , g12 , g21 , g22 , then g is in H if and only if g11 = 12 , g12 = 0, g21 = 0,
and g22 is in O2m−2 (K). Furthermore, x for (x, y) in S is free in Q−1 (0)\{0} and
(e1 , y) is in S if and only if y2 = 1, y1 + y3 y4 + . . . + y2m−1 y2m = 0. Therefore, if
K = Fq , then by using Theorem 9.2.1 we get
card(S) = q 2m−2 (card(Q−1 (0)) − 1) = q 4m−3 [m][m − 1]+ ,
hence
card(O2m (K)) = q 4m−3 [m][m − 1]+ · card(O2m−2 (K))
for m > 1 and card(O2 (K)) = 2q[1] for m = 1. This implies

card(O2m (K)) = 2q m(2m−1) [m] · [2i].
1≤i<m
Still in the case where X = K 2m with K arbitrary for a moment, if there exists
a separable quadratic extension L of K generated by a zero ξ of t2 − at + b for a, b
in K, we put

Q(x) = x2i−1 x2i + (x22m−1 + ax2m−1 x2m + bx22m )
1≤i<m
and write O2m (K) instead of O(Q). If we take g = (w1 w2 . . . w2m ) from M2m (K),
then the condition for g to be in G = O2m (K) becomes Q(wi ) = 0 except for
Q(w2m−1 ) = 1, Q(w2m ) = b, Q(w2i−1 , w2i ) = 1 except for Q(w2m−1 , w2m ) = a,
and Q(wi , wj ) = 0 for all other i < j. If K = Fq , hence L = Fq2 , and if m > 1,
then exactly in the same way as in the previous case but by using the formula in
Theorem 9.2.1 for card(Q−1 (0)) for the above Q this time we get
card(O2m (K)) = card(S) card(O2m−2 (K)), card(S) = q 4m−3 [m]+ [m − 1].
In the case where m = 1, if we put w1 = t (α, γ), w2 = t (β, δ), and N = NL/K , then
the above condition becomes N (α+γξ) = 1, N (β +δξ) = b, and N (α+γξ, β +δξ) =
a. Therefore, η = (β+δξ)(α+γξ)−1 is also a zero of t2 −at+b, hence η = ξ or η = ξ q .
The subgroup of O2 (K) defined by η = ξ is isomorphic to Ker(N ) as g → α + γξ
and O2 (K) has a coset by that subgroup represented by g with α = 1, β = a,
γ = 0, δ = −1. In particular, card(O2 (K)) = 2q[1]+ . Therefore, we get

card(O2m (K)) = 2q m(2m−1) [m]+ · [2i].
1≤i<m
In the case where X = K 2m+1 , since the anisotropic kernel is one-dimensional

over a finite field, we may assume that

Q(x) = x2i−1 x2i + ax20
1≤i≤m
for some a in K × . Since O(Q) does not change even if we multiply any element of
K × to Q, we may further assume that a = 1. We shall write O2m+1 (K) instead of
O(Q). Then if K = Fq , by the same argument as before, we get
card(O2m+1 (K)) = card(S)card(O2m−1 (K)), card(S) = q 4m−1 [2m]
with card(O1 (K)) = 1 or 2 according as q is even or odd. This implies

card(O2m+1 (K)) = 2q m(2m+1) · [2i]
1≤i≤m
with the factor 2 above replaced by 1 if q is even.

There is one more type of classical group. A K-valued K-bilinear function A on
X × X satisfying A(x, x) = 0 for every x in X is called an alternating form on X.
We in fact have
A(x, y) + A(y, x) = A(x + y, x + y) − A(x, x) − A(y, y) = 0
for every x, y in X. If Y is a subspace of X, then clearly A|(Y × Y ) is an alter-

nating form on Y . If S is any subset of X, we define S ⊥ as the subspace of X
of all x satisfying A(x, y) = 0 for every y in S. The alternating form A is called
nondegenerate if X ⊥ = 0. If we denote by J1 the 2 × 2 matrix with 0, 1, −1, 0 as its
entries and by Jm the 2m × 2m matrix with J1 on the diagonal and with 0 as all
148 JUN-ICHI IGUSA
other 2 × 2 entry matrices, then A(x, y) = t xJm y defines a nondegenerate alternat-

ing form on X = K 2m . We shall show that if A is any nondegenerate alternating
form on a vector space X over K, then X has a K-basis {w1 , w2 , . . . , w2m }, hence
dimK (X) = 2m, satisfying A(w2i−1 , w2i ) = 1 for 1 ≤ i ≤ m and A(wi , wj ) = 0 for
all other i < j. That will imply

A xi w i , yi wi = t xJm y
1≤i≤2m 1≤i≤2m
for every x, y in K 2m . We take w1 arbitrarily from X\{0}. The nondegeneracy of A

implies that the K-linear map from X to K defined by y → A(w1 , y) is surjective,
hence A(w1 , w2 ) = 1 for some w2 in X. If we put X = w1 , w2 ⊥ , then X becomes
the orthogonal direct sum of w1 , w2 and X . In fact, every x in X can be written
uniquely as
x = A(x, w2 )w1 − A(x, w1 )w2 + x
with x in X . We observe that if X = 0, then A = A|(X × X ) is nondegenerate,
hence we can apply the same argument to A . We have only to repeat this process.
We introduce, for our later use, the subspace Altn (K) of Mn (K) consisting of all
a such that A(x, y) = t xay becomes an alternating form on K n . If we denote the
(i, j)-entry of a by aij , then this means aii = 0 for 1 ≤ i ≤ n and aij + aji = 0 for
1 ≤ i < j ≤ n. Similarly we denote by Symn (K) the subspace of Mn (K) consisting
of all a such that t a = a, i.e., aij = aji for all i, j.
We now take a nondegenerate alternating form A on X and define the symplectic
group Sp(A) of A as the group of all invertible K-linear maps g from X to itself
satisfying A(gx, gy) = A(x, y) for every x, y in K. We remark, similarly to the
case of a reduced quadratic form, that the invertibility of g is a consequence of
A(gx, gy) = A(x, y) for every x, y in X. At any rate if A(x, y) = t xJm y for X =
K 2m , we shall write Sp2m (K) instead of Sp(A); it consists of all g in M2n (K)
satisfying t gJm g = Jm . By what we have shown Sp(A) is isomorphic to Sp2m (K) if
dimK (X) = 2m. Furthermore, if we denote by S the set of all (x, y) in K 2m × K 2m
satisfying A(x, y) = t xJm y = 1, then the above observation shows that (x, y) can be
considered as the first two columns of an element of Sp2m (K), hence G = Sp2m (K)
acts transitively on S. Furthermore, the fixer H in G of (e1 , e2 ) in S is isomorphic
to Sp2m−2 (K). Since x for (x, y) in S is arbitrary in K 2m \{0} and (e1 , y) is in S if
and only if y2 = 1, if K = Fq , then we get
card(Sp2m (K)) = card(S)card(Sp2m−2 (K)), card(S) = q 4m−1 [2m],
hence
card(Sp2m (K)) = q m(2m+1) · [2i].
1≤i≤m
The above computation shows that if q is even in K = Fq , then Sp2m (K) and
O2m+1 (K) have the same order, hence there exists a bijection from Sp2m (K) to
O2m+1 (K). Actually, they are isomorphic for any perfect field K with char(K) = 2.
The proof of this remarkable fact is simple and it is as follows. We change our
notation and put
Q(x) = x20 + t x x
for x = t (x1 . . . xm ), x = t (xm+1 . . . x2m ). Then O2m+1 (K) is isomorphic to

O(Q). If for similarly defined y , y we put
A(x, y) = t x y − t x y ,
then Sp2m (K) is isomorphic to Sp(A). If we denote the four m × m entry matrices
of any g in M2m (K) by a, b, c, d, then g is in Sp(A) if and only if
t
ac = t ca, t
bd = t db, t
ad − t cb = 1m .
The field K is arbitrary so far. If now K is a perfect field with char(K) = 2, we
define two row vectors u = (u1 . . . um ), v = (v1 . . . vm ) by the condition that u2i ,
vi2 are respectively the i-th diagonal entries of t ac, t bd for 1 ≤ i ≤ m. Then we can
easily verify that the correspondence
 
  1 u v
a b  
 
  → 0 a b 
 
c d  
0 c d
gives an isomorphism from Sp(A) to O(Q).
Finally, there is a twisted form of GLn (K) for K = Fq , i.e. the unitary group
Un (L) where L = Fq2 . If we denote the automorphism x → xq of L over K applied
to all entries of g by g → g , then Un (L) consists of all g in Mn (L) satisfying
t
gg = 1n . We can compute the order of Un (L) similarly as in the other cases and
we get
(1 − (−q)−i ).
2
card(Un (L)) = q n ·
1≤i≤n
We shall not use this fact in our later examples.
9.4 Composition and Jordan algebras

We shall explain some K-algebras. The explanation will be self-contained except
for the fact that we do not prove classification theorems. This does not create any
gap because our purpose is simply to give a good perspective to the list of f (x) for
which we shall compute Z(s).
We recall that a K-algebra A for any field K is a vector space over K in which
a K-bilinear multiplication (a, b) → ab is defined; we shall not assume that (ab)c =
a(bc). A K-algebra A = 0 is called simple if A and 0 are the only two-sided ideals of
A. We shall assume that dimK (A) is finite. We say that C is a composition algebra
over K if firstly, C is a K-algebra with the unit element 1 = 0, i.e., C = 0; secondly,
C is equipped with a reduced quadratic form n; and thirdly,
(1) n(ab) = n(a)n(b)
for every a, b in C. We call C a quaternion (resp. an octonion) algebra over K if
dimK (C) = 4 (resp. dimK (C) = 8). We observe that (1) implies n(1) = n(1)2 ,
150 JUN-ICHI IGUSA
hence n(1) = 1 for otherwise n(1) = 0. But then n(a) = n(a1) = n(a)n(1) = 0,
hence n(a, b) = 0 for all a, b in C, and hence C = C ⊥ ∩ n−1 (0) = 0, a contradiction.
We polarize (1) in b, i.e., we replace b by b1 + b2 , then by b1 , b2 and subtract the
second from the first. In that way we get
n(ab1 , ab2 ) = n(a)n(b1 , b2 ).
If we apply a polarization to the above in a, we get
n(a1 b1 , a2 b2 ) + n(a2 b1 , a1 b2 ) = n(a1 , a2 )n(b1 , b2 )
for all a1 , a2 , b1 , b2 in C. If we put
t(a) = n(a, 1), a = −a + t(a)1,
then 1 = 1 and n(a ) = n(a), hence by polarization n(a , b ) = n(a, b). Furthermore,
by replacing a1 , a2 , b1 , b2 respectively by a, 1, b, c and also by a, c, b, 1 we get
(2) n(ab, c) = n(b, a c) = n(a, cb ).
We rewrite the first polarized form of (1) as n(ab, ac) = n(a)n(b, c). Since n(ab, ac) =
n(a (ab), c) by (2), if we put d = a (ab)−n(a)b, then n(d, c) = 0 for all c in C. By us-
ing (1), (2), and n(a ) = n(a) we also have n(d) = 0. Hence d is in C ⊥ ∩ n−1 (0) = 0,
i.e.,
(3) a (ab) = n(a)b.
If we put b = c = 1 in (2), we get t(a ) = t(a), hence (a ) = a. If we put b = 1 in

(3), we get
a2 − t(a)a + n(a)1 = 0,
hence by polarization ab + ba − t(a)b − t(b)a + n(a, b)1 = 0. On the other hand by
(2) we get n(a, b) = t(ab ) = −t(ab) + t(a)t(b). By putting these together we get
b a − (ab) = 0. Therefore, the K-linear map a → a is an involution of C in the
sense that (ab) = b a and (a ) = a for every a, b in C. For our later use, we also
remark that
t((ab)c) = t(a(bc)),
hence we can simply write it as t(abc). In fact, we have
t((ab)c) = n(ab, c ) = n(b a , c) = n(a , bc) = t(a(bc));
we have used (2), n(a , b ) = n(a, b), and (2) again. We similarly have
t(ab) = t(ba).
In fact, t(ab) = n(a, 1b ) = n(a, b 1) = t(ba). Therefore, we get
t(abc) = t(bca) = t(cab).

We now take a composition algebra C0 equipped with a nondegenerate quadratic

form n0 , put C = C0 + C0 , a direct sum, n(a1 , a2 ) = n0 (a1 ) + n0 (a2 ), and define a
multiplication in C as
(4) (a1 , a2 )(b1 , b2 ) = (a1 b1 − b2 a2 , b2 a1 + a2 b1 ).
Then by using (2) in C0 , we get

n(ab) − n(a)n(b) = n0 (b2 a1 )b1 − b2 (a1 b1 ), a2 ,
in which a = (a1 , a2 ), b = (b1 , b2 ). Therefore, C becomes a composition algebra

if and only if C0 is associative. We shall show in the case where C0 is associative
that C itself is associative if and only if C0 is commutative. Since the if-part
is straightforward, we shall prove the only-if part. If we take a = (a1 , 0), b =
(b1 , 0), c = (1, 1), where a1 , b1 are arbitrary in C0 , then
(ab)c − a(bc) = (a1 b1 , a1 b1 ) − (a1 b1 , b1 a1 ),
hence (ab)c = a(bc) implies a1 b1 = b1 a1 .

We have explained the above partly to give some idea about C but mainly to
simplify the subsequent verifications. Examples of C are as follows:
(C1) C = K, n(a) = a2 .
(C2) C = K + K, (a1 , a2 )(b1 , b2 ) = (a1 b1 , a2 b2 ), n(a1 , a2 ) = a1 a2 .
(C3) C = M2 (K), with the matrix-multiplication, n(a) = det(a).
(C4) C = C0 + C0 , as in (4) for C0 = C in (C3).
The fact that we have a composition algebra in (C1), (C2), (C3) is clear. There-
fore, by the above general observation we also have a composition algebra in (C4).
Furthermore, (C3) is not commutative, hence (C4) is not associative, and hence
the process C0 → C applied to (C4) will not produce a composition algebra. We
also remark that the involution in each case can easily be made explicit and it is as
follows:
a = a, (a1 , a2 ) = (a2 , a1 ), a = J1 t aJ1−1 , (a1 , a2 ) = (a1 , −a2 ),
in which J1 is as in section 9.3, i.e., the 2 × 2 matrix with 0, 1, −1, 0 as its

entries. The fact is that all composition algebras over K can be determined and if
K is algebraically closed, then C is isomorphic to one of the above. We refer to N.
Jacobson [32] for the details (in the case where char(K) = 2).
We shall introduce another class of algebras. We say that A is a Jordan algebra
over K if A is a commutative K-algebra and
(ab)a2 = a(ba2 )
for every a, b in A. If we denote by θ(a) the multiplication by a in A, then the

above Jordan identity becomes θ(a2 )θ(a) = θ(a)θ(a2 ). At any rate, any subalgebra
of a Jordan algebra is clearly a Jordan algebra. In the following, we shall assume
that char(K) = 2. If for any K-algebra A we define a new multiplication as
a ◦ b = (1/2)(ab + ba),
152 JUN-ICHI IGUSA
then we have a ◦ b = b ◦ a and a ◦ a = a2 . If A has an involution a → a and B is the

subspace consisting of symmetric elements a = a, then B becomes a subalgebra
of A under the new multiplication. If C is any K-algebra with involution and
Mn (C) is the K-algebra of n × n matrices with entries in C and with the matrix-
multiplication, then a → t a , where (t a )ij = aji for all i, j, defines an involution
of Mn (C). Therefore, the subspace Hn (C) consisting of hermitian matrices t a = a
forms a subalgebra of Mn (C) under the new multiplication. Suppose now that A
is associative. Then A with the new multiplication becomes a Jordan algebra. In
fact,
(a ◦ b) ◦ a2 = (1/4)(a3 b + a2 ba + aba2 + ba3 ) = a ◦ (b ◦ a2 ).
Therefore, if C is an associative composition algebra over K so that Mn (C) is
also associative, then Hn (C) with the new multiplication is a Jordan algebra. We
shall explain further examples of Jordan algebras. We keep in mind that since
char(K) = 2, an element a of C is symmetric, i.e., a = a, if and only if a is in
K = K1.
We take a quadratic form Q on a vector space X over K, put
A = Ke1 + Ke2 + X,
and define a multiplication in A as
(α1 e1 + α2 e2 + x)(β1 e1 + β2 e2 + y) = γ1 e1 + γ2 e2 + z,
where γi = αi βi + (1/2)Q(x, y) for i = 1, 2 and z = (1/2)((α1 + α2 )y + (β1 + β2 )x).

We shall show that A is a Jordan algebra. Clearly A is a commutative K-algebra.
If we put a = α1 e1 + α2 e2 + x and e = e1 + e2 , then ea = a and further
a2 − (α1 + α2 )a + (α1 α2 − Q(x))e = 0.
Therefore, if we put τ = α1 + α2 , ν = α1 α2 − Q(x), then
(ab)a2 = τ (ab)a − νab = τ a(ba) − νab = a(ba2 ).
If now C is any composition algebra and α1 , α2 , x are the (1, 1), (2, 2), (1, 2)
entries of a in H2 (C) then a → α1 e1 + α2 e2 + x gives an isomorphism from H2 (C)
to A = Ke1 + Ke2 + C where Q(x) = n(x). Therefore, H2 (C) is a Jordan algebra.
We shall show that H3 (C) is also a Jordan algebra.
We take a, b, c from H3 (C) and keep in mind that aij = aji , etc.; we put a◦a◦a =
a ◦ (a ◦ a), e = 13 . We introduce the following quadratic and cubic forms on A:
Q(a) = (1/2)(a211 + a222 + a233 ) + n(a23 ) + n(a31 ) + n(a12 ),

det(a) = a11 a22 a33 + t(a23 a31 a12 ) − a11 n(a23 ) − a22 n(a31 ) − a33 n(a12 ).
Since t(aij ajk aki ) is invariant under even permutations of aij , ajk , aki and is equal
to t((aij ajk aki ) ) = t(aik akj aji ), we see that t(aij ajk aki ) = t(a23 a31 a12 ) for all
distinct i, j, k. Since

Q(a, b) = aii bii + n(aij , bij ),
i i<j
we see that Q is nondegenerate. Furthermore,
(a ◦ b)ii = aii bii + (1/2){n(aij , bij ) + n(aik , bik )},

(a ◦ b)ij = (1/2){(aii + ajj )bij + (bii + bjj )aij + aik bkj + bik akj },
hence
Q(a ◦ b, c) = Q(a, b ◦ c).
We shall use this fact very often. If we replace b above by a and a ◦ a, then we get
(a ◦ a)ii = a2ii + n(aij ) + n(aik ), (a ◦ a)ij = (aii + ajj )aij + aik akj ;
(a ◦ a ◦ a)ii = a3ii + (2aii + ajj )n(aij ) + (2aii + akk )n(aik ) + t(a23 a31 a12 ),
(a ◦ a ◦ a)ij = (a2ii + aii ajj + a2jj + n(a23 ) + n(a31 ) + n(a12 ))aij + Q(a, e)aik akj .
By using these we can easily verify that a satisfies the following basic cubic equation
a ◦ a ◦ a − Q(a, e)a ◦ a + Q1 (a)a − det(a)e = 0,
in which
Q1 (a) = (1/2)Q(a, e)2 − Q(a).
If we take a variable t and define the first polar det1 (a, b) of det(a) as
det(a + tb) = det(a) + det1 (a, b)t + . . . ,
then by using the definition we get
det1 (a, b) = Q(a# , b), a# = a ◦ a − Q(a, e)a + Q1 (a)e.
Similarly, by taking the first polar of the above cubic equation for a we get
(a ◦ a) ◦ b + 2a ◦ (a ◦ b) − Q(b, e)a ◦ a − 2Q(a, e)a ◦ b

+ Q1 (a, b)a + Q1 (a)b − det1 (a, b)e = 0,
in which Q1 (a, b) = Q(a, e)Q(b, e) − Q(a, b). If we denote the LHS of the above
equation by L(a, b), then we get a ◦ L(a, b) − L(a, a ◦ b) = 0 because both terms are
0. By using the cubic equation for a and the above expression for Q1 (a, b), we can
rewrite this equation as
a ◦ (b ◦ (a ◦ a)) − (a ◦ b) ◦ (a ◦ a) = det1 (a, b)a − det1 (a, a ◦ b)e

+ Q1 (a, a ◦ b)a − Q1 (a)Q(b, e)a + det(a)Q(b, e)e.
If we replace det1 (a, b) by Q(a# , b) with a# as above and use the fact that a ◦ a# =
det(a)e, then we get
RHS = (Q1 (a, a ◦ b) − Q(a, e)Q(a ◦ b, e) + Q(a, a ◦ b))a = 0.

154 JUN-ICHI IGUSA
We have thus shown that H3 (C) is a Jordan algebra for any composition algebra
C. If we assume that char(K) = 3, in addition to char(K) = 2, then Q|Ke is
nondegenerate, hence A = Ke ⊕ (Ke)⊥ . The proof can be slightly shortened
by using this fact. At any rate, the fact is that if K is algebraically closed and
char(K) = 2, then every simple Jordan algebra over K is isomorphic to Hn (C) for
some n ≥ 1 for the C in (C1), (C2), (C3), Ke1 + Ke2 + X with a nondegenerate
quadratic form Q on X, or H3 (C) for the C in (C4). We again refer the details to
Jacobson [32]. We might mention, for a comparison, that if A is a simple associative
algebra over any algebraically closed field K, then by a special case of Wedderburn’s
theorem A is isomorphic to Mn (K) for some n ≥ 1.
9.5 Norm forms and Freudenthal quartics

If C is any K-algebra and a is an element of Mn (C) with aij as its (i, j)-entry, then
we put e = 1n and
tr(a) = a11 + a22 + . . . + ann .
We take a Jordan algebra A = Hn (C) and define ak inductively on k > 0 in N as
ak = a ◦ ak−1 with the understanding that a0 = e. We take as C one of the (C1),
(C2), (C3), (C4) in section 9.4 and show that a satisfies an equation of the form
an − tr(a)an−1 + . . . + (−1)n det(a)e = 0,
in which det(a) is a homogeneous polynomial of degree n in the entries of a. Since

the above equation is not unique as stated, we shall make det(a) explicit in each
case and call it the norm form of A. Since we are familiar with quadratic forms, we
shall be interested in the case where n > 2.
If a is any element of Mn (K), where K is an arbitrary field, we denote by det(a)
the usual determinant of a. If φ(t) is the characteristic function of a, i.e., if
φ(t) = det(te − a) = tn − tr(a)tn−1 + . . . + (−1)n det(a),
then by Cayley’s theorem we will have φ(a) = 0. We might recall its proof: If we
put (te − a)# = φ(t)(te − a)−1 and write

φ(t) = αi ti , (te − a)# = bi ti
0≤i≤n 0≤i≤n
with αi in K and bi in Mn (K), then αi e = bi−1 − abi for 0 ≤ i ≤ n, in which

b−1 = bn = 0. This implies

φ(a) = αi ai = (ai bi−1 − ai+1 bi ) = 0.
0≤i≤n 0≤i≤n
We shall now assume that char(K) = 2. If C is as in (C1), hence C = K, then

Hn (C) = Symn (K), the subspace of Mn (K) of symmetric matrices. Therefore, we
can take the determinant of a as det(a). If C is as in (C2), hence C = K + K,
then Hn (C) consists of a = (x, t x) for all x in Mn (K). Furthermore, we see that a
satisfies the same equation in Hn (C) as x in Mn (K). Therefore, we can take the
determinant of x as det(a). If C is as in (C3), hence C = M2 (K), then Hn (C)
consists of all a in M2n (K) satisfying Jn t aJn−1 = a, i.e., such that x = aJn is in
Alt2n (K), in which Jn and Alt2n (K) are as in section 9.3. If x is in Alt2n (K), i.e.
an alternating matrix in M2n (K), we define its Pfaffian Pf(x) as

Pf(x) = ε(σ)xσ(1),σ(2) · · · xσ(2n−1),σ(2n) ,
σ
in which ε(σ) = ±1 according as the permutation σ of {1, 2, . . . , 2n} is even

or odd and indicates the restriction σ(1) < σ(2), . . . , σ(2n − 1) < σ(2n),
σ(1) < σ(3) < . . . < σ(2n − 1); the number of terms is (2n)!/2n n!. We observe that
if we put
ψ(t) = Pf(tJn − x) = tn − τ tn−1 + . . . + (−1)n Pf(x),
then we have
τ = x12 + x34 + . . . + x2n−1,2n = tr(a),
in which tr(a) is defined in Hn (C). Furthermore, a similar argument as in the proof
of Cayley’s theorem shows that ψ(a) = 0. Therefore, we can take Pf(x) = Pf(aJn )
as det(a). If “det” means the usual determinant, then we have found that the norm
form in each case has the following meaning: det(x) for x in Symn (K), det(x) for
x in Mn (K), the Pfaffian Pf(x) for x in Alt2n (K).
Finally, in the case where C is as in (C4), we have only to consider Hn (C) for
n ≤ 3. If n = 2, then
det(a) = a11 a22 − n(a12 ),
which is a hyperbolic form in 10 variables. If n = 3, then
det(a) = a11 a22 a33 + t(a23 a31 a12 ) − a11 n(a23 ) − a22 n(a31 ) − a33 n(a12 ).
We shall obtain a classical expression for this cubic form. If we write an arbitrary
element a of M3 (C) as a = (a1 , a2 ) with a1 , a2 in M6 (K), then a is in H3 (C) if and
only if a1 is in H3 (M2 (K)) and the entries of a2 satisfy (a2 )ii = 0 for 1 ≤ i ≤ 3,
(a2 )ij = −(a2 )ji for 1 ≤ i < j ≤ 3. Therefore, by what we have shown above
y = a1 J3 is in Alt6 (K). We shall denote the (i, j)-entry of y by yij ; also we put
 
z11 z12 . . . z16
(a2 )23 (a2 )31 (a2 )12 =  
z21 z22 . . . z26
and wij = z1i z2j − z1j z2i for 1 ≤ i, j ≤ 6. Then in det(a) those terms which are
free from zij will give Pf(y). Furthermore, we can easily verify that

det(a) − Pf(y) = − yij wij = −t z1 yz2 ,
i<j
in which zi = t (zi1 . . . zi6 ) for i = 1, 2.

156 JUN-ICHI IGUSA
We shall next introduce Freudenthal quartics. We take A = H3 (C) and put

X = K 2 + A2 . An element x of X is of the form (a0 , b0 ; a, b) for some a0 , b0 in K
and a, b in A. We observe that
dimK (X) = 6 · dimK (C) + 8 = 14, 20, 32, 56
for dimK (C) = 1, 2, 4, 8. In the notation of section 9.4 we define the Freudenthal
quartic on X after H. Freudenthal [14] as
f (x) = (a0 b0 − Q(a, b))2 − 4(a0 det(b) + b0 det(a) + Q(a# , b# )).
It appeared in his earlier paper in the following form:
J(y, z) = Pf(y) + Pf(z) − (1/4)tr((yz)2 ) + ((1/4)tr(yz))2 ,
in which y, z are in Alt8 (K). Actually, the above cubic form Pf(y) − t z1 yz2 and
J(y, z) are in E. Cartan’s thesis of 1894 but with some incorrectness about J(y, z),
which was corrected as above by Freudenthal. Furthermore, he and T. A. Springer
developed the theory of Jordan algebras in connection with exceptional simple
groups. At any rate, if C = M2 (K) + M2 (K), then there exists a K-linear map
X → Alt8 (K)2 under which f (x) becomes −4J(y, z). We might give one such map.
We write a = (a1 , a2 ), b = (b1 , b2 ) as before and define y , z in Alt6 (K) with their
(i, j)-entries yij , zij for 1 ≤ i, j ≤ 6 as y = a1 J3 , z = J3 b1 . We next define y, z in
Alt8 (K) with their additional (i, j)-entries yij , zij for i, j = 7, 8 as

y17 y27 . . . y67
= J1 (b2 )23 (b2 )31 (b2 )12 J3 , y78 = b0 ,
y18 y28 . . . y68

z17 z27 ... z67
= (a2 )23 (a2 )31 (a2 )12 , z78 = a0 .
z18 z28 ... z68
Then, after some lengthy computation, we can see that f (x) = −4J(y, z). Since we
shall not use this fact, the details will not be given.
In order to satisfactorily examine Freudenthal quartics in the four cases, we need
to prove a large number of formulas in the theory of Jordan algebras. Instead, we
shall explain only one case where such a preparation will not be necessary. In fact,
we shall use the method we have used in [22], pp. 1021-1023 which depends only
on the usual matrix computation. p n
We know that the action of GLn (K) on K n extends to K for all p. In
particular, for p = 3 we have

(g · x)ijk = gii gjj gkk xi j k
for every g in GLn (K) with gii as its (i, i )-entry and for every i, j, k, in which the
summation is for all i , j , k . We observe that if x = 0, hence xijk = 0 for some i <
j < k and if g is the permutation-matrix representing (1i)(2j)(3k), then (g · x)123 =
3
xijk = 0. We shall assume, from now on, that n = 6, hence dimK ( K n ) = 20,
and denote the 3 × 3 entry matrices of g at (1, 1), (1, 2), (2, 1), (2, 2) respectively
by α, β, γ, δ. We observe that if x123 = 0 in x, then there exists a unique g with
α = δ = 13 , β = 0, such that (g · x)ijk = 0 for all i < j ≤ 3 < k and the unique γ is
given by
−x123 γ = (x23i x31i x12i )i=4,5,6 ;
we shall denote the RHS by a. In other words, a is the element of M3 (K) with
(x23i x31i x12i ) as its (i − 3)-th row for i = 4, 5, 6. Similarly, if x456 = 0, there
exists a unique g with α = δ = 13 , γ = 0 such that (g ·x)ijk = 0 for all i ≤ 3 < j < k
and the unique β is given by
−x456 β = (xi56 xi64 xi45 )i=1,2,3 ;
we shall denote the RHS by b. Finally, for an arbitrary x we put a0 = −x123 ,

b0 = −x456 and use the notation x = (a0 , b0 ; a, b).
We now take x with a0 = 0, denote by g = g0 the element of SL6 (K) for which
α = δ = 13 , β = 0, γ = a−1
0 a and put g0 · x = x = (a0 , b0 ; 0, b ). If for any a in
# #
M3 (K) we put a = Adj(a) so that aa = det(a)13 , then a0 = a0 and
b0 = b0 − tr(γb) − tr(aγ # ) + det(γ)a0

= b0 − a−1 −2
0 tr(ab) − 2a0 det(a),
(b )ij = bij + 2a−1 −1 #
0 (a )ij − a0 (a )ij
#
for 1 ≤ i, j ≤ 3, hence b = b + a−1 #

0 a . Furthermore, if we put
J = a0 det(b ) − (1/4)(a0 b0 )2 ,
then by using the above expressions for a0 , b0 , b and the general formulas
det(a + b) = det(a) + tr(a# b) + tr(ab# ) + det(b),

det(a# ) = det(a)2 , (a# )# = det(a)a,
we see that −4J is equal to
f (x) = (a0 b0 − tr(ab))2 − 4(a0 det(b) + b0 det(a) + tr(a# b# )).
This is the Freudenthal quartic for C = K + K. In order to proceed further we

need the following lemma, the verification of which is straightforward.
Lemma 9.5.1 Let L denote an arbitrary field and g any element of GL6 (L) with
3 × 3 entry matrices α, β, γ, δ. Then
g · (1, 0; 0, 0) = (det(α), det(γ); −γα# , −αγ # ),

g · (0, 1; 0, 0) = (det(β), det(δ); −δβ # , −βδ # ).
We observe that if p, q, p , q are elements of L× satisfying pq = p q , then we

can find g in SL6 (L) satisfying g · (p, q; 0, 0) = (p , q ; 0, 0). In fact, by Lemma
9.5.1 we have only to choose α, δ satisfying det(α) = p /p, det(δ) = q /q and put
158 JUN-ICHI IGUSA
β = γ = 0. We also observe that if q is in L× , we can find g in SL6 (L) satisfying

g · (1, q; 0, 0) = (1, −q; 0, 0). In fact, by Lemma 9.5.1 we have only to choose β, γ
satisfying det(β) = 1/q, det(γ) = −q and put α = δ = 0. We shall show that if
J = 0 in x = (a0 , b0 ; 0, b ) above and θ = (−4J)1/2 , then we can find g in SL6 (L)
for L = K(θ) satisfying g · (1, θ; 0, 0) = x . We shall, for the sake of simplicity, write
a0 , etc. instead of a0 , etc. so that x = (a0 , b0 ; 0, b) and
−4J = (a0 b0 )2 − 4a0 det(b) = θ 2 = 0.
This implies a0 = 0 and either a0 b0 − θ = 0 or a0 b0 + θ = 0. We have remarked that
(1, θ; 0, 0) and (1, −θ; 0, 0) are in the same SL6 (L)-orbit. Therefore, we may assume
that
q = (1/2)a0 (1 − θ −1 a0 b0 ) = 0
and we put p = θ/q. Also by another remark we have only to find g in SL6 (L)
satisfying g · (p, q; 0, 0) = x . We just define g as
α = −θ −1 a0 b, β = 13 , γ = −a−1
0 q13 , δ = θ −2 a0 pb# .
Then g has the required property. In fact, if we multiply θ −1 a0 b to the second
column of g and add to the first column, then the new first column will have 0, −13
as its entry matrices. Since the new matrix is clearly in SL6 (L), we see that g
is in SL6 (L). The verification of g · (p, q; 0, 0) = (a0 , b0 ; 0, b) by Lemma 9.5.1 is
straightforward.
Lemma 9.5.2 If g, q are respectively elements of GL6 (L), L× for any field L and
g · (1, q; 0, 0) = (1, q ; 0, 0), then q = ± det(g)q.
Proof. By Lemma 9.5.1 the condition on g becomes
(∗) det(α) + q det(β) = 1, det(γ) + q det(δ) = q ,
γα# + qδβ # = 0, αγ # + qβδ # = 0.
Suppose first that det(α) = 0. Then the last two equations imply
γ = −qδβ # (α# )−1 , (det(α) + q det(β))βδ # = 0.
By the first equation in (∗), the second equation above becomes βδ # = 0, hence the
fourth equation in (∗) becomes αγ # = 0, hence γ # = 0, and this implies det(γ) = 0.
Then the second equation in (∗) becomes q det(δ) = q , hence det(δ) = 0. Then
βδ # = 0 above implies β = 0, hence det(α) = 1, γα# = 0, hence γ = 0, respectively
by the first and the third equations in (∗). Therefore, we get det(g) = det(δ)
and q = det(g)q. Suppose next that det(α) = 0, hence det(β) = 1/q by the
first equation in (∗). Denote by g0 any element of SL6 (L) with entry matrices
α0 , β0 , γ0 , δ0 satisfying α0 = δ0 = 0 and det(β0 ) = −1/q, det(γ0 ) = q so that
g0 · (1, −q; 0, 0) = (1, q; 0, 0). Then the entry matrices of gg0 are βγ0 , αβ0 , δγ0 , γβ0 ,
in which det(βγ0 ) = 1, and gg0 · (1, −q; 0, 0) = (1, q ; 0, 0). Therefore, by what we
have shown, we get q = − det(g)q.
We shall formulate an immediate consequence of what we have discussed so far
as a proposition; further consequences will be used later to compute Z(s) for f (x).
3
Proposition 9.5.1 Let K denote any field with char(K) = 2, identify K 6 with
K 2 + M3 (K)2 under the K-linear bijection x → (a0 , b0 ; a, b), in which
a0 = −x123 , b0 = −x456 , a = (x23i , x31i , x12i ), b = (xi56 , xi64 xi45 )
respectively for i = 4, 5, 6 and i = 1, 2, 3, and put
f (x) = (a0 b0 − tr(ab))2 − 4(a0 det(b) + b0 det(a) + tr(a# b# )).
Then for every g in GL6 (K), we have
f (g · x) = det(g)2 f (x);
if f (x) = 0 and θ = f (x)1/2 , then there exists an element g of SL6 (K(θ)) satisfying
g · x = (1, θ; 0, 0); and if θ is in L× for any field L, the fixer of (1, θ; 0, 0) in SL6 (L)
consists of all g with its entry matrices α, β, γ, δ satisfying det(α) = det(δ) = 1,
β = γ = 0.
Proof. In order to prove the first part, i.e., f (g · x) = det(g)2 f (x), we can replace K
by its algebraic closure, and we shall use the principle of the irrelevance of algebraic
inequalities in Chapter 1.1. We take t = (g, x) regarding the 56 entries of g, x as
variables and put
R1 (t) = det(g), R2 (t) = x123 , R3 (t) = (g · x)123 , R4 (t) = f (x),
R5 (t) = f (g · x), F (t) = f (g · x) − det(g)2 f (x).
Suppose that R1 (t ) = 0, . . . , R5 (t ) = 0 for any t = (g , x ) in K 56 . Then we

know that there exist elements g0 , g1 of SL6 (K) satisfying g0 · x = (1, θ0 ; 0, 0),
g1 · (g · x ) = (1, θ1 ; 0, 0), in which (θ0 )2 = f (x ), (θ1 )2 = f (g · x ). This implies
g1 g g0−1 · (1, θ0 ; 0, 0) = (1, θ1 ; 0, 0), in which θ0 θ1 det(g1 g g0−1 ) = 0. Then by Lemma
9.5.2 we get
f (g · x ) = (θ1 )2 = (det(g1 g g0−1 )θ0 )2 = det(g )2 f (x ),
i.e., F (t ) = 0. Therefore, by the above principle F (t) = 0, hence F (t ) = 0 for all
t in K 56 .
We shall now go back to the original notation. Suppose that f (x) = 0 and
θ = f (x)1/2 . Then for some permutation matrix g0 we will have (g0 · x)123 = 0.
Since f (g0 · x) = det(g0 )2 f (x) = f (x), the second part follows from what we have
shown earlier. As for the third part, the proof of Lemma 9.5.2 shows that if g is
in the fixer of (1, θ; 0, 0) in GL6 (L) and if det(α) = 0, then det(α) = det(δ) = 1
and β = γ = 0 while if det(α) = 0, then det(g) = −1. Therefore, if g is in SL6 (L),
then we only have the first possibility. Furthermore, every such g is in the fixer of
(1, θ; 0, 0).
160 JUN-ICHI IGUSA
9.6 Gauss’ identity and its corollaries

C. F. Gauss used a remarkable identity to convert a theta series into an infinite
product; he used a special case of that identity in his first sign-determination of the
Gaussian sum. We shall recall his identity with his proof and prove its corollaries
for our later use.
In the original notation of Gauss [15] we put
T = 1 + (an − 1)/(a − 1) · t + (an − 1)(an − a)/(a − 1)(a2 − 1) · t2 + . . . ,
in which a, t are variables. Since T depends on n, we put Θ(n) = T . Then

Θ(0) = 1, Θ(1) = 1 + t, Θ(2) = 1 + (a + 1)t + at2 = (1 + t)(1 + at). Furthermore,
Θ(n + 1) = (1 + an t)Θ(n), hence
T = (1 + t)(1 + at) · · · (1 + an−1 t).
This is the Gauss identity.

We now take m from Z, n from N, and put

Fm,n (a, t) = (1 − am+i t) (1 − ai )
1≤i≤n
with the usual understanding that Fm,0 (a, t) = 1; we extend the definition by
Fm,n (a, t) = 0 for n < 0. If we put
Fm,n (a) = Fm,n (a, 1),
then for m, n ≥ 0 we have

Fm,n (a) = (1 − ai ) (1 − ai ) · (1 − ai )
1≤i≤m+n 1≤i≤m 1≤i≤n
hence Fm,n (a) = Fn,m (a), and Fi,j (a)Fi+j,k (a) = Fi,j+k (a)Fj,k (a). We also observe
that
(∗) Fm,n (a, t) = Fm,n−1 (a, t) + an Fm−1,n (a, t)
and further Fm,n (a, t) = Fm−k,n (a, ak t) for all m, n, k in Z. In this notation the
Gauss identity can be written as

(G0) Fi,j (a)ai(i−1)/2 ti = (1 + ai−1 t)
i+j=n 1≤i≤n
for every n ≥ 0. We might also mention that (∗) for t = 1 is tacitly used in Gauss’
proof, in fact as follows: The coefficient of ti in (1 + an t)Θ(n) is
Fi,n−i (a)ai(i−1)/2 + Fi−1,n−i+1 (a)a(i−1)(i−2)/2+n
for 0 ≤ i ≤ n + 1. Since (i − 1)(i − 2)/2 + n = i(i − 1)/2 + (n − i + 1), by using

(∗) for t = 1 we see that it is equal to the coefficient of ti in Θ(n + 1). Therefore,
Θ(n + 1) = (1 + an t)Θ(n), and the rest is by an induction on n.
We shall show that

2
(G1) (1 − ai+k )(1 − ai+k t) (1 − ai ) ak tk = 1.
0≤k≤n 1≤i≤n−k
By (G0) we have

(1 − ai+k t) = Fi,j (a) ai(i−1)/2 (−ak+1 t)i .
1≤i≤n−k i+j=n−k
Since the other product is Fk,n−k (a), the LHS of (G1) becomes

(−1)i Fk,n−k (a) Fi,j (a) ai(i+1)/2+(i+k)k ti+k .
i+j+k=n
If we put l = i + k, hence j = n − l, by using Fk,n−k (a) Fi,j (a) = Fk,i (a) Fi+k,j (a),
the above expression can be rewritten as

Fk,i (a) ak(k−1)/2 (−1)k · (−1)l Fl,n−l (a) al(l+1)/2 tl .
0≤l≤n k+i=l
According to (G0), we have

Fk,i (a) ak(k−1)/2 (−1)k = (1 − ai−1 ),
k+i=l 1≤i≤l
which represents 1 or 0 according as l = 0 or l > 0. This implies (G1).

We shall next prove
2
(G2) Fm−k,k (a) Fk,n−k (a, t) ak tk = Fm,n (a, t).
0≤k≤n
If we put Dm,n (a, t) = RHS - LHS, then Dm,n (a, t) = 0 for n ≤ 0. Therefore, we
shall assume that n > 0 and prove Dm,n (a, t) = 0 by an induction on n. If we apply
(∗) to Fm,n (a, t) and Fk,n−k (a, t), then we get
2

Dm,n (a, t) = Fm,n−1 (a, t) − Fm−k,k (a) Fk,n−k−1 (a, t) ak tk
0≤k≤n

Fm−k,k (a) Fk−1,n−k (a, t) ak (a−1 t)k .
2
+ an Fm−1,n (a, t) −
0≤k≤n
We observe that in the first {·} the term for k = n is 0, hence {·} is equal
to Dm,n−1 (a, t), which is 0 by induction. On the other hand, if we replace
Fm−1,n (a, t) and Fk−1,n−k (a, t) in the second {·} respectively by Fm,n (a, a−1 t) and
Fk,n−k (a, a−1 t), then it becomes Dm,n (a, a−1 t). Therefore, we get Dm,n (a, t) =
an Dm,n (a, a−1 t). Since a, t are variables and Dm,n (a, t) is a polynomial in t, this
implies
Dm,n (a, t) = Am,n (a)tn
162 JUN-ICHI IGUSA
for some Am,n (a) which is independent of t. If we can show that Am,n (a) = 0, then
we will have Dm,n (a, t) = 0. Since Am,n (a) is the coefficient of tn in Dm,n (a, t), by
going back to its definition we get

amn − (−1)n (1 − ai ) · a−n(n+1)/2 Am,n (a)
1≤i≤n

= (−1)k Fm−k,k (a) · (1 − an−k+i ) · ak(k−1)/2 .
0≤k≤n 1≤i≤k
Therefore, we have only to show that the above RHS is equal to amn . By definition
we have

Fm−k,k (a) (1 − an−k+i ) = Fn−k,k (a) (1 − am−k+i )
1≤i≤k 1≤i≤k
and by (G0) we can write

(1 − am−k+i ) = Fi,j (a) ai(i−1)/2 (−am−k+1 )i .
1≤i≤k i+j=k
Therefore, if we put l = n − k, then the expression which is expected to be amn

becomes
Fl,n−l (a) Fi,j (a) ami+j(j−1)/2 (−1)j .
i+j+l=n
Since Fl,n−l (a) Fi,j (a) = Fi,n−i (a) Fj,l (a), it can be rewritten as

Fi,n−i (a) ami Fj,l (a) aj(j−1)/2 (−1)j .
0≤i≤n j+l=n−i
As in the proof of (G1), we see by (G0) that {·} above represents 1 or 0 according
as n − i = 0 or n − i > 0. Therefore, the above expression is indeed equal to amn .
Finally, we shall show that

Fm−k,k (a) Fk,n−k (a, t) ak −k tk = Fm−1,n (a, t) + tFm,n−1 (a, t).
2
(G3)
0≤k≤n
If we apply (∗) to Fm−k,k (a), the above LHS becomes

Fm−k,k−1 (a) Fk,n−k (a, t) ak −k tk
2
0≤k≤n
2
+ Fm−k−1,k (a) Fk,n−k (a, t) ak tk .
0≤k≤n
By (G2) the second sum is equal to Fm−1,n (a, t). We observe that in the first sum
the term for k = 0 is 0. If we replace k by k + 1, the sum becomes
2
t· Fm−k−1,k (a) Fk+1,n−k−1 (a, t) ak (at)k ,
0≤k<n
in which Fk+1,n−k−1 (a, t) = Fk,n−k−1 (a, at). Therefore, again by (G2), it is equal
to t · Fm−1,n−1 (a, at) = t · Fm,n−1 (a, t).
Chapter 10
Computation of Z(s)
10.1 Z(ω) in some simple cases
We recall that
Z(ω) = ω(f (x)) dx,
X◦
in which X ◦ = OK n
, dx = µn is the Haar measure on X = K n normalized as
µn (X ) = 1, and f (x) is in K[x1 , . . . , xn ]\K. As for ω, it is in Ω0 (K × ), i.e.,
◦
t = ω(π) satisfies 0 < |t| < 1. We have denoted Z(ωs ) by Z(s). This is a p-adic
analogue of
|f (x)|s exp(−π t xx) dx.
Rn
There is a remarkable difference between them about the explicit computability for
a given f (x). In the real case Z(s) is hardly computable; while in the p-adic case it
is a rational function of t and has been computed for a large number of f (x). The
significance of an ever-increasing list of explicitly computed Z(s) is that it allows us
to formulate conjectures and proceed to their proofs. At any rate, in this chapter
we shall explain some Z(s) in the above-mentioned list. We shall start with the
simplest cases after the following general remarks.
×
Suppose that OK acts on X as (u, x) → u · x keeping µn and X ◦ invariant;
suppose further that f (u · x) = um f (x) for some m > 0 in N. Then by replacing x
by u · x in the integral defining Z(ω) we get
Z(ω) = χ(u)m Z(ω)
× ×
for every u in OK , in which χ = ω|OK . Therefore, Z(ω) = 0 unless χm = 1.
◦
We remark that X \{0} is the disjoint union of π e Un for all e in N, where Un =
×
X ◦ \πX ◦ , hence U1 = OK and µn (Un ) = [n] in the notation of Chapter 9.3. We put
[i, j] = 1 − q −i tj , [i, j]+ = 1 + q i tj
so that [i] = [i, 0], [i]+ = [i, 0]+ for every i, j in N.
Suppose now that f (x) = det(x) for x in X = Mn (K), hence dimK (X) = n2 .
If we define u · x as the multiplication by u to the first column x1 of x, then we
will have det(u · x) = u det(x), hence Z(ω) = 0 unless χ = 1. If we denote Z(s) by
Zn (s), then by the same method as in Chapter 6.3, we can show that

Zn (s) = [k]/[k, 1].
1≤k≤n
163
164 JUN-ICHI IGUSA
In fact, as a special case of Lemma 8.2.1, we get Z1 (s) = [1]/[1, 1]. If n > 1, we
write x = (x1 x ) and apply the second remark above to the domain of integration
by dx1 . In that way, we get

Zn (s) = (q −n t)i dx1 | det(x1 x )|sK dx ,
i≥0 Un (X )◦
in which X = Mn,n−1 (K). We observe that every x1 in Un can be written as

x1 = ge1 for some g in SLn (OK ). If we replace x by gx , then {·} above becomes
Zn−1 (s), hence
Zn (s) = Zn−1 (s) · [n]/[n, 1].
If we apply the induction assumption to Zn−1 (s), we get the above expression for
Zn (s).
Suppose next that x is in X = Alt2n (K), hence dimK (X) = n(2n − 1). We then
take the Pfaffian Pf(x) of x as f (x). We recall that Pf(x) is defined in Chapter 9.5
as
Pf(x) = (σ)xσ(1),σ(2) · · · xσ(2n−1),σ(2n) ,
in which σ is a permutation of {1, 2, . . . , 2n} restricted as at that place. It follows

from the definition that Pf(gxt g) = det(g)Pf(x) for every g in GL2n (K). Therefore,
if g is a diagonal matrix in GL2n (OK ) with u, 1, . . . , 1 as its diagonal entries and if
we put u · x = gxt g, then Pf(u · x) = u Pf(x), hence Z(ω) = 0 unless χ = 1. If we
denote Z(s) by Zn (s), then similarly as above we can show that

Zn (s) = [2k − 1]/[2k − 1, 1].
1≤k≤n
In fact, since Pf(x) = x12 for n = 1, we get Z1 (s) = [1]/[1, 1]. In the case where
n > 1, the first row of x in X ◦ is of the form π i (0 t x1 ) for some i in N and x1 in
U2n−1 , hence x1 = g e1 for some g in SL2n−1 (OK ). If we denote by g the element of
SL2n (OK ) with 1, 0, 0, g as its entry matrices and if x in X = Alt2n−2 (K) is the
right-bottom entry matrix of g −1 x t g −1 , then x is in (X )◦ and Pf(x) = π i Pf(x ).
Since d(π i x1 ) = q −(2n−1)i dx1 , therefore, we get

Zn (s) = (q −(2n−1) t)i dx1 |Pf(x )|sK dx
i≥0 U2n−1 (X )◦
= Zn−1 (s) · [2n − 1]/[2n − 1, 1].
By induction this implies the above expression for Zn (s).

As the third simple case we take column vectors x , x in K m and put
x = t (t x , t x ), f (x) = t x x
so that X = K 2m . In order to compute Z(ω) we can use Theorem 8.4.1 and an

explicit form of F (i) in Chapter 8.3 for the above f (x). We can also do it directly
as follows. If we define u · x as the multiplication by u to x , then f (u · x) = uf (x),

hence Z(ω) = 0 unless χ = 1. Furthermore,

Z(s) = (q −m t)i dx |t x x |sK dx .
Um m
OK
i≥0
If in the above {·} we write x = ge1 for some g in GLm (OK ) and replace x by
t −1
g x , then it becomes [1]/[1, 1], hence
Z(s) = [1][m]/[1, 1][m, 1].
Before we proceed further, we shall compute Z(s) in a similar case where
f (x) = t x x + π e x0
for any e in N so that X = K 2m+1 . If we denote Z(s) by ϕ(e) in this case, then
clearly ϕ(0) = [1]/[1, 1]. If e > 0, then we write

ϕ(e) = dx |t x x + π e x0 |sK dx dx0 ,
m m+1
OK OK
m m
and split OK above into Um and πOK . Then, by the repeatedly used argument, we
get
ϕ(e) = [m] · [1]/[1, 1] + q −m t · ϕ(e − 1).
This is a recursion formula by which we can express ϕ(e) in terms of ϕ(0). In that
way we get

|t x x + π e x0 |sK dx = [1] [1, 1][m, 1] · [m] + [0, 1]q −m (q −m t)e
X◦
for every e in N. We might remark that Z(s) in the third case becomes the limit of
the above expression as e → ∞.
As the fourth simple case we take x from X = M2m,2n (K) where m ≥ n and
f (x) = Pf(t xJm x),
in which Jm is as in Chapter 9.3 the element of Alt2m (Z) with 1 as its (2i − 1, 2i)-
entry for 1 ≤ i ≤ m and 0 as its (i, j)-entry for all other i < j. If we define u · x
as the multiplication by u to the first column of x, then f (u · x) = uf (x), hence
Z(ω) = 0 unless χ = 1. If we denote Z(s) by Zm,n (s), then we have

Zm,n (s) = [2(m − k + 1)][2k − 1] [2(m − k + 1), 1][2k − 1, 1].
1≤k≤n
The proof is by now a familiar argument. We write the first column of x as π i x1 for
some i in N and x1 in U2m . If we denote by Sp2m (OK ) the intersection of Sp2m (K)
and GL2m (OK ), then it consists of all g = (w1 w2 . . . w2m ) in M2n (OK ) satisfying
t
w2i−1 Jm w2i = 1 for 1 ≤ i ≤ m and t wi Jm wj = 0 for all other i < j. Therefore by
166 JUN-ICHI IGUSA
the same argument as in Chapter 9.3 we see that every x1 in U2m can be written
as x1 = ge1 for some g in Sp2m (OK ). If we put X = M2m,2n−1 (K), then similarly
as before we get

−2m i
Zm,n (s) = (q t) dx1 |f (x1 x )|sK dx
i≥0 U2m (X )◦

= [2m]/[2m, 1] · |f (e1 x )|sK dx .
(X )◦
We shall denote the above integral over (X )◦ by I. We put Y = M2n−1,2m−1 (K)

and t x = (y1 y2 y ) with column vectors y1 , y2 so that y = (y1 y ) is in Y ◦ . Then
we will have

−(2n−1)i
I= q dy2 |f (e1 t (y1 π i y2 y ))|sK dy .
i≥0 U2n−1 Y◦
If in the above {·} we write y2 = g ∗ e∗1 for some g ∗ in GL2n−1 (OK ), where e∗1 is
the e1 in K 2n−1 , and replace y by g ∗ y, then f (e1 t (y1 π i y2 y )) will be replaced by
det(g ∗ )f (e1 t (y1 π i e∗1 y )). Therefore, we get

I = [2n − 1] · q −(2n−1)i |f (e1 t (y1 π i e∗1 y ))|sK dy.
i≥0 Y◦
We shall denote the above integral over Y ◦ by Ii . The point is that if we put
X = M2m−2,2n−2 (K), x = (e1 t (y1 π i e∗1 y )), and denote by x in (X )◦ the
right-bottom entry matrix of x, then we have
f (x) = Pf(t xJm x) = π i · Pf(t x Jm−1 x ).
Therefore, we get Ii = ti · Zm−1,n−1 (s), hence

Zm,n (s) = [2m][2n − 1] [2m, 1][2n − 1, 1] · Zm−1,n−1 (s)
with the understanding that Zm−1,n−1 (s) = 1 for n = 1. This implies the above-
stated expression for Zm,n (s). We shall see later in this chapter that if we replace
Jm above by an element h of Sym2m (Z) with 1 as its (2i − 1, 2i)-entry for 1 ≤ i ≤ m
and 0 as its (i, j)-entry for all other i < j, and accordingly Pf(t xJm x) by det(t xhx),
then the corresponding Zm,n (s) will have an entirely different expression.
Finally, as the fifth simple case we take x = (y, (z1 z2 )), where y is in Y =
Alt6 (K) and (z1 z2 ) is in M6,2 (K) hence dimK (X) = 27, and
f (x) = Pf(y) − t z1 yz2 .
If g is a diagonal matrix in GL6 (OK ) with u, 1, . . . , 1 as its diagonal entries and if

we put u · x = (gy t g, t g −1 (uz1 z2 )), then f (u · x) = uf (x), hence Z(ω) = 0 unless
χ = 1. We shall show that

Z(s) = [1][5][9] [1, 1][5, 1][9, 1].
We shall use capital letters for the spaces of variables expressed by small letters
such as X, x and Y , y above. We write Z(s) as an integral with respect to dy dz1
followed by an integration by dz2 . If we split (Z2 )◦ \{0} into π i U6 for all i in N,
write z2 in U6 as z2 = ge1 for some g in SL6 (OK ) and replace y, z1 respectively by
t −1
g yg −1 , gz1 , then we get

Z(s) = [6] · q −6i |Pf(y) + π i t y1 w|sK dydw,
i≥0 Y ◦ ×W ◦
in which W = K 5 and (0 t y1 ) is the first row of y. If y is obtained from y by

crossing out its first row and column, hence Y = Alt5 (K), then we write the
above integral as an integral with respect to dy dw followed by an integration by
dy1 . We split (Y1 )◦ \{0} into π j U5 for all j in N, write y1 in U5 as y1 = g e1 for
some g in SL5 (OK ), where e1 is the e1 in K 5 , and replace y , w respectively by
g y t g , t (g )−1 w. If y is obtained from the new y by crossing out its first row and
column, hence Y = Alt4 (K), then we get

Z(s) = [5][6] [5, 1] · q −6i |Pf(y ) + π i x0 |sK dy dx0 .
i≥0 (Y )◦ ×OK
We observe that Pf(y ) is of the form x1 x4 + x2 x5 + x3 x6 . Therefore, the above

integral is ϕ(i) before for m = 3 and, by using that expression for ϕ(i), we get the
expression for Z(s) as stated above.
We might mention that in the above computation K is an arbitrary p-adic field
and that there is no restriction on q. Cases 1, 2, 3 are classical. The computation of
Z(s) in Case 5, i.e., for the norm form of the Jordan algebra H3 (C) for an octonion
K-algebra C, was first made by J. G. M. Mars [39] in an equivalent form assuming
that q is relatively prime to 6. More precisely he computed Weil’s function F ∗ (i∗ )
in that case. Case 4 and one more simple case are in [27]. We might further mention
that one of the reasons why the above cases are called “simple” is that the zeros of
Bernstein’s polynomials of f (x) are all integers. We copy the following list from T.
Kimura [34]:

bf (s) = (s + k), (s + 2k − 1), (s + 1)(s + m),
1≤k≤n 1≤k≤n

(s + 2(m − k + 1))(s + 2k − 1), (s + 1)(s + 5)(s + 9).
1≤k≤n

If we writebf (s) = (s + λ), then the above results can be stated simultaneously
as Z(s) = [λ]/[λ, 1].
10.2 A p-adic stationary phase formula

We shall formulate a general and useful method to compute Z(s). We recall that

Z(s) = |f (x)|sK dx, Re(s) > 0,
X◦
168 JUN-ICHI IGUSA
in which f (x) is in K[x1 , . . . , xn ]\K and the Haar measure dx on X = K n is

normalized so that the total measure of X ◦ = OK n
becomes 1. After multiplying
a suitable power of π to f (x), we may assume that the coefficients of f (x) are in
OK but not all in πOK , hence f¯(x) = f (x) mod π is in Fq [x1 , . . . , xn ]\{0}. We put
t = q −s and use the notation [i, j], etc. as in section 10.1.
Theorem 10.2.1 We take a subset Ē of Fnq and denote by S̄ its subset consisting
of all ā in Ē such that f¯(ā) = (∂ f¯/∂xi )(ā) = 0 for 1 ≤ i ≤ n; we further denote by
E, S the preimages of Ē, S̄ under X ◦ → X ◦ /πX ◦ and by N the number of zeros
of f¯(x) in Ē. Then we have

|f (x)|sK dx = q −n (card(Ē) − N ) + q −n (N − card(S̄)) [1]t [1, 1]
E

+ |f (x)|sK dx.
S
Proof. If we write ā = a mod π for every a in X ◦ , then we have

−n
|f (x)|K dx = q ·
s
|f (a + πx)|K dx +
s
|f (x)|sK dx.
E X◦ S
ā∈Ē\S̄
If f¯(ā) = 0, then f (a+πx) ≡ f (a) ≡ 0 mod π for all x in X ◦ , hence the contribution
of all such ā in Ē is q −n (card(Ē) − N ). If f¯(ā) = 0 and (∂ f¯/∂xi )(ā) = 0 for some
i, then we use Lemma 7.4.3. If we put
gi (x) = π −1 (f (a + πx) − f (a)), gj (x) = xj (j = i),
then g1 (x), . . . , gn (x) are SRP’s in x1 , . . . , xn and
∂(g1 , . . . , gn )/∂(x1 , . . . , xn )(0) = (∂f /∂xi )(a) ≡ 0 mod π,
hence (y1 , . . . , yn ) = (g1 (x), . . . , gn (x)) gives a measure-preserving map from X ◦

to X ◦ . Therefore, we get

|f (a + πx)|sK dx = |π(yi + π −1 f (a))|sK dyi = t · [1] [1, 1],
X◦ OK

hence the contribution of all such ā in Ē is q −n (N − card(S̄)) · [1]t [1, 1].
We have called the above theorem a p-adic stationary phase formula, abbrevi-
ated as SPF, in [30]. The significance of SPF is its wide applicability. Before we
give a more appropriate example or examples, we shall prove the following general
statement by using SPF:
Proposition 10.2.1 If f (x) is a homogeneous polynomial of degree d in
OK [x1 , . . . , xn ] such that f¯(ā) = (∂ f¯/∂xi )(ā) = 0 for 1 ≤ i ≤ n implies ā = 0
and further if N denotes the number of zeros of f¯(x) in Fnq , then

Z(s) = ([1][n]t + (1 − q −n N )[0, 1]) [1, 1][n, d].
Proof. If in SPF we take Ē = Fnq , then S = πX ◦ by assumption. Furthermore,

since f (x) is homogeneous of degree d, we have f (πx) = π d f (x). Therefore, we get

Z(s) = q −n (q −n − N ) + q −n (N − 1)[1]t [1, 1] + q −n td · Z(s).
This can be rewritten as stated in the proposition.
Corollary 10.2.1 If Q(x) is a quadratic form in OK [x1 , . . . , xn ] such that Q̄(x) is
reduced, then

 [1][n, 1]/[1, 1][n, 2] n odd
Z(s) =

[1](1 − χ(Q̄)q −n/2 )/[1, 1](1 − χ(Q̄)q −n/2 t) n even,
in which χ(Q̄) = ±1 according as Q̄ is hyperbolic or not.

In fact, if d = 2, then the condition on f (x) = Q(x) in Proposition 10.2.1 is that
Q̄ on = Fqn is reduced. Therefore by Theorem 9.2.1 we have
 n−1
 q n odd
N=
 n−1
q + χ(Q̄)[1]q n/2 n even.
This implies the formula in the corollary.
We shall compute Z(s) for a variant of f (x) in Proposition 10.2.1. Namely,
instead of that f (x), we take f (x) + π i y for any i in N and compute

ϕ(i) = |f (x) + π i y|sK dxdy.
X ◦ ×OK
We clearly have ϕ(0) = [1]/[1, 1]. In the case where i > 0, if we apply SPF to ϕ(i),
we get  −n i
 q t · ϕ(0) 0<i≤d
ϕ(i) = [n, d]Z(s) +
 −n d
q t · ϕ(i − d) i ≥ d,
in which Z(s) is for f (x). If we write i = dk + i0 , where 0 ≤ i0 < d, then by an
induction on j we get
ϕ(i) = [nj, dj]Z(s) + (q −n td )j · ϕ(d(k − j) + i0 )
for 0 ≤ j ≤ k, hence

 [n(k + 1), d(k + 1)]Z(s) + (q −n td )k · q −n ti0 · [1]/[1, 1] i0 > 0
ϕ(i) =

[nk, dk]Z(s) + (q −n td )k · [1]/[1, 1] i0 = 0.
In particular, if we replace f (x) by a quadratic form Q(x), then we will have

 [n(i + 1)/2, i + 1]Z(s) + (q −n/2 t)i−1 · q −n t · [1]/[1, 1] i odd
ϕ(i) =

[ni/2, i]Z(s) + (q −n/2 t)i · [1]/[1, 1] i even.
170 JUN-ICHI IGUSA
We observe that if we formally equate the above RHS’s, we get

Z(s) = [1][n/2] [1, 1][n/2, 1].
In other words if Q̄ is hyperbolic and only in that case, the above two expressions
become equal, and

|Q(x) + π e y|sK dxdy = [1] [1, 1][n/2, 1] · [n/2] + [0, 1]q −n/2 (q −n/2 t)e
X ◦ ×OK
for every e in N. We have computed this integral already in section 10.1.

Before we proceed further, we remark that quadratic forms Q(x) in
OK [x1 , . . . , xn ] with reduced Q̄ have entirely similar properties as those in
Fq [x1 , . . . , xn ]. We shall show, as an example, that if Q̄ is hyperbolic, hence n = 2m,
then we can find w1 , w2 , . . . , wn in X ◦ = OK n
satisfying

O K wi = X 0 , Q xi w i = x2i−1 x2i
1≤i≤n 1≤i≤n 1≤i≤m
for every x1 , . . . , xn in OK . We first observe that if Q̄ is just reduced and Q̄(ā0 ) = 0

for some a0 in X ◦ with ā0 = 0, then we can find a in X ◦ satisfying Q(a) = 0, a ≡ a0
mod π. In fact π −1 (Q(a0 + πx) − Q(a0 )) is an SRP in x1 , . . . , xn with a unit of
OK as the coefficient of xi for some i, hence it is equal to any element of OK , say
−π −1 Q(a0 ), for some x = a1 in X ◦ . Then Q(a) = 0 for a = a0 + πa1 ≡ a0 mod π.
If now Q(a) = 0 for some a in X ◦ with ā = 0, since b̄ → Q̄(ā, b̄) gives an Fq -linear
surjection from Fnq to Fq , the OK -homomorphism b → Q(a, b) from X ◦ to OK is
also surjective, hence Q(a, b0 ) = 1 for some b0 in X ◦ . If we put b = −Q(b0 )a + b0 ,
we will have Q(a, b) = 1, Q(b) = 0. If we denote the intersection of X ◦ and a, b ⊥
by L, then x − Q(x, b)a − Q(x, a)b is in L for every x in X ◦ , hence
X ◦ = (OK a + OK b) ⊕ L.
Therefore, we have only to put w1 = a, w2 = b and continue this process. We might

also mention that if q is odd, then the square map gives a surjection from 1+πOK to
× × 2
itself by Lemma 8.4.1. This implies the isomorphism OK /(OK ) → F× × 2
q /(Fq ) . If we
× × × 2
denote by χ the character of OK which gives rise to the isomorphism OK /(OK ) →
{±1}, then χ(Q̄) = χ(d(Q)) in Corollary 10.2.1 for the discriminant d(Q) of Q
relative to any OK -basis for X ◦ .
We shall now discuss the case which exemplifies the way SPF can be applied.
We shall show that Z(s) is computable for

f (x) = ci xdi i ,
1≤i≤n
in which d1 , . . . , dn are integers larger than 1 at most one of which is divisible

by the prime factor of q. This condition is clearly satisfied if they form a strictly
increasing sequence of prime numbers with no restriction on q. As for c1 , . . . , cn ,
we may assume that they are in OK but not all in πOK . If we denote by J the set
of all i for which c̄i = 0 so that

f¯(x) = c̄i xdi i = 0
i∈J
and by NJ the number of zeros of f¯(x) in Ē = Fnq , then NJ has been expressed in
terms of Jacobi sums by A. Weil [57]. Furthermore, S̄ is defined by xi = 0 for all i
in J. Therefore, by SPF we get

Z(s) = |f (x)|sK dx = RJ + q −card(J) te · |f1 (x)|sK dx,
X◦ X◦
in which
RJ = (1 − q −n NJ ) + (q −n NJ − q −card(J) )[1]t/[1, 1].
As for e and f1 (x), we replace xi in f (x) by πxi for all i in J and write the new
f (x) as π e f1 (x) with f1 (x) satisfying the same condition as f (x). We can then
apply the same argument to f1 (x). By repeating this process, we get a sequence
f0 (x) = f (x), f1 (x), f2 (x), . . . . If we write

fj (x) = cji xdi i ,
1≤i≤n
then cji differs from ci only by a power of π, hence we can write

Zj (s) = |fj (x)|sK dx = (ord(cj1 ), . . . , ord(cjn ))
X◦
for all j in N. The point is that, in view of ord(cji ) ≤ max(ord(ci ), di ) for all i, j,
the sequence Z0 (s) = Z(s), Z1 (s), Z2 (s), . . . becomes periodic, i.e. Zj (s) = Zj (s)
for some j < j . Then [α, β]Zj (s) for some α, β > 0 in N becomes known, hence
all Zk (s), in particular Z(s), will be known. In the following, we take c1 = c2 =
. . . = cn = 1, {d1 , d2 , . . . , dn } = {2, 3}, {2, 3, 5} and make the corresponding Z(s)
explicit.
If f (x) = x21 + x32 , then NJ = q not only for J = {1}, {2} but also for J = {1, 2}
because the plane curve f −1 {0} is parametrized as x1 = u3 , x2 = −u2 . Therefore,
we get RJ = [1][j, 1]/[1, 1] for j = card(J). If we denote RJ by Rj for j = card(J),
then we get the following sequence:
(0, 0) = R2 + q −2 t2 (0, 1), (0, 1) = R1 + q −1 t(1, 0),
(1, 0) = R1 + q −1 t(0, 2), (0, 2) = R1 + q −1 t2 (0, 0).
This implies

Z(s) = |x21 + x32 |sK dx
X◦
= [1]/[1, 1][5, 6] · (1 − q −2 t(1 − t) − q −5 t5 ).
172 JUN-ICHI IGUSA
If f (x) = x21 + x32 + x53 , then NJ = q 2 for all J. This is clear for card(J) = 1 and also
for J = {1, 2} by what we have shown. If x21 + x53 = 0, x3 = 0, and x1 = ux3 , then
u2 + x33 = 0, hence NJ = q 2 for J = {1, 3}. If x32 + x53 = 0, x3 = 0, and x2 = ux3 ,
then u3 + x23 = 0, hence NJ = q 2 for J = {2, 3}. Finally, if x21 + x32 + x53 = 0, x3 = 0,
and x1 = u1 x33 , x2 = u2 x23 , then (u21 + u32 )x3 + 1 = 0, hence
NJ = q(from x3 = 0) + (q 2 − q)(from x3 = 0) = q 2
also for J = {1, 2, 3}. Therefore, we get RJ = [1][j, 1]/[1, 1] for j = card(J). If we
denote RJ by Rj for j = card(J), then we get the following sequence:
(0, 0, 0) = R3 + q −3 t2 (0, 1, 3), (0, 1, 3) = R1 + q −1 t(1, 0, 2),
(1, 0, 2) = R1 + q −1 t(0, 2, 1), (0, 2, 1) = R1 + q −1 t(1, 1, 0),
(1, 1, 0) = R1 + q −1 t(0, 0, 4), (0, 0, 4) = R2 + q −2 t2 (0, 1, 2),
(0, 1, 2) = R1 + q −1 t(1, 0, 1), (1, 0, 1) = R1 + q −1 t(0, 2, 0),
(0, 2, 0) = R2 + q −2 t2 (0, 0, 3), (0, 0, 3) = R2 + q −2 t2 (0, 1, 1),
(0, 1, 1) = R1 + q −1 t(1, 0, 0), (1, 0, 0) = R2 + q −2 t(0, 2, 4),
(0, 2, 4) = R1 + q −1 t2 (0, 0, 2), (0, 0, 2) = R2 + q −2 t2 (0, 1, 0),
(0, 1, 0) = R2 + q −2 t(1, 0, 4), (1, 0, 4) = R1 + q −1 t(0, 2, 3),
(0, 2, 3) = R1 + q −1 t2 (0, 0, 1), (0, 0, 1) = R2 + q −2 t(1, 2, 0),
(1, 2, 0) = R1 + q −1 t(0, 1, 4), (0, 1, 4) = R1 + q −1 t(1, 0, 3),
(1, 0, 3) = R1 + q −1 t(0, 2, 2), (0, 2, 2) = R1 + q −1 t2 (0, 0, 0).
If we put u = q −1 t, then we can write

Z(s) = |x21 + x32 + x53 |sK dx = [1]/[1, 1][31, 30] ·
X◦
1 − q −3 t(1 − t)(1 + u6 + u10 + u12 + u16 + u18 + u22 ) − q −31 t29 .
In the above two examples the fact that the successive applications of SPF
will terminate by periodicity is known beforehand. In the general case that is
an open problem. We recall that Hironaka’s theorem guarantees the finiteness of
successive monoidal transformations to achieve a desingularization if they are chosen
correctly. We also recall that in the two variable case the process is unique as we
have mentioned in Chapter 3.3. In that case the desingularization of f −1 (0) is so
explicit that it can be used to get precise information about the local zeta fucntion
of f (x). There are several remarkable papers on this subject. We just mention [24],
[26] and above all the papers by D. Meuser [41] and L. Strauss [54].
10.3 A key lemma

We shall formulate a powerful method to compute Z(s) for a relative invariant
f (x) of a subgroup G of GLn (K). Namely, f (x) is an element of K[x1 , . . . , xn ]\K
satisfying f (gx) = ν(g)f (x) with ν(g) in K × for every g in G. This implies ν(gg ) =
ν(g)ν(g ) for every g, g in G. As before we shall assume that the coefficients of
f (x) are in OK but not all in πOK . We define G(OK ) as the intersection of G
and GLn (OK ) and denote by G(Fq ) the image of G(OK ) under the homomorphism
GLn (OK ) → GLn (Fq ). We shall keep on using the notation X ◦ = OK n
, instead
of X(OK ), for X = K . If we denote the images of g, a in G(OK ), X ◦ under
n
G(OK ) → G(Fq ), X ◦ → Fnq respectively by ḡ, ā, then we will have
ga = ḡā.
Theorem 10.3.1 If we choose a subset R of X ◦ such that Fnq becomes a disjoint

union of G(Fq )ξ¯ for all ξ in R, then for Re(s) > 0 we have

|f (x)|K dx =
s ¯
card(G(Fq )ξ) · |f (x)|sK dx.
X◦ ξ∈R ξ+πX ◦
×
Proof. We first observe that ν(g) for every g in G(OK ) is in OK . In fact, if g is in
G(OK ), hence in Mn (OK ), since f (x) is in OK [x1 , . . . , xn ], we see that f (gx) is also
in OK [x1 , . . . , xn ]. Therefore, if ci xi is any term of f (x), then f (gx) = ν(g)f (x)
×
implies that ν(g)ci is in OK for every i. Since ci is in OK for at least one i, we see
that ν(g) itself is in OK . Since ν(g)ν(g ) = 1 and g is in G(OK ), hence ν(g −1 )
−1 −1
×
is in OK , we see that ν(g) is in OK .
◦
Secondly, if a, b are in X , g is in G(OK ), and b̄ = ḡā, then the integrals of
|f (x)|sK over a + πX ◦ and b + πX ◦ are equal. In fact, since ḡā = ga and G(OK )
keeps X ◦ invariant, we have b + πX ◦ = ga + πX ◦ = g(a + πX ◦ ). Furthermore since
the action of G(OK ) on X ◦ is measure preserving, we get

|f (x)|sK dx = |f (x)|sK dx = |f (x)|sK dx.
b+πX ◦ g(a+πX ◦ ) a+πX ◦
We have tacitly used the fact that |f (gx)|K = |ν(g)|K |f (x)|K = |f (x)|K .
We have called the above theorem a key lemma in [28]. Actually, in a slightly
different form it appeared in our paper of 1977 and in the above form with several
applications in [25]. We might mention that it would look more natural to use a
G(OK )-orbital decomposition of X ◦ . However, that approach did not work except
for some very simple cases. The usefulness of the key lemma comes from the facts
that firstly the G(Fq )-orbital decomposition of Fnq is not too difficult to obtain and
secondly in the partial integral of |f (x)|sK over ξ + πX ◦ we can replace f (x) by a
simpler polynomial in a smaller number of variables. This second process has been
formalized as a supplement to the key lemma in [28].
At any rate, we remark that if f¯(ξ)¯ = 0, then the partial integral over ξ + πX ◦
−n
is clearly q . On the other hand, if f (x) is homogeneous of degree d, then the
174 JUN-ICHI IGUSA
partial integral over πX ◦ is q −n td · Z(s). Therefore, in that case, if we denote by

R the subset of R defined by f¯(ξ) ¯ = 0 and ξ¯ = 0, then we will have

[n, d]Z(s) = card(f¯−1 (F×
q ))q −n
+ card(G(F q ) ¯
ξ) · |f (x)|sK dx.
ξ∈R ξ+πX ◦
Furthermore, if f¯(ξ)
¯ = 0 and (∂ f¯/∂xi )(ξ)
¯ = 0 for some i, then the proof of Theorem
10.2.1 shows that
|f (x)|sK dx = [1]q −n t/[1, 1].
ξ+πX ◦
We also remark that if G̃ is a subgroup of GLm (K) for some m, ρ is a homomorphism

from G̃(OK ) to G(OK ), and ρ̄ is a homomorphism from G̃(Fq ) to G(Fq ) which is
compatible with the taking of mod π, i.e.,
ρ̄(g mod π) = ρ(g) mod π
for every g in G̃(OK ), then we can use the decomposition of Fnq into G̃(Fq )-orbits
instead of G(Fq )-orbits. We can also use the obvious fact that if G̃(Fq )ξ̄ denotes
the fixer of ξ¯ in G̃(Fq ) defined by ρ̄(ḡ)ξ¯ = ξ,
¯ then

¯ = card(G̃(Fq )) card(G̃(Fq ) ).
card(ρ̄(G̃(Fq ))ξ) ξ̄
We shall give an application of the key lemma which will illustrate how it works.
We start with some preparations. We take an arbitrary field F and we let GLn (F )
act on Symn (F ) as (g, x) → g ·x = gxt g. We keep in mind that rank(g ·x) = rank(x)
for every g in GLn (F ) and x in Symn (F ). We write n = p + k, where 1 ≤ p ≤ n,
and denote by g1 , g12 , g21 , g2 the p × p, p × k, k × p, k × k entry matrices of g; if
p = n, hence k = 0, it is understood that g = g1 . Also if x1 , x2 are respectively
in Mp (F ), Mk (F ), we denote by x1 ⊕ x2 the element of Mn (F ) with x1 , 0, 0, x2
as its entry matrices. Suppose now that x0 , x0 are in Symp (F ) and det(x0 x0 ) = 0.
Then we have
g · (x0 ⊕ 0) = x0 ⊕ 0
if and only if g1 is in GLp (F ) satisfying g1 · x0 = x0 , g2 is in GLk (F ), g12 is

free in Mp,k (F ), and g21 = 0. We shall show that the GLn (F )-orbit of any x
in Symn (F ) contains an element of the form x1 ⊕ x2 , in which x1 is a diagonal
matrix with det(x1 ) = 0 while all diagonal entries of x2 are 0. Firstly, if g is the
permutation matrix representing (1p), then the (1, 1)-entry of g · x is the (p, p)-
entry of x. Secondly, if x1 , x12 , t x12 , x2 are the p × p, p × k, k × p, k × k entry
matrices of x with det(x1 ) = 0 and if g is the element of SLn (F ) such that t g has
1p , −x−1
1 x12 , 0, 1k as its entry matrices, then
g · x = x1 ⊕ (x2 − t x12 x−1

1 x12 ).
We have only to use these facts repeatedly to verify the above statement.
Lemma 10.3.1 We write n = p + k, where 1 ≤ p ≤ n, and put
Σp = {x ∈ Symn (F ); rank(x) = p}.
Then every GLn (F )-orbit in Σp contains an element of the form ξ = x0 ⊕ 0 with x0

in Symp (F ), hence det(x0 ) = 0. In the special case, where F = Fq the number of
GLn (F )-orbits in Σp is 2 except for the case where p is odd and q is even; in that
case the number is 1. Furthermore, if we put
ck = q −n(n+1)/2 · card(Σp ),
then in all cases we have

ck = q −k(k+1)/2 · [i + k] [2i].
1≤i≤p 1≤2i≤p
Proof. Suppose first that char(F ) = 2. If for every x in Symn (F ) we put Qx (u) =
t
uxu, where u is in F n , then the correspondence x → Qx gives a bijection from
Symn (F ) to the set of all quadratic forms on F n . Furthermore, if we let GLn (F )
act on this set as (g, Q) → (g · Q)(u) = Q(t gu), then the above bijection becomes
equivariant. Therefore by the classical diagonalization of a quadratic form and
by Theorem 9.2.1, we get the lemma except for the expression for ck . If ξ is a
representative of one of the two GLn (F )-orbits in Σp for F = Fq expressed as in
the lemma, then
card(GLn (F ) · x) = card(GLn (F ))/q pk card(GLk (F ))card(O(Qx0 )(F )),
in which Qx0 is on F p . If we use the expressions for card(G(F )) for various G in

Chapter 9.3, then we can easily see that

ck = q −n(n+1)/2 · card(GLn (F ) · ξ)
ξ
is given by the formula in the lemma.

Suppose next that char(F ) = 2 and choose a representative of any GLn (F )-orbit
in Σp of the form x1 ⊕ x2 , in which x1 is a diagonal matrix of degree say i with
det(x1 ) = 0 while all diagonal entries of x2 are 0. Then x2 is in Altl (F ), where
l = n − i, hence g2 · x2 = Jj ⊕ 0 for some g2 in GLl (F ), in which p = i + 2j. We
have thus shown that every GLn (F )-orbit in Σp contains an element ξ of the form
in the lemma with x0 = x1 ⊕ Jj . We shall now take Fq as F . Since every element
of F is a square in F , we may assume that x1 = 1i . A crucial observation is that
there exists an element g of SL3 (F2 ) satisfying
g · (1 ⊕ J1 ) = 13 .
Actually, there are six elements such as g and one of them has 0 only at (2, 3) and
(3, 2), i.e., all other entries are 1. Therefore if p is even, then the GLp (F )-orbit of x0
contains either Jp/2 or ξ0 = 12 ⊕Jp/2−1 while if p is odd, it contains ξ1 = 1⊕J(p−1)/2 .
176 JUN-ICHI IGUSA
We observe that, in the case where p is even, the GLn (F )-orbits of ξ with Jp/2 and
ξ0 as x0 are different because Qξ = 0 for one and Qξ = 0 for another. Therefore,
the number of GLn (F )-orbits in Σp is 2 if p is even and 1 if p is odd.
We shall determine the fixers in GLn (F ) of the above representatives. We have
only to determine the fixers in GLp (F ) of ξ0 , ξ1 . We for a moment change the
notation and replace p by n and write ξ instead of ξ0 , ξ1 . Also we denote by H the
fixer of ξ in GLn (F ). Suppose first that n = 2m, put J = Jm−1 , and denote by a,
b, c, d the entry matrices of g, in which a is in M2 (F ), etc. Then g is in H, i.e.,
gξ t g = ξ for ξ = 12 ⊕ J, if and only if
at a + bJ t b = 12 , at c + bJ t d = 0, ct c + dJ t d = J.
We observe that ct c + dJ t d = J implies that all diagonal entries of ct c are 0,

hence c = (u u) for some u in F 2m−2 . Then ct c = 0, hence dJ t d = J, i.e.,
d is in Sp2m−2 (F ). Similarly, at a + bJ t b = 12 implies that the entries of a are
1 + a0 , a0 , a1 , 1 + a1 for some a0 , a1 in F , hence at c = t c. Then at c + bJ t d = 0
implies t b = Jd−1 c = (v v), in which v = Jd−1 u. This implies bJ t b = 0, hence
at a = 12 , and hence a0 = a1 in a. Conversely, if a0 , u, d are respectively in F ,
F 2m−2 , Sp2m−2 (F ) and a, b, c are defined as above, then g is in H. Furthermore,
the correspondence g → d gives a surjective homomorphism from H to Sp2m−2 (F )
such that the kernel is isomorphic to F 2m−2 × F with the following multiplication:
(u, a0 )(u , a0 ) = (u + u , a0 + a0 + t uJu ).

In particular,

card(H) = q 2m−1 · card(Sp2m−2 (F )) = card(Sp2m (F )) q 2m [2m].
Suppose next that n = 2m + 1 and ξ = 1 ⊕ Jm . Then in a similar but much simpler
way we see that H consists of g = 1 ⊕ d for all d in Sp2m (F ).
If we go back to the original notation, then the order of the fixer of ξ in GLn (F )
is given by
 

 card(Spp (F )) x0 = Jp/2 


 p even

q card(GLk (F )) ·
pk
card(Spp (F ))/q p [p] x0 = ξ0





card(Spp−1 (F )) x0 = ξ1 p odd
Therefore, we get the same ck as before also in this case.
We are now ready to apply the key lemma to the computation of Z(s) for
X = Symn (K) and f (x) = det(x), hence dimK (X) = n(n + 1)/2 and deg(f ) = n.
We write n = p + k, where 1 ≤ p ≤ n, and denote the entry matrices of x by x1 , x12 ,
t
x12 , x2 with x1 in Symp (K), etc. as before. If ξ is an element of X ◦ = Symn (OK )
with x1 = ξ0 , x12 = 0, x2 = 0 such that rank(ξ0 mod π) = p, i.e., det(ξ0 ) is in
×
OK , then the partial integral of |f (x)|sK over ξ + πX ◦ can be written as

q −n(n+1)/2 · dx1 dx12 | det(ξ + πx)|sK dx2 .
Symp (OK )×Mp,k (OK ) Symk (OK )
We shall examine the above {·}. If we denote by g the element of GLn (K) such
that t g has 1p , −π(ξ0 + πx1 )−1 x12 , 0, 1k as its entry matrices, then g is in GLn (OK )
and
g · (ξ + πx) = (ξ0 + πx1 ) ⊕ π(x2 − π t x12 (ξ0 + πx1 )−1 x12 ).
Therefore, if we denote the original Z(s) by Zn (s), then {·} becomes tk Zk (s), hence

| det(x)|sK dx = q −n(n+1)/2 tk Zk (s).
ξ+πX ◦
Therefore, by the key lemma and Lemma 10.3.1 we will have

[n(n + 1)/2, n]Zn (s) = ck tk Zk (s).
0≤k<n
This is a recursion formula by which we can determine Z(s) = Zn (s) starting with
Z0 (s) = 1. The result is as follows:
Proposition 10.3.1 If X = Symn (K) and f (x) = det(x), then


 [2m + 1, 1] n = 2m
Z(s) = 1/[1, 1] · [2i − 1] [2i + 1, 2] ·

1≤i≤m [2m + 1] n = 2m + 1.
Proof. In view of the recursion formula, we have only to show that

ck tk Zk (s)/Zn (s) = 1,
0≤k≤n
in which ck is as in Lemma 10.3.1 with the understanding that cn = q −n(n+1)/2 and

Zk (s) is the Z(s) in the proposition with n replaced by k for 0 ≤ k ≤ n. We separate
cases according as n is even or odd. Then after an elementary computation, we see
that the LHS is equal to

[2i + 2k][2i + 2k + 1, 2] [2i] q −2k (q −1 t2 )k
2
0≤k≤m 1≤i≤m−k
for n = 2m or n = 2m + 1. If in the formal identity (G1) in Chapter 9.6, we replace

n, a, t respectively by m, q −2 , q −1 t2 , then we see that the above expression is indeed
equal to 1.
Proposition 10.3.1 in the case where q is odd is in [25]; the restriction is removed
by the generality of Lemma 10.3.1. At any rate, by T. Kimura [34] we have

bf (s) = (s + (k + 1)/2).
1≤k≤n
178 JUN-ICHI IGUSA
10.4 Z(s) for a Freudenthal quartic

We shall compute Z(s) for the Freudenthal quartic f (x) in Proposition 9.5.1. We
3 6
have identified X = K with K 2 + M3 (K)2 as x = (a0 , b0 ; a, b), in which
a0 = −x123 , b0 = −x456 , a = (x23i x31i x12i ), b = (xi56 xi64 xi45 )
respectively for i = 4, 5, 6 and i = 1, 2, 3; we have dimK (X) = 20 and
f (x) = (a0 b0 − tr(ab))2 − 4(a0 det(b) + b0 det(a) + tr(a# , b# )),
in which a# = Adj(a) = det(a)a−1 , etc. We shall first determine the GL6 (Fq )-
3 6
orbital decomposition of X(Fq ) = Fq . For a moment, we take any field F with
char(F ) = 2. We recall that f (g · x) = det(g)2 f (x) for every g in GL6 (F ) and x
in X. We express g by its 3 × 3 entry matrices α, β, γ, δ as before. We can easily
verify that if β = γ = 0 in g, then
g · (a0 , b0 ; a, b) = (det(α)a0 , det(δ)b0 ; δaα# , αbδ # )
for every α, δ in GL3 (F ).

If ξ is in X(F ) = 3 (F )6 , where F is any field which contains F , and f (ξ) = 0,
then for every g in the fixer of ξ in GL6 (F ) we have det(g) = ±1. Therefore, the
fixer of ξ in SL6 (F ) is the kernel of the homomorphism from the fixer of ξ in
GL6 (F ) to {±1} under g → det(g), hence the corresponding index is at most two.
After this remark, we take ξ0 = (1, θ 3 ; 0, 0) with θ 2 in F × so that f (ξ0 ) = θ 6 is
also in F × . We have shown in Proposition 9.5.1 that the fixer of ξ0 in SL6 (F (θ))
consists of all g with α, δ in SL3 (F (θ)) and β = γ = 0. Furthermore, if g1 is defined
by α = δ = 0, β = θ −1 13 , γ = θ13 , then by Lemma 9.5.1 we see that g1 · ξ0 = ξ0
while det(g1 ) = −1. Therefore, if we denote by H the fixer of ξ0 in GL6 (F (θ)), then
H is the union of SL3 (F (θ)) × SL3 (F (θ)) embedded in GL6 (F (θ)) as (α, δ) → g
above and its coset represented by g1 . In particular, if F = Fq and θ itself is in F ,
say θ = 1, then
card(GL6 (F ) · ξ0 ) = card(GL6 (F ))/2card(SL3 (F ))2 = q 20 · c0 ,

c0 = (1/2)[1][2]+ [3]+ [5].
On the other hand, if θ is not in F , i.e., if θ2 is in F × \(F × )2 , then we take

ξ0 = (−θ 2 , 0; 0, 13 ). If g0 is defined by α = θ13 , β = 13 , γ = −13 , δ = θ −1 13 , then
by Lemma 9.5.1 we see that ξ0 = g0 · (−1/2θ)ξ0 . Therefore, the correspondence
g → g = g0 gg0−1 gives an isomorphism from H to the fixer H of ξ0 in GL6 (F (θ)).
In particular, g1 = g0 g1 g0−1 is in H . Since g1 has α = 13 , δ = −13 , β = γ = 0 as its
entry matrices, it is in GL6 (F ) and det(g1 ) = −1. We observe that the isomorphism
H → H restricted to SL3 (F (θ)) × SL3 (F (θ)) has the following form:
   
α 0 α+δ −θ(α − δ)
g=  → g = (1/2)  
0 δ −θ −1 (α − δ) α+δ
We write α = α0 + θα1 , δ = δ0 + θδ1 with αi , δi in M3 (F ) for i = 0, 1 and observe

that g is in GL6 (F ) if and only if α0 = δ0 , α1 = −δ1 , i.e., if and only if α, δ are
conjugate under the automorphism of F (θ) over F . This shows that the fixer of ξ0
in SL6 (F ) is isomorphic to SL3 (F (θ)). Therefore, if F = Fq , then
card(GL6 (F ) · ξ0 ) = card(GL6 (F ))/2card(SL3 (F (θ)) = q 20 · c0 ,

c0 = (1/2)[1][2][3][5].
We keep in mind that c0 + c0 = [1][10].

We shall next take ξ1 = (0, 0; 13 , 0) = e234 + e315 + e126 in which eijk stands for
ei ∧ ej ∧ ek for 1 ≤ i, j, k ≤ 6, and determine its fixer in GL6 (F ). If for any ξ we
define the p-th polar of f (x) at ξ by

f (ξ + tx) = fp (ξ, x)tp = f (ξ) + f1 (ξ, x)t + . . . ,
p≥0
where t is a variable, then we get f1 (ξ1 , x) = 4x456 . Since for every g in GL6 (F )
we have fp (g · ξ, g · x) = det(g)2 fp (ξ, x), if g · ξ1 = ξ1 and g · x = x , then x456 =
det(g)2 x456 , i.e.,

g4i g5j g6k xijk = det(g)2 x456
1≤i,j,k≤6
for every x in X(F ). We shall for a moment restrict our attention to the submatrix
(γ δ) of g. If we denote by πijk the determinant of the 3 × 3 matrix made up of its
i-th, j-th, k-th columns, then we can rewrite the above condition as

πijk xijk = det(g)2 x456
1≤i<j<k≤6
for every x, hence π456 = det(g)2 = 0. In particular, the 4-th, 5-th, 6-th columns
are linearly independent. Therefore, π145 = π146 = π156 = 0 implies that the
first column is 0. Similarly, we see that the second and the third columns are 0,
hence γ = 0 in g. On the other hand, if v = v1 e1 + v2 e2 + v3 e3 for vi in F ,
then ξ1 v = (v1 e4 + v2 e5 + v3 e6 )e123 . By applying the above g to this we see that
α = det(α)δ, i.e., δα# = 13 . Conversely, if for any α in GL3 (F ) we define δ as
above and take β = γ = 0 in g, then g is in the fixer of ξ1 in GL6 (F ). Furthermore,
if α = δ = 13 and γ = 0 in any g in GL6 (F ), then we can easily verify that
g · ξ1 = ξ1 + tr(β)e123 .
Therefore, the kernel of the surjective homomorphism from the fixer of ξ1 in GL6 (F )
to GL3 (F ) defined by g → α consists of all g with α = δ = 13 , γ = 0, and tr(β) = 0.
If F = Fq , this implies
card(GL6 (F ) · ξ1 ) = card(GL6 (F ))/q 8 card(GL3 (F )) = q 20 · c1 ,

c1 = q −1 [4][5][6].
180 JUN-ICHI IGUSA
We shall next take ξ2 = e126 + e346 . If v = vi ei for vi in F , then
ξ2 v = −(v3 e1236 + v4 e1246 + v5 e1256 + v1 e1346 + v2 e2346 + v5 e3456 ).
Therefore, ξ2 v = 0 if and only if v = v6 e6 . This implies that if g is in the fixer

of ξ2 in GL6 (F ), hence ξ2 (ge6 ) = g(ξ2 e6 ) = 0, we get ge6 = λ−1 e6 for some λ
in F × . Furthermore, if g0 is the element of GL5 (F ) which is obtained from g by
crossing out its 6-th row and column, then g0 · (e12 + e34 ) = λ(e12 + e34 ), i.e.,
g0 (J2 ⊕ 0)t g0 = λ(J2 ⊕ 0). Therefore, g is of the form
 
g11 g12 0
g =  0 g22 0 ,
g31 g32 λ−1
in which g11 is in GL4 (F ) satisfying g11 J2 t g11 = λJ2 , g22 , λ are in F × , and other
entry matrices are free. Conversely, if g is of this form, then g is in GL6 (F ) and
g · ξ2 = ξ2 . We further observe that g → λ defines a surjective homomorphism from
the fixer of ξ2 in GL6 (F ) to F × because the diagonal matrix with λ, 1, λ, 1, 1, λ−1
as its diagonal entries is in the fixer for every λ in F × . Therefore, if F = Fq , then
card(GL6 (F ) · ξ2 ) = card(GL6 (F ))/q 11 [1]2 card(Sp4 (F )) = q 20 · c2 ,

c2 = q −5 [3][5][6]/[1].
Finally, if ξ3 = e123 , then Lemma 9.5.1 shows that the fixer of ξ3 in GL6 (F ) consists
of all g with α in SL3 (F ), δ in GL3 (F ), γ = 0, and β free in M3 (F ). If F = Fq ,
therefore, we get
card(GL6 (F ) · ξ3 ) = card(GL6 (F ))/q 10 [1]card(SL3 (F ))2 = q 20 · c3 ,

c3 = q −10 [2]+ [3]+ [5].
We shall show, by using fp (g · ξ, g · x) = det(g)2 fp (ξ, x) for every g, that the

GL6 (F )-orbits of ξ0 , ξ0 , ξ1 , ξ2 , ξ3 are different. Firstly, f (ξ0 ) is in (F × )2 while
f (ξ0 ) is in F × \(F × )2 . Secondly, f (ξ1 ) = f (ξ2 ) = f (ξ3 ) = 0 and f1 (ξ1 , x) = 0,
f1 (ξ2 , x) = f1 (ξ3 , x) = 0. Thirdly, we can easily see that f2 (ξ2 , x) and f2 (ξ3 , x) are
nondegenerate quadratic forms respectively in five and one variables. Therefore, the
five GL6 (F )-orbits are certainly different. Furthermore, in the case where F = Fq
we have
c0 + c0 + c1 + c2 + c3 = [20] = q −20 · card(X(F )\{0}).
By putting these together we see that X(F )\{0} becomes the disjoint union of the
GL6 (F )-orbits of ξ0 , ξ0 , ξ1 , ξ2 , ξ3 . Therefore, by the key lemma we get

(∗) [20, 4]Z(s) = [1][10] + c1 · [1]t/[1, 1] + ci · |f (ξi + πx)|sK dx.
i=2,3 X◦
We have only to compute the above two partial integrals.

Suppose first that i = 2. If we apply the even permutation {1, 3, 4, 5, 6} →

{3, 5, 6, 4, 1} to the subscripts of ei , then ξ2 = e126 +e346 becomes ξ = −e123 +e156 =
(1, 0; 0, u), in which u = 1 ⊕ 0, and the integral to be computed becomes

|f (ξ + πx)|K dx =
s
da0 da |f (ξ + πx)|sK db0 db .
X◦ OK ×M3 (OK ) OK ×M3 (OK )
In the above {·} we have ξ + πx = (a0 , πb0 ; πa, u + πb), in which a0 = 1 + πa0 .
Therefore, if we define an element g of SL6 (OK ) by α = δ = 13 , β = 0, and
γ = (a0 )−1 πa, then, as we have seen in Chapter 9.5, we have
g · (ξ + πx) = (a0 , πb0 ; 0, u + πb ),
in which b = b + (a0 )−1 πa# and
b0 = b0 − (a0 )−1 tr(a(u + πb)) − 2(a0 )−2 π 2 det(a).
We observe that (b0 , b) → (b0 , b ) gives a measure-preserving bijection from OK ×

M3 (OK ) to itself. Since f (a0 , πb0 ; 0, u+πb ) = (πa0 b0 )2 −4a0 det(u+πb ), therefore
we get

|f (ξ + πx)|sK dx = |(πb0 )2 − det(u + πb)|sK db0 db.
X◦ OK ×M3 (OK )
The above integral can be reduced to an integral in Corollary 10.2.1 by the method
we have repeatedly used in section 10.1. We write b = (b1 b ) with a 3 × 2 submatrix
b and express the integral as an integral by db followed by an integration by db0 db1 .
Since the first column of u + πb can be completed to an element of SL3 (OK ), if b
denotes the bottom 2 × 2 submatrix of b , then the integral by db becomes an
integral by db . In that way we get

|f (ξ + πx)|sK dx = t2 · |b20 + det(b )|sK db0 db
X◦ OK ×M2 (OK )
= t2 · [1][5, 1]/[1, 1][5, 2].
Suppose next that i = 3 and put ξ = −ξ3 = (1, 0; 0, 0). Then by the first part
of the above argument we get

|f (ξ + πx)|K dx =
s
|(πb0 )2 + det(πb)|sK db0 db.
X◦ OK ×M3 (OK )
We shall compute this integral in the following lemma:

Lemma 10.4.1 If we define P (t) as

[1] 1 − (1 + q −1 − q −3 )q −3 t + (1 + q −1 − q −2 − q −3 − q −4 )q −3 t2 + q −12 t3 ,
then
|y02 + π det(y)|sK dy0 dy = P (t)/[1, 1][5, 2][7, 2].
OK ×M3 (OK )
182 JUN-ICHI IGUSA
Proof. We shall denote the integral in question by Z0 and compute it by a successive

application of SPF. In doing so we shall use the following facts: If we let GL3 (F ) ×
GL3 (F ) act on M3 (F ) as (g1 , g2 ) · y = g1 yg2−1 , then M3 (F ) splits into four orbits
Σ0 , Σ1 , Σ2 , Σ3 , in which Σp consists of all y with rank(y) = p and it is represented
by 1p ⊕ 0; if k = 3 − p, the fixer of 1p ⊕ 0 consists of all (g1 , g2 ) with the entry
matrices ai , bi , ci , di of gi satisfying a1 = a2 in GLp (F ), d1 , d2 in GLk (F ), c1 = 0,
b2 = 0, and b1 , c2 free for 0 ≤ p ≤ 3. Furthermore, all partial derivatives of det(y)
vanish at y if and only if y # = 0, i.e., rank(y) ≤ 1. If F = Fq , then the above
structure of the fixer of 1p ⊕ 0 implies
q −9 card(Σp ) = q −k [k + 1]2 . . . [3]2 /[1] . . . [p]

2
= q −9 , q −4 [3]2 /[1], q −1 [2][3]2 /[1], [1][2][3]

respectively for p = 0, 1, 2, 3. Moreover since the map det : M3 (F ) → F is surjective
with card(det−1 (c)) = card(SL3 (F )) or q 9 − card(GL3 (F )) according as c = 0 or
c = 0, the number of (y0 , y) satisfying y02 + det(y) = 0 is
(q − 1) · card(SL3 (F )) + (q 9 − card(GL3 (F )) = q 9 .
Also we shall reduce the integral over M3 (OK ) to that over M2 (OK ) similarly as in
the previous case.
If now we put D = OK × M3 (OK ) and D = OK × M2 (OK ), then we have

Z0 = [1] + q −1 tZ1 , Z1 = |πy02 + det(y)|sK dy0 dy;
D
Z1 = [1][2][3] + [2][3]2 q −1 t/[1, 1] + [3]2 q −4 t/[1] · Z2 + q −9 tZ3 ,

Z2 = |y02 + π det(y )|sK dy0 dy , Z3 = |y02 + π 2 det(y)|sK dy0 dy;
D
D
−1
Z2 = [1] + q tZ21 , Z21 = |πy02 + det(y )|sK dy0 dy ;
D
2 −1 −4
Z21 = [1][2] + [2] q t/[1, 1] + q tZ2 ;

Z3 = [1] + q −1 t2 Z31 , Z31 = |y02 + det(y)|sK dy0 dy;
D
Z31 = [1] + ([1][9] − [3]2 q −4 )q −1 t/[1, 1] + [3]2 [5, 1]q −5 t2 /[1, 1][5, 2] + q −10 t2 Z0 .
At the last stage we have used Corollary 10.2.1 as in the previous case. We have
only to put these together.
If we incorporate the computed integrals for i = 2, 3 into (∗), then after an
elementary computation we get the following result:
Proposition 10.4.1 If q is odd and
f (x) = (a0 b0 − tr(ab))2 − 4(a0 det(b) + b0 det(a) + tr(a# b# ))
for x = (a0 , b0 ; a, b) in X = K 2 + M3 (K)2 , then

|f (x)|sK dx = [1][5]C(t) [1, 1][5, 2][7, 2][10, 2],
X◦
in which C(t) is
[5]+ − (1 + q −1 + q −2 − q −6 )q −5 t + (1 − q −4 − q −5 − q −6 )q −6 t2 + [5]+ q −12 t3 .
We recall that if C is any composition algebra over K and X = K 2 + H3 (C)2 ,
then the corresponding Freudenthal quartic f (x) on X is defined in Chapter 9.5.
We have just computed Z(s) in one case. Actually all Z(s) have been computed in
[25]. If for any m, n in N satisfying m ≥ n and for variables a, t, we put
Cm,n (a, t) = (1+am ) − (1 + am−n + a2m−2n − a2m−n )an+1 t
+ (1 − an − am − a2m−n )am+1 t2 + (1 + am )a2m+2 t3 ,
then for C = (C1), (C2), (C3), (C4) in Chaper 9.4, we have

|f (x)|sK dx = [1][3k/2 + 2]C3k/2+2,k+2 (q −1 , t)/
X◦
[1, 1][k + 3, 2][2k + 3, 2][3k + 4, 2],
in which k = dimK (C) = 1, i.e., k = 2, 4, 8; if k = 1, then it becomes

[1][4][7, 1] [1, 1][4, 2][7, 2].
The computation has been done uniformly. Instead of (∗) we have
[6k + 8, 4]Z(s) = [1][3k + 4] + [1][k + 2][3k/2 + 2][2k + 2]q −1 t/[1, 1]
+ c2 I2 + c3 I3 ,
in which
c2 = q −(k+3) [3k/2][3k/2 + 2][2k + 2]/[k/2],
c3 = q −(3k+4) [k/2 + 1]+ [k + 1]+ [3k/2 + 2],
I2 = [1][k + 3, 1]t2 /[1, 1][k + 3, 2],
I3 = ([1]t2 /[1, 1][k + 3, 2][2k + 3, 2]){1 − (1 + q −k/2 − q −k−1 )q −k/2−2 t
+ (1 + q −k/2 − q −k/2−1 − q −k−1 − q −3k/2−1 )q −k/2−2 t2 + q −3k−6 t3 }
for k = 2, 4, 8. This implies the above result. Actually Z(s) has been computed
also in some “twisted cases.” Furthermore, in the case where k = 8 not only Z(s)
but also Z(ω) was computed earlier via Weil’s function F ∗ (i∗ ). At any rate, by T.
Kimura [34] we have
bf (s) = (s + 1)(s + (k + 3)/2)(s + k + 3/2)(s + 3k/2 + 2)
for k = 1, 2, 4, 8.
We further mention that Z(s) for the norm form has been computed in all twisted
cases. If C is the unique unramified quadratic extension of K, i.e., the field K2 in
the notation of Chapter 11.6, the result is stated in [25]. Furthermore the norm
form of any associative simple K-algebra can be handled in the same way as det(x)
while every octonian K-algebra is isomorphic to the one in (C4). That leaves only
the case where C is either a ramified quadratic extension of K or a quaternion K-
algebra not isomorphic to M2 (K). In those rather difficult cases the computation
has been carried out by M. M. Robinson [46].
184 JUN-ICHI IGUSA
10.5 Z(s) for the Gramian det(t xhx)

We shall start with some observations on a quadratic map. We take an m-dimen-
sional vector space V over any field F with char(F ) = 2 and a nondegenerate
quadratic form Q on V . We put X = V n , where m ≥ 2n, and let G = O(Q) ×
GLn (F ) act on X, Y = Symn (F ) as
(g, g ) · x = gx tg , (g, g ) · y = g · y = g y tg .
We write x = (x1 . . . xn ) with xi in V , denote the (i, j)-entry of y by yij , and define
a quadratic map iX : X → Y as
y = iX (x), yij = Q(xi , xj )
for 1 ≤ i, j ≤ n. We observe that iX is G-equivariant. We shall denote by X
the G-invariant subset of X defined by rank(x) = n, i.e., by the condition that the
column space x1 , . . . , xn of x is n-dimensional. We observe that if x = (x1 . . . xn )
is also in X with iX (x) = iX (x ), then the correspondence xi → xi for 1 ≤ i ≤ n
gives an isometry from x1 , . . . , xn to x1 , . . . , xn , which extends to an element
of O(Q) by the Witt theorem. Therefore, if {η} is a complete set of representatives
of GLn (F )-orbits in Y and if ξ is chosen arbitrary from X ∩ i−1 X (η), then the set
{ξ} forms a complete set of representatives of G-orbits in X . In the following we
shall determine the fixer Gx in G of any x in X .
If rank(iX (x)) = p, then for our purpose we may assume that
y = iX (x) = y0 ⊕ 0,
in which y0 is in Symp (F ) necessarily with det(y0 ) = 0. If we write n = p + k and

denote by g11 , g12 , g21 , g22 the p × p, p × k, k × p, k × k entry matrices of g , then the

fixer GLn (F )y of y in GLn (F ) is defined by g11 in GLp (F ) satisfying g11 y0 t g11 = y0 ,

g22 in GLk (F ), g21 = 0, and g12 free in Mp,k (F ). We shall choose an F -basis for V
depending on x. We write
x1 , . . . , xn = W0 + W
such that Q|W0 is nondegenerate, Q|W = 0, and W is contained in W0⊥ . We then
have dimK (W0 ) = p and dimK (W ) = k. We know that we can find a k-dimensional
subspace W of W0⊥ such that Q|W = 0 and H = W + W is hyperbolic. In that
way we get an orthogonal decomposition
V = W0 ⊕ (W + W ) ⊕ W1 ,
in which m1 = dimK (W1 ) is equal to m−p−2k, hence also to m−n−k = m−2n+p.
If we choose F -bases for W0 , W , W , W1 suitably, then V can be identified with
F m and Q(v) for v in F m can be written as (1/2)t vhx v with
 
h0 0 0 0
0 0 1k 0 
hx =  0 1k 0
,
0
0 0 0 h1
in which h0 , h1 are respectively in Symp (F ), Symm1 (F ) and det(h0 ) det(h1 ) = 0.

Since y = t xhx x, we also have

x0 x11 0 t
x= , x0 = , x11 h0 x11 = y0
0 x21 x22
with x11 , x22 , x21 respectively in GLp (F ), GLk (F ), Mk,p (F ).

We are ready to determine the fixer Gx . If (g, g ) is an element of Gx , hence
gx = x t (g )−1 , then since g21

= 0 in g we will have

α 0
x0 t (g )−1 x−1
0 =
γ δ
−1 −1
with α = x11 t (g11 ) x11 , etc. and
 
α 0 g13 g14
γ δ g23 g24 
g= 0

0 g33 g34 
0 0 g43 g44
with g33 in Mk (F ), etc. Conversely, if g is of this form and if g is defined as above

in terms of α, γ, δ and x0 , then gx = x t (g )−1 . Furthermore the condition that g
is in GLn (F )y becomes α in GLp (F ) satisfying t αh0 α = h0 and δ in GLk (F ). We
shall examine the condition that g is in O(Q), i.e., t ghx g = hx . We observe that if
g14 , g24 , g34 , g43 are all 0, g33 = t δ −1 , g44 = 1m1 , and
g13 = −αh−1t
0 (δ
−1
γ), g23 = −(1/2)γh−1t
0 (δ
−1
γ),
then t ghx g = hx . Therefore, the homomorphism Gx → GLn (F )y defined by

(g, g ) → g is surjective. Furthermore, if α = 1p , γ = 0, δ = 1k , then the con-
dition on g becomes that g13 , g14 , g34 are all 0, g33 = 1k , t g44 h1 g44 = h1 , and g23 ,
g24 , g43 satisfy
g23 + t g23 + t g43 h1 g43 = 0, g24 + t g43 h1 g44 = 0.
This gives a description of the kernel Kx of Gx → GLn (F )y . In fact, if we associate

quadratic forms Q0 , Q1 respectively on F p , F m1 to h0 , h1 as Q is associated with
hx and further if we put
a = (1/2)(g23 − t g23 ), b = g43 , [b, b ] = −(1/2)(t bh1 b − t b h1 b),
then we get a surjective homomorphism Kx → O(Q1 ) as g → g44 such that its kernel
is isomorphic as g → (b, a) to Mm1 ,k (F )×Altk (F ) with the following multiplication:
(b, a)(b , a ) = (b + b , a + a + [b, b ]).
We have seen such a group in the proof of Lemma 10.3.1. At any rate Gx has now
been determined. We keep in mind that
d(Q) = (−1)pm1 d(Q0 )d(Q1 )

186 JUN-ICHI IGUSA
up to a factor in (F × )2 .
We now take Fq as F , write X (F ) instead of X , and choose a complete set of
representatives of G(F )-orbits in X (F ). In doing so, we shall assume that d(Q) is
in (F × )2 if m is even so that, by Theorem 9.2.1, the anisotropic kernel of Q becomes
0. Then with respect to a suitable F -basis for V we can identify V with F m so that
we can write Q(v) = (1/2)t vhv, in which
 
0 1n 0
h = 1n 0 0 
0 0 h
for some h . We then have X = Mm,n (F ) and iX (x) = t xhx for all x in X. We
define Σp as in Lemma 10.3.1 and choose representatives ηp of GLn (F )-orbits in
Σp of the form ηp = y0 ⊕ 0 with y0 in Symp (F ), hence det(y0 ) = 0. We keep in
mind that the number of ηp is 2 for p > 0 and 1 for p = 0. If we define an element
ξp of Mm,n (F ) as ξp = t (ηp 1n 0), then ξp is in X (F ) and (1/2)iX (ξp ) = ηp .
Therefore, {ξp } for all ηp and for 0 ≤ p ≤ n forms a complete set of representatives
of G(F )-orbits in X (F ).
We shall compute card(G(F ) · ξp ) for each ξp . We have card(G(F ) · ξp ) =
card(G)/card(Gx ) for x = ξp , in which card(G) = card(O(Q)) card(GLn (F )) and
Gx has been made explicit. Therefore, if we put
γ = q −m(m−1)/2 · card(O(Q))
and define γ0 , γ1 similarly for O(Q0 ), O(Q1 ), then we can easily verify that

q −mn · card(G(F ) · ξp ) = q −k(k+1)/2 · [k + i] · γ/γ0 γ1 .
1≤i≤p
pm1
Furthermore, since d(Q0 )d(Q1 ) = (−1) d(Q), where d(Q) = 1 if m is even, up to
a factor in (F × )2 , by using the formulas in Chapter 9.3 we get

γ =2· [2i]
1≤2i<m
multiplied by [m/2] if m is even and

1/γ0 γ1 = 1 2 [2i] · [2i]
ηp 1≤2i≤p 1≤2i≤m1
multiplied by [m/2 − k]+ if m, p are both even.

We now take a p-adic field with odd q = card(OK /πOK ), define a quadratic
form Q on V = K m as Q(v) = (1/2)t vhv for any h in Symm (OK ) with det(h)
×
in OK , put X = V n = Mm,n (K) for m ≥ 2n and define f (x) as the Gramian of
iX (x) = t xhx, i.e., as
f (x) = det(t xhx)
× 2
for x in X. We shall assume that d(Q) is in (OK ) if m is even and compute

Z(s) = |f (x)|sK dx
X◦
for Re(s) > 0. If we define an open subset (X )◦ of X ◦ by rank(x mod π) = n,

then the computation of Z(s) will be reduced to that of

Z0 (s) = |f (x)|sK dx.
(X )◦
In fact, if we denote by I the set of all i = (i1 , . . . , in ) in Zn , where 1 ≤ i1 < . . . <

in ≤ m, by πi (x) the determinant of the n × n submatrix of x obtained by crossing
out its k-th rows for k = i1 , . . . , in , and further by hi,j for j = (j1 , . . . , jn ) also in I
the determinant of the n × n submatrix of h obtained by crossing out its k-th rows
for k = i1 , . . . , in and l-th columns for l = j1 , . . . , jn , then we can write

f (x) = det(t xhx) = hi,j πi (x)πj (x).
i,j∈I
Therefore, the reduction of Z(s) to Z0 (s) can be done by the following lemma:
Lemma 10.5.1 We only assume that m ≥ n in X = Mm,n (K) and that f (x) is
any homogeneous polynomial of degree d in πi (x) with coefficients in K. Then

|f (x)|sK dx = 1 [m − k + 1, d] · |f (x)|sK dx.
X◦ 1≤k≤n (X )◦
We shall postpone the proof of this lemma to section 10.6 and quickly finish the
computation of Z(s), i.e., that of Z0 (s). We shall use ηp = y0 ⊕ 0, ξp = t (ηp 1n 0)
for the liftings of the previous ηp , ξp to Symn (OK ), (X )◦ so that, e.g., det(y0 ) ≡ 0
mod π instead of det(y0 ) = 0; also we shall normalize h as before with entries in
OK this time. We observe that if we denote by x22 the k × k submatrix of x with
its (i, j)-entries for p < i, j ≤ n as the entries of x22 , then we will have

y + πy πy12
f (ξp + πx) = 2n · det 0 t 11 ,
π y12 πy22
in which the entries of y11 , y12 , y22 are SRP’s in the entries of x. We further observe
that y22 ≡ (1/2)(x22 + t x22 ) mod π and the above determinant is equal to
π k det(y0 + πy11 ) det(y22 − π t y12 (y0 + πy11 )−1 y12 )
with y0 + πy11 in GLp (OK ). Therefore, we get

|f (ξp + πx)|sK dx = tk · | det(y22 )|sK dy22 .
X◦ Symk (OK )
If we put
ck = q −mn card(G(F ) · ξ¯p )
ηp
for F = Fq and ξ¯p = ξp mod π, then by using the key lemma, the above result on
ck and Proposition 10.3.1 we get the following preliminary result:
188 JUN-ICHI IGUSA
Proposition 10.5.1 We assume that m ≥ 2n in X = Mm,n (K), denote by h

×
an element of Symm (OK ) with det(h) in OK such that (−1)m(m−1)/2 det(h) is in
× 2
(OK ) if m is even, and put

Z(s) = | det(t xhx)|sK dx.
X◦
If further we write n = p + k and put m1 = m − n − k, then we have

Z(s) = 1 [m − i + 1, 2] · Z0 (s), Z0 (s) = ck tk · Ik ,
1≤i≤n 0≤k≤n
in which

ck = q −k(k+1)/2 · [k + i] · [2i] [2i] · [2i]
1≤i≤p 1≤2i<m 1≤2i≤p 1≤2i≤m1
 & '

 [m/2 − k]+ p even
 [m/2] m even
1 p odd
·



1 m odd,
&
[k + 1, 1] k even
Ik = 1 [1, 1] · [2i − 1]/[2i + 1, 2] ·
[k] k odd
1≤2i≤k
for 0 ≤ k ≤ n.
10.6 An integration formula

We shall prove an integration formula which implies Lemma 10.5.1. Since the
formula has some generality, we shall change our notation and state it independently
of that lemma. We take an arbitrary field F and denote by X the subset of Mm,n (F ),
where m ≥ n, consisting of all x with rank(x) = n. Namely the new X is the old
X . The condition means that x can be completed by an additional r = m − n
columns to an element of GLm (F ). We define I and πi (x) for every i in I as in
section 10.5. We order the set I lexicographically and define a map f from X to
F N , where N = card(I), as
f (x) = (πi (x))i∈I = t (πi0 (x) . . . ),
in which i0 = (1, 2, . . . , n). We put G = GLm (F ), H = SLn (F ), and denote the

n × n, n × r, r × n, r × r submatrices of any g in G by g11 , g12 , g21 , g22 . We define a
homomorphism ρ from G to GLN (F ) so that f (gx) = ρ(g)f (x) for every g in G and
x in X. The (i, j)-entry of ρ(g) for i = (i1 . . . , in ), j = (j1 . . . , jn ) is defined by g
exactly in the same way as hi,j is defined by h in section 10.5. If we put Y = f (X),
then G acts on Y as g · y = ρ(g)y for every g in G and y in Y . We see by definition
that the actions of G on X, Y are transitive and equivariant. Furthermore, if we
put
ξ = t (1n 0), η = f (ξ) = t (1 0 . . . 0),
then the fixer Gξ of ξ in G is clearly defined by g11 = 1n , g21 = 0. We shall show

that the fixer Gη of η in G is defined by g11 in H, g21 = 0. Since such an element
is clearly in Gη , we have only to prove the converse. We shall exclude the simple
case where n = 1. If g is in Gη , then f (gξ) = t (1 0 . . . 0) and gξ = t (t g11 t g21 ).
Therefore, if we denote the k-th row of gξ by vk for all k, then v1 , . . . , vn are the
n rows of g11 , and det(g11 ) = 1. In particular, they are linearly independent, hence
vk = c1 v1 +. . .+cn vn with c1 , . . . , cn in F depending on k. If we take k > n and put
i = (1, . . . , j − 1, j + 1, . . . , n, k) for any 1 ≤ j ≤ n, then πi (gξ) = (−1)n−j cj = 0,
hence cj = 0, and hence g21 = 0.
We shall derive two consequences from the above information on Gξ , Gη . Firstly
for every x in X we have f −1 (f (x)) = xH. Since xH is clearly contained in
f −1 (f (x)), we shall prove the converse. We can write x = gξ and any x in f −1 (f (x))
as gg ξ for some g, g in G. Then gg · η = g · η, hence g · η = η. As we have seen,
this implies that (g )11 is in H, (g )21 = 0, hence
x = gg ξ = gξ(g )11 = x(g )11 .
Secondly, if F = Fq , then by Chapter 9.3 we have

card(X(F )) = card(G(F ))/card(Gξ (F )) = q mn · [m − k + 1]
1≤k≤n
while card(Y (F )) = card(X(F ))/card(SLn (F )).

We now take a p-adic field K as F . Then G, H become locally compact groups
and X is a locally compact space because they are either open or closed in locally
compact spaces. We shall show that Y is also locally compact. Since the action of
G on Y is bicontinuous and transitive, it will be enough to show that Y contains a
× N −1
compact open subset. If we put V = Y ∩ (OK × OK ), then V is an open subset
of Y . We shall show that V is compact. If y is in V and x1 is the top n × n
×
submatrix of any x in f −1 (y), then det(x1 ) = yi◦ is in OK , hence x1 is in GLn (K),
and y = f (xx1 )yi◦ . If we denote the k-th row of xx−1
−1
1 by vk for all k, then
vk = c1 v1 + . . . + cn vn with the j-th entry of vk as cj for 1 ≤ j ≤ n. Furthermore,
if we put i = (1, . . . , j − 1, j + 1, . . . , n, k), then
yi = πi (xx−1
1 )yi◦ = (−1)
n−j
cj yi◦
×
with yi◦ in OK and yi in OK , hence cj is in OK . Therefore, xx−1
1 with 1n as its top
n × n submatrix is in Mm,n (OK ), hence

1n ×
V =f OK .
Mr,n (OK )
×
Since OK , OK are compact and f is continuous, we see that V is compact.
Once we know that X, Y are locally compact, then by Lemma 7.3.1 we can
identify X, Y respectively with G/Gξ , G/Gη under g → x = gξ, g → y = g · η. We
shall denote by X ◦ the compact open subset of X consisting of all x in Mm,n (OK )
×
such that rank(x mod π) = n, i.e., with πi (x) in OK for some i. The condition
190 JUN-ICHI IGUSA
means that x can be considered as the left m × n submatrix of an element of

G(OK ) = GLm (OK ). If we put Y ◦ = f (X ◦ ), then G(OK ) acts transitively and
equivariantly on X ◦ , Y ◦ .
We shall introduce relatively invariant measures on the homogeneous spaces X,
Y . We shall first determine the modules of G, Gξ , Gη by using Lemma 7.4.2. We
take any p and denote by µp the normalized Haar measure on K p and on its open
subsets as before. If we put Gp = GLp (K), Hp = SLp (K) so that G = Gm , H = Hn ,
then their modules are 1. In fact,
dg = | det(g)|−p
K µp2 (g)
gives a Haar measure on Gp and d(gg0 ) = dg for every g0 in Gp , hence ∆Gp = 1.

Since the homogeneous space Gp /Hp is a group, in fact the group G1 = K × , it has
an invariant measure. Therefore, by Proposition 7.2.1, we have ∆Gp |Hp = ∆Hp ,
hence ∆Hp = 1. If g is any element of Gξ , hence g11 = 1n , g21 = 0, then
dg = µnr (g12 ) · | det(g22 )|−r

K µr 2 (g22 )
gives a Haar measure on Gξ and d(gg0 ) = | det(g0 )|nK dg for every g0 in Gξ , hence
∆Gξ (g) = | det(g)|nK . If g is an element of Gη , hence g11 is in H, g21 = 0, and if for
a moment µH denotes any Haar measure on H, then
dg = µH (g11 ) · µnr (g12 ) · | det(g22 )|−r

K µr 2 (g22 )
gives a Haar measure on Gη and d(gg0 ) = | det(g0 )|nK dg for every g0 in Gη , hence
∆Gη (g) = | det(g)|nK .
The above information about the modules on G, Gξ , Gη implies, in view of
Proposition 7.2.1, that X, Y have measures µX , µY = 0 satisfying
µX (gx) = | det(g)|nK µX (x), µY (g · y) = | det(g)|nK µY (y)
for every g in G. Actually, the restriction of µmn on Mm,n (K) to its open subset X
is such a measure µX and we simply take µX = µmn |X. Furthermore, we normalize
µH as
µH (H(OK )) = q −(n −1) card(H(F )) =
2
[k],
1<k≤n
in which H(OK ) = SLn (OK ) and F = Fq , and µY as
µY (Y ◦ ) = µX (X ◦ )/µH (H(OK )).
We shall show that

µX (X ◦ ) = [m − k + 1].
1≤k≤n
In general, for every e = (e1 , . . . , en ) in Nn we denote by E(e) the union of all

G(OK )x such that the upper n × n submatrix of x is an upper-triangular matrix in
Mn (OK ) with π e1 , . . . , π en as its diagonal entries and the lower r × n submatrix
of x is 0. If x∗ = gx for some g in G(OK ) has a similar form, then we see that
the upper n × n submatrix of x∗ has the same diagonal entries as x. Therefore,

X0 = X ∩ Mm,n (OK ) becomes the disjoint union of all compact open subsets E(e).
Furthermore, E(0) = G(OK )ξ = X ◦ . We shall show that

µX (E(e)) = [m − k + 1]q −(m−k+1)ek .
1≤k≤n
We denote the first column of any x in Mm,n (K) by x1 . If x is in E(e), then

x1 = π e1 gε1 for some g in G(OK ) and ε1 = t (1 0 . . . 0). Furthermore, suppose
that x1 = π e1 ε1 for some x in Mm,n (OK ) and denote by x the (m − 1) × (n − 1)
submatrix of x obtained by crossing out its first row and column. Then we see that
x is in E(e) if and only if x is in E(e ) for e = (e2 , . . . , en ) relative to X defined
similarly as X for m − 1, n − 1 instead of m, n. Therefore, by an induction on n
we get

µX (E(e)) = [m]q −me1 · µX (E(e )) = [m − k + 1]q −(m−k+1)ek .
1≤k≤n
We are ready to prove the following lemma.
Lemma 10.6.1 If Φ is in D(X), i.e., if Φ is a C-valued locally constant function

on X with compact support, then

Φ(x)µX (x) = Φ(xh)µH (h) µY (f (x)).
X Y H
Proof. By Proposition 7.2.1 we have

ϕ(g)| det(g)|nK dg = ϕ(gh) dk µX (gξ)
G X Gξ

= ϕ(gk ) dk µY (g · η)
Y Gη
for every ϕ in D(G) provided that the Haar measures dg, dk, dk respectively on
G, Gξ , Gη are suitably normalized. If we put

Φ(x) = ϕ(gk) dk,
Gξ
where x = gξ, then we know, cf. loc. cit., that the correspondence ϕ → Φ gives a
C-linear surjection from D(G) to D(X). Furthermore, if we put h = (k )11 , then
we have gk kξ = xh. Therefore, by Proposition 7.2.1 we have

ϕ(gk ) dk = ϕ(gk k) dk µH (h) = Φ(xh)µH (h).
Gη H Gξ H
By putting these together we get the formula in the lemma up to a factor in R×

+
independent of Φ. We shall show that this factor is 1.
192 JUN-ICHI IGUSA
If x0 is in xH ∩ X ◦ for some x in X, then xH ∩ X ◦ = x0 H ∩ X ◦ and some n × n

submatrix say xi of x0 is in GLn (OK ). Therefore, if x0 h for some h in H is in X ◦ ,
then xi h is in Mn (OK ),hence h is in Mn (OK ), and hence in SLn (OK ) = H(OK ).
We have thus shown that xH ∩ X ◦ = ∅ implies xH ∩ X ◦ = x0 H(OK ) for some
x0 in X ◦ and x0 H ∩ X ◦ = x0 H(OK ) for every x0 in X ◦ . Therefore, if in the
integration formula we take the characteristic function of X ◦ as Φ, then its RHS
becomes µH (H(OK ))µY (Y ◦ ). Since the LHS is µX (X ◦ ), the constant factor is 1.
We know that we can replace Φ in the above lemma by any continuous integrable
function on X. Therefore, if Φ, ϕ are continuous functions respectively on X, Y
such that Φ(x)ϕ(f (x)) is integrable on X, then the lemma implies

ϕ(f (x))Φ(x)µX (x) = ϕ(y) Φ(xh)µH (h) µY (y),
X Y H
in which y = f (x). If we take the characteristic function of X ◦ as Φ, then the last

part of the proof of Lemma 10.6.1 implies

ϕ(f (x))µX (x) = µH (H(OK )) ϕ(y)µY (y).
X◦ Y◦
Before we proceed further, we make the following remark. The relatively invariant
measure µY on Y remains relatively invariant under the normalizer of ρ(G) in
GLN (K). In particular, if c is in K × , then µY (cy) differs from µY (y) by a factor in
R×+ independent of y. By Lemma 7.2.1 this factor gives a continuous homomorphism
from K × to R× + . Therefore, it is of the form |c|K for some σ in R independent of c.
σ
On the other hand, if we put g = π1m , then ρ(g)y = π n y and | det(g)|nK = q −mn ,
hence µY (π n y) = q −mn µY (y). By putting these together we see that σ = m, hence
µY (cy) = |c|m
K µY (y)
for every c in K × .
We now express Y as the disjoint union of π k Y ◦ for all k in Z. Then by using
the above remark we can rewrite the integration formula as

(∗) ϕ(f (x))Φ(x)µX (x) = q −mk · ϕ(π k y) Φ(xh)µH (h) µY (y),
X k∈Z Y◦ H
in which f (x) = π k y. In particular, if we take the characteristic function χ of

X0 = X ∩ Mm,n (OK ) as Φ, then (∗) becomes
& '
ϕ(f (x))µX (x) = q −mk · ϕ(π k y) χ(xh)µH (h) µY (y)
X0 k≥0 Y◦ H
in which f (x) = π k y. The above {·} can be made explicit, e.g., as follows: We
observe that it is invariant under x → gx for any g in G(OK ), hence it is independent
of y. We take any k from N and in (∗) we replace ϕ by the constant 1 and Φ by
the characteristic function of the subset of X0 defined by the condition that f (x)
is in π k Y ◦ . We observe that this subset is the disjoint union of E(e) for all e in Nn
satisfying |e| = e1 + . . . + en = k. Therefore we get

µX (E(e)) = q −mk · · · µY (Y ◦ ).
|e|=k
We put all these together and use the fact that Mm,n (OK )\X0 is of measure 0 by
µX (X0 ) = 1. In that way we get the following proposition:
Proposition 10.6.1 Let ϕ denote any C-valued continuous function on the image
of Mm,n (K) under x → (πi (x)) and Um,n the compact open subset of Mm,n (OK )
defined by rank(x mod π) = n. Then we have

ϕ(f (x)) dx = q −(m−k+1)ek ϕ(π |e| f (x)) dx,
Mm,n (OK ) e∈Nn 1≤k≤n Um,n
in which e = (e1 , . . . , en ) and |e| = e1 + . . . + en .
Now Lemma 10.5.1 follows from Proposition 10.6.1. We have only to denote
f (x) in Proposition 10.6.1 say by p(x), express f (x) in Lemma 10.5.1 by h(p(x))
with h(y) homogeneous of degree d, and take ϕ(y) = |h(y)|sK .
10.7 Z(s) for det(t xhx) in product forms

We shall go back to Proposition 10.5.1 and convert Z0 (s), hence also Z(s), into a
product. The product form of Z0 (s) depends on m, n mod 2. We recall that

Z0 (s) = ck tk · Ik ,
0≤k≤n
in which ck and Ik are as in that proposition. We shall make some preliminary

computation. We replace k above by 2k and 2k + 1, and put the two terms for the
same new k together. If n is odd, then the summation in the new k will become
for 0 ≤ k ≤ (n − 1)/2. However, if n is even, the summation will become for
0 ≤ k ≤ n/2 − 1 and an extra term, i.e., cn tn In . It turns out that this extra term
is obtained from the combined general term by substituting n/2 for k. At any rate,
the result can be stated more or less uniformly as follows:

Z0 (s) = A [1, 1] · [i] · Z1 (s),
1≤i≤n

Z1 (s) = Ak q −k(2k+1) t2k
0≤2k≤n

· [2i] [2i] · [2i][2i + 1, 2] ,
m−n−2k<2i<m 1≤2i≤n−2k 1≤i≤k
194 JUN-ICHI IGUSA
in which
 '

 [m/2] n even
 m even
[m/2][m/2, 1]+ n odd '
A=

 [n + 1, 1] n even
 m odd
[m − n + 1, 1] n odd
while Ak = 1 except for the case where m, n are both even, and in that case
Ak = [m/2 − 2k]+ [2k + 1, 1] + [m − n − 2k][n − 2k]q −2k−1 t.
The verifications are similar and straightforward.

We shall convert Z1 (s) into a product. The conversion will be made by using
(G2) and (G3) in Chapter 9.6. We recall that
2
(G2) Fm−k,k (a)Fk,n−k (a, t)ak tk = Fm,n (a, t),
0≤k≤n
2
−k
(G3) Fm−k,k (a)Fk,n−k (a, t)ak tk = Fm−1,n (a, t) + tFm,n−1 (a, t),
0≤k≤n
in which

Fm,n (a, t) = (1 − am+i t)/(1 − ai ), Fm,n (a) = Fm,n (a, 1)
1≤i≤n
for m, n in Z with the understanding that Fm,n (a, t) = 0 for n < 0. In all cases, we
replace q −2 by a and q −1 t2 by t. More precisely, we shall be replacing the variables
a, t in (G2) and (G3) by q −2 , q −1 t2 .
If m, n are not both even, hence Ak = 1, we replace them respectively by

 2(m + n + 1), 2n + 1 m even, n odd
2(m + n) + 1, 2n m odd, n even

2(m + n) + 1, 2n + 1 m odd, n odd.
Then in all three cases we get

2
Z1 (s) = (1 − am+i )/(1 − ai t) · Fm−k,k (a)Fk,n−k (a, t)ak tk
1≤i≤n 0≤k≤n

= (1 − a m+i
)(1 − a m+i
t)/(1 − ai )(1 − ai t),
1≤i≤n
this by (G2). Therefore in the original notation we have


 1≤2i≤n [m − 2i][m − 2i + 1, 2]m even, n odd
Z1 (s) = 1 [2i][2i + 1, 2] ·

1≤2i≤n 1≤2i≤n [m − 2i + 1][m − 2i + 2, 2] m odd.
In the case where m, n are both even, we write

Ak = B1 + B2 q 2k ,
in which B1 , B2 are free from k, i.e.,
B1 = 1 − (1 + q −m/2+n + q −m+2n )q −n−1 t, B2 = [m/2 + 1, 1]+ q −m/2 ,
and split Z1 (s) as

Z1 (s) = B1 · Z1 (s) + B2 · Z2 (s).
If we replace m, n respectively by 2(m + n), 2n, then in the same way as in the
previous cases we get

Z1 (s) = 1/(1 − am+n ) · (1 − am+i )(1 − am+i t)/(1 − ai )(1 − ai t).
1≤i≤n
As for Z2 (s), we have

Z2 (s) = 1/(1 − am+n ) · (1 − am+i )/(1 − ai t)
1≤i≤n
2
−k
· Fm−k,k (a)Fk,n−k (a, t)ak tk ,
0≤k≤n
and the sum over 0 ≤ k ≤ n is equal to Fm−1,n (a, t) + tFm,n−1 (a, t) by (G3).
Therefore, if in the original notation we put
C = [m + 1, 2]B1 + ([m − n + 1, 2] + [n]q −1 t2 )B2 ,

then we have

Z1 (s) = C/[n][n + 1, 2] · [m − 2i][m − 2i + 1, 2]/[2i][2i + 1, 2].
1≤i<n/2
Furthermore if we replace B1 , B2 by their expressions in terms of q −1 , t, then we

get C = Cm/2,n (q −1 , t) for the Cm,n (a, t) in section 10.4, i.e.,
Cm,n (a, t) = (1 + am ) − (1 + am−n + a2m−2n − a2m−n )an+1 t

+ (1 − an − am − a2m−n )am+1 t2 + (1 + am )a2m+2 t3
for m ≥ n. We have thus converted Z0 (s) in Proposition 10.5.1 into the following
definitive form:
Proposition 10.7.1 If m is even, then

Z0 (s) = [m/2] [i] [1, 1] · [2i][2i + 1, 2]
1≤i≤n 1≤2i≤n

 Cm/2,n (q −1 , t)/[m − n][m − n + 1, 2]n even
· [m − 2i][m − 2i + 1, 2]

1≤2i≤n [m/2, 1]+ n odd
196 JUN-ICHI IGUSA
and if m is odd, then

Z0 (s) = [i] [1, 1] · [2i][2i + 1, 2]
1≤i≤n 1≤2i≤n

 [n + 1, 1] n even
· [m − 2i + 1][m − 2i + 2, 2]

1≤2i≤n [m − n + 1, 1] n odd.
× 2
We have assumed that (−1)m(m−1)/2 det(h) is in (OK ) in the case where m is
× 2
even. The fact is that if it is in OK \(OK ) , then we have only to replace all q −m/2 in
the above formulas by −q −m/2 . Propositions 10.5.1 and 10.7.1 are in Part I of [29].
We recall that we have seen the cubic polynomial Cm,n (a, t) in t for (m, n) = (5, 4)
in Proposition 10.4.1, i.e., in the Z(s) for a Freudenthal quartic f (x). We have
mentioned there with a reference that Cm,n (a, t) for (m, n) = (8, 6), (14, 10) also
appear in the Z(s) for other Freudenthal quartics. Although Cm,n (a, t) is a strange
polynomial, this fact can be expected because Freudenthal quartics are similar. We
have now seen that Cm,n (a, t) also appears in the Z(s) for the Gramian det(t xhx),
in fact only in a certain case as stated in Proposition 10.7.1. This is a mystery
for us because we fail to see any similarity between Freudenthal quartics and the
Gramian. At any rate, Cm,n (a, t) has the following formal properties:
Cm,n (a−1 , t−1 ) = a−3m−2 t−3 Cm,n (a, t),

Cm,n (a, 0) = 1 + am ,
Cm,n (a, 1) = (1 + am )(1 − an+1 )(1 − a2m−n+1 ),
Cm,n (a, a−1 ) = (1 + am−1 )(1 − an )(1 − a2m−n ).
We can easily verify the fact that Cm,n (a, t) is the only element of Q(a)[t] with
these properties. Furthermore, we can show that Cm,n (a, t) is irreducible in C(a)[t]
for n > 0 while clearly
Cm,0 (a, t) = (1 + am )(1 − at)(1 − a2m+1 t2 ).
We also mention that bf (s) for f (x) = det(t xhx) was computed as an example
of their general theory in a joint paper [50] by M. Sato, M. Kashiwara, T. Kimura,
and T. Oshima. The result is

bf (s) = (s + (k + 1)/2)(s + (m − k + 1)/2).
1≤k≤n
We observe that, in all examples which we have computed or mentioned, the real
parts of the poles of Z(s) for f (x) are the zeros of bf (s) with the order of each pole
at most equal to the order of the corresponding zero. As we have emphasized in the
Introduction, to convert this experimental fact into a theorem is an open problem.
As we have also mentioned in the Introduction, without the information about the
orders this has been proved by T. Kimura, F. Sato, and X.-W. Zhu [35] in the case
where f (x) is the basic relative invariant of an irreducible regular prehomogeneous
vector space. In their proof Theorem 8.5.1 and a theorem of M. Sato play key roles.
Also in the case where the number of variables in f (x) is 2 and in cases of some
generality, it has been proved by F. Loeser [37], [38].
Now, in spite of the complexity of the expression of Z(s) for the Gramian f (x) =
det(t xhx), it has remarkably simple properties. Firstly, if we denote by degt (Z(s))
the degree of Z(s) as a rational function of t, then we have
degt (Z(s)) = −2n = − deg(f ).
If f (x) is a Freudenthal quartic, then we also have
degt (Z(s)) = −4 = − deg(f ).
If we examine other examples of Z(s), then we will find that the above property is
shared by all Z(s) provided that f (x) is homogeneous and, e.g., non-zero coefficients
×
of f (x) are all units of OK . Secondly, the coefficients of the expressions of Z(s) as
a rational function of a = q −1 and t for f (x) = det(t xhx) are numerical constants
independent of K. Therefore, the process of replacing a, t by a−1 , t−1 , i.e., q by
q −1 , makes sense, and we have
Z(s)|q
→q−1 = t2n Z(s) = tdeg(f ) Z(s).
If f (x) is a Fredenthal quartic, then we similarly have
Z(s)|q
→q−1 = t4 Z(s) = tdeg(f ) Z(s).
If we examine other examples for which the process “q → q −1 ” makes sense, we

will find that the above property is shared by all Z(s) provided that f (x) satisfies
the homogeneity condition, etc. As we have mentioned in the Introduction, we
proposed precisely formulated conjectures on the above properties of Z(s) in [25],
[29] and now they have been settled, hence the conjectures have become theorems,
by J. Denef [9] and J. Denef and D. Meuser [10]. We shall explain the way they
proved their theorems in the next and last chapter.
Chapter 11
Theorems of Denef and

Meuser
11.1 Regular local rings
We have not yet made any explicit use of algebraic varieties. We shall now try to
explain some fragments of algebraic geometry to state Hironaka’s desingularization
theorem in an algebraic form and a theorem on Weil’s zeta functions over finite fields
so that the readers can fully appreciate the proofs by Denef and Meuser of their
theorems which depend heavily on those theorems. We shall go back to Chapter
1.2 and start with some properties of regular local rings.
We take a local ring A, i.e., a noetherian ring A = 0 in which the set of all
nonunits forms a maximal ideal m. It is the same thing to say that A is a noetherian
ring with only one maximal ideal m. We know by Corollary 1.2.1 that the dimension
of m/m2 as a vector space over A/m gives the smallest number of generators of m
as an A-module. Since that will be sufficient for our purpose, we shall assume that
A contains a field K satisfying A = K + m and identify K with A/m. We know
by Theorem 1.2.2 that m0 = A, m, m2 , . . . form a decreasing sequence of ideals of
A with 0 as their intersection. We put Gr (A) = mr /mr+1 and denote by G(A) the
direct sum of G0 (A) = K, G1 (A), G2 (A), . . . . We observe that every Gr (A) is a
vector space over K and G(A) becomes a graded K-algebra. Furthermore, if we
put dimK (m/m2 ) = n and if we introduce a polynomial ring K[x] = K[x1 , . . . , xn ],
where x1 , . . . , xn are variables, then we can define a surjective homomorphism from
the graded K-algebra K[x] to G(A) as follows. We write m = Aa1 + . . . + Aan and
denote the image of ai in G1 (A) by yi for 1 ≤ i ≤ n. Then G1 (A) = Ky1 +. . .+Kyn ,
where y1 , . . . , yn are linearly independent over K, and

Gr (A) = Ky1e1 . . . ynen ,
|e|=r
in which e = (e1 , . . . , en ) is in Nn and |e| = e1 + . . . + en , for all r. Therefore, if

we define a K-linear bijection from K[x]1 to G1 (A) as xi → yi for 1 ≤ i ≤ n, then
it uniquely extends to a surjective K-algebra homomorphism from K[x] to G(A)
mapping K[x]r to Gr (A) for all r. If we denote the kernel of the above homomor-
phism by a, then a is a homogeneous ideal of K[x] and K[x]/a becomes isomorphic
199
200 JUN-ICHI IGUSA
to G(A) as graded K-algebras. Therefore, if we denote Hilbert’s characteristic func-

tion of any finitely generated K[x]-module M by χ(M, t) as in Theorem 1.3.3, then
we will have

dimK (A/mr+1 ) = dimK (Gi (A)) = χ(G(A), r)
i≤r
for all large r. After P. Samuel [47] we call d = deg(χ(G(A), t)) the dimension of A
and write dim(A) = d. We observe that

t+n
χ(G(A), t) = χ(K[x], t) − χ(a, t) = − χ(a, t)
n
and that, by the remark after Theorem 1.3.3, we have

t
deg χ(a, t) − <n
n
if a = 0. Therefore, we get d ≤ n with the equality if and only if a = 0. We shall
summarize the above observations as follows:
Proposition 11.1.1 Let A denote a local ring with m as its maximal ideal such
that A contains a field K satisfying A = K + m and G(A) the graded K-algebra
with Gr (A) = mr /mr+1 for all r. Then there exists a polynomial χ(t) satisfying

dimK (A/mr+1 ) = dimK (Gi (A)) = χ(r)
i≤r
for all large r. If we put
n = dimK (m/m2 ), d = deg(χ(t)) = dim(A),
then d ≤ n. Furthermore, we have d = n if and only if G(A) is isomorphic to a

polynomial ring K[x1 , . . . , xn ] as graded K-algebras.
We call A a regular local ring if d = n. In that case, A is an integral domain. In
fact, if a, b are elements of A both different from 0, their images in Gi (A), Gj (A)
for some i, j are not 0, hence the image of ab in Gi+j (A) is not 0, and hence ab = 0.
Proposition 11.1.2 Suppose that A is a regular local ring with dim(A) = n, ex-
press its maximal ideal m as m = Aa1 + . . . + Aan , and put
p = Aa1 + . . . + Aap
for 0 ≤ p ≤ n. Then A/p is also a regular local ring, hence p is a prime ideal of A,
and dim(A/p) = n − p.
Proof. Since the local ring A is regular, every a in A gives rise to a unique sequence
f0 (x), f1 (x), f2 (x), . . . with fi (x) in K[x]i satisfying

a≡ fi (a) mod mr+1 ,
i≤r
in which fi (a) = fi (a1 , . . . , an ) for all r. We put q = Aap+1 + . . . + Aan so that we

get m = p + q. Since A/p has m/p as its unique maximal ideal, it is a local ring.
Therefore, we have only to show that Gr (A/p) is K-isomorphic to K[xp+1 , . . . , xn ]r
under the correspondence a → fr (x). We have
Gr (A/p) = (m/p)r /(m/p)r+1 = qr /(qr ∩ (p + qr+1 )),
in which p + qr+1 = p + mr+1 . We observe that an element a of A is in qr if and only
if fi (x) = 0 for i < r, fr (x) is in k[xp+1 , . . . xn ]r , and fi (x) for i > r is of degree at
least r in xp+1 , . . . , xn ; that a is in p + mr+1 if and only if fi (x) is at least of degree
1 in x1 , . . . , xp for 0 ≤ i ≤ r. Therefore, the additional condition for a in qr to be in
p + mr+1 is simply fr (x) = 0, hence Gr (A/p) is K-isomorphic to K[xp+1 , . . . , xn ]r
as a → fr (x).
Theorem 11.1.1 Let K[x] = K[x1 , . . . , xn ] denote the ring of polynomials in n
variables x1 , . . . , xn with coefficients in a field K, a = (a1 , . . . , an ) an element
of K n , and f1 (x), . . . , fp (x) elements of K[x] satisfying fi (a) = 0 for all i such
that rank(J(a)) = p for the p × n matrix J(x) with ∂fi /∂xj as its (i, j)-entry for
1 ≤ i ≤ p, 1 ≤ j ≤ n; let further S denote the set of all g(x) in K[x] satisfying
g(a) = 0 and put

A = S −1 K[x], m= A(xi − ai ), p= Afi (x).
1≤i≤n 1≤i≤p
Then A and A/p are regular local rings with dim(A) = n and dim(A/p) = n − p.
Proof. We apply the K-automorphism of K[x] defined by xi → yi = xi − ai for
1 ≤ i ≤ n and reduce the general case to the case where a = 0. Since we can
write A = K[x] + mr for all r, we can identify G(A) with K[x]. Therefore, A
is a regular local ring with dim(A) = n. Furthermore, by assumption there exist
1 ≤ j1 < . . . < jp ≤ n such that if g(x) denotes the determinant of the of p × p
submatrix of J(x) obtained by crossing out its j-th columns for all j = j1 , . . . , jp ,
then g(0) = 0. We observe that every f (x) in K[x] satisfying f (0) = 0 and its first
polar
(∂f /∂x1 )(0)x1 + . . . + (∂f /∂xn )(0)xn
have the same image in m/m2 . Therefore, if we denote xj for j = j1 , . . . , jp by
t1 , . . . , td , then the images of f1 (x), . . . , fp (x), t1 , . . . , td in m/m2 form its K-basis,
hence
m= Afi (x) + Atj .
1≤i≤p 1≤j≤d
It then follows from Proposition 11.1.2 that p is a prime ideal of A and A/p is a
regular local ring with dim(A/p) = d = n − p.
We remark that if we put P = K[x] ∩ p, then P is a prime ideal of K[x]
satisfying S −1 P = p and a minimal representation of the ideal of K[x] generated
by f1 (x), . . . , fp (x) is of the form

K[x]fi (x) = P ∩ Qi ,
1≤i≤p 1<i≤t
202 JUN-ICHI IGUSA
in which the primary ideal Qi of K[x] intersects S for 1 < i ≤ t. This follows from
Lemma 1.2.2 and Proposition 1.2.1.
11.2 Geometric language

We shall fix an algebraically closed field Ω, denote by F any subfield of Ω and by
m, n nonnegative integers. Then, as a set, the affine n-space Aff n is defined as Ωn
with Aff n = {0} for n = 0. Also as a set, the projective n-space Projn is defined as
the factor space of Ωn+1 \{0} by Ω× in the same way as in Chapter 3.1. If a is a
point of Projn represented by (a1 , . . . , an+1 ) in Ωn+1 \{0}, then (a1 , . . . , an+1 ) are
called the homogeneous coordinates of a; they are determined by a up to a common
factor in Ω× . If t1 , . . . , tn+1 are variables and f (t) is a homogeneous polynomial in
Ω[t] = Ω[t1 , . . . , tn+1 ], then we denote f (a1 , . . . , an+1 ) simply by f (a). This will
not cause any problem because we shall be interested only in whether or not f (a) is
0 and in the quotient f (a)/g(a) where g(t) is also homogeneous of the same degree
as f (t) with g(a) = 0. If I is any set of homogeneous polynomials fi (t) in Ω[t], then
a subset of Projn is well defined as the set of all a satisfying fi (a) = 0 for all fi (t) in
I. We call such a set a closed subset of Projn . If J is a similar subset of Ω[t] as I such
that I and J generate the same ideal of Ω[t], then I and J define the same closed
subset of Projn . Therefore, by Hilbert’s basis theorem, i.e., by Theorem 1.3.1, we
may assume that I is finite. If X is a closed subset of Projn defined by I, we take
all homogeneous polynomials in Ω[t] which vanish at every point of X and denote
by I(X) the ideal of Ω[t] generated by them. Then, by Hilbert’s Nullstellensatz,
i.e., by Theorem 1.3.2, I(X) is the root of the ideal of Ω[t] generated by I. If I(X)
has an ideal basis in F [t] = F [t1 , . . . , tn+1 ], then X is called an F -closed subset of
Projn . If I(X) is a prime ideal, then X is called irreducible. Every closed subset X
of Projn can be expressed as a finite union of irreducible closed subsets X1 , X2 , . . . .
If no Xi is contained in Xj for i = j, then the expression is unique, and X1 , X2 , . . .
are called irreducible components of X. An outline of the proof is as follows.
Since I(X) is equal to its root r(I(X)), by Theorem 1.2.1 its minimal representa-
tion I(X) = P1 ∩ P2 ∩ . . . , where P1 , P2 , . . . are necessarily prime ideals of Ω[t] with
no inclusion relation, is unique. Every λ in Ω× gives rise to an Ω-automorphism of
Ω[t] as ti → λti for 1 ≤ i ≤ n + 1, and I(X) is invariant under this action of Ω×
on Ω[t], hence P1 , P2 , . . . will just be permuted. Since the m-th power map from
Ω× to itself is surjective for every m > 0, it has no subgroup of finite index larger
than 1. Therefore, the permutation group is trivial, hence every Pi is invariant.
Then the homogeneous parts of every f (t) in Pi is in Pi , hence Pi is generated by
homogeneous polynomials, and hence they define a closed subset Xi . By definition,
X1 , X2 , . . . are the irreducible components of X.
The complement of a closed subset of Projn is called open. The intersection of a
closed subset and an open subset of Projn , i.e., the difference of two closed subsets,
is called locally closed. We observe that ti = 0, i.e., the complement of the closed
subset of Projn defined by ti , defines an open subset Ui of Projn and Projn becomes
the union of U1 , . . . , Un+1 . If a with homogeneous coordinates (a1 , . . . , an+1 ) is in
Ui , i.e., ai = 0, and if we put
φi (a) = (a1 /ai , . . . , ai−1 /ai , ai+1 /ai , . . . , an+1 /ai ),
then φi gives a bijection from Ui to Aff n for 1 ≤ i ≤ n + 1. Therefore, we can say

that Projn is covered by n + 1 affine n-spaces.
If X is any nonempty locally closed subset of Projn , then we can define closed,
open, and locally closed subsets of X by relative topology. For instance, if I is any
set of homogeneous polynomials in Ω[t], then the intersection of X and the closed
subset of Projn defined by I is a closed subset of X. In the following we shall denote
by g(t), g (t), etc. homogeneous polynomials in Ω[t] all different from 0. If g(t) is
such a polynomial, the open subset of X defined by g(t) = 0, i.e., as the difference
of X and the closed subset of Projn defined by g(t), will be denoted by Xg . We
observe that Xg = ∅ if and only if g(a) = 0 for every a in X; that Xg ∩ Xg = Xgg ;
and that the set of all Xg forms an open base for X. If Xg = ∅ and if f (t) is a
homogeneous polynomial in Ω[t] of the same degree as g(t)k for some k in N, then
the quotient f (t)/g(t)k gives rise to a well-defined Ω-valued function on Xg . We
shall denote by O(Xg ) the Ω-algebra of all such functions on Xg . If now a is any
point of X, then we take the direct limit of O(Xg ) for all g(t) such that g(a) = 0 and
denote it by OX,a . We recall that the above direct limit is defined as follows: We say
that ϕ, ϕ respectively in O(Xg ), O(Xg ), where g(a)g (a) = 0 are equivalent if for
some multiple g (t) of g(t)g (t), also satisfying g (a) = 0, the restrictions of ϕ, ϕ
to Xg are equal. The set of all equivalence classes forms an Ω-algebra, and that is
OX,a . We observe that OX,a is a local ring, i.e., the set ma of all nonunits of OX,a
forms a maximal ideal; the noetherian property of OX,a will become clear later.
The set of OX,a for all a in X is denoted by OX and the pair (X, OX ) or simply
X is called a quasi-projective variety. If X is closed, then it is called a projective
variety. If X is the difference of F -closed subsets, then X is called a quasi-projective
F -variety. As a special case, a projective F -variety is defined. Suppose that X, Y
are quasi-projective varieties not necessarily in the same projective space and f is
a map from X to Y . Then f is called a morphism if for every a in X, b = f (a),
and ψ in OY,b the composition ψ ◦ f is in OX,a . The product of two morphisms is
a morphism. If a morphism is bijective and if the inverse map is also a morphism,
then it is called an isomorphism. If X is contained in Y , then the inclusion map is
a morphism, and X is called a subvariety of Y .
We shall show that the product X = Projm × Projn , hence also the product
of projective varieties, can be considered as a projective variety. We take any
point (a, b) of X and denote the homogeneous coordinates of a, b respectively by
(a1 , . . . , am+1 ), (b1 , . . . , bn+1 ). We order N +1 = (m+1)(n+1) elements cij = ai bj
of Ω lexicographically as (c11 , c12 , . . . ) and regard them as homogeneous coordinates
of a point c of ProjN . Then we get a well-defined map from X to ProjN as (a, b) → c.
We shall examine the image of X. We introduce N + 1 variables tij and put
fij,i j (t) = tij ti j − tij ti j
for 1 ≤ i, i ≤ m + 1, 1 ≤ j, j ≤ n + 1. We denote by I the set of these quadratic

forms in tij and by Y the closed subset of ProjN defined by I. If cij = ai bj as above,
204 JUN-ICHI IGUSA
then fij,i j (c) = 0 for all fij,i j (t) in I. Therefore, c is in Y . Conversely, suppose
that c is any point of Y with homogeneous coordinates (c11 , c12 , . . . ) so that ci0 j0 = 0
for some i0 , j0 . Denote by a, b the points of Projm , Projn with (c1,j0 , . . . , cm+1,j0 ),
(ci0 ,1 , . . . , ci0 ,n+1 ) as their respective homogeneous coordinates. Then we see that
c is the image of (a, b). We have thus shown that Y is the image of X. We shall
show that the surjective map X → Y is injective, hence bijective. Suppose that
(a , b ) in X has the same image as (a, b) and that (a1 , . . . , am+1 ), (b1 , . . . , bn+1 )
are the homogeneous coordinates of a , b . Then we will have ai bj = λai bj for all
i, j with some λ in Ω× . If ai0 bj0 = 0, then ai0 bj0 = 0, hence we may assume that
ai0 = bj0 = ai0 = bj0 = 1. This implies λ = 1, ai = ai bj0 = ai bj0 = ai , and
similarly bj = bj for all i, j, hence (a, b) = (a , b ). We now define OX so that the
bijection X → Y becomes an isomorphism. We might mention that OX can be
defined directly by using doubly homogeneous polynomials. We also remark that
if for any 1 ≤ i0 ≤ m + 1 we define an open subset Ui0 of Projm as before and
put J = {ti0 ,j ; 1 ≤ j ≤ n + 1}, then the above isomorphism X → Y gives rise
to an isomorphism from the product Ui0 × Projm to the difference of Y and the
closed subset of ProjN defined by J. Therefore, Aff m × Projn can be considered as
a quasi-projective variety.
As a rather special case of the above, we regard any closed subset of Aff n as a
quasi-projective variety, and call it an affine variety. If X is such a variety, then
the set I(X) of all f (t) in Ω[t] = Ω[t1 , . . . , tn ] which vanish at every point of X
forms an ideal of Ω[t]. We shall show that if g(t) is any element of Ω[t] different
from 0, then the open subset Xg of X consistsing of all a in X for which g(a) = 0
can be considered as an affine variety. In fact, if s is a variable, then the set Y of
all common zeros of elements of I(X) and h(t, s) = g(t)s − 1 in Ω[t, s] is an affine
variety in Aff n+1 . Furthermore, the correspondence a → (a, b), where b = 1/g(a),
gives a bijection from Xg to Y . If an element of OY,(a,b) is represented by a function
defined by f ∗ (t, s)/g ∗ (t, s) for f ∗ (t, s), g ∗ (t, s) in Ω[t, s] and g ∗ (a, b) = 0 and if e is
at least equal to the degrees in s of f ∗ (t, s), g ∗ (t, s), then
f ∗ (t, 1/g(t))/g ∗ (t, 1/g(t)) = f (t)/g (t),
in which f (t) = g(t)e f ∗ (t, 1/g(t)), g (t) = g(t)e g ∗ (t, 1/g(t)) are in Ω[t] and g (a) =
0. This implies that X → Y is a morphism. Since the projection Y → X is a
morphism, the bijection X → Y is an isomorphism.
We shall make some remarks on the local ring OX,a for any quasi-projective
variety X. Firstly, OX,a will not change even if we replace X by Xg for any
homogeneous polynomial g(t) satisfying g(a) = 0. We observe that Xg for a suitable
g(t) becomes an affine variety. In fact, by definition X can be expressed as X =
X1 \X2 for some closed subsets X1 , X2 of Projn and a is not in X2 . Therefore,
we can find a homogeneous polynomial g0 (t) in I(X2 ) satisfying g0 (a) = 0. If
further a is in Ui , then we can take g(t) = g0 (t)ti . Therefore, after replacing X by
Xg , we may assume that X is an affine variety in Aff n . We change the notation
accordingly and write Ω[t] = Ω[t1 , . . . , tn ]. We shall show that if S denotes the set
of all g(t) in Ω[t] satisfying g(a) = 0, where a = (a1 , . . . , an ), then Oa = OX,a can
be identified with S −1 Ω[t]/S −1 I(X). In particular, Oa is a noetherian ring. We
recall that an element of Oa is an equivalence class of functions on Xg for all g(t)

in S; that two functions on Xg , Xg respectively defined by f (t)/g(t)k , f (t)/g (t)k
for f (t), f (t) in Ω[t], g(t), g (t) in S, and k, k in N are equivalent if for some
multiple g (t) of g(t)g (t), also in S, they give rise to the same function on Xg .

Then h(t) = f (t)g (t)k − f (t)g(t)k vanishes at every point of Xg , hence g (t)h(t)

is in I(X), and hence f (t)/g(t)k − f (t)/g (t)k in in S −1 I(X). Since the converse
−1 −1
is clear, we indeed have Oa = S Ω[t]/S I(X).
After these preliminaries, suppose that X is a quasi-projective variety and for
every a in X put dima (X) = dim(Oa ). Then by Proposition 11.1.1 we have
dima (X) ≤ dimΩ (ma /m2a ).
We say, after O. Zariski [63], that a is a simple point of X if the equality holds, i.e.,
if Oa is a regular local ring. We sometimes denote by Xsmooth the set of all simple
points of X. If X = Xsmooth , then we call X nonsingular or smooth. Furthermore
if d = dima (X) is independent of a in X, then we say that X is d-dimensional
or, more precisely, everywhere d-dimensional. We observe that Aff n and Projn are
both smooth and everywhere n-dimensional.
Finally, suppose that a point a of Projn with homogeneous coordinates
(a1 , . . . , an+1 ) is contained in Ui and Uj , i.e., ai aj = 0. Then we clearly have
F (φi (a)) = F (φj (a)); we denote this field by F (a). If F (a) = F we say that a is an
F -rational point. If X is a quasi-projective F -variety in Projn , we denote by X(F )
the set of all F -rational points of X.

(algebraic form)
We shall start with a simple example. We fix an algebraically closed field Ω and
denote by F any subfield of Ω as in section 11.2. We take the product X × Projn−1 ,
where X = Aff n , and denote by Y its closed subset defined by xi zj − xj zi for
1 ≤ i, j ≤ n, in which (x1 , . . . , xn ) are the coordinates on X and (z1 , . . . , zn ) are
the homogeneous coordinates on Projn−1 all considered as variables. If h denotes
the restriction to Y of the projection from X × Projn−1 to X, then h gives a
bijection from Y \h−1 (0) to X\{0}, in which h−1 (0) = {0} × Projn−1 . In order to
get more precise information about Y and h, we express Projn−1 as the union of
the open subset Zα defined by zα = 0 for 1 ≤ α ≤ n. Then Y becomes the union
of Yα = Y ∩ (X × Zα ) for all α. Since they are all similar, we take α = 1 and put
(t1 , . . . , t2n−1 ) = (x1 , . . . , xn , z2 /z1 , . . . , zn /z1 );
also we denote the image of ti in Ω[t1 , . . . , t2n−1 ]/I(Y1 ) by yi for 1 ≤ i ≤ 2n −

1. Then Y1 becomes isomorphic to Aff n under the map y = (y1 , . . . , y2n−1 ) →
(y1 , yn+1 , . . . , y2n−1 ) and
h(y) = (y1 , . . . , yn ) = (y1 , y1 yn+1 , . . . , y1 y2n−1 ).

206 JUN-ICHI IGUSA
If we use (y1 , yn+1 , . . . , y2n−1 ) as the coordinates on Y1 , then the jacobian of h|Y1
becomes
∂(x1 , . . . , xn )/∂(y1 , yn+1 , . . . , y2n−1 ) = y1n−1 .
Also (y1 , yn+1 , . . . , y2n−1 ) = (x1 , x2 /x1 , . . . , xn /x1 ), hence h|Y1 gives rise to an
isomorphism from Y1 \y1−1 (0) to X\x−1 1 (0). At any rate Y is a closed smooth n-
dimensional subvariety of X × Projn−1 .
We now take f (x) from F [x1 , . . . , xn ]\F , denote by f −1 (0) the set of all a in
X satisfying f (a) = 0, and examine the effect of h : Y → X on the hypersurface
f −1 (0) in X. We shall assume that if f0 (x) is the leading form of f (x), i.e., the
homogeneous part of f (x) of the smallest degree, then N = deg(f0 ) > 0 and not all
partial derivatives of f0 (x) vanish at any point of f0−1 (0) other than 0. We observe
that the preimage (f ◦ h)−1 (0) of f −1 (0) in Y under h is the union of its intersection
with Yα for 1 ≤ α ≤ n. Since they are all similar, we take α = 1 and use the same
notation as above. We then have
f (h(y)) = f (y1 , y1 yn+1 , . . . , y1 y2n−1 ) = y1N f1 (y),
in which
f1 (y) = f0 (1, yn+1 , . . . , y2n−1 ) + y1 f (y)
with f (y) in F [y1 , yn+1 , . . . , y2n−1 ]. In general, if f1 (t), . . . , fp (t) in Ω[t1 , . . . , tn ]
satisfy the condition in Theorem 11.1.1 for K = Ω, then we say that the p hyper-
surfaces f1−1 (0), . . . , fp−1 (0) in Aff n are transversal at a. In that case the theorem
implies that a is a simple point of their intersection. We shall show that f1−1 (0)
and y1−1 (0) are transversal at every point of their intersection.
If b = (b1 , bn+1 , . . . , b2n−1 ) is any point of their intersection, then b1 = 0 and
f1 (b) = f0 (c) = 0, in which c = (1, bn+1 , . . . , b2n−1 ). If f1−1 (0) and y1−1 (0) are not
transversal at b, then
(∂f1 /∂yn+i )(b) = (∂f0 /∂xi+1 )(c) = 0
for 1 ≤ i < n. If we write down Euler’s identity for f0 (x) and evaluate both sides at
c, then we also have (∂f0 /∂x1 )(c) = 0. Since f0 (c) = 0 and c = 0, this contradicts
the assumption that f0−1 (0)\{0} is smooth.
If we further assume that f (x) is homogeneous, hence f (x) = f0 (x), then
f −1 (0)\{0} is smooth. In that case since f1 (y) = f (1, yn+1 , . . . , y2n−1 ), the above
argument shows that f1−1 (0) is smooth. Therefore, if we denote by E1 , E2 the
closed subsets of Y such that their intersections with Y1 are respectively y1−1 (0),
f1−1 (0) and their intersections with Yα are similar, then (f ◦ h)−1 (0) becomes the
union of E1 , E2 and they are both smooth, (n − 1)-dimensional, and have normal
crossings in the sense that they are transversal at every point of their intersection.
Furthermore, the bijection from Y \h−1 (0) to X\{0} is an isomorphism. We also
observe that Y and E1 , E2 are F -varieties.
We shall now introduce the concept of Hironaka’s desingularization of a hyper-
surface by using some terminology which will be explained in a slightly different
language.
Definition 1. Let f (x) denote any element of F [x1 , . . . , xn ]\F , where F is an

arbitrary field and x1 , . . . , xn are variables. Then a Hironaka’s desingularization
of the hypersurface in X = Aff n defined by f (x) is a closed smooth n-dimensional
F -subvariety Y of X × Projm for some m such that the restriction h to Y of the
projection X × Projm → X has the following properties: Firstly, there exists a
finite set E of closed smooth (n − 1)-dimensional F -subvarieties E of Y with normal
crossings. Secondly, as point-sets (f ◦ h)−1 (0) is the union of all E in E and h gives
rise to an isomorphism Y \(f ◦ h)−1 (0) → X\f −1 (0). Thirdly,

of f ◦ h
the divisors
and h∗ (dx1 ∧· · ·∧dxn ) on Y are respectively of the forms NE E and (nE −1)E,
in which NE , nE are positive integers and the summations are for all E in E.
It follows from the definition that h : Y → X is surjective. We shall give a
local description of the Hironaka desingularization. We denote by z1 , . . . , zm+1 the
homogeneous coordinates on Projm and express Projm as the union of the open
subsets Zα defined by zα = 0 for 1 ≤ α ≤ m + 1. Then Y becomes the union of
Yα = Y ∩ (X × Zα ) for all α. We choose α arbitrarily and identify X × Zα with
Aff m+n by using
(x1 , . . . , xn , z1 /zα , . . . , zα−1 /zα , zα+1 /zα , . . . , zm+1 /zα )
as coordinates (t1 , . . . , tm+n ) on Aff m+n . Then Yα is a closed subset of Aff m+n , the

ideal I(Yα ) has a basis in F [t] = F [t1 , . . . , tm+n ], and the local ring Ob of Yα at every
point b = (b1 , . . . , bm+n ) of Yα is regular of dimension n. Therefore, if we denote
the image of ti in Ob by yi for 1 ≤ i ≤ m + n, then yi1 − bi1 , . . . , yin − bin for some
1 ≤ i1 < . . . < in ≤ m + n generate the maximal ideal mb of Ob , and if we denote
the image of yi − bi in mb /m2b by dyi for all i, then dyi1 , . . . , dyin form an Ω-basis
for mb /m2b . If we now write all E’s which contain b by E1 , . . . , Ep , then there exist
f1 (t), . . . , fp (t), g(t) in F [t] with g(b) = 0 such that f1 (y), . . . , fp (y) can be included
in an ideal basis for mb and further Ei ∩ (Yα )g = fi−1 (0) ∩ (Yα )g for 1 ≤ i ≤ p, in
which (Yα )g = Yα \g −1 (0). Finally, we have h(y) = (y1 , . . . , yn ), h∗ (dxi ) = dyi for
1 ≤ i ≤ n, hence (f ◦ h)(y) = f (y1 , . . . , yn ), h∗ (dx1 ∧ . . . ∧ dxn ) = dy1 ∧ . . . ∧ dyn ,
and
(f ◦ h)(y) = ε · fi (y)Ni ,
1≤i≤p

h∗ (dx1 ∧ . . . ∧ dxn ) = η · fi (y)ni −1 · dyi1 ∧ . . . ∧ dyin ,
1≤i≤p
in which ε, η are units of Ob and (Ni , ni ) = (NE , nE ), where E = Ei for 1 ≤ i ≤ p.

In the above example m = n − 1, α = 1, and (i1 , . . . , in ) = (1, n + 1, . . . , 2n − 1);
that E = {E1 , E2 }, f1 (t) = t1 , f2 (t) = f (1, tn+1 , . . . , f2n−1 ), g(t) = ε = η = 1, and
(NE , nE ) = (N, n), (1, 1) respectively for E = E1 , E2 . At any rate, we can state
one form of Hironaka’s desingularization theorem in [20] as follows:
Theorem 11.3.1 If char(F ) = 0, then a Hironaka’s desingularization of any F -
hypersurface in Aff n exists.
We shall now take an algebraic number field k as F and denote any p-adic
completion of k by K, the maximal compact subring of K by OK , the maximal ideal
208 JUN-ICHI IGUSA
of OK by πOK , and put OK /πOK = Fq as before. Furthermore, for any F -closed

subset X of Aff n we denote by F [X] the factor ring of F [x] = F [x1 , . . . , xn ] by its
ideal F [x] ∩ I(X). If f (x) is now in k[x]\k, then by Theorem 11.3.1 a Hironaka’s
desingularization h : Y → X = Aff n of the hypersurface in X defined by f (x) exists.
We observe that if all nonzero coefficients of f (x) are units of OK , by applying the
homomorphism OK → Fq to them, we get an f¯(x) in Fq [x]\Fq . The fact is that
for almost all K, i.e., except for a finite number of K, the above h : Y → X gives
rise to a Hironaka’s desingularization of the hypersurface in X̄ = Aff n defined by
f¯(x). It is understood that the new Aff n is relative to an algebraically closed field
containing Fq . In the following, we shall give some details to this basic fact.
We have expressed X × Projm as the union of X × Zα = Aff m+n and Y as the
union of Yα = Y ∩ (X × Zα ) for 1 ≤ α ≤ m + 1; for each α we choose a finite subset
{gβ (t)} of k[t]\{0}, where k[t] = k[t1 , . . . , tm+n ], and express Aff m+n as the union
of its open subset
Uβ = Aff m+n \gβ−1 (0)
so that Yα becomes the union of Yαβ = Yα ∩ Uβ for all β. We keep in mind that
we can get an arbitrarily fine k-open covering of Aff m+n , hence that of Yα , in that
way. We shall impose three conditions on the choice of {gβ (t)}. The first condition
is that for each β there exist h1 (t), . . . , hm (t) in k[t] such that
Yαβ = h−1 −1
1 (0) ∩ . . . ∩ hm (0) ∩ Uβ .
If we write all E’s which intersect Yαβ by E1 , . . . , Ep , then the second condition
is that there exist f1 (t), . . . , fn (t) in k[t] such that Ei ∩ Yαβ = fi−1 (0) ∩ Yαβ for
1 ≤ i ≤ p and

d(t) = ∂(f1 , . . . , fn , h1 , . . . , hm ) ∂(t1 , . . . , tm+n )
is a unit of k[Uβ ]. The third condition is that if we denote the image of ti in

Ω[t]/I(Yα ) by yi for 1 ≤ i ≤ m + n, then

f ◦ hYαβ = ε · fi (y)Ni ,
1≤i≤p

h∗ (dx1 ∧ . . . ∧ dxn )Yαβ = η · fi (y)ni −1 · df1 (y) ∧ . . . ∧ dfn (y),
1≤i≤p
in which ε, η are units of k[Yαβ ]. These conditions are satisfied for a suitable choice
of {gβ (t)}.
We shall express the above conditions by identities in k[t]; we shall use e0 , e,
etc. to denote nonnegative integers.

First of all, by Hilbert’s Nullstellensatz there
exist gβ (t) in k[t] satisfying gβ (t)gβ (t) = 1. If we put
a = k[t]h1 (t) + . . . + k[t]hm (t),
then k[Uβ ] = k[t, 1/gβ (t)] implies k[Yαβ ] = k[Uβ ]/k[Uβ ]a = k[y, 1/gβ (y)]. Therefore,
every element of k[Yαβ ] is of the form P (y)/gβ (y)e0 for some P (t) in k[t] and e0 ; it
is a unit of k[Yαβ ] if and only if P (t)P (t) ≡ gβ (t)e mod a for some P (t) in k[t] and
e. In particular, ε = ε0 (y)/gβ (y)eε , η = η0 (y)/gβ (y)eη and
ε0 (t)ε0 (t) ≡ η0 (t)η0 (t) ≡ gβ (t)e mod a
for some ε0 (t), ε0 (t), η0 (t), η0 (t) in k[t] and eε , eη , e. Furthermore,
d (t)d(t) = gβ (t)e
for some d (t) in k[t] and e, and

gβ (t)e f (t) ≡ ε0 (t)gβ (t)e0 · fi (t)Ni mod a
1≤i≤p
where e = e0 + eε for some e0 . Finally, if we denote the jacobian matrix of

f1 , . . . , fn , h1 , . . . , hm by t1 , . . . , tm+n as
    
A B 1n B A − BD−1 C 0
 =  
−1
C D 0 D D C 1m
with m × m matrix D = D(t), etc., since d(t) is its determinant, we get d(t) =
det(D) det(A−BD−1 C) provided that det(D) = 0. We shall show that det(D(y)) =
0. If for a moment we denote the (i, j)-entries of C, D by cij , dij , then we have

cij (y) dyj + dij (y) dyn+j = 0
1≤j≤n 1≤j≤m
for 1 ≤ i ≤ m. Since dy1 , . . . , dyn are linearly independent, if det(D(y)) = 0, then

we will have rank(C(y) D(y)) < m. This brings the contradiction that d(y) = 0.
Furthermore, the above proof implies that

df1 (y) ∧ . . . ∧ dfn (y) = det(A − BD−1 C)(y) · h∗ (dx1 ∧ . . . ∧ dxn )Yαβ .
Therefore, the condition on h∗ (dx1 ∧ . . . ∧ dxn )|Yαβ can be written as

det(D)(y) = d(y)η · fi (y)ni −1 , i.e.,
1≤i≤p

gβ (t)e det(D) ≡ d(t)η0 (t)gβ (t)e0 · fi (t)ni −1 mod a
1≤i≤p
where e = e0 + eη . We have used the same e0 , e above because that is permissible

after making them larger. Also, P (t) ≡ 0 mod a means, of course, that P (t) =
A1 (t)h1 (t) + . . . + Am (t)hm (t) for some A1 (t), . . . , Am (t) in k[t]. If we replace the
above congruences mod a by equations of this form, we get a finite number of
identities in k[t]. We now take a p-adic completion K of k such that, in addition to
the coefficients of f (x) being units of OK , the coefficients of all P (t) in k[t], which
appear in the above identities, are all in OK . This condition is satisfied by almost
all K. Furthermore, if we replace all such P (t) by its image P̄ (t) in Fq [t], then it is
rather obvious that the data so obtained will give a Hironaka’s desingularization of
the hypersurface in X̄ = Aff n defined by f¯(x).
210 JUN-ICHI IGUSA
11.4 Weil’s zeta functions over finite fields

We shall explain the main theorem on Weil’s zeta functions over finite fields. Since
the proof will not be given, or rather, can not be given because it is beyond the level
of this book, we shall explain two examples. Although the following first example
is incomparably simpler, these are the examples mentioned in Weil’s paper [57] of
1949 in which he announced the above-mentioned theorem as a conjecture. The
explanation will be with proof except for some topological results. More precisely,
if for any d-dimensional compact C-analytic manifold M we define its Poincaré
polynomial P (M, t) as
P (M, t) = Bi ti ,
0≤i≤2d
in which Bi is the i-dimensional Betti number of M , hence Bi = B2d−i by the

Poincaré duality, then we shall just mention the explicit forms of P (M, t) in the
two examples without proof.
We shall explain the first example. If Q(x) is a reduced quadratic form in
n ≥ 3 variables x1 , . . . , xn with coefficients in Fq and Ω is an algebraically closed
field containing Fq , then Q(x) defines an irreducible smooth (n − 2)-dimensional
projective Fq -variety X in Projn−1 . Furthermore,
card(X(Fq )) = 1 + q + . . . + q n−2 + χ(Q)q n/2−1 ,
in which χ(Q) = 0 if n is odd and χ(Q) = ±1 according as Q(x) is hyperbolic or not

if n is even. This follows immediately from the formula for card(Q−1 (0)) in Theorem
9.2.1. On the other hand, if M is the hypersurface in Pn−1 (C), i.e., Projn−1 for
Ω = C, defined as the set of zeros of x1 x2 +. . .+xn−1 xn or x1 x2 +. . .+xn−2 xn−1 +x2n
according to whether n is even or odd, then it is well known that
P (M, t) = 1 + t2 + . . . + t2(n−2) + (1/2)(1 + (−1)n )tn−2 .
We might mention the proof in the special case where n = 4. If we use the notation
in section 11.2, then Proj1 × Proj1 is isomorphic to the surface in Proj3 defined by
t11 t22 − t12 t21 , hence M is C-bianalytic to P1 (C) × P1 (C), and hence
P (M, t) = P (P1 (C), t)2 = 1 + 2t2 + t4 .
We might also mention that if Q(x) is of the above form, then the open subset of
X defined by x1 = 0 is isomorphic to Aff n−2 .
Before we explain the second example, we shall recall a system of equations
defining the Grassmann variety. We shall follow the presentation in G. B. Gurevich
[17]. Wetake an m-dimensional vector space V over an arbitrary field F and
n
examine
0 < n ≤ m. We shall write products of elements
(V ) for a fixed
of
n (V ) without using .n We first observe that if W is a subspace of V , then
n (W ) is a subspace
of (V
). Furthermore, if W1 , W2 are subspaces
of V , then
(W1 ∩ W2 ) = n (W1 ) ∩ n (W2 ). Therefore, for any xin n (V ) there exists
n
the smallest subspace Wx of V such that x is contained in (Wx ). We shall later
give an explicit description of Wx in terms of x. We observe that dimF (Wx ) ≥ n
if x = 0. We say that x = 0 is decomposable if dimF (Wx ) = n, i.e., if x = v1 . . . vn

for some v1 , . . . , vn in Wx necessarily forming its F -basis. On the other hand if
we choose an F -basis e1 , . . . , em for V , then every x in n (V ) can be uniquely
expressed as

x= xi1 ...in ei1 . . . ein
i1 <...<in
with xi1 ...in in F . We define xi1 ...in for all 1 ≤ i1 , . . . , in ≤ m so that we get an
alternating tensor, which we call the representative tensor of x. We shall find the
necessary and sufficient condition in terms of xi1 ...in for x to be decomposable. We
observe that if x is decomposable, i.e., if x = v1 . . . vn for some v1 , . . . , vn in V ,
hence vj = y1j e1 + . . . + ymj em with yij in F , then the representative tensor of x
becomes
xi1 ...in = πi1 ...in (y),
in which y is the m × n matrix with yij as its (i, j)-entry and πi1 ...in (y) for i1 <
. . . < in is the determinant of the n × n submatrix of y obtained by crossing out its
k-th rows for k = i1 , . . . , in , and they are not all 0.
n
We take x from (V ) and ϕ from the dual space V ∗ of V . If xi1 ...in is the

representative tensor of x, then we define ∂ϕ x in n−1 (V ) as

(∂ϕ x)i1 ...in−1 = xi1 ...in−1 i ϕ(ei )
1≤i≤m
for all i1 , . . . , in−1 . We observe that ∂ϕ x is determined

by x and ϕ independently of
the choice of e1 , . . . , em . Furthermore if x, x are in n (V ) with the representative
tensors xi1 ...in , xi1 ...in and if ∂i = ∂ϕ for ϕ = ϕi in V ∗ for 1 < i ≤ n, then we can
easily verify the formal identity
(∗) (x(∂2 . . . ∂n x ))i1 ...in+1

= (−1)n+1 zi1 ...in+1 ,j2 ...jn ϕ2 (ej2 ) . . . ϕn (ejn ) ,
1≤j2 ,... ,jn ≤m
in which

zi1 ...in+1 ,j2 ...jn = (−1)k xi1 ...ik−1 ik+1 ...in+1 xik j2 ...jn ,
1≤k≤n+1
for i1 < . . . < in+1 .

In the above notation we shall show that Wx and the F -span W of ∂2 . . . ∂n x for
all ϕ2 , . . . , ϕn in V ∗ coincide. If we choose an F -basis e1 , . . . , em for V such that
e1 , . . . , er form an F -basis for Wx , then xi1 ...in = 0 only for i1 , . . . , in ≤ r, hence
∂2 . . . ∂n x is in Wx , and hence W is contained in Wx . If we choose another F -basis
e1 , . . . , em for V such that e1 , . . . , es form an F basis for W , then

xii2 ...in ϕ2 (ei2 ) . . . ϕn (ein ) = 0
1≤i2 ,... ,in ≤m
212 JUN-ICHI IGUSA
for all i > s and for all ϕ2 , . . . , ϕn in V ∗ . This implies that xii2 ...in = 0 for all
i2 , . . . , in if i > s, hence xi1 ...in = 0 only for i1 , . . . , in ≤ s. Therefore x is in
n
(W ), hence Wx is contained in W , hence Wx = W .
We are ready to show that x = 0 in n (V ) is decomposable if and only if its
representative tensor satisfies the following quadratic equations:

(∗∗) (−1)k xi1 ...ik−1 ik+1 ...in+1 xik j2 ...jn = 0
1≤k≤n+1
for all i1 < . . . < in+1 and j2 , . . . , jn . First, suppose that x is decomposable. Then
xv = 0 for all v in Wx , hence for all v = ∂2 . . . ∂n x. Then by (∗) we get (∗∗). Next,
suppose that (∗∗) is satisfied. Then by (∗) we get xWx = 0. If we choose an F -basis
v1 , . . . , vr for Wx , then x can be expressed uniquely as an F -linear combination of
vi1 . . . vin for 1 ≤ i1 < . . . < in ≤ r. Since xvi = 0 for 1 ≤ i ≤ r, we see that r = n
and x is in F v1 . . . vn .
We shall now explain the second example. We shall take Fq , Ω and later C
as F . We denote by Y the open subset of Mm,n consisting of all y with linearly
independent columns and consider ProjN for N + 1 = m n with homogeneous
coordinates xi1 ...in for 1 ≤ i1 < . . . < in ≤ m. Then the correspondence y →
x where xi1 ...in = πi1 ...in (y) gives a morphism from Y to the projective variety
X in ProjN defined by (∗∗), which is called the Grassmann variety. We have
seen in Chapter 10.6, though in different language, that the open subset of X
defined by xi1 ...in = 0 is isomorphic to Mm−n,n for every i1 , . . . , in . Therefore,
X is an irreducible smooth d-dimensional projective F -variety for d = (m − n)n.
Furthermore, if F = Fq , then the results there imply that

card(X(Fq )) = (1 − q m−k+1 )/(1 − q k ).
1≤k≤n
We recall that in the notation of Chapter 9.6 the RHS is Fm−n,n (q). We further
recall the Gauss identity:

Fm−n,n (q) q n(n−1)/2 tn = (1 + q i t).
0≤n≤m 0≤i≤m−1
If we equate the coefficients of tn on both sides, we get

Fm−n,n (q) = q −n(n−1)/2 · q i1 +...+in ,
in which the summation is for 0 ≤ i1 < . . . < in ≤ m − 1. Therefore, if we replace

ik by fk + k − 1 for 1 ≤ k ≤ n, we get

card(X(Fq )) = B2k q k ,
0≤k≤d
in which B2k denotes the number of partitions k = f1 + . . . + fn satisfying 0 ≤ f1 ≤

. . . ≤ fn ≤ m − n. We observe that if we put fi∗ = (m − n) − fn−i+1 for 1 ≤ i ≤ n,
then d − k = f1∗ + . . . + fn∗ and 0 ≤ f1∗ ≤ . . . ≤ fn∗ ≤ m − n, hence B2k = B2d−2k for
0 ≤ k ≤ d. On the other hand, if we take C as Ω and denote the corresponding X

by M , then
P (M, t) = B2k t2k .
0≤k≤d
This fact was known in classical algebraic geometry dealing with Schubert varieties.
We refer to C. Ehresmann [13] for a topological proof.
After these rather special examples, we shall state the main theorem on Weil’s
zeta functions. We recall that an element α of an extension of Q is called an
algebraic integer if it is a zero of a monic polynomial with coefficients in Z.
Theorem 11.4.1 Let X denote a smooth everywhere d-dimensional projective Fq -
variety. Then there exist nonnegative integers Bi for 0 ≤ i ≤ 2d satisfying Bi =
B2d−i and for each i algebraic integers αij = 0, in fact, all their conjugates over Q
in C having the absolute value q i/2 , for 1 ≤ i ≤ Bi with the following properties:
Firstly,
card(X(Fqe )) = (−1)i αij
e
0≤i≤2d 1≤j≤Bi
for all e ≥ 1 and secondly,
{q d /αij ; 1 ≤ j ≤ Bi } = {α2d−i,j ; 1 ≤ j ≤ B2d−i }
for 0 ≤ i ≤ 2d.
As we have stated in the Introduction, this is one of the two theorems which
we shall use without proof. Actually, the deeper part of the theorem stating that
αij are algebraic integers of absolute value q i/2 will not be used. At any rate, the
theorem stated above was conjectured by A. Weil and proved, in the above general
form by A. Grothendieck and P. Deligne. We refer to Deligne [8] for the history up
to his decisive contribution after the work of B. Dwork, Grothendieck and others
all with references. We might at least recall that Weil’s zeta function Z(t) of X is
defined for t near 0 in C as

dlogZ(t)/dt = card(X(Fqe ))te−1 , Z(0) = 1.
e≥1
The first and the second parts of the theorem then imply its rationality

Z(t) = (1 − αij t)εi , εi = (−1)i+1
0≤i≤2d 1≤j≤Bi
and its functional equation
Z(1/q d t) = ±(q d/2 t)χ Z(t),

in which χ = (−1)i Bi = B0 −B1 +. . .+B2d . We observe that the above expression

of Z(t) shows that the formula for card(X(Fqe )) determines the set {αij ; 0 ≤ i ≤
2d, 1 ≤ j ≤ Bi } and further, if we incorporate the statement on the absolute value
of αij , the set {αij ; 1 ≤ j ≤ Bi } for each i.
214 JUN-ICHI IGUSA
Remark. Some irreducible components of X in Theorem 11.4.1 may not be an

Fq -variety and yet Theorem 11.4.1 in the case where X is irreducible implies the
general case. We shall give an example which will illustrate this situation. We take
any θ from Fq3 which is not a zero of
2 2
P (t) = t − tq t + tq + tq t + tq − 2tq ;
such a θ exists for every q and Fq3 = Fq (θ). We denote by σ the Fq -automorphism
of Fq3 defined by σθ = θ q and by a, b the points of Proj3 (Fq3 ) respectively with
(θ, σθ, σ 2 θ, 1), (θ, σ 2 θ, σθ, 0) as their homogeneous coordinates. We further denote
by L, L , L the lines in Proj3 respectively through {a, b}, {σa, σb}, {σ 2 a, σ 2 b}.
Then by the choice of θ we see that L, L , L are mutually disjoint, σ gives a
permutation L → L , L → L , L → L, and their union X is a smooth everywhere
1-dimensional projective Fq -variety. Since card(Proj1 (Fq )) = 1 + q for every q, we
get card(X(Fqe )) = 3(1 + q e ) or 0 according as e ≥ 1 is in 3Z or not. If ζ is any
element of C satisfying 1 + ζ + ζ 2 = 0, then we can write
card(X(Fqe )) = (1 + ζ e + ζ 2e )(1 + q e )
for every e ≥ 1. Therefore, B0 = B2 = 3, B1 = 0, {α01 , α02 , α03 } = {1, ζ, ζ 2 },

and {α21 , α22 , α23 } = {q, ζq, ζ 2 q}. In the general case the situation, hence also the
proof, is entirely similar.
11.5 Degree of Z(s)

We shall start with the following problem proposed in [25]: Assume that a homo-
geneous polynomial f (x) with coefficients in OK has a good reduction mod π; do
we then always have degt (Z(s)) + deg(f ) = 0? If we start with a homogeneous
polynomial f (x) with coefficients in an algebraic number field k, then we can ask
the above relation for almost all p-adic completions K of k thus avoiding the use of
“good reduction mod π”. We recall that if the critical set Cf of f is contained in
{0}, i.e., if the discriminant of f (x) is not 0, then Proposition 10.2.1 implies that the
answer is affirmative. In fact this is one of the examples which motivated the above
problem. Now D. Meuser [42] showed more generally that the answer is affirmative
if Cf is contained in the union of the coordinate hyperplanes. In connection with
other examples, we remark that if f (x) is a basic relative invariant of an irreducible
regular prehomogeneous vector space, this condition on Cf implies deg(f ) ≤ 2. In
the general case, i.e., for an arbitrary homogeneous polynomial f (x) in k[x]\k, the
problem was settled by J. Denef [9]. In the following we shall explain how Denef
solved the problem.
We shall use the same notation as in section 11.3 and, in fact, the setup af-
ter Theorem 11.3.1. We recall that f (x) is arbitrary in k[x]\k, Y is a closed
smooth n-dimensional k-subvariety of X × Projm , and h : Y → X is a Hiron-
aka’s desingularization of the hypersurface in X = Aff n defined by f (x). Further-
more, the p-adic completion K of k satisfies the condition that the homomorphism
OK → Fq = OK /πOK can be applied to the coefficients of a finite number of poly-

nomial identities which describe the above desingularization to give a Hironaka’s
desingularization h̄ : Ȳ → X̄ of the hypersurface in X̄ = Aff n defined by the image
f¯(x) of f (x) in Fq [x]. We observe that Projm (K) can also be defined as the fac-
×
tor space of OK m+1
\πOK m+1
by OK and that it is covered by (Zα )◦ consisting of all
points with homogeneous coordinates in OK and the α-th coordinate equal to 1 for
1 ≤ α ≤ m + 1. We denote Projm (K) with the above open covering by (Projm )◦ ,
X(OK ) = OK n
by X ◦ , put
(X × Projm )◦ = X ◦ × (Projm )◦ , (X × Zα )◦ = X ◦ × (Zα )◦ ,
and define Y ◦ , (Yα )◦ as their respective intersections with Y (K) for all α. We
introduce the coordinates (t1 , . . . , tm+n ) as in section 11.3 and identify (X × Zα )◦
with OK m+n
. If we denote by (Yαβ )◦ the open subset of (Yα )◦ consisting of all b such
m+n
that gβ (b) is a unit of OK , i.e., the set of all b in OK satisfying h1 (b) = . . . =
×
hm (b) = 0 and gβ (b) in OK , then (Yα ) becomes the union of (Yαβ )◦ for all β. We
◦
keep in mind that X ◦ and Y ◦ are compact n-dimensional K-analytic manifolds.

We shall restrict our attention to h|Y ◦ : Y ◦ → X ◦ and also to h̄|Ȳ (Fq ) : Ȳ (Fq ) →
X̄(Fq ) sometimes without explicitly saying so. We denote by dx the normalized
Haar measure on X ◦ and by µY the measure on Y ◦ defined by h∗ (dx1 ∧ . . . ∧ dxn )
as in Chapter 7.4. We take a arbitrarily from X ◦ and for Re(s) > 0 we put

Za (s) = |f (x)|K dx =
s
|f (h(y))|sK µY (y).
a+πX ◦ h−1 (a+πX ◦ )
We observe that Za (s) depends on the image ā of a in X̄(Fq ) rather than a itself.
If we take any b from h−1 (a + πX ◦ ), then b is in (Yαβ )◦ for some α, β. We observe
that if b is any point of Y ◦ with the same image in Ȳ (Fq ) as b, then b is also in
(Yαβ )◦ because b is in (Yα )◦ and gβ (b ) is a unit of OK . Furthermore,
(∂(f1 , . . . , fn , h1 , . . . , hm )/∂(t1 , . . . , tm+n ))(b)
is also a unit of OK . Therefore if we put b = b + πy, where y is a variable in OK

m+n
,
then by Lemma 7.4.3 we see that the correspondence
b → (f1 (b ), . . . , fn (b ), h1 (b ), . . . , hm (b ))

m+n m+n
gives a K-bianalytic map from b + πOK to c + πOK , in which c is the image

of b. Furthermore, if we introduce zi as fi (b ) = ci + πzi for 1 ≤ i ≤ n, where
c = (c1 , . . . , cm+n ), then the set of all b in Y ◦ having the same image as b in
Ȳ (Fq ) becomes K-bianalytic to OK n
under the correspondence b → (z1 , . . . , zn ).
Therefore, if I is the set of all i satisfying f¯i (b̄) = 0, then after replacing π −1 ci + zi
by zi we get

|f (h(b + πy))|K = |ε · fi (b + πy)Ni |K = |πzi |N
K ,
i
i∈I i∈I

µY (b + πy) = |η · fi (b + πy)ni −1 |K · q −n dz = q −n · |πzi |nKi −1 · dz.
i∈I i∈I
216 JUN-ICHI IGUSA
This implies

Ni s+ni −1
Za (s) = q −n · |πzi |K dz,
n
OK
b̄ i∈I
−1
in which the summation is over the set h̄ (ā)(Fq ) and I for each b̄ can be identified
with the set of all Ē’s which contain b̄. Since the integral of |z|s−1K over OK is
(1 − q −1 )/(1 − q −s ), the above formula implies the following theorem of Denef:
Theorem 11.5.1 Take an arbitrary subset I of E and a point a of X ◦ with its
image ā in X̄(Fq ); denote by cI = cI (ā) the number of all b̄ in h̄−1 (ā)(Fq ) such that
b̄ is in Ē(Fq ) if and only if E is in I. Then

Za (s) = |f (x)|sK dx = q −n · cI · ((q − 1)/(q NE s+nE − 1)).
a+πX ◦ I E∈I
Corollary 11.5.1 We have

lim q n Za (s) ≡ 1 mod q
Re(s)→−∞
in the sense that the LHS is in 1 + qZ.

Proof. If char(Fq ) = p, since (1 + qZp ) ∩ Z = 1 + qZ, we have only to show that the
LHS is both in 1 + qZp and in Z. By definition, we have

q n Za (0) = q n · lim |f (x)|sK dx = q n · q −n = 1
s→0 a+πX ◦
and by Theorem 11.5.1 we also have

q n Za (0) = cI · (q − 1)/(q nE − 1) ≡ cI mod qZp .
I E∈I I
On the other hand again by Theorem 11.5.1 we have

lim q n Za (s) = cI (1 − q)card(I) ≡ cI mod q,
Re(s)→−∞
I I

and cI ≡ 1 mod qZp .

Corollary 11.5.2 If f (x) is homogeneous and t = q −s , then
degt (Z(s)) = − deg(f ).
Proof. Put deg(f ) = d. Then, as in Chapter 10.3, we have

Z(s) = (1 − q −n td )−1 · |f (x)|sK dx.
X ◦ \πX ◦
Therefore, by Corollary 11.5.1 we get

lim td Z(s) = − lim qn · |f (x)|sK dx ≡ −(q n − 1) ≡ 1 mod q.
|t|→∞ Re(s)→−∞ X ◦ \πX ◦
Consequently, the above limit is finite and different from 0. Since Z(s) is a rational
function of t, this implies that degt (Z(s)) = −d.
Corollary 11.5.3 If we put

ĒI = Ē, ĒI0 = ĒI \ Ē ,
E∈I E∈I
then we have

Z(s) = q −n · card(ĒI0 (Fq )) · (q − 1)/(q NE s+nE − 1) .
I E∈I
Proof. In the notation of Theorem 11.5.1 we have

cI (ā) = card b̄ ∈ ĒI0 (Fq ); b̄ ∈ h̄−1 (ā) = card(ĒI0 (Fq )).
ā∈X̄(Fq ) ā∈X̄(Fq )
This implies the corollary.

We might mention that if I = ∅, then

card(ĒI0 (Fq )) = card Ȳ (Fq )\ Ē(Fq ) = card X̄(Fq )\f¯−1 (0)(Fq ) .
E∈E
Therefore, the contribution of I = ∅ to Z(s) is 1 − q −n · card(f¯−1 (0)(Fq )).
11.6 The field Ke (a digression)

We shall prove, just for the sake of completeness, some basic facts on algebraic
extensions of p-adic fields. In a fixed algebraically closed field there exists a unique
extension of Fq of any given degree e ≥ 1, i.e., Fqe . The corresponding statement
for a p-adic field is as follows:
Theorem 11.6.1 Let K denote a p-adic field with Fq as its residue class field.
Then in a fixed algebraically closed field there exists a unique extension Ke of K of
any given degree e ≥ 1 with Fqe as its residue class field.
We shall give a classical proof by using Hensel’s lemma. We shall also prove
the fact that any extension of K of finite degree is a p-adic field. We shall start
with a simple remark; for the time being the finiteness of the residue class field
F = OK /πOK will not be used.
We take an element A of Mm,n (OK ) for any m, n ≥ 1 and denote by Ā its
image in Mm,n (F ). If Āx̄ = ȳ is solvable by x̄ in F n for every ȳ in F m , then Ax = y
n m
is solvable by x in OK for every y in OK . In fact, the above solvability of Āx̄ = ȳ
is equivalent to rank(Ā) = m necessarily with m ≤ n, and this is equivalent by
Lemma 7.4.1 to the existence of g, h respectively in GLm (OK ), GLn (OK ) satisfying
A = g(1m 0)h. Then the equation Ax = y can be rewritten as (1m 0)hx = g −1 y.
We may clearly replace hx, g −1 y respectively by x, y. Then the equation becomes
(1m 0)x = y, which is solvable.
If now f (t), etc. are elements of OK [t], where t is a variable, we shall denote
their images in F [t] by f¯(t), etc. We take g0 (t), h0 (t) from OK [t] such that ḡ0 (t),
218 JUN-ICHI IGUSA
h̄0 (t) are relatively prime and ḡ0 (t) = 0, deg(ḡ0 ) = deg(g0 ). Also, we fix an integer
d ≥ deg(g0 h0 ) and take any c(t) from OK [t] with deg(c) ≤ d. Then there exist a(t),
b(t) in OK [t] satisfying
a(t)h0 (t) + b(t)g0 (t) = c(t), deg(a) < deg(ḡ0 ), deg(b) ≤ d − deg(ḡ0 ).
In fact, since ḡ0 (t), h̄0 (t) are relatively prime, we have
ā(t)h̄0 (t) + b̄(t)ḡ0 (t) = c̄(t)
for some ā(t), b̄(t) in F [t]. By replacing ā(t) by its residue mod ḡ0 (t), we may
assume that deg(ā) < deg(ḡ0 ). We then have deg(b̄) ≤ d − deg(ḡ0 ). Therefore, we
have only to apply the above remark with m = n = d + 1 to the coefficients of a(t),
b(t) as x and the coefficients of c(t) as y. We are ready to prove Hensel’s lemma,
which is as follows:
Lemma 11.6.1 Let f (t) denote any element of OK [t]\OK such that its image f¯(t)
in F [t], where F = OK /πOK , splits as f¯(t) = ḡ0 (t)h̄0 (t), in which ḡ0 (t), h̄0 (t) are
relatively prime and ḡ0 (t) = 0. Then there exist g(t), h(t) in OK [t] with deg(g) =
deg(ḡ0 ) satisfying
f (t) = g(t)h(t), ḡ(t) = ḡ0 (t), h̄(t) = h̄0 (t).
Proof. We shall exclude the trivially simple case where f¯(t) = 0, hence h̄0 (t) = 0.
We put d = deg(f ) and choose g0 (t), h0 (t) from OK [t] with ḡ0 (t), h̄0 (t) as their
images in F [t] and satisfying deg(g0 ) = deg(ḡ0 ), deg(h0 ) = deg(h̄0 ). That is clearly
possible and d ≥ deg(f¯) = deg(ḡ0 h̄0 ) = deg(g0 h0 ). We introduce unknown elements
gi (t), hi (t) of OK [t] satisfying
deg(gi ) < deg(ḡ0 ), deg(hi ) ≤ d − deg(ḡ0 )
for i = 1, 2, 3, . . . and put

g(t) = π i gi (t), h(t) = π i hi (t).
i≥0 i≥0
We have only to show that the equation f (t) = g(t)h(t) is solvable. If we put
c1 (t) = π −1 (f (t) − g0 (t)h0 (t)) and ck (t) = 0 for k > 1, then ck (t) is in OK [t] and
deg(ck ) ≤ d for every k. Furthermore, the equation g(t)h(t) = f (t) can be replaced
by the following sequence of equations:

gk (t)h0 (t) + hk (t)g0 (t) = ck (t) − gi (t)hk−i (t)
0<i<k
for k = 1, 2, 3, . . . . The first equation is g1 (t)h0 (t) + h1 (t)g0 (t) = c1 (t), which is
solvable in g1 (t), h1 (t) by the previous observation. Therefore, we shall apply an
induction on k assuming that k > 1. Then in the k-th equation, the RHS is an
already known polynomial in OK [t] of degree at most d, hence for the same reason
it is also solvable in gk (t), hk (t).
Corollary 11.6.1 If f0 (t) is an irreducible monic polynomial in K[t] with f0 (0) in

OK , then necessarily f0 (t) is in OK [t].
Proof. Put d = deg(f0 ) and assume that f0 (t) is not in OK [t]. Choose the smallest
integer e so that f (t) = π e f0 (t) is in OK [t]. Then e > 0, deg(f ) = d, and 0 <
deg(f¯) < d. In Lemma 11.6.1 we can take ḡ0 (t) = f¯(t), h̄0 (t) = 1 and we get
f (t) = g(t)h(t) with 0 < deg(g) < d. This implies that f0 (t) is reducible in K[t], a
contradiction.
Proposition 11.6.1 Let K denote a complete nonarchimedean field with πOK for
some π = πK as the ideal of nonunits of OK ; let L denote a finite extension of K.
Then L is also a complete nonarchimedean field with πL OL for some πL as the ideal
of nonunits of OL . Furthermore, if we define r, e as
r
πOL = πL OL , [OL /πL OL : OK /πOK ] = e,
then [L : K] = re.
Proof. We shall first show that L has a nonarchimedean absolute value |·|L satisfying
|a|L = |a|nK for every a in K, where n = [L : K]. We start with a review of the
norm N from L to K for any field K. We choose a K-basis u1 , . . . , un for L. Then
we get a K-algebra homomorphism ρ : L → Mn (K) as
α(u1 . . . un ) = (u1 . . . un )ρ(α).
Since ρ(α) is unique up to ρ(α) → gρ(α)g −1 for some g in GLn (K) coming from
a change of the K-basis for L, N α = det(ρ(α)) is well defined. Furthermore,
N (αβ) = (N α)(N β) for every α, β in L. We need another property of N α. If f0 (t)
is the irreducible monic polynomial in K[t] of degree d satisfying f0 (α) = 0, then
1, α, . . . , αd−1 form a K-basis for K(α). If we compute the norm N0 α of α from
K(α) to K by using this basis, we get N0 α = (−1)d f0 (0). If we choose a K(α)-basis
v1 , . . . , ve for L, then
v1 , αv1 , . . . , αd−1 v1 , . . . , ve , αve , . . . , αd−1 ve
form a K-basis for L so that n = de. If we compute N α by using this basis, we get
N α = (N0 α)e = (−1)n f0 (0)e .
We now go back to our field K, and in the above notation we put
|α|L = |N α|K = |f0 (0)|eK .
Then |a|L = |a|nK for every a in K. Furthermore, | · |L satisfies AV 1, AV 2 in

Chapter 2.1. We shall show that it also satisfies AV 3 in Chapter 2.2. Since
|αβ|L = |α|L |β|L for every α, β in L, we have only to show that |α + 1|L ≤ 1
if |α|L ≤ 1. Since |α|L ≤ 1 implies |f0 (0)|K ≤ 1, by the above corollary f0 (t) is
in OK [t]. Furthermore, f1 (t) = f0 (t − 1) is also an irreducible monic polynomial
of degree d in K[t] with coefficients in OK [t] and f1 (α + 1) = f0 (α) = 0. Since
f1 (0) = f0 (−1) is in OK , we have |α + 1|L = |f1 (0)|eK ≤ 1.
220 JUN-ICHI IGUSA
We shall show that L is complete, i.e., every Cauchy sequence {αi } in L is

convergent. We say that a sequence is zero if its terms are all zero. If we write

αi = aij uj
1≤j≤n
with aij in K for all i, j then, since K is complete, we have only to show that
{ai1 }, . . . , {ain } are all Cauchy sequences in K. If k denotes the number of nonzero
sequences among them, the statement clearly holds for k ≤ 1. We shall therefore
assume that k > 1 and apply an induction on k. If we can derive a contradiction
from the assumption that one of the k sequences is not a Cauchy sequence, then
the induction will be complete. By converting the double sequence {αi − αj } into
a sequence and replacing it by a subsequence, we may assume that {αi } is a null
sequence in L while the absolute values of all terms of one of the k sequences in
K are at least > 0. After a permutation of u1 , . . . , un , we may assume that
{ai1 }, . . . , {aik } are the nonzero sequences and |aik |K ≥ for all i. Then the
sequence in L with
αi /aik − uk = (aij /aik )uj
1≤j<k
as its i-th term is convergent, hence a Cauchy sequence, and hence the coefficients
of u1 , . . . , uk−1 form Cauchy sequences in K by induction. If we denote their limits
by b1 , . . . , bk−1 , we get b1 u1 + . . . + bk−1 uk−1 + uk = 0. This contradicts the fact
that u1 , . . . , un form a K-basis for L.
Finally, since the image of L× under |·|L is a subgroup of the image of K × under
| · |K , it is discrete. Therefore, the ideal of nonunits of OL can be written as πL OL ,
r
and πOL = πL OL for some positive integer r. On the other hand, if w1 , . . . , we
are the elements of OL such that their images w̄1 , . . . , w̄e in OL /πL OL are linearly
independent over F = OK /πOK , then w1 , . . . , we are linearly independent over K,
hence e ≤ [L : K]. We shall assume that w̄1 , . . . , w̄e form an F -basis for OL /πL OL .
Since
r−1 r−1
1, πL , . . . , πL , π, ππL , . . . , ππL ,...
are elements of OL whose orders are respectively 0, 1, 2, . . . , every element α of
OL can be written as
i
α= aij wj πL
0≤i<r 1≤j≤e
with aij in OK for all i, j. We shall show that the expression is unique. We have
only to show that if the RHS is 0, then aij = 0 for all i, j. Suppose that the RHS is
0 for some aij in OK not all 0. Then, after cancelling a power of π, we may assume
that they are not all in πOK , hence their images āij in OK /πOK are not all 0. If
we take the image of the RHS in OL /πL OL , then we get ā01 w̄1 + . . . + ā0e w̄e = 0,
hence ā0j = 0, and hence a0j = πb0j with b0j in OK for all j. We can repeat the
same argument after cancelling πL , and we get a1j = πb1j with b1j in OK for all
j. By continuing this process, we will see that all aij are in πOK , a contradiction.
If we allow aij to have a power of π as a denominator, then every α in L can be
expressed uniquely as above with aij in K. We have thus shown that [L : K] = re.
In the above proposition if K is a p-adic field, i.e., if F = OK /πOK is finite, then

OL /πL OL is also finite, hence L is a p-adic field. Furthermore, if | · |K is normalized
−1
as |π|K = q −1 , then |·|L is normalized as |πL |L = qL , in which q = card(OK /πOK )
and qL = card(OL /πL OL ). This follows from [L : K] = re. We might mention that
the standard notation for r and e are respectively “e” and “f .” At any rate, we say
that L is unramified if r = 1. Theorem 11.6.1 then states the unique existence of an
unramified extension Ke of K of any given degree e ≥ 1. The proof is as follows:
We shall start with the existence of Ke . We put
∗
P (t) = te − 1 = (t − ζ),
ζ
where e∗ = q e − 1, denote by W the set of all ζ above, and put L = K(W ). We

shall show that L = Ke . We know by Proposition 11.6.1 that L is a p-adic field.
We observe that W forms a cyclic group and it is mapped isomorphically, say, to
W̄ under the homomorphism OL → OL /πl OL . Since Fqe = Fq (W̄ ), it is contained
in OL /πL OL , hence
e = [Fqe : Fq ] ≤ [OL /πL OL : Fq ] ≤ [L : K].
Therefore, if we can show that [L : K] = e, then OL /πL OL = Fqe and L = Ke .

We take a generator ζ of W and denote by f (t) the irreducible monic polynomial
in K[t] satisfying f (ζ) = 0. Since f (t) is a factor of P (t), a power of f (0) is 1,
hence |f (0)|K = 1, and hence f (t) is in OK [t] by Corollary 11.6.1. Furthermore,
deg (f ) = [L : K]. If ζ̄ is the image of ζ in W̄ , then Fq (ζ̄) = Fqe . Therefore, f¯(t)
has a factor of degree e in Fq [t]. Since f¯(t) splilts into a product of distinct linear
factors in Fqe [t], by Lemma 11.6.1 we see that f (t) has a factor of degree e in OK [t],
hence deg(f ) = e.
We shall prove the uniqueness of Ke . Namely, if M is any extension of K of
degree e satisfying OM /πM OM = Fqe , then M = L. We observe that the above
f (t) is in OM [t] and that f (t) splits into linear factors in OM [t] by Lemma 11.6.1.
Therefore ζ is in OM , hence L is contained in M . Since they have the same degree
over K, we get M = L.
11.7 Functional equation of Z(s)

We shall start with the following conjecture proposed in the preliminary but pre-
printed form of Part I of [29] which is mentioned in its introduction:
“(F1) f (x) is a homogeneous polynomial in OK [x1 , . . . , xn ] with good reduction
mod π;
(F2) there exists an element Z(u, v) of Q(u, v) such that for every finite alge-
braic extension L of K and Re(s) > 0 we have

−1 −s
|f (x)|sL dx = Z(qL , qL ).
n
OL
222 JUN-ICHI IGUSA
. . . . It is very likely that the two conditions are sufficient for the functional equa-
tion Z(u−1 , v −1 ) = v deg(f ) · Z(u, v).” The conjecture was investigated by D. Meuser
by using Denef’s formula in Theorem 11.5.1 and, under a certain condition on the
numerical data, was settled by her discovery of the relation between the conjectural
functional equation and the functional equations of Weil’s zeta functions over fi-
nite fields. Later J. Denef succeeded in removing this condition and his proof was
simplified by J. Oesterlé. We shall explain Denef-Meuser’s joint paper [10] on this
subject. As we shall see, their results go much further than the conjecture. We
refer to their paper for more detailed history and further important results.
Lemma 11.7.1 Let Ai denote a subset of a finite set S for all i in an index set I0
and I, J subsets of I0 . Then for any given I, we have

card Ai \ Ai = (−1)card(J\I) card Ai .
i∈I i∈I J⊃I i∈J
Proof. We take a arbitrarily from S and show that a contributes the same number
to both sides of the identity. Now a is contained in Ai1 , . . . , Aip for some i1 , . . . , ip
but not in Ai for any other i. First, suppose that I is not contained in {i1 , . . . , ip },
i.e., a is not contained in Ai for some i in I. Then, clearly, a contributes 0 to both
sides. Next, suppose that I is contained in {i1 , . . . , ip }. We put n = p − card(I).
If n = 0, i.e., I = {i1 , . . . , ip }, then a contributes 1 to both sides. If n > 0, then
a contributes 0 to the LHS. On the other hand, a is contained in Ai for all i in J
if and only if J is a subset of {i1 , . . . , ip }. Therefore, the contribution of a to the
RHS is

n
(−1)j = (1 − 1)n = 0.
j
0≤j≤n
Lemma 11.7.2 Let I0 denote a finite index set, i an element of I0 , and I, J subsets
of I0 ; further, let xi , yJ denote variables. Then we have

(−1)card(J\I) yJ · xi = yI · (xi − 1).
I J⊃I i∈I I i∈I
Proof. We clearly have

yI · (1 + xi ) = yI xi = yI xi .
I i∈I I J⊂I i∈J J I⊃J i∈J
If we permute I, J on the RHS and replace xi , yI by −xi , (−1)card(I) yI on both

sides, then we get the identity in the lemma.
We shall use the same setup as in section 11.5, e.g., k is an algebraic number field,
f (x) is a homogeneous polynomial in k[x]\k of degree d, where k[x] = k[x1 , . . . , xn ],
and h : Y → X = Aff n is a Hironaka’s desingularization of the hypersurface in
Aff n = Ωn defined by f (x). We now need a Hironaka’s desingularization of the
hypersurface in Projn−1 defined by f (x), which can be explained, e.g., as follows.
We observe that every λ in Ω× gives an Ω-automorphism of Ω[x] as xi → λxi for
1 ≤ i ≤ n. This gives rise to an action of Ω× on X under which the hypersurface
in X defined by f (x) is invariant. The fact, which we accept, is that the above
h : Y → X can be constructed in such a way that it is preserved under the action
of Ω× . We shall give some details. We recall that X ∗ = Projn−1 is the factor
space of X\{0} by Ω× . If Y ∗ denotes the factor space of Y \h−1 (0) by Ω× , then Y ∗
becomes a smooth (n − 1)-dimensional k-subvariety of X ∗ × Projm and h gives rise
to a Hironaka’s desingularization h∗ : Y ∗ → X ∗ of the hypersurface in X ∗ defined
by f (x). Furthermore, if E denotes the set of all E in E not contained in h−1 (0),
then for every E in E the factor space E ∗ of E\h−1 (0) by Ω× becomes a smooth
(n − 2)-dimensional k-subvariety of Y ∗ . More generally, for every subset I of E

EI∗ = E∗
E∈I
is either empty or a smooth k-subvariety of Y ∗ of dimension rI = n − card(I) − 1.

Furthermore, if K is a p-adic completion of k which we have used in section 11.5,
then a similarly defined ĒI∗ becomes a smooth rI -dimensional projective Fq -variety
and
card(ĒI∗ (Fq )) = (q − 1)−1 · card (ĒI \h̄−1 (0))(Fq ) .
On the other hand, if we put
xE = (q − 1)/(q NE s+nE − 1),
then by applying Theorem 11.5.1 to the integral in

Z(s) = (1 − q −(ds+n) )−1 · |f (x)|sK dx
X ◦ \πX ◦
we will have

|f (x)|sK dx = q −n card (ĒI0 \h̄−1 (0))(Fq ) · xE .
X ◦ \πX ◦ I⊂E E∈I
We shall now apply Lemmas 11.7.1, 11.7.2 to the above summation in I with I0 = E ,
AE = (Ē\h̄−1 (0))(Fq ), and

yI = card AE .
E∈I
In doing so, we use the obvious fact that if S, S0 are subsets of a set, then the op-
eration S → S\S0 commutes with the taking of union, intersection, and difference.
We see that yI = (q − 1)card(ĒI∗ (Fq )) and that the above summation in I is equal
to
(q − 1) · card(ĒI∗ (Fq )) · (xE − 1).
I⊂E E∈I
Since we shall be using only the desingularization h∗ : Y ∗ → X ∗ of the hypersurface

in X ∗ = Projn−1 defined by f (x), we change our notation and state the above result
in the following self-contained form:
224 JUN-ICHI IGUSA
Theorem 11.7.1 If f (x) is a homogeneous polynomial of degree d > 0 in n vari-

ables with coefficients in an algebraic number field k, then there exists a finite set
E = {E}, where each E is equipped with a pair of positive integers (NE , nE ), such
that for almost all p-adic completion K of k we have
Z(s) =(q − 1)q ds /(q ds+n − 1)

· card(ĒI (Fq )) · (q − 1)/(q NE s+nE − 1) − 1 ,
I⊂E E∈I
in which every ĒI is either empty or a smooth rI -dimensional projective Fq -variety

for rI = n − card(I) − 1.
We keep in mind that we can replace the above K by any finite algebraic ex-
tension L, e.g., by Ke in Theorem 11.6.1. We shall now explain an observation
by J. Oesterlé about the extendability of certain functions on N\{0} to Z\{0}. It
depends on the following lemma:
Lemma 11.7.3 We denote by A the ring of functions on N\{0} generated by n →

αn for all α in C× , i.e., functions of the form

ϕ(n) = mi αin
1≤i≤r
with mi in Z and αi in C× for 1 ≤ i ≤ r, and for some r ≥ 0. If mi = 0 and

αi = αj for all i and i = j, then the set {(mi , αi ); 1 ≤ i ≤ r} is uniquely determined
by ϕ.
Proof. If ϕ is as above and t is a complex variable satisfying |αi t| < 1 for 1 ≤ i ≤ r,

then
exp ϕ(n)tn /n = (1 − αi t)−mi .
n>0 1≤i≤r
We observe that the rational function of t so defined depends only on ϕ and that
it has αi−1 as a pole or a zero of order |mi | according to whether mi is positive or
negative for 1 ≤ i ≤ r.
If now ϕ in A is expressed as in Lemma 11.7.3 possibly with mi = 0 for some
i and αi = αj for some i = j, then the function ϕ# on Z defined by the same
expression for all n in Z depends only on ϕ. In order to prove this fact, we have
only to show that ϕ = 0 implies ϕ# = 0. Let {β} denote the set of distinct
α1 , . . . , αr and for each β define mβ as the sum of all mi for αi = β. Then ϕ = 0
implies mβ = 0 for all β by Lemma 11.7.3, hence

ϕ# (n) = mi αin = mβ β n = 0
1≤i≤r β
for all n in Z, and hence ϕ# = 0. As its immediate consequence, if we denote by

A# the set of all such ϕ# , then A# forms a ring and the correspondence ϕ → ϕ#
gives an isomorphism from A to A# .
In the following we shall restrict n to Z\{0} sometimes without saying so. We

take a complex variable z and introduce the rings B, B # of functions respectively
on C× × (N\{0}), C× × (Z\{0}) consisting of
#
ϕi (n)z in , ϕi (n)z in ,
i i
in which ϕi is in A and ϕ#
i is in A# for all i in N. We observe that B, B # are
commutative

# rings with 1 and the subsets S, S # of B, B # defined by the condition
that ϕi (n)z in = 0 for every n in Z\{0} and for a variable z in C× are both
multiplicative and free from zero divisors. Therefore, C = S −1 B, C # = (S # )−1 B #
are defined, and C consists of elements of the form

Φ(z, n) = ϕi (n)z in / ψi (n)z in ,
i i
in which the denominator is in S. We shall show that the element

# #
Φ# (z, n) = ϕi (n)z in / ψi (n)z in
i i
of C # depends only on Φ(z, n). In fact if Φ(z, n) is expressed similarly as

Φ(z, n) = ϕi (n)z in / ψi (n)z in ,
i i

then we get (ϕi ψj − ψi ϕj ) = 0 where the summation is in i, j in N satisfying

# #
i + j = k for every k in N. This implies (ϕ# #
i (ψj ) − ψi (ϕj ) ) = 0 for the same
summation i + j = k for every k, hence
# #
Φ# (z, n) = (ϕi ) (n)z in / (ψi ) (n)z in .
i i
We now go back to Theorem 11.7.1 and replace K there by L = Ke . If we

denote the corresponding ZL (s) by Ze (s), then we get

Ze (s) = R(e) · card ĒI (Fqe ) RI (e),
I⊂E
in which
R(e) = (q e − 1)q des /(q e(ds+n) − 1),

RI (e) = (q e − 1)/(q e(NE s+nE ) − 1) − 1
E∈I
for all e in N\{0}. We recall that ĒI is either empty or a smooth projective Fq -
variety of dimension rI = n − card(I) − 1. Therefore, Theorem 11.4.1 is applicable
to ĒI . In that way, we get

card(ĒI (Fqe )) = (−1)i αIij
e
0≤i≤2rI 1≤j≤BIi
226 JUN-ICHI IGUSA
for some αIij in C× with the property

(∗) {q rI /αIij ; 1 ≤ j ≤ BIi } = {αI,2rI −i,j ; 1 ≤ j ≤ BI,2rI −i = BIi }
for 0 ≤ i ≤ 2rI . If in the above-explained observation by Oesterlé we use e as n
and q s as z, then the denominator

(q e(ds+n) − 1) (q e(NE s+nE ) − 1)
E∈E
of Ze (s) is in S, hence Ze (s) is in C. Therefore, Ze (s) extends to an element Z(s, e)

of C # where e is arbitrary in Z\{0}. Furthermore, R(e), RI (e) also extend to
elements of C # and, as such, they satisfy
R(−e) = q −(ds−(n−1))e R(e), RI (−e) = q −card(I)e RI (e).
By using these and (∗) above, we can easily verify that
Z(s, −e) = q −des Z(s, e).
In this way, we get the following theorem of Denef and Meuser:
Theorem 11.7.2 In the same situation as in Theorem 11.7.1, if for every L = Ke ,
where e in in N\{0}, we put

Ze (s) = |f (x)|sL dx,
n
OL
then e → Ze (s) extends to a function Z(s, e) on Z\{0} satisfying the functional

equation Z(s, −e) = q −des Z(s, e).
Corollary 11.7.1 If there exists an element Z(u, v) of C(u, v), where u, v are
variables, satisfying Z(q −e , q −es ) = Ze (s) for all e > 0 in e0 Z for some integer e0 >
0, then the obviously unique Z(u, v) satisfies the functional equation Z(u−1 , v −1 ) =
v d Z(u, v).
Proof. In the notation of Theorem 11.7.2, we have
Z(q e , q es ) = Z(s, −e) = q −des Z(s, e) = (q −es )d Z(q −e , q −es )
for all e > 0 in e0 Z and for a variable s. This implies the functional equation in the
corollary.
The above corollary completely settles the conjecture which we have recalled in
the beginning of this section. We might mention that a certain p-adic zeta function
associated with an algebraic group satisfies a functional equation of the same kind;
cf. [29], pp. 708-709. It would be interesting to examine whether or not there exists
a common ground for all such functional equations.
That concludes this introductory book to the theory of local zeta functions. As
the last word, we would like to recommend to the readers to proceed to Denef’s
Bourbaki report [11] and our own expository paper [31] which we have mentioned in
the Introduction. The readers will find in these references not only further important
results but also problems whose solutions will undoubtedly enrich the theory.
Bibliography
[1] M. F. Atiyah, Resolution of singularities and division of distributions,
Comm. Pure Appl. Math., 23 (1970), 145-150.
[2] I. N. Bernstein and S. I. Gel’fand, Meromorphic property of the functions

P λ , Functional Anal. Appl., 3 (1969), 68-69.
[3] I. N. Bernstein, The analytic continuation of generalized functions with

respect to a parameter, Functional Anal. Appl., 6 (1972), 273-285.
[4] A. Borel and J.-P. Serre, Le théorème de Riemann-Roch, Bull. Soc. math.
France, 86 (1958), 97-136.
[5] S. I. Borewicz and I. R. Šafarevič, Zahlentheorie, Birkhäuser (1966).
[6] F. Bruhat, Distributions sur un groupe localement compact et applica-

tions à l’étude des représentations des groupes p-adiques, Bull. Soc. math.
France, 89 (1961), 43-75.
[7] C. Chevalley, The algebraic theory of spinors, Columbia Univ. Press (1954).
[8] P. Deligne, La conjecture de Weil, Inst. Hautes Études Sci. Publ. Math.,
43 (1974), 273-307.
[9] J. Denef, On the degree of Igusa’s local zeta function, Amer. J. Math., 109
(1987), 991-1008.
[10] J. Denef and D. Meuser, A functional equation of Igusa’s local zeta func-
tions, Amer. J. Math., 113 (1991), 1135-1152.
[11] J. Denef, Report on Igusa’s local zeta functions, Sém. Bourbaki 741 (1991),
1-25; Asterisque No. 201-203, 359-386.
[12] L. E. Dickson, Linear groups with an exposition of the Galois field theory,
B. G. Teubner (1901).
[13] C. Ehresmann, Sur la topologie de certains espaces homogènes, Ann. Math.,

35 (1934), 396-443.
227
228 JUN-ICHI IGUSA
[14] H. Freudenthal, Beziehungen der E7 und E8 zur Oktavenebene. I, Prok.

Konkl. Ned. Akad. Wet., A 57 (1954), 218-230.
[15] C. F. Gauss, Hundert Theoreme über die neuen Transscendenten, Werke

III (1876), 461-469.
[16] I. M. Gel’fand and G. E. Shilov, Generalized functions. I, Academic Press

(1964).
[17] G. B. Gurevich, Foundations of the theory of algebraic invariants, Noord-

hoff (1964).
[18] A. Gyoja, Functional equation for Igusa local zeta functions, JAMI lecture
(1993).
[19] D. Hilbert, Über die vollen Invariantensysteme, Math. Ann. 42 (1893), 313-
373.
[20] H. Hironaka, Resolution of singularities of an algebraic variety over a field

of characteristic zero. I-II, Ann. Math., 79 (1964), 109-326.
[21] J. Igusa, Analytic groups over complete fields, Proc. Nat. Acad. Sci., 42
(1956), 540-541.
[22] J. Igusa, A classification of spinors up to dimension twelve, Amer. J. Math.,

92 (1970), 997-1028.
[23] J. Igusa, Complex powers and asymptotic expansions. I, Crelles J. Math.,

268/269 (1974), 110-130; II, ibid., 278/279 (1975), 307-321.
[24] J. Igusa, On the first terms of certain asymptotic expansions, Complex

Analysis and Algebraic Geometry, Iwanami Shoten (1977), 357-368.
[25] J. Igusa, Some results on p-adic complex powers, Amer. J. Math., 106
(1984), 1013-1032.
[26] J. Igusa, Complex powers of irreducible algebroid curves, Geometry Today,

Birkhäuser (1985) , 207-230.
[27] J. Igusa, On functional equations of complex powers, Invent. math., 85

(1986), 1-29.
[28] J. Igusa, On the arithmetic of a singular invariant, Amer. J. Math., 110

(1988), 197-233.
[29] J. Igusa, Universal p-adic zeta functions and their functional equations,
Amer. J. Math., 111 (1989), 671-716.
[30] J. Igusa, A stationary phase formula for p-adic integrals and its applica-
tions, Algebraic Geometry and its Applications, Springer-Verlag (1994),
175-194.
[31] J. Igusa, On local zeta functions, Amer. Math. Soc. Transl. (2) 172, (1996),
1-20.
[32] N. Jacobson, Structure and representations of Jordan algebras, Amer.

Math. Soc. Publ., 39 (1968).
[33] M. Kashiwara, B-functions and holonomic systems (Rationality of b-

functions), Invent. math., 38 (1976), 33-53.
[34] T. Kimura, The b-functions and holonomy diagrams of irreducible regular

prehomogeneous vector spaces, Nogoya Math. J., 85 (1982), 1-80.
[35] T. Kimura, F. Sato, and X.-W. Zhu, On the poles of p-adic complex powers
and b-functions of prehomogeneous vector spaces, Amer. J. Math., 112
(1990), 423-437.
[36] W. Krull, Dimensionstheorie in Stellenringen, Crelles J. Math., 179 (1938),

204-226.
[37] F. Loeser, Fonctions d’Igusa p-adiques et polynômes de Bernstein, Amer.

J. Math., 110 (1988), 1-22.
[38] F. Loeser, Fonctions d’Igusa p-adiques, polynômes de Bernstein, et

polyèdres de Newton, Crelles J. Math., 412 (1990), 75-96.
[39] J. G. M. Mars, Les nombres de Tamagawa de certains groupes exception-

nels, Bull. Soc. math. France 94 (1966), 97-140.
[40] H. Mellin, Abriss einer einheitlichen Theorie der Gamma- und der hyper-
geometrischen Funktionen, Math. Ann., 68 (1910), 305-337.
[41] D. Meuser, On the poles of a local zeta function for curves, Invent. math.,
73 (1983), 445-465.
[42] D. Meuser, On the degree of a local zeta function, Comp. Math., 62 (1987),
17-29.
[43] G. D. Mostow, Self-adjoint groups, Ann. Math., 62 (1955), 44-55.
[44] M. Nagata, Local rings, Interscience Tracts in Pure and Appl. Math., 13
(1962).
[45] L. S. Pontrjagin, Topologische Gruppen. I, B. G. Teubner (1957).
[46] M. M. Robinson, The Igusa local zeta function associated with the singular
cases of the determinant and the Pfaffian, J. Number Theory, 57 (1996),
385-408.
[47] P. Samuel, La notion de multiplicité en algèbre et en géométrie algébrique,

Thèses, Gauthier-Villars (1951).
230 JUN-ICHI IGUSA
[48] M. Sato, Theory of prehomogeneous vector spaces (notes by T. Shintani),

Sugaku-no-ayumi 15-1 (1970), 85-156 (in Japanese).
[49] M. Sato and T. Kimura, A classification of irreducible prehomogeneous

vector spaces and their relative invariants, Nagoya Math. J., 65 (1977),
1-155.
[50] M. Sato, M. Kashiwara, T. Kimura, and T. Oshima, Micro-local analysis

of prehomogeneous vector spaces, Invent. math., 62 (1980), 117-179.
[51] L. Schwartz, Théorie des distributions. I, II, Hermann & Cie (1950,1951).
[52] J.-P. Serre, Classification des variétés analytiques p-adiques compactes,

Topology 3 (1965), 409-412.
[53] H. Späth, Der Weierstrasssche Vorbereitungssatz, Crelles J. Math. 161

(1929), 95-100.
[54] L. Strauss, Poles of a two variable p-adic complex power, Trans. Amer.
Math. Soc., 278 (1983), 481-493.
[55] J. Tate, Fourier analysis in number fields and Hecke’s zeta-functions, The-
sis, Princeton (1950).
[56] A. Weil, L’Intégration dans les groupes topologiques et ses applications,

Hermann & Cie (1940).
[57] A. Weil, Numbers of solutions of equations in finite fields, Bull. Amer.

Math. Soc., VI, 55 (1949), 497-508.
[58] A, Weil, Sur la formule de Siegel dans la théorie des groupes classiques,
Acta Math., 113 (1965), 1-87.
[59] H. Weyl, Gruppentheorie und Quantenmechanik, S. Hirzel (1928).
[60] H. Weyl, The classical groups, Princeton Math. Series 1 (1946).
[61] E. Witt, Theorie der quadratischen Formen in beliebigen Körpern, Crelles

J. Math., 176 (1937), 31-44.
[62] O. Zariski, Foundations of a general theory of birational correspondences,

Trans. Amer. Math. Soc., 53 (1943), 490-542.
[63] O. Zariski, The concept of a simple point of an abstract algebraic variety,

Trans. Amer. Math. Soc., 62 (1947), 1-52.
Index
Absolute value | · |K , 15 Fourier transformation, 78, 120
non-archimedean -, 21 inversion formula, 79, 120
Algebra, 31 Freudenthal quartic, 156
composition -, 149
exterior -, 31 Gauss’ and related identities (G0)-(G3),
filtered -, 51 160-162
graded -, 11 Generalized Gaussian sum, 126
Jordan -, 151 Grassmann variety, 212
tensor -, 31
Haar measure µG , µ, dx, 102
Asymptotic formula for |Γ(s)| as
module ∆G , 103
|Im(s)| → ∞, 87
Heisenberg commutation relation, 45
Baire’s theorem, 106 Hensel’s lemma, 218
Basic relative invariant, 84 Hilbert’s basis theorem, 9
Bernstein’s polynomial bf (s), 47 characteristic function χ(M, t), 11
Nullstellensatz, 10
Characteristic pairs, 40 Hironaka’s desingularization, 206
Complete field, 16 Hironaka’s desingularization theorem,
Complex power ω(f ), 73, 123 39, 207-208
Condition (A), 125, 133
Critical point, set, value, 32 Implicit function theorem (by calcu-
finiteness of the set of critical val- lus of limits), 18, 23
ues, 34
Key lemma, 173
Cubic polynomial Cm,n (a, t), 183, 196
Krull’s theorem, 8
Denef’s formulas for Za (s), Z(s), 216,
Linear groups GLn , SLn , 1, 145
217, 224
formulas for card(GLn (Fq )),
Denef-Meuser’s theorem, 226
card(SLn (Fq )), 145
Differential form (K-analytic), 31
Localization S −1 (S multiplicative), 6
Dimension, local ring, 200
Local ring, 7
manifold, 29
regular, 200
variety, 205
Local singular series, 129
Discriminant d(Q), 142
Local zeta function ZΦ (ω), 71, 73, 123
D-module, 46
Z(ω), Z(s), 73, 124
Bernstein’s finite generation the-
explicit form
orem, 55
K = C, f (x) basic relative invari-
Dominant series, 16
ant, 91
Elementary solution, 81 K = R, f (x) b.r.i. (square free
Exceptional divisor, 36 terms), 93
K = p-adic field
Filtration, 51 x21 + x32 , 171
standard, 54 x21 + x32 + x53 , 172
type (d, e), 53 Q(x) (Q mod π reduced), 169
231
232 JUN-ICHI IGUSA
det(x) (x in Mn ), 163 Primary decomposition theorem, 6

det(x) (x in Symn ), 177
Pf(x), 164 Quadratic form Q, 137
Pf(y) − t z1 yz2 , 166 anisotropic -, 137
Pf(t xJm x), 165 nondegenerate -, 137
Freudenthal quartic (p = 2), reduced -, 137
182-183 formula for card(Q−1 (i)(Fq )) (Q
Gramian det(t xhx) (p = 2), 188, reduced), 143
195-196 Quadratic transformation, 36
Rationality of zeros of b(s), 92

Manifold (K-analytic), 29 Relation of FΦ , FΦ∗ , 128
Measure µα (α differential form), 112- Relation of FΦ , ZΦ , 130
112 Root of an ideal, 5
µα/β , 115
Method of analytic continuation Sato’s b-function b(s), 87
(Gel’fand & Shilov), 66 Schwartz space S(X), 62
Modules, filtered -, 51 its dense subspace G(X), 75
graded -, 11 Serre’s structure theorem, 113
Monoidal transformation (simple cen- Simple point, smoothness, 205
ter), 36 Space of derivations DerF (R, L), 32
Space of (tempered) distributions S(X) ,
Nakayama’s lemma, 8 62
Nerve complex N , 38 its completeness, 64
Noetherian ring, 5 Space of eigendistributions EX (ρ), 108
Normal crossings, 38 continuity of ρ, 101
Numerical data (NE , nE ), 39 Spaces D(X), D(X) (X totally dis-
connected), 98-99
Orthogonal group O(Q) (Q reduced), Stationary phase formula (SPF), 168
139 Symplectic group Sp2n , 148
formula for card(O(Q)(Fq )), 146- formula for card(Sp2n (Fq )), 148
p-adic field, 109 Totally disconnected space, 97

Partition of unity, 74 - group, 98
Poincaré series, 124
Unramified extension Ke (K a p-adic
its rationality, 124
field), 217
Poles of ω(f )(Φ) = ZΦ (ω), 71, 73, 76,
122 Variety, affine -, 204
f relative invariant, 135 projective - & quasi-projective -,
Pontrjagin’s theorem, 106 203
Power series
convergent -, 16 Weierstrass preparation theorem, 24
formal -, 16 Weierstrass product of 1/Γ(s), 88
special restricted - (SRP), 22 Weil’s functions FΦ , FΦ∗ , 125, 127
Prehomogeneous vector space, 83 Weil’s zeta function, 213
irreducible regular -, 95 Witt’s decomposition (Q reduced), 140
regular -, 83 Witt’s theorem, 139
This book is an introductory presentation to the theory of local zeta func-
tions. Viewed as distributions, and mostly in the archimedean case, local
zeta functions are also called complex powers. The volume contains
major results on analytic and algebraic properties of complex powers
by Atiyah, Bernstein, I. M. Gelfand, S. I. Gelfand, and Sato. Chapters
devoted to F-adic local zeta functions present Serre’s structure theorem,
American a rationality theorem, and many examples found by the author. The
presentation concludes with theorems by Denef and Meuser.
Mathematical
Society
www.ams.org
International
Press
www.intlpress.com
AMSIP/14.S

An Introduction To The Theory of Local Zeta Functions: Studies in Advanced Mathematics

Uploaded by

Copyright:

Available Formats

An Introduction To The Theory of Local Zeta Functions: Studies in Advanced Mathematics

Uploaded by

Document Information

Original Title

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

An Introduction To The Theory of Local Zeta Functions: Studies in Advanced Mathematics

Uploaded by

Copyright:

Available Formats

AMS/IP

American Mathematical Society • International Press

American Mathematical Society • International Press

Library of Congress Cataloging-in-Publication Data

2 Implicit function theorems and K-analytic manifolds 15

3 Hironaka’s desingularization theorem 35

5 Archimedean local zeta functions 59

6 Prehomogeneous vector spaces 83

7 Totally disconnected spaces and p-adic manifolds 97

8 Local zeta functions (p-adic case) 117

9 Some homogeneous polynomials 137

10 Computation of Z(s) 163

11 Theorems of Denef and Meuser 199

of the group Ω(K × ) of all continuous homomorphisms ω from K × to C× . By

in which i∗ is in kA , f (x) is a nondegenerate quadratic form on kn , Φ is in S(kA

a diﬀerential operator P with coeﬃcients in k◦ [s, x1 , . . . , xn ], where s is another

f (∂/∂x)f (x)s+1 = b(s)f (x)s

h(x) = c0 (x ) + c1 (x )xn + . . .

a has a neighborhood U◦ with compact closure. By replacing U by its intersection

Since D is assumed to be piecewise smooth, it has a “surface element” dσ. If

The third one is called Lebesgue’s theorem, and it is as follows:

for a convergent sequence of Lebesgue integrable functions. In our later applications

1.2 Noetherian rings

Theorem 1.2.1 If a is an ideal of a noetherian ring A and

is a minimal representation of a, then the set {r(q1 ), . . . , r(qt )} is uniquely deter-

hence qt is redundant. This is a contradiction. Therefore we may assume that

hence qr becomes redundant in the minimal representation of a. This is a contra-

denote a minimal representation of mn. Then mn is contained in qi , hence n is

1.3 Hilbert’s theorems

Lemma 1.3.1 . Let A denote a noetherian ring, M a ﬁnitely generated A-module,

Proof. If we express M as M = Ax1 + . . . + Axn , then the Lemma holds trivially

Lemma 1.3.2 If A is a noetherian ring and x is a variable, then the polynomial

If K is an arbitrary ﬁeld, then 0 and K are the only ideals of K, hence it is a

Theorem 1.3.1 Let A denote the polynomial ring K[x1 , . . . , xn ], where K is a

which contain a. Hilbert’s Nullstellensatz is similar to the above statement. We

then b0 , b1 , . . . , bd−1 are in Z by induction. On the other hand

hence a1 = b0 , a2 = b1 , . . . , ad = bd−1 are in Z. Then

with χ(r) in Z for all large r shows that a0 is also in Z.

Theorem 1.3.3 If A is the polynomial ring K[x1 , . . . , xn ] and M is a ﬁnitely gen-

we always have an = 1. In fact a = 0 implies ar◦ = 0 for some r◦ . If f = 0 is in ar◦ ,

dimK (Ar ) = dimK (Ar f ) ≤ dimK (ar+r◦ ) ≤ dimK (Ar+r◦ )

for every r. This implies the above assertion.

Implicit function theorems

the corresponding factor ring by K ∗ . If a∗ is an element of K ∗ represented by a

◦ i convergent power series forms a subring of K[[x]]. If for ci x in K[[x]] and

and call ci x a dominant series for c i xi .

We observe that if F (y) is in A[[y1 , . . . , ym ]] and fi (x) is in A[[x1 , . . . , xn ]] sat-

with Hk (x, y1 , . . . , yk ) in K[[x, y1 , . . . , yk ]] for 1 ≤ k ≤ m. If further F (x, y) is a

with Hk (x, y, z) in K[[x, y, z]] (resp. K x, y, z ) free from zk+1 , · · · , zm . Since

for some k ≥ 0. If a is in K × and |a|K < r, then

f (a) i−k M |a|K

We know that Im(K × ) contains a null sequence in R. Therefore we can choose a

If |h|K tends to 0, the RHS, hence also the LHS, tends to 0.

∂(F1 , . . . , Fm )/∂(y1 , . . . , ym )(0, 0) = 0,

in which ∂(F1 , . . . , Fm )/∂(y1 , . . . , ym ) is the jacobian, i.e., the determinant of the

Proof of (i). If we write

with unknown coeﬃcients dij in K. Since F (x, f (x)) = 0 can be rewritten as

in which kα = |k| and

in particular |j| + |k| ≤ p, for p = 1, 2, 3, . . . . Since ci0k = 0 for |k| = 1, if p = 1,

h(x) = c0 (x ) + c1 (x )xn + . . .

with Hk (x, y, z) in K[[x, y, z]] (resp. Kx, y, z ) free from zk+1 , · · · , zm . Since

in which Pij is a polynomial in ci j k for |j | + |k | ≤ p with coeﬃcients in N.

with Hij (x, y, y ) in Kx, y, y satisfying Hij (0, 0, 0) = 0 for all i, j. Furthermore,

Finally, we shall show that fi◦ (x) is in Rx . If we put X = x1 + . . . + xn and

This equation in Y has a unique solution in RX which becomes 0 for X = 0.

in the proof of Theorem 2.1.1 (ii) is an SRP in the entries of x, y, y , and it

Corollary 2.3.1 The ring Kx = Kx1 , . . . , xn of convergent power series is

Proof. We put An = Kx1 , . . . , xn and denote the leading degree of any f (x) in

for all gin mb . We observe

on U . We shall sometimes write f ∗ instead of (δf )∗ . In the special case where

for every a in R and b in S. The correspondence ∂ → ∂ # gives an L-linear bijection

in which (∂f1 /∂t)(x , y) = 0. Since F (x , y) = F (x), we also know that

we shall identify (x, t) in fU−1 (U ) with the (x, t) in fU−1

(xi , xp+1 , . . . , xn , u1 , . . . , up−1 ) → (xi , xp+1 , . . . , xn , u1 , . . . , up−1 ).