Algebraic Approach To Differential Equations

www.pdfgrip.
com
Algebraic Approach
to
Differential
Equations
7290 tp.indd 1 6/9/10 8:42 AM

www.pdfgrip.com
This page intentionally left blank

www.pdfgrip.com
Algebraic Approach
to
Differential
Equations
Bibliotheca Alexandrina, Alexandria, Egypt 12 – 24 November 2007
Edited by
Lê Dũng Tráng
ICTP, Trieste, Italy
World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I • CHENNAI
7290 tp.indd 2 6/9/10 8:42 AM

www.pdfgrip.com
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
ALGEBRAIC APPROACH TO DIFFERENTIAL EQUATIONS
Copyright © 2010 by The Abdus Salam International Centre for Theoretical Physics
ISBN-13 978-981-4273-23-7
ISBN-10 981-4273-23-6
Printed in Singapore.
Julia - Algebraic Approach to Diff Eqns.pmd 1 3/24/2010, 11:03 AM

September 21, 2010 10:16 WSPC - Proceedings Trim Size: 9in x 6in 00a˙preface
www.pdfgrip.com
PREFACE
In october 2007, the “Abdus Salam” International Centre for Theoretical Physics
(ICTP) organized a school in mathematics at the Biblioteca Alexandrina in
Alexandria, Egypt. From the 3rd century B.C. until the 4th century A.C. Alexan-
dria was a centre for mathematics. Euclid, Diophante, Eratostene, Ptolemy, Hy-
patia were among those who made the fame of Alexandria and its antique library.
The choice of the Biblioteca Alexandria was symbolic. With the reconstruction
of the library it was natural that one also resumes the universal intellectual ex-
change of the antique library. The will of the director of the Biblioteca, Ismael
Seralgedin made that school possible.
The topic of the school was “Algebraic approach of differential equations”.
This special topic which is at the convergence of Algebra, Geometry and Analysis
was chosen to gather mathematicians of different disciplines in Egypt. This topic
arises from the pioneer work of E. Kolchin, L. Gårding, B. Malgrange and was
formalized by the school of M. Sato in Japan. The techniques used are among the
most recent and modern techniques of mathematics. In these lectures we give an
elementary presentation of the subject. Applications are given and new areas of
research are also hinted. This book allows to understand developments of this. We
hope that this book which gathers most of the lectures given in Alexandria will
interest specialists and show how linear differential systems are studied nowadays.
I especially thank the secretaries Alessandra Bergamo and Mabilo Koutou of
the mathematics section of ICTP and Anna Triolo of the publications section of
ICTP for all the help they gave for the publication of this book.
Lê Dũng Tráng
Erratum
The school on “Algebraic Approach to Differential Equations” was organized

by Lê Dũng Tráng from the ICTP and Egyptian colleagues, Professor Darwish
Mohamed Abdalla from Alexandria University, Professor Fahmy Mohamed
from Al-Azhar University, Professor Yousif Mohamed from the American Uni-
versity in Cairo. Professor Ismail Idris from Ain Shams replaced Professor
Fahmy who had to leave during the conference. Special thanks are going to
Professor Mohamed Darwish for his dedication in organizing the school.
February 4, 2010 17:3 WSPC - Proceedings Trim Size: 9in x 6in 00b˙acknowledgments
www.pdfgrip.com
vii
ACKNOWLEDGMENTS
The school was made possible with the help of Mr Mohamed El Faham,
Deputy Director of the Bibliotheca Alexandrina, Ms Sahar Aly in charge
of the international meetings, Ms Mariam Moussa, Ms Marva Elwakie, Ms
Yasmin Maamoun, Ms Omneya Kamel, Ms Asmaa Soliman and Ms Samar
Seoud, all from Bibliotheca Alexandrina.
From the ICTP side, Ms Koutou Mabilo and Ms Alessandra Bergamo
was in charge of the organisation of the school and Ms Anna Triolo was in
charge of the publication of the proceedings.
Lê Dung Tráng

Head of the mathematics section
ICTP, Trieste, Italy
March 31, 2010 14:5 WSPC - Proceedings Trim Size: 9in x 6in 00c˙contents
www.pdfgrip.com
ix
CONTENTS
Preface v
Acknowledgments vii
D-Modules in Dimension 1 1
L. Narváez Macarro
Modules Over the Weyl Algebra 52

F. J. Castro Jiménez
Geometry of Characteristic Varieties 119

D. T. Lê and B. Teissier
Singular Integrals and the Stationary Phase Methods 136

E. Delabaere
Hypergeometric Functions and Hyperplane Arrangements 210

M. Jambu
Bernstein-Sato Polynomials and Functional Equations 225

M. Granger
Differential Algebraic Groups 292

B. Malgrange
March 31, 2010 14:8 WSPC - Proceedings Trim Size: 9in x 6in 01˙macarro
www.pdfgrip.com
D-MODULES IN DIMENSION 1
´
L. NARVAEZ MACARRO∗
Departamento de Álgebra & Instituto de Matemáticas (IMUS)
Universidad de Sevilla, P.O. Box 1160, 41080 Sevilla, Spain
∗ E-mail: [email protected]
Introduction
These notes are issued from a course taught in the I.C.T.P. School on
Algebraic Approach to Differential Equations, held at Alexandria (Egypt)
from November 12 through November 24, 2007.
These notes are intended to guide the reader from the classical theory
of linear differential equations in one complex variable to the theory of D-
modules. In the first four sections we try to motivate the use of sheaves, in
very concrete terms, to state Cauchy theorem and to express the phenomena
of analytic continuation of solutions. We also study multivalued solutions
around singular points. In sections 5 and 6 we recall the classical result of
Fuchs, the index theorem of Komatsu-Malgrange and Malgrange’s homo-
logical characterization of regularity, which is a key point in understanding
regularity in higher dimension. Section 7 is extracted from the very nice pa-
per2 of J. Briançon and Ph. Maisonobe. It contains the division tools on the
ring of (germs of) linear differential operators in one variable. They allow
us to prove “almost everything” on (complex analytic) D-module theory in
dimension 1 from the classical results. Section 8 tries to motivate the point
of view of higher solutions, a landmark in D-module theory. Sections 9 and
10 deal with holonomic D-modules and the general notion of regularity.
Both sections are technically based on the division tools and so they are
very specific for the one dimensional case, but they give a good flavor of
the general theory. Section 11 is written in collaboration with F. Gudiel
and it contains the local version of the Riemann-Hilbert correspondence in
∗ Partially supported by MTM2007–66929 and FEDER.

www.pdfgrip.com
dimension 1 stated in the paper13 with some complements. In section 12

we sketch the theory of D-modules on a Riemann surface.
We would like to thank the organizers of the I.C.T.P. school, specially

M. Darwish who took care of all practical (and very important) details,
and Lê Dũng Tráng who conceived the school and took the heavy task of
editing the lecture notes.
1. Cauchy Theorem
Let U ⊂ C be an open set. A complex linear differential equation on U is
given by
dn y dy
an + · · · + a1 + a0 y = g, (1)
dz n dz
where the ai and g are holomorphic functions on U and y is an unknown
holomorphic function on U , which in case it exists is called a solution (on
U ) of the equation (1). If the function an does not vanishes identically, we
say that equation (1) has order n.
When g = 0 in (1), we call it an homogeneous complex linear differential
equation. In such a case, the solutions form a complex vector space, i.e.
-) the product of any constant and any solution is again a solution.
-) The sum of two solutions is again a solution.
Remark 1.1. A very basic (and obvious) remark is that a complex linear
differential equation on U as (1) determines, by restriction, a complex linear
differential equation on any open subset V ⊂ U and we may be interested
in searching its solutions, not only on the whole U , but on any open subset
V ⊂ U.
If an (x) 6= 0 for all x ∈ U , then equation (1) is equivalent (in the sense
that they have the same solutions) to
dn y dy dy
n
+ a0n−1 + · · · + a01 + a00 y = g 0 , (2)
dz dz dz
where a0i = aani and g 0 = agn .
Equation (2) is still equivalent to a linear system of order 1
   
Y1 B1
dY  ..   .. 
= AY + B, Y =  .  , B =  .  (3)
dz
Yn Bn
www.pdfgrip.com
with B1 = · · · = Bn−1 = 0, Bn = b and

 
0 1 0 ··· 0
 0 0 1 ··· 0 
 
 
A =  ... ..
.
..
.
..
.
..
. .
 
 0 0 0 ··· 1 
−a00 −a01 −a02 · · · −a0n−1
Correspondences
 
Y1 = y  
 Y2 = y (1)  Y1
   .. 
y 7→  .. ,  .  7→ y = Y1
 . 
(n−1) Yn
Yn = y
establish a bijection between the solutions of (2) and the solutions of (3).
When b = 0 this bijection is an isomorphism of complex vector spaces.
The basic existence theorem for solutions of a linear system of type (3)
is the following result, which ca be found on almost any book of differential
equations (see for instance the book5 no 384).
Theorem 1.1. Let U ⊂ C be an open disc centered at the origin, A a

(n × n) matrix of holomorphic functions on U and B a n-column vector of
holomorphic functions on U . Let us call S the set of solutions of the system
dY
dz = AY + B. Then, the map
Y ∈ S 7→ Y (0) ∈ Cn
is bijective. Moreover, when B = 0 the application above is an isomorphism

of complex vector spaces.
Corollary 1.1. Let U ⊂ C be an open disc centered at the origin and

let a0 , . . . , an holomorphic functions on U with an (z) 6= 0 for all z ∈ U .
Then, for any holomorphic function g on U and any “initial conditions”
v0 , . . . , vn−1 ∈ C there is a unique holomorphic function y on U , which is
a solution of the linear differential equation
dn y dy
an + · · · + a1 + a0 y = g,
dz n dz
and such that
y(0) = v0 , y (1) (0) = v1 , . . . , y (n−1) (0) = vn−1 .

www.pdfgrip.com
2. Sheaves of Holomorphic Functions

Theorem 1.1 can be rephrased in terms of sheaf theory and local systems,
which is in principle nothing but an enlargement of our mathematical lan-
guage. However, this enlargement becomes fundamental in order to under-
stand higher dimensional phenomena and the global behaviour of solutions
of differential equations. Let us start by introducing some provisionala def-
initions.
For each open set V ⊂ C let us denote by O(V ) the complex vector
space of holomorphic functions defined on V .
Definition 2.1. The sheaf of holomorphic functions on an open set U ⊂ C

is the data consisting of all the complex vector spaces O(V ), when V runs
into the set of open subsets of U . It will be denoted by OU , and for each
open set V ⊂ U we will write OU (V ) := O(V ). The following properties
clearly hold:
(a) If V 0 ⊂ V ⊂ U are open sets and f ∈ OU (V ), then f |V 0 ∈ OU (V 0 ).
(b) If V ⊂ U is an open set, {Vi }i∈I is an open covering of V and f : V → C
is a function, we have: f ∈ OU (V ) ⇔ f |Vi ∈ OU (Vi ) for all i ∈ I.
Property (b) above means that for a function, being holomorphic is a

local property.
Definition 2.2. A subsheafb of OU is the data F consisting of a vector

subspace F(V ) ⊂ OU (V ) for each open set V ⊂ U satisfying the following
properties:
(a) If V 0 ⊂ V ⊂ U are open sets and f ∈ F(V ), then f |V 0 ∈ F(V 0 ).
(b) If V ⊂ U is an open set, {Vi }i∈I is an open covering of V and f ∈
OU (V ), we have f ∈ F(V ) ⇔ f |Vi ∈ F(Vi ) for all i ∈ I.
If the data F satisfies property (a) and not necessarily property (b), then
we say that it is a subpresheaf of OU . If F is a subpresheaf of OU , we will
simply write F ⊂ OU .
If F, F0 are subpresheaves of OU , we say that F ⊂ F 0 if F(V ) ⊂ F0 (V )

for any open set V ⊂ U .
Let us note that if F ⊂ OU is a subsheaf and U 0 ⊂ U is an open subset,
then the data F|U 0 defined by F|U 0 (V ) = F(V ) for any open set V ⊂ U 0 is
a Later, we will need the general notion of sheaf, but in this section we only study the
sheaf of holomorphic functions and its subsheaves.

b Here, we only consider subsheaves of complex vector spaces.
www.pdfgrip.com
a subsheaf of OU 0 , that we call the restriction of F to U 0 . Let us also note

that OU |U 0 = OU 0 .
Exercise 2.1. (1) Let F be the data defined by
F(V ) = {f : V → C | f is a constant function} ⊂ OU (V ),
for each open set V ⊂ U . Prove that F is a subpresheaf of OU which is

not a subsheaf. (Hint: what happens with property (b) every time V is not
connected?)
(2) Prove that the data CU defined by
CU (V ) = {f : V → C | f is a locally constant function} ⊂ OU (V ),
for each open set V ⊂ U , is a subsheaf of OU .
Exercise 2.2. Let U ⊂ C be an open set, Σ ⊂ U a closed discrete set and

let us denote by j : U \ Σ ,→ U the inclusion.
(1) Let F be the data defined by
F(V ) = {f ∈ OU (V ) | f = 0 on a neighborhood of any point p ∈ Σ ∩ V }.
Prove that F is a subsheaf of OU , which will be denoted by j! OU \Σ .

(2) Let F be the data defined by
F(V ) = {f ∈ CU (V )| f = 0 on a neighborhood of any point p ∈ Σ ∩ V }.
Prove that F is a subsheaf of OU , which will be denoted by j! CU \Σ .
Exercise 2.3. Let F ⊂ OU be a subpresheaf. Prove that:
(1) There is a unique subsheaf F + ⊂ OU such that:

(a) F ⊂ F+ .
(b) If F0 ⊂ OU is a subsheaf with F ⊂ F 0 , then F+ ⊂ F0 .
The sheaf F+ is called the associated sheaf to F.
(2) Prove that F is a subsheaf of OU if and only if F = F+ .
(3) Prove that (F|U 0 )+ = F+ |U 0 for any open subset U 0 ⊂ U .
Definition 2.3. An endomorphism of OU , L : OU → OU , is the data

consisting of a family of C-linear maps L(V ) : OU (V ) → OU (V ) such
that for any open subsets V 0 ⊂ V ⊂ U and any f ∈ OU (V ) we have
L(V )(f )|V 0 = L(V 0 )(f |V 0 ).
www.pdfgrip.com
Let us denote by End(OU ) the set of endomorphisms L : OU → OU .

The definition of “composition” and “addition” inside End(OU ) is clear and
they define a non-commutative ring structure on End(OU ). Composition in
End(OU ) will be denoted by ◦ or simply by juxtaposition, and addition
by the usual “+”. Moreover, we have an obvious ring homomorphism C →
End(OU ), and so End(OU ) is a non-commutative C-algebra.
If L : OU → OU is an endomorphism and U 0 ⊂ U is an open set, then
we define the restriction of L to U 0 as the endomorphism L|U 0 : OU 0 → OU 0
given by L|U 0 (V ) = L(V ) : OU 0 (V ) = OU (V ) → OU 0 (V ) = OU (V ) for any
open set V ⊂ U 0 . It is clear that the map
L ∈ End(OU ) 7→ L|U 0 ∈ End(OU 0 )
is a homomorphism of C-algebras.
Example 2.1. (a) The family of linear maps

df
f ∈ OU (V ) 7→ ∈ OU (V ), V ⊂U open subset,
dz
d
is an endomorphism of OU that will be denoted by dz : OU → O U .
(b) If h ∈ OU (U ), then the family of linear maps
f ∈ OU (V ) 7→ (h|V )f ∈ OU (V ), V ⊂U open subset,
is an endomorphism that will be denoted by h : OU → OU .
(c) Example (b) gives rise to a ring homomorphism OU (U ) → End(OU ),

which is injective.
Exercise 2.4. Let {Ui }i∈I be an open covering of U and Li ∈ End(OUi ) for
each i ∈ I, such that Li |Ui ∩Uj = Lj |Ui ∩Uj for all i, j ∈ I. Prove that there
is a unique L ∈ End(OU ) such that L|Ui = Li for all i ∈ I.
Remark 2.1. The above exercise indicates that, for a given open set
U ⊂ C, the family End(OV ), V ⊂ U open subset, satisfies the same for-
mal properties as subsheaves of OU (see definition 2.2). In fact, OU , sub-
sheaves of OU , and {End(OV ), V ⊂ U open subset} all are examples of
“abstract sheaves” (of complex vector spaces or C-algebras) (see for in-
stance the book9 ). The family {End(OV ), V ⊂ U open subset} is denoted
by End (OU ), and we write End (OU )(V ) = End(OV ) for any open subset
V ⊂ U.
www.pdfgrip.com
Exercise 2.5. Let L : OU → OU be an endomorphism and let us consider

the data ker L defined by (ker L)(V ) = ker L(V ) ⊂ OU (V ) for each open
set V ⊂ U . Prove that ker L is a subsheaf of OU , that will be called the
kernel of L.
d
Exercise 2.6. (1) Describe the kernel of the endomorphism dz : OC → O C .
d

(2) Prove that ker z dz + 1 : OC → OC = j! CC−{0} .
Exercise 2.7. (1) Let L : OU → OU be an endomorphism and let us con-
sider the data im0 L defined by (im0 L)(V ) = im L(V ) ⊂ OU (V ) for each
open set V ⊂ U . Prove that, in general, im0 L is not a subsheaf of OU .
(Hint: Consider L = dz d
: OC → OC . Is the function z −1 in (im0 L)(C∗ )?
Nevertheless, for each simply connected open set V ⊂ C∗ , the function z −1
belongs to (im0 L)(V ).)
(2) Let us consider the data im L defined by
(im L)(V ) = {g ∈ OU (V ) | ∀p ∈ V, ∃W ⊂ V open neighborhood of p,
∃f ∈ OU (W ) s.t. L(W )(f ) = g|W },
for each open set V ⊂ U . Prove that im L is a subsheaf of OU , that will be
called the image of L. (Note that im L = (im0 L)+ )
d
(3) Compute the image of the endomorphism dz : OC → O C .
Definition 2.4. A (holomorphic) linear differential operator of order ≤ n
on U is an endomorphism L : OU → OU such that there are ai ∈ OU (U ),
0 ≤ i ≤ n, such that for each open set V ⊂ U and each f ∈ OU (V ) we have
dn f df
L(V )(f ) = (an |V ) + · · · + (a1 |V ) + (a0 |V )f,
dz n dz
dn d
or equivalently, the equality L = an dz n + · · · + a1 dz + a0 holds in the ring
End(OU ).
Obviously, if L : OU → OU is a linear differential operator of order ≤ n
and U 0 ⊂ U is an open subset, the restriction L|U 0 is also a linear differential
operator of order ≤ n.
Exercise 2.8. In the above definition, prove that the ai are unique.
Remark 2.2. In the above definition, the functions in (ker L)(V ) are obvi-
ously the same as the solutions on V of the homogeneous linear differential
equation
dn y dy
an n + · · · + a 1 + a0 y = 0.
dz dz
www.pdfgrip.com
In this way, ker L is an object which simultaneously encodes the solutions

of the above differential equation on each open subset of U .
Definition 2.5. A (holomorphic) linear differential operator on U is an

endomorphism L : OU → OU for which there is an open covering {Ui }i∈I
of U and a family of non-negative integers {ni }i∈I such that the restriction
L|Ui is a (holomorphic) linear differential operator of order ≤ ni for each
i ∈ I.
The set of (holomorphic) linear differential operators on U will be de-

noted by D(U ). It is clear that for V ⊂ U ⊂ C open sets, the restriction
to V of any linear differential operator on U is also a linear differential
operator.
Exercise 2.9. (1) Prove that D(U ) is a sub-C-algebra of End(OU ).

(2) Prove that if U is connected, then for any linear differential operator
L on U there exist an integer n ≥ 0 such that L is of order ≤ n. What
happens when U is not connected? Is any differential linear operator on U
of finite order?
(3) Let L : OU → OU be an endomorphism and assume that there is an
open covering {Ui }i∈I such that L|Ui is a (holomorphic) linear differential
operator on Ui for each i ∈ I. Prove that L is also a (holomorphic) linear
differential operator on U .
Remark 2.3. The family {D(V ), V ⊂ U open subset}, as in remark 2.1,

satisfies the same formal properties as subsheaves of OU (see definition 2.2).
It is the another instance of “abstract sheaf”, that will be denoted by DU ,
and which is an “abstract subsheaf” of End (OU ) (see the book9 ).
Definition 2.6. If F ⊂ OU is a subsheaf and p is a point of U , we define

the stalk of F at p, denoted by Fp , as the quotient set M/ ∼, where
M = {(V, f ) | V ⊂ U is an open neighborhood of p, f ∈ F(V )}
and ∼ is the equivalence relation given by
def.
(V, f ) ∼ (V 0 , f 0 ) ⇔ ∃W ⊂ V ∩ V 0 open neighb. of p s.t. f |W = f 0 |W .
The stalk Fp is a complex vector space under the operations:
λ(V, f ) = (V, λf ), (V, f ) + (V 0 , f 0 ) = (V ∩ V 0 , f |V ∩V 0 + f 0 |V ∩V 0 ).
If V ⊂ U is an open subset and f ∈ F(V ), the equivalence class of (V, f ) in
Fp will be called the germ of f at p, and will be denoted by fp .
www.pdfgrip.com
Remark 2.4. The stalk Fp can be described as the inductive limit (or
colimit) of the system F(V ) when V runs into the open neighborhoods of
p contained in U , ordered by the reverse inclusion.
Exercise 2.10. (1) Prove that in the case F = OU , the stalk OU,p is a C-
algebra and that the Taylor expansion centered at p defines an isomorphism
of C-algebras
X∞
∼ 1 di f
Tp : OU,p −
→ C{z}, Tp (fp ) = i
(p)z i ,
i=0
i! dz
where C{z} is the C-algebra of convergent power series in one variable z

with complex coefficients.
(2) Prove that OU,p is a local ring, with maximal ideal mU,p = {ξ ∈
OU,p | ξ(p) = 0}, where ξ(p) = f (p) whenever ξ = (V, f ), f ∈ OU (V ).
(3) Prove that OU,p is a discrete valuation ring (Cf. Atiyah-MacDonald’s
book1 ch. 9), with valuation ν p : OU,p → N ∪ {+∞} defined by ν p (ξ) = r
if ξ ∈ mrU,p − mr+1
U,p , for any ξ 6= 0 and ν p (0) = +∞. In other words, if
ξ = fp , then ν p (ξ) is the vanishing order of f at p, i.e. ν p (fp ) = r with
f (q) = (q − p)r g(q) on a neighborhood of p, g holomorphic and g(p) 6= 0.
Exercise 2.11. Let F ⊂ OU be a subsheaf and p ∈ U . Prove that the stalk

Fp can be considered as a vector subspace of OU,p . Prove also that F = OU
if and only if Fp = OU,p for every p ∈ U .
The following proposition is a version of the analytic continuation prin-

ciple.
Proposition 2.1. Let U ⊂ C be a connected open set. Then the linear map
f ∈ OU (U ) 7→ fp ∈ OU,p is injective for each point p ∈ U .
Proof. Let us assume that fp = 0 and consider the set

W = {q ∈ U | fq = 0 in OU,p } ⊂ U.
It is clear that W is open and p ∈ W 6= ∅.
Let us prove that U − W is also open. If q ∈ U − W , then fq 6= 0 and
there is an open disc D ⊂ U centered at q such that f |D 6= 0. If f (q) 6= 0,
then, for D small enough, f (q 0 ) 6= 0 for all q 0 ∈ D. If f (q) = 0, since zeros
of holomorphic functions (6= 0) in one variable are isolated, we deduce that,
for D small enough, f (q 0 ) 6= 0 for all q 0 ∈ D − {q}. In any case we have
that, for D small enough, fq0 6= 0 for all q 0 ∈ D − {q} and so D ⊂ U − W .
www.pdfgrip.com
10
Since U is connected, we deduce that W = U and f = 0.
Corollary 2.1. Let U ⊂ C be a connected open set and V ⊂ U a non-

empty open set. Then, the restriction map f ∈ OU (U ) 7→ f |V ∈ OU (V ) is
injective.
Definition 2.7. Let L : OU → OU be an endomorphism and p ∈ U . The

stalk of L at p, denoted by Lp : OU,p → OU,p , is the linear map defined by

Lp (fp ) = Lp (V, f ) = (V, L(V )(f )) = (L(V )(f ))p
for every open neighborhood V ⊂ U of p and every f ∈ OU (V ).
Exercise 2.12. (1) If L, L0 : OU → OU are endomorphisms, prove that

(L + L0 )p = Lp + L0p , (L ◦ L0 )p = Lp ◦ L0p .
(2) If L : OU → OU is an endomorphism, L = 0 if and only if Lp = 0 for

all p ∈ U .
Exercise 2.13. In the situation of the above definition, prove that there are
canonical isomorphisms ker Lp ' (ker L)p , im Lp ' (im L)p . Prove also that
L is injectif, i.e. ker L = 0 (resp. L is surjectif, i.e. im L = OU ) if and only
if Lp is injectif (resp. Lp is surjectif) for all p ∈ U .
Example 2.2. Let U ⊂ C be an open set and p ∈ U . For simplicity, let us

assume that p = 0. Let us consider the linear differential operator on U ,
dn d
L = an n
+ · · · + a1 + a0 ,
dz dz
with ai ∈ OU (U ). Let us call ti ∈ C{z} the Taylor expansion at 0 of ai .
Then, under the isomorphism of exercise 2.10, the stalk L0 : OU,0 → OU,0
is identified with the linear endomorphism of C{z} given byc
dn s ds
s ∈ C{z} 7→ tn n
+ · · · + t1 + t0 s ∈ C{z}.
dz dz
Exercise 2.14. Let U ⊂ C be a connected open set and V ⊂ U a non-empty
open set. Prove that the restriction map DU (U ) → D(V ) is injective.
c In definition 7.1, we will study the ring of this kind of linear endomorphisms of C{z}.
www.pdfgrip.com
11
3. Sheaf Version of Cauchy Theorem
Definition 3.1. (1) Let U ⊂ C be a connected open set and F ⊂ OU a

subsheaf. We say that F is constant if for any p ∈ U the map f ∈ F(U ) 7→
fp ∈ Fp is an isomorphism.
(2) Let U ⊂ C be an open set and F ⊂ OU a subsheaf. We say that F
is locally constant, or a local system, if there is an open covering of U ,
S
U = Ui , by connected open sets, such that F|Ui is constant for all i.
Exercise 3.1. Let U ⊂ C be a connected open set. Prove that:
(1) CU is constant subsheaf of OU .
(2) If F ⊂ OU is a constant subsheaf and U 0 ⊂ U is a connected open set,

then the restriction F(U ) → F(U 0 ) is an isomorphism. Conclude that F|U 0
is also a constant subsheaf of OU 0 .
(3) Prove that any restriction of any locally constant subsheaf of OU is

locally constant.
(4) Prove that a subsheaf F ⊂ OU is locally constant if and only if there is

S
an open covering U = Ui such that F|Ui is locally constant for each i.
Exercise 3.2. (1) Prove that any constant subsheaf F ⊂ OU on a connected

open set U ⊂ C is determined by the complex vector subspace F(U ) of
OU (U ). Namely, for any open set V ⊂ U , F(V ) consists of functions which
locally are restrictions of functions in F(U ).
(2) Reciprocally, given a vector subspace E ⊂ OU (U ), prove that there is a

unique constant subsheaf F ⊂ OU such that F(U ) = E.
Exercise 3.3. Let F ⊂ OU be a locally constant subsheaf. Prove that the

function p ∈ U 7→ dimC Fp is locally constant.
If U is connected and F ⊂ OU is a locally constant subsheaf with Fp

finite dimensional vector space for some p ∈ U , then dim C Fq = dimC Fp = r
for all q ∈ U and we call F a locally constant subsheaf (or a local system)
of (finite) rank r.
The proof of the following proposition is a standard argument of general
Topology (see for instance prop. I.2.1 in the paper22 ).
Proposition 3.1. Any locally constant subsheaf F ⊂ OU on a simply con-

nected open set U ⊂ C is constant.
www.pdfgrip.com
12
Definition 3.2. Let U ⊂ C be a connected open set and

dn d
L = an n
+ · · · + a1 + a 0 : OU → O U
dz dz
a linear differential operator of order n, i.e. the function an does not vanish
identically on U . We say that p ∈ U is a regular point of L if an (p) 6= 0.
Otherwise, p will be called a singular point of L. The set of singular points
of L will be denoted by Σ(L).
The theorem 1.1 can be rephrased in the following way.
Theorem 3.1. Let U ⊂ C be a connected open set and L : OU → OU a

linear differential operator of order n. Then the following properties hold:
(1) The restriction (ker L) |U −Σ(L) is a local system of rank n.
(2) The restriction L|U −Σ(L) : OU −Σ(L) → OU −Σ(L) is surjective.
Moreover, for any singular point p ∈ Σ(L), ker Lp is a complex vector space
of dimension ≤ n.
Proof. (1) Let us call L = (ker L) |U −Σ(L) , U 0 = U − Σ(L) and let V ⊂ U 0

be a non-empty open disc. From Cauchy theorem 1.1 we know that for any
non-empty open disc W ⊂ V we have dimC L(W ) = n. In particular, the
restriction L(V ) → L(W ) is an isomorphism and so L|V is a constant sheaf.
(2) Cauchy theorem 1.1 implies that for any non-empty open disc V ⊂ U 0 ,
the map L(V ) : OU 0 (V ) → OU 0 (V ) is surjective. Hence, for any p ∈ U 0 the
map Lp : OU 0 ,p → OU 0 ,p is surjective.
For the last part, using proposition 2.1, it is clear that for any small
open disc V centered at a singular point p, the dimension of (ker L)(V )
is less or equal than the dimension of (ker L)(W ), for any small open disc
W ⊂ V − Σ(L), but for a such W we know that dimC (ker L)(W ) = n.
Corollary 3.1. Let U ⊂ C be a connected and simply connected open

set and L : OU → OU a linear differential operator of order n without
singular points. Then, L(U ) : O(U ) → O(U ) is surjective, i.e. the non-
homogeneous equation L(y) = g has always a holomorphic solution on U
for any g ∈ O(U ).
Proof. The proof of this corollary needs to use a small (and motivating)
argument of sheaf cohomology (see for instance9 ). Let us consider the exact
sequence of sheaves
www.pdfgrip.com
13
L
0→
− ker L →
− OU −
→ OU →
− 0
and the associated long exact sequence of cohomology (cf. loc. cit.)
L(U )
0→
− (ker L)(U ) → − H 1 (U, ker L) →
− OU (U ) −−−→ OU (U ) → − ···
From proposition 3.1 we know that ker L is a constant sheaf, ker L ' CnU ,
and so H 1 (U, ker L) ' H 1 (U, Cn ) = 0 since U is simply connected.
4. Local Monodromy
The universal covering space of C∗ = C − {0}, with base point 1, can be
realized for instance by
− (C∗ , 1),
q : (C, 0) → q(w) = e2πiw .
Base points can be moved inside the set of positive real numbers R∗+ ⊂ C∗
and inside the imaginary axis Ri ⊂ C without ambiguity, since both sets
are contractible.
The group of automorphisms of q is infinite cyclic generated by the
automorphism M : w ∈ C 7→ w + 1 ∈ C.
For any open disk D centered at 0, we write D f∗ = q −1 D∗ and we
also choose q : Df∗ → D as universal covering of D ∗ with base points in
∗
f∗ ∩ (Ri) and D∗ ∩ R∗ respectively. Let us denote by DR the open disk

D +
centered at 0 of radius R ∈]0, +∞].
Definition 4.1. A multivalued holomorphic function on D ∗ is by definition

f∗ .
a holomorphic function on D
∗
The set of multivalued holomorphic functions on DR is denoted by A0R .
It is clearly a conmmutative C-algebra without zero divisors. For 0 < R 0 <
R ≤ +∞ we have restriction maps A0R → A0R0 which are injective and
C-algebra homomorphisms.
Example 4.1. (1) The identity function w ∈ C 7→ w ∈ C is obviously an

element of A0∞ , which will be denoted by Log z. We will also denote by
Log z its restriction to any A0R with R > 0.
(2) Given a fixed complex number α, the function w ∈ C 7→ e2πiαw ∈ C is

also an element of A0∞ , wich will be denoted by z α . We will also denote by
z α its restrictions to any A0R .
www.pdfgrip.com
14
∗
The map f ∈ O(DR ) 7→ f ◦ q ∈ A0R is injective and so we can think
in O(DR ) as a sub-C-algebra of A0R . The automorphism M induces an
∗
automorphism of C-algebras
T : g ∈ A0R 7→ T (g) = g ◦ M ∈ A0R ,
called monodromy operator. It is clear that T commutes with restrictions

∗
and that T (g) = g for any g ∈ O(DR ).
Exercise 4.1. Prove that any multivalued holomorphic function g ∈ A0R

∗
which is “uniform”, i.e. T (g) = g, belongs to O(DR ) and so
∗
O(DR ) = {g ∈ A0R | T (g) = g}.
Definition 4.2. Let g be a multivalued holomorphic function on D ∗ and

U ⊂ D∗ a simply connected open set. A determination of g on U is a
holomorphic function f on U which is obtained as f = g ◦ σ, where σ : U →
f∗ is a holomorphic section of q.
D
f∗ → D∗ is a
Let f = g ◦ σ a fixed determination of g on U . Since q : D
covering space, σ must be a biholomorphic map between U and the open
set σ(U ). Any other holomorphic section of q on U must be of the form
F
M k ◦ σ and q −1 U = k∈Z M k (σ(U )). Hence, any determination of g on U
is of the form T k (g) ◦ σ.
Definition 4.3. We say that a multivalued holomorphic function g on D ∗

is of finite determination if the vector space generated by T k (g), k ∈ Z, is
finite dimensional.
Proposition 4.1. Let g be a multivalued holomorphic function on D ∗ . The

following properties are equivalent:
(a) g is of finite determination.

(b) The vector space generated by the determinations of g on any simply
connected open set U ⊂ D ∗ is finite dimensional.
(c) The vector space generated by the determinations of g on some simply
connected open set U ⊂ D ∗ is finite dimensional.
Proof. The key point is that if we take any simply connected open set
U ⊂ D∗ and we fix a holomorphic section σ : U → D f∗ of q, then σ
must be a biholomorphic map between U and the open set σ(U ) ⊂ D f∗ ,
k
any other holomorphic section of q on U must be of the form M ◦ σ and
www.pdfgrip.com
15
F
q −1 U = k∈Z M k (σ(U )). So, if f = g ◦ σ is a fixed determination of g on
U , then the map
T k (g) 7→ T k (g) ◦ σ = g ◦ M k ◦ σ
is a bijection between the set {T k (g), k ∈ Z} and the set of determinations

of g on U , which clearly preserves linear dependence.
∗
The set of all multivalued holomorphic function on DR of finite deter-
0
mination is a sub-C-algebra of AR , stable by T , and will be denoted by AR .
It is clear that the restriction map A0R → A0R0 sends AR into AR0 .
Example 4.2. (1) Since T (Log z) = 1 + Log z, Log z is a multivalued

holomorphic function of finite determination.
(2) Since T (z α ) = e2πiα z α , z α is a multivalued holomorphic function of

finite determination.
∗
Definition 4.4. Let V ⊂ DR be a convex open neighborhood of R∗+ ∩
DR∗
and let Ve ⊂ D
g ∗
R be the unique connected component of q
−1
V which
intersects the imaginary axis of C. We say that a holomorphic function
∗
f ∈ O(V ) extends to a multivalued holomorphic function on DR if there is
0
a (unique) g ∈ AR such that g|Ve = f ◦ q|Ve . In such a case we say that g is
the multivalued extension of f .
Let us note that in the above definition, f extends to the multivalued

∗
holomorphic function g on DR if and only if f is a determination of g on
V.
∼
Example 4.3. The restriction q|Ve : Ve − → V is biholomorphic. The inverse
−1
function f = q|Ve : V → Ve ⊂ C extends, obviously by definition,
∗
to a multivalued function on DR . In fact, its multivalued extension is the
g
identity function of DR . We have f (1) = 0 and e2πif (z) = (q ◦ f )(z) = z for
∗
all z ∈ V , and so dz = (2πi)e2πif (z) df = (2πi)zdf and

Z z
1 dζ
f (z) = , ∀z ∈ V,
2πi 1 ζ
where the integration path is taken inside the simply connected open set V .
The function f coincides with the usual logarithm “ln” up to the scalar fac-
tor (2πi)−1 . This explains why we denote by “Log z” the identity function
∗
on C considered as “multivalued function” on DR .
www.pdfgrip.com
16
We have then injective maps

∗ ◦q ∇
O(DR ) ,→ O(DR ) ,→ AR ,→ A0R ,→ O(V ) (4)
where the last one associates to any multivalued holomorphic
−1 function g ∈
0
AR its “main determination” on V , ∇(g) = g ◦ q|Ve . The compositions
∗
O(DR ) → O(V ) and O(DR ) → O(V ) are nothing but the restriction maps.
For any radius R0 ∈]0, R] we have a commutative diagram
◦q
O(DR )
/ O(D∗ ) / AR / A0 / O(V ) (5)
R R
rest. rest. rest. rest. rest.

◦q
/ O(D∗ 0 ) / A0 0 / O(V ∩ D∗ 0 )

O(DR0 ) / AR 0
R R R
Exercise 4.2. (1) Prove that ∇(A0R ) is a subspace of O(V ) stable under the
d
action of the derivative dz . Conclude that A0R has a natural structure of
∗ ∗
left D(DR )-module in such a way that ∇ is D(DR )-linear. In particular,
0
AR is a left D(DR )-module.
(2) Prove that the monodromy T : A0R → A0R is D(DR
∗
)-linear.
∗
(3) Prove that AR is a sub-D(DR )-module of A0R .
Proposition 4.2. In the situation of definition 4.4, for any holomorphic

function f ∈ O(V ), the following properties are equivalent:
∗
(a) f extends to a multivalued holomorphic function g on DR of finite
determination.
(b) There is a locally constant subsheaf F ⊂ ODR∗ of finite rank such that
f ∈ F(V ).
Proof. We can assume that f 6= 0.

(a) ⇒ (b): Let us call F e ⊂ Og ∗ the constant subsheaf determined by
DR
the finite dimensional vector subspace E ⊂ O g g∗ k
∗ (D ) generated by T (g),
DR R
k ∈ Z (see exercise 3.2).
∗
For each open subset W ⊂ DR , we define F(W ) ⊂ ODR∗ (W ) as the
vector space of holomorphic functions h on W for which there is an open
S e −1 Wi ) for all i.
covering W = Wi such that h|Wi ◦ q|q−1 Wi belongs to F(q
It is clear that F is a subsheaf of ODR∗ .
∗
Let U ⊂ DR be a simply connected open subset and let us choose a
simply connected open subset U 0 ⊂ D g∗ 0
F R such that q(U ) = U . One has
−1 k 0 k 0 ∼
q U = k∈Z M (U ) and q : M (U ) → U for all k ∈ Z. For each open
www.pdfgrip.com
17
F
set W ⊂ U , let us call W 0 = U 0 ∩ q −1 W and so q −1 W = k∈Z M k (W 0 )
∼
and q : M k (W 0 ) → W for each k ∈ Z. It is easy to see that for a holomor-
phic function h on W , the condition h ◦ q|q−1 W ∈ F(qe −1 W ) is equivalent
e
to the condition h ◦ q|W 0 ∈ F(W 0
). In particular, one has that a holomor-
phic function h on W belongs to F(W ) if and only if h ◦ q|W 0 ∈ F(W e 0
).
Composition with q gives rise to a commutative diagram
∼
F(U ) e 0)
/ F(U
stalk ' stalk

∼
Fq(x) / Fx
for each x ∈ U 0 , where the horizontal arrows are isomorphism because

∼
q : U 0 → U is biholomorphic and the right vertical arrow is an isomorphism
because F e is a constant subsheaf of O g∗ . We deduce that the map F(U ) →
DR
Fy is an isomorphism for each y ∈ U , and so F|U is a constant subsheaf of
OU of finite rank. It is also clear that f ∈ F(V ).
(b) ⇒ (a): For each open set G ⊂ D g ∗ e
R let us define F(G) as the vector
space of holomorphic functions e h on G for which there is an open covering
F ∼
G = i Gi , with q : Gi → q(Gi ), and functions hi ∈ F(q(Gi )) such that
e
h|Gi = hi ◦ q|Gi for all i. It is clear that Fe is a subsheaf of O g ∗ and that
DR
e e e
f := f ◦ q|Ve ∈ F(V ).
e to any open set G ⊂ D
It is not difficult to see that the restriction of F g ∗
R
for which the restriction of q gives a biholomorhic map between G and
q(G) is a locally constant subsheaf of OG of finite rank (the same one
as the rank of F). So, F e is locally constant of finite rank too, and from
proposition 3.1 we deduce that F e is constant of finite rank. In particular,
e
there is a (unique) g ∈ F(DR ) ⊂ A0R such that g|Ve = fe and f extends to
g ∗
the multivalued holomorphic function g. Finally, g is of finite determination

e D
because T k (g) ∈ F( g∗
R ) for all k ∈ Z and this space is finite dimeinsional.
Let L be a linear differential operator on DR of order n with Σ(L) ⊂ {0}:

dn d
L = an + · · · + a1 + a0 , ai ∈ O(DR ), an (z) 6= 0 ∀z 6= 0.
dz n dz
From Cauchy theorem (see 3.1) we know that (ker L)|DR∗ is a locally con-
stant sheaf of rank n.
Proposition 4.3. Under the above hypothesis and with the notations of
definition 4.4, the following properties hold:
www.pdfgrip.com
18
(1) Any multivalued holomorphic function g ∈ A0R annihilated by L is of

finite determination and
{g ∈ A0R | L(g) = 0} = {g ∈ AR | L(g) = 0} ' (ker L)(V ).
In particular dimC ({g ∈ AR | L(g) = 0}) = n.
(2) The maps L : A0R → A0R and L : AR → AR are surjective.
Proof. (1) We have ∇({g ∈ A0R | L(g) = 0}) ⊂ (ker L)(V ) and from
proposition 4.2, we know that (ker L)(V ) ⊂ ∇(AR ). We conclude that
∇
{g ∈ A0R | L(g) = 0} = {g ∈ AR | L(g) = 0} ' (ker L)(V ).
e−2πiw d
(2) For g ∈ A0R , we have d
dz (∇(g)) = ∇(δ(g)), où δ = 2πi dw , and
e
L(∇(g)) = ∇(L(g)) with
e = an (e2πiw )δ n + · · · + a0 (e2πiw ) =
L
−2πniw
dn dn−1 g
an (e2πiw ) e(2πi)n dw n + bn−1 (w) dw ∗
n−1 + · · · + b0 (w) ∈ D(DR ).
e = ∅, we deduce from corollary 3.1 that L

e : O(D
g ∗ g∗
Since Σ(L) R ) → O(DR ) is
0 0
surjective, and so L : AR → AR is surjective.
If g ∈ AR , there is a non-vanishing polynomial P (X) such that
P (T )(g) = 0. We have proved that there is h ∈ A0R such that L(h) = g,
but L(P (T )(h)) = P (T )(g) = 0. We deduce from (1) that P (T )(h) ∈ AR
and h ∈ AR . So, L : AR → AR is surjective.
d
Example 4.4. (1) For L = z dz − α, we have {g ∈ A0R | L(g) = 0} = hz α i.
2
d d
(2) For L = z dz 2 + dz , we have {g ∈ A0R | L(g) = 0} = hLog z, 1i.
Theorem 4.1. Any multivalued holomorphic function g ∈ AR of finite

determination can be expressed as a finite sum
X
g= φα,k z α (Log z)k ,
α∈C,k≥0
∗
where the φα,k ∈ O(DR ) are uniform functions. Moreover, the φα,k are
uniquely determined if we impose that the difference α−α0 is not an integer
whenever φα,k , φα0 ,k 6= 0 for some k (this can be guaranteed for instance if
we restrict ourselves to the set of complex numbers α with −1 ≤ < α < 0).
In fact we have a more precise statement. Let E ⊂ AR be the finite
Q
dimensional vector subspace generated by the T k (g), k ∈ Z, let (X −
λj )rj be the minimal polynomial of the action of T on g (the λj are the
eigenvalues of T |E with λj 6= λl whenever j 6= l), let d(g) be the degree of
www.pdfgrip.com
19
this polynomial, and let us choose complex numbers αj ∈ C with e2πiαj =

λj . Then, there are unique φj,k ∈ O(D∗ ) such that
j −1
X rX
ξ= φj,k z αj (Log z)k .
j k=0
Proof. The key point is that, for λ = e2πiα and k ≥ 0, we have

X k
k−1
α k
(T − λ) z (Log z) = λ z α (Log z)i
i=0
i
and, for any polynomial P (X) ∈ C[X],

P (T ) z α (Log z)k = P (λ)z α (Log z)k + ck−1 z α (Log z)k−1 + · · · + c0 z α ,
where the ci are complex numbers. As a consequence,

(T − λ)k z α (Log z)k = k!λk z α , (T − λ)k+1 z α (Log z)k = 0.
Let us start with uniqueness. Assume that g = 0. We proceed by induction
P P
on r = j (rj − 1). If r = 0, then 0 = g = j φj,0 z αj , and taking Pl =
Q
j6=l (X − λj ) we obtain
X
0 = Pl (T )(g) = Pl (T )(φj,0 z αj ) = Pl (λl )φl,0 z αl
j
and so φl,0 = 0, for each l.

Let us suppose that we have the uniqueness of the coefficients φj,k every
time r ≤ ν and suppose that g = 0 with
j −1
X rX
g= φj,k z αj (Log z)k
j k=0
P
and j (rj − 1) = ν + 1. Let us consider the polynomial P (X) = (X −
Q
λ1 )r1 −1 j6=1 (X − λj )rj . We have
Y
0 = P (T )(g) = · · · = φ1,r1 −1 (r1 − 1)!λr11 −1 (λ1 − λj )rj z α1
j6=1
and so φ1,r1 −1 = 0. To conclude, we apply the induction hypothesis to

j −1
1 −2
rX X rX
0=g= φ1,k z α1 (Log z)k + φj,k z αj (Log z)k .
k=0 j6=1 k=0
Now, let us prove the existence of the φj,k . We proceed by induction on the
degree d(g) of the minimal polynomial of the action of T on g.
www.pdfgrip.com
20
If d(g) = 1 then there is a complex number λ1 6= 0 such that (T −

λ1 )(g) = 0. Consequently, T (z −α1 g) = z −α1 g and φ1,0 := z −α1 ξ is uniform:
g = φ1,0 z α1 .
Assume the result true for any multivalued function h ∈ AR with d(h) ≤
d.
Let g ∈ AR be a multivalued holomorphic function of finite determina-
Q
tion with d(g) = d + 1 and let P (X) = j (X − λj )rj be the minimal poly-
P
nomial of the action of T on g. We have d(g) = j rj = d + 1. Let us write
Q
Q(X) = j6=1 (X−λj )rj and P 0 (X) = P (X)/(X−λ1 ) = (X−λ1 )r1 −1 Q(X).
∗
From the first step of the induction, we know that there exists a ψ ∈ O(DR )
0 α1
such that P (T )(g) = ψz .
We
have P 0 (T ) z α1 (Log z)r1 −1 = Q(T )(T − λ1 )r1 −1 z α1 (Log z)r1 −1 = (r1 −
1)!λr11 −1 Q(λ1 )z α1 and so P 0 (T )(φ1,r1 −1 z α1 (Log z)r1 −1 ) = ψz α1 with
ψ
φ1,r1 −1 = .
(r1 − 1)!λr11 −1 Q(λ1 )

We deduce that P 0 (T ) g − φ1,r1 −1 z α1 (Log z)r1 −1 = 0 and we conclude by
applying the induction hypothesis to g − φ1,r1 −1 z α1 (Log z)r1 −1 .
Remark 4.1. In the course of the proof of the above theorem, we have also
proved that if E ⊂ AR is the finite dimensional vector subspace generated
by the determinations of g and
j −1
X rX
g= φj,k z αj (Log z)k
j k=0
∗
with φj,k ∈ O(DR ) and φj,rj −1 6= 0 for all j, then each φj,rj −1 z αj belongs
to E and it is an eigenvector of T |E with respect to the eigenvalue λj .
Exercise 4.3. Prove that for any complex number λ, the map T − λ : AR →
AR is surjective d .
Remark 4.2. Let τ : C/Z → C be any section of the canonical projection

C → C/Z. The above theorem says that {z α (Log z)k | α ∈ im τ, k ≥ 0} is a
basis of AR as an O(D∗ )-module.
d ActuallyT − λ : A0R → A0R is also surjective, but the proof needs a cohomological
argument (cf. th. (4.1.2) in20 ).
www.pdfgrip.com
21
5. Fuchs Theory
In this section we study the behavior of a linear differential equation, or of
a linear differential operator, in the neighborhood of a singular point.
Definition 5.1. We say that a multivalued holomorphic function g ∈ AR

is regular, or of the Nilsson class (at 0), if in the expression
j −1
X rX
g= φj,k z αj (Log z)k ,
j k=0
the φj,k are meromorphic functions at 0.
It is clear that a g ∈ AR is regular at 0 if and only if its restriction to

some (or to any) AR0 , with 0 < R0 < R, is regular at 0
Let us denote by NR the set of g ∈ AR which are regular (at 0). It is
clear that NR is a sub-C-algebra of AR .
Exercise 5.1. Prove that NR is a sub-D(D)-module of AR . Is NR a sub-

D(D∗ )-module of AR ?.
n
d d
Let L = an dz n + · · · + a1 dz + a0 be a linear differential operator on
D = DR of order n (an 6= 0), and let us assume that 0 is the only singular
point of L.
Definition 5.2. We say that 0 is a regular singular point of L if any g ∈ AR

such that L(g) = 0 is regular at 0.
Remark-Definition 5.1. It is clear that if D 0 ⊂ D is an open disc centered

at 0 and L0 = L|D0 , then 0 is a regular singular point of L if and only if
it is so of L0 . In particular, if L is a linear differential operator on some
open neighborhood of 0, and 0 is a singular point of L, we say that 0 is
a regular singular point of L if it is so for the restriction of L to a small
enough open disc centered at 0. More generally, if L is a linear differential
operator on an open set U ⊂ C and p ∈ U is a singular point of L, we say
that p is a regular singular point of L if 0 is a regular singular point of the
“translated” operator
dn d
L0 = a0n + · · · + a01 + a00
dz n dz
with a0k (z) = ak (z + p), which is defined on the open neighborhood of 0,
U 0 = {z ∈ C | z + p ∈ U }.
www.pdfgrip.com
22
For a function a ∈ O(U ) and a point p ∈ U , let us write ν p (a) for the
vanishing order of a at p. It only depends on the germ ap (see exercise 2.10).
If ap = 0 then ν p (0) = +∞.
n
d
Theorem 5.1. (Fuchs) Let U ⊂ C be an open set, L = an dz n + ··· +
d
a1 dz + a0 a linear differential operator on U of order n ≥ 1 and p ∈ U a
singular point of L. Then, the following properties are equivalent:
(a) p is a regular singular point of L.

(b) max {k − ν p (ak )} = n − ν p (an ).
0≤k≤n
Proof. The proof of this theorem can be found in the book,8 15.3.
6. Index of Differential Operators at Singular Points

n
d d
Let U ⊂ C be a connected open set and L = an dz n + · · · + a1 dz + a0 a
linear differential operator on U of order n. Cauchy theorem 1.1 tells us

that, for any non-singular point p ∈ U of L (an (p) 6= 0), the stalk at p of L,
Lp : OU,p → OU,p , is a surjective map and dimC ker Lp = n. On the other
hand, if p ∈ Σ(L), we have dimC ker Lp ≤ n (see theorem 3.1), but what
about dimC coker Lp ?
We have the following important result, known as Komatsu-Malgrange
index theorem.11,17
Theorem 6.1. Under the above hypothesis, the following properties hold:
(1) dimC coker Lp < ∞.

(2) χ(Lp ) = dimC ker Lp − dimC coker Lp = n − ν p (an ).
The proof of the above theorem consists of a reduction to the case where
dn
the differential operator is of the form L0 = an dz n , where an easy compu-
n
0 d
tation shows that χ(Lp ) = χ dzn + χ(an ) = n − ν p (an ). The reduction is
based on the fact that L can be seen as a compact perturbation of L0 on
convenable Banach spaces.
Let us write O = OU,p , m = mU,p for its maximal ideal and P = Lp :

O → O. We know that Taylor development at p establishes an isomorphism
between O and the ring of convergent power series C{z}, which sends the
ideal m to the ideal (z) (see exercise
2.10). It is easy to see that, for any
n+k k
integer k ≥ 0, we have P m ⊂ m and so P is continuous for the m-adic
topology and induces a linear endomorphism Pb of the m-adic completion
www.pdfgrip.com
23
b of O, which is isomorphic to the (z)-adic completion of C{z}, i.e. to the

O
formal power series ring C[[z]].
The proof of the following theorem is much easier than the proof of
theorem 6.1 and can be found in the paper,17 prop. 1.3 and th. 1.4.
Theorem 6.2. In the above situation, the following properties hold:
(1) The vector spaces ker Pb and coker Pb are finite dimensional and
χ(Pb ) = dimC ker Pb − dimC coker Pb = max {k − ν p (ak )}.

0≤k≤n
(2) The induced map Pe = O/O b b

→ O/O is always surjective and
dimC ker Pe = max {k − ν p (ak )} − (n − ν p (an )).
0≤k≤n
Corollary 6.1. In the above situation, the following properties are equiv-
alent:
(1) p is a regular singular point of L.

(2) χ(Pe ) = 0.
(3) Pe is an isomorphism.
(4) ker Pe = 0.
(5) The canonical maps ker P → ker Pb and coker P → coker Pb are isomor-
phisms.
(6) dimC ker P = dimC ker Pb and dimC coker P = dimC coker Pb .
Proof. From the following commutative diagram
0 −−−−→ O −−−−→ Ob −−−−→ O/O

b −−−−→ 0
  
  
Py by
P ey
P
b −−−−→ O/O
0 −−−−→ O −−−−→ O b −−−−→ 0
we obtain the exact sequence
δ
0 → ker P → ker Pb → ker Pe −→ coker P → coker Pb → coker Pe(= 0) → 0
(6)
and soe χ(P ) − χ(Pb) + χ(Pe ) = 0. From theorems 5.1, 6.1 and 6.2 we have
that (1) ⇔ (2) ⇔ (3) ⇔ (4). On the other hand, from equation (6) we
deduce that (4) ⇔ (5) and (6) ⇒ (4), and finally (5) ⇒ (6) is obvious.
e In fact this is part of the proof of (b) in theorem 6.2.

www.pdfgrip.com
24
Remark 6.1. The above corollary shows that the finite dimensional vector
space ker Lfp is a measure of the non-regularity (or the irregularity) of the
singular point p of L. This point of view is the first step of the notion of
irregularity complexes of holonomic D-modules in higher dimension (see the
paper21 ).
7. Division Tools
The material of this section is taken from the papers.2,13
In this section we work over the ring of convergent power series in one
variable O = C{z}, that we can think as the ring of germs at 0 of holomor-
phic functions defined on a open neighborhood of the origin. Let us denote
by ∂ : O → O the derivative with respect to z.
Definition 7.1. A C-linear endomorphism L : O → O will be called a

linear differential operator of O of order ≤ n if there exist a0 , . . . , an ∈ O
such that, for any g ∈ O we have L(g) = an ∂ n (g) + · · · + a1 ∂(g) + a0 g. In
such a case we will write, as usual, L = an ∂ n + · · · + a1 ∂ + a0 .
By example 2.2, linear differential operators of O are nothing but the

stalk at the origin of linear differential operators defined on an open neigh-
borhood of 0.
Let us denote by F n D ⊂ EndC (O) the set of linear differential operators

S
of order ≤ n and D = n≥0 F n D ⊂ EndC (O). Let us note that the map
a ∈ O 7→ [g ∈ O 7→ ag ∈ O] ∈ EndC (O)
is an injective homomorphism of C-algebras and its image coincides with
F 0 D. From now on, we will identify O = F 0 D. We also set F −1 D = {0}.
For a P ∈ D, with P 6= 0, let us write ord P for its order, i.e ord P = n
means that P ∈ F n D but P ∈ / F n−1 D. For P = 0 we write ord 0 = −∞.
Exercise 7.1. Prove the following recursive description of the F n D:
F 0 D = {P ∈ EndC (O) | [P, a] = P a − aP = 0, ∀a ∈ O},

F n+1 D = {P ∈ EndC (O) | [P, a] ∈ F n D, ∀a ∈ O}.
Exercise 7.2. (see the notes3 ) Prove that:

(1) D is a non-commutative sub-C-algebra of EndC (O).
(2) (F r D)(F s D) ⊂ F r+s D (we say that the family {F n D}n≥0 is a filtration
of the ring D, or that (D, F ) is a filtered ring.)
www.pdfgrip.com
25
(3) The vector space ⊕n≥0 F n D/F n−1 D has a natural structure of ring (in
fact a C-algebra), that we will call the associated graded ring of the filtered
ring (D, F ) and will be denoted by grF D.
(4) If P, Q ∈ D and P, Q 6= 0, then P Q 6= 0 and ord P Q = ord P + ord Q.
(5) If P, Q ∈ D, then ord(P Q − QP ) ≤ ord P + ord Q − 1 and so grF D is
a commutative ring, and that it is isomorphic to the polynomial ring O[ξ].
Exercise 7.3. Prove that the ring D is simple, i.e. it has not any non trivial
two-sided ideal.
Definition 7.2. If P ∈ D is a non-zero operator with ord(P ) = n, we

define its symbol as
σ(P ) = P + F n−1 D ∈ F n D/F n−1 D = grnF D.
It is clear that if P, Q ∈ D are non-zero, then σ(P Q) = σ(P )σ(Q).
Definition 7.3. Given a left ideal I ⊂ D, we define σ(I) as the ideal of

grF D generated by σ(P ), for all P ∈ I, P 6= 0.
Exercise 7.4. Prove that D is left and right noetherian.
Let P be a non-zero linear differential operator (of O) of order n ≥ 0,

Pn P∞
i.e. P = k=0 ak ∂ k , with ak ∈ O and an 6= 0. Let us write ak = l=0 alk z l
and so
Xn X ∞
P = alk xl ∂ k .
k=0 l=0
We call the Newton diagram (or the support) of P the set

supp(P ) = {(l, k) ∈ N2 | alk 6= 0} ⊂ N2 .
Definition 7.4. In the above situation, we define the valuation of P as

ν(P ) = ν 0 (an ) and the exponent of P as exp(P ) = (ν(P ), ord P ).
Exercise 7.5. Prove that if P, Q ∈ D, P, Q 6= 0, then exp(P Q) = exp(P ) +

exp(Q).
Lemma 7.1 (Briançon-Maisonobe2 ). Let P ∈ D, P = 6 0 and exp(P ) =

(v, d). Then, for any A ∈ D there are unique Q, R ∈ D such that A =
QP + R with
ord(A) v−1
X X
R= rlk xl ∂ k + S, with ord(S) < d.
k=d l=0
www.pdfgrip.com
26
The proof of the above lemma is easy, and in fact it is a particular case
of the general division theorems in several variables (see the lectures by F.
Castro). Let us note that the condition on the remainder R is equivalent
to say that

supp(R) ⊂ N2 \ exp(P ) + N2 .
Let us denote by K the field of fractions of the ring O. Any element of K
can be written as a/z r , with a ∈ O and r ≥ 0. We can think of elements of
K as the germs at 0 of meromorphic functions defined on a neighborhood
of 0 and with a pole eventually at 0. The derivative ∂ : O → O extends
obviously to K.
Let DK be the ring of linear differential operators of K, i.e. the subring
of EndC (K) with elements of the form
Xn
ak ∂ k , ak ∈ K.
k=0
The ring is filtered in the obvious way and for any P ∈ DK , P 6= 0, the
definition of its order ord(P ) is clear.
The proof of following lemma is easy.
Lemma 7.2. Let P ∈ DK , P 6= 0. Then, for any A ∈ DK there are unique
Q, R ∈ DK such that A = QP + R with ord(R) < ord(P ).
Corollary 7.1. Let P ∈ D, P 6= 0. Then, for any A ∈ D there are Q, R ∈
D and an integer r ≥ 0 such that xr A = QP + R with ord(R) < ord(P ).
Definition 7.5. Let I ⊂ D be a non-zero left ideal. We define the set
Exp(I) = {exp(P ) | P ∈ I, P 6= 0}.
It is clear that Exp(I) is an ideal of N2 , i.e. Exp(I) + N2 ⊂ Exp(I).
Given a non-zero left ideal I ⊂ D let us write
p = p(I) = min{ord(P ) | P ∈ I, P 6= 0},
and for each d ≥ p,
αd = αd (I) = min{ν(P ) | P ∈ I, P 6= 0, ord(P ) = d}.
Since αp ≥ αp+1 ≥ · · · we can define
q = q(I) = min{d ≥ p | αd = αe , ∀e ≥ d}.
We also define
ν(I) = min{ν(P ) | P ∈ I, P 6= 0}.
It is clear that ν(I) = αq(I) (I).
www.pdfgrip.com
27
Exercise 7.6. With the above notations, prove that

q
[
Exp(I) = (αd , d) + N2 .
d=p
Definition 7.6. With the above notations, a set of elements Fp ,

Fp+1 , . . . , Fq ∈ I with exp(Fd ) = (αd , d) for p ≤ d ≤ q, is called a standard
basis, or a Gröbner basis, of I.
If Fp , Fp+1 , . . . , Fq is a Gröbner basis of I, then p(I) = ord(Fp ) and

ν(I) = ν(Fq ).
For any A ∈ D, and by successive division (lemma 7.1) by the elements
Fq , Fq−1 , . . . , Fp of I, we obtain a unique expression
A = Qp Fp + · · · + Qq−1 Fq−1 + Qq Fq + R
with Qp , . . . , Qq−1 ∈ O, Qq ∈ D and
ord(A) αk −1
X X
R= rlk xl ∂ k + S, with ord(S) < p,
k=p l=0
or in other words
supp(R) ⊂ N2 \ Exp(I).
In particular, A ∈ I ⇔ R = 0 and so any Gröbner basis Fp , Fp+1 , . . . , Fq of
I is a system of generators I.
Exercise 7.7. Prove that if Fp , Fp+1 , . . . , Fq is a Gröbner basis of I, then

σ(I) = (σ(Fp ), . . . , σ(Fq )).
Example 7.1. Let I = D be the total left ideal. It is clear that I is

generated by ∂, z. However, σ(I) = σ(D) = grF D is not generated by
σ(∂) = ξ, σ(z) = z.
Given a left ideal I ⊂ D and a system of generators P1 , . . . , Pr of I,

often we are interested in the module of syzygies (or relations) of the Pi
X
S(P ) = {(Q1 , . . . , Qr ) ∈ Dr | Qi Pi = 0}.
i
r
This module is a sub-D-module of D , and so it is finitely generated.
In general it is not clear how to exhibit a finite number of generators of
S(P ), but the situation is simpler if the Pi form a Gröbner basis of I.
www.pdfgrip.com
28
Let us keep the notations of definition 7.6, and let us assume that the
Fd satisfy the following property:
Fd = z αd ∂ d + terms of lower order.
We say in that case that our Gröbner basis is normalized.
For each d = p + 1, . . . , q, there are unique Qdl ∈ O, l = p, . . . , d − 1 such
that
∂Fd−1 − z αd−1 −αd Fd = Qdp Fp + · · · + Qdd−1 Fd−1 .
We have then the following syzygies of (Fp , Fp+1 , . . . , Fq ):
d−1 d
| {z } |{z}
Rd = (Qdp , Qdp+1 , . . . , −∂ + Qdd−1 , z αd−1 −αd , 0, . . . , 0)
for d = p + 1, . . . , q.
We have the following result (see prop. 3 in2 ). It is a particular case of
a general result valid for Gröbner bases in several variables and in various
settings (see the notes3 ).
Proposition 7.1. The module of syzygies of (Fp , Fp+1 , . . . , Fq ) is generated
by Rp+1 , . . . , Rq .
Proposition 7.2. (Cf. prop. 8.8 in 10 or lemme 10.3.1 in 27 ) Let M be a
left D-module which is finitely generated as O-module. Then it is free (of
finite rank) as O-module.
Proof. We reproduce the proof of lemme 4 in.2 Let B = {e1 , . . . , ep } be a

minimal system of generators of M as O-module and let us write
p
X
∂ei = vij ej , (vij ∈ O) ∀i = 1, . . . , p.
j=1
Let S be the module of syzygies of B:

X
S = {u = (u1 , . . . , up ) ∈ Dp | ui ei = 0}.
i
If B is not a basis, then S 6= 0 and we can define ω = min{ν(u) | u ∈

S, u 6= 0}, where ν(u) = min{ν(ui ) | ui 6= 0}. By Nakayama’s lemma, the
set of classes B = {e1 , . . . , ep } is a basis of the (O/m =)C-vector space
M/mM and so we have ω > 0. Let u ∈ S be a non-vanishing syzygy with
ν(u) = ν(uj0 ) = ω. We have
p
X p
X p
X
0=∂ ui e i = · · · = wj ej , with wj = ∂(uj ) + ui vij ,
i=1 j=1 i=1
www.pdfgrip.com
29
but ν(∂(uj0 )) = ν(uj0 ) − 1 and so ν(wj0 ) = ω − 1, which contradicts the

minimality of ω.
Proposition 7.3. Let I ⊂ D a non-zero left ideal with

q
[
Exp(I) = (αd , d) + N2
d=p
(see exercise 7.6), and Fp , . . . , Fq ∈ I a Gröbner basis of I. Then, the

following properties hold:
(1) For any A ∈ I, there is an integer r ≥ 0 such that xr A ∈ DFp .
(2) I = D(Fp , Fq ).
Proof. We reproduce the proof of prop. 5 in.2 Part (1) is a starightforward

consequence of corollary 7.1. For part (2), let us consider the left D-module
M = I/D(Fp , Fq ). For any A ∈ I, there are unique elements Qp , . . . , Qq−1 ∈
O, Qq ∈ D such that A = Qp Fp + · · · + Qq−1 Fq−1 + Qq Fq , and so M is
generated as O-module by {Fp+1 , . . . , Fq−1 }. But part (a) implies that M
is a torsion O-module, and so, from proposition 7.2, we deduce that M = 0.
Let us note that the ring D is the inductive limit lim D(DR ).
R→0
Example 7.2. Let us see some examples of left D-modules:

(1) O is a left D-module, since D is a subring of EndC (O) and then any
P ∈ D acts on any a ∈ O by P a = P (a).
(2) To any linear differential operator P ∈ D we associate the left D-module
D/DP .
(3) The field K of fractions of O is a left D-module.
(4) The formal power series ring O b = C[[z]] is a left D-module. In fact the
action of any P ∈ D on O is continuous for the m = (z)-adic topology.
(5) Since each A0R is a left D(DR )-module, A0 := lim A0R is a left D-
R→0
∼
module, and the monodromy operator T : A0 − → A0 is D-linear.
(6) A := lim AR is a left sub-D-module of A0 . The elements in A can be
R→0
written as finite sums
X
φα,k z α (Log z)k
α,k
where the φα,k are germs at 0 of holomorphic functions with a possibly

essential singularity at 0, i.e.
∗
φα,k ∈ lim O(DR ).
R→0
www.pdfgrip.com
30
∼
(7) Prove that T : A −→ A induces an automorphism on A/O. Prove also
that for any λ ∈ C, the map T − λ : A → A is surjective (see exercise 4.3).
(8) N := lim NR is a left sub-D-module of A. The elements in N can be
R→0
written as finite sums
X
φα,k z α (Log z)k
α,k
where the φα,k ∈ K.
Let us denote by Mod(D) the abelian category of left D-modules.
Exercise 7.8. (1) Prove that the D-linear map P ∈ D 7→ P (1) ∈ O is

surjective and its kernel is the left ideal generated by ∂. In particular O '
D/D∂.
(2) Prove that the D-linear map P ∈ D 7→ P (z −1 ) ∈ K is surjective and
its kernel is generated by z∂ + 1. In particular K ' D/D(z∂ + 1).
(3) Prove that the D-linear map P ∈ D 7→ P (z −1 ) ∈ K/O is surjective and
its kernel is generated by z. In particular K/O ' D/Dz.
(4) Let a ∈ O be any non-zero element. Prove that O = Da and compute a
Gröbner basis of the left ideal annD a.
Definition 7.7. Let us denote by M0 , M the left D-modules
M0 = A0 /O, M = A/O.
The following proposition is a straightforward consequence of Cauchy the-

orem and Komatsu-Malgrange index theorem.
Proposition 7.4. For any non-zero P ∈ D, the following properties hold:
(1) ker(P : A0 → A0 ) = ker(P : A → A) and dimC ker(P : A0 → A0 ) =

ord(P ).
(2) The maps P : A0 → A0 and P : A → A are surjective.
(3) The maps P : M0 → M0 and P : M → M are surjective.
(4) ker(P : M0 → M0 ) = ker(P : M → M) and dimC ker(P : M0 →
M0 ) = ν(P ).
Proof. Properties (1) and (2) are a simple translation of proposition 4.3.
Property (3) is a consequence of property (2). For property (4), let us
www.pdfgrip.com
31
consider the following commutative diagram:

0 −−−−→ O −−−−→ A0 −−−−→ M0 −−−−→ 0
  
  
Py Py Py
0 −−−−→ O −−−−→ A0 −−−−→ M0 −−−−→ 0.

From theorem 6.1 we know that χ(P : O → O) = ord(P ) − ν(P ), and from
(2) and (3) we deduce that dimC ker(P : M0 → M0 ) = · · · = ord(P ) −
(ord(P )−ν(P )) = ν(P ). A similar argument works for M instead of M0 .
For a left ideal I ⊂ D, let us denote E(I) = {f ∈ A|P f = 0, ∀P ∈ I}

and F (I) = {g ∈ M|P g = 0, ∀P ∈ I}. The following proposition is taken
from prop. 6 in,2 and gives a very precise information about the spaces of
solutions E(I) and F (I).
Proposition 7.5. Let I ⊂ D be a non-zero left ideal and Fp , . . . , Fq a
Gröbner basis of I. Then the following properties hold:
(1) E(I) = ker(Fp : A → A)(= E(DFp )).
(2) F (I) = ker(Fq : M → M)(= E(DFq )).
(3) dimC E(I) = p(I)(= p = ord(Fp )), dimC F (I) = ν(I)(= ν(Fq )).
(4) P ∈ I ⇔ P f = 0, ∀f ∈ E(I) and P g = 0, ∀g ∈ F (I).
Proof. Property (1) is a consequence of proposition 7.3, (1) and the fact
that A has no O-torsion.
For property (2), we only need to prove that any g ∈ M annihilated by
Fq is annihilated by Fp , . . . , Fq . We can assume that our Gröbner basis is
normalized. Then, the definition of the syzygies Rd (see proposition 7.1)
can be written in the following compact form:
     
∂Fp Fp 0
∂Fp+1  Fp+1   0 
     
 ..   ..   .. 
 .  = A . + .  (7)
     
∂Fq−2  Fq−2   0 
∂Fq−1 Fq−1 z αq−1 −αq Fq
with
 
Qp+1
p z αp −αp+1 0 ··· 0 0
Qp+2 Qp+2 z αp+1 −αp+2 ··· 0 0 
 p p+1 
 . .. .. .. .. .. 
A=
 .. . . . . .
,

 q−1 αq−2 −αq−1 
Q p Qq−1
p+1 Q q−1
p+2 · · · Qq−1
q−2 z 
Qqp Qqp+1 Qqp+2 · · · Qqq−2 Qqq−1
www.pdfgrip.com
32
which is a matrix with entries in O. If g = a ∈ ker(Fq : M → M), a ∈ A,

then Fq (a) = b ∈ O and so, by evaluating the equation (7) at a we obtain
     
Fp (a) Fp (a) 0
Fp+1 (a) Fp+1 (a)  0 
d 
 ..



 .
 
  .


 .  = A  .
.  +  .
. 
dz      
Fq−2 (a) Fq−2 (a)  0 
αq−1 −αq
Fq−1 (a) Fq−1 (a) z b
d
and dz (Fi (a)) ∈ O for i = p, . . . , q − 1. By Cauchy’s theorem we deduce
that Fi (a) ∈ O for i = p, . . . , q − 1 and so a ∈ F (I).
Property (3) is a consequence of (2) and proposition 7.4.
For the last property, let us call J ⊂ D the left ideal {P ∈ D | P f = 0, ∀f ∈
E(I), P g = 0, ∀g ∈ F (I)}. It is clear that I ⊂ J. Let A be any element in
J. By division, there are unique Q, T, S ∈ D such that A = QFq + T + S
with
ord(A) v−1
X X
T = rlk xl ∂ k , ord(S) < q = ord(Fq )
k=q l=0
and v = ν(I) = ν(Fq ). So, R = T + S ∈ J and E(I) ⊂ E(DR), F (I) ⊂

F (DR. In particular, by property (3) applied to the ideal DR, we have
ord(R) ≥ p and ν(R) ≥ v and so T = 0. Consequently the classes ∂ l ,
0 ≤ l ≤ q − 1, form a (finite) system of generators of the O-module J/I. On
the other hand, for any A ∈ J there are Q, U ∈ D and an integer r ≥ 0 such
that xr A = QFp + U and ord(U ) < ord(Fp ) = p (see corollary 7.1). We
deduce that U ∈ J and E(I) ⊂ E(DU ). Property (3) again shows that, if
U 6= 0, ord(U ) ≥ dimC E(I) = p. So, U = 0 and J/I is a torsion O-module.
To conclude we apply proposition 7.2.
Remark 7.1. Proposition 7.5 remains true if we replace A and M by A0

and A0 respectively.
Corollary 7.2. Let I ⊂ I 0 ⊂ D be non-zero left ideals. The following

properties are equivalent:
(a) I = I 0.
(b) E(I) = E(I 0 ) and F (I) = F (I 0 ).
(b) p(I) = p(I 0 ) and ν(I) = ν(I 0 ).
(c) p(I) + ν(I) = p(I 0 ) + ν(I 0 ).
www.pdfgrip.com
33
Proof. The equivalence (a) ⇔ (b) comes from property (4) in proposition
7.5. The equivalence (b) ⇔ (c) ⇔ (d) comes from property (3) in proposition
7.5 and the obvious inclusions E(I 0 ) ⊂ E(I), F (I 0 ) ⊂ F (I).
Corollary 7.3. For any non-zero left ideal I ⊂ D, we have lg (D/I) ≤

p(I) + ν(I), and in particular the left D-module D/I is of finite length.
Exercise 7.9. Prove that for any non integer complex number α, the left
D-module D/D(z∂ − α) is simple, i.e. the left ideal D(z∂ − α) is maximal.
Corollary 7.4. For any non-zero left ideal I ⊂ D, the left D-module D/I
is a torsion module.
Proof. Let us take A ∈ D, A ∈ / I, and consider the D-linear map Φ : P ∈

D 7→ Φ(P ) = P A ∈ D/I. Since Dz ⊃ Dz 2 ⊃ Dz 3 ⊃ · · · is an infinite
strictly decreasing sequence of left ideals in D, we have lg(D) = +∞ and
the map Φ cannot be injective. So, there is a P ∈ D, P 6= 0, such that
P A = 0.
8. Generalized Solutions
If we start from a linear differential equation as (1), we may be interested
in searching its solutions, not only holomorphic functions, but possibly
distributions, hyperfunctions, etc.
In order to make sense the sentence “y is a solution” of (1) what we need
is that y is an element of certain space S, g is also an element of the same
space S, and it makes sense the action of any linear differential operator on
elements of S. Algebraically that corresponds to the fact that S is a (left)
D-module.
The solutions of the homogeneous equation associated with (1) in the
space S can be expressed simply as
ker(P : S → S) = {y ∈ S | P y = 0}
n
d d
where P = an dz n + · · · + a1 dz + a0 . But it is clear that
y ∈ ker(P : S → S) 7→ [Q ∈ D/DP 7→ Qy ∈ S] ∈ HomD (D/DP, S) (8)

is an isomorphism of vector spaces, and then the solutions of the homoge-
neous equation can be expressed in some way in terms of the D-module
D/DP . On the other hand, the fact that the equation (1) has solutions for
any g ∈ S exactly means that im(P : S → S) = S, i.e. that P : S → S
is surjective. Algebraically, the obstruction to this surjectivity is measured
by the cokernel coker(P : S → S) = S/ im(P : S → S).
www.pdfgrip.com
34
If instead of having one linear differential equation, we have a system

P11 y1 · · · + P1r yr
+ = g1
.. .... .. .. .. .. (9)
. . . . . . .
Pm1 y1 + · · · + Pmr yr = gm
we can consider the left sub-D-module I ⊂ Dr generated by P i =
(Pi1 , . . . , Pir ), i = 1, . . . , m, and we have as above an isomorphism (for
the solutions of the associated homogeneous system)
ker(P : S r → S m ) ' HomD (Dr /I, S),
where P is the matrix of linear differential operators (Pij ).
But if we are interested in the non-homogeneous system (9), it is not
reasonable to try to solve it for any choice of g1 , . . . , gm , since the existence
of a solution would imply that any time we have a syzygy Q1 P 1 + · · · +
Qm P m = 0, with Qi ∈ D, then Q1 g1 + · · · + Qm gm = 0. So, to measure the
obstruction to solve (9), we have to look not at coker(P : S r → S m ), but
at
X
{(g1 , . . . , gm ) ∈ S m | Qi gi = 0, ∀Q ∈ S(P)}/ im P, (10)
i
where
X
S(P) = {Q ∈ Dm | Qi P i = 0}.
i
In fact, due to the noetherianity of D, S(P) is a finitely generated sub-

D-module of Dm and then the apparently infinite number of conditions
X
Qi gi = 0, ∀Q ∈ S(P) (11)
i
reduce to a finite number of them.

The question now is if it is possible to get an isomorphism of type (8)
for coker(P : S → S), in the one equation case, or for (10) in the general
case of a system of several equations with several unknowns.
The answer is YES and is given by Homological Algebra:
X
{g ∈ S m | Qi gi = 0, ∀Q ∈ S(P)}/ im P ' Ext1D (Dr /I, S), (12)
i
where the ExtiD (M, N )

are complex vector spaces conveniently defined for
any left D-modules M, N and i ≥ 0. For i = 0 we have
Ext0D (M, N ) = HomD (M, N )
www.pdfgrip.com
35
(see the book28 ). In fact, the ExtiD (M, N ) appear as the cohomology of
degree i of a certain complex of vector spaces
R HomD (M, N )
which can be calculated by taking a projective resolution of M or an injec-

tive resolution of N (see for instance ch. II in22 for a quick introduction to
this subject and for a list of references).
Example 8.1. Let us see some examples.

(1) If M = D, then it corresponds to the linear differential equation 0y = g.
Any element y ∈ N is obviously a solution of the homogeneous equation,
and the the compatibility conditions (11) mean that the g must be zero and
the non-homogeneous equation must be actually homogenoeus, and then it
always has solutions (the zero solution). In this case, since D is free as left
D-module, we have
0
R HomD (D, N ) = HomD (D, N ) = · · · → 0 → N → 0 → · · ·
and Ext0D (D, N ) = HomD (D, N ) = N and ExtiD (D, N ) = 0 for i 6= 0.

(2) If M = D/DP , to describe R HomD (M, N ) we take the free resolution
of M
−1 ·P 0
0 → D −→ D → M = D/DP → 0,
−1 ·P 0
M • = · · · → 0 → D −→ D → 0 → · · · ,
and
R HomD (M, N ) = HomD (M • , N ) =

0 HomD (·P,N ) 1
· · · → 0 → HomD (D, N ) −−−−−−−−→ HomD (D, N ) → 0 → · · · =
0 P 1
··· → 0 → N −
→ N → 0 → ··· .
In particular,
HomD (M, N ) = Ext0D (M, N ) = h0 R HomD (M, N ) = ker(P : N → N ),
Ext1D (M, N ) = h1 R HomD (M, N ) = coker(P : N → N )
and ExtiD (M, N ) = 0 for all i 6= 0, 1.

www.pdfgrip.com
36
(3) If N is an injective (see for instance the book28 ) D-module, we can solve
any compatible system with unknowns in N and
R HomD (M, N ) = HomD (M, N ),
i.e. ExtiD (M, N ) = 0 for all i 6= 0.
Definition 8.1. If M is a finitely generated left D-module, we define its

higher holomorphic solutions as the complex of vector spaces
Sol M = R HomD (M, O).
The proof of the following proposition is an interesting application of

the division tools in the ring D and gives a “natural” injective resolution
of the left D-module O.
Proposition 8.1. The following exact sequence of left D-modules
0→O→A→M→0
is an injective resolution of O as a left D-module.
Proof. To prove that A is an injective D-module, we have to check that

for any left ideal I ⊂ D and any D-linear map ϕ : I → A there exists a
D-linear map ϕ e : D → A such that ϕ|e I = ϕ (see any book of Homological
Algebra, for instance28 ). Let us take a Gröbner basis Fp , . . . , Fq of I. We
know from the proposition 7.3, (2) that I = D(Fp , Fq ). Let us write ϕ(Fp ) =
fp , ϕ(Fq ) = fq . Finding ϕ
e is the same as finding f = ϕ(1),
e since ϕ(P
e )=
P ϕ(1)
e for all P ∈ D. On the other hand, the condition ϕ| e I = ϕ exactly
means that Fp f = fp , Fq f = fq .
From proposition 7.3, (1) there exists an integer r ≥ 0 and an operator
Q ∈ D such that xr Fq = QFp , and so xr fq = Qfp . From proposition 7.4,
(2) there exists f ∈ A such that Fp f = fp . We have xr Fq f = QFp f =
Qfp = xr fq , and since A has no O-torsion we deduce that Fq f = fq .
Let us now prove the injectivity of M. Assume that I ⊂ D is a left

ideal and ψ : I → M is a D-linear map. Take a normalized Gröbner basis
Fp , . . . , Fq of I and let us write ψ(Fd ) = gd = fd , d = p, . . . , q. From
proposition 7.4, there exists f ∈ A such that Fq f = fq (and so Fq g = gq
for g = f ∈ M). The generating system of the syzygies of Fp , . . . , Fq 7.1
www.pdfgrip.com
37
gives rise to the relation (see (7) in the proof of proposition 7.5)
     
∂Fp Fp 0
∂Fp+1  Fp+1   0 
     
 ..   ..   .
.. 
 .  = A . + 
     
∂Fq−2  Fq−2   0 
∂Fq−1 Fq−1 z αq−1 −αq Fq
with A a matrix with entries in O. By applying ψ we find
     
∂gp gp 0
∂gp+1  gp+1   0 
     
 ..   ..   .
.. 
 .  = A . + 
     
∂gq−2  gq−2   0 
∂gq−1 gq−1 z αq−1 −αq gq
or
       
fp fp 0 hp
fp+1  fp+1   0  hp+1 
d 
 .. 
 
 ..  
 
.
 
  .. 

 .  = A . + ..  +  . ,
dz        
fq−2  fq−2   0  hq−2 
fq−1 fq−1 z αq−1 −αq fq hq−1
where hd ∈ O for d = p, . . . , q − 1, and
     
fp − F p f fp − F p f hp
fp+1 − Fp+1 f  fp+1 − Fp+1 f  hp+1 
d  ..



 ..
 
  .. 

 .  = A  .  +  .
dz      . 
fq−2 − Fq−2 f  fq−2 − Fq−2 f  hq−2 
fq−1 − Fq−1 f fq−1 − Fq−1 f hq−1
By Cauchy theorem we deduce that fd − Fd f ∈ O for d = p, . . . , q − 1
and so Fd g = gd for d = p, . . . , q − 1. The extension of ψ is given by
ψe : P ∈ D 7→ P g ∈ M.
Example 8.2. We can use the injective resolution of proposition 8.1 to

compute the higher holomorphic solutions of any left D-module M :
0 1
R HomD (M, O) = · · · → 0 → HomD (M, A) → HomD (M, M) → 0 → · · · .
Exercise 8.1. By taking the free resolutions in exercise 7.8 and the injective
resolution of O given in proposition 8.1, compute in two different ways
Sol M for: (1) M = O. (2) M = K. (3) M = K/O.
www.pdfgrip.com
38
Exercise 8.2. Let P = ∂ n + an−1 ∂ n−1 + · · · + a1 ∂ + a0 ∈ D and let us

consider the left D-module M = D/DP associated with the (germ of)
linear differential equation P y = g, where g ∈ S. Let us also consider the
system of (germs of) linear differential equations (see (3))
 
 
y1 0
  0 
y2
   
   .. 
..
P  = .
.
   
yn−1  0
yn g
with
 
−1 0 · · · 0
∂ 0 0
 ∂ −1 · · · 0
0 0 0 
 
 ...
.. .
.. ... .
.. .
.. .
.. 
P= . 
 
 0 0 0 ··· ∂ −1 0 
−a0 −a1 −a2 · · · −an−3 −an−2 ∂ − an−1
and the associated D-module M 0 = Dn /I, where I is the left submodule of

Dn generated by
P1 = (∂, −1, 0, . . . , 0, 0, 0)
2
P = (0, ∂, −1, . . . , 0, 0, 0)
.. .. ..
. . .
n−1
P = (0, 0, 0, . . . , ∂, −1, 0)
P n = (−a0 , −a1 , −a2 , . . . , −an−3 , −an−2 , ∂ − an−1 ).
Prove that the map
n−1
X
(Q0 , . . . , Qn−1 ) ∈ M 0 = Dn /I 7→ Qi ∂ i ∈ M = D/DP
i=0
is an isomorphism of left D-modules, and so
R HomD (M, S) ' R HomD (M 0 , S).
In the above exercise, the isomorphism M ' M 0 is the algebraic coun-

terpart of the classic reduction of an order n linear differential equation to
an order 1 system of linear differential equations described in section 1.
www.pdfgrip.com
39
9. Holonomic D-Modules
In this section, all D-modules considered will be left D-modules.
Definition 9.1. We sayf that a D-module M is holonomic if it is finitely

generated and a torsion module, i.e. for all m ∈ M there is P ∈ D, P 6= 0,
such that P m = 0.
It is clear that any submodule and any quotient of a holonomic D-

module is also holonomic, and that the direct sum of two holonomic D-
modules is again holonomic. In particular the category of holonomic D-
modules is abelian.
Let us denote by Hol(D) the (abelian) category of holonomic (left) D-
modules.
Example 9.1. Any D-module of type D/I, where I ⊂ D is a non-zero

ideal, is holonomic after corollary 7.4.
In fact we have the following result.
Proposition 9.1. Let M be a D-module. The following properties are

equivalent:
(a) M is holonomic.
(b) M is of finite length.
(c) There is a non-zero ideal I ⊂ D such that M ' D/I.
Proof. For (a) ⇒ (b) we proceed by induction on the number of generators

of M . If M = Dm1 is cyclic, then I = annD (m1 ) 6= 0 and M ' D/I is of
finite length by corollary 7.3.
Assume that any holonomic D-module generated by n − 1 elements
is of finite length and take a holonomic D-module M = D(m1 , . . . , mn )
generated by n elements. By induction hypothesis M 0 = D(m2 , . . . , mn )
and M 00 = M/M 0 = Dm1 are of finite length, and so M is also of finite
length.
The implication (b) ⇒ (c) follows from from a general result, which
assures that any left module of finite length over a simple ring R of infinite
length as left R-module is cyclic (cf. 5.7.3 in18 ).
f Holonomic D-modules make sense in several variables, but their definition needs to work
with filtrations (see the notes3 ). The present definition only works in one variable.
www.pdfgrip.com
40
The implication (c) ⇒ (a) is a consequence of corollary 7.4.
Theorem 9.1. Let M be a holonomic D-module. Then Sol M is a complex

of vector space with finite dimensional cohomology. More precisely:
dimC h0 Sol M = dimC HomD (M, O) < +∞,
dimC h1 Sol M = dimC Ext1D (M, O) < +∞, hi Sol M = 0 ∀i 6= 0, 1.
Proof. From proposition 9.1, we know that M ' D/I, where I ⊂ D is

a non-zero left ideal. On the other hand, Sol M can be computed as (see
example 8.2)
0 1
· · · → 0 → HomD (M, A) → HomD (M, M) → 0 → · · · ,
but HomD (M, A) ' HomD (D/I, A) ' E(I) and F (I) ' HomD (M, M) '
HomD (D/I, M). So, the theorem is a consequence of proposition 7.5.
Remark 9.1. For a holonomic D-module it is relatively easy to give a

formula for
χ(R HomD (M, O)) = dimC HomD (M, O) − dimC Ext1D (M, O)
in terms of two integers algebraically associated with M : the multiplicity
e0 of the “null section” and the multiplicity e1 of the “conormal of 0 in the
“characteristic variety” defined by means of filtrations and the theory of
Hilbert polynomials (cf. ch. V in6 ). When M = D/I, then e0 = p(I) and
e1 = ν(I)
χ(R HomD (D/I, O)) = dimC HomD (D/I, A) − dimC HomD /D/I, M) =
dimC E(I) − dimC F (I) = p(I) − ν(I).
10. Regular D-Modules

In this section, all D-modules considered will be left D-modules.
Definition 10.1. Let

n
X ∞
n X
X
k
P = ak ∂ = alk xl ∂ k
k=0 k=0 l=0
be a non-zero linear differential operator (of O) of order n ≥ 0.

(1) We say that P is regular if it satisfies property (b) of theorem 5.1, i.e.
max{k − ν 0 (ak ) | k = 0, . . . , n} = n − ν 0 (an ).
www.pdfgrip.com
41
(2) We define the weight of P as

w(P ) = max{k − l | (l, k) ∈ supp(P )} = max{k − ν 0 (ak ) | k = 0, . . . , n}.
(3) The initial form of P is the operator
X
in(P ) = alk xl ∂ k .
k−l=w(P )
Let us note that P is regular if and only if w(P ) = ord(P ) − ν(P ).
Exercise 10.1. Prove that, if P1 , P2 ∈ D are non-zero linear differential

operators, then:
(a) w(P1 P2 ) = w(P1 ) + w(P2 ).
(b) in(P1 P2 ) = in(P1 ) in(P2 ).
(c) Prove that P1 P2 is regular if and only if P1 and P2 are regular.
n
d
Theorem 5.1 can be rephrased in the following way: Let L = an dz n +
d
· · · + a1 dz + a0 be a linear differential operator of order n on an open disc
D = DR and let P = L0 ∈ D be its stalk at the origin. The following
properties are equivalent:
(a) 0 is a regular singular point of L.
(b) P is regular.
Theorem 10.1. Let I ⊂ D be a non-zero left D-ideal. The following prop-

erties are equivalent:
(a) There is a regular element P ∈ I, P 6= 0.
(b) All the elements of a Gröbner basis of I are regular.
(c) E(I) ⊂ N .
(d) F (I) ⊂ N /O.
(e) b
{η ∈ O/O | P η = 0, ∀P ∈ I} = 0.
Proof. (See II.3.1 in13 ) The equivalence of the first three properties comes
from proposition 7.3, (1), exercise 10.1, proposition 7.5, (1) and theorem
5.1.
Let {Fp , . . . , Fq } be a Gröbner basis of I. We know from proposition
7.5, (2) that F (I) = F (DFq ).
(b) ⇒ (d): Let g = f ∈ M be a class in F (I), i.e. Fq f ∈ O. We can find a
non singular operator P ∈ D (ν(P ) = 0) such that P Fq f = 0, and so, by
theorem 5.1 f ∈ N and g ∈ N /O.
www.pdfgrip.com
42
(d) ⇒ (a): If f ∈ A is annihilated by Fq , then f ∈ F (I) and so f ∈ N .

Hence Fq is regular by theorem 5.1.
b
Since O/O has no O-torsion, we can follow the proof of proposition 7.5, (1)
to prove that

b
{η ∈ O/O | P η = 0, ∀P ∈ I} = ker F b
fp : O/O b
→ O/O .
So, the equivalence (e) ⇔ (a) is a consequence of corollary 6.1.
Remark 10.1. Let us note that for any holonomic D-module M , the vec-
b
tor space HomD (M, O/O) is finite dimensional. For that, it is enough to
consider the case where M = D/I with I ⊂ D a non-zero left ideal. In such
a case we have an isomorphism
b
HomD (D/I, O/O) b
' {η ∈ O/O | P η = 0, ∀P ∈ I},
but the last space is finite dimensional by theorem 6.2, (1).
Definition 10.2. (1) Let M be a holonomic D-module. We define its ir-

b
regularity as the number irr M := dimC HomD (M, O/O) ≥ 0.
b
(2) We say that a holonomic D-module M is regular if HomD (D/I, O/O) =
0, or equivalently, if irr M = 0.
b
Proposition 10.1. The D-module O/O is injective.
Proof. The proof follows the same lines as the proof of the injectivity of
b
A in proposition 8.1, since O/O has no O-torsion either and we can use
theorem 6.2, (2) instead of proposition 7.4, (2).
The following theorem is a straightforward consequence of theorem 10.1

and proposition 10.1.
Theorem 10.2. Let M be a holonomic left D-module. The following prop-

erties are equivalent:
(a) M is regular.
b
(b) R HomD (M, O/O) = 0.
(c) The map HomD (M, N ) → HomD (M, A) induced by the inclusion N ⊂
A is an isomorphism.
(d) The map HomD (M, N /O) → HomD (M, A/O) induced by the inclusion
N /O ⊂ A/O is an isomorphism.
www.pdfgrip.com
43
The proof of the following proposition is also a straightforward conse-

quence of proposition 10.1.
Proposition 10.2. The irregularity irr is an additive function on exact

sequences of holonomic D-modules.
Corollary 10.1. Given a short exact sequence of holonomic D-modules
− M0 →
0→ − M 00 →
− M→ − 0,
M is regular if and only if M 0 and M 00 are regular. In particular the category

of holonomic D-modules is abelian.
Let us denote by RegHol(D) the (abelian) category of regular holonomic

(left) D-modules.
Remark 10.2. The above results are the precursors of the irregularity
complexes along a hypersurface and the notion of regular holonomic module
in higher dimension (see the papers20,21 ).
Additional results and information about regular an irregular holonomic

D-modules in one variable can be found in the paper.29
11. A Local Version of the Riemann-Hilbert

Correspondence in One Variable (in Collaboration with
F. Gudiel Rodrı́guez)
In this section we explain proposition III.4.5 in13 using the description of
simple objects in the category C instead of the more involved description
of indecomposable objects (see the master thesis7 ).
Definition 11.1. Let us call C0 the category defined in the following way:
(1) The objets of C0 are the diagrams (E, F, u, v) where E, F are complex
vector spaces and u : E → F and v : F → E are linear maps such that
IdE + v ◦ u and IdF + u ◦ v are automorphisms.
(2) If O = (E, F, u, v), O 0 = (E 0 , F 0 , u0 , v 0 ) are objets of C0 , a morphism in
C0 from O to O0 is a pair (a, b) of linear maps a : E → E 0 , b : F → F 0
such that u0 ◦ a = b ◦ u, v 0 ◦ b = a ◦ v.
We also call C the full subcategory of C0 whose objects are those (E, F, u, v)
with dimC E, dimC F < +∞.
www.pdfgrip.com
44
Definition 11.2. The functors E0 : Hol(D) →

− C, E : RegHol(D) →
− C are
defined on objects by
E0 (M ) = (HomD (M, A), HomD (M, M), U∗ , V∗ ),
E(M ) = (HomD (M, N ), HomD (M, N /O), U∗ , V∗ )
and on morphisms in the obvious way.
Proposition 11.1. Functors E0 : Hol(D) →
− C and E : RegHol(D) →
− C
are exact.
Proof. The exactness of E0 follows from proposition 8.1. The exactness of

E comes from the fact that for a regular holonomic D-module, the canonical
inclusion E(M ) ,→ E0 (M ) is an isomorphism (see theorem 10.2).
Exercise 11.1. (1) Prove that C0 is an abelian category. Which are the
monomorphisms and the epimorphisms in C0 ?
(2) Prove that C is an abelian subcategory of C0 . Prove that any object in
C has finite length.
(3) Prove that the simple objects in C are isomorphic to one of the following
types: (i) (C, 0, 0, 0); (ii) (0, C, 0, 0); (iii) (C, C, 1, λ), λ 6= −1, 0.
Definition 11.3. We define an “universal” object U 0 in C0 as U 0 =
(A, M, U, V ) with U : A → M the projection map and V : M → A the
“variation” map defined as V (f ) = T (f ) − f . This object contains another
special object U = (N , N /O, U, V ).
The object U 0 is enriched with a (left) D-module structure, since A
and M are left D-modules and U, V are D-linear. So, for any object O =
(E, F, u, v) in C, the abelian group HomC (O, U 0 ) carries a natural structure
of left D-module given by the following operation: for P ∈ D and (a, b) ∈
HomC (O, U 0 ), P (a, b) is defined as (P a, P b) where
P a : x ∈ E 7→ (P a)(x) := P · a(x) ∈ A,
P b : y ∈ F 7→ (P b)(y) := P · b(y) ∈ M.
In that way we define a contravariant left exact additive functor
F0 = HomC0 (−, U 0 ) : C → Mod(D).
Since U is also enriched with a (left) D-module structure, we also have
another contravariant left exact additive functor
F = HomC0 (−, U) : C → Mod(D)
with F ⊂ F0 .
www.pdfgrip.com
45
Exercise 11.2. Let O be a simple object in C. Prove that:

(1) There is an injection ι : O ,→ U.
(2) For any injection ι : O ,→ U, the left D-module FO is generated by ι.
(3) Let ι : O ,→ U be an injection and (E, F, U, V ) = im ι ⊂ U. Let us
define I = {P ∈ D | P f = 0, ∀f ∈ E, P g = 0, ∀g ∈ F }. Prove that I is a
left maximal ideal of D and E(I) = E, F (I) = F .
(4) FO is a simple regular holonomic left D-module.
(Hint: Proceed following the three different types of simple objects in C)
Proposition 11.2. With the above notations, FO is a regular holonomic

left D-module for any object O of C. Moreover, lg FO ≤ lg O.
Proof. The proof goes easily by induction on the length of the object O.
If O is simple, the result has been treated in exercise 11.2.
Assume that the proposition is true anytime that lg O < n and let
O be an object in C with lg O = n. We can find a short exact sequence
i p
0 → O0 → − O− → O00 → 0 in C with lg O0 = 1 and lg O00 = n − 1. By applying
F we obtain a left exact sequence of left D-modules
Fp Fi
− FO00 −→ FO −→ FO0 .
0→
By induction hypothesis FO 0 and FO00 are regular holonomic with lg FO 0 =

1, lg FO00 ≤ n − 1, and so the image of Fi is also regular holonomic and
we conclude that FO is regular holonomic too (see corollary 10.1) with
lg FO = lg FO00 + lg FO0 ≤ n.
As a consequence of the above proposition, we can consider the con-

travariant left exact additive functor
F = HomC0 (−, U) : C → RegHol(D).
Definition 11.4. For any regular holonomic D-module, we define the map
1 2
ξM : M → FEM = HomC0 (EM, U) by ξM (m) = (ξM (m), ξM (m)) with
1
ξM (m) : φ ∈ HomD (M, N ) 7→ φ(m) ∈ N ,
2
ξM (m) : ψ ∈ HomD (M, N /O) 7→ ψ(m) ∈ N /O.
Proposition 11.3. The correspondence which associates to any regular

holonomic D-module M the map ξM is a morphism of functors ξ : Id → FE.
Moreover, ξM is injective for any regular holonomic D-module M .
www.pdfgrip.com
46
Proof. The first part is clear. For the second part, we can restrict ourselves
to the case M = D/I, where I ⊂ D is a non-zero left ideal. In such a case,
EM is canonically isomorphic to O = (E(I), F (I), U, V ) ⊂ U and the map
ξM can be seen as
P ∈ D/I 7→ (P : E(I) → N , P : F (I) → N /O) ∈ FO.
The injectivity of ξM is so a consequence of proposition 7.5, (4).
Lemma 11.1. For any simple regular holonomic D-module M , EM is a

simple object in C.
Proof. We can assume that M = D/I, with I ⊂ D a maximal left ideal and
so EM is isomorphic to O = (E(I), F (I), U, V ). Let O 0 = (E, F, U, V ) be a
simple sub-object of O and J = {P ∈ D | P f = 0, ∀f ∈ E, P g = 0, ∀g ∈ F }.
We know from exercise 11.2, (3) that E(J) = E, F (J) = F .
It is clear that J is a proper ideal containing I, and so I = J. We
conclude that O = O 0 and O is simple.
Proposition 11.4. For any regular holonomic D-module M the map ξM :

M → FEM is an isomorphism.
Proof. We proceed by induction on the length of M . If M is simple, then

EM is simple by lemma 11.1 and FEM is simple by exercise 11.2, (4). So
the injection ξM : M ,→ FEM is an isomorphism.
Assume that ξM is an isomorphism anytime that lg M < n and let M

be a regular holonomic D-module of length n. Let us consider a short exact
sequence of (regular holonomic left) D-modules 0 → M 0 → M → M 00 → 0.
We have a commutative diagram
0 / M0 /M / M 00 /0
' ξM 0 ξM ' ξM 00

0 / FEM 0 / FEM / FEM 00
and so ξM is an isomorphism.
Definition 11.5. For any object O = (E, F, u, v) in C, we define the

map τO : O → EFO = (HomD (FO, N ), HomD (FO, N /O), U∗ , V∗ ) by
τO = (τO1 , τO2 ) with
www.pdfgrip.com
47
τO1 : x ∈ E 7→ τO1 (x) ∈ HomD (FO, N ),

τO1 (x) : (a, b) ∈ FO = HomC0 (O, U) 7→ a(x) ∈ N ,
τO2 : y ∈ E 7→ τO1 (y) ∈ HomD (FO, N /O),
τO2 (y) : (a, b) ∈ FO = HomC0 (O, U) 7→ b(y) ∈ N /O.
Exercise 11.3. Prove that the correspondence which associates to any object
O of C the map τO is a morphism of functors τ : Id → EF.
Exercise 11.4. (1) Prove that for any objects O ⊂ O 0 in C with O0 /O simple,
the induced map FO0 → FO is surjective (Hint: Proceed following the three
different types of simple objects in C and use exercise 7.2, (7)).
(2) Conclude that the functor F : C → RegHol is exact.
Proposition 11.5. For any object O in C the map τO : O → EFO is an

isomorphism.
Proof. Thanks to the exactness of F (and E), we can restrict ourselves to

the case where O is simple, as in the proof of proposition 11.4.
Assume that O is simple. Let ι : O ,→ U be an injection, (E, F, U, V ) =
im ι ⊂ U, and I = {P ∈ D | P f = 0, ∀f ∈ E, P g = 0, ∀g ∈ F }. It is easy to
see that the map τO can be seen as the inclusion
(E, F, U, V ) ,→ (E(I), F (I), U, V )
and so it is an isomorphism by exercise 11.2, (3).
Proposition 11.4 and 11.5 can be summarized in the following theorem.
Theorem 11.1. The functors
E : RegHol(D) →
− C and F:C→
− RegHol(D)
are quasi-inverse contravariant equivalences of categories.
Remark 11.1. (1) Since A and M are in fact left modules over the ring
D∞ of germs at 0 of infinite order linear differential operators (cf.23,26 ),
we can consider F0 as a functor from C to Mod(D∞ ). One can prove that
A = D∞ ⊗D N , M = D∞ ⊗D N /O and that D∞ ⊗D F ' F0 . Let us
call Hol(D∞ ) the full abelian subcategoryg D∞ ⊗D Hol(D) ⊂ Mod(D∞ ).
g One needs to use that the extension D ,→ D∞ is faithfully flat (cf. loc. cit.).
www.pdfgrip.com
48
The functor E0 can be also extended to E0 : Hol(D∞ ) → C and it is a

quasi-inverse of F0 : C → Hol(D∞ ).
(2) One can prove by elementary methods that the category C is equivalent
to the category of germs at 0 ∈ C of perverse sheaves (cf.4,14,25 ).
(1) and (2) are particular cases of the “full” Riemann-Hilbert correspon-
dence in higher dimension (cf. 3.3 in19 and the paper21 ).
12. D-Modules on a Riemann Surface

In this section, we briefly sketch some basic facts of the theory of D-modules
on a Riemann surface. X will be a connected Riemann surface and OX will
denote its sheaf of holomorphic functions. It has the same properties as OU
in definition 2.1.
We also define the notion of subsheaf of OX as in definition 2.2, and the
notion of endomorphism of OX as in definition 2.3.
We have a “generalized sheaf” in the sense that it is not a sheaf of
functions, but a sheaf of rings, given in the following way: for any open set
U ⊂ X we define
Hom C (OX , OX )(U ) := HomC (OU , OU ).
The data Hom C (OX , OX ) satisfies the formal properties of the sheaves of
holomorphic functions (see exercise 2.4). The reader can refer to the book 9
for the general notion of sheaf.
We have an injective morphism of sheaves of rings OX ,→
Hom C (OX , OX ).
To define the sheaf of (holomorphic linear) differential operators, we
have to adapt definition 2.4, because on a Riemann surface we do not have
global coordinates.
Definition 12.1. Let U ⊂ X an open set. A linear differential operator on
U is an endomorphism L : OU → OU which locally, on open sets Ui with
local coordinate zi there are holomorphic functions a0 , . . . , an on Ui (n may
depends on i) such that
dn
L|Ui = an + · · · + a0 .
dzin
The set of linear differential operators on U will be denoted by DX (U ).
Exercise 12.1. Prove that the data DX is a subsheaf of non-commutative
rings of Hom C (OX , OX ).
www.pdfgrip.com
49
The stalk of DX at a point p ∈ X, DX,p is isomorphic to the ring D (by

taking a local coordinate around p).
We have the filtration by the order at the level of sheaves F k DX , k ≥ 0.

The graded sheaf grF DX is locally isomorphic to the sheaf of commutative
rings OX [ξ]. The the sheaf DX has an important property: it is a coherent
sheaf of rings (cf. the paper6 ).
Definition 12.2. A left holonomic DX -module M is a left coherent DX -

module such that Mp is a holonomic DX,p -module for each p ∈ X.
Alternatively, holonomicity can be defined by using local good filtrations

at the sheaf level. In that way we define the characteristic variety Ch M
which is an analytic conical closed subset of the cotangent space T ∗ X, and a
coherent left DX -module is holonomic if and only if dim Ch M = dim X = 1.
We can also define, for any left coherente DX -modules M, N, the sheaf of
complex vector spaces Hom DX (M, N). We also define
Sol(M) = R Hom DX (M, OX ),
in such a way that Sol(M)p = Sol(Mp ) for each point p ∈ X.
Theorem 12.1. Let M be a (left) holonomic DX -module. Then Sol(M ) is

a perverse sheaf, i.e
(1) hi Sol(M ) = ExtiDX (M, OX ) = 0 for all i 6= 0, 1.
(2) There is a closed discrete set Σ ⊂ X (the singular locus of M) such
that:
a) h0 Sol(M )|X\Σ = Hom DX (M, OX )|X\Σ is a locally constant sheaf of
finite rank.
b) h1 Sol(M )|X\Σ = Ext1DX (M, OX )|X\Σ = 0.
c) hi Sol(M)p = hi Sol(Mp ) are finite dimensional spaces for i = 0, 1
and for each p ∈ Σ.
d) h0 Sol(M ) has no section supported by Σ (this is clear because locally
we have h0 Sol(M ) ⊂ OX , and there are no holomorphic functions supported
by a discrete set).
The proof of the above theorem is a direct consequence of theorems 3.1

and 9.1. More details can be found, for instance, in the paper,24 where it
is given an elementary proof of the Riemann-Hilbert correspondence on a
Riemann surface.
www.pdfgrip.com
50
References
1. M.F. Atiyah and I.G. MacDonald. Introduction to commutative algebra.
Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.,
1969.
2. J. Briançon and Ph. Maisonobe. Idéaux de germes d’opérateurs différentiels
à une variable. Enseign. Math., 30 (1984), 7–38.
3. F. J. Castro Jiménez. Modules over the Weyl Algebra. This volume.
4. P. Deligne. Lettre à R. MacPherson. 1981.
5. E. Goursat. Cours d’analyse mathématique, Tome II. Gauthier-Villars, Paris,
1933.
6. J.M. Granger and Ph. Maisonobe. A basic course on differential modules,
in16 vol. I, 103–168.
7. F. Gudiel Rodrı́guez. Sobre una variante local del problema de Riemann-
Hilbert en una variable. Tesis de Licenciatura, Facultad de Matemáticas,
Universidad de Sevilla, October 1991.
8. O. Ince. Ordinary Differential Equations. Dover, New York, 1950.
9. B. Iversen. Cohomology of Sheaves. Universitext. Springer Verlag, Heidelberg,
1986.
10. N.M. Katz. Nilpotent connections and the monodromy theorem: Applications
of a result of Turritin. Inst. Hautes Études Sci. Publ. Math., 39 (1970), 175–
232.
11. H. Komatsu. On the index of ordinary differential operators. J. Fac. Sci.
Univ. Tokyo Sect. IA Math., 18 (1971), 379–398.
12. Lê Dũng Tráng (editor). Introduction à la théorie algébrique des systèmes
différentiels. Travaux en cours, 34. Hermann, Paris, 1988.
13. Ph. Maisonobe. Germes de D-modules à une variable et leurs solutions, in 12
97–134.
14. Ph. Maisonobe. Faisceaux pervers sur C relativement à {0} et couple E → ←F ,
in12 135–146.
15. Ph. Maisonobe and L. Narváez Macarro (editors). Éléments de la théorie des
systèmes différentiels géométriques. Cours du CIMPA, École d’été de Séville
(1996). Séminaires et Congrès, vol. 8. Soc. Math. France, Paris, 2004.
16. Ph. Maisonobe and C. Sabbah (editors). Eléments de la théorie des systèmes
différentiels (vol. I, II). Summer school at CIMPA, Nice, 1990. Travaux en
cours, 45, 46. Hermann, Paris, 1993.
17. B. Malgrange. Sur les points singuliers des équations différentielles. L’Enseig.
Math., XX (1974), 147–176.
18. J.C. McConnell and J.C. Robson. Noncommutative Noetherian Rings. John
Wiley & Sons, Chichester, 1987.
19. Z. Mebkhout. Une autre équivalence de catégories. Compositio Math., 51
(1984), 63–88.
20. Z. Mebkhout. Le théorème de comparaison entre cohomologies de De Rham
d’une variété algébrique complexe et le théorème d’existence de Riemann.
Inst. Hautes Études Sci. Publ. Math., 69 (1989), 47–89.
21. Z. Mebkhout. Le théorème de positivité, le théorème de comparaison et le
théorème d’existence de Riemann, in15 165–308.
www.pdfgrip.com
51
22. Z. Mebkhout and L. Narváez-Macarro. Le théorème de constructibilité de

Kashiwara, in16 vol. II, 47–98.
23. Z. Mebkhout and L. Narváez Macarro. Le théorème de continuité de la divi-
sion dans les anneaux d’opérateurs différentiels. J. Reine Angew. Math., 503
(1998), 193–236.
24. L. Narváez Macarro. Systèmes différentiels linéaires sur une surface de Rie-
man, in12 50–96.
25. L. Narváez-Macarro. Cycles évanescents et faisceaux pervers. II. Cas des
courbes planes réductibles. In J.P. Brasselet, editor, Singularities (Lille,
1991), volume 201 of London Math. Soc. Lecture Note Ser., pages 285–323.
Cambridge Univ. Press, Cambridge, 1994.
26. L. Narváez Macarro and A. Rojas León. Continuous division of linear differ-
ential operators and faithful flatness of D∞X over DX , in
15
129–148.
27. F. Pham. Singularités des systèmes différentiels de Gauss-Manin, volume 2
of Progress in Math. Birkhäuser Boston, Mass., 1979. With contributions by
Lo Kam Chan, Philippe Maisonobe and Jean-Étienne Rombaldi.
28. J.J. Rotman. Notes on Homological Algebra. Math. Studies. Van Nostrand
Reinhold Co., N.Y., 1970.
29. C. Sabbah. Introduction to algebraic theory of linear systems of differential
equations, in16 vol. I, 1–80.
February 4, 2010 18:13 WSPC - Proceedings Trim Size: 9in x 6in 02˙jimenez
www.pdfgrip.com
52
MODULES OVER THE WEYL ALGEBRA
´
FRANCISCO J. CASTRO JIMENEZ
Department of Algebra
University of Seville
Campus Reina Mercedes
41012 Seville (Spain)
E-mail: [email protected]
We develop some basic results on modules over the Weyl algebra including the
existence of the Bernstein polynomial and the relationship between logarithmic
modules and the so-called Logarithmic Comparison Theorem. We use computer
algebra system Macaulay 2 to explicitly compute some invariants arising in the
whole subject.
Keywords: Weyl algebra; Linear Differential Operator; Characteristic variety;

Holonomic module; Bernstein-Sato polynomial; Logarithmic derivation; Loga-
rithmic differential form; Logarithmic An –module.
Introduction
These notes are an enlarged version of the lecture notes given at the School
on Algebraic Approach to Differential Equations that took place at Biblio-
theca Alexandrina (Alexandria, Egypt) from 12th to 24th November 2007. a
The school was organized by the Mathematics Section of the ICTP (The
Abdus Salam International Centre for Theoretical Physics, Trieste).
The content of these notes is the following. Section 1 is devoted to the
definition of the complex Weyl algebra, the ring of linear differential oper-
ators with polynomial coefficients, and the study of different filtrations on
the Weyl algebra and on left An –modules. The associated graded structures
will be used in Section 2 to define the characteristic variety, the dimension
and the multiplicity of a finitely generated An –module. In this section we
prove the Bernstein’s inequality and study the class of holonomic modules.
a Part of this material has been previously taught in the Aachen Summer School 2007
Algorithmic D-module theory that took place from 3rd to 7th September 2007, at the
Söllerhaus in Kleinwalsertal (Austria).
www.pdfgrip.com
53
One of the deepest result of this section, due to J. Bernstein, states that
the An –module of rational functions, with poles along a hypersurface in
Cn , is holonomic. Following J. Bernstein we also introduce the Bernstein
polynomial (also known as Bernstein-Sato polynomial) associated with a
given polynomial f ∈ C[x] := C[x1 , . . . , xn ].
In Section 3 we consider the class of logarithmic An –modules. These
modules are associated with a polynomial f ∈ C[x] and are defined by using
differential operators of order 1 arising from the logarithmic derivations
with respect to f .
Some parts of the lectures were devoted to algorithmic questions and
explicit computations in the theory of finitely generated modules over the
Weyl algebra. These explicit computations are possible because the theory
of Groebner bases can be extended from the polynomial ring to the Weyl
algebra. We have added in an Appendix some basic results on the Division
Theorem for differential operators and the theory of Groebner bases in An
and other rings of differential operators. Many of the used algorithms are
due to T. Oaku and N. Takayama37 and have been implemented in the
D-modules package for the Computer Algebra system Macaulay 2. 28
While giving the lectures and writing these notes we have supposed that
the students and the readers have some familiarity with basic notions in
Commutative Algebra and Algebraic Geometry. In particular they should
have a good elementary knowledge of the theory of commutative rings and
their modules, as contained for instance in the first three chapters of Atiyah-
Macdonald.4 They should also know the basic definitions and results in
the theory of affine algebraic varieties at the level of the first chapter of
R. Hartshorne.30 This knowledge must include the Nullstellensatz and the
theory of Hilbert functions and polynomials.
1. The Weyl Algebra

1.1. Linear differential operators
For the sake of simplicity we are going to consider the complex field C as
base field. Nevertheless, in what follows many algebraic results also hold
for any base field K of characteristic zero.
Let n ≥ 1 be an integer number and C[x] = C[x1 , . . . , xn ] be the ring of
polynomials in n variables with complex coefficients.
Let EndC (C[x]) be the C-algebra of endomorphisms of the C-vector
space C[x]. As the product in this algebra is just the composition of endo-
morphisms then EndC (C[x]) is a noncommutative ring with unit.
www.pdfgrip.com
54
Let f ∈ C[x] be a polynomial. The multiplication by f

φf
C[x] −→ C[x]
defined by φf (g) = f g for all g ∈ C[x], is an endomorphism.
The unit of the ring EndC (C[x]) is the identity map which coincides
with φ1 . We will denote φ1 just by 1 if no confusion is possible.
The partial derivative with respect to xi
∂
i ∂x
C[x] −→ C[x]
∂
is also an endomorphism. We will denote ∂i instead of ∂xi for i = 1, . . . , n.
∂f
So, for any f ∈ C[x] we have ∂i (f ) = ∂x i
.
Definition 1.1. Let n ≥ 1 be an integer. The n-th complex Weyl alge-

bra, denoted by An (C), is the subalgebra of EndC (C[x]) generated by the
endomorphisms
φx 1 , . . . , φ x n , ∂ 1 , . . . , ∂ n .
We will adopt the convention A0 (C) = C and we will simply write An =
An (C).
Remark 1.1. An element in An is nothing but a finite linear combination,

with coefficients in C, of words in the generators φx1 , . . . , φxn , ∂1 , . . . , ∂n .
Each of these words must be identified with the corresponding endomor-
phism built up by composing the generators appearing in the word.
For any f ∈ C[x] we have
(∂i ◦ φxi )(f ) = ∂i (xi f ) = f + xi ∂i (f ) = f + (φxi ◦ ∂i )(f ).
This means that the equality ∂i ◦ φxi = φxi ◦ ∂i + 1 holds in EndC (C[x]) and
therefore in An . In particular An is a noncommutative ring (for n ≥ 1).
More generally, for any f, g ∈ C[x] and 1 ≤ i ≤ n we have
(∂i ◦ φg )(f ) = ∂i (gf ) = ∂i (g)f + g∂i (f ) = φ∂i (g) (f ) + (φg ◦ ∂i )(f ).
That is: the equality ∂i ◦ φg = φg ◦ ∂i + φ∂i (g) holds in EndC (C[x]) and
therefore in An . The last equality is known as Leibniz’s rule.
Exercise 1.1. Prove that the following equalities hold in An :

∂i ◦ φxj = φxj ◦ ∂i for all 1 ≤ i, j ≤ n with i 6= j.
∂i ◦ ∂j = ∂j ◦ ∂i for all 1 ≤ i ≤ j ≤ n.
φxi ◦ φxj = φxj ◦ φxi for all 1 ≤ i ≤ j ≤ n.
www.pdfgrip.com
55
Proposition 1.1. The map C[x] −→ An defined by f 7→ φf is an injective

morphism of rings (and of C-algebras).
Proof. The proof follows from the equalities
φf +g (h) = (f + g)h = f h + gh = φf (h) + φg (h) = (φf + φg )(h)
and
φf g (h) = (f g)h = f (gh) = φf (φg (h)) = (φf ◦ φg )(h)
which hold for any f, g, h ∈ C[x].
Notation 1.1. The notations above are not easy to use. Hence, we will
simply write xi instead of φxi . This identification is justified by Proposition
1.1. We will also write P Q instead of P ◦ Q for the product in An .
For α = (α1 , . . . , αn ) ∈ Nn we will write xα = xα αn
1 · · · xn both for the
1
monomial in C[x] and the corresponding element in An . We will also write

∂ α = ∂1α1 · · · ∂nαn ∈ An . By convention we have x0i = 1 and ∂i0 = 1 for
i = 1, . . . , n.
An element xα ∂ β ∈ An for α, β ∈ Nn is called a monomial in An .
Proposition 1.2.
(1) Let f ∈ C[x] and β ∈ Nn . The product ∂ β f in An satisfies the equality

X β
β
∂ f= ∂ σ (f )∂ β−σ
σ
σβ

β β!
where σ β stands for σi ≤ βi for i = 1, . . . , n, = σ!(β−σ)! and
σ
β! = β1 ! · · · βn !
γ γ
(2) If β, γ ∈ Nn then we have ∂ β (xγ ) = β! xγ−β where = 0 if
β β
the relation β γ doesn’t hold.
(3) If α, α0 , β, β 0 ∈ Nn then we have
0 0 0 0
xα ∂ β xα ∂ β = xα+α ∂ β+β
X 0
β α 0 0
+ σ! xα+α −σ ∂ β+β −σ .
σ σ
σβ,σα0 ,σ6=0
Proof. (1) It follows from the case n = 1 and the distributivity of the
product with respect to the sum in An . For n = 1 (writing t and ∂t instead
www.pdfgrip.com
56
of x1 and ∂1 ) the formula

j
X
∂tj f = ∂tk (f )∂tj−k
k=0
can be proved by induction on j.

(2) The formula follows by induction on n. The case n = 1 can be proved
by induction on β1 .
(3) It follows from (1) and (2).
Proposition 1.3. The set of monomials B = {xα ∂ β | α, β ∈ Nn } is a basis

of the C-vector space An . Each nonzero element P in An can be written in
an unique way as a finite sum
X
P = pαβ xα ∂ β
α,β
P
for some nonzero complex numbers pαβ . Moreover, P = β pβ (x)∂ β with
P
pβ (x) = α pαβ xα .
Proof. It is enough to prove the first statement since the second statement
follows from it.
Any word in the generators x1 , . . . , xn , ∂1 , . . . , ∂n is a product of mono-
mials (see Remark 1.1). By Proposition 1.2 (3.) a product of monomials is
a linear combination with coefficients in C of elements in B. This proves
that B is a generating system for the vector space An . Let us prove now
that B is linearly independent. Let us consider
X
a non trivial linear combination of monomials in B. Let β 0 ∈ Nn be the
smallest element, with respect to the lexicographical order,b appearing as
0 P
the exponent of ∂ in P . We have that P (xβ ) = (β 0 )!( α pαβ 0 xα ) because
0
if β 0 is strictly smaller than β in the lexicographic order then ∂ β (xβ ) = 0.
Because of the choice of β 0 there exists α ∈ Nn such that pαβ 0 6= 0 and then
0
P (xβ ) is nonzero. In particular the endomorphism P ∈ An is nonzero.
Remark 1.2. There exists a natural action of An on the polynomial ring

C[x] since each element P ∈ An is an endomorphism of the C-vector space
C[x]. This natural action induces on C[x] a structure of left An –module.
b Given β, β 0 ∈ Nn we say that β 0 is smaller than or equal to β with respect to the
lexicographic order if there exists 1 ≤ i ≤ n − 1 such that βj0 = βj for 1 ≤ j ≤ i and

0
βi+1 ≤ βi+1 .
www.pdfgrip.com
57
P
Due to Proposition 1.3 the action of an element P = β pβ (x)∂ β ∈ An
on an element g ∈ C[x] can be written as
X ∂ β1 +···+βn (g)
P (g) = pβ (x) β1
β ∂x1 · · · ∂xβnn
and this justifies the name of Linear Differential Operators with polynomial
coefficients for the elements in An .
In fact C[x] is finitely generated as An -module. This can be proved just
by considering the map
An
φ : C[x] −→
An (∂1 , . . . , ∂n )
defined by φ(g) = g = g + An (∂1 , . . . , ∂n ) where An (∂1 , . . . , ∂n ) is the left
ideal generated by ∂1 , . . . , ∂n .
This map is a morphism of left An –modules and it is injective by Propo-
sition 1.3. Let’s see that φ is also surjective. Consider P ∈ An and write it
as
X
P = pβ (x)∂ β .
β
It is clear that φ(p0 (x)) = φ(P (1)) = P .
1.2. Order and total order

P
We will denote |β| = i βi for each β ∈ Nn .
Definition 1.2. For a nonzero operator

X X
P = pαβ xα ∂ β = pβ (x)∂ β ∈ An ,
α,β β
the maximum of |β| for pβ (x) 6= 0 is called the order of P . This nonnegative
integer is denoted by ord(P ). The maximum of |α| + |β| for pαβ 6= 0 is
called the total order of P and it is denoted by ordT (P ). We will write
ord(0) = ordT (0) = −∞.
Definition 1.3. The principal symbol of the operator

X X
P = pαβ xα ∂ β = pβ (x)∂ β
α,β β
is the polynomial
X
σ(P ) = pβ (x)ξ β ∈ C[x, ξ] = C[x1 , . . . , xn , ξ1 , . . . , ξn ]
|β|=ord(P )
www.pdfgrip.com
58
where ξ1 , . . . , ξn are new variables. The principal total symbol of P is the

polynomial
X
σ T (P ) = pαβ xα ξ β ∈ C[x, ξ].
|α+β|=ordT (P )
Remark 1.3. Notice that σ(P ) ∈ C[x, ξ] is a homogeneous polynomial of

degree ord(P ) in the ξ-variables while σ T (P ) ∈ C[x, ξ] is a homogeneous
polynomial of degree ordT (P ). In general we have σ(P ) 6= σ T (P ). In general
we do not have neither σ(P + Q) = σ(P ) + σ(Q) nor σ T (P + Q) = σ T (P ) +
σ T (Q). Nevertheless, we have the following Proposition.
Proposition 1.4. For P, Q ∈ An one has
(1) ord(P Q) = ord(P ) + ord(Q) and σ(P Q) = σ(P )σ(Q).
(2) ordT (P Q) = ordT (P ) + ordT (Q) and σ T (P Q) = σ T (P )σ T (Q).
(3) ord(P Q−QP ) ≤ ord(P )+ord(Q)−1 and ordT (P Q−QP ) ≤ ordT (P )+
ordT (Q) − 2.
(4) ord(P + Q) ≤ max{ord(P ), ord(Q)} (and similarly for ordT ).
(5) If ord(P ) = ord(Q) and σ(P )+σ(Q) 6= 0 then σ(P +Q) = σ(P )+σ(Q)
(and similarly for ordT and σ T ).
We are assuming −∞ + k = k + (−∞) = −∞ for k ∈ Z ∪ {−∞}.
Proof. (1), (2) and (3) follow from Proposition 1.2. Parts (4) and (5) follow
from the very Definitions 1.2 and 1.3.
From (1) in Proposition 1.4 we have:

Corollary 1.1. An is an integral domain.
Definition 1.4. For each left (or right) ideal I ⊂ An the graded ideal
associated with I is the ideal gr(I) of C[x, ξ] generated by the family of
principal symbols of elements in I
gr(I) = C[x, ξ]{σ(P ) | P ∈ I}.
The total graded ideal associated with I is the ideal grT (I) of C[x, ξ] gen-
erated by the family of principal total symbols of elements in I
grT (I) = C[x, ξ]{σ T (P ) | P ∈ I}.
Remark 1.4. If I = An P is the principal left ideal generated by an op-
erator P ∈ An then gr(I) is the principal ideal in C[x, ξ] generated by the
principal symbol σ(P ), so we have gr(An P ) = C[x, ξ]σ(P ). We have an
analogous result for grT (I).
www.pdfgrip.com
59
Proposition 1.5. Let I ⊂ An be a left ideal and F (x, ξ) be a polynomial

in gr(I) (resp. grT (I)). If F (x, ξ) is homogeneous with respect to the ξ-
variables (resp. homogeneous) then it is the principal symbol σ(P ) (resp.
the total principal symbol σ T (P )) for some P ∈ I.
Proof. Both cases being analogous we will only write the proof for the first
one. Assume F ∈ gr(I) is not zero. As gr(I) is generated by {σ(P ) | P ∈ I}
there are P1 , . . . , Pr in I such that
X
F = Hi σ(Pi )
i
for some polynomials Hi ∈ C[x, ξ]. Let’s denote by d (resp. di ) the ξ-degree
of F (resp. σ(Pi )). Then we can assume Hi ξ–homogeneous of degree d − di
(in particular Hi = 0 if d < di ). Let’s denote by Qi (x, ∂) any element in
An such that σ(Qi ) = Hi . As
X
F = σ(Qi )σ(Pi )
i
is a nonzero ξ–homogeneous polynomial of degree d then (see Proposition

1.4)
!
X X
σ Q i Pi = σ(Qi )σ(Pi )
i i
P
and i Qi Pi is the wanted operator in I.
Proposition 1.6. The ring An is left and right Noetherian domain.
Proof. We will proof that any left ideal I ⊂ An is finitely generated. A

similar proof can be done for right ideals. We can assume I 6= (0). By
definition of grT (I) there are polynomials σ T (P1 ), . . . , σ T (Pr ) with Pi ∈
I generating grT (I). Let us denote J the left ideal in An generated by
P1 , . . . , Pr . We will prove that J = I. Assume there exists P ∈ I \ J. We
can also assume P is of minimal total order. As σ T (P ) ∈ grT (I) then there
are homogeneous polynomials H1 , . . . , Hr ∈ C[x, ξ] such that
X
σ T (P ) = Hi σ T (Pi ).
i
We can also assume deg(Hi ) + ord (Pi ) = ordT (P ). Denote by Qi any

T
element in An with σ T (Qi ) = Hi . Then the operator

X
P 0 := P − Q i Pi
i
www.pdfgrip.com
60
has total order strictly smaller than ordT (P ) and then P 0 should be in J.
This implies P ∈ J which is a contradiction. This proves J = I.
Remark 1.5. Replacing in the above proof grT by gr, σ T by σ, ordT by

ord and homogeneous polynomials by ξ-homogeneous polynomials we get
another proof of Proposition 1.6.
Exercise 1.2. Prove that An is a simple ring.

Quick answer.- Let J be a nonzero two-sided ideal in An . Let P be a nonzero
element of minimal total order in J and write d = ordT (P ). If d = 0
then P ∈ C and as P is nonzero we have J = An . Assume d > 0. Let
(α, β) ∈ N2n be such that |α + β| = d and the coefficient pαβ of xα ∂ β in P
is nonzero. Assume there is i such that βi > 0.
Claim.- The operator Q = [xi , P ] = xi P − P xi belongs to J and it is
nonzero.
Let us prove the claim. The first part follows because J is a two-sided ideal.
Let us write P = P 0 + P 00 where P 0 (resp. P 00 ) is the sum of the monomials
in P of total order equal to d (resp. less or equal to d − 1). We can write
X̀
P0 = Pj (x, ∂ 0 )∂ij
j=0
where Pj = Pj (x, ∂ 0 ) doesn’t depend on ∂i , P` 6= 0 and, moreover, all

monomials in Pj have total order d − j. We also have ` ≤ d and since
βi > 0 we have ` > 0. From the equality Q = [xi , P 0 ] + [xi , P 00 ] and by
Proposition 1.4 we have ordT ([xi , P 00 ]) ≤ d − 2. Moreover, any monomial
in [xi , Pj ∂ij ] = jPj ∂ij−1 has total order less or equal than d − 1 and `P` ∂i`−1
is not zero. That proves the claim.
The claim contradicts the minimality of d. Then β should be zero as long
as pαβ 6= 0 and |α + β| = d. Assume now there exists i with αi > 0 and
|α| = d. In a similar way, using Q0 = [∂i , P ] we get a contradiction with the
minimality of d. So P should be a nonzero element in C and then J = An .
Remark 1.6. The only invertible elements in An are the nonzero constants
(i.e. the elements in C \ {0}), since if 1 = P Q for some P, Q ∈ An then
0 = ordT (1) = ordT (P ) + ordT (Q) and this implies that both P, Q must be
nonzero elements in C.
Exercise 1.3. If φ : S → S 0 is a ring morphism then ker(φ) is a two-sided

ideal in S.
www.pdfgrip.com
61
Corollary 1.2. Each ring morphism φ : An → S (S a possibly noncom-

mutative ring) is injective.
1.3. Filtrations on An
Let us denote
Fk (An ) = {P ∈ An | ord(P ) ≤ k}
and
Bk (An ) = {P ∈ An | ordT (P ) ≤ k}
for k ∈ Z.
If no confusion arises we will simply write Fk = Fk (An ) and Bk =
Bk (An ).
Proposition 1.7. The following properties hold:
(1) Bk = Fk = {0} for k ≤ −1.

(2) Bk ⊂ Fk for k ∈ Z.
(3) Bk ⊂ Bk+1 , Fk ⊂ Fk+1 for k ∈ Z.
(4) Bk B` ⊂ Bk+` , Fk F` ⊂ Fk+` for k, ` ∈ Z.
S S
(5) A n = k B k = k Fk .
(6) 1 ∈ B0 = C, 1 ∈ F0 = C[x].
2n + k
(7) Each Bk is a C–vector space with dimension .
k

n+k
(8) Each Fk is a free C[x]-module with rank .
k
Definition 1.5. The family (Fk )k (resp. (Bk )k ) is called the order filtration
(resp. the total order filtration) on An .
1.4. The graded rings grB (An ) and grF (An )
Bk
Exercise 1.4. Each quotient Bk−1 is a C–vector space with dimension

2n + k − 1
.
2n − 1
Quick answer.- The residue classes
xα ∂ β = xα ∂ β + Bk−1
www.pdfgrip.com
62
with |α+β| = k generate the quotient vector space Bk /Bk−1 and, moreover,
they are linearly independent since a linear combination
X
λαβ xα ∂ β
|α+β|=k
belongs to Bk−1 if and only if all λαβ are 0.
Exercise
1.5.
Each quotient Fk /Fk−1 is a free C[x]–module with rank
n+k−1
.
n−1
Quick answer.- The residue classes

∂ β = ∂ β + Fk−1
with |β| = k generate the quotient C[x]–module Fk /Fk−1 and, moreover,
they are linearly independent over C[x], since a C[x]–linear combination
X
λβ (x)∂ β
|β|=k
belongs to Fk−1 if and only if each λβ (x) is 0.
Proposition 1.8. The Abelian group

M Bk
grB (An ) :=
Bk−1
k∈Z
!
M Fk
F
resp. gr (An ) :=
Fk−1
k∈Z
has a natural structure of commutative ring with unit.
Proof. We will write down the case of the total order filtration (Bk )k , the
other one being analogous. Let us consider, for k, ` ∈ Z, the map
Bk B` Bk+`
µk` : × →
Bk−1 B`−1 B`+k−1
defined by
µk` (P + Bk−1 , Q + B`−1 ) = P Q + Bk+`−1
for P ∈ Bk , Q ∈ B` .
The map µk` is well defined: P Q + Bk+`−1 does not depend on the
chosen representatives P ∈ Bk and Q ∈ B` .
www.pdfgrip.com
63
We will simply denote P = P + Bk−1 , Q = Q + B`−1 and P Q instead

of µk` (P , Q).
From Proposition 1.4 it follows that for P ∈ Bk and Q ∈ B` we have
P Q−QP ∈ Bk+`−1 (in fact in this case we also have P Q−QP ∈ Bk+`−2 but
we do not need this stronger property here) and then P Q = QP . Moreover,
we also have P (Q R) = P (QR) = P (QR) = (P Q)R = (P Q) R = (P Q)R.
We define a map
µ0 : grB (An ) × grB (An ) → grB (An )
by bilinearity:
X X X
µ0 ( Pk , Q` ) = Pk Q `
k ` k,`
where Pk ∈ Bk and Q` ∈ B` for all k, `.

P P P P
We will simply denote ( k Pk )( ` Q` ) instead of µ0 ( k Pk , ` Q` ).
The map µ0 is well defined and defines a product on grB (An ). To this
end we can see that the binary operation defined by µ0 is associative and
commutative (since the corresponding properties hold for the maps µk` ).
As µ0 is defined by bilinearity it is distributive with respect to the sum.
Let us denote by 1 the residue class of 1 modulo B−1 = {0}. It is clear
P P
that 1( k Pk ) = k Pk and then it is the unit of the commutative ring
grB (An ).

Bk Fk
Remark 1.7. The family of Abelian groups Bk−1 (resp. Fk−1 ) is a
k k
B F
grading on the ring gr (An ) (resp. gr (An )).
Recall that we have denoted C[x, ξ] = C[x1 , . . . , xn , ξ1 , . . . , ξn ].
Proposition 1.9. The graded ring grB (An ) is isomorphic to the polyno-
mial ring C[x, ξ] endowed with the grading defined by the degree of the poly-
nomials.
Proof. Let us consider, for k ∈ N, the isomorphism (of vector spaces)
ηk : Bk /Bk−1 → C[x, ξ]k
defined
  
X X
ηk  pαβ xα ∂ β  + Bk−1  = pαβ xα ξ β .
|α+β|≤k |α+β|=k
www.pdfgrip.com
64
Here C[x, ξ]k denotes the C–vector space of homogeneous polynomials of

degree k. The family ηk yields by bilinearity a natural isomorphism
η : grB (An ) → C[x, ξ]
of graded rings.
Remark 1.8. The polynomial ring C[x, ξ] can be also endowed with the
grading defined by the degree in ξ:
M
C[x, ξ] = C[x, ξ](k)
k
with
X
C[x, ξ](k) = C[x]ξ β .
|β|=k
We will call this grading the ξ–grading on C[x, ξ].
Proposition 1.10. The graded ring grF (An ) is isomorphic to the polyno-
mial ring C[x, ξ] endowed with the ξ–grading.
Proof. Let us consider, for k ∈ N, the isomorphism (of vector spaces)

ηk0 : Fk /Fk−1 → C[x, ξ](k)
defined by
 
X X
ηk0  pβ (x)∂ β  = pβ (x)ξ β .
|β|≤k |β|=k
Here C[x, ξ](k) denotes the free C[x]–module of ξ–homogeneous polynomials

of degree k (see Remark 1.8). The family ηk0 yields by bilinearity a natural
isomorphism
η 0 : grF (An ) → C[x, ξ]
of graded rings (when considering the ξ-grading on C[x, ξ]).
Remark 1.9. Notice that if P ∈ Bk \Bk−1 (resp. P ∈ Fk \Fk−1 ) then ηk (P )

(resp. ηk0 (P )) is nothing but the principal total symbol (resp. the principal
symbol) of P : ηk (P ) = σ T (P ) (resp. ηk0 (P ) = σ(P )) (see Definition 1.3).
Notation 1.2. From now on, we will identify the graded rings grB (An )
and C[x, ξ] (endowed with the degree of the polynomials) (resp. grF (An )
and C[x, ξ] (endowed with the ξ–degree)) by mean of the isomorphism η
(resp. η 0 ).
www.pdfgrip.com
65
1.5. B–filtrations on An -modules

From now on, all the An –modules (resp. ideals) will be left modules (resp.
left ideals) unless otherwise specified.
Definition 1.6. Let M be an An -module. A B–filtration on M is a family

Γ = (Mk )k∈N of finitely dimensional C–vector subspaces of M such that:
i) Mk ⊂ Mk+1 for all k ∈ N.
S
ii) k Mk = M .
iii) Bk M` ⊂ Mk+` for all (k, `).
That is not the more general definition of filtration on a module. There

exist filtrations indexed by Z instead of N but we will not need them here.
Nevertheless, if in these notes we write Γ = (Mk )k∈Z for a B–filtration on
M it is assumed to be Mk = {0} for k < 0.
Remark 1.10.
(1) The total order filtration (Bk (An ))k is a B–filtration on the left An –
module An .
(2) Let us denote, for k ∈ Z,
Bk (C[x]) = {f ∈ C[x] | deg(f ) ≤ k}.
The family (Bk (C[x]))k is a B–filtration on the left An -module C[x].
(3) Let I ⊂ An be an ideal and denote Bk (I) = Bk (An ) ∩ I for k ∈ Z. The
family (Bk (I))k is a B–filtration on I considered as a left An –module
(see also Subsection 1.9).
(4) Let I ⊂ An be an ideal and define

An Bk (An ) + I
Bk =
I I
for k ∈ N. The family (Bk (An /I))k is a B–filtration on the left An –
module An /I. It will be called the induced B–filtration on An /I (see
also Subsection 1.9).
1.6. F –filtrations on An -modules

We can also define in a similar way as before F –filtrations on An –modules.
Definition 1.7. An F –filtration on a An –module M is a family Γ =

(Mk )k∈N of finitely generated C[x]–submodules of M such that:
i) Mk ⊂ Mk+1 for all k ∈ N.
www.pdfgrip.com
66
S
ii) k Mk = M .
iii) Fk M` ⊂ Mk+` for all (k, `).
As before, if in these notes we write Γ = (Mk )k∈Z for an F –filtration on

M it is assumed to be Mk = {0} for k < 0.
Remark 1.11.
(1) The order filtration (Fk (An ))k is an F –filtration on the left An –module
An .
(2) Let us denote, for k ∈ Z, Fk (C[x]) = C[x] for k ≥ 0 and Fk (C[x]) = 0.
The family (Fk (C[x]))k is an F –filtration on the left An –module C[x].
(3) Let I ⊂ An be an ideal and denote Fk (I) = Fk (An ) ∩ I for k ∈ Z. The
family (Fk (I))k is an F –filtration on I considered as a left An –module
(see also Subsection 1.9).
(4) Let I ⊂ An be an ideal and denote

An Fk (An ) + I
Fk =
I I
for k ∈ Z. The family (Fk (An /I))k is an F –filtration on the left An –
module An /I. It will be called the induced F –filtration on An /I (see
also Subsection 1.9).
1.7. The Γ–order and the Γ–symbol map

Let Γ = (Mk )k be a filtrationc on an An -module M . For each nonzero
m ∈ M we call the Γ–order of m and we denote by ordΓ (m) the integer k
such that m ∈ Mk \ Mk−1 .
Let us denote
σkΓ : Mk → Mk /Mk−1
the canonical projection (which is a C–linear map). We have σkΓ (m) =
m + Mk−1 for m ∈ Mk .
If Γ is an F –filtration then σkΓ is also a morphism of C[x]–modules.
The map σkΓ is called the k-th Γ–symbol map associated with the filtra-
tion Γ.
If M = An and Γ = (Bk )k is the total order filtration on An (also called
the B–filtration on An ), the corresponding k-th symbol map will be also
denoted by σkB .
cA filtration on an An –module M is either a B–filtration or an F –filtration.

www.pdfgrip.com
67
If Γ = (Fk )k is the order filtration on An (also called the F –filtration

on An ), the corresponding k-th symbol map will be also denoted σkF .
Remark 1.12. We will use here the notations of Propositions 1.9 and 1.10.
If P ∈ Bk \ Bk−1 (resp. P ∈ Fk \ Fk−1 ) then
(ηk ◦ σkB )(P ) = σ T (P )

resp. (ηk0 ◦ σkF )(P ) = σ(P ) .
1.8. Graded associated module

Let M be an An –module and Γ = (Mk )k a B–filtration (resp. F –filtration
) on M .
As each quotient Mk /Mk−1 is an Abelian group (and a C–vector space)
the direct sum
M Mk
grΓ (M ) :=
Mk−1
k
is also an Abelian group (and a C–vector space).

For m ∈ Mk the class
Mk
m + Mk−1 ∈
Mk−1
will be simply denoted by m if no confusion arises.
P
An element in grΓ (M ) is a finite sum k mk where each mk belongs to
Mk .
Proposition 1.11. Let M be an An –module and Γ = (Mk )k be a B–

filtration (resp. F –filtration) on M . The Abelian group grΓ (M ) has a nat-
ural structure of grB (An )–module (resp. grF (An )–module).
Proof. We will only treat the case of the B–filtration the other one being
analogous.
Let us consider the map
ν : grB (An ) × grΓ (M ) → grΓ (M )
defined by bilinearity from the maps
Bk M` Mk+`
νk : × →
Bk−1 M`−1 Mk+`−1
defined by
νk (Pk , m` ) = Pk m` .
www.pdfgrip.com
68
It is straightforward to show that the map ν defines on grΓ (M ) a struc-

ture of grB (An )–module.
Definition 1.8. The graded module grΓ (M ) will be called the associated
graded module to the filtration Γ = (Mk )k on M .
1.9. Induced filtrations

Let us recall that a filtration on an An –module M will be either a B–
filtration or an F –filtration on M .
Let M be an An –module, N ⊂ M a submodule of M and Γ = (Mk )k a
filtration on M . For each k ∈ Z let us denote Nk := Mk ∩N and (M/N )k :=
(Mk + N )/N .
The following proposition is easy to prove.
Proposition 1.12.
(1) The family Γ0 = (Nk )k is a filtration on N . It will be called the induced

filtration by Γ on N .
(2) The family Γ00 = ((M/N )k )k is a filtration on M/N . It will be called
the induced filtration by Γ on M/N .
Let I be an ideal in An . Using the notations in Remark 1.10 the family

(Bk (I))k (resp. (Bk (An /I))k ) is the induced B–filtration on I (resp. on
An /I). We also have the analogous statement for the F –filtration (using
Remark 1.11).
Proposition 1.13. Let M be an An –module, N ⊂ M a submodule of M

and Γ = (Mk )k a filtration on M . Then there exists a canonical exact
sequence of graded modules
0 00
0 → grΓ (N ) → grΓ (M ) → grΓ (M/N ) → 0.
Proof. For each k ∈ Z we have an exact sequence of C-vector spaces
0 → Nk → Mk → (M/N )k → 0
since (M/N )k = MkN+N ' MM k

k ∩N
=M Nk . Then for each k ∈ Z there exists a
k
canonical exact sequence of vector spaces

Nk Mk Mk + N
0→ → → → 0.
Nk−1 Mk−1 Mk−1 + N
www.pdfgrip.com
69
Corollary 1.3. Let I be an ideal in An . We have canonical isomorphisms

B An grB (An ) F An grF (An )
gr ' and gr ' .
I grB (I) I grF (I)
Proof. It follows from Proposition 1.13 applied to the canonical exact se-
quence
An
0 → I −→ An −→ →0
I
where each term of the sequence is endowed with its corresponding B–
filtration or F –filtration.
By Corollary 1.3 there exists a canonical injective map grB (I) →

gr (An ) which is a morphism of graded modules and then grB (I) can be
B
identified with a graded submodule of grB (An ). This means that grB (I) is
a graded ideal –modulo this identification– of the graded ring grB (An )).
Let us recall (see Proposition 1.9) that there exists a natural isomor-
phism of graded rings
η : grB (An ) → C[x, ξ]
whose k-th homogeneous component
ηk : Bk /Bk−1 → C[x, ξ]k
is defined by
  
X X
ηk  pαβ xα ∂ β  + Bk−1  = pαβ xα ξ β .
|α+β|≤k |α+β|=k
Here C[x, ξ]k denotes the C–vector space of homogeneous polynomials

of degree k.
Proposition 1.14. With the above notations, the ideal η(grB (I)) is the
homogeneous ideal of C[x, ξ] generated by the family {σ T (P ) | P ∈ I}.
We have an analogous result for grF (I) and the family {σ(P ) | P ∈ I}
(using the notation of Proposition 1.10).
Recall that we have identified grB (An ) with C[x, ξ] (see Notation 1.2).
Let I be an ideal of An and let us denote M = An /I and Γ00 = (Bk (M ))k
00
the induced B–filtration on M (see Remark 1.10). By Corollary 1.3 grΓ (M )
is isomorphic as C[x, ξ]–module to
www.pdfgrip.com
70
C[x, ξ]
.
grB (I)
We have an analogous result for the F –filtration.
1.10. Good filtrations
Proposition 1.15. Let M be an An –module and Γ = (Mk )k a B–filtration

on M . If grΓ (M ) is a finitely generated C[x, ξ]–module then M is Noethe-
rian.
Proof. Since An is a Noetherian ring it is enough to prove that M is

finitely generated. Assume m1 , . . . , mr is a homogeneous generating system
of grΓ (M ). Assume mi ∈ Mki for some ki ∈ Z, i = 1, . . . , r. We will prove
that the set {m1 , . . . , mr } generates M .
Let us denote M 0 the submodule of M generated by the mi . We will prove
that any m ∈ M is in fact in M 0 . We use induction on ordΓ (m) (see Subsec-
tion 1.7). There is nothing to prove if ordΓ ≤ 0. Assume that any element
m0 ∈ M such that ordΓ (m0 ) ≤ k is in M 0 for some integer k > 0. Let
m ∈ M be such that ordΓ (m) = k + 1. Let us write
X
m= f i mi
i
for some homogeneous polynomials fi ∈ C[x, ξ] where deg(fi ) = k + 1 − ki .

Let us write
X
m0 = m − Pi (x, ∂)mi
i
where Pi = Pi (x, ∂) is a differential operator satisfying σ T (Pi ) = fi . The

residue class of m0 modulo Mk+1 is zero and then, by induction, m0 is in
M 0 . Then m ∈ M 0 .
Definition 1.9. Let M be an An –module and Γ = (Mk )k a B–filtration .

We say that Γ is a good filtration if grΓ (M ) is a finitely generated grB (An )–
module.
Exercise 1.6. Let I be an ideal of An . Prove that:

i) The induced B–filtration on I is a good filtration.
ii) The induced B–filtration on An /I is a good filtration.
iii) Any finitely generated An –module M admits a good B–filtration.
www.pdfgrip.com
71
Both i) and ii) have direct proofs. Hint for iii): If m1 , . . . , mr is a

P
generating system for M , define Mk := j Bk mj for k ∈ N. The family
(Mk )k is a good B–filtration on M .
Proposition 1.16. Let M be an An –module and Γ = (Mk )k a B–filtration

on M . The following conditions are equivalent:
i) Γ is a good B–filtration on M .
ii) There exists k0 ∈ N such that Mk+` = B` Mk for all ` ≥ 0 and for all
k ≥ k0 .
Proof. ii) ⇒ i). It is enough to prove that grΓ (M ) is generated by

M0 ⊕ M1 /M0 ⊕ · · · ⊕ Mk0 /Mk0 −1 since each M` is a vector space of fi-
nite dimension. To see this, if m ∈ Mk and k > k0 then write k = k0 + i
for i = k − k0 > 0. Since Mk = Bi Mk0 we have
r
X
m= P j mj
j=1
where Pj ∈ Bi and m1 , . . . , mr is a basis (or simply a finite generating

system) of the C–vector space Mk0 . We can write
X X X
m+Mk−1 = m = Pj mj +Mk−1 = (Pj +Bi−1 )(mj +Mk0 −1 ) = P j mj .
j j j
i) ⇒ ii). Let m1 , . . . , mr be a homogeneous system of generators of grΓ (M ).

Suppose mj ∈ Mkj \ Mkj −1 for j = 1, . . . , r. Write k0 := max{kj }. We will
prove by induction on ` that Mk+` = B` Mk for all ` ≥ 0 and all k ≥ k0 .
There is nothing to prove for ` = 0. Suppose the result is true for ` − 1 for
some ` > 0. Let us consider m ∈ Mk+` for k ≥ k0 . We can write
X
m = m + Mk+`−1 = f j mj
j
for some homogeneous polynomial fj ∈ C[x, ξ] of degree k + ` − kj . Let us

write
X
m0 = m − Pj (x, ∂)mj
j
for some Pj ∈ Bk+`−kj such that σ T (Pj ) = fj . It is clear that m0 ∈ Mk+`−1

and by induction m0 ∈ B`−1 Mk . Since k − kj ≥ 0 we also have Bk+`−kj =
B` Bk−kj . Then
www.pdfgrip.com
72
X
m = m0 + Pj (x, ∂)mj .
j
Since Pj (x, ∂) ∈ Bk+`−kj = B` Bk−kj then Pj (x, ∂)mj ∈ B` Bk−kj Mkj ⊂

B` Mk .
Exercise 1.7. Let us define Bk0 = B2k for k ∈ Z. Prove that (Bk0 )k is a B–
filtration on the An –module An . Prove that (Bk0 )k is not a good filtration.
(Hint: For i > 0 and k > 0 we have Bi Bk0 = Bi B2k $ Bi+k 0
= B2i+2k ).
Proposition 1.17. Let M be an An –module; Γ = (Mk )k and Γ0 = (Mk0 )k

two B–filtrations on M . We have:
i) If Γ is a good B–filtration then there exists k1 ∈ N such that
0
Mk ⊂ Mk+k 1
for all k ∈ N.
ii) If Γ and Γ0 are both good filtrations then there exists k2 ∈ N such that
0 0
Mk−k 2
⊂ Mk ⊂ Mk+k 2
for all k ∈ N.
Proof. Let us prove first that ii) follows from i). So, let us assume i) is
0
proved. Then there exists j1 ∈ N such that Mk ⊂ Mk+j 1
and there exists
0
j2 ∈ N such that Mk ⊂ Mk+j2 and both inclusions hold for all k. Let us
define k2 to be the maximum of j1 and j2 . This k2 satisfies ii).
Let us prove i). By Proposition 1.16 there exists k0 ≥ 0 such that Mk+` =
B` Mk for all ` ≥ 0 and for all k ≥ k0 . As Mk0 is a C–vector space of finite
dimension there exists k1 ∈ N such that Mk0 ⊂ Mk0 1 . If k ≥ k0 we have
Mk = Mk−k0 +k0 = Bk−k0 Mk0 ⊂ Bk−k0 Mk0 1 ⊂ Mk−k
0
0 +k1
0
⊂ Mk+k 1
.
If 0 ≤ k ≤ k0 then Mk ⊂ Mk0 ⊂ Mk0 1 ⊂ Mk+k
0
1
.
Definition 1.10. Let M be an An –module and Γ = (Mk )k an F –filtration.

We say that Γ is a good filtration if grΓ (M ) is a finitely generated grF (An )–
module.
Exercise 1.8. If I is an ideal of An , prove that:

i) The induced F –filtration on I is a good filtration.
ii) The induced F –filtration on An /I is a good filtration.
iii) Any finitely generated An –module M admits a good F –filtration.
www.pdfgrip.com
73
i) and ii) have direct proofs. Hint for iii). If m1 , . . . , mr is a generating

P
system for M , define Mk := j Fk mj for k ∈ N. The family (Mk )k is a
good F –filtration on M .
Remark 1.13. Propositions 1.15, 1.16 and 1.17 remain true if one replaces
B–filtrations by F –filtrations.
1.11. Rational functions

The ring C[x] has a natural structure of left An -module (see Remark 1.2)
P
since each linear differential operator P = β pβ (x)∂ β ∈ An acts on any
polynomial g ∈ C[x] just by
X ∂ β1 +···+βn (g)
P (g) = pβ (x) ∈ C[x].
∂xβ1 1 · · · ∂xβnn
Let us consider C(x) the fraction field of the domain C[x]. Elements in
C(x) are rational functions, that is, quotients fg of polynomials g, f ∈ C[x]
with f 6= 0.
For any of these rational functions fg its partial derivative ∂i ( fg ) is noth-
ing but
∂i (g)f − g∂i (f )
f2
and then it is also a rational function. This can be extended to an action

of An on C(x) defining C(x) as a left An –module.
For any nonzero polynomial f = f (x) ∈ C[x] the ring
g
C[x]f = { ∈ C(x) | g ∈ C[x], m ∈ N}
fm
is a C[x]-module. In fact, C[x]f has also a natural structure of left An -

module since ∂i ( fgm ) ∈ C[x]f for all g ∈ C[x] and all m ∈ N.
If f ∈ C \ {0} then C[x]f is simply the ring C[x]. If f is not a constant,
then the elements in C[x]f are called rational functions with poles along
the affine hypersurface V(f ) := {a ∈ Cn | f (a) = 0}.
If f ∈ C[x] \ C then the C[x]–module C[x]f is not finitely generated.
Nevertheless, we have
Theorem 1.1 (J. Bernstein6 ). The An –module C[x]f is finitely gener-

ated.
www.pdfgrip.com
74
This theorem was done by J. Bernstein.6 It is deeply related to the

existence of the global Bernstein (or global Bernstein-Sato) polynomial as-
sociated with f . We will give a stronger version of this result in Theorem
2.7.
Bibliographical note
Most of the material of this Section appears in the already cited article
Bernstein6 and in the books by J.E. Björk7 and by S.C. Coutinho.21
2. Characteristic Variety. Holonomic An –Modules

2.1. Classical characteristic vectors
Let us consider a linear partial differential equation
 
X
P (x, ∂)(u) =  pβ (x)∂ β  (u) = v
β∈Nn
with polynomial coefficients pβ (x) ∈ R[x] := R[x1 , . . . , xn ].

A vector ξ0 ∈ Rn is called characteristic for P at a point x0 ∈ Rn if
σ(P )(x0 , ξ0 ) = 0. Here σ(P ) is the principal symbol of P (see Definition
1.3). The set of all such ξ0 is called the characteristic variety of the operator
P (or of the equation P (u) = v) at x0 ∈ Rn and is denoted by Charx0 (P ).
Notice that here, in contrast to some textbooks, the zero vector could
be characteristic. More generally, the classical characteristic variety of the
operator P is by definition the set
Char(P ) = {(x0 , ξ0 ) ∈ Rn × Rn | σ(P )(x0 , ξ0 ) = 0}.
Assume ord(P ) ≥ 1, then P is said to be elliptic at a point x0 ∈ Rn if

P has no nonzero characteristic vectors at x0 (i.e. if Charx0 (P ) ⊂ {0}) and
it is said to be elliptic (on Rn ) if Char(P ) ⊂ Rn × {0}.
Pn
The Laplace operator i=1 ∂i2 is elliptic on Rn .
Pn
The characteristic variety of the wave operator P = ∂12 − i=2 ∂i2 is
nothing but the hyperquadric defined in Rn × Rn by the equation ξ12 −
Pn 2
i=2 ξi = 0.
Characteristic vectors are important in the study of singularities of so-
lutions as can be seen in any classical book on Differential Equations.
To define the characteristic vectors for a Linear Partial Differential
System
www.pdfgrip.com
75
P11 (u1 ) + · · · + P1m (um ) = v1

.. .. .. .. (1)
. . . .
P`1 (u1 ) + · · · + P`m (um ) = v`
is more involved and in general the naive approach of simply considering
the principal symbols of the equations turns out to be unsatisfactory (see
Example 2.1).
We will use graded ideals and Groebner bases for Linear Differential Op-
erators (see Appendix A and Subsection 2.2) to define and to compute the
characteristic variety of a system as in (1). Here `, m are nonzero integers,
Pij are Linear Differential Operators, uj are unknown and vi are given data
(e.g. functions, distributions, hyperfunctions, . . . ).
2.2. Characteristic variety

Let us assume that the operators Pij in the System (1) are in the Weyl alge-
bra An and remember we are assuming An to be defined over the complex
field C (see Subsection 1.1). With the System (1) we associate the quotient
An –module
Am n
An (P1 , . . . , P` )
where Pi = (Pi1 , . . . , Pim ) ∈ Am n and An (P1 , . . . , P` ) denotes the submodule
of Amn generated by P 1 , . . . , P ` .
Let us assume first that System (1) has only one unknown u = u1 so
that the associated An –module is nothing but An (PA1 ,...,P n
`)
.
If J ⊂ C[x, ξ] = C[x1 , . . . , xn , ξ1 , . . . , ξn ] is a polynomial ideal we denote
by VC (J) (or simply by V(J)) the affine algebraic variety defined in C2n by
J, that is
VC (J) = V(J) = {(a, b) ∈ C2n | g(a, b) = 0, ∀g(x, ξ) ∈ J}.
Recall that for any (left) ideal I ⊂ An we denote by gr(I) the ideal of
C[x, ξ] generated by the family of principal symbols of elements in I (see
Definition 1.4).
Definition 2.1. Let I ⊂ An be a left ideal. The characteristic variety of

the left An –module An /I is defined as
Char(An /I) = VC (gr(I)).
Remark 2.1. If I = An P is a principal ideal then the characteristic variety

of An /I coincides with the classical characteristic variety of the operator
www.pdfgrip.com
76
P since in this case the graded ideal gr(I) ⊂ C[x, ξ] is generated by the
principal symbol σ(P ).
In general, if the ideal I ⊂ An is generated by a family P1 , . . . , Pm ,

the ideal grF (I) could be strictly bigger than the ideal generated by
σ(P1 ), . . . , σ(Pm ). One example is given just bellow.
Example 2.1. If I = A2 (P1 , P2 ) with P1 = x1 ∂1 + x2 ∂2 and P2 =

x1 ∂2 + x22 ∂1 then gr(I) = hξ1 , ξ2 i. This ideal strictly contains the ideal
hσ(P1 ), σ(P2 )i = hx1 ξ1 + x2 ξ2 , x1 ξ2 + x22 ξ1 i.
The following Macaulay 2 script can be used to compute generators
of gr(I) by using the Macaulay 2 command charIdeal. We need the
D-modules.m2 package to this end (see Ref. 28).
The input command lines are i1 :, i2 : ... while the output ones are
o4 =, o5 =, ... (a semicolon ; at the end of an input line prevents it from
being printed).
Input i1 defines the ring R as the polynomial ring in the variables x, y
with coefficients in Q (we are using x = x1 , y = x2 ).
Input i2 loads the D-modules.m2 package. Input i3 defines W as the 2 nd
Weyl algebra over the field Q (i.e. the algebra of linear differential operators
with polynomial coefficients in R).
Inputs i4 and i5 define the operators P1 and P2 as above and the (left)
ideal I in W generated by these two operators. In the Weyl algebra W the
expressions dx, dy stand for ∂1 , ∂2 respectively.
The Macaulay 2 expression charIdeal I computes a generating sys-
tem of the graded ideal gr(I). Notice the additional line of output labelled
with o6 : Output lines labelled with colons (:) provide information about
the type of output. In this case, the symbol QQ [x, y, dx, dy] denotes
the graded ring associated with the F –filtration on the Weyl algebra W (see
Subsection 1.4). In particular the expressions dx, dy, when considered in
the polynomial ring QQ [x, y, dx, dy], stand for the commutative vari-
ables ξ1 , ξ2 .
Output o6 = shows that gr(I) = hξ1 , ξ2 i ⊂ Q[x1 , x2 , ξ1 , ξ2 ].
Macaulay 2, version 1.2 with packages: Elimination,
IntegralClosure, LLLBases, PrimaryDecomposition, ReesAlgebra,
SchurRings, TangentCone
i1 : load "D-modules.m2";
i2 : R=QQ[x,y];
i3 : W=makeWA R;
i4 : P1=x*dx+y*dy, P2=x*dy+y^2*dx
www.pdfgrip.com
77
2
o4 = (x*dx + y*dy, y dx + x*dy)
o4 : Sequence
i5 : I=ideal(P1,P2)
2
o5 = ideal (x*dx + y*dy, y dx + x*dy)
o5 : Ideal of W
i6 : charIdeal I
o6 = ideal (dy, dx)
o6 : Ideal of QQ[x, y, dx, dy]
Let us add some extra computation in Macaulay 2. The following script

gives a new insight into the previous ideal I.
Input i7: defines J as the (left) ideal in W generated by ∂1 , ∂2 . The
output o7 : tells us that the (left) ideal J is considered in the ring W. Notice
the similarity of the lines o6 = and o7 = although the first output represents
an ideal in a polynomial ring and the second one an ideal in a Weyl algebra.
The string == stands for the binary operator testing equality of ideals.
The last part of the script proves the equality I = J in W giving a new
explanation for the equality gr(I) = hξ1 , ξ2 i ⊂ C[x1 , x2 , ξ1 , ξ2 ].
i7 : J=ideal(dx,dy)
o7 = ideal (dx, dy)
o7 : Ideal of W
i8 : J==I
o8 = true
Remark 2.2. Let M be an An –module provided with an F –filtration (resp.

a B–filtration) Γ = (Mk )k . The annihilating ideal AnngrF (An ) (grΓ (M )) is a
ξ–homogeneous ideal in grF (An ) (resp. a homogeneous ideal in grB (An )).
We will prove it for the F –filtration (the other case being analogous). To
this end, let us consider a nonzero element G = G(x, ξ) ∈ grF (An ) = C[x, ξ]
P
(see Proposition 1.10) annihilating grΓ (M ). We can write G = j Gj as
sum of its ξ–homogeneous components Gj ∈ C[x, ξ](j) (see Remark 1.8).
Let Pj be an element in Fj (An ) such that σ(Pj ) = Pj + Fj−1 (An ) = Gj .
P
For each k ∈ N we have G MMk−1 k
= 0 and then ( j (Pj +
Fj−1 (An )) MMk−1
k
= 0. Then for each j, k ∈ N we have Pj Mk ⊂
Mk
Mj+k−1 and therefore Gj = Pj + Fj−1 (An ) annihilates Mk−1 for all k.
www.pdfgrip.com
78
Then Gj ∈ AnngrF (An ) (grΓ (M )) for each j. This proves that the ideal
AnngrF (An ) (grΓ (M )) is ξ–homogeneous.
Proposition 2.1. Let M be a finitely generated An –module provided with

two good F –filtrations Γ = (Mk )k and Γ0 = (Mk0 )k ) (see Subsection 1.10).
Then
q q
AnngrF (An ) (grΓ (M )) = AnngrF (An ) (grΓ0 (M )).
Proof. Let’s write

J = AnngrF (An ) (grΓ (M ))
and
0
J 0 = AnngrF (An ) (grΓ (M )).
√ √
By symmetry it is enough to prove the inclusion
√ J ⊂ J 0 . By Remark
√ 2.2,
the ideal J is homogeneous. Then also J is homogeneous. Let G ∈ J be
a ξ–homogeneous element of ξ–degree ν ≥ 0. There exists an integer ` > 0
such that G` ∈ J. Let P be an element in Fν (An ) such that σ(P ) = G. We
have P ` Mk ⊂ M`ν+k−1 and then, for all p ∈ N we also have
P p` Mk ⊂ Mp`ν+k−p .
By Proposition 1.17 there exists k2 ∈ N such that Mk−k2 ⊂ Mk0 ⊂ Mk+k2
for all k ∈ N. In particular, for p = 2k2 + 1 and for all k ∈ N we have
P (2k2 +1)` Mk0 ⊂ P (2k2 +1)` Mk+k2 ⊂ M(2k2 +1)`ν+k+k2 −2k2 −1 ⊂ M(2k
0
2 +1)`ν+k−1
.
0
This √proves that σ(P (2k2 +1)` ) = G(2k2 +1)` annihilates grΓ (M ) and then
G ∈ J 0.
The following Proposition can be proven similarly to the previous one.

two good B–filtrations Γ = (Mk )k and Γ0 = (Mk0 )k ) (see Subsection 1.10).
Then
q q
AnngrB (An ) (grΓ (M )) = AnngrF (Bn ) (grΓ0 (M )).
Definition 2.2. Let M be a finitely generated An –module. The character-

istic variety of M is defined as
Char(M ) = V(AnngrF (An ) (grΓ (M )))
for a good F –filtration Γ on M .
www.pdfgrip.com
79
By Proposition 2.1 the definition of Char(M ) doesn’t depend on the

choice of the good F –filtration.
Remark 2.3. If M = An /I for some ideal I in An then by Corollary 1.3

we have

An grF (An )
grF '
I grF (I)
and then both Definitions 2.1 and 2.2 coincide.
2.3. Dimension of an An –module

The Krull dimension of a Zariski closed subset Z in Cn (denoted by dim(Z))
is by definitiond the maximum of the lengths m of decreasing chains
Z ⊇ Z0 ⊃ · · · ⊃ Zm
of irreducible Zariski closed subsets in Z.

By Hilbert’s Nullstellensatz the Krull dimension of Z equals the Krull
dimension of the C–algebra C[x]/J if J ⊂ C[x] is any ideal verifying V(J) =
Z. The Krull dimension of a ringe is the maximum of the lengths of chains
of prime ideals in the ring (see e.g. [30, Chapter I, Proposition 1.7]). By
convention the Krull dimension of the empty set (and of the zero ring) is
−1.
Krull dimension can be calculated by using Groebner basis computa-
tions in polynomial rings (see e.g. [24, Section 15.10.2]). For example, in
Macaulay 2 the string dim I computes the dimension of the quotient ring
R/I if I is an ideal in the ambient polynomial ring R.
i1 : R=QQ[x,y,z];
i2 : f=x^2*y^2+x*z-y; g=x^2+y^2+z^2;
i4 : I=ideal(f,g)
2 2 2 2 2
o4 = ideal (x y + x*z - y, x + y + z )
o4 : Ideal of R
i5 : dim I
o5 = 1
d See e.g. [30, Ch.I, page 5]

e i.e. a commutative ring with unit
www.pdfgrip.com
80
In the previous script, output o5 tells us that the Krull dimension of

the quotient ring Q[x,y,z]
I is 1, the ideal I being generated by polynomials
f = x2 y 2 + xz − y and g = x2 + y 2 + z 2 .
Definition 2.3. Let M be a finitely generated An –module. The dimension

of M (denoted dim(M )) is the Krull dimension of Char(M ) the character-
istic variety of M .
Example 2.2. Assume I = An P for some P ∈ An . Then if P is a

nonzero constant then An /I = (0) and its dimension is −1. If P = 0
then Char(An /I) = Char(An ) = Cn × Cn and then its dimension is 2n.
If P ∈ An \ C then σ(P ) ∈ C[x, ξ] is a non-constant polynomial and the
Krull dimension of Char(An /I) = VC (σ(P )) is 2n − 1. Thus, in this case,
dim(An /I) = 2n − 1.
If I is an ideal in a Weyl algebra An , the Macaulay 2 command dim I

computes the dimension of the quotient module An /I.
The following Macaulay 2 script shows that the dimension of the A3 –
module A3 /I is 3, where I ⊂ A3 is the ideal generated by the two operators
P = x3 ydxdy +dz and Q = ydydz −zy. We are using as usual x, y, z instead
of x1 , x2 , x3 and dx, dy, dz instead of ∂1 , ∂2 , ∂3 .
Input i3 defines W as the Weyl algebra over the polynomial ring in
the three variables x,y,z and with rational coefficients. Output o5 shows
that the dimension of W/I is 3. This last result is somehow unexpected
because the ideal I is defined by 2 elements in a Weyl algebra over three
variables.
i2 : R=QQ[x,y,z];
i3 : W=makeWA R;
i4 : P=x^3*y*dx*dy+dz,Q=y*dy*dz-z*y;
i5 : I = ideal (P,Q);
i6 : dim I
o6 = 3
i7 : charIdeal I
3
o7 = ideal (dz, y, x dx)
o7 : Ideal of QQ[x, y, z, dx, dy, dz]

www.pdfgrip.com
81
Output o7 = gives a system of polynomials defining the characteristic

variety of the module W/I. From this computation it is clear that the Krull
dimension of this variety is 3.
2.4. Hilbert polynomial

Let M be a finitely generated An –module provided with a B–filtration
Γ = (Mk )k . Let us denote by
gHFgrΓ (M ) : N −→ N
the Hilbert functionf of the grB (An )–module grΓ (M ). By definition we have

Mν
gHFgrΓ (M ) (ν) = dimC
Mν−1
for all ν ∈ N. Let us notice that by Definition 1.6 each Mν (and hence each
quotient Mν /Mν−1 ) is a finite dimensional vector space.
Theorem 2.1 (Hilbert, Serre). With the notation above, there exists
a unique polynomial gHPgrΓ (M ) (t) ∈ Q[t] such that gHFgrΓ (M ) (ν) =
gHPgrΓ (M ) (ν) for ν ∈ N, ν big enough. Furthermore, the degree of
gHPgrΓ (M ) (t) equals d−1 where d = dim V(AnngrB (An ) (grΓ (M ))) . More-
over the degree of gHPgrΓ (M ) (t) is less than or equal to 2n − 1.
Proof. See [44, Ch. VII, §12] or [30, Ch.1, Th. 7.5].
Definition 2.4. The polynomial gHPgrΓ (M ) (t) ∈ Q[t] is called the Hilbert
polynomial of the graded module grΓ (M ).
Remark 2.4. We will denote by

HFM,Γ : N −→ N
the map defined by
ν
X
HFM,Γ (ν) = gHFgrΓ (M ) (k)
k=0
for all ν ∈ N.
By induction on ν and using the exact sequence
Mν
0 → Mν−1 → Mν → →0
Mν−1
it is easy to prove that HFM,Γ (ν) = dimC (Mν ).
f Also called characteristic function in [44, Ch. VII, §12]

www.pdfgrip.com
82
If φ : Z → Z is any map, we denote by ∆φ : Z → Z the map defined by

∆φ(ν) = φ(ν + 1) − φ(ν).
By definition we have ∆HFM,Γ (ν) = gHFgrΓ (M ) (ν + 1).
Definition 2.5. A numerical polynomial is a polynomial Q(t) ∈ Q[t] (in

one variable t) such that Q(ν) ∈ Z for all ν ∈ Z, ν big enough.
Proposition 2.3.
i) If Q(t) ∈ Q[t] is a numerical polynomial, then there are integers
c0 , . . . , cd such that
d
X
t
Q(t) = ck
k
k=0
where

t t(t − 1) · · · (t − k + 1)
= .
k k!
ii) If φ : Z → Z is any map, and if there exists a numerical polynomial
Q(t) such that ∆φ(ν) = Q(ν) for ν ∈ N, ν big enough, then there exists
a numerical polynomial R(t) such that φ(ν) = R(ν) for ν ∈ N, ν big
enough. Moreover, if the leading term of Q(t) is ad td then the leading
ad d+1
term of R(t) is d+1 t .
Proof. See [44, Ch. VII, §12] or [30, Ch.1, Proposition 7.3].
Corollary 2.1. Let M be a finitely generated An –module provided with a

B–filtration Γ = (Mk )k . There exists a unique polynomial HPM,Γ (t) ∈ Q[t]
such that HFM,Γ (ν) = HPM,Γ (ν) for all ν ∈ N, ν big enough. Moreover, the
degree of HPM,Γ (t) equals the Krull dimension of VC (AnngrB (An ) (grΓ (M )))
and hence it is less than or equal to 2n.
Proof. It follows from Theorem 2.1, Proposition 2.3 and the fact that
∆HFM,Γ (ν) = gHFgrΓ (M ) (ν + 1) for all ν ∈ N.
Definition 2.6. The polynomial HPM,Γ (t) ∈ Q[t] is called the Hilbert
polynomial of M with respect to the B–filtration Γ.

two good B–filtrations Γ = (Mk )k and Γ0 = (Mk0 )k . Then the leading
terms of HPM,Γ (t) and HPM,Γ0 (t) coincide. In particular deg(HPM,Γ (t)) =
deg(HPM,Γ0 (t)).
www.pdfgrip.com
83
Proof. Assume that

HPM,Γ (t) = ad td + ( lower terms in t)
and
0
HPM,Γ0 (t) = a0d0 td + ( lower terms in t)
with ad and a0d0 nonzero. By Proposition 1.17 there exists k2 ∈ N such that
for all k ∈ N we have Mk−k2 ⊂ Mk0 ⊂ Mk+k2 . Then for k big enough we
have
HPM,Γ (k − k2 ) ≤ HPM,Γ0 (k) ≤ HPM,Γ (k + k2 ).
Dividing by k d and taking the limit when k → ∞ we get d = d0 and
ad = a0d0 .
Definition 2.7. Let M be a finitely generated An –module. The multiplic-

ity of M is e(M ) = ad · d! where ad td is the leading term of the polynomial
HPM,Γ (t) for a (or any) good B–filtration Γ on M . g
Remark 2.5. If M 6= (0) then the multiplicity e(M ) is a strictly positive

integer. It follows from Proposition 2.3 and the fact that HPM,Γ (k) ∈ N for
k big enough.
An important result relating the dimension of a finitely generated An –

module and the degree of its Hilbert polynomial, with respect to any good
B–filtration, is the following
Theorem 2.2 (Th. 3.1., Bernstein5 ). Let M be a finitely generated

An –module provided with a good B–filtration Γ. Then dim(M ) =
deg(HPM,Γ (t)).
Concerning the statement of this Theorem let us remark that while the
Hilbert polynomial HPM,Γ (t) and the multiplicity e(M ) are defined using
a good B–filtration Γ of M , the dimension of M , dim(M ), which is the
Krull dimension of the characteristic variety Char(M ), is defined using a
good F –filtration on M .
To each good B–filtration Γ = (Mk )k on a finitely generated An –module
M , we can also associate the algebraic variety defined in C2n by the homo-
geneous ideal AnngrB (An ) (grΓ (M )) of the graded ring grB (An ) (see Propo-
sition 1.9).
g The notation e(M ) appears in [6, Def. 1.1].

www.pdfgrip.com
84

Let us denote ∇(M ) := VC AnngrB (An ) (grΓ (M )) which is an homoge-
neous affine algebraic variety in the affine space C2n .
By Proposition 2.2 the algebraic variety ∇(M ) is independent of the
choice of the good B–filtration Γ on M .
The variety ∇(M ) can be different from the characteristic variety
Char(M ). The following is an example of this situation. Let’s consider
P = x21 + ∂12 in the Weyl algebra A1 (C) and define M = AA11P . Then
the characteristic variety of M is the line ξ1 = 0 in the plane C2 (with
coordinates x1 , ξ1 ) while ∇(M ) is V(x21 + ξ12 ) ⊂ C2 .
Exercise 2.1.
i) Let I = An P a proper principal ideal in An . Compute the dimension

dim(An /I) and the multiplicity e(An /I) (see Example 2.2).
ii) Prove that the dimension of the An –module C[x] is n and that its
multiplicity is 1.
iii) Prove that e(An ) = 1.
Quick answer.- i) We have dim(An /I) = dim(V(σ(P ))) = 2n − 1 (see

Example 2.2). The leading term of the Hilbert polynomial of the graded
grB (An )–module
00 grB (An )
grΓ (An /I) '
hσ T (P )i
d
is (2n−2)! t2n−2 where d = ordT (P ). Here Γ00 stands for the induced B–
filtration on An /I. Thus e(An /I) = d.
ii) We have an isomorphism C[x] ' An (∂A n
1 ,...,∂n )
(see Remark 1.2). It
B
is easy to prove that the ideal gr (An (∂1 , . . . , ∂n )) ⊂ C[x, ξ] is gen-
erated by (ξ1 , . . . , ξn ). The Hilbert polynomial
of the graded module
t+n−1
C[x, ξ]/hξ1 , . . . , ξn i ' C[x] is and then e(C[x]) = 1.
n−1
iii) It follows from thefact that theHilbert polynomial of the graded module
t + 2n − 1
C[x, ξ] ' grB (An ) is .
2n − 1
Theorem 2.3. Let M be a finitely generated An –module and N ⊂ M a

submodule. Then
i) dim(M ) = max{dim(N ), dim(M/N )}.

ii) If dim(N ) = dim(M/N ) then e(M ) = e(N ) + e(M/N ).
iii) dim(M ) ≤ 2n.
www.pdfgrip.com
85
Proof. Let us consider Γ = (Mk )k a good B–filtration on M and denote

by Γ0 and Γ00 the induced B–filtrations on N and M/N respectively (see
Subsection 1.9). From the proof of Proposition 1.13 we have HFM,Γ (k) =
HFN,Γ0 (k) + HFM/N,Γ00 (k) for all k ∈ N. Then, for k big enough, we have
HPM,Γ (k) = HPN,Γ0 (k) + HPM/N,Γ00 (k) and thus
HPM,Γ (t) = HPN,Γ0 (t) + HPM/N,Γ00 (t).
The last equality proves i) and ii). Part iii) follows from the very definition
of dim(M ) since the Krull dimension of any algebraic set in C2n is less than
or equal to 2n.
2.5. Bernstein’s inequality

Theorem 2.4 (Bernstein’s inequality). Let M be a nonzero finitely
generated An -module. Then dim(M ) ≥ n.
Proof. [A. Joseph’s proof].

Claim.- Let M be a finitely generated An –module and let (Mk )k be a
B–filtration with M0 6= 0. Then the C-linear map
φi : Bi →HomC (Mi , M2i )
defined by φi (P )(m) = P m is injective for all i ≥ 0.
Let’s assume the claim. Let m1 , . . . , m` be a finite system of generators
P
of M and Γ = (Mk )k the good B-filtration defined by Mk = j Bk mj . (see
P
Exercise 1.6). Then M0 = j Cmj 6= 0. From the claim we have
dimC (Bk ) ≤ dimC (HomC (Mk , M2k )) = dimC (Mk ) dimC (M2k ).
For k a big enough integer we have

2n + k
= dimC (Bk )) ≤ HPM,Γ (k)HPM,Γ (2k)
k
and so 2 deg(HPM,Γ (t)) ≥ 2n and dim(M ) ≥ n. Here HPM,Γ (t) is the
Hilbert polynomial of M with respect to the B–filtration Γ.
Let’s prove the claim by induction on i. For i = 0 we have B0 = C
and for λ ∈ C \ {0} the C-linear map φ0 (λ) is nonzero since M0 6= (0) (by
assumption). Assume i > 0 and the claim proved for i − 1. Let P ∈ Bi
nonzero. Assume P Mi = (0). Since Mi 6= (0) then P is non-constant.
Therefore, either there exists j ∈ {1, . . . , n} such that xj appears in at
least one monomial in P or there exists k ∈ {1, . . . , n} such that ∂k appears
in at least one monomial in P . In the first case we have [P, ∂j ] 6= 0 and in
the second one we have [P, xk ] 6= 0.
www.pdfgrip.com
86
By Proposition 1.4 we have that [P, ∂j ] and [P, xk ] belong to Bi−1 .

Moreover
[P, ∂j ]Mi−1 ⊂ P ∂j Mi−1 + ∂j P Mi−1 = (0)
and analogously [P, xk ]Mi−1 = (0). Then by induction hypothesis [P, ∂j ]
and [P, xk ] should be zero. Which is a contradiction. That proves the claim.
Exercise 2.2. Compute the dimension and the multiplicity of the quotient
An
An –module An (∂k+1 ,...,∂n ) for each k = 0, . . . , n − 1.
Quick answer.- Let’s write Ik = An (∂k+1 , . . . , ∂n ) and M (k) = AIkn . It is

easy to prove (e.g. using Buchberger’s algorithm in An , see Appendix A and
Remark 4.11) that grB (Ik ) = C[x, ξ](ξk+1 , . . . , xn ). The Hilbert polynomial
of the graded quotient module
grB (An ) C[x, ξ]
B
'
gr (Ik ) C[x, ξ](ξk+1 , . . . , xn )

t+n+k−1
is . Then the leading coefficient of the Hilbert polynomial
n+k−1
An tn+k
HP (t) of the An –module Ik is (n+k)! . So, dim(M (k) ) = n+k and e(M (k) ) =
1.
2.6. Holonomic An –modules

Definition 2.8. A finitely generated An –module M is said to be holonomic
if either M = (0) or dim(M ) = n.
Remark 2.6. For P ∈ An \ C the quotient An /An P is holonomic if and
only if n = 1 (see Example 2.2).
Example 2.3.
(1) Let I be a proper ideal in A1 and P a nonzero element in I. Let us
write J = A1 P . Let us consider the exact sequence of finitely generated
A1 –modules
0 −→ I/J −→ A1 /J −→ A1 /I −→ 0.
The quotient A1 /J is holonomic (see Remark 2.6). By applying Theo-
rems 2.3 and 2.4 we get that A1 /I is holonomic. However, the ideal I is
not holonomic (considered as A1 –module): otherwise, using the exact
sequence of An –modules
0 −→ I −→ A1 −→ A1 /I −→ 0
www.pdfgrip.com
87
one gets that A1 is also holonomic and this is not true because dim A1 =
2.
(2) Assume that M is a finitely generated A1 –module, say M =
Pr
`=1 A1 m` , for some m1 , . . . , mr ∈ M . Then M is the sum of modules
of type A1 /I` where I` = AnnA1 (m` ). From Theorem 2.3 we have that
M is holonomic if and only if all the I` are nonzero.
(3) The An –module C[x] = C[x1 , . . . , xn ] is isomorphic to An (∂A n
1 ,...,∂n )
and
then it is holonomic (see Exercise 2.1).
Theorem 2.5.
(1) Let M be a finitely generated An –module and N a submodule of M .
Then M is holonomic if and only if N and M/N are holonomic. If M
is holonomic then e(M ) = e(N ) + e(M/N ).
P
(2) If M` , ` = 1, . . . , r is a holonomic An –module then ` M` is holonomic
and
r
X
e(⊕r`=1 M` ) = e(M` ).
`=1
Proof. (1) There is nothing to prove if N = (0) or N = M . Assume

N is a proper submodule of M and M 6= (0). From Theorem 2.3 we get
dim(M ) = max{dim(N ), dim(M/N )} and therefore if N and M/N are
holonomic then M is also holonomic.
From Theorem 2.4 we have dim(N ) ≥ n and dim(M/N ) ≥ n.
Assume M is holonomic. Then we have n = dim(M ) = dim(N ) =
dim(M/N ). Again applying Theorem 2.3 we get e(M ) = e(N ) + e(M/N ).
Part (2) follows from (1) by induction on r.
Theorem 2.6. Let M be a holonomic An –module. Then we have:

(1) M is a torsion module, i.e. for each m ∈ M there exists P ∈ An ,
P 6= 0, such that P m = 0.
(2) M is an Artinian module of finite length. Moreover, the length of M is
less or equal than e(M ).
Proof. (1) We can assume M 6= (0). Take m ∈ M , m 6= 0. Let us consider

the morphism of An –modules
φ : An −→ M
defined by φ(P ) = P m. The image of φ, Im(φ), is a nonzero An –module
(since m ∈ Im(φ) ⊂ M ) and, moreover, it is holonomic (see Theorem 2.5).
Since An is non-holonomic the kernel ker(φ) is nonzero.
www.pdfgrip.com
88
(2) Let M = M0 ⊃ M1 ⊃ M2 ⊃ · · · a decreasing chain of An –

submodules of M . By Theorem 2.5 each Mi is holonomic. If there exists
i such that Mi = (0) then the chain is stationary. Assume the chain non-
stationary and that each Mi is nonzero. From the exact sequence
0 −→ Mi+1 −→ Mi −→ Mi /Mi+1 −→ 0
we get that e(Mi ) = e(Mi+1 ) + e(Mi /Mi+1 ) (see Theorem 2.5). Then we
Pr
have e(M ) = e(Mr+1 ) + i=0 e(Mi /Mi+1 ) ≥ r + 1, for each r ≥ 0. This
is a contradiction. Moreover, the length of M should be less or equal than
e(M ).
Remark 2.7. There are finitely generated An –modules of finite length

–and even irreducible– which are non-holonomic.
An example of that –due to J.T. Stafford– is the following. Consider
M = A2 (C)/A2 (C)P with P = x2 ∂1 ∂2 − ∂2 + x1 + x2 . We have that
dim(M ) = 3 and then M is non-holonomic (see Example 2.2).
J. T. Stafford proved that M is irreducible as A2 –module [41, Th. 1.1].
Exercise 2.3. Prove that if a holonomic An –module has multiplicity 1 then

it is irreducible. Prove that C[x] is irreducible.
Quick answer.- Assume M is a holonomic An –module with e(M ) = 1 and

consider N ⊂ M a nonzero submodule of M . By Theorem 2.5 we have
1 = e(M ) = e(N ) + e(M/N ). Since the integer e(N ) is strictly positive (see
Remark 2.5) we have e(N ) = 1 and e(M/N ) = 0. The last equality implies
M = N . So M is irreducible. The last part of the exercise follows since C[x]
is holonomic and e(C[x]) = 1 (see Exercise 2.1).
Proposition 2.5 (Cor. 1.4, Bernstein6 ). Let M be an An –module en-

dowed with a B–filtration Γ = (Mk )k such that there are two rational num-
bers c1 , c2 satisfying
c1 n
dimC Mj ≤ j + c2 (j + 1)n−1
n!
for j ∈ N, j big enough. Then M is finitely generated. Moreover, it is
holonomic and e(M ) ≤ c1 .
Proof. First of all we will prove that any nonzero finitely generated sub-
module N of M is holonomic and e(N ) ≤ c1 . Since N is finitely generated
it admits a good B–filtration say ΓN = (Nk )k . Consider the B–filtration
www.pdfgrip.com
89
Γ0 = (Mk ∩ N )k . By Proposition 1.17 there exists a positive integer r such

that Nj ⊂ Mj+r ∩ N for all j. We have
c1
dimC (Nj ) ≤ dimC (Mj+r ) ≤ (j + r)n + c2 (j + r + 1)n−1
n
and then the degree of the Hilbert polynomial of N with respect to ΓN is
less than or equal to n. Applying 2.4 we get dim(N ) = n and then N is
holonomic. Moreover, from the previous inequality we also get e(N ) ≤ c1 .
We will prove now that M is finitely generated. If M is nonzero let us
consider a nonzero element m1 ∈ M . Denote M1 = An m1 . If M 6= M1
let us consider a nonzero m2 ∈ M \ M1 and denote M2 = An m1 + An m2 .
Assume we can construct an infinite increasing chain
M1 ⊂ M 2 ⊂ M 2 ⊂ · · · ⊂ M i ⊂ · · ·
of finitely generated submodules of M . From the first part of the proof
we deduce that each Mi is holonomic and e(Mi ) ≤ c1 . We also have that
e(Mi ) ≥ i for each i, which is a contradiction. So there exists a finite
generating set m1 , . . . , mr of M .
2.7. C[x]f is holonomic

Let f be a nonzero polynomial in C[x].
Theorem 2.7 (§2, Bernstein6 ). The An –module C[x]f is holonomic.
Proof. Put N = C[x]f and deg(f ) = d ≥ 0.

For each k ∈ N define
Nk = {g/f k ∈ N | deg(g) ≤ (d + 1)k}.
Let’s prove first that the family Γ = (Nk )k is a B–filtration on N .
It’s clear that Nk ⊂ N` for k ≤ `.
Assume g/f k ∈ Nk . We have deg(xi g) = deg(g) + 1 ≤ (d + 1)k + 1 ≤
(d + 1)(k + 1). That proves the inclusion xi Nk ⊂ Nk+1 . We also have

g ∂i (g)f − kg∂i (f )
∂i k
=
f f k+1
and deg(∂i (g)f −kg∂i (f )) ≤ d+deg(g)−1 ≤ d−1+(d+1)k ≤ (d+1)(k+1).
That proves ∂i Nk ⊂ Nk+1 . Then B1 Nk ⊂ Nk+1 . Since B` = (B1 )` we have
B` Nk ⊂ Nk+` .
We will now prove N = ∪k Nk . To this end take g/f k ∈ N and assume
deg(g) = m. We have
www.pdfgrip.com
90
g gf m
k
= k+m
f f
and deg(gf m ) = m + dm ≤ (d + 1)(k + m). That proves g/f k ∈ Nm+k .
We have proved that (Nk )k is a B–filtration on N = C[x]f .
We will now prove that this filtration satisfies the hypothesis of Propo-
sition 2.5 for adequate c1 , c2 .
The dimension of the C–vector space Nk is bounded by the number of
monomials xα in C[x] with degree |α| ≤ (d + 1)k. This number is

(d + 1)k + n 1
= (d + 1)n k n + p(k)
n n!
where p(t) is polynomial in t with rational coefficients and degree less than
or equal to n − 1. Then there exists an integer number c2 > 0 such that

(d + 1)k + n 1 (d + 1)n k n
dimC (Nk ) ≤ = (d+1)n k n +p(k) ≤ +c2 (k+1)n−1
n n! n!
for k 0. Then by Proposition 2.5 C[x]f is holonomic.
Remark 2.8. From the above proof we can also deduce, applying Propo-
sition 2.5, that the multiplicity of C[x]f is bounded by (d + 1)n . This bound
is far to be sharp. See Exercise 2.4.
Exercise 2.4. Let us write f = x1 . Prove that:

i) C[x]f = An f1 .
ii) AnnAn (1/f ) = An (x1 ∂1 + 1, ∂2 , . . . , ∂n ).
iii) dim(C[x]f ) = n, e(C[x]f ) = 2.
Answer.- Recall that the annihilating ideal AnnAn (1/f ) is by definition the
ideal {P ∈ An | P (1/f ) = 0}.
i) By definition we have that An f1 ⊂ C[x]f . The equality

1 (−m)g
g∂1 = m+1
xm
1 x1
holds for any g ∈ C[x] and any m ∈ N. This equality proves that any
g
rational function of type xm+1 belongs to An x11 and then it proves the
1
equality C[x]f = An f1 .
ii) The inclusion An (x1 ∂1 + 1, ∂2 , . . . , ∂n ) ⊂ AnnAn (1/f ) is obvious. Let
us consider an operator P = P (x, ∂) ∈ An annihilating x11 . We can write
P = Q 2 ∂2 + · · · + Q n ∂n + P 1
www.pdfgrip.com
91
P
for some Q2 , . . . , Qn , P1 ∈ An and P1 = ` a` (x)∂1` for some a` (x) ∈ C[x].
The operator P1 annihilates x11 since P also does.
We can write
P1 = Q(x1 ∂1 + 1) + S(x0 , ∂1 ) + r(x)

P
for some Q, S(x0 , ∂1 ) ∈ An , r(x) ∈ C[x] and S := S(x0 , ∂1 ) = k>0 bk (x0 )∂1k
for some bk (x0 ) ∈ C[x0 ] := C[x2 , . . . , xn ].
We have 0 = P1 ( x11 ) = S( x11 ) + r(x)x1 . Assuming that S is nonzero, let us
write d > 0 the degree of S with respect to ∂1 . The order of the pole of S( x11 )
at x1 = 0 is d + 1 while r(x) x1 has a pole of order at most 1. This implies
that d = 0 which is a contradiction. Then we have S = 0 and r(x) = 0
since r(x)
x1 = 0. This proves that P = Q2 ∂2 + · · · Qn ∂n + Q(x1 ∂1 + 1) ∈
An (x1 ∂1 + 1, ∂2 , . . . , ∂n ).
iii) Let us denote I = An (x1 ∂1 + 1, ∂2 , . . . , ∂n ) and let us write J ⊂
C[x, ξ] the ideal generated by (x1 ξ1 , ξ2 , . . . , ξn ). We have the inclusion J ⊂
grF (I) and then the inclusion V(grF (I)) ⊂ V(J). The last affine algebraic
set has Krull dimension n. Then the Krull dimension of V(grF (I)) is less
than or equal to n. So, dim(An /I) = dim(V(grF (I))) ≤ n and then, from
Theorem 2.4, we get dim(An /I) = dim(C[x]f ) = n.
Let us now compute the multiplicity of C[x]f . We will prove first the
equality grB (I) = J. It is easy to prove that each nonzero P ∈ I can be
written as
P = Q2 ∂2 + · · · + Qn ∂n + Q(x1 ∂1 + 1)
with ordT (Q) = ordT (P ) − 2 and ordT (Qi ) = ordT (P ) − 1 for i = 2, . . . , n

(see, e.g. Theorem A.1).
Then, by Proposition 1.4, we have
n
X
σ T (P ) = σ T (Qi )ξi + σ T (Q)x1 ξ1
i=2
and then σ T (P ) ∈ J. This proves the equality grB (I) = J. The leading term
of the Hilbert polynomial gHPgrΓ (C[x]f ) (t) of the quotient graded module
C[x, ξ]
' grΓ (C[x]f )
J
2
equals (n−1)! tn−1 (here Γ denotes the induced B–filtration on An /I '
C[x]f ). This proves that e(C[x]f ) = 2.
www.pdfgrip.com
92
2.8. The Bernstein polynomial

The Bernstein (or the Bernstein-Sato) polynomial associated with a given
polynomial f ∈ C[x] has been introduced in its general form in Ref. 6 (and
independently by M. Sato in Ref. 38).
Let f be a nonzero polynomial in C[x]. Let s be a new variable and
C(s) the field of rational functions on s. We denote by An (s) the Weyl
algebra over the field C(s) and An [s] := An ⊗C C[s]. Denote by C(s)[x]f f s
the free C(s)[x]f –module of rank 1 with basis the formal symbol f s . This
free module admits a natural structure of left An (s)–module by defining
∂i f s = sf −1 ∂i (f )f s
for i = 1, . . . , n (the action of C(s)[x] being the natural one).
Proposition 2.6 (§2, Bernstein6 ). The An (s)–module C(s)[x]f f s is

holonomic.
Proof. Put N = C(s)[x]f f s and deg(f ) = d ≥ 0.

For each k ∈ N define

g(s, x)
Nk = ∈ N | deg(g) ≤ (d + 1)k .
fk
It can be proved, in a similar way to the proof of Proposition 2.7, that the
family Γ = (Nk )k is a B–filtration on N .
We will now prove that this filtration satisfies the hypothesis of Propo-
sition 2.5 for adequate c1 , c2 .
The dimension of the C(s)–vector space Nk is bounded by the number
of monomials xα in C(s)[x] with degree |α| ≤ (d + 1)k. This number is

(d + 1)k + n 1
= (d + 1)n k n + p(k)
n n!
where p(t) is a polynomial in t with rational coefficients and degree less
than or equal to n − 1. Then there exists a integer c2 > 0 such that

(d + 1)k + n 1
dimC(s) (Nk ) ≤ = (d + 1)n k n + p(k)
n n!
(d + 1)n k n
≤ + c2 (k + 1)n−1
n!
for k ∈ N, k big enough. Then by Proposition 2.5 the An (s)–module N =
C(s)[x]f f s is holonomic.
www.pdfgrip.com
93
Theorem 2.8. Let f be a nonzero polynomial in C[x]. There exists a

nonzero polynomial b(s) ∈ C[s] and a differential operator P (s) ∈ A n [s]
such that the equality
P (s)f f s = b(s)f s
holds in C(s)[x]f f s .
Proof. The module An (s)f s is an An (s)–submodule of C(s)[x]f f s and

then, by Proposition 2.6, it is holonomic and furthermore of finite length
(see Theorems 2.5 and 2.6). Then the descending sequence
An (s)f s ⊇ An (s)f f s ⊇ · · · ⊇ An (s)f ` f s ⊇ · · ·
is stationary. Thus there exists ` ∈ N such that
f ` f s ∈ An (s)f `+1 f s .
So, there exists Q(s) ∈ An (s) such that f ` f s = Q(s)f `+1 f s . Then f s =
Q(s − `)f f s . Let b(s) ∈ C[s] a nonzero polynomial such that P (s) :=
b(s)Q(s − `) ∈ An [s]. Thus we have b(s)f s = P (s)f f s .
For a given nonzero polynomial f in C[x] the set of polynomials c(s) ∈

C[s] such that there exists an operator P (s) ∈ An [s] such that P (s)f f s =
c(s)f s is an ideal in C[s]. We will denote this ideal by Bf .
Definition 2.9. Let f be a nonzero polynomial in C[x]. The monic gen-

erator of the ideal Bf is denoted by bf (s) and it is called the Bernstein
polynomial (or the Bernstein-Sato polynomial) of f .
The computation of bf (s) is difficult although there exists an algorithm

computing the Bernstein polynomial bf (s) for a given polynomial f ∈ C[x]
(see T. Oaku;36 see also M. Noro34 ).
A variant of this algorithm has been implemented in the D-modules
package for Macaulay 2. We can use this implementation to make some
experiments.
Macaulay 2, version 1.2 with packages: Elimination, IntegralClosure,
LLLBases, PrimaryDecomposition, ReesAlgebra, SchurRings, TangentCone
i1 : load "D-modules.m2"
i2 : R=QQ[x,y,z];
i3 : W=makeWA R;
i4 : f=x^2+y^2+z^2;
i5 : globalBFunction f
www.pdfgrip.com
94
2
o5 = 2s + 5s + 3
o5 : QQ[s]
The previous command globalBFunction f computes the Bernstein

polynomial of the given polynomial f. We have to notice here that, in order
to simplify the output, Macaulay 2 clears the denominators of the Bern-
stein polynomial (and so, the output of globalBFunction is not necessarily
monic).
The previous script tells us that the Bernstein polynomial of x21 +x22 +x23
is just the polynomial s2 + (5/2)s + 3/2.
Let us continue with the following computation
i6 : f=x^3+y^3+z^3;
5 4 3 2
o7 = 9s + 63s + 173s + 233s + 154s + 40
o7 : QQ[s]
i8 : factor o7
2
o8 = (s + 1) (s + 2)(3s + 4)(3s + 5)
The command factor factorizes the given polynomial. Let us make

some other experiments
i9 : g=x^2-y^3;
i10 : globalBFunction g
3 2
o10 = 36s + 108s + 107s + 35
o10 : QQ[s]
i11 : factor o13
o11 = (s + 1)(6s + 5)(6s + 7)
i12 : f=x^2-y^3+1;
o13 = s + 1
o13 : QQ[s]
The last two examples show that a little change on the expression of
the polynomial f can produce completely different Bernstein polynomials.
www.pdfgrip.com
95
The expansion of the polynomial g is very close to the one of f although

the corresponding Bernstein polynomials are very different.
For each polynomial f ∈ R the Bernstein polynomial bf (s) depends on
the singularities of the hypersurface V(f ) = {a ∈ Cn | f (a) = 0}. Bern-
stein polynomial is a very useful invariant in singularity theory. Two main
references in this topic are Malgrange33 and Kashiwara.31
Most of the material of this Section appears in the articles Bernstein6 and
Ehlers23 and in the books Björk7 and Coutinho.21
3. Logarithmic An –Modules
In this Sectionh unless otherwise stated, we will denote R = C[x] =
C[x1 , . . . , xn ].
3.1. Logarithmic derivations

Let us denote by DerC (R) the R–module of C-derivations of the ring R. An
element δ ∈ DerC (R) can be written as
n
X
δ= ai (x)∂i
i=1
for some ai (x) ∈ R. Moreover, DerC (R) is a free R–module of rank n the
set {∂1 , . . . , ∂n } being one of its bases. Elements in DerC (R) are also called
vector fields on Cn with polynomial coefficients.
If f ∈ R is not a constant we can define the notion of (global) logarithmic
derivation (or logarithmic vector field) with respect to the hypersurface
D = V(f ) = {a ∈ Cn | f (a) = 0} ⊂ Cn , as follows:
P
Definition 3.1 (K. Saito39 ). A vector field δ = ni=1 ai (x)∂i with coef-
ficients ai (x) ∈ R is said to be logarithmic with respect to D if δ(f ) ∈ Rf .
The R–module of logarithmic vector fields (or logarithmic derivations)

with respect to D ⊂ Cn is denoted by DerR (− log D). The R–module of
logarithmic vector fields with respect to D ⊂ Cn has been also denoted in
h Part of this Section follows the talk Computational methods for testing the range of
validity of the Logarithmic Comparison Theorem given by the author at the Workshop
Geometry and analysis on complex algebraic varieties held at RIMS, Kyoto University,
from 11th to 15th December 2006.
www.pdfgrip.com
96
the literature as DerR (log D). The sign before log D in our notation will be
explained latter (see Remark 3.4).
For each polynomial g ∈ R, we can also denote
n
X
DerR (− log g) = {δ = ai (x)∂i ∈ DerC (R) | δ(g) ∈ Rg}.
i=1
Exercise 3.1.
(1) Prove the equality DerR (− log gh) = DerR (− log g) ∩ DerR (− log h) for
g, h ∈ R.
(2) Prove that if f, g ∈ R and V(f ) = V(g) then DerR (− log f ) =
DerR (− log g). (Hint: Use Nullstellensatz).
Exercise 3.1 justifies the notation DerR (− log D) for the R–module of
logarithmic vector fields with respect to the hypersurface D = V(f ).
Exercise 3.2. Prove that DerR (− log D) is a Lie algebra (i.e. prove that it
is closed under the Lie bracket [−, −]).
Let us write fi = ∂i (f ) for i = 1, . . . , n. To each logarithmic derivation

P δ(f )
δ = i ai (x)∂i one can associate the syzygy (− f , a1 (x), . . . , an (x)) of
the polynomials (f, f1 , . . . , fn ). This defines a map
: DerR (− log D) −→ SyzR (f, f1 , . . . , fn )
where SyzR (f, f1 , . . . , fn ) is the R–module of syzygies of (f, f1 , . . . , fn ).
Exercise 3.3. Prove that the previous map is an isomorphism of R–

modules.
Remark 3.1. As R is a Noetherian ring then SyzR (f, f1 , . . . , fn ) and

DerR (− log D) are finitely generated R-modules.
Moreover, for each f ∈ R one can compute, by using Groebner bases in
R, a system of generators of the syzygy module SyzR (f, f1 , . . . , fn ) (and
then of the R–module DerR (− log D) by using the isomorphism ) (see
e.g. [1, Section 3.4]).
Let’s treat the case f = x2 − y 3 in Macaulay 2. In the following script,
R denotes the polynomial ring with variables x, y and with coefficients if
the field of rational numbers (which is denoted as QQ in Macaulay 2).
Input i5 : computes a system of generators of the syzygy module of
∂f ∂f
(f, ∂x , ∂y ). Notice that output o5 : gives a submodule of the free module
W3 and that this submodule is generated by vectors in R3 . Thus, these two
vectors also generate the R–module of syzygies of (f, ∂f ∂f
∂x , ∂y ).
www.pdfgrip.com
97
Input i6 : computes a (1 × 2)–matrix which entries are the logarithmic

vector fields with respect to f associate to each previously computed syzygy.
Input i7 : gives the ideal in the Weyl algebra W generated by these two
logarithmic vector fields.

i2 : R=QQ[x,y];
i3 : W=makeWA R;
i4 : f=x^2-y^3
3 2
o4 = - y + x
o4 : W
i5 : kernel matrix({{f,diff(x,f),diff(y,f)}})
o5 = image {3} | 6 0 |
{1} | -3x -3y2 |
{2} | -2y -2x |
3
o5 : W-module, submodule of W
i6 : matrix({{0,dx,dy}}) * gens o5
o6 = | -3xdx-2ydy -3y2dx-2xdy |
1 2
o6 : Matrix W <--- W
i7 : ideal o6
2
o7 = ideal (- 3x*dx - 2y*dy, - 3y dx - 2x*dy)
o7 : Ideal of W
Exercise 3.4. Prove that one has an exact sequence of R–modules
DerR (− log D)
0 → f DerC (R) → DerR (− log D) → →0
f DerC (R)
where the morphism f DerC (R) → DerR (− log D) is an inclusion.
Exercise 3.5.
(1) Assume f = x1 ∈ R and D = V(x1 ) ⊂ Cn . Prove the equality

DerR (− log D) = Rx1 ∂1 ⊕ R∂2 ⊕ · · · ⊕ R∂n .
www.pdfgrip.com
98
(2) Assume f = x1 x2 · · · xr for some 1 ≤ r ≤ n, and D = V(f ) ⊂ Cn .

Prove the equality DerR (− log D) = Rx1 ∂1 ⊕ Rx2 ∂2 ⊕ · · · ⊕ Rxr ∂r ⊕
R∂r+1 ⊕ · · · ⊕ R∂n .
To each nonzero polynomial f ∈ C[x] we associate the quotient An –

module defined as
An
M log f := .
An Der(− log f )
For example, if f = x1 x2 · · · xr then
An
M log f = .
An (x1 ∂1 , x2 ∂2 , . . . , xr ∂r , ∂r+1 , . . . , ∂n )
(1)
3.2. The ideal AnnAn ( f1 )
Let f be a nonzero polynomial in R. We denote
g log f ) := {δ + δ(f ) | δ ∈ Der(− log f )}.

Der(−
f
)
For each δ ∈ Der(− log f ), the operator δ + δ(f
f annihilates the rational
)
function f1 . We notice here that the operator δ + δ(ff has order 1 (see
Definition 1.2).
Reciprocally, if an operator P ∈ An annihilates f1 and ord(P ) = 1 then
we can write
P = η + a0 (x)
Pn
for some polynomial a0 (x) ∈ R and some derivation η = i=1 ai (x)∂i (with
ai (x) in R for i = 1, . . . , n). Then, from P (1/f ) = 0 we get f a0 (x) = η(f ).
)
So, η ∈ Der(− log f ) and P = η + η(f f .
(1)
We denote by AnnAn ( f1 ) (or simply Ann(1) ( f1 )) the left ideal in An
)
generated by the operators of the form δ + δ(f
f for some δ in Der(− log f ).
That is:

1 g log f ).
Ann(1) = An Der(−
f
To each nonzero polynomial f ∈ C[x] we have associated (see Subsection
3.1) the quotient An –module
An
M log f := .
An Der(− log f )
www.pdfgrip.com
99
Moreover, to the polynomial f ∈ R we can also associate a new quotient

An –module:
flog f := An An
M = .
g log f )
An Der(− Ann(1) (1/f )
Both modules M log f and M flog f will be called the logarithmic An –

modules associated with the polynomial f ∈ R. These modules encode
information about the singularities of the hypersurface V(f ) ⊂ Cn defined
by f ∈ R.
Since Ann(1) ( f1 ) is included in the annihilating ideal Ann( f1 ) we have
a natural surjective morphism of An –modules
flog f = An φf An
M −→
Ann(1) (1/f ) Ann(1/f )
defined by

1 1
φf P + Ann(1) = P + Ann( ).
f f
The morphism φf is an isomorphism if and only

1 1
Ann(1) = Ann
f f
(we say in this case that the annihilating ideal of 1/f is generated by op-
erators of order 1).
It is an open question to characterize the class of polynomials f ∈ R
such that the Ann(1/f ) is generated by operators of order 1.
Annihilating ideals can be computed using Groebner bases in An (see
Oaku and Takayama37).
We will use Macaulay 2 to compute some examples.
The next script computes a system of generators for the annihilating
ideal Ann(1/f ) ⊂ A2 for f = x41 +x52 . The computation gives two generators
for Ann(1/f ), namely P = 5x1 ∂1 + 4x2 ∂2 + 20 and Q = 5x42 ∂1 − 4x31 ∂2 . So,
the annihilating ideal of 1/f is generated by operators of order 1 and then,
in this case, the morphism
A2 A2
φf : −→
Ann(1) (1/f ) Ann(1/f )
is an isomorphism.
www.pdfgrip.com
100
In the next script the command RatAnn f computes a generating system

for the annihilating ideal Ann(1/f ).
i2 : R=QQ[x,y];
i3 : W=makeWA R;
i4 : f=x^4+y^5
5 4
o4 = y + x
o4 : W
i5 : RatAnn f
4 3
o5 = ideal (5x*dx + 4y*dy + 20, 5y dx - 4x dy)
o5 : Ideal of W
We will change slightly the polynomial f just by adding the monomial

x1 x42 . So, we get a new polynomial g = x41 + x52 + x1 x42 .
We are going to use Macaulay 2 to compute, using the following script,
a system of generators of the rational function 1/g.
This computation is performed using the command Ann=RatAnn g. The
value of the string Ann is then a system of generators of the annihilating
ideal Ann(1/g).
In order to easily manage the output o7 we use the command toString
Ann. Output o8 gives a string representation of the previous system of
generators of the annihilating ideal Ann(1/g). This system does not have
to be minimal. The fourth generator has order 2. To prove that Ann(1/g)
is not generated by operators of order 1 we will compute generators of
Ann(1) (1/g). This is performed from input i9 until input i11.
The value of the name Ann1, in the next script, is then a system of gen-
erators of the ideal Ann(1) (1/g). Then we compare the ideals Ann(1) (1/g)
and Ann(1/g), by using the command Ann==Ann1. The corresponding out-
put being false means that the inclusion Ann(1) (1/g) ⊂ Ann(1/g) is strict
and then the morphism
A2 A2
φg : −→
Ann(1) (1/g) Ann(1/g)
is not an isomorphism.
www.pdfgrip.com
101
i6 : g=f+x*y^4
4 5 4
o6 = x*y + y + x
o6 : W
i7 : Ann=RatAnn g;
i8 : toString Ann
o8 = ideal(4*x^2*dx+5*x*y*dx+3*x*y*dy+4*y^2*dy+16*x+20*y,
16*x*y^2*dx+4*y^3*dx+12*y^3*dy-125*x*y*dx-4*x^2*dy+5*x*y*dy-100*y^2*dy+64*y^2-500*y,
4*x*y^3*dx+5*y^4*dx-y^4*dy-4*x^3*dy,
-64*x^2*y*dx^2+36*y^3*dx^2-96*x*y^2*dx*dy-32*y^3*dx*dy-36*y^3*dy^2+500*x^2*dx^2
+125*x*y*dx^2-36*x^2*dx*dy+720*x*y*dx*dy+100*y^2*dx*dy+24*x^2*dy^2-29*x*y*dy^2
+260*y^2*dy^2-368*x*y*dx-72*y^2*dx-264*y^2*dy+2425*x*dx+625*y*dx-105*x*dy
+1495*y*dy-192*y-300)
i9 : kernel matrix({{g,diff(x,g),diff(y,g)}})
o9 = image {5} | -16x-20y -16y2-100x |

{4} | 4x2+5xy 4xy2+y3+25x2 |
{4} | 3xy+4y2 3y3-x2+20xy |
3
i10 : matrix({{-1,dx,dy}}) * gens o9
o10 = | 4x2dx+5xydx+3xydy+4y2dy+16x+20y 4xy2dx+y3dx+3y3dy+25x2dx-x2dy+20xydy+16y2+100x |
1 2
i11 : Ann1=ideal o10
2 2 2 3 3 2 2
o11 = ideal (4x dx + 5x*y*dx + 3x*y*dy + 4y dy, 4x*y dx + y dx + 3y dy + 25x dx - x dy + 20x*y*dy)
o11 : Ideal of W
i12 : Ann==Ann1
o12 = false
Notice that the input i9 computes a generating system of the syzygy

∂g ∂g
module SyzR (g, ∂x , ∂y ) while the input i10 computes a generating system
(1)
of the ideal Ann (1/g).
It is not easy to determine when the modules M log f and M flog f are
holonomic.
For any non constant polynomial f ∈ C[x1 , x2 ] both A2 –modules M log f
and M flog f are holonomic (see Calderón11 ).
The A3 –modules M log f and M flog f are not holonomic for f = (xz +
4 5 4
y)(x + y + xy ) (see [20, Example 6.4.]). Let us compute the dimension
of these two A3 –modules using Macaulay 2.
www.pdfgrip.com
102
i2 : R=QQ[x,y,z]; W=makeWA R;
i3 : f=(x*z+y)*(x^4+y^5+x*y^4)
2 4 5 5 6 5 4
o3 = x y z + x*y z + x*y + y + x z + x y
o3 : W
i4 : kernel matrix({{f,diff(x,f),diff(y,f),diff(z,f)}})
o4 = image {7} | -76x-96y x -xz-19x-24y -2y2z-38y2-150x+120y |

{6} | 16x2+20xy 0 4x2+5xy 8xy2+2y3+30x2-25xy |
{6} | 12xy+16y2 0 3xy+4y2 6y3-2x2+25xy-20y2 |
{6} | -4xz-4yz -xz-y xz2-xz 2y2z2-2y2z+5yz+2x+5y |
4
i5 : matrix({{0,dx,dy,dz}})* gens o4;
o5 = | 16x2dx+20xydx+12xydy+16y2dy-4xzdz-4yzdz -xzdz-ydz xz2dz+4x2dx+5xydx+3xydy+4y2dy-xzdz

--------------------------------------------------------------------------------------
2y2z2dz+8xy2dx+2y3dx+6y3dy-2y2zdz+30x2dx-25xydx-2x2dy+25xydy-20y2dy+5yzdz+2xdz+5ydz |
1 4
i6 : Ilog=ideal o5;
i7 : toString o6
o7 = matrix {{16*x^2*dx+20*x*y*dx+12*x*y*dy+16*y^2*dy-4*x*z*dz-4*y*z*dz, -x*z*dz-y*dz,

x*z^2*dz+4*x^2*dx+5*x*y*dx+3*x*y*dy+4*y^2*dy-x*z*dz,
2*y^2*z^2*dz+8*x*y^2*dx+2*y^3*dx+6*y^3*dy-2*y^2*z*dz+30*x^2*dx-25*x*y*dx-2*x^2*dy
+25*x*y*dy-20*y^2*dy+5*y*z*dz+2*x*dz+5*y*dz}}
i8 : dim Ilog
o8 = 4
In the previous script, input i4 : computes a generating system of the

syzygy module SyzR (f, ∂f ∂f ∂f
∂x , ∂y , ∂z ) and inputs i5 : to i7 : compute a
system of generators of the ideal A3 Der(− log f ), which is denoted here as
Ilog. Notice that W stands for the Weyl algebra A3 . As dim Ilog is 4 that
means that the A3 –module
A3
M log f =
A3 Der(− log f )
is not holonomic.
In the next script we will first compute a system of generators of the
ideal Ann(1) (1/f ) for the same f = (xz + y)(x4 + y 5 + xy 4 ). The value of
the string Ann1 is a generating system of the ideal Ann(1) (1/f ).
www.pdfgrip.com
103
i9 : matrix({{-1,dx,dy,dz}})* gens o4
o9 =
| 16x2dx+20xydx+12xydy+16y2dy-4xzdz-4yzdz+76x+96y -xzdz-ydz-x xz2dz+4x2dx+5xydx+3xydy+4y2dy-xzdz+xz+19x+24y
-----------------------------------------------------------------------------------------------------------
2y2z2dz+8xy2dx+2y3dx+6y3dy-2y2zdz+2y2z+30x2dx-25xydx-2x2dy+25xydy-20y2dy+5yzdz+38y2+2xdz+5ydz+150x-120y |
1 4
i10 : Ann1=ideal o9;
i11 : dim Ann1
o11 = 4
As the dimension of the ideal Ann1 is 4 then the A3 –module
flog f = A3
M (1)
Ann (1/f )
is not holonomic.
3.3. Logarithmic differential forms

For each q ∈ N, let us denote by ΩqR the R–module of differential q-forms
with coefficients in R and let us write ΩqR,f = Rf ⊗R ΩqR the R–module of
rational differential q–forms with poles along D = V(f ).
Both Ω•R and Ω•R,f are complexes of C–vector spaces once endowed with
the exterior derivative and in fact Ω•R is a subcomplex of Ω•R,f .
If D = V(f ) = V(g) then Ω•R,f = Ω•R,g and this complex (endowed with
the exterior derivative) is called the complex of (global) rational differential
forms with respect to D and it is denoted by Ω•R (∗D) (or simply Ω• (∗D)).
Remark 3.2. The natural map
DerC (R) × Ω1 (∗D) −→ Rf

P P P
defined by (δ = i ai (x)∂i , ω = i gi dxi ) 7→ ai gi ∈ Rf is R–bilinear
and non-degenerate.
Definition 3.2 (K. Saito39 ). A rational differential form ω ∈ Ωq (∗D)

is said to be logarithmic with respect to D if ω and dω have at most a
first-order pole along D.
The R–module of logarithmic q-forms with respect to D is denoted by

ΩqR (log D).
If f ∈ R is a reduced equation of D then ω ∈ Ωq (∗D) is logarithmic
with respect to D if and only if f ω ∈ ΩqR and f dω ∈ Ωq+1
R .
www.pdfgrip.com
104
Exercise 3.6. Prove that ω ∈ Ωq (∗D) is logarithmic with respect to D

if and only if f ω ∈ ΩqR and df ∧ ω ∈ Ωq+1 R . In particular, prove that
P
ω = i bfi dxi ∈ Ω1R (log D) (with bi ∈ R) if and only if
bj bi
fi − fj ∈ R
f f
∂f df
for all 1 ≤ i, j ≤ n. Here fi := ∂xi . Prove that f ∈ Ω1 (log D).
Remark 3.3. We have a natural exact sequence

1 1
1 1 f ΩR
0 → Ω1R (log D) → ΩR → 1 →0
f ΩR (log D)
and the last quotient is a torsion R–module.
Ω•R (log D) is a sub-complex of Ω•R (∗D). The corresponding inclusion,

which is a morphism of complexes of vector spaces, is denoted by
ιD : Ω•R (log D) → Ω•R (∗D).
Proposition 3.1. ΩqR (log D) is a finitely generated R–module for all q.
Proof. Recall that f ∈ R is a reduced equation for the hypersurface D =

V(f ) ⊂ Cn . We have the inclusion
1 q
ΩqR (log D) ⊂ Ω
f R
and f1 ΩqR is a free R–module with basis { f1 dxi1 ∧ · · · ∧ dxiq | 1 ≤ i1 < · · · <
iq ≤ n}. As R is a Noetherian ring, f1 ΩqR is a Noetherian module and its
submodules are finitely generated.
If A is a commutative ring and M is an A–module, the dual of M is

M := HomA (M, A). An A–module M is called reflexive if M ' M ∗ .
∗
Exercise 3.7. Prove that the dual of DerC (R) is naturally isomorphic to Ω1R .
Proposition 3.2. The R–modules DerR (− log D) and Ω1R (log D) are dual
to each other. Both modules are then reflexive.
Remark 3.4. Because of the result in Proposition 3.2, V. Goryunov and

D. Mond said (see [26, page 207]) that Kyoji Saito suggested, following the
conventions of algebro-geometric notation, that the module of logarithmic
derivations should have − log D in parentheses rather than log D.
www.pdfgrip.com
105
Proof. [of Proposition 3.2] We consider the R-bilinear map

h−, −i : DerR (− log D) × Ω1R (log D) −→ Rf
defined by
* + P
X X bi a i bi
i
ai ∂i , dxi = .
f f
i i
This bilinear form is just the restriction of the one of Remark 3.2. First of
P P
all, we will prove that if δ = i ai ∂i ∈ DerR (− log D) and ω = i bfi dxi ∈
Ω1 (log D) then hδ, ωi ∈ R, i.e.
X
ai bi ∈ Rf.
i
P
Let us write g = i ai bi . Recall (see Exercise 3.6) that if ω =
P bi 1
i f dxi ∈ Ω (log D) then
f i bj − f j bi
∈R
f
for all i, j. Let us write fi bj − fj bi = hij f for some hij ∈ R. Then, for any
i = 1, . . . , n,
X X X
fi g = f i ( a j bj ) = a j f i bj = aj bi fj = bi δ(f ) ∈ Rf.
j j j
So we have
J · Rg ⊂ Rf
and V(f ) ⊂ V(J) ∪ V(g) where J = R(f1 , . . . , fn ) is the Jacobian ideal
associated with f .
We can write V(f ) = V(J, f ) ∪ V(g, f ). Recall that V(J, f ) is the set
of singular points of the hypersurface V(f ) and that it is a proper Zariski
closed subset of V(f ). So any irreducible component of V(f ) should be
included in V(g, f ) ⊂ V(g). By Hilbert’s Nullstellensatz g ∈ Rf .
There is a natural injective R-module morphism
Ω1 (log D) → (DerR (− log D))∗
which associates to any logarithmic 1-form ω the R-module morphism
φω : DerR (− log D) → R
defined by φω (δ) = hδ, ωi. We will prove that this natural injective mor-
phism is also surjective.
www.pdfgrip.com
106
Assume ϕ ∈ (DerR (− log D))∗ . We will see that the rational form
X ϕ(f ∂i )
ω= dxi
i
f
belongs to ∈ Ω1R (log D) and, moreover, ϕ = φω .

P
First of all, we will prove that ω = i ϕ(ff∂i ) dxi belongs to Ω1R (log D).
By, Exercise 3.6 it is enough to prove that
fi ϕ(f ∂j ) − fj ϕ(f ∂i ) ∈ Rf
for all i, j. To this end, fi ϕ(f ∂j ) − fj ϕ(f ∂i ) = ϕ(fi f ∂j ) − ϕ(fj f ∂i ) =

f ϕ(fi ∂j − fj ∂i ) ∈ Rf .
By definition of ω we have ϕ(f ∂i ) = φω (f ∂i ) for all i. Moreover, for
P
δ = i ai ∂i ∈ DerR (− log D) we have
X X
f ϕ(δ) = ϕ(f δ) = ai ϕ(f δi ) = ai φω (f ∂i ) = f φω (δ)
i i
and then ϕ(δ) = φω (δ) for all δ ∈ DerR (− log D). That proves ϕ = φω and,
moreover, the natural morphism
Ω1R (log D) → (DerR (− log D))∗
is an isomorphism of R–modules.
We will prove in a similar way that the natural injective R-module
morphism
DerR (− log D) → (Ω1R (log D))∗
which associates to any logarithmic vector field δ the R-module morphism
φδ : Ω1R (log D) → R
defined by φδ (ω) = hδ, ωi is an isomorphism.

Assume ϕ ∈ (Ω1R (log D))∗ . We will see that the vector field
X
δ= ϕ(dxi )∂i
i
belongs to DerR (− log D) and, moreover, ϕ = φδ .

We have ϕ( df df
f ) ∈ R and then f ϕ( f ) = ϕ(df ) ∈ Rf . We also have
X ∂f X ∂f
δ(f ) = ϕ(dxi ) = ϕ( dxi ) = ϕ(df ) ∈ Rf.
i
∂xi i
∂xi
Let’s prove that ϕ = φδ . By definition of δ we have ϕ(dxi ) = φδ (dxi ).

www.pdfgrip.com
107
P gi
Assume ω = i f dxi ∈ Ω1R (log D) for some gi ∈ R. We have
X X
f ϕ(ω) = ϕ(f ω) = ϕ( gi dxi ) = gi ϕ(dxi )
i i
X
= gi φδ (dxi ) = φδ (f ω) = f φδ (ω)
i
and then ϕ(ω) = φδ (ω).
Recall that the complex of (global) logarithmic differential forms

Ω•R (log D) is a sub-complex of Ω•R (∗D) (both endowed with the exterior
derivative) and that we denoted by ιD : Ω•R (log D) → Ω•R (∗D) the corre-
sponding inclusion.
Definition 3.3 (Calderón et al.12 Granger et al.27 ). The hypersur-

face D ⊂ Cn satisfies the Global Logarithmic Comparison Theorem if ιD
is a quasi-isomorphism. In this case we will say that D satisfies GLCT or
that GLCT holds for D.
An open question is to characterize the class of hypersurfaces D =

V(f ) ⊂ Cn such that D satisfies GLCT. This question is related to the
characterization of polynomials f ∈ R such that Ann(1/f ) is generated by
operators of order 1. These questions (and the extension of the previous
notions and results to the local analytic case) are treated in many classical
as well as many recent research papers. The interested reader can consult
the references 39, 17, 11, 12, 18, 19, 43, 13, 27, 2, 42.
Appendix A. Division Theorems and Groebner Bases in Rings

of Differential Operators
In this Appendix we first recall the definition of the rings of germs of linear
differential operators with holomorphic and formal coefficients. Then we
recall the main results on the division theorems and the theory of Groebner
bases in these rings of differential operators and in the Weyl algebra.
More rings of linear differential operators

Let C{x} = C{x1 , . . . , xn } (resp. C[[x]] = C[[x1 , . . . , xn ]]) be the ring of
convergent power series (resp. of formal power series) in n variables with
complex coefficients. The ring C{x} can also be viewed as the ring of germs
of holomorphic functions at the origin in Cn .
www.pdfgrip.com
108
We will denote by D (resp. D)b the ring of linear differential operators

with coefficients in C{x} (resp. C[[x]]).
By definition, any nonzero element P in D (resp. in D)b can be written
in a unique way as a finite sum
X
P = pβ (x)∂ β
β∈Nn
where pβ (x) ∈ C{x} (resp. C[[x]]) for all β ∈ Nn and, as in the case of the
Weyl algebra, ∂ β stands for ∂1β1 · · · ∂nβn .
As any power series pβ (x) can be written as
X
pβ (x) = pαβ xα
α∈Nn
with pαβ ∈ C, then each element P ∈ D, unlike in the case of the Weyl
algebra, can be written as a, possibly infinite, sum
X
α,β
b
and we have a similar result for D.
Some notions in Section 1 can be easily extended from the Weyl alge-
bra to D and D. b Among these notions we have the order ord(P ) and the
principal symbol σ(P ) of an operator P in D or D b (see Definitions 1.2 and
1.3).
The ring D is then filtered by the order of its elements. We will denote
this filtration by
(Fk (D))k∈Z .
F
The associated graded ring gr (D) is naturally isomorphic to the ring
C{x}[ξ] = C{x}[ξ1 , . . . , ξn ]
which is a polynomial ring with coefficients in C{x}. With any (left) ideal
I in D we can associate (see Definition 1.4) its graded ideal
grF (D) = C{x}[ξ] · {σ(P ), | P ∈ I}
which is homogeneous with respect to the ξ–variables. We also have anal-
b
ogous results for D.
Alike An the ring D is left and right Noetherian. The proof of the
noetherianity of An uses the total order in An (see Proposition 1.6). By
Remark 1.5 we can also give a proof of the same result by using the order
b
in An instead of the total order. This last proof is also valid in D and in D.
b
For another proof of the noetherianity of An (also valid for D and D) see
Remark 4.10.
www.pdfgrip.com
109
Division theorem in An
Definition A.1. A well ordering ≺ on Nn is said to be a monomial order

if it is compatible with the sum: α ≺ β implies α + γ ≺ β + γ for all γ ∈ Nn .
Remark 4.5. For any monomial order ≺ on Nn one has 0 = (0, . . . , 0) ≺ α

for all α ∈ Nn . Moreover, for α, β ∈ Nn such that αi ≤ βi for all i one has
α ≺ β. In other words, any monomial order refines the componentwise order
on Nn .
We usually translate any order ≺ on Nn to an order —also denoted by

≺— on the set of monomial {xα | α ∈ Nn } just by writing xα ≺ xβ if and
only if α ≺ β.
Example 4.1. (1) The lexicographical or lexicographic order (denoted by

<lex ) on Nn is defined as follows:
(α1 , . . . , αn ) <lex (β1 , . . . , βn )
if and only if the first nonzero component of
(α1 − β1 , . . . , αn − βn )
is negative. The well ordering <lex is a monomial order.

(2) Let ≺ be a monomial order on Nn . Let L : Qn → Q be a linear form
with non negative coefficients. The binary relation ≺L defined on Nn by:

L(α) < L(β)
α ≺L β if and only if
or L(α) = L(β) and α ≺ β
is a monomial order on Nn .
P
Let P = P (x, ∂) = β∈Nn pβ (x)∂ β be a differential operator in An . The
operator P can be rewritten as
X
αβ
P
just by considering the polynomial pβ (x) as pβ (x) = α pαβ xα , with pαβ ∈
C.
The Newton diagram of P is the set
N (P ) := {(α, β) ∈ N2n | pαβ 6= 0}.
Let us fix a monomial order ≺ on N2n .

www.pdfgrip.com
110
Definition A.2. We call privileged exponent i with respect to ≺ of a

nonzero operator P —and we denote it by exp≺ (P )— the maximum
(α, β) ∈ N2n , with respect to ≺, such that pαβ 6= 0. In other words
exp≺ (P ) = max N (P ).
≺
We will simply write exp(P ) if no confusion is possible.
Proposition 4.3. Let P, Q ∈ An . We have:

1) exp(P Q) = exp(P ) + exp(Q).
2) If exp(P ) 6= exp(Q) then exp(P + Q) = max≺ {exp(P ), exp(Q)}. More
generally, for any family P1 , . . . , Pm ∈ An such that exp(Pi ) 6= exp(Pj ) for
P
all i, j, i 6= j we have exp( i Pi ) = max≺ {exp(Pi ) | i = 1, . . . , m}.
With each m–tuple ((α1 , β 1 ), . . . , (αm , β m )) of elements in N2n , we as-

sociate a partition j
{∆0 , ∆1 , . . . , ∆m }
of N2n in the following way. We set:
∆1 = (α1 , β 1 ) + N2n
∆i+1 = ((αi+1 , β i+1 ) + N2n ) \ (∆1 ∪ · · · ∪ ∆i ) if i ≥ 1
∆0 = N2n \ (∪m i
i=1 ∆ ).
The following theorem generalizes the division theorem for polynomials

in C[x] (see e.g. [22, p. 9] or [1, Th. 1.5.9]).
Theorem A.1 (Division in An ). Let (P1 , . . . , Pm ) be an m–tuple of

nonzero elements of An and let {∆0 , ∆1 , . . . , ∆m } be the partition of N2n
associated with (exp(P1 ), . . . , exp(Pm )). Then, for any P in An , there exists
a unique (m + 1)–tuple (Q1 , . . . , Qm , R) of elements in An , such that:
(1) P = Q1 P1 + · · · + Qm Pm + R.
(2) exp(Pi ) + N (Qi ) ⊂ ∆i , i = 1, . . . , m.
(3) N (R) ⊂ ∆0 .
i Thisnotion generalizes the one of privileged exponent of a power series, due to H.

Hironaka. It was introduced in Lejeune and Teissier32 (see also Aroca et al.3 ).
j The word partition is used here in a broad sense, which means that an element of the
family may be empty.

www.pdfgrip.com
111
Proof. Let us prove uniqueness first. Assume that two (m + 1)–tuples,

(Q1 , . . . , Qm , R) and (Q01 , . . . , Q0m , R0 ), satisfy the conditions of the theo-
rem. We have:
m
X
(Qi − Q0i )Pi + R − R0 = 0 (1)
i=1
If Qi 6= Q0i then exp((Qi − Q0i )Pi ) ∈ ∆i . If R 6= R0 then exp(R − R0 ) ∈ ∆0 .

Since ∆0 , ∆1 , . . . , ∆m is a partition of N2n , the equality (1) is only possible
if Qi = Q0i for any i and if R = R0 (see Proposition 4.3). That proves the
uniqueness.
It is clear that it is enough to prove the existence for the monomials
xα ∂ β ∈ An . We will use induction on (α, β). If xα ∂ β = 1 (i.e. if α = β =
(0, . . . , 0)), then either exp(Pi ) 6= 0 ∈ N2n for all i and in this case it is
Pm
enough to write 1 = i=1 0Pi + 1, or there exists an integer j such that
exp(Pj ) = 0 ∈ N2n . In this case Pj is a nonzero constant because 0 is the
first element in N2n with respect to the well ordering ≺ and, moreover,
∆0 = ∅. Assume that j is minimal. We write
X
1= 0 · Pi + (1/Pj )Pj + 0.
i6=j
This proves the existence at the first step of the induction. Assume that
the result is proved for any (α0 , β 0 ) strictly less than some (α, β) 6= 0 ∈ N2n .
Let j ∈ {0, 1, . . . , m} be such that (α, β) ∈ ∆j . If j = 0 we write
m
X
xα ∂ β = 0 · Pi + xα ∂ β .
i=1
If j ≥ 1 let (γ, δ) = (α, β) − exp(Pj ) ∈ N2n .

We can write
1 γ δ
xα ∂ β = x ∂ Pj + G j
cj
where cj is the coefficient of the privileged monomial of Pj and all the
monomials in Gj are smaller (with respect to ≺) than (α, β) (see Exercise
1.2). By the induction hypothesis there exists (Q01 , . . . , Q0m , R0 ) satisfying
the conditions of the theorem for P = Gj . In particular we have:
X 1 γ δ
xα ∂ β = Q0i Pi + ( x ∂ + Q0j )Pj + R0 .
cj
i6=j
This proves the result for (α, β). Thus, existence is proved for any P ∈ An .
www.pdfgrip.com
112
Remark 4.6. The linear differential operator Qi in the theorem is called

the i–th quotient and R is called the remainder of the division of P by
(P1 , . . . , Pm ). The remainder will be denoted by R(P ; P1 , . . . , Pm ).
Remark 4.7. It follows from the proof of Theorem A.1 that for any
division P = Q1 P1 + · · · + Qm Pm + R as in the theorem we have
max{maxi {exp≺ (Qi Pi )}, exp≺ (R)} = exp≺ (P ).
Groebner bases in An
The theory of Groebner bases in the polynomial ring C[x] (see Buch-
berger9,10 ) can be extended to the Weyl algebra An and also to the rings D
and D b (see Castro14,15 and Briançon and Maisonobe8 ). The book M. Saito
et al.40 contains many applications of Groebner basis theory to the study
of modules over the Weyl algebra An .
Let us fix a monomial order ≺ on N2n . For each non zero ideal I of An
let Exp≺ (I) denote the set
{exp≺ (P )|P ∈ I \ {0}}.
If no confusion is possible we write Exp(I) instead of Exp≺ (I). From Propo-

sition 4.3 we have: Exp(I) + N2n = Exp(I).
Proposition 4.4. Let m > 0 be an integer and E ⊂ Nm such that E +

Nm = E. Then there exists a finite subset F ⊂ E such that
[
E= (α + Nm )
α∈F
Proof. This is a version of Dickson’s lemma. The proof is by induction on

m. For m = 1 a (finite) family of generators of E is given by the smallest
element of E (for the usual ordering in N). Assume that m > 1 and that
the result is true for m − 1. Let E ⊂ Nm be such that E + Nm = E. We
can assume that E is non empty. Let α ∈ E. For any i = 1, . . . , m and
j = 0, . . . , αi we consider the bijective mapping
φi,j : Ni−1 × {j} × Nm−i −→ Nm−1

(β1 , . . . , βi−1 , j, βi+1 , . . . , βm ) 7→ (β1 , . . . , βi−1 , βi+1 , . . . , βm )
and we denote
Ei,j = φi,j (E ∩ (Ni−1 × {j} × Nm−i )).

www.pdfgrip.com
113
It is clear that Ei,j + Nn−1 = Ei,j and by the induction hypothesis there is
a finite subset Fi,j ⊂ Ei,j generating Ei,j . The finite set
 
[ [
F = {α}  (φi,j )−1 (Fi,j )
i,j
generates E. This proof is taken from Galligo.25
Remark 4.8. The previous proposition can be rephrased as follows:

Any monomial ideal in C[x] (Or more generally in a polynomial ring
K[x1 , . . . , xn ] with coefficients in a field K) is finitely generated. This is
a particular case of the Hilbert basis theorem.
In the same way, we can see that any increasing sequence (Ek )k of
subsets of Nm , stable under the action of Nm , is stationary. We shall often
use this property called the noetherian property for Nm .
We can adapt the proof above to show that, given E ⊂ Nn as in Propo-
sition 4.4, we can find in any set of generators, a finite subset of generators
of E. This proves in particular that in any system of generators made of
monomials of a monomial ideal in the polynomial ring C[x], we can find a
finite subset of generators. This is Dickson’s lemma.
Definition A.3. Let I be a nonzero ideal of An . We call any family

P1 , . . . , Pm of elements in I such that
m
[
Exp≺ (I) = (exp≺ (Pi ) + N2n )
i=1
a Groebner basis of I, relative to ≺ (or a ≺–Groebner basis of I).
Remark 4.9. There always exists a Groebner basis of I by definition of

Exp(I) and by Proposition 4.4.
We have the following two corollaries of Theorem A.1.
Corollary A.1. Let I be a nonzero ideal of An and let P1 , . . . , Pm be a

family of elements of I. The following conditions are equivalents:
1) P1 , . . . , Pm is a Groebner basis of I.
2) For any P in An , we have: P ∈ I if and only if R(P ; P1 , . . . , Pm ) = 0
(see Remark 4.6).
Corollary A.2. Let I be a nonzero (left) ideal of An and let P1 , . . . , Pm

be a Groebner basis of I. Then P1 , . . . , Pm is a system of generators of I.
www.pdfgrip.com
114
Remark 4.10. Corollary A.2 proves in particular that An is a left-

Noetherian ring (see also Proposition 1.6).
Remark 4.11. Assume the monomial order ≺ satisfies the following prop-
erty: for any P ∈ An we have exp≺ (P ) = exp≺ (σ(P )). Then, for any ideal
I in An and any finite subset {P1 , . . . , Pm } ⊂ I, the family {P1 , . . . , Pm }
is a Groebner basis of I if and only if the family {σ(P1 ), . . . , σ(Pm )} is a
Groebner basis of grF (I) ⊂ C[x, ξ].
If L : Qn × Qn −→ Q is the linear form defined by
P
L(a1 , . . . , an , b1 , . . . , bn ) = i bi and ≺ is any monomial order on N2n then
the monomial order ≺L (see Example 4.1) satisfies the desired condition
exp≺ (P ) = exp≺ (σ(P )) for all P ∈ An .
Buchberger’s algorithm in An
Buchberger’s algorithm for polynomials (see Buchberger10 ) can be easily
adapted to the Weyl algebra (see Briançon and Maisonobe8 and Castro14
(see also Saito et al.40 )).
Considering as input a monomial order ≺ in N2n and a finite set F =
{P1 , . . . , Pm } of differential operators in An , one can compute a Groebner
basis, with respect to ≺, of the ideal I ⊂ An generated by F. So, one can
also compute a finite set of generators of the subset Exp≺ (I) ⊂ N2n .
The Division Theorem A.1 and the theory of Groebner bases can be
extended for vectors and for submodules of free modules Arn for any integer
r > 0 (see Castro14 ).
Moreover, the Division Theorem and the Groebner basis notion can be
also considered, in a straightforward way, for right ideals (or more generally
for right submodules of a free module Arn ).
Similarly to the commutative polynomial case, Groebner bases in An are
used to compute, in an explicit way, some invariants in An -module theory.
Most of the algorithms in this subject appears in Oaku and Takayama.37
In particular, Groebner bases in An are used:
a) to compute a generating system of SyzAn (P1 , . . . , Pm ), the An –module
of syzygies of a given family P1 , . . . , Pm in a free module Arn (r ≥ 1).
b) to solve the membership problem (i.e. to decide if a given vector P ∈ Arn
belongs to the submodule generated by P1 , . . . , Pm ) and to decide if two
submodules of Arn are equal.
c) to compute the graded ideal and the total graded ideal associated with
a (left) ideal I in An (see Definition 1.4) and to compute the dimension
of a quotient module An /I.
www.pdfgrip.com
115
d) to decide if a finitely presented An –module is holonomic (i.e. to decide

if its characteristic variety has dimension n. See Definition 2.8).
e) to construct a finite free resolution of a given finitely presented An –
module.
f) to decide if a finite complex of free An –modules is exact.
Many computer algebra systems can handle this kind of computations.

Among the most used should be mentioned Macaulay28 Risa/Asir35 and
Singular.29
Remark 4.12 (Division theorem and Groebner bases in D and D). b

A Division Theorem (analogous to Theorem A.1) can be proved for ele-
ments in D or in D b (see Briançon and Maisonobe 8 and Castro 14 ). This
is not straightforward from the Weyl algebra case because Definition A.2
of privileged exponent doesn’t work for general operators in D or in D. b
Nevertheless, Groebner bases also exist for left (or right) ideals in D (and
b and the analogous of Corollaries A.1 and A.2 also hold in D and D.
in D) b
b
This proves in particular that D and D are Noetherian rings. We will not
give here the details and refer the interested reader to the references above.
Most of the material of this Appendix appears in Castro,14,15 Briançon and
Maisonobe,8 Saito et al.40 and Castro and Granger.16
References
1. Adams W.W. and Loustaunau Ph. An introduction to Gröbner bases. Grad-
uate Studies in Mathematics, 3. American Mathematical Society, 1994.
2. Álvarez-Montaner J., Castro-Jiménez F.J., Ucha-Enrı́quez J.M. Localiza-
tion at hyperplane arrangements: combinatorics and D-modules, J. Algebra,
available on-line 28 Dec 2006; http://dx.doi.org/10.1016/j.jalgebra.
2006.12.006
3. Aroca J. M., Hironaka H. and Vicente J.L. The theory of maximal contact.
Volume 29 of Memorias del Instituto “Jorge Juan”, Madrid, 1975.
4. Atiyah M.F. and Macdonald I.G. Introduction to Commutative Algebra.
Addison-Wesley, 1969.
5. Bernstein J. Modules over the ring of differential operators. Study of the fun-
damental solutions of equations with constant coefficients. Funkcional. Anal.
i Priložen. 5 (1971), no. 2, 1–16.
6. Bernstein J. Analytic continuation of generalized functions with respect to a
parameter. Funkcional. Anal. i Priložen. 6 (1972), no. 4, 26–40.
7. Björk J-E. Rings of Differential Operators. North-Holland, Amsterdam 1979.
www.pdfgrip.com
116
8. Briançon J. and Maisonobe Ph. Idéaux de germes d’opérateurs différentiels

à une variable, L’Enseignement Math. 30, (2004), 7–38.
9. Buchberger B. An Algorithm for Finding the Bases Elements of the Residue
Class Ring Modulo a Zero Dimensional Polynomial Ideal (German), PhD
thesis, (1965), Univ. of Innsbruck (Austria).
10. Buchberger B. An Algorithmical Criterion for the Solvability of Algebraic
Systems of Equations (German), Aequationes Mathematicae 4(3),(1970),
374–383.
11. Calderón-Moreno F.J. Logarithmic Differential Operators and Logarithmic
De Rham Complexes relative to a Free Divisor. Ann. Sci. E.N.S., 4e série, t.
32, 1999, 701-714.
12. Calderón-Moreno F.J., Mond D., Narváez-Macarro L. and Castro-Jiménez
F.J. Logarithmic Cohomology of the Complement of a Plane Curve. Com-
ment. Math. Helv. 77 (2002), no. 1, 24–38.
13. Calderón-Moreno F.J. and Narváez-Macarro L. Dualité et comparaison sur
les complexes de de Rham logarithmiques par rapport aux diviseurs libres.
Ann. Inst. Fourier (Grenoble) 55 (2005), no. 1, 47–75.
14. Castro F. Théorème de division pour les opérateurs différentiels et calcul des
multiplicités, Thèse de 3eme cycle, Univ. of Paris VII, (1984).
15. Castro-Jiménez F. Calculs effectifs pour les idéaux d’opérateurs différentiels,
in Actas de la II Conferencia Internacional de Geometrı́a Algebraica. La
Rábida, Travaux en Cours 24, (1987), Hermann, Paris.
16. Castro Jiménez F. J. and Granger, M. Explicit calculations in rings of
differential operators. Éléments de la théorie des systèmes différentiels
géométriques, 89–128, Sémin. Congr., 8, Soc. Math. France, Paris, 2004.
17. Castro-Jiménez F.J., Mond, D. and Narváez-Macarro L. Cohomology of the
complement of a free divisor. Trans. Amer. Math. Soc. 348 (1996), no. 8,
3037–3049.
18. Castro-Jiménez, F.J. and Ucha-Enrı́quez J.M. Explicit comparison theorems
for D-modules. J. Symbolic Comput., Special Issue on Effective Methods in
Rings of Differential Operators, 32 (2001) no. 6, 677–685.
19. Castro Jiménez F. J. and Ucha Enrı́quez J. M. Testing the Logarithmic Com-
parison Theorem for free divisors. Experimental Math. 13:4, 441-449, 2004.
20. Castro-Jiménez, F.J. and Ucha-Enrı́quez J.M. Gröbner bases and logarithmic
D-modules. J. Symbolic Comput., 41 (2006), 317–335.
21. Coutinho, S.C. A primer on D-modules. Student Texts 33, London Mathe-
matical Society. Cambridge University Press 1995.
22. Cox D., Little J. and O’Shea D. Using algebraic geometry. Graduate Texts
in Mathematics, 185, Springer-Verlag, New York, 2005.
23. Ehlers F. The Weyl algebra in Borel A. et al. Algebraic D-modules. Perspec-
tives in Mathematics, 2. Academic Press, Inc., Boston, MA, 1987. xii+355
pp.
24. Eisenbud, D. Commutative algebra. With a view toward algebraic geometry.
Graduate Texts in Mathematics, 150. Springer-Verlag, New York, 1995.
25. Galligo A. Théorème de division et stabilité en Géométrie Analytique locale.
Ann. Inst. Fourier, 29,2:107–184, 1979.
www.pdfgrip.com
117
26. Goryunov V. and Mond D. Tjurina and Milnor numbers of matrix singular-
ities. J. London Math. Soc. (2) 72 (2005) 205-224.
27. Granger M., Mond D., Nieto A. and Schulze M. Linear free divisors. arXiv
:math/0607045.
28. Grayson D. and Stillman M. Macaulay2: a software system for research
in algebraic geometry. Available at http://www.math.uiuc.edu/Macaulay2
Leykin, A. and Tsai, H. D-modules package for Macaulay 2. Available at
http://www.math.cornell.edu/~htsai
29. Greuel G.-M., Pfister G. and Schönemann H. Singular a computer algebra
system for polynomial computations, Centre for Computer Algebra, Univer-
sity of Kaiserslautern, http://www.singular.uni-kl.de.
30. Hartshorne R. Algebraic Geometry. Graduate Texts in Mathematics.
Springer-Verlag. 1977.
31. Kashiwara M. B-functions and holonomic systems. Rationality of roots of
B-functions. Invent. Math. 38, 1 (1976) 33–53.
32. M. Lejeune-Jalabert and B. Teissier. Quelques calculs utiles pour la résolution
des singularités, Volume of Prépub. Centre de Mathématiques, Ecole Poly-
technique, 1971.
33. Malgrange, B. Le polynome de Bernstein d’une singularite isolée. Lecture
Notes in Math. 459, (1976), Springer-Verlag, 98-119.
34. Noro M. An efficient modular algorithm for computing the global b-function.
Mathematical software (Beijing, 2002), 147–157. World Sci. Publishing, River
Edge, NJ, (2002).
35. Noro M., Shimoyama T. and Takeshima T. A Computer Algebra System
Risa/Asir. Available at http://www.math.kobe-u.ac.jp/Asir/index.
html.
36. Oaku T. An algorithm of computing b-functions. Duke Math. J. 87 (1997),
no. 1, 115–132
37. Oaku T. and Takayama N. Algorithms for D-modules – restriction, tensor
product, localization, and local cohomology groups. Journal of Pure and Ap-
plied Algebra, 156 (2001) pp. 495-518.
38. M. Sato. Theory of prehomogeneous vector spaces (algebraic part)—the En-
glish translation of Sato’s lecture from Shintani’s note. Notes by T. Shintani.
Translated from the Japanese by M. Muro. Nagoya Math. J. 120 (1990),
1–34.
39. Saito K. Theory of logarithmic differential forms and logarithmic vector fields.
J. Fac Sci. Univ. Tokyo 27:256-291, 1980.
40. Saito M., Sturmfels B. and Takayama N., Gröbner deformations of hypergeo-
metric differential equations. Algorithms and Computation in Mathematics,
6. Springer-Verlag, Berlin, (2000).
41. Stafford, J.T. Nonholonomic modules over Weyl algebras and enveloping al-
gebras. Invent. Math., 79, (1985), no. 3, 619-638.
42. Terao H. and Yuzvinsky S. Logarithmic forms on affine arrangements,
Nagoya Math. J. 139 (1995), 129–149.
43. Torrelli, T. On meromorphic functions defined by a differential system of
order 1. Bull. Soc. Math. France 132 (2004), no. 4, 591–612.
www.pdfgrip.com
118
44. Zariski, O. and Samuel, P. Commutative algebra. Vol. I and II. Springer
Verlag, 1975
April 6, 2010 15:50 WSPC - Proceedings Trim Size: 9in x 6in 03˙teissier
www.pdfgrip.com
119
GEOMETRY OF CHARACTERISTIC VARIETIES
ˆ DUNG
LE ˜ ´
TRANG
The Abdus Salam International Centre for Theoretical Physics,
Trieste, Italy
BERNARD TEISSIER
Inst. Math. de Jussieu, 175 Rue Chevalleret, 75013 Paris, France
0. Introduction
In this paper we give a quick presentation of the characteristic variety of a
complex analytic linear holonomic differential system. The fact that we can
view a complex analytic linear holonomic differential system on a complex
manifold as a holonomic DX -module, allows us to present the characteristic
variety in an algebro-geometric way as we do here.
We do not define what is a sheaf; for this we refer to the famous book of
R. Godement.2 We also do not define properly DX -modules although one
can find in the lectures of F. Castro the definition of left modules on the
Weyl algebra of operators on Cn with polynomial coefficients which give a
local version of DX -modules.
Also we consider categories, complexes in abelian categories, derived
functors and hypercohomologies1 which are the natural language to be used
here. Of course, we cannot define all these notions. One should view these
notes as a provocation rather than a self-contained exposition. We hope
they will encourage the reader to learn more in the subject.
1. Whitney Conditions
In Ref. 19, §19 p. 540, H. Whitney introduced Whitney conditions. The gen-
eral idea is to find conditions for the attachment of a non singular analytic
space having an analytic closure along a non singular part of its bound-
ary which ensure that the closure is “locally topologically trivial” along
www.pdfgrip.com
120
the boundary, that is, locally topologically a product of the boundary by a

“transverse slice”. Whitney’s approach can be deemed to be based on the
fundamental fact, which he discovered, that any analytic set is, at each of its
points, “asymptotically a cone”. More precisely, given any closed complex
analytic subspace X ⊂ U ⊂ CN and a point of X which we may take as
the origin 0, given any sequence of non singular points xn ∈ X converging
to 0, we may consider the tangent spaces TX ,xn of X and the secant lines
0xn . They respectively define points in the Grassmanian G(N, d) and the
projective space PN −1 , which are compact. Therefore, possibly after taking
a subsequence we may assume the limits T = limn TX ,xn and ` = limn 0xn
exist, when xn tends to 0. Then we have ` ⊂ T , which ensures that X is
locally “cone-like” with respect to the “vertex” 0.
Whitney’s idea may have been that the topological triviality of the
closed analytic space X along a part Y would be ensured by the condition
that X should be “cone-like” along the “vertex” Y, which is a way to ensure
that locally X is transversal to the boundaries of small tubes around Y. It
gives this:
Let X be a complex analytic subset of CN and Y be a complex analytic
subset of X . One says that X satisfies the Whitney condition along Y at
the point y ∈ Y, if:
(1) the point y is non-singular in Y;
(2) for any sequence (xn )n∈N of non-singular points of X which converges
to y such that the tangent spaces Txn (X ) have a limit T and, for any
sequence (yn )n∈N of points of Y which converges to y, such that the
lines yn xn have a limit `, we have ` ⊂ T .
Examples. Consider the complex algebraic subset X of C3 defined by the
equation:
X 2 − Y Z 2 = 0.
The line L = {X = Y = 0} is contained in X . The surface X satisfies
the Whitney condition along L at any point y ∈ L \ {0}, but not at the
point {0}, because, for the sequence (0, yn , 0) of non-singular points of X ,
the limit of tangent spaces is the plane T = {Z = 0}, and for the points
(0, 0, yn ) the limit ` is the line {X = 0, Y = −Z} which is not contained in
T.
Consider the complex algebraic subset X of C3 defined by the equation
(see Fig. 2):
X 2 − Y 3 − Z 2 Y 2 = 0.
www.pdfgrip.com
121
2 2
X-YZ =0
Fig. 1.
2 3 2 2
X -Y - Z Y = 0
Fig. 2.
The surface X satisfies Whitney condition along L = {X = Y = 0}

at any point y ∈ L \ {0}, but not at {0}, because, one may consider the
sequence of non-singular points (0, zn2 , zn ) of X and (0, 0, zn ) of L as zn
tends to 0. The limit ` is the line which contains (0, 0, 1), the limit T is the
plane orthogonal to (0, 0, 1).
2. Stratifications
Let X be a complex analytic subset of CN . A partition S = (Xα )α∈A is a
stratification of X , if:
(1) The closure X α of Xα in X and X α \Xα are complex analytic subspaces
of X ;
(2) The family (Xα ) is locally finite;
(3) Each Xα is a complex analytic manifold;
www.pdfgrip.com
122
(4) If, for a pair (α, β), we have Xα ∩ X β 6= ∅, then Xα ⊂ X β .

The subsets Xα are called the strata of the stratification S. The condition
(4) is called the frontier condition (see e.g. Ref. 13, Définition (1.2.3)).
Examples. Consider the complex algebraic subset of C3 defined by the

equation:
X 2 − Y 3 − Z 2 Y 2 = 0.
One can consider this surface as a deformation of complex plane algebraic
curves parametrized by Z. The singular points of these curves are on the Z-
axis which if given by X = Y = 0. For Z = 0, one has a cusp X 2 − Y 3 = 0.
For Z 6= 0, one has a strophoıd with a singular point at the origin.
One has a stratification by considering the strata given by the non-
singular points, which are the points of the surface outside of the line X =
Y = 0 on the surface, and by the singular points which are the points of
the line X = Y = 0.
Now consider the complex algebraic subset of C3 defined by the equation
(see Fig. 3):
X 2 − Y 2 Z 3 = 0.
L1
2 2 3
X -Y Z =0
L2
Fig. 3.
It is a surface whose singular points lie on two lines L1 = {X = Y = 0}

and L2 = {X = Z = 0}. We can define a partition of this surface by
considering S0 := X 0 , the subset of non-singular points of the surface, the
punctured line S1 = L1 \ {0} and the line S2 = L2 .
This partition does not define a stratification, because it does not satisfy
the frontier condition. However, one can consider instead the partition given
by S0 , S1 , the punctured line S20 = L2 \ {0} and the origin {0}. This is a
stratification.
www.pdfgrip.com
123
More generally one can prove that, by choosing a “refinement” of a

partition satisfying (1), (2) and (3), one obtains the frontier condition (see
Ref. 19).
A complex analytic partition S = (Xα )α∈A of a complex analytic space

X is a partition of X which satisfies (1), (2) and (3) above.
Then, a complex analytic partition S 0 = (Xα0 )α∈A0 of X is finer than
the partition S = (Xα )α∈A of X , if any stratum Xα of S is the union of
strata of S 0 . One may quote a proposition of H. Whitney in the following
way:
Proposition 2.1. For any complex analytic partition S = (Xα )α∈A of

X , there is a finer complex analytic partition which satisfies the frontier
condition.
Remark also that there is a coarsest finer partition with connected strata;
this is obvious by taking the connected components of strata. In stratifica-
tion theory one often assumes that the strata are connected.
3. Constructible Sheaves
All the sheaves that we consider are sheaves of complex vector spaces.
First, let us define local systems.
Definition 3.1. Let A be a topological space. A sheaf F on A is called a

local system on A if it is locally isomorphic to a constant sheaf.
For instance, a constant sheaf is a local system. In fact, when A is an

arcwise connected space, a local system F on A defines a homomorphism
ρF
π1 (A, a) → AutC Fa . This homomorphism is defined in the following way:
let γ be a loop at a; one can extend a section of Fa by continuity along
γ, since there is a neighbourhood of a on which F is constant; because we
can cover the image of γ by a finite number of open sets over which F is
constant we define a map of Fa into itself determined by γ; one can show
that this map depends only on the homotopy class of γ and is a complex
linear automorphism of Fa .
The correspondence F 7→ ρF defines an equivalence of category between
the category of local systems on A and the category of representations of
the fundamental group π1 (A, a) in finite dimensional vector spaces.
www.pdfgrip.com
124
Example: In the case A is the circle S1 , the fundamental group π1 (A, a)

is the group of relative integers Z. Local systems of rank one are given by
maps C → C given by z 7→ z k , with k ∈ Z.
Now, we have an important concept which, in some sense, generalises
complex analytic partitions:
Definition 3.2. A sheaf L over a complex analytic set X is constructible

if there is a complex analytic partition S = (Xα )α∈A of X , such that the
restriction of L over each Xα is a local system.
Examples: a) Let f : X → Y be a proper algebraic map between two

algebraic varieties, the sheaf on Y whose stalks are the k-th cohomology
Hk (f −1 (y), C) of the fibers f −1 (y) of f is a constructible sheaf.
b) The cohomologies of the solutions RHomD (M, O) of a holonomic D-
module M over a complex space CN are constructible sheaves (Kashiwara’s
constructibility Theorem).
c) Let Z be an algebraic subvariety of X and i : Z → X be the inclusion.
Let z ∈ Z. The local cohomology (Rk i! )(C)z ' HkZ∩Bz (X ∩ Bz , C), where
Bz is a good neighbourhood of z in X , defines a constructible sheaf on Z.
These examples are not easy to prove. The first and third ones are
difficult theorems on the topology of algebraic maps, the second one is a
basic theorem of the theory of D-modules, also difficult to prove.
4. Whitney Stratifications
Let S = (Xα )α∈A be a stratification of a complex analytic set X . We
say that S is a Whitney stratification of X if, for any pair (Xα , Xβ ) of
strata, such that Xα ⊂ X β , the complex analytic set X β satisfies Whitney
condition along Xα at any point of Xα .
As we announced above, the interest for Whitney stratifications comes
from the fact that they imply local topological triviality. Namely, a theorem
of J. Mather14 and R. Thom18 gives:
Theorem 4.1. Let S = (Xα )α∈A be a Whitney stratification of a complex

analytic subset X of CN . For any point x ∈ Xα there is an open neighbour-
hood U of x in CN such that X ∩U is homeomorphic to (Xα ∩U )×(Nα ∩U )
where Nα is a slice of Xα at x in X by a transversal affine space in CN .
This theorem shows that locally on X , along any strata, the analytic
set is topologically a product.
A theorem of H. Whitney19 gives that:
www.pdfgrip.com
125
Theorem 4.2. Let S = (Xα )α∈A be a stratification of a complex analytic

set X . There is a stratification of X which is finer than S and is a Whitney
stratification for X .
As a consequence on a singular compact space one can observe only

finitely many different topological types of embedded germs of complex
spaces, while in general there are continua of different analytic types.
Examples. In the examples given before the stratification of:

X2 − Y Z2 = 0
given by the non-singular part X \ Y of X , Y \ {0} and {0} gives a Whitney
stratification of X which is finer than X \ Y and Y.
We have the same for:
X 2 − Y 3 − Z 2 Y 2 = 0.
For the example given by:
X 2 − Y 2 Z 3 = 0,
a Whitney stratification is given by X \ L1 ∪ L2 , L1 \ {0}, L2 \ {0} and {0}.
A theorem of B. Teissier shows that a Whitney stratification can be
characterized algebraically and is very useful to know if a stratification is
a Whitney stratification. In order to give this criterion, we need to define
Polar Varieties.
Let X be a complex analytic subset of CN . Let x be a point of X . Assume
for simplicity that X is equidimensional at x. Consider affine projections of
X into Ck+1 , for 1 ≤ k ≤ dimx (X ). Then, one can prove:
Theorem 4.3. There is a non-empty Zariski open set Ωk in the space of

projections of CN into Ck+1 , such that for any p ∈ Ω, there is an open
neighbourhood U , such that either the critical locus of the the restriction of
p to the non-singular part of X ∩ U is empty or the closure of the critical
locus of the restriction of p to the non-singular part of X ∩ U is reduced
and has dimension k at x and its multiplicity at x is an integer which is
independent of p ∈ Ωk .
In the case where the critical locus of the restriction of p to the non-
singular part of X ∩ U is not empty for p ∈ Ωk , the closure of the critical
locus in X ∩U is called “the” polar variety Pk (X , x, p) of X at x of dimension
k, defined by p and the multiplicity m(Pk (X , x, p)) is called the k-th polar
multiplicity of X at the point x. For Pk (X , x, p) this is an abuse of language
www.pdfgrip.com
126
since only the “equisingularity type” of Pk (X , x, p) is well-defined for p ∈

Ωk , but the theorem just stated shows that there is no abuse as far as the
multiplicity is concerned. When the critical locus of the restriction of p is
empty for p ∈ Ωk , we say that mk (X , x) = 0.
Then, the criterion of B. Teissier is the following (Ref. 17):
Theorem 4.4. Let X be a complex analytic set. Let S = (Xα )α∈A be a

stratification of X with connected strata. Suppose that for any pair (X α , Xβ )
such that Xα ⊂ X β , the multiplicities m(Pk (X β ), y) are constant for
y ∈ Xα , for 1 ≤ k ≤ dimy (X β ), then, the stratification S is a Whitney
stratification. Conversely, any Whitney stratification with connected strata
has this property.
Beware that this theorem is true with stratifications: for instance the
frontier condition is important. One can consider the case of the surface
defined by:
X 2 − Y 2 Z 3 = 0.
Examples. Let X be a surface, i.e. a complex analytic set of dimension 2.
If, x ∈ X is a non-singular point, the 2-nd polar variety at x is X itself and,
by definition, its multiplicy is 1. At x the 1-st polar variety is empty, so
m1 = 0. If x ∈ X is singular, again the 2-nd polar variety at x is X itself and
m2 = mx (X) > 1. For almost all singular points, except a finite number
locally, the 1-st polar curve is empty, so m1 = 0. So, a Whitney stratification
of X is given by the non-singular part X 0 of X , the nonsingular part of the
singular locus minus the points where the polar curve is not empty, the
points where the polar curve is not empty and finally the singular points
of the singular locus.
For the surface given by:
X 2 − Y 3 − Z 2Y 2 = 0
the stratification given by X \Y, Y \{0} and {0} is a Whitney stratification.
A consequence of this theorem is the existence of a minimal Whitney
stratification refining a given one (see Ref. 17):
Theorem 4.5. Let X be a complex analytic set. Let S = (Xα )α∈A be a

complex analytic partition of X . Then there is a unique coarsest refinement
of S which is a Whitney stratification of X with connected strata.
If one takes as partition the non-singular part of X , the non-singular part

of the singular locus, and so on, one sees that every complex-analytic space
www.pdfgrip.com
127
has a unique “minimal” Whitney stratification in the sense that any other
Whitney stratification is a refinement of it.
5. Milnor Fibrations
Let f : U ⊂ CN → C be a complex analytic function defined on a neigh-
bourhood of the origin 0 in CN such that the image f (0) of the origin 0 by
f is 0.
One can prove:
Theorem 5.1. There is ε0 > 0, such that, for any ε such that 0 < ε < ε0 ,
there is η() > 0, such that for any η such that 0 < η < η(ε), the map
ϕε,η : Bε ∩ f −1 (S)η → Sη ,
induced by f over the circle of radius η centered at the origin 0 in the
complex plane C, is a locally trivial smooth fibration.
See Ref. 16 for the case where f has an isolated critical point at 0 and
Ref. 5 for the general case. We call the fibration given by the theorem the
Milnor fibration of f at 0.
When f has a critical point at 0, the fibers of ϕε,η have a non-trivial
homotopy. Since the “fiber” at 0, i.e., Bε ∩ f −1 (0) is contractible by Ref. 16,
one usually calls vanishing cycles the cycles of a fiber of ϕε,η , i.e. the ele-
ments of the homology H∗ (Bε ∩ f −1 (t), C) for t ∈ Sη .
One calls neighbouring cycles the elements of the cohomology H∗ (Bε ∩
f −1 (t), C). The theorem above shows that these definitions do not depend
on t ∈ Sη . One can prove that it does not depend on ε, η chosen conve-
niently.13
One may observe that the complex cohomology H∗ (Bε ∩ f −1 (t), C) is
the sheaf cohomology of Bε ∩ f −1 (t) with coefficients in the constant sheaf
C.
We also have on any analytic set a theorem similar to the one above
(see Ref. 9):
Theorem 5.2. Let f : X → C be a complex analytic function on a complex
analytic subset of CN . Suppose that 0 ∈ X and f (0) = 0. There is ε0 > 0,
such that, for any ε such that 0 < ε < ε0 , there is η() > 0, such that for
any η such that 0 < η < η(ε), the map
ϕε,η : Bε ∩ X ∩ f −1 (S)η → Sη ,
induced by f over the circle of radius η centered at the origin 0 in the
complex plane C, is a locally trivial topological fibration.
www.pdfgrip.com
128
In particular, this theorem shows that the homology

H∗ (Bε ∩ X ∩ f −1 (t))
or the cohomology
H∗ (Bε ∩ X ∩ f −1 (t), C)
does not depend on t ∈ Sη and on ε, η chosen appropriately.
6. Local Constructible Sheaves and Whitney Conditions

Since the constant sheaf C is a constructible sheaf, one may also consider the
neighbouring cycles of a constructible sheaf L along the function f : X → C
as the sheaf cohomology H∗ (Bε ∩ X ∩ f −1 (t), L).
Let p be a projection CN into Ck+1 which defines a k-th polar variety
Pk (X , x, p) of X at a point x ∈ X . Then, one has neighbourhood U and V
of x and p(x) in X and Ck+1 , such that p induces a map π : U → V . One
can show that π(Pk (X , x, p) ∩ U ) is a complex analytic subset of V and π
induces a locally trivial topological fibration of π −1 (V \ π(Pk (X , x, p) ∩ U ))
over V \ π(Pk (X , x, p) ∩ U ).
The sheaf (R` π∗ )(CU ), whose fiber at y ∈ V is the `-th cohomol-
ogy of π −1 (y), is a constructible sheaf. The Euler characteristic χk (X , x)
of the general fiber of π is called the k-th vanishing Euler characteris-
tic of X at x. At x, one has dimx (X ) Euler characteristics χ(X , x) :=
(χ1 (X , x), . . . , χdimX (X , x)).
For simplicity assume that X is equidimensional. Then, one has a char-
acterization of a Whitney stratification by a result of Lê and Teissier
(Ref. 13, Théorème (5.3.1)) similar to the one of Teissier given above:
Theorem 6.1. Let X be an equidimensional complex analytic set. Let S =

(Xα )α∈A be a stratification of X . Suppose that, for any pair (Xα , Xβ ), such
that Xα ⊂ X β , the Euler characteristics (χ1 (Xβ , y), . . . , χdimy Xβ (Xβ , y))
are constant for y ∈ Xα , then, the stratification S is a Whitney stratifica-
tion.
7. Neighbouring Cycles
We saw in Section 5 shows that any analytic function on an open set defines
locally a locally trivial fibration on a circle S. The last theorem of Section
5 shows that this extends to functions on any complex analytic sets.
Let f : X → C be a complex analytic function on a complex analytic
subset of CN . On the fiber f −1 (f (x)) of the function f through any point
www.pdfgrip.com
129
x of X one can define a sheaf Rk (ψf −f (x) )(C), with Rk (ψf −f (x) )(C)y '
Hk (Fy , C) where Fy is a fiber of the Milnor fibration defined above at y ∈
f −1 (f (x)) by f .
This sheaf Rk (ψf −f (x) )(C) is a constructible sheaf on the fiber of f over
f (x). One calls it the sheaf of k-th neighbouring cycles of f at x.
Similarly, for any constructible sheaf L, one can define the sheaf of
neighbouring cycles Rk (ψf −f (x) )(L) of f at x, where Rk (ψf −f (x) )(L)y '
Hk (Fy , L).
When f is defined on Cn+1 and has an isolated singular point at
x, the sheaf Rk (ψf −f (x) )(C) is non-zero when k = n or 0. In this case,
R0 (ψf −f (x) )(C) is the constant sheaf on f −1 (x) and Rn (ψf −f (x) )(C) is a
sheaf whose value at x is Cµ and which is zero on a neighbourhood of x
outside {x}. In this special case, when n ≥ 1, one also call Rn (ψf −f (x) )(C)
the sheaf of n-neighbouring cycles of f at x.
When the complex analytic space X is a Milnor space (see Ref. 10) and
the function f : X → C has isolated singularities, i.e. there is a Whitney
stratification of X , such that the restriction of f to the strata has maximal
rank except at isolated points, one can define the same type of sheaf over the
fiber above the image of a singularity. For instance, complete intersection
spaces, e.g., hypersurfaces, are Milnor spaces.
Because of the fibration theorem, neighbouring cycles and vanishing
cycles at a point y of f −1 (f (x)) are endowed with the monodromy of the
fibration. We have the important theorem (see e.g. Ref. 8, Theorem I p. 89,
or Ref. 4):
Theorem 7.1. The monodromy automorphism of neighbouring (or van-

ishing) cycles is a quasi-unipotent automorphism, i.e. its eigenvalues are
roots of unity.
8. Constructible Complexes
Constructible sheaves over a complex analytic set X make a category where
objects are constructible sheaves over X , morphisms are morphisms of
sheaves. Unit morphisms are identities of constructible sheaves and the
composition is the composition of sheaf morphisms.
One can notice that, for each morphism of constructible sheaves over
X , one can define the kernel, the image and the cokernel of the morphism.
A complex of sheaves of complex vector spaces over X is a sequence
of morphisms (ϕn )n∈Z , say φn : En → En+1 , such that, for every n ∈ Z,
ϕn ◦ ϕn−1 = 0.
www.pdfgrip.com
130
Complexes of sheaves of complex vector spaces over X make a category

whose objects are complexes of sheaves of complex vector spaces over X
and morphisms from (ϕn )n∈Z into (ψn )n∈Z are families (hn )n∈Z of sheaf
morphisms such that, for any n ∈ Z,
hn+1 ◦ ϕn = ψn ◦ hn .
In practice we are interested in bounded complexes (ϕn )n∈Z of sheaves

of complex vector spaces over X such that for all n ∈ Z except a finite
number ϕn is the trivial morphism from the null-sheaf 0 into itself. These
complexes make also a full subcategory D b (X ) of the preceding one. Now
let (ϕn )n∈Z ∈ Db (X ), since ϕn ◦ ϕn−1 = 0, the image of ϕn−1 is a subsheaf
of the kernel of ϕn . The quotient of the kernel sheaf of ϕn by the image
sheaf of ϕn−1 is by definition the n-th cohomology of (ϕn )n∈Z .
We say that the complex (ϕn )n∈Z is constructible over X , if it is bounded
and, for any n ∈ Z, the n-th cohomology of (ϕn )n∈Z is a constructible sheaf
over X .
Constructible complexes over make a full subcategory Dcb (X ) of Db (X ).
Given a constructible complex K over X , one can consider the hyperco-
homology H∗ (X , K) of K (see Ref. 1, Chap. XVII).
Using the notations of 7, one can consider the neighbouring cycles
Rk (ψf −f (x) )(K) of K along f at x, where
Rk (ψf −f (x) )(K)y ' Hk (Fy , K).
9. Vanishing Cycles
Let X be a complex analytic space and K a constructible complex over
X . Let x be a point of X and f : U → C be a complex analytic function
defined on an open neighbourhood U of x in X . We have defined the k-th
neighbouring cycles of K along f − f (x) at x as the sheaf Rk (ψf −f (x) )(K),
where Rk (ψf −f (x) )(K)y ' Hk (Fy , K), over the space f −1 (f (x)).
It is convenient to define the sheaf Rk (ψf −f (x) )(K) as the k-th cohomol-
ogy of a bounded complex R(ψf −f (x) )(K). The definition of the complex
R(ψf −f (x) )(K) uses derived categories and is rather abstract. Then, there
is a natural morphism from the restriction K|f −1 (f (x)) to R(ψf −f (x) )(K).
There is a natural triangle in the appropriate derived category:
K|f −1 (f (x)) → R(ψf −f (x) )(K)
+1
- .
R(φf −f (x) )(K)
www.pdfgrip.com
131
The complex R(φf −f (x) )(K) is called the complex of vanishing cycles of
K along f − f (x). The cohomology gives a long exact sequence:
→ Hk (K|f −1 (f (x))) → Rk (ψf −f (x) )(K) → Rk (φf −f (x) )(K)
→ Hk+1 (K|f −1 (f (x))) →
The sheaf Rk (φf −f (x) )(K) is the sheaf of k-th vanishing cycles of K along
f.
In special cases these sheaves are easy to interpret. Let X = Cn+1 and
f : Cn+1 → C be complex analytic function with an isolated critical point
at 0. Let U be an open neighbourhood of 0 in Cn+1 where f has the only
critical point 0. Let the complex K be the complex having one term in
degree 0 equal to the constant sheaf C over Cn+1 and the sheaf 0 in other
degrees. The results of J. Milnor in Ref. 16 show that:


C if k=0

k
R (ψf −f (0) )(K|U ) = 0 if k 6= 0, n


Cµ at 0 if k = n and 0 at x 6= 0
Therefore if n ≥ 2, the complex R(ψf −f (0) )(K|U ) is the complex with the
constant sheaf C|U in degree 0, the sheaf with one non-trivial stalk Cµ at
0 in degree n, and all morphisms are zero.
The complex R(φf −f (0) )(K|U ) has only one term in degree n which is
the sheaf with one non-trivial stalk Cµ at 0, all the other terms in degree
6= n being 0.
One can observe that in this special case of isolated critical point the
complex of vanishing cycles of the complex K|U along f consists of a sheaf
non-trivial in the degree equal to the dimension of f −1 (0) which has only
one non-trivial stalk over the isolated critical point 0.
It can be proved that it is true for any space X satisfying the Milnor
condition (see Ref. 10, §5) for functions having isolated singularities in the
general sense of (Ref. 10, §1) and for any complex KX equal to the constant
sheaf CX over X in degree 0 and to the trivial sheaf 0 in other degrees.
The support of a constructible sheaf L on the complex analytic space
X is the complex analytic subspace Y closure of the set of points x where
Lx 6= 0.
One says that a constructible complex K on CN satisfies the support
condition if the codimension of the support of its i-th cohomology Hi (K) is
≥ i.
The Verdier dual of a constructible complex K on CN is the derived
complex RHomCCN (K, CCN ). Beware that the Verdier dual of a complex on
www.pdfgrip.com
132
CN is not the complex of duals. On the other hand, there is also a notion of
duality for a constructible complex on a complex analytic space. However
in this general case the construction is not as “simple” as the Verdier dual.
The construction of the Verdier dual is usually difficult. Properties of the
Verdier dual can be found in Ref. 3. One can prove that the Verdier dual
of a constructible complex is also a constructible complex.
One says that the constructible complex K on CN satisfies the cosupport
condition, if the Verdier dual Ǩ of K satisfies the support condition.
A constructible complex K on CN is perverse, if it satisfies the sup-
port condition and the cosupport condition. We shall call perverse sheaf a
perverse constructible complex.
One can prove that more generally, if K is a perverse sheaf on CN , the
complex of vanishing cycles along a function f which has an isolated critical
point in the general sense of Ref. 10, §1 at 0 for all the restrictions of f to
the closures of the strata on which all the cohomologies Hk (K) are locally
constant, in an open neighbourhood U of 0, consists in the degree equal
to the dimension n of f −1 (0), i.e. for Rn (ψf −f (0) )(K)|U , of a sheaf which
has only one non-trivial stalk over the isolated critical point 0 and 0 in the
other degrees:
Rn (φf −f (0) )(K)0 = Ck
Rn (φf −f (0) )(K)y = 0, for y 6= 0
One has the following result due to P. Deligne (see e.g. Ref. 11):
Theorem 9.1. Let K be a perverse sheaf on CN . Let x ∈ CN . For almost

all affine functions ` of CN , the sheaf of vanishing cycles of K along ` in
an open neighbourhood of x is either zero or a complex which is zero in all
degrees except in degree equal to the dimension of N − 1, where it has a
non-trivial stalk only at x.
We can define:
Definition 9.1. Let K be a perverse sheaf on CN . The subvariety V(K) of

the cotangent bundle T ∗ (CN ) of CN is the characteristic variety of K if,
a point (x, `) ∈ V(K) if and only if it belongs to the closure of the points
(y, l) for which there is a neighbourhood U where Rk (φl−l(y) )(K)|U = 0 for
k 6= n = N − 1 and RN −1 (φl−l(y) )(K)|U is a non-trivial skyscraper sheaf
with a non-zero fiber at y.
www.pdfgrip.com
133
10. Holonomic D-Modules

Let X a complex analytic manifold of complex dimension n. On X consider
the sheaf D of complex analytic differential operators. Notice that this sheaf
is filtered by the sheaves D(k) of differential operators of order ≤ k.
We consider sheaves on X which are coherent left D-modules. We shall
call these sheaves simply D-modules.
For instance, if X is a Riemann surface, in this meeting L. Narvaez
considered D-modules over a complex analytic manifold of dimension 1.
Since the sheaf D itself is coherent, a D-module M is locally of finite
presentation, i.e. for any x ∈ X , there is an open neighbourhood Ux over
which one has an exact sequence:
p ψ ϕ
(D)|Ux → (D|Ux )q → M|Ux → 0.
One can notice that ϕ((D(k)|Ux )q ) = M(k) defines a good filtration of M

(see the lectures of F. Castro).
The complex analytic space Specan ⊕k≥0 (D(k)|Ux )/(D(k − 1)|Ux ) cor-
responding to the commutative graded ring gr(D|Ux )associated to the fil-
tration of D|Ux by the D(k)|Ux is the cotangent space of Ux . The support
of the gr(D|Ux )-module ⊕k≥0 M|Ux (k)/M|Ux (k − 1) is the characteristic
variety Ch(M|Ux ) of the D|Ux -modules M|Ux . One can show that this
definition does not depend on the local finite presentation of M.
We shall say that the D-module M is holonomic if the dimension of the
characteristic variety Ch(M) is equal to the dimension n of the manifold
X.
As it is done in Ref. 7, we obtain the following relation between the char-
acteristic variety of a holonomic D-module and the topology (see Proposi-
tion 10.6.5 of Ref. 7):
Theorem 10.1. Let M be a holonomic D-module on a complex analytic

manifold X. There is a Whitney stratification (Xα )α∈A of X, such that
the characterisitc variety of M is contained in the union of the conormal
∗
bundles TX α
X of the strata Xα in X.
Recall that if S is a submanifold of X, the conormal TS∗ X of S in X is

the bundle of (s, `) ∈ T ∗ X, such that ` is a linear form which vanishes on
Ts (S). It can be seen that, when X is a manifold and S is submanifold, the
conormal bundle TS∗ X of S in X is a Lagrangean submanifold of T ∗ X.
In fact we can obtain a more precise result.
www.pdfgrip.com
134
As it is done for n = 1 in the lectures of L. Narvaez, we define the

analytic solutions of D-module M to be the complex RHomD (M, O). This
is the stable (derived) version of the “naive” definition HomD (M, O).
M. Kashiwara proved in Ref. 6 that:
Theorem 10.2. If the D-module M is holonomic, the analytic solutions

RHomD (M, O) of M is a constructible complex which satisfies the support
condition.
A theorem of Z. Mebkhout (see Ref. 15) shows that:
Theorem 10.3. The Verdier dual of the analytic solutions RHomD (M, O)
of the holonomic D-module M is the analytic solutions of a holonomic D-
module, dual of M.
It follows that:
Corollary 10.1. The analytic solutions RHomD (M, O) of the holonomic

D-module M is a perverse sheaf.
From this result and the ones of the preceding section we can state a
result of D. T. Lê and Z. Mebkhout:12
Theorem 10.4. The characteristic variety of an holonomic D-module M

is the characteristic variety of the perverse sheaf of its analytic solutions.
References
1. H. Cartan and S. Eilenberg, Homological Algebra (Princeton University Press,
1956).
2. R. Godement, Topologie algébrique et théorie des faisceaux, Actualités Sci.
Ind. No. 1252. Publ. Math. Univ. Strasbourg No. 13 (Hermann, Paris, 1958).
3. M. Goresky and R. MacPherson, Intersection Homology II, Invent. Math.
72, 77 (1983).
4. Periods of Integrals on Algebraic Manifolds, Mimeographed notes (U.C.
Berkeley, 1965-1966).
5. H. Hamm and Lê Dũng Tráng, Ann. Sci. École Norm. Sup. 6, 317 (1973).
6. M. Kashiwara, Pub. Res. Inst. Math. Sci. 10, 563 (1975).
7. M. Kashiwara and P. Schapira, Acta Math. 142, 1 (1979).
8. A. Landman, Trans. Amer. Math. Soc. 181, 89 (1973).
9. Lê Dũng Tráng, Some remarks on relative monodromy, in real and complex
singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math.,
Oslo, 1976), pp.397-403, Sijthoff and Noordhoff, Alphen aan den Rijn (1977).
10. Lê Dũng Tráng, J. Algebraic Geom. 1, 83 (1992).
11. Lê Dũng Tráng, Bol. Soc. Mat. Mex. 4, 229 (1998).
www.pdfgrip.com
135
12. Lê Dũng Tráng and Z. Mebkhout, C. R. Acad. Sci. Paris Sér. I Math. 296,
129 (1983).
13. Lê Dũng Tráng and B. Teissier, Cycles évanescents, sections planes et condi-
tions de Whitney II, in Singularities, Part 2 (Arcata, Calif., 1981), pp.65-103,
Proc. Sympos. Pure Math. 40, Amer. Math. Soc., Providence, RI (1983).
14. J. Mather, Notes on topological stability, Harvard booklet (1971).
15. Z. Mebkhout, Théorème de dualité pour les DX -modules cohérents, C. R.
Acad. Sci. Paris Sér. A-B 285 (1977), A785 - A787/ Math. Scand. 50, 25
(1982).
16. J. Milnor, Singular Points of Complex Hypersurfaces, Annals of Mathematics
Studies, No. 61 (Princeton University Press, Princeton, N.J., 1968).
17. B. Teissier, Variétés polaires II. Multiplicités polaires, sections planes, et
conditions de Whitney, in Algebraic geometry (La Rábida, 1981), pp.314-
491, Lecture Notes in Math. 961 (Springer, Berlin, 1982).
18. R. Thom, Bull. Amer. Math. Soc. 75, 240 (1969).
19. H. Whitney, Ann. of Math. 81, 496 (1965).
February 4, 2010 18:28 WSPC - Proceedings Trim Size: 9in x 6in 04˙delabaere
www.pdfgrip.com
136
SINGULAR INTEGRALS AND THE STATIONARY

PHASE METHODS
E. DELABAERE
Université Nantes, Angers, Le Mans
Université d’Angers, UMR CNRS 6093
2 Boulevard Lavoisier, 49045 Angers Cedex 01, France
∗ E-mail: [email protected]
The paper is based on a course given in 2007 at an ICTP school in Alexan-

dria, Egypt. It aims at introducing young scientists to methods to calculate
asymptotic developpments of singular integrals.
Keywords: Singular integrals, Asymptotic in the complex domain, Steepest de-

scent, Singularities. AMS class. : 30E15, 30E20, 32Cxx, 32S30, 34E05, 34Mxx,
41A60
1. Introduction
Classically a singular integral is an integral whose integrand reaches an
infinite value at one or more points in the domain of integration. Even so,
such integrals can converge as shown in the following example.
Example Z : Hilbert transform For f ∈ L2 (R), we would like to define the

1 f (x − y)
integral dy. Since y = 0 is a singular point for the integrand,
π R y
we define instead
Z
1 f (x − y)
H[f ](x) = lim dy.
ε→0 π |y|>ε y
It can be shown that H[f ] is well defined in L2 (R). In fact one gets an
automorphism of L2 (R),
H : L2 (R) → L2 (R), H2 = −Id,
which is called the Hilbert transform.29
In this lecture, we shall be interested in analytic functions defined by
integrals. One more example is provided by the Gauss hypergeometric func-
tions.
www.pdfgrip.com
137
Example : Gauss hypergeometric function We briefly discuss here

the Gauss hypergeometric functions (see also Le Dung Trang, M. Jambu,
M. Yoshida in this volume). For a ∈ C we write as usual
Γ(a + n)
(a)n = = a(a + 1) · · · (a + n − 1), (a)0 = 1.
Γ(a)
(Γ is the Gamma function). We consider for c 6= 0, −1, −2, · · · ,
X∞
(a)n (b)n n
F (a, b; c; z) = z (1)
n=0
(c)n n!
which converges absolutely for z ∈ C, |z| < 1. We have also, for < c > < b >
0 and |z| < 1, the Euler’s formula :
Z 1
Γ(c)
F (a, b; c; z) = tb−1 (1 − t)c−b−1 (1 − zt)−a dt (2)
Γ(b)Γ(c − b) 0
Indeed, since for |tz| < 1
X∞
(a)n n n
(1 − tz)−a = t z
n=0
n!
one has, for |z| < 1 (normal convergence)

Z 1
Γ(c)
tb−1 (1 − t)c−b−1 (1 − zt)−a dt
Γ(b)Γ(c − b) 0
X∞ Z 1
(a)n Γ(c) n
= z tn+b−1 (1 − t)c−b−1 dt
n=0
n! Γ(b)Γ(c − b) 0
and we recognize the Euler’s Beta function. Therefore, for < c > < b >
0, |z| < 1,
Z 1
Γ(c)
tb−1 (1 − t)c−b−1 (1 − zt)−a dt
Γ(b)Γ(c − b) 0
X∞
(a)n Γ(n + b)Γ(c − b) n
= z
n=0
n! Γ(n + c)
X∞
(a)n (b)n n
= z .
n=0
(c)n n!
As a consequence of the Lebesgue dominated convergence theorem, we note
that the integral representation (2) allows to extend analytically F (a, b; c; z)
in z to | arg(1 − z)| < π (the cut plane C\]1, +∞[). This raises the following
questions :
www.pdfgrip.com
138
• what it the behaviour of F (a, b; c; z) at z = 1 ?

• what it the behaviour of F (a, b; c; z) at z = ∞ ?
Here part of the answers comes from the fact that the Gauss hypergeo-
metric function F (a, b; c; z) is solution of the following Gauss hypergeomet-
ric equation,
h d2 d i
z(1 − z) 2 + c − (a + b + 1)z − ab f = 0. (3)
dz dz
Now (cf. L. Narváez, this volume):
• From the Cauchy existence theorem for linear differential equa-

tions, any solution f of (3) extends analytically in z as a multival-
ued holomophic function on C\{0, 1} (or on CP1 \{0, 1, ∞}).
• Actually equation (3) is an example of a linear differential equation
of Fuchsian type : the singular points 0, 1, ∞ are regular singular
points (to see what happens at ∞, make the change of variable
1
z = ).
Z
This ensures that for any solution f of (3), one can write locally f
as a finite sum,
X
f (z) = φα,k (X)X α (Log X)k , φα,k ∈ C{X}.
α∈C,k∈N
1
(Here X stands for X = z, X = z − 1 and X = respectively
z
depending on where we localize.)
Remark : to calculate the above decomposition at ∞ explicitely for the

Gauss hypergeometric function F (a, b; c; z) it is convenient to use its Mellin-
Barnes integral representation18 (see also M. Granger, this volume, for
Mellin integrals).
Roughly speaking, in this course we shall be mainly concerned by an-
alytic functions defined as Laplace transforms of solutions of linear dif-
ferential equations with regular singular points, for instance the following
function,
Z ∞
I(a, b; c; k) = e−kz F (a, b; c; z) dz
0
where we integrate along a path on C\{1}. For such a function, we shall be

interested in the (Poincaré) asymptotics at infinity in k.
www.pdfgrip.com
139
The paper is base on a course given at the ICTP “School on Algebraic

approach to differential equations”, Alexandria - Egypt, November 12-24,
2007. It aims at introducing the reader into a subject where analysis and
geometry are linked. It is written so as to be self-contained up to some
other courses given during the school and written in this volume.
The structure of the paper is as follows.
In section 2 we consider some gentle examples of analytic functions
defined as Laplace transforms in dimension 1 so as to introduce the reader
into the subject. In section 3 we concentrate on properties of integrals of
holomorphic differential forms along cycles so as to derive, in section 4
the asymptotics of Laplace-type integrals. Two appendices remind some
results in homology theory and some properties of analytic spaces which
are required in the paper.
2. Laplace-Type Integrals and Asymptotics: Some

Examples
2.1. A basic example
We start this section with the following Laplace-type integral
 Z

 I(k) = e−kf σ

R
(4)


 2
f (z) = z , σ = dz
which defines a holomorphic function in <(k) > 0 (Lebesgue dominated
convergence theorem). We are interested in the asymptotics of I(k) when
|k| → +∞.
2.1.1. First method

By a direct calculation, one gets
Z √
2 ds π
I(k) = e−s =
R k 1/2 k 1/2
(s = k 1/2 z)
Of course, this exact result provides the asymptotics of I(k).
2.1.2. Second method

By symmetry, we reduce I(k) into a Laplace transform :
www.pdfgrip.com
140
Z +∞ Z +∞
−kf dt Γ(1/2)
I(k) = 2 e σ = e−kt = 1/2
0 0 t1/2 k
(symmetry) (f = t)
For the last equality, we have used the following formula:
Z +∞ l
−kt α l d Γ(α + 1)
e t (ln t) dt =
0 dα k α+1 (5)
l ∈ N, <(α) > −1.
2.1.3. Third method : the stationary phase method

The first two methods rely on exact calculations. The third method develops
another viewpoint and prepares the reader for some generalisations.
First step : localisation We recall that were are interested in the asymp-
totics of I(k) when |k| → +∞, <(k) > 0. When |k| 1, we note that the
integrand z ∈ R 7→ e−kf (z) has a support which is essentially concentrated
near z = 0 where df = 0, approaching the Dirac δ-distribution, see Fig. 1.
This is why we localize near that point. This can be done in the following
√ √
way : we introduce a t0 > 0 and we note that f −1 ([0, t0 ]) = [− t0 , t0 ].

− t 1/2
0 0 t 1/2
0

f

0 t0
(b)

(a)
Fig. 1. (a) The graph of z ∈ R 7→ |e−kf (z) | for k = 100. (b) The map f .
www.pdfgrip.com
141
We thus write


 I(k) = I(k, t0 ) + R(k, t0 )

Z Z (6)


 I(k, t0 ) = e−kf σ, R(k, t0 ) = e−kf σ
0≤f ≤t0 f >t0
We introduce also the following open sectorial neighbourdhood of in-

finity of aperture π − 2δ,
π π
Σr,δ = {k ∈ C, |k| > r, | arg k| < − δ}, r > 0, 0 < δ < . (7)
2 2
(see Fig. 2) so that,
∀ k ∈ Σr,δ , <(k) > |k| sin(δ) > r sin(δ).
Therefore,
∃ C > 0 , ∃ M > 0, ∀ k ∈ Σr,δ , |R(k, t0 )| ≤ M.e−C|k| (8)
(C = C(t0 , δ) = t0 sin(δ)), and in particular (since for all N ∈ N,
|k|N e−C|k| → 0 when |k| → +∞) :
C
∀N ∈ N, ∃C > 0, ∀ k ∈ Σr,δ , |R(k, t0 )| ≤ . (9)
|k|N +1
Remark : In fact, (8) implies the following more precise result (use the
Stirling formula21 ):
Γ(N + 2)
∃C > 0, ∀N ∈ N, ∀ k ∈ Σr,δ , |R(k, t0 )| ≤ C N +1 . (10)
|k|N +1
0 Σ r, δ
Fig. 2. The sectorial neighbourdhood of infinity Σr,δ .

www.pdfgrip.com
142
Second step : reduction to an incomplete Laplace transform We

now concentrate on
Z
√ √
I(k, t0 ) = e−kf σ, Γt0 = [− t0 , t0 ].
Γ t0
For k fixed, we notice that e−kf σ ∈ Ω1 (C), where Ω1 (C) is the space of
holomorphic differential 1-forms on C.
[ Recall : if z = x + iy and z̄ = x − iy, (x, y) ∈ R2 , one introduces the

differential 1-forms dz = dx + idy and dz̄ = dx − idy. If g ∈ C 1 (U, C) with
U ⊂ C an open set, then
∂g ∂g
dg = dz + dz̄.
∂z ∂ z̄
Now
∂g ∂g
g ∈ O(U ) ⇔ = 0 ⇔ dg = dz
∂ z̄ ∂z
Cauchy-Riemann
and by definition,
ω ∈ Ω1 (U ) ⇔ ∃ h ∈ O(U ), ω = hdz.
We recall also that if U is a simply connected open subset of C, then every
closed differential 1-form is exact (Poincaré lemma). We note also that if
ω ∈ Ω1 (U ), then ω is closed. ]
From the fact that e−kf σ ∈ Ω1 (C), there exists ϕ ∈ Ω0 (C) = O(C) such
that dϕ = e−kf σ (we are just saying that e−kf (z) has a primitive ϕ). Thus,
Z Z Z
I(k, t0 ) = e−kf σ = dϕ = ϕ
Γ t0 Γ t0 γ(t0 ) (11)
Stokes formula
where γ(t0 ) is the boundary of Γt0 , that is the formal difference of two
points,
√ √
γ(t0 ) = [ t0 ] − [− t0 ].
In other words (11) is just a pedantic way of writing:
√ √
I(k, t0 ) = ϕ( t0 ) − ϕ(− t0 ). (12)
We note Dt0 = D(t0 , η) ⊂ C the open disc centred at t0 with radius
/ Dt0 (0 < η < t0 ). We note also by S 1 = ∂Dt0 its
small enough so that 0 ∈
boundary with its natural orientation, Fig. 3, and we consider for t ∈ Dt0 :
√
I1 (k, t) = ϕ( t).
www.pdfgrip.com
143
f 1/2
0 t0 0 t0 t 1/2
t
S1
Fig. 3.
Obviously I1 (k, t) ∈ O(Dt0 ) ∩ C 0 (Dt0 ) so that, by the Cauchy formula,

Z √
1 ϕ( s)
∀ t ∈ Dt0 , I1 (k, t) = ds.
2iπ S 1 s − t
By the inverse function theorem, f realizes a biholomorphic mapping be-
√
tween a neighbourhood of t0 and Dt0 . In other words, one can take f as
a coordinate : taking s = f (z) = z 2 , we get
I I
1 ϕ(z) 1 ϕ df
∀ t ∈ Dt0 , I1 (k, t) = 2zdz =
2iπ z2 − t 2iπ f −t
√ √
where we integrate along an oriented loop surrounding t0 and t, see Fig.
3.
This integral representation has the following consequence : since the path
of integration does not depend on t, we obtain: for t ∈ Dt0 ,
I
d 1 ϕ df
I1 (k, t) =
dt 2iπ (f − t)2
Also,

ϕ dϕ ϕ df
d = −
f −t f − t (f − t)2
and by Stokes
I
1 ϕ
d = 0.
2iπ f −t
Therefore
I
d 1 dϕ
∀ t ∈ D t0 , I1 (k, t) = .
dt 2iπ f −t
We remark now that, as far as f can be chosen as a coordinate, one can
write:
dϕ = df.Ψ
www.pdfgrip.com
144
where Ψ is a holomorphic 0-form, usually denoted as a quotient form

dϕ σ
Ψ= = e−kf ,
df df
which is the meromorphic function here
e−kf (z)
Ψ(z) = .
2z
What we get is finally:
I
d 1 df.Ψ e−kf √
∀ t ∈ D t0 , I1 (k, t) = = |
dt 2iπ f −t 2z f =t,z= t
Cauchy formula
√
What have been done for I1 (k, t) can be done as well for I2 (k, t) = ϕ(− t),
t ∈ Dt0 , so that :
Z ! Z Z
d d dϕ σ
∀ t ∈ D t0 , I(k, t) = ϕ = |f =t = e−kt |f =t
dt dt γ(t) γ(t) df γ(t) df
(13)
which is nothing but writing

d 1 1 e−kt
∀ t ∈ D t0 , I(k, t) = e−kt √ − √ = √ .
dt 2 t −2 t t
To write I(k, t0 ) as an incomplete Laplace transform, what remains to do
now is to notice that
lim I(k, t) = 0
t>0,t→0
(see (12), we recall that ϕ ∈ O(C)), so that I(k, t0 ) can we written as the
following incomplete Laplace transform :
Z t0 Z ! Z t0
−kt σ dt
I(k, t0 ) = e |f =t dt = e−kt √ . (14)
0 γ(t) df 0 t
Final step From (14) one has

Z +∞
dt Γ(1/2)
I(k, t0 ) = e−kt √ + r(k, t0 ) = 1/2 + r(k, t0 ) (15)
0 t k
where
Z +∞
dt
r(k, t0 ) = − e−kt √
t0 t
www.pdfgrip.com
145
is an exponentially decreasing function in k ∈ Σr,δ : the reasoning is

quite similar to what have been done for R(k, t0 ) (in fact here r(k, t0 ) =
−R(k, t0 )) so that
C
∀N ∈ N, ∃C > 0, ∀ k ∈ Σr,δ , |r(k, t0 )| ≤ . (16)
|k|N +1
Finally, (6), (15), (9) and (16) imply :
Γ(1/2) C
∀N ∈ N, ∃C > 0, ∀ k ∈ Σr,δ , |I(k) − |≤ .
k 1/2 |k|N +1
2.2. Airy and the steepest-descent method in dimension 1

We turn to the Airy equation
d2 y
− xy = 0 (17)
dx2
which has ∞ as an irregular singular point (set y(x) = Y (X) in (17) with
d2 Y dY
X = 1/x, then X 5 + 2X 4 − Y = 0).
dX 2 dX
By Fourier transformation, one easily obtains the following particular solu-
tion of (17),
Z +∞
1 s3
)
Ai(x) = ei(xs+ 3 ds, (18)
2π −∞
known as the Airy function. Applying the Cauchy existence theorem for
linear differential equations, we know that Ai(x) extends as a holomorphic
function on C.
Our aim in this subsection is to analyse the asymptotic behaviour of
the Airy function when x → +∞. More precisely we shall demonstrate the
following result (see also Ref. 5 for an hyperasymptotic viewpoint):
Proposition 2.0.1. If Σr,δ is the sectorial neighbourdhood of infinity de-

fined by (7), then
XN
√ an C
∀N ∈ N, ∃C > 0, ∀ k ∈ Σr,δ , 2 πk 1/6 e2k/3 Ai(k 2/3 ) − n
≤ N +1 .
n=0
k |k|
3 Γ(n + 1/6)Γ(n + 5/6)

where the an belong to C (explicitely, an = (− )n ).
4 2πΓ(n + 1)
www.pdfgrip.com
146

Fig. 4. Trace on the real of the Airy function.
2.2.1. Prepared form

We assume for the moment that x > 0. In (18) we first make the change of
variable
s = iu
then we set
k = x3/2 , ζ = uk −1/3 .
We thus get :
Z
k 1/3 ζ3
for k > 0, Ai(k 2/3 ) = e−kg(ζ) dζ, g(ζ) = ζ − (19)
2iπ C 3
where the path C is drawn on Fig. 5. It is easy to see that (19) is equivalent
to writing:
Z
2/3 k 1/3 ζ3
for k > 0, Ai(k ) = e−kg(ζ) dζ, g(ζ) = ζ − (20)
2iπ C1 3
where the path ZC1 is drawn on Fig. 5. (By integration by part, first write
Z
1 2ζ
e−kg(ζ) = e−kg(ζ) dζ then check that this last integral is
C k C (1 − ζ 2 )2
equal to the integral along C1 ).
2.2.2. Integrability and space of allowed paths of integration

We pause a moment. Let us consider the integral
Z Z Z
I(k) = e−kg(ζ) dζ = c? (e−kg(ζ) dζ) = e−kg(c(s)) c0 (s)ds (21)
c R R
www.pdfgrip.com
147
π/6
C C1 c
0 a b
ΣR(A0 )
Fig. 5. On the left the paths C and C1 . On the right the open set ΣR (A0 ). On this
right picture, a, b and c are relative 1-cycles of C with respect to ΣR (A0 ), and d is a
relative 1-boundary. The relative 1-cycles a and b are homologous and thus represent the
same class in H1 (C, ΣR (A0 )). This 1-homology group is a free Z-module generated by
the classes of a and c.
where c : s ∈ R 7→ c(s) ∈ C is assumed to be smooth-piecewise,

lims→±∞ |c(s)| = +∞ and with moderate variations (|c0 (s)| has moder-
ate growth). (C and C1 are such paths).
To make this integral I(k) absolutely convergent, we would like that
<(−kg(ζ)) → −∞ when ζ → ∞ along the endless path of integration
c. This translates into the condition that <(kζ 3 ) → −∞ at infinity along c.
We write this in another way. If
θ = arg(k) ∈ S = R/2πZ (22)
we introduce the following open subsets of S,
[2
π 2π θ π 2π θ
Aθ,δ = ] + j − + δ, + j − − δ[, δ > 0 small enough
j=0
6 3 3 2 3 3
[2
π 2π θ π 2π θ
Aθ = ] + j− , + j − [,
j=0
6 3 3 2 3 3
(23)
and the associated family of sectorial neighbourdhoods of infinity
ΣR (Aθ,δ ) = {ζ ∈ C, |ζ| > R, arg(ζ) ∈ Aθ,δ }, R > 0.
(24)
ΣR (Aθ ) = {ζ ∈ C, |ζ| > R, arg(ζ) ∈ Aθ }, R > 0.
www.pdfgrip.com
148
Then, the integral I(k) will be absolutely convergent if apart from a com-
pact set the (support of the) path c belongs to ΣR (Aθ,δ ) for some R > 0
and δ > 0.
To precise the space of such allowed paths c, we now introduce the
1-homology group of the pair (C, ΣR (Aθ,δ )),
H1 (C, ΣR (Aθ,δ )) = H1 (C, ΣR (Aθ,δ ); Z).
(see Appendix A for some notions on homology theory, see also Fig. 5).
We see that H1 (C, ΣR (Aθ,δ )) ∼
= Z2 . Also, if R0 ≥ R, so that ΣR0 (Aθ,δ ) ⊆
ΣR (Aθ,δ ) one has a natural isomorphism
RRR0 : H1 (C, ΣR0 (Aθ,δ )) → H1 (C, ΣR (Aθ,δ ))
since the pairs (C, ΣR0 (Aθ,δ )) and (C, ΣR (Aθ,δ )) are homotopic. This allows
to define the following 1-homology group by inverse limit:
A
H1 θ,δ (C) = lim H1 (C, ΣR (Aθ,δ ))
←
(“we make R → +∞”) which is still a free Z-module of rank 2. Finally we

define the 1-homology group we have in mind by inductive limit
A
H1Aθ (C) = lim H1 θ,δ (C) (25)
→
(“we make δ → 0”), a free Z-module of rank 2 which deserves to be the

space of equivalent classes of our allowed paths of integration.
Note indeed that if c and c0 are two endless paths (smooth-piecewise, with
moderate variations) which represent the same element in H1Aθ (C), then
(by integrability and by Stokes since e−kg(ζ) dζ ∈ Ω1 (C) is closed)
Z Z
e−kg(ζ) dζ = e−kg(ζ) dζ.
c c0
2.2.3. The steepest-descent method

We return to our integral (20) and we recall that we assume k > 0. From
what have been said previously, in (20) we can replace C1 by any other
path of integration (smooth-piecewise, with moderate variations) belong-
ing to the same class of homology [C1 ] ∈ H1Aθ (C).
This is that freedom that we are going to use in the following station-
ary phase method8,13 known as the Riemann-Debye “steepest-descent” or
“saddle-point” methoda .
a Itis amazing how Riemann introduced this method to obtain what is now known as
the Riemann-Siegel formula for the zeta function, see Ref. 4.
www.pdfgrip.com
149

C1
C2

Fig. 6. The vector field ∇ <(kg) , some steepest-descent curves of <(−kg) for k > 0,
and the path C2 .
In the homology class of C1 we look for a path C2 (or a linear com-

bination of paths) which follows a level curve of =(kg) : if kg(z) =
P (x, y) + iQ(x, y), such a curve is an integral curve of the vector field
∂Q ∂Q
∂x − ∂y. By the Cauchy-Riemann equalities, this vector field is also
∂y ∂x
∂P ∂P
the gradient field ∇ <(kg) = ∂x + ∂y whose (oriented) level curves
∂x ∂y
are the steepest-ascent curves of <(kg), or equivalently the steepest-descent
curves of <(−kg). See Fig. (6). Note that g has two critical points ζ = ±1
where dg = 0. These are nondegenerate saddle points (since g is holomor-
phic, g as well as =(g) and <(g) are harmonic functions and therefore they
satisfy the maximum principle : any critical point is necessarily a saddle
point).
Consequence : In the homology class [C1 ] ∈ H1Aθ (C) of C1 , we can choose
a path C2 (smooth-piecewise, with moderate variations) such that:
• C2 is a level curve of =(kg),
• the support of C2 (i.e., its image) contains the saddle point ζ = +1.
See Fig. (6). In practice to obtain C2 , one can consider the deformation of
C1 under the flow Φ of the gradient field <(kg) : if C1 : s ∈ R 7→ c(s) ∈ C,
consider the homotopy map
h : R+ × R → C, h(τ, s) := Φ (τ, C1 (s)) .
Write C2 : s ∈ R 7→ C2 (s) ∈ C and
assume
that C2 (0) = 1. Then for
k > 0 the map s ∈ R 7→ −k g C(s) − g(1) defines a real function with
www.pdfgrip.com
150
the following variations :
s −∞ 0 +∞
0
−k g C(s) − g(1) % &
−∞ −∞
In (20) thus replacing C1 by C2 , one gets

Z
k 1/3 ζ3
Ai(k 2/3 ) = e−kg(ζ) dζ, g(ζ) = ζ − (26)
2iπ C2 3
and we note that the integral defines a holomorphic function in k ∈ Σr,δ

(Σr,δ is the sectorial neighbourdhood of infinity defined by (7)).
We are now in position to copy what we done in 2.1.
2.2.4. Localisation
In (26) we set ζ = 1 + z so as to be centred on the critical point:
k 1/3 e−2k/3
Ai(k 2/3 ) = I(k)
2iπ
for k ∈ Σr,δ , Z (27)
−kf (z) 2z3
I(k) = e dz, f (z) = −z −
Γ 3
where Γ is the translated of C2 .
We now fix a t0 > 0 and we note Γt0 the restriction of Γ (precisely its
support) to f −1 ([0, t0 ]), see Fig. 7.
Γ
z2(t 0 )
Γt 0 f
0 0
t0
z1(t 0)
B D
Fig. 7. The path Γ and its restriction Γt0 . The path Γ is mapped twice on R+ , resp.
Γt0 twice on [0, t0 ], by f .
www.pdfgrip.com
151
We write
Z
I(k) = I(k, t0 ) + R(k, t0 ), I(k, t0 ) = e−kf (z) dz. (28)
Γ t0
Concerning R(k, t0 ) one easily shows that
∃ C > 0 , ∃ M > 0, ∀ k ∈ Σr,δ , |R(k, t0 )| ≤ M.e−C|k|
and in particular
C
∀N ∈ N, ∃C > 0, ∀ k ∈ Σr,δ , |R(k, t0 )| ≤ (29)
|k|N +1
or more precisely
Γ(N + 2)
∃C > 0, ∀N ∈ N, ∀ k ∈ Σr,δ , |R(k, t0 )| ≤ C N +1 . (30)
|k|N +1
This allows to concentrate on I(k, t0 ).
2.2.5. First method

We briefly mention this first method just for completeness. Since the origin
is a nondegenerate critical point for f , by the complex Morse lemma one
can find a local coordinate Z such that
f (z) = Z 2
Explicitely,
1 5 1
z = iZ + Z 2 − iZ 3 − Z 4 + · · · ∈ C{Z}.
6 72 27
Therefore, for t0 > 0 small enough,
Z √t0
2
I(k, t0 ) = √ e−kZ h(Z) dZ,
− t0
X 1 5 4
h(Z) = hn Z n = i + Z − iZ 2 − Z 3 + · · · ∈ C{Z}.
3 24 27
n≥0
As we shall see later (see 2.2.8), to get the asymptoticsZ of I(k, t0 ) and I(k)
P
when |k| → +∞ in Σr,δ reduces in exchanging and and in integrating
on R:
√ −1/2 √
π 5 π −3/2
I(k) ∼ i −i k +···
k 48
www.pdfgrip.com
152

Fig. 8. Trace on the real of L.
2.2.6. Second method : reduction to an incomplete Laplace transform

(1)
Z
We have here in mind to represent I(k, t0 ) = e−kf (z) dz as an incom-
Γ t0
plete Laplace transform, that is in a way to make the change of variable
t = f (z). For doing that we may introduce the complex algebraic curve
L = {(z, t) ∈ C2 , t = f (z)}.
This is an analytic submanifold of C2 , dimC L = 1, see Fig. 8.
(z, t) ∈ L
If one considers π1 : ↓ then (L, π1 ) is nothing but the graph of
z∈C
f . If one considers instead
(z, t) ∈ L\{(0, 0), (−2, −4/3)} (z, t) ∈ L
π2 | : ↓ π2 : ↓
t ∈ C\{0, −4/3} t∈C
then (L\{(0, 0), (−2, −4/3)}, π2|) is a 3-sheeted covering of C\{0, −4/3}
which is the Riemann surface of the inverse function z(t), 0 and −4/3
being algebraic branch points of order 2. ((L, π2 ) is the so-called ramified
Riemann surface).
However for our purpose we do not need such a global information, since
the support of Γt0 is localized near z = 0.
www.pdfgrip.com
153
We thus introduce an open disc D = D(0, η), resp. an open ball B =

B(0, ε) with 0 < ε < 2, see Fig.7 and Fig. 8. We can choose 0 < η small
enough (η < 43 ) so that:
• for any t0 ∈ D? = D\{0}, the set B ∩ f −1 (t0 ) consists in two

distinct points z1 (t0 ) and z2 (t0 ).
• for t near t0 , each of these two functions z1 (t) and z2 (t) represents
a germ of holomorphic functions at t0 (this is a consequence of the
implicit function theorem since t0 6= 0 is not a critical value of f ).
In other words, z1 (t) (say) extends to a multivalued function z(t) ∈ O^D ? ,t0
with two determinations z1 (t) and z2 (t) (cf. L. Narváez, this volume).
It is not hard to calculate z(t) explicitely:

1 1 5
z(t) = t − t2 + · · · +t 1/2
−i + it + · · ·
6 27 72
where the series expansions belong to C{t} with 4/3 for their radius of
convergence.
We now go back to I(k, t0 ) as defined by (28), where we assume that
t0 > 0, t0 ∈ D? . Making the change of variable z = z(t) one has
Z
∨ ∨ 1
I(k, t0 ) = e−kt J (t) dt, J (t) = −
g
Γ t0
z(t) 2 + z(t)
where the path Γf

t0 drawn on Fig. 9 should be thought of as a path on
(L, π2 ).
J(t) = −1
z1(t)(2+z1(t))
0
t0
J(t) = −1
z2(t)(2+z2(t))
Fig. 9. g
The path Γ t0 .
www.pdfgrip.com
154
This reduces into

Z t0
I(k, t0 ) = e−kt J(t) dt,
0
∞
X 5 385 2 (31)
−1/2
J(t) = bn tn−1/2 =t i − it + it + · · ·
n=0
24 3456
where the series expansion belongs to C{t} with 4/3 for its radius of con-
vergence.
2.2.7. Third method : reduction to an incomplete Laplace transform

(2)
Rather than the previous method, we can repeat what has been done in
2.1.3. With the notations and hypotheses of the previous subsection, we
have
Z
z3
I(k, t0 ) = e−kf σ, with σ = dz, f (z) = −z 2 −
Γ t0 3
so that e−kf σ ∈ Ω1 (B) and thus e−kf σ is closed. Since B is simply con-
nected we deduce that e−kf σ is exact : there exists ϕ ∈ Ω0 (B) = O(B)
such that dϕ = e−kf σ. By Stokes,
Z Z
I(k, t0 ) = dϕ = ϕ (32)
Γ t0 γ(t0 )
where γ(t0 ) is the boundary of Γt0 ,
γ(t0 ) = [z2 (t0 )] − [z1 (t0 )]. (33)
For t near t0 we introduce a small circle S 1 of radius small enough

(S 1 ⊂ D? ) which surrounds t0 and t in the positive oriented way, Fig. 10.
To this circle S 1 and to γ(t) we associate δt γ(t) = δ2 − δ1 which consists
in the formal difference of the two closed paths drawn of Fig. 10, where δ2
surrounds z2 (t0 ), z2 (t), and δ1 surrounds z1 (t0 ), z1 (t).
By Cauchy and taking f as a coordinate, one gets:
Z Z
1 ϕdf
I(k, t) = ϕ= .
γ(t) 2iπ δt γ(t) f −t
This implies that (see 2.1.3):

Z Z
d 1 dϕ 1 df dϕ
I(k, t) = =
dt 2iπ δt γ(t) f − t 2iπ δt γ(t) f − t df
www.pdfgrip.com
155
z1(t)
z1(t0 )
t
f
0 t0 0
z2(t)
S1 z2(t0 )
D
B
Fig. 10.
where the quotient form is the following meromorphic function:

dϕ e−kf (z)
=− .
df z(2 + z)
By Cauchy, this means that
Z Z
d dϕ σ
I(k, t) = |f =t = e−kt |f =t .
dt γ(t) df γ(t) df
Using the fact that

lim I(k, t) = 0
t>0,t→0
one finally concludes that :

Z t0 Z
σ
I(k, t0 ) = e−kt J(t) dt, J(t) = |f =t . (34)
0 γ(t) df
2.2.8. The asymptotics

We first analyse the asymptotics of I(k, t0 ) for k ∈ Σr,δ . From (31),
Z t0 ∞
!
X
−kt n−1/2
I(k, t0 ) = e bn t dt
0 n=0
so that for any N ∈ N,
Z +∞ N
X Z +∞ N
X
−kt n−1/2 −kt
I(k, t0 ) = e bn t dt − e bn tn−1/2 dt
0 n=0 t0 n=0
Z t0 X∞
−kt
+ e bn tn−1/2 dt
0 n=N +1
www.pdfgrip.com
156
Using (5), this means that

N
X Γ(n + 1/2)
I(k, t0 ) − bn
n=0
k n+1/2
∞
X Z t0 N
X Z +∞
−kt n−1/2
= bn e t dt − bn e−kt tn−1/2 dt.
n=N +1 0 n=0 t0
Making the change of variable t = t0 s one obtains:

N
X
Γ(n + 1/2)
I(k, t0 ) − bn
=
n=0
k n+1/2
∞
X Z 1 XN Z +∞
−kt0 s n−1/2
b n t0 n+1/2
e s ds − b n t0 n+1/2
e−kt0 s tn−1/2 ds.
n=N +1 0 n=0 1
n−1/2 N +1/2
Since s ≤s in the integrals, it follows that for k ∈ Σr,δ ,
N ∞
!
X Γ(n + 1/2) X n+1/2 Γ(N + 3/2)
I(k, t0 ) − bn n+1/2
≤ |bn |t0 N +3/2 .
k
n=0 n=0 t0 sin(δ)|k|
(35)
Using (28) and (30), we can conclude for the asymptotics that:
N
X Γ(n + 1/2)
∃C > 0, ∀N ∈ N, ∀ k ∈ Σr,δ , I(k) − bn
n=0
k n+1/2
Γ(N + 3/2)
≤ C N +3/2 . (36)
|k|N +3/2
Note this result (36) is stronger than the usual Poincaré asymptotics (see
Malgrange21).
With this result and (27) we get the proposition 2.0.1.
2.2.9. Remark 1 : geometric monodromy

Let us go back to the function J(t) defined by (34), namely
Z
σ
J(t) = |f =t , γ(t) = [z2 (t)] − [z1 (t)].
γ(t) df
This function J(t) can be viewed as defining a germ of holomorphic func-

tions at t0 which extends to a multivalued function on D ? . So we now think
of J as a germ in O^ D ? ,t0 .
www.pdfgrip.com
157
Let λ be the following loop in D ? ,

λ : [0, 1] → D? , λ : s 7→ t = t0 e2iπs .
This loop λ generates the first homotopy group π1 (D? , t0 ). We note T =
ρ(λ) the associated monodromy operator, where
^
ρ : π1 (D? , t0 ) → AutO(D? )−alg. (O D ? ,t0 )
(cf. L. Narváez, this volume). Then

T (J) = −J
because z1 (t) and z2 (t) are exchanged when t follows the loop λ, so that
γ(t0 ) is transformed into −γ(t0 ).
Thus J is of finite determination and one infers from the geometry that
T (t1/2 J) = t1/2 J,
that is t1/2 J is uniform.
Since we know by other means that J is a multivalued function with
moderate growth at 0, we deduce that J belongs to the Nilsson class at
t = 0 (cf. L. Narváez, this volume). This implies that J(t) is solution of a
holomorphic linear differential equation L(J) = 0, L ∈ EndC (OD ) with a
regular singular point at 0. By inverse Laplace transformation, there is a
relationship between this linear differential equation and the Airy equation
(17) we started with. It can be shown that

−1/2 1 5 1 3
J(t) = it F , , ;− t
6 6 2 4
where F (a, b; c; z) is the Gauss hypergeometric function.
2.2.10. Remark 2 : an example of local system

In 2.2.2 we introduced the family of homology groups
Fθ = H1Aθ (C), θ ∈ S = R/2πZ,
each of them being a free Z-module of rank 2.
More generally, for any U ⊂ S, a connected open arc of length < π, one
can defined
\
AU = Aθ ⊂ S
θ∈U
and the group

F(U ) = H1AU (C) ∼
= Z2 .
www.pdfgrip.com
158
Fig. 11. Continous deformation of a relative 1-cycle a when θ moves along S.
The data of all these groups (F(U )) with the obvious isomorphisms
ρV,U : F(U ) → F(V ), V ⊂U
makes a sheafb of groups F on the topological space S which is locally

constant (F is a local system). Note that F is not a constant sheaf, see Fig.
11.
2.2.11. Exercise
z3 z4
One considers f (z) = + . Note that z = 0 is a degenerate singular
3 4
point of order 2 for f , whereas z = −1 is a nondegenerate singular point,
see Fig. 12.
For k ∈ Σr,δ (for some r > 0 and 0 < δ < π/2) we define
Z +∞ Z +i∞
−kf (z)
I1 (k) = e dz, I2 (k) = e−kf (z) dz.
−∞ −i∞
Show that for k ∈ Σr,δ :
b ForU ⊂ S, one constructs a section Γ ∈ F (U ) by considering a covering (U i )i∈I of U

made of open connected arcs of length < π and a family (Γi )i∈I of elements of F (Ui )
such that for all i, j ∈ I one has
ρUi ∩Uj ,Ai (Γi ) = ρUi ∩Uj ,Uj (Γj ).

www.pdfgrip.com
159

Fig. 12. The vector field ∇ <(kf ) , and some steepest-descent curves of <(−kf ) for
z 3 z 4
k > 0 and f (z) = + .
3 4
(1) I1 (k) is asymptotic to

k/12
√ 1 31 1
e 2π + +··· .
k 1/2 12 k 3/2
(2) I2 (k) is asymptotic to
√
2π 3 1 32/3 Γ(2/3) 1 91
− + + +···
9Γ(2/3) k 1/3 6 k 2/3 8k
With the methods described in this subsection, we know that the
asymptotics of I2 (k) is governed by those of
Z t0 Z
−kt dz
I(k, t0 ) = e J(t) dt, J(t) = |f =t .
0 γ(t) df
with t0 > 0 small enough, and a convenient γ(t). With the notations of
2.2.9, show that T 3 (J) = J where T is the monodromy operator.
2.3. An example in higher dimension

2.3.1. Two division lemmas
We start with two lemmas (see Refs. 19 and 25).
Lemma 2.0.1 (Local division of forms). We assume that f is a C ∞
function on the open set U ⊂ Rn (resp. a holomorphic function on U ⊂ Cn ).
www.pdfgrip.com
160
Let ω be a C ∞ differential p-form (resp. a holomorphic differential p-form,

ω ∈ Ωp (U )) such that
df ∧ ω = 0
on U . If z0 ∈ U is not a critical point of f (df (z0 ) 6= 0) then there exists a

neighbourhood Uz0 ⊂ U of z0 and a C ∞ differential (p − 1)-form Ψ on Uz0
(resp. a holomorphic differential (p − 1)-form, Ψ ∈ Ωp−1 (Uz0 )) such that
ω = df ∧ Ψ.
Moreover, the restriction of Ψ to any hypersurface Uz0 ∩f −1 (t) is a uniquely

defined C ∞ (p − 1)-form (resp. holomorphic (p − 1)-form).
One usually notes Ψ as a quotient form
ω
Ψ=
df
which is called a Leray-Gelfand quotient form.
Proof. We just consider the case p = n so that ω = g(z)dz1 ∧ · · · ∧ dzn .

• From the implicit function theorem, one can choose a local system of
coordinates s = (s1 , · · · , sn ) such that f = s1 is a coordinate while ω =
h(s)ds1 ∧ · · · ∧ dsn . Defining Ψ = h(s)ds2 ∧ · · · ∧ dsn gives the result. To
make things explicit, there exists a C ∞ (resp. holomorphic) diffeomorphism
φ : s ∈ (V, 0) ⊂ Rn 7→ z ∈ (Uz0 , z0 ) such that
φ? ◦ f (s) = f ◦ φ(s) = s1 .
Meanwhile,

φ? ◦ ω(s) = φ? ◦ g(s) det dφ(s) ds1 ∧ · · · ∧ dsn .
Defining

φ? ◦ Ψ(s) = φ? ◦ g(s) det dφ(s) ds2 ∧ · · · ∧ dsn
one gets

φ? ◦ ω = d φ? ◦ f ∧ φ? ◦ Ψ = φ? ◦ df ∧ φ? ◦ Ψ = φ? ◦ df ∧ Ψ .
• Assume that ω = df ∧ Ψ1 = df ∧ Ψ2 , then using the above local system of

coordinates one has ds1 ∧ (Ψ1 − Ψ2 ) = 0 where f = s1 , so that Ψ1 − Ψ2 =
ds1 ∧ (· · · ) and finally i?t ◦ (Ψ1 − Ψ2 ) = 0, where it : Uz0 ∩ f −1 (t) → Uz0 is
the canonical injection.
www.pdfgrip.com
161
Lemma 2.0.2 (Global division of forms). With the notations of the

previous lemma, we assume that df ∧ ω = 0 on U and furthermore
∀ z ∈ U, df (z) 6= 0.
ω
Then there exists a C ∞ (p − 1)-form on U such that
df
ω
ω = df ∧ .
df
ω
Moreover, the restriction of to any hypersurface U ∩f −1 (t) is a uniquely
df
defined C ∞ (p − 1)-form (resp. holomorphic (p − 1)-form).
Proof. We recall that on a C ∞ -manifold (separated with countable basis),

every open covering has a C ∞ partition of 1 subordinate to it. Lemma 2.0.2
is thus a consequence of Lemma 2.0.1.
ω
Note that in the holomorphic case is only a C ∞ (p − 1)-form on U as a
df
rule. However its restriction to a level hypersurface U ∩ f −1 (t) is holomor-
phic thanks to its uniqueness and Lemma 2.0.1.
2.3.2. An application
We consider here the integral
Z
I(k) = ω, ω = e−kf σ
Rn
where z = (z1 , · · · , zn ), σ = g(z)dz1 ∧ · · · ∧ dzn (Rn with its canonical
orientation). We assume that f ∈ C ∞ (Rn , R) and g ∈ Cc∞ (Rn , R).
Assuming that the support of the differential form ω does not meet the
singular locus of f (the set of critical points) and using Lemma 2.0.2, one
obtains :
Z Z Z Z ! Z Z !
ω ω −kt σ
ω= df ∧ = dt = e dt
Rn Rn df R γ(t) df R γ(t) df
(Fubini)
where γ(t) is level hypersurface f = t (naturally oriented as the boundary
of f ≤ t).
As a matter of fact, the above equality extends to the case where the support
of ω meets some singular fibres f = t since:
• the set of critical values of the smooth function f is a Lebesgue null set
(Sard’s theorem).
• each fibre f = t is a Lebesgue null set for the Lebesgue measure in Rn .
www.pdfgrip.com
162
2.3.3. An example
By way of a simple example we consider for k > 0 (or even k ∈ Σr,δ for
some r > 0 and 0 < δ < π/2),
Z
I(k) = e−kf σ, f (z1 , z2 ) = z1a + z2b , (a, b) ∈ (2N? )2
R2
and σ = g(z1 , z2 )dz1 ∧ dz2 , g ∈ C[z1 , z2 ]. Writing

X
g(z1 , z2 ) = gp,q z1p z2q
p,q
with (p, q) ∈ N2 , it is straighforward (by Fubini) to calculate I(k),

X p+1 q+1
gp,q Γ( a )Γ( b ) δp,q = 0 if p or q is odd
I(k) = 4δp,q p+1 q+1 , (37)
p,q
ab k a + b δp,q = 0 otherwise
Also from what precedes (since f is a positive function),

Z +∞ Z
−kt σ
I(k) = e J(t) dt, J(t) =
0 γ(t) df
with γ(t) the closed curve γ(t) = {(z1 , z2 ) ∈ R2 , f (z1 , z2 ) = t} and from
(37) by inverse Laplace transformation,
X gp,q p + 1 q + 1 p+1 + q+1 −1
J(t) = 4δp,q B( , )t a b
p,q
ab a b
with B the Euler Beta function. See also M. Granger, this volume.
3. Integrals of Holomorphic Differential Forms along Cycles

We would like to extend the constructions seen in §2 to multidimensional
integrals defined in the complex field. Here we shall define and analyse the
functions
Z
J(t) = ω
γ(t)
where:
• ω is a holomorphic differential (n − 1)-form on U ⊂ Cn ,
• γ(t) is a (n − 1)-cycle on the level hypersurface f = t,
• f : U ⊂ Cn → C is holomorphic.
The notions of homology theory which are used in the sequel are recalled
in the appendix A, see also Refs. 12, 16 and 28.
www.pdfgrip.com
163
3.1. Integrals along cycles on fibres

3.1.1. Definition and first property
We assume here that f : Cn → C is a (non constant) holomorphic function
(f ∈ O(Cn )) and
Xthat ω ∈ Ω
n−1
(Cn ) is a holomorphic differential (n − 1)-
form : ω = gI (z)dzI where z = (z1 , · · · , zn ), gI ∈ O(Cn ) and if
|I|=n−1
I = (i1 , · · · , ip ) then dzI = dzi1 ∧ · · · ∧ dzip .
We assume that t0 ∈ C is not a critical value of f so that f −1 (t0 ) is an
holomorphic submanifold of Cn of dimension dimC f −1 (t0 ) = n − 1.
Let γ(t0 ) ∈ Zn−1 (f −1 (t0 )) be a (n − 1)-cycle of f −1 (t0 ).
Z
Proposition 3.0.2. With the above hypotheses, the integral ω de-
γ(t0 )
pends only on the homology class [γ(t0 )] ∈ Hn−1 (f −1 (t0 )).
Z Z
Proof. We first mention that ω is a short way of writing ω
γ(t0 ) i? γ(t0 )
where i : Zf −1 (t0 ) ,→ C n
Z is the canonical injection.
We have ω= i? ω where i? ω ∈ Ωn−1 (f −1 (t0 )) is a holomorphic
i? γ(t0 ) γ(t0 )
differential (n − 1)-form on f −1 (t0 ). Since dimC f −1 (t0 ) = n − 1, this means
that i? ω is of maximal order, so that i? ω is closed. We conclude with the
Corollary A.4.1.
3.1.2. The Ehresmann fibration theorem

Z
We would like to think of the integral ω with [γ(t)] ∈ Hn−1 (f −1 (t)) as a
γ(t)
continuous function of t. This requires to being able to deform continuously
γ(t0 ) to a nearby γ(t) for t near t0 . This relies on some fibration theorems.
We first mention the following result, see Ref 12:
Theorem 3.1 (Ehresmann fibration theorem). If M and N are dif-
ferentiable (resp. C ∞ ) manifolds and if f : M → N is a proper submersion,
then f is a locally trivial differentiable (resp. C ∞ ) fibration.
If M is a manifold with boundary ∂M , and if both f : M → N and
f | : ∂M → N are proper submersions, then both f and f | are locally trivial
differentiable (resp. C ∞ ) fibrations.
So when f is a submersion (for any m ∈ M , rank Tm f = dim N ) and

proper (f −1 (a compact) is compact), then for every t0 ∈ N , there exist
www.pdfgrip.com
164
a neighbourhood U of t0 and a differentiable (resp. C ∞ ) diffeomorphism

Φ : f −1 (U ) → U × f −1 (t0 ) (a so-called local trivialization map) such that
the following diagram commutes:
Φ
f −1 (U ) −→ U × f −1 (t0 )
f| & . pr1
U
(in other words, f is the projection of a differentiable (resp. C ∞ ) fibre

bundle).
If M is a manifold with boundary ∂M , and if both f : M → N and
f | : ∂M → N are proper submersions, then the above diagram respects the
boundary and the interior M \∂M of M .
We give also the following result12 :
Proposition 3.1.1. If f : M → N is a locally trivial differentiable (resp.

C ∞ ) fibration and if N is contractible, then this fibration is trivial.
For a given f ∈ O(Cn ) the Ehresmann fibration theorem cannot be

applied directly since in general (when n ≥ 2) such a map is not proper (just
think of f (z1 , z2 ) = z1 − z2 ) and furthermore f has as a rule a nonempty
set Singf ⊂ Cn of critical points.
We now consider a z0 ∈ / Singf . We introduce the open ball B =
B(z0 , ε) ⊂ Cn and the closed ball B whose boundary is ∂B = {z ∈
Cn , kz − z0 k = ε}. We introduce also the open disc D = D(t0 , η) ⊂ C
where t0 = f (z0 ) and we note
X = B ∩ f −1 (D), X = B ∩ f −1 (D).
We assume that X contains no critical point of f : this is true at least for

ε > 0 small enough since z0 ∈ / Singf .
−1
Also we assume that f (t0 ) and ∂B intersect transversally : apart from its
singular points, the locus f −1 (t0 ) can be seen as a C ∞ submanifold of R2n
of real dimension 2(n − 1) while ∂B can be considered as a C ∞ submanifold
of R2n of real dimension 2n − 1. Then f −1 (t0 ) and ∂B cut transversally if
∀ z ∈ f −1 (t0 ) ∩ ∂B, Tz (f −1 (t0 )) + Tz ∂B = R2n .
It can be shown that this is true at least for ε > 0 small enough. Now by
the implicit function theorem, f −1 (t) and ∂B will intersect transversally
for any t ∈ D provided that η > 0 is chosen small enough.
www.pdfgrip.com
165
With these hypotheses, one obtains that X = B ∩ f −1 (D) is a C ∞ manifold

with boundary ∂X = ∂B ∩ f −1 (D). Applying the Ehresmann fibration
X
theorem to f | ↓ one obtains:
D
Theorem 3.2. We note B = B(z0 , ε) ⊂ Cn an open ball and D =

D(t0 , η) ⊂ C with t0 = f (z0 ). We note
X = B ∩ f −1 (D), X = B ∩ f −1 (D)
and we assume that X contains no critical point of f . Then ε and η can be

X X
chosen so that f | ↓ and f | ↓ are locally C ∞ trivial fibrations.
D D
As a matter of fact, because D is contractible, we deduce from Propo-

X
sition 3.1.1 that the fibre bundle f | ↓ is trivial : there exists a C ∞ diffeo-
D
morphism Φ such that the following diagram commutes:
Φ
X −→ D × Xt0
f| & . pr1 , Xt = B ∩ f −1 (t) (38)
D
3.1.3. Applications
We assume that the conditions of Theorem 3.2 are fulfilled. We start with
a given (n − 1)-cycle γ(t0 ) of Xt0 ,
X
γ(t0 ) = ni σi ∈ Zn−1 (Xt0 )
and with (38) we define for t ∈ D :

X
γ(t) = ni Φ−1 −1
t (σi ) = Cq−1 (Φt )(γ(t0 ))
where we have written Φ−1 t (.) = Φ

−1
(t, .) and used the notations of A.3.
What we get is a (so called) “horizontal deformation” of γ(t0 ), that is γ(t0 )
has been deformed continuously (in fact in a C ∞ manner) into a (n−1)-cycle
γ(t) of Xt .
www.pdfgrip.com
166
Remark The C ∞ diffeomorphim Φ−1

t : Xt0 → Xt gives rise to an isomor-
phism of homology groups
Φ−1
t ? : Hq (Xt0 ) → Hq (Xt ), ∀q
G
and the disjoint union Hq (Xt ) of all these groups makes a local system
t∈D
on D which is a constant sheafc .
Proposition 3.2.1. If ω ∈ Ωn−1 (X) is a holomorphic differential (n − 1)-

form in X, and with the above notations and hypotheses, the integral J(t) =
Z
ω defines a C ∞ function in D.
γ(t)
Z
Proof. By Proposition 3.0.2, the integral J(t) = ω depends only on
γ(t)
the homology class [γ(t)] ∈ Hn−1 (Xt ). In the homology class [γ(t0 )] ∈
Hn−1 (Xt0 ) we choose a piecewise differentiable (n − 1)-cycle γt0 and we
note
X X
γ(t) = ni Φ−1
t (σi ) = ni σi (t).
By construction, each (q − 1)-simplex σi (t) is a C ∞ function of t and since

Z Z
ω= σi (t)? ω
σi (t) ∆q−1
one concludes with the dominated Lebesgue theorem.
Proposition
Z 3.2.2. With the above notations and hypotheses, the integral
J(t) = ω defines a holomorphic function in D.
γ(t)
Proof. To be holomorphic is a local property. For t0 ∈ D we consider t ∈ D

near t0 . We introduce a small oriented circle S 1 in D which surrounds t0
and t in the natural way (cf. Fig. 10).
We construct a cycle δt γ(t) ∈ Zn (X\Xt ) in the following way: if
X
γ(t) = n i σi , S 1 : ∆1 → D
G
c Note F = Hq (Xt ). If U ∈ D is an open neighbourhood of t0 , then F (U ) is the set
t∈D
of the sections γ : t ∈ U 7→ γ(t) ∈ F defined as γ(t) = Φ−1
t ? (γ(t0 )) ∈ Hq (Xt ) where
γ(t0 ) is some given element of Hq (Xt0 ).
www.pdfgrip.com
167
γ(s) γ(t)
Fig. 13. The cycles γ(t) ∈ Zn−1 (Xt ) and δt γ(t) ∈ Zn (X\Xt ).
we define (with Φ as in (38))

δt σi : ∆1 ⊗ ∆n−1 → X\Xt , δt σi = Φ−1 (S 1 , σi )
(∆1 ⊗ ∆n−1 is here the oriented product of ∆1 and ∆n−1 ) andd
X
δt γ(t) = ni (δt σi ).
[
The reader may think of δt γ(t) as the union γ(s) ⊂ X\Xt , see Fig. 13.
s∈S 1
In that way, one gets a homomorphism
δt : Hn−1 (Xt ) → Hn (X\Xt )
which is just a particular case of the so-called Leray coboundary or tube
operator1,19,25 . With this definition, we now gives a lemma:
Lemma 3.2.1. For t near t0 ,

Z
1 df ∧ ω
J(t) = .
2iπ δt γ(t) f −t
Z Before to give the proof (from Ref. 1), it is worth noting that the integral
df ∧ ω
only depends on the homology class [δt γ(t)] ∈ Hn (X\Xt )
δt γ(t) f − t
df ∧ ω
because ∈ Ωn (X\Xt ) is closed.
f −t
d Note that δt σi can be seen as an element of Cn (X\Xt ) in the sense of appendix A :
Pn n
Pn
just consider for instance the map j=0 λj ej ∈ ∆ 7→ λ0 e0 + (1 − λ0 )e1 , j=1 (λj +

λ0 1 n−1
n
)ej−1 ∈ ∆ × ∆ whose inverse map is obvious.
www.pdfgrip.com
168
Proof. By Fubini one gets

Z Z Z !
1 df ∧ ω 1 ds
= ω
2iπ δt γ(t) f −t 2iπ S1 γ(s) s−t
We write this as
Z
1 df ∧ ω
2iπ δt γ(t) f − t
Z Z ! Z Z Z !
1 ds 1 ds
= ω + ω− ω
2iπ S1 γ(t) s − t 2iπ S1 γ(s) γ(t) s−t
| {z } | {z }
A1 A2
Z
By the Cauchy formula (because ω does not depend on s) one obtains
γ(t)
Z Z Z
1 ds
A1 = ω= ω.
2iπ S1 s−t γ(t) γ(t)
Z
By Proposition 3.2.1 we know that ω is a C ∞ function in t, so that by
γ(t)
Taylor,
Z Z
ω− ω = Cste.(s − t) + O([s − t|2 )
γ(s) γ(t)
where Cste is a constant complex number. Using this result in A2 then

making the radius of the circle S 1 tend to 0 one gets A2 = 0.
This lemma proves Proposition 3.2.2: in the integral representation

Z
1 df ∧ ω
J(t) =
2iπ δt γ(t) f − t
the cycle δt γ(t) does not depend on t for t near t0 in D. Since the integrand
is holomorphic in t, one concludes with the dominated Lebesgue theorem.
3.2. The polynomial case

In this subsection we assume that f : Cn → C is a (non constant) polyno-
mial function.
www.pdfgrip.com
169
3.2.1. A fibration theorem and consequences

In the polynomial case we have the following fibration theorem6,31 :
Theorem 3.3. Assume that f : Cn → C is a polynomial function. Then
there exists a finite set Atypf = {t1 , · · · , tN } ⊂ C of “atypical values” such
Cn \f −1 (Atypf )
that the restriction map f | ↓ is a locally C ∞ trivial fibration.
C\Atypf
Remark To make a link between Theorems 3.3 and 3.1, let us introduce a
compactification of f which can be defined as follows. If d = deg f , define
z
G(z, z0 , t) = z0d f ( ) − tz0d = f d (z) + z0 f d−1 (z) + · · · + z0d f 0 (z) − tz0d
z0
(z0d f ( zz0 ) is the projectivization of f by the new variable z0 ; we have noted
by f l the homogeneous part of degree l of f ) and introduce the set
M = {((z : z0 ), t) ∈ PnC × C, G(z, z0 , t) = 0}
where PnC is the n-dimensional complex projective space. We consider the
embedding e : z = (z1 , · · · ,zn ) ∈ Cn
7→ (z : 1) = (z1 : · · · : zn : 1) ∈ PnC
and the map E : z ∈ Cn 7→ e(z), f (z) ∈ PnC × C, so that M is the Zariski
closure of E(Cn ) in PnC × C. We thus get a commutative diagram
E
Cn −→ M 3 ((z : z0 ), t)
f↓ ↓p
id
C −→ C3t
A compactification of f : Cn → C is given by the proper map p : M → C
which is a locally trivial fibration except over a set of points, and this
translates to the fibration f (see Refs. 31 and 6.. One has to see M as a
Whitney stratified space – see B. Teissier, this volume – and instead of
Theorem 3.1 one has in fact to use the Thom-Mather first isotopy lemma).
Of course f (Singf ) ⊂ Atypf . However, Atypf may contains other values
which come from a set Singf∞ of singular points at infinity. This is related
to the fact that the hypersurface M ⊂ PnC × C has a singular locus, namely
Σ×C where Σ is the following algebraic subset of the hyperplane at infinity
H ∞ = {x0 = 0} ⊂ PnC ,
Σ = {(z : z0 ) ∈ PnC , gradf d = f d−1 = z0 = 0} ⊂ M. (39)
What have beenGdone in §3.1.2 can be repeated. Theorem 3.3 implies
that the union Hn−1 (f −1 (t), K) (K = Z or C) makes a local
t∈C\Atypf
www.pdfgrip.com
170
system of groups on C\Atypf (a locally constant sheaf but not constant

one in general) and, for a given ω ∈ Ωn−1 (Cn ), the mapping
Z
−1
Lω : [γ(t)] ∈ Hn−1 (f (t)) 7→ ω (40)
γ(t)
defines a representation of this local system into the sheaf OC\Atypf of

holomorphic functions on C\Atypf :
Corollary 3.3.1. For ω ∈ Ωn−1 (Cn ) we consider

Z
J(t) = ω, [γ(t)] ∈ Hn−1 (f −1 (t))
γ(t)
as defining a germ of holomorphic functions at t0 ∈ C\Atypf . Then J(t)

extends to a multivalued function on C\Atypf .
Using the fibration Theorem 3.3 one can consider the continuous hor-
izontal deformation of a (n − 1)-cycle when t follows a closed path λ ∈
π1 (C\Atypf , t0 ). We thus have an action

M : λ ∈ π1 (C\Atypf , t0 ) 7→ ρ(λ) = Aut Hn−1 (f −1 (t0 ), K)
where the automorphism M(λ) is the geometric monodromy associated

with the loop λ. We can define also the action
^ ,t )
ρ : π1 (C\Atypf , t0 ) → AutO(C\Atypf )−alg. (OC\Atyp f 0
and we see that Zanalysing the action of the monodromy operator ρ(λ) on
the germ J(t) = ω just reduces by the representation (40) in analysing
γ(t)
the action of the geometric monodromy operator M(λ) on [γ(t0 )]:
Z Z
ρ(λ) ω = ω. (41)
γ(t0 ) M(λ)[γ(t0 )]
3.2.2. The finite rank case

∼ µ
We now assume
that Hn−1 (Xt0 ) = Z for t0 ∈ C\Atypf , so that
Hn−1 (f −1 (t)) makes a local system of free Z-modules of rank
t∈C\Atypf
µ on C\Atypf .
This is always true when n = 1, but also for instance when the set of atypi-
cal values Atypf arises only from isolated critical points at finite distance (in
C[z]
particular Atypf = f (Singf )). In that case and for n ≥ 2, µ = dimC
(∂f )
www.pdfgrip.com
171
is the total Milnor number ((∂f ) is the Jacobian ideal) and is the
Xsum of
the Milnor numbers at each singular point (cf. Ref. 30), µ = µz ? ,
z? ∈Singf
C{z − z? }
µz? = dimC . In a more general case for n ≥ 2, still assuming that
(∂f )
f has only isolated singular points including those at infinity (for instance
(39) defines a finite set), then for any t ∈ C\Atypf , Hn−1 (Xt ) ∼ = Zµ ⊕ A
where µ is the total Milnor number while A is some finitely generated
abelian group.7 (We shall return to that point in §4).
Let us then define a basis B(t0 ) = (γ1 (t0 ), · · · , γµ (t0 )) of (n − 1)-
cycles whose classes generate Hn−1 (f −1 (t0 )) as a free Z-module of rank
µ. The continuous horizontal deformation of B(t0 ) along a loop λ ∈
π1 (C\Atypf , t0 ) provides another basis M(λ) B(t0 ) of Hn−1 (f −1 (t0 )) :
the geometric monodromy operator M(λ) is represented in the basis B(t0 )
by an invertible matrix M (λ) ∈ GLµ (Z).
With Atypf = {t1 , · · · , tN }, the fundamental group π1 (C\Atypf , t0 ) of
C\Atypf with respect to the base point t0 is the free group generated by
the family of loops (λi )1≤i≤N drawn on Fig. 14 : the loop λi follows a path
li which goes from t0 to some tei near ti , then λi turns around ti in the
positive oriented direction and finally goes back to t0 following li in the
inverse way. Z
We return to J(t) = ω as in Corollary 3.3.1. By analytic continu-
γ(t)
ations along the path li , we can assume that t0 = tei and think of J(t) as
an element of O^Dt? ,t0 where Dti is a small disc centred on ti . From the fact
i
−1
that Hn−1 (f (t0 )) is a free Z-module of rank µ, we know that the geomet-
ric monodromy operator M(λi ) is the zero of a polynomial (for instance its
ti
λi
t0
Fig. 14. The generators of the fundamental group π1 (C\Atypf , t0 ).

www.pdfgrip.com
172
characteristic polynomial, by Cayley-Hamilton). By (41) this translates into

the fact that J(t) is of finite determination at ti (The characteristic polyno-
mial P ∈ Z[X] provides a relationship between the various determinations
J, ρ(λi )(J), ..., ρ(λi )µ (J)). Then (cf. L. Narváez, this volume):
Proposition 3.3.1. We assume that f : Cn → C is a polynomial function

with Atypf = {t1 , · · · , tN } for its set of atypical values. For t0 ∈ C\Atypf
we assume that Hn−1 (Xt0 ) ∼ = Z µ and we fix ω ∈ Ωn−1 (C n
Z ). Then any
analytic continuations li J near any ti of the germ J(t) = ω, [γ(t)] ∈
γ(t)
Hn−1 (f −1 (t)) of analytic functions at t0 is of finite determination at ti and
one can write locally li J as a finite sum
X
li J(t) = φα,l (t − ti )α Log l (t − ti )
α∈C,l∈N
where the φα,l are uniform near ti .
Next we give a (consequence of a) result due to Nilsson23,24 :
Theorem 3.4. We assume that ω is a polynomial differential (n − 1)-

form. Then, with the notations and hypotheses of Proposition 3.3.1, for
each i = 1, · · · , N , and for any sectorial neighbourhood Σ(ti ; r, a, b) of ti ,
Σ(ti ; r, a, b) = {t ∈ C, 0 < |t − ti | < r, a < arg t < b}
with r > 0 small enough, b − a < 2π, there exist M ∈ N and C > 0 such
that
∀ t ∈ Σ(ti ; r, a, b), |li J(t)| ≤ C|t − ti |−M .
Proposition 3.3.1 and Theorem 3.4 imply that each li J(t) belongs to the
Nilsson class (cf. L. Narváez, this volume). Consequently:
Corollary 3.4.1. With the hypotheses of Proposition 3.3.1 and Theorem

3.4, the multivalued function J(t) is a solution of a linear differential equa-
tion with at most regular singular points at t1 , · · · , tN .
3.2.3. Examples
First example We consider f : (p, q) ∈ C2 7→ p2 + V (q), V (q) ∈ C[q]. In
classical mechanics, p is the momentum and V (q) stands for the potential
function. For a non-critical E (E= the energy) one has H1 (f −1 (E)) ∼
= Zm−1
where m is the degree of V .
www.pdfgrip.com
173
γ1
γ2
Fig. 15. The level manifold f −1 (E) viewed as the Riemann surface of p = (E −V (q))1/2
for V (q) = q 3 and E = 1. The homology classes of the 2 cycles γ1 and γ2 drawn generate
H1 (f −1 (E)).
We now assume that V (q) = q 3 so that E = 0 is the sole critical value.

We are interested in the following “period integral”
Z
J(E) = ω
γ(E)
where ω is a polynomial differential 1-form (when ω = pdq the period inte-

gral is the “action” in classical mechanics). One can thus apply Proposition
3.3.1 and Theorem 3.4.
We analyse the geometric monodromy; We consider for E0 = 1 the basis
B(E0 ) = ([γ1 (E0 )], [γ2 (E0 )]) of H1 (f −1 (E)) as drawn on Fig. 15. If λ0 is
the natural loop which generates π1 (C? , 1), then the associated geometric
monodromy operator M(λ0 ) satisfies
M(λ0 ) : ([γ1 (E0 )], [γ2 (E0 )]) 7→ ([γ2 (E0 )], [γ2 (E0 ) − γ1 (E0 )]).
(Just see how the zeros of q 3 = E are exchanged). The geometric mon-
odromy operator M(λ represented in the basis B(E0 ) by an invert-
0 ) is thus
0 −1
ible matrix M (λ0 ) = ∈ GL2 (Z), whose characteristic polynomial
1 1
is P (X) = X 2 − X + 1. Thus, if J(E) is defined as a germ of holomorphic
functions at E0 , one has
ρ(λ0 )2 (J) − ρ(λ0 )(J) + J = 0.
In particular ρ(λ0 )6 (J) = J so that J(E) reads:
X
J(E) = an E n/6
n
where n ∈ N apart from a finite set of negative values. One can be more
precise: since the eigenvalues of M(λ0 ) are X1 = e2iπ/6 , X2 = e−2iπ/6
www.pdfgrip.com
174
one deduces that there exists l0 ∈ Z such that

X 5
X 7
J(E) = αl E 6 +l + βl E 6 +l
l≥l0 l≥l0
(cf. L. Narváez and M. Granger, this volume, see also Ref. 1). In fact l0 ≥ 0
as we shall see in a moment (cf. Theorem 3.6), and the result extends when
ω is a holomorphic 1-form.
Second example We now consider the so-called Broughton’s polynomial,6

f : (z1 , z2 ) ∈ C2 7→ z2 (1 + z1 z2 ). This polynomial has no critical point,
however its set of atypical values is Atypf = {0} : the level set f −1 (0) is
the disjoint union of the two connected components locus, z2 = 0 which
is isomorphic to C and {z1 = −1/z2, z2 ∈ C? } which is isomorphic to C? ,
while for t 6= 0 the fibre f −1 (t) has only one connected component
for t 6= 0, f −1 (t) = {z1 = (t − z2 )/z22 , z2 ∈ C? } (42)
which is isomorphic to C? .
Using (39) we can remark that (z1 : z2 : z0 ) = (1 : 0 : 0) is the sole
candidate for being a singular point a infinity (see Ref. 30).
From (42) we see that H1 (f −1 (t)) ∼
= Z for t 6= 0. Indeed, H1 (f −1 (t)) can
be generated for instance by the oriented cycle
γ(t) = {z1 = (t − z2 )/z22 , |z2 | = 1} (43)
As a consequence, the geometric monodromy operator M(λ0 ) associated

with the natural loop λ0 which generates π1Z(C? , 1) is the identity. Therefore,
if ω is a polynomial 1-form, then J(t) = ω with [γ(t)] ∈ H1 (f −1 (t))
γ(t)
?
is a holomorphic univalued function on C and byX Proposition 3.3.1 and
Theorem 3.4, there exists n0 ∈ Z such that J(t) = an tn . In fact
n≥n0
X
J(t) = a n tn .
n≥0
This result can be obtained by simple calculations by first observing that

for t 6= 0 the mapping
Ψt : (z1 , z2 ) 7→ (t−1 z1 , tz2 )
defines a diffeomorphism from the level complex curve f −1 (1) to the level
complex curve f −1 (t). This means that, starting with a 1-cycle γ(t0 ) on
a generic fibre f −1 (t0 ), its horizontal deformation γ(t) will extend toward
www.pdfgrip.com
175
infinity in the z1 direction when t → 0. Let us see how this geometric

information translates for J(t). One has
Z Z
J(t) = ω= Ψ?t ω.
γ(t) γ(1)
In particular, using also (43),

Z Z
for ω = z1m z2n dz1 then ω = tn−m−1
γ(t) γ(1)
I
(1 − z2 )m (z2 − 2)
ω = tn−m−1 dz2
z22m−n+3
Z Z (44)
for ω = z1m z2n dz2 then ω = tn−m+1
γ(t) γ(1)
I
(1 − z2 )m
ω = tn−m+1 dz2 ,
z22m−n
H m
thus the conclusion ( (1−z 2 ) (z2 −2)
z22m−n+3
dz2 = 0 if n − m − 1 < 0 and
H (1−z2 )m
z 2m−n
dz2 = 0 if n − m + 1 < 0).
2
3.3. Localisation near an isolated singularity

We assume in this subsection that f : (Cn , 0) → (C, 0) represents a germ
of holomorphic functions at 0, f (0) = 0. We assume furthermore that 0 is
an isolated critical point for f .
The following theorem is due to Milnor,22 see also Ref. 12 :
Theorem 3.5 (Milnor). We note B = B(0, ε) ⊂ Cn the open ball and

D = D(0, η) ⊂ C the open disc. We set
X = B ∩ f −1 (D), X = B ∩ f −1 (D), Xt = B ∩ f −1 (t).
Then for ε > 0 small enough and for and 0 < η = η(ε) ε the restrictions
X\f −1 (0) X\f −1(0)
f| ↓ and f | ↓ are locally C ∞ trivial fibrations. Also :
D? D?
(1) The Milnor fibration does not depend of (ε, η), in the sense that two
Milnor fibrations given by two allowed pairs (ε1 , η1 ) and (ε2 , η2 ) are
www.pdfgrip.com
176
equivalent : X(ε1 ,η1 ) and X(ε2 ,η2 ) are diffeomorphic and the following
diagram commutes:
≈
X(ε1 ,η1 ) \f −1 (0) −→ X(ε2 ,η2 ) \f −1 (0)
f| ↓ ↓ f|
? ≈
D η1 −→ Dη?2
(2) X is contractible.
(3) For t ∈ D? and Hn−1 (Xt ) ∼ = Z µ for n ≥ 2, H0 (Xt ) ∼
= Z µ+1 for n = 1,
C{z}
where µ = dimC is the Milnor number at 0.
(∂f )
In the conditions of the theorem, X is called a Milnor ball while the
homology group Hn−1 (Xt ) is called the vanishing homology group of the
singularity z = 0.
Thanks to this theorem 3.5, what have been done in the previous sec-
tions can be repeated for
Z
J(t) = ω, [γ(t)] ∈ Hn−1 (Xt )
γ(t)
where ω ∈ Ωn−1 (X). More precisely one has the following properties, see
Malgrange,20 see also Refs. 1, 2 and 3.
Z
n−1
Theorem 3.6. If ω ∈ Ω (X) and J(t) = ω, [γ(t)] ∈ Hn−1 (Xt ),
γ(t)
then J(t) viewed as a germ of holomophic function at t0 ∈ D? extends as a
multivalued holomorphic function J ∈ O ^D ? ,t0 and J is of finite determina-
tion at 0.
Also J belongs to the Nilsson class at 0. More precisely, for any sectorial
neighbourhood Σ(r, a, b) of 0,
Σ(r, a, b) = {t ∈ D, 0 < |t| < r, a < arg t < b} ⊂ D
with r > 0 small enough, b − a < 2π, there exists C > 0 such that
∀ t ∈ Σ(r, a, b), |J(t)| ≤ C.
Moreover
lim J(t) = 0
t→0, t∈Σ(r,a,b)
and
X
J(t) = φα,l tα Log l (t), φα,l ∈ C{t}.
α∈Q?+ , l∈[[0,n−1]]
www.pdfgrip.com
177
4. On the Asymptotics of Laplace-Type Integrals

We would like to analyse the asymptotics when |k| → +∞ of Laplace-type
integrals
Z
I(k) = e−kf σ
Γ
where f ∈ O(Cn ) and σ ∈ Ωn (Cn ). What we have in mind is to generalise
the steepest-descent method discussed in §2.2 on the Airy example.
4.1. Allowed chains of integration

In a similar way to what we have done in §2.2, we first have to define a
space of allowed endless contours of integration running between valleys at
infinity where <(kf ) → +∞.
We assume that
R
θ = arg(k) ∈
2πZ
is fixed and we introduce the following half-planes: for any r > 0 we set
+ + − −
SR = SR (θ) = {t ∈ C, <(teiθ ) ≥ R}, SR = SR (θ) = {t ∈ C, <(teiθ ) ≤ R}
(45)
For f ∈ O(Cn ) we introduce the family Ψ = Ψ(θ) of closed subsets A ∈ Cn
defined as follows :
−
A ∈ Ψ ⇔ ∀ R > 0, A ∩ f −1 (SR ) is compact. (46)
This family Ψ obviously satisfies the properties (A.1) of §A.5. This means
that Ψ is a family of supports in Cn in the sense of homology theory which
allows to define the chain-complex (C•Ψ (Cn ), ∂• ) of Z-modules and its as-
sociated homology groups
ZqΨ (Cn )
HqΨ (Cn ) = .
BqΨ (Cn )
X
Since for any c = ni σi ∈ CnΨ (Cn ) one has [c] ∈ Ψ one deduces from (46)
i
that
<(kf (z)) → +∞ when kzk → +∞, z ∈ [c].
In particular |e−kf (z) | = e−<(kf (z)) is exponentially decreasing when kzk →
+∞, z ∈ [c]. Nevertheless, to make the integral
Z X Z X Z
e−kf σ = ni e−kf σ = ni e−kf (σi ) σi? (σ)
c i σi i ∆n
www.pdfgrip.com
178
X
absolutly convergent for some c = ni σi ∈ CnΨ (Cn ), one needs also to
i
control the growth of the differential forms σi? (σ). This why we restrict
ourself to the polynomial case in the following theorem, see Pham26,27 :
Theorem 4.1. Assume that f ∈ C[z] is a polynomial function on Cn and

that σ is a polynomial differential n-form on Cn , σ = gdz1 ∧ · · · ∧ dzn ,
g ∈ C[z]. Then
Z
(1) The integral e−kf σ along a n-cycle Γ ∈ ZnΨ (Cn ) is well defined for
Γ
|k| > 0, arg(k) = θ and depends only on the homology class [Γ] ∈
HnΨ (Cn ).
(2) One has the following isomorphism
HqΨ (Cn ) ∼
= lim
←−
+
Hq (Cn , f −1 (SR )), ∀q
R≥R0
where the inverse limit is considered for R ≥ R0 large enough
We say more about the item 2. of the Theorem 4.1. When f is a

polynomial function we know from Theorem 3.3 that there exists a finite
Cn \f −1 (Atypf )
set Atypf ⊂ C such that the restriction map f | ↓ is a lo-
C\Atypf
cally C ∞ trivial fibration. In particular, if R0 > max |ti | then for any
ti ∈Atypf
+ +
R0 ≥ R > R0 , the pairs (Cn , f −1 (SR n
0 )) and (C , f
−1
(SR )) are homotopy
equivalent so that
∀ R0 ≥ R > R0 , ∀q, +
Hq (Cn , f −1 (SR )) ∼ +
= Hq (Cn , f −1 (SR 0 )).
+
We detail that point. Because SR 0
is contractible, there exists a C ∞ -
diffeomorphism Φ such that the following diagram commutes (for a chosen
+
t0 ∈ S R 0
):
+ Φ +
f −1 (SR 0
) −→ SR 0
× f −1 (t0 )
f| & . pr1 (47)
+
SR 0
Now for R0 ≥ R > R0 it is easy to defined a homeomorphism h such that

+ +
h : (C, SR ) → (C, SR 0 ), h|C\S + = id.
R0
www.pdfgrip.com
179
(for instance if teiθ = a + ib, define


 a if a ≤ R0
h : (a, b) ∈ R2 7→ (h1 (a), b), h1 (a) = R 0 − R0
 R0 + (a − R0 ) if a ≥ R0
R − R0
This being done we lift this homeomorphism by the fibration f , setting:
 n −1 + n −1 +
 H : (C , f (SR )) → (C , f (SR0 ))
+
H = id on restriction to Cn \f −1 (SR )
 −1
0
H = Φ ◦ (h × id) ◦ Φ elsewhere

+ + Φ + + h×id + + Φ−1 + +
(f −1 (SR ), f −1 (SR )) → (SR , SR ) × f −1 (t0 ) → (SR , SR 0) × f
−1
(t0 ) → (f −1 (SR ), f −1 (SR 0 ))
0 0 0 0
f | & . pr1 pr1 | & . f |
+ + h + +
(SR , SR ) → (SR , SR 0)
0 0
By its very construction H is a homeomorphism.

Z
Instead of working with the integrals e−kf σ with [Γ] ∈ HnΨ (Cn ),
Γ
Theorem 4.1 allows us to consider rather the integrals
Z
+
I(k) = e−kf σ, [Γ] ∈ Hn (Cn , f −1 (SR ))) (48)
Γ
for some R ≥ R0 . This means working with a class of functions defined

modulo some exponentially decreasing functions when |k| → +∞, arg k =
+
θ. Indeed, assume that Γ, Γ0 ∈ Zn (Cn , f −1 (SR ))) define the same class in
n −1 +
Hn (C , f (SR )),
+
Γ − Γ0 = c + ∂b, c ∈ Cn (f −1 (SR )), b ∈ Cn+1 (Cn ).
Z Z
We remark that e−kf σ = d(e−kf σ) = 0 because e−kf σ ∈ Ωn (Cn ) is
∂b b
+
closed. Moreover, since c ∈ Cn (f −1 (SR )) one has, for some fixed r > 0:
Z
∃C = C(r) > 0, e−kf σ ≤ Ce−R|k| , |k| ≥ r, arg(k) = θ.
c
Also, Theorem 4.1 implies that a n-cycle Γ ∈ ZnΨ (Cn ) can be represented
+
by a relative n-cycle Γ ∈ Zn (Cn , f −1 (SR ))) and, thanks to the convergence
:
Z Z
∃C > 0, e−kf σ − e−kf σ ≤ Ce−R|k| , |k| ≥ r, arg(k) = θ.
Γ Γ
Consequently :
www.pdfgrip.com
180
Corollary 4.1.1. We assume that f ∈ C[z] and that σ = gdz1 ∧ · · · ∧ dzn ,

g ∈ C[z]. We fix some R >ZR0 and r > 0. Then for |k| ≥ r, arg(k) = θ the
space of functions I(k) = e−kf σ, [Γ] ∈ HnΨ (Cn ) can be identified with
ΓZ
+
the space of functions I(k) = e−kf σ, [Γ] ∈ Hn (Cn , f −1 (SR )) modulo the
Γ
exponentially decreasing functions of type R in |k|.
+
Remark : By the long exact homology sequence of the pair (Cn , f −1 (SR ))
(cf. Proposition A.0.3), one easily obtains that (for n ≥ 2)
+
Hn (Cn , f −1 (SR )) ∼ +
= Hn−1 (f −1 (SR ))
because Cn is contractible. Then using (47) one deduces that
+
Hn (Cn , f −1 (SR )) ∼
= Hn−1 (f −1 (t)), t∈
/ Atypf . (49)
4.2. The steepest-descent method

We still assume that f ∈ C[z] so that the restriction map
Cn \f −1 (Atypf )
f| ↓ is a locally C ∞ trivial fibration, where Atypf =
C\Atypf
{t1 , · · · , tN } ⊂ C is the finite set of atypical values. Our aim here is to
+
analyse the homology group Hn (Cn , f −1 (SR ))), having in mind to extend
the steepest-descent method discussed in §2.2 for the Airy example.
Following ideas developed in Ref. 27, in the t-plane we draw the family
(Li )1≤i≤N of closed half-lines Li = ti + e−iθ R+ for all ti ∈ Atypf . We as-
sume also that θ has been chosen generically so that no Stokes phenomenon
is currently occuring, i.e. all these half-lines are two-by-two disjoint. To ev-
ery ti ∈ Atypf we associate a closed neighbourhood Ti of Li , retractable by
deformation onto Li . It will be assumed that all these Ti are disjoint from
one another, as shown in Fig. 16.
+ +F +
We construct a deformation-retraction of dr : (C, SR ) → (SR i Ti , S R )
: this means a continous map of pairs such that dr|S + F Ti = id while i ◦ dr
+F
R i
+ +
is homotopic to the identity with i : (SR i Ti , SR ) ,→ (C, SR ) the canon-
F
ical injection. Here we add furthermore the condition that dr(C\ i Ti ) =
+ F
SR \ i Ti .
F
Since C\ i Ti is contractible, one easily lift dr by the trivialisation
Φ
f −1 (C\ ti Ti ) → C\ ti Ti × f −1 (t0 )
+
f | & . pr1 (for a chosen t0 ∈ SR ), thus ob-
C\ ti Ti
www.pdfgrip.com
181
t1
Ti
ti Li
tN
+
SR
Fig. 16. The family of half-lines Li and their closed neighbourhoods Ti for θ = 0.
+ S +
taining a deformation-retraction Dr : (Cn , f −1 (SR )) → ( i f −1 (SR ∪
−1 +
Ti ), f (SR )), by setting:
 G
 H = id on restriction to f −1 (Ti )
i

H = Φ−1 ◦ (dr × id) ◦ Φ elsewhere
F Φ F dr×id + F Φ−1 + F
f −1 (C\ i Ti ) → C\ i Ti × f −1 (t0 ) → SR \ i Ti × f −1 (t0 ) → f −1 (SR \ i Ti )
f | & . pr1 pr1 | & . f |
F dr + F
C\ i Ti → SR \ i Ti
This gives the isomorphism (by Proposition A.0.4):

[
+ + +
Hn (Cn , f −1 (SR )) = Hn f −1 (SR ∪ Ti ), f −1 (SR )
i
+ + S +
Next, defining AR = intSR \ i (intSR ∩ Ti ) where intSR is the interior
+ +S +
of SR , we note that (SR i Ti \AR , SR \AR ) is a deformation retract of
+S +
(SR i Ti , SR ), Fig. 17.
+F +
Now by excision (Corollary A.1.1) the pair (SR
F F + i Ti \AR , SR \AR ) is homo-
topic to ( i Ti , i (SR ∩ Ti )). Lifting this information through the fibration
f as previously done, one gets the isomorphism
G G
+ +
Hn (Cn , f −1 (SR )) = Hn f −1 (Ti ), f −1 (Ti ∩ SR )
i i
Applying the relative Mayer-Vietoris exact homology sequence (Theorem

A.2) one deduces that
www.pdfgrip.com
182
t1 t1
Ti Ti
ti Li ti
tN tN
+ S +
Fig. 17. On the left the pair (SR i Ti \AR , SR \AR ), on the right the pair
F F +
( i Ti , i (SR ∩ Ti )).
M
+ +
Hn (Cn , f −1 (SR )) = Hn f −1 (Ti ), f −1 (Ti ∩ SR ) (50)
ti ∈Atypf
What we have obtained can be formulated as follows (compare to §2.2.3):

+
Proposition 4.1.1. Any relative n-cycle Γ ∈ Zn (Cn , f −1 (SR )) can be
represented
+ P
in its homology class [Γ] ∈ Hn (Cn , f −1 (SR ))) by a sum Γ = i Γi with
+
Γi ∈ Zn (f −1 (Ti ), f −1 (Ti ∩ SR )).
4.3. Localisation
For each ti ∈ Atypf , let Di be an open disc centred at ti with a small
radius ηi and Diτ = Di ∩ {t, <((t − ti )eiθ ) ≥ τ }, 0 < τ < ηi . Then by a
deformation-retraction whose definition is left to the reader we get
M
+
Hn (Cn , f −1 (SR )) = Hn f −1 (Di ), f −1 (Diτ ) (51)
ti ∈Atypf
At this stage, we have localised our problem at “the target”. What we

would like to do is to localised the analysis at the source, like what we have
done in §2.2.4. For that we have to make some further assumptions on f :
Hypothesis : We assume that the (non-constant) polynomial function
f : Cn → C (n ≥ 2) has only isolated critical points and no singular point
at infinity. In particular Singf is a finite subset of Cn and we have the
following topological triviality property at infinity :30 for any t0 ∈ C and
for any R = R(t0 ) > 0 large enough, there exists η(R) > 0 such that for
any 0 < η ≤ η(R) the fibres f −1 (t), t ∈ Dη cut ∂BR transversally where
www.pdfgrip.com
183
BR is the open ball of radius R centred at 0, and the restriction maps

(Cn \BR ) ∩ f −1 (Dη ) ∂BR ∩ f −1 (Dη )
f| ↓ and f | ↓ are C ∞ trivial fibrations.
Dη Dη
For each ti ∈ Atypf we note {zij } the subset of Singf above ti (f (zij ) =
ti ). We introduce also
Xij = Bij ∩ f −1 (Di ), τ
Xij = Xij ∩ f −1 (Diτ ) (52)
where Bij ⊂ Cn is an open ball centred at zij with radius ε small enough so
that the Bij are disjoints. Assuming that ηi has been chosen small enough,
one can assume that the Xij are Milnor balls (cf. Theorem 3.5).
For R > R0 > R(ti ) we note
Yi = f −1 (Di ) ∩ BR , Yiτ = f −1 (Diτ ) ∩ BR
and
Zi = f −1 (Di ) ∩ (Cn \BR0 ), Ziτ = f −1 (Diτ ) ∩ (Cn \BR0 ).
The couple {(Yi , Yiτ ), (Zi , Ziτ )} of pairs is an excisive couple of pairs (apply
Theorem A.1) so that the relative Mayer-Vietoris exact homology sequence
(Theorem A.2) can be applied:
i
· · · → Hn (Yi ∩ Zi , Yiτ ∩ Ziτ ) →
?
Hn (Yi , Yiτ ) ⊕ Hn (Zi , Ziτ )
j?
→ Hq (f −1 (Di ), f −1 (Diτ ))
∂
→ Hn−1 (Yi ∩ Zi , Yiτ ∩ Ziτ ) → · · ·
The topological triviality property at infinity implies that Hn (Zi , Ziτ ) =
Hn (Yi ∩ Zi , Yiτ ∩ Ziτ ) = Hn−1 (Yi ∩ Zi , Yiτ ∩ Ziτ ) = 0 so that

Hn f −1 (Di ), f −1 (Diτ ) = Hn (Yi , Yiτ )
and we thus concentrate on Hn (Yi , Yiτ ). We follow the reasoning of Pham,27

see also Broughton6,7 : by the exact homology sequence of a triple (Corollary
A.2.1) one obtains (∀ q):
F i?
· · · → Hn ( j Xij ∪ Yiτ , Yiτ ) −→ Hn (Yi , Yiτ ) −→
F ∂ F (53)
Hn (Yi , j Xij ∪ Yiτ ) −→ Hn−1 ( j Xij ∪ Yiτ , Yiτ ) → · · ·
and by excision (by Corollary A.1.1) then by the relative Mayer-Vietoris
exact homology sequence one has
G G G M
Hq ( Xij ∪ Yiτ , Yiτ ) = Hq ( Xij , τ
Xij )= τ
Hq (Xij , Xij ) (54)
j j j j
www.pdfgrip.com
184
Therefore (53) means that

G
if Hq (Yi , Xi ∪ Yiτ ) = 0 with Xi = Xij (55)
j
then
M
Hq (Yi , Yiτ ) = τ
Hq (Xij , Xij ).
j
In the pair (Yi , Xi ∪ Yiτ ) we excise Xi \∂Xi where ∂Xi is the boundary of
Xi in Yi :
G
Hq (Yi , Xi ∪ Yiτ ) = Hq (Yi \Xi , ∂Xi ∪ (Yiτ \Xτi )) with Xτi = τ
Xij .
j
(56)
By the exact homology sequence of a triple one has
i
· · · → Hq (∂Xi ∪ (Yiτ \Xτi ), ∂Xi ) −→
?
Hq (Yi \Xi , ∂Xi )
→ Hq (Yi \Xi , ∂Xi ∪ (Yiτ \Xτi ))
∂
−→ Hq−1 (∂Xi ∪ (Yiτ \Xτi ), ∂Xi ) → · · · (57)
so that by (56) condition (55) reduces in the isomorphism
Hq (Yi \Xi , ∂Xi ) = Hq (∂Xi ∪ (Yiτ \Xτi ), ∂Xi ). (58)
But by excision and denoting by ∂Xτi the boundary of Xτi in Yiτ , (58) reads
Hq (Yi \Xi , ∂Xi ) = Hq (Yiτ \Xτi , ∂Xτi ) (59)
and this last equality is true by a deformation-retraction argument, because
Yi
f | ↓ is a locally trivial fibration (by the Ehresmann fibration theorem 3.1,
Di
since Yi is a manifold), thus a trivial fibration because Di is contractible.
What we have obtained is in particular the following proposition.
Proposition 4.1.2. Assume that the (non-constant) polynomial function

f : Cn → C (n ≥ 2) has only isolated critical points and no singular point
at infinity (in the sense of the above hypothesis). Then
M M
+
Hn (Cn , f −1 (SR )) = τ
Hn (Xij , Xij ) (60)
ti ∈Atypf zij ∈Singf
where Xij is the Milnor ball associated with the critical point zij ∈ Singf ,
f (zij ) = ti .
Remark : To see what happens in a more general case, see Tibăr.30

www.pdfgrip.com
185
4.4. The asymptotics

To simplify we assume here that arg(k) = θ = 0. From what precedes and
under some convenient
Z hypotheses, we have reduced the analysis of our
integral I(k) = e−kf σ, [Γ] ∈ HnΨ (Cn ) into the analysis of the class of
Γ
functions
Z
I(k) = e−kf σ, [Γ] ∈ Hn (X, X τ ). (61)
Γ
Up to a translation at the source and at the target, we can now assume

that f : (Cn , 0) → (C, 0) represents a germ of holomorphic functions at 0
which is an isolated critical point and
X = B ∩ f −1 (D), X τ = B ∩ f −1 (Dτ )
with
B = B(0, ε) ⊂ Cn , D = D(0, η) ⊂ C ,
τ
D = D ∩ {t, <(t) ≥ τ }, 0 < τ < η
such that X is a Milnor ball (see Theorem 3.5). For later purpose it will be
convenient to note
D+ = D ∩ {t, <(t) > 0}, X + = B ∩ f −1 (D+ ).
Since we have localised the problem, one can also assume that σ ∈ Ωn (X).
We mention that if Γ, Γ0 ∈ Zn (X, X τ ) define the same class in
Hn (X, X τ ), then for some fixed r > 0:
Z Z
∃C = C(r) > 0, e−kf σ − e−kf σ ≤ Ce−τ |k| ,
Γ Γ0
|k| ≥ r, arg(k) = 0
R R
(see the reasoning in §4.1) so that Γ e−kf σ and Γ0 e−kf σ have the same
Poincaré asymptotics when k → +∞.
4.4.1. Reduction to an incomplete Laplace transform

We know from Theorem 3.5 that X is contractible. This imply that
Hn (X) = Hn−1 (X) = 0 (we assume that n ≥ 2) and from the exact
homology sequence of the pair (X, X τ ) (Proposition A.0.3) one has
j? ∂ i
0 = Hn (X) −→ Hn (X, X τ ) −→
?
Hn−1 (X τ ) −→
?
Hn−1 (X) = 0.
www.pdfgrip.com
186
X+
+
Also, since D is contractible, the restriction map f | ↓ is a trivial fibra-
D+
tion. This induces an isomorphism (because D τ ⊂ D+ is also contractible)
Hn−1 (X τ ) = Hn−1 (Dτ × Xt0 ) = Hn−1 (Xt0 )
for any chosen t0 ∈ Dτ . Putting things together, we have obtained an
isomorphism
∂t
Hn (X, X τ ) −→
0
Hn−1 (Xt0 ), t0 ∈ D τ .
Consequence: in the class of a given element [Γ] ∈ Hn (X, X τ ) one can
choose a chain Γt0 ∈ Cn (X) whose boundary ∂Γt0 = γ(t0 ) belongs to
Zn−1 (Xt0 ), with t0 ∈ Dτ .
X+
τ
In what follows we fix a t0 > 0 in D . By the trivial fibration f | ↓
D+
the cycle γ(t0 ) can be horizontally deformed into γ(t) ∈ Zn−1 (Xt ), t ∈ D+ .
Then :
Theorem 4.2. With the above notations and hypotheses,
Z Z t0
−kf
I(k, t0 ) = e σ= e−kt J(t) dt
Γ t0 0
with
Z X
σ
J(t) = |f =t = ψα,l tα−1 Log l (t), ψα,l ∈ C{t}.
γ(t) df
α∈Q?+ , l∈[[0,n−1]]
(62)
Proof. We follow Malgrange.20 The holomorphic differential form e−kf σ ∈

Ωn (X) (for a fixed k) is closed (being of maximal order) and X being a Stein
contractible manifold (Theorem 3.5 and Proposition B.1.1), one deduces
from Corollary B.3.1 the existence of ϕ ∈ Ωn−1 (X) such that
e−kf σ = dϕ.
Thus by the Stokes theorem
Z Z Z
−kf
I(k, t0 ) = e σ= dϕ = ϕ. (63)
Γ t0 Γ t0 γ(t0 )
We note
Z
+
for t ∈ D , I(k, t) = ϕ. (64)
γ(t)
www.pdfgrip.com
187
Note that I(k, t) extends as a multivalued holomorphic function I(k, t) ∈

^
O D ? ,t0 in t for which Theorem 3.6 can be applied:
X
I(k, t) = φα,l tα Log l (t), φα,l ∈ C{t}. (65)
α∈Q+ , l∈[[0,n−1]]
By Lemma 3.2.1 we know that

Z
+ 1 df ∧ ϕ
for t ∈ D , I(k, t) =
2iπ δt γ(t) f −t
where δt : Hn−1 (Xt ) → Hn (X\X t ) is the

Leray coboundary operator.
ϕ dϕ df ∧ ϕ
Therefore using the remark that d = − ,
f −t f − t (f − t)2
Z
+ ∂I 1 df ∧ ϕ
for t ∈ D , (k, t) =
∂t 2iπ δt γ(t) (f − t)2
Z Z
1 dϕ 1 σ
= = e−kf .
2iπ δt γ(t) f −t 2iπ δt γ(t) f −t
Introducing X ? = X\{0} we note that the restriction f |X ? is a submersion.

Also, σ ∈ Ωn (X) being of maximal order, it can be written locally as
σ = df ∧ β. But X ? is a Stein manifold (by Proposition B.1.2) so that
Proposition B.3.1 implies that
σ n−1 σ
∃ ∈ ΩX (X ? ), σ = df ∧ .
df df
Therefore, by Fubini and the Cauchy formula,

Z
∂I σ
for t ∈ D+ , (k, t) = e−kt |f =t .
∂t γ(t) df
From (65) one finally gets that

Z Z !
t0
−kt σ
I(k, t0 ) = e |f =t dt
0 γ(t) df
with
Z X
σ
|f =t = ψα,l tα−1 Log l (t), ψα,l ∈ C{t}.
γ(t) df
α∈Q?+ , l∈[[0,n−1]]
www.pdfgrip.com
188
4.4.2. The asymptotics

From Theorem 4.2 we know that
Z Z t0
I(k, t0 ) = e−kf σ = e−kt J(t)
Γ t0 0
with
X
J(t) = ψα,l tα−1 Log l (t), ψα,l ∈ C{t}.
α∈Q?+ , l∈[[0,n−1]]
Since this is a finite sum, to analyse the asymptotics it is enough to consider

the case when
∞
X
J(t) = bn tα−1+n Log l (t)
n=0
for some α ∈ Q?+ and l ∈ [[0, n − 1]] and we assume that the series ex-
pansion converges for |t| < τ , τ > t0 . We also assume that Log (t) is the
principal determination of the logarithm function (real for t > 0).
We adapt what we have done in §2.2.8. For k in the sectorial neighbourd-
hood of infinity Σr,δ as defined in (7) one has, for any N ∈ N:
Z t0 ∞
!
X
−kt α−1+n l
I(k, t0 ) = e bn t Log (t) dt
0 n=0
Z +∞ N
X
I(k, t0 ) = e−kt bn tα−1+n Log l (t) dt
0 n=0
Z +∞ N
X
− e−kt bn tα−1+n Log l (t) dt
t0 n=0
Z t0 ∞
X
−kt
+ e bn tα−1+n Log l (t) dt
0 n=N +1
By (5) this means that

N
X l
Γ(α + n) d
I(k, t0 ) − = bn
n=0
k α+n dα
∞
X Z t0 N
X Z +∞
−kt α−1+n
bn e t l
Log (t) dt − bn e−kt tα−1+n Log l (t) dt.
n=N +1 0 n=0 t0
www.pdfgrip.com
189
Therefore, making the change of variable t = t0 s,

XN l
d Γ(α + n)
I(k, t0 ) − bn
n=0
dα k α+n
Xl X∞ Z 1
l
= Log p (t0 ) bn t0 α+n e−kt0 s sα−1+n Log l−p
(s) ds
p 0
p=0 n=N +1
N Z !
X +∞
α+n −kt0 s α−1+n l−p
− b n t0 e s Log (s) ds.
n=0 1
Z 1
1
We now introduce 0 < ≤ In the integral we have sα−1−+n ≤
2.
Z +∞ 0
sα−+N while in the integral we have sα−1++n ≤ sα−+N . Then,

1
there exists A > 0 such that for k ∈ Σr,δ ,
XN l
d Γ(α + n)
I(k, t0 ) − bn
n=0
dα k α+n
∞
!
X Γ(N + α − + 1)
≤ A (1 + |Log (t0 )|) l
|bn |tn+α
0 N +α−+1 . (66)
n=0 t0 sin(δ)|k|
In other words, for a given Σr,δ ,
1
∀ ∈]0, ], ∃C > 0, ∀N ∈ N, ∀ k ∈ Σr,δ ,
XN 2l
d Γ(α + n) Γ(N + α − + 1)
I(k, t0 ) − bn ≤ C N +α−+1 .
n=0
dα k α+n |k|N +α−+1
(67)
This provides the asymptotics we were looking for.
4.5. An example
We illustrate the previous considerations with the following simple example.
We take
f (z1 , z2 ; β) = β 2 z1 + z2 + z1 z22 , β ∈ C?
which is a deformation of the Broughton’s polynomial (see §3.2.3). This
polynomial has two atypical values which are the critical values t = ±iβ.
Each generic fibre f −1 (t), t ∈ C\{±iβ} has only one connected component
for t 6= ±iβ, f −1 (t) = {z1 = (t − z2 )/(β 2 + z22 ), z2 ∈ C\{±iβ}} (68)
∼ Z, H1 (f −1 (t)) ∼
so that H0 (f −1 (t)) = = Z2 and Hq (f −1 (t)) ∼ = 0 otherwise.
www.pdfgrip.com
190
Note that the singular fibre f −1 (iβ) = {z1 ∈ C, z2 = iβ} ∪ {z1 =

−1/(iβ +z2 ), z2 ∈ C\{−iβ}} resp. f −1 (−iβ) = {z1 ∈ C, z2 = −iβ}∪{z1 =
−1/(−iβ + z2 ), z2 ∈ C\{iβ}} can be written as the one point union
(wedge) of two connected spaces that intersect at the nondegenerate crit-
i i
ical point (z1 , z2 ) = ( , iβ) resp. (z1 , z2 ) = (− , −iβ). In particular
2β 2β
H0 (f −1 (±iβ)) ∼= Z, H1 (f −1 (±iβ)) ∼
= Z and Hq (f −1 (±iβ)) ∼
= 0 otherwise.
We would like to analyse the asymptotics when k → +∞ of the following
integral
Z
IΓ (k; β) = e−kf (z1 ,z2 ;β) g(z1 , z2 ) dz1 ∧ dz2
Γ
Z
z1
= e−βkf (z1 ,z2 ;1) g( , βz2 ) dz1 ∧ dz2
e
Γ β
Ψ(β) ∼ Z2 for t ∈
where [Γ] ∈ H2 (C2 ) and g ∈ C[z1 , z2 ]. Since H1 (f −1 (t)) =
C\{±iβ}, we deduce from Theorem 4.1 and (49) that
Ψ(β)
H2 (C2 ) ∼
= Z2 . (69)
Ψ(β)
From (68) it is easy to see that H2 (C2 ) is generated by the following
two cycles Γ± (the so-called Lefschetz thimbles),
Γ = m + Γ+ + m − Γ− , m± ∈ Z
Γ+ = {z1 = (iβ + r − z2 )/(β 2 + z22 ), r ∈ [0, +∞[, |z2 − iβ| = β} (70)
Γ− = {z1 = (−iβ + r − z2 )/(β 2 + z22 ), r ∈ [0, +∞[, |z2 + iβ| = β}

Ψ(β)
Note that the data H2 (C2 ) makes a local system on C? (in fact
β∈C?
a constant sheaf of Z-modules).
One can recover (69) using Proposition 4.1.2:
H2Ψ (C2 ) ∼ τ
= H2 (X+ , X+ τ
) ⊕ H2 (X− , X− )
where X+ (resp. X− ) is the Milnor ball associated with the nondegenerate

i i
critical point (z1 , z2 ) = ( , iβ) (resp. (z1 , z2 ) = (− , −iβ)). From our
2β 2β
previous theoretical analysis, the asymptotics of IΓ (k; β) just reduces in a
local analysis near each of the two critical points. Simple calculations gives
www.pdfgrip.com
191
for instance:
for g(z1 , z2 ) = a + bz1 + cz2 ,
i b b
!
−iβk
a+ 2β + icβ β −3
IΓ (k; β) = m+ e π + O(k ) +
2(βk)2 βk
i b b
!
−a + 2 β + icβ β
+m− e+iβk π − + O(k −3 )
βk 2(βk)2
(71)
where m± are integers which depend only [Γ]. It can be shown that the
remainder terms O(k −3 ) are in fact zero functions. We do this by direct
calculation, using (70) which provides
Z +∞ I
−iβk −kr iβ + r − z2 dz2
IΓ+ (k; β) = e e a+b + cz2 dr
0 β 2 + z22 β + z22
2
that is
Z +∞
−iβk −kr π π π
IΓ+ (k; β) = e e a + b(i 2 + 3 r) + ciπ dr.
0 β 2β 2β
so that
i b b
!
−iβk
a+ 2β + icβ β
IΓ+ (k; β) = e π + .
βk 2(βk)2
Similarly,
i b b
!
−a + 2β + icβ β
IΓ− (k; β) = e+iβk π − .
βk 2(βk)2
4.6. To go further
By their very nature, the complex singular integrals considered in this paper
are crossroads for different scientific communities, from pure mathemati-
cians to physicists and chemists. In such various situations one often needs
to enlarge the methods presented in this paper, for instance for Laplace-
type integrals with boundaries. Also, it allows a theoretical and numeri-
cal control of the so-called Stokes phenomenon with a hyperasymptic and
resurgent viewpoint. To go futher in that direction, the reader may consult
Refs. 10 and 11 and references therein.
www.pdfgrip.com
192
Acknowledgments
This paper is based on a course given at the ICTP “School on Algebraic
approach to differential equations”, Alexandria - Egypt, November 12-24,
2007. The author would like to warmly thank the organizers of this so
pleasant School in a so charming place, especially Professors Lê Dung Trang
and Mohamed Darwish, as well as the students and colleagues for their
many interesting questions and helpful discussions.
Appendix A. Short Introduction to Singular Homology

For this appendix we mainly refer to Refs. 12, 16 and 28. In what follows
X is a topological space.
Appendix A.1. Simplex and chain
• The standard (oriented) q-simplex in the affine space Rq+1 is by

definition
Xq q
X
q
∆ ={ λj e j , λj = 1, 0 ≤ λj ≤ 1} ⊂ Rq+1
j=0 j=0
that is the convex hull of (e0 , · · · , eq ), where (e0 , · · · , eq ) is the standard

basis of Rq+1 . (The orientation is chosen so that if a is a generic point
of ∆q , then a basis (v1 , · · · , vq ) ∈ Ta ∆q is positively oriented if the basis
Xq
1
(v0 = ej , v1 , · · · , vq ) determines the standard orientation of Rq+1 ).
j=0
q + 1
• A singular q-simplex of X is a continuous map
σ : ∆q 7→ X
• A singular q-chain of X is a formal linear combination
m
X
c= n i σi , ni ∈ Z (resp. ni ∈ K = R or C)
i=1
where the σi are singular q-simplices.

• We note by Cq (X) = the set of all singular q-chains. We remark
that Cq (X) is a Z-module
X (resp. a K-vector space).
Example: C0 (X) = { ni [xi ]} where the xi are points of X.
f inite
• The support of a singular q-simplex σ is |σ| = σ(∆q ). The support
m
X S
of a singular q-chain c = ni σi is |σ| = |σi |.
i=1
www.pdfgrip.com
193
Note that the support of a singular q-chain is always a compact subset of

X. Also, since any standard q-simplex is path connected, the support of
any q-simplex is contained in a path connected component of X.
Appendix A.2. Boundary operator
• For q > 0 we define for i = 0, · · · , q,

q−1
X i−1
X q−1
X
Fiq : ∆q−1 → ∆q , Fiq : λj ej 7→ λj e j + λj ej+1
j=0 0 i
which maps ∆q−1 onto the face [e0 , · · · , ebi , · · · , eq ] of ∆q .

Example: for q = 2,
F02 : λ0 e0 + λ1 e1 7→ λ0 e1 + λ1 e2
F12 : λ0 e0 + λ1 e1 7→ λ0 e0 + λ1 e2
F22 : λ0 e0 + λ1 e1 7→ λ0 e1 + λ1 e1
• If σ : ∆q 7→ X is a singular q-simplexe, the i − th-face σ (i) of σ is the
singular (q − 1)-simplex
?
σ (i) = Fiq ◦ σ = σ ◦ Fiq : ∆q−1 → X.
This allows to define the boundary of σ :
q
X
∂q σ = (−1)j σ (j) .
j=0
m
X
• If c = ni σi is a singular q-chain, then one defines the boundary of c
i=1
by
m
X
∂q c = n i ∂ q σi .
i=1
In this way one have a homomorphism (of Z-modules, resp. of K-vector

spaces)
∂q : Cq (X) → Cq−1 (X)
which is the boundary operator.
• We set:

 Cq (X) = 0 for q < 0

∂q = 0 for q ≤ 0
Then it can be shown that:
www.pdfgrip.com
194
Proposition A.0.1.
∀ q, ∂q ◦ ∂q+1 = 0.
Definition A.0.1. We note for every q,
Zq (X) = {q-cycles} = ∂q−1 (0) = ker ∂q ,

Bq (X) = {q-boundaries} = ∂q+1 Cq+1 (X) = im ∂q+1 .
Two q-chains c1 and c2 are said to be homologous c1 ∼ c2 if c1 − c2 is a

q-boundary.
Proposition A.0.1 say that the sequence

∂q+1 ∂q
· · · → Cq+1 (X) −→ Cq (X) −→ Cq−1 (X) → · · ·
is a chain complex (C• (X), ∂• ) of Z-modules (resp. of K-vector spaces).

This allows to define:
Definition A.0.2. For every q,

Zq (X) Zq (X)
Hq (X) = (resp. Hq (X; K) = )
Bq (X) Bq (X)
is the q-th (singular) homology group of X (Z-module, resp. K-vector
space).
The homology class of a q-cycle c is usually denoted by [c]

M ∈ Hq (X).
The graded (by the dimensions q) group H• (X) = Hq (X) is the
q
total (singular) homology group of X.
Z0 (X) ∼ m
Example H0 (X) = = Z where m is the number of path-
B0 (X)
connected
X components of X. Indeed, one has Z0 (X) =XC0 (X) =
{ ni [xi ]} where the xi are points of X and B0 (X) = { ni ([x2i ] −
f inite f inite
[x1i ])} where for each i, x1i and x2i are connected by a continuous path in
X.
More generally, if X is a Hausdorff topological space and if {Xi } are its
path-connected components, then
M
Hq (X) = Hq (Xi ), ∀ q.
i
www.pdfgrip.com
195
Appendix A.3. Homomorphism induced by a continuous map
We consider two topological spaces X and Y and a continuous map

f : X → Y.
To a given singular q-simplex σ : ∆q 7→ X of X there is a natural way of
associating a singular q-simplex of Y ,
f ◦ σ : ∆q 7→ Y.
We thus have a homomorphism (of Z-modules, resp. K-vector spaces)
X X
Cq (f ) : Cq (X) → Cq (Y ), Cq (f ) n i σi = ni f ◦ σ i
f inite f inite
and it can be shown that

∂q ◦ Cq (f ) = Cq−1 (f ) ◦ ∂q .
We thus have a chain map C(f ) which allows to define the homomorphism
of homology groups:
f? : Hq (X) → Hq (Y ).
Of course, if Z is another topological space and if g : Y → Z is a continuous
map, then (g ◦ f )? = g? ◦ f? . Also:
Proposition A.0.2. If f, g : X → Y are homotopic continuous maps, that

is there exists a continuous map
H : X × [0, 1] → Y, H(x, 0) = f (x), H(x, 1) = g(x),
then f? = g? .
Example Let A ⊂ X and i : A ,→ X the canonical injection. Assume that

one has a continuous map f : X → A which is a deformation-retraction,
that is:
(1) f is the identity on A : f ◦ i = idA (f is a retraction).
(2) i ◦ f is homotopic to the identity : there exists a continuous map H :
X × [0, 1] → X such that H(., 0) = idX and H(., 1) = i ◦ f .
(For instance f : x ∈ Rn+1 \{0} 7→ x/kxk ∈ S n , H : (x, t) ∈ Rn+1 \{0} ×
[0, 1] 7→ (1 − t)x + tx/kxk ∈ Rn+1 \{0}).
From condition 1. one obtains that f? ◦ i? : Hq (A) → Hq (A) is the identity
map. Using Proposition A.0.2, condition 2. implies that i? ◦ f? : Hq (X) →
Hq (X) is also the identity map.
www.pdfgrip.com
196
Consequence : if A ⊂ X is a deformation-retract of X, then the ho-

mologies H• (A) and H• (X) are isomorphic. (For instance H• (S n ) =
H• (Rn+1 \{0})).
In particular if X is contractible, that is can be retract by deformation to
a point, then the homology of X is isomorphic to the homology of a point
: ∀ q 6= 0, Hq (X) = 0, H0 (X) = Z.
Appendix A.4. Homology of a pair
Let A ⊂ X be a subspace of the topological space X and we note i : A ,→

X the canonical injection. This mapping induces a natural homomorphism
i? : Cq (A) → Cq (X) so that the space Cq (A) can be seen as a sub Z-module
(resp. a sub K-vector space) of Cq (X).
Cq (X)
• The quotient space Cq (X, A) = is the space of relative q-
Cq (A)
chains of X with respect to A.
• A q-chain c ∈ Cq (X) is said to be a relative q-cycle of X with
respect to A if ∂q c ∈ Cq−1 (A).
We note
Zq (X, A) = {relative q-cycles}.
• A q-chain c ∈ Cq (X) is said to be a relative q-boundary of X with
respect to A if there exists a q-chain c0 ∈ Cq (A) such that c and c0 are
homologous in X : c − c0 ∈ Bq (X).
One notes
Bq (X, A) = {relative q-boundaries}.
(See §2.2.2, Fig. 5).
We remark that Bq (X, A) is a sub Z-module, (resp. sub K-vector space)
of Zq (X, A). (Indeed, if c ∈ Bq (X, A), then there exists c0 ∈ Cq (A), there
exists d ∈ Cq+1 (X) such that c = c0 + ∂q+1 d. Therefore, ∂q c = ∂q c0 ∈
Cq−1 (A)).
Definition A.0.3. For every q, the quotient group

Zq (X, A) Zq (X, A)
Hq (X, A) = (resp. Hq (X, A; K) = )
Bq (X, A) Bq (X, A)
is the q-th relative (singular) homology group (Z-module, resp. K-vector
space) of X with respect to A, or also the q-th (singular) homology group
of the pair (X, A).
www.pdfgrip.com
197
The canonical injection i : A ,→ X induces for any q the natural homo-

morphism
i? : Hq (A) → Hq (X).
Since Zq (X) ⊂ Zq (X, A) and Bq (X) ⊂ Bq (X, A) (note that we have equal-
ity when A = ∅) one has also in a natural way the homomorphism
j? : [c] ∈ Hq (X) 7→ [c] ∈ Hq (X, A).
The following boundary homomorphism
∂? : [c] ∈ Hq (X, A) 7→ [∂q c] ∈ Hq−1 (A)
is also well-defined : take c ∈ Zq (X, A), that is c ∈ Cq (X) and ∂q c ∈
Cq−1 (A). The homology class [c] ∈ Hq (X, A) reads [c] = {c + c0 + d, c0 ∈
Cq (A), d ∈ Bq (X)}. Therefore
∂q (c + c0 + d) = ∂q c + ∂ q c0
∈ Zq−1 (A) ∈ Bq−1 (A)
With these definitions, one has:
Proposition A.0.3. The following homology sequence of the pair (X, A)

is exact:
i ? j? ? ∂
· · · → Hq (A) −→ Hq (X) −→ Hq (X, A) −→ Hq−1 (A) → · · ·
What have been said in §A.3 can be extended for pairs. Let X and Y
be two topological spaces and A ⊂ X, B ⊂ Y . From a continuous map of
pairs,
f : x ∈ (X, A) 7→ y ∈ (Y, B), f (A) ⊂ B,
one can associated a homomorphism of homology groups
f? : Hq (X, A) → Hq (Y, B).
Example In §2.2.2 we have introduced the pairs (C, ΣR (Aθ,δ )). For R0 ≥
R, (C, ΣR0 (Aθ,δ )) is a sub-pair of (C, ΣR (Aθ,δ )) and one has an homeomor-
phism
R0
z ∈ (C, ΣR0 (Aθ,δ ))
h : z ∈ (C, ΣR (Aθ,δ )) →
R
from which is associated an isomorphism
h? : Hq (C, ΣR (Aθ,δ ) → Hq (C, ΣR0 (Aθ,δ )).
One has the analogue of Proposition A.0.2:
www.pdfgrip.com
198
Proposition A.0.4. If f, g : (X, A) → (Y, B) are homotopic continuous

maps of pairs, that is if there exists a continuous map
H : (X, A) × [0, 1] → (Y, B), H(x, 0) = f (x), H(x, 1) = g(x),
then f? = g? .
Example Let (Y, B) be a subpair of (X, A) and i : (Y, B) ,→ (X, A) the

canonical injection. Assume that the continuous map f : (X, A) → (Y, B)
is a deformation-retraction, that is:
(1) f is the identity on Y : f ◦ i = idY (f is a retraction).

(2) i ◦ f is homotopic to the identity.
Then the homologies H• (X, A) and H• (Y, B) are isomorphic.
Definition A.0.4. An inclusion map (Y, B) ,→ (X, A) between topological

pairs is called an excision map if Y \B = X\A.
The couple {X1 , X2 } made by two subsets of a given topological space is
an excisive couple of subsets if the inclusion chain map C• (X1 ) + C• (X2 ) ⊂
C• (X1 ∪ X2 ) induces an isomorphism of homology.
The couple {(X1 , A1 ), (X2 , A2 )} of pairs in a given topological space is an
excisive couple of pairs if both {X1 , X2 } and {A1 , A2 } are excisive couple
of subsets.
Theorem A.1. • If X1 ∪ X2 = intX1 ∪X2 X1 ∪ intX1 ∪X2 X2 , then {X1 , X2 }

is an excisive couple.
• {X1 , X2 } is an excisive couple if and only if the excision map (X1 , X1 ∩
X2 ) ,→ (X1 ∪ X2 , X2 ) induces an isomorphism of homology.
Corollary A.1.1. We assume that U ⊂ A ⊂ X are subsets such that the

closure U of U is included in the interior intA of A. Then U can be excised,
that is
Hq (X, A) = Hq (X\U, A\U ), ∀ q.
Proof. The couple {X\U, A} is an excisive couple so that the excision map
(X\U, A\U ) ,→ (X, A) induces an isomorphism of homology.
Theorem A.2 (Mayer-Vietoris). • Assume that {X1 , X2 } is an excisive

couple. Then the following Mayer-Vietoris homology sequence is exact:
www.pdfgrip.com
199
(iX1 ? ,−iX2 ? ) jX1 ? +jX2 ?

· · · → Hq (X1 ∩ X2 ) −→ Hq (X1 ) ⊕ Hq (X2 ) −→ Hq (X1 ∪ X2 )
∂
−→ Hq−1 (X1 ∩ X2 ) → · · ·
i X1 j X1 i X2 j X2
where X1 ∩ X2 ,→ X1 ,→ X1 ∪ X2 and X1 ∩ X2 ,→ X2 ,→ X1 ∪ X2 are the
canonical injections.
• Assume that {(X1 , A1 ), (X2 , A2 )} is an excisive couple of pairs. Then the
following relative Mayer-Vietoris homology sequence is exact:
? i
· · · → Hq (X1 ∩ X2 , A1 ∩ A2 ) → Hq (X1 , A1 ) ⊕ Hq (X2 , A2 )
j?
→ Hq (X1 ∪ X2 , A1 ∪ A2 )
∂
→ Hq−1 (X1 ∩ X2 , A1 ∩ A2 ) → · · ·
Note that for a given triple (X, A, B), B ⊂ A ⊂ X, the couple of pairs
{(X, B), (A, A)} and {(X, B), (A∪, A ∪ B)} are always an excisive couple of
pairs. One thus deduces from the relative Mayer-Vietoris homology exact
sequence that:
Corollary A.2.1. • Assume that B ⊂ A ⊂ X. Then the following homol-

ogy sequence of the triple (X, A, B) is exact:
?i j? ∂
· · · → Hq (A, B) −→ Hq (X, B) −→ Hq (X, A) −→ Hq−1 (A, B) → · · ·
• Assume that A, B are two subsets of X such that {A, B} is an excisive
couple. Then the following homology sequence of the triad (X, A, B) is
exact:
? i j? ∂
· · · → Hq (A, A∩B) −→ Hq (X, B) → Hq (X, A∪B) −→ Hq−1 (A, A∩B) → · · ·
Appendix A.5. Homology with support in a family
It is sometimes needed to extend the definition of homology to homology

defined by a family of supports (see25 ).
Let X be a Hausdorff topological space assumed to be locally compact
(i.e., every point has a local base of compact neighbourhoods) and para-
compact (i.e., every open cover admits an open locally finite refinement).
We introduce a family Φ of closed subsets of X which satisfies the fol-
lowing properties:

 (i) A, B ∈ Φ ⇒ A ∪ B ∈ Φ
(ii) B closed ⊂ A ∈ Φ ⇒ B ∈ Φ (A.1)

(iii) Any A ∈ Φ has a neighbourhood which belongs to Φ
www.pdfgrip.com
200
To define what is a a q-chain c with support in Φ one modifies the

definition of a q-chain as follows :
X
• c is a formal linear combination c = ni σi where the sum may be
i
infinite but is locally finite: any point x ∈ X has a neighbourhood
Ux ⊂ X which meets only a finite number of supports [[σi ] (the σi are
usual singular q-simplices). This implies that [c] = [σi ] is a closed
i
subset of X.
• moreover [c] ∈ Φ.
The set of q-chains with support in Ψ makes the group CqΦ (X) of q-
chains with support in Φ. The boundary operator ∂q can be defined in
an obvious way on CqΦ (X) (∂q defines a map from CqΦ (X) into Cq−1 Φ
(X)
Φ
by (ii) of (A.1)). To the chain-complex (C• (X), ∂• ) of Z-modules one can
then associate the homology groups
ZqΦ (X)
HqΦ (X) =
BqΦ (X)
and HqΦ (X) is called the q-th homology group with support in Φ.
Examples • The family Φ = c of all compact subsets of X satisfies

(A.1) and the associated homology coincides with the singular homology :
Hqc (X) = Hq (X).
• The family Φ = F of all closed subsets of X satisfies (A.1) : H•F (X)
is the homology with closed support (or Borel-Moore homology).
Appendix A.6. Homology and fibre bundle
We just mention here the following fundamental tool in algebraic topol-

ogy:
Theorem A.3 (Homotopy lifting Theorem). Assume that M and N

are topological spaces such that N is a paracompact Hausdorff space. As-
M
sume that f ↓ is the projection of a fibre bundle. Then f has the homotopy
N
lifting property : for any given topological space X and for
– any homotopy H : X × [0, 1] → N ,
– any continuous map h f0 : X → M lifting h0 = H|X×{0} ,
www.pdfgrip.com
201
M
f0 % ↓ f , there exists a homotopy H
h e : X × [0, 1] → M lifting
h0
X × {0} −→ N
M
H, e % ↓ f , with h
H f0 = H|
e X×{0} .
H
X × I −→ N
Appendix A.7. Integration
We assume here that X is a C ∞ -manifold.

m
X
• If ω is a differential q-form on X and if c = ni σi is a q-chain, then
i=1
by definition
Z m
X Z m
X Z
ω= ni ω= ni σi? ω
c i=1 σi i=1 ∆q
(σi? ω = σi? ◦ ω) with the following remarks:
• we remind the reader that the ∆q are oriented;

• to define the σi? ω, we assume that each σi is differentiable instead of
continuous, that is each σi is the restriction to ∆q of a C ∞ function
Xm
σei : U 7→ X, ∆q ⊂ U ⊂ Rq+1 . In that case the q-chain c = ni σi is
i=1
said to be piecewise differentiable.
Note that any continuous function σ : ∆q 7→ X can be approximated

by a differentiable map. (Any continous function with compact support can
be uniformaly approximated by Cc∞ functions, by regularisation).
This has the following consequence :
Proposition A.3.1. Each homology class [c] ∈ Hq (X) of a q-cycle can

be represented by a piecewise differentiable q-cycle c, and each piecewise
differentiable null homologous q-cycle is the boundary of a piecewise differ-
entiable (q + 1)-chain.
In this course, when an integral of a differential form along a chain is

considered, one always assume that this chain is piecewise differentiable.
www.pdfgrip.com
202
Theorem A.4 (Stokes theorem). If ϕ is a differential (q − 1)-form on

X and if c is a q-chain, then
Z Z
dϕ = ϕ.
c ∂c
Corollary A.4.1.
Z If ϕ is a closed differential q-form on X (dϕ = 0), then
the integral ϕ along a q-cycle c only depends on the homology class
c
[c] ∈ Hq (X).
Appendix B. Stein Manifolds and Some Consequences

We refer to Refs. 15 and 14 for this appendix.
Appendix B.1. Definition and main properties
In what follows it will be assumed that one works with paracompact

complex manifolds. We start with the following definition of Stein mani-
folds:
Definition B.0.1. A complex manifold X is called :

• holomorphically spreadable if for any point x0 ∈ X there are holomorphic
functions f1 , · · · , fN on X such that x0 is isolated in the set
N (f1 , · · · , fN ) = {x ∈ X, f1 (x) = · · · = fN (x) = 0}.
• holomorphically convex if for any compact set K ⊂ X the holomorphically
b = {x ∈ X, |f (x)| ≤ supX |f | for every f ∈ O(X)} is also
convex hull K
compact.
• a Stein manifold if X is connected and is holomorphically spreadable and
holomorphically convex.
Typical Stein manifolds are domains of holomorphy:
Theorem B.1. We assume that D ⊂ Cn is a domain, that is D is a

connected open set of Cn . Then:
D is a Stein manifold
m
D is holomorphically convex
m
D is a domain of holomorphy, that is there exists a function h ∈ O(D)
that is completely singular at every point of ∂D.
www.pdfgrip.com
203
We provide some example of domains of holomorphy.
Proposition B.1.1. • Every domain in C is a domain of holomorphy.

• Every affine convex open subset of Cn is a domain of holomorphy (for
instance Cn and any open ball of Cn are domains of holomorphy).
• Every cartesian product of domains of holomorphy is a domain of holo-
morphy.
• If G1 ⊂ Cn and G2 ⊂ Cp are domains of holomorphy and if f : G1 → Cp
is a holomorphic map, then f −1 (G2 ) ∩ G1 is a domain of holomorphy.
Example : A Milnor ball is a domain of holomorphy (and thus a Stein

manifold).
Next we give some properties for Stein manifolds.
Proposition B.1.2. • Every closed submanifold of a Stein manifold is

Stein.
• If X is a Stein manifold and if f ∈ OX (X), then X\N (f ) is Stein.
• If X is a complex manifold and if U1 , U2 ⊂ X are two open Stein manifold,
then U1 ∩ U2 is Stein.
• Every cartesian product of Stein manifolds is a Stein manifold.
• If f : X → Y is a holomorphic submersion between complex manifolds.
If X is Stein and if Z ⊂ Y is a Stein submanifold, then f −1 (Z) is a Stein
manifold.
To go further, one needs the so-called theorem B of Cartan-Serre. We

first give some definitions.
We consider a complex manifold X and a sheaf F of OX -modules. Let
s1 , · · · , sp ∈ F(U ) and consider the homomorphism
p
X
σU : (g1x , · · · , gpx ) ∈ (OX |U )p 7→ σU (g1x , · · · , gpx ) = gix six ∈ F|U
i=1
(B.1)
(we note F|U the restriction sheaf to the open set U ⊂ X). One says that
F|U is generated by the sections s1 , · · · , sp if σU is surjective. (Every stalk
Fx , x ∈ U is generated as a Ox -module by the germs s1x , · · · , spx ).
The sheaf F is said to be finite at x ∈ X if there exists a neighbourhood U
of x and a finite number of sections s1 , · · · , sp ∈ F(U ) which generate F|U .
The sheaf F is said to be finite on X if it is finite at every point x ∈ X.
If σU is an OX |U -homomorphism as defined by (B.1), then the following
www.pdfgrip.com
204
sheaf of OX |U -modules in (OX |U )p

[ p
X
p
Rel(s1 , · · · , sp ) = ker σU = {(g1x , · · · , gpx ) ∈ (Ox ) , gix six = 0}
x∈U i=1
is called the sheaf of relations of s1 , · · · , sp .

The sheaf F is a finite relation sheaf at x ∈ X if for every open neigh-
bourhood U of x and for arbitrary sections s1 , · · · , sp ∈ F(U ) the sheaf of
relations Rel(s1 , · · · , sp ) is finite at x.
The sheaf F is said to be a finite relation sheaf if it is a finite relation
sheaf at every x ∈ X. The OX -sheaf F over X is said to be coherent
if it is finite and a finite relation sheaf.
Example : if X is a complex manifold of dimension n, then the OX -sheaves
n
ΩpX are coherent (since ΩpX is locally free of rank ).
p
Theorem B.2 (Theorem B). We consider a complex manifold X. Then

X is Stein if and only if for every coherent OX -sheaf F over X one has
H q (X, F) = 0, ∀q > 0.
Here we do not define cohomology with values in a sheaf. It will be

enough for our purpose to mention that
H 0 (X, F) = F(X)
where F(X) is the set of all global sections of the sheaf F.
Appendix B.2. Some applications
Appendix B.2.1. First applicaton
Let X be a complex manifold of dimension n. We note ΩpX the OX -sheaf

of germs of holomorphic p-forms on X. Denoting by C the constant sheaf,
one has the following sequence of sheaf:
i d d d d d
0 → C → OX = Ω0X → Ω1X → Ω2X → · · · → ΩnX → 0 (B.2)
By the holomorphic Poincaré lemma we know that if U ⊂ X is a star-

shaped open set then any holomorphic closed p-form is exact; consequently
(B.2) is an exact sequence (in which case (B.2) is a so-called resolution of
the constant sheaf C). Furthermore if we assume that X is Stein, then the
OX -sheaves ΩpX are coherent sheaves and by Theorem B.2 we get
www.pdfgrip.com
205
∀ q > 0, ∀p ≥ 0, H q (X, ΩpX ) = 0.
In such case the resolution (B.2) is said to be acyclic and one has the
following de Rham theorem:
Theorem B.3 (de Rham theorem). We assume that X is a complex

manifold of dimension n. If (B.2) is an acyclic resolution of the constant
sheaf C then there exists natural C-isomorphisms

H 0 (X, C) ∼
= ker d : O(X) → Ω1 (X)

ker d : Ωp (X) → Ωp+1 (X)
H p (X, C) ∼
= , p ≥ 1.
Im d : Ωp−1 (X) → Ωp (X)
This is true in particular when X is a Stein manifold.
In this theorem H • (X, C) stands for the singular cohomology de-

duced from the singular homology by duality (one defines the
set of q-cochains as C q (X) = Hom(Cq (X), C) and the cobound-
ary operator δ as the dual map of the boundary operator ∂,
< σ, δC >=< ∂σ, C >). By the so-called universal coefficient theorem for
cohomology (cf.28 ) one has the exact sequence
0 → Ext(Hq−1 (X), C) → H q (X, C) → Hom(Hq (X), C) → 0 (B.3)
where Ext(Hq−1 (X), C) = 0 if Hq−1 (X) is a free Z-module. In particular

if X is contractible then ∀ q 6= 0, Hq (X) = 0, H0 (X) = Z so that ∀ q 6=
0, H q (X, C) = 0, H 0 (X, C) = C. Therefore by Theorem B.3:
Corollary B.3.1. If X is a contractible Stein manifold then every closed

holomorphic form is exact.
Appendix B.2.2. Second application
Assume that X is a complex manifold of dimension n and f : X → T ⊂

C a (non constant) holomorphic map, where T is an open connected set.
www.pdfgrip.com
206
We define the following sheaf of holomorphic relative formse :

ΩqX
Ω0X/T = OX , ΩqX/T = , q ≥ 1. (B.4)
df ∧ Ωq−1
X
Note that we have a (de Rham relative) chain complex

dX/T dX/T dX/T dX/T
Ω0X/T −→ Ω1X/T −→ · · · −→ ΩX/T
n−1
−→ ΩnX/T
where the differential dX/T is defined in a natural way (dX/T [ω] = dX/T [ω +
df ∧ Ψ] = [d(ω + df ∧ Ψ)] = [dω − df ∧ dΨ] = [dω]).
From the very definition (B.4) one has the following exact sequence of
sheaves,
i
0 → df ∧ Ωq−1
X → ΩqX → ΩqX/T → 0. (B.5)
We know that the ΩqX are coherent sheaves of OX -modules. Since df ∧ Ωq−1X
is the image of Ωq−1
X by the OX -linear map df ∧ . : Ωq−1
X → ΩqX one deduces
that df ∧ Ωq−1
X are also coherent sheaves. In the exact sequence (B.5) since
two sheaves are coherent, then the third sheaf ΩqX/T is also coherent as a
consequence of the so-called “Three lemma”.14
We now add the assumption that X 0 ⊂ X is a Stein (sub)manifold.
From the short exact sequence of sheaves (B.5) we derive the following long
sequence of cohomology:
· · · → H 0 (X 0 , df ∧ Ωq−1 0 0 q
X ) → H (X , ΩX )
→ H 0 (X 0 , ΩqX/T ) → H 1 (X 0 , df ∧ Ωq−1
X ) → ··· (B.6)
and by Theorem B.2 we know that H 1 (X 0 , df ∧ Ωq−1 X ) = 0. This implies

that the homomorphim ΩqX (X 0 ) → ΩqX/T (X 0 ) is a surjective map (∀ q).
We have also the following result which is used in this course:
Proposition B.3.1. We assume that X is a complex manifold of dimen-

sion n and that f : X → T ⊂ C a (non constant) holomorphic map where
e If F is a sheaf over X and G a subsheaf the collection of quotient spaces

F (U ) F
makes as a rule only a presheaf and is the sheaf generated by this
G(U ) U open ⊂X G
F (U )
presheaf. The collection of canonical homomorphisms F (U ) → allows to define a
G(U )
F F
sheaf homomorphism φ : F → which is surjective, that is for all x, φx : Fx →
G Gx
F
is surjective. However the homomorphisms φ(U ) : F (U ) → (U ) are not surjective in
G
general.
www.pdfgrip.com
207
T is an open connected set. We assume furthermore that X 0 ⊂ X is a Stein

(sub)manifold and that the restriction f |X 0 is a submersion.
dω
Then if ω ∈ ΩX n−1
(X 0 ) is dX/T -closed, then there exists n−1
∈ ΩX (X 0 )
df
dω
such that dω = df ∧ .
df
Proof. The following sequence of sheaves

df ∧. df ∧. df ∧. df ∧.
0 → Ω0X −→ Ω1X −→ · · · −→ ΩX
n−1
−→ ΩnX → 0
is exact on X 0 : this is a consequence of Lemma 2.0.1 since f |X 0 is a sub-
mersion. This implies that the following short sequence of sheaves is also
exact on X 0 :
n−2 df ∧.
n−1 df ∧.
0 → ΩX/T −→ ΩX −→ df ∧ Ωn−1
X →0
We thus derive the following long sequence of cohomology,
· · · → H 0 (X 0 , ΩX
n−1
) → H 0 (X 0 , df ∧Ωn−1 1 0 n−2
X ) → H (X , ΩX/T ) → · · · , (B.7)
and since X 0 is Stein while Ωn−2

X/T is a coherent sheaf we deduce that the
(df ∧ .)? : H 0 (X 0 , Ωn−1 0 0 n−1

X ) → H (X , df ∧ ΩX )
is a surjective homomorphism.
n−1
Now, if ω ∈ ΩX (X 0 ) is dX/T -closed, then dω can be seen as an element
n−1 dω
of H 0 (X 0 , df ∧ ΩX ) and the above result provides the existence of ∈
df
dω
ΩXn−1
(X 0 ) such that dω = df ∧ .
df
References
1. V. Arnold, A. Varchenko, S. Gussein-Zadé, Singularities of differentiable
maps. Vol. II. Monodromy and asymptotics of integrals. Monographs in
Mathematics, 83. Birkhäuser Boston, Inc., Boston, MA (1988).
2. V. Arnold, V. Goryunov, O. Lyashko, V. Vassiliev, Singularities I. Ency-
clopaedia of Mathematical Sciences, Vol. 6, Springer-Verlag, Berlin and New
York (1993).
3. V. Arnold, V. Goryunov, O. Lyashko, V. Vassiliev, Singularities II. Ency-
clopaedia of Mathematical Sciences, Vol. 39, Springer-Verlag, Berlin and New
York (1993).
4. M.V. Berry, The Riemann-Siegel expansion for the zeta function: high orders
and remainders. Proc. Roy. Soc. London Ser. A 450 Proc. Roy. Soc. Lond.
A 434, no. 1939, 439–462 (1995).
www.pdfgrip.com
208
5. M.V. Berry, C.J. Howls, Hyperasymptotics for integrals with saddles. Proc.
Roy. Soc. Lond. A 434, 657-675 (1991).
6. S.A. Broughton, On the topology of polynomial hypersurfaces. Proc. Symp.
of pure Math. 40, A.M.S. Pub., 167-178 (1983).
7. S.A. Broughton, Milnor numbers and the topology of polynomial hypersur-
faces. Invent. Math. 92, No.2, 217-241 (1988).
8. E.T. Copson, Asymptotic expansions. Cambridge University Press (1965).
9. E. Delabaere, C.J. Howls, Hyperasymptotics for multidimensional Laplace in-
tegrals with boundaries. Toward the exact WKB analysis of differential equa-
tions, linear or non-linear (Kyoto, 1998) 119, 145–163, Kyoto Univ. Press,
Kyoto (2000).
10. E. Delabaere, C.J. Howls, Global asymptotics for multiple integrals with
boundaries. Duke Math. J. 112 (2002), no. 2, 199–264.
11. E. Delabaere, C.J. Howls, Addendum to the hyperasymptotics for multidimen-
sional Laplace integrals. In Analyzable functions and applications, 177–190,
Contemp. Math., 373, Amer. Math. Soc., Providence, RI, (2005).
12. W. Ebeling, Functions of several complex variables and their singularities.
Translated from the 2001 German original by Philip G. Spain. Graduate
Studies in Mathematics, 83. American Mathematical Society, Providence,
RI, (2007).
13. M.V. Fedoryuk, Metod perevala (The saddle-point method). zdat. “Nauka”,
Moscow (1977).
14. H. Grauert, R. Remmert, Theory of Stein spaces. Translated from the Ger-
man by Alan Huckleberry. Reprint of the 1979 translation. Classics in Math-
ematics. Springer-Verlag, Berlin, (2004).
15. K. Fritzsche, H. Grauert, From holomorphic functions to complex manifolds.
Graduate Texts in Mathematics 213, Springer-Verlag, New York, (2002).
16. M. J. Greenberg, Lectures on algebraic topology. W. A. Benjamin, Inc., New
York-Amsterdam (1967).
17. C.J. Howls Hyperasymptotics for multidimensional integrals, exact remainder
terms and the global connection problem. Proc. Roy. Soc. Lond. A 453, 2271-
2294 (1997).
18. D. Kaminski, R.B. Paris, Asymptotics and Mellin-Barnes integrals. Encyclo-
pedia of Mathematics and its Applications, 85. Cambridge University Press,
Cambridge, (2001).
19. J. Leray, Le calcul différentiel et intégral sur une variété analytique complexe
(Problème de Cauchy III). Bull. Soc. Math. France, 87, 81-180 (1959).
20. B. Malgrange, Intégrales asymptotiques et monodromie. Ann. Sci. cole Norm.
Sup. (4) 7 (1974), 405–430 (1975).
21. B. Malgrange, Sommation des séries divergentes. Expositiones Mathemati-
cae 13, 163-222 (1995).
22. J. Milnor, Singular points of complex hypersurfaces. Ann. Math. Stud. 61.
Princeton Univ. Press: Princeton (1968).
23. N. Nilsson, Some growth and ramification properties of certain integrals on
algebraic manifolds. Ark. Mat. 5, 463–476 (1965).
www.pdfgrip.com
209
24. N. Nilsson, Monodromy and asymptotic properties of certain multiple inte-

grals. Ark. Mat. 18, no. 2, 181–198. (1980).
25. F. Pham, Intégrales singulières. Savoirs actuels, EDP Sciences/CNRS Edi-
tions (2005).
26. F. Pham, Vanishing homologies and the n variable saddlepoint method. Proc.
Symposia in Pure Math. 40, part 2 319–333 (1983).
27. F. Pham, La descente des cols par les onglets de Lefschetz, avec vues sur
Gauss-Manin. in Systèmes différentiels et singularités, Astérisque 130 (1985).
28. E.H. Spanier, Algebraic topology. Corrected reprint of the 1966 original.
Berlin: Springer-Verlag (1995).
29. E. M. Stein, Singular integrals and differentiability properties of functions.
Princeton Mathematical Series, No. 30 Princeton University Press, Princeton,
N.J. (1970).
30. M. Tibăr, Polynomials and vanishing cycles. Cambridge Tracts in Mathe-
matics 170, Cambridge University Press (2007).
31. J-L. Verdier, Stratifications de Whitney et théorème de Bertini-Sard. Inven-
tiones math. 36, 295-312 (1976).
February 4, 2010 18:44 WSPC - Proceedings Trim Size: 9in x 6in 05˙jambu
www.pdfgrip.com
210
HYPERGEOMETRIC FUNCTIONS
AND HYPERPLANE ARRANGEMENTS
MICHEL JAMBU
Laboratoire J.A. Dieudonné, UMR 6621, Université de Nice, France
This is a short presentation of some relations between the arrangements of

hyperplanes and the theory of hypergeometric functions. All of this paper is
coming from the illuminating monograph written by P. Orlik and H. Terao9
where all the detailed proofs could be found. Moreover, we focus our interest
on the part of this theory which consists of the computation of the coho-
mology groups, part of the hypergeometric pairing. We mainly consider the
non-resonant weights and we just give some insight on the resonant case.
1. Classical Hypergeometric Functions

1.1. Classical hypergeometric series (Gauss)
The series which has become known as the ordinary hypergeometric se-
ries or the Gauss series is denoted
ab x a(a + 1)b(b + 1) x2
F [(a, b); c; x] = 1 + + +···
c 1! c(c + 1) 2!
When we introduce Appel’s notation (a, n) = a(a + 1)(a + 2) · · · (a + n − 1)
we can write
X (a, n)(b, n) xn
F [(a, b); c; x] =
(c, n) n!
n≥0
Notice that:
• If a = 0 or b = 0 or is negative, then F is a polynomial.
• If c = 0 or is negative, then F is not defined.
• The series is convergent for | x |< 1.
The sum of the series inside its circle of convergence is called the hyperge-
ometric function, and the same name is used for its analytic continuation
outside the circle of convergence.
www.pdfgrip.com
211
Barnes constructed more general hypergeometric series with p numera-

tors parameters (a) = (a1 , . . . , ap ) and q denominators parameters (c) =
(c1 , . . . , cq )
X (a1 , n) · · · (ap , n) xn
p Fq [(a); (c); x] =
(c1 , n) · · · (cq , n) n!
n≥0
Therefore, the original Gauss series is 2 F1 . It is also possible to express

many functions as hypergeometric series. Let us give some few examples:
X xn
ex = = 0 F0 [x]
n!
n≥0
X xn
(1 − x)−a = (a, n) = 1 F0 [a; x]
n!
n≥0
x2 x4 2
cos(x) = 1 − + +··· = 0 F1 [1/2; −x /4]
2! 4!
x3 x5
sin(x) = x − + + · · · = x.0 F1 [3/2; −x2 /4]
3! 5!
x2 x3
log(1 + x) = x − + + · · · = x.2 F1 [(1, 1); 2; −x]
2 3
X xn
Li2 (x) = = x.3 F2 [(1, 1, 1); (2, 2); x]
n2
n≥0
1.2. Hypergeometric differential equation (Gauss)

The differential equation:
d2 y dy
x(x − 1) 2
+ [c − (1 + a + b)] − aby = 0
dx dx
is satisfied by F [(a, b); c; x].
1.3. Hypergeometric integral (Euler)

Z ∞
Let be the Gamma function Γ(x) = e−u ux−1 du and set
0
Z 1
I(x) = ua−1 (1 − u)c−a−1 (1 − xu)−b du
0
www.pdfgrip.com
212
Then I(x) converges if Re(a) > 0 and Re(c − a) > 0.

X xn
Moreover, (1 − xu)−b = (b, n)un
n!
n≥0
X Z 1
n
so I(x) = (b, n)x ua+n−1 (1 − u)c−a−1 du
n≥0 0
X (b, n) Γ(a + n)Γ(c − a)
= xn
n! Γ(c + n)
n≥0
Γ(c − a)Γ(a)
= F [(a, b); c; x]
Γ(c)
Thus we obtain the integral representation
Z 1
Γ(c − a)Γ(a)
F [(a, b); c; x] = ua−1 (1 − u)c−a−1 (1 − xu)−b du
Γ(c) 0
provided that | x |< 1, Re(a) > 0, and Re(c − a) > 0.
2. Modern Approach
This point of view is mainly due to Aomoto and Kita1 on one hand, and
Gelfand7 and Varchenko11 on the other hand.
Let Mx = C \ {0, 1, x−1 }, x 6= 0, 1, and λ = (λ1 , λ2 , λ3 ) ∈ C3 . Let
Φ(u; λ; x) = (1 − u)λ1 uλ2 (1 − xu)λ3
which defines a multivalued holomorphic function on Mx .

In order to write down suitable integrals using homology and cohomology,
we must introduce twisted version where twisting comes from the change of
Φ as we prolong it by analytic continuation while moving around 0, 1, x−1 ,
so we have to introduce a rank one local system L on Mx given by the
representation
ρ : π1 (Mx ) −→ Aut(C) = Gl(1; C) ' C∗
√
γj 7−→ exp(−2π −1λj ); j = 1, 2, 3
for any meridian loop γj about the hyperplane Hj , j = 1, 2, 3 where

H1 = {1}, H2 = {0}, H3 = {x−1 }.
More generally, let us define the multidimensional hypergeometric
functions as follows.
www.pdfgrip.com
213
Let A = {H1 , . . . , Hn } be an affine arrangement of hyperplanes in V '

[n
l
C and M = V \ Hi . For each hyperplane Hi , choose a degree one
i=1
polynomial αi such that Hi = Kerαi and let λ = (λ1 , . . . , λn ) ∈ Cn be a
collection of weights. Define
n
Y
Φ(u; λ) = αλi i := Φλ
i=1
Φλ is a multivalued holomorphic function on, M

A generalized hypergeometric integral is of the form
Z
Φλ η
σ
where σ is a suitable domain of integration and η is a holomorphic form on

M . As for the classical hypergeometric integral, we have to introduce a rank
√
one local system Lλ on M defined with the monodromy exp(−2π −1λi )
around the hyperplanes Hi , i = 1, . . . , n.
The need to calculate the local system cohomoloy H • (M, Lλ ) arises in sev-
eral problems: the Aomoto-Gelfand theory of multivariable hypergeometric
integrals;1,7 representation theory of Lie algebras and quantum groups and
solutions of the Knizhnik-Zamolodchikov differential equations in confor-
mal field theory;11 determining the cohomology groups of the Milnor fiber
of the non-isolated hypersurface singularity at the origin obtained by coning
the arrangement.
3. Local Systems
Definition 3.1. A locally constant sheaf on M of complex vector spaces

of dimension r is called complex local system of rank r on M .
Example 3.1. Let r = 1, M = C∗ , and U an open set in C∗ . Define

√
L(U ) = C{branch of u on U }. Then L(C∗ ) = {0}, L(U ) ' C if U is sim-
ply connected and Lx ' C. Then we get the monodromy representation
ρ : π1 (C∗ ; 1) −→ Gl(1; C) ' C∗
γ 7−→ −1
where γ is a generator of π1 (C∗ ; 1).

www.pdfgrip.com
214
Example 3.2. Let r = 1, M = C∗ , and U an open set in C∗ . Define

1 √
L(U ) = C{branch of u−λ on U } where λ = − √ logµ , µ = e−2π −1λ .
2π −1
Then L(C∗ ) = {0}, L(U ) ' C if U is simply connected and Lx ' C. Then
we get the monodromy representation
ρ : π1 (C∗ ; 1) −→ Gl(1; C) ' C∗
γ 7−→ µ
where γ is a generator of π1 (C∗ ; 1).
Remark 3.1. There is a bijection between the isomorphism classes of

rank r local systems on M and the isomorphism classes of representations
π1 (M ) → Gl(r; C).
Our central object is the local system Lλ on M defined as follows.

Let Lλ (U ) = {f : U → C, holomorphic such that df + ωλ ∧ f = 0} where
dΦλ
ωλ = = d(log(Φλ ).
Φλ
Proposition 3.1. Lλ defines a local system of rank 1 on M with the
monodromy
√
representation ρ : π1 (M ) −→ Gl(1; C) ' C∗ where ρ(γi ) 7→
−2π −1λi
e and γi are the generators of π1 (M ).
Proof. Denote ∇λ = d + ωλ ∧ .
dΦλ −1
∇λ (Φ−1 −2
λ ) = −Φλ dΦλ + .Φ = 0
Φλ λ
dΦλ
If ∇λ (f ) = 0, then df = −f ωλ = −f . Thus d(f Φλ ) = df Φλ +f dΦλ = 0
Φλ
so f ∈ CΦ−1
λ . Cover M with contractible open sets to see that Lλ is locally
constant.
Let us point out that for any m = (m1 , . . . , mn ) ∈ Zn , the local system
associated to λ + m coincides with that associated to λ.
4. Hypergeometric Pairing
We have to interpret hypergeometric integrals as the result of the hyperge-
ometric pairing.
www.pdfgrip.com
215
The rank one local system Lλ on M defines the cohomology groups

H p (M, Lλ ). On the other hand, the dual local system L∨ λ defines the homol-
ogy groups Hp (M, L∨ λ ). In fact, it is shown that we have a perfect pairing
H p (M, Lλ ) × Hp (M, L∨
λ ) −→ C
Z
(η, σ) 7−→ Φλ η
σ
In this paper, we focus on the determination of the cohomology groups

H p (M, Lλ ).
5. Cohomology Groups H p (M, Lλ )

Let O denote the sheaf of germs of holomorphic functions on M and let
Ω• be the de Rham complex of germs of holomorphic differentials on M ,
where Ω0 = O. Then we see that ∇λ : Ω0 → Ω1 is a flat connection whose
kernel is Lλ . Extend to a derivation of degree one. The sequence
∇λ 1 ∇λ ∇λ l
0 → Lλ −→ Ω0 −→ Ω −→ · · · −→ Ω →0
is exact.
Theorem 5.1. (Holomorphic de Rham theorem)
H p (M, Lλ ) ' H p (Γ(M ; Ω• ), ∇λ )
where Γ denotes global sections on M .

S
Let Ωp (∗A) be the group of p-rational forms on M with poles on ni=1 Hi .
Then Ωp (∗A) ⊂ Γ(M ; Ω• ). But ωλ ∈ Ωp (∗A) then (Ω• (∗A), ∇) is a com-
plex.
Theorem 5.2. (Algebraic de Rham theorem (Deligne, Grothendieck))
H p (Γ(M ; Ω. ), ∇λ ) ' H p (Ωp (∗A), ∇λ )
The problem is still difficult. We have to reduce the case of arbitrary poles
Sn
on N = i=1 Hi to the case of order one (i.e. logarithmic). Following
Deligne’s results, we must compactify M with a normal crossing divisor.
We embed V ⊂ CP l by adding the infinite hyperplane, H∞ . Define the
projective closure of A, as A∞ . The divisor N (A∞ ) may have non-normal
crossings. Before going further, let us review some basic notions on hyper-
plane arrangements.
www.pdfgrip.com
216
6. Hyperplane Arrangements
6.1. Generalities
We refer the reader to Ref. 8 as a general reference on arrangements.
Let A = {H1 , . . . , Hn } be an arrangement of hyperplanes over C, where Hi
are affine hyperplanes of Cl . A is also called l-arrangement. A is said to be
[n
T
central if H∈A H 6= ∅. Define the complement M (A) = Cl \ Hi and
i=1
L(A) the poset of nonempty intersections of hyperplanes of A with reverse
order
X ≤ Y if Y ⊆ X
Notice that the rank denoted rk satisfies rk(X) = codim(X). An element

X ∈ L(A) is called an edge of A. The rank of A is the maximal number
of linearly independent hyperplanes in A. In the following, we will denote
r = rk(A).
S
Let N (A) = H∈A H be the divisor of A. For each hyperplane Hi , choose a
Yn
degree one polynomial αi such that Hi = Kerαi . The product Q(A) = αi
i=1
is a defining polynomial for A.
The affine arrangement A gives rise to a central arrangement cA in Cl+1 ,
called the cone over A. Let Q̃ be the homogenized Q(A) with respect to
the new variable u0 . Then Q(cA) = u0 Q̃.
Conversely, given a central arrangement A in V = Cl+1 and H ∈ V , we
define an affine arrangement dH A, called the decone of A with respect to
H.
Embed V = Cl in the complex projective space CP l and call the comple-
ment of V the infinite hyperplane denoted H ∞ . Let H be the projective clo-
sure of H and write N (A∞ ) = (∪H∈A )H) ∪ {H ∞ }. Then Q(A∞ ) = Q(cA).
We construct the projective quotient PA and choose coordinates so that
PH = keru0 is the hyperplane at infinity. By removing it, we obtain an
affine arrangement dH A.
Let µ : L(A) → Z be the MöbiusX function of L(A) defined by µ(V ) = 1,
and for X > Y by the recursion µ(Y ) = 0.
Y ≤X
Definition 6.1. The characteristic polynomial of A is defined as

X
χ(A, t) = µ(X)tdimX .
X∈L(A)
www.pdfgrip.com
217
Given an edge X ∈ L(A), we define the subarrangement AX of A by

AX = {H ∈ A | X ⊆ H} and the arrangement AX called restriction of
AX to X by AX = {X ∩ H | H ∈ A and X ∩ H 6= ∅}.
The deletion-restriction triple (A, A0 , A00 ) is a nonempty arrangement A
and H ∈ A together with A0 = A \ {H} and A00 = AH .
6.2. Orlik-Solomon algebra

The Orlik-Solomon algebra is combinatorially defined. Let A =
{H1 , . . . , Hn } be an affine arrangement and let {e1 , . . . , en } be a set. Let
V
E = C (e1 , . . . , en ) be the free exterior algebra over C. If S = (i1 , . . . , ip )
is an ordered p-tuple, denote the product ei1 ∧ · · · ∧ eip by eS . Then the
Orlik-Solomon algebra is defined as
A• (A) := E/J
T
where J ideal generated by all eS with i∈S Hi = ∅ and the relations of
the form:
X s
(−1)i−1 ei1 . . . ebij . . . eis
j=1
for all 1 ≤ i1 < . . . < is ≤ n such that rk(Hi1 ∩ · · · ∩ His ) < s, i.e.
the hyperplanes {Hi1 , · · · , His } are dependent and where b indicates an
omitted factor.
T
Notice that for a central arrangement, i∈S Hi 6= ∅ for any S.
We will denote aH (resp. aS ) the image of eH (resp. eS ) under the natural
projection and denote bp (A) = dimAp (A) the p-th Betti number of A• .
P p •
Let P (A, t) = p≥0 bp (A)t be the Poincaré polynomial of A . It is
P
shown that P (A, t) = X∈L(A) µ(X)(−t)codim(X) .
Moreover, P (cA, t) = (1 + t)P (A, t).
P (A, t) is closely related to the characteristic polynomial χ(A, t). As an
important property of this algebra is the following theorem due to Orlik
and Solomon.
Theorem 6.1. The cohomology algebra and the Orlik-Solomon algebra are
isomorphic as graded algebras
A• (A) ∼
= H • (M (A); C).
6.3. Brieskorn algebra

Define the Brieskorn algebra denoted B• (A) as the C-algebra generated
dαH
by 1 and the forms ωH = = dlog(αH ), H ∈ A.
αH
www.pdfgrip.com
218
The inclusion B• (A) ⊂ Ω•M induces isomorphisms of graded algebras

B• (A) ' H • (M (A), C)
6.4. Dense edges

Let A be a central arrangement in V . We call A decomposable if A is the
disjoint union of two nonempty subarrangements A1 and A2 .
Definition 6.2. X ∈ L(A) is called dense edge in A if and only if the

central arrangement AX is not decomposable.
[Let Dj (A) denote the set of
dense edges of dimension j and let D(A) = Dj .
j≥0
The divisors N (A) and N (A∞ ) do not have normal crossings along a dense
edge.
Example 6.1. An l-arrangement A is called a general position ar-

rangement if for every subset {H1 , . . . , Hq } ⊆ A with q ≤ l, then
r(H1 ∩ . . . ∩ Hq ) = q and when q > l, H1 ∩ . . . ∩ Hq = ∅. For such a
arrangement A, the set of dense edges D(A) = ∅.
Lemma 6.1. Let A be an nonempty central arrangement with H ∈ A. If

A0 and A00 are decomposable, then A is decomposable.
Example 6.2. Let A be the Selberg arrangement defined by

Q(A) = u1 (u1 − 1)u2 (u2 − 1)(u1 − u2 ).
Label the hyperplanes in the order given by the factors in Q and write j in
place of Hj . Let
Q(A∞ ) = u0 u1 (u1 − u0 )u2 (u2 − u0 )(u1 − u2 ) = u0 Q(A).
Then D1 (A) = {1, 2, 3, 4, 5}, D2 (A) = {135, 245}. The additional dense
edges in its projective closure A∞ are {∞, 12∞, 34∞}.
6.5. The β invariant

In higher dimensions, it is difficult to determine the dense edges. The fol-
lowing provides a numerical criterion to decide which edges are dense. Re-
call that the characteristic polynomial of the arrangement A is defined as
P
χ(A, t) = X∈L(A) µ(X)tdim(X) .
www.pdfgrip.com
219
Remark 6.1. Let A be a central arrangement, then χ(A, 1) = 0.
Definition 6.3. Let A be an arrangement of rank r. Define its beta in-

variant by
β(A) = (−1)r χ(A, 1)
Remark 6.2. β(A) =|e(M )|, where e(M ) is the Euler characteristic of the
complement. β(A) is a combinatorial invariant.
Remark 6.3. If A is a complexified real arrangement (i.e. the polynomials

defining the hyperplanes have real coefficients), then β(A) is the number of
bounded chambers of the complement of the real arrangement.
Theorem 6.2. Let A be an arrangement and let X ∈ L(A). The following

conditions are equivalent:
1. X is dense,
2. AX is not decomposable,
3. β(dAX ) 6= 0,
4. β(dAX ) > 0.
The proof mainly uses the properties of the triple (A, A0 , A00 ) where A0 =
A \ {H} for a suitable H.
7. Resonance
X
Let λ∞ = − λH be the weights of H∞ . For X ∈ L(A∞ ), define λX ∈ C
H∈A
by
X
λX = λH , H ∈ A.
X⊂H
P
Let ωλ = H∈A λH ωH = d(logΦλ ). Since ωλ ∧ ωλ = 0, wedge product with
ωλ provides a finite dimensional subcomplex (B• , ωλ ∧) of (Ω• (∗A), ∇λ ):
ωλ ∧ 1 ωλ ∧ ωλ ∧ l
0 −→ B0 −→ B −→ . . . −→ B −→ 0
Theorem 7.1 (Refs. 4,10). Assume that λX ∈

/ Z>0 , for every dense edge
X ∈ L(A∞ ). Then, for every p
H p (M, Lλ ) ' H p (B• (A), ωλ ∧).

www.pdfgrip.com
220
Notice that the map aH 7→ ωH induces an isomorphism of graded algebras

A• (A) → B •
X(A).
Let aλ = λH aH . Since aλ ∧ aλ = 0, exterior product with aλ provides
H∈A
a complex (A• (A), aλ ∧)
aλ ∧ 1 aλ ∧ aλ ∧ l
0 −→ A0 (A) −→ A (A) −→ . . . −→ A (A) −→ 0
Corollary 7.1 (Ref. 9). Assume that λX ∈
/ Z>0 , for every dense edge
X ∈ L(A∞ ) Then, for every p
H p (M, Lλ ) ' H p (B• (A), ωλ ∧) ' H p (A• (A), aλ ∧).
Example 7.1. Let A be a general position arrangement. Then for all p < l,
H p (A• (A), aλ ∧) = 0.
This completes the transformation of the analytic problem into a problem

in combinatorics.
We mention the following easy result which explains the assumption on the
P
weights. This result explains why it is assumed λ∞ = − H∈A λH .
P
Proposition 7.1 (Refs. 5,12). If H∈A∞ λH 6= 0, then H • (A• , aλ ∧) =
0.
Proof. Let ∂ be the map defining by ∂(ai aS ) = aS − ai ∂aS . Let denote dλ

P
the left multiplication by aλ . It follows that dλ ∂ + ∂dλ = ( H∈A∞ λH )id ,
P
so that ( H∈A∞ λH )−1 ∂ is a chain contraction of (A• , aλ ∧). Thus
H • (A• , aλ ∧) = 0.
The case λ = 0 implies the well-known result that A• (A) is isomorphic to

the ordinary cohomology H • (M ; C).
The following result is due to S. Yuzvinsky.
Theorem 7.2 (Ref. 12). Assume that λX ∈
/ Z≥0 , for every dense edge
X ∈ L(A∞ ). Then
H q (A• (A), aλ ∧) = 0 for q 6= r
and
dimH r ((A• (A), aλ ∧) = dimH r (M, Lλ ) =| e(M ) |
Up to now, we considered some systems of weights which defined a subset

of Cn :
W(A) = {λ ∈ Cn | λX ∈
/ Z≥0 for every dense edge X of A∞ }
www.pdfgrip.com
221
For λ ∈ W(A), the local system cohomology groups are independent of the
weights. There exists a maximal dense open subset U(A) ⊇ W(A) where
the local system cohomology groups are independent of the weights. We
call weights λ ∈ U(A) non-resonant. Notice that there are in general
non-resonant weights which do not lie in the set W(A).
Much is known in such a case which is called of non-resonant weights. But
substantially less is known about resonant weights. This interest has been
a motivating factor in many works and we will refer to the paper of D.
Cohen and P. Orlik.3 They study the local system cohomology using strat-
ified Morse theory following D. Cohen.2 They construct a universal com-
plex (KΛ• (A), ∆• (x)), where x = (x1 , . . . , xn ) are non-zero complex vari-
ables, Λ = C[x±1 ±1
1 , . . . , xn ] denote the ring of complex Laurent polynomials,
q q
KΛ (A) = Λ ⊗C KΛ (A) ' Λbq (A) where bq (A) = dimAq (A) = dimH q (M ; C)
is the q-th Betti number of M with trivial local coefficients C and ∆• (x) are
Λ-linear with the property that the specialization xj 7→ tj = exp(−2πiλj )
calculates H • (M, Lλ ).
There is also a similar universal complex, called the Aomoto complex
(A•R (A), ay ∧) where y = (y1 , . . . , yn ) are variables, R = C[y1 , . . . , yn ] is the
polynomial ring, where AqR (A) = R⊗C Aq (A) and boundary maps are given
by p(y) ⊗ η 7→ yi p(y) ⊗ aHi ∧ η.
For λ ∈ Cn , the specialization y 7→ λ of the Aomoto complex yields the
Orlik-Solomon algebra (A• (A), aλ ∧).
Theorem 7.3 (Ref. 3). For any arrangement A, the Aomoto complex
(A•R (A), ay ∧) is chain equivalent to the linearization of the universal com-
plex (KΛ• (A), ∆• (x)).
Recall that, for m ∈ Zn , the local system associated to λ + m coincides

with that associated to λ. Finally, some lower and upper bounds could be
determined for arbitrary weights as following:
supm∈Zn dimH p (A• , aλ+m ∧) ≤ dimH p (M, Lλ ) ≤ dimH p (M ; C)
Thus a system of weights λ ∈ Cn is non-resonant if the Betti numbers of
M with coefficients in the local system Lλ are minimal.
8. The NBC Complex

Let A be an affine arrangement of hyperplanes over C, A = {H1 , . . . , Hn },
define the linear order on A by Hi ≺ Hj if i < j. An inclusion-minimal
dependent set is called a circuit. A broken circuit is a set S for which
there exist H ≺ min(S) such that {H} ∪ S is a circuit. The collection of the
www.pdfgrip.com
222
empty set and the nonempty subsets of A which have nonempty intersection
and contain no broken circuit is called NBC(A). Since NBC(A) is closed
under taking subsets, it forms a pure (r−1)-dimensional simplicial complex.
Example 8.1. Let A be the Selberg arrangement as defined above. The

1-simplices of NBC(A) are {{1; 3}, {2, 3}, {1, 4}, {2, 4}, {1, 5}, {2, 5}}.
Theorem 8.1 (Ref. 9). Let A be an l-arrangement of hyperplanes of rank

r ≥ 1. Then NBC(A) has the homotopy type of a wedge of spheres,
W r−1
β(A S .
Idea of the proof: If v is a vertex of NBC(A) then its star, denoted st(v),
consists of all the simplexes whose closure contains v. The closure st(v) is a
cone with cone point v. Let (A, A0 , A00 ) be a triple with respect to the last
hyperplane Hn . Then prove that NBC(A00 ) ' st(Hn ) ∩ NBC(A0 ). Consider
the Mayer-Vietoris sequence for the excisive couple {st(Hn ), NBC(A0 )} to
get the long exact sequence and use the fact that st(Hn ) is contractible.
Theorem 8.2 (Ref. 9). Let A be an l-arrangement of hyperplanes of rank
r ≥ 1. Then
H p (NBC(A)) = 0 if p 6= r − 1
= free of rank β(A) if p = r − 1
Idea of the proof: Use induction on r and consider the long exact sequence
of the previous theorem.
A maximal independent set is called a frame. An (r − 1)-dimensional sim-

plex of NBC(A) is called an nbc frame. Following Ziegler,13 let define a
subset βnbc(A) of nbc(A) of cardinality β(A).
Let define a βnbc(A) frame B as a frame which is a nbc frame such that
for every H ∈ B, there exists H 0 ≺ H in A with (B \ {H} ∪ {H 0} is a frame.
Let βnbc(A) be the set of all nbc(A) frames. When A is empty, we agree
that βnbc(A)= ∅.
Let (A, A0 , A00 ) be a triple with respect to the last hyperplane Hn .
Then there is a disjoint union βnbc(A) = βnbc(A0 ) ∪ βnbc(A00 ) where
βnbc(A00 ) = {{ν(B 00 ), Hn } | b00 ∈ βnbc(A00 )} and ν(X) = min(AX ).
For an nbc frame B ∈nbc(A), B ∗ ∈ C r−1 (NBC(A)), denotes the (r − 1)-
cochain dual to B.
Theorem 8.3 (Ref. 6). The set {[B ∗ ] | B ∈ βnbc(A)} is a basis for
H r−1 (NBC(A)).
www.pdfgrip.com
223
Example 8.2. Let A be the Selberg arrangement as defined above. Then

βnbc(A) = {{2, 4}, {2, 5}}. So the cohomology classes [{2, 4}]∗ and [{2, 5}]∗
form a basis for H 1 (NBC(A)).
Let us back to the combinatorial complex (A• (A), aλ ∧).

Theorem 8.4 (Ref. 9). Let A be an affine arrangement with projective
closure A∞ . Assume that λX =
6 0 for every dense edge X ∈ D(A∞ ), then
H p (A• (A), aλ ∧) ' H p−1 (NBC(A), C)
Let B = {Hi1 , · · · , Hir } be a βnbc frame and X1 > · · · > Xr where

r
\ ^r
Xp = Hik for 1 ≤ p ≤ r. Define ζ(B) = ωλ (Xp ) where ωλ (X) =
k=p p=1
P
H∈AX λH ωH ∈ B1 (A).
Theorem 8.5 (Ref. 9). Let A be an affine arrangement of rank r with
projective closure A∞ . Assume that λX ∈
/ Z≥0 for every X ∈ D(A∞ ).
Then the set
{ζ(B) ∈ H r (M, Lλ ) | B ∈ βnbc(A)}
is a basis for the only nonzero local system cohomology group, H r (M, Lλ ).
Thus, there is an explicit isomorphism between the only non trivial coho-
mology group H r (M, Lλ ) and H r−1 (NBC(A), C) under the non resonance
conditions.
Example 8.3. Let A be the Selberg arrangement once more. Assume the
weights of the dense edges satisfy suitable conditions. Then
ζ({2, 4}) = (λ2 ω2 + λ4 ω4 + λ5 ω5 )λ4 ω4 = λ2 λ4 ω24 − λ4 λ5 ω45
ζ({2, 5}) = (λ2 ω2 + λ4 ω4 + λ5 ω5 )λ5 ω5 = λ2 λ5 ω25 + λ4 λ5 ω45

provide a basis for H 2 (M, Lλ ).
References
1. K. Aomoto and M. Kita, Hypergeometric Functions (in Japanese), (Springer-
Verlag, 1994).
2. D. Cohen, Adv. Math. 97, 231 (1993).
3. D. Cohen and P. Orlik, Mathematical Research Letters 7, 299 (2000).
4. H. Esnault, V. Schechtman and E. Viehweg, Invent. Math. 109, 557 (1992);
Erratum, ibid 112, 447 (1993).
5. M. Falk, Annals of Combinatorics 1, 135 (1997).
www.pdfgrip.com
224
6. M. Falk and H. Terao, Trans. AMS 349, 189 (1997).

7. I. M. Gelfand, Soviet Math. Dokl. 33, 573 (1986).
8. P. Orlik and H. Terao, Arrangements of Hyperplanes, Grundlehren Math.
Wiss., Vol. 300 (Springer-Verlag, Berlin, 1992).
9. P. Orlik and H. Terao, Arrangements and Hypergeometric Integrals, MSJ
Memoirs, Mathematical Society of Japan 9 (2001).
10. V. Schechtman, H. Terao and A. Varchenko, J. Pure Appl. Algebra 100, 93
(1995).
11. A. N. Varchenko, Multidimensional Hypergeometric Functions and Repre-
sentation Theory of Lie groups, Adv. Studies in Maths. Physics 21 (World
Scientific Publishers, 1995).
12. S. Yuzvinsky, Comm. Algebra 23, 5339 (1995).
13. G. Ziegler, J. Alg. Combinatorics 1, 283 (1992).
www.pdfgrip.com
225
BERNSTEIN-SATO POLYNOMIALS AND

FUNCTIONAL EQUATIONS
MICHEL GRANGER
Université d’Angers, LAREMA, UMR 6093 du CNRS
2 Bd Lavoisier 49045 Angers France
March 23, 2010
These notes are an expanded version of the lectures given in the frame
of the I.C.T.P. School held at Alexandria in Egypt from 12 to 24 November
2007.
Our purpose in this course was to give a survey of the various aspects,
algebraic, analytic and formal, of the functional equations which are sat-
isfied by the powers f s of a function f and involve a polynomial in one
variable bf (s) called the Bernstein-Sato polynomial of f . Since this course
is intended to be useful for newcomers to the subject we give enough signif-
icant details and examples in the most basic sections, which are sections 1,
2, and also 4. The latter is devoted to the calculation of the Bernstein-Sato
polynomial for the basic example of quasi-homogeneous polynomials with
isolated singularities. This case undoubtedly served as a guide in the first
developments of the theory.
We particularly focused our attention on the problem of the meromor-
s
phic continuation of the distribution f+ in the real case, which in turn moti-
vated the problem of the existence of these polynomials, without forgetting
related questions like the Mellin transform and the division of distributions.
See the content of section 3. The question of the analytic continuation prop-
erty was brought up as early as 1954 at the congress of Amsterdam by I.M.
Gelfand. The meromorphic continuation was proved 15 years later indepen-
dently by Atiyah and I.N. Bernstein-S.I. Gel’fand who used the resolution
of singularities. The existence of the functional equations proved by I.N.
Bernstein in the polynomial case allowed him to give a simpler proof which
does not use the resolution of singularities. His proof establishes at the
same time a relationship between the poles of the continuation and the
zeros of the Bernstein Sato polynomial. The already known rationality of
the poles gave a strong reason for conjecturing the famous result about the
rationality of the zeros of the b-function which was proved by Kashiwara
and Malgrange.
www.pdfgrip.com
226
We want also to mention another source of interest for studying func-

tional equations due to Mikio Sato. It concerns the case of the semi-
invariants of prehomogeneous actions of an algebraic group, and especially
of a reductive group. In the latter case the functional equation is of a very
particular type and the name b-function frequently employed as a shortcut
for the Bernstein-Sato Polynomial, comes from this theory. We give the
central step of the proof of the existence theorem in the reductive case in
section 2.4.
Let us summarize the contents of the different sections. In section 1, we
give the basic definitions and elementary facts about b-functions with em-
phasis on first hand examples. In section 2 we recall the proof of Bernstein
for the existence theorem in the polynomial case. Although also treated
by F. Castro in this volume we give it for the sake of completeness and
also to make it clear that the case of multivariable Bernstein-Sato polyno-
mials can be solved with the same proof in the algebraic case. In section
3 we give a detailed proof of the analytic continuation property using the
functional equation and we make a comparison with the proof which uses
the Mellin transform and asymptotic expansions in the way Atiyah and
Bernstein-Gel’fand first did. In section 4, we give a proof of the calculation
of the Bernstein polynomial in the quasi-homogeneous case. In so doing we
give a large view of a preprint less accessible to the public than a published
version which treats directly the more complicate case of a singularity non-
degenerate with respect to its Newton polygon. In section 5 we give the
main steps of the proof of the existence theorem for the local analytic case.
This proof is originally due to Kashiwara and uses a fairly large amount of
material from analytic D-module theory. In order to make this section 5
more readable we gathered a summary of the necessary material in section
7 refering to the literature for the details. Finally in section 6 we give an
account without proof of a very fundamental property of Bernstein-Sato
polynomials, the fact that their roots are negative rational number. This
result using different methods, is basically due to B. Malgrange and M.
Kashiwara.
I am aware of the fact that these notes do not cover all the aspects of
the subject or recent developments like the theory of the V filtration in the
continuity of Kashiwara and Malgrange results, the microlocal aspects, the
computational aspects in the algebraic case, the relative b-functions and
their link with deformations, the prehomogeneous space aspects. I refer
the reader to the bibliography and the references it contains for further
reading.
I wish to end this introduction first by thanking Professor Lê Duńg
Tráng who conceived and organized the school, as well as the local organizers
www.pdfgrip.com
227
and especially Professor Mohamed Darwish for their hospitality and the way
they took care of all the practical details of our stay in Alexandria.
Contents
1 Introduction to Functional Equations 228
1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
1.2 A review of a number of elementary facts about b-functions 230
1.3 A first list of examples . . . . . . . . . . . . . . . . . . . . . 231
1.4 Remarks on variants of the definition . . . . . . . . . . . . . 232
2 On the existence of a functional equation in the

polynomial case 233
2.1 Holonomic modules . . . . . . . . . . . . . . . . . . . . . . . 233
2.2 Bernstein equation . . . . . . . . . . . . . . . . . . . . . . . 235
2.3 Generalisations . . . . . . . . . . . . . . . . . . . . . . . . . 236
2.4 Semi-invariants of prehomogeneous spaces . . . . . . . . . . 238
3 Bernstein-Sato polynomials and analytic continuation

of f s 241
3.1 Roots of b and analytic continuation of Y (f )f s . . . . . . . 241
3.2 Asymptotic expansion and Mellin transforms . . . . . . . . 243
3.3 Application to analytic continuation . . . . . . . . . . . . . 245
3.4 Application to the division of distribution . . . . . . . . . . 247
4 Quasi-homogeneous and semi-quasi-homogeneous

isolated singularities 248
4.1 Quasi-homogeneous polynomials . . . . . . . . . . . . . . . . 248
4.2 Semi-quasi-homogeneous germs . . . . . . . . . . . . . . . . 251
4.3 Calculation of the Bernstein-Sato polynomial of a
quasi-homogeneous polynomial . . . . . . . . . . . . . . . . . 253
5 Proof of the existence of the b-function for a an

analytic germ 256
5.1 The proof in the analytic case . . . . . . . . . . . . . . . . . 256
5.2 An application: Holonomicity of the module O[ f1 ] . . . . . . 259
5.3 Links between various Bernstein-Sato polynomials . . . . . . 260
6 Rationality of the zeroes over any algebraically closed

field of characteristic zero 261
6.1 The rationality of the zeros . . . . . . . . . . . . . . . . . . 261
6.2 Derived results . . . . . . . . . . . . . . . . . . . . . . . . . 262
www.pdfgrip.com
228
7 Introduction to analytic differential modules.

Abridged version 263
7.1 The ring of differential operators . . . . . . . . . . . . . . . 263
7.1.1 Definition of D(U ) on an open subset of Cn . . . . . 263
7.1.2 Behaviour under change of coordinates . . . . . . . . 264
7.1.3 Sheaf of differential operators on an analytic
manifold . . . . . . . . . . . . . . . . . . . . . . . . . 266
7.1.4 Principal symbols, and graded associated sheaves . . 266
7.1.5 Coherence . . . . . . . . . . . . . . . . . . . . . . . . 268
7.2 Coherent D-modules or differential systems . . . . . . . . . 270
7.2.1 What is a differential system? . . . . . . . . . . . . . 270
7.2.2 Regular connections . . . . . . . . . . . . . . . . . . 271
7.3 Good filtrations and coherence conditions . . . . . . . . . . . 273
7.4 Characteristic variety of a differential system . . . . . . . . 276
7.4.1 Case of a monogeneous module M = D I . . . . . . . 276
7.4.2 General case . . . . . . . . . . . . . . . . . . . . . . 277
7.4.3 A finiteness property . . . . . . . . . . . . . . . . . . 278
8 Complements 282
8.1 Appendix A: Mellin and Laplace transforms . . . . . . . . . 282
8.1.1 Integrals depending on a parameter . . . . . . . . . . 282
8.1.2 Analyticity of Mellin transforms . . . . . . . . . . . 282
8.2 Appendix B: Regular sequences, and application to the
annihilator AnnD f s , in the isolated case . . . . . . . . . . . 285
1 Introduction to Functional Equations
1.1 Definitions
Let us consider the ring O = C{x1 , · · · , xn } of germs of functions defined
by a convergent power series at the origin of Cn :
X
f (x1 , · · · , xn ) = aα xα with xα = xα αn
1 · · · xn
1
α∈Nn
Let D be the ring of differential operators with coefficients in O i.e. the

set of sums P of monomials in the variables xi and in the partial derivatives
denoted ∂i or ∂∂x
i
X β
P = P (x, ∂ x ) = fβ (x)∂ βx with ∂ x = ∂1β1 · · · ∂nβn .
β∈Nn
www.pdfgrip.com
229
Given a germ of an analytic function f ∈ O = C{x1 , · · · , xn } there

exists a non zero polynomial e(s) and an analytic differential operator
P (s) ∈ D[s] polynomial in the indeterminate s such that
P (s)f s+1 = e(s)f s (1)
The set of polynomials e(s) for which an equation of the type (1) exists
is clearly an ideal Bf of C(s), which is principal since C[s] is a principal
domain.
The Bernstein-Sato polynomial of f is by definition the monic generator
of this ideal denoted by bf (s) or simply b(s):
Bf = C[s] · bf (s)
Here are some variants and extensions of this definition
1) Algebraic case. In all what follows we denote K a field of char-
acteristic zero. When f ∈ K[X1 , · · · , Xn ] wePconsider the Weyl algebra
β
An (K) i.e. the set of operators P (x, ∂ x ) = β∈N fβ (x)∂ x with polyno-
mial coefficients fβ ∈ K[X1 , · · · , Xn ]. The global or algebraic Bernstein-
Sato polynomial is the monic generator of the ideal of polynomials e(s)
included in a functional equation as in (1) but with a polynomial operator
P (s) ∈ An (K)[s].
2) We may also consider the formal analogue where f and the coefficients
of P (s) are in K[[X1 , · · · , Xn ]].
3) A generalisation: let f1 , · · · , fp be p elements in C{x1 , · · · , xn }. Then
there exists a non zero polynomial b(s1 , · · · , sp ) ∈ C[s1 , · · ·, sn ], and a func-
tional equation:
s s +1
b(s1 , · · · , sp )f1s1 · · · fp p = P (s1 , · · · , sp )f1s1 +1 · · · fp p (2)
with
b ∈ C[s1 , · · · , sp ] , P ∈ D[s1 , · · · , sp ]
The set of polynomials b(s1 , · · · , sp ) as in (2) is an ideal B(f1 ,··· ,fp ) of
K[s1 , · · · , sp ]. There is an algebraic variant of this notion of a Bernstein-
Sato ideal B(f1 ,··· ,fp ) .
History of the existence theorem

This polynomial was simultaneously introduced by Mikio Sato in a dif-
ferent context in view of giving functional equations for relative invariants
of prehomogeneous spaces and of studying zeta functions associated with
them, see [33], [34]. The name b-function comes from this theory and the so-
called a,b,c functions of M. Sato, see [35] for definitions. The existence of a
www.pdfgrip.com
230
nontrivial equation as stated in (1), was first proved by I.N. Bernstein in the
polynomial case, see [6]. The polynomials bf (s) are called Bernstein-Sato
polynomials in order to take this double origin into account. The analytic
local case is due to Kashiwara in [21]. An algebraic proof by Mebkhout
and Narvaez can be found in [32]. The formal case is given by Björk in his
book [9]
Nontrivial polynomials as in (2) were first introduced by C. Sabbah,
see [36]. In the algebraic case the proof is a direct generalisation of the
proof of Bernstein. For the analytic case see [36], completed by [3] where is
shown the necessity of using a division theorem proved in [1]. There is by
[36], [3] a functional equation (2) in which the polynomial b(s1 , · · · , sp ) is a
product of a finite number of affine forms. It is as far as I know an unsolved
problem to state the existence of a system of generators of Bf1 ,···fp made
of polynomials of this type.
1.2 A review of a number of elementary facts about

b-functions
• The functional equation (1) is an identity in O[s, f1 ] · f s which is the
rank one free module over the ring O[s, f1 ], with s as an indeterminate.
The generator is denoted f s viewed as a symbol, in order to signpost
the fact that we give this module the D[s]-module structure in which
the action of the derivatives on a generic element g(x, s) · f s is:
∂ ∂g s ∂f
· g(x, s)f s = + g · ∂xi f s
∂xi ∂xi f
We may notice that the multiplication by s on this module is D-linear.
• We consider the submodule D[s] · f s generated by f s . The equation
(1) means that the action of s on the quotient:
D[s]·f s D[s]·f s
s̃ : D[s]·f s+1 −→ D[s]·f s+1
which is the D-linear map
[P (s)f s ] → [sP (s)f s ].

D[s]·f s
admits a minimal polynomial hence that the module D[s]·f s+1 is finite
over D.
It is a remarkable fact due to Kashiwara that D[s] · f s itself is finite
over D. We sometimes denote f m · f s = f s+m , which corresponds to
www.pdfgrip.com
231
the intuitive meaning and the change of s into s+m induces a D-linear
automorphism of O[s, f1 ] · f s , which restrict to D[s] · f s → D[s] · f s+m
given by P (s)f s → P (s + m)f s+m .
• Recall that there is a unique maximal ideal in O:
M = {f ∈ O | f (0) = 0}
- If f ∈
/ M is a unit in O then bf = 1. Similarly, in the algebraic case
bf = 1 if f is a constant.
- If f is in the maximal ideal of O in the local analytic case, or if f is
not a constant in the polynomial case, we obtain by setting s = −1
in the functional equation: P (−1) · 1 = b(−1) f1 , and this implies
b(−1) = 0. We write usually in this case b(s) = (s + 1)b̃(s).
Setting s = −1 in the equation now gives P (−1) · 1 = 0 and therefore
n
X n
X
∂ ∂
P (−1) = Ai , P (s) = (s + 1)Q(s) + Ai
i=1
∂ xi i=1
∂xi
Carrying this over to the functional equation leads to the following

result:
Lemma 1.1 The polynomial b̃(s) is the minimal polynomial of the

D[s]·f s
action of s on (s+1) D[s]·f s+1 . This is the same as the unitary minimal
polynomial such that there is a functional equation:

n
X ∂ s
b̃(s)f s = Q(s) · f + Ai (s) ·f
i=1
∂xi
We summarize this fact by writing b̃(s)f s ∈ D[s](f + J(f )) · f s , with

∂f ∂f
J(f ) the jacobian ideal of f generated by ∂x 1
, · · · , ∂x n
.
1.3 A first list of examples

• When f is smooth we have bf (s) = s + 1. This can be seen easily
by reducing the calculation to the case f = x1 . The converse is true:
the equality bf (s) = s + 1 may only happen in the smooth case. The
result can be found in [10] by Briançon and Maisonobe.
www.pdfgrip.com
232
• When f = xα αn
1 · · · xn we obtain by a straightforward calculation:
1
n αn
1 Y ∂ α1 ∂
Qn αi ··· · f s+1
i=1 αi i=1 ∂x1 ∂ xn
Yn i −1
αY !
k
= s+1− · fs
i=1
α i
k=0
It is an easy exercise to prove that this equation does yield the minimal
polynomial for a monomial.
• Let f = x21 +· · ·+x2n and let ∆ be the Laplacian operator then there is
a functional equation leading to the polynomial b(s) = (s + 1)(s + n2 ).
To be precise:
∆f s+1 = (s + 1)(4s + 2n) · f s
The minimality of the polynomial is not so obvious but can be deduced
from calculations in the section 4 below. This is an example of a semi
invariant for the action on Cn of the complex orthogonal group which
is reductive.
• Finding by blind calculatory means a functional equation in more
general cases is virtually impossible. The case of f = x2 + y 3 is
already challenging. In section 4 we shall treat the case of all quasi-
homogeneous singularities which includes all the Pham-Brieskorn poly-
nomials
xa1 1 + · · · + xann
1.4 Remarks on variants of the definition

Let f ∈ K[x1 , · · · , xn ] be a polynomial with coefficients in K. For a given
a ∈ K n we may consider various Bernstein-Sato polynomials:
1) The usual Bernstein-Sato polynomial bf = balg , such that:
bf (s)f s ∈ An (K)[s] · f s+1
2) The local algebraic Bernstein-Sato polynomial bloc , such that:
bloc,a (s)f s ∈ An (C)ma [s] · f s+1
with a functional equation having its coefficients in the algebraic local ring
at a.
3) If K = C we may consider the local analytic Bernstein-Sato polyno-
mial ban at a, characterised by
ban,a (s)f s ∈ DCn ,a [s] · f s+1
www.pdfgrip.com
233
Obviously we have the divisibility relations ban,a |bloc,a , and bloc,a |balg ,
but there is in in fact a much more precise relation
Proposition 1.2
ban,a = bloc,a , bf = lcmn (bloc,a ) when K is algebraically closed

a∈K
The proof of these results may be found in a more general setting in [11], in-
cluding the case of multivariables Bernstein-Sato polynomials b(s1 , · · ·, sp ).
We shall give the proof of proposition 1.2 in section 5.3.
Remark 1.3 If we consider the ring of differential operators with coef-

ficients in a ring of analytic functions O(U ) on an open set in Cn this
proposition 1.2 is no longer valid. It is not even true that for f ∈ O(U )
there is always a functional equation globally written on U .
2 On the existence of a functional equation in the poly-

nomial case
In this section we recall the proof by Bernstein of the existence of the
functional equation in the polynomial case. I put off giving a sketch of
the proof in the analytic case which requires more sophisticated tools, to
the last part of these notes. This section also covered in the course of
F. Castro, is inserted here for the sake of completeness and because it is
a natural and basic question. I also give a first approach in the algebraic
frame to fundamental notions (dimension, multiplicity, holonomicity) which
will be reformulated with more sophisticated tools in the analytic case.
2.1 Holonomic modules

We consider the Bernstein filtration of An (K):
 
 X 
Γk (An (K)) = aα,β xα ∂ β
 n

(α,β∈N ),|α|+|β|≤k
Let us first recall general results about holonomic modules on the Weyl
algebra An (K):
Theorem 2.1 1) Let M be a finitely generated module on the Weyl algebra

An (K), endowed with a good filtration F• (M ). Then there is a polynomial
χ(Γ, M ), such that for large k:
dim Fk (M ) = χ(Γ, M )(k)

www.pdfgrip.com
234
2) The leading term of this polynomial is independent of the choice of a

d
filtration. If we denote it e(M ) kd! , then the coefficient e(M ) is an integer
called the multiplicity of M , and the degree d is called the dimension of M .
Proposition 2.2 (basic properties) Let M be a finitely generated mod-
ule on the Weyl algebra An (K) and N a sub-module then we have:
d(M ) ≤ max(d(N ), d(M/N ))
if d(N ) = d(M/N ) then e(M ) = e(M ) + e(M/N )
Theorem 2.3 (Bernstein inequality) For any finitely generated module
we have the inequalities:
n ≤ d(M ) ≤ 2n
Proposition-definition 2.4 A module M such that d(M ) = n is called
holonomic if d(M ) = n. A holonomic module has finite length.
Let us end this section by a characterization of finitely generated mod-
ules which will be useful in the next section.
Lemma 2.5 Let M be an An (K)-module with a filtration F compatible
with the Bernstein filtration on An (K) and such that for some constants
c1 > 0 and c2 > 0 and any j ∈ N
kn
dim F k (M ) ≤ c1 + c2 (k + 1)n−1
n!
then M is finitely generated and holonomic with multiplicity e(M ) ≤ c 1 .
Proof Assume first that M is finitely generated: By the hypothesis

on M we may choose a good filtration Ω. The fact that Ω is good implies
the existence of an integer q such that for all k ∈ N:
Ωk (M ) ⊂ Fk+q (M )
n
In particular χ(Ω, M )(k) ≤ c1 (k+q)
n! + c2 (k + q + 1)n−1 , and the degree
of χ is at most n, so that d(M ) = n and M has the minimal dimension.
Looking at the leading term we also see that e(M ) ≤ c1 .
Reduction to the finite case: Let N ⊂ M be a finitely generated
submodule. By applying part 1 to N with the filtration F• ∩ N we see that
N is holonomic of multiplicity ≤ c1 . This implies because of 2.4 that any
strictly ascending sequence:
0 6= N1 ⊂ N2 ⊂ · · · ⊂ Nr
of submodules of M has length r ≤ c1 , so that M must be finitely generated.
And we are reduced to part 1.
www.pdfgrip.com
235
2.2 Bernstein equation

Theorem 2.6 Let f ∈ K[x1 , · · · , xn ] be a non zero polynomial.There is a
non zero polynomial b ∈ K[s] in one indeterminate s, and a differential
operator P (s) ∈ An (K)[s] such that
P (s)f s+1 = b(s)f s
Proof We work in K[x1 , · · · , xn , f1 ], given with its natural structure of a

module over the algebra An (K), and also with the modules K[s][x1 , · · · , xn , f1 ]f s
and K(s)[x1 , · · · , xn , f1 ]f s seen as modules over the algebras An (K)[s] and
An (K)(s). We remark that An (K)[s] is not a Weyl algebra over a field but
that An (K)(s) is the Weyl algebra for the field K(s) so that we can apply
dimension theory to this field as well.
Lemma 2.7 The module M = K[x1 , · · · , xn , f1 ] is a holonomic module

over An (K).
Let N be the total degree of the polynomial f . We define a filtration on

M:
g(x)
Fk (M ) = { | deg g ≤ k(N + 1)}
f (x)k
The space Fk (M ) is isomorphic to the space of all polynomials of degree
≤ k(N + 1) in n variables. Thus:

k(N + 1) + n
dim Fk (M ) =
n
and looking at this expression as a polynomial in k with a constant c2

depending on N and n we obtain:
kn
dim Fk (M ) ≤ (N + 1)n + c2 (k + 1)n−1
n!
It remains to be noticedS that Fk (M ) is a filtration of M as an An (K)-
module and that M = k∈Z Fk (M ). The conclusion of the lemma follows
after 2.5, moreover with the inequality:
e(M ) ≤ (deg f + 1)n
Lemma 2.8 The module Mf = K[x1 , · · · , xn , f1 ](s)f s is a holonomic mod-

ule over An (K(s)).
www.pdfgrip.com
236
The proof is similar except for the fact that the filtrations and dimensions
concern vector spaces over the ring of fractions K(s). We set
g(x, s) s
Fk (Mf ) = { f | degx g ≤ k(N + 1)}
f (x)k
In order to check that we obtain a filtration, assume that degx g ≤ k(N +1).
Then we have
∂g ∂f
∂ g(x, s) s ∂xi ∂xi s
· f = + g f
∂xi f (x)k fk f k+1
∂g
and the verification to be made is just that deg(f · ∂x i
+ g · ∂∂f
xi
) ≤ (k +
1)(N + 1)
The proof of the existence of a functional equation can now be ended
in the following way: The module Mf contains the descending sequence of
submodules
Mf ⊃ An (K(s)) · f · f s ⊃ · · · ⊃ An (K(s)) · f m · f s ⊃ · · ·
Since the module Mf is holonomic hence of finite length this sequence is

stationary. There is an integer m such that An (K(s)) · f m · f s = An (K(s)) ·
f m+1 · f s or equivalently:
f m · f s ∈ An (K(s)) · f m+1 · f s
This means the existence of a differential operator P1 (s) with coefficients

in K(s) such that
f m · f s = P1 (s) · f m+1 · f s (3)
We may also notice that the map An (K(s)) · f s −→ An (K(s)) · f m · f s ,
given by
P (s)f s → P (s + m) · f m · f s
is an isomorphism of An (K)-modules, its inverse being Q(s) · f s+m −→
Q(s − m)f s . We convert the equation (3) into an equation f s = P2 (s) · f f s ,
(s)
with P2 (s) = P1 (s − m), and setting P2 (s) = Pb(s) with P now having
polynomial coefficients, we obtain the desired functional equation.
2.3 Generalisations
Theorem 2.9 Let f1 , · · · , fp ∈ K[x1 , · · · , xn ] be p non zero polynomials.
There is a non zero polynomial b ∈ K[s] depending on the p variables
s = (s1 , · · · , sp ), and a differential operator P (s) ∈ An (K)[s1 , · · · , sp ] such
that
P (s)f1s1 +1 · · · fpsp +1 = b(s)f1s1 · · · fpsp
www.pdfgrip.com
237
These functional equations were considered by Claude Sabbah in [36]

in the analytic context. The proof for the polynomial case is a simple
adaptation of the proof in the previous section, while the proof in the
analytic case is much more intricate. For the proof in the analytic case, see
the forthcoming section 5 in the case p = 1 and for p ≥ 2 see the remarks
at the end of section 1.1.
Here is a proof in the algebraic case: Let us consider the following
module over the Weyl algebra An (K)(s1 , · · · , sp ) of the field of fractions
K(s1 , · · · , sp ):
1
Mf1 ,··· ,fp = An (K(s1 , · · · , sp ))[x1 , · · · , xn , ] · f1s1 · · · fpsp
f1 · · · fp
The action of the derivative is given is a way similar to the one dimensional
case at the beginning of the proof of theorem 2.6. We use the abbreviations
s s +1
s = (s1 , · · ·, sp ), f s = f1s1 · · ·fp p , f s+1 = f1s1 +1 · · ·fp p and f1 = f1 ···f
1
p
.
p
X ∂fk
∂ ∂g
· g(x, s)f s = +g sk ∂xi · f s
∂xi ∂xi fk
k=1
Let M be a holonomic An (K)-module and let u ∈ M . We are also

interested in equations of the following type:
P (s)uf s+1 = b(s)uf s
This is an equation in the module M ⊗K[x] K[s][x1 , · · · , xn , f1 ]f s which is
a module over An (K)[s], with the natural action of the partial derivatives
given by:
∂ ∂ ∂
(u ⊗ g(x, s)f s ) = ( · u) ⊗ g(x, s)f s + u ⊗ · g(x, s)f s
∂xi ∂xi ∂xi
Theorem 2.10 Let f ∈ K[x1 , · · · , xn ] be a non zero polynomial, and let u
be an element in a holonomic An (K)-module. Then there is a polynomial
b ∈ K[s], and a differential operator P (s) ∈ An (K)[s] such that
P (s) · (u ⊗ f s+1 ) = b(s)u ⊗ f s
The proof is quite similar to the previous ones. We consider a filtration
Fk (M ), and the good filtration of K(s)[x1 , · · · , xn , f1 ]f s already considered
in the previous section. We shall work in the An (K(s))-module:
1
K(s) ⊗K[s] (M ⊗K[x] K[s][x1 , · · · , xn , ]f s )
f
1
= M (s) ⊗K(s)[x] K(s)[x1 , · · · , xn , ]f s
f
www.pdfgrip.com
238
where M (s) = M ⊗K K(s) has the obviously good filtration Fk (M )⊗K K(s).
Then we see easily that the filtration:
1 s X 1
Fk (M (s) ⊗ K(s)[x, ]f ) = Fp (M (s)) ⊗ Fq (K(s)[x, ]f s )
f f
p+q=k
is good and that its Hilbert polynomial χ is related to the Hilbert polyno-
mials χ1 and χ2 of M and K(s)[x, f1 ]f s by the inequality:
X
χ(k) ≤ χ1 (p)χ2 (q)
p+q=k
from which the result follows: the module M (s)⊗K(s)[x, f1 ]f s is holonomic

and we just have to look now at the decreasing sequence of submodules:
h 1 i
An (K(s)) · u ⊗ K(s)[x, ]f m f s
f
to obtain the conclusion in the same way as above.
Final remarks.- In the analytic case the simple algebraic proof devel-
opped above fails already in the standard case with p = 1 and u = 1 ∈ O.
The reason as explained in [32] is that there is not a simple way to involve
the field K(s), because the formation of rings of analytic operators with
series coefficients does not commute with base change K(s)⊗K as in the
algebraic case. In loc.cit. this failure is in a sense repaired in the case
p = 1, and a purely algebraic proof of the existence of a Bernstein-Sato
polynomial is given, but with much more sophisticated tools than in the
proof above. I am not aware of such a proof in the analytic multifunctions
case.
2.4 Semi-invariants of prehomogeneous spaces

We refer for the detailed definitions and the main properties of prehomo-
geneous spaces to the book of T. Kimura [23].
ρ
Recall that a prehomogeneous space is just an algebraic action G //
GL(V ) of an algebraic group G on a K-vector space V which admits a
Zariski open orbit U . A semi invariant is a rational function f ∈ K(V )
such that there exists a one dimensional representation or character χ :
G → K ? = GL(1) such that
∀x ∈ V, ∀g ∈ G, f (ρ(g)x) = χ(g)f (x)
On a prehomogeneous space the character χ determines the semi invariant

f , in particular a semi invariant is associated to the character 1 if and only
www.pdfgrip.com
239
if it is constant. The equations of the one codimensional components Si of

the complement V \ U of the open orbit, are irreducible homogeneous poly-
nomials and are semi invariants. In fact all the irreducible semi invariants
are of this type. See [23, Theorem 2.9.] for details.
Let us now focus on the case of a reductive complex algebraic group.
One shows that such a group is the Zariski closure of a compact subgroup
H that we may assume included in the unitary group U (n), after an appro-
priate change to coordinates (x) known as unitary. In that situation one
shows that the dual action
ρ? =t ρ−1
G // GL(V )
is a prehomogeneus vector space. One also shows by a straightforward

calculation in the dual coordinates (y) that if h ∈ H, then ρ? (h) = ρ(h) the
complex conjugate and that if f (x) is a semi invariant polynomial associated
with a character χ, f ? (y) = f (y) is a semi invariant associated with χ−1 .
Proposition 2.11 In the above situation with homogeneous polynomial
semi invariants f, f ? of degree d, there exists a non zero polynomial b(s) of
degree d such that
∂ ∂
f ? (Dx )f (x)s+1 = b(s)f (x)s Dx = transpose( ,··· , )
∂x1 ∂xn
P ∂
Proof The chain rule ∂y ∂u
i
(g ·x) = j g −1 j,i ∂x j
(u(g ·x)) yields the usual
formula for base change on vectors in accordance with the base change
y = g.x on coordinates:
 ∂   ∂ 
∂y1 ∂x1
 ..  t −1  .. 
 . = g  .  (4)
∂ ∂
∂yn ∂xn
Let us condense formula (4) into

∂
Dy = t g −1 Dx = ρ? (g)Dx or componentwise : = (t g −1 Dx )i
∂yi
Kimura writes Dρ(g)x for the more explicit ρ? (g)Dx which is just a col-
∂
umn of differential operators which are linear combinations of the ∂x j
and
f ? (Dρ(g)x ), is just f ? (ρ? (g)Dx ). In a similar way we associate to any poly-
nomial differential operator its image by g −1
X ∂ i1 ∂ in
P (Dy ) = CI ( ) ···( )
∂y1 ∂yn
I=(i1 ,··· ,in )∈N
www.pdfgrip.com
240
X
P (Dρ(g)x ) = cI (ρ? (g)Dx )i11 · · · (ρ? (g)Dx )inn
I=(i1 ,··· ,in )∈N
obtained by applying the substitution (4) into f ? (Dy ).

Now the chainrule formula reads:
∂u X ∂
◦ ρ(g) = g −1 j,i (u ◦ ρ(g)) = ρ? (g)Dx i (u ◦ ρ(g))
∂yi j
∂x j
This reformulation allows us by a straighforward induction on the degree

of P to write:
P (Dy )(u) ◦ ρ(g) = P (Dρ(g)x )(u ◦ ρ(g))
therefore in particular:
1 ?
f ? (Dy )(u) ◦ ρ(g) = f ? (ρ? (g)Dx )(u ◦ ρ(g)) = f (Dx )(u ◦ ρ(g)) (5)
χ(g)
If u(x) is a semi invariant satisfying u(ρ(g)x) = χu (g)u(x) we get from the

formula (5) and using the k-linearity of f ? (Dx ):
∂ 1 ? χu (g) ?
f ?( )(u)(ρ(g)(x)) = f (Dx )(χu (g)u(x)) = f (∂x )(u)(x)
∂y χ(g) χ(g)
In particular applied to u = f (y)s+1 , we obtain
∂ χ(g)s+1 ?
f ?( )(f (y)s+1 ) ◦ ρ(g) = f (∂x )(f (x)s+1 )
∂y χ(g)
so that
1
? ∂ s+1 1
f ( )(f (y) ) = f ? (∂x )(f (x)s+1 )
f (y)s ∂y |y=ρ(g)x f (x) s
? s+1
This proves that x → f (∂xf)(f (x)
(x)s
)
is an absolute invariant hence a con-
stant b(s) depending on s. The fact that b(s) is a polynomial is clear and
we refer to [23] for the statement on its degree.
We refer the interested reader to [35] and its bibliography for more
informations on these b-functions as well as for the method of calculation
of bf in the case of an irreducible singular locus S by microlocal calculus
and holonomy diagramm. See also recent work about reducible examples
involving linear free divisor, in [16], [17].
www.pdfgrip.com
241
3 Bernstein-Sato polynomials and analytic

continuation of f s
3.1 Roots of b and analytic continuation of Y (f )f s

In this section, we assume that the fonction f is real on an open subset U
of Rn :
f : Rn → R
and admits a global Bernstein-Sato polynomial denoted by b. We can al-
ways assume this is true on a small enough neighbourhood of each point.
We shall see after the next proposition how to deal with a function for
which we drop this restriction.
For <s > 0 let us define the locally integrable function Y (f )f s or f+
s
(
exp(s log f ) if f (x) > 0
Y (f )f s =
0 if f (x) ≤ 0
Proposition 3.1 As a distribution Y (f )f s admits an analytic continua-

tion with poles on the set:
A − N := {s ∈ C | ∃i ∈ N, b(s + i) = 0}
where A is the set of zeros of b.
Proof Recall that a distribution is defined by its value for any test func-
tion ϕ ∈ Cc∞ with a compact support in the domain of f :
Z
hY (f )f s , ϕi = ϕ(x)Y (f )f s dx
Rn
convergent and holomorphic for <s > 0.

We may write for all s with <s > 0:
Z Z
b(s) ϕ(x)Y (f )f s dx = ϕ(x)b(s)f s dx
Rn f >o
Z
= ϕ(x)Y (f )P (s)f s+1 dx
Rn
Z
= P (s)? (ϕ(x))Y (f )f s+1 dx (6)
Rn
The last equality requires a bit more explanation:

www.pdfgrip.com
242
The operator P (s)? is the adjoint operator of P (s) defined by:

X ∂ β
P (s) = aβ (x, s)
∂xi
X ∂ β
P (s)? = (−1)|β| aβ (x, s)
∂xi
We first prove the last equality in (6) for <s large enough. Using the
fact that if <s ≥ ord(P (s)) := d the function Y (f )f s+1 is of class C d at
least, we have
Y (f )P (s)f s+1 = P (s)(Y (f )f s+1 )
since both members are zero along the hypersurface f −1 (0). With each
monomial in P (s) we perform a succession of integrations by parts in which
the integrated term is zero. By adding up the equalities obtained in this
way equation (6) follows.
Now (6) is valid for any s with <s > 0 by analytic continuation.
We have obtained the following equality:
1
hY (f )f s , ϕi = b(s) hY (f )f s+1 , P (s)? (ϕ(x))i
Since the right hand side is well defined and holomorphic in the open
set:
{s | <s > −1} \ b−1 (0),
we have an analytic continuation of
Z
hY (f )f s , ϕi = ϕ(x)Y (f )f s dx,
Rn
as a meromorphic function defined on the half-plane {s | <s > −1} with

poles included in the zeros of b.
Iterating this process we obtain similarly:
Z
1
< Y (f )f s , ϕ > = P (s + p − 1)?
b(s + p − 1) · · · b(s) Rn
· · · P (s)? (ϕ(x))Y (f )f s+p dx
Since the right-hand side of this equation has a meaning as soon as

<s > −p, we obtain a meromorphic continuation of the distribution Y (f )f s
to the half plane <s > −p, in which a necessary condition for s to be a pole
is
b(s + i) = 0, for some i ∈ {0, · · · , p − 1}
www.pdfgrip.com
243
Doing this for all p gives the result.

If we consider an arbitrary analytic function defined on an open subset
of Rn we can no longer assume the existence of a global Bernstein-Sato poly-
nomial, but we can assume it locally i.e. on each element of a sufficiently
fine open covering of U . Finally we obtain:
Corollary 3.2 Let f : U → R a real analytic function defined on an open

set U ⊂ Rn . Then the map s → Y (f )f s considered as a distribution depend-
ing holomorphically on s in the half-plane <s > 0, admits an meromorphic
continuation to the whole plane C. The precise meaning is that its restric-
tion to any compact set K ⊂ U admits a discrete set of poles (depending
on K) contained in a set of the form:
A − N := {s ∈ C | ∃i ∈ N, s + i ∈ A}
for some finite subset A ⊂ Q<0 .
N.B. In the last statement we anticipate the fact that the roots of the
Bernstein-Sato polynomial of an analytic germ are negative rational num-
bers. See section 6.
3.2 Asymptotic expansion and Mellin transforms

In this section we wish to show another aspect of the meromorphic continu-
ation of f s that we have just obtained: its interpretation as the continuation
of a Mellin transform.
Assume that f has no singular point outside of f −1 (0). Then there is a
unique differential (n − 1)-form ω defined on the open set f > 0 such that
dx = df ∧ ω.
Using Fubini theorem we have:

Z Z Z ∞ Z
s s s
f ϕdx = f ϕdf ∧ ω = t dt ϕω|f =t ,
f >0 f >0 0 f =t
In this way we describe the function s → hY (f )f s , ϕi as the Mellin

transform ΓF (s + 1) 1 of the following function in which we denote
1 We may also notice that the stationnary phase integrals are Fourier transforms of
the same function by:

Z Z +∞ Z
dx
eiτ f ϕdx = eiτ t dt( ϕ )
−∞ f =t df
www.pdfgrip.com
244
dx
df := ω|f =t :
Z
dx
F (t) = ϕ
f =t df
The definition of the Mellin transform is:
Z +∞
ΓF (s) = ts−1 F (t)dt
0
The function F (t) has an asymptotic expansion with terms of the form
aα,q tα (log)k . A precise version of this result can be found in Jeanquartier
[18]:
Let f : X → R be a real analytic function, on a manifold X. The Dirac
distribution δt (f ) is defined setting Xt = f −1 (t) by the following formula:
Z
hδt (f ), ϕdxi = ω.
Xt
where ω is a n − 1 form such that ϕdx = df ∧ ω in a neighbourhood of

supp(ϕ) ∩ Xt
Théorème 3.3 For any relatively compact open subset U ⊂ X of X, there

is a positive integer q and distributions Aj,k ∈ D0 (U ) such that δt (f ) has
j−q
in the scale of functions t q (log t)k an asymptotic expansion:
X j−q
δt (f ) ∼ Aj,k t q (log t)k
t→0,t>0
j=1,2,··· ; 0≤k≤n−1
At the level of usual functions F (t) = hδt (f ), ϕdxi has an asymptotic

expansion:
X j−q
F (t) ∼ aj,k t q (log t)k ,
t→0,t>0
j=1,2,··· ; 0≤k≤n−1
We are mainly interested in the case X = Rn , but even in this case the
statement in [18] with an arbitrary variety X is necessary. This variety
may even be a non-oriented one and the test function becomes a n-form of
odd type. This is already hidden in the case of Rn , because the proof in
[18], see also [19, Proposition 4.4.] uses the resolution of singularities.
We shall not give a detailed proof of this theorem but just make the cal-
culation for the crucial step which concerns the function f (x) = xk11 · · · xk` `
on the hypercube {x ∈ Rn | 0 < xi ≤ a, i = 1 . . . n}. Using a comple-
tion argument one shows that it is sufficient to consider test functions with
separated variables i.e. of the type g(x) = g1 (x1 ) · · · g` (x` ).
www.pdfgrip.com
245
We have
x1 df
dx = dx1 ∧ · · · ∧ dxn = ∧ dx2 ∧ · · · ∧ dxn
k1 f
Z
x1
F (t) = g1 (x1 ) · · · g` (x` ) dx2 ∧ · · · ∧ dxn
f (x)=t k1 f
Z t k1
F (t) = g1 1 g2 (x2 )
0<xi ≤a;1≤i≤n xk22 · · · xk` `
t
k1
1
k k
x2 2 ···x` `
· · · g` (x` ) dx2 ∧ · · · ∧ dxn .
k1 t
Z
1 k1 −1 t k1
F (t) = t 1 g1 k`
1 g2 (x2 )
k1 0<xi ≤a;1≤i≤n xk22 · · · x`
1 k1
· · · g` (x` ) 1 dx2 ∧ · · · ∧ dxn .
xk22 · · · xk` `
We make an induction on `. The case ` = 2 which already gives a precise

idea of the general case is written as follows2 :
Z t 1
1 k1 −1 a 1
F (t) = t 1 k1 g1 k
k1
g2 (x2 ) k2 dx2 .
k1 t 2 x2 2
ak1 x2k1
From this we derive an asymptotic expansion with exponents of the type

−1 + kj11 + kj22 , with jp > 0 and possibly degree-one logarithmic terms.
The asymptotic expansion in the complex domain for integrals on the
fiber f −1 (t) of an holomorphic map, has been intensively studied by Daniel
Barlet and also by Henri Maire. We will not consider this question here but
refer to [4] for an existence theorem, and also to further papers, in which
D. Barlet studied a number of consequence for the poles of f λ .
3.3 Application to analytic continuation

We shall see in the appendix 8.1 on Mellin transforms that we can derive
from the asymptotic expansion of the distribution t → δt f a meromorphic
2 Notice also the easiest case of one variable in which the exponents −1 + j
k
already
clearly appear: Z Z
dx ϕ(x) 1 1
ϕ(x) k = k−1
= t−1+ k ϕ(t k )
k
x =t dx k
x =t kx
www.pdfgrip.com
246
continuation for hY (f )f s , ϕi. In fact we have:
hY (f )f s , ϕi = I(s) + J(s)
Z γ Z +∞
s
with I(s) = t F (t)dt ; J(s) = ts F (t)dt
0 γ
and J(s) is an entire function on C.

But conversely although there is a theory of an inverse Mellin transform
in the case of f (t) having compact support in ]0, +∞[, we cannot hope in
the situation of a locally integrable I, to recover an asymptotic expansion
at 0 from its Mellin transform.
The following example is noticed in [18]:
Example Let f (t) = sin( 1t ). The integral:
Z 1
I(s) = ts f (t)dt
0
which is a priori convergent for Res > −1 admits an analytic continua-

tion as an entire function on the whole complex plane: indeed a double
integration by part yields for <s > 2 a formula:
I(s − 2) = s(s + 1)I(s) + as + b (a = sin 1; b = cos 1)
which allows us to define I(s) by analytic continuation successively on each

strip < > −2 · · · < > −2p, as in the case of the usual function Γ, when we
use the identity Γ(z + 1) = zΓ(z). But in the present case case no pole at
all appears.
Atiyah proved in [2], as well as Bernstein and Gel’fand in [5], using this
type of method that the distribution Y (f )f s has an analytic meromorphic-
continuation in C, with poles of order at most n.
The details are explained in the appendix A, see section 8.1. The first
and easiest case (without multiple poles) of the method for finding mero-
morphic continuations out of aymptotic developments is explained in detail
in theorem 8.3. The case of higher order poles which come from terms
in the asymptotic expansion involving powers of log t is also treated more
succinctly by an argument of derivation.
Conclusion
We described in the two preceding sections two ways for obtaining mero-
morphic continuations of the distribution Y (f )f s .
• By proving the existence of an asymptotic expansion and using its
Mellin transform. We obtain that the poles are rational of order at
most the dimension of the ambient space.
www.pdfgrip.com
247
• By the existence of Bernstein-Sato polynomial. Here the precision is:

the poles are in A0 \ N, A0 the set of rational roots of the b function.
As we announced in corollary 3.2 and will prove in the forthcoming section

6 the roots of the b-function are in Q<0 . This gives another proof of the
rationality of the poles. Similarly the upper bound on the order of the poles
is a consequence of the fact that the multiplcity of the roots of b are at most
n. This result comes from the geometric interpretation of these roots as
exponents r of the eigenvalues of the monodromy e2πir . Ultimaltely the
result on the order of poles comes from the monodromy theorem which
states that these eigenvalues are roots of unity.
3.4 Application to the division of distribution

Proposition 3.4 Let f : X → C be a real analytic function on a real ana-
lytic manifold X, with f not identically zero on each connected component
of X. Then there is a distribution T such that T f = 1.
Proof Assume first that f ≥ 0. Then the distribution Y (f )f 0 is just the

constant 1.
Applying the theorem of meromorphic continuation we get a Laurent
expansion of f s around s = −1 as a distribution of the type
P
∞
fs = Ak (s + 1)k
−m
with Ak a distribution.
The expansion of f · f s = f s+1 has no pole at s = −1, its value at
s = −1 being f 0 = 1, This expansion is also
∞
X
f s+1 = f Ak (s + 1)k
−m
This implies the following relations:
f Ak = 0 for k < 0 and f A0 = 1
and the latter provides an inverse to f as a distribution, as intented. In the

general case we take an inverse T to |f |2 and the relation giving the result
is 1 = T |f |2 = T f · f
www.pdfgrip.com
248
4 Quasi-homogeneous and
semi-quasi-homogeneous isolated
singularities
This section is inpired by the preprint [7] in which we give an algorithm
for the calculation of the Bernstein-Sato polynomial non only for quasi-
homogeneous polynomials, but also in the more general situation of a
semi-quasi-homogeneous function with an isolated singularity. In [8] we
generalised this calculation to any function which is nondegenerate with
respect to its Newton polygon. We will give in detail below the proof in
the quasi-homogeneous case with hints for the extension to the semi-quasi-
homogeneous germs.
In all these examples we see by a direct calculation that the coefficients
of the monic Bernstein-Sato polynomial are in Q.
As a preliminary let us recall the example of a monomial in which we
see directly that the roots of bf (s) are rational.
When f = xa1 1 · · · xann := xa , bf divides the following product:
n
Y i −1
aY
k
Πa (s) = (s + 1 − )
i=1
ai
k=0
In fact we have the equality bf = Πa (s). Consider a functional equation

X
aα,β,j sj xα ∂ β xas+a = e(s)xas
α,β∈N,
By an identification of the terms having the same degree we can restrict

the differential operator to a linear combination of monomials of the type
sj xβ−a ∂ β .
This implies βi ≥ ai for all i such that aα,β,j 6= 0. Since any such β
yields ∂ β xas+a = eβ (s)xas with eβ (s) a multiple of Πa (s), the polynomial
Πa (s) divides e(s) as expected.
4.1 Quasi-homogeneous polynomials

We consider a system of weights w = (w1 , · · · , wn ) ∈ (Q?+ )n . and for I ∈ Nn
we denote hw, Ii = w1 I1 + · · · wn In
Definition 4.1 The polynomial f ∈ C[x1 , · · ·, xn ] is quasi-homogeneous of
weight (we say also w-degree) ρ if its expansion has the form:
X
f= f I xI
hw,Ii=ρ
www.pdfgrip.com
249
We denote C[X]ρ the finite dimensional space of quasi-homogeneous poly-

nomials of degree ρ
Definition 4.2 The function germ f defines an isolated singularity if {0}

is an isolated point in the set defined by the equations
∂f ∂f
= ··· = =0
∂x1 ∂xn
∂f ∂f
The ideal J(f ) = ( ∂x 1
, · · · , ∂x n
) generated by the partial derivatives is
called the Jacobian ideal.
Examples:
f (x, y) = xa + y b , f (x, y) = xa + xy b , are quasi-homogeneous isolated
singularities.
f (x, y) = x4 +y 5 +xy 4 , has a non-quasi-homogeneous isolated singularity
at (0, 0).
Basic example (Pham-Brieskorn polynomials.)
f (x1 , · · · , xn ) = xa1 1 + · · · + xann

1
This polynomial is quasi-homogeneous of weight 1 for the weights wi = ai .
The jacobian ideal J(f ) is generated by the monomials
1 0
f = xai i −1
a i xi
and the quotient is finite dimensional generated by the set of monomials:
M = {xk11 · · · xknn , with 0 ≤ ki ≤ ai − 2}
The main result proved in subsection 4.3 is:
Proposition 4.3 Let f be a quasi homogeneous polynomial with an isolated

singularity at (0, 0) and w-weight equal to 1. Let M be a monomial basis
1 ,···,xn ]
P
of the Jacobian quotient C[xJ(f ) . Let |w| = wi , and let Π be the set of
weights without repetition of elements of M . Then:
Y
bf (s) = (s + 1) (s + |w| + ρ)
ρ∈Π
We have the following elementary properties:

O
• The hypothesis of isolated singularity implies that the quotient J(f ) is
finite-dimensional. Therefore M is finite as implicitly stated in propo-
sition 4.3.
www.pdfgrip.com
250
• Let us denote by Eρ a supplementary subspace to J(f ) ∩ C[x]ρ in

C[x]ρ , generated by monomials. Then
σ
M
E= Eρ
0
O
is a set of representative of J(f ) .
• This reflects a grading:

M σ
O C[x]ρ
= .
J(f ) 0
J(f ) ∩ C[x]ρ
• The dimension of Eρ depends only on the weights w1 , · · · , wn if we

normalize the weight of f as being 1.
• Let σ be the maximum of the weights ρ such that Eρ 6= 0 or equiva-
lently such that O>ρ := {g ∈ O|ρ(g) > ρ} is contained in J(f ). It is
well known that σ is equal to the weight of the Hessian of f :
∂2f Xn
σ = ρ det( ) =n−2 wi
∂xi ∂xj i=1
2 2
In the Pham-Brieskorn case σ = (1 − a1 ) + · · · + (1 − an )
• We also have by a theorem of Milnor-Orlik [31]:

Yn
1
µ = dim(O/J(f )) = ( − 1)
i=1
wi
To finish these preliminaries let us quote a very simple division lemma:

Lemma 4.4 For any u ∈ O quasi-homogeneous of degree ρ, then there are
a unique v ∈ Eρ and (λ1 , · · · , λn ) ∈ On such that
n
X ∂f
u=v+ λi
i=1
∂xi
with the following conditions (using the convention that ρ(0) = +∞):
ρ(v) = ρ(u) or v = 0
ρ(λi ) = ρ(u) − 1 + wi or λi = 0
www.pdfgrip.com
251
4.2 Semi-quasi-homogeneous germs

In this section we mention briefly the notion of a semi-quasi-
homogeneous germ for which a generalisation of proposition (4.3) for the
calculation of the Bernstein-Sato polynomial can be derived. Given the
system
P of positive weights w1 , · · ·, wn let us define the weight of a series
f = fI xI ∈ O := C{x1 , · · ·, xn } as the rational number ρ = ρ(f ) :=
min{hw, Ii | I 6= 0}. We define the initial part inw f of f as the quasi
homogeneous polynomial, sum of its lowest degree terms:
X
inw f = f I xI
hw,Ii minimal
Definition 4.5 The function germ f is called semi-quasi-

homogeneous of weight ρ if its initial part in w f has an isolated singularity
at the origin.
Let us list again a number of elementary properties similar to those in
the quasi-homogeneous case:
• Thanks to the definition of initial parts the sequence ( ∂inf ∂x1 , · · · ,
∂inf
∂xn ), is a regular sequence, generating the Jacobian ideal of inf .
This implies that the initial ideal inw (J(f )) i.e. the ideal generated
by {inw g, g ∈ J(f ), }, is in fact the Jacobian ideal J(inf ).
O O
• This implies also that J(f ) and In(J(f )) are of the same finite dimen-
sion.
• Recall that according to subsection 4.1 there is a grading
C[x] L C[x]ρ
in(J(f )) = in(J(f ))∩C[x]ρ where Eρ is a supplementary subspace to
inw (J(f )) ∩ C[x]ρ in C[x]ρ . The vector space:
σ
M
E= Eρ
0
C[x]
is a set of representative of In(J(f )) .
• By the hypothesis of semi-quasi-homogeneity, E is also a set of rep-

O
resentatives of J(f ) . In fact we have a decreasing filtration of O
by Oρ the set of series of weight ≥ ρ, and an induced filtration
C[x]
C{x1 , · · ·, xn }ρ ∩ J(f ) on J(f ). The graded module in(J(f )) is just
the graded module associated with this filtration:
O gr(O) C[x]
gr = =
J(f ) gr(J(f )) in(J(f ))
www.pdfgrip.com
252
• The dimension of Eρ depends only on the weights w1 , · · · , wn as long

as we normalize the degree of inw f as being 1. This can be seen by
considering the graded Koszul complex associated with the regular
sequence (inw ∂∂f
x1
, · · · , inw ∂∂f
xn
). It is well known too that the weight
σ which is by definition the maximal weight such that Eρ 6= 0 is equal
to the weight of the Hessian of f so that:
∂2f Xn
σ = ρ det( ) =n−2 wi
∂xi ∂xj i=1
• We have by Milnor-Orlik:
Yn
1
µ = dim(O/J(f )) = ( − 1)
i=1
w i
The proof relies on the following simple division lemma in which we have
fixed a graded supplementary space E to inJ(f ) as above:
Lemma 4.6 For any u ∈ O there are unique v ∈ E and (λ1 , · · · , λn )

∈ On such that
Xn
∂f
u=v+ λi
i=1
∂x i
with the following conditions (using the convention that ρ(0) = +∞):
ρ(v) ≥ ρ(u) and ρ(λi ) ≥ ρ(u) − 1 + αi
Semi-quasi-homogeneity and µ-constant deformation.

Consider a semi-quasi-homogeneous germ with initial part f0 . We set
wi = kdi and assume k1 , · · ·, kn to be relatively prime, and f to have weight
1, i.e. degree d with respect to the weights ki . We may consider the
deformation of f0 :
f (x, t) = t−d f (tk1 x1 , · · ·, tkn xn )
This is a deformation of isolated singularities with a constant Milnor num-

ber. All the germs ft = f (•, t), for t 6= 0 are isomorphic, and therefore have
the same Bernstein-Sato polynomial bft = bf which is in general different
from the polynomial bf0 (s) calculated in proposition 4.3 and subsection 4.3.
According to the result of Lê Duñg Tráng and C.P. Ramanujan, see [25],
this situation implies for n 6= 3 at least that the topological type of the
germ ft is constant. It also implies that the monodromies of f and f0 are
www.pdfgrip.com
253
conjugate, see [24]. We shall see in section 6 that this allows a comparison
of the roots of the two Bernstein-Sato polynomials bf , bf0 , because the roots
of bf are deeply linked to the eigenvalues of the monodromy. By a result
due to B. Malgrange in the isolated case:
∃k ∈ Z | bf (r + k) = 0 ⇐⇒ λ = e2iπr
is an eigenvalue of the monodromy of f
Therefore the polynomial bf is a product of factors which are all of the

form (s + |w| + ρ + k), with ρ ∈ Π and k ∈ Z can be thought as a ”shift”
with respect to the quasi-homogeneous case. The number of factors may be
different but all the Bernstein-Sato polynomials have only simple roots and
in fact all the shifts are negative bounded by −n ≤ k ≤ 0. We refer to [7] for
a precise statement and an explicit algorithm for the calculation. See also
the published version [8] which deals directly with the non-degenerate case
(for which the fact that the roots have multiplicity one is lost in general).
In both papers the generic value of the Bernstein-Sato polynomial of a
deformation is calculated in the case of quasi-homogeneous curves.
4.3 Calculation of the Bernstein-Sato polynomial of a

quasi-homogeneous polynomial
In this section we assume that f is a quasi homogeneousP polynomial with
∂
isolated singularities and of w weight equal to 1. Let χ = wi xi ∂x i
, called
the Euler vector field. The quasi-homogeneity condition means that
χ(f ) = f
Let u be a quasi-homogeneous polynomial of weight ρ. We shall use

lemma 4.4 which is a w-homogeneous equality in C[x]ρ in order to carry out
first an easy calculation in the D-module D[s]f s . We obtain a polynomial
which is a priori a multiple of the b-function.
Pn
Lemma 4.7 Let u ∈ O be any function germ. We set |w| = i=1 wi .
Then, for any ρ ∈ Q:
n
X ∂
(s + |w| + ρ)uf s = wi · xi u + (ρ · u − χ(u)) f s
i=1
∂xi
It is just a matter of an elementary calculation noticing that

n
X ∂
wi · xi = χ + |w|, χ(uf s ) = suf s + χ(u)f s
i=1
∂xi
www.pdfgrip.com
254
n
X ∂
wi · xi uf s = (χ + |w|) · uf s = (χ(u) + (|w| + s)u)f s
i=1
∂xi
and reordering.
Let Π = ρ(E) = {ρ, Eρ 6= 0} be the set of weights of elements of E,
which is a finite subset of [0, σ] ∩ Q. When u ∈ E is a quasi-homogeneous
element of weight ρ = ρ(u), we can write χ(u) = ρ · u and by applying the
division lemma to each of the monomials xi u we get:
n n
X ∂ X ∂
(s + |w| + ρ)uf s = wi · xi u f s = · vi ( mod DJ(f ) · f s )
i=1
∂xi i=1
∂xi
where vi ∈ E is quasi -homogeneous of weight ρ + wi .

By iterating this process applied to all the vi or equivalently by a de-
scending induction on the weight of u we obtain an equality for a quasi
homogeneous u:
Y X ∂f s
(s + |α| + ρ)uf s = Ai f
∂xi
ρ≥ρ(u),ρ∈Π
involving operators Ai ∈ D. Notice moreover that

∂f s ∂
(s + 1) f = · f s+1
∂xi ∂xi
Putting all this together we obtain:
Y X ∂
(s + 1) (s + |w| + ρ)uf s = Ai · f s+1 ∈ D[s]f s+1
∂xi
ρ≥ρ(u),ρ∈Π
and in particular setting u = 1:

Y
(s + 1) (s + |w| + ρ)f s ∈ D[s]f s+1
ρ∈Π
Q
which proves the fact that (s + 1) ρ∈Π (s + |w| + ρ) is a multiple of the
Bernstein-Sato polynomial of f .
Theorem 4.8 Let f be a quasi homogeneous isolated singularity of weight

1 for the system of weight (w1 , · · · , wn ) and E a graded space representative
of the quotient O/J(f ). Then the Bernstein-Sato polynomial of f is equal
to: Y
bf (s) = (s + 1) (s + |w| + ρ)
ρ∈ρ(E)
www.pdfgrip.com
255
Proof Let b(s) = (s + 1)b̃(s) be the Bernstein-Sato polynomial of f . We

need to prove that b̃(−|w| − ρ) = 0, for all the weight in Π. Let u ∈ E
be a monomial of weight ρ = ρ(u). Using the functional equation and
multiplying it by u on one side and using the lemma 4.7 on the other side
we obtain two equations:
n
X ∂
(s + |w| + ρ)uf s = wi · xi u f s
i=1
∂xi
Xn
∂f s
b̃(s)uf s = Q(s)f + Ai f
i=1
∂xi
If b̃(−|w| − ρ) 6= 0, then the polynomials s + |w| + ρ and b̃(s) would be

relatively prime so that we would have a Bézout identity c1 (s)(s + |w| +
ρ) + c2 (s)b̃(s) = 1 and therefore:
n
X X n
s
∂ ∂f s
uf = c1 (s)( αi · xi u) + c2 (s)(Q(s)f + Ai ) f
i=1
∂x i i=1
∂x i
Using the fact that sf s = χ(f s ) and that f ∈ J(f ) we finally obtain an
identity:
Xn n
X
∂f ∂
uf s = Bi · fs + · Ci f s
∂xi ∂xi
i=1 i=1
with Bi , Ci ∈ D. This implies the fact that the operator

n
X X ∂n
∂f
Q=u− Bi − · Ci .
i=1
∂xi i=1 ∂xi
is in the annihilator of f . But we know that the annihilator of f s in D

s
∂ ∂f ∂ ∂f
is generated by the operators ∂x i ∂xj
− ∂x j ∂xi
which would imply Q ∈
P I
I∈N ∂ · J(f ) and hence also u ∈ J(f ) which is a contradiction.
In the last argument we used the following proposition:
Proposition 4.9 If f is a germ of an isolated singularity the annihilator
of f s in D (not in D[s]!) is generated by the following operators:
∂ ∂f ∂ ∂f ∂f ∂ ∂f ∂
− = −
∂xi ∂xj ∂xj ∂xi ∂xi ∂xj ∂xj ∂xi
∂f ∂f
This result is a consequence of the fact that ( ∂x 1
, · · ·, ∂x n
) is a regular
sequence. For the sake of completeness and of accessibility we shall give a
detailed proof of this result in appendix B, in section 8.2.
www.pdfgrip.com
256
5 Proof of the existence of the b-function for a an an-

alytic germ
The existence theorem announced in this section is much more difficult to
prove than its algebraic counterpart. We already explained at the end of
section 2.3 the difficulty of the purely algebraic proof given in [32]. We
are going now to give the basic step of a proof for analytic germs which
uses tools from sheaf theory, coherent D-modules and their characteristic
variety. The original source for this proof can be found in the paper of M.
Kashiwara [21], see also [22] and these notes are inspired by the version of it
that we gave in [15, Chapter VI]. We shall not develop these tools in detail
here in order to focus on the plan of the proof itself, but refer to section
7 for an abridged version, and to [15] for more details about the theory
of analytic D-modules, especially for the notions of characteristic variety,
multiplicities, dimension, and holonomic modules.
5.1 The proof in the analytic case

Let us first enumerate a number of basic ingredients for the proof of the
existence of a functional equation
1. The following theorem will be accepted without proof in these notes,

and we refer to [15]:
Theorem 5.1 Let M be a coherent D-module with c =

codimM. Then there exists a filtration by coherent D-
submodules:
M = Fc M ⊃ Fc+1 M ⊃ · · · ⊃ Fn M
such that Fk (M) = {m ∈ M/codimDm ≥ k} is the maximal subsheaf
of codimension k of M.
The fact that the maximal (n − k)-dimensional coherent subsheaf

of M is, if it exists, given by the condition codimDm ≥ k on the
germ of section m, and that this condition defines a subsheaf of D-
module can be seen at once. The point is to prove that the so-defined
subsheaf is indeed D-coherent. This extra condition contains in fact
an essential difficulty, and its proof requires an analysis of the bid-
uality spectral sequence
of M, attached to the derived functors of
Hom Hom(•, D), D see [15, sections V.5., V.6.]
2. Proposition 5.2 Let M be a germ of an holonomic module and ϕ :

M −→ M a D-linear map. Then ϕ admits a minimal polynomial.
www.pdfgrip.com
257
This means that there exists a polynomial b ∈ C[s] such that b(ϕ) = 0.
The proof is by induction on the number of components (of dimension
n) of the characteristic variety of M. We postpone to section 7 an
exposition of the necessary material on analytic D-modules and a
sketch of the proof. See also [15, Proposition 23].
Corollary 5.3 Let f ∈ O be an analytic germ. Then if the D-

D[s]f s
module M := D[s]f s+1 is holonomic, the germ f admits a nontrivial
Bernstein-Sato polynomial.
Indeed the map s : M → M given by the multiplication by s is D-

linear and the existence of a functional equation associated with e(s)
is equivalent to e(s)(M) = 0.
3. Let M be a D-module, not necessary a coherent one. Then M is

coherent if and only if it admits locally good filtrations.
Theorem 5.4 For any f ∈ O there is a nontrivial functional equation.
Proof We start with a particular case:

Case 1: f is Euler-homogeneous. We assume here that f is an element
of its Jacobian
P ideal which is equivalent to the fact that there is a vector
∂
field χ = ai (x) ∂x i
such that χ(f ) = f.
The key step in the proof of the theorem is the following lemma:
Lemma 5.5 Under the hypothesis of case 1, the module N := D[s] · f s is

coherent and of dimension n+1. Such a module is said to be subholonomic.
Proof of the lemma: Because of the existence of the vector field χ, we

may write:
sf s = sχ(f )f s−1 = χ(f s )
This shows that N = Df s is of finite type over D. In order to prove that it
is coherent we just have to exhibit a good filtration, for which a candidate
is Nk := D(k)f s , if we can prove that this module is O-coherent. But Nk
is clearly of finite type over O, generated by the finite set of all ∂ α · f s , for
|α| ≤ k, and in order to prove its coherence we just have to notice that it is
included as a submodule in a free and finite type hence coherent O-module
because we have: h X i
Nk ⊂ si f −j O f s
i+j≤2k
www.pdfgrip.com
258
Now let us consider according to the theorem 5.1 the maximal subholonomic
submodule Ñ ⊂ N . I claim that if x ∈ e and N
/ f −1 (0), then the germ of N
at x are equal. The reason is that the operator:
∂f ∂ ∂f ∂ ∂ ∂f ∂ ∂f
◦ − ◦ = ◦ − ◦
∂xi ∂xj ∂xj ∂xi ∂xj ∂xi ∂xi ∂xj
is in the annihilator of f s so that the germ at x of the characteristic variety
of N is contained in and in fact equal to the set defined by the equations:
∂f ∂f
ξj − ξi = 0 , 1 ≤ i < j ≤ n.
∂xi ∂xj
∂f
This set is smooth of dimension n + 1 because ∂xi0 (x) 6= 0 for some index
N
i0 . Now this proves that e
N
is a coherent D-module with support in f −1 (0).
Therefore by the nullstellensatz and since the section [f s ] belongs to an
O-coherent module, namely O · [f s ], we have f k · [f s ] = 0 which means that
f k+s ∈ Ne . Furthermore there is an isomorphism:
N −→ Df s+k : P f s → P f s+k
Therefore N is isomorphic to Df s+k which is isomorphic to a submodule

e . This implies that N is already subholonomic and that N = N
of N e . This
ends the proof of the lemma.
Under the hypothesis of case 1 we deduce from the lemma the exis-
tence of a nontrivial equation with the help of corollary 5.3 by proving the
holonomicity of M. This is done as follows. We have an exact sequence:
0 → N0 → N → M → 0
with N 0 = Df s+1 . Since N 0 and N are isomorphic, we have at each point

x? of the characterictic variety of M the following relations between the
local dimensions and multiplicities:
dx? (N ) = dx? (N 0 ) ex? (N ) = ex? (N 0 ) ,

hence dx? (M) < dx? (N ) = dx? (N 0 )
which implies dx? (M) = n. The module M is holonomic and therefore a

nontrivial functional equation exists.
General case. We set g(t, x) = (1 + t)f (x). The germ g is Euler
∂
homogeneous using the vector field χ = (1 + t) ∂t and we have by the case
1 an equation:
P (x, ∂x , t, ∂t )(s)[(1 + t)f ]s+1 = b(s)(1 + t)s f s

www.pdfgrip.com
259
We may order P with respect to the powers ∂t and we notice that

∂ s+1 1
g = (s + 1)g s+1
∂t 1+t
Therefore by descending induction on deg ∂ P (s) we may write another
∂t
∂
equation with an operator independent of ∂t
P (x, ∂x , t)(s)[(1 + t)f ]s+1 = b(s)(1 + t)s f s
and this leads directly to the desired equation by P (x, ∂x , 0)(s)f s+1 =
b(s)f s
5.2 An application: Holonomicity of the module O[ f1 ]

Theorem 5.6 The module O[ f1 ] is coherent as a module over D.
Proof Let b(s) be the Bernstein-Sato polynomial of f and let U be an

open set containing the origin in which there is a functional equation with
converging analytic coefficients. For any k ∈ N we have therefore an equa-
tion:
1 1
P (−k − 1) k = b(−k − 1) k+1
f f
Let k0 be an integer such that b(`) 6= 0 if ` < k0 . Then for any k ≥ k0 we
have:
1 1 1 1
= P (−k − 1) k ∈ D k
f k+1 b(−k − 1) f f
Finally this proves that O[ f1 ] = D · f 1k0 and O[ f1 ] is of finite type over DU
In order to obtain the coherence we now just need to produce a good fil-
tration of M = O[ f1 ]. The filtration by the O-submodules Mk = DU (k) f 1k0
works. The subsheaf Mk is indeed finitely generated over O hence coherent
1
being a submodule of the O-module OU f k+k 0
which is itself coherent being
free of rank one.
Theorem 5.7 The module O[ f1 ] is an holonomic D-module.
Proof Les us consider for a given λ ∈ C the following D-modules:

N
Nλ := , D · fλ
(s − λ)N
the second one being viewed as a submodule of O[ f1 ] · f λ with its obvious

action. There is a surjective map Nλ −→ D · f λ and therefore it is sufficient
www.pdfgrip.com
260
to prove that Nλ is holonomic because we have seen that for some integer
k0 this surjection becomes:
1 1
Nk0 −→ D · k
= O[ ]
f 0 f
Now the annihilator of uλ := f s (mod (s − λ)) contains the following ele-

ments:
∂f ∂ ∂f ∂ ∂ ∂f
− , f −λ
∂xi ∂xj ∂xj ∂xi ∂xi ∂xi
from which we derive that Nλ is holonomic outside f −1 (0).

Considering the maximal holonomic submodule of Nλ we prove that Nλ
is holonomic exactly in the same way as for the proof of the subholonomicity
of Df s in the lemma 5.5.
5.3 Links between various Bernstein-Sato

polynomials
We are going to prove the results announced in the introduction concerning
the link beween various types of global and local Bernstein-Sato polyno-
mials and deduce some properties. We anticipate in this section the result
about the rationality of the zeroes of the Bernstein-Sato polynomial which
is explained independently in next section 6.
Let f ∈ K[x1 , · · · , xn ] and let P ⊂ K[x1 , · · · , xn ] be a prime ideal. We
denote by bf,P the monic generator of the ideal consisting of polynomial e(s)
for which there is a functional equation P (s)f s+1 = b(s)f s with operators
having their coefficients in the local ring K[x1 , · · · , xn ]P . In other words
we allow denominators p(x) ∈ K[x1 , · · · , xn ] \ P.
Clearly if P1 ⊂ P2 , we have the divisibility relations bf,P1 bf,P2 bf .
Proposition 5.8 The global Bernstein-Sato polynomial bf of a non zero

polynomial f ∈ K[x1 , · · · , xn ] is equal to the lcm of all the bf,P when P
runs over all prime ideals. And in fact it is enough to take only maximal
ideals.
In particular if K is algebraically closed we have:
bf (s) = lcm{bloc,a | a ∈ K n}
where bloc,a = bf,Ma is the Bernstein-Sato polynomial attached to the max-

imal ideal at a ∈ K n
www.pdfgrip.com
261
Proof The divisibility relation
b(s) := lcm{bf,P |P maximal } bf (s)
is obvious and in view of the converse we notice that the set of polynomials
g(x), such that
g(x)b(s)f s ∈ An (K) · f s+1
is an ideal of I of K[x1 , · · · , xn ]. Since for any maximal ideal P we have
(s) s+1
a functional equation b(s)f s = Pg(x) f , with g ∈
/ P and P (s) ∈ An (K)[s]
we obtain g ∈ I \ P and therefore I * P. This result being true for any
maximal ideal we must have I = K[x1 , · · · , xn ] so that I contains 1 and
therefore bf b as required.
There is also a link with the analytic Bernstein-Sato polynomial in the
case of K = C.
Proposition 5.9 Let f ∈ C[x1 , · · · , xn ]. The local analytic Bernstein-Sato

polynomial ban,a at a ∈ Cn is equal to the local algebraic one:
ban,a = bf,Ma
We shall admit this result for which we refer to [11]. The proof relies on
the faithful flatness of OCN ,a over C[x1 , · · · , xn ]Ma
6 Rationality of the zeroes over any

algebraically closed field of characteristic zero
6.1 The rationality of the zeros

There is a deep result due to Malgrange [27] in the isolated case and to
Kashiwara [21] in general:
Theorem 6.1 The roots of the local analytic Berstein polynomial are neg-
ative rational numbers.
This results is linked to the monodromy theorem and had been tested
beforehand in various contexts.
- In the quasi-homogeneous case the rationality of the roots can be
checked directly as was first noticed by Kashiwara. We gave a proof and
an explicit calculation in section 4.
- In [27], B. Malgrange gave a partial proof in the case of isolated singu-
larities. He used the monodromy theorem which states that the eigenvalues
of the monodromy are roots of unity and he proved that the spectrum of
www.pdfgrip.com
262
the monodromy is exactly the set of all e2iπr when r runs over all the roots
of the b function. The statement is more precise in that b appears as the
minimal polynomial of the action of t∂t on a saturation of the Brieskorn
lattice. The negativity of the roots relies essentially on results in [28].
- The proof of M. Kashiwara in [21] is quite different and is based on the
resolution of singularities, the obvious case of a monomial and the theory
of direct images of D-modules.
- In [30] B. Malgrange gave a proof for the general case of non isolated
singularities in the spirit of [27]. This is the starting point of the theory of
V -filtrations due to Malgrange and Kashiwara.
6.2 Derived results

Proposition 6.2 Le K be an arbitrary field of characteristic zero and let
f ∈ K[x1 , · · · , xn ]. Then the roots of the Bernstein-Sato polynomial as
well as the roots of any local Bernstein-Sato polynomial b f,a (s) are negative
rational numbers.
Proof
Lemma 6.3 Let K ⊂ L be a field extension and let f ∈ K[x1 , · · · , xn ].

Then the Bernstein-Sato polynomial bK,f of f is equal to the Bernstein-
Sato polynomial of the same f seen as having coefficients in L
It is immediate to see that bL,f bK,f . Conversely consider a functional

equation
bL,f f s = P (s)f s+1
with P (s) ∈ An (L) and let (ej )i∈J be a basis of L over K. We have
decompositions, involving only a finite number of terms in J:
X X
bL,f = bj (s)ej and P (s) = Pj (s)ej ,
j∈J j∈J
such that for all j, bj ∈ K[s] and Pj ∈ An (K). Since f has all its coefficients
in K these decompositions are transmitted to the functional equation and
we obtain for all j:
bj (s)f s = Pj (s)f s+1
and therefore bK,f bj , so that bK,f bL,f as expected.
Let us now come to the proof of the theorem itself: Let k ⊂ K be
the subfield of K generated by Q and all the coefficients of f . Because of
the lemma the Bernstein-Sato polynomial of f is equal to its Bernstein-
Sato polynomial as a polynomial in k[x1 , · · · , xn ]. Now since k is a finitely
www.pdfgrip.com
263
generated extension of Q there exists an embedding k ,→ C, and therefore

bf may be considered as the Bernstein-Sato polynomial of an element of
C[x1 , · · · , xn ].
Because of the results of the section 5.3, this polynomial is the lcm of
local analytic Bernstein-Sato polynomials. The roots of bf have then the
required properties by the results of Kashiwara.
N.B. This proof fails for the Bernstein equation of a ”true formal ” i.e.
non convergent power series. The existence is proved in the book [9] of J.E.
Björk. The rationality of the roots is likely but I do not know of a reference
or a proof.
7 Introduction to analytic differential

modules. Abridged version
7.1 The ring of differential operators
The aim of this section is to define linear differential operators with analytic
coefficients on a complex analytic manifold X. For that we shall look at
operators on open subsets U ⊂ Cn , and at their behaviour by change of
coordinates.
We shall see that the natural definition for an operator on a real or
complex variety P ∈ D(X), is as an operator on the sheaf OX . We refer to
the book of R. Godement [14] for general facts about sheaves and to [15]
for a more expanded version of this subject.
7.1.1 Definition of D(U ) on an open subset of Cn

Let U ⊂ Cn be open and let O(U ) be the ring of holomorphic functions on
U.
Definition 7.1 A linear differential operator with analytic coefficients in

R = O(U ) resp in. R = C{x1 , · · · , xn } is a C-linear map P : R → R,
described by a finite sum of the type:
X
f→ aα (x)Dα (f )
α∈Nn
with aα ∈ R and Dα = ∂1α1 · · · ∂nαn , ∂i = ∂x

∂
i
. We set |α| = α1 + · · · + αn
and define the order of P as the natural integer ord(P ) := sup{m | ∃α ∈
Nn , |α| = m, aα (x) 6= 0}.
We denote D(U ), resp. D0 the ring of linear operators on U resp. of
germs of operators. We also write D(m)(U ) resp. D(m)0 for the sub-
modules of operators of order ≤ m.
www.pdfgrip.com
264
The coefficients aα in the expansion of an operator P are unique. Indeed

if |α| = m = ord(P ), we easily get P (xα ) = α!aα , and therefore the highest
order coefficient are determined in a unique way by P . We conclude by
induction.
As easyconsequence
we obtain that D(m)(U ) is a free O(U )-module of
m+n
dimension
n
Exercises:
i) Verify that [Dα , a] P
= Dα ◦ a − a ◦ Dα has order ≤ |α| − 1.
ii) Assume that P = |α|≤p aα (x)Dα has order exactly p, and similarly
that Q has order q. Prove that [P, Q] = P ◦ Q − Q ◦ P has order ≤ p + q − 1.
Define the symbols of P and Q as the functions on U × Cn :
X X
f (x, ξ) = σ(P ) = aα (x)ξ α , g(x, ξ) = σ(Q)(x, ξ) = aβ (x)ξ β
|α|=p |β|=q
Let σp+q−1 ([P, Q]) be either 0 or the symbol of [P, Q] if its order is p+q −1.
Then we have
Xn
∂f ∂g ∂f ∂g
σp+q−1 ([P, Q]) = −
i=1
∂ξ i ∂x i ∂x i ∂ξi
We shall prove that this formula has an intrinsic meaning on any variety.
Towards the sheaf D
The correspondence U → D(U ) defines a pre-sheaf of rings on Cn : If
V ⊂ U , and P ∈ D(U ), the restriction ρU,V P ∈ D(V ) is just the restricted
action of P on O(V ).
The sheaf DCn associated with this presheaf is called the sheaf of linear
differential operators. Any section has locally a finite order, and because
of the unicity of the expansion and by analytic continuation for the aα ’s,
the set of section of D is exactly D(U ) on any connected open set.
7.1.2 Behaviour under change of coordinates

In order to define operators on an analytic variety it is necessary to deal
with changes of coordinates. Let
ϕ
U // V
be a diffeomorphism between two open subsets of Cn . We denote ψ = ϕ−1 .

ϕ?
It induces an isomorphism of rings O(V ) // O(U ) , by ϕ? (f ) = f ◦ ϕ,
whose inverse is ψ ? .
www.pdfgrip.com
265
Given a linear map P ∈ HomC (O(U ), O(U )), we define ϕ? (P ) := ψ ? ◦

P ◦ ϕ? : O(V ) → O(V ). This means that that ϕ? (P ) is inserted in the
following commutative diagram:
P
O(U ) // O(U )
OO OO
ϕ? ϕ?
ϕ? (P )
O(V ) // O(V )
We get an isomorphism of rings whose inverse is ψ?

ϕ?
HomC (O(U ), O(U )) // HomC (O(V ), O(V ))
Proposition 7.2 The map P → ϕ? (P ), can be restricted to an isomor-

phism of rings from D(U ) to D(V ). On the set of vector fields it coin-
cides with the direct image induced by the tangent map of ϕ: ϕ? (ξ)(y) :=
dx ϕ(ξ(x)), ξ ∈ Θ(U ) , y = ϕ(x).
Proof The equality ϕ? (P Q) = ϕ? (P )ϕ? (Q) is a direct consequence of the

above diagram and of the definition of P Q as a composition of operators
from O(U ) to itself.
In order to prove that ϕ? D(U ) ⊂ D(V ) we just have to verify that ϕ?
preserves vector fields.
It can be done by the following explicit calculation:
∂ ∂
ϕ? (a(x) )(h)(y) = a(ϕ−1 (y)) (h ◦ ϕ)(ϕ−1 (y))
∂xj ∂xj
X ∂ϕi ∂h
= a(ϕ−1 (y)) (ϕ−1 (y))
∂xj ∂yi
or erasing the variables:

∂ X ∂ϕi ∂
) = (a ◦ ψ)
ϕ? (a ( ◦ ψ)
∂xj ∂xj ∂yi
P
And for an operator P = |α|≤ aα Dα it gives the formula:
n hX
∂ iαj
X Y n
∂ϕi
ϕ? (P ) = aα (ψ(y))
j=1 i=1
∂xj ∂yi
|α|≤m
from which incidentally we see that the map ϕ? preserves the order of
operators.
www.pdfgrip.com
266
7.1.3 Sheaf of differential operators on an analytic

manifold
An example. Consider the manifold X = P1 with the two coordinate sets
(U, z), (V, t), and change of coordinates z → t = z1 on U ∩ V ∼
= C? . There
is a global vector field ξ ∈ Θ(P1 ) defined by:
(
∂
ξ|U = ∂z
ξ|V = −t2 ∂t
∂
By the maximum principle O(P1 ) = C, so that the action of ξ is zero. This

shows that a differential operator is not always determined by its action on
global functions. However we see easily that the same ξ is determined by
its action on all the O(U )’s, for U ⊂ P1 an open subset.
Exercise: Describe a similar situation with the elliptic curve X = C/Z2
(Hint: there is an atlas whose coordinate changes are only translations).
InPgeneral we notice that if U is an open subset of Cn the operator
P = α∈Nn xα ∂ β acts as well on all the rings O(V ) for V ⊂ U . In fact
the definition of P as acting on the ring O(U ) is clearly equivalent to its
definition as a morphism of sheaves OU → OU .
Definition 7.3 1) A differential operator P ∈ D(X) on a complex analytic

manifold X is a C-linear homomorpism of sheaves:
P
OX // OX
whose restriction to each coordinate neighbourhood U ⊂ X is a linear dif-

ferential operator in the sense of 7.1.1.
2) We may define D(U ) in the same way for any open set U ⊂ X. The
sheaf DX is the sheaf of non-commutative rings U → DX (U ) = D(U ).
In this definition the restriction is just the restriction D(U ) → D(V ) of

morphism of sheaves P → P|OV and the axioms of sheaves are straightfor-
ward. We leave it as a routine exercise.
7.1.4 Principal symbols, and graded associated sheaves

Proposition 7.4 Let U ∈ Cn be an open set. Let
X
D(m)(U ) := { aα (x)Dα }
|α|≤m
www.pdfgrip.com
267
be the set of operators with order ≤ m. Then we obtain a ring filtration i.e.
D(m1 )(U )D(m2 )(U ) ⊂ D(m1 + m2 )(U ) and the associated graded ring
M D(m)(U )
grD(U ) =
D(m − 1)(U )
∂
is commutative isomorphic to O(U )[ξ1 , · · · , ξn ] with ξi the class of ∂xi .
D(m)(U )
Proof Denote by σm (P ) the class of P ∈ D(m)(U ) in the quotient D(m−1)(U ).
If σm (P ) 6= 0, that is if P ∈ D(m)(U ) \ D(m − 1)(U ) we denote it σ(P ) and
call it the principal symbol of P .
P
It is clear that σ(P ) = |α|=m aα (x)ξ α , and therefore O(U ) and ξ1 , · · · , ξn
generate grD(U ) and this ring is commutative because these generators
pairwise commute, the only nontrivial case being:
∂ ∂
[ξi , a] = σ1 ( )σ0 (a) − σ0 (a)σ1 ( )
∂xi ∂xi
∂ ∂ ∂a
= σ1 ( ◦a−a◦ ) = σ1 ( )=0
∂xi ∂xi ∂xi
The last assertion follows from the unicity of the coefficients aα of an op-
erator hence also of a symbol of order m.
Let T ? X → X be the conormal bundle on X.
The set of operators DX (m) of order ≤ m, is organized in a sheaf of
locally finite modules over the sheaf OX of holomorphic sections. We obtain
the following intrinsic formulation of the proposition 7.4:
Proposition 7.5 There is a coherent sheaf of rings

M DX (m)
grDX =
DX (m − 1)
m∈N
which is isomorphic to the sheaf π? O[T ? X] whose section on U are holomor-

phic functions on T ? U , polynomial in the fibers.
The fact that a principal symbol σ(P ) identifies with a function on

T ? U ⊂ T ? X, is a matter of control of the behaviour by a coordinate change,
and a direct consequence of the proof of proposition 7.2. Heuristically the
case of a vector field v ∈ Θ(U ) is clear because the evaluation on linear
forms is a function on the conormal bundle of U . We leave the details and
the extension to operators as an exercise.
www.pdfgrip.com
268
Consequence: Let us define the Poisson bracket by the formula:

Xn
∂f ∂g ∂f ∂g
{f, g} = −
i=1
∂ξi ∂xi ∂xi ∂ξi
Although we used coordinates it has an intrinsic meaning on T ? X, this

being
{f, g} = σp+q−1 ([P, Q]),
if f = σp P and g = σq Q.
7.1.5 Coherence
Theorem 7.6 The sheaves DX and grDX are coherent sheaves of rings.
We refer to the paper of J.-P. Serre for the definition of coherence of a
sheaf of rings and details about this notion. Recall that this notion is of
particular interest in that it reduces the notion of a coherent module, to
the property of having locally a finite presentation. Left coherent modules
are differential systems in the case of D. For DX the statement is valid for
left coherence and for right coherence as well.
We shall admit the following three results, more details being available
in [15]:
Lemma 7.7 Let K ⊂ Cn be a compact polycylinder, then the ring O(K)
of holomorphic functions in a neighbourghood of K is noetherian.
Theorem 7.8 Theorem of Cartan and Oka. For any complex analytic
manifold the sheaf OX is coherent.
Theorem 7.9 Theorem A of H. Cartan. Let F be a coherent sheaf of
O-modules on a neighbourhood U of a polycylinder K.
1) The O(K)-module Γ(K, F) is finitely generated
2) The sheaf F|K is (finitely!) generated by the set Γ(K, F) of its global
sections.
This last claim means that for any x ∈ K, and any germ of a section of F,
s ∈ Fx there are a1 , · · · , aq ∈ Fx and s1 , · · · , sq ∈ Γ(K, F) such that
q
X
s= ai si,x
i=1
where we denote by si,x the germ of si at x.

In view of giving the sketch of a proof of the coherence theorem we need
some facts about sections on a polycylinder K.
www.pdfgrip.com
269
1. The homomorphism induced by restrictions:
lim F(U ) → F(K)

U ⊃K,U open
is an isomorphism. This fact can be found in [14].
2. The module grD(U ) of sections of grD over an open chart domain U

is just the quotient DD X (m)(U )
X (m−1)(U )
and similarly over a compact poly-
cylinder we have
Γ(K, DX (m))
grD(K) '
Γ(K, DX (m − 1))
The reason is that on a chart we have a split exact sequence of sheaves

of free OX -modules:
// DX (m − 1) // DX (m) DX (m)
0 //
DX (m−1)
// 0
Theorem 7.10 The ring grDX (K) is noetherian and DX (K) is left and
right noetherian.
Proof By the description above and proposition 7.4 we have an isomor-

phism grDX (K) ' O(K)[ξ1 , · · · , ξn ], and the first noetherianity is just a
consequence of lemma 7.7 and of the usual transfer theorem.
As for the left (or right) noetherianity of DX (K) we can deduce it from
the noetherianity of grDX (K) = gr(DX (K)) by a proof similar to the proof
of the transfer property: take a left ideal I in DX (K) and prove its finiteness
from the finiteness of grI. Details are left as an exercise, and we refer to
our course [15, page 111].
Proof of the coherence theorem. We have to prove that the kernel of

any morphism of left D-modules:
φ : (DU )q → (DU )p
is coherent. Since the morphism φ is just the multiplication on the right

by a matrix of operators (Q1 , · · · , Qq ) 7→ (Q1 , · · · , Qq ) Ri,j , and setting
k0 = max(ordRi,k ) we find that φ((DU (`))q ) ⊂ (DU (` + k0 ))p . Therefore
ker φ∩DU (`)q appears as the kernel of a morphism of locally free finite type
modules:
DU (`)q → DU (` + k0 )p
www.pdfgrip.com
270
and hence is coherent for each `, by Cartan-Oka theorem. The union over
` being ker φ.
Therefore by theorem A of Cartan,
for any polycylinder K ⊂ U , and any
x ∈ K the fiber ker φ ∩ DU (`))q x is generated by Γ(K, ker φ ∩ DU (`))q ⊂
Γ(K, ker φ), from which ker φx is generated by Γ(K, ker φ).
By the left exactness of Γ(K, •), this gives an exact sequence:
0 // Γ(K, ker φ) // Γ(K, (DU )q ) ' DX (K)q

// Γ(K, (DU )p ) ' DX (K)p
and by the noetherianity of DX (K) this proves that Γ(K, ker φ) is generated
by a finite number of sections say s1 , · · · , sr .
Using again the theorem A of Cartan we conclude that the morphism:
DX (K)r → (ker φ)|K
is surjective which ends the proof.
7.2 Coherent D-modules or differential systems

7.2.1 What is a differential system?
Definition 7.11 A differential system over a complex analytic variety is
a sheaf of coherent DX -modules.
Since DX is a coherent sheaf we have local finite presentations of M
q φ p
D|U // D|U // M|U // 0

as cokernels of maps φ(Q1 , · · · , Qq ) = (Q1 , · · · , Qq ) Ri,j .
Let us consider solutions u = tr (u1 , · · · , up ) in OU , of the differential
system associated with the matrix of φ
Ri,1 u1 + · · · + Ri,p up = 0, for i = 1, · · · , q
There is a bijection between this set of solutions and the set of left D-
q
linear maps MU → OU . Indeed for any solution u and any Q ∈ D|U , we

have φ(Q)(u) = QRi,j u = 0, this meaning that the D-linear homomor-
p P
phism D|U → OU , given by (P1 , · · · Pp ) → Pj uj , is zero on Imφ, hence
factors through MU . This leads to:
Definition 7.12 The sheaf of solutions of M is the sheaf Sol(M) = Hom(M, O)
of germs of homomorphisms from M to O.
www.pdfgrip.com
271
7.2.2 Regular connections

Lemma 7.13 Let M be a coherent DX -module which is also coherent as
an OX -module. Then the fiber of M is a locally free OX module of finite
rank.
Proof We may assume that X is a neighbourhood of 0 ∈ Cn . Consider a

minimal system of generators (e1 , · · · , ep ) of the fiber M0 , and let us prove
that (e1 , · · · , ep ) is free over O0 . Otherwise we would have a linear relation:
p
X
(?) ui ei = 0 with ui ∈ OX,0
i=1
Define ν as the minimal valuation amongst the val(ai ), reached say on a1 ,

that is:
∀i , ui ∈ (x1 , · · · , xn )ν , and u1 ∈ (x1 , · · · , xn )ν \ (x1 , · · · , xn )ν+1
If ν > 0 then we can choose a variable xk such that the initial part inν u1
∂u1
depends on it. In that case the valuation of ∂x k
is exactly ν − 1.
Let us derive (?):
Xp Xp
∂ui ∂
ei + ui ei
i=1
∂xk i=1
∂xk
Rewriting each ∂x∂ k ei as a linear combination of the ei we obtain a new

linear relation between the generators ei . In this relation the coefficient of
e1 , has the form:
∂u1
u01 = + L(u1 , · · · , un )
∂xk
where L(u1 , · · · , un ) is a linear combination of the coefficients ui . This
implies that the valuation of this coefficient u01 is now ν − 1. By repeating
this process we might have assumed that ν = 0. But in that case u1 would
be a unit in the local ring OX,0 and e1 a linear combination of e2 , · · · , en
contrary to the minimality.
Now assume that e1 , · · · , en are defined as sections of M on the open
neighbourhood U of 0. Since M is locally of finite type, we know that they
generate the restriction of M to some V ⊂ U containing the origin.
Arguing similarly with the sheaf of their relations over O whose fiber at
0 is zero we can prove that M is free with basis {ei } on some W ⊂ V.
www.pdfgrip.com
272
Theorem 7.14 Let M be a coherent DX module which is also of finite type

and coherent 3 over OX . Then M is a locally isomorphic to some (OX )r as
a DX module, with OX being equipped with its canonical structure over DX .
We call a D-module which is locally free over OX a regular connection4 .
By the lemma we have a local isomorphism M → (OX|U )r of OX|U -
modules, with a basis (e1 , · · · , er ).
Due to the fact that O ' D·(∂1D ,··· ,∂n ) , the statement of the theorem is
equivalent to the existence of an horizontal basis m1 , · · · , mr that is a basis
such that for all (j, k) we have ∂j mk = 0.
We refer to our course [15] theorem 4 page 137 to 139.
Exercise: Show that the one variable case in 1) is just Cauchy theorem.
∂
Hint: write the action of ∂x on M as
   
m1 m1
   
∇  ...  = A(x)  ... 
mr mr
where A(x) =Pai,j (x) is an n × n matrix with holomorphic coefficients.

Show that ∇( gi (x)mi ) = 0 is equivalent to the linear differential system:
   
g1 g1
∂  .  tr  . 
 ..  = − A(x)  .. 
∂x
gr gr
The proof in higher dimension is a more technical n-dimensional version

of the Cauchy theorem. The following corollary is the first step toward the
finiteness property in proposition 5.2 used for the proof of the existence of
the Bernstein-Sato polynomial.
Corollary 7.15 Let M be a regular connection and let ϕ : M → M be a

D-linear homomorphism. Then the germ of ϕ at any point p ∈ X admits a
minimal polynomial.
According to theorem 7.14 we may assume that Mp = Or as a Dp -module.

Then ϕ is described by a matrix (fi,j ) so that the action on the vectors of
the canonical basis is
X n
ϕ(j ) = fi,j j
i=1
3 Infact the coherence hypothesis over OX is redundant as we shall see from the
existence of good filtrations, see next section theorem 7.18.
4 We shall see in the next section that a regular connection is the same as a D -module
X
whose characteristic variety is the zero section TX ? X of the cotangent bundle.
www.pdfgrip.com
273
From the fact that for any k, ∂k j = 0, we deduce easily that for all i, j, k,
∂k (fi,j ) = 0, so that ϕ is represented by an r × r matrix with constant
coefficients and as such admits a minimal polynomial bϕ ∈ C[s], bϕ (ϕ) = 0
7.3 Good filtrations and coherence conditions

Consider the differential system M = D I associated with a left ideal I and
consider its OX -submodules induced as quotients of the subsheaves DX (m)
of operators of order ≤ m.
Following this model we are led to the definition below:
Definition 7.16 Let M be a DX -module. A filtration on M is a collection

of OX -submodules Mk such that
S
1. M = k∈N Mk
2. Mk ⊂ Mk+1
3. ∀k, ` ∈ N, DX (`)Mk ⊂ Mk+`
Definition 7.17 We say that {Mk } is a good filtration on M if

1. Each Mk is a coherent OX -module.
2. ∃k0 ∈ N , ∀` ∈ N , DX (`)Mk0 = Mk0 +`
L
Exercise Verify that grM = k∈N MMk−1 k
is a grD-module in the obvi-
ous way.
Examples
• Let M be a regular connection. By definition it is an OX -coherent
module and we can take Mk = M for all k.
• O[ x1 ], sheaf of meromorphic functions on X = C with finite order pole
at 0, filtered by the set O[ x1 ]≤k of functions with poles of order ≤ k.
Exercises on good filtrations
In both exercises ii) and iii), remember that if F is coherent as a sheaf
of O-module, then a submodule F 0 ⊂ F is coherent if it admits locally a
finite number of generators.
i) Let Mk be a locally good filtration on M. Then for any compact
subset K of X we can find an integer with the property of the definition,
globally on K.
ii) Prove that the sub DX -module D · f s ⊂ O[ f1 , s]f s has a good filtra-
tion.
www.pdfgrip.com
274
Hint: Consider the filtration of DX [s](m) by the total order with respect
to (∂, s) and prove that DX [s](m)f s is OX -coherent.
iii) In the proof of theorem 5.6 we found k0 such that O[ f1 ] = DX f 1k0 .
Prove in detail that (DX (m) f 1k0 )m∈N is good. A key step towards the
holonomicity of O[ f1 ]
Théorème 7.18 Any coherent left DX -module admits locally a good filtra-
tion
We just give here an idea of the proof. Given x ∈ X consider a local

presentation:
Φ π
(D|U )q // (D|U )p // M|U // 0
in a neighbourhood U of x. We set Mk = π(D|U (k)p ). For this filtration

the condition 2. of definition 7.17 is satisfied with k0 = 0 and each Mk is
O-coherent. For the details see [15].
In the same reference the reader will find treated with some details a
result converse to the previous result, starting from a coherence criterion
associated with graded modules, and then a theorem which asserts the
equivalence between the local existence of a good filtration and the coher-
ence of a DX -module. This is the main reason why we did not limit our
definition of a good filtration to a priori coherent DX -modules.
Proposition 7.19 Let M be a left DX -module equipped with a filtration
{Mk }. Then this filtration is good if and only if grM is a coherent grD-
module.
About this result an easy part of the proof is that definition 7.17 implies
the finiteness of the fiber grMx over grDx .
Théorème 7.20 A left D-module M is coherent if and only if it admits
good filtrations on a sufficiently small neighbourhood of each point.
This result allowed us to prove that O[ f1 ], and D[s]f s are coherent D-
modules since we refered to it in section 5 and in theorem 5.6. For the
second module we notice that the goodness of an appropriate filtration
does not use the existence of the Bernstein-Sato polynomial. All the better
so since this coherence is a piece of the proof of the existence of a local
Bernstein-Sato polynomial!
A corollary of the proposition 7.19 is the behaviour by exact sequences
which leads to an Artin type property necessary in the proof of theorem
7.20:
www.pdfgrip.com
275
Proposition 7.21 Let
π
0 // M // N // P // 0
be an exact sequence of left D-modules and let {Nk } be a good filtration

on N . Then the induced and quotient filtrations {Mk = Nk ∩ M} and
Pk = π(Nk ) are good.
The induced and quotient filtrations are exactly made for the exactness of
the sequence of sheaves
// grM // grN π // grP

0 // 0
By the proposition 7.19 grN is coherent and it is sufficient to prove the

coherence of grP, see the details in [15, Proposition 14].
Exercises i) Translate the above proposition into the following Artin type
formula:
∃k1 ∈ N, ∀` ∈ N, Nk1 +` ∩ M = DX (`)(Nk1 ∩ M)
ii) Prove that any good filtration {Mk } has a local expression of the follow-
ing type: in a sufficiently small neighbourhood
r
U of any given x ∈ X, there
is a presentation of M as a quotient DN of a free module by a coherent
submodule N and a r-uple of integers n = (n1 , · · · , nr ) ∈ Z such that we
have:
Mr
Mk = π(D[n](k)) , with D[n](k) = D(k − ni )
i=1
Hint: choose a local system of homogeneous generators of grM which also

generates M.
It is technically important to state a result of local comparison between

two good filtrations:
Proposition 7.22 Let Mk = Fk (M) and M0k = Fk0 (M) be two good fil-
trations on a left DX -module M. Then, for any x ∈ X, there is a neigh-
bourhood U of X and an integer k0 ∈ N such that
∀` ∈ N , F` (M|U ) ⊂ Fk0 0 +` (M|U ) ⊂ F2k0 +` (M|U )
We shall omit the easy proof, left as an exercise. See also [15].
www.pdfgrip.com
276
7.4 Characteristic variety of a differential system

D
7.4.1 Case of a monogeneous module M = I
Let I be a coherent left ideal of D. We denote I the graded ideal of D := grD

associated with the natural induced filtration on I:
M I ∩ DX (`) M I ∩ DX (`) + DX (` − 1)
I = grI = '
I ∩ DX (` − 1) DX (` − 1)
`∈N
We are in fact in the situation of the proposition 7.21

π
0 // I // D // M // 0
DX (`)
with the natural filtrations M` = π(DX (`)) = I∩D X (`)
and I` = I ∩ DX (`)
L M` D
We can write gr(M) := I∩DX (`) ' I according to elementary results
on quotients of modules:
Exercise: Verify that the following sequences are exact:
0 // I` // D(`) // M` // 0
// gr` I D(`) π M`
0 //
D(`−1)
//
M`−1
// 0
Using proposition 7.21 we can set the following

Definition 7.23 The sheaf I is a coherent graded sheaf of ideals in D ⊂
OT ? X . This ideal defines a conical subset of T ? X that we denote char(M)
and call the characteristic variety associated with I.
Some Remarks:
- Conical means that (x, ξ) ∈ char(M) ⇒ ∀λ ∈ C , (x, λξ) ∈ char(M)
- Let fi be a local system of homogeneous generators of I, that is I|U =
(f1 , · · · , fr )grD|U , then we have a set of homogeneous equations for char(M):
f1 (x, ξ) = · · · = fr (x, ξ) = 0
But if I = D · (P1 , · · · , Pr ), the symbols σ(P1 ), · · · , σ(Pr ), may fail to

generate I.
We shall prove soon in a more general setting that this variety is intrinsic
for M. Here is an example:
D
Let M = D·x . Let δ be the class of 1. Since xδ = 0, we have δ = −xδ 0 ,
so that M = D · δ = D · δ 0
D
This gives two presentations M = I1 = ID2 , with I1 = D · x and I2 the
0
annihilator of δ .
www.pdfgrip.com
277
∂
Exercise Prove that I2 = D · x2 + D · (x ∂x + 2)
Verify that I1 = OT ? X · x and I2 = OT ? X · (x2 , xξ) define the same
subset of T ? X, the conormal set to the origin T0? X.
Let us remark as a transition that I = grI is the annihilator of grM.
In the next section we generalize this observation.
7.4.2 General case

Let us assume first that M has a global good filtration {M` }`∈N .
We set:
I = Ann(grM) = {f ∈ grD , f · grM = 0}
Remark 7.24 The set I is a homogeneous ideal of D which means that:

M
I= I` where I` = I ∩ gr` D.
This is a general fact about annihilators of graded modules over graded

rings, which we leave as an exercise.
Remark 7.25 If f = σ(P ) ∈ Id , with P of order d, and m ∈ Mk , the

equation f m = 0 is equivalent to
Mk+d
P m = 0 , in
Mk+d−1
that is:
P · Mk ⊂ Mk+d−1
In view of defining the characteristic variety in term of I the following result

is important:
Proposition 7.26 Let M be a coherent D-module with a good filtration

{Mk }. Then the ideal I = Ann(grM) is a coherent sheaf of ideals of D.
To complete the definition we would like to glue the sets of zeros of I on

all coordinate sets where we have a good filtration. This requires verifying
that this set of zeros as a reduced analytic set is independent of the chosen
good filtration. This is obtained in the following theorem.
Théorème 7.27 Let M be a left coherent DX -module. Then the radical

√ p
I = AnngrF M
does not depend on the choice of a good filtration F .

www.pdfgrip.com
278
The proof is identical to the proof for the algebraic case and we omit it.
See the course of F. Castro-Jimenez in this volume.
√ fact (see Gunning and Rossi): if I ⊂ D
We shall also admit the following
is a coherent sheaf of ideals then I is also a coherent sheaf. We are now
ready to give as a conclusion the final definition:
Definition 7.28 Let M be a coherent left DX module on a complex an-
alytic manifold X. Then there is a coherent sheaf of homogeneous ideals
J ⊂ grDX such that for any local good filtration of the restriction of M to
an open subset U ⊂ X, we have:
p
J = AnngrF M
The conical subset of T ? X defined as the zero locus of J is called the char-
acteristic variety char M of the differential system M.
7.4.3 A finiteness property

Various properties of the Characteristic variety char(M) ⊂ T ? X, such as
the dimension, the Bernstein-Sato inequality, the involutivity, the multi-
plicity and the dimension at a point in T ? X should still be developped
and we shall give in this last section a very partial view of these subjects
adapted to the use we made of them in section 5.
As we already noticed in section 5 theorem 5.1 and proposition 5.2 are
indeed the two pillars on which the existence of the b-function is built.
For the analysis of the local dimension and the multiplicity at any point
of the characteristic variety of M, for the study of the homological dimen-
sion of D and the way all these ingredients allow us to build the filtration
by the dimension of a coherent module as stated in theorem 5.1, we refer
to [15, Chapter V].
In this last section we shall now give a partial account of the list of
results which lead quite naturally to a proof of the proposition 5.2 as in
[15, IV.2.2.].
Theorem 7.29 : Bernstein inequality. Let M 6= (0) be a coherent DX -
module on an n-dimensional manifold X. Then for any x ∈ X such that
the germ at (x, 0) of char(M) is non zero we have:
dim(x,0) char(M) ≥ n = dim X.
It is easy to prove that the support of M which is the setlocus {x ∈ X |
?
Mx 6= 0} of non zero stalks is equal to char(M) ∩ TX X. The result follows
by descending induction on the dimension of this support using a lemma on
the structure of the systems whose support is contained in a hypersurface.
See [15, III.3.] for more details on coherence conditions.
www.pdfgrip.com
279
Lemma 7.30 Let M be a DU -module, with U ⊂ Cn whose support is

contained in the hyperplane H = {xn = 0}. Then the kernel M of the map
xn : M → M is a coherent DU ∩H -module and we have isomorphisms:
∞
M
M|U ∩H ' ∂nk M , and char(M) = char(M) × {(0, ξn )}
k=0
Definition 7.31 A coherent DX -module is called holonomic if the charac-

teristic variety char(M) has dimension n.
Let us notice that Bernstein inequality, in the way we stated it in 7.29

does not guarantee the same lower bound
dx? char(M) ≥ n (7)
for the local dimension at any point x? ∈ char(M) which is out of the zero
?
section TX X.
This generalized Bernstein inequality is also true. As a consequence the
components of char(M) have dimension at least n, and in particular the
characteristic variety of a holonomic module is of pure dimension n. The
inequality (7) follows from a much deeper result: the involutivity of the
characteristic variety. Recall that the cotangent bundle is equipped with
a canonical 1-form θ and a canonical 2-form ω = dθ whose expressions in
local coordinates are:
X n
X
θ= ξi dxi , ω = dθ = dξi ∧ dxi .
i=1
The 2-form ω defines at each point x? of T ? X a nondegenerate bilinear

form. Given a smooth subvariety V ⊂ T ? X and p ∈ V we write (Tp V )⊥ ⊂
Tp (T ? X) for the orthogonal subspace to Tp V with respect to ω.
Theorem 7.32 The characteristic variety of any coherent module is invo-

lutive.
Involutivity means that for any p in the smooth part of char(M), we have:
(Tp char(M))⊥ ⊂ Tp char(M) (8)
A proof of the involutivity theorem can be found in [29]. As explained by

Bernard Malgrange in this reference the first proof of the involutivity is
due to M. Kashiwara, T. Kawai and M. Sato in [20]. This proof as well
as the proof by Malgrange in [29] or the proof inspired by a suggestion
of B. Malgrange in [15] uses a localisation of operators in T ? X known as
www.pdfgrip.com
280
microlocalisation. For a purely algebraic proof valid for a much larger class
of non commutative rings a basic reference is the paper of O. Gabber [13].
Because of the nondegeneracy of ω, Bernstein inequality follows from
the inequality of dimensions 2n − dim Tp char(M) ≤ dim Tp char(M) at
any smooth point hence everywhere. For a holonomic module we deduce
from the inclusion (8) that there is in fact an equality (Tp char(M))⊥ =
Tp char(M) at each smooth point. A variety with this property is called
Lagrangian.
The structure of these varieties in the conical case is easy to deduce
from a study of the symplectic form ω and we obtain (see [15, IV.1.] for
details):
Theorem 7.33 The characteristic variety of a holonomic module M is a

conical involutive subset of T ? X. As such it can be written as a union
[
charM = TS?α X
α∈A
where (Sα )α∈A is the locally finite family of closed analytic sets consisting
in the projections of the components of char(M) by the map T ? X → X.
In this theorem TS?α X is the closure of the conormal to the smooth F part
of
S S α . We can also define a refinement of the support of M, α∈A α =
S
β∈B Y β , into a Whitney stratification (Y β ) β∈B . We have then an inclusion:
G
char(M) ⊂ TY?β X
β∈B
because the second member of this inclusion is closed as a consequence of the

Whitney condition. For another aspect of the geometry of the characteristic
variety, involving sheaves of solutions, the reader can consult the lectures
notes [26] of Lê Duñg Tráng and Bernard Teissier in this volume.
The structure of a D-module whose characteristic variety is the conormal
bundle to a smooth subvariety can be described completely from a local view
point. We assume for this purpose that Y = {x ∈ Cn | x1 = · · · = xp =
0} ⊂ X = Cn . A typical module having TY? X as a characteristic variety is:
DX
BY (X) =
DX (x1 , · · ·, xp , ∂x∂p+1 , · · ·, ∂x∂ n )
Proposition 7.34 Let M be a D-module whose characteristic variety is

the conormal bundle TY? X. Then M is isomorphic to the direct sum of r
copies of the D-module BY (X)
www.pdfgrip.com
281
Proof We can apply lemma 7.30 repeatedly. The iterated kernel M :=

{m ∈ M|Y | x1 m = · · · = xp m = 0} is a coherent DY -module, and because
of the relation charM|Y = {(0p , ξ1 , · · ·, ξp )} × charM, we find that:
charM|Y = TY? Y
the conormal section of Y . It is easy to deduce from this fact that M|Y is
a connection see [15, IV.2.], hence is isomorphic to a OYr as a DY -module.
The proposition follows by application of an iterated version of lemma 7.30.
M
M|Y ' ∂1k1 · · ·∂pkp OYr ' BY (X)r
k∈Np

The above proposition can be thought as a generalisation of theorem
7.14 which concerns the case Y = X. Similarly the generalisation of corol-
lary 7.15 is:
Corollary 7.35 Let M be a DX -module whose characteristic variety is the

conormal bundle TY? X to a smooth submanifold Y ⊂ X and let ϕ : M → M
be a DX -linear homomorphism. Then the germ of ϕ at any point p ∈ X
admits a minimal polynomial.
The proof is identical to the proof of corollary 7.15 once we have proved
by a direct calculation that any morphism BY (X) → BY (X) is a dilatation
m 7→ λm, λ ∈ C.
Let us now recall proposition 5.2, which we are now able to prove as
announced at the beginning of this last section 7.4.3
Proposition 5.2 Let M be a germ of a holonomic module and ϕ : M −→

M a D-linear map. Then ϕ admits a minimal polynomial.
We use the notation of theorem 7.33 and consider a component Sα0 of
the highestS possible dimension in the support of M and a smooth point
p ∈ Sα0 \ α6=α0 Sα . Then we may apply corollary 7.35 to the germ of M
at p. Therefore there exists a polynomial b1 (s) ∈ C[s] such that b1 (ϕp ) = 0
and therefore b1 (ϕ)M does not contain any more p in its support. Since Sα0
is irreducible we conclude that in the characteric variety char(b1 (ϕ)cM ) ⊂
charM we may drop one component TSα0 X.
Proposition 5.2 follows by induction on the number of local components
of charM.
www.pdfgrip.com
282
8 Complements
8.1 Appendix A: Mellin and Laplace transforms

8.1.1 Integrals depending on a parameter
We consider Ω ⊂ C an open subset of C and f : I × Ω with I ⊂ R an
interval, a continuous function which is holomorphic with respect to the
second variable.
Proposition 8.1 Assume that for a compact subset K ⊂ Ω there is an

integrable function ϕ : I → R+ such that:
∀(t, z) ∈ I × Ω , |f (t, z)| ≤ ϕ(t)
then the function

Z
g(z) = f (t, z)dt
I
is holomorphic on Ω its derivatives being

Z n
(n) ∂ f
g (z) = n
(t, z)dt
I ∂z
8.1.2 Analyticity of Mellin transforms

Let f : ]0, +∞[→ C be a continuous function (or simply a locally integrable
one). The Mellin transform of f is the function Γf defined in some open
set Ω ⊂ C (possibly empty!), by the integral:
Z +∞
Γf (s) = f (x)xs−1 dx
0
If we set x = e2πt , et s = −iz, this integral can also be interpreted as a

Laplace transform of g(t) = f (e2πt ), t ∈ R.
Z +∞ Z +∞
2πt(−iz−1) 2πt
Γf (s) = 2π g(t)e e dt = 2π g(t)e−2πitz dt
∞ ∞
Thus we have Γf (s) = 2πĝ(z) in which the Laplace transform of g is defined

as: Z +∞
ĝ(z) = g(t)e−2iπtz dt
∞
www.pdfgrip.com
283
Theorem 8.2 Assume that the integral defining Γf (s) is absolutely con-
vergent for <(s) = a and for <(s) = b > a. Then this integral is also
absolutely convergent for a ≤ <(s) ≤ b, and the function Γf is holomor-
phic and bounded in the open strip B := {s ∈ C , a < <(s) < b} and
continuous in the closed strip B.
Since |xs−1 | = x<s−1 , the absolute convergence of Γf (s) depends only
on <s and for any x > 0 and any s ∈ B, we have:
|xs−1 | = x<s−1 ≤ xa−1 + xb−1
This implies the convergence of the integral and by section 8.1.1 the conti-
nuity on B (using the Lebesgue convergence theorem) and holomorphy on
B. Furthermore Γf is bounded since
|Γf (s)| ≤ Γ|f | (a) + Γ|f | (b)
The derivative of Γf is:
Z +∞
Γ0f (s) = f (x)(log x)xs−1 dx
0
Examples
1. Let f : [0, +∞[−→ R be rapidly decreasing at +∞, which means that
all xn Dα (f ) are bounded, and continuous at 0. Then Γf (s) is well
defined for <s > 0
2. Let f be with compact support [a, b] ⊂]0, +∞[. Then Γf (s) is an
entire function defined in the whole complex plane.
Both examples are direct consequences of section 8.1.1.
We may refine and generalise the first example to a large class of ex-
amples for which there is a meromorphic continuation to a half-plane or to
the whole complex plane in some cases.
Theorem 8.3 Let f : ]0, +∞[−→ C be a locally integrable function rapidly
decreasing at infinity, and which admits an asymptotic expansion at zero of
the form
f (x) = a1 xs1 + · · · + an xsn + O(xsn+1 )
with <s1 ≤ · · · ≤ <sn < <sn+1 . Then the integral Γf (s) is convergent for
<s > −<s1 and admits an analytic continuation as a meromorphic function
on the punctured half-plane:
{s ∈ C | <s > −<sn+1 } \ {−s1 , · · · , −sn }
with simple poles having as residues
Res−sk (Γ( f ) = ak
www.pdfgrip.com
284
R∞
Proof For any s ∈ C the integral 1 f (x)xs−1 ds is convergent and defines
a holomorphic entire function Rby section 8.1.1. Therefore in what follows
1
we deal only with the integral 0 f (t)dt.
−s1
Furthermore g(s) = x f (x) is continuous at 0 with g(0) = a1 . Then
by example 1 above the Mellin transform Γg (s̃) of g is defined in the half
plane s̃ > 0, which is equivalent to the fact that Γf is defined by a conver-
gent integral on the half plane <(s + s1 ) > 0.
Let us write the asymptotic expansion of f and its integral on [0, 1]:
f (x) = a1 xs1 + · · · + an xsn + xsn+1 h(x), h(x) bounded.
Z 1 Z 1 Z 1
s+s1 −1
f (x)x s−1
dx = a1 x dx + · · · + an xs+sn −1 dx
0 0 0
Z 1
+ xs+sn+1 −1 h(x)dx
0
Z 1 Z 1
a1 an
f (x)xs−1 dx = +···+ + xs+sn+1 −1 h(x)dx
0 s + s1 s + sn 0
This ends the proof of the theorem because, again by example 1 above (or
rather a slight generalisation), the function
Z 1
s→ xs+sn+1 −1 h(x)dx
0
is holomorphic in the half plane <(s + sn+1 ) > 0.
Corollary 8.4 If f is C ∞ on [0, +∞[ and rapidly decreasing at +∞ then

Γf can be extended to a meromorphic function on C \ (−N) such that
f (n) (0)
Res−n (Γf ) =
n!
More generally if we have an asymptotic expansion at any order of the
type:
X
f (x) ∼ ai xsi (log x)`
i∈N,`≤ni
for some sequence with (strictly) increasing <si tending to +∞ then the
function Γf can be extended to a meromorphic function on C \ {−si } with
poles at −si of order ≤ ni . We just have to use exactly the same argument
at any order and use the formula for the derivative of the Mellin transform.
www.pdfgrip.com
285
8.2 Appendix B: Regular sequences, and application

to the annihilator AnnD f s , in the isolated case
In this appendix we want to show the following theorem, about the an-
nihilator of f s over D in the case of an isolated singularity. We denote
by g1 , · · · , gn a regular sequence in O = C{x1 , · · · , xn }, and A the ideal
(g1 , · · ·, gn ).
The following proposition is stated in Yano’s paper [38] without proof.
Since I do not know a reference for a proof I give one after hand-written
notes [12] which Joël Briançon communicated to me a long time ago...
Proposition 8.5 The following conditions on g1 , · · · , gn are equivalent:

O
i) For i = 2, · · · , n gi+1 is a non zero divisor in (g1 ,··· ,gi ) (which is the
definition of a regular sequence.)
ii) The kernel of the natural map sending ξi to gi :
ϕ L ν
O[ξ1 , · · · , ξn ] / RA := ν∈N A Tν
is the ideal of O[ξ1 , · · · , ξn ] generated by the set of elements {gi ξj −gj ξi | 1 ≤

i < j ≤ n}.
These two conditions are also,Pas stated in Yano’s paper, equivalent to a

third one: Let A be the ideal Ogi , then the ring homomorphism:
O ψ
/
L Aν
A [ξ1 , · · · , ξn ] ν∈N
Aν+1
determined by ψ(ξi ) = gi ∈ AA2 is an isomorphism.

We will not need this latter fact in these notes. The fact that the
∂f ∂ ∂f ∂
annhihilator in D of f s is generated by the operators ∂x i ∂xj
− ∂x j ∂xj
when
f is an isolated singularity is then an easy induction on the order of the
operator P using the fact that if P f s = 0 the principal symbol σ(P )(x, ξ)
∂f
is in the kernel of the map ϕ, and the proposition 8.5 applied to gi = ∂x i
.
Proof We first recall the relations between the elements of a regular se-
quence in an arbitrary commutative ring R
Lemma 8.6 If g1 , · · · , gn is a regular sequence in R then the module of

relations: X
Rel := {(λ1 , · · · , λn ) ∈ Rn | λi g i }
www.pdfgrip.com
286
is generated by the elements gi j − gj i , with 1 , · · · , n the canonical basis

of Rn .
An equivalent formulation is to say that there exists an antisymetric n × n

matrix A, such that λ = Ag, with λ = tr (λ1 , · · · , λn ) and g = tr (g1 , · · · , gn )
The lemma is easily proved by induction on n. First by the regular
sequence condition we know that λn = 0 in (g1 ,···O,gn−1 ) . Therefore we may
write λn = α1 g1 + · · · + αn−1 gn−1 , and we get a relation between the n − 1
first gi :
n−1
X
(λi + αi gn )gi
i=1
to which we may apply the induction hypothesis. This gives the desired
result (details left to the reader).
P
Lemma 8.7 Let |J|=p µJ g J = 0 be a relation in O between the mono-
mials of a fixed degree p in the gi ’s. Then there is a matrix H = aJ,i with
coefficients in O such that:
n
X
µJ = aJ,i gi
i=1
and for any n-multiindex L of length p + 1,

X
aJ,i = 0,
J+i =L
Proof We set in an arbitrary way:

X
µJ = µJ,k , for any J = (j1 , · · · , jn ).
jk >0
We get
X X X
0 = µJ g J = µI+k ,k fk f I
|J|=p |I|=p−1 k
X X
I
:= νI f with νI = µI+k ,k fk
|I|=p−1 k
By an induction hypothesis on p this implies:

www.pdfgrip.com
287
n
X
νI = bI,k gk
k=1
X
bI,k = 0
I+k =J
which can be rewritten in the following way:
n
X
(µI+k ,k − bI,k )fk = 0
k=1
and using the case p = 1 which is exactly lemma 8.6 leads to:
X
µI+k ,k − bI,k = hIk,i fi
a formula which involves several antisymmetric matrix hI = (hIk,i ). Then

by an appropriate summation:
X X
(µJ,k − bJ−k ,k ) = µJ − bI,k = µJ
jk >0 I+k =J
and finally
X X X n
X
µJ = (µI+k ,k − bI,k ) = ( hIk,i fi ) = aJ,i fi
I+k =J I+k =J i=1
P
with aJ,i = I+k =J hIk,i This proves the lemma 8.7 because of the following
last calculation:
www.pdfgrip.com
288
X X
aJ,i = hIk,i + hIi,k = 0
J+i =L I+k +i =L

Now we return finally to the proof of the proposition 8.5:
i) ⇒ ii) P
Let us take
L νa homogeneous element of degree p, µJ ξJPin the kernel
ν
of
P O[ξ] → A T . By the lemma 8.7 we may write µ J = i aJ,i gi with
a
J+i =L J,i = 0 for any fixed L of length p + 1.
Let us choose J and i such that aJ,i 6= 0. We change the relation if
there is an index jk such that j + k 6= 0 and k < i. Then for such a triple
(J, k, i) we have:
0
gi ξ J = gi ξk ξ J−k = (gi ξk − gk ξi )ξ J−k + gk ξ J
with J 0 = (j1 , · · · , jk − 1, · · · , ji + 1, · · · , jn )
Now modulo the ideal of O generated by the element gi ξk − gk ξi we can
replace the term aJ,iP gi ξJ by aJ,i gk ξJ 0 , noticing that because of i + J =
k +J 0 the condition J+i =L aJ,i = 0 is preserved. Iterating this proces we
are reduced to a situation where the only remaining aJ,i are for multiindices
J of the type:
J = (0, · · · , 0, ji , · · · , jn )
P
Then all the aJ,i are zero because of the relation J+i =L aJ,i = 0 and the
fact that in this situation there is for any L a unique J.
O
ii) ⇒ i) We have to prove that gi+1 is not a zero divisor in (gi ,...,g i)
. It
O
is enough to do this for i + 1 = n. If u · gn = 0 in (gi ,...,gn−1 ) this means
Pn Pi
that ugn = i=1 ui gi or that (umodA)ξi+1 = k=1 (ui modA)gk ∈ Kerψ.
By a direct application of ii) this implies that u ∈ cA.
References
[1] A. Assi, F.J. Castro-Jimenez and M. Granger, The standard fan of an
analytic D-module. JPAA n◦ 164 (2001) pp 3-31.
[2] M.F. Atiyah, Resolution of singularities and division of distributions.
Comm in pure and applied mathematics vol 23, 1970, p. 145-150.
[3] R. Bahloul, Démonstration constructive de l’existence de polynômes de
Bernstein-Sato, Compositio Math. 41 (2005) 175-191.
[4] D. Barlet, Développememts asymptotiques des fonctions obtenues par
intégration sur les fibres. Inventiones Mathematicae, 68, p. 129-174
(1982).
www.pdfgrip.com
289
[5] I. N. Bernstein, S.I. Gel’fand Meromorphy of the function Pλ . Funct.

Anal. Appl. 1, p. 84-86, 1968.
[6] I. N. Bernstein, The analytic continuation of generalised functions with

respect to a parameter. Funct. Anal. Appl. 6, 273-285, 1972.
[7] J. Briançon, M. Granger and Ph. Maisonobe, Sur le polynôme de

Bernstein des singularités semi quasi-homogènes. Prépublication de
l’Université de Nice n◦ 138. Nov. 1986.
[8] J. Briançon, M. Granger, Ph. Maisonobe and M. Miniconi, Algorithme

de calcul du polynôme de Bernstein: cas non dégénéré. Annales de
l’Institut Fourier, tome 39, Fasc. 3.1989. pp. 553-610.
[9] J.E. Björk, Rings of Differential Operators. Volume 21 of North-

Holland Math. Libr., Amsterdam, Noth-Holland, 1979.
[10] J. Briançon and Ph. Maisonobe, Caractérisation de l’existence du

polynôme de Bernstein relatif. Progress in Mathematics, vol. 134, 1996.
215-236. Acte du colloque de La Rabida.
[11] J. Briançon and Ph. Maisonobe, Remarques sur l’idéal de bernstein

associé à des polynômes. Prépublications n◦ 560 de l’Université de Nice.
Mai 2002.
[12] J. Briançon, Suites régulières et relations. Hand written notes.
[13] O. Gabber, The integrability of the characteristic variety. American

Journal of Math., 103, (1981), p. 445-468.
[14] R. Godement, Théorie des faisceaux. Hermann, Paris, 1964.
[15] M. Granger and Ph. Maisonobe, A basic course on differential modules.

In: Eléments de la théorie des systèmes différentiels, tome 1: DX -
modules cohérents. Travaux en cours n◦ 46 Hermann, 1993, tome I,
pp. 103-167.
[16] M. Granger and M. Schulze, On the symmetry of b-functions of linear

free divisors. arXiv:0807.0560. to appear in the publications of RIMS
(Kyoto).
[17] I de Gregorio, David Mond and Christian Sevenheck, Linear free divi-
sors and Frobenius manifolds. Compositio Mathematica, volume 145,
n◦ 5, p. 1305-1350, 2009
www.pdfgrip.com
290
[18] P. Jeanquartier, Développement asymptotique de la distribution de

Dirac attaché à une fonction analytique. note aux CRAS t.271 (9 dec
1970)
[19] P. Jeanquartier, Transformation de Mellin et développements.

L’Enseignement Mathématique, Vol.25 (1979)
[20] M. Kashiwara, T. Kawai and M. Sato, Microfunctions and pseudo-

differential equations. Springer Lectures notes in Math., n◦ 287 (1973),
p.264-529.
[21] M. Kashiwara, B-functions and holonomic systems. Inventiones Math-

ematicae 38, 33-53 (1976).
[22] M. Kashiwara, On the holonomic systems of linear differential equa-

tions II. Inventiones Mathematicae 49, 121-135 (1978).
[23] T. Kimura, Introduction to Prehomogeneous Vector Spaces. Transla-

tions of Mathematical monographs, vol 215, Amer. Math. Soc. (2003)
[24] Lê Duñg Tráng and C.P. Ramanujan, Topologie des singularités des
hypersurfaces complexes. In Singularités à Cargèse, Astérisque n ◦ 7 et
8. SMF Paris (1973).
[25] Lê Duñg Tráng and C.P. Ramanujan, The invariance of Milnor’s num-
ber implies the invariance of the topological type. Am. Journ. of Math.
98(1976) n◦ 1 , 67-78.
[26] Lê Duñg Tráng and B. Teissier, Geometry of characteristic varieties.

This volume.
[27] B. Malgrange, Le polynôme de Bernstein d’une singularité isolée. Lec-

tures notes in Mathematics 459, 98-119 Springer (1974).
[28] B. Malgrange, Intégrales asymptotiques et monodromie. Annales sci-

entifiques de l’EN.S. 4e série, tome 7, no 3 (1974), p. 405-430.
[29] B. Malgrange, L’involutivité des caractéristiques des systèmes

différentiels et microdifférentiels. in Séminaire Bourbaki, Springer Lec-
tures notes in Math., n◦ 522 (1978).
[30] B. Malgrange, Polynômes de Bernstein-Sato et cohomologie

évanescente. in Analysis and topology on singular spaces, II, III
(Luminy, 1981) 243–267, Astérisque, 101-102, Soc. Math. France,
Paris, 1983.
www.pdfgrip.com
291
[31] J. Milnor and P. Orlik, Isolated singularities defined by weighted ho-

mogeneous polynomials. 1969, Topology 9, (1970), p.385-393
[32] Z. Mebkhout and L. Narvaez-Maccarro, La théorie du polynôme
de Bernstein-Sato pour les algèbres de Tate et de Dwork-Monsky-
Washnitzer. Annales scientifiques de l’ENS t.24, n◦ 2 (1991), 227-256.
[33] M. Sato, Theory of prehomogeneous vector spaces. noted by T. Shintani
in Japanese, Sugaku non ayumi 15, 85-157 (1970.)
[34] M. Sato and T. Shintani, On zeta-function associated with prehomoge-
neous vector spaces. Annals of Math., 100, p. 131-170 (1974).
[35] M. Sato, M. Kashiwara, and T. Kimura and T. Oshima, Micro-Local
Analysis of Prehomogeneous Vector Spaces. Inventiones Mathematicae
62 p 117 - 179, (1980)
[36] C. Sabbah, Proximité evanescente I. La structure polaire d’un D-
module. Compositio Math. 62 (1987) 283-319.
[37] J.-P. Serre Faisceaux algébriques cohérents. Annals of Mathematics, 61
(1955)
[38] T. Yano, On the theory of b-functions. Publ. RIMS, Kyoto Univ. 14
(1978) 111-202.
March 1, 2010 17:57 WSPC - Proceedings Trim Size: 9in x 6in 07˙malgrange
www.pdfgrip.com
292
DIFFERENTIAL ALGEBRAIC GROUPS
B. MALGRANGE
UFR de Mathématiques, UMR 5582,
Institut Fourier, Université Grenoble 1 -CNRS,
38402 Saint-Martin d’Hères Cedex, France
Given a family of algebraic groups G → S depending on a parameter S, one

studies systems of partial differential equations with variables in S and values
in G, which are compatible with the group structure. Two cases are especially
studied: the commutative case, and the simple case.
Keywords: Systems of partial differential equations. Differential groups; differ-

ential Lie algebras; connections
1. Introduction
Roughly speaking, the question is the following: given an algebraic variety
S over C, and a “family of algebraic groups π : G → S with parameter in
S”, study differential equations over the sections of π which are compatible
with the group structure, in a suitable sense.
The question is implicit in the work of E. Cartan on “infinite Lie
groups”: the differential groups are, in fact, a special case of the “intransi-
tive case” of his theory.1 But it has been developed in an independent way
by E. Kolchin2 and his students, namely P. Cassidy and A. Buium. It has
been used for a “parametric differential Galois theory” by P. Cassidy and
M. Singer3 (see other references later in the text).
My aim here is to give a rough idea of this theory. This work should not
be considered as original, except for the presentation of the definitions and
of some results.
2. Systems of partial differential equations

I will use the usual language of the theory of schemes (here, over C), as in4
or.5 A “variety” means a reduced scheme of finite type over C; in general I
will suppose it is separated and irreducible (if it is not the case I will say it
explicitly). If V is a variety, I will write often “a ∈ V ” for “a ∈ V (C)”.
www.pdfgrip.com
293
2.1 Suppose we are given two smooth (= nonsingular) algebraic varieties

X and S and a morphism π : X → S surjective and smooth (= everywhere
of maximal rank, or submersive), we want first to define “systems of partial
differential equations on the sections of π”.
For that purpose, one introduces, for all k ≥ 0, the space Jk (π) of jets of
order k of sections of π. This definition is well known in differential geom-
etry, and is developed in many treatises (however, in differential geometry,
it is generally used in an analytic or C ∞ context; but there is no problem
to transpose it in an algebraic context, and I will not insist on this point).
Then, one has the following definition
Definition 2.2. A system of p.d.e.’s on the sections of π is a projective

system {Yk } of closed subschemes of (Jk (π)) with the following properties
(i) For k ≥ 1, denote by pk the projection Jk (π) → Jk−1 (π). Then, one has
pk Yk ⊂ Yk−1 , and the map OYk−1 → pk∗ OYk is injective.
(ii) Yk+1 is contained in the first prolongation pr1 Yk of Yk .
To explain what mean these properties, by taking local charts, we can

make the following hypotheses:
(a) S is a finite étale covering of Cn − Z, Z a hypersurface (i.e. Z is defined
by f = 0, f ∈ C[s1 . . . sn ].
(b) X is contained as closed subvariety in S × Cp .
First, if X = S × Cp , the space Jk (π) is the space whose coordinates
are (s, xα α
j ), with s = (s1 , . . . , sn ) ∈ S, xj ∈ C, with 1 ≤ j ≤ p, α =
n
(α1 , . . . , αn ) ∈ N , |α| = α1 + · · · + αn ≤ k. Denote by Ak the space of
regular functions on Jk (π), i.e. Ak = O(S)[xα j ], O(S) the space of regular
functions on S. Then Yk is defined by an ideal Jk ⊂ Ak , and the conditions
(i) and (ii) mean the following
(i) One has Jk+1 ∩ Ak = Jk (note that we have an obvious inclusion Ak ⊂
Ak+1 )
(ii) One has Jk+1 ⊃ pr1 Jk , where pr1 Jk is defined in the following way.
First note that the vector fields ∂s∂ 1 , . . . , ∂s∂n are well defined on S (there
are vector fields on Cn , which can be lifted to S). Then, one defines the
derivation Di : Ak → Ak+1 by
∂f X ∂f
Di f = + xα+ε
j
i
, εi = (0, . . . , 1, . . . , 0)
∂si ∂xα j
Then pr1 Jk is the ideal of Ak+1 generated by Jk and Di Jk , 1 ≤ i ≤ n.

www.pdfgrip.com
294
The union J = ∪Jk is an ideal of A = limAk (= ∪Ak ). Property (i)

→
means that Jk = J ∩ Ak . Property (ii) means that J is a differential ideal,
i.e. that Di J ⊂ J , 1 ≤ i ≤ n.
Now, the case where X is a closed subvariety of S × Cp reduces im-
mediately to a special case of the preceding one: one considers just the
differential ideals which contain the equations of X (and, of course, their
derivatives).
2.3. A few remarks on these definitions

(i) I will not make the (very boring) verification that the property (ii),
as defined, is compatible with the change of charts. Without giving
details, I mention another definition of the first prolongation. One has
an inclusion Jk+1 → J1 (Jk (π)) (the “J1 ” is taken on the projection
“source”). Then, one has pr1 Yk = J1 (Yk )× J1 (Jk ) Jk+1 (I omit the “π”).
(ii) The reason of the definition of the Di is the following: if s 7→ {Xj (s)}
is a section of π, then, in the local coordinates considered above J
give a system of p.d.e. by substituting ∂ α Xj to xα α α1 αn
j (∂ = ∂s1 . . . ∂sn ).
α ∂ α
Then, by the chain rule, one has Di f (s, ∂ Xj ) = ∂si [f (s, ∂ Xj )].
(iii) Of course, this section is a solution of the system (in short, a solution
of J ) if, for all f ∈ J , one has f (s, ∂ α Xj ) = 0.
In general, with the definition 2.2, a solution is defined in the following
way: a section σ of π is a solution of Y = {Yk } iff, ∀ k, j k σ is a section
of Yk .
Note also the following fact: when we speak of solutions, we can take
algebraic sections, but also analytic germ of sections, (or even formal
sections). In this paper, unless otherwise stated explicitly, solution will
mean “germ of analytic solution”.
(iv) If we have a differential ideal J , its solutions depend only on its radical
J red = {f ; f k ∈ J for some k}. Now, by a well-known result by Ritt6
or,7 J red is also a differential ideal.
Applying this result in the general case 2.2, this means the following: if
Y = {Yk } is a system of p.d.e.’s, then Y red = {Ykred } is also a system of
p.d.e.’s (since, by definition, Ykred is defined by the radical of the sheaf
of ideals defining Yk ). Therefore, from now on, we can limit ourselves
to reduced systems of p.d.e.’s
3. Differential Groups
Suppose now that X is a “group over S”. More precisely, writing G instead
of X, it means the following: first, we assume, as before that G and S are
www.pdfgrip.com
295
smooth and that π : G → S is smooth and surjective. Furthermore, we

assume given 3 maps
(i) µ : G × G → G, compatible with the projections over S

S
(ii) λ : G → G, compatible with π
(iii) ε : S → G section of π
We assume also that these morphisms give respectively, in each fiber

π −1 (s), s ∈ S(C) the product, the inverse, and the unit of a group structure
(I leave to the reader to write the explicit conditions).
We give now a (reduced) system of p.d.e.’s Y = {Yk } on π. We are
looking to conditions ensuring the following property:
3.1 If σ and σ 0 are sections (i.e. germs of analytic sections) of π which are
solutions of Y , then µ(σ, σ 0 ) and λ(σ) are also solutions of Y . [I suppose,
of course, that σ and σ 0 are germs at the same point s ∈ S(C).]
An obvious, but important observation is the following: it is sufficient
to impose these conditions for the germs at points s ∈ V , where V is a
dense set in S(C), (dense, for the “transcendental”, or analytic topology);
actually the conditions to be a solution are closed: if they are satisfied on
V , they will be satisfied everywhere. A fortiori, it is sufficient to impose
those conditions at the points of a Zariski open dense set V .
In particular, the condition (3.1) will be satisfied by the following defi-
nition
Definition 3.2. Given π : G → S, and Y = {Yk } as before, we say that

they define a structure of differential (algebraic) group if there is a Zariski
open dense set U ⊂ S such that the restrictions Yk |U satisfy the following
properties
(i) The Yk |U are smooth, and the mappings Yk+1 |U → Yk |U smooth and
surjective.
(ii) Over U , Yk+1 is contained in the prolongation pr1 Yk |U , with equality
if k 0.
(iii) Yk |U is a subgroup over S of Jk (π)|U .
Only (iii) needs an explanation: the structure of group over S of G

extends in an obvious way to Jk (π). Therefore, the notion of subgroup over
U of Jk (π) is just the obvious one (with however, the slight modification
that we do not suppose necessarily Yk irreducible). It just means that the
“µ, λ, ε” of Jk (π) restricts to Yk |U .
www.pdfgrip.com
296
A priori, the conditions of (3.2) could seem rather restrictive. In fact,

they are reasonably general; the reasons imply some arguments of differen-
tial algebra which are beyond the scope of this course. A more general case,
for Lie pseudogroups instead of groups is explained in8 ). However, I do not
know if they are absolutely general, in other words if (3.1) implies (3.2).
4. Differential Lie Algebras

Given S as before, and π̇ : L → S a (finite dimensional) vector bundle on
S, a structure of Lie algebra of L over S is, of course, a bilinear map [·, ·] :
L× L → L satisfying the Jacobi identity. Now, linear differential systems on
S
S could be defined as before, but we would need “linear subspaces of Jk (π)
with variable fibers”. It is simpler to work in terms of DS -modules, where
DS is the sheaf of scalar linear differential operators on S (in the sequel,
I will say “DS -module” for “sheaf of DS -modules”). Denote by Diff(L, 1lS )
the sheaf of linear differential operators from sections of L to functions.
Then Diff(L, 1lS ) is a left DS -module, and a linear differential system on L
is simply defined by a coherent DS -submodule M of Diff(L, 1lS ).
The solutions of M are the sections of L annihilated by M . Here, we are
interested in their stability by [·, ·]. The question is simpler than the case
of groups, and we will find a necessary and sufficient condition.
For that purpose, we have to recall a generic property of D-modules.
To simplify the notations, I write D for DS and Diff for Diff(L, 1lS ). These
spaces are provided with filtrations, respectively Dk and Diffk , by differen-
tial operators of order ≤ k. Put Mk = M ∩ Diffk , and Nk = Diffk /Mk . Put
also D̄k = Dk /Dk−1 , D̄ = ⊕D̄k , and also N̄k = Nk /Nk−1 , N̄ = ⊕N̄k . Then
the result is the following.
Proposition 4.1. There is a Zariski-open dense subset U of S such that

the Nk |U and the N̄k |U are locally free over OS (= the sheaf of regular
functions on S).
The filtrations {Mk } and {Nk } are good, therefore N̄ is a coherent mod-
ule over D̄ (see the courses on D-modules in this school). Now, according
to a result of Grothendieck,9 (see exposé 4, lemma 6.7 of “generic flatness”)
there is a Zariski open dense subset U of S such that N̄ |U is flat, and even
free over OS |U . Then the N̄k are also flat; since they are coherent, it implies
that they are locally free. It follows easily that the Nk are also locally free.
For another argument, see.10 On Mk , this means that, over U “Mk is a
subbundle of Diffk over S”, of constant rank over each point 0 ∈ S.
www.pdfgrip.com
297
Now, Jk (L) is naturally the dual of Diffk : in local coordinates, if S is

étale over Cn −Z, and L ' S×Cp , then a section of Diffk is written Σaj,α ∂ α ,
aj,α functions on S, and a section of Jk (π), is a collection xα j of functions
on S; the duality is given by Σaj,α xα
j . In this duality, the orthogonal of Mk
(= the dual of Nk ) is Yk , the space of jets of order k of solutions of the
differential system defined by M .
The properties of Mk are translated in the following:
(i) The Yk are vector bundles (in the usual sense, of constant rank), and
the projections Yk+1 → Yk are surjective.
(ii) Yk+1 is contained in pr1 Yk with equality for k 0.
(This last property is the translation of the fact that one has Mk+1 ⊃
D1 Mk , with equality for k 0. I omit the details.)
Now, M being given, I fix a U with these properties. Note that the bracket
[·, ·] extends obviously to Jk (π).
Definition 4.2. In the preceding situation, we say that M defines a struc-

ture of differential Lie algebra if, for all k, Yk is stable by the bracket of
Jk (π). This implies obviously that, on S (not only on U ), the solutions are
stable by [·, ·]. Conversely, suppose that this is true, then, on U (4.2) is
satisfied because of the following property.
4.3. In the preceding situation, given a point s ∈ U , and a p ∈ Yk (C) over

s, there exists a germ of analytic solution of M at s whose jet of order k
at s is p.
This is a special case of a theorem of existence of analytic solutions for
p.d.e.’s. See the precise statement and the proof in,11 th.III.4.1.
4.4. Convention
Taking into account 3.2 and 4.2, we will now work “generically on S”, i.e.
we will identify structures which coincide on a (not precised) Zariski open
dense subset of S. For instance, if necessary, we can suppose that 3.2 or 4.2
are true, not only on U , but on S itself.
4.5. Lie groups and Lie algebras

Given a Lie group Γ, recall that one defines its Lie algebra as the tangent
space at identity Te Γ. It is supplied by a structure of Lie algebra by the
following trick: one identifies it with the space of left invariant vector fields
www.pdfgrip.com
298
on Γ; then the Lie bracket on vector fields furnishes the required structure
(one could also take the right invariant vector fields. The consideration of
the inverse g 7→ g −1 shows easily that one would obtain the opposite Lie
algebra law on Te Γ).
π
Similarly, if we have a group over S : G −→ S in the sense of §3, one
has a structure of Lie algebra over S on the bundle of vertical vectors on
G along ε (or equivalently, on the normal bundle along the section ε). I
denote it by Lie G.
Suppose that we have an algebraic subgroup G0 ⊂ G over S (I suppose
again G0 smooth, and the map π|G0 smooth surjective; but I dot not suppose
necessarily G0 connected). Then, to G0 correspond a Lie subalgebra L0 over
S, i.e. a subbundle (of “constant rank”, as explained after 4.1), stable by
the bracket of L.
If we have a Lie subalgebra L0 of L in the preceding sense, there does
not exist necessarily a corresponding G0 . There can exist also several G0 ;
but, in this case, their connected components of identity (= ε) are identical.
Now, suppose that G → S is provided with a structure of differential
group, as in 3.2. Restricting S if necessary to a Zariski open dense set, we
can suppose that the properties (i) to (iii) are satisfied on S.
We will provide Lie G = L with a structure of differential Lie algebra.
For that purpose, denote π 0 the projection L → S, and note that there is
a canonical isomorphism Lie Jk (π) ' Jk (π 0 ) (I leave the simple verification
to the reader). Then the collection of Lie Yk give a collection of subbundles
of Jk (π 0 ); and one verifies that they define a structure of differential Lie
algebra on L.
I omit the verification (there is essentially one point to verify, i.e. the
commutation of “take the first prolongation” and “take the Lie algebra”).
Of course, a special case is the trivial one, where we take Yk = Jk (π), and
then Lie Yk = Jk (π 0 ). In this case, the solutions are all the sections of π,
and similarly for the Lie algebra.
I will denote by G̃ this differential group, and I will consider the other
Y = {Yk } as subgroups of G̃. Following a terminology of Kolchin, I will say
that “Y is dense in G̃” if (in restriction to some U Zariski open dense set
of S), one has Y0 = G(= J0 (π)). The main purpose of these lectures is to
give, in several cases, a description of dense differential subgroups of G̃. If
there is no possible confusion, I write G instead of G̃.
4.6. N.B. As I said at the beginning the differential algebraic groups

have been defined and studied by E. Kolchin and his students. But their
www.pdfgrip.com
299
definition is technically quite different of the one given here, as, instead of
a variety S, they work on a differential field K, which they suppose gen-
erally differentially closed. Therefore, the translation of their results in the
present language is not always easy. In the first approximation, it would
be simpler to consider that one has two theories, close to each other, but
distinct. Some effort to connect them more closely would be useful.
5. First Examples
For a study of these examples in the point of view of differential fields,
see.12
5.1. Additive groups

I recall first some simple facts (see any course on algebraic groups). Let A1
be the “affine space over C”, i.e. C itself provided with the ring C[x] (in
other words, A1 = specC[x]). Now the additive group Ga is simply A1 with
the addition (x, x0 ) 7→ x + x0 .
The only closed subgroups of Gm a are its vector subspaces. This can be
seen as follows: first the Lie algebra is Cm (i.e. the same space) with the
trivial bracket; its Lie subalgebras are therefore the vector subspaces. To
this Lie algebra corresponds one connected subgroup, i.e. the same vector
space. To finish, we have to see that all the closed subgroups of Gm a are
connected; taking the quotient by the connected component of 0, i.e. by a
vector subspace we are reduced to prove the following result: Gm a has no
discrete (closed) subgroups; but such a group would be finite and Cm has
no finite subgroups.
Now, let S be a smooth algebraic variety as before, and take G = S×Gm a ,
with the obvious projection π over S. Applying similar arguments to the
Jk (π), one finds the following result: the differential algebraic subgroups of
G are the linear partial differential systems on S with value in Cm , in other
words the left DS -submodules of DSm . Of course, here, according to our con-
ventions, we identify two such systems which coincide on a Zariski dense
open subset of S. In other words, we get only D-modules at the generic
point of S.
π
5.2. Here is a remark which will be useful in the next sections. Let G −→ S
be a group over S, with G connected. Then, if Y = {Yk } is a differential
group dense in G, Y is connected (i.e. the Yk are connected). Therefore, it
www.pdfgrip.com
300
is determined by its Lie algebra. This is seen by recurrence on k, by using

the following fact: for all k ≥ 0, the kernel ker(Yk+1 → Yk ) is connected;
actually this is a subgroup over S of ker(Jk+1 (π) → Jk (π)). But this is a
family over S of additive groups, and therefore its subgroups are connected.
5.3. Multiplicative group

The multiplicative group Gm is specC[x, x−1 ]; the underlying space is there-
fore C∗ . The law is the multiplication (x, x0 ) 7→ xx0 .
I am interested in differential subgroups of S × Gm . First, it is easy
to see that the non-dense ones are simply the subgroups S × µp . µp the
group of p-roots of unity. To determine the dense subgroups, I first look
at the Lie algebra S × C (with trivial bracket). Then, the corresponding
differential Lie algebra is a linear differential system, in other words an ideal
(= sheaf of ideals) J ⊂ DS . According to 5.2 such a J gives at most one
differential subgroup, we have to determine those which give actually one
such subgroup.
To do that, I look first at the analytic situation: using the exact sequence
exp
0 → 2πiZ → C −→ C∗ → 0, one finds that S × C is a covering of S × C∗
(I write here S for S an , i.e. S considered as an analytic variety). Lifting
the system from S × Gm to S × C, we find a connected differential system
with Lie algebra J ; therefore, on S × C, our system is J itself. To descent
to S × C∗ , it is necessary that the 2πik, k ∈ Z, are solutions. This implies
that all the constants C ∈ C are solutions.
We will write it more explicitly. We can suppose that S is a finite étale
covering of Cn −Z, Z a hypersurface. Denote by s1 , . . . , sn the coordinates in
Cn . The condition on J is the following: if p = Σaα ∂ α is in J , then one has
a0 = 0. In other words, one has J ⊂ J0 , J0 the ideal generated by the ∂i .
Now, it is easy to see that this condition is sufficient. To do that, for
α = (α1 , . . . , αn ), define Dα = D1α1 . . . Dnαn , where the Di ’s are defined in
§2. If x is the coordinate in C∗ , we write Dα x = xα (I will write xi for xεi ).
Then, one verifies easily that the system obtained on S ×C∗ is generated
by the following equations: for each monomial α 6= 0, choose an αi 6=
0 and write α = β + εi . Then, to each p = Σaα ∂ α ∈ J , we associate
Σaα Dα (x−1 xi ). The differential ideal obtained will not depend on the choice
of β since one has Dj (x−1 xi ) = Di (x−1 xj ).
6. Family of Elliptic Curves

Let S be as before, a smooth algebraic variety over C, and let Γ be an
abelian variety (e.g. an elliptic curve). Writing Γ = Cn /Λ, Λ a lattice, we
www.pdfgrip.com
301
can easily describe the differential subgroups of G = S × Γ by the same

method used in 5.3. I will not give the details.
Here, we are interested in a more general case, where G is a family of
abelian varieties over S, varying with s ∈ S. This subject is very rich and
related to many other ones; here we will give only a very short introduction
to it.
For simplicity, I will limit myself to a family of elliptic curves, e.g. the
Legendre family. It is defined in the affine coordinates (y, z) by the equation
z 2 = y(y − 1)(y − s), s ∈ C − {0, 1} = S.
First I recall some basic facts on elliptic curves, and especially on this
family (see any course on elliptic curves)
(i) Fix s ∈ S; we add to the curve a point at ∞ [by passing to homoge-

neous coordinates Y, Z, U , the equation is Z 2 U = Y (Y − U )(Y − sU )
and the point is (0 : 1 : 0)]. Then, we have a projective non singular
curve GS of genus one; if we choose an origin, it has a structure of
commutative group. If we choose the origin at infinity, the law is given
by the following rule: a + b + c = 0 iff a, b, and c are on the same line
in P2 (C).
Now, varying s ∈ S, we get a group G over S, whose restriction to s
is Gs .
(ii) For s ∈ S, Gs , as analytic variety, is the quotient of C by a lattice (=
a discrete subgroup of rank 2) Λs .
In particular, only the ratio of the periods is well defined, modulo the
action of P G`(2, Z). But we have a canonical choice: we can choose t,
the coordinate of C in order to have dt = (inverse image of) dy dy
z , z
the differential of first kind on Gs . This is the celebrated theorem of
Abel on inversion of elliptic integrals.
R
Then Λs is the lattice of periods γ dy z , γ ∈ H1 (Gs , Z). Locally on
S, for the transcendental topology, we have a canonical isomorphism
H1 (Gs , Z) ' H1 (Gs0 , Z), and the periods “vary holomorphically with
s”.
Finally, the group law of Gs is the (quotient of) the addtion (t, t0 ) 7→
t + t0 .
(iii) Note also that, for general s and s0 , Gs is not isomorphic to Gs0 .
[The condition of isomorphism is ϕ(s) = ϕ(s0 ), ϕ(s) = (s1 − s + 2)3 /
[(s + 1)(2s − 1)(s − 2)]2 . This can be seen by reducing the equation,
by translation in y, to the form z 2 = y 3 + ay + b, in which case the
condition for isomorphism is a3 /b2 = a03 /b02 . I will not use this fact.]
www.pdfgrip.com
302
Now, to find a dense structure of differential group on G, we can try to do

as in 5.3. We have a linear system on Lie G = S ×C, i.e. a left ideal J of DS .
Considering now S×C as the covering of G, we look at conditions to descend
this system to G. To have that, it is necessary, when s varies, that the periods
are solutions of J . But the equation satisfied by the periods is classical:
this is the
special case of the Gauss hypergeometric00equation of parameters
1 1
,
2 2 , 1 . Explicitly, this equation is (6.1) s(1 − s)t + (1 − 2s)t0 − 41 t = 0.
Denote J0 the ideal generated by this equation. Then, the condition
is that J ⊂ J0 . Incidentally J0 is just the special case of what is called
“Gauss-Manin” or “Picard-Fuchs” equation for general families of varieties.
See the other courses of this school.
Now comes the difficult point, i.e. to prove that, under these conditions,
the differential system on G defined by J is algebraic. I will speak only
on J0 (in the other cases, there are probably similar results; but I have no
proof or reference).
For J0 , the result is proved in13 and.14 The first reference uses the
formalism of “crystals” by Grothendieck; the second one is more elementary.
The subject is closely related to the so-called “maximal extension by an
additive group” of an elliptic curve, or, more generally, an abelian variety. It
is also related to isomonodromic deformations of rank one bundles provided
with an integrable connection. See loc. cit.
Following,14 I will call “multiplicative Gauss-Manin connection” the dif-
ferential group on G corresponding to J0 .
The subject is also closely related to another one, seemingly very dif-
ferent: the theory of “algebraically integrable hamiltonian systems”. On
this subject, see e.g.15 or.16 Again, I will give only a simple example, the
pendulum.
This is the motion of a particle on a vertical circle, under the action of
the weight. With suitable normalizations, the circle is given in the (x, y)
2
plane by the equation x2 + y − 12 = 14 or x2 + y 2 − y = 0; the force
is vertical, along the y-axis. Then, t being the time, the energy integral is
x02 + y 02 + y = s.
Derivating the first equation, one finds xx0 + y − 21 y 0 = 0; eliminating
x and x0 , one gets y 02 = 4y(y − 1)(y − s). Further change of time reduces
finally to y 02 = y(y − 1)(y − s).
For s fixed, this is just the motion on the curve Gs , with dy dt = z or
dt = dy z ; therefore, it is linearized by the covering C → G s considered before
(if one prefers, it is given by the corresponding Weiestrass p function).
www.pdfgrip.com
303
The variation of period when s varies satisfies the equation (6.1) (in
general, in the theory of integrable systems, the authors omit to consider
this point).
The multiplicative Gauss-Manin connection can also be interpreted as
the differential Galois group of the hamiltonian system of the pendulum; I
hope to explain that in a future publication.
7. Connections
Our next aim is to study the simple (or semi simple) case. For that purpose,
some preliminaries are necessary.
7.1. Connections in the sense of Ehresmann (or “foliated

bundles”)
This is simpler to explain first in the analytic case. First I do it locally; let
U (resp. V ) be an open subset for the usual (or “transcendental”) topology
of Cn (resp. Cp ), and let π be the projection U × V → U . Denote the coor-
dinates on U (resp. V ) by s1 , . . . , sn (resp. x1 , . . . , xp ). Then a π-connection
(in the sense of Ehresmann) is simply a system of p.d.e.’s of the type
∂xj /∂si = aij (s, x), aij holomorphic on U × V
This system is also expressed in the following way:

∂
P ∂
(i) By the family of vector fields ∂si + aij ∂x j
= ξi .
j P
(ii) By the family of 1-form orthogonal to the ξi : ωj = dxj − aij dsi .
i
The integrability condition is just the Frobenius condition for the ξi ’s, or
the ωj ’s. Due to the special form of ξi , it means just that [ξi , ξj ] = 0 ∀ i, j.
Explicitly, one has

∂ajk ∂aik X ∂ajk ∂aik
− + ai` − aj` =0
∂si ∂sj ∂x` ∂x`
`
This is equivalent to give a foliation on U × V with leaves étale over U or,

as one says, a “foliated bundle”.
Now, this notion can be defined directly without local coordinates: let
S and X be smooth analytic C-varieties and π : X → S a surjective sub-
mersion. Let a ∈ X, with π(a) = s; then we have an exact sequence
www.pdfgrip.com
304
π0
0 → va X → Ta X −→ Ts S → 0
(π 0 the tangent map to π; v mean here “vertical”).
Then, a connection can be defined as a splitting of this exact sequence,
depending analytically on a ∈ X. One sees at once that this is equivalent,
in local coordinates, to the notion defined above. Considering this splitting,
either as a lifting of Ts S to Ta X, or as a projection of Ta X to va X, we get the
interpretations (i) and (ii) above. In particular, the second interpretation
gives a form Ω on X with values in vertical vectors.
We say that this connection is “flat” or “integrable” or “without curva-
ture” if the integrability condition considered above is satisfied. Although
I will not need this fact, I mention that the integrability condition can be
written [Ω, Ω] = 0, where [·, ·] is the Nijenhuis bracket on vector valued
forms; one could also, more generally, define the curvature as [Ω, Ω].
To end this section, note that the notion of connection and its integra-
bility can be defined similarly in the algebraic context: here, S and X are
smooth algebraic varieties over C, and π : X → S is smooth and surjective.
Of course, here, we require that the data defining the connection, e.g. the
form Ω, are algebraic. Again, a flat connection defines a foliated bundle
(this is just a different name for the same notion).
7.2. Connections and differential equations

An algebraic connection (flat or not) on π : X → S is a special kind of first
order differential equation: more precisely, they are the closed subvarieties
Y1 ⊂ J1 (π) such that the projection Y1 → X is an isomorphism.
The condition of flatness is equivalent to the fact that the first prolon-
gation Y2 = pr1 , Y1 is surjective over Y1 ; it is sufficient to verify this in
local coordinates for the underlying analytic connection, which is almost
obvious.
Then, if the connection is flat, all the prolongations (defined by recur-
rence by Yk+1 = pr1 Yk ) are isomorphic to each other by the projections
Yk+1 → Yk . Again, it is sufficient to prove this in local analytic coordi-
nates; in that case, it follows from the following well-known fact: with the
notations of the beginning of 7.1, if the integrability condition is satisfied,
by a fibered local analytic change of coordinates, or can reduce the system
to the case aij = 0.
Then we can identify the connection with the differential system {Yi },
with Y0 = X. The solutions of the connection, called also “horizontal sec-
tions” coincide with the solutions of this differential system.
www.pdfgrip.com
305
7.3. From now on, the connections will be supposed flat

We will now consider the case where, on X, there is some additional struc-
ture; we want that the connection be “adapted” to this structure. We will
not try to give general definitions, but only examples.
7.3.1. Suppose X → S is a vector bundle. Then we will say that the connec-
tion is adapted if the corresponding differential system is a linear system
(= the Yk are linear subspaces of Jk (π).) It is sufficient that Y1 is a linear
subspace of Jj (π).
To give explicit conditions, I suppose that S is an étale covering of

C − Z, Z a hypersurface, and that X = S × Cp .
n
Then the equations of the connection can be written

∂x
+Ai (s)x = 0, x = (x1 , . . . , xp )T , Ai with values in End(Cp ) = G`(p, C).
∂si
If we write Ω = ΣAi (s)dsi , the integrability condition is dΩ + Ω ∧ Ω = 0.
7.3.2. Suppose further that X is a Lie algebra over S. We will say that
the connection is compatible with the structure of Lie algebra if the corre-
sponding differential system is a Lie subalgebra of X → S. It is necessary
and sufficient to require that Y1 is a Lie subalgebra over S of J1 (π).
Explicitly, in the case considered above, the condition is that Ai (s) is,
for s ∈ S, a derivation of the Lie algebra, i.e. Ai (s)[x, y] = [Ai (s)x, y] +
[x, Ai (s)y].
7.4. We are now interested in the case where X = G is a group over S,

in the sense of §3. We will say that a connection on π : G → S is adapted
(or compatible) with the group structure if the corresponding differential
system is a differential subgroup of G (note that, here, it is necessarily
dense). Again here, the following fact is easy to verify: the connection is
compatible with the group structure iff the corresponding subvariety Y1 ⊂
J1 (π) is a subgroup of J1 (π) over S.
If one has G = S × Γ, Γ an algebraic group over C, the connection can
be written dγ − Ω(γ, s), Ω regular on G with values in the vertical vector
fields. The condition of compatibility is Ω(γγ 0 , s) = γΩ(γ 0 , s) + Ω(γ, s)γ 0 .
(This expression is clear if the group is linear. Otherwise, one should inter-
pret γ· as “left translation by γ” and ·γ 0 as “right translation by γ 0 ”.)
If we prefer to have a form with values on Lie Γ, write γ −1 dγ −
−1
γ Ω(γ, s), etc.
www.pdfgrip.com
306
7.4.1. Example [cf.14 ]

Take Γ = Gm × Ga = C∗ × C with variables (y, x), and S = C. Suppose
the connection is given by (dx − xds, dy − xyds). It is easy to verify its
compatibility with the group structure.
If we integrate these equations or, as one says, if we take the flow of the
connection we get x = x0 es−s0 , y = y0 ex−x0 . We get an analytic automor-
phism of Γ, which is not algebraic.
This is related to the following fact: for the group Γ, the functor AutΓ
is not representable (in particular, there are infinitesimal automorphisms
which are not in the Lie algebra of the group AutΓ, cf.14 for a systematic
discussion of this problem.
7.4.2. In the same spirit, let me mention briefly some other facts, also dis-
cussed by Buium in.14
Let G → S be a connected group over S, and let Ω be a flat connection
on G → S, compatible with the group structure.
(i) The fibers Gs are all analytically isomorphic. This follows from a result
by Hamm, see.14
(ii) If G is affine over S, then the Gs are algebraic. This follows from
a theorem of Hochschild-Mostow,17 which says that two connected
affine algebraic C-groups analytically isomorphic are also algebraically
isomorphic (but it does not mean that all analytic isomorphisms are
algebraically isomorphic, cf.18 ).
(iii) If G is not affine, the result is not true. Actually we have seen implicitly
a counter-example in §6: the “multiplicative Gauss-Manin connection”
considered on G (notations of §6) is not a connection on G. But it is a
second order differential operator, which can be considered as a con-
nection on J1 (π) = G1 . Each fiber is a commutative group, extension
of G by an additive group (its “universal” extension, cf. loc. cit.). If
Gs is not isomorphic to Gs0 , G1s is not isomorphic to G1s0 (use the
Chevalley-Barsotti theorem: the connected maximal affine subgroup is
unique, and therefore also the quotient, which is abelian). But they are
analytically isomorphic according to Hamm’s theorem. Actually, it is
not difficult to see that they are all analytically isomorphic to (C∗ )2 .
7.5. Connections on principal bundles

Let Γ be a connected algebraic group over C, and let π : P → S be a
(right) principal bundle, or, if one prefers, an S × Γ torsor. A connection
www.pdfgrip.com
307
on P is compatible with the structure of principal bundle, by definition, if

the corresponding variety Y1 ⊂ J1 (π) is stable by the obvious action of Γ
on J1 (π).
If the bundle is trivial, i.e. P = S × Γ, with the obvious right action of
Γ, the form of the connection Ω verifies Ω = dγ + αγ, α a form with values
on Lie Γ (for the notation αγ, see the beginning of 7.4). The integrability
condition can be written dα + α ∧ α = 0, in the case where Γ is linear. In
general, one should write 12 [α, α] instead of α ∧ α.
Now comes a remark which will be used in §8. In the preceding situation,
denote by Aut P the bundle of fibered automorphisms of P commuting
with the action of Γ. Then the connection Ω gives a connection on Aut P
compatible with its group structure.
I shall verify it only in the case where P = S × Γ, which is the case I will
use. Then, one has Aut P = S ×Γ since the automorphisms of Γ commuting
with right translations are just left tranlations. If Ω = dγ + αγ, then the
connection on Aut P is given by dγ + γα − αγ. It is immediate to verify
directly that the product of horizontal sections is a horizontal section.
The corresponding connection on Lie Aut P is given by “the same for-
mula”, better written here dγ + [γ, α]. One has the following result.
Theorem 7.6. If Γ is semi-simple and connected, all connections on S × Γ
(resp. S× Lie Γ) compatible with the group structure (resp. with the Lie
algebra structure) are obtained in this way.
It is sufficient to prove the result for Lie algebras (since connected dif-
ferential subgroups, in particular those coming from a connection on S × Γ
are determined by the Lie algebra). If we write L = Lie Γ, a connection
compatible with the structure of Lie algebra on S × L is given by d` + α, α
a form on S with values in EndL, satisfying α[`, `0 ] = [α(`), `0 ] + [`, α(`0 )].
Now the result follows from the fact that all derivations of L are interior.
7.7. Connections with respect to a foliation

We need a more general definition of connections. Let us give a foliation
of S, i.e. a subsheaf N ⊂ Ω1s , which is “a subbundle” (i.e. Ω1s /N is locally
free; cf. 4.1) and satisfies the Frobenius condition dN ⊂ N ∧ Ω1s .
Let F = N ⊥ be the subbundle of T (S) orthogonal to N . Then, roughly
speaking an F -connection on π : X → S is something like a connection,
but one derives only along F .
In the analytic case, in suitable local coordinates, F is defined by ds1 =
· = dsk = 0. With the same notations as in 7.1, this means that we will have
www.pdfgrip.com
308
only derivations with respect to sk+1 , . . . , sn , i.e. a system of equations of

the type
∂xj
= aij (s, x), k + 1 ≤ i ≤ n, 1 ≤ j ≤ p.
∂si
In other words, s1 , . . . , sk act here only as parameters.
In the global analytic or algebraic case π : X → S, we have to consider
TF (X) = π 0−1 (F ) (π 0 , the differential of π), and to consider splittings of
the sequence 0 → v(X) → TF (X) → F → 0.
I will indicate briefly how the considerations of the preceding sections
of §7 generalize here. First, the integrability condition is generalized in an
obvious way.
Again, an F -connection defines a (smooth) subvariety Y1 of J1 (π). If the
integrability condition is satisfied, then all the prolongations Yk are smooth
and surjective on each other. Therefore, a flat (= integrable) F -connection
defines a differential system on π : X → S.
The compatibility with a structure of group, Lie algebra, or principal
bundle is defined as above. Finally the theorem 7.6 is still true with “F -
connection” instead of “connection”. The verification is not difficult, and I
leave it to the reader.
8. Simple Groups
For the general theory of linear algebraic groups, see e.g.19 or.20 Here, the
result is the following.
Theorem 8.1. Let S be a smooth algebraic variety over C as before, and

let L be a simple Lie algebra over C. Then the only structures of differential
Lie algebra on S × L, dense in S × L, are the connections with respect to
foliations of S.
As for groups, “dense” means, of course, that there are no equations of

order 0.
Suppose now that Γ is a connected algebraic group over C, Lie with
Γ = L. Then, combining 8.1 and 7.6 (extended to F -connections), we have
a description of dense differential groups on S × Γ: all of them are F -
connections for a suitable F . In particular, we have a bijection between
dense differential Lie algebras on S × L, and dense differential Lie groups
of S × Γ.
The proof follows an argument by Kiso.21 See historical remarks 8.3.
I will follow the notations of §4 (with here S × L instead of L). Suppose
www.pdfgrip.com
309
that we have a dense differential Lie algebra given by a DS -module M ⊂

Diff(S × L, 1ls ) = “Diff”. We consider its induced filtration Mk . As in §4,
we write Nk = Diffk /Mk , N̄k = Nk /Nk−1 .
By generic flatness (prop. 4.1), we can suppose, by restricting S, that
the Nk and N̄k are locally free OS -modules. On the other hand, by restrict-
ing again S, we can suppose that it is finite and étale over Cn − Z, Z a
∂
hypersurface. Then N̄ = ⊕N̄k is a OS [ξ1 , . . . , ξn ]-module, with ξi = gr ∂s i
.
∗
Consider now the duals Nk over OS of Nk ; it can be identified with
Yk ⊂ Jk (π), the space of k-jets of solutions of M (here, π is the projection
S × L → S). Similarly, the dual of N̄k is Ȳk = ker(Yk → Yk−1 ). By the
hypothesis of density, we have Ȳ0 = Y0 = S × L.
Now, we have two structures on Ȳ = ⊕Ȳk
(a) A structure of OS [ξ1 , . . . , ξ1 ]-module, obtained by duality over OS of
the structure of N̄ . It is graded by the opposite degree −k.
(b) A structure of graded Lie algebra, i.e. [Ȳk , Ȳ` ] ⊂ Ȳk+` , obtained from
the structure of Lie algebras of the Yk ’s.
Now, the proof goes as follows.
(i) Ȳ1 ⊂ J¯1 (π) = L ⊗ T ∗ S has the following form (after restricting S):
There exists E, vector subbundle of T ∗ S such that Ȳ1 = L ⊗ E.
This follows from the next lemma
Lemma 8.1. Let V be a vector space over C, and consider the rep-
resentation ad ⊗ id of L on L ⊗ V . Then the invariant subspaces of
L ⊗ V are the L ⊗ W , W a vector subspace of V .
Since L is simple, the representation is completely reducible. Therefore,
it suffices to study the irreducible case.
Then, let F be an irreducible invariant subspace of L ⊗ V ; choose a
basis v1 , . . . , vp of V . For a ∈ F , write a = `1 ⊗ v1 + · · · + `p ⊗ vp ,
and denote ui the map a 7→ `i from F to L. As F is irreducible, ui is
identically 0, or is an isomorphism. Now, if ui and uj are isomorphisms,
according to Schur lemma, one has uj u−1 i = λ · id, λ ∈ C. The result
follows at once.
(ii) One has Ȳk = L ⊗ S k E, as subspace of J¯k (π) = L ⊗ S k (T ∗ S).
Again the proof can be seen fiber by fiber. I will give it for k = 2;
the general case is similar. On one side, the module structure implies,
ξi Ȳ2 ⊂ Ȳ1 = L ⊗ E, 1 ≤ i ≤ n. I leave to the reader to verify that this
means Ȳ2 ⊂ L ⊗ S 2 E. On the other side, the fact that [L, L] = L and
[Ȳ1 , Ȳ1 ] ⊂ Ȳ2 imply Ȳ2 ⊃ L ⊗ S 2 E.
www.pdfgrip.com
310
These results show that our system of equations D is generated by the

first order ones; choose a basis of vector fields X1 , . . . , Xq orthogonal
to L; then a basis of the first order equations is Xi + ai , ai functions
on S with values in EndL (for Xi fixed, there is a unique ai , because
there are no equations of order 0). To end the proof, we have two facts
to prove
(iii) The Xi ’s (or E) verify the Frobenius condition. Then our system is an
F connection, F the foliation defined by E.
(iv) This connection is integrable.
One could prove (iii) by using the “integrability of characteristics” (see

the other courses of this school). But it is simpler to prove simultaneously
(iii) and (iv): consider [Xi + ai , Xj + aj ]: this is a first order equation which
belongs to our system. Therefore, it has to be a linear combination on OS
of the Xi + ai . Both results follow immediately; this ends the proof of 8.1.
8.2. Generalizations
Let me indicate them very briefly.
(i) Let Γ be a connected semi-simple group over C. Then, the dense dif-
ferential group structures on Γ can be obtained as follows.
First, one looks at differential Lie algebra structures on Lie Γ =
L1 ×· · ·×Lq , Li simple. Arguments similar to 8.1 show that these struc-
tures are products of Fi -connections on S × Li (the Fi can be different
from each other). Then, if Γ0i is the connected component of the group
Aut Li , this differential structure can be lifted to Γ0 = Γ01 × · · · × Γ0p .
Call Y 0 = {Yk0 } this structure. Finally, Γ is a finite covering of Γ0 by
the adjoint action; then one lifts Y 0 to S × Γ by Yk = Yk0 × Γ [note that
Γ0
one has Jk (π) = Jk (π 0 ) × Γ, with π (resp. π 0 ) the projection S × Γ → S
Γ0
(resp. S × Γ0 → S 0 )]. I leave the details to the reader.
(ii) More generally, let G be a semi-simple group defined on C(S), the field
of rational functions of S. Then, if we replace C(S) by a suitable finite
extension, i.e. if we replace S by S 0 , étale finite covering of U ⊂ S,
Zariski dense, we are reduced to (i). However, to finish, we would have
to examine descent conditions to go back from S 0 to S. I will not
examine this here.
www.pdfgrip.com
311
8.3. Historical remarks

Theorems type 8.1 seem to have been considered for the first time by E.
Cartan.1 They come in his theory of “infinite Lie groups”, as “intransitive
simple groups”. Here is what he says:
“Les groupes infinis simples qui ne sont isomorphes à aucun groupe transitif
. . . se partagent en deux catégories
(a) Les groupes simples proprement dits: on les obtient en prenant un
groupe simple transitif (fini ou infini), et en faisant dépendre de la
manière la plus générale possible les éléments arbitraires de p variables
invariantes par le groupe.
(b) Les groupes simples improprement dits . . . ”
My comment: Cartan considers only the local analytic situation with re-
spect to the parameter (here, denoted S). Now, locally, a flat F -connection
is simply a (trivial) connection with respect to some of the variables. Cartan
neglects them, seeming to consider that “this does not change the group”.
This question has been re-examined by Kiso.21 He gives also a local
analytic statement with respect to the parameters. However, his method
gives the more precise result stated here.
Note that Kiso, and Morimoto22 study also the case of intransitive “infi-
nite groups”, whose transitive part is simple, or even primitive. The results,
as claims Cartan, are similar. In the algebraic case, they could probably also
be expressed in terms of F -connections.
Finally, the question of simple, or, more generally, semi-simple differ-
ential algebraic groups has also been studied by Cassidy.12,23 The results
are very similar. But, as I said in 4.6, I found difficult to compare precisely
these results with those of Kiso, mainly because they are stated in the very
different context of differentially closed differential fields.
References
1. E. Cartan, Ann. Ecole Normale Supérieure 26, 93 (1909).
2. E. Kolchin, Differential Algebraic Groups (Academic Press, New York, 1985).
3. P. Cassidy and M. Singer, Galois theory of parameterized differential equa-
tions, Differential equations and quantum groups, IRMA Lecture Notes in
Math. and Th. Physics 99, Strasbourg (2007) European Math. Soc.
4. R. Hartshorne, Algebraic Geometry, (Graduate Texts in Maths. 52, Springer-
Verlag, 1977).
5. D. Mumford, The Red Book of Varieties and Schemes, Lecture Notes in Math.
1358 (Springer-Verlag, 1988).
6. J. F. Ritt, Amer. Math. Soc. Coll. Publ. 33 (1950) and Dover (1965).
www.pdfgrip.com
312
7. E. Kolchin, Differential Algebra and Algebraic Groups (Academic Press, New

York, 1973).
8. B. Malgrange, Monographie de l’Ens. Math. 38, 465 (2001).
9. A. Grothendieck, Séminaire de Géométrie Algébrique, Buressur-Yvette
(1960-1961).
10. H. Goldschmidt, Inv. Math. 9, 165 (1970).
11. B. Malgrange, Systèmes différentiels involutifs, Panoramas et synthèses,
No.19 Soc. Math. France (2005).
12. P. Cassidy, Amer. J. Math. 94, 891 (1972).
13. B. Mazur and W. Messing, Universal extensions and one dimensional crys-
talline cohomology, Lecture Notes in Math. 330 (Springer-Verlag, 1974).
14. A. Buium, Differential algebraic groups of finite dimension, Lecture Notes in
Math, 1506 (Springer-Verlag, 1991).
15. F. Calogero, Classical many-body problems amenable to exact treaments,
Lecture Notes in Physics, Monographs (Springer-Verlag, 2001).
16. O. Babelon, D. Bernard and M. Talon, Introduction to Classical Integrable
Systems, Cambridge Monographs on Math. Physics (Cambridge University
Press, 2003).
17. G. Hochschild and D. Mostow, Amer. J. Math. 93, 111 (1961).
18. G. Hochschild and D. Mostow, J. Algerbra 25-1, 146 (1973).
19. A. Borel, Linear Algebraic Groups, 2nd edition (Graduate Texts in Maths.
126, Springer-Verlag, 1991).
20. T. A. Springer, Linear Algebraic Groups, (Progress in Maths. 9, Birkhäuser,
1981).
21. K. Kiso, Jap. J. Math. 5-1, 101 (1979).
22. T. Morimoto, J. Math. Soc. Japan 29, 35 (1977).
23. P. Cassidy, J. Algebra 121-1, 169 (1989).

Algebraic Approach To Differential Equations

Uploaded by

Copyright:

Available Formats

Algebraic Approach To Differential Equations

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

Algebraic Approach To Differential Equations

Uploaded by

Copyright:

Available Formats

www.pdfgrip.

7290 tp.indd 1 6/9/10 8:42 AM

This page intentionally left blank

7290 tp.indd 2 6/9/10 8:42 AM

British Library Cataloguing-in-Publication Data

ALGEBRAIC APPROACH TO DIFFERENTIAL EQUATIONS

Julia - Algebraic Approach to Diff Eqns.pmd 1 3/24/2010, 11:03 AM

Lê Dũng Tráng

The school on “Algebraic Approach to Differential Equations” was organized

Lê Dung Tráng

Modules Over the Weyl Algebra 52

Geometry of Characteristic Varieties 119

Singular Integrals and the Stationary Phase Methods 136

Hypergeometric Functions and Hyperplane Arrangements 210

Bernstein-Sato Polynomials and Functional Equations 225

Differential Algebraic Groups 292

∗ Partially supported by MTM2007–66929 and FEDER.

dimension 1 stated in the paper13 with some complements. In section 12

We would like to thank the organizers of the I.C.T.P. school, specially

with B1 = · · · = Bn−1 = 0, Bn = b and

Theorem 1.1. Let U ⊂ C be an open disc centered at the origin, A a

is bijective. Moreover, when B = 0 the application above is an isomorphism

Corollary 1.1. Let U ⊂ C be an open disc centered at the origin and

y(0) = v0 , y (1) (0) = v1 , . . . , y (n−1) (0) = vn−1 .

2. Sheaves of Holomorphic Functions

Definition 2.1. The sheaf of holomorphic functions on an open set U ⊂ C

Property (b) above means that for a function, being holomorphic is a

Definition 2.2. A subsheafb of OU is the data F consisting of a vector

If F, F0 are subpresheaves of OU , we say that F ⊂ F 0 if F(V ) ⊂ F0 (V )

sheaf of holomorphic functions and its subsheaves.

a subsheaf of OU 0 , that we call the restriction of F to U 0 . Let us also note

Exercise 2.1. (1) Let F be the data defined by

F(V ) = {f : V → C | f is a constant function} ⊂ OU (V ),

for each open set V ⊂ U . Prove that F is a subpresheaf of OU which is

CU (V ) = {f : V → C | f is a locally constant function} ⊂ OU (V ),

for each open set V ⊂ U , is a subsheaf of OU .

Exercise 2.2. Let U ⊂ C be an open set, Σ ⊂ U a closed discrete set and

F(V ) = {f ∈ OU (V ) | f = 0 on a neighborhood of any point p ∈ Σ ∩ V }.

Prove that F is a subsheaf of OU , which will be denoted by j! OU \Σ .

F(V ) = {f ∈ CU (V )| f = 0 on a neighborhood of any point p ∈ Σ ∩ V }.

Prove that F is a subsheaf of OU , which will be denoted by j! CU \Σ .

Exercise 2.3. Let F ⊂ OU be a subpresheaf. Prove that:

(1) There is a unique subsheaf F + ⊂ OU such that:

(2) Prove that F is a subsheaf of OU if and only if F = F+ .

(3) Prove that (F|U 0 )+ = F+ |U 0 for any open subset U 0 ⊂ U .

Definition 2.3. An endomorphism of OU , L : OU → OU , is the data

Let us denote by End(OU ) the set of endomorphisms L : OU → OU .

L ∈ End(OU ) 7→ L|U 0 ∈ End(OU 0 )

Example 2.1. (a) The family of linear maps

(b) If h ∈ OU (U ), then the family of linear maps

f ∈ OU (V ) 7→ (h|V )f ∈ OU (V ), V ⊂U open subset,

is an endomorphism that will be denoted by h : OU → OU .

(c) Example (b) gives rise to a ring homomorphism OU (U ) → End(OU ),

Exercise 2.5. Let L : OU → OU be an endomorphism and let us consider

In this way, ker L is an object which simultaneously encodes the solutions

Definition 2.5. A (holomorphic) linear differential operator on U is an

The set of (holomorphic) linear differential operators on U will be de-

Exercise 2.9. (1) Prove that D(U ) is a sub-C-algebra of End(OU ).

Remark 2.3. The family {D(V ), V ⊂ U open subset}, as in remark 2.1,