Sie sind auf Seite 1von 233

CONTEMPORARY

MATHEMATICS
555

Commutative Algebra
and Its Connections
to Geometry
Pan-American Advanced Studies Institute
August 3 –14, 2009
Universidade Federal de Pernambuco, Olinda, Brazil

Alberto Corso
Claudia Polini
Editors

American Mathematical Society


Commutative Algebra
and Its Connections
to Geometry
This page intentionally left blank
CONTEMPORARY
MATHEMATICS
555

Commutative Algebra
and Its Connections
to Geometry
Pan-American Advanced Studies Institute
August 3–14, 2009
Universidade Federal de Pernambuco, Olinda, Brazil

Alberto Corso
Claudia Polini
Editors

American Mathematical Society


Providence, Rhode Island
Editorial Board
Dennis DeTurck, managing editor
George Andrews Abel Klein Martin J. Strauss

2010 Mathematics Subject Classification. Primary 05C90, 13C40, 13D02, 13D07, 13D40,
14C05, 14J60, 14M12, 14N05.

Photographs courtesy of Alberto Corso.

Library of Congress Cataloging-in-Publication Data


Pan-American Advanced Studies Institute (2009 : Universidade Federal de Pernambuco)
Commutative algebra and its connections to geometry : Pan-american Advanced Studies In-
stitute, August 3–14, 2009, Universidade Federal de Pernambuco, Olinda, Brazil / Alberto Corso,
Claudia Polini, editors.
p. cm. — (Contemporary mathematics ; v. 555)
Includes bibliographical references.
ISBN 978-0-8218-4959-0 (alk. paper)
1. Combinatorial group theory—Congresses. 2. Algebra, Homological—Congresses. 3. Ge-
ometry, Algebraic—Congresses. I. Corso, Alberto. II. Polini, Claudia, 1966– III. Title.

QA182.5.P36 2009
512.44—dc23
2011027283

Copying and reprinting. Material in this book may be reproduced by any means for edu-
cational and scientific purposes without fee or permission with the exception of reproduction by
services that collect fees for delivery of documents and provided that the customary acknowledg-
ment of the source is given. This consent does not extend to other kinds of copying for general
distribution, for advertising or promotional purposes, or for resale. Requests for permission for
commercial use of material should be addressed to the Acquisitions Department, American Math-
ematical Society, 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can
also be made by e-mail to reprint-permission@ams.org.
Excluded from these provisions is material in articles for which the author holds copyright. In
such cases, requests for permission to use or reprint should be addressed directly to the author(s).
(Copyright ownership is indicated in the notice in the lower right-hand corner of the first page of
each article.)

c 2011 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Copyright of individual articles may revert to the public domain 28 years
after publication. Contact the AMS for copyright status of individual articles.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 16 15 14 13 12 11
Contents

Preface vii
List of Mini Courses at PASI 2009 xi
List of Talks at PASI 2009 xiii
PASI 2009 Photos xv
Hilbert depth of powers of the maximal ideal
Winfried Bruns, Christian Krattenthaler, and Jan Uliczka 1
A note on reductions of monomial ideals in k[x, y](x,y)
C.-Y. Jean Chan and Jung-Chen Liu 13
Plücker-Clebsch formula in higher dimension
Ciro Ciliberto and Vincenzo Di Gennaro 35
Invariants of ideals generated by pfaffians
Emanuela De Negri and Elisa Gorla 47
Hilbert polynomial and the intersection of ideals
Juan Elias and Jordi Martı́nez-Borruel 63
Polynomial vector fields with algebraic trajectories
Viviana Ferrer and Israel Vainsencher 71
Minimal free resolutions for certain affine monomial curves
Philippe Gimenez, Indranath Sengupta, and Hema Srinivasan 87
Uniform bounds for Hilbert coefficients of parameters
Shiro Goto and Kazuho Ozeki 97
Absolute integral closure
Craig Huneke 119
A property of the Frobenius map of a polynomial ring
Gennady Lyubeznik, Wenliang Zhang, and Yi Zhang 137
A note on the variety of pairs of matrices whose product is symmetric
Mahdi Majidi-Zolbanin and Bart Snapp 145
Combinatorics of symbolic Rees algebras of edge ideals of clutters
José Martı́nez-Bernal, Carlos Renterı́a-Márquez,
and Rafael H. Villarreal 151

v
vi CONTENTS

Reconciling Riemann-Roch results


Paul Roberts and Anurag K. Singh 165
Hilbert functions of Cohen-Macaulay local rings
Maria Evelina Rossi 173
Some homological properties of almost Gorenstein rings
Janet Striuli and Adela Vraciu 201
Preface

Commutative algebra has several sources – among them number theory, invari-
ant theory and algebraic geometry. Besides these classical sources, the field has
undergone a striking evolution over the last quarter of a century, reaching out to
applicable disciplines such as combinatorics, computer algebra, optimization and
statistics. Commutative algebra is taught, and actively pursued as a research field,
not only in developed countries, but also in several developing countries – in partic-
ular in Brazil and other Latin American countries such as Argentina and Mexico.
In Brazil the area has since its inception been associated to algebraic geometry.
This association has been very fruitful and, in particular, it has produced a joint
research group (ALGA) involving partners in the scientific centers of the country.
Many of the results published in this volume were presented during the NSF-
sponsored Pan-American Advanced Studies Institute (PASI) in Commutative Al-
gebra and its Connections to Geometry held in Olinda (Brazil) during the period
August 3-14, 2009, under the patronage of the Universidade Federal de Pernam-
buco. This institution is located in the center of the Brazilian Northeast, a vast,
less developed region of the country. It has an established mathematics department
with a long tradition as a research institution that has produced, for instance, a
number of mathematicians currently holding positions at leading US universities.
The goals of this event were to reinforce the above mentioned successful Brazil-
ian partnership throughout less privileged regions of Brazil and of Latin America in
general by introducing young mathematicians from Brazil and other Latin Ameri-
can countries to fundamental techniques and recent developments in the field and
promoting the collaboration between mathematicians from these areas and devel-
oped countries. It is our belief that the school contributed to the training of young
researchers from North and South America in a timely atmosphere of international
cooperation. For algebra in Latin America it had the effect of building regional
expertise to overcome the problem of isolation. For both the US and the Latin
American participants, the school and conference provided international experi-
ence and exposure, and new opportunities for collaborations. We expect that the
meeting has in fact enlarged the circle of US researchers visiting Latin American
universities on a regular basis. This in turn may help stem the now classical brain-
drain of scientists from these countries. Yet another direction in which the school
and conference succeeded was to enhance the collaboration between mathematicians
within Latin America and to trigger initiatives for further interchange between the
various countries of the continent.
We would also like to point out the fact that the PASI program was also an
occasion to honor Wolmer Vasconcelos of Rutgers University, one of the most dis-
tinguished senior commutative algebraists. Wolmer Vasconcelos grew up in the
vii
viii PREFACE

vicinity of Recife, where he also received his undergraduate education. He gradu-


ated with a Ph.D. in mathematics from the University of Chicago and spent most
of his professional life at Rutgers University. So far he has supervised 17 doctoral
students, authored 6 books, and written over 125 papers as a single author or
jointly with a total of 40 collaborators. Wolmer Vasconcelos is a member of the
Brazilian Academy of Sciences. Through his highly original contributions he has
been a major influence in the development of modern commutative algebra.

The school and conference, co-organized by Craig Huneke, Aron Simis, Bernd
Ulrich and the editors of the present volume, brought together more than 70 peri-
doctoral students from all over the world, but mostly from the Americas, and
approximatively 50 senior researchers in the areas of commutative algebra and al-
gebraic geometry. The focus of the program was on the following clusters of topics
central to modern commutative algebra:
• the homological conjectures;
• problems in positive and mixed characteristic;
• tight closure and its interaction with birational geometry;
• integral dependence and blowup algebras;
• equisingularity theory;
• Hilbert functions and multiplicities;
• combinatorial commutative algebra;
• Gröbner bases and computational algebra.
Homological conjectures, positive and mixed characteristic, tight closure. Charac-
teristic p methods, where p is a prime number, have been a powerful tool in both
commutative algebra and algebraic geometry. Via reduction to characteristic p they
also apply to the case of arbitrary characteristic. A culmination of this approach
is the theory of tight closure conceived by Hochster and Huneke in the eighties,
that led to powerful new results and sometimes strikingly simple proofs of known
theorems. Very recently one of the main open problems in this area has been solved
by Brenner and Monsky – the question of whether tight closure commutes with lo-
calization. Ein, Hara, Mustaţă, K. Smith, K. Watanabe and others have uncovered
intriguing connections between tight closure and seemingly unrelated notions from
birational geometry, such as rational singularities, multiplier ideals and arc spaces.
These connections are being explored further, and have already led to deep results
in both algebra and geometry.
The homological conjectures are a system of interrelated conjectures proposed
by Hochster in the seventies that grew out of work of Serre, Peskine and Szpiro on
intersection multiplicities. They have been solved for rings containing a field, but
are open – and notoriously difficult – in the case of mixed characteristic. Fairly
recently Heitmann was able to prove the direct summand conjecture in dimension
three. His methods are currently being developed further by Lyubeznik, Roberts,
Singh and others, who have uncovered deep properties of local cohomology modules
with the aim of proving the conjectures in general.
Blowup algebras, integral dependence, equisingularity theory. There has been an
abundance of new results about Rees algebras, or blowup algebras, and their struc-
ture over the past two decades. Rees algebras are the algebraic objects appearing
in the process of resolving singularities, and they have been used in Kawasaki’s
celebrated proof of the existence of Macaulifications, a weak version of resolution
PREFACE ix

of singularities. Recent efforts, by Busé, Chardin, Cox, Jouanolou, Kustin, Polini,


Simis, Ulrich, Vasconcelos and others, have focused on describing Rees algebras
explicitly in terms of generators and relations, a problem known in the applied
mathematics community as ‘implicitation’.
Rees algebras are also the environment in which integral dependence of ideals
and modules can be studied. Integral dependence is a notion of central importance
in modern commutative algebra, and has been the subject of two recent mono-
graphs, one by Vasconcelos and one by Huneke and Swanson. The concept appears
naturally in the study of multiplicities, Hilbert functions, cores and the aforemen-
tioned multiplier ideals. It also figures prominently in equisingularity theory, an
area particularly well represented in Brazil.
Equisingularity theory strives to providing numerical criteria for when the mem-
bers of a family of analytic sets are ‘similar’ to each other. Such criteria are provided
by the equisingularity conditions of Whitney and Thom. Through work of Gaffney,
Kleiman and Teissier these conditions are expressed in terms of integral dependence
of Jacobian modules, which in turn can be translated into numerical conditions on
multiplicities, at least in the case of isolated complete intersection singularities. In
the light of recent progress in multiplicity theory one can hope that these results
could be extended to more general types of singularities.
Combinatorial and computational commutative algebra, Gröbner bases, Hilbert func-
tions. Since the seminal work of Hochster and Stanley in the seventies and eighties,
combinatorial methods have played an important role in commutative algebra. The
objects studied in this area include Stanley-Reisner rings of simplicial complexes,
ideals of graphs, and toric varieties, which, being described by combinatorial data,
are a useful testing group for statements about varieties in general. More recently,
through work of Dickenstein, E. Miller and others, hypergeometric systems of linear
partial differential equations have entered the realm of combinatorial commutative
algebra as well.
A fundamental tool in computational algebra is Gröbner bases, a device that
reduces arbitrary ideals to ideals generated by monomials, which are combinatorial
objects and thereby more easily accessible computationally. Gröbner bases also
play a role theoretically in the study of Hilbert functions, multiplicities and graded
free resolutions. A recent breakthrough in this area has been the solution, by
Eisenbud and Schreyer, of a longstanding conjecture relating the degree shifts in a
free resolution to the multiplicity.
We end this Preface by expressing our gratitude to the contributors of this
volume for their enthusiasm in the project. The anonymous referees, who worked
very closely with us, also deserve special credit for all their time spent in reading
and correcting the original manuscripts. We are aware of the many demands on
our time that the academic profession requires from each of us! Finally, we wish to
express our heartfelt gratitude to the following institutions for their generous finan-
cial support: CAPES, CNPq, Department of Energy, through its Office of Basic
Energy Sciences, FACEPE (State Foundation of Pernambuco), National Science
Foundation (NSF), through the grant NSF-OISE 0819049, Millenium Program,
Third World Academy of Science (TWAS).

Alberto Corso and Claudia Polini


Lexington and Notre Dame · June 1, 2011
This page intentionally left blank
List of Mini Courses at PASI 2009

Elimination theory: Interactions between geometric modeling


and commutative algebra
David Cox, Amherst College (USA)
Combinatorial commutative algebra
Jürgen Herzog, Universität Duisburg-Essen (Germany)
Tight closure theory and problems in positive characteristic
Craig Huneke, University of Kansas (USA)
Equisingularity theory
Steven Kleiman, Massachusetts Institute of Technology (USA)
Hilbert functions and Hilbert coefficients in local rings
Maria Evelina Rossi, Università di Genova (Italy)
Tropical algebra
Bernd Sturmfels, University of California at Berkeley (USA)
Rees algebras, integral closures and adjoint ideals
Bernd Ulrich, Purdue University (USA)

xi
This page intentionally left blank
List of Talks at PASI 2009

Higher Fano manifolds


Carolina Araujo, IMPA (Brazil)

Reflexivity and rigidity for complexes


Luchezar Avramov, University of Nebraska (USA)

Implicit equations of multigraded hypersurfaces


Nicolás Botbol, Universidad de Buenos Aires (Argentina)

Normaliz: Algorithms for rational cones and affine monoids


Winfred Bruns, Universität Osnabrück (Germany)

Cremona geometry of plane curves


Ciro Ciliberto, Università di Roma (Italy)

Tropicalisation of rational varieties


Alicia Dickenstein, Universidad de Buenos Aires (Argentina)

Boji-Soederberg theory and the size of free resolutions


Daniel Erman, University of California at Berkeley (USA)

The Poincaré problem for subschemes invariant under Pfaff fields on projective
spaces
Eduardo Esteves, IMPA (Brazil)

Danilov Gizatullin surfaces and Ga -actions


Hubert Flenner, Universität Bochum (Germany)

On certain maximal curves over finite fields


Arnaldo Garcia, IMPA (Brazil)

Gotzmann coefficients of Hilbert functions


Anthony Geramita, Queen’s University (Canada)

On the hyperhomology of the small Gobelin for codimension 2


Xavier Gomez Mont, CIMAT (Mexico)

Cohen-Macaulayness versus vanishing of the first Hilbert coefficient of parameters:


Towards a problem of Wolmer Vasconcelos
Shiro Goto, Meiji University (Japan)

Gonality of ACM curves in P3


Robin Hartshorne, University of California at Berkeley (USA)
xiii
xiv LIST OF TALKS AT PASI 2009

A tight closure theory that commutes with localization in equal characteristic


Melvin Hochster, University of Michigan (USA)
A property of the ring of polynomials over a perfect field of characteristic p > 0
Gennady Lyubeznik, University of Minnesota (USA)
Blowup algebras and elimination theory
Claudia Polini, University of Notre Dame (USA)
Toric ideals of graphs and digraphs
Enrique Reyes, Instituto Politecnico Nacional/CINVESTAV (Mexico)
Oil fields and Hilbert schemes
Lorenzo Robbiano, Università di Genova (Italy)
Fontaine rings and local cohomology
Paul Roberts, University of Utah (USA)
Counting compatibly Frobenius split ideals
Karl Schwede, University of Michigan (USA)
Stanley decompositions of squarefree monomial ideals
YiHuang Shen, Purdue University (USA)
Local cohomology with determinantal support
Anurag Singh, University of Utah (USA)
Growth of Bass numbers
Janet Striuli, Fairfield University (USA)
Polynomial vector fields with algebraic trajectories
Israel Vainsencher, UFMG (Brazil)
On the ideal theory of graphs (fifteen years later)
Rafael H. Villarreal, Instituto Politecnico Nacional/CINVESTAV
(Mexico)
The a-invariants of normal graded Gorenstein rings and varieties with even
canonical class
Kei-ichi Watanabe, Nihon University (Japan)
PASI 2009 Photos

Photo 1. Mini course lecturers with Wolmer Vasconcelos

Photo 2. PASI 2009 organizers with Wolmer Vasconcelos

xv
xvi PASI 2009 PHOTOS

Photo 3. Group photo


Contemporary Mathematics
Volume 555, 2011

Hilbert depth of powers of the maximal ideal

Winfried Bruns, Christian Krattenthaler, and Jan Uliczka

Abstract. The Hilbert depth of a module M is the maximum depth that


occurs among all modules with the same Hilbert function as M . In this note
we compute the Hilbert depths of the powers of the irrelevant maximal ideal
in a standard graded polynomial ring.

1. Introduction
In [5] and [7] the authors have investigated the relationship between Hilbert
series and depths of graded modules over standard graded and multigraded poly-
nomial rings. In this paper we will consider only the standard graded case, i.e.,
finitely generated graded modules over polynomial rings R = K[X1 , . . . , Xn ] for
which K is a field and deg Xi = 1 for i = 1, . . . , n.
We refer the reader to [4] for the basic theory of Hilbert functions and
 series.
Let us just recall that the Hilbert function of a graded R-module M = k∈Z Mk
is given by
H(M, k) = dimK Mk , k ∈ Z,
and that the Laurent series

HM (T ) = H(M, k) T k
k∈Z

is called the Hilbert series of M . The Hilbert series is the Laurent expansion at 0
of a rational function as in (1.1) with aLaurent polynomial in the numerator.
Let us say that a Laurent series k∈Z ak T
k
is positive if ak ≥ 0 for all k.
Hilbert series are positive by definition, and it is not surprising that positivity is
the central condition in the following theorem that summarizes the results of [7]. It
describes the maximum depth that a graded module with given Hilbert series can
have.

2010 Mathematics Subject Classification. Primary 05E40, 13C15; Secondary 33C20.


The second author was partially supported by the Austrian Science Foundation FWF, grants
Z130-N13 and S9607-N13, the latter in the framework of the National Research Network “Analytic
Combinatorics and Probabilistic Number Theory”.

2011
c 0000
c Mathematical
American (copyright Society
holder)

1
2 WINFRIED BRUNS, CHRISTIAN KRATTENTHALER, AND JAN ULICZKA

Theorem 1.1. Let R = K[X1 , . . . , Xn ] as above, and let M = 0 be a finitely


generated graded R-module with Hilbert series
QM (T )
(1.1) HM (T ) = , QM (T ) ∈ Z[T, T −1 ].
(1 − T )n
Then the following numbers coincide:
(1) max{depth N : HM (T ) = HN (T )},
(2) the maximum d such that HM (T ) can be written as
n
Qe (T )
(1.2) HM (T ) = , Qe (T ) ∈ Z[T, T −1 ],
(1 − T )e
e=d
with positive Laurent polynomials Qe (T ),
(3) max{p : (1 − T )p HM (T ) positive},
(4) n − min{q : QM (T )/(1 − T )q positive}.
The crucial point of the proof of Theorem 1.1 is to show that every positive
Laurent series that can be written in the form (1.1) has a representation of type
(1.2) (with d ≥ 0).
In [7], Theorem 1.1(1) is used to define the Hilbert depth Hdepth M of M ,
whereas [5] bases the definition of Hilbert depth on (2). In view of the theorem,
this difference is irrelevant in the standard graded case, but in the multigraded case
the equivalence of 1.1(1) and a suitable generalization of (2) is widely open, and
can be considered as a Hilbert function variant of the Stanley conjecture (see [5] for
this connection). (Note that Theorem 1.1(3) and (4) cannot be transferred easily
to the multigraded situation.)
The Hilbert depth of the maximal ideal m = (X1 , . . . , Xn ) is known:
n
Hdepth m = .
2
This was observed in [7] and will be proved again below. (In fact, by a theorem of
Biró et al. [3], the (multigraded) Stanley depth of m is given by n/2.)
In general, Hilbert depth is hard to compute since one almost inevitably en-
counters alternating expressions whose nonnegativity is to be decided. Not even
for all syzygy modules of m Hilbert depth is known precisely; see [5]. Nevertheless,
the main result of our paper shows that the Hilbert depths of the powers of m can
be determined exactly.
Theorem 1.2. For all n and s one has
 
n
Hdepth ms = .
s+1
That n/(s + 1) is an upper bound is seen easily. The Hilbert series of ms is

n+s−1 s n + s s+1
T + T + ···
s s+1
Thus the coefficient of T s+1 in (1 − T )p Hms (T ) is

n+s n+s−1
−p ,
s+1 s
and the difference is negative unless p ≤ (n + s)/(s + 1) = n/(s + 1). That the
condition p ≤ n/(s + 1) is sufficient for the positivity of (1 − T )p Hms (T ) will be
shown in the remainder of this paper.
HILBERT DEPTH OF POWERS OF THE MAXIMAL IDEAL 3

The first step in the proof is the computation of (1 − T )r Hms (T ) since, in view
of Theorem 1.1(3), we want to find the maximum r for which this series is positive.
Proposition 1.3. For any integer 0 < r < n, we have

n+s−1 s
(1.3) (1 − T )r Hms (T ) = T
s
⎡ ⎤
 n + k − 1 − r

r+s−1 
s−1
r

n+j−1 ⎦ k
+ ⎣ + (−1)k−1 (−1)j T
k j=0
k − j j
k=s+1



n+k−1−r k
+ T .
k
k=r+s

This is easily proved by induction on r or by the binomial expansion of (1−T )r .


The critical term in (1.3) is the one in the middle row. In the next section
we will find an alternative expression for it. The positivity of this expression for
r ≤ n/(s+1) will be stated in Proposition 3.1. Its proof is the subject of Section 3.

Acknowledgement. We are indebted to Jiayuan Lin for pointing out a gap in the
proof of Lemma 3.5 in the first version of this article, for providing an argument
that fixes this gap, and for giving us the permission to reproduce this argument
here.

2. Binomial identities
One of the ingredients in the proof of Theorem 1.2 via Proposition 3.1 in the
next section are two alternative expressions for the sum over j in the critical term in
(1.3). We provide two proofs: a direct one using well-known identities for binomial
and hypergeometric series, and a — perhaps more illuminating — algebraic one
which shows that the three expressions give the Hilbert function of a certain module.
Lemma 2.1. For all positive integers n, s, r, k, we have
k

k−j n + j − 1 r
(−1)
j=s
j k−j


s−1

n+k−r−1 k−1 j r n+j−1


(2.1) = + (−1) (−1)
k j=0
k−j j

 r

n+k−r−1 r−t n−t+s−1


(2.2) = + (−1)k+s .
k t=1
k−s s−1
 
Proof. We start with the direct proof. Using the short notation T k f (T )
for the coefficient of T k in the power series f (z), we have


∞

k
n+j−1 r   n+j−1 j
(−1)k−j = T k (1 − T )r T
j=s
j k−j j=s
j
⎛ ⎞
s−1


 k n + j − 1
= T (1 − T )r ⎝(1 − T )−n − T j⎠
j=0
j
4 WINFRIED BRUNS, CHRISTIAN KRATTENTHALER, AND JAN ULICZKA

s−1


    n+j−1 j
= T k (1 − T )r−n − T k (1 − T )r T
j=0
j

s−1

n+k−r−1 r n+j−1
= − (−1)k−j ,
k j=0
k−j j

which proves (2.1).


In order to see the equality between (2.1) and (2.2), we have to prove

s−1

r

r n+j−1 r−t n−t+s−1


(2.3) (−1)j = (−1)s+1 .
j=0
k−j j t=1
k−s s−1

In the sum on the left-hand side, we reverse the order of summation (that is, we
replace j by s−j −1), and we rewrite the resulting sum in hypergeometric notation.
Thus, we obtain that the left-hand side of (2.3) equals


 
s+1 r n+s−2 1, 1 − s, 1 + k − r − s
(−1) 3 F2 ;1 .
k−s+1 s−1 2 − n − s, 2 + k − s
Next we apply the transformation formula (see [2, Ex. 7, p. 98])
   
a, b, c Γ(e) Γ(d + e − a − b − c) a, d − b, d − c
3 F2 ;1 = 3 F2 ;1
d, e Γ(e − a) Γ(d + e − b − c) d, d + e − b − c
to the 3 F2 -series, to obtain
 
s+1 (n)s−1 (−k + r + s)1+k−s 1 − k − n + r, 1, 1 − n
(−1) 3 F2 ;1 .
(1 − n + r) (1)k−s (1)s−1 2 − n − s, 2 − n + r
Now we apply the transformation formula ([6, Eq. (3.1.1)])
   
a, b, −N (e − b)N −N, b, d − a
3 F2 ;1 = 3 F2 ;1
d, e (e)N d, 1 + b − e − N
where N is a nonnegative integer. After some simplification, one sees that the
resulting expression agrees with the right-hand side of (2.3).
Now we discuss the algebraic proof. We consider the u-th syzygy M of the
residue class ring S = R/(X1 , . . . , Xr ) of R = K[X1 , . . . , Xn ]. There are two exact
sequences from which the Hilbert function of M can be computed since S is resolved
by the Koszul complex of the sequence X1 , . . . , Xr . We simply break the Koszul
complex into two parts, inserting M as the kernel or cokernel, respectively, at the
appropriate place:

u−1
0→M → F (−u + 1) → · · · → F → R → S → 0, F = Rr ,

r 
r−1 
u
0→ F (−r) → F (−r + 1) → · · · → F (−r) → M → 0.
The computation of the Euler characteristic of the first complex in degree k yields
the equation

u−1


n−r+k−1  n+k−i−1
i r
H(M, k) = (−1) u
− (−1)
n−r−1 i=0
i k−i
HILBERT DEPTH OF POWERS OF THE MAXIMAL IDEAL 5

⎛ ⎞


k

u⎝ n−r+k−1 r n+j−1 ⎠
(2.4) = (−1) − (−1)k−j ,
n−r−1 k−j j
j=k−u+1

where we pass from the first to the second line by the substitution j = k − i. In
the same degree we obtain for the second complex

k−u

l r n+k−u−l−1
H(M, k) = (−1)
u+l n−1
l=0

k−u

k−u j r n+j−1
(2.5) = (−1) (−1) ,
j=0
k−j j

where we have substituted the summation index l by k −u−j in the second line. On
the other hand, we can also compute the Hilbert function by [5, Proposition 3.7].
To this end, we fix r. For n = r, loc. cit. then yields
 r

r−t r−t+k−u
H(M, k) = .
t=1
u−1 k−u

For the ring extension from K[X1 , . . . , Xr ] to K[X1 , . . . , Xn ], we have to replace


the dimensions of the symmetric powers of K r−t by those of K n−t ; therefore
r

r−t n−t+k−u
(2.6) H(M, k) =
t=1
u−1 k−u

in the general case. Setting s = k − u + 1, one arrives at (2.1) by equating (2.4)


and (2.5), while equating (2.5) and (2.6) leads to (2.2). 

3. The proof of positivity


In view of Proposition 1.3 and Lemma 2.1, for the proof of Theorem 1.2 we have
to show the inequality that we state below in Proposition 3.1. Its proof requires
several auxiliary lemmas, provided for in Lemmas 3.3–3.5. The actual proof of
Proposition 3.1 (and, thus, of Theorem 1.2) is then given at the end of this section.
Proposition 3.1. Let n and s be positive integers, and let r = n/(s + 1).
Then, for all k = s + 1, s + 2, . . . , s + r − 1, we have

 r

n+k−r−1 r−t n−t+s−1


(3.1) ≥ .
k t=1
k−s s−1

Remark 3.2. The assertion of the proposition is trivially true if r − 1 ≤ k − s.


In the following, we make frequent use of the classical digamma function ψ(x),
which is defined to be the logarithmic derivative of the gamma function Γ(x), i.e.,
ψ(x) = Γ (x)/Γ(x).
Lemma 3.3. For all real (!) numbers n, k, s, t with n, t ≥ 1, s ≥ 2, s + 2 ≤ k ≤
r + s − t − 1, we have
ψ(n + k − r) − ψ(k + 1) > ψ(r − t − k + s + 1) − ψ(k − s + 1),
where, as before, r = n/(s + 1).
6 WINFRIED BRUNS, CHRISTIAN KRATTENTHALER, AND JAN ULICZKA

Proof. We want to prove that


(3.2) ψ(n + k − r) − ψ(r − t − k + s + 1) + ψ(k − s + 1) − ψ(k + 1) > 0
for the range of parameters indicated in the statement of the lemma. Since ψ(x)
is monotone increasing for x > 0 (this follows e.g. from [1, Eq. (1.2.14)]), the left-
hand side of (3.2) is monotone increasing in t. It therefore suffices to prove (3.2)
for t = 1, that is, it suffices to prove
(3.3) ψ(n + k − r) − ψ(r − k + s) + ψ(k − s + 1) − ψ(k + 1) > 0.
Next we claim that the left-hand side of (3.3) is monotone increasing in k. To
see this, we differentiate the left-hand side of (3.3) with respect to k, to obtain
(3.4) ψ  (n + k − r) + ψ  (r − k + s) + ψ  (k − s + 1) − ψ  (k + 1).
Since ψ(x) is monotone increasing, the first two terms in (3.4) are positive. More-
over, ψ(x) is a concave function for x > 0 (this follows also from [1, Eq. (1.2.14)]),
whence ψ  (k − s + 1) − ψ  (k + 1) > 0. This proves that the expression in (3.4) is
positive, that is, that the derivative with respect to k of the left-hand side of (3.3)
is positive. This establishes our claim.
As a result of the above argument, we see that it suffices to prove (3.3) for the
smallest k, that is, for k = s + 2. In other words, it suffices to prove
(3.5) ψ(n + s − r + 2) − ψ(r − 2) + ψ(3) − ψ(s + 3) > 0.
We now investigate the behaviour of the left-hand side of (3.5) as a function of
n, which we denote by f (n) (ignoring the dependence of the expression on s at this
point). Clearly, as long as n stays strictly between successive multiples of s + 1,
r = n/(s+1) does not change, and f (n) is monotone increasing in n in this range.
However, if n changes from n = (s + 1), say, to something just marginally larger,
then r jumps from  to  + 1, thereby changing the value of f discontinuously. The
limit value limn↓(s+1) f (n) is given by
lim f (n) = ψ((s + 1) + s −  + 1) − ψ( − 1) + ψ(3) − ψ(s + 3).
n↓(s+1)

By the argument above, we know that f (n) stays above this value for (s + 1) <
n ≤ ( + 1)(s + 1). Let us examine the difference of two such limit values:
lim f (n) − lim f (n) = ψ((s + 1) + s −  + 1) − ψ( − 1)
n↓(s+1) n↓(+1)(s+1)

− ψ(( + 1)(s + 1) + s − ) + ψ()


1
(3.6) = ψ(( + 1)s + 1) − ψ(( + 1)s + s + 1) + ,
−1
where we used [1, Eq. (1.2.15) with n = 1] to obtain the last line. By [1,
Eq. (1.2.12)], we have ψ  (1) = −γ, where γ is the Euler–Mascheroni constant.
Making use of the integral representation
 1
1 − xa−1
ψ(a) = −γ + dx, (a − 1) > 0
0 1−x
(see [1, Theorem 1.6.1(ii) after change of variables x = e−z ]), we estimate
 1 (+1)s
x − x(+1)s+s
ψ(( + 1)s + 1) − ψ(( + 1)s + s + 1) = − dx
0 1−x
HILBERT DEPTH OF POWERS OF THE MAXIMAL IDEAL 7

 1
1 − xs
=− x(+1)s dx
0 1−x
 1
≥ −s x(+1)s dx
0
s 1 1
≥− >− >− .
( + 1)s + 1 +1 −1
This shows that the difference in (3.6) is (strictly) negative, that is, that the left-
hand side of (3.5) becomes smaller when we “jump” from (slightly above) n =
(s + 1) to (slightly above) n = ( + 1)(s + 1), while the values in between stay
above the limit value from the right at n ↓ ( + 1)(s + 1). Therefore, it suffices to
prove (3.5) in the limit n → ∞. By recalling the asymptotic behaviour

1
(3.7) ψ(x) = log x + O , as x → ∞,
x
of the digamma function (cf. [1, Cor. 1.4.5]), we see that this limit of the left-hand
side of (3.5) is log s + ψ(3) − ψ(s + 3), so that it remains to prove
(3.8) log s + ψ(3) − ψ(s + 3) > 0.
Also here, we look at the derivative of the left-hand side with respect to s:
1
− ψ  (s + 3).
s
By [1, Eq. (1.2.14)], this can be rewritten in the form

1  1
− .
s m=0 (s + m + 3)2
The infinite sum can be interpreted as the integral of the step function
1
x →
x2
between x = s + 2 and x = ∞. The function being bounded above by the function
x → 1/x2 , we conclude
 ∞
1 1 dx 1 1
− ψ  (s + 3) > − 2
= − > 0.
s s s+2 x s s + 2
In other words, the derivative with respect to s of the left-hand side of (3.8) is
always positive, hence it suffices to verify (3.8) for s = 2:
1 4
log 2 + ψ(3) − ψ(5) = log 2 − − > 0,
3 5
where we used again [1, Eq.(1.2.15)].
This completes the proof of the lemma. 
Lemma 3.4. Let the real numbers n, k0 , s be given with n ≥ 1, s ≥ 2, s + 2 ≤
k0 ≤ r + s − t − 1. Suppose that (3.1) holds for this choice of n, k0 , s. Then it also
holds for k in an interval [k0 , k0 + ε) for a suitable ε > 0.
Proof. We extend the binomial coefficients in (3.1) to real values of k, by
using gamma functions. To be precise, we extend the left-hand side of (3.1) to
Γ(n + k − r)
,
Γ(k + 1) (n − r − 1)!
8 WINFRIED BRUNS, CHRISTIAN KRATTENTHALER, AND JAN ULICZKA

and the right-hand side to


r

(r − t)! n−t+s−1
.
t=1
Γ(k − s + 1) Γ(r − t − k + s + 1) s−1
In abuse of notation, we shall still use binomial coefficient notation, even if we allow
real values of k.
We now compute the derivative at k = k0 on both sides of (3.1). On the
left-hand side, this is

  n + k0 − r − 1
ψ(n + k0 − r) − ψ(k0 + 1) ,
k0
while on the right-hand side this is


r
  r−t n−t+s−1
ψ(r − t − k0 + s + 1) − ψ(k0 − s + 1) .
t=1
k0 − s s−1
By using Lemma 3.3, we can estimate the derivative of the right-hand side as
follows:


r
  r−t n−t+s−1
ψ(r − t − k0 + s + 1) − ψ(k0 − s + 1)
t=1
k0 − s s−1


r
 r−t n−t+s−1
< ψ(n + k0 − r) − ψ(k0 + 1)
t=1
k0 − s s−1

  n + k0 − r − 1
< ψ(n + k0 − r) − ψ(k0 + 1) .
k0
The last expression is however exactly the derivative of the left-hand side of (3.1).
Hence, the left-hand side of (3.1) must also exceed the right-hand side in a small
“neighbourhood” [k0 , ε) to the right of k0 . This proves the lemma. 
Lemma 3.5. For all positive integers n and s with n > 3s + 3 and s ≥ 2, we
have




n+s−r+1 n+s+1 n+s r n+s−1


(3.9) 2 ≥ −r + ,
s+2 s+2 s+1 2 s
where, as before, r = n/(s + 1).
Proof. We proceed in a spirit similar to the one in the proof of Lemma 3.4.
We regard both sides of (3.9) as functions in the real variable n. The reader should
note that the assumption that n > 3s + 3 implies that r ≥ 4, a fact that will be
used frequently without further mention.
Let first n be strictly between (s + 1) and ( + 1)(s + 1), for some fixed non-
negative integer . Then r =  + 1, so that both sides of (3.9) become polynomial
(whence continuous) functions in n. The derivative of the left-hand side of (3.9)
with respect to n (in the interval ((s + 1), ( + 1)(s + 1))) equals

  n+s−r+1
(3.10) 2 ψ(n + s − r + 2) − ψ(n − r) ,
s+2
while the derivative of the right-hand side of (3.9) with respect to n equals

2n + 2s + 1 − r(s + 2) (n + s − 1)!
(3.11) ψ(n + s) − ψ(n) + N (n, s),
N (n, s) (s + 2)! (n − 1)!
HILBERT DEPTH OF POWERS OF THE MAXIMAL IDEAL 9

where

r
N (n, s) = (n + s)(n + s + 1) − r(n + s)(s + 2) + (s + 1)(s + 2).
2
We claim that1
2n + 2s + 1 − r(s + 2)
(3.12) ψ(n + s − r + 2) − ψ(n − r) > ψ(n + s) − ψ(n) + .
N (n, s)
This would imply that, provided (3.9) holds for some n in the (closed) interval
[(s + 1), ( + 1)(s + 1)], then the derivative of the left-hand side of (3.9) would be
larger than the derivative of the right-hand side of (3.9) at this n, and hence the
function on the left-hand side of (3.9) would grow faster than the right-hand side of
(3.9) for n in ((s + 1), ( + 1)(s + 1)]. In turn, this would mean that it would suffice
to show the validity of (3.9) for n ↓ (s + 1) (that is, for n = (s + 1) and r =  + 1),
to conclude that (3.9) holds for the whole interval ((s + 1), ( + 1)(s + 1)].
We next embark on the proof of (3.12). Using [1, Eq. (1.2.15)], we see that
  s+1
1  1 s−1
ψ(n + s − r + 2) − ψ(n − r) − ψ(n + s) − ψ(n) = −
i=0
n − r + i i=0 n + i
s−1


1 1 1 1
= + + −
n − r n − r + 1 i=0 n − r + i + 2 n + i

2 1  s−1
r−2
= + +
n − r + 1 (n − r)(n − r + 1) i=0 (n − r + i + 2)(n + i)

2 1  s−1
r−2
> + +
n − r + 1 (n − r)(n − r + 1) i=0 (n + i − 1)(n + i)
s−1


2 1 1 1
> + + (r − 2) −
n − r + 1 (n − r)(n − r + 1) i=0
n+i−1 n+i

2 1 1 1
> + + (r − 2) −
n − r + 1 (n − r)(n − r + 1) n−1 n+s−1
2 1 (r − 2)s
> + +
n − r + 1 (n − r)(n − r + 1) (n − 1)(n + s − 1)
2 (r − 2)s + 1
> + .
n − r + 1 (n − 1)(n + s − 1)
Hence, if we are able to prove that
2 (r − 2)s + 1 2n + 2s + 1 − r(s + 2)
(3.13) + ≥ ,
n − r + 1 (n − 1)(n + s − 1) N (n, s)
the inequality (3.12) will follow immediately. We now claim that
(3.14) (r − 1)N (n, s) ≥ (n − 1)(n + s)
and
(3.15) 2N (n, s) + ((r − 2)s + 1)s ≥ (n − r + 1)(2n + 2s + 1 − r(s + 2)).
1Our original proof had a weaker inequality at this point, which however turns out to be
not sufficient. This gap was pointed out by Jiayuan Lin. In addition, he provided the following
argument establishing (3.12), and he kindly gave us the permission to reproduce it here.
10 WINFRIED BRUNS, CHRISTIAN KRATTENTHALER, AND JAN ULICZKA

If, for the moment, we assume the validity of (3.14) and (3.15), then we infer

2 (r − 2)s + 1 1 ((r − 2)s + 1)(n − r + 1)


+ = 2+
n − r + 1 (n − 1)(n + s − 1) n−r+1 (n − 1)(n + s − 1)

1 ((r − 2)s + 1)(n − r + 1)


≥ 2+
n−r+1 (r − 1)N (n, s)

1 ((r − 2)s + 1)s


≥ 2+
n−r+1 N (n, s)
2n + 2s + 1 − r(s + 2)
≥ ,
N (n, s)
which is exactly (3.13). Here we used (3.14) to obtain the second line, the simple
fact that
n−r+1 (r − 1)(s + 1) − r + 1
≥ =s
r−1 r−1
to obtain the third line, and (3.15) to obtain the last line. In summary, (3.14) and
(3.15) together would imply (3.13), and hence (3.12).
To see (3.14), we rewrite it explicitly in the form

r  
(3.16) (r −1) (s+1)(s+2) ≥ (n+s) n−1−(r −1)(n+s+1)+r(r −1)(s+2) .
2
We write n = r(s + 1) − n0 , with 0 ≤ n0 ≤ s. If we substitute this in the inequality
above, then the right-hand side of (3.16) turns into
  
r(s + 1) + s − n0 r(s + 1) − 1 − (r − 1)(r + 1)(s + 1) + (r − 2)n0 + r(r − 1)(s + 2) .
We consider this as a quadratic function in n0 . It has its unique maximum at
r 2 s − rs − r − 3s 3rs − r − 3s 2rs − 4s
≥ ≥ = s.
2(r − 2) 2(r − 2) 2(r − 2)
It is therefore monotone increasing on the interval [0, s] and consequently attains
its maximal value on the interval [0, s] at n0 = s. So it suffices to verify (3.16) at
n0 = s. After simplification, this turns out to be equivalent to

r
(s + 1)(s(r − 3) − 2) ≥ 0,
2
which, by our assumptions on r and s, is trivially true.
To see (3.15), we again substitute r(s + 1) − n0 for n (with 0 ≤ n0 ≤ s) to
obtain the equivalent inequality
rs(r − 3) + s − 1 + (rs − 2s + 1)n0 > 0,
which is trivially true by our assumptions on r, s, and n0 .
Altogether, we have now established (3.12). Hence, the conclusion of the para-
graph following (3.12) that it suffices to prove (3.9) for n = (s + 1) and r =  + 1
holds as well.
We substitute n = (s + 1) and r =  + 1 in (3.9):

( + 1)s ( + 1)(s + 1)
2 ≥
s+2 s+2


( + 1)(s + 1) − 1  + 1 ( + 1)(s + 1) − 2
− ( + 1) + ,
s+1 2 s
HILBERT DEPTH OF POWERS OF THE MAXIMAL IDEAL 11

and, after simplification, we obtain the equivalent inequality


(( + 1)s − 1)! 1 (( + 1)(s + 1) − 2)! ((s + 1) − 2)
2 ≥ .
(s − 2)! 2 ((s + 1) − 1)!
We shall actually establish the stronger inequality
(( + 1)s − 1)! 1 (( + 1)(s + 1) − 2)!
(3.17) 2 ≥ .
(s − 2)! 2 ((s + 1) − 2)!
In order to do so, we regard the functions in (3.17) again as functions in real
variables, more precisely, as functions in the real variable , while we think of s as
being fixed.
We first verify that (3.17) holds for the smallest possible value of  = r − 1,
that is, for  = 3. For that value of , the inequality (3.17) becomes
(4s − 1)! 1 (4s + 2)!
2 ≥ ,
(3s − 2)! 2 (3s + 1)!
or, equivalently,
4(3s + 1)3s(3s − 1) ≥ (4s + 2)(4s + 1)4s,
which is indeed true for s ≥ 2.
Next we compute the derivative of both sides of (3.17) with respect to . On
the left-hand side, we obtain
(( + 1)s − 1)!
(3.18) 2s (ψ(( + 1)s) − ψ(s − 1)) ,
(s − 2)!
while on the right-hand side we obtain
s+1 (( + 1)(s + 1) − 2)!
(3.19) (ψ(( + 1)(s + 1) − 1) − ψ((s + 1) − 1)) .
2 ((s + 1) − 2)!
Using [1, Eq. (1.2.15)], it is straightforward to see that
ψ(( + 1)s) − ψ(s − 1) > ψ(( + 1)(s + 1) − 1) − ψ((s + 1) − 1).
s+1
Furthermore, we have 2s > 2 , so that
s+1
2s (ψ(( + 1)s) − ψ(s − 1)) > (ψ(( + 1)(s + 1) − 1) − ψ((s + 1) − 1))
2
for all  ≥ 3. Since we already know that (3.17) holds for  = 3, it then follows that
the derivative of the left-hand side of (3.17) (see (3.18)) is always larger than the
derivative of the right-hand side (see (3.19)). This establishes (3.17) and completes
the proof of the lemma. 
Proof of Proposition 3.1. The assertion is true for k = s + 1 because of
the choice of r. By comparing (2.1) and (2.2) in Lemma 2.1, the assertion for
k = s + 2, which reads

 r

n+s−r+1 r−t n−t+s−1


≥ ,
s+2 t=1
2 s−1
can be rewritten as


s+2

n+s−r+1 n+s−r+1 n+j−1 r


≥− + (−1)s−j+2 ,
s+2 s+2 j=s
j s−j+2
12 WINFRIED BRUNS, CHRISTIAN KRATTENTHALER, AND JAN ULICZKA

or, equivalently, as




n+s−r+1 n+s+1 n+s r n+s−1


2 ≥ −r + .
s+2 s+2 s 2 s
Remembering Remark 3.2, we see that it is enough to show this for r > 3, that
is, for n > 3s + 3. Lemma 3.5 shows that the above inequality indeed holds for
that range of n. Lemma 3.4 then implies that the assertion must be true for all
k ≥ s + 2. 

References
[1] G. E. Andrews, R. A. Askey and R. Roy, Special functions, Encyclopedia of Math. And Its
Applications, vol. 71, Cambridge University Press, Cambridge, 1999.
[2] W. N. Bailey, Generalized hypergeometric series, Cambridge University Press, Cambridge,
1935.
[3] C. Biró, D. M. Howard, M. T. Keller, W. T. Trotter and S. J. Young, Interval partitions and
Stanley depth, J. Combin. Theory Ser. A 117 (2010), 475–482.
[4] W. Bruns and J. Herzog, Cohen-Macaulay rings, rev. ed., Cambridge Studies in Advanced
Mathematics 39, Cambridge University Press, 1998.
[5] W. Bruns, C. Krattenthaler and J. Uliczka, Stanley decompositions and Hilbert depth in the
Koszul complex, J. Commut. Algebra 2 (2010), 327–359.
[6] G. Gasper and M. Rahman, Basic hypergeometric series, second edition, Encyclopedia of
Math. And Its Applications, vol. 96, Cambridge University Press, Cambridge, 2004.
[7] J. Uliczka, Remarks on Hilbert series of graded modules over polynomioal rings, Manuscr.
Math. 132 (2010), 159–168.

Universität Osnabrück, Institut für Mathematik, 49069 Osnabrück, Germany


E-mail address: wbruns@uos.de

Fakultät für Mathematik, Universität Wien, Nordbergstrasze 15, A-1090 Vienna,


Austria
E-mail address: Christian.Krattenthaler@univie.ac.at

Universität Osnabrück, Institut für Mathematik, 49069 Osnabrück, Germany


E-mail address: Jan.Uliczka@uos.de
Contemporary Mathematics
Volume 555, 2011

A Note on Reductions of Monomial Ideals in k[x, y](x,y)

C-Y. Jean Chan and Jung-Chen Liu

We dedicate this paper to Professor Wolmer Vasconcelos.

Abstract. We consider monomial ideals in the two-dimensional localized


polynomial ring k[x, y](x,y) where k is an infinite field. The goal of this paper is
to determine a sufficient condition under which an ideal containing xa y b +xc y d
is a reduction of an ideal containing xa y b and xc y d . The main theorem states
and provides a means to verify a condition that implies the integral closure of
an ideal is a monomial ideal. As a corollary, we form an algorithm to obtain
a minimal reduction of an arbitrary non-principal monomial ideal. We also
demonstrate an application of the main results on the monomial ideals dis-
cussed in this paper to the computation of the Buchsbaum-Rim multiplicity
of a module under certain conditions.

1. Introduction and Motivation


Let R be a commutative ring with identity and let J ⊆ I be ideals of R. I is
integral over J, if for every element u in I, there exist elements ai in J i such that
un + a1 un−1 + a2 un−2 + · · · + an−1 u + an = 0.
J is a reduction of I, if I m+1 = JI m for some positive integer m. When R is
Noetherian, these two definitions are equivalent; in other words, I is integral over
J if and only if J is a reduction of I (c.f. [RS, 1.1]). The reduction of ideals was
first introduced by Northcott and Rees [NR] and Rees [R] extended the notion
to modules. Since then the reduction of ideals and modules have been discussed
extensively.
It is known that the ideals with reduction relation have the same integral clo-
sure and that they, if m-primary, have the same Hilbert-Samuel multiplicity. The
main aim of this paper is to discuss the reductions of monomial ideals in a two-
dimensional localized polynomial ring R = k[x, y](x,y) over an infinite field. In
particular, for a given ideal I in R, there is a monomial ideal I ∗ naturally arising
from I (see definition below). Theorem 3.3 provides an algorithm to determine if

1991 Mathematics Subject Classification. Primary 13B22, Secondary 13A30, 13B25, 13F20.
Key words and phrases. monomidal ideals, reductions, minimal reductions, integral closures,
Hilbert Samuel multiplicity, monomial modules, Buchsbaum-Rim multiplicity.
The first author was partially supported by FRCE Type B Grant #48780 and Early Career
Investigator’s Grant #C61368 of Central Michigan University.
1 2011
c American Mathematical Society

13
14
2 C-Y. JEAN CHAN AND JUNG-CHEN LIU

I is a reduction of I ∗ . This is desirable since the integral closure and the Hilbert-
Samuel multiplicity (if defined) both have graphical interpretations. So one can
easily obtain this information for I if I ∗ is integral over I.
For the convenience of discussions in the latter sections, we elaborate here the
graph associated to a monomial ideal and some relevant information one can read
from it. Note that each monomial in a localized polynomial ring k[x1 , . . . , xd ](x1 ,...,xd )
over a field k corresponds to a point in Zd . If a is a monomial ideal in the above
local ring, we define the graph of a, commonly called the Newton polyhedron, to be
the set
Ga = {(a1 , . . . , ad ) | ai ≥ αi ∀i for some xα
1 · · · xd ∈ a} ⊂ R .
1 αd d

Then, the integral closure of a is again a monomial ideal (c.f. [SH, 1.4.2]) and is
generated by those monomials corresponding to the integral lattice points in the
convex hull of the graph of a (c.f. [SH, 1.4.6]). If we further assume that a is
m-primary where m is the unique maximal ideal in k[x1 , . . . , xd ](x1 ,...,xd ) , then the
Hilbert-Samuel multiplicity is
(1.1) e(a) = d! × the volume of the complement of the convex hull of Ga in Rd≥0 .

(c.f. [R2, Exercise 2.8 (b)]).


In this paper, we assume that R is two-dimensional, i.e., R = k[x, y](x,y) . Let

I be the ideal generated by fi = μij xaij y bij with μij = 0 and i = 1, . . . , m.
All the monomials occurring in the fi for all i together generate a monomial ideal,
denoted I ∗ . A simple exercise shows that such monomial ideal is independent from
the choices of the generating set {fi }. Indeed I ∗ is the smallest monomial ideal
containing I (see also Remark 3.1). We focus on finding a sufficient condition under
which I becomes a reduction of I ∗ . In such case, the integral closure and, if I is
m-primary, the Hilbert-Samuel multiplicity of I can be obtained straightforwardly
from the graph of I ∗ as stated in the previous paragraph.
A reduction of an ideal is minimal if it is minimal with respect to inclusion.
In a Noetherian local ring (R, m), a reduction b of an ideal a is minimal if b is
generated by  elements where  is the analytic spread of a (i.e., the dimension of
the fiber cone of a) (c.f. [SH, 8.3.5]). It is also known that ht a ≤  ≤ dim R
(c.f. [SH, 8.3.9]). Moreover when the residue field is infinite, then a reduction b
is minimal if and only if b is minimally generated by  elements (c.f. [SH, 8.3.5,
8.3.7]). In a Noetherian unique factorization local domain of dimension d with
d > 1, all non-principal ideals have analytic spread 2 ≤  ≤ d. (An easy exercise
shows that principal ideals in an UFD are integrally closed.) Most examples in this
paper are taken from k[x, y](x,y) which is a unique factorization local domain.
In particular, if a is an m-primary ideal in a local ring of dimension d with
infinite residue field, then a reduction b of a is minimal if and only if b is gener-
ated by d elements. In principal, b can be chosen by taking d “general enough”
combinations on the generators of a. Example 4.4, however, shows that false com-
binations exist easily. Let R = k[x, y](x,y) with |k| = ∞. Deducing from the main
result, Theorem 3.3, we give an algorithm of obtaining a minimal reduction of an
arbitrary non-principal monomial ideal, not necessarily m-primary (Corollary 3.7).
It was brought to our attention that V. C. Quiñonez has also achieved the same
result as Corollary 3.7 in her research report [Q]. In Subsection 4.2, we briefly
discuss the main theorem in [Q] and give a proof using Lemma 3.4
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 15
3

For an arbitrary monomial ideal a in R = k[x, y](x,y) , it is known that the


ideal b generated by the monomials corresponding to the vertices of the graph Ga
of a is a reduction of a. Singla [S, 2.1] proves that such b is the unique minimal
monomial reduction of a, where b being a minimal monomial reduction means
that no monomial ideal properly contained in b is a reduction of a. Note that
the minimal monomial reduction of a non-principal monomial ideal a is almost
never a minimal reduction unless Ga contains only two vertices. For example, for
the ideal a1 = (xy 5 , x2 y 4 , x3 y), the vertices of Ga1 are (1, 5) and (3, 1), and so
the ideal b1 = (xy 5 , x3 y) is the unique minimal monomial reduction as well as a
minimal reduction of a1 since b1 is generated by two elements. In fact from the
discussion in the previous paragraphs, every non-principal ideal in R = k[x, y](x,y)
has analytic spread two, so a reduction is minimal if and only if it is generated by
exactly two elements. For the ideal a2 = (xy 5 , x2 y 2 , x3 y), a2 is its own minimal
monomial reduction by inspecting Ga2 but not a minimal reduction of itself since a2
is minimally generated by three elements. On the other hand, b2 = (xy 5 +x3 y, x2 y 2 )
is a reduction of a2 by Corollary 3.7 and is not a minimal monomial reduction.
In [S] Singla studies monomial reductions of monomial ideals in the polynomial
ring k[x1 , x2 , . . . , xn ] and proves that every monomial ideal has a unique minimal
monomial reduction. In this paper we intend to find, for a non-monomial ideal a,
conditions that guarantee the existence of a monomial ideal which is integral over
a.
The notion of the Hilbert-Samuel multiplicity of an m-primary ideal is general-
ized to a module satisfying certain conditions by Buchsbaum and Rim in 1960s [BR].
It has attracted attention of both algebraists and geometers. Many nice properties
of m-primary ideals are then extended to modules, especially those in the reduc-
tion theory (see Section 2). Buchsbaum-Rim multiplicity is also used to classify
singularities of certain types (c.f. [G]).
Attempting to generalize the computation in (1.1) to Buchsbaum-Rim multi-
plicity, Jones [J] considers R = k[x, y](x,y) and proves that the Buchsbaum-Rim
multiplicity of so-called monomial modules (see definition in Section 5) of lower
rank has a graphical computation. In particular, Jones breaks her computations
into seven cases and for each case she gives a graphical interpretation. We present
a formula that summarizes the seven individual cases in [J]. The multiplicity for-
mula presented in this paper should be viewed as an improvement of the result in
[J] instead of recovering. In fact, our formula is achieved by using Theorem 3.3,
Corollary 3.7, and some computations in [J].
The paper is arranged as follows. Section 2 gives definitions of notation and
reviews of some propositions that will be applied often. Section 3 proves the main
result in monomial ideals. The proof of Theorem 3.3 utilizing properties on fiber
cones is broken down to several lemmas since each lemma has its own independent
interest. Section 4 mainly consists of examples and an application of Lemma 3.4 to
Quiñonez’s main theorem in [Q]. Indeed we suggest the readers first to look at the
examples in Section 4 before jumping into the proof of Theorem 3.3. The examples
in Section 4 are meant to provide better intuition for the condition in Theorem 3.3
which is very straightforward once visualized. Also we wish to point out that the
condition in Theorem 3.3 is a sufficient condition. It is not clear, at least to the
authors, whether or not it is possible to establish a necessary condition. These are
discussed in Section 4 as well.
16
4 C-Y. JEAN CHAN AND JUNG-CHEN LIU

Finally, Section 5 is devoted to the relationship between the Buchsbaum-Rim


multiplicity of a module and the Hilbert-Samuel multiplicity of certain ideals related
to the module. The discussion in this section leads to the formula in Theorem 5.3
that is also proved in our joint work with B. Ulrich [CLU] in a more general setting.
Here the authors would like to provide the original idea and the motivation of how
such result is formulated using the reduction theory on monomial ideals discussed
in Section 3. We refer the readers to [CLU] for more generalized results.
Acknowledgement. We would like to express our gratitude to William Heinzer
for helpful discussions on the reduction of ideals and for his endless encouragement
throughout the course of writing this paper.

2. Notation and Background


Assume R is a Noetherian local ring with maximal ideal m. Given an ideal a
of R, there are two rings that are used frequently in the study of reduction, namely
the Rees algebra R[at] of a and the fiber cone R[at]/mR[at] of a. It can be seen
directly from the definition that a subideal b ⊆ a is a reduction if and only if
the Rees algebra R[at] of a is integral over the Rees algebra R[bt] of b. In fact,
a similar result holds for their fiber cones, i.e., a subideal b ⊆ a is a reduction if
and only if the fiber cone of a is integral over that of b via the homomorphism
induced by the inclusion R[bt] ⊆ R[at] (c.f. [SH, 8.2.4]). In the proof of our main
result Theorem 3.3, we use this equivalent condition on the fiber cones to verify the
reduction relation.
If a is an m-primary ideal of R, then there exists a polynomial Pa (n) of degree
d, where d is the dimension of R, such that Pa (n) = (R/an ) for all large n. This is
d
called the Hilbert-Samuel polynomial and the coefficient of nd! is called the Hilbert-
Samuel multiplicity, denoted e(a). If we further assume that R is Cohen-Macaulay
with infinite residue field, then given an m-primary ideal a, it is known that an m-
primary subideal b of a is a reduction of a if and only if e(a) = e(b). Furthermore,
if b is a minimal reduction of a, then
(2.1) e(a) = e(b) = (R/b).
Let F be a free R-module of rank r and let M be a submodule of F with
(F/M ) < ∞. Buchsbaum and Rim [BR, 3.4] prove that there exists a poly-
nomial λ(n) such that for all large n ∈ N, λ(n) = (Sn (F )/Rn (M )), where
S(F ) = ⊕n≥0 Sn (F ) is the symmetric algebra of F and R(M ) is the R-subalgebra
of S(F ) generated by the image of M in S1 (F ). It is also proved in [BR, 3.4] that
nd+r−1
the polynomial λ(n) has degree d + r − 1 unless M = F . The coefficient of (d+r−1)!
in this polynomial is defined to be the Buchsbaum-Rim multiplicity br(M ). It is a
positive integer whenever M = F and only depends on F/M [BR, 3.3]. Note that
if r = 1 and M = F , then M is an m-primary ideal of R and br(M ) = e(M ), so
the Buchsbaum-Rim multiplicity can be viewed as a generalization of the Hilbert-
Samuel multiplicity.
Let U be a submodule of M . We write R(U ) to be the R-subalgebra of S(F )
generated by U . We say that U is a reduction of M if R(M ) is integral over R(U ) as
rings. A free module has no proper reduction. Similar to the ideal case, a minimal
reduction of M is a reduction that is minimal with respect to inclusion. When
M = F and the residue field of R is infinite, a reduction U of M is minimal if and
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 17
5

only if its minimal number of generators is d + r − 1 (c.f. [BR, 3.5], [R, 2.1 and
2.2], [EHU, page 707]).
Reductions of modules are closely related to Buchsbaum-Rim multiplicities. If
U is a reduction of M then br(U ) = br(M ) [KT, 6.3(i)], and the converse holds
in case R is universally catenary and equidimensional with d > 0 (c.f. [KR, 4.11],
[KT, 6.3(ii)], [K, 2.2], [SUV1, 5.5]). Furthermore, similar to the result (2.1) in
ideals, if R is a Cohen-Macaulay local ring with infinite residue field and if U is a
minimal reduction of M , then
(2.2) br(M ) = br(U ) = (F/U )
(c.f. [BR, 4.5(2)]).
Suppose M is a submodule of a free module F of rank r such that F/M ∼ = I/J
for two m-primary ideals I and J; for instance, if R is a Cohen-Macaulay ring of
dimension at least two, one can choose I and J to be the Bourbaki ideal of F and
M , respectively (c.f. [B, Chapter 7, no. 4], [SUV2, 3.2(a),(c)]). If r ≥ 2 and if M
is generated by three elements, then M = F or r = d = 2 [BR, 3.5]. In this case,
I and J can be chosen to be the unit ideal or an m-primary complete intersection.
Since M is its own minimal reduction, by (2.2), we have the following equalities,
br(M ) = (F/M ) = (R/J) − (R/I) = e(J) − e(I).
From this we observe that not only does the Buchsbaum-Rim multiplicity gener-
alize the Hilbert-Samuel multiplicity by definition and share parallel properties in
the reduction theory as described earlier but also the two multiplicities are con-
nected in such a special case. Such a relation was generalized in [J] when I and
J are monomial ideals with small number of generators. In Section 5, we take the
results in [J] one step further by formulating its outcome. The formula obtained in
Theorem 5.3 also motivates the work in [CLU] for arbitrary modules over a two
dimensional Gorenstein local ring.

3. Reductions of Monomial Ideals


For the rest of the paper, we let R = k[x, y](x,y) be the polynomial ring k[x, y]
localized at the maximal ideal (x, y). We also assume that k is an infinite field.
Note that by multiplying a suitable unit to a generator, we see that every ideal
in R is generated by polynomials in k[x, y]. For every element  f iinj k[x, y], f can
be written as a finite sum in distinct monomials; i.e., f = ηij x y with ηij ∈ k
where we assume no repeated like terms in the expression. We use the following
notation for the collection of the finitely many monomials occurring in f

Γ(f ) = {xi y j | f = ηij xi y j and ηij = 0}.

Remark 3.1. Let I be an ideal of R and suppose it is generated by f1 , . . . , fm ∈


k[x, y]. If I  is a monomial ideal containing I, then it is clear that I  contains
Γ(f1 ) ∪ · · · ∪ Γ(fm ). Hence the smallest monomial ideal containing I is generated
by Γ(f1 ) ∪ · · · ∪ Γ(fm ). We denote this monomial ideal by I ∗ .
We are interested in conditions under which I ∗ becomes integral over I.
Question 3.2. Under what conditions is I a reduction of I ∗ ?
18
6 C-Y. JEAN CHAN AND JUNG-CHEN LIU

Theorem 3.3 states a sufficient condition, in terms of monomials in Γ(f1 ) ∪ · · · ∪


Γ(fm ), for I to be a reduction of I ∗ . As an application, Corollary 3.7 provides a
minimal reduction of a given monomial ideal.
Theorem 3.3. Let R = k[x, y](x,y) and |k| = ∞. Let I be an ideal of R
generated by f1 , . . ., fm ∈ k[x, y]. Assume that the following is true: for all i =
1, 2, . . . , m and for any two distinct monomials xa y b and xc y d in Γ(fi ) with c < a
and b < d, there exists xr y s ∈ Γ(fj ) for some j such that the point (r, s) lies on the
left hand side of the line through (a, b) and (c, d). Then I is a reduction of I ∗ .
Prior to proving this theorem, we discuss several supporting lemmas.
Lemma 3.4. Let k[u1 , u2 , . . . , un ] be a k-algebra and consider its k-subalgebra
k[η1 u1 + η2 u2 + · · · + ηn un ] for nonzero η1 , . . . , ηn in k. For each i = 1, 2, . . . , n,
α β
suppose there are positive integers αij , βij such that ui ij uj ij = 0 for all j = i.
Then ui is integral over k[η1 u1 + η2 u2 + · · · + ηn un ]. Consequently, k[u1 , . . . , un ] is
integral over k[η1 u1 + η2 u2 + · · · + ηn un ].
Proof. First, we show that the lemma can be reduced to proving the case
α β
where η1 = η2 = · · · = ηn = 1. Note that ui ij uj ij = 0 implies (ηi ui )αij (ηj uj )βij =
0. Therefore, once the case where η1 = η2 = · · · = ηn = 1 is proved, then by
replacing u by η u for all  = 1, 2, . . . n, we have k[η1 u1 , . . . , ηn un ] is integral over
k[η1 u1 + η2 u2 · · · + ηn un ]. Since η1 , . . . , ηn are all units in k, k[u1 , u2 , . . . , un ] =
k[η1 u1 , η2 u2 , . . . , ηn un ]. This gives the general case. Hence, it suffices to show that
k[u1 , u2 , . . . , un ] is integral over k[u1 + u2 + · · · + un ] under the same hypothesis.
We prove the statement (with η1 = · · · = ηn = 1) by induction on n and
assume n ≥ 2 since the assertion is trivial for n = 1. We first show the base case
n = 2. For simplicity on the notation, in this case we write uα 1 α2
1 u2 = 0. This
α1 α2 +1
implies u1 u2 = 0 so we may assume that α2 is odd. By a straightforward
computation,

1
1 +α2
= uα α2 α2
1 (u2 + u1 ) = u
1 α1
1 
α {[(u1 + u2 ) − u1 ]α2 + uα
1 }
2

−1
= u1 [(u1+ u2 ) − 12 (u1 + u2 ) 2 u1 + · · ·
α1 α2 α
2 −1
+ αα 2
2 −1
(u1 + u2 )uα
1 ].
We conclude
 2 
uα 1 +α2
− αα (u1 + u2 )u1α1 +α2 −1 + · · ·
1 2 −1  
+ α12 (u1 + u2 )α2 −1 uα1
1 +1
− (u1 + u2 )α2 uα
1 = 0.
1

Thus u1 is integral over k[u1 + u2 ] and so is u2 .


Assume n ≥ 3 and suppose the assertion holds for all k-algebras with n − 1
or less generators. For the k-algebra k[u1 , u2 , . . . , un ] with n generators, choose
α = max{αij , βij }, then we have uαi uj = 0 for all i = j. Consider the k-algebras
α
i,j

k[u1 + · · · + un−1 + un ] ⊆ k[u1 + · · · + un−1 , un ] ⊆ k[u1 , . . . , un−1 , un ].


We claim that
(1) k[u1 , . . . , un−1 , un ] is integral over k[u1 + · · · + un−1 , un ], and
(2) k[u1 + · · · + un−1 , un ] is integral over k[u1 + · · · + un−1 + un ].
For (1), since uα i uj = 0 for all i = j, by the induction hypothesis, we have that
α

k[u1 , . . . , un−1 ] is integral over k[u1 + u2 + · · · + un−1 ] and so k[u1 , . . . , un−1 , un ] is


integral over k[u1 +u2 +· · ·+un−1 , un ]. For (2), consider the element (u1 +u2 +· · ·+
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 19
7

un−1 )(n−1)α uα n . Note that after expanding (u1 +u2 +· · ·+un−1 )


(n−1)α
, all terms are
αn−1
of the form u1 u2 · · · un−1 with α1 +α2 +· · ·+αn−1 = (n−1)α. Hence at least one
α1 α2

of the αi is larger than or equal to α. Therefore, (u1 +u2 +· · · +un−1 )(n−1)α uα n = 0.


Thus, by the case of n = 2, k[u1 +· · ·+un−1 , un ] is integral over k[u1 +· · ·+un−1 +un ].
At last, it follows from (1) and (2) that k[u1 , . . . , un ] is integral over k[u1 + · · · + un ]
and the proof is complete. 
The following two lemmas are deduced from algebraic inequalities regarding
relative positions of a point and a line. They both play important roles in the proof
of Theorem 3.3. Although only one implication in Lemma 3.5 will be needed, we
prove an equivalent statement for completion.
Lemma 3.5. Let (a, b), (c, d), (e, f ) ∈ Z2≥0 with a > c and b < d. Then the point
(e, f ) lies within the convex region with vertices (a, b), (c, d), (0, d), (0, 0), (a, 0),
including all boundaries except the line segment connecting (a, b) and (c, d), if and
only if there exist nonnegative integers α, β, γ, δ such that γ, δ are not both zero and
that
 a b α  c d β  α+β
x y x y = xγ y δ xe y f .
Simply speaking, the above monomial equality holds if and only if (e, f ) is in the
following shaded region:

(c, d)
(0, d)

(a, b)

(0, 0) (a, 0)

Proof. Suppose that (e, f ) is in the assumed region. If 0 ≤ e ≤ c < a and


0 ≤ f ≤ d and (e, f ) = (c, d), then xc y d is a proper multiple of xe y f , so we may
set α = 0, β = 1, γ = c − e, and δ = d − f to achieve the monomial equality.
Symmetrically, if 0 ≤ e ≤ a and 0 ≤ f ≤ b < d and (e, f ) = (a, b), we may take
α = 1, β = 0, γ = a − e, and δ = b − f . The remaining case is when (e, f ) lies
in the interior region of the triangle with vertices (a, b), (c, d), (c, b). The fact that
(e, f ) is on the left hand side of the line through (a, b) and (c, d) implies
f (a − c) < b(e − c) + d(a − e).
By setting α = e − c, β = a − e and δ = b(e − c) + d(a − e) − f (a − c), we have
(xa y b )α (xc y d )β = y δ (xe y f )α+β
in which α, β, δ are all positive integers for c < e < a.
Conversely, assume the existence of α, β, γ and δ. Then we have
(3.1) (a − e)α + (c − e)β = γ ≥ 0
(3.2) (b − f )α + (d − f )β = δ ≥ 0
and γ and δ are not both zero. This implies α and β are not both zero either. First
we claim that e ≤ a. Suppose not, i.e., e > a; this implies (a − e)α + (c − e)β < 0
20
8 C-Y. JEAN CHAN AND JUNG-CHEN LIU

since α and β are nonnegative and not both zero. This contradicts (3.1). Similarly,
one can show that f ≤ d. Therefore, to complete the proof, it suffices to show that
(e, f ) does not lie in the upper right triangular region (including the boundaries)
with vertices (a, b), (c, d) and (a, d). Suppose the contrary, then
c≤e≤a

b≤f ≤d

(3.3) af + bc + de − ad − be − cf ≥ 0.
First, if e = a, then (3.1) becomes (c − e)β = γ and this implies β = γ = 0 since
c − e = c − a < 0 and β, γ ≥ 0. Consequently, (3.2) becomes (b − f )α = δ and this
implies δ = 0. This contradicts to the condition that γ and δ are not both zero.
Therefore, e  a. Suppose f = b. From (3.3) one conclude (a − e)(b − d) ≥ 0. But
this cannot be true because b < d and e < a. So f  b. Symmetrically, f < d and
e > c. Therefore, c < e < a and b < f < d. Now, multiplying (3.2) by (a − e), we
have
(3.4) (a − e)(b − f )α + (a − e)(d − f )β = (a − e)δ.
Replacing (a − e)α in (3.4) using (3.1), (3.4) becomes
(3.5) (b − f )(e − c)β + (a − e)(d − f )β = (a − e)δ + (f − b)γ.
Since e < a and b < f and since δ and γ are nonnegative and not both zero, the
right hand side of (3.5) is positive. Thus, (b − f )(e − c) + (a − e)(d − f ) > 0
since β ≥ 0. This contradicts (3.3). Hence the point (e, f ) must be in the desired
region. 

Lemma 3.6. Let (a, b), (c, d), (e, f ) ∈ Z2≥0 with a > c and b < d. If the point
(e, f ) lies within the triangular region bounded by the x-axis, the line x = a, and
the line through (c, d) and (a, b) excluding the vertical side and hypotenuse, i.e., the
shaded region in Figure 3.6.1,

(c, d) (c, d)

(a, b) (0, d)
(a, b)

(0, 0) (a, 0) (0, 0)

Figure 3.6.1. Figure 3.6.2.

then there exist positive integers α, β, γ such that


 a b α+β  α  c d β
x y = xγ xe y f x y .
Symmetrically, if the point (e, f ) is in the shaded region in Figure 3.6.2, then there
exist positive integers α, β, δ such that
 c d α+β  α  e f β
x y = y δ xa y b x y .
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 21
9

Proof. Note that if (e, f ) is in the shaded region in Figure 3.6.1, then (a, b)
is on the right hand side of the line through (c, d) and (e, f ). This implies

a(d − f ) > c(b − f ) + e(d − b).

Take α = d − b, β = b − f and γ = a(d − f ) − c(b − f ) − e(d − b), then we have


the desired equality. Symmetrically, if (e, f ) is in the shaded region in Figure 3.6.2,
then (c, d) is on the right hand side of the line through (e, f ) and (a, b). We take
α = c − e, β = a − c, and δ = d(a − e) − b(c − e) − f (a − c) to obtain the desired
equality. 

Now, we are ready to prove the main theorem of this section, Theorem 3.3.

n1
Proof. We express the generators of I as the following: f1 = j=1 η1j xa1j y b1j ,
n2 nm
f2 = j=1 η2j x
a2j b2j
y , . . . , fm = j=1 ηmj x
amj bmj
y with ηij = 0 in k. Then
∗ aij bij
by Remark 3.1 I is the ideal generated by x y for all i = 1, . . . , m and
j = 1, . .. , ni . Consider the polynomial ring R[Uij ] = R Uij | i = 1, . . . , m, j =
1, . . . , ni and the ring homomorphism

ϕ : R[Uij ] −→ R[I ∗ t]
Uij −→ xaij y bij t

Then the Rees algebra of I ∗ is R[I ∗ t] ∼


= R[Uij ]/ ker ϕ and the fiber cone is

R[I ∗ t] ∼ R[Uij ]
(3.6) =
mR[I ∗ t] (mR[Uij ] + ker ϕ)

where m = (x, y)R is the maximal ideal of R.


Let uij denote the homomorphic image of Uij in R[Uij ]/(mR[Uij ] + ker ϕ). In
order to show that I is a reduction of I ∗ , it suffices to show that the fiber cone
R[I ∗ t]/mR[I ∗ t] of I ∗ is integral over the fiber cone R[It]/mR[It]  nof I. This is
j=1 1j u1j , · · · ,
1
equivalent to showing that the k-algebra k[uij ] is integral over k η
nm 
j=1 ηmj umj (c.f. Section 2 and [SH, 8.2.4]). Therefore by Lemma 3.4, for each
 β
uij , it is enough to prove that for all  = j, uα ij ui = 0 for some positive integers
α and β . Note that Uij (resp. Ui ) corresponds to xaij y bij t (resp. xai y bi t)
in the isomorphism (3.6). Without loss of generality, we assume aij ≥ ai . If
aij > ai and bij ≥ bi , then xaij y bij t = xaij −ai y bij −bi (xai y bi t). This shows
Uij − xaij −ai y bij −bi Ui ∈ ker ϕ and so Uij ∈ mR[Uij ] + ker ϕ. That is uij = 0 and
so is true uij ui = 0. Similarly for aij = ai and bij < bi (resp. bij > bi ), one can
show ui = 0 (resp. uij = 0) and so uij ui = 0.
The last case is that aij > ai and bij < bi . By the assumption of the theorem,
there exists xahs y bhs ∈ Γ(fh ) such that (ahs , bhs ) lies on the left hand side of the
line through (aij , bij ) and (ai , bi ) as shown in the following shaded region:
22
10 C-Y. JEAN CHAN AND JUNG-CHEN LIU

(ai , bi )

(aij , bij )

(0, 0)

Figure 3.3.1.

We divide Figure 3.3.1 into the following three parts:

(ai , bi ) (ai , bi )

(0, bi ) (aij , bij ) (aij , bij )

(0, 0) (aij , 0) (0, 0) (aij , 0)

Figure 3.3.2. Figure 3.3.3.

(ai , bi )

(0, bi ) (aij , bij )

(0, 0)

Figure 3.3.4.

If (ahs , bhs ) lies in the shaded region in Figure 3.3.2, then, by Lemma 3.5, there
exist nonnegative integers α, β, γ, δ with γ, δ not both zero such that
(xaij y bij t)α (xai y bi t)β = xγ y δ (xahs y bhs t)α+β .
α β α+β α β
Thus, we have Uij Ui − xγ y δ Uhs ∈ ker ϕ. That is Uij Ui ∈ mR[Uij ] + ker ϕ and
β
uα u
ij i = 0. If (a , b
hs hs ) lies in the shaded region in Figure 3.3.3, then, by Lemma 3.6,
there exist positive integers α, β, γ such that
 a b α+β  α  a b β
x ij y ij = xγ xahs y bhs x i y i ,
and this implies uα+β
ij = 0 as above. Similarly, if (ahs , bhs ) lies in the shaded region
in Figure 3.3.4, we may apply Lemma 3.6 and get  uα+β
i = 0 for somepositive inte-
n1 nm
gers α, β. These prove that uij is integral over k j=1 η1j u1j , · · · , j=1 ηmj umj

for all i, j. Hence I is a reduction of I . The proof is completed. 
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 23
11

As an immediate application of Theorem 3.3, we find a minimal reduction of


a given monomial ideal I. Note that every monomial ideal has a unique minimal
monomial generating set. The monomials corresponding to the vertices of the
convex hull of the graph of I are part of the irredundant monomial generators of
I. Indeed, if we let H denote the ideal generated by the monomials corresponding
to the vertices of the convex hull of the graph of I, then all monomial generators
of I are integral over H and thus H is a reduction of I. In order to find a minimal
reduction of I, it is enough to find a minimal reduction of H. The ideal H is called
least monomial reduction in [Q] and minimal monomial reduction in [S], but it is
not necessarily a minimal reduction as pointed out in Introduction.
Note that every principal ideal is integrally closed and is its own minimal reduc-
tion. So we are interested in monomial ideals minimally generated by at least two
elements. Although the following corollary was mentioned in [Q] and [SH], it was
among the initial results of our project proved in 2001. We later extend the study
on the monomial ideals and obtain Theorem 3.3 in its current form. Corollary 3.7
and a special case of Lemma 3.4 were quoted and utilized in Lu’s master thesis ([L],
2003) under the second author Liu’s supervision.
Corollary 3.7. Let R = k[x, y](x,y) and |k| = ∞. Let I be a monomial
ideal minimally generated by at least two elements. Suppose the convex hull of the
graph of I has vertices (a1 , b1 ), (a2 , b2 ), . . ., (an , bn ) with a1 > a2 > · · · > an and
b1 < b2 < · · · < bn . Then
  ai bi  
K= x y , xai y bi
1≤i≤n 1≤i≤n
i is odd i is even

is a minimal reduction of I.
Proof. Notice that for any two odd (resp. even) indices j < k, there exists an
even (resp. odd) index i such that j < i < k. Since (a1 , b1 ), (a2 , b2 ), . . ., (an , bn )
are vertices of a convex graph, (ai , bi ) is on the left of the line through (aj , bj ) and
(ak , bk ). By Theorem 3.3, K is a reduction of K ∗ . Since K ∗ ⊆ I and since their
graphs have the same convex hull, K ∗ is a reduction of I. Thus, K is a reduction of
I. Furthermore, as pointed out in Introduction, since I is non-principal, a reduction
of I is minimal if it is generated by two elements. Hence K is indeed a minimal
reduction of I. 
Independently, V. C. Quiñonez also obtains the result of Corollary 3.7 by prov-
ing a more general theorem on reductions of ideals. In Section 4, we recover
Quiñonez’s main theorem as an application of Lemma 3.4.

4. Examples and Remarks


4.1. Examples. This section is mainly dedicated to discuss the conditions
that make I ∗ an integral extension of I.
Let R = k[x, y](x,y) and |k| = ∞. We first give an example to illustrate Theo-
rem 3.3 and Corollary 3.7. Theorem 3.3 provides a sufficient condition under which
the ideal I ∗ is integral over I. Additional remarks are made here on the condition
in Theorem 3.3. Examples 4.2 and 4.3 show that this is not a necessary condition
for I to be a reduction of I ∗ . Example 4.3 indicates that a seemingly most intuitive
extension is still not necessary. In Example 4.4 we discuss another extension which
24
12 C-Y. JEAN CHAN AND JUNG-CHEN LIU

may likely make the condition necessary but we do not have neither a proof nor
enough evidence to make such conjecture.

Example 4.1. Consider the ideal I = (x11 + x10 y + x2 y 4 + xy 7 , x10 y 2 + x3 y 3 +


y , x5 y 2 + xy 7 ). Note that the monomials occurring in the first generator of I are
8

x11 , x10 y, x2 y 4 , xy 7 . From Figure 4.1.1, we see that for every line through two of
the above 4 monomials, either x5 y 2 or y 8 is on the left of the line; for example,
x5 y 2 is on the left of the line through x11 and x2 y 4 , and y 8 is on the left of the
line through x2 y 4 and xy 7 . Similarly, from Figure 4.1.2, we see that for each of the
three lines determined by the three monomials occurring in the second generator
of I, either x5 y 2 or x2 y 4 is on the left of the line. From Figure 4.1.3, x3 y 3 is on the
left of the line through those two monomials occurring in the third generator of I.
Hence, by Theorem 3.3, I ∗ = (x11 , x10 y, x5 y 2 , x3 y 3 , x2 y 4 , xy 7 , y 8 ) is integral over I.

y8
y8
xy 7

x2 y 4 x2 y 4
x3 y 3
x10 y 2
5 2
x y 10
x y
x5 y 2

x11
Figure 4.1.1. Figure 4.1.2.

xy 7

x3 y 3

x5 y 2

Figure 4.1.3.

On the other hand, from Figure 4.1.4, we see that the monomials corresponding
to the vertices of the convex hull of the graph of I ∗ are x11 , x5 y 2 , x3 y 3 , x2 y 4 , y 8 .
Hence, by Corollary 3.7, (x11 + x3 y 3 + y 8 , x5 y 2 + x2 y 4 ) is a minimal reduction of
I ∗.
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 25
13

y8
xy 7

x2 y 4
x3 y 3
x5 y 2
x10 y

x11
Figure 4.1.4.

Example 4.2. Consider the ideal I = (x4 , x3 y + xy 3 , y 4 ). Although I does


not satisfy the assumption in Theorem 3.3, I is indeed a reduction of I ∗ . In fact,
because the convex hull of the graph of I ∗ has vertices (4, 0) and (0, 4), the ideal
K = (x4 , y 4 ) is a reduction of I ∗ . Thus, I ∗ is integral over I. This example tells us
that the hypothesis given in Theorem 3.3 is not a necessary condition for I being
a reduction of I ∗ .
Example 4.3. Consider two ideals J1 = (x3 + x2 y + xy 2 , y 3 ) and J2 = (x3 +
x y, xy 2 + y 3 ). Note that J1 ∗ = J2 ∗ = m3 and it is integrally closed. Because
2

R is a two-dimensional regular local ring, m3 has reduction number one (c.f. [H,
5.1]). So an ideal K is a reduction of m3 if and only if Km3 = m6 . One can
check that J1 m3 = m6 and J2 m3  m6 . Hence, J1 is a reduction of J1 ∗ but J2 is
not a reduction of J2 ∗ . Note that both J1 and J2 satisfy the following condition:
for all i = 1, 2, . . . , m and for any two distinct monomials xa y b and xc y d in Γ(fi )
with c < a and b < d, there exists xr y s ∈ Γ(fj ) for some j such that the point
(r, s) lies on the line through (a, b) and (c, d), where f1 , . . . , fm generate the ideal.
This example tells us that one should not expect to generalize the condition in
Theorem 3.3 to including the line connecting (a, b) and (c, d).
Example 4.4. Consider the ideal I = (x4 , x2 y + xy 2 , y 4 ). Then the ideal
I = (x4 , x2 y, xy 2 , y 4 ) and I ∗ has reduction number one, again since it is integrally

closed in the two-dimensional regular local ring R (c.f. [H, 5.1]). It can be checked
directly t hat II ∗  (I ∗ )2 . Thus, I is not a reduction of I ∗ . On the other hand,
we know that K1 = (x4 + y 4 , x2 y + xy 2 ) is a reduction of I, also through a direct
computation: K1 I 2 = I 3 . We observe that x4 , y 4 are in Γ(x4 + y 4 ) and the term
x2 y +xy 2 is a combination of two monomials corresponding to two points on the left
hand side of the line through (4, 0) and (0, 4). It would be interesting to determine
whether or not Theorem 3.3 can be extended to state “if the left side of the line
contains points whose combination is in I, then K1 is a reduction of I.” However,
the authors are not able to verify this statement.
By Corollary 3.7, K2 = (x4 +xy 2 , x2 y+y 4 ) is a minimal reduction of I ∗ while K1
is not a reduction of I ∗ . Although the set consisting of (a1 : · · · : a4 : b1 : · · · : b4 ),
with the fact that a1 x4 + a2 x2 y + a3 xy 2 + a4 y 4 and b1 x4 + b2 x2 y + b3 xy 2 + b4 y 4
generate a minimal reduction of I ∗ , form a dense open set in P3 × P3 , this example
shows there are ample exceptions.
4.2. An application of Lemma 3.4. In [Q] Quiñonez discusses minimal
reductions of monomial ideals in k[[x, y]] and proves the following Theorem 4.5 on
the reduction of ideals in general. It states that if a nice partition on a generating
26
14 C-Y. JEAN CHAN AND JUNG-CHEN LIU

set of an ideal (not necessarily monomial) is available to result certain relationships


among the generators, then the partition provides a reduction. The existence of such
a partition may not be determined easily in general but in two-dimensional local
rings such as k[x, y](x,y) and k[[x, y]], the partition consisting of even indices and odd
indices as stated in Corollary 3.7 satisfies the conditions required in Theorem 4.5
and hence a minimal reduction of the ideal is obtained. In the following, we recover
Quiñonez’s main theorem as an application of Lemma 3.4 in the previous section.
Theorem 4.5 (Quiñonez, [Q, 3.3] ). Let I = (mi )0≤i≤r be an ideal in a
Noetherian local ring (R, m). Assume that there exists a partition on r elements:
{0, . . . , r} = 0≤α≤s Sα , where s ≤ r, such that if i, j ∈ Sα , i = j, then mli mlj ∈
 
I 2l m for some integer l. Let J = i∈Sα mi 0≤α≤s , then J is a reduction of I.

With the same strategy as proving Theorem 3.3, this theorem can be proved as
a corollary of Lemma 3.4. Precisely, consider the polynomial ring R[U0 , U1 , . . . , Ur ]
and the ring homomorphism
ϕ : R[U0 , U1 , . . . , Ur ] −→ R[It]
.
Ui −→ mi t
Then the Rees algebra of I is R[It] ∼ = R[U0 , . . . , Ur ]/ ker ϕ and the fiber cone of I
is
R[It] ∼ R[U0 , U1 , . . . , Ur ]
= .
mR[It] (mR[U0 , U1 , . . . , Ur ] + ker ϕ)
Let ui denote the homomorphic image of Ui in R[U0 , . . . , Ur ]/(mR[U0 , . . . , Ur ] +
ker ϕ). In order to show that J is a reduction of I, it suffices to show that
the fiber cone R[It]/mR[It] of I is integral over the fiber cone R[Jt]/mR[Jt] of
J.This is equivalent to showing that the k-algebra k[u0 , . . . , ur ] is integral over
k i∈Sα mi t | 0 ≤ α ≤ s , where k = R/m is the residue field of R. Note that with
the assumption that if i, j ∈ Sα , i = j, then mli mlj ∈ I 2l m for some integer l, we
have (mi t)l (mj t)l ∈ m(It)2l and this implies uli ulj = 0. Now, apply Lemma 3.4 to
complete the proof.

5. Application to Multiplicities
In this section, we apply Theorem 3.3 to revisit the computations of Buchsbaum-
Rim Multiplicity in Jones [J]. We also present a formula that summarizes the seven
individual cases concluded in [J]. As before, R = k[x, y](x,y) with |k| = ∞. The
modules under consideration are finitely generated over R and arise from the fol-
lowing Setting 5.1. These are called monomial modules (c.f. [R1, Section 4]).
Setting 5.1. Let I = (xs , y t ) and J = (xs+i , y j+t , xd y e+t ) be monomials
in R. Let F be a free module of rank two with free basis e1 , e2 . Consider the
homomorphism φ : F −→ I defined by φ(e1 ) = xs and φ(e2 ) = y t . Then φ induces
a short exact sequence

φ
(5.1) 0 −→ M −→ F −→ I/J −→ 0
where M is the kernel of the induced map. In this case, M can be identified with
the submodule of F generated by the columns of the matrix
i

x 0 0 −y t
.
0 y j xd y e xs
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 27
15

Clearly, F/M ∼ = I/J in Setting 5.1. We call the above matrix a presenting
matrix of M and denote it by M by abusing the notation since presenting matrices
of a module are not unique. However, the Fitting ideal of F/M , used most often
in the present discussion, does not depend on the choice of the presenting matrix
of M .
In [J], Jones computes the Buchsbaum-Rim multiplicity br(M ) of M as in
Setting 5.1 and compares br(M ) and e(J) − e(I). Applying reduction theory on
modules, Jones classifies the module M into seven cases and gives graphical inter-
pretation of the relation between br(M ) and e(J) − e(I). However, the algebraic
meaning of some crucial graphs is lacking. In this section, we make use of the Fitting
ideals of modules to revisit Jones’s classification and reinterpret the Buchsbaum-
Rim multiplication of M as the Hilbert-Samuel multiplicities of I, J, and certain
zeroth Fitting ideals of modules closely related to M .
The following result of Rees will be used frequently in our approach. This the-
orem enables us to transfer reduction relations between modules to those between
ideals and vice versa.

Theorem 5.2. (Rees, [R, 1.2]) Let N ⊆ M ⊆ F ∼ = Rr with (F/N ) < ∞. Then
N is a reduction of M if and only if Fitt0 (F/N ) is a reduction of Fitt0 (F/M ).

Recall that the zeroth Fitting ideal Fitt0 (F/M ) is the ideal generated by all
as in Setting 5.1, so
2 × 2-minors of the matrix M

Fitt0 (F/M ) = (xi y j , xi+s , xi+d y e , y t+j , xd y e+t ) .

Note that this is a monomial ideal so its Hilbert-Samuel multiplicity is easy to


compute using (1.1). Also, given a submodule N of M or a submodule L of F
containing M , Theorem 3.3 and Theorem 5.2 provide means to determine if N is a
reduction of M or if L is integral over M .

Theorem 5.3. Let R = k[x, y](x,y) with |k| = ∞. Let I, J, F and M be as in


Setting 5.1 such that F/M ∼
= I/J. Then,

br(M ) = e(J) − e(I) − [e(Fitt0 (F/N )) − e(Fitt0 (F/M ))] ,

where N is the submodule of F lifted from either (xs+i , y t+j ) or (xs+i +y t+j , xd y t+e )
whichever is a minimal reduction of J.

Proof. As in Section 3, we continue to use the correspondence between mono-


mials and lattice points of Z2 in R2 . Recall that J is minimally generated by three
elements xs+i , y t+j and xd y t+e . By Corollary 3.7, depending on the relative posi-
tion of the point (d, e + t) and the line segment connecting (s + i, 0) and (0, j + t),
either (xs+i , y t+j ) or (xs+i + y t+j , xd y t+e ) is a minimal reduction of the monomial
ideal J.
Before we start the discussion, it should be pointed out that the four points
corresponding to the four monomials y t+j , xd y t+e , xd+i y e and xi y j form a paral-
lelogram in all cases in this proof. This parallelogram is applied to classify various
situations. We will show how the formula is achieved via several steps.
Step 1. We discuss the case where K1 = (xs+i , y j+t ) is a minimal reduction of
J. In other words, we have the following graph:
28
16 C-Y. JEAN CHAN AND JUNG-CHEN LIU

(0, t + j)
(d, e + t)

(s + i, 0)

Figure A.

Recall the map φ in Setting 5.1 and notice that the pre-image of K1 under φ is the
submodule N1 of F with the following presenting matrix

1 = xi 0 −y t
N .
0 yj xs

The Fitting ideal Fitt0 (F/N1 ) is monomial and

Fitt0 (F/M ) = Fitt0 (F/N1 ) + (xi+d y e , xd y e+t ).

It is already clear that xd y e+t is integral over (xi+s , y j+t ) ⊆ Fitt0 (F/N1 ). Thus
whether or not N1 is a minimal reduction of M depends on whether or not xi+d y e
is integral over Fitt0 (F/N1 ) by Theorem 5.2.
Step 1.1. Depending on the location of the point (i, j), the shape of the convex
hull of the graph of Fitt0 (F/N1 ) is one of the following two:

P (0, t + j) P (0, t + j)
(d, e + t)

(i, j) (i, j)
(i + d, e)

Q(s + i, 0) Q(s + i, 0)

Figure A1. Figure A2.

If it is the case of Figure A1, the entire parallelogram with vertices (i, j),
(0, t + j), (d, e + t), (i + d, e) is on the right hand side of the line segment P Q and so
xi+d y e is integral over Fitt0 (F/N1 ). We note that A1 also includes the case where
the positions of (i, j) and (d, e + t) are exchanged. In the case of Figure A2, there
are two possibilities depending on the point (i + d, e) being on the right or left of
the line segment connecting (i, j) and (s + i, 0):
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 29
17

P (0, t + j) (d, e + t) P (0, t + j) (d, e + t)

(i, j)

(i + d, e)
(i, j)
(i + d, e)

Q(s + i, 0) Q(s + i, 0)

Figure A2.1. Figure A2.2.

In the case of Figure A2.1, xi+d y e is also integral over Fitt0 (F/N1 ). If it is
the case of Figure A2.2, then xi+d y e is not integral over Fitt0 (F/N1 ) and thus N1
is not a reduction of M . We deal with this case in the next step. Therefore, we
conclude that in the cases of Figure A1 and Figure A2.1, N1 is a minimal reduction
of M and by (2.2)

br(M ) = br(N1 ) = (F/N1 ) = (I/K1 )


= (R/K1 ) − (R/I)
= e(J) − e(I)

Step 1.2. For the case of Figure A2.2, the convex hull of Fitt0 (F/M ) is as in
Figure B.

P (0, t + j) (d, e + t)

(i + d, e)
(i, j)

Q(s + i, 0)

Figure B.

The result in [J] shows that the multiplicity of such module M is

br(M ) = e(J) − e(I) − 2(dark area)

where the dark area refers to the triangular area in both Figures C1 and C2.
30
18 C-Y. JEAN CHAN AND JUNG-CHEN LIU
P (0, t + j) P (0, t + j)

X(d, e + t) X(d, e + t)
(0, t) T (s, t) T (s, t)

Y (i, j)

Z(i + d, e)

(s, 0) Q(s + i, 0) Q(s + i, 0)

Figure C1. Figure C2.

Figure C1 is an original graph in [J]. We add a few more points corresponding to the
generators of the convex hull of the graph of Fitt0 (F/M ) to obtain an alternative
graph as shown in C2. We utilize parallel lines in Figure C2 and observe that
P XT and Y ZQ have the same area. By (1.1), it is clear that
2(dark area) = e(xs+i , y t+j , xi y j ) − e(xs+i , y t+j , xi y j , xd+i y e )
(5.2)
= e(Fitt0 (F/N1 )) − e(Fitt0 (F/M )).

The second equality is straightforward since xd y e+t is integral over (xs+i , y t+j , xi y j ,
xd+i y e ) by the assumption. Thus, we have
(5.3) br(M ) = e(J) − e(I) − [e(Fitt0 (F/N1 )) − e(Fitt0 (F/M ))].
Note that (5.3) also holds for the cases in the previous step since N1 in Step 1.1 is
a minimal reduction of M .
Recall that K1 = (xs+i , y t+j ) is assumed to be a minimal reduction of J in
this current Step 1. We would like to emphasize that the module N1 in (5.3) is a
submodule in the rank two free module lifted by a minimal reduction (xs+i , y t+j )
of J. This observation is a key to the conclusion of the next step.
Step 2. We discuss the case where K2 = (xs+i + y j+t , xd y e+t ) is a minimal
reduction of J. In other words, we have the following graph:

P (0, t + j)

(d, e + t)

Q(s + i, 0)

Figure D.

Note that the pre-image of K2 under φ is the submodule N2 of F with a presenting


matrix i

x 0 −y t
N2 =
y j xd y e xs
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 31
19

The zeroth Fitting ideal Fitt0 (F/N2 ) of F/N2 is (xi+d y e , xi+s + y j+t , xd y e+t ). No-
tice that the point xd y e+t is on the left hand side of the line P Q, so by Theorem 3.3,
the monomial ideal a = (xi+d y e , xi+s , y j+t , xd y e+t ) is integral over Fitt0 (F/N2 ).
Moreover, Fitt0 (F/M ) = a + (xi y j ). Thus, in order to determine if N2 is a minimal
reduction of M , i.e., Fitt0 (F/N2 ) is a reduction of Fitt0 (F/M ), it is equivalent to
determining if a is a reduction of Fitt0 (F/M ). And it suffices to check if xi y j is
integral over a. Depending on the location of the point (i + d, e), the shape of the
convex hull of the graph of a is one of the following three:

(0, t + j) (i, j) (0, t + j)


(i, j)

(d, e + t)
(i + d, e) (i + d, e)
(d, e + t)

(s + i, 0) (s + i, 0)

Figure D1. Figure D2.

(0, t + j)

(d, e + t)
(i, j)

(i + d, e)

(s + i, 0)

Figure D3.

Step 2.1. Note that in the cases of Figures D1 and D2, the entire parallelogram
with vertices (i, j), (0, t + j), (d, e + t), (i + d, e) is inside the convex hull of the
graph of a, so xi y j is integral over a and hence N2 is a minimal reduction of M .
Therefore again by (2.2) and the fact that F/N2 ∼ = I/K2 ,

br(M ) = br(N2 ) = (F/N2 ) = (I/K2 )


= (R/K2 ) − (R/I)
= e(K2 ) − e(I) = e(J) − e(I).

The last two equalities hold due to the fact that K2 is a minimal reduction of J.
This shows
br(M ) = e(J) − e(I).
Step 2.2. In the case of Figure D3, using the same parallelogram as above,
the point (i, j) is not in the convex hull of the graph of a. So N2 is not a minimal
reduction of M . As Step 1.2, we quote the result in [J] which shows br(M ) as
follows:
32
20 C-Y. JEAN CHAN AND JUNG-CHEN LIU
P (0, t + j) P (0, t + j)

Y (i, j)
X(d, e + t) X(d, e + t)
(0, t) T (s, t) T (s, t)

Z(i + d, e)

(s, 0) Q(s + i, 0) Q(s + i, 0)

Figure E1. Figure E2.

(5.4) br(M ) = e(J) − e(I) − 2(dark area) + 2(light area)


where the dark area is the area of P XT and the light area is that of P XQ in
both Figures E1 and E2. Similar to the discussion in Step 1.2, by moving P XT
to its similar triangle Y ZQ, we now have
(5.5)
br(M ) = e(J) − e(I) − [e(xs+i , y t+j , xi y j ) − e(Fitt0 (F/M ))] + [e(xs+i , y t+j ) − e(J)].
We would like to take a moment to look at (5.5) closer. Recall that in the
formula (5.2) in Step 1.2, (xs+i , y t+j , xi y j ) is the Fitting ideal of the submodule
N1 of F lifted from the ideal K1 = (xs+i , y t+j ) which is a minimal reduction of
J in the case of Step 1.2. However, in the present case, the ideals (xs+i , y t+j )
and J have no reduction relation. Formulas (5.3) and (5.5) together inspire us to
take a minimal reduction of J and take its lifting in F into consideration. More
precisely, in the right hand side of (5.5), we consider K2 in place of (xs+i , y t+j ) and
replace (xs+i , y t+j , xi y j ) by Fitt0 (F/N2 ). Then, we compare e(J) − e(I) − br(M )
and e(Fitt0 (F/N2 )) − e(Fitt0 (F/M )). Note that by (5.4), we have
e(J) − e(I) − br(M ) = 2(the area of P XT − the area of P XQ).
On the other hand, although the ideal Fitt0 (F/N2 ) is not monomial, we recall that
the monomial ideal a is integral over Fitt0 (F/N2 ); thus, e(Fitt0 (F/N2 )) = e(a).
Moreover, the difference between the convex hull of the graph of a and that of
Fitt0 (F/M ) is the triangular region P Y Z. Hence
e(Fitt0 (F/N2 )) − e(Fitt0 (F/M )) = e(a) − e(Fitt0 (F/M ))
= 2(the area of P Y Z).
Two quantities in comparison become 2(the area of P XT − the area of P XQ)
and 2(the area of P Y Z). By straightforward computation, we find that both
quantities are equal to td − i(j − e) and we have
(5.6) br(M ) = e(J) − e(I) − [e(Fitt0 (F/N2 )) − e(Fitt0 (F/M ))] .
We note that (5.6) is also satisfied for the cases discussed in Step 2.1 where N2 is
indeed a minimal reduction of M .
Hence, the desired formula follows either Step 1 or Step 2 depending on which
minimal reduction J possesses. 
A NOTE ON REDUCTIONS OF MONOMIAL IDEALS IN k[x, y](x,y) 33
21

Theorem 5.3 describes the Buchsbaum-Rim multiplicity of a monomial module


as the Hilbert-Samuel multiplicities of certain ideals closely related to the module.
The graphical argument applied in the proof presented here is limited to the mono-
mial modules of rank two with small number of generators as in Setting 5.1. For
modules of higher rank or with bigger number of generators, although one can try
to do similar observation and work as done in [J], one has to discuss a lot more
cases and their classification is much more complicated. For instance, in the case
where the module M ∼ = I/J has rank two as in Setting 5.1 but with J generated by
four monomials instead of three, there are more than 50 cases (see [L]). This also
shows a formulated result such as Theorem 5.3 is desirable in order to extend the
outcome to modules of higher ranks. One also notice that the terms involved in the
formula in Theorem 5.3 are defined even if I and J are not monomial ideals. For
modules of higher rank, and not being restricted to monomial quotients, we refer
to [CLU, 2.4 and Section 3] where linkage theory of ideals are utilized.

References
[B] N. Bourbaki, Commutative Algebra, Herman, Paris, 1972.
[BR] D. A. Buchsbaum and D. S. Rim, A generalized Koszul complex. II. Depth and multiplicity,
Trans. Amer. Math. Soc. 111 (1963), 197-224.
[CLU] C-Y. J. Chan, J.-C. Liu, and B. Ulrich, Buchsbaum-Rim multiplicity as Hilbert-Samuel
multiplicity, J. Algebra, 319 (2008), 4413-4425.
[EHU] D. Eisenbud, C. Huneke and B. Ulrich, What is the Rees algebra of a module?
Proc. Amer. Math. Soc. 131 (2002), 701-708.
[G] T. Gaffney, Multiplicities and equisingularity of ICIS germs, Invent. Math. 123 (1996),
209-220.
[H] C. Huneke, Complete ideals in two-dimensional regular local rings, in Commutative Algebra
(Berkeley, CA, 1987) Math. Sci. Res. Inst. Publ., 15, New York, Springer (1989), 325-338.
[J] E. Jones Computations of Buchsbaum-Rim multiplicities, J. Pure Appl. Algebra 162 (2001),
37-52.
[K] D. Katz, Reduction criteria for modules, Comm. Algebra 23 (1995), 4543-4548.
[KR] D. Kirby and D. Rees, Multiplicities in graded rings I: The general theory, Contemp.
Math. 159 (1994), 209-267.
[KT] S. Kleiman and A. Thorup, A geometric theory of the Buchsbaum-Rim multiplicity, J.
Algebra 167 (1994), 168-231.
[L] S.-Y. Lu, Computations of Samuel multiplicities and Buchsbaum-Rim multiplicities, Master
Thesis, National Taiwan Normal University, 2003.
[NR] D. G. Northcott and D. Rees, Reductions of ideals in local rings, Proc. Cambridge Philos.
Soc. 50 (1954), 145-158.
[Q] V. C. Quiñonez, Minimal reductions of monomial ideals, Research Reports in Mathemat-
ics, Number 10 (2004), Department of Mathematics, Stockholm University, available at
http://www.math.su.se/reports/2004/10/.
[R] D. Rees, Reduction of modules, Math. Proc. Cambridge Philos. Soc. 101 (1987), 431-450.
[RS] D. Rees and R. Y. Sharp, On a theorem of B. Teissier on multiplicities of ideals in local
rings, J. London Math. Soc. (2) 18 (1978), 449-463.
[R1] P. C. Roberts, Multiplicities and Chern classes, Contemporary Mathematics 159 (1994),
333-350.
[R2] P. C. Roberts, Multiplicities and Chern Classes in Local Algebra, Cambridge University
Press, Cambridge, 1998.
[S] P. Singla, Minimal monomial reductions and the reduced fiber ring of an extremal ideal,
Illinois J. Math. 51 (2007), 1085-1102.
[SH] I. Swanson and C. Huneke, Integral Closure of Ideals, Rings, and Modules, Cambridge
University Press, Cambridge, 2006.
[SUV1] A. Simis, B. Ulrich and W.V. Vasconcelos, Codimension, multiplicity and integral
extensions, Math. Proc. Cambridge Philos. Soc. 130 (2001), 237-257.
34
22 C-Y. JEAN CHAN AND JUNG-CHEN LIU

[SUV2] A. Simis, B. Ulrich and W.V. Vasconcelos, Rees algebras of modules, Proc. London
Math. Soc. 87 (2003), 610-646.

Department of Mathematics, Central Michigan University, Mt. Pleasant, MI 48809,


U.S.A.
E-mail address: chan1cj@cmich.edu

Department of Mathematics, National Taiwan Normal University, Taipei, Taiwan


E-mail address: liujc@math.ntnu.edu.tw
Contemporary Mathematics
Volume 555, 2011

Plücker–Clebsch formula in higher dimension

Ciro Ciliberto and Vincenzo Di Gennaro

Abstract. Let S ⊂ Pr (r ≥ 5) be a nondegenerate, irreducible, smooth,


complex, projective surface of degree d. Let δS be the number of double
pointsof ageneral projection of S to P4 . In the present paper we prove that
δS ≤ d−22
, with equality if and only if S is a rational scroll. Extensions to
higher dimensions are discussed.

1. Introduction
Let X be a nondegenerate, irreducible, smooth, projective variety of dimension
n ≥ 1 in the complex projective space Pr with r ≥ 2n + 1. A general projection
to Ps , with 2n + 1 ≤ s ≤ r, induces an isomorphism of X with its image. A
general projection to P2n induces an isomorphism of X with its image, except for
a finite set of points of X, which correspond to a certain number δX of improper
double points of the image, i.e. double points with tangent cone formed by two
linear spaces of dimension n spanning P2n . The double point formula (see [7], pg.
166) expresses δX in terms of invariants of X. When  X is a curve of genus g and
degree d, the double point formula says that δX = d−1 − g. This is the classical
d−1 2
Plücker–Clebsch formula. In particular δX ≤ 2 and equality holds if and only
if X is a rational curve.
In §2 of this paper we prove a similar result for surfaces:
Theorem 1.1. Let S ⊂ Pr , with r ≥ 5, be a nondegenerate, irreducible, smooth,
projective surface of degree d. Then
 
d−2
(1.1) δS ≤
2
with equality if and only if S is a rational scroll.
In § 3 we examine the higher dimensional case, in the attempt of proving a
similar theorem, but we obtain only partial results (see Proposition 3.4 and Remark
3.5, (iii) and (iv)). In this
 context
 there is some evidence supporting a conjecture
to the effect that δS ≤ d−2 2 − g, where g is the sectional genus of the surface S.
We are able to prove this only in some cases, e.g. when the Kodaira dimension

2010 Mathematics Subject Classification. Primary 14J99; Secondary 14M20, 14N15.


Key words and phrases. Projective surface, Double point formula, Rational normal scroll.

1
35
36
2 CIRO CILIBERTO AND VINCENZO DI GENNARO

 g3.5, (i) and (ii)). However in §4 we show the


κ(S) of S is not positive (see Remark
intermediate inequality δS ≤ d−2 2 − 2 (see Theorem 4.1).

2. Proof of Theorem 1.1


Remark 2.1. (i) We say that S ⊂ Pr is a scroll if it is a P1 –bundle over a
smooth curve C, and the restriction of OS (1) to a fibre is OP1 (1). If C ∼ = P1 , the
scroll is said to be rational. In this case S is isomorphic, via projection, to a rational

normal scroll S  ⊂ Pr with r  = deg(S) + 1 (see [6]).
(ii) Let HS be the general hyperplane section of a smooth surface S ⊂ Pr . The
line bundle OS (KS + HS ) is spanned, unless S is either a scroll, or S ∼ = P2 and

OS (HS ) = OP2 (i), with 1 ≤ i ≤ 2, in which cases OS (KS + HS ) is not effective (see
[10], Theorem (0.1)).
(iii) Assume S ⊂ Pr , with r ≥ 5, is a smooth, irreducible, nondegenerate
surface. The double point formula says that
d(d − 5)
δS = − 5(g − 1) + 6χ(OS ) − KS2 ,
2
where g and KS denote the sectional genus and a canonical divisor of S, and
χ(OS ) = 1 + pa (S), where pa (S) is the arithmetic genus of S. Hence
 
d−2
(2.1) δS − = 6χ(OS ) − KS2 − 5(g − 1) − 3.
2
Note that δS = 0 if and only if S is secant defective, i.e. dim(Sec(S)) < 5, where
Sec(X) denotes the secant variety of a variety X, i.e. the Zariski closure of the
union of all lines spanned by distinct points of X. A theorem of Severi’s implies
that a smooth surface is secant defective if and only if it is the Veronese surface of
degree 4 in P5 (see [2, 11]).
Proposition 2.2. In the above setting, if OS (KS + HS ) is not spanned, then
(1.1) holds, with equality if and only if S is a rational scroll.
Proof. If S is the Veronese surface of degree 4 in P5 , we have δS = 0 and
the assertion holds. Otherwise S isa scroll,
 hence χ(OS ) = 1 − g, KS2 = 8(1 − g).
Plugging into (2.1) we obtain δS − 2 = −3g and the assertion follows.
d−2

In order to prove Theorem 1.1, it suffices to prove the following:
Proposition 2.3. Assume that OS (KS + HS ) is spanned. Then (1.1) holds
with strict inequality.
We collect some preliminaries in the following lemma.
Lemma 2.4. Let S ⊂ Pr (r ≥ 5) be a nondegenerate, irreducible, smooth,
projective surface of degree d and sectional genus g. Denote by e the index of
speciality of the general hyperplane section HS of S, i.e. e := max {t ∈ Z :
H 1 (HS , OHS (t)) = 0}.
(i) If g > d − 2 then (1.1) holds with strict inequality.
(ii) If pa (S) ≤ 0 then (1.1) holds, with equality if and only if S is a rational
scroll.
(iii) If e = 1 then (1.1) holds, with equality if and only if S is a rational scroll.
(iv) If e = 1 then pa (S) ≤ g − d2 .
PLÜCKER–CLEBSCH FORMULA IN HIGHER DIMENSION 37
3

Proof. (i) Let NS,Pr be the normal bundle of S in Pr . We have


(2.2) c2 (NS,Pr (−1)) = 2(KS2 − 6χ(OS )) + 4KS HS + 6HS2 ≥ 0
the last inequality holding because NS,Pr (−1) is spanned. Then 6χ(OS ) − KS2 ≤
4(g − 1) + d. Plugging into (2.1) we obtain
 
d−2
(2.3) δS − ≤d−g−2
2
and the assertion follows.
(ii) By Proposition 2.2 it suffices to prove that: if pa (S) ≤ 0 and OS (KS + HS )
is spanned then (1.1) holds with strict inequality. If OS (KS + HS ) is spanned then
(KS + HS )2 ≥ 0, therefore KS2 ≥ d − 4(g − 1). Plugging into (2.1) we get
 
d−2
(2.4) δS − ≤ 6pa (S) − g − d + 4.
2
 
If pa (S) ≤ 0 then δS − d−2
2 ≤ −g − d + 4, which is negative unless d = 4, in which
case (ii) is trivial.
(iii) From the Poincaré residue sequence tensored with OS (i − 1)
0 → ωS (i − 1) → ωS (i) → ωHS (i − 1) → 0
we get

+∞
(2.5) pg (S) ≤ h1 (HS , OHS (i)),
i=1

where pg (S) := h (S, OS (KS )) is the geometric genus of S. If e ≤ 0 from (2.5) we


0

deduce pa (S) ≤ pg (S) ≤ 0, and (1.1) holds by (ii). If e ≥ 2 then 2g − 2 − 2d ≥ 0,


 
i.e. g ≥ d + 1, and by (i) we have (1.1) with strict inequality. When δS = d−2 2
the previous argument yields e ≤ 0, so pa (S) ≤ 0, and S is a rational scroll by (ii).
(iv) If e = 1, from (2.5) we deduce pa (S) ≤ h1 (HS , OHS (1)). By Clifford
Theorem we have h0 (HS , OHS (1)) ≤ 1 + d2 . Therefore pa (S) ≤ h1 (HS , OHS (1)) =
h0 (HS , OHS (1)) − (1 − g + d) ≤ g − d2 . 
We are now in position to prove Proposition 2.3.
Proof of Proposition 2.3. Consider the adjunction map φ : S → PR de-
fined by |KS + HS |. By [10], Theorem (0.2) and [1], Lemma 1.1.3, Lemma 10.1.1
and Theorem 10.1.3, we know that if dim(φ(S)) ≤ 1 then S is birationally ruled.
Hence pa (S) ≤ 0, and Proposition 2.3 follows by Lemma 2.4, (ii).
Suppose dim(φ(S)) = 2. By Lemma 2.4, (ii), we may assume pa (S) > 0. Hence
κ(S) ≥ 0, so we may apply [1], Lemma 10.1.2, which ensures that (KS + HS )2 ≥
2(pa (S) + g − 2), i.e. KS2 ≥ 2χ(OS ) − 2(g − 1) + d − 4. Hence by (2.1) it suffices to
prove that 2χ(OS ) − 2(g − 1) + d − 4 > 6χ(OS ) − 5(g − 1) − 3, i.e. that
(2.6) 4pa (S) < 3g + d − 8.
To prove this, note that by Lemma 2.4, (i), (iii) and (iv), we may assume g ≤ d−2,
e = 1 and pa (S) ≤ g − d2 . So we have
4pa (S) ≤ 4g − 2d < 3g + d − 8.
This proves (2.6), concluding the proof of Proposition 2.3 and of Theorem 1.1. 
38
4 CIRO CILIBERTO AND VINCENZO DI GENNARO

Remark 2.5. Theorem 1.1 holds even if S ⊂ P4 is an irreducible, nondegenerate


surface of degree d with only δS improper double points as singularities. Again (1.1)
holds, with equality if and only if S is the projection in P4 of a smooth rational
normal scroll in Pd+1 . There is no difficulty in adapting the above argument, hence
we will not dwell on this.

3. Results in higher dimension


Let X ⊂ Pr be a nondegenerate, irreducible, smooth, projective variety of
dimension n and degree d, with r ≥ 2n + 1. In view of Plücker–Clebsch formula
in the 1–dimensional case and of Theorem 1.1 for n = 2, one may ask whether in
general
 
d−n
(3.1) δX ≤ ,
2
with equality if and only if X is a rational scroll, i.e. the projection of a rational
normal scroll in Pd+n−1 (see [6]).
Remark 3.1. When X is a scroll, i.e. when X is a Pn−1 –bundle over a smooth
curve of genus g and the restriction
  of OX(1) to a fibre is OPn−1 (1), then the double
n+1
point formula gives δX − d−n 2 = − 2 g.

Before proceedings further, let us recall a geometric interpretation of


vX := cn (NX,Pr (−1))
where NX,Pr is the normal bundle of X in Pr . Note that vX ≥ 0 because NX,Pr (−1)
is spanned.
Let Tan(X) be the tangential variety of X, i.e. the Zariski closure of the
union of all tangent spaces to X at smooth points, which makes sense even if X is
singular. Denote by tX the degree of Tan(X). One has dim(Tan(X)) ≤ 2n and, if
strict inequality holds, X is called tangentially defective, whereas X is called secant
defective if dim(Sec(X)) < 2n + 1.
Note that Tan(X) is contained in Sec(X). If X is smooth, then it is secant
defective if and only if Tan(X) = Sec(X) (see [11]). Hence, if X is smooth and
tangentially defective, then it is also secant defective, but the converse does not
hold in general.
If X is not tangentially defective, there are finitely many tangent spaces to X
containing a general point of Tan(X). Let wX be their number. It is a question,
on which we will not dwell here, whether wX = 1 if X is smooth and not secant
defective. However, one may have wX > 1 when X is either secant defective or
singular: e.g., consider the cases X is the Veronese surface of degree 4 in P5 , where
wX = 2, and X is a surface lying on a 3–dimensional, nondegenerate cone with
vertex a line in P5 , where wX can be as large as we want.
The following lemma is known to the experts. We give a proof for completeness.
Lemma 3.2. If X ⊂ Pr , with r ≥ 2n, is a smooth, irreducible, nondegenerate
variety of dimension n, then vX = 0 if and only if X is tangentially defective,
whereas vX = tX wX ≥ tX if X is not tangentially defective.
Proof. Consider a general projection φ : X → P2n−1 and let Z be the rami-
fication scheme of φ on X. By the generality assumption about φ, Z is reduced of
PLÜCKER–CLEBSCH FORMULA IN HIGHER DIMENSION 39
5

finite length Z . One has Z = 0 if and only if X is tangentially defective, whereas


Z = tX wX otherwise. Look at the exact sequence

(3.2) 0→ OX → NX,Pr (−1) → Nφ (−1) → 0
r−2n+1

where Nφ is the normal sheaf to the morphism φ (see [3], p. 358, sequence (2.2)).
This is locally free of rank n − 1 off Z, where there is torsion, with lenght equal to
Z . Taking Chern classes of the sheaves in (3.2), the assertion follows. 

Remark 3.3. Let X ⊂ Pr , with r ≥ 2n + 1, be a smooth, irreducible, nonde-


generate of dimension n. Then δX = 0 if and only if X is secant defective. If HX
is a general hyperplane section of X, then X is tangentially defective if and only if
HX is secant defective, i.e. vX = 0 if and only if δHX = 0. This is a consequence
of Terracini’s Lemma (see [11]).
Going back to the question about the validity of (3.1), the arguments in the
proof of Theorem 1.1, based on Surface Theory, do not apply for n ≥ 3. However,
comparing δX with the analogous number for a general hyperplane section of X,
we may prove the following:
Proposition 3.4. Let X ⊂ Pr be a nondegenerate, irreducible, smooth, projec-
tive variety, of dimension n ≥ 2 and degree d, with r ≥ 2n+1. Let Y := HX ⊂ Pr−1
be a general hyperplane section of X and C ⊂ Pr−n+1 a general curve section of
X. Let g and e be the genus and the index of speciality of C. Then:
(i) 2(δY − δX ) = vX ;
    d−(n−1)
(ii) δX − d−n2 − δ Y − 2 = d − n − 12 vX ;

(iii) δX ≤ δY , with equality if and only if δY = 0;


     
(iv) δX − d−n 2 − δY − d−(n−1)
2 ≤ d − n, with equality as in (iii);
d−n n−2
(v) δX − 2 ≤ (n − 2)(d − n) + 2 with equality only if n = 2 and X is a
rational scroll;
   
(vi) δX − d−n 2 < −g + (n − 1)(d − n) + n−12 .
  d−n
In particular, if either g > (n − 1)(d − n) + n−1 2 , or e ≥ 2(n − 1), then δX < 2 .

Proof. Let ϕ : X → P2n and ψ : Y → P2(n−1) be general projections. Denote


by c(TX )−1 = 1 + s1 + · · · + sn the inverse total Chern class of X, with si ∈ Ai (X)
(we abuse notation and denote in the same way elements of An (X) and their degree:
we did this already a few times above). By the double point formula ([7], p. 166),
2(δY − δX ) is equal to
n  


2n
(3.3) c(ϕ TP2n )c(TX )−1 n − c(ψ ∗ TP2(n−1) )c(TY )−1 n−1 = HXi
sn−i .
i=0
i

Since c(NX,Pr ) = (1 + HX )r+1 c(TX )−1 , we have


 n  
n−i r − n − i n−i
(3.4) cn (NX,Pr (−1)) = (−1) HX ci (NX,Pr )
i=0
n − i
40
6 CIRO CILIBERTO AND VINCENZO DI GENNARO

⎡ ⎤
n   i  
r − n − i n−i ⎣ r + 1
= (−1)n−i HX HX sj ⎦
i−j

i=0
n − i j=0
i − j
 i   
 
n
i−h r − 2n + i − h r+1 i
= (−1) HX sn−i .
i=0
i − h h
h=0
For any i ∈ {0, . . . , n} and any r ≥ 2n + 1 one has
   i   
2n i−h r − 2n + i − h r+1
= (−1) .
i i−h h
h=0

Comparing (3.3) with (3.4) we obtain (i).


Property (ii) follows from (i). Properties (iii) and (iv) follow from (i) and (ii)
and Remark 3.3.
Iterating n − 2 times (iv), and denoting by S the general surface section of X,
to which we apply Theorem 1.1, we obtain
(3.5)        
d−n d−2 n−2 n−2
δX − ≤ δS − +(n−2)(d−n)+ ≤ (n−2)(d−n)+ .
2 2 2 2
If n = 2, equality between the extremes holds only if X is a rational scroll by
Theorem 1.1. If n > 2 the equality cannot hold. Otherwise S is a rational scroll,
therefore also X is a rational scroll (see [6]), hence the leftmost term in (3.5) is zero
(see Remark 3.1), whereas the rightmost term is not.
Property (vi) follows
 by applying (ii) to S in the middle term of (3.5), by
recalling that g = d−12 − δC and noting that vS > 0 because S, being smooth, is
not tangentially defective (see Lemma 3.2 and (5.37) of [8]).
For the final assertion
n−1notice
 that if e ≥ 2(n − 1) then g ≥ (n − 1)d + 1, hence
−g + (n − 1)(d − n) + 2 < 0. 
Remark 3.5. (i) In the setting of Proposition 3.4, assume X is not tangentially
defective. Then none of its general linear section is tangentially defective. In view
of Proposition 3.4, (ii), one may ask whether
(3.6) vX ≥ 2(d − n)
or, rather
(3.7) tX ≥ 2(d − n).
If so, applying (3.6) to the successive general linear sections of X one would deduce
 
d−n
(3.8) δX ≤ − g.
2

 δX = 0 and (3.8) holds in this


Note that, if X is tangentially defective, then
d−n
case, since Castelnuovo’s bound implies g < 2 .
In conclusion one is lead to the following:
Question. Let X ⊂ Pr be a nondegenerate, irreducible, smooth, projective
variety of dimension n and degree d, with r ≥ 2n + 1. Is it true that (3.8) holds,
with equality if and only if g = 0, hence X is a rational scroll?
When n = 1 the inequalities (3.6), (3.7) and (3.8) are obvious. In fact in this
case wX = 1 and therefore tX = vX . Moreover c1 (NX,Pr (−1)) = KX + 2HX , hence
PLÜCKER–CLEBSCH FORMULA IN HIGHER DIMENSION 41
7

 
vX = tX = 2g − 2 + 2d ≥ 2(d − 1), and d−1 2 − δX = g is the Plücker–Clebsch
formula.
In case n = 2, in view of Proposition 3.4, (ii), one has vX > 0 because X, being
smooth, is never tangentially defective and, by (2.2), (3.6) reads KX 2
− 6χ(OX ) +
4g − 2 ≥ 0.  
If X is smooth and nondegenerate in P4 then (3.8) becomes g ≤ d−2 2 , which
  
holds by Castelnuovo’s bound. On the contrary, if X ⊂ P3 then g = d−1 2 > d−2
2
for d ≥ 3, so in this case (3.8) is false.
(ii) In case n = 2, if pa (X) ≤ 0, e.g. if κ(X) ≤ 0, then (3.8) holds. In fact,
taking into account the proof of Proposition 2.2, in order to prove (3.8) one may
assume that OX (KX + HX ) is spanned. In this case, by (2.4), one has
 
d−2
δX − ≤ 6pa (X) − g − d + 4 ≤ −g − d + 4 < −g
2
as soon as d ≥ 5 (the case d = 4 is obvious). Further cases in which we are
able to prove (3.8) are the following (we omit the proof): complete intersections;
surfaces contained in a threefold of minimal degree; surfaces contained in a smooth
hypersurface of P5 of degree t with d ≥ t3 (in order to prove (3.8), one may assume
X ⊂ P5 , and any smooth surface in P5 is contained in some smooth hypersurface);
surfaces for which 3χ(OX ) ≥ d2 ; arithmetically Cohen-Macaulay surfaces contained
in a threefold
d−2 of degree s, with d s. Actually in the first three cases we find
δX − 2 ≤ −2g, and in a smooth threefold of minimal degree in P5 there are
   
surfaces X for which δX − d−22 = −2g with g > 0, and surfaces X with δX − d−2 2 >
−3g.
(iii) By Proposition 3.4, for varieties X (if any) for which (3.1) fails, one has
g ≤ (n − 1)d and e ≤ 2(n − 1) − 1. On the other hand, using a similar argument as
in the proof of (2.5), one may prove that
+∞ 
 
i−1 1
(3.9) pg (X) ≤ h (C, OC (i)).
i=n−1
n−2

Since h1 (C, OC (i)) ≤ g, then


2(n−1)−1    
 i−1 2(n − 1)
(3.10) pg (X) ≤ g≤ (n − 1)d = O(d).
i=n−1
n−2 n−1
(iv) As a consequence we can prove that if X ⊂ Pr is arithmetically Cohen–
Macaulay and d r then (3.1) holds with strict inequality. In fact, when X is
arithmetically Cohen-Macaulay, equality holds in (3.9), and for any i ≥ 0 one has
+∞
h1 (C, OC (i)) = j=i+1 (d − hΓ (j)) (Γ is the general 0-dimensional linear section of
X and hZ is, as usual, the Hilbert function of a projective scheme Z). We deduce
+∞ 
 
i−1
pg (X) = (d − hΓ (i)).
i=n
n−1
 
When δX ≥ d−n 2 , (3.10) says that pg (X) ≤ O(d), so we have
+∞ 
 
i−1
(d − hΓ (i)) ≤ O(d).
i=n
n−1
One sees that this is impossible if d r.
42
8 CIRO CILIBERTO AND VINCENZO DI GENNARO

(v) With the same notation as in Proposition 3.4, one has: if n ≥ 3, δY −


d−(n−1)
2 ≤ −g and pg (X) > 0, then (3.1) holds with strict inequality. In fact
(3.9) yields e ≥ 2, thus g > d − n. Hence from Proposition 3.4, (iv), we get
   
δX − d−n2 ≤ δY − d−(n−1)
2 + d − n ≤ −g + d − n < 0.

4. A stronger inequality
In this section we improve Theorem 1.1 as follows:
Theorem 4.1. With the same notation as in Theorem 1.1, denote by g the
sectional genus of S. Then one has
 
d−2 g
(4.1) δS − ≤− ,
2 2
and equality holds if and only if S is a rational scroll in Pr .
Remark 4.2. (i) By Theorem 1.1, it suffices to prove the assertion when g > 0.
By (2.1) this is equivalent to prove that: when g > 0, then
9
(4.2) KS2 > 6pa (S) − g + 8.
2

(ii) By the proof of Proposition 2.2 and Remark 3.5, (ii), we may assume that
OS (KS + HS ) is spanned and pa (S) > 0. In particular g > d+1 2 ≥ 3, otherwise
e ≤ 0, hence pa (S) ≤ 0 by (2.5). Moreover by (2.3) we see that if g > 2(d − 2) then
(4.1) holds with strict inequality. So we may also assume g ≤ 2(d − 2).
Lemma 4.3. If g > 0 and 3pa (S) < 5
2g + d − 9 then (4.1) holds with strict
inequality.
Proof. Consider the adjunction map φ : S → PR . By Riemann–Roch one has
(4.3) R = h0 (S, OS (KS + HS )) − 1 = pa (S) + g − 1
since hi (S, OS (KS + HS )) = 0, 1 ≤ i ≤ 2, by Kodaira vanishing theorem. Let Σ
be the image of S via φ and let σ be its degree. Except for a few cases in which
pa (S) ≤ 0 ([10], pg. 593-594), one knows that Σ is a smooth surface, birational to
S via φ. In particular we have σ ≥ R − 1, i.e.
(4.4) σ = (KS + HS )2 = KS2 + 4g − 4 − d ≥ R − 1.
Set
σ − 1 = m(R − 2) + , with 0 ≤ ≤ R − 3.
By (4.4) we see that m ≥ 1. The case m = 1 is not possible. In fact, by
Castelnuovo Theory [5, 9] (see Remark 3.5, (iii) and (iv) above), we know that

+∞
(4.5) pg (Σ) ≤ (i − 1)(σ − hΓ (i)),
i=1
where Γ is the general 0-dimensional linear section of Σ. Moreover
(4.6) σ ≥ hΓ (i) ≥ min {σ, i(R − 2) + 1} for any i ≥ 1.
If m = 1 then σ ≤ 2(R − 2), and from (4.5) and (4.6) we get pg (Σ) = 0, against our
assumption pg (Σ) = pg (S) ≥ pa (S) > 0 (compare also with ([1], Lemma 10.1.2)).
When m = 2, by (4.5) and (4.6), we get
pa (S) ≤ pg (S) = pg (Σ) ≤ σ − hΓ (2) ≤ σ − (2(R − 2) + 1)
PLÜCKER–CLEBSCH FORMULA IN HIGHER DIMENSION 43
9

which implies KS2 ≥ 3pa (S) − 2g + d − 1 (use (4.3) and (4.4)). Since 3pa (S) <
2 g + d − 9, we deduce (4.2), hence the assertion holds.
5

Finally assume m ≥ 3. Since σ ≥ m(R − 2) + 1, by (4.3) and (4.4) we have


KS2 ≥ mpa (S) + (m − 4)g + d + 5 − 3m ≥ 3pa (S) − 2g + d − 1 because pa (S) > 0
and g ≥ 3 (see Remark 4.2, (ii)). We conclude as in the case m = 2. 

Lemma 4.4. Let X ⊂ Pr (r ≥ 2n + 1) be a nondegenerate, irreducible projective


variety of dimension n which is neither tangentially defective, nor secant defective.
Then
(4.7) tX ≥ 2(r − 2n + 1).
Proof. The main remark is that Tan(X) is singular along X, as a local com-
putation shows.
First we examine the case r = 2n + 1. Let  be a general secant line of X. Then
 contains two distinct points of X, which are singular points of Tan(X). Moreover
 is not contained in Tan(X), otherwise X would be secant defective against the
assumption. It follows that tX ≥ 4.
Next assume r > 2n + 1 and argue by induction on r. Fix a general point
x ∈ X, and denote by X  ⊂ Pr−1 the image of X via the projection φ from x. By
the Trisecant Lemma ([2], Proposition 2.6, pg. 158), φ induces a birational map
of X to X  . It also induces a generically finite map, of a certain degree ν, from
Tan(X) to its image V . Otherwise Tan(X) would be a cone of vertex x for a general
x ∈ X, and this would imply that X is degenerate, for the set of vertices of a cone
is a linear space. In particular we have dim(V ) = dim(Tan(X)) = 2n. Since the
general tangent space to X projects to the general tangent space to X  , one has
V = Tan(X  ), thus X  is not tangentially defective. The same argument says that
φ induces a generically finite map of Sec(X) to Sec(X  ), thus X  , as well as X is
not secant defective. By induction, we have tX  ≥ 2(r − 2n). Since x is a point of
Tan(X) of multiplicity μ ≥ 2 we deduce tX = μ + νtX  ≥ 2 + tX  ≥ 2 + 2(r − 2n) =
2(r − 2n + 1). 

It is a nice problem, on which we do not dwell here, to determine all varieties


X, which are neither tangentially defective, nor secant defective, for which equality
holds in (4.7). If X is smooth and one believes that (3.8), or rather (3.7), holds
(see the discussion in Remark 3.5, (i)), then X should conjecturally be a rational
normal scroll. There are however singular varieties X reaching the bound (4.7),
which are not rational normal scrolls.
Corollary 4.5. Let S ⊂ Pr (r ≥ 5) be a nondegenerate, irreducible, smooth,
projective surface of degree d and sectional genus g. Let C be a general hyperplane
section of S. Then
  
+∞
d−2
δS − ≤ −pa (S) + h1 (C, OC (i)).
2 i=2

Proof. We may assume S is linearly normal, i.e. r = h0 (S, OS (1)) − 1. On


the other hand, from Proposition 3.4, (ii), we see that
 
d−2 1
δS − = −g + d − 2 − vS .
2 2
44
10 CIRO CILIBERTO AND VINCENZO DI GENNARO

Moreover by Lemma 4.4 we have


vS ≥ tS ≥ 2(r − 3) = 2(h0 (S, OS (1)) − 4).
Hence
 
d−2
(4.8) δS − ≤ −g + d − 2 − (h0 (S, OS (1)) − 4).
2
Since χ(OS (1)) = χ(OS ) + χ(OC (1)), we also have
h0 (S, OS (1)) − 4 = h1 (S, OS (1)) − h2 (S, OS (1)) + (pa (S) + d − 2 − g).
From (4.8) we deduce
 
d−2
δS − ≤ −pa (S) + h2 (S, OS (1)) − h1 (S, OS (1)).
2
With an argument similar to the one used to prove (2.5), one sees that h2 (S, OS (1))
+∞
= h0 (S, ωS (−1)) ≤ i=2 h1 (C, OC (i)). The assertion follows. 
We are now in position to prove Theorem 4.1.

Proof of Theorem 4.1. By Remark 4.2, (ii), we may assume d+1 2 < g ≤
2(d − 2).
2 < g ≤ d. In this case e ≤ 1. Put gm := |OC (1)|,
First we examine the range d+1 s

and set
2g − 2 = (k − 1)m + h, with 0 ≤ h ≤ m − 1.
By Comessatti’s bound [4] one knows that
2k(m − 1) − 2g
(4.9) s≤ + 1.
k(k + 1)
In our case s = h0 (C, OC (1)) − 1, m = d and k = 2. So h0 (C, OC (1)) − 1 ≤
3 d − 3 g + 3 . By Riemann-Roch Theorem we deduce h (C, OC (1)) ≤ 3 g − 3 d + 3 .
2 1 1 1 2 1 1

Hence from (2.5) we get


5
3pa (S) ≤ 2g − d + 1 < g+d−9
2
(the second inequality holds since we may assume d ≥ 5). This proves Theorem 4.1
2 < g ≤ d by Lemma 4.3.
in the range d+1
Next, assume d < g ≤ 3d+1 2 . Then e ≤ 2. By applying (4.9) to |OC (i)|, with
1 ≤ i ≤ 2, we prove that h1 (C, OC (1)) ≤ 56 g− 12 d+ 12 and h1 (C, OC (2)) ≤ 23 g− 23 d+ 13 .
From (2.5) we get
9 7 5 5
3pa (S) ≤ g − d + < g + d − 9.
2 2 2 2
The second inequality holds when d > 8 because g ≤ 3d+1 2 and when d ≤ 8 by
Castelnuovo’s bound. This proves Theorem 4.1 in the range d < g ≤ 3d+1 2 by
Lemma 4.3.
Finally assume 3d+12 < g ≤ 2(d − 2). Then e ≤ 3. Using again (4.9) one sees
that h1 (C, OC (2)) ≤ 23 g − 23 d + 13 and h1 (C, OC (3)) ≤ 23 g − d + 13 . By Remark 4.2,
(i), and Corollary 4.5, we may assume
1
pa (S) ≤ g + h1 (C, OC (2)) + h1 (C, OC (3)).
2
PLÜCKER–CLEBSCH FORMULA IN HIGHER DIMENSION 45
11

So
11 5
g − 5d + 2 < g + d − 9
3pa (S) ≤
2 2
(the second inequality holds because g ≤ 2(d − 2)). This concludes the proof of
Theorem 4.1. 

References
[1] Beltrametti, M.C., Sommese, A.J.: The adjunction theory of complex projective varieties, de
Gruyter Expositions in Mathematics, 1995.
[2] Chiantini, L. Ciliberto, C.: Weakly defective varieties, Trans. Amer. Math. Soc. 354, No. 1,
151-178 (2002).
[3] Ciliberto, C.: On the Hilbert scheme of curves of maximal genus in a projective space, Math.
Z. 194, 351–363 (1987).
[4] Comessatti, A.: Limiti di variabilità della dimensione e dell’ordine d’una gnr sopra una curva

di dato genere, Atti R. Ist. Veneto Sci. Lett. Arti 74 (1914/1915), 1685-1709.
[5] Eisenbud, D., Harris, J.: Curves in projective space, Presse de l’Université de Montreal, 1982.
[6] Eisenbud, D., Harris, J.: On varieties of minimal degree (a centennial account), in “Algebraic
geometry”, Bowdoin, 1985 (Brunswick, Maine, 1985), 3–13, Proc. Sympos. Pure Math., 46,
Part 1, Amer. Math. Soc., Providence, RI, 1987.
[7] Fulton, W.: Intersection theory, Ergebnisse der Mathematik und ihrer Grenzgebiete; 3.Folge,
Bd. 2, Springer-Verlag, 1984.
[8] Griffiths, P., Harris, J.: Algebraic geometry and local differential geometry, Ann. Scient. Éc.
Norm. Sup. 4e série, 12, no 3 (1979), 355-452.
[9] Harris, J.: A bound on the geometric genus of projective varities, Ann. Scuola Norm. Sup.
Pisa Cl. Sci. (4), 8, 35-68 (1981).
[10] Sommese, A.J. - Van de Ven, A.: On the Adjunction Mapping, Math. Ann., 278, 593-603
(1987).
[11] Zak, F. L.: Tangents and secants of algebraic varieties, Translated from the Russian manu-
script by the author, Translations of Mathematical Monographs, 127, American Mathematical
Society, Providence, RI, 1993, viii+164.

Università di Roma “Tor Vergata”, Dipartimento di Matematica, Via della Ricerca


Scientifica, 00133 Roma, Italy.
E-mail address: cilibert@axp.mat.uniroma2.it

Università di Roma “Tor Vergata”, Dipartimento di Matematica, Via della Ricerca


Scientifica, 00133 Roma, Italy.
E-mail address: digennar@axp.mat.uniroma2.it
This page intentionally left blank
Contemporary Mathematics
Volume 555, 2011

Invariants of ideals generated by pfaffians

Emanuela De Negri and Elisa Gorla

Abstract. Ideals generated by pfaffians are of interest in commutative alge-


bra and algebraic geometry, as well as in combinatorics. In this article we
compute multiplicity and Castelnuovo-Mumford regularity of pfaffian ideals of
ladders. We give explicit formulas for some families of ideals, and indicate
a procedure that allows to recursively compute the invariants of any pfaffian
ideal of ladder. Our approach makes an essential use of liaison theory.

Introduction
Pfaffians are the natural analogue of minors when working with skew-symmetric
matrices. Ideals generated by pfaffians are studied in the context of commutative
algebra and algebraic geometry, as well as in combinatorics. Many are the reasons
for such an interest, e.g., many ideals generated by pfaffians are Gorenstein (see,
e.g., [KL] and [D]). Conversely, due to a famous result ([BE]) of Buchsbaum and
Eisenbud, any Gorenstein ideal of height 3 of a polynomial ring over a field is gener-
ated by the maximal pfaffians of a suitable skew-symmetric matrix of homogeneous
forms. Ideals generated by pfaffians arise naturally in algebraic geometry as, e.g.,
ideals of pfaffians in a generic skew-symmetric matrix define Schubert cells in or-
thogonal Grassmannians. Moreover, some Grassmannians are defined by pfaffians,
as well as some of their secant varieties.
In this article, we compute numerical invariants of pfaffian ideals of ladders.
Pfaffian ideals of ladders are, informally speaking, ideals generated by pfaffians
which only involve indeterminates in a ladder of a skew-symmetric matrix of in-
determinates. The size of the pfaffians is allowed to vary in different regions of
the ladder. This family was introduced by the authors in [DGo], and contains the
classically studied ideals of 2t-pfaffians of a matrix or of a ladder. It is a very large
family, and a natural one to study from the point of view of liaison theory, since all
the ideals in this family arise from ideals of 2t-pfaffians in a ladder while performing

1991 Mathematics Subject Classification. Primary 13C40, 13H10.


The second author was supported by the Swiss National Science Foundation under grant
no. 123393. Part of this work was done while the authors were attending the conference “PASI
2009 in Commutative Algebra and its Connections to Geometry, honoring Wolmer Vasconcelos”,
which took place in Olinda (Brazil) in August 2009. The authors wish to thank the organizers,
the speakers and the participants to the conference for the stimulating working environment that
they created.

c 0000
2011
c De(copyright
Negri andholder)
Gorla

1
47
48
2 EMANUELA DE NEGRI AND ELISA GORLA

elementary G-biliaisons. In [DGo] we proved that these ideals are prime, normal
and Cohen-Macaulay. The main result of the paper was a proof that any pfaffian
ideal of ladder can be obtained from an ideal generated by indeterminates via a
finite sequence of ascending G-biliaisons. In particular they are glicci, i.e., they
belong to the G-liaison class of a complete intersection. The G-biliaison steps were
described very explicitly. Therefore, as a biproduct, it is possible to recursively
compute numerical invariants of pfaffian ideals of ladders such as the multiplicity,
the Hilbert function, the h-vector, as well as a graded free resolution. In some cases
it is also possible to compute the graded Betti numbers and in particular the Castel-
nuovo Mumford regularity. Although it is possible to perform these computations
in any specific example, it is in general hard to produce explicit formulas. In this
paper, we derive explicit formulas for some classes of pfaffian ideals of ladders.
The paper is organized as follows. In Section 1 we fix the notation and define
the classes that we study. We also recall the main result of [DGo] on which our
approach is based. In Section 2 we give explicit or recursive formulas for the
multiplicity of the ideals that we study. In Theorem 2.5 we give a simple numerical
condition which forces the multiplicity of a pfaffian ideal of ladder to decompose
as the product of the multiplicities of two pfaffian ideals relative to subladders. In
Section 3 we compute Castelnuovo-Mumford regularities. In Section 4 we show
how to use our approach to compute the graded Betti numbers of ideals of pfaffians
of maximal size of a generic skew-symmetric matrix. We also give a simple proof
that the h-vectors of these ideals are of decreasing type. The ideals generated by
pfaffians of maximal size of a generic skew-symmetric matrix are Gorenstein ideals
of height 3, so the results are well-known. However, we are able to give a very
simple, self-contained proof.

1. Some classes of pfaffian ladder ideals


Let X = (xij ) be an n × n skew-symmetric matrix of indeterminates. In other
words, the entries xij with i < j are indeterminates, xij = −xji for i > j, and
xii = 0 for all i = 1, ..., n. Let R = K[X] = K[xij | 1 ≤ i < j ≤ n] be the
polynomial ring associated to X.
Definition 1.1. A ladder Y of X is a subset of the set {(i, j) ∈ N2 | 1 ≤ i, j ≤
n} with the following properties :
(1) if (i, j) ∈ Y then (j, i) ∈ Y,
(2) if i < h, j > k and (i, j), (h, k) belong to Y, then (i, k), (i, h), (h, j), (j, k)
belong to Y.
We do not assume that a ladder Y is connected, nor that X is the smallest skew-
symmetric matrix having Y as ladder. We can assume without loss of generality
that the ladder Y is symmetric.
It is easy to see that any ladder can be decomposed as a union of square
subladders
(1.1) Y = X1 ∪ . . . ∪ Xs
where
Xk = {(i, j) | ak ≤ i, j ≤ bk },
for some integers 1 ≤ a1 ≤ . . . ≤ as ≤ n and 1 ≤ b1 ≤ . . . ≤ bs ≤ n such that
ak < bk for all k. We say that Y is the ladder with upper corners (a1 , b1 ), . . . , (as , bs ),
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 49
3

and that Xk is the square subladder of Y with upper outside corner (ak , bk ). We
allow two upper corners to have the same first or second coordinate, but we assume
that no two upper corners coincide. Notice that with this convention a ladder does
not have a unique decomposition of the form (1.1). In other words, a ladder does
not correspond uniquely to a set of upper corners (a1 , b1 ), . . . , (as , bs ). However,
the upper corners determine the subladders Xk , hence the ladder Y according to
(1.1).
Let t be a positive integer. A 2t-pfaffian is the pfaffian of a 2t × 2t submatrix
of X. Given a ladder Y we set Y = {xij ∈ X | (i, j) ∈ Y, i < j}. We let I2t (Y )
denote the ideal generated by the set of the 2t-pfaffians of X which involve only
indeterminates of Y . In particular I2t (X) is the ideal generated by the 2t-pfaffians
of X. We regard all the ideals as ideals in K[X].
Whenever we consider a ladder Y, we assume that it comes with its set of upper
corners and the corresponding decomposition as a union of square subladders as in
(1.1).
The following family of ideals has been introduced and studied in [DGo]:
Definition 1.2. Let Y = X1 ∪ . . . ∪ Xs be a ladder as in Definition 1.1.
Let Xk = {xij | (i, j) ∈ Xk , i < j} for k = 1, . . . , s. Fix a vector t = (t1 , . . . , ts ),
t ∈ {1, . . . ,  n2 }s . The pfaffian ideal I2t (Y ) is by definition the sum of pfaffian
ideals I2t1 (X1 ) + . . . + I2ts (Xs ) ⊆ K[X]. We refer to these ideals as pfaffian ideals
of ladders.
Remarks 1.3 (Remarks 1.5, [DGo]). We can assume without loss of generality
that
2tk ≤ bk − ak + 1, for 1 ≤ k ≤ s.
Moreover, we can assume that
ak − ak−1 > tk−1 − tk and bk − bk−1 > tk − tk−1
for 2 ≤ k ≤ s.
In [DGo], pfaffian ideals of ladders are proved to be prime, normal, and Cohen-
Macaulay. A formula for their height is given.
Notation 1.4. For a ladder Y with upper corners (a1 , b1 ), . . . , (as , bs ) and
t = (t1 , . . . , ts ), we denote by Ỹ the ladder with upper corners (a1 + t1 − 1, b1 − t1 +
1), . . . , (as + ts − 1, bs − ts + 1).

The ladder Ỹ computes the height of the ideal I2t (Y ) as follows:


Proposition 1.5 (Proposition 1.10, [DGo]). Let Y be the ladder with upper
corners (a1 , b1 ), . . . , (as , bs ) and t = (t1 , . . . , ts ). Let Ỹ be as in Notation 1.4. Then
the height of I2t (Y ) equals the cardinality of {(i, j) ∈ Ỹ | i < j}.
We now recall the definition of biliaison.
Definition 1.6. Let I, I  , J be homogeneous, unmixed ideals in K[X], with
ht(I) = ht(I  ) = ht(J) + 1. Assume that R/J is Cohen-Macaulay and generically
Gorenstein, i.e., (R/J)P is Gorenstein for any minimal associated prime P of J.
We say that I is obtained from I  by a G-biliaison of height  on J if I/J and
I  /J() represent the same element in the ideal class group of K[X]/J.
50
4 EMANUELA DE NEGRI AND ELISA GORLA

In other words, I is obtained from I  by a G-biliaison of height  on J if


there exist homogeneous polynomials f, g ∈ R with deg(g) = deg(f ) + , such that
f I + J = gI  + J as ideals of R.
The main result of [DGo] is that ladder pfaffian ideals belong to the G-biliaison
class of a complete intersection. In particular, they are glicci. We briefly recall the
single G-biliaison step which is described in the proof of [DGo, Theorem 2.3]. With
the notation of Definition 1.2, let Y  be the subladder of Y with upper corners
(a1 , b1 ), . . . , (ak−1 , bk−1 ), (ak + 1, bk − 1), (ak+1 , bk+1 ), . . . , (as , bs ),
and let t = (t1 , . . . , tk−1 , tk − 1, tk+1 , . . . , ts ). Let Z be the subladder of Y obtained
by removing the entry (ak , bk ) and its symmetric. Equivalently, Z is the ladder with
upper corners
(a1 , b1 ), . . . , (ak−1 , bk−1 ), (ak , bk − 1), (ak + 1, bk ), (ak+1 , bk+1 ), . . . , (as , bs ).
Let u = (t1 , . . . , tk−1 , tk , tk , tk+1 , . . . , ts ). One has:
Theorem 1.7 (Theorem 2.3, [DGo]). Let I = I2t (Y ), I  = I2t (Y  ) and J =
I2u (Z) be ideals of K[X]. Then I is obtained from I  via an elementary G-biliaison
of height 1 on J.
More precisely, with the above notation we have
f I + J = gI  + J
where f ∈ I  is a 2(tk − 1)-pfaffian, g ∈ I is a 2tk -pfaffian, and f, g ∈ J.
When discussing biliaison, we will refer without distinction to the ideals and
to the varieties associated to them.
In this paper we deal with special classes of pfaffian ideals of ladders, and we
compute some of their numerical invariants using the biliaison step described in
Theorem 1.7. The same technique gives a recursive procedure to determine such
invariants for any pfaffian ideal of ladder. However, it is in general hard to deduce
explicit formulas.
We now introduce the classes we are going to study. First we consider the ideal
Lnt = I2t (Y ) where Y is the ladder with upper corners (1, n − 1) an (2, n) and
t = (t, t). Clearly Lnt is generated by the 2t-pfaffians of the ladder obtained from
X by deleting the entries (1, n) and (n, 1).

(1, n − 1)
• (2, n)

2t-pfaffians
Lnt :

Then we restrict our attention to some ideals generated by pfaffians whose size
is maximal or submaximal, in a sense that we are going to specify. In particular, we
consider the ideals generated by maximal and by submaximal pfaffians of a skew-
symmetric matrix of indeterminates. More precisely, we denote by Mt the ideal
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 51
5

generated by the 2t-Pfaffians of a (2t + 1) × (2t + 1) matrix and by SMt the ideal
generated by the 2t-pfaffians of a (2t + 2) × (2t + 2) matrix.
Moreover we consider ideals generated by pfaffians of two different sizes in
different regions of a matrix. Here we regard nested matrices as a ladder. In
particular, we consider Nt = I2t (Y ) where Y is the ladder with upper corners
(1, 2t − 1) and (1, 2t + 1), and t = (t − 1, t). So Nt is the ideal generated by the
2t-pfaffians of a skew-symmetric matrix of size 2t + 1 and the (2t − 2)-pfaffians of
its first 2t − 1 rows and columns. We denote by SNt the ideal I2t (Y ) where Y is
the ladder with upper corners (1, 2t − 1) and (1, 2t + 2), and t = (t − 1, t). This
is the ideal generated by the 2t-pfaffians of a skew-symmetric matrix of size 2t + 2
and the (2t − 2)-pfaffians of its first 2t − 1 rows and columns.

(1, 2t − 1) (1, 2t + 1) (1, 2t − 1)


• • • • (1, 2t + 2)

(2t − 2)-pfaff. (2t − 2)-pfaff.


Nt :
SNt :

2t-pfaff.
2t-pfaff.

We let Lt (k) = I2t (Y ), where Y is the ladder with upper corners (1, 2t +
1), (2, 2t + 2), (3, 2t + 3), . . . , (k, 2t + k), and t = (t, . . . , t). Notice that Lt (1) = Mt ,
and Lt (2) = L2t+2t .

(1, 2t + 1) (1, 2t + 1)
• (2, 2t + 2) • (2, 2t + 2)
• • ..
• .
• (k, 2t + k)

2t-pfaffians 2t-pfaffians
Lt (2) : Lt (k):

Moreover, given two integers j and k we let Yjk be the ladder with the j + k
upper outside corners (1, 2t − 1), (2, 2t), (3, 2t + 1), . . . , (j, 2t + j − 2), (j, 2t + j), (j +
1, 2t + j + 1), . . . , (j + k − 1, 2t + j + k − 1). We consider the ideal
Lt (j, k) := I2t (Yjk ), where t = (t − 1, . . . , t − 1, t, . . . , t).
     
j k

Notice that Lt (0, k) = Lt+1 (k, 0). Moreover, this class contains most of the classes
that we have already introduced. More precisely: Lt (k) = Lt (0, k), Mt = Lt (0, 1),
SMt = Lt (1, 0), and Nt = Lt (1, 1).
52
6 EMANUELA DE NEGRI AND ELISA GORLA

(1, 2t − 1)

(2, 2t)

(3, 2t + 1) (3, 2t + 3)
• •
(4, 2t + 4)
(2t − 2)-pfaff. •
• (5, 2t + 5)
(6, 2t + 6)

Lt (j, k) :

j = 3, k = 4 2t-pfaffians

Given two integers j and k, we let Zjk be the ladder with the j +k upper outside
corners (1, 2t − 1), (2, 2t), (3, 2t + 1), . . . , (j, 2t + j − 2), (j + 1, 2t + j + 1), . . . , (j +
k, 2t + j + k). We consider the ideal

Ht (j, k) := I2t (Zjk ), where t = (t − 1, . . . , t − 1, t, . . . , t).


     
j k

It is Lt (k) = Ht (0, k) = Ht+1 (k, 0).

(1, 2t − 1)

(2, 2t)

(3, 2t + 1)

(4, 2t + 4)
(2t − 2)-pfaff. •
• (5, 2t + 5)
(6, 2t + 6)

Ht (j, k) :

j = 3, k = 3 2t-pfaffians
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 53
7

2. Multiplicity of pfaffian ladder ideals


In this section we give some formulas for the multiplicity of the ideals introduced
in the previous section. Throughout the section, we denote by e(I) the multiplicity
of R/I for any ideal I ⊂ R = K[X]. All the formulas that we produce are obtained
as a finite sum of positive contributions. Therefore they are well suited to give lower
bounds for the multiplicity. In the sequel we will need the following well know fact,
which we prove for completeness.
Proposition 2.1. Let H, I, J ⊂ K[X] be homogeneous, unmixed ideals. As-
sume that H is Cohen-Macaulay and that I is obtained from J via an elementary
G-biliaison of height  ∈ Z on H. Then
e(I) = e(J) + e(H).
Proof. Let U, S, T be the schemes associated to H, I, J, respectively. Under
our assumptions, U is arithmetically Cohen-Macaulay and S, T are generalized
divisors on U . Moreover, S is linearly equivalent to T + h as generalized divisors
on U , where h denotes the hyperplane section class on U . In particular
e(I) = deg(S) = deg(T ) +  deg(U ) = e(J) + e(H).


We denote by Itn the ideal generated by the 2t-pfaffians of an n × n skew-


symmetric matrix of indeterminates. In [K, Theorem 7] Krattenthaler proved that
 2(t − 1) + i + j
(2.1) e(Itn ) = .
i+j
1≤i≤j≤n−2t+1

In particular for the ideals Mt and SMt one has:


 2(t − 1) + i + j  2(t − 1) + i + j
e(Mt ) = , e(SMt ) = .
i+j i+j
1≤i≤j≤2 1≤i≤j≤3

From the results in [DGo] one can easily deduce a formula for the multiplicity
of the ideal Lnt .
Proposition 2.2.
   2(t − 1) + i + j
(n−2t+2)!
e(Lnt ) = (2n−4t+4)!
(2n−2t+2)!
n!
− (n−1)!
(2t−3)! i+j
1≤i≤j≤n−2t+2

Proof. By Theorem 1.7 the ideal Itn+1 is obtained from It−1


n−1
via an elemen-
n
tary G-biliaison of height 1 on Lt . Hence by Proposition 2.1
e(Lnt ) = e(Itn+1 ) − e(It−1
n−1
).
Substituting (2.1) we obtain e(Lnt ) =
 1    
(2t − 2 + i + j) − (2t − 4 + i + j) .
i+j
1≤i≤j≤n−2t+2 1≤i≤j≤n−2t+2 1≤i≤j≤n−2t+2

Since  
(2t − 4 + i + j) = (2t − 2 + i + j)
1≤i≤j≤n−2t+2 0≤i≤j≤n−2t+1
54
8 EMANUELA DE NEGRI AND ELISA GORLA

by means of direct computation one gets


 
(2t − 2 + i + j) − (2t − 4 + i + j) =

1≤i≤j≤n−2t+2  
1≤i≤j≤n−2t+2
  .
(2(t − 1) + i + j) (n + i) − (2t − 2 + j)
1≤i≤j≤n−2t+1 1≤i≤n−2t+2 0≤j≤n−2t+1

The result now follows from the equality


 1 
(2(t − 1) + i + j) =
i+j
1≤i≤j≤n−2t+2 1≤i≤j≤n−2t+1
 (2(t − 1) + i + j)  1
.
i+j n − 2t + 2 + i
1≤i≤j≤n−2t+1 1≤i≤n−2t+2


The case of ideals generated by maximal pfaffians of a matrix has been exten-
sively studied. In particular it is well known that
(2.2) e(Lt (1)) = e(Mt ) = 1 + 22 + 32 + · · · + t2
(see [HTV, Section 6], and [HT, Theorem 5.6 and the following example]).
We deduce the following formulas from Theorem 1.7.
Proposition 2.3.

t
e(Lt (2)) = 1 + [2s(1 + 22 + · · · + s2 ) − s3 ]
s=2
and

t−1
e(Nt ) = 1 + [2s(1 + 22 + · · · + s2 ) − s3 ] + t(1 + 22 + · · · + (t − 1)2 ).
s=2

Proof. By Theorem 1.7 the ideal Lt (2) is obtained from Nt via an elementary
G-biliaison of height 1 on Mt + (f ), where f is a 2t-pfaffian which is regular modulo
Mt . Thus by Proposition 2.1 one has
(2.3) e(Lt (2)) = e(Nt ) + e(Mt + (f )) = e(Nt ) + te(Mt ).
Moreover the ideal Nt is obtained from Lt−1 (2) via an elementary G-biliaison of
height 1 on Mt−1 + (g), where g is a 2t-pfaffian which is regular modulo Mt−1 .
Therefore
(2.4) e(Nt ) = e(Lt−1 (2)) + te(Mt−1 )
and combining (2.3) and (2.4) one gets
(2.5) e(Lt (2)) = e(Lt−1 (2)) + te(Mt−1 ) + te(Mt ).
Finally by (2.5) and (2.2), after solving the recursion one obtains

t
t
e(Lt (2)) = 1 + [s(e(Ms−1 ) + e(Ms ))] = 1 + [2s(1 + 22 + · · · + s2 ) − s3 ].
s=2 s=2
The formula for e(Nt ) follows from substituting the formula for e(Lt (2)) and (2.2)
in (2.4). 
We now deduce a formula for the multiplicity of ideals generated by submaximal
pfaffians.
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 55
9

Corollary 2.4.

t
r
e(SMt ) = t + [2s(1 + 22 + · · · + s2 ) − s3 ].
r=2 s=2

Proof. Since SMt is obtained from SMt−1 via an elementary G-biliaison of


height 1 on Lt (2), one has e(SMt ) = e(SMt−1 ) + e(Lt (2)). By solving the recursion
and using Proposition 2.3, one obtains the result. 

Let Y = Y1 ∪ Y2 be a ladder which is union of two smaller ladders. Let


I1 = I2t1 (Y1 ) and I2 = I2t2 (Y2 ) be pfaffian ideals associated to the ladders Y1 and
Y2 , and let the upper corners of Y be the union of the upper corners of Y1 and Y2 .
Let t = t1 ⊕ t2 be the vector obtained by appending the vector t2 to the vector
t1 and let I = I2t (Y ) = I1 + I2 be the pfaffian ideal associated to the ladder Y. If
Y1 ∩ Y2 = ∅, one can easily show that
(2.6) e(I) = e(I1 )e(I2 ).
The following theorem gives a sufficient condition on the ladder so that (2.6) holds.
Theorem 2.5. Let Y, Y1 , Y2 be ladders, Y = Y1 ∪ Y2 . Let I1 = I2t1 (Y1 ) and
I2 = I2t2 (Y2 ) be pfaffian ideals of ladders associated to Y1 and Y2 . Let t = t1 ⊕ t2
and let I = I2t (Y ) = I1 + I2 be the corresponding pfaffian ideal of ladder. Let
Ỹ, Y˜1 , Y˜2 be defined as in Notation 1.4, and let Ỹ , Y˜1 , Y˜2 be the corresponding sets
of indeterminates. If Y˜1 ∩ Y˜2 = ∅, then
e(I) = e(I1 )e(I2 ).
Proof. Let Z = Y1 ∩ Y2 , R1 = K[Y1 ]/I1 , and R2 = K[Y2 ]/I2 . We have
K[Y ]/I ∼
= R1 ⊗K R2 /J
where J is generated by |Z| linear forms (which identify the corresponding indeter-
minates in Y1 and Y2 ). If Y˜1 ∩ Y˜2 = ∅, then
ht I = ht I1 + ht I2
hence
ht J = dim R1 ⊗ R2 − dim K[Y ]/I = |Y1 | − ht I1 + |Y2 | − ht I2 − |Y | + ht I = |Z|.
Since R1 ⊗ R2 is a Cohen-Macaulay ring, J is generated by a regular sequence and
e(I) = e(R1 ⊗K R2 /J) = e(I1 )e(I2 ).


We now give an example of a family of pfaffian ideals of ladders whose multi-


plicity can be computed directly from Theorem 2.5.
Proposition 2.6.
e(Ht (j, k)) = e(Lt−1 (j))e(Lt (k)).
Proof. Let Y = Zjk be the ladder with the j + k upper corners (1, 2t −
1), (2, 2t), . . . , (j, 2t + j − 2), (j + 1, 2t + j + 1), . . . , (j + k, 2t + j + k). Let Y1 be
the ladder with the j upper corners (1, 2t − 1), . . . , (j, 2t + j − 2) and let Y2 be the
56
10 EMANUELA DE NEGRI AND ELISA GORLA

ladder with the k upper corners (j + 1, 2t + j + 1), . . . , (j + k, 2t + j + k). Clearly


Y = Y1 ∪ Y2 . Let
t1 = (t − 1, . . . , t − 1), t2 = (t, . . . , t), t = t1 ⊕ t2 = (t − 1, . . . , t − 1, t, . . . , t).
           
j k j k

Then Y˜1 is the ladder with upper outside corners (t − 1, t + 1), (t, t + 2), . . . , (t + j −
2, t + j) and Y˜2 is the ladder with upper outside corners (t + j, t + j + 2), . . . , (t +
j + k − 1, t + j + k + 1). Hence Y˜1 ∩ Y˜2 = {(t + j, t + j)} and Y1 ∩ Y2 = ∅. By
Theorem 2.5 it follows that
e(Ht (j, k)) = e(I1 )e(I2 )
where I1 = I2t1 (Y1 ) and I2 = I2t2 (Y2 ). The thesis follows from the observation
that I1 = Lt−1 (j) and I2 = Lt (k). 
Combining Proposition 2.6 and Proposition 2.8, we obtain a formula for the
multiplicity of the ideals Lt (j, k).
Proposition 2.7. For j, k ≥ 1 we have

k−1
e(Lt (j, k)) = e(Lt−1 (j + k)) + te(Lt−1 (j + k − 1)) + e(Lt−1 (j + k − 1 − l))e(Lt (l)).
l=1

Proof. We proceed by induction on k ≥ 1. By Theorem 1.7, Lt (j, 1) is ob-


tained from Lt−1 (j + 1) via an elementary G-biliaison on Lt−1 (j) + (f ), where f is
a 2t-pfaffian which does not belong to Lt−1 (j). Hence by Proposition 2.1
e(Lt (j, 1)) = e(Lt−1 (j + 1)) + te(Lt−1 (j)).
This proves the thesis for k = 1. To establish the formula for k ≥ 2, observe that
Lt (j, k) is obtained from Lt (j + 1, k − 1) via an elementary G-biliaison of height 1
on Ht (j, k − 1). Hence by Proposition 2.1 and Proposition 2.6
(2.7) e(Lt (j, k)) = e(Lt (j + 1, k − 1)) + e(Lt−1 (j))e(Lt (k − 1)).
By induction hypothesis e(Lt (j + 1, k − 1)) =

k−2
e(Lt−1 (j + k)) + te(Lt−1 (j + k − 1)) + e(Lt−1 (j + k − 1 − l))e(Lt (l))
l=1
and the thesis follows. 
Explicit formulas for e(Lt (1)) and e(Lt (2)) were given in (2.2) and in Proposi-
tion 2.3. Since L1 (k) is generated by indeterminates, e(L1 (k)) = 1 for any k. The
following formula allows us to calculate e(Lt (k)) recursively, for t ≥ 2 and k ≥ 3.
Proposition 2.8. For t, k ≥ 2 we have e(Lt (k)) =

k−2
e(Lt−1 (k)) + t[e(Lt (k − 1)) + e(Lt−1 (k − 1))] + e(Lt−1 (k − 1 − l))e(Lt (l)).
l=1

Proof. By Theorem 1.7, Lt (k) is obtained from Lt (1, k − 1) via an elementary


G-biliaison of height 1 on Lt (k − 1) + (f ), where f is a 2t-pfaffian which does not
belong to Lt (k − 1). Hence by Proposition 2.1 and Proposition 2.7
e(Lt (k)) = Lt (1, k − 1) + te(Lt (k − 1)) =
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 57
11


k−2
e(Lt−1 (k)) + t[e(Lt (k − 1)) + e(Lt−1 (k − 1))] + e(Lt−1 (k − 1 − l))e(Lt (l)).
l=1

Remarks 2.9. (1) Proposition 2.8 allows us to compute the multiplicity
of the ideals Lt (k) for any values of t and k. This can in fact be done
recursively, using as a starting point that e(L1 (k)) = 1 for any k, and
the explicit formulas for the multiplicities of Lt (1) = Mt and Lt (2) which
appear in (2.2) and in Proposition 2.3, respectively.
(2) Proposition 2.7 allows us to compute the multiplicity of the ideals Lt (j, k)
for any values of t, j, k. One can in fact use Proposition 2.8 to compute
the multiplicities of Lt (1), . . . , Lt (k − 1) and Lt−1 (j), . . . , Lt−1 (j + k).
(3) Since Lt (k) = Lt (0, k), the multiplicity of Lt (j, k) for j = 0 is computed
in Proposition 2.8. In fact, the formula obtained in Proposition 2.8 corre-
sponds to the formula computed in Proposition 2.7 for j = 0, taken “cum
grano salis”.
(4) The formula given in Proposition 2.7 is false for k = 0.
Finally, we express the multiplicity of SNt in terms of the multiplicities of SMt
and Lt (1, 2). The latter two can be computed by Proposition 2.4 and Proposi-
tion 2.7.
Proposition 2.10. For t ≥ 1 we have

t
t−1
e(SNt ) = e(Ls (1, 2)) + s e(SMs ) + 1.
s=2 s=2

Proof. We proceed by induction on t. If t = 1, then SN1 is generated by


indeterminates and e(SN1 ) = 1.
Let Y denote the ladder with upper corners (1, 2t − 1) and (2, 2t + 1). Then
I2(t−1) (Y ) is the ideal generated by the 2(t−1)-pfaffians of Y. By Theorem 1.7, SNt
is obtained from I2(t−1) (Y ) via an elementary G-biliaison of height 1 on Lt (1, 2).
In turn, I2(t−1) (Y ) is obtained from SNt−1 via an elementary G-biliaison of height
1 on SMt−1 + (f ), where f is a 2(t − 1)-pfaffian which does not belong to SMt−1 .
Therefore, by Proposition 2.1
e(SNt ) = e(Lt (1, 2)) + (t − 1)e(SMt−1 ) + e(SNt−1 )
and the thesis follows by induction hypothesis. 
Remark 2.11. From the proof of Proposition 2.10 it also follows that

t−1
e(I2(t−1) (Y )) = e(SNt ) − e(Lt (1, 2)) = [e(Ls (1, 2) + se(Ms )] + 1.
s=2

3. Castelnuovo-Mumford regularity
In this section we use biliaison to compute the Castelnuovo-Mumford regularity
of some of the ideals considered in the previous section. For an ideal I of R = K[X],
we denote by βi,j (I) the (i, j)−th graded Betti number of I, regarded as an R-
module. The Castelnuovo-Mumford regularity of a Cohen-Macaulay ideal I of
height h = ht(I) is
reg(I) = max{j | βh−1,j (I) = 0} − h + 1.
58
12 EMANUELA DE NEGRI AND ELISA GORLA

It is well known that reg(Mt ) = 2t − 1.


The following result allows us to recursively compute the Castelnuovo Mumford
regularities of ideals obtained one from the other by biliaison.
Theorem 3.1. Let H, I, J ⊂ R be homogeneous, Cohen-Macaulay ideals. As-
sume that I is obtained from J via an elementary G-biliaison of height  ∈ Z on
H. If reg(J) < reg(H), then
reg(I) = reg(H) +  − 1.
Proof. Since I is obtained from J via an elementary G-biliaison of height
 ∈ Z on H, there are homogeneous polynomials f, g with deg(f ) +  = deg(g) =: t
such that
f I + H = gJ + H ⊂ R.
Let h = ht I = ht J = ht H + 1. Applying the Mapping Cone construction to the
short exact sequence
0 −→ H[−t] −→ H ⊕ J[−t] −→ gJ + H −→ 0
we have that
reg(gJ + H) = max{j | βh−1,j (gJ + H) = 0} − h + 1 =
max{reg(H) + h − 2, reg(J) + h − 1} + t − h + 1 = reg(H) + t − 1.
The last equality follows from the assumption that reg(J) < reg(H). The previous
equality follows from the observation that, since J and H are Cohen-Macaulay
ideals,
max{j | βh−2,j (H) = 0} = reg(H) + h − 2 ≥
reg(J) + h − 1 > max{j | βh−2,j (J) = 0}
therefore no cancellation involving a direct summand R[−reg(H) + h − 2] can take
place in the free resolution of gJ + H.
In an analogous fashion, we can produce a free resolution for gJ + H = f I + H
by applying the Mapping Cone construction to the short exact sequence
0 −→ H[−t + ] −→ H ⊕ I[−t + ] −→ f I + H −→ 0.
Since
max{j | βh−1,j (f I + H) = 0} = reg(H) + t + h − 2 >
reg(H) + h − 2 + t −  = max{j | βh−2,j (H[−t + ]) = 0},
it must be
reg(H) + t + h − 2 = max{j | βh−1,j (I[−t + ]) = 0} = reg(I) + h − 1 + t − ,
hence
reg(I) = reg(H) +  − 1.

We now derive formulas for the Castelnuovo-Mumford regularity of some pfaf-
fian ideals of ladders. They are all easy consequences of Theorem 3.1.
Proposition 3.2. For t ≥ 1 we have
reg(Lt (2)) = 3t − 2
and for t ≥ 2
reg(Nt ) = 3t − 4.
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 59
13

Proof. We compute the regularity of Lt−1 (2) and Nt for t ≥ 2. We proceed


by induction on t ≥ 2. If t = 2, L1 (2) is generated by indeterminates, hence
reg(L1 (2)) = 1. By Theorem 1.7, N2 is obtained from L1 (2) via an ascending G-
biliaison of height 1 on M1 + (p), where p is a 4-pfaffian which is regular modulo
M1 . Since reg(L1 (2)) = 1 < 2 = reg(H), by Theorem 3.1 we have
reg(N2 ) = 2.
We now assume by induction hypothesis that reg(Lt−2 (2)) = 3t−8 and reg(Nt−1 ) =
3t − 7, and compute the regularity of Lt−1 (2) and Nt . By Theorem 1.7, the
ideal Lt−1 (2) is obtained from Nt−1 via an elementary G-biliaison of height 1 on
Mt−1 + (f ), where f is a 2(t − 1)-pfaffian which is regular modulo Mt−1 . Since
reg(Nt−1 ) = 3t − 7 < 3t − 5 = reg(Mt−1 + (f )), by Theorem 3.1
reg(Lt−1 (2)) = 3t − 5.
By Theorem 1.7, the ideal Nt is obtained from Lt−1 (2) via an elementary G-biliaison
of height 1 on Mt−1 + (g), where g is a 2t-pfaffian which is regular modulo Mt−1 .
Since reg(Lt−1 (2)) = 3t − 5 < 3t − 4 = reg(Mt−1 + g), by Theorem 3.1 we have
reg(Nt ) = 3t − 4.

Proposition 3.3. For t ≥ 1 we have
reg(SMt ) = 3t − 2.
Proof. We proceed by induction on t ≥ 1. If t = 1, SM1 is generated by
indeterminates, hence reg(SM1 ) = 1. By Theorem 1.7 the ideal SMt is obtained
from SMt−1 via an elementary G-biliaison of height 1 on Lt (2). By induction
hypothesis and Proposition 3.2
reg(SMt−1 ) = 3t − 5 < 3t − 2 = reg Lt (2).
Therefore, by Theorem 3.1
reg(SMt ) = reg(Lt (2)) = 3t − 2.


4. The Gorenstein height 3 case


The ideal Mt generated by the 2t-pfaffians of a generic skew-symmetric matrix
of size 2t + 1 is a Gorenstein ideal of height 3. A classical result due to Buchsbaum
and Eisenbud [BE] states that any Gorenstein ideal of height 3 is obtained by spe-
cialization from Mt , for some t. An alternative proof for many classically known
results on Gorenstein ideals of height 3 can therefore be given by combining spe-
cialization with a liaison approach analogous to what we have done in the previous
sections.
In this section we wish to give a taste of what can be obtained following such an
approach. In particular, we use G-biliaison to compute the graded Betti numbers
of the ideal Mt and to prove that its h-vector is of decreasing type. We start by
recalling some definitions and fixing the notation.
Let I be a homogeneous ideal of R = K[X]. The Hilbert function of R/I is
defined as
HI (m) = dimK (R/I)m
60
14 EMANUELA DE NEGRI AND ELISA GORLA

for every integer m. Clearly HI (m) = 0 for m < 0. The formal power series

HSI (z) = HI (m)z m
m∈Z

is called the Hilbert series of R/I. It is well known that the Hilbert series of R/I
is of the form
hI (0) + hI (1)z + . . . + hI (s)z s
HSI (z) = ,
(1 − z)d
where d = dim(R/I) and hi ∈ Z for every i. The vector
hI = (hI (0), . . . , hI (s)) ∈ Zs
is called h-vector of I. Moreover we denote by ΔHI the first difference of HI , that
is
ΔHI (m) = HI (m) − HI (m − 1).
Definition 4.1. Let h = (h0 , h1 , . . . , hs ) ∈ Zs
a) h is unimodal if there exists t ∈ {1, . . . s} such that h1 ≤ h2 ≤ · · · ≤ ht ≥ ht+1 ≥
· · · ≥ hs .
b) h is of decreasing type if whenever ht > ht+1 , then hj > hj+1 for every j > t.
Notice that every h-vector of decreasing type is unimodal.
Proposition 4.2. The h-vector of Mt is of decreasing type.
Proof. Let X be a (2t+1)×(2t+1) skew-symmetric matrix of indeterminates
and let R = K[X] be the corresponding polynomial ring. Denote by h(t,t) (m) the
m-th entry of the h-vector of a complete intersection generated by two forms of
degree t.
We follow the notation of Section 1 and consider the ideals Mt , Mt−1 and
L2t+1
t =: I. It is clear that I is generated by two 2t-pfaffians which form a complete
intersection. By Theorem 1.7, Mt−1 is obtained from Mt via an elementary G-
biliaison of height 1 on I. In other words, there are homogeneous polynomials f, g
of degree t − 1 and t respectively, such that
f Mt + I = gMt−1 + I ⊂ R.
By the additivity of the Hilbert function on the two short exact sequences
0 −→ I[−t + 1] −→ I ⊕ Mt [−t + 1] −→ f Mt + I −→ 0
0 −→ I[−t] −→ I ⊕ Mt−1 [−t] −→ gMt−1 + I −→ 0
one obtains that HMt (d − t + 1) − HI (d − t + 1) = HMt−1 (d − t) − HI (d − t) for any
d ∈ Z. By setting m = d − t + 1, we get
HMt (m) = HMt−1 (m − 1) + ΔHI (m).
Since dim R/I − 1 = dim R/Mt−1 = dim R/Mt , one has
hMt (m) = hMt−1 (m − 1) + h(t,t) (m).
Solving the recursion one obtains

t
hMt (m) = hM1 (m − t + 1) + h(j,j) (m − t + j).
j=2

This proves that the h-vector of Mt is obtained by summing the h-vectors of suitable
complete intersections. Notice that the h-vectors involved in the summation are
INVARIANTS OF IDEALS GENERATED BY PFAFFIANS 61
15

shifted in such a way, that the maximum is always attained at the same point.
Therefore, their sum hMt is of decreasing type. 

We can easily compute the graded Betti numbers of Mt as follows.

Proposition 4.3. A minimal free resolution of Mt has the form


0 −→ R[−2t − 1] −→ R[−t − 1]2t+1 −→ R[−t]2t+1 −→ Mt −→ 0.

Proof. We prove the statement by induction on t ≥ 1. If t = 1, the ideal M1


is generated by three distinct indeterminates, hence a minimal free resolution has
the form
0 −→ R[−3] −→ R[−2]3 −→ R[−1]3 −→ M1 −→ 0.
Assume now that t ≥ 2 and consider the ideals Mt , Mt−1 and L2t+1 t . We denote
L2t+1
t by I for brevity. It is clear that I is generated by two 2t-pfaffians which
form a complete intersection. By Theorem 1.7, Mt−1 is obtained from Mt via
an elementary G-biliaison of height 1 on I. Moreover, there are homogeneous
polynomials f, g of degree t − 1 and t respectively, such that
f Mt + I = gMt−1 + I ⊂ R.
By induction hypothesis Mt−1 has a minimal free resolution of the form
0 −→ R[−2t + 1] −→ R[−t]2t−1 −→ R[−t + 1]2t−1 −→ Mt−1 −→ 0.
Let
0 −→ F3 −→ F2 −→ F1 −→ Mt −→ 0
be a minimal free resolution of Mt . Applying the Mapping Cone to the two short
exact sequences
0 −→ I[−t + 1] −→ I ⊕ Mt [−t + 1] −→ f Mt + I −→ 0

0 −→ I[−t] −→ I ⊕ Mt−1 [−t] −→ gMt−1 + I −→ 0


one obtains free resolutions for the ideal J = f Mt + I = gMt−1 + I of the form
R[−3t + 1] R[−t]2
0 −→ ⊕ −→ R[−2t] 2t+2
−→ ⊕ −→ J −→ 0
R[−3t] R[−2t + 1]2t−1
and
R[−3t + 1] R[−2t] ⊕ R[−2t + 1]2 R[−t]2
0→ ⊕ → ⊕ → ⊕ → J → 0.
F3 [−t + 1] F2 [−t + 1] F1 [−t + 1]
The first free resolution must be minimal, hence
(4.1) F3 ⊇ R[−2t − 1], F2 ⊇ R[−2t]2t+1 , and F1 ⊇ R[−t]2t+1 .
Since no cancellation is possible among F1 [−t + 1], F2 [−t + 1] and F3 [−t + 1] in the
second free resolution of J, we deduce that all the containments in (4.1) must be
equalities. 
62
16 EMANUELA DE NEGRI AND ELISA GORLA

References
A. L. Avramov. “A class of factorial domains”, Serdica 5 (1979), 378–379.
BE. D. Buchsbaum, D. Eisenbud. “Algebra structures for finite free resolutions, and some
structure theorems for ideals of codimension 3”, Amer. J. Math. 99 (1977), no. 3,
447–485.
BCG. N. Budur, M. Casanellas, E. Gorla. “Hilbert functions of irreducible arithmetically
Gorenstein schemes”, J. Algebra 272 (2004), no. 1, 292–310.
C. A. Conca, “Gröbner bases of ideals of minors of a symmetric matrix”, J. Algebra 166
(1994), no. 2, 406–421.
CN. A. Corso, U. Nagel. “Monomial and toric ideals associated to Ferrers graphs”, Trans.
Amer. Math. Soc. 361 (2009), no. 3, 1371–1395.
D. E. De Negri. “Pfaffian ideals of ladders”, J. Pure Appl. Alg.125 (1998), 141–153.
D1. E. De Negri. “Some results on Hilbert series and a-invariant of Pfaffian ideals”, Math.
J. Toyama Univ. 24 (2001), 93–106.
DGo. E. De Negri, E. Gorla. “G-Biliaison of ladder Pfaffian varieties”, J. Algebra 321 (2009),
no. 9, 2637–2649.
K. C. Krattenthaler, “The major counting of nonintersecting lattice paths and generating
functions for tableaux”, Mem. Amer. Math. Soc. 115 (1995).
HT. J. Herzog, N. V. Trung. “Gröbner bases and multiplicity of determinantal and Pfaffian
ideals”, Adv. Math. 96 (1992), 1–37.
HTV. J. Herzog, N. V. Trung, G. Valla. “On hyperplane sections of reduced irreducible vari-
eties of low codimension”, J. Math. Kyoto Univ. 34 (1994), no. 1, 47–72.
KL. H. Kleppe, D. Laksov. “The algebraic structure and deformation of Pfaffian schemes”,
J. Algebra 64 (1980), 167–189.

Università di Genova, Dipartimento di Matematica, Via Dodecaneso 35, IT-16146


Genova, Italia.
E-mail address: denegri@dima.unige.it

Universität Basel, Departement Mathematik, Rheinsprung 21, CH-4051 Basel, Switzer-


land.
E-mail address: elisa.gorla@unibas.ch
Contemporary Mathematics
Volume 555, 2011

Hilbert polynomial and the intersection of ideals

Juan Elias and Jordi Martı́nez-Borruel

In honor to W. V. Vasconcelos in occasion of his 70-th birthday.

Abstract. We prove that if two perfect ideals I, J are geometrically linked


and the tangent cone defined by them satisfy some condition, then we have a
full control of the Hilbert coefficients of I ∩J in terms of the Hilbert coefficients
of I, J and I + J. As a corollary we provide a purely algebraic proof of the
Sally’s conjecture on the monotony of the Hilbert function, extending to the
non-equicharacteristic case the previous proof given by the first author.

Introduction
It is well known that in the local case the Hilbert polynomial behaves bad
with respect the intersection of ideals; see Example 1.6. Moreover, given I, J two
ideals geometrically linked in a Cohen-Macaulay ring R, what can be said about the
Hilbert polynomial of R/I∩J with respect the Hilbert polynomials of R/I and R/J?
In view of the well known formula about their multiplicities e0 (I∩J) = e0 (I)+e0 (J),
there exists similar formulae for the higher Hilbert coefficients ei ?
In this paper we show that, under a good condition on the tangent cones, there
is an easy link between the Hilbert coefficients of I, J, I +J and I ∩J, Corollary 1.3.
In particular, if I, J are geometrically linked then we get the Hilbert coefficients of
I ∩ J in terms of the Hilbert coefficients of I, J and I + J, Proposition 1.5. See [3],
and [4], for the corresponding result in the graded case.
In the section two we carefully analyze the one-dimensional case giving a purely
algebraic proof, avoiding the use of infinitely near points, of some previous results
of [2]. As corollary we give in Theorem 2.8 a short proof of Sally’s conjecture, [8],
already settled by the author in [2]. See also the contribution of M. E. Rossi in this
volume, [7]
Notations. Let R be a d-dimensional Noetherian local ring with maximal ideal
m. We assume that the residue field k = R/m is infinite. For an ideal I ⊂ R we

2010 Mathematics Subject Classification. Primary 13H15, 13H10; Secondary 14B05.


Key words and phrases. Hilbert polynomial, liaison.
The first author is partially supported by MTM2010-20279-C02-01.

1
63
64
2 JUAN ELIAS AND JORDI MARTÍNEZ-BORRUEL

define the Hilbert function of R/I as


 
0 mn + I
HR/I (n) = dimk ,
mn+1 + I
and the first iterated Hilbert function by

n  
1 0 R
HR/I (n) = HR/I (i) = dimk .
i=0
mn+1 + I

By a well known result of Hilbert there exist integers ej (I) ∈ Z such that the
polynomial
d  
z+d−j
h1R/I (z) = (−1)j ej (I) ,
j=0
d−j
known as the Hilbert polynomial of R/I, satisfies
h1R/I (n) = HR/I
1
(n)
for all n  0. The numbers e0 (I), e1 (I), . . . , ed (I) are the Hilbert coefficients of
R/I, e0 (I) is the multiplicity of R/I.
We denote by grm (R) = ⊕n≥0 mn /mn+1 the associated graded ring to R. Then
we define I ∗ , the ideal of initial forms of I, to be the graded ideal of grm (R) such
that grm (R/I) = grm (R)/I ∗ . From the exact sequence
mn ∩ (mn+1 + I) mn mn + I
0 −→ In∗ := −→ −→ −→ 0
mn+1 mn+1 mn+1 + I

we get I ∗ = ∗
n≥0 In . The tangent cone of X = Spec (R/I) is the projective
sub-scheme of Proj (grm (R)) defined by Con (X) = Proj (grm (R/I)).

1. Hilbert polynomial and liaison


Let I, J, K be ideals of R such that K ⊆ I ∩ J. For all n ∈ N we consider the
finite length R-module
(I + mn+1 ) ∩ (J + mn+1 )
Tn = .
K + mn+1

Proposition 1.1. Let I, J, K be ideals of R such that K ⊆ I ∩ J. For all


n ≥ 0 it holds
1
λk (Tn ) = HR/K 1
(n) + HR/I+J (n) − HR/I
1
(n) − HR/J
1
(n).
In particular λk (Tn ) is an asymptotically polynomial function on n for all n  0.
The following conditions conditions are equivalent:
(i) Tn = 0, for n  0,
(ii) h1R/K = h1R/I + h1R/J − h1R/I+J .
If we write X = Spec (R/I), Y = Spec (R/J) and Z = Spec (R/K) then the above
conditions imply the following equivalent conditions:
(iii) In∗ ∩ Jn∗ = Kn∗ , for n  0,
(iv) Con (X) ∪ Con (Y ) = Con (Z).
If depth (R/K) > 0 then the above four conditions are equivalent.
HILBERT POLYNOMIAL AND THE INTERSECTION OF IDEALS 65
3

Proof. From the exact sequence of R-modules


R R R R
0 −→ Tn −→ −→ ⊕ −→ −→ 0
K + mn+1 I + mn+1 J + mn+1 I + J + mn+1
we get
1 1 1 1
HR/K (n) + HR/I+J (n) = HR/I (n) + HR/J (n) + λk (Tn ).
From this identity we get the first part of the claim and that (i) is equivalent to
(ii).
For all n ≥ 0 let us consider the exact sequence of R-modules
In∗ ∩ Jn∗ mn + [(I + mn+1 ) ∩ (J + mn+1 )]
0 −→ −→ T n −→ −→ 0.
Kn∗ K + mn
If Tn = 0 for n  0 the from the above exact sequence we get (iii). From the
definition of the tangent cone it is easy to deduce the equivalence between (iii) and
(iv).
Now, if In∗ ∩ Jn∗ = Kn∗ for n  0 then, from last exact sequence again, we have
mn + [(I + mn+1 ) ∩ (J + mn+1 )] (I + mn ) ∩ (J + mn )
0 −→ Tn = → = Tn−1 .
K + mn K + mn
Hence we get a decreasing sequence of finite length R-modules
Tn−1 ⊇ Tn ⊇ Tn+1 ⊇ . . . ,
so we have Tn+1 = Tn for n ≥ n0 .
Since k is infinite and depth (R/K) > 0, there exists a superficial element x no
zero-divisor in R/K. So, for n ≥ n1 we have that the multiplication by x defines
the monomorphism:
·x
0 −−−−→ R
K+mn −−−−→ R
K+mn+1
that induces a monomorphism
·x
0 −−−−→ Tn−1 −−−−→ Tn .
Since the R-modules Tn have bounded length modules for n ≥ n0 . Then for all
n ≥ Max{n0 , n1 }
Tn = xTn−1 = xTn
and by Nakayama’s lemma, Tn = 0 for n ≥ Max{n0 , n1 }. 

Remark 1.2. Notice that in general we have (I ∩ J)∗ ⊂ I ∗ ∩ J ∗ , in the last


Proposition we have showed that if we have the asymptotic equality of the graded
pieces of these ideals then we can control the variation of the Hilbert polynomials.
On the other hand, there are two cases where the tangent cone can be computed or
controlled easily, this is the case of hypersurfaces and the case of one-dimensional
Cohen-Macaulay local rings. We will study the one-dimensional case in the second
section of this paper, see Example 1.7 for the case of hypersurfaces. On the other
hand is not difficult to prove that the condition depth (R/K) > 0 is necessary in
Proposition 1.1.

Corollary 1.3. Let K ⊂ I ∩ J be ideals of R such that depth (R/K) > 0. Let
us suppose
(i) dim(R/I) = dim(R/J) = dim(R/K) = d, dim(R/I + J) = t, and
66
4 JUAN ELIAS AND JORDI MARTÍNEZ-BORRUEL

(iii) Con (X) ∪ Con (Y ) = Con (Z).


Then it holds
(a) ei (K) = ei (I) + ei (J), for ≤ i < d − t,
(b) ei (K) = ei (I) + ei (J) + (−1)t−d+1 ei+t−d (I + J), for d − t ≤ i ≤ d:
Proof. From last theorem we have h1K (n) + h1I+J (n) = h1I (n) + h1J (n) and
writing this equality in terms of the Hilbert coefficients, we get the claim. 
Definition 1.4. Let us assume that R is regular. Two perfect ideals of the
same height I, J ⊂ R, without common minimal primes, are geometrically linked
by a Gorenstein ideal K if and only if I ∩ J = K.

From [5], Remarque 1.4, we know that dim(R/I + J) = dim(R/K) − 1, and by


the last corollary we obtain:

Proposition 1.5. Let I, J be two perfect ideals geometrically linked by a


Gorenstein ideal K. If Con (X) ∪ Con (Y ) = Con (Z) then
(i) e0 (K) = e0 (I) + e0 (J),
(ii) ei (K) = ei (I) + ei (J) + ei−1 (I + J), for all i ≥ 1.

In the next example we prove that Con (X) ∪ Con (Y ) = Con (Z) is a necessary
condition in the two last results.
Example 1.6. Consider R = k[x, y](x,y) , I = (y − x2 ), J = (y). Then K =
I ∩ J = (y 2 − yx2 ). Then, we have Con (R/I) ∪ Con (R/J)  Con (R/K) and
e1 (K) = 1 = 2 = e1 (I) + e1 (J) + e0 (I + J).

In the next remark we study the hypersurface case.


Remark 1.7. Let us consider now R = k[X1 , · · · , Xd ](X1 ,··· ,Xd ) , I = (F ), and
J = (G), where F, G ∈ S = k[X1 , · · · , Xd ] are coprimes polynomials of order d1
and d2 respectively. Then I ∩ J = (F.G), I ∗ = (F ∗ ), J ∗ = (G∗ ), (I ∩ J)∗ = (F ∗ G∗ ),
and I ∗ ∩ J ∗ = (mcm (F ∗ , G∗ )). If we take K = I ∩ J then
 
In∗ ∩ Jn∗ ∼ S
=
Kn∗ H n−(d1 d2 −t)
where H = F ∗ G∗ /mcm (F ∗ , G∗ ) is a degree t homogenous polynomial. Hence
In∗ ∩ Jn∗ = (I ∗ ∩ J ∗ )n for n  0 if and only if t = 0 or, equivalently, gcd(F ∗ , G∗ ) = 1.
Notice that under these hypotheses, I, J are geometrically linked by K.

2. 1-dimensional rings
Although the condition Con (R/I) ∪ Con (R/J) = Con (R/I ∩ J) of Proposi-
tion 1.1 is necessary in order to control the behavior of the Hilbert coefficients, as
we saw in the Example 1.6, that condition is difficult to check or to impose by di-
mension restrictions. In the one-dimensional case we can easily control the relative
position of Con (X) and Con (Y ), as we will see in this section. See [6] for the
computation of the tangent cone of the codimension two monomial curves.
In this section we assume that R is a regular local ring of dimension d, and we
denote by X1 , · · · , Xd a regular system of parameters of R. Then the associated
HILBERT POLYNOMIAL AND THE INTERSECTION OF IDEALS 67
5

graded ring to R is isomorphic to the polynomial ring S = k[X1 , · · · , Xd ]. For a


graded ideal L ⊂ S we denote by HS/L0
the Hilbert function of the quotient S/I,
i.e.
0
HS/L = dimk ((S/L)n ) = dimk (Sn /Ln ).
for all n ≥ 0.
The first part of the following result is well known but we insert it here for the
sake of completeness.

Proposition 2.1. (i) Given ideals I, J of R such that depth (R/I) ≥ 1 and
depth (R/J) ≥ 1, then
depth (R/I ∩ J) ≥ 1.
In particular, if I, J are perfect ideals of height d − 1 then I ∩ J is also a perfect
ideal of height d − 1.
(ii) Given ideals I, J ⊂ R of non-maximal height, let us consider the schemes
X = Spec (R/I), Y = Spec (R/J) defined by them. If Con (X) ∩ Con (Y ) = ∅
then I, J do not share minimal primes.
Proof. (i) This is a straight depth counting in the standard exact sequence
R R R R
0 −→ −→ ⊕ −→ −→ 0.
I ∩J I J I +J
(ii) Let us assume that p is a common minimal prime of I and J. The minimal
primes of I, J have no maximal height, so we have that the height of p is also non-
maximal. Since dim(grk (R/p)) = dim(R/p) > 0 we get that In∗ ∩ Jn∗ ⊂ p∗n  Sn for
all n ≥ 0. Hence ∅  Con (R/p) ⊂ Con (X) ∩ Con (Y ). 
Remark 2.2. Notice that (i) of the last result cannot be generalized to higher
depths as the the following classical example shows: R = k[X, Y, Z, T ](X,Y,Z,T ) , and
the height two perfect ideals I = (X, Y ), and J = (Z, T ). It is well known that
R/I ∩ J has exactly depth one.

Proposition 2.3. Let I, J be height d − 1 perfect ideals of R. Let X =


Spec (R/I), Y = Spec (R/J) the one-dimensional Cohen-Macaulay schemes defined
by the ideals I, J, respectively. If Con (X) ∩ Con (Y ) = ∅ then
Con (R/I) ∪ Con (R/J) = Con (R/I ∩ J).
Proof. From the definition of the tangent cone we get that the condition
Con (X) ∩ Con (Y ) = ∅ is equivalent to
In∗ + Jn∗ = Sn
for all n  0. From this identity we get
0 0 0 0 0 0
HR/I (n) + HR/J (n) =HS/I ∗ (n) + HS/J ∗ (n) = HS/I ∗ ∩J ∗ (n) + HS/I ∗ +J ∗ (n)

0
=HS/I ∗ ∩J ∗ (n)

for all n  0. From the fact (I ∩ J)∗ ⊂ I ∗ ∩ J ∗ we get


∗ ∩J ∗ (n) ≤ HS/(I∩J)∗ (n)
0 0 0 0
(1) HR/I (n) + HR/J (n) = HS/I
for all n  0.
68
6 JUAN ELIAS AND JORDI MARTÍNEZ-BORRUEL

On the other hand, since I, J be height d−1 perfect ideals of R without common
minimal primes, Proposition 2.1 (ii), from the additivity of the multiplicity we get
e0 (I ∩ J) = e0 (I) + e0 (J).
From this identity and the inequality (1) we deduce
∗ ∩J ∗ (n) = HS/(I∩J)∗ (n) = e0 (I ∩ J)
0 0 0 0
e0 (I) + e0 (J) = HR/I (n) + HR/J (n) = HS/I
for all n  0. Hence
In∗ ∩ Jn∗ = (I ∩ J)∗n
for all n  0, so
Con (R/I) ∪ Con (R/J) = Con (R/I ∩ J).


From the last result and Proposition 1.5 we deduce


Corollary 2.4. Let I, J be height d−1 perfect ideals of R. Let X = Spec (R/I),
Y = Spec (R/J) the one-dimensional Cohen-Macaulay schemes defined by the ideals
I, J, respectively. If Con (X) ∩ Con (Y ) = ∅ then
(i) e0 (I ∩ J) = e0 (I) + e0 (J),
(ii) e1 (I ∩ J) = e1 (I) + e1 (J) + e0 (I + J).

Remark 2.5. Notice that the condition Con (X)red ∩ Con (Y )red = ∅ is equi-
valent to Con (X) ∩ Con (Y ) = ∅. In the one-dimensional case the first condition is
easy to check or impose as we will see in the next result.

Definition 2.6. Given an ideal I ⊂ R we define the initial degree s(I) of I by


s(I) = Max{n | I ⊂ mn }.
Notice that s(I) is also the least integer n such that In∗ = 0.
Given a nonzero vector a = (a1 , a2 , · · · , ad ) ∈ kd we denote by L(a1 , a2 , · · · , ad )
the height d − 1 perfect ideal defined by the 2 × 2-minors of the matrix
 
a1 · · · a d
X1 · · · Xd
Notice that if ad = 0 then the ideal L(a1 , a2 , · · · , ad ) is generated by the linear
regular sequence of forms ad X1 − a1 Xd , · · · , ad Xd−1 − ad−1 Xd ; and L∗ is the homo-
geneous ideal of S generated by the above linear forms. In a geometric framework,
i.e. R = [X1 , · · · , Xd ](X1 ,··· ,Xd ) , R/L is the local ring of regular functions at the
origin of the straight line Y = {(a1 , a2 , · · · , ad )t | t ∈ k} ⊂ kd .
In the next result we study the variation of the Hilbert polynomial of R/I by
”adding a line”, i.e. the variation of the Hilbert polynomial by intersecting I with
L(a1 , a2 , · · · , ad ). See [2], Lemma 2.1, by a geometric proof of this result by using
the device of infinitely near points.
HILBERT POLYNOMIAL AND THE INTERSECTION OF IDEALS 69
7

Proposition 2.7. Let I be a height d − 1 perfect ideal of R, and let H ∈ I ∗


be an homogeneous form of degree s = s(I). Let L = L(a1 , a2 , · · · , ad ) be the ideal
defined by a nonzero vector a = (a1 , a2 , · · · , ad ) ∈ kd . If H(a1 , a2 , · · · , ad ) = 0 then
  
0 0 n+d−1 
HI∩L (n) = Max HI (n) + 1,
d−1
for all n ≥ 0.
Proof. Let us assume that ad = 0. Then HXdn−s ∈ In∗ but HXdn−s ∈ L∗n .
Hence
(I ∩ L)n ⊂ In∗ ∩ L∗n  In∗
for all n ≥ s. From this we get that for all n ≥ s
(2) 0
HI∩L (n) ≥ HI0 (n) + 1.
Let X = Spec (R/I), Y = Spec (R/L) the one-dimensional Cohen-Macaulay
schemes defined by the ideals I, L, respectively. Since Con (X) is a subscheme of
the hypersurface defined by H and H(a1 , a2 , · · · , ad ) = 0 we get that Con (X)red ∩
Con (Y )red = ∅, so Con (X) ∩ Con (Y ) = ∅. Since H(a1 , a2 , · · · , ad ) = 0 and H is
an hypersurface of degree s = s(I) we deduce that e0 (I + J) = s. By Corollary 2.4
we obtain e0 (I ∩ J) = e0 (I) + 1, and
e1 (I ∩ L) = e1 (I) + s.
From these identities and (2) we get that
0
HI∩L (n) = HI0 (n) + 1
for all n ≥ s. From this it is easy to deduce the claim. 

In the next result we prove Sally’s conjecture, [8], let’s recall that in [2] we
proved that conjecture in the equicharacteristic case, here we extend the proof to
the general case. See [2] for some historical hints on Sally’s conjecture.

Theorem 2.8 (Sally’s Conjecture). Let A be a one-dimensional Cohen-Macaulay


local ring of embedding dimension three. Then the Hilbert function of A is non-
decreasing, i.e.
HA0
(n + 1) ≥ HA0
(n)
for all n ≥ 0.
Proof. Let us assume that HI0 (t) > HI0 (t + 1) for some t ∈ N. Since the
residue field is infinite, by the last result there exist finitely many ideals Li =
L(ai1 , ai2 , ai3 ) ∈ k3 \ {0}, i = 1, · · · , r = t+2
2 − H 0
I (t), such that t + 1 = s(K),
K = I ∩ L1 ∩ · · · ∩ Lr , and
 
0 t+2
HK (t + 1) < .
2

In particular dimk (Kt+1 ) ≥ t + 3, so the minimal number of generators if K is at
least t + 3. Notice that K is a height two perfect ideal, Proposition 2.1. Then by
Burch’s result, [1], we know that the minimal number of generators of K is less
or equal that its initial degree plus one, i.e. t + 2. Hence we get a contradiction
0
on the number of generators of K, so there is not an integer t such that HA (t) >
0
HA (t + 1). 
70
8 JUAN ELIAS AND JORDI MARTÍNEZ-BORRUEL

References
[1] L. Burch, On ideals of finite homological dimension in local rings, Math. Proc. Camb. Phil.
Soc. 64 (1968), 941–946.
[2] J. Elias, The conjecture of Sally on the Hilbert function for curve singularities, J. Algebra
160 (1993), no. 1, 42–49.
[3] R. M. Miró-Roig, Determinantal ideals, Progress in Mathematics, vol. 264, Birkhäuser Verlag,
Basel, 2008.
[4] U. Nagel, On Hilbert function under liaison, Matematiche (Catania) 46 (1991), no. 2, 547–558
(1993).
[5] C. Peskine and L. Szpiro, Liaison des varietes algébriques I, Inv. Math. 26 (1974), no. 271-302,
271–302.
[6] L. Robbiano and G. Valla , On the equations defining tangent cones, Math. Proc. Camb. Phil.
Soc. 88 (1980).
[7] M.E. Rossi, Hilbert functions of local rings, Contemporary Math., This volume.
[8] J. Sally, Number of generators of ideals in local rings, Lec. Notes in Pure and Appl. Math.
New York. 35 (1978).

`
Departament d’Algebra i Geometria, Universitat de Barcelona. Gran Via 585,
08007 Barcelona, Spain
E-mail address: elias@ub.edu

EESA/CPD, Institut del Teatre, Plaça Margarida Xirgu s/n, 08004 Barcelona,
Spain
E-mail address: jordimab@gmail.com
Contemporary Mathematics
Volume 555, 2011

Polynomial vector fields with algebraic trajectories

Viviana Ferrer and Israel Vainsencher

Abstract. It is known after Jouanolou that a general holomorphic foliation


of degree ≥ 2 in the complex projective plane has no algebraic leaf. We give
formulas for the degrees of the subvarieties of the parameter space of one-
dimensional foliations in complex projective spaces that correspond to folia-
tions endowed with some invariant subvariety of degree 1 or 2 and dimension
≥ 1.

Introduction
Holomorphic foliations are an offspring of the geometric theory of polynomial
differential equations. Following the trend of many branches in Mathematics, in-
terest has migrated to global aspects. Instead of focusing on just one curve, or
surface, or metric, or differential equation, try and study their family in a suitable
parameter space. The geometry within the parameter space of the family acquires
relevance. For instance, the family of hypersurfaces of a given degree correspond
to points in a suitable projective space; geometric conditions imposed on hyper-
surfaces, e.g., to be singular, usually correspond to interesting subvarieties in the
parameter space, e.g., the discriminant. Hilbert schemes have their counterpart in
the theory of polynomial differential equations, to wit, the spaces of foliations.
Let us recall that, while a general surface of degree d ≥ 4 in P3 contains no line
–in fact, only complete intersection curves are allowed, those that do contain some
line correspond to a subvariety of codimension d − 3 and degree d+1 4
4 (3d + 6d +
3

17d + 22d + 24)/4! in a suitable P .


2 N

Similarly, while a general holomorphic foliation, say in the complex plane P2 ,


of degree d ≥ 2 has no algebraic leaf, those that do have, say an invariant line,
correspond to a subvariety of codimension d − 1 and degree 3 d+3 4 in a suitable
P .
N

Our goal is to give similar closed formulas for the degrees of the subvarieties
of the parameter space of one-dimensional foliations of degree d on the complex
projective space of dimension n that correspond to foliations endowed with some
invariant subvariety of degree 1 or 2 and dimension ≥ 1. Imposing a linear invariant
subvariety is easy, essentially due to the absence of degenerations. The classical

1991 Mathematics Subject Classification. Primary 14C17, 37F75; Secondary 32M25.


Key words and phrases. holomorphic foliation, invariant subvarieties, enumerative geometry.
The authors were partially supported by CNPQ.
1

71
72
2 VIVIANA FERRER AND ISRAEL VAINSENCHER

spaces of complete quadrics help us handle the quadratic case. For higher degree,
we don’t know the answer.
.

1. The space of foliations


The main reference for this material is Jouanolou [7]. We call a vector field of
degree d in Pn a global section of T Pn ⊗ OPn (d − 1), for some d ≥ 0.
Denote by Sd the space Symd (Cn+1 )∨ of homogeneous polynomials of degree d
in the variables z0 , . . . , zn . We write ∂i = ∂/∂zi , thought of as a vector field basis
for Cn+1 = S1∨ . Recalling Euler sequence
0 → OPn (d − 1) −→ OPn (d) × Cn+1 −→ T Pn (d − 1) → 0
and taking global sections we get the exact sequence

0 → Sd−1 −→ Sd ⊗ S1∨ −→ Vdn := H 0 (Pn , T Pn (d − 1)) → 0.


ι
(1.1)

Here ι(H) = HR, with R = zi ∂i the radial vector field. Any degree d vector
field X ∈ Vdn can be written in homogeneous coordinates as
(1.2) X = F0 ∂0 + · · · + Fn ∂n ,
where the Fi ’s denote homogeneous polynomials of degree d, modulo multiples of
the radial vector field. A vector field X induces a distribution of directions in T Pn .
Any nonzero multiple of X yields the same distribution.
The space of foliations of degree d in Pn is the projective space
(1.3) PN = P(Vdn )
of dimension d+n d+n−1
N = (n + 1) n − n − 1.
We shall often abuse notation and denote by the same symbol X both the foliation
and a vector field. The singular scheme of X is the scheme of zeros of the section
X : OPn → T Pn (d − 1). With X as in (1.2), the singular scheme is given by the
2×2 minors of the matrix  
F0 F 1 · · · F n
.
z0 z1 · · · zn

1.1. Invariant subvarieties. Let X be a foliation in Pn . An irreducible


subvariety Z ⊂ Pn is said to be invariant by X if
X (p) ∈ Tp Z
for all p ∈ Z \ (Sing(Z) ∪ Sing(X )). If Z is reducible, it is invariant by X if and
only if each irreducible component of Z is invariant by X . If Z is defined by a
saturated ideal IZ := G1 , . . . , Gr , invariance means
dGi (X ) = X (Gi ) ∈ IZ
for all i = 1, . . . , r. The hypothesis of saturation is necessary, see [3, p. 5]. It can
be easily checked that the condition above does not depend on the representative
of X in Sd ⊗ S1∨ .
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 73
3

2. Foliations with an invariant k-plane


We show that the locus in PN corresponding to foliations with an invariant k-
plane is the birational image of a natural projective bundle over the Grassmannian
of k-planes in Pn .

2.1. Consider the tautological exact sequence of vector bundles over the
Grassmannian G := G(k, n) of projective k-planes in Pn ,
(2.1) 0 → T → G × Cn+1 → Q → 0
where T is of rank k + 1. The projectivization
P(T ) = {(W, p) ∈ G × Pn | p ∈ W }
is the universal k-plane. Write the projection maps
P(T ) ⊂ G × Pn
q NNN
p1 qqq NNpN2
qq q NNN
q NN'
x qq
q
G Pn .
We denote by N the normal bundle to P(T ) in G × Pn . We have the exact sequence
over P(T ),
(2.2) 0 → TP(T )/G → p∗2 T Pn → N → 0.
It is easy to see that
N = p∗1 Q ⊗ OT (1).
Proposition 2.1. Notation as in (1.1) and (2.1), there exists a vector subbundle
E ⊂ G × Vdn
such that
(i) P(E) = {(W, X ) ∈ G × PN | W is invariant by X }.
Set Y := q2 (P(E)), where q2 : P(E) → PN is theprojection. Then
(ii) the codimension of Y in PN is (n − k)( k+dd − (k + 1)) and
(iii) the degree of Y is given by the top-dimensional Chern class,
cg (Q ⊗ Symd (T ∨ ))
where g := dim G.
Proof. Consider the following diagram of maps of vector bundles over P(T ),

TP(T )/G (d − 1)



Vdn = H 0 (Pn , T Pn (d − 1))
ev / p∗2 T Pn (d − 1)
UUUU
UUUU
UU
ϕ UUUUU 
U*
p∗1 Q ⊗ OT (d).
The map of evaluation yields a surjective map of vector bundles
ϕ : Vdn  p∗1 Q ⊗ OT (d).
74
4 VIVIANA FERRER AND ISRAEL VAINSENCHER

The kernel of ϕ is the vector bundle


kerϕ = {(W, p, X ) | X (p) ∈ Tp W }.
Observe that a k-plane W is invariant by X if and only if (W, p, X ) ∈ kerϕ for
all p ∈ W = p−1
1 (W ). Put in other words, we ask ϕ to vanish along the fibers of
p1 : P(T ) → G. It follows from [1, p. 16] that there exists a map of vector bundles
over G,
(♥) ϕ : Vdn / / p1∗ (p∗1 Q ⊗ OT (d)) = Q ⊗ Symd (T ∨ )

such that
E := ker(ϕ ) = {(W, X ) | X (p) ∈ Tp W ∀p ∈ W }.
The projectivization P(E) ⊂ G × PN is clearly as stated in (i). Let q1 : P(E) → G
and q2 : P(E) → PN be the maps induced by projection. It can be shown that q2 :
P(E) → PN is generically injective: the general vector field of degree d ≥ 2 with an
invariant k-plane has exactly one invariant k-plane, see Lemma 2.3 and Lemma 2.4
below. Write H for the hyperplane class of PN . Set u = dim Y = dim P(E). We
have q2∗ H = c1 OE (1) =: h. We may compute
   
deg Y = Hu ∩ Y = hu = q1∗ (hu ) = sg (E).
PN P(E) G G

The assertions (ii) and (iii) follow noticing that the Segre class satisfies
sg (E) = cg (Q ⊗ Symd (T ∨ ))
in view of the exact sequence arising from (♥),
E / / Vdn / / Q ⊗ Symd (T ∨ ).


In the case of invariant hyperplane (i.e., k = n − 1) we have explicitly


Theorem 2.2. The degree of the subvariety Y of the space of foliations in Pn that
admit an invariant hyperplane is given by 
d+n 
deg Y = n .
n

Proof. We have the following exact sequence over G(n − 1, n) = P̌n ,
0 → Sd−1 ⊗ Q∨ → Sd = Symd (Cn+1 )∨ → Symd (T ∨ ) → 0.
Twisting by Q = OP̌n (1) we obtain:
0 → Sd−1 → Sd ⊗ Q → Symd (T ∨ ) ⊗ Q → 0.
From this we can compute cn (Symd (T ∨ ) ⊗ Q) = cn (Sd ⊗ Q). Setting H = c1 (Q),
the hyperplane class in P̌n , the sought-for degree is just the coefficient of H n in
d+n
(1 + H)( n ) . 

To compute cg (Q ⊗ Symd (T ∨ )) for any fixed k, n, see the script in § 5.


Lemma 2.3. In the situation of the previous proposition, in order to prove that q2
is generically one to one, it suffices to find a point y ∈ Y such that the fiber q2−1 (y)
consists of one reduced point.
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 75
5

Proof. If we prove the existence of such point y there exists a non empty
open set U in Y such that the fiber over each point in U has dimension zero.
Therefore q2 is generically finite. Shrinking U we may assume (i) U is affine, say
with coordinate ring A and (ii) the restriction of q2 over q2 −1 U is finite. It follows
that q2 −1 U is affine, with coordinate ring B which is an A-module of finite type.
Now for each point u ∈ U with corresponding maximal ideal mu ⊂ A, the fiber
q2 −1 u = Spec(B/mu B) is finite and consists of dimC (B/mu B) points counted with
multiplicity. By semicontinuity, this vector space dimension attains a minimum
over an open subset. Since the fiber over y consists of one reduced point, that
mininum is precisely one and we are done. 

Lemma 2.4. Notation as in Proposition 2.1, for d > 1, the projection q2 : P(E) →
PN is generically injective.
Proof. It’s sufficient to prove that there exists an open set U1 ⊂ Y such that
for each point y ∈ U1 , q2−1 (y) consists of just one point. Indeed, in this case we
deduce that dim P(E) = dim Y. Since P(E) (resp. Y) are smooth (resp. generically
smooth) varieties we have that
dq2 : T P(E) −→ T Y
is generically of maximal rank, i.e., there exists an open set U2 ⊂ Y such that
if y ∈ U2 , and x ∈ q2−1 (y), then dx q2 is surjective, equivalently dx q2 is injective.
Summarizing we find an open set U := U1 ∩U2 such that if y ∈ U then q2−1 (y) = {x},
and x is a reduced point in the fiber. It follows by Lemma 2.3, that q2 is generically
injective.
Let’s prove the existence of U1 above. Define
Y2 := {X ∈ Y | X leaves invariant two subspaces W = W  }.
It is sufficient to prove that dim Y2 < dim Y. Indeed, in this case we define
U1 := Y \ Y2 .
Now, for each 1 ≤ s ≤ min{k, n − k} set
G2s := {(W, W  ) ∈ G × G | cod(W ∩ W  ) = n − k + s},
 2s = {(X , (W, W  )) | X leaves invariant W and W  , (W, W  ) ∈ G2s }.
Y
 2s → Y ⊂ PN denote the projection and define Y2s := p(Y
Let p : Y  2s ). Then
we have, for each s
 2s
Y
{ @@
{ @@p
{{{ @@
{ @@
}{{
G2s Y
and Y2 = ∪s Y2s . Also denote YW ⊂ Y the set of foliations which leave W invariant,
and for each (W, W  ) ∈ G2s , define YW W  := YW ∩ YW  .
With this notation it is easy to prove that for each s, Y 2s → G2s is a fibration
with fiber dimension equal to the dimension of YW W  .
We claim that  
k+d
dim YW W ≤ dim YW − s
 .
d
76
6 VIVIANA FERRER AND ISRAEL VAINSENCHER

In fact, suppose that W = Z(Z0 , . . . , Zn−k−1 ). Then W is invariant by a field of


degree d,
∂ ∂
X = F0 + · · · + Fn
∂Z0 ∂Zn
if and only if
(2.3) F0 , . . . , Fn−k−1 ∈ Z0 , . . . , Zn−k−1 .
Take W  such that cod(W ∩ W  ) = n − k + s. Acting with the stabilizer of
W in P GLn+1 we can suppose that W  = Z(Z0 , . . . , Zn−k−1−s , Zi1 . . . , Zis ), where
i1 , . . . , is ∈ {0, . . . , n−k −1}. Now the condition of W  to be invariant by X implies
that, for each j = 1, . . . , s we have
Fij ∈ Z0 , . . . , Zn−k−1−s , Zi1 , . . . , Zis .
These conditions are independent of the others in (2.3). On the other hand it is easy
to count the new conditions imposed by Fij ∈ Z0 , . . . , Zn−k−1−s , Zi1 , . . . , Zis : this
  k+d
number is k+dd . So we have that codYW YW W ≥ s d . This proves the claim.


Using the claim it is easy to prove that for each s, dim Y2s < dim Y. Therefore
dim Y2 < dim Y as claimed.


3. Foliations with an invariant conic in P2


We construct a compactification of the space of foliations that leave invariant
a smooth conic. This compactification is obtained as the birational image of a
projective bundle over the variety of complete conics.
3.1. The incidence variety. The parameter space for the family of conics is
P5 = P(S2 ). We have the natural trilinear map
S2 ⊗ Sd ⊗ S1∨ → Sd+1
G ⊗ F ⊗ ∂i → (∂i G)F.
It induces the map of vector bundles over P5 ,

ϕ : O
P5 (−1) ⊗ Sd ⊗ S1 → Sd+1
given by ϕ(G, X ) := X (G) = Fi ∂i G, where G is the equation of the conic and
X = Fi ∂i , Fi ∈ Sd .
Notice that ϕ(G, HR) = 2GH, for all H ∈ Sd−1 . Recalling (1.1),

Vdn = Sd ⊗ S1∨ Sd−1 R
we see that ϕ induces a map
Sd+1
ψ : OP5 (−1) ⊗ Vdn −→ ·
OP5 (−1) ⊗ Sd−1
This fits into the commutative diagram,
 / kerψ
(3.1) kerϕ

 
OP5 (−1) ⊗ Sd−1 · R / / OP5 (−1) ⊗ Sd ⊗ S1∨ / / OP5 (−1) ⊗ V n
d

 ϕ ψ
  
OP5 (−1) ⊗ Sd−1 / / Sd+1 // Sd+1
OP5 (−1)⊗Sd−1 ·
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 77
7

Twisting by OP5 (1) we obtain the commutative diagram


 /M
(3.2) M

 
Sd−1 · R / / Sd ⊗ S ∨
1
/ / Vdn

 θ θ
  
Sd−1 / / OP5 (1) ⊗ Sd+1 // OP5 (1)⊗Sd+1
Sd−1 ,

where


:= ker θ.
M := ker θ  M
The map θ : Sd ⊗ S1∨ → OP5 (1) ⊗ Sd+1 appearing in (3.2) is surjective only
over the open subset consisting of smooth conics. In fact, for G ∈ P5 the rank of
the image of θG depends on the singularities of the conic:
⎧ d+3
⎨ total, 2 , if G is smooth;
rank θG = d+3
− 1, if G is a line pair;
d+2
⎩ 2
rank Symd = 2 , if G is a double line.
d+2
The minimal rank, r = 2 , is achieved along
V := ν2 (P(S1 )) ⊂ P(S2 ),
the Veronese variety of double lines. It turns out that over the open subset
U ⊂ P5 \ V
of smooth conics, the restriction M
U is a vector subbundle of the trivial bundle

Vd . The projectivization P(MU ) is the incidence variety


n

{(C, X ) | C is invariant by X } ⊂ U × PN .
Let us denote by Y the closure of its image in PN . We see that Y consists of (limits
of) foliations that admit an invariant smooth conic. Our strategy to find its degree
is summarized in the following.
Theorem 3.1. Let π : B → P5 be the blowup of P5 along the Veronese V. There
exists a subbundle
E ⊂ B × Vdn

U (cf. 3.2). In particular, the
such that the restriction Eπ−1 U coincides with π ∗ M
image of P(E) in P = P(Vd ) is equal to Y.
N n

Proof. Consider the pullback by π of the maps ϕ and ψ (cf. 3.1)


ϕB : π ∗ (OP5 (−1) ⊗ Sd ⊗ S1∨ ) → π ∗ Sd+1 ,

ψB : π ∗ (OP5 (−1) ⊗ Vdn ) → π ∗ ( Sd−1 ⊗O


Sd+1
5 (−1)

P

By Lemma 3.2 below it’s enough to prove that the ideals generated by the k × k
minors of ϕB are locally principal for all k ≥ 1. Indeed, in this case J := ImϕB is
locally free. Therefore we obtain a factorization of ϕB ,
78
8 VIVIANA FERRER AND ISRAEL VAINSENCHER

π ∗ (OP5 (−1) ⊗ Sd ⊗ S1∨ )


ϕB
/ π ∗ Sd+1
SSS O
SSS
SSS
SSS ?
ϕ SSS
SS) )
J,

and π ∗ (OP5 (−1) ⊗ Sd−1 ) = ϕB (π ∗ (OP5 (−1) ⊗ Sd−1 · R)) ⊂ J . This factorization
induces a factorization of ψB

π ∗ (OP5 (−1) ⊗ Vdn )


ψB
/ π∗( S d+1
SSS OP5 (−1)⊗Sd−1 )
SSS O
SSS
SSS ι
 SSS
ψ SSS
S) )
J ,
where
J
J :=
π ∗ (OP5 (−1) ⊗ Sd−1 )
is a vector bundle. Define
(3.3)
E := π ∗ OP5 (1) ⊗ kerψ.

over the
It follows that E is a subbundle of B × Vdn that coincides with π ∗ M
open set π −1 (U ), where U ⊂ P5 is the open set of smooth conics. Indeed, over
π −1 (U ) = U the map  
∗ Sd+1
ι:J →π
OP5 (−1) ⊗ Sd−1
is an isomorphism. Therefore kerψ = kerψB over U .
To prove that the ideals generated by the k × k minors of a local matrix rep-
resentation of ϕB are principal for all k ≥ 1, consider
ϕ0 : OP5 (−1) ⊗ S1∨ → S1 ,
the universal symmetric map that gives the matrix of the conic. We are blowing-
up the Veronese, which is the scheme of zeros of the ideal of 2 × 2-minors of ϕ0 .
Therefore, up on B, we have that the ideal of 2×2 minors of ϕ0B is locally principal,
say generated
  by t. Thus we can write the matrix of ϕ0 locally in the form A =
10 0
0 t 0 (cf. [10, Lemma 1]). Let C = Z(z02 + tz12 + tsz22 ) be the associated conic.
0 0 ts
Choosing an appropriate ordering of the basis of Sd ⊗S1∨ and of Sd+1 the matrix
of ϕC can be put in the following form:
⎛ ⎞
2In(d) 0 0 B1 B 3
Ad = ⎝ 0 2tId+1 0 0 B4 ⎠ ,
0 0 2ts 0 0
where all the entries of B1 are multiples of t, and the entries of B3 , B
 4 are
 multiples
of ts. Here Im stands for the identity matrix of size m, and n(d) = d+2 2 . It follows
that the ideals Ji of⎧i × i-minors of Ad are:
⎨ Ji = 1 for i = 1, . . . , n(d);
Jn(d)+j = tj for j = 1, . . . , d + 1;

Jn(d+1) = td+2 s .
In particular these ideals of minors are principal as claimed. 
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 79
9

Lemma 3.2. Let R be a local Noetherian domain, and ϕ : Rn → Rm a ho-


momorphism of free, finitely generated R-modules. Suppose that the ideals k ×
k minors of ϕ are principal for all k. Then J := Imϕ is free.
Proof. Let  a11 ... a1n

A= .. . . ..
. . .
am1 ... amn
be the m × n matrix associated to ϕ, i.e. the columns of A generate J . By
hypotheses for k = 1, the ideal of coefficients of A is principal:
a11 , . . . , aij , . . . , amn = b
We may assume b = 0. Let bij := bij . We may suppose a11 = b. Let J  be the
a

module generated by the columns of 


1 ... b1n
B= .. . . .. .
. . .
bm1 ... bmn

Equivalently (by elementary  1operations)


 J  is isomorphic to the submodule gen-
0
erated by the columns of 0 B  , where for each k, the ideal generated by the
(k − 1) × (k − 1) minors of B  coincides with the ideal generated by the k × k minors
of B.
Applying the inductive hypotheses, we have that ImB  is free. Thus J  is free.
Since R is a domain we have J = bJ   J  and we conclude that J is free. 

Remark 3.3. The variety B is the well-known variety of complete conics (see [9],
[11]). It is constructed to solve the indeterminacies of the map

e1 : P5  Pˇ5
2
sending a conic to the envelope of its tangent lines. Equivalently e1 (A) =∧ A, where
A is the symmetric matrix of the conic. We have that B is equal to the closure of
the graph of e1 in P5 × Pˇ5 .
3.2. A parameter space for foliations with invariant smooth conic.
Consider the projective bundle associated to E cf. (3.1, p. 7), and let q2 : P(E) → PN
denote the projection. Then
Y = q2 (P(E)) ⊂ PN
is a compactification for the parameter space of foliations with an invariant smooth
conic. It’s not difficult to show that q2 is generically injective (cf. Lemma 3.5
below), so the degree of Y is given by the top dimensional Segre class s5 (E). It will
be computed using Bott’s formula.
Theorem 3.4. Notation as above, let Y be the compactification for the parameter
space of 1-dimensional foliations of degree d on P2 with an invariant smooth conic.
Then the degree of Y is given by
1
(d − 1)d(d + 1)(d7 + 25d6 + 231d5 + 795d4 + 1856d3 + 2468d2 + 2256d + 768)
25 5!
and its codimension is equal to 2(d − 1).
Proof. From the definition of E (cf. 3.1) we see that rank(E) = d(d + 2). As
q2 is generically injective it is easy to see that cod Y = 2(d − 1).
80
10 VIVIANA FERRER AND ISRAEL VAINSENCHER

To compute the degree we use Bott’s formula,


  sT (Ep ) ∩ [p]
(3.4) s5 (E) ∩ [B] = 5

T
cT5 (Tp B)
p∈B

(cf. [2]) where the torus T := C acts on B with isolated fixed points. We recall
that each T -equivariant Chern class cTi appearing on the right hand-side is simply
the i-th symmetric function on the weights of the action of T on the fiber of the
vector bundle at the fixed point. The action of T on B is induced by the action of
T on S1 , given by
t · zi = twi zi
for a suitable choice of weights w0 , w1 , w2 ∈ Z. This action induces actions on
2 5
P(S1 ), on P(S2 ) = P5 and on P(Sym2 ∧ S1 ) = P̌ in such a way that the map
e1 : P5  Pˇ5 (see 3.3) is T -equivariant. Thus we obtain an action of T on B =
closure of Graph e1 ⊂ P5 × P̌5 .
It is easy to see that if we choose the weights in such a way that all sums
wi + wj with 0 ≤ i ≤ j ≤ 2
are pairwise distinct, we obtain the following six isolated fixed points in P5 :
z02 , z0 z1 , z0 z2 , z12 , z1 z2 , z22 .
It remains to find the fixed point on the fiber of π : B → P5 over each fixed
point in P5 . If A ∈ {z0 z1 , z0 z2 , z1 z2 }, then π −1 (A) has just one point. For example,
π −1 (z0 z1 ) = (z0 z1 , ž22 ). Here we put ži for the dual basis.
Take A ∈ {z02 , z12 , z22 }, say A = z02 . Recall that the exceptional divisor of the
blowup is E := P(N ) where N = NV P5 stands for the normal bundle. We have an
explicit description for N see [11, Proposition 4.4.]. Notation as in (2.1, p. 3) with
k = 1, n = 2, we have
NV P5 = OP̌2 (2) ⊗ Sym2 (T ∨ ).
The fiber of OP̌2 (2) (resp. Sym2 (T ∨ )) over z0 is the dual space ž02 (resp. z12 , z1 z2 ,
z22 ). Thus we get
π −1 (z02 ) = Ez02 = P(ž02 ⊗ z12 , z1 z2 , z22 ).
We see there are three fixed points in the fiber of each fixed point A ∈ V. For
A = z02 these points are
ž02 ⊗ z12 , ž02 ⊗ z1 z2 , ž02 ⊗ z22 .
Summarizing, we have twelve fixed points in B:
(1) (zi zj , žk2 ) with 0 ≤ i < j ≤ 2; k = i, j; these three lie off E;
(2) (zi2 , ži2 ⊗ zj zk ) with j, k = i; j < k;
(3) (zi2 , ži2 ⊗ zj2 ) with i = j.
Next we compute the fibers of E over each type of fixed point. Suppose that
B ∈ B is a fixed point. The strategy is to take a curve B(t) ∈ B such that
lim B(t) = B
t→0

and such that A(t) := π(B(t)) ∈ P5 is a curve of smooth conics for t = 0. Therefore

A(t) ,
EB will be obtained as the limit of EB(t) = M

A(t) = EB .
lim M
t→0
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 81
11

This enables us to use the well-known space of vector fields of degree d that
leave invariant a smooth conic C = Z(G) (see [3]). This space is
(♠) {Fij ( ∂z
∂G
i
∂j − ∂G
∂zj ∂i ) | Fij ∈ Sd−1 },
modulo multiples of the radial vector field.
We will adopt the following notation: for each subset J := {v0 , . . . , vk } ⊂
{z0 , z1 , z2 } we set Mm (J) = {v0m , v0m−1 v1 , . . . , vkm }, the canonical monomial basis
of Symm (J). We write simply Mm for Mm ({z0 , z1 , z2 }).
Set Xi,j := zi ∂i − zj ∂j . Notice this is a vector of weight 0, since t · zi = twi zi
whereas t · ∂i = t−wi ∂i .
We now describe suitable 1-parameter families of smooth conics abutting each
type of fixed point.
(1) B1 = z0 z1 . We take A(t) = z0 z1 + tz22 ∈ P5 . Using the characterization (♠) we
obtain that the space M
A(t) of vector fields leaving A(t) invariant is given by
{F10 (z1 ∂0 − z0 ∂1 ), F20 (z1 ∂2 − 2tz2 ∂0 ), F21 (z0 ∂2 − 2tz2 ∂1 ) | Fij ∈ Sd−1 }.
Taking limit as t → 0, we find a basis for EB1 :
{F1 X0,1 , F2 ∂2 | F1 ∈ Md−1 , F2 ∈ Md \ {z2d }},
Clearly this basis consists of T -eigenvectors.
(2) B2 = (z02 , ž02 ⊗ z1 z2 ). In this case, we take A(t) = z02 + tz1 z2 . With the same
procedure as above, we obtain the following basis (of eigenvectors) for EB2 :
{F1 z0 ∂1 , F2 z0 ∂2 , F3 X1,2 | F1 , F2 ∈ Md−1 , F3 ∈ Md−1 ({z1 , z2 })}.
(3) B3 = (z02 , ž02 ⊗ z12 ). In this case a curve of smooth conics that approximates B3
is A(t) = z02 + tz12 + t2 z22 . As before, we obtain the following basis of eigenvectors
for EB3 :
{F1 z0 ∂z∂ 1 , F2 ∂2 | F1 ∈ Md−1 , F2 ∈ Md \ {z2d }}.
In order to handle the denominator in Bott’s formula (3.4) we need, for each fixed
point B, a base consisting of eigenvectors of TB B.
If B ∈ E then TB B  Tπ(B) P(S2 ). For example, for B1 = z0 z1 we have
TB1 B  z0 z1 ∨ ⊗ z02 , z0 z2 , z12 , z1 z2 , z22 .
If B ∈ E, then B = (A, [v]) with A ∈ V and v ∈ NA , v = 0. Now
NA
TB B = TA V ⊕ Hom(v , ) ⊕ v
v
as C -modules. For B2 = (z02 , ž02 ⊗ z1 z2 ) we have:
TB2 B = Tz02 V ⊕ ž02 ⊗ z1 z2 ∨ ⊗ ž02 ⊗ z12 , ž02 ⊗ z22 ⊕ ž02 ⊗ z1 z2
where Tz02 V = z02 ∨ ⊗ z0 z1 , z0 z2 . Similarly, for B3 = (z02 , ž02 ⊗ z12 ) we have
TB3 B = Tz02 V ⊕ ž02 ⊗ z12 ∨ ⊗ ž02 ⊗ z1 z2 , ž02 ⊗ z22 ⊕ ž02 ⊗ z12 .
The explicit calculation in Bott’s formula is better left for a script (see § 5).
Finally we use Lemma 3.6 below which enables us to restrict the computation
just for the first sixteen values of d = 2, . . . , 17 and then interpolate the answers
obtained.

82
12 VIVIANA FERRER AND ISRAEL VAINSENCHER

Lemma 3.5. Notation as in Theorem 3.4, for d > 1, the projection q2 : P(E) → PN
is generically injective.
Proof. By Lemma 2.3 it is sufficient to find, for d ≥ 2, a degree d vector field
X with a single invariant conic C and such that C is a reduced point in the fiber
q2−1 (X ) (equivalently, such that d(C,X ) q2 is injective).
For each degree d, it is possible to find such a vector field. For d = 2, it is not
difficult to see that the field
X = (Z2 Z0 − Z12 )∂Z0 − (Z0 Z1 − Z22 )∂Z1 + (Z1 Z2 − Z02 )∂Z2
and the conic C = Z(P ) with P = Z02 − Z12 + Z22 have the required properties.
For d = 3 the pair
X = (Z22 Z0 − Z13 )∂Z0 + (Z0 Z12 − Z23 )∂Z1 + (Z1 Z22 − Z02 Z2 )∂Z2
and C = Z(Z02 + Z12 + Z12 ) does the job.
For d ≥ 4 we will construct, using the example of Jouanolou, a field of degree
d with a unique invariant conic.
Jouanolou proves in [7, p. 157] that the field
Y := Z2e ∂Z0 + Z0e ∂Z1 + Z1e ∂Z2
has no invariant algebraic subset if e ≥ 2 .
Let P be an irreducible polynomial of degree 2. Setting X := P · Y we produce
a field of degree d ≥ 4 that has C = Z(P ) as unique invariant conic (in fact C is in
the singular set of X , but this is sufficient for us). It is not difficult to verify this
and the fact that d(C,X ) q2 is injective.


Lemma 3.6. The sum in the right-hand side of Bott’s formula


  sT (EB ) ∩ [B]
s5 (E) ∩ [B] = 5
,
T
cT5 (TB B)
B∈B

is a combination of wi s with polynomial coefficients in d of degree ≤ 15.


Proof. For each fixed point B let {ξ1 (d), . . . , ξm(d) (d)} denote the set of
weights of the action of T on EB . Since sT5 (EB ) is a polynomial in {cTk (EB ) |
k = 1, . . . , 5} it’s enough to prove that each
cTk (EB ) = σk (ξ1 (d), . . . , ξm(d) (d))
is a combination of wi s with polynomial coefficients in d of degree ≤ 3k.
Recalling Newton’s identities

kσk = ki=1 (−1)i+1 σk−i pi
where m(d)
pk (ξ1 (d), . . . , ξm(d) (d)) := i=1 ξi (d)k
we see that it suffices to prove that pk (ξ1 (d), . . . , ξm(d) (d)) is a combination of wi s
with polynomial coefficients in d of degree ≤ k + 2.
On the other hand, a careful analysis of the weights appearing in the basis of
EB at each fixed point shows that these weights can be separated into sets of the
form
{weights of Me (J)} or {weights of Me (J)} + w
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 83
13

where w is a (fixed) combination of wi s; e = d, d − 1 and J = z0 , z1 , z2 or


J = zi , zj , i = j. From this the reader may convince her(him)self that it’s enough
to prove the following
 
Claim: Let m = m(d, n) := d+n n and {ξn,1 (d), . . . , ξn,m (d)} be the weights
associated to the basis Md ({z0 , . . . , zn }) of Symd (z0 , . . . , zn ). Then
m
pnk (d) := ξn,i (d)k
i=1
is a combination of wi s with polynomial coefficients in d of degree ≤ k +n.
To prove the claim we proceed by induction on n ≥ 1 and on k ≥ 0.
For n = 1,
Md ({z0 , z1 }) = {z0d , z0d−1 z1 , . . . , z0 z1d−1 , z1d }
so that m(d, 1) = d + 1. We have

d+1 
d
p1k (d) = ξ1,i (d)k = (iw0 + (d − i)w1 )k
i=1 i=0

d d  k  
k
= (i(w0 − w1 ) + dw1 ) = k
(i(w0 − w1 ))j (dw1 )k−j
i=0 i=0 j=1
j

k   d
k
= (dw1 )k−j (w0 − w1 )j ij .
j=1
j i=0
d
The sum i=0 ij is polynomial in d of degree j +1, therefore p1k (d) is a combination
of wi s with polynomial coefficients in d of degree ≤ k + 1.
For k = 0, we have pn0 (d) = m(d, n), a polynomial in d of degree n.
For the general case, decompose the basis Md ({z0 , . . . , zn }) as the union
z0 Md−1 ({z0 , . . . , zn }) ∪ z1 Md−1 ({z1 , . . . , zn }) ∪ z2 Md−1 ({z2 , . . . , zn }) ∪ · · · ∪ {znd }.
Then the weights are:
w0 + {ξn,i (d − 1)} ∪ w1 + {ξn−1,i (d − 1)} ∪ w2 + {ξn−2,i (d − 1)} ∪ · · · ∪ {dwn }.
Hence we can write

m(d,n)

m(d−1,n)
pnk (d) = (ξn,i (d))k = (w0 + ξn,i (d − 1))k +
i=1 i=1


m(d−1,n−1)

m(d−1,n−2)
(w1 + ξn−1,i (d − 1))k + (w2 + ξn−2,i (d − 1))k + · · · + (dwn )k
i=1 i=1
k 
  k  
k k
= w0j pnk−j (d − 1) + w1j pn−1
k−j (d − 1) +
j=0
j j=0
j

k  
k
w2j pn−2
k−j (d − 1) + · · · + (dwn ) .
k

j=0
j

By induction we conclude that pnk (d) − pnk (d − 1) is a combination of wi s with


polynomial coefficients in d of degree ≤ k + n − 1, and this implies that pnk (d) is a
combination of wi s with polynomial coefficients in d of degree ≤ k + n.

84
14 VIVIANA FERRER AND ISRAEL VAINSENCHER

4. Higher dimension
The varieties of complete quadrics can also be employed to construct a com-
pactification of the space of 1-dimensional foliations in Pn that leave invariant a
smooth quadric of arbitrary dimension. For example, in the case of conics and
quadrics in P3 we obtain the following.
Theorem 4.1. Let Y1 (resp.Y2 ) denote the closure in PN of the variety of 1-
dimensional foliations in P3 that have an invariant smooth conic (resp. quadric
surface). Then we have the formulas for the degrees and codimensions,
4 
(i) deg Y1 = 2
(d − 1)d 207d14 + 2763d13 + 15447d12 + 54395d11 + 114847d10
8!3
+207891d9 + 256737d8 + 225801d7 + 164937d6 + 182101d5 + 38993d4 +

316221d3 + 248856d2 − 118908d − 332640 ;
codimY1 = 4(d − 1);
1 
(ii) deg Y2 = 9
(d−1)d(d+1) d24 + 81d23 + 3151d22 + 77949d21 + 1369333d20
9!(3!)
+18084843d19 + 185031133d18 + 1481854743d17 + 9251138050d16 +
44737976160d15 + 168507293704d14 + 503603726976d13 + 1212870415960d12 +
2353394912904d11 + 3628929239056d10 + 4249158105672d9 + 3232639214668d8 +
413912636928d7 − 2874493287072d6 − 3885321416832d5 − 1115680433472d4 +

4477695012864d3 + 8264265366528d2 + 8139069775872d + 4334215495680
codim Y2 = (d − 1)(d + 5).

5. Scripts
The formula for invariant k-planes is computed using schubert/maple, [8].
The calculations for Bott’s formula uses Singular [6]. The scripts can be found in
[4].

References
[1] A. Altman, S. Kleiman. Foundations of the Theory of Fano Schemes. Compositio Math. 34
(1977), 3–47.
[2] M. Brion. Equivariant cohomology and equivariant intersection theory, Notes by Alvaro Ritta-
tore. NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 514, Representation theories and algebraic
geometry (Montreal, PQ, 1997), 1–37, Kluwer Acad. Publ., Dordrecht, 1998; arxiv 9802063.
[3] E. Esteves. The Castelnuovo-Mumford regularity of an integral variety of a vector field on
projective space, Math. Res. Lett. 9, no. 1, ( 2002), 1–15.
[4] V. Ferrer, I. Vainsencher. Polynomial vector fields with algebraic trajectories.
http://www.mat.ufmg.br/israel/Publicacoes/Vivis, or arxiv/1003.3997, (2010).
[5] W. Fulton. Intersection Theory. Springer-Verlag. New York. (1985).
[6] G.-M. Greuel, G. Pfister, H. Schönemann. Singular 3-1-1 – A Computer Algebra System for
Polynomial Computations. http://www.singular.uni-kl.de, (2010).
[7] J.P. Jouanolou. Equations de Pfaff algébriques. Lecture Notes in Math., 708. Springer-Verlag,
(1979).
[8] S.Katz, S.A Strømme.schubert. A Maple package for Intersection The-
ory.http://linus.mi.uib.no, (2001).
[9] S. Kleiman, A. Thorup. Complete bilinear forms. Algebraic geometry (Sundance, UT, 1986),
253–320, Lecture Notes in Math., 1311, Springer, Berlin, (1988).
[10] D. Laksov The geometry of complete linear maps. Ark. Mat. 26, no. 2, 231–263, (1988).
POLYNOMIAL VECTOR FIELDS WITH ALGEBRAIC TRAJECTORIES 85
15

[11] I. Vainsencher. Schubert calculus for complete quadrics. Enumerative geometry and classical
algebraic geometry (Nice, 1981), pp. 199–235, Progr. Math.,24, Birkhäuser, Boston, Mass.
(1982).

GAN. Departamento de Matemática, UFF, Rua Mário Santos Braga, s/n, 4 andar,
CEP 24020-140, Niterói, RJ - Brazil.
E-mail address: vivisferrer@gmail.com

ICEX. Departamento de Matemática, UFMG, Avenida Antonio Carlos 6627, CEP


31270-901, Belo Horizonte Brazil.
E-mail address: israel@mat.ufmg.br
This page intentionally left blank
Contemporary Mathematics
Volume 555, 2011

Minimal free resolutions for certain affine monomial curves

Philippe Gimenez, Indranath Sengupta, and Hema Srinivasan

This paper is dedicated to Wolmer V. Vasconcelos.

Abstract. Given an arbitrary field k and an arithmetic sequence of positive


integers m0 < . . . < mn , we consider the affine monomial curve in An+1 k
parameterized by X0 = tm0 , . . . , Xn = tmn . In this paper, we conjecture
that the Betti numbers of its coordinate ring are completely determined by
n and the value of m0 modulo n. We first show that the defining ideal of
the monomial curve can be written as a sum of two determinantal ideals.
Using this fact, we describe the minimal free resolution of the coordinate ring
in the following three cases: when m0 ≡ 1 mod n (determinantal), when
m0 ≡ n mod n (almost determinantal), and when m0 ≡ 2 mod n and n = 4
(Gorenstein of codimension 4).

Introduction
Let k denote an arbitrary field and R be the polynomial ring k[X0 , . . . , Xn ].
Consider the k-algebra homomorphism ϕ : R → k[t] given by ϕ(Xi ) = tmi , i =
0, . . . , n. Then the ideal P := ker ϕ ⊂ R is the defining ideal of the monomial curve
in An+1k given by the parametrization X0 = tm0 , . . . , Xn = tmn . The k-algebra of
the semigroup Γ ⊂ N generated by m0 , . . . , mn is k[Γ] := k[tm0 , . . . , tmn ]  R/P,
which is one-dimensional and P is a perfect ideal of codimension n. It is well known
that P is minimally generated by binomials. Moreover, P is a homogeneous ideal
and k[Γ] is the homogeneous coordinate ring if we give weight mi to the variables
Xi . Henceforth, homogeneous and graded would mean homogeneous and graded
with respect to this weighted graduation.
Assume now that the positive integers m0 , . . . , mn satisfy the following prop-
erties:

1991 Mathematics Subject Classification. Primary 13D02; Secondary 13A02, 13C40.


Key words and phrases. Monomial curves, arithmetic sequences, determinantal ideals, Betti
numbers, minimal free resolutions.
The first author is partially supported by MTM2010-20279-C02-02, Ministerio de Educación
y Ciencia, Spain.
The second author thanks DST, Government of India for financial support for the
project “Computational Commutative Algebra”, reference no. SR/S4/MS: 614/09, and for the
BOYSCAST 2003 Fellowship. This collaboration is the outcome of the second author’s visits to
the University of Missouri, Columbia, USA during Fall 2007 and to the University of Valladolid,
SPAIN for two months in 2009. He thanks both the institutions for their support.
1 2011
c American Mathematical Society

87
88
2 PHILIPPE GIMENEZ, INDRANATH SENGUPTA, AND HEMA SRINIVASAN

(i) gcd (m0 , . . . , mn ) = 1;


(ii) 0 < m0 < · · · < mn and mi = m0 + id for every i ∈ [1, n], i.e., the integers
form an arithmetic progression with common difference d;
(iii) m0 , . . . , mn generate the semigroup Γ := Nmi minimally, where
 0≤i≤n
N = {0, 1, 2, . . .}, i.e., mj ∈
/ Nmi for every i ∈ [0, n].
0≤i≤n; i=j

Definition 1. A sequence of positive intergers (m) = m0 , . . . , mn is called an


arithmetic sequence if it satisfies the conditions (i),(ii) and (iii) above.
Let us write m0 = an + b such that a, b are positive integers and b ∈ [1, n].
Note that b ∈ [1, n] and condition (iii) on m0 , . . . , mn ensures that a ≥ 1; otherwise
m0 = b and mb = m0 + bd = (1 + d)b contradicts minimally condition (iii).
Let (m) = m0 , . . . , mn be an arithmetic sequence. We say that the monomial
curve in An+1
k parameterized by X0 = tm0 , . . . , Xn = tmn is the monomial curve
associated to the arithmetic sequence (m) and denote it by C(m). A minimal
binomial generating set for the defining ideal P of C(m) was given in [P], and it
was rewritten in [MS] to prove that P is not a complete intersection if n ≥ 3. An
explicit formula for the type of k[Γ] is given in [PS, Corollary 6.2] under a more
general assumption of almost arithmetic sequence on the integers (m) = m0 , . . . , mn
and it follows from this result that if (m) = m0 , . . . , mn is an arithmetic sequence
then k[Γ] is Gorenstein if and only if b = 2. In this paper, we prove in Theorem 1.1
that ideal P has the special structure that it can be written as a sum of two
determinantal ideals, one of them being the defining ideal of the rational normal
curve. We exploit this structure to construct an explicit minimal free resolution of
the graded ideal P, in the cases when m0 ≡ 1 or n modulo n, and when m0 ≡ 2
modulo n and n ≤ 4. Note that a minimal free resolution for P had already been
constructed for n = 3 in [S1] using a Gröbner basis for P. In [S2] the following
question was posed : do the total Betti numbers of P depend only on the integer
m0 modulo n ? In this article, we answer this question in affirmative for the above
cases. This question will be addressed in general in our work in progress [GSS].

1. The defining ideal


By [P] and [MS], one knows that the number of elements in a minimal set of
generators of the ideal P depends only on m0 modulo n. Our first result shows
that P has an additional structure that will be helpful in the sequel.
Theorem 1.1. Let (m) = m0 , . . . mn be an arithmetic sequence and P be the
defining ideal of the monomial curve C(m) associated to m. Then
⎛ ⎞ ⎛ ⎞
X0 X1 ··· Xn−1 a
Xn X0 ··· Xn−b
P = I2 (⎝ ⎠)+I2 (⎝ ⎠).
X1 X2 ··· Xn X0a+d Xb ··· Xn

Proof. It is known from [P] and [MS], that P is minimally generated by the
following set of binomials
{δij | 0 ≤ i < j ≤ n − 1} ∪ {Δ1, j | j = 2, . . . , n + 2 − b},
such that
δi, j := Xi Xj+1 − Xj Xi+1 , 0≤i<j ≤n−1
MINIMAL FREE RESOLUTIONS FOR CERTAIN AFFINE MONOMIAL CURVES 89
3

and
Δ1, j := Xb+j−2 Xna − Xj−2 X0a+d , j = 2, . . . , n + 2 − b.
 
The binomials δij are precisely the n2 generators of the ideal of 2 × 2 minors of
the matrix  
X0 X1 ··· Xn−1
A= .
X1 X2 ··· Xn
On the other hand, the binomials Δ1,j are the 2 × 2 minors from the first and the
jth column of the matrix
 
Xna X0 ··· Xn−b
B= .
X0a+d Xb ··· Xn

Since the rest of the 2 × 2 minors of B are already in the ideal I2 (A), one gets
that P = P1 + P2 , where P1 and P2 are the determinatal ideals generated by the
maximal minors of the matrices A and B respectively. 

The actual generators also depend only on the first term m0 , the common
difference d and the length n of the arithmetic sequence (m). This is no surprise as
these three numbers do determine the arithmetic sequence.  However, we also have
that the number of minimal generators of the ideal P is n2 + n − b + 1 and hence
only depends on n and b. We conjecture that the following statement is true:
Conjecture 1.2. Let (m) = m0 , . . . , mn be an arithmetic sequence. Then all
the Betti numbers of the homogeneous coordinate ring k[Γ] of the affine monomial
curve C(m) in An+1
k associated to (m) are determined by n and the value of m0
modulo n.
In the next section, we discuss the conjecture in cases b = 1, 2 and n and
construct a minimal resolution when b = 1 or n using the aforesaid determinantal
structures. In the first special case, b = 1, we have that P is a determinantal ideal.
The second special case is b = n, that is when
 
Xna X0
B := ,
X0a+d Xn

and therefore the ideal P2 is the principal ideal generated by X0a+d+1 − Xna+1 . The
third special case is b = 2, that is when k[Γ] is Gorenstein. In this case, we wiil
give the Betti numbers when the codimension n is 4.

2. The minimal resolution


2.1. First case: determinantal (m0 ≡ 1 modulo n). Assume that b = 1.
In this case,
 
Xna X0 ··· Xn−1
B :=
X0a+d X1 ··· Xn
and hence I2 (A) ⊂ I2 (B). Thus, P = P2 is a determinantal ideal of codimension
n generated by the maximal minors of the 2 × (n + 1) matrix B. Therefore, the
homogeneous coordinate ring k[Γ] is minimally resolved by the Eagon-Northcott
complex. The resolution is given by
0 → ∧n+1 Rn+1 ⊗ Dn−1 (R2 ) → · · · → ∧3 Rn+1 ⊗ D1 (R2 ) → ∧2 Rn+1 → ∧2 R2 → k[Γ] → 0
90
4 PHILIPPE GIMENEZ, INDRANATH SENGUPTA, AND HEMA SRINIVASAN

such that Ds−1 (R2 ) = (Ss−1 (R2 ))∗ denotes the R-module which is the dual of the
symmetric algebra of R2 and the module Ss−1 (R2 ) is free of rank s, with a basis
the set of monomials of total degree (s − 1) in λ0 , λ1 (symbols representing basis
elements of R2 ). Therefore, up to an identification, Ds−1 (R2 ) = (Ss−1 (R2 ))∗ is a
free R-module of rank s, with a basis {λv00 λv11 | v0 , v1 ∈ N , v0 + v1 = s − 1}. For
every s = 1, . . . , n, let Gs denote ∧s+1 Rn+1 ⊗ Ds−1 (R2 ), which is a R-free module
s+1 generated by the basis elements (ei1 ∧ · · · ∧ eis+1 ) ⊗ λ0 λ1 , for every
of rank s n+1 v0 v1

1 ≤ i1 < i2 < · · · < is+1 ≤ n + 1 and v0 , v1 ∈ N , v0 + v1 = s − 1. The differentials


ds : Gs → Gs−1 are given by the following formulae:

d1 (ei1 ∧ ei2 ) = Xi1 −1 Xi2 − Xi2 −1 Xi1 , ∀ 1 ≤ i1 < i2 ≤ n + 1,

and for s = 2, . . . , n and 1 ≤ i1 < · · · < is+1 ≤ n + 1,


s+1
v0 −1 v1
ds ((ei1 ∧ · · · ∧ eis+1 ) ⊗ λv00 λv11 ) = (−1)j+1 Xij −1 ((ei1 ∧ · · · ∧ e
ij ∧ · · · eis+1 ) ⊗ λ0 λ1 )
j=1


s+1
v0 v1 −1
+ (−1)j+1 Xij ((ei1 ∧ · · · ∧ e
ij ∧ · · · eis+1 ) ⊗ λ0 λ1 )
j=1

such that summands on the right hand side involve only non-negative powers of
λ0 and λ1 .

We have therefore proved the following:

Theorem 2.1. The minimal free resolution of the homogeneous coordinate ring
k[Γ] of the affine monomial curve C(m) in An+1 k associated to the arithmetic se-
quence of integers (m) = m0 , . . . , mn with the property m0 ≡ 1 modulo n is

n+1 n+1 n+1


0 → Rn −→
n
R(n−1)(
d
n ) d−→
n−1
· · · −→ Rs( s+1 ) −→
ds
· · · −→ R( 2 ) −→
d1 ϕ
R −→ k[Γ] → 0.

n+1
In particular, the Betti numbers are β0 = 1 and βs = s s+1 for every s ∈ [1, n].

Remark 2.2. The case a = 1 (and b = 1) corresponds to semigroups Γ of


maximal embedding dimension, i.e., those semigroups such that the inequality
m(Γ) ≥ e(Γ) is an equality, being m(Γ) := m0 = an + b and e(Γ) := n + 1
the multiplicity and the embedding dimension respectively.

Note that the above argument gives indeed the graded resolution (with respect
to the grading given by deg(Xi ) = mi = m0 + id):

Corollary 2.3. Under the hypothesis of Theorem 2.1, the minimal graded free
resolution of k[Γ] is given by
MINIMAL FREE RESOLUTIONS FOR CERTAIN AFFINE MONOMIAL CURVES 91
5

 

n
n+1
0 −→ R(−(a + n + d)m0 + kd − d)
2
k=1
n−1   
d
 n+1
−→
n
R(−nm0 + kd − d)
2
k=1
⎛ n−1 ⎞
  
n−1
⊕⎝ R(−(a + d + n − 1)m0 + kd − ri d) ⎠
1≤r1 <...rn−1 ≤n k=1 i=1

dn−1
−→ ···
..
. ⎛  s−1 ⎞
  
s
···
ds
−→ ⎝ R(−sm0 + kd − ri d) ⎠ ⊕
1≤r1 <...<rs ≤n k=1 i=1
⎛  s−1 ⎞
  
s−1
⎝ R(−(a + s − 1 + d)m0 + kd − ri d) ⎠
1≤r1 <...<rs−1 ≤n k=1 i=1

ds−1
−→ ···
..
. ⎛ ⎞ n−1 
 
···
d2
−→ ⎝ R(−2m0 − (r1 + r2 − 1)d)⎠ ⊕ R(−(a + 1 + d)m0 − kd)
1≤r1 <r2 ≤n k=0

d ϕ
−→
1
R −→ k[Γ] → 0 .

2.2. Second case: almost determinantal (m0 ≡ n modulo n). In this


case, we show that k[Γ] can be resolved minimally by a mapping cone using the
resolution for R/P1 , which is the homogeneous coordinate ring of the rational nor-
mal curve. The minimal free resolution for R/P1 is the Eagon-Northcott complex
given by
E : 0 −→ ∧n Rn ⊗ (Sn−2 (R2 ))∗ −→ · · · −→ ∧2 Rn ⊗ (S0 (R2 ))∗ −→ ∧2 R2 −→ R/P1 −→ 0.
n
Since Es := ∧s+1 Rn ⊗ (Ss−1 (R2 ))∗ = Rs(s+1) , for every s = 1, . . . , n − 1, E0 :=
∧2 R2 = R, the resolution E takes the form
n n n
0 −→ Rn−1 −→ R(n−2)(n−1) −→ · · · −→ Rs(s+1) −→ · · · −→ R( 2 ) −→ R −→ R/P1 −→ 0.
The differentials d1s : Es → Es−1 are defined similarly as ds above, for every
s = 1, . . . , n − 1.
Now consider the short exact sequence of R-modules
Δ1, 2
0 −→ R/(P1 : Δ1, 2 ) −→ R/P1 −→ R/P1 + (Δ1, 2 ) −→ 0
Δ1, 2
where the map R/(P1 : Δ1, 2 ) −→ R/P1 is the multiplication by Δ1,2 . The colon
ideal (P1 : Δ1, 2 ) is exactly equal to P1 , since P1 is a prime ideal and Δ1, 2 ∈ P1 ,
being a part of a minimal generating set for the ideal P = P1 + P2 . Therefore, the
above short exact sequence becomes,
Δ1, 2
0 −→ R/P1 (−(a + d + 1)m0 ) −→ R/P1 −→ R/P1 + (Δ1, 2 ),
92
6 PHILIPPE GIMENEZ, INDRANATH SENGUPTA, AND HEMA SRINIVASAN

taking gradation into account. We define the graded complex homomorphism

ψ : E(−(a + d + 1)m0 ) −→ E

as the multiplication by Δ1, 2 , which is a lift of the map

Δ1, 2
R/P1 (−(a + d + 1)m0 ) −→ R/P1 .

In fact, the complex homomorphism ψ is simply multiplication by Δ1, 2 . The map-


n n
ping cone F(ψ) is given by the free modules Fs = Es−1 ⊕ Es = R(s−1)( s ) ⊕ Rs(s+1) ,
for every s = 0, . . . , n (with E−1 = En = 0) and the differentials d2s : Fs → Fs−1 ,
defined as d2s (x, y) = (ψs−1 (y) + d1s (x), −d1s−1 (y)), for every s = 1, . . . , n.

R/P1 +(Δ1, 2 ) is resolved by the mapping cone F(ψ). It is minimal because each
of the maps in ψ is of positive degree and the resolution E is minimal. Hence, the
mapping cone F(ψ) is the minimal free resolution of the homogeneous coordinate
ring k[Γ] = R/P. We have therefore proved the following:

Theorem 2.4. The minimal free resolution of the homogeneous coordinate ring
k[Γ] of the affine monomial curve C(m) in An+1 k associated to the arithmetic se-
quence of integers (m) = m0 , . . . , mn with the property m0 ≡ n modulo n is

d2 n d2n−1 n d2 ϕ
0 → Rn−1 −→
n
R(n−2)(n−1)+(n−1) −→ · · · −→ R1+( 2 ) −→
1
R −→ k[Γ] → 0.

n n  n

In particular, the Betti numbers are β0 = 1, β1 = 1+ 2 and βs = (s−1) s +s s+1
for every s ∈ [2, n].

As in Section 2.1, one can be more precise and give the graded resolution. First
note that the graded minimal resolution of P1 is given by


n−1 
n+1 d1n−1
0 −→ R(−nm0 + kd − d) −→ · · ·
2
k=1
 s−1 
d1s    d1s−1
··· −→ R(−sm0 + kd − ri d) −→ ···
1≤r1 <...<rs ≤n k=1
d1  d1
··· −→
2
R(−2m0 − (r1 + r2 − 1)d) −→
1
R −→ R/P1 → 0 .
1≤r1 <r2 ≤n
MINIMAL FREE RESOLUTIONS FOR CERTAIN AFFINE MONOMIAL CURVES 93
7

Corollary 2.5. Under the hypothesis of Theorem 2.4, the minimal graded free
resolution of k[Γ] is given by


n−1 
n+1
0 −→ R(−(a + n + d + 1)m0 + kd − d)
2
k=1
n−1  
d2  n+1
−→
n
R(−nm0 + kd − d) ⊕
2
k=1
⎛ n−2 ⎞
  
⎝ R(−(a + n + d)m0 + kd − ri d) ⎠
1≤r1 <...<rn−1 ≤n k=1

d2n−1
−→ ···
..
. ⎛  s−1 ⎞
d2s   
··· −→ ⎝ R(−sm0 + kd − ri d) ⎠ ⊕
1≤r1 <...<rs ≤n k=1
⎛  s−2 ⎞
  
⎝ R(−(a + s + d)m0 + kd − ri d) ⎠
1≤r1 <...<rs−1 ≤n k=1

d2s−1
−→ ···
..
. ⎛ ⎞
d22 
··· −→ ⎝ R(−2m0 − (r1 + r2 − 1)d)⎠ ⊕ R(−(a + 1 + d)m0 )
1≤r1 <r2 ≤n

d21 ϕ
−→ R → k[Γ] → 0.

2.3. Third case: Gorenstein of codimension 4 (m0 ≡ 2 modulo n and


n = 4). When b = 2, the ideal P is Gorenstein of height n and hence the first
and the last Betti numbers are known. This provides a very interesting family of
Gorenstein ideals of height n generated by (n − 1)(n + 2)/2 binomials.

When n = 2, P is a complete intersection and when n = 3, P is the ideal


of 4 × 4 pfaffians of a 5 × 5 skew symmetric matrix. For n = 4, the ideal P is
a height 4 Gorenstein ideal minimally generated by 9 elements and therefore has
Betti numbers 9, 16, 9 and 1.

In fact, we can show that with this weighted grading, the shifts in a graded
resolution of R/P are as follows:

Theorem 2.6. Given two non-negative integers a and d, consider the monomial
curve C(m) in A5k associated to the arithmetic sequence (m) = 4a + 2, 4a + 2 +
d, 4a + 2 + 2d, 4a + 2 + 3d, 4a + 2 + 4d. Set q := 2a + 1. Then the defining ideal P
of C(m) is a height four Gorenstein ideal whose minimal graded free resolution is
94
8 PHILIPPE GIMENEZ, INDRANATH SENGUPTA, AND HEMA SRINIVASAN

0 → R(−q(q + 2d + 9) − 9d)
 9   7 
 
→ R(−8q − kd) ⊕ R(−q(q + 2d + 5) − kd) ⊕ R(−q(q + 2d + 5) − 5d)
k=7 k=3
 

7
→ R(−6q − 4d) ⊕ R (−6q − kd)
2
⊕ R(−6q − 8d) ⊕ R(−q(q + 2d + 3) − d)
k=5
 

4
⊕ R (−q(q + 2d + 3) − kd)
2
⊕ R(−q(q + 2d + 3) − 5d)
k=2
 6   
 
2
→ R(−4q − kd) ⊕ R(−4q − 4d) ⊕ R(−q(q + 2d + 1) − kd) → P → 0.
k=2 k=0

Proof. We know that P is Gorenstein of height 4 and by Theorem 1.1, it is


generated by the 2 × 2 minors of A and B, where
 
X0 X1 X2 X3
A=
X1 X2 X3 X4

and  
X4a X0 X1 X2
B= .
X0a+d X2 X3 X4
Moreover, the 9 minimal generators of P are the six 2×2 minors of A whose degrees
are 2m0 + kd, k = 2, 3, 4, 4, 5, 6 for m0 := 4a + 2 = 2(2a + 1) = 2q, and the 3 minors
of B involving the first column which are of degrees m0 (a + d + 1) + kd, k = 0, 1, 2.
Note that for both determinantal ideals I2 (A) and I2 (B), the Eagon-Northcott
complex provides a minimal resolution. The 8 determinantal relations from the
Eagon-Northcott resolution of I2 (A) and the 6 determinantal relations from the
Eagon-Northcott resolution of I2 (B) involving the first column will necessarily be
among the 16 minimal syzygies of P. The degrees of the other two relations are
determined by the symmetry of the resolution. Thus the remaining two relations
must be of degrees q(q + 3) + (2q + 2)d and q(q + 3) + (2q + 4)d. This determines
the rest of the degrees in the resolution. 

We note that when d = 1, this ideal can never be determinantal even though
it is height 4 and generated by 9 elements. That is, this ideal is not the ideal of
2 × 2 minors of a 3 × 3 matrix for sum of the degrees of such an ideal must be
even. For in this case, the ideal is generated by 9 elements whose degrees add up
to 27d + km0 . Since m0 is even, this number is odd whenever d is odd. Hence if
d is odd, it cannot be a determinantal ideal. However, it is not clear if it can be
determinantal for even values of d. Of course, the conjecture only says that the
Betti numbers are determined by b.

References
[GSS] P. Gimenez, I. Sengupta and H. Srinivasan, Minimal graded free resolution for monomial
curves defined by an arithmetic sequence (2011), in progress.
[MS] A. K. Maloo and I. Sengupta, Criterion for complete intersection of certain monomial curves,
in: Advances in Algebra and Geometry (Hyderabad, 2001), 179–184, Hindustan Book Agency,
New Delhi, 2003.
MINIMAL FREE RESOLUTIONS FOR CERTAIN AFFINE MONOMIAL CURVES 95
9

[P] D. P. Patil, Minimal sets of generators for the relation ideals of certain monomial curves,
Manuscripta Math. 80 (1993) 239–248.
[PS] D. P. Patil and I. Sengupta, Minimal set of generators for the derivation module of certain
monomial curves, Comm. Algebra 27(11) (1999) 5619–5631.
[S1] I. Sengupta, A minimal free resolution for certain monomial curves in A4 , Comm. Algebra
31(6) (2003) 2191–2809.
[S2] I. Sengupta, Betti numbers of certain affine monomial curves, In EACA-2006 (Sevilla), F.-J.
Castro Jiménez and J.-M. Ucha Enrı́quez Eds, 171–173. ISBN: 84-611-2311-5.

Department of Algebra, Geometry and Topology, Faculty of Sciences, University


of Valladolid, 47005 Valladolid, Spain.
E-mail address: pgimenez@agt.uva.es

School of Mathematical Sciences, RKM Vivekananda University, Belur, India.


Current address: Department of Mathematics, Jadavpur University, Kolkata, WB 700 032,
India.
E-mail address: sengupta.indranath@gmail.com

Mathematics Department, University of Missouri, Columbia, MO 65211, USA.


E-mail address: SrinivasanH@missouri.edu
This page intentionally left blank
Contemporary Mathematics
Volume 555, 2011

Uniform bounds for Hilbert coefficients of parameters

Shiro Goto and Kazuho Ozeki

Abstract. Let A be a Noetherian local ring with d = dim A > 0. This paper
shows that the Hilbert coefficients {eiQ (A)}1≤i≤d of parameter ideals Q have
uniform bounds if and only if A is a generalized Cohen-Macaulay ring. The
uniform bounds are huge; the sharp bound for e2Q (A) in the case where A is a
generalized Cohen-Macaulay ring with dim A ≥ 3 is given.

1. Introduction
The purpose of this paper is to study the problem of when the Hilbert coef-
ficients of parameter ideals in a Noetherian local ring have uniform bounds, and
when this is the case, to ask for their sharp bounds.
Let A be a Noetherian local ring with maximal ideal m and d = dim A > 0. For
simplicity, we assume that the residue class field A/m of A is infinite. Let A (M )
denote, for an A-module M , the length of M . Then for each m-primary ideal I in
A, we have integers {eiI (A)}0≤i≤d such that the equality
   
n+d n+d−1
A (A/I n+1 ) = e0I (A) − e1I (A) + · · · + (−1)d edI (A)
d d−1
holds true for all n  0, which we call the Hilbert coefficients of A with respect
to I. We say that our local ring is a generalized Cohen-Macaulay ring, if the local
cohomology modules Him (A) are finitely generated for all i = d. Let
hi (A) = A (Him (A))
for each i ∈ Z.
With this notation our first purpose is to study the problem of when the set
Λi (A) = {eiQ (A) | Q is a parameter ideal in A}
is finite for all 1 ≤ i ≤ d. Our research has started from the joint work [GhGHOPV],
where the authors intensively studied the problem of when the set Λ1 (A) is finite
and proved that A is a generalized Cohen-Macaulay ring, once the set Λ1 (A) is

1991 Mathematics Subject Classification. Primary 13D40; Secondary 13H10, 13H15.


Key words and phrases. Hilbert function, Hilbert coefficient, parameter ideal, associated
graded ring, Rees algebra.
The first author is partially supported by Grant-in-Aid for Scientific Researches (C) in Japan
(22540054). The second author is supported by a grant from MIMS (Meiji Institute for Advanced
Study of Mathematical Sciences) and Grant-in-Aid for Young Scientists (B) in Japan (22740026).
1 2011
c American Mathematical Society

97
98
2 SHIRO GOTO AND KAZUHO OZEKI

finite and A is unmixed ([GhGHOPV, Proposition 4.2]). The present research is


a continuation of [GhGHOPV] and the first main result is stated as follows.
Theorem 1.1. Let A be a Noetherian local ring with d = dim A ≥ 2. Then the
following conditions are equivalent.
(1) A is a generalized Cohen-Macaulay ring.
(2) The set Λi (A) is finite for all 1 ≤ i ≤ d.
We shall prove Theorem 1.1 in Section 2. Let us note here a few comments
about the proof. The proof of the implication (2) ⇒ (1) is based on [GhGHOPV,
Proposition 4.2] but the heart of the proof of the implication (1) ⇒ (2) is, in
the case where A is a generalized Cohen-Macaulay ring, the existence of uniform
bounds of the Castelnuovo-Mumford regularity reg G(Q) of the associated graded
rings G(Q) of parameter ideals Q. So, let us briefly recall the definition of the
Castelnuovo-Mumford regularity. Let Q be a parameter ideal in A and let
R(Q) = A[Qt], R (Q) = A[Qt, t−1 ], and G(Q) = R (Q)/t−1 R (Q)
respectively denote the Rees algebra, the extended Rees algebra, and the associated
graded ring of Q, where t is an indeterminate over A. Let M = mR + R+ be the
unique graded maximal ideal in R = R(Q). For each i ∈ Z let
ai (G(Q)) = sup{n ∈ Z | [HiM (G(Q))]n = (0)}
and put
regG(Q) = sup{ai (G(Q)) + i | i ∈ Z},
which we call the Castelnuovo-Mumford regularity of the graded ring G(Q). Then,
if A is a generalized Cohen-Macaulay ring, we have that
(1) |e1Q (A)| ≤ I(A) and
(2) |eiQ (A)| ≤ 3 · 2i−2 (r + 1)i−1 I(A) for all 2 ≤ i ≤ d
for every parameter ideals Q in A (see Theorem 2.2), where
  d − 1
d−1
r = reg G(Q) and I(A) = hj (A).
j=0
j

Therefore, thanks to the uniform bounds [LT, Theorem 2.3] of reg G(Q) for param-
eter ideals Q in a generalized Cohen-Macaulay ring A, we readily get the finiteness
in the set Λi (A) for all 1 ≤ i ≤ d.
Although the finiteness problem of Λi (A) is settled affirmatively, both the
bounds in Theorem 2.2 and [LT, Theorem 2.3] are given in the exponential or-
der and very huge. It seems to be a different problem to ask for the sharp bounds
for the values of eiQ (A) of parameter ideals Q, which is our second purpose of the
present research.
When A is a generalized Cohen-Macaulay ring with d = dim A ≥ 2, one has
the inequalities
 d − 2
d−1
0 ≥ e1Q (A) ≥ − hj (A)
j=1
j − 1
for every parameter ideal Q in A ([MSV, Theorem 3.6], [GNi, Lemma 2.4]), where
d−1   j
the equality e1Q (A) = − j=1 d−2j−1 h (A) holds true if and only if Q is a standard
parameter ideal in A ([S, Korollar 3.2], [GO, Theorem 2.1]), provided depth A > 0.
The reader may consult [GhGHOPV] for the characterization of local rings which
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 99
3

contain parameter ideals Q with e1Q (A) = 0. Thus the behavior of the first Hilbert
coefficients e1Q (A) for parameter ideals Q are rather satisfactorily understood.
The second purpose is to study the natural question of how about e2Q (A). First,
we will show that in the case where dim A = 2 and depth A > 0, even though A is
not necessarily a generalized Cohen-Macaulay ring, the inequality
−h1 (A) ≤ e2Q (A) ≤ 0
holds true for every parameter ideal Q in A. We will also show that e2Q (A) = 0 if
and only if the ideal Q is generated by a system a, b of parameters which forms a
d-sequence in A. When A is a generalized Cohen-Macaulay ring with dim A ≥ 3,
we shall show that the inequality
d−1 
    d − 3
d−2
d−3 j
− h (A) ≤ e2Q (A) ≤ hj (A)
j=2
j − 2 j=1
j − 1

holds true for every parameter ideal Q (Proposition 3.6). The following theorem
which is the second main result of this paper shows that the upper bound
 d − 3
d−2
e2Q (A) ≤ hj (A)
j=1
j − 1
d−2   j
is sharp, clarifying when the equality e2Q (A) = j=1 d−3j−1 h (A) holds true.

Theorem 1.2. Suppose that A is a generalized Cohen-Macaulay ring with d =


dim A ≥ 3 and depth A > 0. Let Q be a parameter ideal in A. Then the following
two conditions are equivalent.
d−2   j
(1) e2Q (A) = j=1 d−3 j−1 h (A).
(2) There exist elements a1 , a2 , · · · , ad ∈ A such that
(a) Q = (a1 , a2 , · · · , ad ),
(b) the sequence a1 , a2 , · · · , ad is a d-sequence in A, and
(c) Q·Hjm (A/(a1 , a2 , · · · , ak )) = (0) for all j ≥ 1 and k ≥ 0 with j + k ≤
d − 2.
When this is the case, we furthermore have the following :
d−i   j
(i) (−1)i eiQ (A) = j=1 d−i−1 j−1 h (A) for 3 ≤ i ≤ d − 1 and
(ii) edQ (A) = 0.
At this moment we do not know the sharp uniform bound for e3Q (A) for pa-
rameter ideals Q in a generalized Cohen-Macaulay ring A with dim A ≥ 3.
Let us briefly note how this paper is organized. We shall prove Theorem 1.1 in
Section 2. Theorem 1.2 will be proven in Section 4. Section 3 is devoted to some
preliminary steps for the proof of Theorem 1.2. We will closely study in Section
3 the problem of when e2Q (A) = 0 in the case where dim A = 2. We shall prove
d−2 d−3 j
Theorem 1.2 also in Section 3. The equality e2Q (A) = j=1 j−1 h (A) holds,
even if the parameter ideal Q is not standard; in Section 4 we will construct a
non-standard parameter ideal Q which satisfies condition (1) in Theorem 1.2.
In what follows, unless otherwise specified, let A be a Noetherian local ring
with maximal ideal m and d = dim A > 0. We throughout assume that the field
A/m is infinite. For each m-primary ideal I in A, we put
R(I) = A[It], R (I) = A[It, t−1 ], and G(I) = R (I)/t−1 R (I),
100
4 SHIRO GOTO AND KAZUHO OZEKI

where t is an indeterminate over A. Let M = mR + R+ be the unique graded


maximal ideal in R = R(I). We denote by HiM (∗) (i ∈ Z) the ith local cohomology
functor of R(I) with respect to M. Let L be a graded R-module. For each n ∈ Z
let [HiM (L)]n stand for the homogeneous component of HiM (L) with degree n. We
denote by L(α), for each α ∈ Z, the graded R-module whose grading is given by
[L(α)]n = Lα+n for all n ∈ Z.

2. Proof of Theorem 1.1


Let A be a Noetherian local ring with maximal ideal m, d = dim A > 0, and
infinite residue class field A/m.
The purpose of this section is to prove Theorem 1.1. To begin with, let us
note the following result of Linh and Trung [LT], which gives a uniform bound for
reg G(Q) for parameter ideals Q in a generalized Cohen-Macaulay ring.
Theorem 2.1 ([LT], Theorem 2.3). Suppose that A is a generalized Cohen-
Macaulay ring and let Q be a parameter ideal in A. Then
(1) reg G(Q) ≤ max{I(A) − 1, 0}, if d = 1.
(2) reg G(Q) ≤ max{(4I(A))(d−1)! − I(A) − 1, 0}, if d ≥ 2.
The following result is the key for our proof of the implication (1) ⇒ (2) in
Theorem 1.1.
Theorem 2.2. Suppose that A is a generalized Cohen-Macaulay ring. Let Q
be a parameter ideal in A and put r = reg G(Q). Then
(1) |e1Q (A)| ≤ I(A).
(2) |eiQ (A)| ≤ 3 · 2i−2 (r + 1)i−1 I(A) for 2 ≤ i ≤ d.
To prove Theorem 2.2 we need the following lemma, which is fairy well-known.
However, since we cannot find good references, let us include a sketch of proof for
completeness.
Lemma 2.3. Let Q be a parameter ideal in A.
(1) Let C = A/H0m (A). Then reg G(Q) ≥ reg G(QC).
(2) Assume that d ≥ 2 and suppose that a ∈ Q is superficial with respect to
Q. Let Q = Q/(a) in A = A/(a). Then reg G(Q) ≥ reg G(Q).
Proof. (1) This follows from the fact that the kernel of the canonical epimor-
phism G(Q) → G(QC) of graded rings is finitely graded.
(2) Let f = at and we look at the canonical epimorphism
ϕ
G(Q)/f G(Q) → G(Q)
of graded rings. Then the kernel of ϕ is finitely graded, because a is superficial
with respect to Q, whence we have
reg(G(Q)/f G(Q)) ≥ reg G(Q),
where

ai (G(Q)/f G(Q)) = sup n ∈ Z | [HiM (G(Q)/f G(Q)]n = (0)
and
reg(G(Q)/f G(Q)) = sup {ai (G(Q)/f G(Q)) + i | i ∈ Z} .
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 101
5

On the other hand, thanks to the exact sequence


f
0 → [(0) :G(Q) f ](−1) → H0M (G(Q))(−1) → H0M (G(Q)) → H0M (G(Q)/f G(Q)) → · · ·
f
· · · → Hi−1
M (G(Q)/f G(Q)) → HM (G(Q))(−1) → HM (G(Q)) → HM (G(Q)/f G(Q))
i i i

of local cohomology modules induced from the exact sequence


f
0 → [(0) :G(Q) f ](−1) → G(Q)(−1) → G(Q) → G(Q)/f G(Q) → 0,
we have
max {ai (G(Q)), ai+1 (G(Q)) + 1} ≥ ai (G(Q)/f G(Q))
for all i ∈ Z. Hence
reg G(Q) ≥ reg(G(Q)/f G(Q)) ≥ reg G(Q)
as required. 
Proof of Theorem 2.2. We proceed by induction on d. Thanks to [GNi,
d−1   i
Lemma 2.4], we have e1Q (A) = −h0 (A), if d = 1, and e1Q (A) ≥ − i=1 d−2 i−1 h (A),
if d ≥ 2. Thus |e1Q (A)| ≤ I(A), whence our assertion holds true for the case where
d = 1.
We may assume that d ≥ 2 and that our assertion holds true for d − 1. We may
also assume that depth A > 0. In fact, let W = H0m (A), C = A/W , and assume
that
|eiQC (C)| ≤ 3 · 2i−2 (r + 1)i−1 I(C)
for 2 ≤ i ≤ d, where r = regG(QC). We then have

|eiQ (A)| ≤ |eiQC (C)| + h0 (A)


≤ 3 · 2i−2 (r + 1)i−1 I(C) + h0 (A)

d−1 
 d − 1
≤ 3 · 2 (r + 1)
i−2 i−1
h (C) + h0 (A)
i

i=1
i

d−1 
 d − 1
= 3 · 2 (r + 1)
i−2 i−1
h (A) + h0 (A)
i

i=1
i

d−1 
 d − 1
≤ 3 · 2 (r + 1)
i−2 i−1 i
h (A)
i=0
i
≤ 3 · 2i−2 (r + 1)i−1 I(A),
because r ≤ r by Lemma 2.3 (1), hi (A) = hi (C) for i ≥ 1, and 3·2i−2 (r +1)i−1 ≥ 1.
Thus, passing to the ring C, we may assume that depthA > 0.
Let a ∈ Q be a superficial element with respect to Q. We put A = A/(a),
Q = QA, and r  = reg G(Q). Let k ≥ 0 be an integer. Then, thanks to the exact
sequence
a k+1
0 → [Qk+1 : a]/Qk → A/Qk → A/Qk+1 → A/Q → 0,
we get
A (Qk /Qk+1 ) = A (A/Qk+1 ) − A (A/Qk )
k+1
= A (A/Q ) − A ([Qk+1 : a]/Qk ). (∗)
102
6 SHIRO GOTO AND KAZUHO OZEKI

Hence


n
A (A/Qn+1 ) = A (Qk /Qk+1 )
k=0
n
k+1 
n
= A (A/Q )− A ([Qk+1 : a]/Qk ) (∗∗)
k=0 k=0

for all n ≥ 0.

Claim 1. Qk+1 : a = Qk for all k > r.

Proof of Claim 1. We have by a classical theorem of Serre (see [BH, The-


orem 4.4.3]) that


d−1  
k k+1 i i k+d−1−i
A (Q /Q )= (−1) eQ (A)
i=0
d−1−i

for all k > r and


d−1  
k+1 k+d−1−i
A (A/Q )= (−1)i eiQ (A)
i=0
d−1−i

k+1
for all k ≥ r . Hence A (Qk /Qk+1 ) = A (A/Q ) for all k > r, because r ≥ r  by
Lemma 2.3 (2) and eQ (A) = eQ (A) for 0 ≤ i ≤ d − 1. Thus the equality (∗) shows
i i

that A ([Qk+1 : a]/Qk ) = 0 for all k > r. 

Since


d  
n+d−i
A (A/Qn+1 ) = (−1)i eiQ (A)
i=0
d−i

for all n ≥ r (cf. [BH, Theorem 4.4.3]), we get


d−1  
n+d−i
(−1)d edQ (A) = A (A/Q n+1
)− i i
(−1) eQ (A)
i=0
d−i
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 103
7

for all n ≥ r. Hence by the equality (∗∗) we see

(−1)d edQ (A)



n d−1  
 k+1 n  n+d−i
= A (A/Q )− A ([Q k+1
: a]/Q ) −
k
(−1)i eiQ (A)
i=0
d−i
k=0 k=0
n n  n 
d−1  
k+1 k+d−1−i
= A (A/Q )− A ([Q k+1
: a]/Q ) −
k i i
(−1) eQ (A)
d−1−i
k=0 k=0 k=0 i=0

 
n
k+1 
d−1
k+d−1−i n
= A (A/Q )− (−1)i eiQ (A) − A ([Qk+1 : a]/Qk )
i=0
d − 1 − i
k=0 k=0
r

  
 k+1 
d−1
k+d−1−i
r
= A (A/Q )− (−1)i eiQ (A) − A ([Qk+1 : a]/Qk )
i=0
d − 1 − i
k=0 k=0
r

   r 
d−1   
k+1 k+d−1 k+d−1−i
= A (A/Q ) − eQ (A)
0
− i i
(−1) eQ (A)
d−1 i=1
d−1−i
k=0 k=0

r
− A ([Qk+1 : a]/Qk ),
k=0

k+1 d−1  
because A (A/Q )= i=0 (−1) eQ (A) k+d−1−i
i i
d−1−i for all k ≥ r  and [Qk+1 : a] =
Qk for all k > r by Claim 1. On the other hand, thanks to [LT, Lemma 1.1] and
[HH, Theorem 1.1], we have

   
k+1 k+d−1 k+d−2
0 ≤ A (A/Q )− e0Q (A) ≤ I(A)
d−1 d−2

for all k ≥ 0, and thanks to [LT, Corollary 1.4] we have

 
k+d−2
A ([Qk+1 : a]/Qk ) ≤ I(A)
d−2

for all k ≥ 0. The hypothesis of induction on d readily shows that

|e1Q (A)| = |e1Q (A)| ≤ I(A) ≤ I(A)

and

|eiQ (A)| = |eiQ (A)| ≤ 3 · 2i−2 (r  + 1)i−1 I(A) ≤ 3 · 2i−2 (r + 1)i−1 I(A)
104
8 SHIRO GOTO AND KAZUHO OZEKI

for 2 ≤ i ≤ d − 1, because r  ≤ r by Lemma 2.3 and I(A) ≤ I(A). Therefore, for


the case where i = d, we get
r
    r  
k+1 k+d−1 k+d−2
|eQ (A)| ≤
d
A (A/Q ) − eQ (A)
0
+ |eQ (A)|
1
d−1 d−2
k=0 k=0
r 
 d−1   r
k+d−1−i
+ |eiQ (A)| + A ([Qk+1 : a]/Qk )
i=2
d − 1 − i
k=0 k=0
r 
  r 
 
k+d−2 k+d−2
≤ I(A) + I(A)
d−2 d−2
k=0 k=0
r 
 d−1    r  
k+d−1−i k+d−2
+ 3 · 2i−2 (r + 1)i−1 I(A) + I(A)
d−1−i d−2
k=0 i=2 k=0
     
r +d−1 r +d−1
= I(A) + I(A)
d−1 d−1

d−1     
r +d−i r+d−1
+ 3 · 2 (r + 1) I(A)
i−2 i−1
+ I(A)
i=2
d−i d−1
  
d−1  
r+d−1 i−1 r + d − i
≤ 3 I(A) + 3 · 2 (r + 1)
i−2
I(A)
d−1 i=2
d−i
(since r ≤ r by Lemma 2.3 and I(A) ≤ I(A))

d−1
≤ 3(r + 1)d−1 I(A) + 3 · 2i−2 (r + 1)i−1 (r + 1)d−i I(A)
i=2
 
m+n
(since ≤ (m + 1)n for all integers m, n ≥ 0)
n

d−1
= 3(r + 1)d−1 I(A) + 3 · 2i−2 (r + 1)d−1 I(A)
i=2



d−1
= 3(r + 1)d−1 I(A) 1 + 2i−2
i=2
= 3 · 2d−2 (r + 1)d−1 I(A)

d−1
(since 2i−2 = 2d−2 − 1).
i=2

This completes the proof of Theorem 2.2 and, thanks to Theorem 2.1, that of the
implication (1) ⇒ (2) of Theorem 1.1 as well. 

We are now in a position to prove the implication (2) ⇒ (1) in Theorem 1.1.

Proof of Theorem 1.1. We may assume that A is complete. Also we may


assume A is not unmixed, because Λ1 (A) is a finite set (cf. [GhGHOPV, Propo-
sition 4.2]). Let U denote the unmixed component of the ideal (0) in A. We put
B = A/U and t = dimA U (≤ d − 1). We must show that B is a generalized
Cohen-Macaulay ring and t = 0.
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 105
9

Let Q be a parameter ideal in A. Then by the exact sequence 0 → U → A →


B → 0 we have
A (A/Qn+1 ) = A (B/Qn+1 B) + A (U/Qn+1 ∩ U )
for all integers n ≥ 0. The function A (U/Qn+1 ∩ U ) is a polynomial in n  0 with
degree t and there exist integers {siQ (U )}0≤i≤t with s0Q (U ) = e0Q (U ) such that
t  
n+t−i
A (U/Qn+1
∩ U) = i i
(−1) sQ (U )
i=0
t−i
for all n  0, whence

d    t  
n+d−i n+t−i
A (A/Qn+1 ) = (−1)i eiQ (B) + (−1)i siQ (U ) .
i=0
d−i i=0
t−i
Consequently
⎧ d−i t−i
⎨ (−1)d−i eQ (B) + (−1)t−i sQ (U ) if 0 ≤ i ≤ t,
(−1)d−i ed−i
Q (A) =

(−1)d−i ed−i
Q (B) if t + 1 ≤ i ≤ d.
Therefore, if t < d − 1, we have e1Q (A) = e1Q (B), so that Λ1 (B) = Λ1 (A) is a finite
set. If t = d − 1, we get −e1Q (A) = −e1Q (B) + s0Q (U ). Since e1Q (A), e1Q (B) ≤ 0 and
s0Q (U ) = e0Q (U ) ≥ 1, Λ1 (B) is a finite set also in this case. Thus the set Λ1 (B) is
finite in any case, so that the ring B is a generalized Cohen-Macaulay ring.
We now assume that t ≥ 1 and choose a system a1 , a2 , · · · , ad of parameters
in A so that (at+1 , at+2 , · · · , ad )U = (0). Let  ≥ 1 be an integer such that m is
standard for the ring B and choose integers n ≥ . We look at parameter ideals
Q = (an1 , an2 , · · · , and ) of A. Then
t  
d−t d−t t−1 j
(−1) eQ (B) = h (B)
j=1
j−1

by [S, Korollar 3.2], which is independent of the integers n ≥ . Therefore, since


s0Q (U ) = e0(an1 ,an2 ,··· ,ant ) (U ) = nt ·e0(a1 ,a2 ,··· ,at ) (U ) ≥ nt ,
we see
(−1)d−t ed−t
Q (A) = (−1)
d−t d−t
eQ (B) + s0Q (U )
t  
t−1 j
= h (B) + nt ·e0(a1 ,a2 ,··· ,at ) (U )
j=1
j − 1
≥ nt ,
whence the set Λd−t (A) cannot be finite. Thus t = 0 and A a generalized Cohen-
Macaulay ring. 

3. The second Hilbert coefficients e2Q (A) of parameters


Let A be a Noetherian local ring with maximal ideal m, d = dim A > 0,
and infinite residue class field A/m. In this section we study the second Hilbert
coefficients e2Q (A) of parameter ideals Q. The purpose is to find the sharp bound
106
10 SHIRO GOTO AND KAZUHO OZEKI

for e2Q (A). The bound |e2Q (A)| ≤ 3(r + 1)I(A) given by Theorem 2.2 is too huge in
general and far from the sharp bound.
Let us begin with the following.
Lemma 3.1. Suppose that d = 2 and depth A > 0. Let Q = (x, y) be a param-
eter ideal in A and assume that x is superficial with respect to Q. Then
  
[(x ) : y  ] ∩ Q
e2Q (A) = −A ≤0
(x )
for all   0.
Proof. Let   0 be an integer which is sufficiently large and put I = Q .
Let G = G(I) and R = R(I) be the associated graded ring and the Rees algebra
of I, respectively. We put M = mR + R+ . Then [HiM (G)]n = (0) for all integers
i ∈ Z and n > 0, thanks to [H, Lemma 2.4]. We put a = x and b = y  . Then the
element a remains superficial with respect to I and the equality I 2 = (a, b)I holds
true, whence
a2 (G) = sup{n ∈ Z | [H2M (G)]n = (0)} < 0.
We furthermore have the following.
Claim 2. [HiM (R)]0 ∼= [HiM (G)]0 as A-modules for all i ∈ Z. Hence H0M (G) =
(0), so that f = at ∈ R is G-regular.
Proof of Claim 2. Let L = R+ and apply the functors HiM (∗) to the fol-
lowing canonical exact sequences
p
0→L→R→A→0 and 0 → L(1) → R → G → 0,
where p denotes the projection, and get the exact sequences
(1) · · · → Hi−1
m (A) → HM (L) → HM (R) → Hm (A) → · · · and
i i i

(2) · · · → Hi−1
M (G) → HM (L)(1) → HM (R) → HM (G) → HM (L)(1) → · · ·
i i i i+1

of local cohomology modules. Then by exact sequence (2) we get the isomorphism
[Hi (L)]n+1 ∼
M = [Hi (R)]n M

for n ≥ 1, because [Hi−1


M (G)]n = [HM (G)]n = (0) for n ≥ 1, while we have the
i

isomorphism
[HiM (L)]n+1 ∼
= [HiM (R)]n+1
for n ≥ 1, thanks to exact sequence (1). Hence [HiM (R)]n ∼ = [HiM (R)]n+1 for n ≥ 1,
which implies [HM (R)]n = (0) for all i ∈ Z and n ≥ 1, because [HiM (R)]n = (0)
i

for n  0. Thus by exact sequence (1) we get [HiM (L)(1)]n = (0) for all i ∈ Z and
n ≥ 0, so that by exact sequence (2) we see [HiM (R)]0 ∼ = [HiM (G)]0 as A-modules
for all i ∈ Z. Considering the case where i = 1 in exact sequence (2), we have the
embedding
0 → H0M (G) → H1M (L)(1),
so that [H0M (G)]0 = (0), because [H1M (L)(1)]0 = [H0M (L)]1 = (0). Hence H0M (G) =
(0), so that f is G-regular, because (0) :G f is finitely graded. 
Thanks to [B, Theorem 4.1], Claim 2 shows that

2
e2Q (A) = (−1)i A ([HiM (G)]0 ) = −A ([H1M (G)]0 ),
i=0
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 107
11

since a2 (G) < 0. Therefore to prove


 
[(x ) : y  ] ∩ Q
e2Q (A) = −A ,
(x )
it is suffices to check that
[(a) : b] ∩ I
[H1M (G)]0 ∼
=
(a)
as A-modules.
Let A = A/(a) and I = IA. Then G/f G ∼ = G(I), because f = at is G-regular
(cf. Claim 2). We now look at the exact sequence
f
0 → H0M (G(I)) → H1M (G)(−1) → H1M (G)
of local cohomology modules which is induced from the exact sequence
f
0 → G(−1) → G → G(I) → 0
of graded G-modules. Then, since [H1M (G)]n = (0) for all n ≥ 1, we have an
isomorphism
[H0M (G(I))]1 ∼
= [H1M (G)]0
of A-modules and the vanishing [H0M (G(I)]n = (0) for n ≥ 2.
Look now at the homomorphism
[(a) : b] ∩ I
ρ: → [H0M (G(I))]1
(a)
of A-modules defined by ρ(x) = xt for each x ∈ [(a) : b] ∩ I, where x and xt
denote the images of x in A and xt ∈ [R(I)]1 in G(I), respectively. We will show
that the map ρ is an isomorphism. Take ϕ ∈ [H0M (G(I))]1 and write ϕ = xt with
x ∈ I. Since [H0M (G(I))]2 = (0), we have bt · xt = bxt2 = 0 in G(I), whence
bx ∈ [(a) + I 3 ] ∩ I 2 = [(a) ∩ I 2 ] + I 3 = aI + bI 2 (recall that I 2 = (a, b)I and that
a is super-regular with respect to I). So, we write bx = ai + bj with i ∈ I and
j ∈ I 2 . Then, since b(x − j) = ai ∈ (a), we have x − j ∈ [(a) : b] ∩ I, whence
ϕ = xt = (x − j)t. Thus the map ρ is surjective.
To show that the map ρ is injective, take x ∈ [(a) : b] ∩ I and suppose that
ρ(x) = xt = 0 in G(I). Then
x ∈ [(a) : b] ∩ [(a) + I 2 ] = (a) + [((a) : b) ∩ I 2 ].
To conclude that x ∈ (a), we need the following.
Claim 3. Let n ≥ 2 be an integer. Then [(a) : b] ∩ I n ⊆ (a) + [((a) : b) ∩ I n+1 ].
Proof of Claim 3. Take y ∈ [(a) : b] ∩ I n . Then, since by ∈ (a), we see bt ·
ytn = bytn+1 = 0 in G(I). Hence ytn ∈ [H0M (G(I))]n , because bt is a homogeneous
parameter for the graded ring G(I). Recall now that n ≥ 2, whence [H0M (G(I))]n =
(0), so that ytn = 0. Thus y ∈ (a) + I n+1 , whence y ∈ (a) + [((a) : b) ∩ I n+1 ], as
claimed. 
Since x ∈ (a) + [((a) : b) ∩ I 2 ], thanks to Claim 3, we get x ∈ (a) + I n+1 for all
n ≥ 1, whence x ∈ (a), so that the map ρ is injective. Thus
[(a) : b] ∩ I
[H1M (G)]0 ∼
=
(a)
108
12 SHIRO GOTO AND KAZUHO OZEKI

as A-modules. 
Theorem 3.2. Suppose that d = 2 and depth A > 0. Let Q = (x, y) be a
parameter ideal in A and assume that x is superficial with respect to Q. Then
−h1 (A) ≤ e2Q (A) ≤ 0
and the following three conditions are equivalent.
(1) e2Q (A) = 0.
(2) x, y forms a d-sequence in A.
(3) x , y  forms a d-sequence in A for all integers  ≥ 1.
Passing to the ring A/H0m (A), thanks to Theorem 3.2, we readily get the fol-
lowing.
Corollary 3.3. Suppose that d = 2 and let Q be a parameter ideal in A. Then
h0 (A) − h1 (A) ≤ e2Q (A) ≤ h0 (A).
Proof of Theorem 3.2. By Lemma 3.1 we have
  
[(x ) : y  ] ∩ (x, y)
e2Q (A) = −A ≤0
(x )
for all integers   0. To show
−h1 (A) ≤ e2Q (A),
we may assume H1m (A) is finitely generated. Then since
[(a) : b] ∩ Q (a) : b
⊆ ⊆ H0m (A/(a)) ∼
= H1m (A),
(a) (a)
we get −h1 (A) ≤ e2Q (A).
Let us consider the second assertion.
(1) ⇒ (3). Take an integer N ≥ 1 so that
  
[(x ) : y  ] ∩ (x, y)
e2Q (A) = −A =0
(x )
for all  ≥ N (cf. Lemma 3.1); hence
[(x ) : y  ] ∩ (x, y) = (x ).
Claim 4. [(x ) : y  ] ∩ (x, y) = (x ) for all  ≥ 1.
Proof of Claim 4. We may assume that 1 ≤  < N . Take τ ∈ [(x ) :
y ] ∩ (x, y) . Then, since y N (xN − τ ) = y N − xN − (y  τ ) ∈ (xN ), we have xN − τ ∈


[(xN ) : y N ] ∩ (x, y)N = (xN ). Thus τ ∈ (x ), because x is A-regular (recall that
depth A > 0 and x is superficial with respect to Q). 
Since x is A-regular and [(x ) : y  ] ∩ (x , y  ) = (x ) by Claim 4, we readily see
that x , y  is a d-sequence in A.
(3) ⇒ (2) This is clear.
(2) ⇒ (1) It is well-known that e2(x,y) (A) = 0, if depth A > 0 and the system
x, y of parameters forms a d-sequence in A; see Proposition 3.4 below. 
The results in the following proposition are, more or less, known but let us
indicate brief proofs for the sake of completeness, because the results play important
roles in our proof of Theorem 1.2.
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 109
13

Proposition 3.4. Suppose that d > 0 and let Q = (a1 , a2 , · · · , ad ) be a param-


eter ideal in A. Let G = G(Q) and R = R(Q). Let fi = ai t ∈ R for 1 ≤ i ≤ d.
Assume that the sequence a1 , a2 , · · · , ad forms a d-sequence in A. Then we have
the following, where Qi = (a1 , a2 , · · · , ai ) for 0 ≤ i ≤ d.
(1) e0Q (A) = A (A/Q) − A ([Qd−1 : ad ]/Qd−1 ).
(2) (−1)i eiQ (A) = h0 (A/Qd−i ) − h0 (A/Qd−i−1 ) for 1 ≤ i ≤ d − 1 and
(−1)d edQ (A) = h0 (A).
d  
(3) A (A/Qn+1 ) = i=0 (−1)i eiQ (A) n+d−i for all n ≥ 0, whence A (A/Q) =
d i i
d−i

i=0 (−1) eQ (A).


(4) f1 , f2 , · · · , fd forms a d-sequence in G.
(5) H0M (G) = [H0M (G)]0 ∼ = H0m (A), where M = mR + R+
(6) [HM (G)]n = (0) for all n > −i and i ∈ Z, whence reg G = 0.
i

Proof. Let us check assertions (1), (2), and (3). Let W = H0m (A) and put
A = A/(a) and Q = QA, where a = a1 . Let k ≥ 0 be an integer. Then by virtue
of the exact sequence
a k+1
0 → [Qk+1 : a]/Qk → A/Qk → A/Qk+1 → A/Q →0
of A-modules and the equality
Qk+1 : a = Qk + W
(recall that W = (0) : a and (a) ∩ Qk+1 = aQk for all k ≥ 0, since a1 , a2 , · · · , ad
forms a d-sequence in A), we get
A (Qk /Qk+1 ) = A (A/Qk+1 ) − A (A/Qk )
k+1
= A (A/Q ) − A (W/[W ∩ Qk ])
k+1
= A (A/Q ) − A (W ) + A (Qk ∩ W ).
Hence

n 
n
k+1
A (A/Qn+1 ) = A (Qk /Qk+1 ) = A (A/Q ) − h0 (A)·n ()
k=0 k=0

for all n ≥ 0, because Qk ∩ W = (0) if k ≥ 1.


Suppose now that d = 1. Then
A (A/Qn+1 ) = A (A/Q)(n + 1) − h0 (A)(n + 1) + h0 (A)
= {A (A/Q) − h0 (A)}(n + 1) + h0 (A)
k+1
for all n ≥ 0, because A (A/Q ) = A (A/Q) for all k ≥ 0. Hence
e0Q (A) = A (A/Q) − A ([(0) : a1 ]) and − e1Q (A) = h0 (A).

Assume that d ≥ 2 and that our assertions hold true for d − 1. Then Q =
(a2 , a3 , · · · , ad ) and the sequence a2 , a3 , · · · , ad forms a d-sequence in A, where ai
denotes the image of ai in A. Then the hypothesis of induction on d guarantees
that

d−1  
k+1 i i k+d−1−i
A (A/Q )= (−1) eQ (A)
i=0
d−1−i
110
14 SHIRO GOTO AND KAZUHO OZEKI

for all k ≥ 0. Hence by equality () we have



n
k+1
A (A/Qn+1 ) = A (A/Q ) − h0 (A)(n + 1) + h0 (A)
k=0

d−1  

n  k+d−1−i
= (−1)i eiQ (A) − h0 (A)(n + 1) + h0 (A)
d − 1 − i
k=0 i=0

d−1  
n+d−i
= i i
(−1) eQ (A) − h0 (A)(n + 1) + h0 (A)
i=0
d − i

d−2  
n+d−i
= i i
(−1) eQ (A) + {(−1)d−1 ed−1 (A) − h0 (A)}(n + 1) + h0 (A)
i=0
d − i Q

for all n ≥ 0. Thus (−1)d edQ (A) = h0 (A) and by the hypothesis of induction on d
we get that
e0Q (A) = e0Q (A) = A (A/Q) − A ([(a2 , a3 , · · · , ad−1 ) : ad ]/(a2 , a3 , · · · , ad−1 ))
= A (A/Q) − A ([Qd−1 : ad ]/Qd−1 ),

(−1)i eiQ (A) = (−1)i eiQ (A)


= h0 (A/(a2 , a3 , · · · , ad−i )) − h0 (A/(a2 , a3 , · · · , ad−i−1 ))
= h0 (A/Qd−i ) − h0 (A/Qd−i−1 )
for 1 ≤ i ≤ d − 2, and
(−1)d−1 ed−1
Q (A) = (−1) eQ (A) − h0 (A)
d−1 d−1

= h0 (A) − h0 (A)
= h0 (A/Q1 ) − h0 (A/Q0 ).
(4) This follows from the standard argument; use the fact that (a1 , a2 , · · · , ai )∩
Qn = (a1 , a2 , · · · , ai )Qn−1 for all 1 ≤ i ≤ d and n ∈ Z.
(5) We get H0M (G) = (0) :G f1 by assertion (4). Since [(0) : a1 ] ∩ Q = (0) and
(a1 ) ∩ Qn = a1 Qn−1 for all n ∈ Z, we also have [(0) :G f1 ] = [(0) :G f1 ]0 . Hence
H0M (G) = [H0M (G)]0 ∼ = W , because [(0) :G f1 ]0 ∼
= (0) : a1 = W .
(6) We proceed by induction on d. We may assume that d ≥ 2 and that our
assertion holds true for d − 1. Then, we have the isomorphism G/f1 G ∼ = G(Q) of
graded rings, since (a1 ) ∩ Qn = a1 Qn−1 . Because H0M (G) = (0) :G f1 , we have the
exact sequence
f1
0 → H0M (G)(−1) → G(−1) → G → G(Q) → 0
of graded G-modules. Hence applying the local cohomology functors HiM (∗) to it,
we get the long exact sequence
f1
0 → H0M (G) → H0M (G(Q)) → H1M (G)(−1) → H1M (G) → · · ·
f1
M (G(Q)) → HM (G)(−1) → HM (G) → HM (G(Q)) → · · · .
· · · → Hi−1 i i i

Then since [HiM (G(Q))]n = (0) if n > −i by the hypothesis of induction on d, it


is standard to check that [HiM (G)]n = (0) if n > −i, which completes the proof of
Proposition 3.4. 
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 111
15

Let us note one example of local rings A which are not generalized Cohen-
Macaulay rings but every parameter ideal in A is generated by a system of param-
eters that forms a d-sequence in A.
Example 3.5. Let R be a regular local ring with the maximal ideal n and
d = dim R ≥ 2. Let X1 , X2 , · · · , Xd be a regular system of parameters of R. We
put p = (X1 , X2 , · · · , Xd−1 ) and D = R/p. Then D is a DVR. Let A = R  D
denote the idealization of D over R. Then A is a Noetherian local ring with the
maximal ideal m = n × D, dim A = d, and depth A = 1. We furthermore have the
following.
(1) Λi (A) = {0} for all 1 ≤ i ≤ d such that i = d − 1.
(2) Λ0 (A) = {n | 0 < n ∈ Z} and Λd−1 (A) = {(−1)d−1 n | 0 < n ∈ Z}.
(3) After renumbering, every system of parameters in A forms a d-sequence.
The ring A is not a generalized Cohen-Macaulay ring, because H1m (A) (∼ = H1n (D))
is not a finitely generated A-module.
Proof. Let Q be a parameter ideal in A and put q = p(Q), where p : A →
R, p(a, x) = a denotes the projection map. We then have, thanks to the exact
sequence
j p
0 →p D → A → R → 0
of A-modules (here p D denote the R-module D which is considered to be an A-
module via p and j(x) = (0, x) for each x ∈ D),
A (A/Qn+1 ) = R (R/qn+1 ) + D (D/qn+1 D)
   
n+d n+1
= R (R/q)· + D (D/qD)·
d 1
   
n+d n+1
= e0q (R) + e0q (D)
d 1
for all integers n ≥ 0, so that e0Q (A) = e0q (R), (−1)d−1 ed−1 Q (A) = eq (D) ≥ 1, and
0

eiQ (A) = 0 if i = 0, d − 1.
Let ni ≥ 1 (1 ≤ i ≤ d) be integers and put q = (X1n1 , X2n2 , · · · , Xdnd )R. We

look at the parameter ideals Q = qA in A. Then e0Q (A) = e0q (R) = di=1 ni and
Q (A) = eq (D) = n1 . Hence Λ0 (A) = {n | 0 < n ∈ Z} and Λd−1 (A) =
(−1)d−1 ed−1 0

{(−1) n | 0 < n ∈ Z}.


d−1

Let f1 , f2 , · · · , fd be a system of parameters in A. For each 1 ≤ i ≤ d, let us


write fi = (ai , xi ) with ai ∈ R and xi ∈ D. We assume that ai D ⊆ a1 D for all
1 ≤ i ≤ d. Then f1 , f2 , · · · , fd is a d-sequence in A. In fact, let 1 ≤ i ≤ j ≤ d be
integers and ϕ ∈ (f1 , f2 , · · · , fi−1 ) : fi fj . If i = 1, then f1 (fj ϕ) = 0, so that fj ϕ =
0, because f1 is A-regular. Suppose that i ≥ 2. Then p(ϕ) ∈ (a1 , a2 , · · · , ai−1 ), since
(ai aj )p(ϕ) ∈ (a1 , a2 , · · · , ai−1 ) and ai aj is R/(a1 , a2 , · · · , ai−1 )-regular. Therefore,
considering the canonical exact sequence
j p
0 →p [D/(a1 , a2 , · · · , ai−1 )D] → A/(f1 , f2 , · · · , fi−1 ) → R/(a1 , a2 , · · · , ai−1 ) → 0
of A-modules, we see that ϕ ≡ (0, x) mod(f1 , f2 , · · · , fi−1 ) for some x ∈ D. Hence
fj ϕ ≡ (0, aj x) mod(f1 , f2 , · · · , fi−1 ), which implies fj ϕ ∈ (f1 , f2 , · · · , fi−1 ), be-
cause aj x ∈ a1 D. 
We now consider the case where dim A ≥ 3
112
16 SHIRO GOTO AND KAZUHO OZEKI

Theorem 3.6. Suppose that A is a generalized Cohen-Macaulay ring with d =


dim A ≥ 3. Let Q = (a1 , a2 , · · · , ad ) be a parameter ideal in A. Then


d−1 
d−3 j   d − 3
d−2
− h (A) ≤ eQ (A) ≤
2
hj (A).
j=2
j−2 j=1
j − 1

We have Q·Hjm (A/(a1 , a2 , · · · , ak )) = (0) for all k ≥ 0 and j ≥ 1 with j + k ≤ d − 2,


d−2   j
j−1 h (A) and if a1 , a2 , · · · , ad forms a superficial sequence with
if e2Q (A) = j=1 d−3
respect to Q.

Proof. First, we shall prove that


d−1 
  
d−2 
d−3 d−3 j
− hj (A) ≤ e2Q (A) ≤ h (A)
j=2
j−2 j=1
j−1

and that Q·Hjm (A) = (0) for all 1 ≤ j ≤ d − 2, once the equality e2Q (A) =
d−2 d−3 j 0 2
j=1 j−1 h (A) holds true. Let W = Hm (A) and put C = A/W . Since eQC (C) =
eQ (A) and h (C) = h (A) for j ≥ 1, passing to the ring C, we may assume that
2 j j

depthA > 0. We may also assume that each ai is superficial with respect to Q. We
put a = ai , A = A/(a), and Q = QA.
Suppose that d = 3. Then by the long exact sequence
a a
0 → H0m (A) → H1m (A) → H1m (A) → H1m (A) → H2m (A) → H2m (A) → · · ·

of local cohomology modules, we have

h0 (A) = A ([(0) :H1m (A) a]) and h1 (A) = A ([(0) :H1m (A) a]) + A ([(0) :H2m (A) a]).

Therefore, thanks to Corollary 3.3, we get

e2Q (A) = e2Q (A) ≤ h0 (A) = A ([(0) :H1m (A) a]) ≤ h1 (A)

and

e2Q (A) = e2Q (A) ≥ h0 (A) − h1 (A)


= A ([(0) :H1m (A) a]) − {A ([(0) :H1m (A) a]) + A ([(0) :H2m (A) a])}
= −A ([(0) :H2m (A) a]) ≥ −h2 (A).

In particular, aH1m (A) = (0), if e2Q (A) = h1 (A). By the symmetry among ai we get
ai H1m (A) = (0) for all i, whence Q·H1m (A) = (0).
Assume that d ≥ 4 and that our assertion holds true for d−1. Let 1 ≤ j ≤ d−2
and look at the exact sequence
a a
Hjm (A) → Hjm (A) → Hjm (A) → Hj+1
m (A) → Hm (A)
j+1

of local cohomology modules and we have

hj (A) = A ([(0) :Hjm (A) a]) + A ([(0) :Hj+1


m (A)
a]).
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 113
17

Hence the hypothesis of induction on d shows that


 d − 4
d−3
e2Q (A) = e2Q (A) ≤ hj (A)
j=1
j − 1


d−3
d−4

= {A ([(0) :Hjm (A) a]) + A ([(0) :Hj+1 a])}
j=1
j−1 m (A)


d−2
d−3

= A ([(0) :Hjm (A) a])
j=1
j−1


d−2 
d−3 j
≤ h (A)
j=1
j−1

and
d−2 
 
d−4
e2Q (A) = e2Q (A) ≥ − hj (A)
j=2
j−2
d−2 
 
d−4
= − {A ([(0) :Hjm (A) a]) + A ([(0) :Hj+1 a])}
j=2
j−2 m (A)

d−1 
 
d−3
= − A ([(0) :Hjm (A) a])
j=2
j−2
d−1 
 
d−3 j
≥ − h (A).
j=2
j−2
d−3   j
Hence aHjm (A) = (0) for 1 ≤ j ≤ d − 2 and e2Q (A) = j=1 d−4 2
j−1 h (A), if eQ (A) =
d−2 d−3 j j
j=1 j−1 h (A). Then by the symmetry among ai , we get ai Hm (A) = (0) for
1 ≤ i ≤ d and 1 ≤ j ≤ d − 2, so that Q·Hjm (A) = (0) for 1 ≤ j ≤ d − 2.
Now let us check by induction on d that Q·Hjm (A/(a1 , a2 , · · · , ak )) = (0) for
d−2 d−3 j
all k ≥ 0 and j ≥ 1 with j + k ≤ d − 2, if e2Q (A) = j=1 j−1 h (A) and if
a1 , a2 , · · · , ad forms a superficial sequence with respect to Q. We may assume that
d ≥ 4 and that the assertion holds true for d − 1. Similarly as before, we may also
assume that depth A > 0. Then, because
d−3 
 
d−4 j
e2Q/(a1 ) (A/(a1 )) = h (A/(a1 ))
j=1
j−1

as we have shown above, we get by the hypothesis on d that


Q·Hjm (A/(a1 , a2 , · · · , ak )) = Q·Hjm (A/(a2 , a3 , · · · , ak )A) = (0)

for all k, j ≥ 1 with k + j ≤ d − 2. Thus Q·Hjm (A/(a1 , a2 , · · · , ak )) = (0) for all


k ≥ 0 and j ≥ 1 with j + k ≤ d − 2, because Q·Hjm (A) = (0) for 1 ≤ j ≤ d − 2. 

The following result guarantees the implication (2) ⇒ (1) and the last assertion
in Theorem 1.2.
114
18 SHIRO GOTO AND KAZUHO OZEKI

Proposition 3.7. Suppose that A is a generalized Cohen-Macaulay ring with


d = dim A ≥ 3 and let Q = (a1 , a2 , · · · , ad ) be a parameter ideal in A. Assume that
the sequence a1 , a2 , · · · , ad forms a d-sequence in A and Q·Hjm (A/(a1 , a2 , · · · , ak )) =
(0) for all k ≥ 0 and j ≥ 1 with j + k ≤ d − 2. Then
d−i 
 
d−i−1 j
(−1)i eiQ (A) = h (A)
j=1
j−1

for 2 ≤ i ≤ d − 1 and (−1)d edQ (A) = h0 (A).

Proof. We proceed by induction on d. Suppose that d = 3. Then −e3Q (A) =


h0 (A) and
e2Q (A) = h0 (A/(a1 )) − h0 (A)
= {h1 (A) + h0 (A)} − h0 (A) = h1 (A)
by Proposition 3.4, because h0 (A/(a1 )) = h0 (A) + h1 (A) (notice that H0m (A) =
(0) : a1 and Q·H1m (A) = (0)).
Let us assume that d ≥ 4 and that our assertion holds true for d − 1. Notice
that a1 , a2 , · · · , ad is a superficial sequence with respect to Q = (a1 , a2 , · · · , ad ) by
Proposition 3.4 (4). Let A = A/(a1 ) and let ai denote the image of ai in A. Then,
since a2 , a3 , · · · , ad forms a d-sequence in A and
Q·Hjm (A/(a2 , a3 , · · · , ak ) = (0)
for all j, k ≥ 1 with j + k ≤ d − 2, by the hypothesis of induction on d we get
 d − i − 2
d−i−1
(−1)i eiQ (A) = (−1)i eiQ (A) = hj (A)
j=1
j − 1

 
d−i−1
d−i−2

= {hj (A) + hj (A)}
j=1
j−1
d−i 
 
d−i−1 j
= h (A)
j=1
j−1

for 2 ≤ i ≤ d − 2 and
(−1)d−1 ed−1
Q (A) = (−1) eQ (A) − h0 (A)
d−1 d−1

= h0 (A) − h0 (A)
= {h0 (A) + h1 (A)} − h0 (A) = h1 (A)
by Proposition 3.4 (2) (notice that hi (A) = hi (A) + hi+1 (A) for 0 ≤ i ≤ d −
3, because H0m (A) = (0) : a1 and Q·Him (A) = (0) for 1 ≤ i ≤ d − 2). Since
(−1)d edQ (A) = h0 (A) by Proposition 3.4 (2), this completes the proof of Proposition
3.7. 

4. Proof of Theorem 1.2


Let A be a Noetherian local ring with maximal ideal m and infinite residue
class field A/m. The purpose of this section is to prove Theorem 1.2. Thanks to
Theorem 3.6 and Proposition 3.7, we have only to show the following.
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 115
19

Theorem 4.1. Suppose that A is a generalized Cohen-Macaulay ring with d =


dim A ≥ 3 anddepth jA > 0. Let Q be a parameter ideal in A and assume that
d−2
e2Q (A) = j=1 d−3
j−1 h (A). Then Q is generated by a system of parameters which
forms a d-sequence in A.
For each ideal a in A (a = A) let U(a) denote the unmixed component of a.
When a = (a) with a ∈ A, we write U(a) simply by U(a). We have

U(a) = [(a) :A mn ] ,
n≥0

if A is a generalized Cohen-Macaulay ring with dim A ≥ 2 and a is a part of a


system of parameters in A (cf. [STC, Section 2]). The following result is the key
in our proof of Theorem 4.1.
Proposition 4.2. Suppose that A is a generalized Cohen-Macaulay ring with
d = dim A ≥ 2 and depth A > 0. Let Q = (a1 , a2 , · · · , ad ) be a parameter ideal
in A. Assume that ad H1m (A) = (0) and that the sequence a1 , a2 , · · · , ad−1 forms a
d-sequence in the generalized Cohen-Macaulay ring A/U(ad ). Then
U(a1 ) ∩ [Q + U(ad )] = (a1 ).
Proof. Let x ∈ U(a1 ) ∩ [Q + U(ad )]. Notice that
U(a1 ) ∩ [Q + U(ad )] = (a1 ) + [U(a1 ) ∩ U(ad )].
In fact, let x ∈ U(a1 )∩[Q+U(ad )]. We put B = A/U(ad ). Let x and ai respectively
denote the images of x and ai in B. Then
x ∈ U(a1 ) ∩ (a1 , a2 , · · · , ad−1 )B ⊆ (a1 )B,
since the sequence a1 , a2 , · · · , ad−1 forms a d-sequence in B. Hence
x ∈ [(a1 ) + U(ad )] ∩ U(a1 ) = (a1 ) + [U(a1 ) ∩ U(ad )].
Let us write x = y + z with y ∈ (a1 ) and z ∈ U(a1 ) ∩ U(ad ). We will show that
z ∈ (a1 ). Since ai is A-regular, we have
U(ai )/(ai ) = H0m (A/(ai )) → H1m (A)
for i = 1, d, whence ad U(a1 ) ⊆ (a1 ), because ad H1m (A) = (0) by our assumption.
Choose an integer  > 0 so that a1 H1m (A) = (0). Then, since a1 U(ad ) ⊆ (ad ), we
have
ad z ∈ (a1 ) and a1 z ∈ (ad ).
Let us write
ad z = a1 u and a1 z = ad v with u, v ∈ A.
1 u = ad v, we have v ∈ U(a1 ). Notice that
Then, since a1 ad z = a+1 2 +1

U(a+1 +1 0 +1 1


1 )/(a1 ) = Hm (A/(a1 )) → Hm (A),

because a+1 1 ) ⊆ (a1 ), since ad Hm (A) = (0).


is A-regular. We then have ad U(a+1 +1 1
1
Therefore, because ad v ∈ (a1 ), we may write ad v = a1 w with w ∈ A, whence
+1 +1

a1 ad z = ad ·ad v = a+1


1 ad w, so that z = a1 w ∈ (a1 ). Thus x = y + z ∈ (a1 ), which
completes the proof of Proposition 4.2. 

We are now ready to prove Theorem 4.1.


116
20 SHIRO GOTO AND KAZUHO OZEKI

Proof of the Theorem 4.1. We proceed by induction on d. Choose ele-


ments a1 , a2 , · · · , ad ∈ A so that Q = (a1 , a2 , · · · , ad ) and for each 0 ≤ i ≤ d−2, the
i + 2 elements a1 , a2 , · · · , ai , ad−1 , ad form a superficial sequence with respect to Q.
We will show that there exist b2 , b3 , · · · , bd ∈ A such that b1 = ad−1 , b2 , b3 , · · · , bd
forms a d-sequence in A and Q = (b1 , b2 , · · · , bd ). We put A = A/(a1 ), Q = QA,
and C = A/H0m (A) (= A/U(a1 )).
Suppose that d = 3. Then
e2QC (C) = e2Q (A) − h0 (A) = e2Q (A) − h0 (A) = h1 (A) − h0 (A) = 0,

because h1 (A) = h0 (A) (recall that Q·H1m (A) = (0) by Proposition 3.6). Hence,
thanks to Proposition 3.2, a2 , a3 forms a d-sequence in C, because a2 is superficial
for the ideal QC = (a2 , a3 )C. Therefore, since a1 H1m (A) = (0), we have
U(a2 ) ∩ [Q + U(a1 )] = (a2 ),
by Proposition 4.2. Let Q = (a2 , a3 , b3 ) and B = A/U(a2 ). Then since e2QB (B) = 0,
by Proposition 3.2 the sequence b2 = a3 , b3 forms a d-sequence in B, because b2
is superficial for QB. Therefore, since U(a2 ) ∩ Q ⊆ U(a2 ) ∩ [Q + U(a1 )] = (a2 ),
the sequence b2 , b3 forms a d-sequence in A/(a2 ), so that b1 = a2 , b2 , b3 forms a
d-sequence in A, because b1 is A-regular.
Assume that d ≥ 4 and that our assertion holds true for d − 1. Then, thanks
to Theorem 3.6 and its proof, we have
d−3 
 
d−4 j
e2Q (A) = e2Q (A) = e2QC (C) ≤ h (C)
j=1
j−1
d−3 
 
d−4
= hj (A)
j=1
j−1
d−3 
 
d−4
= {hj (A) + hj+1 (A)}
j=1
j−1
d−2 
 
d−3 j
= h (A) = e2Q (A),
j=1
j−1

because Q·Hjm (A) = (0) for 1 ≤ j ≤ d − 3. Hence


 d − 4
d−3
e2QC (C) = hj (C).
j=1
j − 1

Therefore, because QC = (a2 , a3 , · · · , ad )C and the sequence a2 , a3 , · · · , ai , ad−1 , ad


is superficial in the ideal QC for all 1 ≤ i ≤ d − 2 where aj denotes the image of aj
in C, the hypothesis of induction on d yields that there exist γ2 , γ3 , · · · , γd−1 ∈ C
such that the sequence γ1 = ad−1 , γ2 , γ3 , · · · , γd−1 forms a d-sequence in C and
QC = (γ1 , γ2 , · · · , γd−1 )C. Let us write γj = cj for each 2 ≤ j ≤ d − 1 with cj ∈ Q,
where cj denote the image of cj in C. We put q = (a1 , ad−1 , c2 , c3 , · · · , cd−1 ). Then
q is a parameter ideal in A, a1 H1m (A) = (0), and ad−1 , c2 , c3 , · · · , cd−1 forms a
d-sequence in C. Therefore
U(ad−1 ) ∩ [Q + U(a1 )] = U(ad−1 ) ∩ [q + U(a1 )] = (ad−1 )
UNIFORM BOUNDS FOR HILBERT COEFFICIENTS OF PARAMETERS 117
21

by Proposition 4.2, whence


U(ad−1 ) ∩ Q = (ad−1 ).
Let B = A/U(ad−1 ). We then have
d−3 
 
d−4 j
e2QB (B) = h (B)
j=1
j−1

for the same reason as for the equality



d−3 
d−4 j
e2QC (C) = h (C)
j=1
j−1
d−3 d−4
(in fact, to show e2QC (C) = j=1 j−1 hj (C), we only need that a1 is superficial
with respect to Q). Therefore, by the hypothesis of induction on d, we may choose
elements β2 , β3 , · · · , βd ∈ B so that QB = (β2 , β3 , · · · , βd )B and the sequence
β2 , β3 , · · · , βd forms a d-sequence in B. We put b1 = ad−1 and write βj = bj
with bj ∈ Q for 2 ≤ j ≤ d, where bj denotes the image of bj in B. We now put
q = (b1 , b2 , · · · , bd ). Then q is a parameter ideal in A and because U(b1 )∩Q = (b1 ),
we get
Q ⊆ [q + U(b1 )] ∩ Q = q + [U(b1 ) ∩ Q] ⊆ q + (b1 ) = q ;
hence Q = q . Thus the sequence b2 , b3 , · · · , bd forms a d-sequence in A/(b1 ), so
that b1 , b2 , · · · , bd forms a d-sequence in A, because b1 is A-regular. This complete
the proof of Theorem 4.1 and that of Theorem 1.2 as well. 

Let A be a generalized Cohen-Macaulay ring with d = dim A ≥ 3 and Q a


parameter ideal in A. Then by [S, Korollar 3.2]
d−1 
 
d−2 j
e1Q (A) = − h (A) and
j=1
j−1


d−2 
d−3 j
e2Q (A) = h (A),
j=1
j−1
if Q is standard.  We conversely have that Q is standard, if depth A > 0 and
d−1
e1Q (A) = − j=1 d−2 j 2
j−1 h (A) ([GO, Theorem 2.1]). This is no more true for eQ (A);
  
Q is not necessarily standard, even though the equality e2Q (A) = d−2 d−3 j
j=1 j−1 h (A)
holds in Theorem 1.2. Let us note one example.
Example 4.3. Let A be a 3-dimensional generalized Cohen-Macaulay ring with
depth A = 2. Assume that mH2m (A) = (0) and let a ∈ m such that a is A-regular but
aH2m (A) = (0). Choose b, c ∈ A so that b, c form a standard system of parameters
in A/(a). Let Q = (a, b, c). Then e2Q (A) = 0 = h1 (A) by Proposition 3.4 (2), since
a, b, c is a d-sequence in A. However Q is not standard, since Q·H2m (A) = (0).
For a concrete example, let R be a 3-dimensional regular local ring with max-
imal ideal n and let M = SyzR 2 2
2 (R/n ) be the second syzygy module of R/n . We
look at the idealization A = R  M of M over R. Then A is a 3-dimensional
generalized Cohen-Macaulay ring with depth A = 2 and H2m (A) ∼ = H2n (M ) ∼
= R/n2 ,
so that mHm (A) = (0).
2
118
22 SHIRO GOTO AND KAZUHO OZEKI

Acknowledgements
The authors are most grateful to Hoang Le Truong and Ngo Viet Trung for their
inspiring discussions during the 5-th Japan-Vietnam Joint Seminar on Commutative
Algebra (January 5-9, 2010, Institute of Mathematics Hanoi). Our Theorem 1.1 is
deep in debt from their suggestions.

References
[B] C. Blancafort, On Hilbert functions and cohomology, J. Algebra 192 (1997), 439–
459.
[BH] W. Bruns and J. Herzog, Cohen-Macaulay rings, Cambridge studies in advanced
mathematics, 39, Cambridge University Press, Cambridge, New York, Port Chester,
Melborne Sydney, 1993.
[GhGHOPV] L. Ghezzi, S. Goto, J. Hong, K. Ozeki, T. T. Phuong, and W. V. Vasconcelos,
Cohen–Macaulayness versus the vanishing of the first Hilbert coefficient of param-
eter ideals, J. London Math. Soc. (2) 81 (2010), 679–695.
[GNi] S. Goto and K. Nishida, Hilbert coefficients and Buchsbaumness of associated graded
rings, J. Pure and Appl. Algebra, 181 (2003), 61–74.
[GO] S. Goto and K. Ozeki, Buchsbaumness in local rings possessing first Hilbert coeffi-
cients of parameters, Nagoya Math. J. 199 (2010), 95–105.
[H] L. T. Hoa, Reduction numbers and Rees Algebras of powers of ideal, Proc. Amer.
Math. Soc. 119 (1993), 415–422.
[HH] F. Hayasaka and E. Hyry, On the Buchsbaum-Rim function of a parameter module,
J. Algebra 327 (2011), 307–315.
[LT] C. H. Linh and N. V. Trung, Uniform bounds in generalized Cohen-Macaulay rings,
J. Algebra 304 (2006), 1147–1159.
[MSV] M. Mandal, B. Singh, and J. K. Verma, On some conjectures about the Chern
numbers of filtrations, J. Algebra 325 (2011), 147–162.
[S] P. Schenzel, Multiplizitäten in verallgemeinerten Cohen-Macaulay-Moduln, Math.
Nachr. 88 (1979), 295–306.
[STC] P. Schenzel, N. V. Trung, and N. T. Cuong, Verallgemeinerte Cohen-Macaulay-
Moduln, Math. Nachr. 85 (1978), 57–73.

Department of Mathematics, School of Science and Technology, Meiji University,


1-1-1 Higashi-mita, Tama-ku, Kawasaki 214-8571, Japan
E-mail address: goto@math.meiji.ac.jp

Organisation for the Strategic Coordination of Research and Intellectual Prop-


erty, Meiji University, 1-1-1 Higashi-mita, Tama-ku, Kawasaki 214-8571, Japan
E-mail address: kozeki@math.meiji.ac.jp
Contemporary Mathematics
Volume 555, 2011

Absolute Integral Closure

Craig Huneke

Abstract. This paper is an expanded version of three lectures the author gave
during the summer school, “PASI: Commutative Algebra and its Connections
to Geometry,” from August 3-14, 2009 held in Olinda, Brazil.

Contents
1. Introduction
2. Basic Concepts
3. Dimension One
4. Regular Sequences
5. R+ is Cohen-Macaulay in Positive Characteristic
6. Applications
References

1. Introduction

This paper is based on three talks given at the PASI conference in Olinda,
Brazil in the summer of 2009. One point of the paper is to introduce students
to one aspect of characteristic p methods in commutative algebra. Such methods
have been among the most powerful in the field. The basic method of reduction to
characteristic p is used to prove results for arbitrary Noetherian rings containing
a field; the field can be characteristic 0. Thus, even though one is often working
in positive characteristic, one’s main interest might well be in rings containing the
rationals. In the late 1980’s, “tight closure theory” was discovered by M. Hochster
and myself. This theory synthesized many existing reduction to characteristic p
proofs into one theory, which has now grown to encompass a large number of
different directions. In this paper, we’ll concentrate on a result which Karen Smith

2010 Mathematics Subject Classification. Primary 13A35, 13B22, 13C14, 13D02, 13D45.
Key words and phrases. absolute integral closure, Cohen-Macaulay.
The author was partially supported by the NSF, DMS-0756853.

c 2011
XXXX
c American Mathematical Society

1
119
120
2 CRAIG HUNEKE

has refered to as a “crown jewel” of tight closure theory, namely the fact that
the absolute integral closure of a complete local domain of positive characteristic
is Cohen-Macaulay. We first introduce basic concepts, and discuss some rather
amazing properties of absolute integral closures. Then we prove some classical
theorems in dimension one. The main part of the paper will give a recent proof of
the Cohen-Macaulayness of the absolute integral closure in positive characteristic.
Applications will be given in Section 6. Some further thoughts about directions to
go will be presented in the last section. For unexplained results or definitions, we
refer the reader to Eisenbud’s book [7].

2. Basic Concepts
We begin with a definition:
Definition 2.1. Let R be a domain, and let R ⊂ S, a ring. The integral
closure of R in S is the set of all elements s ∈ S which satisfy an equation of the
form
sn + r1 sn−1 + ... + rn = 0,
where ri ∈ R for all i.
The set of elements in S which are integral over R form a ring, called the
integral closure of R in S. When S is the fraction field of R, the resulting ring
is called the integral closure of R. The ring R is said to be integrally closed if R
is equal to its integral closure. For example, a well-known result in commutative
algebra says that every UFD is integrally closed. So, e.g., the integers are integrally
closed. Another basic result states that if R is integrally closed, so is the ring of
polynomials R[X] as well as the ring of formal power series R[[X]]. An obvious
induction shows that also R[X1 , ..., Xn ] and R[[X1 , ..., Xn ]] are integrally closed for
all n ≥ 1.
The main object we will study is given in the following definition.
Definition 2.2. Let R be an integral domain with fraction field K. Let K
be a fixed algebraic closure of K. The integral closure of R in K, denoted R+ , is
called the absolute integral closure of R.
We simply say “R-plus” to mean the absolute integral closure. This ring has
been well-studied in several different contexts.
For other interesting results concerning R+ which we will not be writing about,
see [1],[2], and [24].

Discussion 2.3. Whenever we study R+ , we might as well assume that R is


integrally closed itself; if S is the integral closure of R, then R ⊂ S ⊂ R+ , and
S + = R+ . More generally, if S is any finite integral extension of R which is a
domain, then the fraction field of S is algebraic over the fraction field of R, and
hence there is an isomorphic copy of S which contains R and sits inside R+ . By
an abuse of language, we’ll just assume that S ⊂ R+ . In this case, S + = R+ .
Studying R+ is basically studying all integral extensions of R at the same time.

Remark 2.4. Continuing the discussion from above, if R is local and complete
in the m-adic topology and contains a field, then R is actually a finite integral
extension of a formal power series ring over a field k. Namely, the Cohen structure
ABSOLUTE INTEGRAL CLOSURE 121
3

theorem gives that there exists a field k inside R such that the composition of
maps k → R → R/m is an isomorphism. Moreover, if x1 , ..., xd is a system of
parameters for R, then the complete local subring of R generated by these elements,
k[[x1 , ..., xd ]] = A is isomorphic with a formal power series ring over k, and R is
module-finite over A. Hence A+ = R+ . Thus, if we fix a copy of the residue field,
say k, sitting inside R, there is really only one R+ for each dimension; it is the
absolute integral closure of the formal power series ring over the field k. In other
words, for each dimension, we are really just studying one ring, the absolute integral
closure of the power series ring over k. There is a classical theorem which gives
the structure of such a ring in dimension one over a field of characteristic 0. See
Theorem 3.1.

Discussion 2.5. It is not difficult to prove that if R is an integral domain, then


a domain S integral over R is isomorphic with R+ if and only if monic polynomials
over S factor into monic linear polynomials over S. Using this criterion, one can
easily prove two very basic properites of R+ .
• Firstly, if q ∈ Spec(R), then (Rq )+ ∼
= (R+ )q , where the latter localization is
at the multiplicatively closed set R − q.
• Secondly, if Q ∈ Spec(R+ ), then R+ /Q ∼= (R/(Q∩R))+ . Thus every quotient
+
of R by a prime ideal is itself the absolute integral closure of an appropriate
quotient of R.
We leave these statements as exercises below.
These properties are very useful in understanding R+ . For example, if we
replace R by (R/P )Q for primes P ⊂ Q, then the plus closure of this new ring
is simply (R+ /P  )Q for any prime P  of R+ which lies over P . To give a specific
example, the main result of Section 4 states that the absolute integral closure of
an excellent local domain in positive characteristic is Cohen-Macaulay in a suitable
sense. The second property discussed above then says that the same is true modulo
every prime ideal of the absolute integral closure! This is remarkable.
M. Artin [3] proved another amazing property of R+ :
Theorem 2.6. Let R be a domain. If P, Q are prime ideals in R+ , then either
P + Q = R+ , or P + Q is prime.
This is very far from what happens in rings we typically study. For example
in the polynomial ring k[x, y], the ideals P = (x + y 2 ) and Q = (x) are prime, but
the sum is not. The proof we give is taken from [11].
Proof. To prove Artin’s result, let ab ∈ P + Q. Set z = b − a. Notice that
a2 + za = ab = u + v for some u ∈ P, v ∈ Q. The equation X 2 + zX = u has a
solution x ∈ R+ . Since u ∈ P , it follows that either x + z ∈ P or x ∈ P . But
(x2 + zx) − (a2 + za) = u − (u + v) ∈ Q, so either x − a ∈ Q or x + a + z ∈ Q. In
each of the four different cases, we obtain that either a ∈ P + Q or b ∈ P + Q. For
example, if x + z ∈ P and x − a ∈ Q, then b = (x + z) − (x − a) ∈ Q. 

Notice that if R is complete and local, then every finite extension domain of
R is also local. In particular, R+ is a local ring, i.e., has a unique maximal ideal.
122
4 CRAIG HUNEKE

Therefore the possibility that P + Q = R cannot happen in this case, and the sum
of every set of prime ideals is prime.

Exercises

1. Let R be a domain with fraction field K, and let L be an algebraic extension


of K. Prove that L is algebraically closed if and only if every monic polynomial
with coefficients in R has a root in L.
2. Let R be an integral domain. Prove that a domain S integral over R
is isomorphic with R+ if and only if every monic polynomial over S factors into
monic linear polynomials over S.
3. Prove that Artin’s theorem can be extended to primary ideals: the sum of
any two primary ideals in R+ is again primary or the whole ring.
4. Let R be a domain of positive dimension. Prove that R+ is not Noetherian.
5. Suppose that R is an integral domain, and R ⊆ S ⊆ R+ . Prove that
R = S+.
+

6. If q ∈ Spec(R), then (Rq )+ = ∼ (R+ )q , where the latter localization is at the


multiplicatively closed set R − q.
7. If Q ∈ Spec(R+ ), then R+ /Q = ∼ (R/(Q ∩ R))+ . Thus every quotient of R+
by a prime ideal is itself the absolute integral closure of an appropriate quotient of
R.

3. Dimension One

Let k be a field of characteristic 0, and consider the formal power series ring
k[[t]], whose fraction field is denoted k((t)). A famous theorem due to Puiseux, but
apparently known to Newton, is that the algebraic closure of the field k((t)) is the
union of the fields k((t1/n )) for n ≥ 1. A recent paper by Kedlaya ([15]) gives a
characterization of this algebraic closure in the case k has positive characteristic; it
is difficult to describe. Chevalley [6] pointed out that the  Artin-Schreirer polyno-
mial xp − x − t−1 has no root in the Newton-Puiseux field n k((t1/n )). In fact, as
we shall see, there are remarkable differences in R+ depending on the characteristic
of the ground field.
We give a proof of the Newton-Puiseux theorem, following a treatment in [17].
Theorem 3.1. Let k be an algebraically closed field of characteristic 0, and let
k((t)) be the fraction field of the formal power series ring k[[t]]. Then an algebraic
closure of k((t)) is the Newton-Puiseux field ∪n k((t1/n )).
Proof. We need to prove that every monic polynomial P (z) ∈ B[z] is re-
ducible, where B = ∪n k[[t1/n ]]. Write P = z n + a1 z n−1 + ... + an , where ai ∈ B.
By using a transformation z  = z + n1 a1 , we can assume without loss of generality
that a1 = 0. This is called the “Tschirnhausen transformation”. Furthermore, by
replacing k[[t]] by k[[t1/n ]] for some large n, we can change variables and assume
that all ai ∈ k[[t]]. Let rk = ord(ak ), so that ak = trk uk (t), where uk (t) is a unit.
Here “ord” denotes the t-adic order of an element.
ABSOLUTE INTEGRAL CLOSURE 123
5

We want to factor P (z), and to do so we will use Hensel’s lemma1 ; it suffices


to factor P (z) into two relatively prime polynomials after going modulo t. This is
always possible unless after reduction mod t, P (z) becomes of the form (t − α)n
for some α ∈ k. However, since a1 = 0, the only possible such α is 0. Thus we
are done unless for every k, 2 ≤ k ≤ n, ai (0) = 0. We want to make a change of
variables where this does not occur.
Set z = tr y, for some rational r to be chosen later. Substituting, we see that
P (z) = trn y n + a2 tr(n−2) y n−2 + ... + an . We wish to factor out trn from every
term in such a way that at least one term becomes a unit. The power of t dividing
the ith term is tri t(n−i)r , so what we need to do is to choose r in such a way that
ri + (n − i)r ≥ rn, i.e., so that ri ≥ ri. We set r = min{ rii }. Then we can rewrite
P (z) = trn Q(y, t), where Q = y n + b2 y n−1 + ... + bn . By again replacing t by t1/m
for suitably large m, we can assume that each bj ∈ k[[t]]. But now for at least one
bi , bi (0) = 0; this occurs for every term where ri /i = r. It follows that we can
factor Q over the residue field into relatively prime polynomials, and by Hensel’s
Lemma this lifts to a factorization of Q and hence also of P . 

This theorem has the following almost immediate corollary:


Corollary 3.3. Let k be an algebraically closed field of characteristic 0. Then
k[[t]]+ = ∪n k[[t1/n ]].
Proof. The ring on the right side of the equation is clearly integral over
k[[t]], and the fraction field of it is clearly the Newton-Puiseux field, which is an
algebraic closure of k((t)). To finish the proof of the corollary, we need to prove
that ∪n k[[t1/n ]] is integrally closed. But since it is a union of discrete valuation
rings, each integrally closed, it is also. 

Example 3.4. As an example, consider the equation X 2 − X = t−1 . Using the


quadratic formula and Taylor series shows that the roots of this polynomial are a
power series in t−1/2 . One can solve recursively for the coefficients.

The situation in positive characteristic is drastically different. We use the


discussion in the paper [15] in what follows. A generalized power series is an
expression of the form, i∈Q ci ti , with ci ∈ k, such that the set of i with xi = 0
is a well-ordered subset of Q, i.e., every non-empty subset has a least element.
Such generalized power series form a ring in a natural way; it makes sense to
multiply and add them in the obvious way. Abyhankar pointed out that with this
generalization of the idea of a power series, the Chevalley polynomial has the root
t−1/p + t−1/p + ....
2

Are there enough generalized power series to obtain an algebraic closure of


k((t))? The following result was proved independently by Huang [14], Rayner [19],
and Ştefănescu [25]:

1 The version of Hensel’s lemma we use is the following theorem:


Theorem 3.2. Let (R, m) be a local ring which is complete in the m-adic topology. If f (T )
is a monic polynomial with coefficients in R such that f (T ) factors modulo m into the product
of two relatively prime monic polynomials of positive degree, then this factorization lifts to a
factorization of f (T ) into two monic polynomials.
124
6 CRAIG HUNEKE

Theorem 3.5. Let k be an algebraically closed field of positive


characteristic,
and let K be the set of generalized power series of the form f = i∈S ci ti , with
ci ∈ k, where the set S has the following properties:
(1) S is a subset of Q which depends on f .
(2) Every nonempty subset of S has a least element.
(3) There exists a natural number m such that every element of mS has denom-
inator a power of p.
Then K is an algebraically closed field containing k((t)).
Kedlaya [15] gives a construction of the algebraic closure of k((t)) when k is
algebraically closed of positive characteristic. By the above theorem, one needs to
identify generalized power series which are algebraic over k((t)). A glimpse of the
issues which arise can be seen in the following result of Huang [14] and Ştefănescu
[25]:

Theorem 3.6. The series ∞ i=0 ci t
−1/pi
with ci ∈ Fp , the algebraic closure of
the field with p elements, is algebraic over Fp ((t)) if and only if the sequence {ci }
is eventually periodic.
A consequence of the main result of [15] is the following nice theorem, which
perhaps can be proved directly.
Theorem
 3.7. Let k be an algebraically closed field of positive characteristic,
and let i ci ti be a generalized
 power series which is algebraic over k((t)). Then
for every real number α, i<α ci ti is also algebraic over k((t)).
We now move away from the local case. Another classical result in dimension
one concerns the ring of all algebraic integers, Z+ . To prove this theorem, we first
recall some results about class groups.
Let D be a ring of algebraic integers. This means that the fraction field K of
D is a finite extension of Q, and D is the integral closure of Z in K. Necessarily D
is a one-dimensional integrally closed domain, i.e., a Dedekind domain. Moreover,
a classical result is that D has a torsion class group. In particular this means that
every ideal I in D has a power which is a principal ideal. This fact leads to the
following theorem about Z+ .
Theorem 3.8. Every finitely generated ideal of Z+ is principal (a domain with
this property is said to be a Bézout domain).
Proof. Let I ⊂ Z+ be generated by t1 , ..., tm . There is a finite field extension
L of Q containing all of these elements. The integral closure of Z in L, say D, is a
Dedekind domain which contains t1 , ..., tm . Let J be the ideal they generate in D.
By the discussion above, some power of J is principal, say J n = (d). We claim then
1 
that d n Z+ = I. Let i ∈ I. We can write i = j ej tj , where the ej ∈ Z+ . Again,
there is a finite extension T of D, with T a Dedekind domain, such that i ∈ I ∩ T =
(t1 , ..., tm )T = JT . In a Dedekind domain there is unique factorization into prime
ideals. Write JT = P1a1 · · · Pkak . Then J n T = dT = P1na1 · · · Pknak . By unique
1 1
factorization, it follows that d n T = P1a1 · · · Pkak = JT , so that i ∈ d n Z+ . 
ABSOLUTE INTEGRAL CLOSURE 125
7

4. Regular Sequences

We summarize some of the basic notions we will use to analyze R+ .


Definition 4.1. A sequence of elements x1 , ..., xd in a ring R is said to be a
regular sequence if rxi ∈ (x0 , ..., xi−1 ) implies that r ∈ (x0 , ..., xi−1 ) for all 1 ≤ i ≤
d − 1 (here we set x0 = 0), and (x1 , ..., xd )R = R.
Another definition we will need is the following.
Definition 4.2. Let (R, m) be a Noetherian local ring of dimension d. Ele-
ments x1 , ..., xd are said to be a system of parameters if the nilradical of the ideal
they generate is m. A Noetherian local ring is said to be Cohen-Macaulay if some
(equivalently every) system of parameters forms a regular sequence.
We are aiming at a theorem which shows that in characteristic p, R+ has regular
sequences having maximal length. It turns out that this is equivalent to being flat
in an appropriate sense. The next proposition proves this.

Proposition 4.3. Let A = k[[x1 , ..., xd ]], where k is a field of characteristic


p > 0. Then the following two conditions are equivalent:
(1) x1 , ..., xd form a regular sequence on A+ (by an abuse of language we say
that A+ is Cohen-Macaulay).
(2) A+ is flat over A.
Proof. We prove the equivalence. First assume (1). If A+ is not flat over A,
choose i ≥ 1 as large as possible so that TorA i (A/P, A ) = 0 for some prime P in
+

A. Such a choice is possible because A is regular and large Tors vanish. If y1 , ..., ys
is a maximal regular sequence in P , then one can embed A/P in A/(y1 , ..., ys )
with cokernel C. But since our assumption forces TorR +
i+1 (C, A ) = 0 (as C has
+
a prime filtration), and y1 , ..., ys form a regular sequence on A , we obtain that
TorA +
i (A/P, A ) = 0, a contradiction.
To see that y1 , ..., ys form a regular sequence, extend them to a system of
parameters, and let B = k[[y1 , ..., yd ]]. Then B + = A+ , and our hypothesis says
that the y  s form a regular sequence.
Assume (2). Flat maps preserve regular sequence in general. 
Our method of studying regular sequences relies on local cohomology. We only
need the description below.
For x ∈ R, let C • (x; R) denote the complex 0 → R → Rx → 0, graded so that
the degree 0 piece of the complex is R, and the degree 1 is Rx . If x1 , ..., xn ∈ R,
let C • (x1 , x2 , ..., xn ; R) denote the complex C • (x1 ; R) ⊗R ... ⊗R C • (xn ; R), where in
general recall that if (A• , dA ) and (B • , dB ) are complexes, then the tensor product
of these  complexes, (A ⊗R B, Δ) is by definition the complex whose ith graded
piece is j+k=i Aj ⊗ Bk and whose differential is determined by the map from
Aj ⊗Bk → (Aj+1 ⊗Bk )⊕(Aj ⊗Bk+1 ) given by Δ(x⊗y) = dA (x)⊗y+(−1)k x⊗dB (y).
The modules in this complex, called the C̆ech cohomology complex, are
 
0→R→⊕ Rxi → ⊕ Rxi xj → ... → Rx1 x2 ···xn → 0
i i<j

where the differentials are the natural maps induced from localization, but with
signs attached. If M is an R-module, we set C • (x1 , x2 , ..., xn ; M ) = C • (x1 , x2 , ..., xn ; R)⊗R
126
8 CRAIG HUNEKE

M . We denote the cohomology of C • (x1 , x2 , ..., xn ; M ) by HIi (M ), called the ith


local cohomology of M with respect to I = (x1 , ..., xd ). It is a fact that this module
only depends on the ideal generated by the xi up to radical. We summarize some
useful information concerning these modules.
Proposition 4.4. Let R be a Noetherian ring, I and ideal and M an R-module.
Let ϕ : R → S be a homomorphism and let N be an S-module.
(1) If ϕ is flat, then HIj (M )⊗R S ∼ j
= HIS (M ⊗R S). In particular, local cohomology
commutes with localization and completion.
(2) (Independence of Base) HIj (N ) ∼ j
= HIS (N ), where the first local cohomology
is computed over the base ring R.
Proof. Choose generators x1 , ..., xn of I. The first claim follows at once from
the fact that C • (x1 , ..., xn ; M ) ⊗R S = C • (ϕ(x1 ), ..., ϕ(xn ); M ) ⊗R S), and that S
is flat over R, so that the cohomology of C • (x1 , ..., xn ; M ) ⊗R S is the cohomology
of C • (x1 , ..., xn ; M ) tensored over R with S.
The second claim follows from the fact that
C • (x1 , ..., xn ; N ) = C • (x1 , ..., xn ; R) ⊗R N = (C • (x1 , ..., xn ; R) ⊗R S) ⊗S N
= C • (ϕ(x1 ), ..., ϕ(xn ); S) ⊗R N = C • (ϕ(x1 ), ..., ϕ(xn ); N ).


5. R+ is Cohen-Macaulay in Positive Characteristic

Let R be a commutative ring containing a field of characteristic p > 0, let I ⊂ R


be an ideal, and let R be an R-algebra. The Frobenius ring homomorphism f :
r→r p
R −→ R induces a map f∗ : HIi (R )−→HIi (R ) on all local cohomology modules
of R called the action of the Frobenius on HIi (R ). For an element α ∈ HIi (R ) we
denote f∗ (α) by αp . This follows since the Frobenius extends to localization of R
in the obvious way, and commutes with the maps in the C̆ech cohomology complex,
which are simply signed natural maps.
The main result is that if R is a local Noetherian domain which is a homomor-
phic image of a Gorenstein local ring and has positive characteristic, then R+ is
Cohen-Macaulay in the sense that every system of parameters of R form a regular
sequence in R+ . To prove this result we use the proof given in [13]. The original
proof, with slightly different assumptions, was given in 1992 in [11], as a result
of developments from tight closure theory. Although tight closure has now disap-
peared from the proof, it remains an integral part of the theory. A critical point is
that we must find some way of annihilating nonzero local cohomology classes. The
next lemma is essentially the only way known to do this.
Lemma 5.1. Let R be a commutative Noetherian domain containing a field of
characteristic p > 0, let K be the fraction field of R and let K be the algebraic
closure of K. Let I be an ideal of R and let α ∈ HIi (R) be an element such that
2 t
the elements α, αp , αp , . . . , αp , . . . belong to a finitely generated R-submodule of
HIi (R). There exists an R-subalgebra R of K (i.e. R ⊂ R ⊂ K) that is finite
as an R-module and such that the natural map HIi (R)−→HIi (R ) induced by the
natural inclusion R−→R sends α to 0.
ABSOLUTE INTEGRAL CLOSURE 127
9

i=t pi
Proof. Let At = i=1 Rα be the R-submodule of HIi (R) generated by
t
α, αp , . . . , αp . The ascending chain A1 ⊂ A2 ⊂ A3 ⊂ . . . stabilizes because R is
Noetherian and all At sit inside a single finitely generated R-submodule of HIi (R).
s
Hence As = As−1 for some s, i.e. αp ∈ As−1 . Thus there exists an equation
s s−1 s−2
αp = r1 αp + r2 αp + · · · + rs−1 α with ri ∈ R for all i. Let T be a variable and
s s−1 s−2
let g(T ) = T p − r1 T p − r2p − · · · − rs−1 T . Clearly, g(T ) is a monic polynomial
in T with coefficients in R and g(α) = 0.
Let x1 , . . . , xd ∈ R generate the ideal I. Recall that we can calculate the local
cohomology from the C̆ech cohomology complex C • (R),
di−1 d
0−→C 0 (R)−→ . . . −→C i−1 (R) −→ C i (R) −→
i
C i+1 (R)−→ . . . −→C d (R)−→0
where C 0 (R) = R and C i (R) = ⊕1≤j1 <···<ji ≤d Rxj1 ···xji , and HIi (M ) is the ith
cohomology module of C • (R).
Let α̃ ∈ C i (R) be a cycle (i.e. di (α̃) = 0) that represents α. The equality
g(α) = 0 means that g(α̃) = di−1 (β) for some β ∈ C i−1 (R). Since C i−1 (R) =
rj ,...,j
⊕1≤j1 <···<ji−1 ≤d Rxj1 ···xji−1 , we may write β = ( e11 ei−1 i−1 ) where rj1 ,...,ji−1 ∈ R,
xj ···xj
1 i−1
rj1 ,...,ji−1
the integers e1 , . . . , ei−1 are non-negative, and e e ∈ Rxj1 ···xji−1 .
xj 1 ···xj i−1
1 i−1
Zj1 ,...,ji−1 rj1 ,...,ji−1
Consider the equation g( e e )− e e = 0 where Zj1 ,...,ji−1 is a vari-
xj 1 ···xj i−1 xj 1 ···xj i−1
1 i−1 1 i−1
ei−1 ps
able. Multiplying this equation by (xej11 · · · xji−1 )
produces a monic polynomial
equation in Zj1 ,...,ji−1 with coefficients in R. Let zj1 ,...,ji−1 ∈ K be a root of this
equation and let R be the R-subalgebra of K generated by all the zj1 ,...,ji−1 s, i.e.
by the set {zj1 ,...,ji−1 |1 ≤ j1 < · · · < ji−1 ≤ d}. Since each zj1 ,...,ji−1 is integral over
R and there are finitely many zj1 ,...,ji−1 s, the R-algebra R is finite as an R-module.
˜ = ( zej1 ,...,jei−1
Let α̃ i−1 ) ∈ C
i−1
(R ). The natural inclusion R−→R makes C • (R)
xj ···xj
1
1 i−1

into a subcomplex of C • (R ) in a natural way, and we identify α̃ ∈ C i (R) and


β ∈ C i−1 (R) with their natural images in C i (R ) and C i−1 (R ) respectively. With
this identification, α̃ ∈ C i (R ) is a cycle representing the image of α under the
natural map HIi (R)−→HIi (R ), and so is α = α̃ − di−1 (α̃) ˜ ∈ C i (R ). Since g(α̃)˜ =
β and g(α̃) = di−1 (β), we conclude that g(α) = 0. Let α = (ρj1 ,...,ji ) where
ρj1 ,...,ji ∈ Rxj ···xj . Each individual ρj1 ,...,ji satisfies the equation g(ρj1 ,...,ji ) = 0.
1 i
Since g(T ) is a monic polynomial in T with coefficients in R, each ρj1 ,...,ji is an
element of the fraction field of R that is integral over R. Let R be obtained from
R by adjoining all the ρj1 ,...,ji . This subcomplex is exact because its cohomology
groups are the cohomology groups of R with respect to the unit ideal. Since α is
a cycle and belongs to this exact subcomplex, it is a boundary, hence it represents
the zero element in HIi (R ). 
We recall that for a Gorenstein local ring A of dimension n, local duality says
that there is an isomorphism of functors D(Extn−i ∼ i
A (−, A)) = Hm (−) on the category
of finite A-modules, where D = HomA (−, E) is the Matlis duality functor (here E
is the injective hull of the residue field of A in the category of A-modules).
Theorem 5.2. [13] Let R be a commutative Noetherian local domain containing
a field of characteristic p > 0, let K be the fraction field of R and let K be the
algebraic closure of K. Assume R is a homomorphic image of a Gorenstein local
128
10 CRAIG HUNEKE

ring A. Let m be the maximal ideal of R. Let i < dim R be a non-negative integer.
There is an R-subalgebra R of K (i.e. R ⊂ R ⊂ K) that is finite as an R-module
and such that the natural map Hmi
(R)−→Hm i
(R ) is the zero map.

Proof. The proof comes from [13]. Let n = dim A and let N = Extn−i A (R, A).
Clearly N is a finite A-module.
Let d = dim R. We use induction on d. For d = 0 there is nothing to prove,
so we assume that d > 0 and the theorem proven for all smaller dimensions. Let
P ⊂ R be a non-maximal prime ideal. We claim there exists an R-subalgebra RP
of K such that RP is a finite R-module and for every RP -subalgebra R∗ of K (i.e.
RP ⊂ R∗ ⊂ K) such that R∗ is a finite R-module, the image I ⊂ N of the natural
∗ ∗
map Extn−iA (R , A)−→N induced by the natural inclusion R−→R vanishes after
localization at P , i.e. IP = 0. Indeed, let dP = dim R/P . Since P is different from
the maximal ideal, dP > 0. As R is a homomorphic image of a Gorenstein local
ring, it is catenary, hence the dimension of RP equals d − dP , and i < d implies
i−dP < d−dP = dim RP . By the induction hypothesis applied to the local ring RP ,
there is an RP -subalgebra R̃ of K, which is finite as an RP -module, such that the
natural map HPi−dP (RP )−→HPi−dP (R̃) is the zero map. Let R̃ = RP [z1 , z2 , . . . , zt ],
where z1 , z2 , . . . , zt ∈ K are integral over RP . Multiplying, if necessary, each zj
by some element of R \ P , we can assume that each zj is integral over R. We
set RP = R[z1 , z2 , . . . , zt ]. Clearly, RP is an R -subalgebra of K that is finite as
R-module.
Now let R∗ be both an RP -subalgebra of K (i.e. RP ⊂ R∗ ⊂ K) and a finite
R-module. The natural inclusions R−→RP −→R∗ induce natural maps

Extn−i n−i P
A (R , A)−→ExtA (R , A)−→N.

This implies that I ⊂ J , where J is the image of the natural map


ϕ : Extn−iA (R , A)−→N . Hence it is enough to prove that JP = 0. Localiz-
P

ing this map at P we conclude that JP is the image of the natural map ϕP :
AP (R̃, AP )−→ExtAP (RP , AP ) induced by the natural inclusion RP −→R̃ (by
Extn−i n−i

a slight abuse of language we identify the prime ideal P of R with its full preimage
in A). Let DP (−) = HomAP (−, EP ) be the Matlis duality functor in the cate-
gory of RP -modules, where EP is the injective hull of the residue field of RP in
the category of RP -modules. Local duality implies that DP (ϕP ) is the natural
map HPi−dP (RP )−→HPi−dP (R̃) which is the zero map by construction (note that
i − dP = dim AP − (n − i)). Since ϕP is a map between finite RP -modules and
DP (ϕP ) = 0, it follows that ϕP = 0. This proves the claim.
Since N is a finite R-module, the set of the associated primes of N is finite.
Let P1 , . . . , Ps be the associated primes of N different from m. For each j let RPj
be an R-subalgebra of K corresponding to Pj , whose existence is guaranteed by the
above claim. Let R = R[RP1 , . . . , RPs ] be the compositum of all the RPj , 1 ≤ j ≤ s.
Clearly, R is an R-subalgebra of K. Since each RPj is a finite R-module, so is R.
Clearly, R contains every RPj . Hence the above claim implies that IPj = 0 for
every j, where I ⊂ N is the image of the natural map Extn−i A (R, A)−→N induced
by the natural inclusion R−→R. It follows that not a single Pj is an associated
prime of I. But I is a submodule of N , and therefore every associated prime of
I is an associated prime of N . Since P1 , . . . , Ps are all the associated primes of N
different from m, we conclude that if I = 0, then m is the only associated prime of
ABSOLUTE INTEGRAL CLOSURE 129
11

I. Since I, being a submodule of a finite R-module N , is finite, and since m is the


only associated prime of I, we conclude that I is an R-module of finite length.
Writing the natural map Extn−i
A (R, A)−→N as the composition of two maps

Extn−i
A (R, A)−→I−→N,

the first of which is surjective and the second injective, and applying the Matlis
i i
duality functor D, we get that the natural map ϕ : Hm (R)−→Hm (R) induced by
i i
the inclusion R−→R is the composition of two maps Hm (R)−→D(I)−→Hm (R),
the first of which is surjective and the second injective. This shows that the image
of ϕ is isomorphic to D(I) which is an R-module of finite length since so is I. In
particular, the image of ϕ is a finitely generated R-module. Let α1 , . . . , αs ∈ Hm i
(R)
generate Imϕ.
The natural inclusion R−→R is compatible with the Frobenius homomorphism,
i.e. with the raising to the pth power on R and R. This implies that ϕ is compatible
i i
with the action of the Frobenius f∗ on Hm (R) and Hm (R), i.e. ϕ(f∗ (α)) = f∗ (ϕ(α))
for every α ∈ Hm (R), which, in turn, implies that Imϕ is an f∗ -stable R-submodule
i

of Hm i
(R), i.e. f∗ (α) ∈ Imϕ for every α ∈ Imϕ. We finish the proof by applying
Lemma 5.1 to each element of a finite generating set α1 , ..., αs of Imϕ. Applying
Lemma 5.1 we obtain a R-subalgebra Rj of K (i.e. R ⊂ Rj ⊂ K) such that Rj is
i i
a finite R-module and the natural map Hm (R)−→Hm (Rj ) sends αj to zero. Let
R = R[R1 , . . . , Rs ] be the compositum of all the Rj . Then R is an R-subalgebra of


K and is a finite R-module since so is each Rj . The natural map Hm i


(R)−→Hm i
(R )
sends every αj to zero, hence it sends the entire Imϕ to zero. Thus the natural
map Hm i
(R)−→Hm i
(R ) is zero. 

Corollary 5.3. Let R be a commutative Noetherian local domain containing


a field of characteristic p > 0. Assume that R is a homomorphic image of a
Gorenstein local ring. Then the following hold:
i
(a) Hm (R+ ) = 0 for all i < dim R, where m is the maximal ideal of R.
(b) Every system of parameters of R is a regular sequence on R+ .
Proof. (a) R+ is the direct limit of the finitely generated R-subalgebras R ,
hence Hm i
(R+ ) = lim Hm i
(R ). But Theorem 5.2 implies that for each R there is
−→
R such that the map Hm (R )−→Hm
 i i
(R ) in the inductive system is zero. Hence
the limit is zero.
(b) Let x1 , ..., xd be a system of parameters of R. We prove that x1 , ..., xj is
a regular sequence on R+ by induction on j. The case j = 1 is clear, since R+
is a domain. Assume that j > 1 and x1 , . . . , xj−1 is a regular sequence on R+ .
i
Set It = (x1 , ..., xt ). The fact that Hm (R+ ) = 0 for all i < d and the short exact
sequences
xt
0−→R+ /It−1 R+ −→ R+ /It−1 R+ −→R+ /It R+ −→0
q
for t ≤ j − 1 imply by induction on t that Hm (R+ /(x1 , ..., xt )R+ ) = 0 for q < d − t.
In particular, Hm (R /(x1 , ..., xj−1 )R ) = 0 since 0 < d − (j − 1). Hence m is not
0 + +

an associated prime of R+ /(x1 , ..., xj−1 )R+ . This implies that the only associated
primes of R+ /(x1 , ..., xj−1 )R+ are the minimal primes of R/(x1 , ..., xj−1 )R. Indeed,
if there is an embedded associated prime, say P , then P is the maximal ideal of
the ring RP whose dimension is bigger than j − 1 and P is an associated prime of
(R+ /(x1 , ..., xj−1 )R+ )P = (RP )+ /(x1 , ..., xj−1 )(RP )+ which is impossible by the
130
12 CRAIG HUNEKE

above. Hence every element of m not in any minimal prime of R/(x1 , ..., xj−1 )R,
for example, xj , is a regular element on R+ /(x1 , ..., xj−1 )R+ . 

Discussion 5.4. It is important to understand the huge differences between


characteristic p, characteristic 0, and mixed characteristic. Suppose that k has
characteristic 0. Let R be a complete Noetherian local integrally closed domain
with residue field k, and fraction field K. If L is any finite field extension of K and
S is the integral closure of R in L, then the reduced trace map2 gives a splitting of
R from S, i.e., S ∼= R ⊕ N as an R-module for some module R-module N . Then
i i i i
Hm (R) splits out of Hm (S), so that the map Hm (R)−→Hm (S) is never zero unless
i
Hm (R) = 0. Thus the exact opposite holds in characteristic 0.
What happens in mixed characteristic is a great mystery. One of the great
results in recent years was that of Heitmann. He proved the following theorem:
Theorem 5.5. Let (R, m) be a complete three-dimensional Noetherian local
integrally closed domain of mixed characteristic p ∈ N. (This means that p ∈ m.)
1
Then for all n ≥ 1, p n annihilates the local cohomology Hm
2
(R+ ).
Heitmann’s theorem is slightly stronger than this, but this is the essential result
of [8]. In characteristic p we know this local cohomology module is zero, but this
is not known in mixed characteristic.
One can hope that if R has mixed characteristic p, then R+ /pR+ is Cohen-
Macaulay. Of course, this ring has positive characteristic. However, perhaps this is
too much to hope for. The next best result would be to conjecture that R+ / pR+
is Cohen-Macaulay, a question raised by Lyubeznik. He has a partial result in this
direction [16]:
Theorem 5.6. Let (R, m) be a Noetherian local excellent domain √ of mixed
characteristic
 p. Assume the dimension of R is at least 3. Set R = R/ pR and
R+ = R+ / pR+ . Then Hm 1
(R+ ) = 0, and every part of a system of parameters
a, b of R form a regular sequence on R+ .

Another interesting question is whether or not one must use inseparable exten-
sions to trivialize local cohomology. In fact, Anurag Singh [23] found the following
nice trick to change inseparable elements to separable.
Proposition 5.7. Let R be an excellent domain of characteristic p > 0, and
let I be an ideal of R. Suppose that z ∈ R is such that z q ∈ I [q] , where q = pe
is a power of p. Then there exists an integral domain S, which is a module-finite
separable extension of R, such that z ∈ IS.

The point here is that there clearly a finite inseparable extension of R,
say T ,
such that z ∈ IT . Simply take qth roots of the elements aj such that z = aj xqj
where xj ∈ I.

2 The reduced trace is defined as 1 T r


n L/K where L is a finite field extension of a field K, and
[L : K] = n. This map fixes the ground field K. If R is an integrally closed domain with fraction
field K and S is the integral closure of R in L, then the reduced trace sends S to R and fixes R.
ABSOLUTE INTEGRAL CLOSURE 131
13


Proof. Write z = 1≤j≤n aj xqj where xj ∈ I as above. Consider the equa-
tions for 2 ≤ i ≤ n,
Uiq + Ui xq1 − ai = 0.
These are monic separable equations and therefore have roots ui in a separable
field extension of the fraction field of R. Let S be the integral closure of the ring
R[u2 , ..., un ]. Since R is excellent, S is finite as an R-module. We claim that z ∈ IS.
Set 
u1 = (z − xi ui )/x1 .
2≤i≤n
Note that u1 is an element of the fraction field of S. Taking qth powers we see that

uq1 = a1 + ui xqi .
2≤i≤n

Therefore u1 is integral over S. As S is integrally closed, u1 ∈ S. This implies that



z= ui x i
1≤i≤n

and so z ∈ IS. 

Discussion 5.8. There is an interesting property pertaining to our main the-


orem. Suppose that (R, m) is a complete local Noetherian domain of positive char-
acteristic, and let x1 , ..., xd form a regular sequence. If x1 , ..., xd is a system of
parameters, and if R is not Cohen-Macaulay, then there is a non-trivial relation
r1 x1 + ... + rd xd = 0. Non-trivial means that it does not come from the Koszul
relations. Since R+ is Cohen-Macaulay, we can trivialize this relation in R+ , and
therefore in some finite extension ring S of R, R ⊆ S ⊆ R+ . But Theorem 5.2 does
not say whether or not there is a fixed finite extension ring T , R ⊆ T ⊆ R+ in
which all relations on all parameters of R become simultaneously trivial. Even if
such a ring T exists, this does not mean T is itself Cohen-Macaulay; new relations
coming from elements of T may be introduced. However, there is a finite extension
which simultaneously trivializes all relations on systems of parameters. This fact
has been proved by Melvin Hochster and Yongwei Yao [12].

6. Applications

The existence of a big Cohen-Macaulay algebra has a great many applications.


In some sense it repairs the failure of a ring to be Cohen-Macaulay. Hochster
proved and used the existence of big Cohen-Macaulay modules (the word “big”
refers to the fact the modules may not be finitely generated) to prove many of the
homological conjectures. For a modern treatment, see [10]. In general, if you can
prove a theorem in the Cohen-Macaulay case, a good strategy is to immediately
try to use R+ to prove it in general. We give several examples of this phenomena
in this section. As examples, we will prove some of the old homological conjectures
using this approach; this is not new, but there are currently a growing number of
new homological conjectures, and it could be that characteristic p methods apply.
Of course, some of the homological conjectures deal directly with systems of
parameters. These are easy to prove once one has a Cohen-Macaulay module. For
example, the next theorem gives the monomial conjecture.
132
14 CRAIG HUNEKE

Theorem 6.1. Let R be a local Noetherian ring of dimension d and positive


characteristic p. Let x1 , ..., xd be a system of parameters. Then for all t ≥ 1,
(x1 · · · xd )t is not in the ideal generated by xt+1 t+1
1 , ..., xd .

Proof. We use induction on the dimension d of R. The case d = 1 is trivial.


Suppose by way of contradiction that d > 1 and (x1 · · · xd )t ∈ (xt+1 t+1
1 , ..., xd ). This
is preserved after completion, and is further preserved after moding out a minimal
prime P such that the dimension of the completion modulo P is still d. After these
operations, the images of the elements xi still form a system of parameters as well.
Thus we may assume that R is a complete local domain. We apply Theorem 5.2 to
conclude that x1 , ..., xd is a regular sequence in R+ . Write

(x1 · · · xd )t = si xt+1
i ,
i

where si ∈ R. Then xtd ((x1 · · · xd−1 ) − sd xd ) ∈ (xt+1


t t+1
1 , ..., xd−1 ). Since the powers
of the xi also form a regular sequence in R+ , we conclude that (x1 · · · xd−1 )t −
sd xd ∈ (xt+1 t+1 +
1 , ..., xd−1 )R . It follows that there is a Noetherian complete lo-
cal domain S containing R and module-finite over R such that (x1 · · · xd−1 )t ∈
1 , ..., xd−1 , xd )S. But now (x1 · · · xd−1 ) is in the ideal (x1 , ..., xd−1 ) in the
(xt+1 t+1 t t+1 t+1

ring S/xd S, which has dimension d − 1. Our induction shows that this is impossi-
ble. 
Next, we apply Theorem 5.2 it to various intersection theorems. One of the
first such intersection conjectures was:
Conjecture 6.2. Let (R, m) be a local Noetherian ring, and let M, N be two
finitely generated nonzero R-modules such that M ⊗R N has finite length. Then
dim N ≤ pdR (M ).
Of course there is nothing to prove if the projective dimension of M is infinite.
We prove (see [9]):
Theorem 6.3. Let (R, m) be a local Noetherian ring of positive prime char-
acteristic p, and let M, N be two finitely generated nonzero R-modules such that
M ⊗R N has finite length. Then
dim N ≤ pdR (M ).
Proof. One can begin by making some easy reductions. These types of reduc-
tion are very good practice in commutative algebra. First, note that the assumption
that the tensor product has finite length is equivalent to saying that I + J is m-
primary, where I = Ann(N ) and J = Ann(M ). Then we can choose a prime P
containing I such that dim(R/P ) = dim(N ), and observe that we can replace N by
R/P without loss of generality. It is more difficult to change M , since the property
of being finite projective dimension does not allow many changes.
Let’s just suppose for a moment that R/P is Cohen-Macaulay. Since P + J is
m-primary, we can always choose x1 , ..., xd ∈ J whose images in B = R/P form a
system of parameters (and thus are a regular sequence in R/P ). If the projective
dimension of M is smaller than d = dim(R/P ), then TorR d (B/(x1 , ..., xd )B, M ) =
0. Notice that TorR 0 (B, M ) = 0. We claim by induction that for 0 ≤ i ≤ d,
TorRi (B/(x1 , ..., xi )B, M ) = 0. When i = d we arrive at a contradiction. Sup-
pose we know this for i < d. Set Bi = B/(x1 , ..., xi ). The short exact sequence
ABSOLUTE INTEGRAL CLOSURE 133
15

0−→Bi −→Bi −→Bi+1 −→0 obtained by multiplication by xi+1 on Bi induces a map


of Tors when tensored with M . Since all xi kill M , we obtain a surjection of
TorR R
i+1 (Bi+1 , M ) onto Tori (Bi , M ). This finishes the induction.
Of course, we don’t know that R/P is Cohen-Macaulay, and in general it won’t
be. But now suppose that we are in positive characteristic. We can first complete R
before beginning the proof. Now R/P is a complete local domain, and S = (R/P )+
is Cohen-Macaulay in the sense that x1 , .., xd form a regular sequence in this ring.
The same proof works verbatim, provided we know that S ⊗R M = 0. But this is
easy; it is even nonzero after passing to the residue field of S. 

As a corollary, we get a favorite of the old Chicago school of commutative


algebra, the zero-divisor conjecture (now a theorem):

Theorem 6.4. Let (R, m) be a Noetherian local ring of characteristic p, and


let M be a nonzero finitely generated R-module having finite projective dimension.
If x is a non-zerodivisor on M , then x is a non-zerodivisor on R.
Proof. This proof is taken from [18]. First observe that the statement of the
theorem is equivalent to saying that every associated prime of R is contained in an
associated prime of M . We induct on the dimension of M to prove this statement.
If dim M = 0, then the only associated prime of M is m, which clearly contains
every prime of R. Hence we may assume that dim M > 0. Let P ∈ Ass(R). First
suppose that there is a prime Q ∈ Supp(M ), Q = m, such that P ⊆ Q. Then we
can change the ring to RQ and the module to MQ . By induction, PQ is contained
in an associated prime of MQ , so lifting back gives us that P is in an associated
prime of M . We have reduced to the case in which R/P ⊗R M has finite length. By
Theorem 6.3, dim(R/P ) ≤ pdR (M ) = depth(R) − depth(M ). Since P is associated
to R, dim(R/P ) ≥ depth(R) (exercise). It follows that the depth of M is 0, and
hence the maximal ideal is associated to M (and contains P ). 

Both of the theorems proved above in characteristic p are in fact true in full
generality. The work of [18] showed that both follow from the new intersection
theorem, which has been proved for all Noetherian local rings by Paul Roberts. See
[20], [21], [22]. For a full discussion of these conjectures and others, see the recent
article of Hochster [10].
For a completely different type of application, we consider an old result of
Grothendieck’s concerning when the punctured spectrum of a local ring is con-
nected. Recall that if (R, m) is a local ring, the punctured spectrum of R is the
space Spec(R) − {m}. There is a beautiful proof of Grothendieck’s result in all
characteristics due to Brodmann and Rung [4]. The main point here is that if R
is Cohen-Macaulay and the xi are parameters, then there is a very easy proof. It
turns out that one can always assume that the xi are parameters, and then the
proof of the Cohen-Macaulay case directly generalizes to one in characteristic p
using R+ . The exact statement is:
Theorem 6.5. Let (R, m) be a complete local Noetherian domain of dimen-
sion d, and let x1 , ..., xk ∈ m, where k ≤ d − 2. Then the punctured spectrum of
R/(x1 , ..., xk ) is connected.
134
16 CRAIG HUNEKE

Proof. We take the proof from [11]. First assume that the xi are parame-
ters. Let I and J give a disconnection of the punctured spectrum of R/(x1 , ..., xk ).
Choose elements u + v, y + z which together with the xi form parameters such
that u, y ∈ I and v, z ∈ J. Modulo (x1 , ..., xk ) one has the relation y(u + v) −
u(y + z) = 0. Since the parameters form a regular sequence in R+ , we ob-
tain that y ∈ (y + z, x1 , ..., xk )R+ . Similarly, z ∈ (y + z, x1 , ..., xk )R+ . Write
y = c(y + z) modulo (x1 , ..., xk ) and z = d(y + z) modulo (x1 , ..., xk ). Then
(1 − c − d)(y + z) ∈ (x1 , ..., xk )R+ so that 1 − c − d ∈ (x1 , ..., xk )R+ . At least one of
c or d is a unit in R+ , say c. But then y is not a zerodivisor modulo (x1 , ..., xk )R+
and this implies that J ⊆ (x1 , ..., xk )R+ . Then I must be primary to mR+ which
is a contradiction since the height of I is too small.
It remain to reduce to the case in which the xi are parameters.
We claim that any k-elements are up to radical in an ideal generated by k-
elements which are parameters. The key point is to prove this for k = 1. Suppose
that x = x1 is given. If x already has height one we are done. If x is nilpotent,
choose y to be any parameter in I. So assume that x is not in every minimal prime.
For n  0, 0 : xn = 0 : xn+1 , and changing x to xn , we obtain that x is not a zero
divisor on R/(0 : x). Then there is an element s ∈ 0 : x such that y = x + s has
height one, and we may multiply s by a general element of I to obtain that y ∈ I.
But xy = x2 so that x is nilpotent on (y). Inductively choose y1 , ..., yk−1 which
are parameters such that the ideal they generate contains x1 , ..., xk−1 up to radical.
Replace R by R/(y1 , ..., yk−1 ), and repeat the k = 1 step.
Now if I and J disconnect the punctured spectrum of R/(x1 , ..., xk ) choose any
ideals I  and J  of height at least k, not primary to the maximal ideal such that
I  contains I up to radical and J  contains J up to radical. Choose parameters in
I  ∩ J  such that the xi are in the radical K of these parameters. Then K + I and
K + J disconnect the punctured spectrum of R/K. 

References
1. I.M. Aberbach, The vanishing of TorR +
1 (R , k) implies that R is regular, Proc. Amer. Math.
Soc. 133 (2004), 27-29.
2. I. M. Aberbach and M. Hochster, Failure of coherence in R+ and finite Tor dimension, J.
Pure Appl. Algebra 122 (1997), 171-184.
3. M. Artin, On the joins of Hensel rings, Advances in Math. 7 (1971), 282–296.
4. M. Brodmann and J. Rung, Local cohomology and the connectedness dimension in algebraic
varieties, Comment. Math. Helv. 61 (1986), 481–490.
5. M. Brodmann and R. Sharp, Local Cohomology, Cambridge University Press, 1998
6. C. Chevalley, Introduction to the Theory of Algebraic Functions of One Variable, Amer. Math.
Soc., 1951.
7. D. Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate Texts
in Mathematics, 150 (1995), Springer-Verlag, New York, Berlin, Heidelberg.
8. R. Heitmann, The direct summand conjecture in dimension three, Ann. of Math. (2) 156
(2002), 695–712.
9. M. Hochster, Topics in the homological theory of modules over commutative rings, CBMS 24
(1975), American. Math. Soc.
10. M. Hochster, Homological conjectures, old and new. Illinois J. Math. 51 (2007), 151–169.
11. M. Hochster and C. Huneke, Infinite integral extensions and big Cohen-Macaulay algebras,
Ann. of Math. 135 (1992), 53-89.
12. M. Hochster and Y. Yao, A weak embedding theorem for modules of finite phantom projective
dimension, in preparation.
13. C. Huneke and G. Lyubeznik, Absolute integral closure in positive characteristic. Adv. Math.
210 (2007), 498–504.
ABSOLUTE INTEGRAL CLOSURE 135
17

14. M.-F. Huang, Ph.D. thesis, Purdue University, 1968.


15. K. Kedlaya, The algebraic closure of the power series field in positive characteristic. Proc.
Amer. Math. Soc. 129 (2001), 3461–3470.
16. G. Lyubenzik, A property of the absolute integral closure of an excellent local domain in
mixed characteristic, preprint (2009).
17. K. J. Nowak, Some elementary proofs of Puiseux’s theorems. Univ. Iagel. Acta Math. 38
(2000), 279–282.
18. C. Peskine and L. Szpiro, Dimension projective finie et cohomologie locale, I.H.E.S. 42 (1973),
47–119.
19. F. Rayner, An algebraically closed field, Glasgow J. Math. 9 (1968), 146–151.
20. P. Roberts, Le théoréme intersection, C. R. Acad. Sc. Paris Sér. I 304 (1987), 177-180.
21. P. Roberts, Intersection theorems, in Commutative Algebra, Math. Sci. Research Inst. Publ.
15, Springer- Verlag, New York-Berlin-Heidelberg, 1989, 417-436.
22. P. Roberts, Multiplicities and Chern classes in local algebra, Cambridge Tracts in Mathematics
133, Cambridge University Press, Cambridge, 1998.
23. A. Singh, Separable integral extensions and plus closure, Manuscripta Math. 98 (1999), 497-
506.
24. K.E. Smith, Tight closure of parameter ideals, Invent. Math. 115 (1994), 41-60.
25. D. Ştefănescu, Generalized Puiseux expansions and their Galois groups, Illlinois J. Math. 41
(1997), 129–141.

Department of Mathematics, University of Kansas, Lawrence, KS 66045


E-mail address: huneke@math.ku.edu
URL: http://www.math.ku.edu/~huneke
This page intentionally left blank
Contemporary Mathematics
Volume 555, 2011

A property of the Frobenius map of a polynomial ring

Gennady Lyubeznik, Wenliang Zhang and Yi Zhang

Abstract. Let R = k[x1 , . . . , xn ] be a ring of polynomials in a finite number


of variables over a perfect field k of characteristic p > 0 and let F : R → R be
the Frobenius map of R, i.e. F (r) = r p . We explicitly describe an R-module
isomorphism HomR (F∗ (M ), N ) ∼ = HomR (M, F ∗ (N )) for all R-modules M and
N . Some recent and potential applications are discussed.

1. Introduction
The main result of this paper is Theorem 3.3 which is a type of adjointness
property for the Frobenius map of a polynomial ring over a perfect field. The
interest in this fairly elementary result comes from its striking recent applications
(see [7, Sections 5 and 6] and [8]) and also from the fact that despite extensive
inquiries we have not been able to find it in the published literature.
Especially interesting is the application in [8] where a striking new result on
local cohomology modules in characteristic p > 0 is deduced from Theorem 3.3.
There is no doubt that the same result is true in characteristic 0, but the only
currently known (to us) proof is in characteristic p > 0, based on Theorem 3.3.
This has motivated what promises to be a very interesting search for some new
technique to extend the result in [8] to characteristic 0.
Our Theorem 3.3 ties in nicely with the very recent and beautiful theory of
Cartier modules of Blickle and Böckle [1]. In particular, the isomorphism (8) (see
below) on which our Theorem 3.3 rests is implicit in [1, Theorem 5.9]. Theorem
3.3 is likely to find applications to Cartier modules over polynomial rings.
We believe the results of this paper hold a potential for further applications
and we discuss some of them in the last section.

2. Preliminaries
Let R be a regular (but not necessarily local) UFD, let Rs and Rt be two copies
of R (the subscripts stand for source and target) and let F : Rs → Rt be a finite

1991 Mathematics Subject Classification. Primary 13A35, 13P05, 13D45.


Key words and phrases. the Frobenius, polynomial ring, characteristic p > 0, local
cohomology.
The first and third authors gratefully acknowledge NSF support through grants DMS-0202176
and DMS-0701127. The second author is supported in part by a 2009 Spring/Summer Research
Fellowship from Department of Mathematics, University of Michigan.

1
137
138
2 GENNADY LYUBEZNIK, WENLIANG ZHANG AND YI ZHANG

ring homomorphism. Let


F∗ : Rt −mod → Rs −mod
be the restriction of scalars functor (i.e. F∗ (M ) for every Rt -module M is the
additive group of M regarded as an Rs -module via F ) and let
F ! , F ∗ : Rs −mod → Rt −mod
be the functors defined by F ! (N ) = HomRs (Rt , N ) and F ∗ (N ) = Rt ⊗Rs N for
every Rs -module N .
F ∗ (N ) and F ! (N ), for an Rs -module N , have a structure of Rs -module via
the natural Rs -action on Rt , while F∗ (M ), for an Rt -module M , retains its old
Rt -module structure (from before the restriction of scalars). Thus F ∗ (N ), F ! (N ),
and F∗ (M ) are both Rs - and Rt -modules.
It is well-known that F ∗ is left-adjoint to F∗ , i.e. there is an isomorphism of
Rt -modules
(1) HomR (F ∗ (N ), M ) ∼
t = HomR (N, F∗ (M ))
s

f →(n → f (1 ⊗ n))
(r ⊗ n → rf (n)) ←f
which is functorial in M and N (see, for example, [4, II.5, p.110]). This holds
without any restrictions on R.
The main result of this paper is based on a different type of adjointness (see (8)
below) which is certainly not well-known (we could not find an explicit reference).
Unlike (1) it is not quite canonical but depends on a choice of a certain isomorphism
φ (described in (6) below). And it requires the conditions we imposed on R, namely,
a regular UFD.
Since Rt , being regular, is locally Cohen-Macaulay, it follows from the Auslander-
Buchsbaum theorem that the projective dimension of Rt as Rs -module is zero, i.e.
Rt is projective, hence locally free, as an Rs -module. It is a standard fact that in
this case F ! is right adjoint to F∗ , [4, Ch. 3, Exercise 6.10] i.e. for every Rt -module
M and every Rs -module N there is an Rt -module isomorphism
(2) HomR (F∗ (M ), N ) ∼
s =HomR (M, F ! (N ))
t

(m → f(1)) ←f
where f = f (m) : Rt → N . This isomorphism is functorial both in M and in N .
def
Consider the Rt -module H = HomRs (Rt , Rs ). Since Rt is a locally free Rs -
module of finite rank, it is a standard fact [4, Ch 2, Exercise 5.1(b)] that there is
an isomorphism of functors
(3) ∼ H ⊗R −,
F! = s

i.e. for every Rs -module N there is an Rt -module isomorphism


(4) H ⊗R N ∼ s= HomR (Rt , N )s

h ⊗ n → (r → h(r)n)
and this isomorphism is functorial in N . Replacing F ! by H ⊗Rs − in (2) produces
an Rt -module isomorphism
(5) HomRs (F∗ (M ), N ) ∼
= HomRt (M, H ⊗Rs N ).
which is functorial in M and N .
A PROPERTY OF THE FROBENIUS MAP 139
3

It follows from either [5, Kor. 5.14] or [3, Th. 3.3.7] that locally H is the
canonical module of Rt . In particular, the rank of H as Rt -module is 1 and therefore
H is Rt -module isomorphic to some ideal I of Rt . According to [5, Kor. 6.13] the
quotient Rt /I is locally Gorenstein of dimension dimRt −1, hence I has pure height
1. Since Rt is a UFD, I is principal, i.e. there is an Rt -module isomorphism
(6) φ : Rt → H.
Hence φ induces an isomorphism of functors
φ⊗id
(7) Rt ⊗Rs − ∼
= H ⊗Rs −.
Replacing H ⊗Rs − by F ∗ = Rt ⊗Rs − in (5) via (7) produces an Rt -module
isomorphism
(8) HomRs (F∗ (M ), N ) ∼
= HomRt (M, F ∗ (N ))
which is functorial in M and N .
While the isomorphisms (2), (3), (4) and (5) are canonical, the isomorphisms
(6), (7) and (8) depend on a choice of φ. Every Rt -module isomorphism φ : Rt → H
is obtained from a fixed φ by multiplication by an invertible element of Rt , i.e.
φ = c · φ where c ∈ Rt is invertible. Therefore the isomorphisms (7) and (8) are
defined up to multiplication by an invertible element of Rt .
Since the element 1 ∈ Rt generates the Rs -submodule F (Rs ) of Rt and does not
belong to mRt for any maximal ideal m of Rs , the Rs -module Rt /F (Rs ) is projec-
tive. Hence applying the functor HomRs (−, Rs ) to the injective map F : Rs → Rt
produces a surjection H → HomRs (Rs , Rs ). Composing it with the standard Rs -
ψ→ψ(1)
module isomorphism HomRs (Rs , Rs ) −→ Rs produces an Rs -module surjection
H → Rs . Composing this latter map with the isomorphism φ from (6) produces
an Rs -module surjection
(9) ψ : Rt → Rs .
If N is an Rs -module, applying − ⊗Rs N to ψ produces an Rs -module surjection
(10) ψN : Rt ⊗Rs N → N.
It is not hard to check that the isomorphism (8) sends g ∈ HomRt (M, F ∗ (N )) to
ψN ◦ g ∈ HomRs (F∗ (M ), N ).

3. The Main Result


For the rest of this paper R is a ring of polynomials in a finite number of vari-
ables over a perfect field k of characteristic p > 0 and F : Rs → Rt is the standard
Frobenius map, i.e. F (r) = r p . The main result of this paper (Theorem 3.3) is
an explicit description of the isomorphism (8) in terms of polynomial generators of
R. The recent applications [7, 8] crucially depend on this explicit description. We
keep the notation of the preceding section.
Let x1 , . . . , xn be some polynomial generators of R over the field k, i.e. R =
k[x1 , . . . , xn ]. We denote the multi-index i1 , . . . , in by ī. Since k is perfect, Rt is a
def
free Rs -module on the pn monomials eī = xi11 · · · xinn where 0 ≤ ij < p for every j. If

ij < p−1, then xj eī = eī where ī is the multi-index i1 , . . . , ij−1 , ij +1, ij+1 , . . . , in .
If ij = p−1, then xj eī = xpj eī where ī is the multi-index i1 , . . . , ij−1 , 0, ij+1 , . . . , in .
140
4 GENNADY LYUBEZNIK, WENLIANG ZHANG AND YI ZHANG

Let {fī ∈ H|0 ≤ ij < p for every j} be the dual basis of H, i.e. fī (ei¯ ) = 1 if
ī = i¯ and fī (ei¯ ) = 0 otherwise. If ij > 0, then xj fī = fī where ī is the multi-
index i1 , . . . , ij−1 , ij − 1, ij+1 , . . . , in . If ij = 0, then xj fī = xpj fī where ī is the
multi-index i1 , . . . , ij−1 , p − 1, ij+1 , . . . , in (here xj ∈ Rt ).
Denote the multi-index p − 1, . . . , p − 1 by p − 1 and let p − 1 − ī be the multi-
index p − 1 − i1 , . . . , p − 1 − in .
Proposition 3.1. (cf. [2, Remark 3.11]) The Rs -linear isomorphism φ : Rt →
H that sends eī to fp−1−ī is Rt -linear.
Proof. All we have to show is that φ(xj eī ) = xj φ(eī ) for all indices j and multi-
indices ī. This is straightforward from the definition of φ and the above description
of the action of xj on eī and fī . 
Clearly, F ∗ (N ) = Rt ⊗Rs N = ⊕ī (eī ⊗Rs N ), as Rs -modules. Thus every Rs -
linear map g : M → F ∗ (N ) has the form g = ⊕ī (eī ⊗Rs gī ) where gī : M → N are
Rs -linear maps (i.e. gī : F∗ (M ) → N because M with its Rs -module structure is
F∗ (M )).
Lemma 3.2. An Rs -linear map g : M → F ∗ (N ) as above is Rt -linear if and
only if gī (−) = gp−1 (ep−1−ī (−)) for every ī (here ep−1−ī ∈ Rt acts on (−) ∈ F∗ (M )
via the Rt -module structure on F∗ (M )).
Proof. Assume g is Rt -linear. Then g commutes with multiplication by every el-
ement of Rt and in particular with multiplication by ep−1−ī . That is g(ep−1−ī (−)) =
ep−1−ī g(−). Since ep−1−ī eī = ep−1 , the p − 1-component of ep−1−ī g(−) is ep−1 ⊗Rs
gī (−) while the p − 1-component of g(ep−1−ī (−)) is ep−1 ⊗Rs gp−1 (ep−1−ī (−)). Since
the two p − 1-components are equal, gī (−) = gp−1 (ep−1−ī (−)).
Conversely, assume gī (−) = gp−1 (ep−1−ī (−)) for every ī. To show that g is
Rt -linear all one has to show is that g commutes with the action of every xj ∈ Rt ,
i.e. the ī-components of g(xj (−)) and xj g(−) are the same for all ī. If ij > 0, then
the ī-component of xj g(−) is eī ⊗Rs gī (−) where ī is the index i1 , . . . , ij−1 , ij −
1, ij+1 , . . . , in while the ī-component of g(xj (−)) is eī ⊗Rs gī (xj (−)). But the fact
that gī (−) = gp−1 (ep−1−ī (−)) for every ī implies gī (−) = gī (xj (−)).
If ij = 0, then the ī-component of xj g(−) is eī ⊗Rs (s xj gī (−)) where s xj
denotes the element of Rs corresponding to xj ∈ Rt (i.e. F (s xj ) = xpj ) and
ī is the index i1 , . . . , ij−1 , p − 1, ij+1 , . . . , in while the ī-component of g(xj (−))
is gī (xj (−)). But the fact that gī (−) = gp−1 (ep−1−ī (−)) for every ī implies
s xj gī (−) = gī (xj (−)). 
Finally we are ready for the main result of the paper which is the following
explicit description of the isomorphism (8) for the Frobenius map.
Theorem 3.3. For every Rt -module M and every Rs -module N there is an
Rt -linear isomorphism
HomRs (F∗ (M ), N ) ∼
= HomRt (M, F ∗ (N ))
(11) gp−1 (−) ←(g = ⊕ī (eī ⊗Rs gī (−)))
(12) g → ⊕ī (eī ⊗Rs g(ep−1−ī (−))).
Proof. As is pointed out at the end of the preceding section, the isomorphism
(8) sends g ∈ HomRt (M, F ∗ (N )) to ψN ◦g ∈ HomR (F∗ (M ), N ). It is straightforward
A PROPERTY OF THE FROBENIUS MAP 141
5

to check that with φ as in Proposition 3.1 the map ψ of (9) sends ep−1 to 1. This
implies that ψN ◦ g = gp−1 and finishes the proof that formula (11) produces the
isomorphism of (8) in one direction. The fact that the other direction of this
isomorphism is according to formula (12) has essentially been proven in Lemma
3.2. 

4. Potential Applications
We continue to assume that R is a polynomial ring over a perfect field.
The notion of F -finite modules was introduced in [6]. An F -finite module is
determined by a generating morphism, i.e. an R-module homomorphism β : M →
F ∗ (M ) where M is a finite R-module. For simplicity assume the R-module M has
finite length, i.e. the dimension of M as a vector space over k, which we denote by
d, is finite. Then the dimension of F ∗ (M ) as a vector space over k equals pn · d.
The number pn can be huge even for quite modest values of p and n. Thus the
target of β may be a huge-dimensional vector space even if d, p and n are fairly
small. But the R-module F∗ (M ) has dimension d as a k-vector space, hence the
map β̃ : F∗ (M ) → M that corresponds to β under the isomorphism of Theorem 3.3,
is a map between two d-dimensional vector spaces. Huge-dimensional vector spaces
do not appear! This should make the map β̃ easier to manage computationally
than the map β. Of course the isomorphism of Theorem 3.3 means that many
properties of β could be detected in β̃; for example, β is the zero map if and only if
β̃ is. Therein lies the potential for using Theorem 3.3 to make computations more
manageable.
The functors ExtiR (−, R) commute with both F ∗ and F∗ . More precisely, for
every finitely generated Rt -module M and every finitely generated Rs -module N
there exist functorial Rt -module isomorphisms
κi : ExtiRt (F ∗ (N ), Rt ) ∼
= F ∗ (ExtiRs (N, Rs ))

λi : F∗ (ExtiRt (M, Rt )) ∼
= ExtiRs (F∗ (M ), Rs ).
Indeed, for i = 0, M = Rt and N = Rs a straightforward composition of the Rt -
r  ⊗r→r  r p
module isomorphism Rt ⊗Rs Rs −→ Rt and the standard R-module isomor-
phisms HomR (R, R) ∼ = R for R = Rt , Rs produce κ0 while an additional Rt -module
isomorphism HomRs (Rt , Rs ) = H ∼ = Rt produces λ0 . This implies that if − stands
for a complex of finite free R-modules, then HomR (−, R) commutes with F ∗ and
F∗ . Taking now finite free resolutions of M and N in the categories of Rt - and
Rs -modules respectively and considering that F ∗ and F∗ , being exact, commute
with the operation of taking the (co)homology of complexes, we get κi and λi for
every i.
It is straightforward to check that there is a commutative diagram
HomRt (F ∗ (N ), M ) −−−−→ HomRs (N, F∗ (M ))
⏐ ⏐
⏐ ⏐
 
HomRt (ExtiRt (M, Rt ), ExtiRt (F ∗ (N ), Rt )) HomRs (ExtiRs (F∗ (M ), Rs ), ExtiRs (N, Rs ))
⏐ ⏐
⏐ ⏐
 
HomRt (ExtiRt (M, Rt ), F ∗ (ExtiRs (N, Rs ))) −−−−→ HomRs (F∗ (ExtiRt (M, Rt )), ExtiRs (N, Rs ))
142
6 GENNADY LYUBEZNIK, WENLIANG ZHANG AND YI ZHANG

where the top horizontal map is the isomorphism (1), the bottom horizontal map is
the isomorphism of Theorem 3.3, the bottom vertical maps are induced by κi and
λi and, finally, the top vertical maps are defined by sending every f ∈ Hom(L, L )
to the map Exti (L , R) → Exti (L, R) functorially induced by f .
In other words, if a pair of maps F ∗ (N ) → M and N → F∗ (M ) correspond to
each other under (1), then the induced maps ExtiRt (M, Rt ) → F ∗ (ExtiRs (N, Rs ))
and F∗ (ExtiRt (M, Rt )) → ExtiRs (N, Rs ) correspond to each other under isomor-
phism of Theorem 3.3.
The commutativity of the diagram can be verified via the following sequence
of steps. First one verifies the commutativity in the special case that M and N
are, respectively, finite free Rt - and Rs -modules. Next one replaces M and N by
their finite free resolutions M • and N • ; in this case Exti would be replaced by the
Hom complex of the corresponding complex into either Rt or Rs and what is now
Hom in the above diagram would be understood as chain maps of complexes; the
commutativity of the resulting diagram follows from the special case where M and
N are free. Finally, the above commutative diagram is induced on the homology of
complexes by the diagram for complexes which has just been described.
Of course there is a similar diagram with the isomorphism of Theorem 3.3 in
the top row; we are not going to use it in the rest of the paper.
An important example of F -finite modules are local cohomology modules HIi (R)
of R with support in an ideal I ⊂ R. A generating morphism of HIi (R) is the
composition
κi
f : ExtiR (R/I, R) → ExtiR (F ∗ (R/I), R) ∼
= F ∗ (ExtiR (R/I, R))
r  ⊗r̄→r  r̄ p
where the first map is induced by the isomorphism F ∗ (R/I) ∼
= R/I [p] fol-
lowed by the natural surjecton R/I → R/I. The following proposition holds the
[p]

potential for simplifying computations involving local cohomology modules.


Proposition 4.1. Let I ⊂ R be an ideal and let the composition
κ
f : ExtiR (R/I, R) → ExtiR (F ∗ (R/I), R) ∼
i
= F ∗ (ExtiR (R/I, R))
be as above.The map that corresponds to f under the isomorphism of Theorem 3.3
is the composition
λ
g : F∗ (ExtiR (R/I, R)) ∼
i
= ExtiR (F∗ (R/I), R) → ExtiR (R/I, R)
where the second map in the composition is nothing but the map induced on ExtiR (−, R)
r→r p
by the natural Frobenius map R/I ∼
= F∗ (R/I).
Proof. This is immediate from the above commutative diagram considering
r  ⊗r→r  r p r→r p
that the maps F ∗ (R/I) −→ R/I and R/I → F∗ (R/I) correspond to each
other under the isomorphism (1). 
In addition to potential use in computation, the material of this section has
already been used in a proof of a theoretical result [7, Section 5].

References
[1] M. Blickle and G. Böckle, Cartier modules: finiteness results, preprint, 2009
arXiv:0909.2531v1.
A PROPERTY OF THE FROBENIUS MAP 143
7

[2] M. Blickle, K Schwede, S. Takagi and W. Zhang, Discreteness and Rationality of F -jumping
Numbers on Singular Varieties, preprint, 2009, arXiv:0906.4679
[3] W. Bruns and J. Herzog, Cohen-Macaulay Rings, Cambridge University Press, 1993.
[4] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics, 52, Springer 1977.
[5] J. Herzog and E. Kunz, Der Kanonische Modul eines Cohen-Macaulay-Rings, Lecture Notes
in Mathematics 238, Springer 1971.
[6] G. Lyubeznik, F -modules: Applications to Local Cohomology and D-modules in Character-
istic p > 0, J. reine angew. Math., 491 (1997) 65 - 130.
[7] W. Zhang, Lyubeznik Numbers of Projective Schemes, preprint, 2010 arXiv:1001.3662.
[8] Y. Zhang, A Property Of Local Cohomology Modules Of Polynomial Rings, preprint 2010
arXiv:1001.3363.

Dept. of Mathematics, University of Minnesota, Minneapolis, MN 55455


E-mail address: gennady@math.umn.edu

Dept. of Mathematics, University of Michigan, Ann Arbor, MI 48109.


E-mail address: wlzhang@umich.edu

Dept. of Mathematics, University of Minnesota, Minneapolis, MN 55455


E-mail address: zhang397@math.umn.edu
This page intentionally left blank
Contemporary Mathematics
Volume 555, 2011

A note on the variety of pairs of matrices whose product is


symmetric

Mahdi Majidi-Zolbanin and Bart Snapp

This paper is dedicated to Wolmer Vasconcelos.

Abstract. We give different short proofs for a result proved by C. Mueller


in [9]: Over an algebraically closed field pairs of n × n matrices whose product
is symmetric form an irreducible, reduced, and complete intersection variety of
dimension (3n2 + n)/2. Our work is connected to the work of Brennan, Pinto,
and Vasconcelos in [2].

1. Introduction and notation


A central research theme in classical algebraic geometry is to study the scheme
defined by a given ideal in a polynomial ring in order to determine whether it
has desirable properties such as being normal, nonsingular, complete intersection,
smooth, Gorenstein, Cohen-Macaulay, irreducible, or reduced. Many interesting
results have been obtained in connection to such questions. However, there are
also long-standing open problems, that continue to generate interest and provide
motivation for further research in this area. One of these open problems is the
following
Conjecture 1.1 (M. Artin, M. Hochster). Let X = (xij ) and Y = (yij ) be square
n × n matrices in 2n2 indeterminates and let I be the ideal generated by entries of
XY − Y X in the polynomial ring R = k[x11 , . . . , xnn , y11 , . . . , ynn ], where k is an
algebraically closed field. Then Spec (R/I) is reduced, and Cohen-Macaulay.
It is easy to show Spec (R/I) is Cohen-Macaulay when n = 1 and n = 2.
However, Macaulay, see [1], was used to completely establish this when n = 3
(though several partial results were obtained in [12]), and Macaulay 2, see [6], was
used to establish this when n = 4.
Several authors, including T. Motzkin and O. Taussky [8], M. Gerstenhaber [4],
and R. Guralnick [5] have shown that commuting pairs of n × n matrices with
entries in k, regarded as points in the 2n2 -dimensional affine space over k, form an
irreducible variety of dimension n2 + n. Since the coordinate ring of this variety is

2010 Mathematics Subject Classification. Primary 14M10, 13C40; Secondary 15B99.


Key words and phrases. Algebraic variety, Complete intersection, Irreducible, Reduced, Jacobian
module, Symmetric matrices, Commuting variety.
1 2011
c American Mathematical Society

145
146
2 MAHDI MAJIDI-ZOLBANIN AND BART SNAPP

R/ rad(I), the scheme Spec (R/ rad(I)) is referred to as the commuting variety in
the literature.
In [2] a special commuting variety is studied, the variety of commuting pairs
of symmetric matrices. It is shown that this variety is reduced, irreducible and a
complete intersection. Two symmetric matrices A and B commute if and only if
their product is symmetric; that is, AB = BA if and only if AB = B t At , where X t
is the transpose of a matrix X.
Partially inspired by the work in [2], we consider the ideal J generated by
entries of XY − Y t X t in the polynomial ring
R = k[x11 , . . . , xnn , y11 , . . . , ynn ].
This ideal was studied by C. Mueller in [9]. As a special case of his results, he finds
that the scheme Spec (R/J) is reduced, irreducible and a complete intersection of
dimension (3n2 + n)/2. In this paper we use completely different methods to give
proofs of these results. We should note that closed points of Spec (R/J) can be
identified with pairs of n × n matrices over k, whose product is symmetric.
In what follows, unless otherwise stated k will be an arbitrary field. We let
Mn (k) be the affine space of all n × n matrices with entries in k and Sn (k) be the
affine space of all symmetric n × n matrices with entries in k. We will use the
notation X t to denote the transpose of a matrix X.

2. Pairs of matrices whose product is symmetric


We begin by showing that if k is algebraically closed, pairs of n × n matri-
ces whose product is symmetric regarded as points in the 2n2 -dimensional affine
space form an irreducible variety. We will use an idea similar to the proof of [5,
Theorem 2]. First we need a few elementary facts.
Lemma 2.1. Let A ∈ Mn (k). Then there is a nonsingular symmetric n × n matrix
T such that AT is symmetric.
Proof. This is essentially proved in [11, Theorem 1]. By that theorem there
is a nonsingular symmetric matrix T , such that AT = T At . Since T is symmetric,
the previous equality shows that AT is symmetric. 
Proposition 2.2. Let B ∈ Mn (k), and T be a nonsingular n × n matrix. Let x be
an indeterminate. Then det (B + xT ) is not identically zero.
Proof. Suppose det (B + xT ) is identically zero. Then columns of the matrix
B + xT are linearly dependent. Hence, there are scalars α1 , . . . , αn ∈ k, not all
zero, such that
n n
αi VB,i = −x αi VT,i ,
i=1 i=1
where {VB,i } and {VT,i } are columns of B and T , respectively. Since this identity
holds for all x ∈ k, we must have

n
αi VT,i = 0,
i=1

which is a contradiction since T is nonsingular and its columns are linearly inde-
pendent. 
THE VARIETY OF MATRICES WHOSE PRODUCT IS SYMMETRIC 147
3

Lemma 2.3. Let A, B ∈ Mn (k). If det(B) = 0 and AB is symmetric, then there


is a symmetric matrix S ∈ Sn (k) such that A = B t S.
Proof. To say AB is symmetric means AB = B t At . Multiplying both sides
of this equality by B −1 from the right, we obtain
A = B t At B −1 .
We end the proof by showing that S := At B −1 is a symmetric matrix. To see this
multiply both sides of AB = B t At by B −1 on the right, and (B −1 )t on the left.
This gives us (B −1 )t A = At B −1 , hence S t = S. 
Theorem 2.4. Let k be an algebraically closed field and let
V n = {(A, B) ∈ Mn (k) × Mn (k) : AB is symmetric}.
Then V n , viewed as a subset of the 2n2 -dimensional affine space over k, is an
irreducible variety.
Proof. Consider a pair (A, B) ∈ V n . By Lemma 2.1 there is a nonsingular
matrix T such that AT is symmetric. Thus, (A, B + xT ) is contained in V n for
every x ∈ k. Moreover, except for finitely many values of x, det (B + xT ) is nonzero
(because by Proposition 2.2 the determinant of B + xT is a nonzero polynomial in
x and vanishes for only finitely many values of x). Thus (A, B) is in the closure of
the set of elements in V n where the second term is nonsingular.
Now consider the map ϕ : Sn (k) × Mn (k) −→ V n , defined by
ϕ(S, B) = (B t S, B).
The image of ϕ is irreducible (since the domain is) and is dense (since by Lemma 2.3
it contains all elements of V n where the second term is nonsingular). Thus its
closure is equal to V n and is irreducible. 
Working from the fact that V n is irreducible, we will use a technique introduced
in [2] to show that the variety of pairs of matrices whose product is symmetric is
reduced and a complete intersection. This technique relies on the notion of Jacobian
module and some other related results. Let I = (f1 , . . . , fm ) be an ideal in the
polynomial ring R = k[x11 , . . . , xnn , y11 , . . . , ynn ] with a given set of generators
f1 , . . . , fm , such that each fi is a sum of quadratic terms of the form xij ykl . Then,
the Jacobian matrix Φ of the fi ’s with respect to the variables yij is a matrix with
entries in the ring A = k[x11 , . . . , xnn ] and can be used to define an A-module
E := coker(Φ)
Φ 2
Am −→ An −→ E −→ 0.
The Jacobian module of R/I is the A-module E. The importance of the Jacobian
module is the following, see [2, Proposition 1.2]:
Proposition 2.5. Let R = k[x11 , . . . , xnn , y11 , . . . , ynn ]. Let X = (xij ) and Y =
(yij ) be square n×n matrices in 2n2 indeterminates and let J be the ideal generated
by entries of XY −Y t X t in R. Let E be the Jacobian module of R/J. Then R/J 
S(E). Moreover, if k is algebraically closed and V n is as defined in Theorem 2.4,
then the affine coordinate ring of V n is isomorphic to Spec (S(E))red .
The following theorem follows from [13, Proposition 1.4.1] and [10, Theo-
rem 3.4]:
148
4 MAHDI MAJIDI-ZOLBANIN AND BART SNAPP

Theorem 2.6. Let R be a polynomial ring over an arbitrary field, and let E be a
finitely generated R-module of projective dimension one. Then the following condi-
tions are equivalent:
(a) S(E) is an integral domain.
(b) Spec (S(E)) is irreducible.
Moreover, under these conditions, S(E) is a local complete intersection over R.
Theorem 2.7. Let k be an algebraically closed field and R and J be as defined in
Proposition 2.5. Then the scheme Spec(R/J) is reduced and is a complete inter-
section.
Proof. This will follow from Proposition 2.5 and Theorems 2.4 and 2.6, once
we show that the Jacobian module of R/J has projective dimension one. The
following argument is similar to the proof of [2, Lemma 3.2]. By definition of E we
have a presentation
Φ 2
An(n−1)/2 −→ An −→ E −→ 0,
where A = k[x11 , . . . , xnn ]. It suffices to show that the presentation matrix Φ of E
is injective. This will be achieved by showing that Φ has rank n(n − 1)/2. If we
specialize the matrix X (as defined in Proposition 2.5) to a generic diagonal matrix
(specializing the matrix X will not increase the rank of Φ), it is easy to see that
the entries of Z = XY − Y t X t specialize to

zij = Xii Yij − Xjj Yji
so that the corresponding Jacobian matrix has full rank. 
Remark 2.8. Mueller has shown in [9] that the scheme Spec(R/J) as defined in
Theorem 2.7 is also normal.

3. A connection with the variety of commuting pairs of symmetric


matrices
In char k = 2 we can give another proof for the fact that the variety of pairs of
n × n matrices whose product is symmetric is a complete intersection. This proof
is based on results from [2] which we summarize in the following theorem:
Theorem 3.1 ([2]). Let k be an algebraically closed field of characteristic = 2,
and let Xs and Ys be generic symmetric n × n matrices in n(n + 1) indeterminates.
Let Is be the ideal generated by entries of Xs Ys − Ys Xs in the polynomial ring in
n(n+1) indeterminates over k, which we denote by S. Then Spec (S/Is ) is reduced,
irreducible and a complete intersection.
Theorem 3.1 was proved in [2, Theorem 3.1]. Even though it appears that this
proof is only written for characteristic 0, it works in fact in any characteristic = 2.
The only place in their proof where characteristic 0 seems to have been used is in
showing that every symmetric matrix commutes with a symmetric nonderogatory
matrix [2, Lemma 3.5], which they need for proving the irreducibility of Spec (S/Is ).
We would like to rewrite the proof given in [2, Lemma 3.5] for all characteristics
= 2. While all we need to do is to change some references, the result needed is
rather technical and it is most convenient to simply reproduce the proof. We recall
that a square matrix A is called nonderogatory if its minimal polynomial coincides
with its characteristic polynomial. This condition is also equivalent to saying that
every matrix that commutes with A can be expressed as a polynomial in A.
THE VARIETY OF MATRICES WHOSE PRODUCT IS SYMMETRIC 149
5

Lemma 3.2 ([2, Lemma 3.5]). Let B be a an element of Sn (k), where k is an


algebraically closed field of characteristic = 2. Then there exists a nonderogatory
element of Sn (k) that commutes with B.
Proof. By the Jordan decomposition Theorem, see [3, Theorem 24.9], there
are matrices D and N , such that B = D + N , where D is diagonalizable, N is
nilpotent, and both D and N can be expressed as polynomials in B, therefore both
are symmetric. It follows that D is in fact orthogonally similar to a diagonal matrix,
see [7, Theorem 70]. Let O be an orthogonal matrix that diagonalizes D. Then
s
O t BO = (λi Ii + Ni ) ,
i=1

with Ni t
= O Ni O nilpotent and symmetric. Then the matrix
 s 


O (μi Ii + Ni ) O t
i=1
with distinct μi ’s, is symmetric, nonderogatory and commutes with B. 
We are now ready to present our second proof of the fact that the variety of
pairs of n × n matrices whose product is symmetric is a complete intersection.
Theorem 3.3. Let k be an algebraically closed field of characteristic = 2, and let
R, X, Y and J be as defined in Proposition 2.5. Then the ring R/J is a complete
intersection of dimension (3n2 + n)/2.
Proof. Let S be the polynomial ring over k whose indeterminates are exactly
those found in the set
{xij : i ≤ j} ∪ {yij : i ≤ j}.
We see that R has 2n2 indeterminates and that S has n2 +n indeterminates. Define
a surjective homomorphism ϕ : R → S where
 
xij if i ≤ j, yij if i ≤ j,
xij → and yi,j →
xji if i > j, yji if i > j.
Note that the images of X and Y under ϕ are both symmetric matrices. Hence
(3.1) ϕ(XY − Y t X t ) = ϕ(X)ϕ(Y ) − ϕ(Y )ϕ(X).
By [2, Theorem 3.1] the distinct nonzero entries of the matrix
ϕ(X)ϕ(Y ) − ϕ(Y )ϕ(X)
form a regular sequence in S of length (n2 − n)/2. Since given any regular sequence
in S, the preimage of this sequence will form a regular sequence in R/ ker(ϕ), it
follows from (3.1) that the distinct (up to sign) nonzero entries of XY − Y t X t form
a regular sequence of length (n2 − n)/2 in R/ ker(ϕ).
Next we examine ker(ϕ). It is easy to see that that ker(ϕ) is generated by the
elements of the set
{xij − xji : i < j} ∪ {yij − yji : i < j}
and that these generators are algebraically independent. Hence they form a regular
sequence in R of length (n2 − n). Since these generators and the distinct (up to
sign) nonzero entries of XY − Y t X t are homogeneous elements forming a regular
sequence, we conclude that the distinct (up to sign) nonzero entries of XY − Y t X t
150
6 MAHDI MAJIDI-ZOLBANIN AND BART SNAPP

also form a regular sequence of length (n2 − n)/2 in R. Thus, R/J is a complete
intersection ring of dimension (3n2 + n)/2. 

References
1. D. Bayer and M. Stillman, Macaulay user manual.
2. J. P. Brennan, M. V. Pinto and W. Vasconcelos, The Jacobian module of a Lie algebra,
Transactions of American Mathematical Society, Vol. 321, Pages 183-196, 1990.
3. C. W. Curtis, Linear algebra, an introductory approach, Springer-Verlag, 1984.
4. M. Gerstenhaber, On dominance and varieties of commuting matrices, Annals of Mathemat-
ics, Vol. 73, Pages 324-348, 1961.
5. R. M. Guralnick, A note on commuting pairs of matrices, Linear and Multilinear Algebra,
Vol. 31, Pages 71-75, 1992.
6. F. Hreinsdottir, A case where choosing a product order makes the calculations of a Gröbner
basis much faster, Journal Symbolic Computation, Vol. 18, Pages 373-378, 1994.
7. I. Kaplansky, Linear algebra and geometry, a second course, Chelsea Publishing Company,
1974.
8. T. Motzkin and O. Taussky, Pairs of matrices with property L. II, Transactions of American
Mathematical Society, Vol. 80, Pages 387-401, 1955.
9. C. C. Mueller, On the varieties of pairs of matrices whose product is symmetric, PhD disser-
tation, The University of Michigan, 2007. This dissertation is available electronically at:
http://www.math.lsa.umich.edu/∼hochster/mueller.thesis.pdf.
10. A. Simis and W. Vasconcelos, On the dimension and integrality of symmetric algebras, Math-
ematische Zeitschrift, Vol. 177, Pages 341-358, 1981.
11. O. Taussky and H. Zassenhaus, On the similarity transformation between a matrix and its
transpose, Pacific Journal of Mathematics, Vol. 9, No. 3, Pages 893-896, 1959.
12. M. Thompson, Topics in the ideal theory of commutative Noetherian rings, PhD dissertation,
The University of Michigan, 1985.
13. W. Vasconcelos, Arithmetic of blowup algebras, Cambridge University Press, 1994.

Department of Mathematics, CUNY LaGuardia Community College, NY 11101


E-mail address: mmajidi-zolbanin@lagcc.cuny.edu

Department of Mathematics, The Ohio State University, Columbus, OH 43210


E-mail address: snapp@math.osu.edu
Contemporary Mathematics
Volume 555, 2011

Combinatorics of symbolic Rees algebras of edge ideals of


clutters

José Martı́nez-Bernal, Carlos Renterı́a-Márquez, and Rafael H. Villarreal

This paper is dedicated to Wolmer Vasconcelos.

Abstract. Let C be a clutter and let I be its edge ideal. We present a com-
binatorial description of the minimal generators of the symbolic Rees algebra
Rs (I) of I. It is shown that the minimal generators of Rs (I) are in one to one
correspondence with the indecomposable parallelizations of C. From our de-
scription some major results on symbolic Rees algebras of perfect graphs and
clutters will follow. As a byproduct, we give a method, using Hilbert bases,
to compute all indecomposable parallelizations of C and all the corresponding
vertex covering numbers.

1. Introduction
Let C be a clutter with finite vertex set X = {x1 , . . . , xn }, i.e., C is a family of
subsets of X, called edges, none of which is included in another. The set of vertices
and edges of C are denoted by V (C) and E(C) respectively. A basic example of a
clutter is a graph. Let R = K[x1 , . . . , xn ] be a polynomial ring over a field K. The
edge ideal of C, denoted
 by I = I(C), is the ideal of R generated by all square-free
monomials xe = xi ∈e xi such that e ∈ E(C). The assignment C → I(C) gives a
natural one to one correspondence between the family of clutters and the family of
square-free monomial ideals.
The blowup algebra studied here is the symbolic Rees algebra of I:
Rs (I) = R ⊕ I (1) t ⊕ · · · ⊕ I (i) ti ⊕ · · · ⊂ R[t],
where t is a new variable and I (i) is the ith symbolic power of I. Recall that the
ith symbolic power of I is defined as
I (i) = S −1 I i ∩ R,
where S = R \ ∪sk=1 pi , the ideals p1 , . . . , ps are the minimal primes of I and S −1 I i
is the localization of I i at S. In our situation the ith symbolic power of I can be

2010 Mathematics Subject Classification. Primary 13A30; Secondary 13F20, 05C65, 05C75.
The second author was Partially supported by COFAA-IPN and SNI. The third author was
partially supported by CONACyT grant 49251-F and SNI.
1

151
152
2 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

expressed using systems of linear inequalities (see Eq. (2.1) in Section 2). Closely
related to Rs (I) is—another blowup algebra—the Rees algebra of I:

R[It] = R ⊕ It ⊕ · · · ⊕ I i ti ⊕ · · · = K[{x1 , . . . , xn , xe t| e ∈ E(C)}] ⊂ R[t].

Blowup algebras are interesting objects of study in algebra and geometry [30].
The study of symbolic powers of edge ideals from the point of view of graph
theory and combinatorics was initiated in [28] and further elaborated on in [29, 32].
A breakthrough in this area is the translation of combinatorial problems (e.g., the
Conforti-Cornuéjols conjecture [6], the max-flow min-cut property, the idealness
of a clutter, or the integer rounding property) into algebraic problems of blowup
algebras of edge ideals [3, 8, 10, 11, 16, 17, 22].
By a result of Lyubeznik [21], Rs (I) is a K-algebra of finite type generated
by a unique minimal finite set of monomials. The main theorem of this paper is a
description—in combinatorial optimization terms—of this minimal set of generators
of Rs (I) as a K-algebra. Before stating the theorem, we need to recall some more
terminology and notations.
A subset C of X is called a vertex cover of C if every edge of C contains at least
one vertex of C. A vertex cover C is called a minimal vertex cover if no proper
subset of C is a vertex cover. The number of vertices in a minimum vertex cover of
C, denoted by α0 (C), is called the vertex covering number of C. The dual concept of
a vertex cover is a stable set, i.e., a subset C of X is a vertex cover of C if and only
if X \ C is a stable set. The number of vertices in a maximum stable set, denoted
by β0 (C), is called the stability number of C. Notice that α0 (C) + β0 (C) = n.
A clutter C is called indecomposable if it cannot be decomposed as a disjoint
union of induced subclutters C1 , C2 such that α0 (C) = α0 (C1 ) + α0 (C2 ) (see Def-
inition 2.2). Erdös and Gallai [13] introduced this notion for graphs. A clutter
obtained from C by a sequence of deletions and duplications of vertices is called a
parallelization (see Definition 2.3). If a = (ai ) is a vector in Nn , we denote by C a
the clutter obtained from C by successively deleting any vertex xi with ai = 0 and
duplicating ai − 1 times any vertex xi if ai ≥ 1 (see Example 2.4).
Our main result is:
Theorem 2.6 Let 0
= a = (ai ) ∈ Nn , b ∈ N. Then xa1 1 · · · xann tb is a minimal
generator of Rs (I), as a K-algebra, if and only if C a is an indecomposable clutter
and b = α0 (C a ).
There are two cases where a combinatorial description of the symbolic Rees
algebra is known. If the clutter C has the max-flow min-cut property, then by a
result of [17], we have I i = I (i) for all i ≥ 1, i.e., Rs (I) = R[It] and a minimal
generator of Rs (I) is either a vertex xi or an “edge” xe t with e ∈ E(C). If C is a
perfect graph, then the minimal generators of Rs (I) are in one to one correspon-
dence with the cliques (complete subgraphs) of C [32]. Both cases will follow from
our combinatorial description of Rs (I) (see Corollaries 2.10 and 4.2 respectively).
As a byproduct, in Section 3 we give a method—based on the computation of
Hilbert bases of polyhedral cones—to compute all indecomposable parallelizations
of any clutter C along with all the corresponding vertex covering numbers. In
particular our method allows to compute all indecomposable induced subclutters of
any clutter C. This means that the symbolic Rees algebra of I encodes combinatorial
information of the clutter which can be decoded using a computer program, such
as Normaliz [4], which is able to compute Hilbert bases of polyhedral cones.
SYMBOLIC REES ALGEBRAS OF EDGE IDEALS 153
3

Harary and Plummer [19] studied some properties of indecomposable graphs.


They showed that if a connected graph is separated by the points of a complete
subgraph, then G is decomposable. All indecomposable graphs with at least three
vertices contain at least one odd cycle, and the join of two indecomposable graphs
is indecomposable [19]. Indecomposable graphs were first studied from an algebraic
point of view in [12]. To the best of our knowledge there is no structure theorem
for indecomposable graphs.
Indecomposable subgraphs occur naturally in the theory of perfect graphs.
Indeed, a graph G is perfect if and only if the indecomposable parallelizations of G
are exactly the complete subgraphs or cliques of G (see Proposition 4.1). This was
first shown in [12] using the main result of [5]. For graphs, we can use our methods
to locate all induced odd cycles (odd holes) and all induced complements of odd
cycles (odd antiholes) of length at least five. Indeed, odd holes of any length and
odd antiholes of length at least five are indecomposable subgraphs (see Lemma 3.5),
and thus by Theorem 2.6 they correspond to minimal generators of the symbolic
Rees algebra of the edge ideal of the graph. Odd holes and odd antiholes play a
major role in graph theory. In [5] it is shown that a graph G is perfect if and only
if G is a Berge graph, i.e., if and only if G has no odd holes or odd antiholes of
length at least five. In commutative algebra odd holes occurred for the first time
in [26], and later in the description of I(G){2} , the join of an edge ideal of a graph
G with itself [27]. They also occurred in the description of the associated primes
of powers of ideals of vertex covers of graphs [15].
The problem of finding a minimum vertex cover of a graph is a classical op-
timization problem in computer science and is a typical example of an NP-hard
problem. From the point of view of computational complexity theory, finding all
indecomposable subgraphs of a given graph using our method is a hard problem be-
cause to apply this method we must know all minimal vertex covers (see Section 3).
Thus, although our results provide some tools for computing, the contributions of
this paper could be more interesting from the theoretical point of view.
Throughout the paper we introduce most of the notions that are relevant for
our purposes. For unexplained terminology, we refer to [7, 25, 30, 31].

2. Symbolic Rees algebras of edge ideals


In this section we will give a combinatorial description of the minimal generators
of the symbolic Rees algebra of the edge ideal of a clutter using the notion of a
parallelization of a clutter and the notion of an indecomposable clutter. We continue
using the definitions and terms from the introduction.
Let C be a clutter with vertex set X = {x1 , . . . , xn } and let I = I(C) be its edge
ideal. We denote by Υ(C) the clutter whose edges are the minimal vertex covers of
C. The clutter Υ(C) is called the blocker of C or the Alexander dual of C. As usual,
we use xa as an abbreviation for xa1 1 · · · xann , where a = (ai ) ∈ Nn . 
If C is a subset of X, its characteristic vector is the vector v = xi ∈C ei , where
ei is the ith unit vector in Rn . Let C1 , . . . , Cs be the minimal vertex covers of C and
let uk be the characteristic vector of Ck for 1 ≤ k ≤ s. In our situation, according
to [31, Proposition 7.3.14], the bth symbolic power of I has a simple expression:

I (b) = pb1 ∩ · · · ∩ pbs


(2.1) = ({xa | a, uk ≥ b for k = 1, . . . , s}),
154
4 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

where pk is the prime ideal of R generated by Ck and , denotes the standard


inner product. In particular, if b = 1, we obtain the primary decomposition of I
because I (1) = I. Thus the height of I equals α0 (C), the vertex covering number
of C. This is a hint of the rich interaction between the combinatorics of C and the
algebra of I.
Next, in Lemma 2.1, we give a simple description of Rs (I) that was first ob-
served in the discussion of symbolic Rees algebras given in [14, p. 75], see also [20].
Let a = (ai )
= 0 be a vector in Nn and let b ∈ N. From Eq. (2.1) we get that xa tb
is in Rs (I) if and only if
a, uk ≥ b for k = 1, . . . , s.
If a, b satisfy this system of linear inequalities, we say that a is a b-vertex cover of
Υ(C). Often we will call a b-vertex cover simply a b-cover . Thus the symbolic Rees
algebra of I is equal to the K-subalgebra of R[t] generated by all monomials xa tb
such that a is a b-cover of Υ(C), as was first shown in [14, Theorem 3.5]. The notion
of a b-cover occurs in combinatorial optimization (see for instance [25, Chapter 77,
p. 1378] and the references there) and algebraic combinatorics [14, 20]. We say
that a b-cover a of Υ(C) is decomposable if there exists an i-cover c and a j-cover
d of Υ(C) such that a = c + d and b = i + j. If a is not decomposable, we call a
indecomposable. The indecomposable 0 and 1 covers of Υ(C) are the unit vectors
e1 , . . . , en and the characteristic vectors v1 , . . . , vq of the edges of C respectively.

Lemma 2.1. A monomial xa tb


= 1 is a minimal generator of Rs (I), as a K-
algebra, if and only if a is an indecomposable b-cover of Υ(C). In particular, the
following equality holds:
(2.2) Rs (I) = K[{xa tb | a is an indecomposable b-cover of Υ(C)}].

Proof. It follows from the discussion above, by decomposing any b-cover into
indecomposable ones. 

Let S be a set of vertices of a clutter C. The induced subclutter on S, denoted


by C[S], is the maximal subclutter of C with vertex set S. Thus the vertex set of
C[S] is S and the edges of C[S] are exactly the edges of C contained in S. Notice
that C[S] may have isolated vertices, i.e., vertices that do not belong to any edge of
C[S]. If C is a discrete clutter, i.e., all the vertices of C are isolated, we set I(C) = 0
and α0 (C) = 0.
Let C be a clutter and let X1 , X2 be a partition of V (C) into nonempty sets.
Clearly, one has the inequality
(2.3) α0 (C) ≥ α0 (C[X1 ]) + α0 (C[X2 ]).
If C is a graph and equality occurs, Erdös and Gallai [13] call C a decomposable
graph. This motivates the following similar notion for clutters.

Definition 2.2. A clutter C is called decomposable if there are nonempty vertex


sets X1 , X2 such that X is the disjoint union of X1 and X2 , and α0 (C) = α0 (C[X1 ])+
α0 (C[X2 ]). If C is not decomposable, it is called indecomposable.

Examples of indecomposable graphs include complete graphs, odd cycles and


complements of odd cycles of length at least five (see Lemma 3.5).
SYMBOLIC REES ALGEBRAS OF EDGE IDEALS 155
5

Definition 2.3. (Schrijver [25]) The duplication of a vertex xi of a clutter C


means extending its vertex set X by a new vertex xi and replacing E(C) by
E(C) ∪ {(e \ {xi }) ∪ {xi }| xi ∈ e ∈ E(C)}.
The deletion of xi , denoted by C \ {xi }, is the clutter formed from C by deleting
the vertex xi and all edges containing xi . A clutter obtained from C by a sequence
of deletions and duplications of vertices is called a parallelization.
It is not difficult to verify that these two operations commute. If a = (ai ) is a
vector in Nn , we denote by C a the clutter obtained from C by successively deleting
any vertex xi with ai = 0 and duplicating ai − 1 times any vertex xi if ai ≥ 1 (for
graphs cf. [18, p. 53]).
Example 2.4. Let G be the graph whose only edge is {x1 , x2 } and let a = (3, 3).
We set x1i = xi for i = 1, 2. The parallelization Ga is a complete bipartite graph
with bipartition V1 = {x11 , x21 , x31 } and V2 = {x12 , x22 , x32 }. Note that xki is a vertex,
i.e., k is an index not an exponent.

x1 s x11 s x21 s x31 s x11 s x2 s x31 s


 HH 1 
 @ @ H @ 
@
G(3,1) @H H@
s  @s H@
G G(3,3)
s s
 H s
x2 x12 x12 x22 x32

Fig. 1. Graph Fig. 2. Duplications of x1 Fig. 3. Duplications of x2

Proposition 2.5. ([9, Lemma 2.15], [25, p. 1385, Eq. (78.6)]) Let C be a
clutter with n vertices and let Υ(C) be the blocker of C. If a = (ai ) ∈ Nn , then
  
 
min ai  C ∈ Υ(C) = α0 (C a ).

xi ∈C

We come to the main result of this section.


Theorem 2.6. Let C be a clutter with vertex set X = {x1 , . . . , xn } and let
0
= a = (ai ) ∈ Nn , b ∈ N. Then xa tb is a minimal generator of Rs (I(C)), as a
K-algebra, if and only if C a is an indecomposable clutter and b = α0 (C a ).
Proof. We may assume that a = (a1 , . . . , am , 0, . . . , 0), where ai ≥ 1 for
i = 1, . . . , m. For each 1 ≤ i ≤ m the vertex xi is duplicated ai − 1 times, and the
vertex xi is deleted for each i > m. We denote the duplications of xi by x2i , . . . , xai i
and set x1i = xi for 1 ≤ i ≤ m. Thus the vertex set of C a can be written as
X a = {x11 , . . . , xa1 1 , . . . , x1i , . . . , xai i , . . . , x1m , . . . , xamm } = X a1 ∪ X a2 ∪ · · · ∪ X am ,
where X ai = {x1i , . . . , xai i } for 1 ≤ i ≤ m and X ai ∩ X aj = ∅ for i
= j.
⇒) Assume that xa tb is a minimal generator of Rs (I(C)). Then, by Lemma 2.1,
a is an indecomposable b-cover of Υ(C). First we prove that b = α0 (C a ). There is
k such that ak
= 0. We may assume that a − ek
= 0. By Proposition 2.5 we need
only show the equality
  
 
b = min ai  C ∈ Υ(C) .

xi ∈C
156
6 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

As a is a b-cover of Υ(C), the minimum is greater than or equal to b. If the minimum


is greater than b, then we can write a = (a − ek ) + ek , where a − ek is a b-cover and
ek is a 0-cover, a contradiction to the indecomposability of a.
Next we show that C a is indecomposable. We proceed by contradiction. Assume
that C a is decomposable. Then there is a partition X1 , X2 of X a such that α0 (C a ) =
α0 (C a [X1 ]) + α0 (C a [X2 ]). For 1 ≤ i ≤ n, we set
i = |X ai ∩ X1 | and pi = |X ai ∩ X2 |
if 1 ≤ i ≤ m and i = pi = 0 if i > m. Consider the vectors  = (i ) and
p = (pi ). Notice that a has a decomposition a =  + p because one has a partition
X ai = (X ai ∩ X1 ) ∪ (X ai ∩ X2 ) for 1 ≤ i ≤ m. To derive a contradiction we now
claim that  (resp. p) is an α0 (C a [X1 ])-cover (resp. α0 (C a [X2 ])-cover) of Υ(C).
Take an arbitrary C in Υ(C). The set
 
Ca = {x1i , . . . , xai i } = X ai
xi ∈C xi ∈C

is a vertex cover of C a . Indeed, if fk is any edge of C a , then fk has the form


j j j
(2.4) fk = {xkk11 , xkk22 , . . . , xkkrr } (1 ≤ k1 < · · · < kr ≤ m; 1 ≤ jki ≤ aki )
for some edge {xk1 , xk2 , . . . , xkr } of C. Since {xk1 , xk2 , . . . , xkr } ∩ C
= ∅, we get
fk ∩ Ca
= ∅. Thus Ca is a vertex cover of C a . Therefore Ca ∩ X1 and Ca ∩ X2 are
vertex covers of C a [X1 ] and C a [X2 ] respectively because E(C a [Xi ]) is contained in
E(C a ) for i = 1, 2. Hence using the partitions
 
Ca ∩ X1 = (X ai ∩ X1 ) and Ca ∩ X2 = (X ai ∩ X2 )
xi ∈C xi ∈C

we obtain
 
α0 (C a [X1 ]) ≤ |Ca ∩ X1 | = i and α0 (C a [X2 ]) ≤ |Ca ∩ X2 | = pi .
xi ∈C xi ∈C

This completes the proof of the claim. Consequently a is a decomposable b-cover


of Υ(C), where b = α0 (C a ), a contradiction to the indecomposability of a.
⇐) Assume that C a is an indecomposable clutter and b = α0 (C a ). To show
that xa tb is a minimal generator of Rs (I(C)) we need only show that a is an inde-
composable b-cover of Υ(C). To begin with, notice that a is a b-cover of Υ(C) by
Proposition 2.5. We proceed by contradiction assuming that there is a decomposi-
tion a =  + p, where  = (i ) is a c-cover of Υ(C), p = (pi ) is a d-cover of Υ(C), and
b = c + d. Each X ai can be decomposed as X ai = X i ∪ X pi , where X i ∩ X pi = ∅,
i = |X i |, and pi = |X pi |. We set
X  = X 1 ∪ · · · ∪ X m and X p = X p1 ∪ · · · ∪ X pm .
Then one has a decomposition X a = X  ∪ X p of the vertex set of C a . We now
show that α0 (C a [X  ]) ≥ c and α0 (C a [X p ]) ≥ d. By symmetry, it suffices to prove
the first inequality. Take an arbitrary minimal vertex cover C of C a [X  ]. Then
C ∪ X p is a vertex cover of C a because if f is an edge of C a contained in X  , then
f is covered by C , otherwise f is covered by X p . Hence there is a minimal vertex
cover Ca of C a such that Ca ⊂ C ∪ X p . Since C[{x1 , . . . , xm }] is a subclutter of
C a , there is a minimal vertex cover C1 of C[{x1 , . . . , xm }] contained in Ca . Then
SYMBOLIC REES ALGEBRAS OF EDGE IDEALS 157
7

the set C1 ∪ {xi | i > m} is a vertex cover of C. Therefore there is a minimal vertex
cover C of C such that C ∩ {x1 , . . . , xm } ⊂ Ca . Altogether one has:
(2.5) C ∩ {x1 , . . . , xm } ⊂ Ca ⊂ C ∪ X p =⇒
(2.6) C ∩ {x1 , . . . , xm } ⊂ Ca ∩ {x1 , . . . , xm } ⊂ (C ∪ X p ) ∩ {x1 , . . . , xm }.
We may assume that Ca ∩ {x1 , . . . , xm } = {x1 , . . . , xs }. Next we claim that X ai ⊂
Ca for 1 ≤ i ≤ s. Take an integer i between 1 and s. Since Ca is a minimal
vertex cover of C a , there exists an edge e of C a such that e ∩ Ca = {x1i }. Then
(e \ {x1i }) ∪ {xji } is an edge of C a for j = 1, . . . , ai , this follows using that the edges
of C a are of the form described in Eq. (2.4). Consequently xji ∈ Ca for j = 1, . . . , ai .
This completes the proof of the claim. Thus one has X i ⊂ X ai ⊂ Ca for 1 ≤ i ≤ s.
Hence, by Eq. (2.5), and noticing that X i ∩ X p = ∅, we get X i ⊂ C for 1 ≤ i ≤ s.
So, using that i = 0 for i > m, we get

s  
α0 (C a [X  ]) ≥ |C | ≥ i ≥ i = i ≥ c.
i=1 xi ∈C∩{x1 ,...,xm } xi ∈C

Therefore α0 (C a [X  ]) ≥ c. Similarly α0 (C a [X p ]) ≥ d. Thus


α0 (C a [X  ]) + α0 (C a [X p ]) ≥ c + d = b,
and consequently by Eq. (2.3) we have the equality
α0 (C a [X  ]) + α0 (C a [X p ]) = α0 (C a ).
Thus we have shown that C a is a decomposable clutter, a contradiction. 
Let C be a clutter. A set of edges of C is called independent if no two of them
have a common vertex. We denote the maximum number of independent edges of
C by β1 (C), this number is called the matching number of C. In general the vertex
covering number and the matching number satisfy β1 (C) ≤ α0 (C).
Definition 2.7. If β1 (C) = α0 (C), we say that C has the König property.
Lemma 2.8. If C is an indecomposable clutter with the König property, then
either C has no edges and has exactly one isolated vertex or C has only one edge
and no isolated vertices.
Proof. Let f1 , . . . , fg be a set of independent edges and let X  = ∪gi=1 fi ,
where g = α0 (C). Note that g = 0 if C has no edges. Then V (C) has a partition

V (C) = (∪gi=1 fi ) ∪ ∪xi ∈V (C)\X  {xi } .


As C is indecomposable, we get that either g = 0 and V (C) = {xi } for some vertex
xi or g = 1 and V (C) = fi for some i. Thus in the second case, as C is a clutter,
we get that C has exactly one edge and no isolated vertices. 
Corollary 2.9. Let C be a clutter and let I = I(C) be its edge ideal. Then
all indecomposable parallelizations of C satisfy the König property if and only if
I i = I (i) for i ≥ 1.
Proof. ⇒) It suffices to prove that R[It] = Rs (I). Clearly R[It] ⊂ Rs (I).
To prove the reverse inclusion take a minimal generator xa tb of Rs (I). If b = 0,
then a = ei for some i and xa tb = xi . Thus xa tb ∈ R[It]. Assume b ≥ 1. By
Theorem 2.6 we have that C a is an indecomposable clutter such that b = α0 (C a ).
As C a is indecomposable and satisfies the König property, using Lemma 2.8, it is
158
8 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

 that E(C ) =a {e} consists of a single edge e of C,


a
not hard to see that b = 1 and
i.e., x t = xe t, where xe = xi ∈e xi . Thus x tb ∈ R[It].
a b

⇐) Since R[It] = Rs (I), by Theorem 2.6 we obtain that the only indecompos-
able parallelizations are either induced subclutters of C with exactly one edge and
no isolated vertices or subclutters consisting of exactly one isolated vertex. Thus
in both cases they satisfy the König property. 
A clutter C is called Mengerian if all its parallelizations have the König property.
A clutter C satisfies the max-flow min-cut property if the linear program:
max{ 1, y | y ≥ 0, Ay ≤ a}
has an integral optimal solution for all a ∈ Nn , where A is the incidence matrix of
the clutter C and 1 is the vector of all ones. The columns of A are the characteristic
vectors of the edges of C. It is well known that a clutter is Mengerian if and only
if it satisfies the max-flow min-cut property [25, Chapter 79].
Thus the last corollary can be restated as:
Corollary 2.10. [17, Corollary 3.14] Let C be a clutter and let I be its edge
ideal. Then C has the max-flow min-cut property if and only if I i = I (i) for i ≥ 1.
The following was the first deep result in the study of symbolic powers of edge
ideals from the viewpoint of graph theory.
Corollary 2.11. [28, Theorem 5.9] Let G be a graph and let I be its edge
ideal. Then G is bipartite if and only if I i = I (i) for i ≥ 1.
Proof. ⇒) If G is a bipartite graph, then any parallelization of G is again
a bipartite graph. This means that any parallelization of G satisfies the König
property because bipartite graphs satisfy this property [7, Theorem 2.1.1]. Thus
I i = I (i) for all i by Corollary 2.9.
⇐) Assume that I i = I (i) for i ≥ 1. By Corollary 2.9 all indecomposable
induced subgraphs of G have the König property. If G is not bipartite, then G has an
induced odd cycle, a contradiction because induced odd cycles are indecomposable
[19] and do not satisfy the König property. 
Corollary 2.12. Let C be a clutter with vertex set X = {x1 , . . . , xn } and let
S ⊂ X. Then
 the induced clutter H = C[S] is indecomposable if and only if the
monomial ( xi ∈S xi )tα0 (H) is a minimal generator of Rs (I(C)).

xi ∈S ei . Since C = C[S], the result follows from Theo-
a
Proof. Let a =
rem 2.6. 
Corollary 2.13. Let C be a clutter with n vertices and let A be its incidence
matrix. If the polyhedron Q(A) = {x| x ≥ 0; xA ≥ 1} has only integral vertices,
then α0 (C a ) ≤ n − 1 for all indecomposable parallelizations C a of C.
Proof. Let v1 , . . . , vq be the characteristic vectors of the edges of C and let I i
be the integral closure of I i , where I is the edge ideal of C. As Q(A) is integral, by
[17, Corollary 3.13] we have that I i = I (i) for i ≥ 1, where
I i = ({xa | a ∈ iB ∩ Zn })
and B = Qn+ + conv(v1 , . . . , vq ), see [31]. Thus we have the equality R[It] =
Rs (I), where R[It] is the integral closure of R[It] in its field of fractions. Take any
SYMBOLIC REES ALGEBRAS OF EDGE IDEALS 159
9

indecomposable parallelization C a of C and consider the monomial m = xa tb , where


b = α0 (C a ). By Theorem 2.6 m is a minimal generator of Rs (I). Now, according
to [14, Corollary 3.11], a minimal generator of R[It] has degree in t at most n − 1,
i.e., b ≤ n − 1. 
We end this section showing some very basic properties of indecomposable
clutters. If e is a edge of a clutter C, we denote by C \ {e} the spanning subclutter
of C obtained by deleting e and keeping all the vertices of C.
Definition 2.14. A clutter C is called vertex critical if α0 (C \ {xi }) < α0 (C)
for all xi ∈ V (C). A clutter C is called edge critical if α0 (C \ {e}) < α0 (C) for all
e ∈ E(C).
The next lemma is not hard to prove.
Lemma 2.15. Let xi be a vertex of a clutter C and let e be an edge of C.
(a) If α0 (C \ {xi }) < α0 (C), then α0 (C \ {xi }) = α0 (C) − 1.
(b) If α0 (C \ {e}) < α0 (C), then α0 (C \ {e}) = α0 (C) − 1.
Definition 2.16. A clutter C is called connected if there is no U ⊂ V (C) such
that ∅  U  V (C) and such that e ⊂ U or e ⊂ V (C) \ U for each edge e of C.
Proposition 2.17. If a clutter C is indecomposable, then it is connected and
vertex critical.
Proof. Assume that C is disconnected. Then there is a partition X1 , X2 of
V (C) such that
(2.7) E(C) ⊂ E(C[X1 ]) ∪ E(C[X2 ]).
For i = 1, 2, let Ci be a minimal vertex cover of C[Xi ] with α0 (C[Xi ]) vertices. Then,
by Eq. (2.7), C1 ∪ C2 is a minimal vertex cover of C. Hence α0 (C[X1 ]) + α0 (C[X2 ])
is greater than or equal to α0 (C). So α0 (C) is equal to α0 (C[X1 ]) + α0 (C[X2 ]), a
contradiction to the indecomposability of C. Thus C is connected.
We now show that α0 (C \ {xi }) < α0 (C) for all i. If α0 (C \ {xi }) = α0 (C), then
V (C) = X1 ∪X2 , where X1 = V (C)\{xi } and X2 = {xi }. Note that C[X1 ] = C\{xi }.
As α0 (C[X1 ]) = α0 (C) and α0 (C[X2 ]) = 0, we contradict the indecomposability of
C. Thus α0 (C \ {xi }) < α0 (C) and C is vertex critical. 
Proposition 2.18. If C is a connected edge critical clutter, then C is indecom-
posable.
Proof. Assume that C is decomposable. Then there is a partition X1 , X2 of
V (C) into nonempty vertex sets such that α0 (C) = α0 (C[X1 ]) + α0 (C[X2 ]). Since
C is connected, there is an edge e ∈ E(C) intersecting both X1 and X2 . Pick a
minimal vertex cover C of C \ {e} with less than α0 (C) vertices. As E(C[Xi ]) is a
subset of E(C \ {e}) = E(C) \ {e} for i = 1, 2, we get that C covers all edges of
C[Xi ] for i = 1, 2. Hence C must have at least α0 (C) vertices, a contradiction. 
From Propositions 2.17 and 2.18 we obtain:
Corollary 2.19. The following hold for any connected clutter :
edge critical =⇒ indecomposable =⇒ vertex critical.
The next result can be used to build indecomposable clutters.
160
10 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

Proposition 2.20. Let D be a clutter obtained from a clutter C by adding


a new vertex v and some new edges containing v and some vertices of V (C). If
a = (1, . . . , 1) ∈ Nn is an indecomposable α0 (C)-cover of Υ(C) such that α0 (D) =
α0 (C) + 1, then a = (a, 1) is an indecomposable α0 (D)-cover of Υ(D).
Proof. Clearly a is an α0 (D)-cover of Υ(D). Assume that a = a1 + a2 ,
where ai
= 0 is a bi -cover of Υ(D) and b1 + b2 = α0 (D). We may assume that
a1 = (1, . . . , 1, 0, . . . , 0) and a2 = (0, . . . , 0, 1, . . . , 1). Let ai be the vector in Nn
obtained from ai by removing its last entry. Set v = xn+1 . Take a minimal vertex
cover Ck of C and consider Ck = Ck ∪{xn+1 }. Let uk (resp. uk ) be the characteristic
vector of Ck (resp. Ck ). Then
a1 , uk = a1 , uk ≥ b1 and a2 , uk + 1 = a2 , uk ≥ b2 ,
and consequently a1 is a b1 -cover of Υ(C). If b2 = 0, then a1 is an α0 (D)-cover of
Υ(C), a contradiction; because if u is the characteristic vector of a minimal vertex
cover of C with α0 (C) elements, then we would obtain α0 (C) ≥ u, a1 ≥ α0 (D),
which is impossible. Thus b2 ≥ 1, and a2 is a (b2 −1)-cover of Υ(C) if a2
= 0. Hence
a2 = 0, because a = a1 + a2 and a is indecomposable. This means that a2 = en+1 is
a b2 -cover of Υ(D), a contradiction. Therefore a is an indecomposable α0 (D)-cover
of Υ(D), as required. 

3. Indecomposable parallelizations and Hilbert bases


Let C be a clutter with vertex set X = {x1 , . . . , xn } and let C1 , . . . , Cs be the
minimal vertex covers of C. For 1 ≤ k ≤ n, we denote the characteristic vector of
Ck by uk .
The Simis cone of I = I(C) is the rational polyhedral cone:
Cn(I) = He+1 ∩ · · · ∩ He+n+1 ∩ H(u
+
1 ,−1)
∩ · · · ∩ H(u
+
s ,−1)
.
Here Ha+ denotes the closed halfspace Ha+ = {x| x, a ≥ 0} and Ha stands for the
hyperplane through the origin with normal vector a. Simis cones were introduced
in [14] to study symbolic Rees algebras of square-free monomial ideals. The term
Simis cone is intended to do homage to Aron Simis [26, 27, 28]. The Simis cone
is a pointed rational polyhedral cone. By [24, Theorem 16.4] there is a unique
minimal finite set of integral vectors
H = {h1 , . . . , hr } ⊂ Zn+1
such that Zn+1 ∩ R+ H = NH and Cn(I) = R+ H (minimal relative to taking
subsets), where R+ H denotes the cone generated by H consisting of all linear com-
binations of H with non-negative real coefficients and NH denotes the semigroup
generated by H consisting of all linear combinations of H with coefficients in N.
The set H is called the Hilbert basis of Cn(I). The Hilbert basis of Cn(I) has the
following useful description.
Theorem 3.1. [24, p. 233] H is the set of all integral vectors 0
= h ∈ Cn(I)
such that h is not the sum of two other non-zero integral vectors in Cn(I).
Corollary 3.2. Let H be the Hilbert basis of Cn(I). Then
(3.1) H = {(a, b)| xa tb is a minimal generator of Rs (I)}
(3.2) = {(a, α0 (C))| C a is an indecomposable parallelization of C}
and Rs (I) is equal to the semigroup ring K[NH] of NH.
SYMBOLIC REES ALGEBRAS OF EDGE IDEALS 161
11

Proof. The first equality follows from Lemma 2.1 and Theorem 3.1. The
second equality follows from Theorem 2.6. The equality K[NH] = NH was first
observed in [14, Theorem 3.5]. 
This result is interesting because it allows to compute all indecomposable par-
allelizations of C and all indecomposable induced subclutters of C using Hilbert
bases. In particular, as is seen in Corollary 3.3, we can use this result to decide
whether any given graph or clutter is indecomposable (see Example 3.4).
The indecomposable subclutters can be computed using the next consequence
of Corollary 3.2.
Corollary 3.3. Let C be a clutter and let α = (a1 , . . . , an , b) be a vector in
{0, 1}n × N. Then α is in the Hilbert basis of Cn(I(C)) if and only if the induced
subclutter H = C[{xi | ai = 1}] is indecomposable with b = α0 (H).
Example 3.4. Consider the graph G shown below. Let I be the edge ideal
of G and let H be the Hilbert basis of Cn(I). Using Corollary 3.2, together with
Normaliz [4], it is seen that G has exactly 61 indecomposable parallelizations and
49 indecomposable subgraphs. Since α0 (G) = 6 and the vector (1, . . . , 1, 6) is not
in H we obtain that G is a decomposable graph.

x x
s9 s 10
Q Q
 Q  Q
s
x5  Qs x s x7
 Qs x8
A x2 s A x 
6
x1As As 3 s x
4

Fig. 4. Decomposable graph G

The vector a = (1, . . . , 1, 2, 7) is in H, i.e., G(1,...,1,2) is indecomposable and has


covering number 7.

s x10
sx9
J sx10
Q
J Q
 Q  Q
s
x5  Qs
x Js x7 Qs x8
A x2   6
A x 
x1As s As 3 sx4

Fig. 5. Indecomposable graph G(1,...,1,2)


The next result, together with Corollary 3.3, allows to locate all induced odd
cycles (odd holes) and all induced complements of odd cycles (odd antiholes).
Lemma 3.5. Let Cn = {x1 , . . . , xn } be a cycle. (a) If n ≥ 5 is odd, then the
complement Cn of Cn is an indecomposable graph, (b) if n is odd, then Cn is an
indecomposable cycle, and (c) any complete graph is indecomposable.
Proof. (a) Assume that G = Cn is decomposable. Then there are disjoint
sets X1 , X2 such that V (G) = X1 ∪ X2 and α0 (G) = α0 (G[X1 ]) + α0 (G[X2 ]).
Since β0 (G) = 2, it is seen that G[Xi ] is a complete graph for i = 1, 2. We
may assume that x1 ∈ X1 . Then x2 must be in X2 , otherwise {x1 , x2 } is an
162
12 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

edge of G[X1 ], a contradiction. By induction it follows that x1 , x3 , x5 , . . . , xn are


in X1 . Consequently {x1 , xn } is an edge of G[X1 ], a contradiction. Thus G is
indecomposable. (b) This was observed in [19]. (c) Follows readily from the fact
that the covering number of a complete graph in r vertices is r − 1. 
Example 3.6. Consider the graph G of Fig. 6, where vertices are labeled
with i instead of xi . Using Corollary 3.2, together with Normaliz [4], it is seen
that G has exactly 21 indecomposable parallelizations, 20 of which correspond to
indecomposable subgraphs. Apart from the seven vertices, the nine edges, one
triangle and three pentagons, the only indecomposable parallelization of G which
is not a subgraph is the duplication shown in Fig 7.
t
1 1
t
@@ @@
@ t1
X @
 X XXX
5 t 6 t @t2 5 t @
 @6 t @t2
7 t 7 t
 HH 
 HH
4 t  HHt3 4 t HHt3

Fig. 6. Decomposable graph G Fig. 7. Indecomposable graph G(2,1,1,1,1,1,1)

Example 3.7. Consider the graph G of Fig. 8. Using Corollary 3.2 and
Normaliz [4], it is seen that G has exactly 103 indecomposable parallelizations,
92 of which correspond to indecomposable subgraphs. The only indecomposable
parallelization Ga which do not delete vertices is that obtained by duplication of
the five outer vertices, i.e., a = (2, 2, 2, 2, 2, 1, 1, 1, 1, 1) and α0 (Ga ) = 11.
s1
@ L
L@
!6s L @
a
! ! D a L a@
5s! ! 10 s  D Ls7 aa @s2
LZZ @  D D 
L Z @  D  D
L  Z9 s @D s
8
D
H
H 
L  @ D

H
4 Ls HD s 3
@
Fig. 8. Decomposable graph G

4. Symbolic Rees algebras and perfect graphs


We now turn our attention to the indecomposability of graphs and its connec-
tion with the theory of perfect graphs. Examples of indecomposable graphs include
complete graphs, odd cycles, and complements of odd cycles of length at least 5
(see Lemma 3.5).
Let us recall the notion of a perfect graph that was introduced by Berge [2,
Chapter 16]. A colouring of the vertices of a graph G is an assignment of colours
to the vertices of G in such a way that adjacent vertices have distinct colours.
The chromatic number of G is the minimal number of colours in a colouring of G.
A graph is perfect if for every induced subgraph H, the chromatic number of H
SYMBOLIC REES ALGEBRAS OF EDGE IDEALS 163
13

equals the size of the largest complete subgraph of H. We refer to [6, 18, 25] for
the theory of perfect graphs.
The next result shows that indecomposable graphs occur naturally in the theory
of perfect graphs.
Proposition 4.1. [12, Proposition 2.13] A graph G is perfect if and only if
the indecomposable parallelizations of G are exactly the complete subgraphs of G
Let G be a graph. We denote a complete subgraph of G with r vertices by Kr .
The empty set is regarded as an independent set of vertices whose characteristic
vector is the zero vector. A clique of G is a subset of the set of vertices that induces
a complete subgraph. The support of a monomial xa = xa1 1 · · · xann , denoted by
supp(xa ), is the set supp(xa ) = {xi | ai > 0}. If ai ∈ {0, 1} for all i, xa is called a
square-free monomial.
The next major result shows that the symbolic Rees algebra of the edge ideal
of a perfect graph G is completely determined by the cliques of G. This was first
shown in [32] using polyhedral geometry.
Corollary 4.2. [32, Corollary 3.3] If G is a perfect graph, then
Rs (I(G)) = K[{xa tb | xa is square-free ; G[supp(xa )] = Kb+1 }].
Proof. Let xa tb be a minimal of Rs (I(G)). By Theorem 2.6 Ga is an inde-
composable graph and b = α0 (Ga ). As G is perfect, by Proposition 4.1, we obtain
that Ga is a complete subgraph of G with b + 1 vertices. 
Since complete graphs are perfect, an immediate consequence is:
Corollary 4.3. [1] If G is a complete graph, then
Rs (I(G)) = K[{xa tb | xa is square-free ; deg(xa ) = b + 1}].

References
1. C. Bahiano, Symbolic powers of edge ideals, J. Algebra 273 (2004), no. 2, 517–537.
2. C. Berge, Graphs and hypergraphs, North-Holland Mathematical Library, Vol. 6, North-
Holland Publishing Co., Amsterdam-London; American Elsevier Publishing Co., Inc., New
York, 1976.
3. J. P. Brennan, L. A. Dupont and R. H. Villarreal, Duality, a-invariants and canonical modules
of rings arising from linear optimization problems, Bull. Math. Soc. Sci. Math. Roumanie
(N.S.) 51 (2008), no. 4, 279–305.
4. W. Bruns and B. Ichim, Normaliz 2.0, Computing normalizations of affine semigroups 2008.
Available from http://www.math.uos.de/normaliz.
5. M. Chudnovsky, N. Robertson, P. Seymour and R. Thomas, The strong perfect graph theorem,
Ann. of Math. (2) 164 (2006), no. 1, 51–229.
6. G. Cornuéjols, Combinatorial optimization: Packing and covering, CBMS-NSF Regional Con-
ference Series in Applied Mathematics 74, SIAM (2001).
7. R. Diestel, Graph Theory, Graduate Texts in Mathematics 173, Springer-Verlag, New York,
2nd ed., 2000.
8. L. A. Dupont, C. Renterı́a and R. H. Villarreal, Systems with the integer rounding property
in normal monomial subrings, An. Acad. Brasil. Ciênc 82 (2010), no. 4, 801-811.
9. L. A. Dupont, E. Reyes and R. H. Villarreal, Cohen-Macaulay clutters with combinatorial
optimization properties and parallelizations of normal edge ideals, São Paulo J. Math. Sci. 3
(2009), no. 1, 61–75.
10. L. A. Dupont and R. H. Villarreal, Edge ideals of clique clutters of comparability graphs and
the normality of monomial ideals, Math. Scand. 106 (2010), no. 1, 88–98.
11. L. A. Dupont and R. H. Villarreal, Algebraic and combinatorial properties of ideals and
algebras of uniform clutters of TDI systems, J. Comb. Optim. 21 (2011), no. 3, 269-292.
164
14 MARTÍNEZ-BERNAL, RENTERÍA-MÁRQUEZ, AND VILLARREAL

12. L. A. Dupont and R. H. Villarreal, Symbolic Rees algebras, vertex covers and irreducible
representations of Rees cones, Algebra Discrete Math. 10 (2010), no. 2, 64–86.
13. P. Erdös and T. Gallai, On the minimal number of vertices representing the edges of a graph,
Magyar Tud. Akad. Mat. Kutató Int. Közl. 6 (1961), 181–203.
14. C. Escobar, R. H. Villarreal and Y. Yoshino, Torsion freeness and normality of blowup rings
of monomial ideals, Commutative Algebra, Lect. Notes Pure Appl. Math. 244, Chapman &
Hall/CRC, Boca Raton, FL, 2006, pp. 69–84.
15. C. A. Francisco, H. T. Hà and A. Van Tuyl, Associated primes of monomial ideals and odd
holes in graphs, J. Algebraic Combin. 32 (2010), no. 2, 287-301.
16. I. Gitler, E. Reyes and R. H. Villarreal, Blowup algebras of square–free monomial ideals and
some links to combinatorial optimization problems, Rocky Mountain J. Math. 39 (2009), no.
1, 71–102.
17. I. Gitler, C. Valencia and R. H. Villarreal, A note on Rees algebras and the MFMC property,
Beiträge Algebra Geom. 48 (2007), no. 1, 141–150.
18. M. C. Golumbic, Algorithmic graph theory and perfect graphs, second edition, Annals of
Discrete Mathematics 57, Elsevier Science B.V., Amsterdam, 2004.
19. F. Harary and M. D. Plummer, On indecomposable graphs, Canad. J. Math. 19 (1967),
800–809.
20. J. Herzog, T. Hibi and N. V. Trung, Symbolic powers of monomial ideals and vertex cover
algebras, Adv. Math. 210 (2007), 304–322.
21. G. Lyubeznik, On the arithmetical rank of monomial ideals, J. Algebra 112 (1988), 86–89.
22. J. Martı́nez-Bernal, E. O’Shea and R. H. Villarreal, Ehrhart clutters: regularity and max-flow
min-cut, Electron. J. Combin. 17 (2010), no. 1, R52.
23. A. Schrijver, On total dual integrality, Linear Algebra Appl. 38 (1981), 27–32.
24. A. Schrijver, Theory of Linear and Integer Programming, John Wiley & Sons, New York,
1986.
25. A. Schrijver,Combinatorial Optimization, Algorithms and Combinatorics 24, Springer-Verlag,
Berlin, 2003.
26. A. Simis, Combinatoria Algebrica, XVIII Coloquio Brasileiro de Matematica, IMPA, 1991
(Apendice. Palimpsesto 2: Potencias simbolicas, 2.1).
27. A. Simis and B. Ulrich, On the ideal of an embedded join, J. Algebra 226 (2000), no. 1, 1–14.
28. A. Simis, W. V. Vasconcelos and R. H. Villarreal, On the ideal theory of graphs, J. Algebra,
167 (1994), 389–416.
29. S. Sullivant, Combinatorial symbolic powers, J. Algebra 319 (2008), no. 1, 115–142.
30. W. V. Vasconcelos, Arithmetic of Blowup Algebras, London Math. Soc., Lecture Note Series
195, Cambridge University Press, Cambridge, 1994.
31. R. H. Villarreal, Monomial Algebras, Dekker, New York, N.Y., 2001.
32. R. H. Villarreal, Rees algebras and polyhedral cones of ideals of vertex covers of perfect graphs,
J. Algebraic Combin. 27 (2008), no. 3, 293–305.

Departamento de Matemáticas, Centro de Investigación y de Estudios Avanzados


del IPN, Apartado Postal 14–740, 07000 Mexico City, D.F.
E-mail address: jmb@math.cinvestav.mx

Departamento de Matemáticas, Escuela Superior de Fı́sica y Matemáticas, Insti-


tuto Politécnico Nacional, 07300 Mexico City, D.F.
E-mail address: renteri@esfm.ipn.mx

Departamento de Matemáticas, Centro de Investigación y de Estudios Avanzados


del IPN, Apartado Postal 14–740, 07000 Mexico City, D.F.
E-mail address: vila@math.cinvestav.mx
Contemporary Mathematics
Volume 555, 2011

Reconciling Riemann-Roch results

Paul Roberts and Anurag K. Singh

In honor of the contributions of Wolmer Vasconcelos

Abstract. In the course of their work on the homological conjectures, Peskine


and Szpiro proved a Riemann-Roch formula for graded modules; we show that
this agrees with the Hirzebruch-Riemann-Roch formula in the case of graded
modules over polynomial rings.

In 1974 Peskine and Szpiro [PS] proved a number of conjectures on intersec-


tion multiplicities of graded modules, using a formula that they developed for the
Hilbert polynomials of such modules. This formula was considered to be a kind
of Riemann-Roch formula, and indeed it was one of the inspirations for the lo-
cal Chern characters and the Riemann-Roch formula for local rings, developed by
Baum, Fulton, and MacPherson [BFM].
In this paper we do not discuss the connections to questions on multiplicities,
but look instead at the formula of Peskine and Szpiro and examine the extent to
which it may be considered a Riemann-Roch formula; in other words, we compare
it to the Riemann-Roch formula of Hirzebruch. In the case of perfect complexes
over polynomial rings, we show that the formulae agree in a precise sense.

1. The Peskine-Szpiro formula


Let A be an N-graded ring such that A0 is a field K, and A is generated over A0
by finitely many elements of A1 . For n an integer, A(n) will denote the module A
with the shifted grading A(n)k = An+k for each k. By a perfect complex F• we
mean a bounded complex
(1.1) 0 −→ Fs −→ Fs−1 −→ · · · −→ F0 −→ 0 ,
where each Fi is a finite direct sum of copies of A(n)—for varying n—and such that
the homomorphisms in F• preserve degrees. Set

βi
Fi = A(nij ) for each i .
j=1

2010 Mathematics Subject Classification. Primary 14C40; Secondary 13A02, 13D22, 13H15.
P.R. and A.K.S. were supported in part by grants from the National Science Foundation.

2011
c 0000
c Mathematical
American (copyright Society
holder)

1
165
166
2 PAUL ROBERTS AND ANURAG K. SINGH

Recall that if M is a finitely generated graded A-module, the Hilbert polynomial


of M is the polynomial PM (x) with the property that PM (n) agrees with rankK Mn
for sufficiently large integers n. The Peskine-Szpiro result is a formula for the
alternating sum of the Hilbert polynomials of the homology modules of F• in terms
of the integers nij :
For each integer k  0, set
1  
s βi
i
ρk = (−1) nkij .
k! i=0 j=1

Then the Peskine-Szpiro formula is


 s  (k)
(1.2) (−1)i PHi (x) = ρk PA (x) ,
i=0 k0
(k)
where Hi denotes the i-th homology module of the complex F• and PA (x) is the
k-th derivative of the polynomial PA (x).

2. Chern classes
We briefly review some material that may be found in [Ha] or [Fu]. Let X be
a nonsingular projective variety of dimension d over a field K. A cycle of codimen-
sion k on X is an element of the free abelian group generated by closed irreducible
subvarieties of X having codimension k. The group CHk (X) consists of cycles
of codimension
 k modulo rational equivalence. Cycles of codimension d have the
form i ni Pi for points Pi of X, and one has a group homomorphism
 
(2.1) deg : CHd (X) −→ Z where deg ni Pi = ni .
i i
The intersection pairing on X provides

d
CHr (X)
r=0

with the structure of a commutative ring, the Chow ring of X, denoted CH(X).
Extending the correspondence between invertible sheaves and divisors, for each
locally free sheaf F on X—say of rank r—there exist Chern classes ci = ci (F)
in CHi (X), where c0 = 1 and ci = 0 for all i > r. The Chern polynomial of F is
ct (F) = 1 + c1 t + c2 t2 + · · · + cr tr .
For an exact sequence of locally free sheaves
(2.2) 0 −→ F  −→ F −→ F  −→ 0 ,
the Whitney sum formula states that
ct (F) = ct (F  )ct (F  ) .
For the purposes of a Riemann-Roch formula, one wants to associate to locally free
sheaves, invariants that are additive on exact sequences. Towards this, factor the
polynomial ct (F) formally as

r
ct (F) = (1 + αi t) ;
i=1
RECONCILING RIEMANN-ROCH RESULTS 167
3

the αi are the Chern roots of F. Working in CH(X)Q = CH(X) ⊗Z Q, define the
Chern character of F as
r
1
ch(F) = eαi , where ex = 1 + x + x2 + · · · .
i=1
2
Since ch(F) is a symmetric function of the Chern roots α1 , . . . , αr , and the Chern
classes c1 , . . . , cr are precisely the elementary symmetric polynomials in α1 , . . . , αr ,
one can express ch(F) in terms of the Chern classes; the first few terms, as may be
found in [Ha, page 432] or [Fu, Example 3.2.3], are
1 1
ch(F) = r + c1 + (c21 − 2c2 ) + (c31 − 3c1 c2 + 3c3 ) + · · · .
2 6
The Whitney sum formula now yields
ch(F) = ch(F  ) + ch(F  ) ,
i.e., Chern characters are additive on short exact sequences.
The Chern character of a tensor product of locally free sheaves is
(2.3) ch(F ⊗ F  ) = ch(F) ch(F  ) .
The formal power series
1 − e−x x x2 x3
= 1− + − + ···
x 2! 3! 4!
is an invertible element of Q[[x]]; we denote its inverse by
(2.4) Q(x) = x/(1 − e−x ) .
The occurrence of this power series will be justified in Example 3.1. For now, we
conclude this section with one last definition: the Todd class of a locally free sheaf
F with Chern roots α1 , . . . , αr is

r
td(F) = Q(αi ) .
i=1
Once again, this is a symmetric function of α1 , . . . , αr , so it can be expressed in
terms of the Chern classes; indeed, one has
1 1 1
td(E) = 1 + c1 + (c21 + c2 ) + (c1 c2 ) + · · · .
2 12 24
For an exact sequence (2.2), the Whitney sum formula implies
(2.5) td(F) = td(F  ) td(F  ) .

3. The Riemann-Roch Theorem


Let F be a locally free sheaf on a nonsingular projective variety X of dimen-
sion d. The Euler characteristic of F is the alternating sum of the ranks of the
sheaf cohomology groups H i (X, F), i.e.,

d
χ(F) = (−1)i rankK H i (X, F) .
i=0
Let TX be the tangent sheaf of X. The Riemann-Roch theorem of Hirzebruch states
χ(F) = deg [ch(F) td(TX )]d ,
168
4 PAUL ROBERTS AND ANURAG K. SINGH

where [−]d denotes the component in CHd (X)Q and deg : CHd (X)Q −→ Q extends
the homomorphism deg : CHd (X) −→ Z of (2.1); see [Ha, Appendix A].
Let L• be a bounded complex of locally free sheaves on X, say
0 −→ Ls −→ Ls−1 −→ · · · −→ L0 −→ 0 .
The Euler characteristic of L• is

s
χ(L• ) = (−1)i χ(Li ) .
i=0
 i
Setting ch(L• ) = i (−1) ch(Li ), the Riemann-Roch theorem takes the form
χ(L• ) = deg [ch(L• ) td(T )]d .
Example 3.1. We examine the Riemann-Roch theorem in the case X is projec-
tive space Pd , and provide some justification for the choice of the power series Q(x)
used in the definition of the Todd class; see (2.4).
Any subvariety of degree k in Pd is linearly equivalent to k times a linear space
of the same dimension, so
CH(Pd )Q = Q[h]/(hd+1 ) ,
where h is the class of a hyperplane; the class of a point in Pd is identified with
hd ∈ CHd (Pd )Q .
By the exact sequence
0 −→ OPd −→ OPd (1)d+1 −→ TPd −→ 0
and (2.5), one has
 
td(TPd ) = td OPd (1)d+1 = td(h)d+1 .
Taking F to be OPd in the Riemann-Roch theorem, we see that
1 = deg [td(TPd )]d = deg [td(h)d+1 ]d .
Thus, td(h) is a power series such that xd occurs with unit coefficient in td(h)d+1 .
It turns out that
Q(x) = x/(1 − e−x )
is the only power series in Q[[x]] with the property that xd occurs with unit coeffi-
cient in Q(x)d+1 for each d  0; this can be proved using the recursion (4.3).
It follows from the above discussion that
td(TPd ) = Q(h)d+1 .

4. A comparison
We now reconcile the Riemann-Roch theorem of Hirzebruch with the formula
of Peskine and Szpiro; first, some notation:
We use  p(x) to denote the coefficient of xk in a polynomial or formal power
k
series p(x).
Let A be a standard graded polynomial ring over a field. A perfect complex of
A-modules (1.1) defines a complex of locally free sheaves on Proj A, which is Pd .
The sheaves that occur are direct sums of OPd (n) for varying n.
RECONCILING RIEMANN-ROCH RESULTS 169
5

We now translate the Peskine-Szpiro formula into these terms. The modules
Fi = j A(nij ) in (1.1) define sheaves j OPd (nij ) on Pd . The complex F• thus
yields a complex L• of locally free sheaves Li , where

βi
Li = OPd (nij ) .
j=1

The Chern polynomial of OPd (1) is 1 + ht, for h the class of a hyperplane. Hence
ch(OPd (1)) = eh ,
and using (2.3), it follows that the Chern character of OPd (n) is
ch(OPd (n)) = enh
n2 h2 n3 h3 nd hd
= 1 + nh + + + ···+ in Q[h]/(hd+1 ) .
2! 3! d!
Since Chern characters are additive on short exact sequences—in particular, on
direct sums—the Chern character of L• is

s
ch(L• ) = (−1)i ch(Li )
i=0


s 
βi
= (−1)i ch(OX (nij ))
i=0 j=1


s 
βi
 n2ij h2 ndij hd 
= (−1)i 1 + nij h + + ··· + .
i=0 j=1
2! d!

Collecting coefficients, we obtain that for each k the coefficient of hk is



s 
βi
nkij
(−1)i .
i=0 j=1
k!

Since this is precisely ρk , we see that


ch(L• ) = ρ0 + ρ1 h + ρ2 h2 + · · · + ρd hd .
Thus, the quantities ρk in the Peskine-Szpiro formula occur as the components of
the Chern character in the Riemann-Roch Theorem. We now look at the other half
of the theorem, namely the Todd class.
In the Riemann-Roch Theorem for projective space Pd , the Todd class of the
tangent bundle is td(TPd ) = Q(h)d+1 , where Q(x) = x/(1 − e−x ); see Example 3.1.
Since hd+1 = 0, this may be written as
td(TPd ) = a0 + a1 h + · · · + ad hd ,
where the ak are rational numbers. As deg hd = 1 in CHd (Pd )Q , the Riemann-Roch
Theorem states that
χ(L• ) = ρd a0 + ρd−1 a1 + · · · + ρ0 ad , where ak =  Q(x)d+1 .
k

To compare this with the Peskine-Szpiro formula (1.2), we need to know the
relation between the Hilbert polynomial of A(n) and the Euler characteristic of
170
6 PAUL ROBERTS AND ANURAG K. SINGH

OPd (n). This is very simple: if M is a finitely generated graded A-module with
Hilbert polynomial PM (x) and F = M the associated coherent sheaf, then

PM (n) = χ(F(n)) for all integers n ,

where F(n) = F ⊗ OPd (n). In particular, the Euler characteristic χ(F) equals
PM (0). Hence, the Peskine-Szpiro formula yields


s  (k)
χ(L• ) = (−1)i PHi (0) = ρk PA (0) .
i=0 k0

(d−i)
Setting bi = PA (0) for i = 0, . . . , d, we have

χ(L• ) = ρd b0 + ρd−1 b1 + · · · + ρ0 bd .

In this formula, bk is the constant term of the (d − k)-th derivative of the


polynomial PA (x), i.e.,
bk = (d − k)!  PA (x) .
d−k
x+d
Since A is the polynomial ring in d + 1 variables, its Hilbert polynomial is d ,
and so


x+d
bk = (d − k)!  .
d−k d
Thus, we have two similar formulae for the Euler characteristic of L• , involving
the sequences a0 , a1 , . . . , ad and b0 , b1 , . . . , bd ; the first sequence is given by the first
−x
d+1 coefficients of the power series Q(x)d+1 , where Q(x) = x/(1−e  ); the second
x+d
sequence is derived from the coefficients of the polynomial d . It is by no means
a priori obvious that these sequences are related; the remainder of this section is
devoted to giving a direct proof that these sequences are indeed the same.

Proposition 4.1. Let d be a nonnegative integer. Then, for each integer k


with 0  k  d, the coefficient of xk in the formal power series

d+1
x
1 − e−x

agrees with the coefficient of xd−k in the polynomial




x+d (d − k)!
(d − k)! = (x + 1)(x + 2) · · · (x + d) .
d d!
Proof. Set


x+d
q(d, k) = (d − k)!  .
d−k d
Note that q(0, 0) = 1, and that q(d, k) = 0 unless 0  k  d. We claim that

d−k
(4.1) q(d, k) = q(d − 1, k − 1) + q(d − 1, k) .
d
RECONCILING RIEMANN-ROCH RESULTS 171
7

This holds since


d−k
q(d − 1, k − 1) + q(d − 1, k)

d

x+d−1 d−k x+d−1
= (d − k)!  + (d − k − 1)! 
d−k d−1 d d−k−1 d−1
(d − k)! (d − k)!
=  (x + 1) · · · (x + d − 1) +  (x + 1) · · · (x + d − 1)
(d − 1)! d−k d! d−k−1
(d − k)! (d − k)!
=  (x + 1) · · · (x + d − 1)d +  (x + 1) · · · (x + d − 1)x
d! d−k d! d−k
(d − k)!
=  (x + 1) · · · (x + d − 1)(x + d)
d! d−k
= q(d, k) .
Next, consider the polynomials

qd = q(d, k)xk , where d  0 .
k

Using the recursion relation (4.1), it follows that


 d−k 
qd = q(d − 1, k − 1)xk + q(d − 1, k)xk
d
k k
  kx 
= x q(d − 1, k − 1)xk−1 + q(d − 1, k)xk − q(d − 1, k)xk−1
d
k k k
x 
= xqd−1 + qd−1 − qd−1 ,
d

where qd−1 denotes the derivative of qd−1 with respect to x. Thus, we have
x 
(4.2) qd = (1 + x)qd−1 − q for d  1 , and q0 = 1 .
d d−1
The polynomials qd —and hence the numbers q(d, k)—are determined by (4.2).
Next, we claim that the formal power series

d+1
x
Qd = , where d  0 ,
1 − e−x
satisfy a similar recursion, namely
x 
(4.3) Qd = (1 + x)Qd−1 − Q for d  1 .
d d−1
The derivative of x/(1 − e−x ) may be computed as


−1
d x d 1 − e−x
=
dx 1 − e−x dx x

−2
−x
1 − e−x xe − 1 + e−x
= −
x x2
−x −x
1 − e − xe
= .
(1 − e−x )2
172
8 PAUL ROBERTS AND ANURAG K. SINGH

Hence

d
d
x  x x d x
(1 + x)Qd−1 − Qd−1 = (1 + x) −
d 1 − e−x d dx 1 − e−x

d
d−1

x x 1 − e−x − xe−x
= (1 + x) − x
1 − e−x 1 − e−x (1 − e−x )2

d

x 1 − e−x − xe−x
= 1 + x −
1 − e−x 1 − e−x

d+1
x
= ,
1 − e−x
which proves (4.3).
The proposition asserts that the coefficients of xk in Qd and qd agree for each
k with 0  k  d, i.e., that
 
Qd − qd ∈ xd+1 Q[[x]] .
We prove this by induction on d; the case d = 0 is readily checked. Assuming the
result for d − 1, we have
Qd−1 − qd−1 = xd E for some E ∈ Q[[x]] .
But then, using (4.3) and (4.2), we have
  x  

Qd − qd = (1 + x) Qd−1 − qd−1 − Q − qd−1
d d−1
x  d 
= (1 + x)xd E − x E
d
x  
= (1 + x)xd E − dxd−1 E + xd E 
d
 1 
= x d+1
E− E . 
d
References
[BFM] P. Baum, W. Fulton, and R. MacPherson, Riemann-Roch for singular varieties, Inst.
Hautes Études Sci. Publ. Math. 45 (1975), 101–145.
[Fu] W. Fulton, Intersection Theory, Springer-Verlag, Berlin, 1984.
[Ha] R. Hartshorne, Algebraic geometry, Graduate Texts in Mathematics 52, Springer-Verlag,
New York-Heidelberg, 1977.
[PS] C. Peskine and L. Szpiro, Syzygies et multiplicités, C. R. Acad. Sci. Paris Sér. A Math.
278 (1974), 1421–1424.
[Sz] L. Szpiro, Sur la théorie des complexes parfaits, in: Commutative algebra (Durham 1981),
London Math. Soc. Lecture Note Series 72 1982, 83–90.

Department of Mathematics, University of Utah, 155 South 1400 East, Salt Lake
City, UT 84112, USA
E-mail address: roberts@math.utah.edu

Department of Mathematics, University of Utah, 155 South 1400 East, Salt Lake
City, UT 84112, USA
E-mail address: singh@math.utah.edu
Contemporary Mathematics
Volume 555, 2011

Hilbert Functions of Cohen-Macaulay local rings

Maria Evelina Rossi

Abstract. This survey is based on a series of lectures given by the author


at the School in Commutative Algebra and its Connections to Geometry, Uni-
versidade Federal de Pernambuco, Olinda (Brasil), August 2009, in honor of
V.W. Vasconcelos.

Contents
Introduction
1. Basic facts
2. Superficial elements
3. 1-dimensional local Cohen-Macaulay rings
4. Hilbert coefficients and classical bounds
5. Hilbert Functions of Artinian Gorenstein rings
References

Bibliography

Introduction

The notion of Hilbert function is a central tool in commutative algebra and in


algebraic geometry and is becoming increasingly important in combinatorics and in
computational algebra. The Hilbert function of the homogeneous coordinate ring
of a projective variety V was classically called the postulation of V and it is a rich
source of discrete invariants of V and of its embedding. The dimension, the degree
and the arithmetic genus of V, can be computed from the Hilbert function or from
the Hilbert polynomial of its coordinate ring.
In this survey we mainly deal with the Hilbert function of Cohen-Macaulay
local rings and our aim is to introduce the reader to some aspects of this area of
dynamic mathematical activity.

2010 Mathematics Subject Classification. 13A02, 13A30,13C14, 13C15, 13H15.


Key words and phrases. Hilbert function, associated graded ring, Cohen-Macaulay ring,
Gorenstein local ring, superficial sequence, multiplicity, Hilbert coefficients.
The author is personally grateful to the organizers of the School for giving her the opportunity
to teach the favorite topic. A particular thank you goes to W. Vasconcelos because his teaching
and his example have influenced her mathematical life greatly.
1 2011
c American Mathematical Society

173
174
2 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

The Hilbert function of a local ring (A, m, k) is, by definition, the Hilbert func-
tion of the associated graded ring grm (A). The standard graded k-algebra grm (A)
arises from a relevant geometric construction and it has been studied extensively.
Namely, if A is the localization at the origin of the coordinate ring of an affine
variety V passing through 0, then grm (A) is the coordinate ring of the tangent cone
of V , which is the cone composed of all lines that are limiting positions of secant
lines to V in 0. The P roj of this algebra can also be seen as the exceptional set of
the blowing-up of V in 0.
Despite the fact that the Hilbert function of a Cohen-Macaulay graded standard
algebra is well understood by means of Macaulay’s Theorem, very little is known
in the local case. One of the main problems is whether geometric and homological
properties of the local ring A can be carried on the corresponding tangent cone
grm (A). For example if a given local domain has fairly good properties, such as
normality or Cohen-Macaulayness, its depth provides in general no information on
the depth of the associated graded ring. It could be interesting to remind that a
still wide open problem is to characterize the Hilbert function of an affine curve
in A3 whose defining ideal is a complete intersection, while the Hilbert function
of any complete intersection of homogeneous forms is well known in terms of the
degrees of the generators.
In this presentation basic facts will be introduced, the use of fruitful techniques
will be stressed in order to present easier proofs of known facts and to get a taste of
some open problems. Starting from classical results by S. Abhyankar, D. Northcott
and J. Sally, we present results on the coefficients of the Hilbert polynomial and, in
several cases, we discuss algebraic and geometric properties of the local ring in terms
of this asymptotic information. We shall focus our attention on the one-dimensional
case because it plays an important role in our approach, in particular we provide
tricky proofs proving the non-decreasing of the Hilbert function for several classes
of one-dimensional Cohen-Macaulay local rings. The survey ends with results and
techniques concerning the Hilbert function of Artinian Gorenstein local k-algebras.
This investigation is strongly motivated by the interest related to the study of the
Punctual Hilbert scheme and of the rationality of the Poincaré series of Gorenstein
local rings (see for example [6, 7, 8, 9, 16, 17, 36]).
For more details and complete proofs we will refer to the monograph [47] writ-
ten jointly with G. Valla, and to a recent paper [14] with J. Elias. For further
reading on the same topic we would also suggest a series of lectures on problems
and results on Hilbert functions of graded algebras given by G. Valla in Bellaterra
(Spain) (see [56]), a survey by A. Corso and C. Polini on the Hilbert coefficients
with a view toward blowup algebras (see [10]) and a survey on the Hilbert Func-
tions of filtered modules by the author, 4-th Joint Seminar Japan-Vietnam, Meij
University (2009) (see [40]). We present here several examples, all performed with
the help of CoCoA [61].

1. Basic facts
Let (A, m, k) be a Noetherian local ring with maximal ideal m and residue field
k. Denote by μ(I) the minimal number of generators of an ideal I of A. The Hilbert
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 175
3

function of A is, by definition, the numerical function HFA : N → N such that


HFA (n) := dimk mn /mn+1 = μ(mn )
for every n ∈ N. Hence HFA is also the Hilbert function of the homogeneous
k-standard algebra
grm (A) = ⊕n≥0 mn /mn+1
which is the associated graded ring of m.

The generating function of HFA is a power series HSA (z) = n≥0 HFA (n)z n .
This series, by the Hilbert-Serre Theorem, is rational and it can be written as
hA (z)
HSA (z) = ,
(1 − z)d
where hA (z) is a polynomial with integer coefficients such that hA (1) = 0. More-
over, d is the Krull dimension of A, which is also the Krull dimension of grm (A).
The numerator hA (z) is also called the h-polynomial of A. For large n, HFA (n)
agrees with a polynomial HPA (X) with rational coefficients and degree d − 1. It is
called the Hilbert polynomial of A. We will denote by λ(M ) the length of M as A-
module. Classically, another polynomial has been introduced, the Hilbert-Samuel
polynomial of A, that is λ(A/mn+1 ) for n  0. It is denoted by PA (X) and
 d  
i X +d−i
(1.1) PA (X) = (−1) ei (m)
i=0
d−i
 
where X+j j := (X+j)(X+j−1)···(X+1)
j! .
The integers e0 (m), e1 (m), . . . , ed (m) are uniquely determined by m and are known
as the Hilbert coefficients of A. In particular e0 (m) is the multiplicity of A. If there
is not risk of confusion, we shall simply write ei instead of ei (m). We can prove that
for every i ≥ 0
(i)
h (1)
ei = A
i!
(0)
where 0! = 1 and hA (1) = hA (1).
 m)
If (A,  denotes the completion of A with respect to the m-adic filtration, it
is well-known that

 (A)  grm (A)
grm
as graded rings, hence studying the Hilbert function we may assume that A is
complete. If A is equicharacteristic (for example if A is a k-algebra), we also
assume A = R/I where I is an ideal in the power series R = k[[x1 , . . . , xn ]]. This
assumption is not restrictive by the well-known Cohen’s theorem.
If a ∈ A is a non zero element and a ∈ mr , a ∈ mr+1 , then we denote by

a := a ∈ mr /mr+1 and call it the initial form of a in grm (A). If (R, n) is a local
ring and A = R/I, it is easy to prove that
grm (A)  grn (R)/I ∗
where I ∗ is the ideal generated by the initial forms of elements of I in grn (R). If
R = k[[x1 , . . . , xn ]], then nr /nr+1  k[x1 , . . . , xn ]r , the k-vector space generated by
all the forms of degree r, hence grn (R)  k[x1 , . . . , xn ].
176
4 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

If A is a Cohen-Macaulay graded k-algebra, the Hilbert Function is well-


understood: its Hilbert series is
hA (z)
HSA (z) =
(1 − z)d
where, by Macaulay’s theorem, hA (z) is the Hilbert series of an Artinian graded
k-algebra. The same does not hold true if A is a Cohen-Macaulay local ring.
The following example gives a measure of the complexity of the problem.
Example 1.1. Consider the coordinate ring A of the monomial curve parametrized
by (t6 , t7 , t15 ). Then A is a Cohen-Macaulay local domain of dimension one. In par-
ticular, A = k[[x, y, z]]/I where I = (y 3 − xz, x5 − z 2 ) is generated by a regular
sequence. Now, grm (A)  k[x, y, z]/I ∗ where I ∗ = (xz, z 2 , y 3 z, y 6 ). Hence:
(1) A is a domain, but grm (A) is not even reduced;
(2) A is a complete intersection, but grm (A) is not a complete intersection;
(3) A is a 1-dimensional Cohen-Macaulay ring, but depth grm (A) = 0.
(4) The Hilbert series of A is
1 + 2z + z 2 + z 3 + z 5
HSA (z) =
1−z
and A is Cohen-Macaulay, but its Hilbert function is not admissible for a
Cohen-Macaulay graded algebra.

Another interesting example (dim A = 2) was given by J. Herzog, M.E. Rossi and
G. Valla in [22].
Example 1.2. Let A = k[[x, y, w, t]]/I be where I = (x3 − y 7 , x2 y − xt3 − w6 ).
Then A is a 2-dimensional complete intersection, I ∗ = (x3 , x2 y, x2 t3 , xt6 , x2 w6 , xy 9 −
xw6 t3 , xy 8 t3 , y 7 t9 ), dim grm (A) = 2, but depth grm (A) = 0.

Notice that in the graded case the h-polynomial of a complete intersection is


always symmetric, this is no longer true in the local case.
S. Kleiman proved that there is a finite number of admissible Hilbert func-
tions for graded domains with fixed multiplicity and dimension. The analogous of
Kleiman’s result does not hold in the local case. The following example shows that
the class of local domains of dimension two and multiplicity 4 does not have a finite
number of Hilbert functions.
Example 1.3. Let r > 1 and
Ar := k[[x, y, w, t]]/℘r
where
℘r = (wr tr − xy, x3 − w2r y, y 3 − t2r x, x2 tr − y 2 wr )
is a prime ideal in k[[x, y, w, t]]. The associated graded ring of Ar is the standard
graded algebra
grm (Ar ) = k[x, y, w, t]/(xy, x3 , y 3 , x2 tr − y 2 wr ),
and
1 + 2z + 2z 2 − z r+2
HSAr (z) = .
(1 − z)2
Hence, for all r, we have e0 = 4, d = 2, but the Hilbert function depends on r.
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 177
5

Nevertheless, V. Srinivas and V. Trivedi in [53] and [54] proved that the number
of Hilbert Functions of Cohen-Macaulay local rings with given multiplicity and
dimension is finite (a different proof was given by M.E. Rossi, N.V. Trung and G.
Valla in [42]). This is a very interesting result and it produces upper bounds on
the Hilbert coefficients. If (A, m) is a Cohen-Macaulay local ring of dimension d
and multiplicity e0 , then
3(i!)−i
|ei | ≤ e0 − 1 for all i ≥ 1
(see [53] Theorem 1, [42], Corollary 4.2). The result had been extended by several
authors without assuming the Cohen-Macaulyness of A and in terms of homological
degrees.
The problem of characterizing the admissible numerical functions for Cohen-
Macaulay local rings is largely open. M. E. Rossi, G. Valla and W. Vasconcelos
in [48] proved that, if (A, m) is a Cohen-Macaulay local ring of dimension d and
multiplicity e0 , then
1 + (e0 − 1)z
HSA (z) ≤
(1 − z)d
and the equality has been characterized. From this result we achieve other bounds,
unfortunately very far from being sharp.
After the pionering work of D. G. Northcott in the 50’s, several papers were
written in order to better understand the Hilbert function of a Cohen-Macaulay
local ring in relation with its Hilbert coefficients. Here we shall present some of
them.

2. Superficial elements

A fundamental tool in local algebra is the notion of superficial element. This notion
goes back to the work of P. Samuel ([60] p.296).
Definition 2.1. An element a ∈ m, is said to be superficial for m if there exists
a non-negative integer c such that
(mn+1 : a) ∩ mc = mn
for all n ≥ c.
For every a ∈ m and n ≥ c, mn is contained in (mn+1 : a) ∩ mc . It is the other
inclusion that makes superficial elements special. It is clear that, if A is Artinian,
then every element is superficial, hence, in this section we will assume dim A =
d > 0. From the very definition, we deduce that, if a is a superficial element, then
a ∈ m \ m2 . We give now equivalent conditions for an element to be superficial. The
following development of the theory of superficial elements basically follows Kirby’s
work in [31].
Consider N := Hm 0
(A) = {x ∈ A | ∃ n, mn x = 0A } the 0-th local cohomology of
A with support in m. Let G = grm (A) and denote by H the 0-th local cohomology
HG0
+
(G) of G with support in G+ = ⊕i>0 mi /mi+1 . Further Ass(G) will denote the
set of the associated primes of G.
Theorem 2.2. Let a ∈ m \ m2 , the following conditions are equivalent:
1. a is superficial for m.
178
6 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

m
2. a∗ ∈
/ i=1 Pi with Pi ∈ Ass(G), Pi = G+ .
3. H :G a∗ = H.
4. N : a = N and mj+1 ∩ (a) = amj for all large j.
5. (0 :G a∗ )j = 0 for all large j.
6. mj+1 : a = mj + (0 :A a) and mj ∩ (0 :A a) = 0 for all large j.
We notice that, if the residue field A/m is infinite, the existence of superficial
elements is given by Condition 2. Hence from now on it will be useful to assume
that the residue field is infinite. With regard to this, we read the following standard
fact which says that our assumption is not restrictive.
Proposition 2.3. Let (A, m) be a local ring, B = A[x] where x is an indeter-
minate and ℘ = mB. Then, for every integer n ≥ 0,
HFA (n) = HFB℘ (n).
Since A and B℘ have the same Hilbert function, they have same dimension,
multiplicity and embedding dimension. Moreover, since B℘ is a flat extension of A,
if A is Cohen-Macaulay, then B℘ is Cohen-Macaulay with a larger residue field.
If depth A > 0, then every superficial element for m is also A-regular and a is
superficial for m if and only if mj+1 : a = mj for all large j. Moreover Proposition
2.2 5. says that a is superficial if and only if a∗ is an homogeneous filter-regular
element in G. We refer to [55] for the definition and the properties of homogeneous
filter-regular elements. As the geometric meaning of superficial elements we refer
to the papers by Bondil and Le (see [4], [5]).

We collect in the following theorem important results on superficial elements.


They will be very useful in order to control numerical invariants of A under generic
hyperplane sections.
Theorem 2.4. Let a be a superficial element for m and let d > 0 be the dimen-
sion of A. We have
1. dim(A/aA) = d − 1
2. a is A-regular ⇐⇒ depth A > 0
3. j ≥ 1, depth A/aA ≥ j ⇐⇒ depth A ≥ j + 1
4. a∗ is grm (A)-regular ⇐⇒ depth grm (A) > 0
5. Sally’s machine: j ≥ 1, depth grm/(a) (A/(a)) ≥ j ⇐⇒ depth grm (A) ≥
j+1
6. ej (A) = ej (A/(a)) for every j = 0, . . . , d − 2
7. ed−1 (A/(a)) = ed−1 (A) + (−1)d−1 λ(0 : a)
HSA/(a) (z)
8. a∗ is a grm (A)-regular element ⇐⇒ HSA (z) = 1−z ⇐⇒ a is
A-regular and ed (A) = ed (A/(a)).

We remark that if depth A > 0, then a is superficial for m if and only if


ej (A) = ej (A/(a)) for every j = 0, . . . , d − 1.
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 179
7

We recall that Property 5. is also named Sally’s machine because it was proved
by J. Sally in dimension one and stated in the same paper for higher dimension.
It is a very useful trick in inductive arguments and it was proved by S. Huckaba
and T. Marley in [24]. Property 3. is well known when a is a regular element,
but it seems ignored in literature for superficial elements, a proof can be found in
[46]. The remaining properties can be easily proved by the definition of superficial
element and by means of the following result due to B. Singh. First, we recall the
definition of the Hilbert-Samuel function
j
HFA1 (j) := λ(A/mj+1 ) = HFA (n).
n=0

In particular HFA1 (n) − HFA1 (n − 1) = HFA (n). We will denote by PA (X) the
corresponding polynomial function which is called the Hilbert-Samuel polynomial.
Lemma 2.5. Let a ∈ m. Then, for every j ≥ 0,
1
HFA (j) = HFA/aA (j) − λ(mj+1 : a/mj ).
In the following we let h be the embedding codimension of A (embcodim(A)),
namely
h := HFA (1) − d.
Remark that if a is a superficial element, then embcodim(A) = embcodim(A/(a)).
An usual trick in reducing problems to positive depth is the following. Let
Asat := A/Hm 0 0
(A) where Hm (A) denotes the 0-th local cohomology of A. We remark
that e0 (A ) = e0 (A). If a ∈ m is a superficial element for m, then a ∈ Asat is a
sat
0
superficial element for m/Hm (A).
We relate the Hilbert coefficients of A with those of Asat :
• ei (A) = ei (Asat ) 0 ≤ i ≤ d − 1;
• ed (A) = ed (Asat ) + (−1)d λ(Hm
0
(A))

0
If A is 1-dimensional, S. Goto and K. Nishida gave a nice interpretation of λ(Hm (A))
in terms of the first Hilbert coefficient of an ideal generated by a parameter of A,
see [20] Lemma 2.4. Concerning the problem in higher dimension, very interesting
results by L. Ghezzi, S. Goto, J. Hong, K. Ozeki, T. Phuong and W. Vasconcelos
were recently proved, see [19]. On this topic we point out also [32] by M. Mandal
and J. Verma.

A sequence of elements a1 , . . . , ar (r ≤ d) is a superficial sequence for m if for every


j = 1, . . . , r the element aj is superficial for m/(a1 , . . . , aj−1 ).
Let a1 , . . . , ad be a (maximal) superficial sequence for m and let J = (a1 , . . . , ad ),
then
(2.1) mn+1 = Jmn for n  0.
The above equality says that J is a minimal reduction of m. If the residue field is
infinite, there is a 1-1 correspondence between maximal superficial sequences and
minimal reductions (see for example [26]). Every minimal reduction J of m can be
generated by a maximal superficial sequence and, conversely, the ideal generated
by a maximal superficial sequence is a minimal reduction of m.
180
8 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

The above results can be extended from the m-adic filtrations to the q-good
(or stable) filtrations, where q is an m-primary ideal. In particular to the classical
q-adic filtration (see [47]).

3. 1-dimensional local Cohen-Macaulay rings


Assume (A, m) to be a Cohen-Macaulay local ring of dimension one and embed-
ding codimension h. J. Elias (see [12]) characterized the admissible Hilbert-Samuel
polynomials
PA (X) = e0 (X + 1) − e1
   
for any pair (e0 , e1 ) such that e0 − 1 ≤ e1 ≤ e20 − h2 . Nevertheless the problem of
determining the admissible Hilbert functions of one-dimensional Cohen-Macaulay
local rings is still far from being solved. The question has a clear geometric meaning
related to singularities of affine curves which are arithmetically Cohen-Macaulay.
In the following theorem we collects some classical results due to J. Herzog and R.
Waldi, D.G. Northcott and D. Kirby giving easier proofs.
Theorem 3.1. Let (A, m) be a one-dimensional Cohen-Macaulay local ring.
For every j ≥ 0 we set vj = λ(mj+1 /a mj ) where a is a superficial element for m.
Then
(1) HFA (j) = e0 − vj (≤ e0 ).
(2) If HFA (n) = e0 for some integer n, then HFA (j) = e0 for every j ≥ n.
(3) e1 = j≥0 vj .
(4) HFA (j) ≥ min{e0 , j + 1}.
(5) If HFA (n) = n + 1 for some integer n > 0, then HFA (j) = min{j + 1, e0 }
for every j ≥ n.
 
(6) e0 − 1 ≤ e1 ≤ e20 .
Proof. Recall that e0 = λ(A/aA). From the diagram
A ⊃ mj+1
∪ ∪
aA ⊃ amj
we easily get
(3.1) HFA (j) = e0 − vj
where vj = λ(mj+1 /amj ). It is clear that HFA (j) ≤ e0 and if HFA (n) = e0 for
some integer n, then HFA (j) = e0 for every j ≥ n. since vj = 0 for every j ≥ n.
n
Moreover for n  0, vn = 0. As a consequence, HFA1 (n) = j=0 HFA (j) =
 
e0 (n + 1) − j≥0 vj = e0 (n + 1) − e1 . Hence, e1 = j≥0 vj and (1) (2) and (3) are
proved.
Now if there exists j < e0 such that HFA (j) ≤ j, then by Macaulay’s Theorem
(see e.g. Theorem 1.3 [56]), we should have HFA (n) ≤ j < e0 for every n, which
is a contradiction because HFA (n) = e0 for large n. Hence (4) and (5) follow. We
remark that by (3), e1 ≥ v0 = λ(m/(a)) = e0 − 1 and hence
  0 −1
e  
e0
e0 − 1 ≤ e1 = vj = (e0 − HFA (j)) ≤ (e0 − (j + 1)) = .
j=0
2
j≥0 j≥0


HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 181
9

We remark that Theorem 3.1 (5) was also proved by J. Herzog and V. Waldi
(see [23]) by using different methods.

The following statements follow by Theorem 3.1 and they underline the general
philosophy that extremal values of the Hilbert coefficients (with respect to some
bound) answer to the whole Hilbert function. The proof is an easy exercise.
Exercise 3.2. Let (A, m) be a Cohen-Macaulay local ring of dimension one
and multiplicity e0 . The following facts hold:
  e0 −1 i
z
(1) e1 = e20 ⇐⇒ HSA (z) = i=0 1−z
 e0  e0 −1 i
1+2z+ i=2 z
(2) e1 = 2 − 1 ⇐⇒ HSA (z) = 1−z

It is clear that, if we introduce more invariants of A, we can get more precise bounds.
For example if we involve the embedding codimension of A, h = HFA (1) − 1, then
we have v1 = λ(m2 /am) = λ(m/(a)) + λ((a)/am) − λ(m/m2 ) = e0 − 1 − h. Hence
(3.2) e1 ≥ v0 + v1 = 2e0 − (h + 2).
By using sophisticated devices introduced by Matlis in [33], J. Elias proved
that
   
e0 h
(3.3) e1 ≤ −
2 2
and, for every triple (e0 , e1 , h) satisfying the inequalities (3.2) and (3.3), one can
produce an affine curve with such invariants (see [12]).
In a more general setting, a different proof of (3.3) was given by G. Valla and the
author, by using the Ratliff-Rush filtration (see [45]). In both proofs the techniques
are quite complicate and we do not present here them. Nevertheless, the inequality
(3.3) is an easy exercise assuming an extra (unnecessary) condition.
Exercise 3.3. Let (A, m) be a one-dimensional Cohen-Macaulay local ring,
with embedding dimension h and multiplicity e0 . Assume that the Hilbert function
  
of A is not decreasing, i.e. HFA (n+1) ≥ HFA (n) for every n. Then e1 ≤ e20 − h2 .

If HFA (n + 1) ≥ HFA (n) for every n, we say that the Hilbert function is
non-decreasing. It is a natural question to ask whether the Hilbert function of a
one-dimensional Cohen-Macaulay ring is not decreasing. Clearly, this is the case if
grm (A) is Cohen-Macaulay, but this is not a necessary requirement (see Example
1.1).
Unfortunately, it can happen that HFA (2) = μ(m2 ) < HFA (1) = μ(m). The
first example was given by J. Herzog and R. Waldi in 1975. If we consider the
semigroup ring
A = k[[t30 , t35 , t42 , t47 , t148 , t153 , t157 , t169 , t181 , t193 ]]
of multiplicity e0 = 30 and embedding dimension HFA (1) = 10, we have HFA (2) =
9 < HFA (1). In 1980 F. Orecchia proved that, for all embedding dimension b ≥ 5,
there exists a reduced one-dimensional local ring of embedding dimension b and
decreasing Hilbert function. L. Roberts in 1982 built ordinary singularities with
locally decreasing Hilbert function and embedding dimension at least 7.
182
10 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

We recall that Gupta and Roberts proved that there exists a one-dimensional
Cohen-Macaulay local ring of embedding dimension 4 and multiplicity 32 with
locally decreasing Hilbert function.
It is interesting to ask what happens if the embedding dimension is < 4. It is
clear that if HFA (1) = 1, then A is regular. If HFA (1) = 2, then by Theorem 3.1
(5) we conclude that HFA (n) = min{n + 1, e0 }, hence it is non-decreasing. The
only unknown case seems to be HFA (1) = 3.

In 1978 J. Sally stated the following conjecture.


Conjecture: If A is a one-dimensional Cohen-Macaulay local ring with embedding
dimension three, then HFA is non-decreasing.
J. Elias gave a positive answer to the problem stated by J. Sally for the equichar-
acteristic rings of embedding dimension three (see [13]). Following the geometrical
idea by J. Elias, we present here an elementary, but tricky proof.
We assume A = R/I where R = k[[x1 , . . . , xn ]] with maximal ideal n =
(x1 , . . . , xn ). We may assume k infinite and I ⊆ n2 . Denote by t(I) the initial
degree of I, namely
 
n+j−1
t(I) = min{j : HFA (j) = }.
j
It is well known that grm (A)  grn (R)/I ∗ = k[X1 , . . . , Xn ]/I ∗ where I ∗ is the ideal
generated by the initial forms of the elements of I.
Let P = (a1 , . . . , an ) = (0, . . . , 0) in kn and, if ai = 0, denote by L = L(P ) the ideal
of R generated by the elements ai xj − aj xi for j = 1, . . . , n. Remark that L∗ is the
ideal of k[X1 , . . . , Xn ] generated by the elements ai Xj − aj Xi . In particular L∗ is
the defining ideal of the point P = (a1 , . . . , an ) ∈ Pn−1 k , we will write L∗ = L∗ (P ).
With the above notation we can prove the following result.
Lemma 3.4. Let A = R/I be a one-dimensional Cohen-Macaulay local ring.
There exists P = (a1 , . . . , an ) ∈ kn such that if we let J = I ∩ L(P ) then
HFR/J (j) = HFA (j) + 1 for every j ≥ t(I).

Proof. Denote by I<t(I)> the ideal generated by the homogeneous part of

degree t(I) of I . First we prove that there exists P = (a1 , . . . , an ) = (0, . . . , 0) ∈ kn

 
such that I<t(I)> ⊆ L∗ (P ). Let X = {P1 , . . . , Ps } be a set of s ≥ n+t(I)−1
t(I)
points of
Pn−1
k with maximal Hilbert function (it is well known that for every integers n, t(I)
such X exists). Since the ideal I(X) = ∩si=1 L∗ (Pi ) has initial degree t(I) + 1, there

exists P ∈ X such that I<t(I)> ⊆ L∗ (P ). We may assume P = (a1 , . . . , an ) with
a1 = 0. Hence L = L(P ) = (a2 x1 − a1 x2 , . . . , an x1 − a1 xn ) and n = (x1 ) + L.

From I<t(I)> ⊆ L∗ (P ) we deduce that there exists a generator f of I such that
t(I)
f = x1 + α with α ∈ L ∩ nt(I) . It follows
t(I)
(3.4) I + L = (x1 ) + L.
t(I) t(I)
In fact I ⊆ nt(I) ⊆ (x1 ) + L. On the other hand we remark that x1 = f − α ∈
I + L. Hence, by (3.4), dimk R/(I + L) = t(I). Now we claim that, for every
j ≥ t(I) − 1, we have
(3.5) nj+1 + J = (nj+1 + L) ∩ (nj+1 + I).
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 183
11

By a repeated use of the “modular law”:


(nj+1 + L) ∩ (nj+1 + I) = nj+1 + L ∩ (nj+1 + I) = nj+1 + L ∩ ((xj+1
1 ) + I)

and the conclusion still follows by the modular law since, by (3.4), xj+1
1 ∈ I+L∩nj+1
for every j ≥ t(I) − 1. Using (3.5), it is well defined the following exact sequence of
R-modules (of finite length):
0 → R/(nj+1 + J) → R/(nj+1 + I) ⊕ R/(nj+1 + L) → R/(nj+1 + I + L) → 0.
Since the length is an additive function:
1
HFR/J 1
(j) = HFR/I 1
(j) + HFR/L (j) − t(I) = HFR/I
1
(j) + j − t(I).
1
It follows HFR/J (j) = HFR/J (j) − HFR/J
1
(j − 1) = HFR/I (j) + 1 for every j − 1 ≥
t(I) − 1, hence j ≥ t(I).


The above Lemma has a clear geometric meaning. It describes the behavior of
the Hilbert function by adding to the curve corresponding to Spec(R/I), a straight
line Spec(R/L).
We remark that R/I ∩ L is still one-dimensional and Cohen-Macaulay (primes
avoidance) of the same embedding dimension of R/I.
Theorem 3.5. Let A = R/I be a one-dimensional Cohen-Macaulay local ring
of embedding dimension three. Then the Hilbert function of A is non-decreasing.
Proof. Assume that there exists n0 such that HFA (n0 ) > HFA (n0 + 1). By
a repeated use of Lemma 3.4, we may assume that n0 = t(I) − 1. It follows
   
t(I) + 2 t(I) + 2
μ(I) ≥ − HFA (t(I)) ≥ − HFA (t(I) − 1) + 1 =
2 2
   
t(I) + 2 t(I) + 1
= − + 1 = t(I) + 2.
2 2
On the other hand, from the Hilbert-Burch Theorem, we deduce that μ(I) ≤ t(I) +
1, a contradiction. 

We characterize other interesting classes of one-dimensional Cohen-Macaulay local


rings with non-decreasing Hilbert function. By Theorem 3.1 (1), we deduce that
for every superficial element a ∈ m we have:
(3.6) e0 = μ(m) + λ(m2 /am) = h + 1 + λ(m2 /am)
where h is the embedding codimension, hence
e0 ≥ h + 1 = μ(m) = HFA (1)
which is the one-dimensional version of the well-known Abhyankar’s inequality (see
[1]). We prove now that the Hilbert function of a one-dimensional Cohen-Macaulay
local ring of small multiplicity is non-decreasing.
Theorem 3.6. Let A = R/I be a one-dimensional Cohen-Macaulay local ring
of multiplicity e0 ≤ h + 3. Then the Hilbert function of A is non-decreasing.
184
12 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

Proof. Let a be a superficial element for m and consider the corresponding


Artinian reduction B = A/aA. If we assume HFB (3) = 0, equivalently m3 ⊆ am,
then λ(m2 /am) = λ(m2 /am + m3 ) ≤ 2 by (3.6). Consider the standard graded
k-algebra
T = ⊕i≥0 mi /(a mi−1 + mi+1 ).
Because HFT (2) = λ(m2 /am) ≤ 2, by Macaulay’s Theorem we get HFT (i) ≤ 2 for
every i ≥ 2. But if HFT (i0 ) ≤ 1 for some i0 ≥ 2, then HFT (i) ≤ 1 for every i ≥ i0 .
Hence HFT is not locally increasing and we conclude by Theorem 3.1 (1), indeed
for i ≥ 2 HFA (i) = e0 − λ(mi+1 /ami ) = e0 − HFT (i).
Assume now HFB (3) = 0, since e0 = i≥0 HFB (i) = h + 3, necessarily HFB (2) =
HFB (3) = 1. Hence λ(m2 /am) = 2 by (3.6) and λ(m2 /am+m3 ) = HFB (2) = 1. We
deduce that m2 /am is a cyclic module (it is not a k-vector space) and there exists
w ∈ m \ m2 such that m2 = am + (w2 ). In fact if w2 ∈ am for every w ∈ m \ m2 ,
then it is easy to prove that m3 ⊆ am. Therefore for every j ≥ 1
mj+1 = amj + (wj+1 ).
·w
Since the map mj+1 /amj → mj+2 /amj+1 is an epimorphism, we conclude because
HFA (j) = e0 − λ(mj+1 /amj ) ≤ HFA (j + 1) = e0 − λ(mj+2 /amj+1 )
for every j.


The above situation cannot be extended to multiplicity e0 = h + 4. S. Molinelli


and G. Tamone produced the following example with multiplicity e0 = h + 4 and
locally decreasing Hilbert function. Consider
A = k[[t13 , t19 , t24 , t44 , t49 , t54 , t55 , t59 , t60 , t66 ]].
Then HFA (2) = 9 < HFA (1) = 10.

Starting from one-dimensional Cohen-Macaulay local rings with locally decreas-


ing Hilbert functions, it is possible to produce examples with several peaks and val-
leys. The technique had been pointed out by J. Elias. Let A = k[[x1 , . . . , xr ]]/I and
B = k[[y1 , . . . , ys ]]/J be one dimensional Cohen-Macaulay local rings and consider
the ideal
K = [I + (y1 , . . . , ys )] ∩ [J + (x1 , . . . , xr )] ⊆ k[[x1 , . . . , xr , y1 , . . . , ys ]] = C
Then
HFC/K (n) = HFA (n) + HFB (n)
for every n ≥ 1. The crucial point is that
K ∗ = [I +(y1 , . . . , ys )]∗ ∩[J +(x1 , . . . , xr )]∗ = [I ∗ +(y1∗ , . . . , ys∗ )]∩[J ∗ +(x∗1 , . . . , x∗r )].

Several computations lead us to think that these pathologies cannot be realized if


the local ring is Gorenstein. We state the following problem:
Problem 3.7. Is the Hilbert function of a Gorenstein local ring of dimension
one not decreasing?
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 185
13

The problem is open even if we consider complete intersections. Very partial results
have been proved. Puthenpurakal gave a positive answer in codimension two case
(see [37]). Some results appeared for Gorenstein monomial curves in A4 (see [2]).

4. Hilbert coefficients and classical bounds


In this section we will extend the interest to higher dimensions. After the pio-
neering work done by Northcott (see [35]), several efforts have been made also to
better understand the Hilbert function of a d-dimensional Cohen-Macaulay local
ring (A, m) starting from the asymptotic information given by the Hilbert coeffi-
cients. J. Sally carried on Northcott’s work (see for example [49], [50]) by proving
interesting results on this topic and by asking challenging questions as well. Sev-
eral improvements along this line have been obtained in the last years (J. Elias,
C. Huneke, S. Itoh, A. Ooishi, C. Polini, T. Puthenpurakal, M. E. Rossi, G. Valla,
W. Vasconcelos, J. Verma, etc.), often extending the framework to Hilbert func-
tions associated to an m-primary ideal or, more in general, to a stable filtration of
A-modules (see [47]).
The first Hilbert coefficient e0 is the multiplicity and, due to its geometric mean-
ing, has been studied very deeply. From the algebraic point of view e0 = λ(A/J)
where J is the ideal generated by a maximal superficial sequence, in particular a
minimal reduction of m. The other coefficients are not as well understood, either
geometrically or in terms of how they are related to algebraic properties of the local
ring.
We present here elementary proofs of two classical bounds due to Abhyankar
and Northcott respectively (see [1] and [35]). As usual, we denote by ei the Hilbert
coefficients and by h the embedding codimension of A, i.e. h = HFA (1) − d =
μ(m)−d. We recall that, since (A, m) is Cohen-Macaulay, every superficial sequence
is also a regular sequence in A.
Theorem 4.1. (Abhyankar’s inequality) Let (A, m) be a d-dimensional Cohen-
Macaulay local ring. Then
e0 ≥ h + 1.
Proof. Since e0 and h do not change modulo a superficial sequence (see The-
orem 2.4 (7)), we may assume d = 0. Then
HSA (z) = 1 + hz + · · · + hs z s
where s ≥ 1 and hi ≥ 0. Thus e0 = 1 + h + · · · + hs ≥ 1 + h. 

The above result shows that e0 = 1 if and only if h = μ(m) − d = 0, hence


A is regular. We remark that being A Cohen-Macaulay is essential. Let A =
k[[x, y]]/(x2 , xy), then A is not Cohen-Macaulay and e0 = 1 < h + 1 = 2. Hence
we deduce that the standard graded k-algebra G = k[x, y]/(x2 , xy) cannot be the
associated graded ring of a Cohen-Macaulay local ring.
Theorem 4.1 extends to the local Cohen-Macaulay rings the well known lower
bound for the degree of a reduced and irreducible non-degenerate variety X in Pn :
deg X ≥ codim X + 1.
186
14 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

The varieties for which the bound is attained are called varieties of minimal degree
and they are completely classified. In particular, they are always arithmetically
Cohen-Macaulay.
In the following result, the first inequality has been proved by J. Elias and G.
Valla in [15], the second is a classical result due to Northcott (see [35]).
Theorem 4.2. Let (A, m) be a d-dimensional Cohen-Macaulay local ring. Then
e1 ≥ 2e0 − h − 2 ≥ e0 − 1.
Proof. By Theorem 2.4, parts 6. and 7. and the fact that the embedding
codimension h does not change modulo superficial elements, we may assume d = 1.
Then the result follows by (3.2). The second inequality is a consequence of Theorem
4.1. 
Theorem 4.3. Let (A, m) be a d-dimensional Cohen-Macaulay local ring. The
following facts are equivalent:
(1) e0 = h + 1
(2) mn+1 = Jmn for every n ≥ 1 and for every minimal reduction J of m.
1+hz
(3) HSA (z) = (1−z)d

(4) e1 = e0 − 1
Proof. Let J be a minimal reduction of m (or equivalently the ideal generated
by a maximal superficial sequence). From the diagram
m ⊃ m2
∪ ∪
J ⊃ Jm
we get λ(m/J) = e0 − 1 = HFA (1) − d + λ(m2 /Jm). Hence, if e0 = h + 1, then m2 =
Jm and therefore (1) implies (2). If we assume (2), by the well-known Valabrega-
Valla criterion, we have that grm (A) is Cohen-Macaulay. Hence by Theorem 2.4,
part 8., we may reduce the computation of HSA to that of the Artinian reduction
B = A/J. Since the embedding codimension does not change and (m/J)2 = 0 we
get (3). The implication (3) =⇒ (4) is clear. If we assume (4), by Theorem 4.2,
we get e1 = 2e0 − h − 2 = e0 − 1, therefore e0 = h + 1.


Hence the next step will be the study of the Hilbert functions of Cohen-Macaulay
local rings of almost minimal multiplicity, that is e0 = h + 2. The problem is con-
siderably more difficult. Also from the geometric point of view, varieties satisfying
deg X = codim X + 2 are not necessarily arithmetically Cohen-Macaulay.
We propose in the following exercise the easy part of the general result.
Exercise 4.4. Let (A, m) be a d-dimensional Cohen-Macaulay local ring. The
following facts are equivalent:
(1) e0 = h + 2 and grm (A) is Cohen-Macaulay
2
(2) HSA (z) = 1+hz+z
(1−z)d
(3) e1 = e0
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 187
15

In the above exercise the Cohen-Macaulyness of grm (A) is essential. In fact if we


consider A = k[[t4 , t5 , t11 ]], then e0 = 4 = h + 2, but grm (A) is not Cohen-Macaulay
3
and HSA (z) = 1+hz+z
(1−z) .

We come now to a result proved by G. Valla and the author [45], and indepen-
dently by Wang [59], which gives a positive answer to a longstanding conjecture
stated by J. Sally in [50].
Theorem 4.5. Let (A, m) be a d-dimensional Cohen-Macaulay local ring. The
following facts are equivalent:
(1) e0 = h + 2
1+hz+z s
(2) HSA (z) = (1−z)d
for some integer 2 ≤ s ≤ h + 1.
Further, if either of the above conditions holds, then depthgrm (A) ≥ d − 1.
Sally proved the above result in dimension one and, for every integer s, she gave
an example:
A2 = k[[te , te+1 , te+3 , . . . , t2e−1 ]]
e e+1 e+s+1 e+s+2
As = k[[t , t , t ,t , . . . , t2e−1 , t2e+3 , t2e+4 , . . . , t2e+s ]], 3 ≤ s ≤ e − 2
Ae−1 = k[[t2 , te+1 , t2e+3 , t2e+4 , . . . , t3e−1 ]].
Sally’s conjecture can be proved by reducing the problem to the two-dimensional
case. In spite of the fact that in the one-dimensional case the proof of Theorem 4.5
is very easy, in dimension two the result is complicated and a crucial tool in the
proof given in [45] involves the use of the Ratliff-Rush filtration (see [38]) which
we recall below.
Let (A, m) be a Cohen-Macaulay local ring. For every n ≥ 0 we consider the chain
of ideals

mn ⊆ mn+1 : m ⊆ mn+2 : m2 ⊆ · · · ⊆ mn+k : mk ⊆ · · ·


This chain stabilizes at an ideal which we will denote by


n :=
m (mn+k : mk ).
k≥1

Hence there exists a positive integer t, depending on n, such that m


n = mn+k : mk

for every k ≥ t. It is clear that m = A and, for every non negative integers i and j,
0


i , m
mi ⊆ m
j ⊆ m

i m  
i+j , m
i .
i+1 ⊆ m

Furthermore, if a is superficial, for every i ≥ 0 we have


m
i .
i+1 : a = m


i = mi for large integers i. If J is an
It is easy to prove that if depth A > 0, then m
ideal generated by a maximal superficial sequence, we may define, for every i ≥ 0,

σi := λ(m 
i+1 /J mi+1 ).

For example σ0 = e0 − 1. With the above notation one can prove the following
equalities, proved by S. Huckaba and T. Marley (see [24]) in the case dim A ≤ 2.
 
(4.1) e1 = σi and e2 = iσi
i≥0 i≥1
188
16 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI


i /m
:= ⊕i≥0 m
The main tool is to consider the graded k-algebra G i+1 . The Hilbert

polynomial of G is the Hilbert polynomial of A and the advantage is that G has
positive depth even if grm (A) has depth zero.
The crucial point in the proof of Theorem 4.5 is to show that, for every Cohen-
Macaulay local ring of dimension at most two, the reduction number r(m) of m is
bounded by the following linear function of e1 and e0
r(m) ≤ e1 − e0 + 2,
equivalently we have
me1 −e0 +3 = Jme1 −e0 +2
for every minimal reduction J of m (see [39]).

We do not know if the above equality can be extended to higher dimensions.


More in general, it is natural to ask the following problem
Problem 4.6. Let (A, m) be a Cohen-Macaulay local ring of dimension d. Does
there exist a linear function f (e0 , e1 , . . . , ed−1 , d) such that the reduction number of
m is bounded by f (e0 , e1 , . . . , ed−1 , d)?

Using (4.1), we investigate the successive Hilbert coefficients and we present an


easy proof of a lower bound on e2 proved by J. Sally in [52].
Proposition 4.7. Let (A, m) be a Cohen-Macaulay local ring of dimension
d ≥ 2. Then
e2 ≥ e1 − e0 + 1 ≥ 0.
Proof. By Theorem 2.4 we prove the result in the case d = 2. Then by (4.1)
we have   
e2 = iσi = σi − σ0 + (i − 1)σi ≥ e1 − e0 + 1.
i≥1 i≥0 i≥1
By Theorem 4.2 it follows that e2 ≥ 0. 

Clearly, if e2 = 0, then e2 = e1 − e0 + 1 and e1 = e0 − 1. Hence, by Theorem 4.3,


the Hilbert function of A is known and grm (A) is Cohen-Macaulay. The following
example (due to H. J. Wang) shows that the only equality e2 = e1 − e0 + 1 does
not force grm (A) to be Cohen-Macaulay.
Example 4.8. Consider the two-dimensional Cohen-Macaulay local ring
A = k[[x, y, t, u, v]]/(t2 , tu, tv, uv, yt − u3 , xt − v 3 )
with maximal ideal m. We have
1 + 3z + 3z 3 − z 4
HSA (z) = .
(1 − z)2
Hence one has e0 = h + 3 = 6, e1 = 8, e2 = 3 = e1 − e0 + 1. The associated graded
ring gr m (A) has depth zero. Notice that e3 < 0.

It seems that the normality of the ideal m (i.e. mn is integrally closed for every
n) yields non trivial consequences on the Hilbert coefficients of A and, ultimately,
on depth grm (A) and the Hilbert function.
In [11] A. Corso, C. Polini and M.E. Rossi proved the following result.
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 189
17

Theorem 4.9. Let (A, m) be a Cohen-Macaulay local ring of dimension d.


Assume m is a normal ideal and e2 = e1 − e0 + 1, then
1 + hz + (e0 − h − 1)z 2
HSA (z) = and m3 = Jm2
(1 − z)d
for every minimal reduction J of m. In particular grm (A) is Cohen-Macaulay.
This result gives a positive answer to a question raised by G. Valla in [56]. The
key point is a theorem proved by S. Itoh on the normalized Hilbert coefficients of
ideals generated by a system of parameters (see [29], [30]).

Unfortunately, the positivity of the Hilbert coefficients stops with e2 . Indeed,


M. Narita showed that e3 can be negative (see also Example 4.8). However, a
remarkable result of S. Itoh says that e3 ≥ 0 provided m is normal (see [29]). If
equality holds, then grm (A) is Cohen-Macaulay. A nice proof of this result was
also given by S. Huckaba and C. Huneke in [25] where the ideal is assumed to be
asymptotically normal. A different proof has been also produced by A. Corso, C.
Polini and M.E. Rossi by using techniques developed in this survey (see Theorem
4.1. in [11]).
As far as we know there are not negative answers to the following natural question.
Problem 4.10. If m is normal, then is ei ≥ 0 for every i ?

The previous problem is related to the asymptotic behavior of the associated


graded ring of the powers of the maximal ideal and it has some relation with the
Grauert-Riemenschneider Vanishing Theorem. For interesting questions related to
the Hilbert coefficients of normal ideals we refer to [29] and [30].
Finally, we remark that if e0 = h + 3, then Example 4.8 shows that it is no
longer true that depth gr m (A) ≥ d − 1. J. Sally proved that if A is Gorenstein and
e0 = h + 3, then grm (A) is Cohen-Macaulay (see [51]). Later M.E. Rossi and G.
Valla (see [44]) proved the following result
Theorem 4.11. Let (A, m) be a Cohen-Macaulay local ring of dimension d and
Cohen-Macaulay type τ. If e0 = h + 3 and τ < h, then

1 + hz + z 2 + z s
HSA (z) =
(1 − z)d
where 2 ≤ s ≤ τ + 2. In particular depth grm (A) ≥ d − 1.
We state the following problem.

Problem 4.12. Let (A, m) be a Cohen-Macaulay local ring of dimension d and


multiplicity e0 = h + 3. Then is depth grm (A) ≥ d − 2?

Notice that if e0 = h + 3, then τ ≤ h + 1, hence it remains to explore the cases


τ = h, h + 1.

Interesting problems on the Hilbert functions also come from the Artinian case,
under the assumption that A is Gorenstein or, more in general, level. In the next
section we present an interesting approach by using Macaulay’s Inverse System.
190
18 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

5. Hilbert Functions of Artinian Gorenstein rings

Let (A, m, k) be an Artinian local ring. The socle-degree s of A is the largest


 0. The local ring A is s-level of type τ if
integer for which ms =
0 : m = ms and dimk ms = τ
A is Gorenstein if A has type 1.

Since HFA (n) = 0 for n > s, we will write the Hilbert function of the Artinian
local ring A as a finite sequence of integers HFA = {HFA (0), . . . , HFA (s)}.
Very little is known about the Hilbert function of A, if we assume that A
is Gorenstein or level. We recall that, in the graded case, the Hilbert function
of an Artinian Gorenstein standard k-algebra is symmetric, but this is no longer
true if we assume that A is local. For instance, consider the complete intersection
I = (xy, x2 − y 3 ) ∈ R = k[[x, y]], then A = R/I has Hilbert function {1, 2, 1, 1}
which is not symmetric.
It is interesting to recall that, however we fix an integer e ≥ 4, there exists
an ideal I = (f, g) which is a complete intersection of two elements of valuation
two in R with multiplicity e0 (R/I) = e (for example see [3]). In the corresponding
homogeneous setting the multiplicity would be 4!
In the following we will consider numerical functions H = {1, h1 , h2 , . . . , hs }, hi
positive integers, which verify Macaulay’s theorem and our aim is to characterize
those which are admissible for an Artinian Gorenstein local ring. The following
nice result has been proved by several authors, among others by A. Iarrobino, S.
Kotari, S. Goto et al.
Theorem 5.1. A numerical function H = {1, 2, h2 , . . . , hs }, hi positive inte-
gers, is admissible for an Artinian Gorenstein local ring A = k[[x, y]]/I if and only
if
|hi+1 − hi | ≤ 1
for every i = 1, . . . , s (consider h1 = 2 and hi = 0 if i > s).
An interesting generalization to level algebras was presented by V. Bertella (see
[3]). She proved that a numerical function H = {1, 2, h2 , . . . , hs } is admissible for
a s-level Artinian local ring of type τ if and only if
|hi+1 − hi | ≤ τ.

A very short proof of the above results was recently given by M.E. Rossi and L.
Sharifan as a consequence of a result on the minimal free resolutions of local rings
of given Hilbert Function (see [41]).
A characterization of the Hilbert functions of Artinian Gorenstein local rings
of codimension three (H = {1, 3, h2 , . . . , hs }) is still an open problem, even if we
assume A = k[[x, y, z]]/I where I is a complete intersection.
Another motivation concerning the study of Artinian local rings comes from
recent papers by J. Elias and G. Valla (see [16], [17] ), by G. Casnati and R.
Notari (see [7], [8]), by D.A. Cartwright, D. Erman, M. Velasco and B. Viray (see
[6]) and by B. Poonen (see [36]), where the authors studied the classification up to
isomorphisms of the Artinian Gorenstein k-algebras of a given Hilbert function.
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 191
19

For example an Artinian Gorenstein local ring A with Hilbert function {1, 2, 2, 1}
allows only two models, namely those corresponding to the ideals I = (x2 , y 3 ) and
I = (xy, x3 − y 3 ) and both are homogneous. But if we move to the next case with
symmetric Hilbert function {1, 2, 2, 2, 1} we will see that one has three different
models, namely two ideals which are homogeneous I = (x2 , y 4 ), I = (xy, x4 − y 4 )
and one which is not homogeneous, the ideal I = (x4 + 2x3 y, y 2 − x3 ). It is in-
teresting to say that there is a finite number of isomorphism classes of Artinian
Gorenstein algebras of multiplicity ≤ 9. The first case with a 1-dimensional family
of models corresponds to e0 = 10 and Hilbert function {1, 2, 2, 2, 1, 1, 1}. This case
has been studied by Elias and Valla in [17].
This classical problem has an important motivation related to the study of the
Hilbert scheme Hilbd (PkN ) parametrizing scheme of dimension 0 and fixed degree
d in PkN . Since it is known that any zero-dimensional Gorenstein scheme of degree
d can be embedded as an arithmetically Gorenstein non-degenerate subscheme in
Pkd−2 , it is natural to study the open locus

HilbaG d−2
d (Pk ) ⊆ Hilbd (Pkd−2 ).

d−2
The scheme HilbaG d (Pk ) has a natural stratification which reduce the problem
to understand the intrinsic structure of Artinian Gorenstein k-algebras of degree
d. Since such an algebra is the direct sum of local, Artinian, Gorenstein k-algebras
of degree at most d, it is natural to begin with the inspection of some elementary
bricks.

The aim of this section is to investigate these problems by means of the In-
verse System. In 1916 Macaulay established a one-to-one correspondence between
Gorenstein Artin algebras and suitable polynomials. This correspondence has been
deeply studied in the homogeneous case, among other authors, by A. Iarrobino in
a long series of papers. For a modern treatment and a list of references, we refer
to a book by Iarrobino and Kanev [28]. Every Artinian Gorenstein graded algebra
A = k[x1 , . . . , xn ]/I of socle degree s, corresponds, up to scalar, to a form of degree
s in a polynomial ring P = k[y1 , . . . , yn ]. In particular this correspondence is bi-
jective and it is compatible with the action of the linear group GLn (k). The study
of the geometric objects arising from this correspondence between forms and ideals
is a classical theme in algebraic geometry and commutative algebra. Notice that
from a categorial point of view, Macaulay’s correspondence is a particular case of
Matlis duality.
J. Emsalem and A. Iarrobino presented the role of Macaulay’s Inverse System
in the study of Artinian local algebras (see [30] and [18]). We follow their approach.

Let R = K[[x1 , . . . xn ]] be the ring of the formal power series with maximal ideal
n = (x1 , . . . , xn ) and let P = K[y1 , . . . , yn ] be a polynomial ring. P has a structure
of R-module by means of the following action

◦: R×P −→ P
(f, g) → f ◦ g = f (∂y1 , . . . , ∂yn )(g)
192
20 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

where ∂yi denotes the partial derivative with respect to yi . This action can be
defined bilinearly from that for monomials. If we denote by xα = xα 1 · · · xn then
1 αn

⎨ (β−α)! y β−α if βi ≥ αi for i = 1, · · · , n
β!

x ◦y =
α β

0 otherwise
β! n βi !
(β−α)! = i=1 (βi −αi )! .

We remark that for every f, h ∈ R and g ∈ P, (f h) ◦ g = f ◦ (h ◦ g). Let now s


be a positive integer and denote by P≤s the set of polynomials of degree ≤ s. Thus
ns+1 ◦ g = 0 if and only if g ∈ P≤s .

Now consider the pairing induced by ◦:


, : R×P −→ k
(f, g) → (f ◦ g)(0)
which will give a canonic bijection between the ideals of R and R-submodules of P.
Let I ⊂ R be an ideal, we define
I ⊥ := {g ∈ P | f, g = 0 ∀f ∈ I}.
Since (nI ◦ g)(0) = 0 if and only if I ◦ g = 0, it follows that
I ⊥ = {g ∈ P | I ◦ g = 0 }.
In particular I ⊥ is an R-submodule of P. In fact if g ∈ I ⊥ , then f ◦ g ∈ I ⊥ for every
f ∈ R. Observe that, if I = ns+1 , then I ⊥ coincides with P≤s . Notice that I ⊥ is
finitely generated if and only if A = R/I has finite length.

Conversely, for every R-submodule M of P, define


AnnR (M ) := {g ∈ R | g, f  = 0 ∀f ∈ M }.
Since M is an R-submodule of P one can prove that
AnnR (M ) = {g ∈ R | g ◦ M = 0 }.
It is easy to see that AnnR (M ) is an ideal of R. If M is cyclic, i.e. M = f R = R◦f
with f ∈ P, then we will write AnnR (f ). We remark that M = f R is a k-vector
space generated by the polynomial f and all its derivatives of every order.
Let I be an ideal of R such that A = R/I has finite length. Let m = n/I be the
maximal ideal of A. The action  ,  induces the following isomorphism of k-vector
spaces (see [18] Proposition 2 (a)):
(5.1) (R/I)∗  I ⊥
where ( )∗ denotes the dual induced by  , . Hence dimk R/I(= e0 (R/I)) =
dimk I ⊥ .

There is a one-to-one correspondence between the ideals I ⊆ R such that R/I


is Artinian and the finitely generated R-submodules M of P. The correspondence
is defined sending I to I ⊥ , conversely M goes to AnnR (M ). For a reference, see
also [18] Proposition 2, Corollary 2.

A more precise result can be formulated for Artinian Gorenstein rings as follows
(see [27], Lemma 1.2.).
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 193
21

Theorem 5.2. A local ring A = R/I is an Artinian Gorenstein local ring


of socle degree s if and only if its dual module I ⊥ is a cyclic R-submodule of P
generated by a polynomial F ∈ P of degree s.

In other words, the above result says that there is the following one-to-one corre-
spondence of sets:
⎧ ⎫ ⎧ ⎫
⎨ I ⊆ R ideal such that ⎬ ⎨ M = R ◦ F R-cyclic submodule of P ⎬
R/I is Artinian Gorenstein ←→ with degree F = s.
⎩ ⎭ ⎩ ⎭
with socledegree(R/I) = s.

I −→ I⊥
AnnR (F ) ←− M =< F >R
If F, G ∈ P, then < F >R =< G >R if and only if G = u ◦ F for some unit u in R.
Now, let AF denote the Artinian Gorenstein algebra associated to F ∈ P, i.e.
AF = R/AnnR (F ).

The isomorphism (5.1) preserves the length and the Hilbert function (see [27]
page 10). As in the graded case, it is possible to compute the Hilbert function via
the Inverse System.

If we let
I ⊥ ∩ P≤i + P<i
(5.2) (I ⊥ )i :=
P<i
then, by (5.1), we can prove
(5.3) HFR/I (i) = dimk (I ⊥ )i .

Let’s consider the following examples.


Example 5.3. Let R = k[[x, y]] and P = k[x, y]. Consider the R-submodule of
P generated by F = y 6 + xy 4 and G = x4 + y 3 . We have
I = AnnR (< F, G >R ) = AnnR (< F >R )∩AnnR (< G >R ) = (x2 y, x5 , y 5 −30xy 3 ).
An easy computation shows that
< F, G >R =< F, G, x4 , y 4 , x3 , xy 2 , y 3 , x2 , xy, y 2 , x, y, 1 >k
as k-vector space. Hence by (5.3) we get
HFR/I (z) = 1 + 2z + 3z 2 + 3z 3 + 2z 4 + z 5 + z 6 .
We remark that A is a Cohen-Macaulay local ring of type 2 and the generators of
its socle are in degrees 4 and 6, respectively the degrees of F and G.
Example 5.4. Let I = (xy, y 2 − x3 ) ⊆ R = k[[x, y]]. It is easy to see that
I ⊥ =< x3 + 3y 2 >R
and < x3 + 3y 2 >R =< x3 + 3y 2 , x2 , x, y, 1 >k as k-vector space. Hence
HFR/I (z) = 1 + 2z + z 2 + z 3
We remark that the Hilbert function of A = R/I is not symmetric even if A is
Gorenstein.
194
22 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

From now on we assume that A is an Artinian Gorenstein local k-algebra where


k is algebraically closed of characteristic zero. In [27] A. Iarrobino studied a special
filtration of G = grm (A) consisting of a descending sequence of ideals:
G = C(0) ⊇ C(1) ⊇ · · · ⊇ C(s − 2)
whose successive quotients
Q(a) = C(a)/C(a + 1),
a = 0, . . . , s−2, are reflexive G-modules (see [27], Theorem 1.5). In particular Q(a)
has symmetric Hilbert function HFQ(a) with offset center s−a 2 (see also Proposition
1.9 [27]). The reflexivity of Q(a) as k-vector space is induced from the nonsingular
pairing on A where the ideals (0 : mi ) of A correspond to (mi )⊥ . In the graded case
0 : mi = ms+1−i , hence the Hilbert function of A is symmetric. When A is not
graded, (0 : mi ) = ms+1−i but the duality still gives some information.
The homogeneous ideals C(a) of G are defined piecewise as
0 : ms+1−a−i ∩ mi
C(a)i =
0 : ms+1−a−i ∩ mi+1

Theorem 5.5. Let (A, m) be an Artinian Gorenstein local ring with socle degree
s ≥ 2. With the above notation, we have:
(1) HFQ(a) (i) = HFQ(a) (s − a − i) for every i ≥ 0
s−2
(2) HFA (i) = a=0 HFQ(a) (i) for every i ≥ 0.
The decomposition of the Hilbert function in symmetric functions HFQ(a) will be
called Q-decomposition.
Notice that Q(0) = G/C(1) is the unique (up to isomorphism) homogeneous quotient
of G which is Gorenstein with the same socle degree s.
It is known that if HFA (n) is symmetric, then G = Q(0) and G is Gorenstein.
Hence
G is Gorenstein ⇐⇒ HFA (n) is symmetric ⇐⇒ G = Q(0).
(see [27], Proposition 1.7 and [18], Proposition 7).

The G-module Q(0) plays a crucial role and it can be computed in terms of the
corresponding polynomial in the inverse system. Let F ∈ P be a polynomial of
degree s and denote by Fs the form of highest degree in F, that is F = Fs + . . .
terms of lower degree, then
Q(0)  R/AnnR (Fs ).
Example 5.6. We consider I = (x4 , x3 − y 2 ) ⊆ R = K[[x, y]]. In this case

I =< y 3 + x3 y >R and
HSR/I (z) = 1 + 2z + 2z 2 + 2z 3 + z 4
which is symmetric, hence grm (A) is Gorenstein. Indeed in this case the associated
graded ring is grm (A) = P/(x4 , y 2 ). Notice that A = R/I is not canonically graded,
that is A  grm (A) as k-algebras.
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 195
23

The following example shows that the Q-decomposition is not uniquely deter-
mined by the Hilbert function.
Example 5.7. Consider the numerical function
H = {1, 3, 3, 2, 1, 1}.
It is admissible for an Artinian Gorenstein algebra. In this case s = 5 and it admits
the following Q-decompositions, a = 0, 1, 2, 3:

H 1 3 3 2 1 1
H 1 3 3 2 1 1
Q(0) 1 1 1 1 1 1
Q(0) 1 1 1 1 1 1
Q(1) 1 2 1
Q(1) 1 1 1
Q(2) 0 0
Q(2) 1 1
Q(3) 1
Both can be realized. Let R = K[[x, y, z]] and consider the following Artinian
Gorenstein rings:
(1) A = R/I where I = AnnR (x5 + x3 z + x2 y 2 + y 4 + z 3 )
(2) B = R/J where J = AnnR (x5 + x2 y 2 + xy 3 + 2y 4 + zx2 + z 2 )
Then H = HFA = HFB and they correspond respectively to the above different
Q-decompositions, in particular we deduce that A and B are not isomorphic.
In codimension two the Hilbert function determines the Q-decomposition. Since
in codimension two being Gorenstein is equivalent being a complete intersection,
Iarrobino in [27] asked the following question.

Problem 5.8. Can there be more than one Q-decomposition for the Hilbert
function of a complete intersection?

In this last part of the survey, we shall use the Macaulay’s correspondence in
problems of classification.

It is reasonable to think what should be convenient to classify cyclic submodules


of P instead of Artinian Gorenstein algebras A. Emsalem studied this problem in
[18], Section C.

Given I and J ideals of R = k[[x1 , . . . , xn ]], there exists a k-algebras isomor-


phism
φ : R/I → R/J
if and only if φ comes from an automorphism of R sending I to J. We recall that
the automorphisms of R act as replacement of xi by zi = φ(xi ), i = 1, · · · , n, such
that n = (x1 , . . . , xn ) = (z1 , . . . , zn ) (see [17]). We encode the isomorphism φ by
the list xi −→ zi , i = 1, · · · , n.
The automorphisms of R induce univocally the automorphisms of R/ns+1 which
are special homomorphisms of k-vector spaces of finite dimension. If we assume
ns+1 ⊂ I, J, we can say that there exists a k-algebras isomorphism
φ : R/I → R/J
if and only if φ comes from an automorphism of R/ns+1 sending I/ns+1 to J/ns+1 .
196
24 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

Passing to the dual as K-vector spaces induced by  , ,


Hom(φ) = φ∗ : (R/J)∗ → (R/I)∗
is an isomorphism of the k-vector spaces where (R/I)∗ = I ⊥ and (R/J)∗ = J ⊥ .
Hence, in terms of the corresponding polynomials F, G ∈ P, we can prove that
(5.4) φ(AF ) = AG if and only if (φ∗ )−1 (< F >R ) =< G >R .

Then φ(AF ) = AG if and only if (φ∗ )−1 (F ) = u ◦ G where u is a unit in R.

We want to use this approach for the classification of the Artinian Gorenstein
local algebras of low socle degree. Assume k is algebraically closed. It is easy
to classify, up to isomorphism, the Artinian Gorenstein local rings of socle degree
two and embedding dimension n. In fact, using the techniques of this section, the
problem can be easily rephrased in terms of the classification of the hypersurfaces of
degree two in Pn−1 . In particular we can prove that A  AF with F = y12 +· · ·+yn2 ∈
P = k[y1 , . . . , yn ].
We have thus the following result.
Proposition 5.9. An Artinian local k–algebra A of embedding dimension n is
Gorenstein with socle degree two if and only if A ∼
= R/I where
I = (xi xj , x2n − x21 )1≤i<j≤n .
Hence if m3 = 0, then A  grm (A). It is very rare that a local ring is isomorphic
to its associated graded ring. Nevertheless we will see that this happens even if
m4 = 0 and the Hilbert function is symmetric.
If A  grm (A), accordingly with Emsalem in [18], we say that A is canonically
graded. The following result is surprising.

Theorem 5.10. ([14] Theorem 3.3.) Let A = R/I be an Artinian Gorenstein


local k-algebra with Hilbert function {1, n, n, 1}. Then A is canonically graded.

Corollary 5.11. The classification of Artinian Gorenstein local k-algebras


with Hilbert function HFA = {1, n, n, 1} is equivalent to the projective classification
of the hypersurfaces V (F ) ⊂ Pn−1
K where F is a degree three non degenerate form
in n variables.

If n = 1 then it is clear that A ∼


= K[[x]]/(x4 ), so there is a only one analytic
model. By using Theorem 5.10, here we present the classification in the case n = 2.

Proposition 5.12. Let A be an Artinian Gorenstein local k-algebra with Hilbert


function HFA = {1, 2, 2, 1}. Then A is isomorphic to one and only one of the fol-
lowing algebras
Model A = R/I Inverse system F Geometry of C = V (F ) ⊂ P1K
(x31 , x22 ) y12 y2 Double point plus a simple point
(x1 x2 , x31 − x32 ) y1 − y23
3
Three different points
HILBERT FUNCTIONS OF
OF COHEN-MACAULAY
COHEN-MACAULAY LOCAL
LOCALRINGS
RINGS 197
25

Proof. By the previous theorem, we have A  grm (A) = K[y1 , y2 ]/Ann(F )


where F ∈ K[y1 , y2 ] is a degree three form in the variables y1 , y2 . Since K is
an algebraic closed field, F can be decomposed as product of three linear forms
L1 , L2 , L3 , i.e. F = L1 L2 L3 . We set d = dimK L1 , L2 , L3 , so we only have to
consider three cases. If d = 1, then we can assume F = y13 , but this has not the
right Hilbert function.
If d = 2, then we can assume F = y12 y2 . It is easy to see that Ann(y12 y2 ) =
(x1 , x22 ). If d = 3 then we can assume F = y13 − y23 . In this case we get Ann(y13 −
3

y23 ) = (x1 x2 , x31 −x32 ). Since V (y12 y2 ) (resp. V (y13 −y23 )) is a degree three subscheme
of P1K with two (resp. three) point basis we get that the algebras of the statement
are not isomorphic. 

The classification of Artinian Gorenstein local rings with Hilbert function HFA =
{1, 3, 3, 1} had been studied by Casnati and Notari in [7] and by Cartwright et al.
in [6]. In [14] J. Elias and the author present a different proof by using Theorem
5.10 and the geometric models of the varieties defined by them are described.
Gorenstein local rings with symmetric Hilbert function are not necessarily
canonically graded. The following exercise shows that we cannot extend the main
result of this section to higher socle degrees.
Exercise 5.13. Let A be an Artinian Gorenstein local local k-algebra with
Hilbert function HFA = {1, 2, 2, 2, 1}. Then A is isomorphic to one and only one
of the following rings:
(a) R/I with I = (x4 , y 2 ) and I ⊥ = x3 y and A is isomorphic to its associated
graded ring,
(b) R/I with I = (x4 , −x3 + y 2 ) and I ⊥ = x3 y + y 3 . The associated graded
ring is of type (a) and it is not isomorphic to R/I,
(c) R/I with I = (x2 + y 2 , y 4 ) and I ⊥ = xy(x2 − y 2 ) and A is isomorphic
to its associated graded ring.
The computation can be performed by using the techniques of this section, a dif-
ferent approach can be found in [15].

Next we consider the Artinian Gorenstein local rings A of socle degree three.
Then by using Theorem 5.5, we know that
HFA = {1, m, n, 1}
with m ≥ n. J. Elias and M.E. Rossi recently proved the following result (see [14],
Theorem 4.1.).
Theorem 5.14. There exists an isomorphism between the Artinian Gorenstein
local k-algebras (A, m) and (B, n) with Hilbert function {1, m, n, 1}, m ≥ n if and
only if QA (0)  QB (0) as graded k-algebras.
Since QA (0) is Gorenstein with Hilbert function {1, n, n, 1}, we obtain the fol-
lowing result which reduces the problem of the classification of the Artinian Goren-
stein algebras of socle degree three to the classification of the projective cubics.
Corollary 5.15. The classification of Artinian Gorenstein local k-algebras
with Hilbert function {1, m, n, 1}, m ≥ n, is equivalent to the projective classification
of cubic hypersurfaces H = V (F ) ⊂ Pn−1
k .
198
26 MARIA
MARIA EVELINA
EVELINA ROSSI
ROSSI

The crucial point in the proof of Theorem 5.14 is the following. Let A be an Artinian
Gorenstein local k–algebra with Hilbert function {1, m, n, 1}, then A ∼
= R/AnnR (F )
where
2
F = F3 + yn+1 + · · · + ym
2
∈P
with F3 a non-degenerate degree three form in Pn = k[y1 , . . . , yn ] (F = F3 if n = m).
Non-degenerate in Pn = k[y1 , . . . , yn ] means that F3 ∈ Pn and R/AnnR (F3 ) has
embedding dimension n.

Starting from the above result we can prove a structure theorem for Artinian
Gorenstein local k–algebra A = R/I with maximal ideal m, embedding dimension
m and socle degree three. Denote by Rn := k[[x1 , . . . , xn ]] with n ≤ m the subring
of R = k[[x1 , . . . , xm ]].
Theorem 5.16. Let A be an Artinian local k–algebra of embedding dimension
m and let n = HFA (2).
A is Gorenstein of socle degree three if and only if n ≤ m and there exists a non-
degenerate cubic form F3 ∈ Pn such that A ∼ = R/I where

⎨ AnnRn (F3 )R + (xi xj , x2j − 2σ(x1 , · · · , xn ))i<j, n+1≤j≤m if n < m
I=

AnnR (F3 )R if n = m
being σ ∈ Rn any form of degree 3 such that σ ◦ F3 = 1.
For a proof we refer to [9].
Taking advantage of the classification of the cubics in P2 , from the above results
we may deduce the classification, up to isomorphism, of the Artinian Gorenstein
local k-algebras of socle degree and embedding dimension ≤ 3 (see [14]).

References
[1] Abhyankar S., Local rings of high embedding dimension, Amer. J. Math. 89 (1967), 1073-1077.
[2] Arslan F., Mete P., Hilbert functions of Gorenstein monomial curves, Proc. Amer. Math. Soc.
135 (2007), 1993-2002.
[3] Bertella V., Hilbert function of local Artinian level rings in codimension two, J. Algebra 321
(2009), no. 5, 1429-1442.
[4] Bondil R., Geometry of superficial elements, Annales de la Faculte’ des sciences de Toulouse,
Ser. 6, 14, no. 2 (2005), 185-200.
[5] Bondil R., Le’ Dung Trang, Characterisations des elements superficials d’un ideal, C.R. Acad.
Sci. Paris, 332, Ser. I (2001), 717–722.
[6] Cartwright D. A., Erman D., Velasco M., Viray B., Hilbert schemes of 8 points in A8 , Algebra
and Number Theory, 3(7) (2009), 763–795.
[7] Casnati G., Notari R., On the Gorenstein locus of some punctual Hilbert schemes, J. of Pure
and Applied Algebra Vol. 213, Issue 11, (2009), 2055–2074.
[8] Casnati G., Notari R., On some Gorenstein loci in Hilb6 (Pk4 ), J. of Algebra 308 (2007),
493–523.
[9] Casnati G., Elias J., Notari R., Rossi M. E., Poincaré series and deformations of Gorenstein
local algebras, to appear in Comm. Alg.
[10] Corso A., Polini C., Hilbert Coefficients of ideals with a view Toward Blowup Algebras,
Chapter 2, Syzygies and Hilbert functions, Lecture Notes in Pure and Applied Mathematics,
Vol. 254 (2007), 41–66.
[11] Corso A., Polini C., Rossi M. E., Depth of associated graded rings via Hilbert coefficients of
ideals, J. Pure and Applied Algebra 201 (2005), 126-141.
HILBERT FUNCTIONS OF COHEN-MACAULAY LOCAL RINGS 199
27

[12] Elias J., Characterization of the Hilbert-Samuel polynomial of curve singularities, Comp.
Math., 74(1990), 135-155.
[13] Elias J., The conjecture of Sally on the Hilbert function for curve singularities, J. of Algebra,
160 (1993), 42–49.
[14] Elias J., Rossi M. E., Isomorphism classes of Artinian local rings via Macaulay’s inverse
system, to appear in Trans. Amer. Math. Soc. (2010), arXiv:0911.3565.
[15] Elias J., Valla G., Rigid Hilbert functions, J. Pure and Applied Algebra 71 (1991), 19-41.
[16] Elias J., Valla G., Structure theorems for certain Gorenstein ideals, Michigan Journal of
Math. 57 (2008), 269-292.
[17] Elias J., Valla G., Isomorphism classes of certain Artinian Gorenstein Algebras, to appear in
Algebras and Representation Theory, DOI 10.1007/s10468-009-9196-8 (2009).
[18] Emsalem J., Geometrie des points epais, Bull. Soc. Math. France, 106 (1978), 399–416.
[19] Goto S., Ghezzi L., Hong J., Ozeki K., Phuong T. T., Vasconcelos W. V., Cohen-Macaulayness
versus the vanishing of the first Hilbert coefficient of parameter ideals, to appear in the J. of
the London Mathematical Society, arXiv:1002.0391 (2009)
[20] Goto S., Nishida K., Hilbert coefficients and Buchsbaumness of associated graded rings. J.
Pure Appl. Algebra 181 (2003), no. 1, 61–74.
[21] Guerrieri A. and Rossi M. E., Hilbert coefficients of Hilbert filtrations, J. Algebra 199, no.1,
(1998), 40–61.
[22] Herzog J., Rossi M. E., Valla G., On the depth of the symmetric algebra. Trans. Amer. Math.
Soc. 296 (1986), no. 2, 577–606.
[23] Herzog, J., Waldi, R., A note on the Hilbert function of a one-dimensional Cohen-Macaulay
ring. Manuscripta Math. 16 (1975), no. 3, 251–260.
[24] Huckaba S., Marley T., Hilbert coefficients and the depths of associated graded rings, J.
London Math. Soc. 56 (1997), 64–76.
[25] Huckaba S., Huneke C., Normal ideals in regular rings, J. Reine Angew. Math. 510 (1999),
63-82.
[26] Huneke C., Swanson I., Integral Closure of Ideals, Rings, and Modules, London Mathematical
Lecture Note Series 336, Cambridge University Press (2006).
[27] Iarrobino A., Associated graded algebra of a Gorenstein Artin algebra, Vol. 514 of. Memoirs
of the American Mathematical Society (1994).
[28] Iarrobino A., Kanev V., Power sums, Gorenstein algebras, and determinantal loci, Lecture
Notes in Mathematics, vol. 1721, Springer-Verlag, Berlin, 1999.
[29] Itoh S., Coefficients of normal Hilbert polynomials. J. Algebra 150 (1992), no. 1, 101–117.
[30] Itoh S., Hilbert coefficients of integrally closed ideals, J. Algebra 176 (1995), 638-652.
[31] Kirby D., A note on superficial elements of an ideal in a local ring, Quart. J. Math. Oxford
(2), 14 (1963), 21-28.
[32] Mandal M., Verma J., On the Chern number of an ideal, Proc. Amer. Math. Soc. 138 (2010),
1995–1999.
[33] Matlis E., One-dimensional Cohen-Macaulay rings, Lecture Notes in Math., Springer, Berlin
(1973).
[34] Narita M., A note on the coefficients of Hilbert characteristic functions in semi-regular local
rings, Proc. Cambridge Philos. Soc., 59 (1963), 269–275.
[35] Northcott D. G., A note on the coefficients of the abstract Hilbert function, J. London Math.
Soc., 35 (1960), 209-214.
[36] Poneen B., The module space of commutative algebras of finite rank, J. Eur. Math. Soc. 10
(2008), 817–836.
[37] Puthenpurakal T., The Hilbert function of a maximal Cohen-Macauly module,
arXiv:math/0409051v2 [math.AC].
[38] Ratliff L. J., Rush D., Two notes on reductions of ideals, Indiana Univ. Math. J. 27 (1978)
929-934.
[39] Rossi M. E., A bound on the reduction number of a primary ideal, Proc. of the American
Math. Soc., 128, Number 5 (1999), 1325–1332.
[40] Rossi M. E., Hilbert function and free resolutions of filtered modules, Proceedings 4th Japan-
Vietnam Joint Seminar on Commutative Algebra, Meiji University, Institute of Mathematics
(2009), 8–44.
[41] Rossi M. E., Sharifan L., Consecutive cancellations of Betti numbers of local rings, Proc.
Amer. Math. Soc. 138 (2009), 61–73.
200
28 MARIA EVELINA ROSSI
Bibliography

[42] Rossi M. E., Trung N. V., G. Valla, Castelnuovo-Mumford regularity and extended degree,
Trans. Amer. Math. Soc. 355 (2003), no. 5, 1773–1786
[43] Rossi M. E., Valla G., A conjecture of J.Sally, Communications in Algebra, 24 (13) (1996),
4249–4261.
[44] Rossi M. E., Valla G., Cohen-Macaulay local rings of embedding dimension e + d − 3, Proc.
London Math. Soc. 80 (2000), 107-126.
[45] Rossi M. E., Valla G., The Hilbert function of the Ratliff-Rush filtration, J. Pure Appl.
Algebra, 201 (2005), 25-41.
[46] Rossi M. E., Valla G., On the Chern number of a filtration, Rend. Semin. Mat. Univ. Padova
121 (2009), 201–222.
[47] Rossi M. E., Valla G., Hilbert Functions of Filtered Modules, Lecture Notes of the Unione
Matematica Italiana, Springer, n. 9 (2010).
[48] Rossi M. E., Valla G., Vasconcelos W.V., Maximal Hilbert Functions, Result. Math. 39 (2001)
99-114.
[49] Sally J., On the associated graded ring of a local Cohen-Macaulay ring, J. Math. Kyoto Univ.
(1977) 17, no. 1, 19–21.
[50] Sally J., Cohen-Macaulay local rings of embedding dimension e+d-2, J. Algebra 83 (1983),
no. 2, 393-408.
[51] Sally J., Good embedding dimensions for Gorenstein singularities, Math. Ann., 249 (1980),
95-106.
[52] Sally J., Hilbert coefficients and reduction number 2., J. Algebraic Geom. 1 (1992), no. 2,
325–333.
[53] Srinivas V. and Trivedi V., On the Hilbert function of a Cohen-Macaulay local ring, J.
Algebraic Geom. 6 (1997), 733-751.
[54] Trivedi V., Finiteness of Hilbert functions for generalized Cohen-Macaulay modules, Comm.
Algebra 29 (2) (2001), 805813.
[55] Trung N.V., Filter-regular sequences and multiplicity of blow-up rings of ideals of the principal
class. J. Math. Kyoto Univ. 33 (1993), no. 3, 665–683.
[56] Valla G., Problems and results on Hilbert functions of graded algebras. Six lectures on com-
mutative algebra (Bellaterra, 1996), 293–344, Progr. Math., 166, Birkhauser, Basel, 1998.
[57] Vasconcelos W. V. , Arithmetic of Blowup Algebras, London Mathematical Society, Lecture
Note Series 195, Cambridge University Press (1994).
[58] Vasconcelos W. V. , Hilbert Functions, Analytic Spread, and Koszul Homology, Contempo-
rary Mathematics, Vol 159 (1994), 410–422.
[59] Wang H. J., On Cohen-Macaulay local rings with embedding dimension e + d − 2, J. Algebra
190, (1997), no.1, 226-240.
[60] Zariski O., Samuel P., Commutative Algebra, Vol II, Van Nostrand, New York, 1960.
[61] CoCoATeam, CoCoA: a system for doing Computations in Commutative Algebra, Available
at http://cocoa.dima.unige.it.

Department of Mathematics, University of Genova, Via Dodecaneso 35, 16146 Gen-


ova, Italy
E-mail address: rossim@dima.unige.it
Contemporary Mathematics
Volume 555, 2011

Some Homological Properties of Almost Gorenstein Rings

Janet Striuli and Adela Vraciu

Abstract. We show that the residue field k is a direct summand of the second
syzygy of the canonical module for some almost Gorenstein rings. This implies
that over a Teter ring the only totally reflexive modules are the free ones. We
provide an example of an almost Gorenstein ring which has infinitely many
non-isomorphic totally reflexive modules.

Introduction
Totally reflexive modules are the object of extensive research activity, and yet
it is an open problem to determine conditions that are necessary and sufficient
for the existence of a non-free totally reflexive module, see for example [5]. The
starting point of our investigation was to consider the problem for some artinian
local rings, and in particular for the class of Teter rings, for which we show that
the only totally reflexive modules are the free ones, see Corollary 2.1.
Teter rings are a particular example of almost Gorenstein rings, defined in [10]:
Definition 0.1. An artinian local ring (R, m, k) is almost Gorenstein if the
inclusion 0 :R (0 :R I) ⊆ I :R m holds for every ideal I ⊆ R.
Our investigation led us to the study of the syzygies of the canonical module
for a certain class of almost Gorenstein rings, for which we prove the following:

Main Theorem. Let (R, m, k) be a local noetherian ring which is almost


Gorenstein with canonical module ωR . Assume that R is not Gorenstein, and write
R = S/J, where S is an artinian Gorenstein ring. Denote by c the dimension of
the k-vector space (J :S m)/(mJ :S m) and assume c > 0. Then the vector space
kc is a direct summand of the second syzygy of the canonical module ωR .

Issues concerning direct summands of syzygies have played an important role


in criteria for regularity, see [11], and finite G-dimension, see [15]. In particular
the result of [15] implies that for a local ring, the canonical module cannot appear
as a summand of a syzygy module of the residue field unless the ring is regular.

1991 Mathematics Subject Classification. Primary 13D02, Secondary 13C99.


Key words and phrases. Commutative algebra, totally reflexive modules, canonical module,
almost gorenstein ring.
The first author was supported in part by NSF Grant DMS #0901427.
The second author was partly supported by NSA grant #H98230-09-1-0057.
1

201
202
2 JANET STRIULI AND ADELA VRACIU

The Main Theorem gives information on the size of the minimal free resolutions
of the canonical module for a certain class of almost Gorenstein rings. A sequence
{ai }i∈N has exponential growth if there exists an integer A > 1 such that ai ≥ Ai
for all i  0. Studies on the exponential growth of the sequence of the Betti
numbers of the canonical module can be found for example in [13], [9]. As an
immediate consequence of the above theorem, we obtain a new family of ring for
which the Betti numbers of the canonical module have exponential growth. This
follows immediately from the fact that the Betti numbers of the residue field have
exponential growth if the ring is not a complete intersection, see for example [4].
The paper is organized in the following way. In Section 1 we prove the Main
Theorem and in Section 2 we give some examples of almost Gorenstein rings. The
connection between the Main Theorem and the existance of totally reflexive modules
is given in Section 3 where we also give, in contrast, an example of an almost
Gorenstein ring that admits a totally reflexive modules. In Section 4 we consider
almost Gorenstein rings that are quotients of a polynomial ring by a monomial
ideal, and we show that k is a direct summand of the first or the second syzygy of
the canonical module.
In the following (R, m, k) will denote a local noetherian ring with maximal ideal
m and residue field k.

1. The canonical module over almost Gorenstein rings


In this section we will prove the Main Theorem. Given two ideals of R, I and
J, we will often use the colon ideal I :R J. When I is generated by a single element
I = (f ), to abbreviate the notation I :R J will be denoted by f :R J. Similarly,
when J = (f ) we will write I :R f instead of I :R (f ). We will often use that
0 :R (0 :R I) = I for every ideal I ⊂ R, provided that R is a Gorenstein artinian
ring. For easy reference, we collect two properties of the colon ideal in the following
Lemma 1.1. Let (R, m, k) a noetherian local ring and I1 and I2 two ideals of
R. Then the following hold:
(1) (0 :R I1 ) :R I2 = (0 :R I1 I2 ) = (0 :R I2 ) :R I1 ;
(2) If R is Gorenstein and artinian, then 0 :R (I1 :R I2 ) = I2 (0 :R I1 ).
Proof. (1) is straightforward. For (2), as the ring R is Gorenstein, it is enough
to show that
0 :R (0 :R (I1 :R I2 )) = 0 :R (I2 (0 :R I1 )).
But 0 :R (I2 (0 :R I1 )) = (0 :R (0 :R I1 )) :R I2 by (1) and applying twice the
assumption that R is Gorenstein, we obtain (0 :R (0 :R I1 )) :R I2 = I1 :R I2 = 0 :R
(0 :R (I1 :R I2 )). 

For any artinian ring R, one may assume, by the Cohen Structure Theorem,
that R is a quotient S/J where S is a Gorenstein artinian ring. If S is a Gorenstein
ring, then 0 :S (0 :S I) = I for all ideals I in S. Therefore without loss of generality
we may assume that J = 0 :S K for some ideal K ⊆ S. The following result is an
adaptation of Proposition 4.1 in [10].
Lemma 1.2. Let (S, mS ) be a Gorenstein artinian local ring and let K be an
ideal minimally generated by f1 , . . . , fn such that the ring R = S/(0 :S K) is almost
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 203
3

Gorenstein, but not Gorenstein. Denote by Ki the ideal (f1 , . . . , fˆi , . . . , fn ), where
the element fi is dropped from the list f1 , . . . , fn . Then the equality
mS = fi :S Ki + (Ki (fi :S Ki )) :S fi ,
holds for all i ∈ {1, . . . , n}.
In particular, the equality
mS = fi :S Ki + Ki :S fi ,
holds for all i ∈ {1, . . . , n}.
Proof. The last statement follows from the first, as (fi :S Ki )Ki ⊂ Ki ⊂ mS .
Without loss of generality we may assume that i = 1. Let I = (0 :S f1 ) and
denote by J = (0 :S K). As J ⊆ I and S/J is almost Gorenstein, one has the
inclusion
(1.1) J :S (J :S I) ⊆ I :S mS .
We first show that J :S (J :S I) = (0 :S K(f1 : K)). Indeed, the following equalities
hold:
J :S (J :S I) = (0 :S K) :S (J :S I), by definition of J,
= (0 :S K(J :S I)), applying Lemma 1.1 (1),
= (0 :S K((0 :S K) :S I))), by definition of J,
= (0 :S K((0 :S I) :S K)), applying Lemma 1.1(1),
= (0 :S K(f1 :S K)), as I = (0 :S f1 ) and S is Gorenstein.

On the other hand, the right hand side of inclusion (1.1) can be written as
I :S mS = (0 :S f1 ) :S mS = 0 :S f1 m,
where the first equality holds by the definition of I and the second equality holds
by Lemma 1.1(1).
Now inclusion (1.1) becomes (0 :S K(f1 :S K)) ⊆ 0 :S f1 m and this, together
with the assumption that S is Gorenstein, implies that
f1 mS = K(f1 :S K).
n
In particular, for every element x ∈ mS we can write nxf1 = i=1 ui fi , with ui ∈
(f1 :S K) = (f1 : K1 ) and hence (x − u1 )f1 = i=2 ui fi . This implies that
(x − u1 )f1 ∈ (f1 : K1 )K1 . Finally x is an element of
((f1 :S K1 )K1 ) :S f1 + f1 :S K1 .
As x is an arbitrary element in the maximal ideal m, we have the thesis.


Remark 1.3. The conditions in Lemma 1.2 are not sufficient for a ring to
be almost Gorenstein. Take for example S = k[x, y, z, u]/(x2 , y 2 , z 2 , u2 ), f1 = xz,
f2 = yz, f3 = xu, f4 = yu so that R = k[|x, y, z, u|]/(x2 , y 2 , z 2 , u2 , xy, zu). For the
ideal I = (u) we obtain 0 :R (0 :R I) = mR but I :R mR = (u, yz, xz), showing that
R is not almost Gorenstein.
On the other hand it is easy to check that Ki :S fi = mS for every i = 1, . . . , 4.
204
4 JANET STRIULI AND ADELA VRACIU

Now we give the proof of the Main Theorem, Theorem 1.4. We will use ΩiR (M )
to denote the ith syzygy of an R-module M .
Theorem 1.4. Let (R, mR , k) be a local noetherian ring which is almost Goren-
stein with canonical module ωR . Assume that R is not Gorenstein, and write
R = S/J, where (S, mS , k) is an artinian Gorenstein local ring. Let c = dim[k](J :S
mS )/(mS J :S mS ) and assume c > 0. Then the vector space kc is a direct summand
of the second syzygy of the canonical module ωR .
Proof. In the following, denote by y  the image in R of an element y ∈ S.
Since S is Gorenstein, we may assume that J = (0 :S K), for some ideal K =
(f1 , . . . , fn ). The canonical module ωR is given by
HomS (R, S) = HomS (S/(0 :S K), S) ∼
= 0 :S (0 :S K) = K,
where equality holds as S is Gorenstein. Let

... / Rp ∂2
/ Rm ∂1
/ Rn ∂0
/K /0

be a minimal presentation of the canonical module.


By the last statement in Lemma 1.2, we can choose a set of minimal gen-
erators x1 , . . . , xe of the maximal ideal mS , such that xi ∈ f1 :S (f2 , . . . fn ) or
xi ∈ (f2 , . . . , fn ) :S f1 for every i = 1, . . . , e.
Let u ∈ (J :S mS ) \ (mS J :S mS ). There exists an h ∈ {1, . . . , e} such that
xh u ∈ / mS J, and there is a relation a1 f1 + a2 f2 + · · · + an fn = 0 in S, such that
either a1 = xh or a2 = xh . The column vectors D = (a1 , . . . , an ) is part of a
minimal generating set for the module of first syzygies. After a choice of basis, we
may assume that D1 = D , D2 , . . . , Dm 
are the columns of ∂1 . Let D1 , D2 , . . . , Dm
denote the liftings of these vectors to S n , and let D denote the matrix with columns
D1 , . . . , Dm .
We claim that the m-tuple u ∈ Rm which has u in the first entry and zero in
all the other entries is part of a minimal generating set for the module Ω2R (ωR ). To
prove the claim, denote by B  = (bij ) the matrix representing ∂2 . It is clear by the
choice of u that u is in the kernel of ∂1 . Assume that u is not part of a minimal
set of generators of the second syzygy and denote by u the lift of u to S m where
all the entries are equal to zero except the first entry which is equal to u. This
implies that we can write the u as follow
⎛ ⎞ ⎛ ⎞
b11 b1p
⎜ ⎟ ⎜ ⎟
u = c1 ⎝ ... ⎠ + · · · + cp ⎝ ... ⎠ + j
bm1 bmp
where ci are elements of the maximal ideal mS and j ∈ JS m . Moreover we have
b1i D1 + · · · + bmi Dm ∈ JS n

for all i = 1, . . . , p. This implies that ∂1 u = cj (b1j D1 + · · · + bmj Dm ) ∈ mJS n .
On the other hand, either the first component or the second component of Du,
according to whether xh appears in the first component or the second component
of D1 , is equal to xh u which by assumption is not in mS J.
If u1 , . . . , uc are elements in (J :S mS ) such that their representatives in (J :S
mS )/(mS J :S mS ) are a basis, one can then construct vectors ui ∈ Rm using the
same procedure as above, and the same argument shows that they are part of a
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 205
5

minimal system of generators of Ω2R (ωR ). The R-module spanned by these vectors
is isomorphic to kc , and it is a direct summand of Ω2R (ωR ).

Before closing the section, we record a remark which gives a condition equivalent
to the hypothesis in the Main Theorem; it will be used in the later sections.
Remark 1.5. Assume R = S/J it is the quotient of an artinian Gorenstein
ring S, and we write J = 0 :S K, for some ideal K ⊂ S, then we have
mS J :S mS = J :S mS ⇔ mS K :S mS = K :S mS .
Indeed, one has J :S mS = mS J :S mS if and only if

(0 :S K) : mS = mS (0 :S K) :S mS ,
or equivalently, as S is Gorenstein, if and only if
(1.2) 0 :S ((0 :S K) :S mS ) = 0 :S (mS (0 :S K) :S mS ).
The first term of the equality (1.2) is equal to 0 :S (0 :S mS K) by (1.1)(1)
applied to K and mS , and hence it is equal to mS K as S is Gorenstein.
For the second term in the equality (1.2) the following equalities hold:
0 :S (mS (0 :S K) :S mS ) = mS [0 :S mS (0 :S K)], by Lemma 1.1(2),
= mS ((0 :S (0 :S K)) :S mS ), by Lemma 1.1(1),
= mS (K :S mS ) as S is Gorenstein.
In particular (1.2) holds if and only if mS (K :S mS ) = mS K or, equivalently, if and
only if K :S mS = mS K :S mS .

2. Examples of Almost Gorenstein Rings


Remark 2.1. A ring R is called Teter if R = S/(δ) where S is a Gorenstein
artinian ring with socle element generated by δ. By Theorem 2.1 and Proposition
1.1 of [10] a Teter ring is an almost Gorenstein ring.
In [10], the authors prove that quotiens of Cohen-Macaulay rings of finite
Cohen-Macaulay type via a special system of parameters are almost Gorenstein.
We adapt their proof to show the following
Proposition 2.2. Let (R, m, k) be a Cohen-Macaulay ring such that the equal-
ity m Ext1R (M, R) = 0 holds for all maximal Cohen-Macaulay module M . Then
R/(xx) is an almost Gorenstein ring for all systems of parameters x .
Proof. Let I be any ideal of R containing the ideal generated by x . We need
x) :R I) ⊆ I :R m. Assume that I is generated by f1 , . . . , fn
x) :R ((x
to show that (x
and consider the short exact sequence

n
R R
0→ → → N → 0,
x) : I
(x x)
(x
where the first non-zero map is given by u → (f1 u, . . . , fn u). Applying the functor
HomR ( , R/(x x)) to the short exact sequence we obtain:
R R R n R R R
0 → HomR (N, ) → HomR ( , ) → HomR ( , ) → Ext1R (N, ).
x)
(x x)
I (x x) : I (x
(x x) x)
(x
206
6 JANET STRIULI AND ADELA VRACIU

The cokernel of the middle map is the cokernel of:


x ) :R I
(x x) :R ((x
(x x) :R I)
⊕ →
x)
(x x)
(x
given by (u1 , . . . , un ) → f1 u1 + · · · + fn un . The cokernel is therefore isomorphic
x x):R I)
to (x ):R ((x
I and embeds in Ext1R (N, R/(x x)). As (xx) ⊆ annR Ext1R (N, (xR
x) ), we

R
1
obtain the isomorphism Ext (N, R/(x x)) ∼= Ext
R
d+1
(N, R) which is isomorphic to
x x
Ext1R (Ωd (N ), R) and therefore annihilated by m. This implies that m (xx):R ((x
I
x):R I)
=
0 and therefore the thesis. 

3. Almost Gorenstein rings and totally reflexive modules


We begin with the definition of totally reflexive modules.
Definition 3.1. An R-module M is totally reflexive if and only if M ∗∗ ∼
=M
and ExtiR (M, R) = 0 = ExtiR (M ∗ , R), for all i > 0.
The following lemma is well-known by the experts. We include the proof for
easy reference.
Lemma 3.2. Let (R, m, k) be a local ring with canonical module ωR . If k is
a direct summand of any syzygy of ωR then there are no non-free totally reflexive
modules.
Proof. Let X be a totally reflexive module. By definition, ExtiR (X, F ) = 0
for every free module F and for every i > 0. Applying the functor HomR (X, )
to the short exact sequence 0 → Ω1R (ωR ) → F → ωR → 0, one obtains the equal-
ities Ext1R (X, ωR ) = Exti+1 i
R (X, ΩR (ωR )) for every R-module M . In particular,
i+1
ExtR (X, k) = 0 if k is a direct summand of ΩiR (ωR ). This shows that X has fi-
nite projective dimension and therefore it is free, by the Auslander-Bridger formula
(see for example Theorem 1.4.8 [7]) and the Auslander-Buchsbaum formula (see for
example Theorem 1.3.3 [6]). 
Remark 3.3. In [12] Theorem 1.6 and Remark 1.8 (e) it is shown that if a
local ring R can be written as a quotient S/J , where (S, mS ) is a local ring such
that dimk (J :S mS )/(mS J :S mS ) ≥ 2 then there are no non-free totally reflexive
modules. The Main Theorem and Lemma 3.2 show that for almost Gorenstein rings
the conclusion holds even in the case when dimk (J :S mS )/(mS J :S mS ) ≥ 1.
Corollary 3.4. Let R be a Teter ring, then R does not admit totally reflexive
modules which are not free.
Proof. let S be an artinian Gorenstein ring such that R = S/(δ), where δ
generates the socle of S. As δ :S mS strictly contains mS δ :S mS = 0 :S mS , by
Remark 2.1 the conditions of the Main Theorem are satisfied and therefore the
corollary follows from Lemma 3.2. 
Teter rings are the ring of smallest Gorenstein colength, for a definition see [1].
The following example shows that it is possible to have totally reflexive modules
over rings of Gorenstein colength 2.
Example 3.5. The ring R = k[|x, y, z|]/(x2 , y 2 , z 2 , yz) has totally reflexive
modules which are not free. On the other hand, let S = k[|x, y, z|]/(x2 , y 2 , z 2 ) and
J = (yz)S then J :S mS = (mS J :S mS ). The ring R has Gorenstein colength 2.
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 207
7

We conclude the section with an example of an almost Gorenstein ring that


admits a totally reflexive module, and therefore infinitely many by [8]. For the
argument we use some facts which we collect in the following three remarks.
Remark 3.6. In [3], Theorem 3.1, the authors prove that if (R, m) is a local
ring and y = y1 , . . . , yd is a regular sequence in m2 then R/(yy ) has a totally reflexive
module.
Remark 3.7. Let (R, m) be a local ring. Let M be a finitely generated R-
module and 0 → Ω1R (M ) → F → M → 0 be the beginning of a minimal free
resolution of M . For every element x of the maximal ideal, denote by μx the
multiplication by x. If for every x in a minimal set of generators of m there exists
a linear map φx such that the diagram:
0 / Ω1 (M ) /F /M /0
yy
R
yy
μx yy
yy φx
 |y
0 / Ω1 (M ) /F /M /0
R

commutes, then m Ext1R (M, N ) = 0 for all modules N .


Remark 3.8. Let (R, m) be a local Cohen-Macaulay ring with canonical mod-
ule ωR . For every R-module N , denote by N ∨ the R-module HomR (N, ωR ). Let
M and L be two maximal Cohen-Macaulay modules. There exists an isomorphism
φ : Ext1R (M, L) → Ext1R (L∨ , M ∨ ) such that φ(ξ1 ) = ξ2 where
ξ1 : 0 → L → X → M → 0
and
ξ 2 : 0 → M ∨ → X ∨ → L∨ → 0
is obtained by applying HomR ( , ωR ) to ξ1 .
Example 3.9. The ring R = C[[x, y, z, u, v]]/(xz − y 2 , xv − yu, yv − zu) is of
finite Cohen Macaulay type and its only indecomposable maximal Cohen-Macaulay
modules are R, the ideals ωR ∼ = α = (x, y), α2 = (x2 , y 2 , xy), β = (x, y, u) and the
1
R-module ΩR (β). For a proof of this see for example [14]. In the following we
show that the maximal ideal m annihilates all the R-modules Ext1R (M, R) for M
maximal Cohen-Macaulay.
For the ideal α, the first syzygy Ω1R (α) is generated by
[−v, u], [−z, y], [−y, x].
The following list gives the maps of Remark 3.7

y z−y −y 0
φx = φy =
−x −y
+x x 0

z 0 v − y −z
φz = φv =
−y 0
−u + x y
y 0
φu =
−u 0
In particular, by Remark 3.7, we have that m Ext1R (α, N ) = 0 for every R-module
N.
For every maximal Cohen-Macaulay module M , the following holds:
annR (Ext1R (M, R)) = annR (Ext1R (ωR , M ∨ ) = annR (Ext1R (α, M ∨ )) = m,
208
8 JANET STRIULI AND ADELA VRACIU

where the second equality follows from Remark 3.8. Now the sequence x2 , v 2 , z 2 +u2
is a system of parameters of R contained in m2 . Therefore R/(x2 , v 2 , z 2 + u2 ) is
almost Gorenstein by Lemma 2.2 and has a totally reflexive module by Remark 3.6.

4. The monomial case


The main result of this section deals with artinian almost Gorenstein rings
which are obtained as quotients of polynomial rings by monomial ideals.
Ad
Theorem 4.1. Let S = k[x1 , . . . , xd ]/(xA
1 , . . . , xd ), and let f1 , . . . , fn be
1

monomials in S such that R = S/0 :S (f1 , . . . , fn ) is almost Gorenstein. Then


the residue field is a direct summand of the first or second syzygy of the canonical
module ωR .
The proof of Theorem 4.1 will be given after we prove the following:
Ad
Theorem 4.2. Let S = k[x1 , . . . , xd ]/(xA 1 , . . . , xd ), and let f1 , . . . , fn be
1

monomials in S such that


(1) fi does not divide fj for every i = j;
(2) xi divides fj for all i ∈ {1, . . . , d} and for all j ∈ {1, . . . n};
n
(3) (x1 , . . . , xu ) ⊆ i=1 fi :S (f1 , . . . , fn ).
Then, one of the following conclusions holds:
(A) There exists an i ∈ {1, . . . , n} and a j ∈ {1, . . . , u} such that
fi
∈ (f1 , . . . , fn ) :S (x1 , . . . , xu );
xj
(B) There exist mutually disjoint sets S1 , . . . , Sn ⊆ {1, . . . , u} such that for all
i ∈ {1, . . . , n} and all j ∈ {1, . . . , u}, xj fi = 0 ⇔ j ∈ Si .
Proof. Before we proceed with the proof, we establish some claims that we
N
will use later. Write each fj = Πni=1 xi ji , with Nji < Ai .
Claim 1: If xi fj ∈ (fk ), for some integers i, j, k then one of the following cases hold:

Nji = Nki − 1
(i)
Njl ≥ Nkl , for every l = i
(ii) Nji = Ai − 1
Moreover, for fixed j, k, the first case can hold for at most one i.
Proof of Claim 1: Note that (ii) is equivalent to xi fj = 0 in S. If 0 = xi fj ∈ (fk ),
then (i) is obtained by comparing the exponents of each variable for xi fj and fk .
The fact that Nji = Nki − 1 is due to the assumption that fk does not divide fj .
For the last statement, assume that there are two indeces i1 and i2 such that

Nji1 = Nki1 − 1
Njl ≥ Nkl , for every l = i1
and
Nji2 = Nki2 − 1
Njl ≥ Nkl , for every l = i2
then, Nki2 − 1 = Nji2 ≥ Nki2 which is a contradiction.
Claim 2: If conclusion B holds but A does not hold, then we have the following:
(i) each set Si has cardinality at least 2;
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 209
9

(ii) for every k ∈ Si we have xk ∈ (fi ) :S (f1 , . . . , fn ), and xk ∈


/ fj :S
(f1 , . . . , fn ) for all j = i.
Proof of Claim 2: Assume that there exist indeces i and k such that Si = {xk }.
Since xk fj = 0 for all j = i, it follows that case (A) holds, as
fi
∈ fi :S (f1 , . . . , fn ).
xk
For (ii), let k ∈ Si . Assume that xk ∈ fj :S (f1 , . . . , fn ) for some j = i. Then
0 = xk fi ∈ (fj ). As we may assume (i), there exists an l ∈ Si such that l = k. By
Claim 1, we have Nil ≥ Njl . As l ∈ / Sj , we have xl fj = 0, and thus Njl = Al − 1.
This contradicts the fact that Nil < Al − 1.
The proof of the theorem goes by induction on the number of variables d, the
case d = 1 being obvious. Assume that the theorem holds for d − 1 variables. We
now induct on the number n of polynomials. Assume that the theorem holds in the
case of n − 1 polynomials.
Claim 3: If there exists k ∈ {1, . . . , u} such that xk fi = 0 for all i ∈ {1, . . . , n},
then we are done by induction on the number of variables. In particular, whenever
conclusion B holds for a subset of {f1 , . . . , fn } with respect to a subset {x1 , . . . , xs }
of {x1 , . . . , xu }, we may assume that the sets S1 , S2 , . . . , asserted in Conclusion B
form a partition of {1, . . . , s}.
Indeed, we can write fi = xkAk −1 fi , with fi ∈ k[x1 , . . . , xˆk , . . . , xd ]. Assump-
tions (1), (2), (3) hold for {f1 , . . . , fn } viewed as monomials in d − 1 variables. If
conclusion (A) holds for {f1 , . . . , fn }, then it also holds for {f1 , . . . , fn }. Similarly,
if conclusion (B) holds for {f1 , . . . , fn }, then it also holds for {f1 , . . . , fn } (with the
same choice of the sets Si ).
Claim 4: Assume that conclusion B holds for {f1 , . . . , fn−1 } with respect to a set
of variables {x1 , . . . , xs }, with s ≤ u. Let S1 , . . . , Sn−1 
⊂ {1, . . . , s} be the sets
asserted in conclusion B. Let k ∈ {1, . . . , s}, and let i ∈ {1, . . . , n − 1} be such that
k ∈ Si .
Then we have either xk ∈ fi :S (f1 , . . . , fn ), or xk ∈ fn :S (f1 , . . . , fn ). Among
the k’s for which the first situation occurs, we can have xk fn = 0 for at most one
such k.
Proof of Claim 4: By Claim 2 (ii), we cannot have xk ∈ fj :S (f1 , . . . , fn−1 )
for any i = j ≤ n − 1. Thus, we have either xk ∈ fi :S (f1 , . . . , fn ), or xk ∈
fn :S (f1 , . . . , fn ). For the last part of the claim, assume that xk1 fn ∈ (fi1 ), and
xk2 fn ∈ (fi2 ), with k1 ∈ Si1 , and k2 ∈ Si2 . We need to show that one of xk1 fn
or xk2 fn is zero. If i1 = i2 , this follows from Claim 1. Assume that i1 = i2 and
xk1 fn = 0. Then Nnk1 = Ni1 k1 − 1, Nnl ≥ Ni1 l for all l = k1 . In particular,
Nnk2 ≥ Ni1 k2 . Since k2 ∈ / Si1 , we have xk2 fi1 = 0, and thus xk2 fn = 0.
Claim 5: If
(x1 , . . . , xu ) ⊆ fl :S (f1 , . . . , fn ).
l=i
for some i ∈ {1, . . . , n}, then conclusion A holds.
n−1
Proof of Claim 5: Assume that (x1 , . . . , xu ) ⊆ i=1 fi :S (f1 , . . . , fn ). The as-
sumptions (1),(2),and (3) in the theorem are satisfied for {f1 , . . . , fn−1 } with re-
spect to the variables {x1 , . . . , xu }, and by the induction hypothesis either A of B
holds. If (A) holds for {f1 , . . . , fn−1 } then it also holds for {f1 , . . . , fn }, and we are
done.
210
10 JANET STRIULI AND ADELA VRACIU

Assume that (B) holds for {f1 , . . . , fn−1 }. Let {1, . . . , u} = S1 ∪ . . . ∪ Sn−1

be
the partition asserted in conclusion (B). By Claim 4, for each k ∈ {1, . . . , u} we
have either xk ∈ fi :S fn or xk ∈ fn :S fi , where i ∈ {1, . . . , n − 1} is such that
k ∈ Si .
If the first situation occurs for all k ∈ {1, . . . , u}, then Claim 3 shows that
xk fn = 0 for all values of k except one, say k0 . Then conclusion A holds, with
fn
∈ (f1 , . . . , fn ) :S (x1 , . . . , xu ).
xk 0
Assume that there exists a k0 such that xk0 fi0 ∈ (fn ) holds, where i0 is such
that k0 ∈ Si0 . Note that xk0 fi0 = 0, so we have Ni0 l ≥ Nnl for all l = k0 . By Claim
2(i), we may assume that Si0 has cardinality at least two. Let k ∈ Si0 , k = k.
Since Nnk ≤ Ni0 k < Ak − 1, it follows that xk fn = 0 for all k0 = k ∈ Si0 .
Also, by Claim 1, we cannot have xk fi0 ∈ (fn ). The only remaining possibility
is that 0 = xk fn ∈ (fi0 ), and therefore Nnl ≥ Ni0 l for all l = k . In particular,
/ Si0 . It follows that conclusion A holds, with
xj fn = 0 for all j ∈
fn
∈ (f1 , . . . , fn ) :S (x1 , . . . , xu ).
xk 0
Indeed, for k ∈ Si0 , k = k0 we have xk fn ∈ (fi0 ), and Nnk0 = Ni0 k0 + 1, from
which we see that
fn
xk ∈ (fi0 ).
xk 0

Claim 5 allows us to rename the variables so that we may assume that


(4.1) x 1 , . . . , xs ∈
/ fn :S (f1 , . . . , fn−1 , fn )

(4.2) xs+1 , . . . , xu ∈ fn :S (f1 , . . . , fn−1 , fn )


We apply the induction hypothesis to {f1 , . . . , fn−1 } with respect to the variables
{x1 , . . . , xs }.
Assume that conclusion B holds for {f1 , . . . , fn−1 } with respect to {x1 , . . . , xs },
but A does not. Let {1, . . . , s} = S1 ∪ . . . ∪ Sn−1

be the partition asserted by B.
We claim that
(4.3) xl f1 = . . . xl fn−1 = 0, for all s + 1 ≤ l ≤ u.
Indeed, assume by way of contradiction that there exists an l ∈ {s + 1, . . . , u} and
an i ≤ n − 1 such that xl fi = 0. Since xl fi ∈ (fn ), we must have Nik ≥ Nnk ∀k = l.
In particular, for k ∈ Si , we have Nik < Ak − 1, and thus Nnk < Ak − 1, which
means that xk fn = 0. Since we may assume that Si has cardinality at least two,
Claim 4 shows that there exists a k ∈ Si with xk fi ∈ (fn ). The fact that both xj fi
and xk fi are nonzero elements in (fn ) contradicts Claim 1.
Equation 4.1 and Claim 4 show that we have two possibilities:
(1) There exists a k ∈ {1, . . . , s} with 0 = xk fn ∈ (fi ), where k ∈ Si , and
xl fn = 0 for all l ∈ {1, . . . , s}, l = k. Then we also have xl fn = 0 for all l ∈
{s + 1, . . . , u}, because Nnl ≥ Nil , and Equation 4.3 shows that Nil = Al − 1. It
follows that conclusion A holds, as
fn
∈ (f1 , . . . , fn ) : (x1 , . . . , xu ).
xk
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 211
11

(2) xk fn = 0 for all k ∈ {1, , . . . , s}. If xl fn = 0 for all l ∈ {s + 1, . . . , u}, then


conclusion B holds for {f1 , . . . , fn }, {x1 , . . . , xu }, with Si = Si for i ≤ n − 1, and
Sn = {s + 1, . . . , u}. Otherwise, assume that xl fn = 0 for some l ∈ {s + 1, . . . , u}.
Use Equation 4.3 to see that xl fi = 0 for all i ∈ {1, . . . , n}, and thus we are done
by induction on the number of variables, by Claim 3.
Now assume that conclusion A holds for {f1 , . . . , fn−1 } with respect to the
variables {x1 , . . . xs }. Without loss of generality, we may assume that
f1
(4.4) ∈ (f1 , f2 , . . . fn−1 ) :S (x1 , . . . , xs ).
x1
If
f1
(4.5) xl ∈ (fn ), for every s + 1 ≤ l ≤ u,
x1
then conclusion A would hold for {f1 , . . . , fn }, {x1 , . . . , xu }, and we would be done.
We know that xl f1 ∈ (fn ) for all s + 1 ≤ l ≤ u by equation 4.2. If xl f1 = 0 for all
s + 1 ≤ l ≤ u, or if N11 > Nn1 then equation 4.5 holds. Without loss of generality
we may assume that
(4.6) N11 ≤ Nn1
and xl f1 = 0 for some s + 1 ≤ l ≤ u. By Claim 1, there exists just one value of l,
say l = s + 1 such that xl f1 = 0 (since we have xl f1 ∈ (fn ) for all l ≥ s + 1). So we
may assume
(4.7) xs+1 f1 = 0, N11 = Nn1 and xl f1 = 0, for all s + 2 ≤ l ≤ u
Claim 6: With the above assumptions, the following holds:
(4.8) x2 f1 = . . . xs f1 = 0
If, say, x2 f1 = 0, then
f1
0 = x2
∈ (fi )
x1
for some i ≤ n − 1, which implies that N11 > Ni1 and N1s+1 ≥ Nis+1 . As, by
equation 4.2, xs+1 fi ∈ (fn ) then we obtain the following two possibilities:
(1) either xs+1 fi = 0, which implies xs+1 f1 = 0, contradicting 4.7; or
(2) Ni1 ≥ Nn1 , which implies N11 > Nn1 , contradicting 4.6.
This proves Claim 6.
Because of Claim 5, we may assume that there exists an index j, such that
1 ≤ j ≤ s and
(4.9) xj ∈ f1 :S (f2 , . . . , fn )
We may assume that
(4.10) x1 f1 = 0, and therefore x1 fn = 0(sinceN11 = Nn1 ).
Otherwise, by 4.7 and 4.8, xl f1 = 0 for all l ∈ {1, . . . , s, s + 2, . . . , u}, and it follows
that condition A holds:
f1
∈ (f1 , . . . , fn ) : (x1 , . . . , xu ).
xs+1
The following cases finish the proof of the theorem.
(1) Assume j = 1. Since x1 fn ∈ (f1 ) and N11 = Nn1 , by 4.7, then x1 fn = 0
contradicting 4.10.
212
12 JANET STRIULI AND ADELA VRACIU

(2) Assume j ≥ 2. We may assume j = 2. By 4.7 and 4.8 we have xl f1 = 0


for all l = 1, s + 1. We may assume that x1 f1 = 0, by 4.10.
(a) Assume that x2 fn = 0. We know x1 ∈ fi :S (f1 , . . . , fn ) for some
i ∈ {1, . . . , n − 1}. As 0 = x2 fn ∈ (f1 ) and x1 fn = 0, by Claim (1) it
follows that 2 ≤ i ≤ n − 1 (because N12 > Nn2 ≥ Ni2 , so i = 1). For
such an i, we claim that
fi
(4.11) ∈ (f1 , . . . , fn ) : (x1 , . . . , xu ).
x1
First notice that Ni1 = Nn1 + 1 = N11 + 1, since 0 = x1 fn ∈ (fi ) and
by 4.6. Moreover, as x2 fn = 0, by multiplying x1 fn by x2 we obtain
that 0 = x2 fi ∈ (f1 ) (we have x2 fi ∈ (f1 ) by equation 4.9, and we
x2 fi
have x2 fi = 0 because Ni2 ≤ Nn2 ). Moreover, ∈ (f1 ), since
x1
Ni1 > N11 ). If xl fi = 0 for some l ∈ / {1, 2, s + 1}, then xl f1 = 0,
contradicting 4.6 and 4.7. As xs+1 ∈ (fn ) : (f1 , . . . , fn ), we obtain
xs+1 fi ∈ (fn ) and since Ni1 = Nn1 + 1 also xs+1 xf1i ∈ (fn ).
(b) Assume that x2 fn = 0. If x2 fi = 0, for all i ∈ {1, . . . , u} then we are
done by Claim 3. So we may assume that there is a t ∈ / {1, n} such
that x2 ft = 0 and x2 ft ∈ (f1 ). Therefore N12 = Nt2 + 1. As xl ft ∈
(fn ) for every s + 1 ≤ l ≤ u, if xl ft = 0 then A2 − 1 = Nn2 ≤ Nt2
which contradicts x2 ft = 0. Therefore we have that xl ft = 0 for all
s + 1 ≤ t ≤ u. Also, as 0 = x2 ft ∈ (f1 ), we have Ntk ≥ N1k for all
k = 2. As xl f1 = 0 for all l
notin{1, s + 1}, it follows that xl ft = 0 for all l ∈ / {1, 2}. If also
x1 ft = 0 then conclusion A holds as
ft
∈ (f1 , . . . , fn ) :S (x1 , . . . , xu ).
x2
Assume that x1 ft = 0. Recall that x1 ∈ fi :S (f1 , . . . , fn ) for some
i ≤ n − 1. We claim that
fi
∈ (f1 , . . . , fn ) :S (x1 , . . . , xu ).
x1
As 0 = x1 ft ∈ (fi ), we have Nil ≤ Ntl for all l = 1. As x2 ft = 0
this implies that x2 fi = 0. As x2 fi ∈ (f1 ) by equation 4.9, and since
xl f1 = 0 for l ∈
/ {1, s+1}, we obtain that xl fi = 0 for l ∈
/ {1, 2, s+1}.
To prove the claim, it is therefore enough to prove that xf1i x2 ∈ (f1 )
and xf1i xs+1 ∈ (fn ). As 0 = x1 f1 ∈ (fi ) we obtain Ni1 = N11 + 1 =
Nn1 + 1, where the last equality follows from 4.7. This, together
with the fact that x2 fi ∈ (f1 ) by equation 4.9, and xs+1 fi ∈ (fn ) by
equation 4.2 concludes the claim.

Now we give the proof of Theorem 4.1
Proof. We may apply Theorem 4.2 to {f1 , . . . , fn }, {x1 , . . . , xd }.
Indeed, the assumption that R = S/0 :S (f1 , . . . , fn ) is almost Gorenstein
implies hypothesis (3) of Theorem 4.2 by Lemma 4.3 . We may assume without loss
of generality that (1) holds by choosing f1 , . . . , fn to be a minimal set of generators
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 213
13

for the ideal they generate. In order to establish hypothesis (2), note that R does
not change if we replace S by S  = k[x1 , . . . , xd ]/(x1A1 +1 , . . . , xdAd +1 ), and f1 , . . . , fn
by f1 , . . . , fn , where fi = (x1 · · · xd )fi .
If (A) holds, then we may apply Theorem 1.4 to conclude that a copy of the
residue field k splits off the second syzygy of ωR . Take K = (f1 , . . . , fn ) ⊂ S. From
conclusion (A) of Theorem 4.2, we have
fi
∈ (K :S mS ) \ (mS K :S mS )
xj
which, by Remark 1.5, implies J :S mS = mS J :S mS , where J = 0 :S K, and now
Theorem 1.4 applies.
If (B) holds, we will check that k is a direct summand of the first syzygy of ωR .
Let S1 , . . . , Sn be the sets asserted in Conclusion (B). We have

1 , . . . , xd ) :S (f1 , . . . , fn ) = (x1 , . . . , xd ) + (xj xj  | j, j not in the same Si ).
Ad Ad
(xA 1 A1

The relations on the generators f1 , . . . , fn of ωR are xj fi = 0 for j ∈/ Si , and


Aj −1 Aj −1
(Πj∈Si xj )fi − (Πj∈Si xj )fi = 0. Note that the latter relations are killed by
the maximal ideal, thus each of them generates a copy of k which splits off the first
syzygy.


1 , . . . , xd ) and let f1 , . . . , fn ∈ S be
Ad
Lemma 4.3. Let S = k[x1 , . . . , xd ]/(xA 1

monomials such that S/0 :S (f1 , . . . , fn ) is almost Gorenstein. Then we have


(x1 , . . . , xd ) ⊆ Σi fi :S (f1 , . . . , fn ).
Proof. We will use Nik to denote the exponent of the variable xk in the
monomial fi .
By Lemma 1.2, the almost Gorenstein assumption implies
xi ∈ f1 :S (f2 , . . . , fn ) + [f1 :S (f2 , . . . , fn )][(f2 , . . . , fn )] :S f1
for all i = 1, . . . , d.
Without loss of generality, we will show that x1 ∈ fj :S (f1 , . . . , fn ) for some
j ∈ {1, . . . , n}.
If x1 fi = 0 for all i = 1, . . . n then the conclusion follows. So there exist a j
such that x1 fj = 0. Denote by S the set of indexes j such that x1 fj = 0. There
are two cases: either xk fj = 0 for all k = 1 and for all j ∈ S or there exists a k = 1
such that xk fj = 0 for some j ∈ S,
In the first case we have that f1 = x1N1 1 xA 2
2 −1
. . . xdAd −1
Assume that for a such that x1 fj = 0 one has xk fj = 0 for all k = 1.
Choose a j (say j = 1) such that x1 fj = 0, and xk fj = 0 for some k = 1. If no
such j exists, it is easy to see that x1 ∈ f1 :S (f2 , . . . , fn ) (where we assume that
x1 f1 = 0 and x2 f1 = . . . = xd f1 = 0).
Assume x1 ∈ / f1 :S (f2 , . . . , fn ). Then there exists a j ∈ {2, . . . , n}, say j = 2,
such that x1 f1 = a2 f2 for some a2 ∈ f1 :S (f2 , . . . , fn ).
Assume that x2 f1 = 0.
We know that x2 f1 ∈ / (f2 ) by comparing the exponents of x1 (N1k ≥ N2k for
k = 1, and N11 = N21 − 1). So x2 ∈ / f2 :S (f1 , . . . , fn ), thus there exists a j such
that x2 f2 = aj fj with aj ∈ f2 :S (f1 , . . . , fn ). Note that j = 1, since x2 f2 ∈ / (f1 )
(by comparing the exponents of xj , j > 2 -if there are more than 2 variables).
214
14 JANET STRIULI AND ADELA VRACIU

We may assume j = 3. We will use aik to denote the exponent of the variable xk
in the monomial ai . We have a2 ∈ f1 :S (f2 , . . . , fn ), so in particular a2 f3 ∈ (f1 ).
This means that either a2 f3 = 0, or, by comparing exponents in each variable,
a2k + N3k ≥ N1k for all k.
We claim that a2 f3 cannot equal zero in S. If a2 f3 = 0, we must have a2k +
N3k ≥ Ak for some k. Since a2k = N1k − N2k for all k = 1, and since N2k ≥ N3k
for k = 2, we see that a2k + N3k ≤ N1k ≤ Ak for all k = 1, 2. For k = 1, we
have a21 = 0, so a21 + N31 = N31 < A1 . For k = 2, we have N22 = N32 − 1, so
a22 + N32 = N12 − N32 + 1 + N32 = N12 + 1, which is less that A2 since we are
assuming x2 f1 = 0. This concludes the proof of the claim.
Now we have a2k + N3k ≥ N1k for all k.
For k = 1, this means N1k − N2k + N3k ≥ N1k , thus N3k ≥ N2k . Since we
already knew the inequality in the other direction, it follows that N3k = N2k for
k = 1, 2. Thus, we have a3k = 0 for k = 1, 2, and we also know that A32 = 0,
a31 + N21 − N31 , where N21 = N11 + 1 and N31 ≥ N11 , so that a31 can be at
most one. It follows that a3 = x1 , and by our assumption on a3 we now have
x1 ∈ f2 :S (f1 , . . . , fn ) as desired. 

References
1. H. Ananthnarayan, The Gorenstein colength of an Artinian local ring, J. Algebra 320 (2008),
no. 9, 3438–3446. MR 2455508 (2009h:13032)
2. Maurice Auslander and Mark Bridger, Stable module theory, Memoirs of the American Mathe-
matical Society, No. 94, American Mathematical Society, Providence, R.I., 1969. MR 0269685
(42 #4580)
3. Luchezar L. Avramov, Vesselin N. Gasharov, and Irena V. Peeva, Complete intersection di-
mension, Inst. Hautes Études Sci. Publ. Math. (1997), no. 86, 67–114 (1998). MR 1608565
(99c:13033)
4. Luchezar L. Avramov, Infinite free resolutions, Six Lectures in Commutative Algebra (Bel-
laterra, 1996), 1–118, Progr. Math., 166, Birkhuser, Basel (1998). MR 1648664 (99m:13022)
5. Luchezar L. Avramov and Alex Martsinkovsky, Absolute, relative, and Tate cohomology of
modules of finite Gorenstein dimension, Proc. London Math. Soc. (3) 85 (2002), no. 2, 393–
440. MR 1912056 (2003g:16009)
6. Winfried Bruns and Jürgen Herzog, Cohen-Macaulay rings, Cambridge Studies in Ad-
vanced Mathematics, vol. 39, Cambridge University Press, Cambridge, 1993. MR 1251956
(95h:13020)
7. Lars Winther Christensen, Gorenstein dimensions, Lecture Notes in Mathematics, vol. 1747,
Springer-Verlag, Berlin, 2000. MR 1799866 (2002e:13032)
8. Lars Winther Christensen, Greg Piepmeyer, Janet Striuli, and Ryo Takahashi, Finite Goren-
stein representation type implies simple singularity, Adv. Math. 218 (2008), no. 4, 1012–1026.
MR 2419377 (2009b:13058)
9. Lars Winther Christensen, Janet Striuli and Oana Veliche, Growth in the minimal injective
resolution of a local ring, J. Lond. Math. Soc. (2), 81, (2010), no. 1, 24–44. MR 2580452
10. Craig Huneke and Adela Vraciu, Rings that are almost Gorenstein, Pacific J. Math. 225
(2006), no. 1, 85–102. MR 2233726 (2007f:13035)
11. Sankar Dutta, Syzygies and homological conjectures, Commutative Algebra (Berkeley, CA,
1987), 139–156, Math. Sci. Res. Inst. Publ. 15, Springer, New York, 1989.
12. David Jorgensen, Meri Hughes, and Liana Sega, Acyclic complexes of finitely generated free
modules over local rings, preprint (2008).
13. David A. Jorgensen and Graham J. Leuschke, On the growth of the Betti sequence of the
canonical module, Math. Z., 256, (2007), no. 3, 647–659. MR 2299575 (2008a:13018)
14. Idun Reiten, Cohen-Macaulay modules over isolated singularities, Séminaire d’algèbre Paul
Dubreil et Marie-Paule Malliavin, 37ème année (Paris, 1985), Lecture Notes in Math., vol.
1220, Springer, Berlin, 1986, pp. 25–32. MR 926295 (89c:14003)
SOME HOMOLOGICAL PROPERTIES OF ALMOST GORENSTEIN RINGS 215
15

15. Takahashi, Ryo, Syzygy modules with semidualizing or G-projective summands, J. Algebra,
295, (2006), no. 1, 179–194. MR 2188856 (2006j:13010)

Department of Mathematics and Computer Science, Fairfield University, Fairfield,


CT 06824
E-mail address: jstriuli@fairfield.edu

Department of Mathematics, University of South Carolina, Columbia, SC 29208


E-mail address: vraciu@math.sc.edu
This volume contains papers based on presentations given at the Pan-American Advanced
Studies Institute (PASI) on commutative algebra and its connections to geometry, which
was held August 3–14, 2009, at the Universidade Federal de Pernambuco in Olinda, Brazil.
The main goal of the program was to detail recent developments in commutative algebra
and interactions with such areas as algebraic geometry, combinatorics and computer
algebra. The articles in this volume concentrate on topics central to modern commutative
algebra: the homological conjectures, problems in positive and mixed characteristic, tight
closure and its interaction with birational geometry, integral dependence and blowup alge-
bras, equisingularity theory, Hilbert functions and multiplicities, combinatorial commuta-
tive algebra, Gröbner bases and computational algebra.

CONM/555 AMS on the Web


www.ams.org